12865 lines
481 KiB
C++
12865 lines
481 KiB
C++
/*
|
|
* Cppcheck - A tool for static C/C++ code analysis
|
|
* Copyright (C) 2007-2022 Cppcheck team.
|
|
*
|
|
* This program is free software: you can redistribute it and/or modify
|
|
* it under the terms of the GNU General Public License as published by
|
|
* the Free Software Foundation, either version 3 of the License, or
|
|
* (at your option) any later version.
|
|
*
|
|
* This program is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
* GNU General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU General Public License
|
|
* along with this program. If not, see <http://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
//---------------------------------------------------------------------------
|
|
#include "tokenize.h"
|
|
|
|
#include "check.h"
|
|
#include "errorlogger.h"
|
|
#include "library.h"
|
|
#include "mathlib.h"
|
|
#include "platform.h"
|
|
#include "preprocessor.h"
|
|
#include "settings.h"
|
|
#include "standards.h"
|
|
#include "summaries.h"
|
|
#include "symboldatabase.h"
|
|
#include "templatesimplifier.h"
|
|
#include "timer.h"
|
|
#include "token.h"
|
|
#include "utils.h"
|
|
#include "valueflow.h"
|
|
|
|
#include <algorithm>
|
|
#include <cassert>
|
|
#include <cctype>
|
|
#include <cstdlib>
|
|
#include <cstring>
|
|
#include <iostream>
|
|
#include <iterator>
|
|
#include <exception>
|
|
#include <memory>
|
|
#include <set>
|
|
#include <stack>
|
|
#include <stdexcept>
|
|
#include <type_traits>
|
|
#include <unordered_map>
|
|
#include <unordered_set>
|
|
#include <utility>
|
|
#include <vector>
|
|
|
|
#include <simplecpp.h>
|
|
|
|
//---------------------------------------------------------------------------
|
|
|
|
namespace {
|
|
// local struct used in setVarId
|
|
// in order to store information about the scope
|
|
struct VarIdScopeInfo {
|
|
VarIdScopeInfo()
|
|
: isExecutable(false), isStructInit(false), isEnum(false), startVarid(0) {}
|
|
VarIdScopeInfo(bool isExecutable, bool isStructInit, bool isEnum, nonneg int startVarid)
|
|
: isExecutable(isExecutable), isStructInit(isStructInit), isEnum(isEnum), startVarid(startVarid) {}
|
|
|
|
const bool isExecutable;
|
|
const bool isStructInit;
|
|
const bool isEnum;
|
|
const nonneg int startVarid;
|
|
};
|
|
}
|
|
|
|
/** Return whether tok is the "{" that starts an enumerator list */
|
|
static bool isEnumStart(const Token* tok)
|
|
{
|
|
if (!tok || tok->str() != "{")
|
|
return false;
|
|
return (tok->strAt(-1) == "enum") || (tok->strAt(-2) == "enum") || Token::Match(tok->tokAt(-3), "enum class %name%");
|
|
}
|
|
|
|
template<typename T>
|
|
static void skipEnumBody(T **tok)
|
|
{
|
|
T *defStart = *tok;
|
|
while (Token::Match(defStart, "%name%|::|:"))
|
|
defStart = defStart->next();
|
|
if (defStart && defStart->str() == "{")
|
|
*tok = defStart->link()->next();
|
|
}
|
|
|
|
const Token * Tokenizer::isFunctionHead(const Token *tok, const std::string &endsWith) const
|
|
{
|
|
return Tokenizer::isFunctionHead(tok, endsWith, isCPP());
|
|
}
|
|
|
|
const Token * Tokenizer::isFunctionHead(const Token *tok, const std::string &endsWith, bool cpp)
|
|
{
|
|
if (!tok)
|
|
return nullptr;
|
|
if (tok->str() == "(")
|
|
tok = tok->link();
|
|
if (Token::Match(tok, ") ;|{|[")) {
|
|
tok = tok->next();
|
|
while (tok && tok->str() == "[" && tok->link()) {
|
|
if (endsWith.find(tok->str()) != std::string::npos)
|
|
return tok;
|
|
tok = tok->link()->next();
|
|
}
|
|
return (tok && endsWith.find(tok->str()) != std::string::npos) ? tok : nullptr;
|
|
}
|
|
if (cpp && tok->str() == ")") {
|
|
tok = tok->next();
|
|
while (Token::Match(tok, "const|noexcept|override|final|volatile|mutable|&|&& !!(") ||
|
|
(Token::Match(tok, "%name% !!(") && tok->isUpperCaseName()))
|
|
tok = tok->next();
|
|
if (tok && tok->str() == ")")
|
|
tok = tok->next();
|
|
while (tok && tok->str() == "[")
|
|
tok = tok->link()->next();
|
|
if (Token::Match(tok, "throw|noexcept ("))
|
|
tok = tok->linkAt(1)->next();
|
|
if (Token::Match(tok, "%name% (") && tok->isUpperCaseName())
|
|
tok = tok->linkAt(1)->next();
|
|
if (tok && tok->originalName() == "->") { // trailing return type
|
|
for (tok = tok->next(); tok && !Token::Match(tok, ";|{|override|final"); tok = tok->next())
|
|
if (tok->link() && Token::Match(tok, "<|[|("))
|
|
tok = tok->link();
|
|
}
|
|
while (Token::Match(tok, "override|final !!(") ||
|
|
(Token::Match(tok, "%name% !!(") && tok->isUpperCaseName()))
|
|
tok = tok->next();
|
|
if (Token::Match(tok, "= 0|default|delete ;"))
|
|
tok = tok->tokAt(2);
|
|
|
|
return (tok && endsWith.find(tok->str()) != std::string::npos) ? tok : nullptr;
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
/**
|
|
* is tok the start brace { of a class, struct, union, or enum
|
|
*/
|
|
static bool isClassStructUnionEnumStart(const Token * tok)
|
|
{
|
|
if (!Token::Match(tok->previous(), "class|struct|union|enum|%name%|>|>> {"))
|
|
return false;
|
|
const Token * tok2 = tok->previous();
|
|
while (tok2 && !Token::Match(tok2, "class|struct|union|enum|{|}|;"))
|
|
tok2 = tok2->previous();
|
|
return Token::Match(tok2, "class|struct|union|enum");
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
|
|
Tokenizer::Tokenizer() :
|
|
list(nullptr),
|
|
mSettings(nullptr),
|
|
mErrorLogger(nullptr),
|
|
mSymbolDatabase(nullptr),
|
|
mTemplateSimplifier(nullptr),
|
|
mVarId(0),
|
|
mUnnamedCount(0),
|
|
mCodeWithTemplates(false), //is there any templates?
|
|
mTimerResults(nullptr)
|
|
#ifdef MAXTIME
|
|
, mMaxTime(std::time(0) + MAXTIME)
|
|
#endif
|
|
, mPreprocessor(nullptr)
|
|
{}
|
|
|
|
Tokenizer::Tokenizer(const Settings *settings, ErrorLogger *errorLogger) :
|
|
list(settings),
|
|
mSettings(settings),
|
|
mErrorLogger(errorLogger),
|
|
mSymbolDatabase(nullptr),
|
|
mTemplateSimplifier(nullptr),
|
|
mVarId(0),
|
|
mUnnamedCount(0),
|
|
mCodeWithTemplates(false), //is there any templates?
|
|
mTimerResults(nullptr)
|
|
#ifdef MAXTIME
|
|
,mMaxTime(std::time(0) + MAXTIME)
|
|
#endif
|
|
, mPreprocessor(nullptr)
|
|
{
|
|
// make sure settings are specified
|
|
assert(mSettings);
|
|
|
|
mTemplateSimplifier = new TemplateSimplifier(this);
|
|
}
|
|
|
|
Tokenizer::~Tokenizer()
|
|
{
|
|
delete mSymbolDatabase;
|
|
delete mTemplateSimplifier;
|
|
}
|
|
|
|
|
|
//---------------------------------------------------------------------------
|
|
// SizeOfType - gives the size of a type
|
|
//---------------------------------------------------------------------------
|
|
|
|
nonneg int Tokenizer::sizeOfType(const Token *type) const
|
|
{
|
|
if (!type || type->str().empty())
|
|
return 0;
|
|
|
|
if (type->tokType() == Token::eString)
|
|
return Token::getStrLength(type) + 1U;
|
|
|
|
const std::map<std::string, int>::const_iterator it = mTypeSize.find(type->str());
|
|
if (it == mTypeSize.end()) {
|
|
const Library::PodType* podtype = mSettings->library.podtype(type->str());
|
|
if (!podtype)
|
|
return 0;
|
|
|
|
return podtype->size;
|
|
} else if (type->isLong()) {
|
|
if (type->str() == "double")
|
|
return mSettings->sizeof_long_double;
|
|
else if (type->str() == "long")
|
|
return mSettings->sizeof_long_long;
|
|
}
|
|
|
|
return it->second;
|
|
}
|
|
//---------------------------------------------------------------------------
|
|
|
|
// check if this statement is a duplicate definition
|
|
bool Tokenizer::duplicateTypedef(Token **tokPtr, const Token *name, const Token *typeDef) const
|
|
{
|
|
// check for an end of definition
|
|
const Token * tok = *tokPtr;
|
|
if (tok && Token::Match(tok->next(), ";|,|[|=|)|>|(|{")) {
|
|
const Token * end = tok->next();
|
|
|
|
if (end->str() == "[") {
|
|
if (!end->link())
|
|
syntaxError(end); // invalid code
|
|
end = end->link()->next();
|
|
} else if (end->str() == ",") {
|
|
// check for derived class
|
|
if (Token::Match(tok->previous(), "public|private|protected"))
|
|
return false;
|
|
|
|
// find end of definition
|
|
while (end && end->next() && !Token::Match(end->next(), ";|)|>")) {
|
|
if (end->next()->str() == "(")
|
|
end = end->linkAt(1);
|
|
|
|
end = (end)?end->next():nullptr;
|
|
}
|
|
if (end)
|
|
end = end->next();
|
|
} else if (end->str() == "(") {
|
|
if (tok->previous()->str().compare(0, 8, "operator") == 0) {
|
|
// conversion operator
|
|
return false;
|
|
} else if (tok->previous()->str() == "typedef") {
|
|
// typedef of function returning this type
|
|
return false;
|
|
} else if (Token::Match(tok->previous(), "public:|private:|protected:")) {
|
|
return false;
|
|
} else if (tok->previous()->str() == ">") {
|
|
if (!Token::Match(tok->tokAt(-2), "%type%"))
|
|
return false;
|
|
|
|
if (!Token::Match(tok->tokAt(-3), ",|<"))
|
|
return false;
|
|
|
|
*tokPtr = end->link();
|
|
return true;
|
|
}
|
|
}
|
|
|
|
if (end) {
|
|
if (Token::simpleMatch(end, ") {")) { // function parameter ?
|
|
// look backwards
|
|
if (Token::Match(tok->previous(), "%type%") &&
|
|
!Token::Match(tok->previous(), "return|new|const|struct")) {
|
|
// duplicate definition so skip entire function
|
|
*tokPtr = end->next()->link();
|
|
return true;
|
|
}
|
|
} else if (end->str() == ">") { // template parameter ?
|
|
// look backwards
|
|
if (Token::Match(tok->previous(), "%type%") &&
|
|
!Token::Match(tok->previous(), "return|new|const|volatile")) {
|
|
// duplicate definition so skip entire template
|
|
while (end && end->str() != "{")
|
|
end = end->next();
|
|
if (end) {
|
|
*tokPtr = end->link();
|
|
return true;
|
|
}
|
|
}
|
|
} else {
|
|
// look backwards
|
|
if (Token::Match(tok->previous(), "typedef|}|>") ||
|
|
(end->str() == ";" && tok->previous()->str() == ",") ||
|
|
(tok->previous()->str() == "*" && tok->next()->str() != "(") ||
|
|
(Token::Match(tok->previous(), "%type%") &&
|
|
(!Token::Match(tok->previous(), "return|new|const|friend|public|private|protected|throw|extern") &&
|
|
!Token::simpleMatch(tok->tokAt(-2), "friend class")))) {
|
|
// scan backwards for the end of the previous statement
|
|
while (tok && tok->previous() && !Token::Match(tok->previous(), ";|{")) {
|
|
if (tok->previous()->str() == "}") {
|
|
tok = tok->previous()->link();
|
|
} else if (tok->previous()->str() == "typedef") {
|
|
return true;
|
|
} else if (tok->previous()->str() == "enum") {
|
|
return true;
|
|
} else if (tok->previous()->str() == "struct") {
|
|
if (tok->strAt(-2) == "typedef" &&
|
|
tok->next()->str() == "{" &&
|
|
typeDef->strAt(3) != "{") {
|
|
// declaration after forward declaration
|
|
return true;
|
|
} else if (tok->next()->str() == "{") {
|
|
return true;
|
|
} else if (Token::Match(tok->next(), ")|*")) {
|
|
return true;
|
|
} else if (tok->next()->str() == name->str()) {
|
|
return true;
|
|
} else if (tok->next()->str() != ";") {
|
|
return true;
|
|
} else {
|
|
return false;
|
|
}
|
|
} else if (tok->previous()->str() == "union") {
|
|
if (tok->next()->str() != ";") {
|
|
return true;
|
|
} else {
|
|
return false;
|
|
}
|
|
} else if (isCPP() && tok->previous()->str() == "class") {
|
|
if (tok->next()->str() != ";") {
|
|
return true;
|
|
} else {
|
|
return false;
|
|
}
|
|
}
|
|
if (tok)
|
|
tok = tok->previous();
|
|
}
|
|
|
|
if ((*tokPtr)->strAt(1) != "(" || !Token::Match((*tokPtr)->linkAt(1), ") .|(|["))
|
|
return true;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
void Tokenizer::unsupportedTypedef(const Token *tok) const
|
|
{
|
|
if (!mSettings->debugwarnings)
|
|
return;
|
|
|
|
std::ostringstream str;
|
|
const Token *tok1 = tok;
|
|
int level = 0;
|
|
while (tok) {
|
|
if (level == 0 && tok->str() == ";")
|
|
break;
|
|
else if (tok->str() == "{")
|
|
++level;
|
|
else if (tok->str() == "}") {
|
|
if (level == 0)
|
|
break;
|
|
--level;
|
|
}
|
|
|
|
if (tok != tok1)
|
|
str << " ";
|
|
str << tok->str();
|
|
tok = tok->next();
|
|
}
|
|
if (tok)
|
|
str << " ;";
|
|
|
|
reportError(tok1, Severity::debug, "simplifyTypedef",
|
|
"Failed to parse \'" + str.str() + "\'. The checking continues anyway.");
|
|
}
|
|
|
|
Token * Tokenizer::deleteInvalidTypedef(Token *typeDef)
|
|
{
|
|
Token *tok = nullptr;
|
|
|
|
// remove typedef but leave ;
|
|
while (typeDef->next()) {
|
|
if (typeDef->next()->str() == ";") {
|
|
typeDef->deleteNext();
|
|
break;
|
|
} else if (typeDef->next()->str() == "{")
|
|
Token::eraseTokens(typeDef, typeDef->linkAt(1));
|
|
else if (typeDef->next()->str() == "}")
|
|
break;
|
|
typeDef->deleteNext();
|
|
}
|
|
|
|
if (typeDef != list.front()) {
|
|
tok = typeDef->previous();
|
|
tok->deleteNext();
|
|
} else {
|
|
list.front()->deleteThis();
|
|
tok = list.front();
|
|
}
|
|
|
|
return tok;
|
|
}
|
|
|
|
namespace {
|
|
struct Space {
|
|
Space() : bodyEnd(nullptr), bodyEnd2(nullptr), isNamespace(false) {}
|
|
std::string className;
|
|
const Token * bodyEnd; // for body contains typedef define
|
|
const Token * bodyEnd2; // for body contains typedef using
|
|
bool isNamespace;
|
|
std::set<std::string> recordTypes;
|
|
};
|
|
}
|
|
|
|
static Token *splitDefinitionFromTypedef(Token *tok, nonneg int *unnamedCount)
|
|
{
|
|
std::string name;
|
|
bool isConst = false;
|
|
Token *tok1 = tok->next();
|
|
|
|
// skip const if present
|
|
if (tok1->str() == "const") {
|
|
tok1->deleteThis();
|
|
isConst = true;
|
|
}
|
|
|
|
// skip "class|struct|union|enum"
|
|
tok1 = tok1->next();
|
|
|
|
const bool hasName = Token::Match(tok1, "%name%");
|
|
|
|
// skip name
|
|
if (hasName) {
|
|
name = tok1->str();
|
|
tok1 = tok1->next();
|
|
}
|
|
|
|
// skip base classes if present
|
|
if (tok1->str() == ":") {
|
|
tok1 = tok1->next();
|
|
while (tok1 && tok1->str() != "{")
|
|
tok1 = tok1->next();
|
|
if (!tok1)
|
|
return nullptr;
|
|
}
|
|
|
|
// skip to end
|
|
tok1 = tok1->link();
|
|
|
|
if (!hasName) { // unnamed
|
|
if (tok1->next()) {
|
|
// use typedef name if available
|
|
if (Token::Match(tok1->next(), "%type%"))
|
|
name = tok1->next()->str();
|
|
else // create a unique name
|
|
name = "Unnamed" + MathLib::toString((*unnamedCount)++);
|
|
tok->next()->insertToken(name);
|
|
} else
|
|
return nullptr;
|
|
}
|
|
|
|
tok1->insertToken(";");
|
|
tok1 = tok1->next();
|
|
|
|
if (tok1->next() && tok1->next()->str() == ";" && tok1->previous()->str() == "}") {
|
|
tok->deleteThis();
|
|
tok1->deleteThis();
|
|
return nullptr;
|
|
} else {
|
|
tok1->insertToken("typedef");
|
|
tok1 = tok1->next();
|
|
Token * tok3 = tok1;
|
|
if (isConst) {
|
|
tok1->insertToken("const");
|
|
tok1 = tok1->next();
|
|
}
|
|
tok1->insertToken(tok->next()->str()); // struct, union or enum
|
|
tok1 = tok1->next();
|
|
tok1->insertToken(name);
|
|
tok->deleteThis();
|
|
tok = tok3;
|
|
}
|
|
|
|
return tok;
|
|
}
|
|
|
|
/* This function is called when processing function related typedefs.
|
|
* If simplifyTypedef generates an "Internal Error" message and the
|
|
* code that generated it deals in some way with functions, then this
|
|
* function will probably need to be extended to handle a new function
|
|
* related pattern */
|
|
Token *Tokenizer::processFunc(Token *tok2, bool inOperator) const
|
|
{
|
|
if (tok2->next() && tok2->next()->str() != ")" &&
|
|
tok2->next()->str() != ",") {
|
|
// skip over tokens for some types of canonicalization
|
|
if (Token::Match(tok2->next(), "( * %type% ) ("))
|
|
tok2 = tok2->linkAt(5);
|
|
else if (Token::Match(tok2->next(), "* ( * %type% ) ("))
|
|
tok2 = tok2->linkAt(6);
|
|
else if (Token::Match(tok2->next(), "* ( * %type% ) ;"))
|
|
tok2 = tok2->tokAt(5);
|
|
else if (Token::Match(tok2->next(), "* ( %type% [") &&
|
|
Token::Match(tok2->linkAt(4), "] ) ;|="))
|
|
tok2 = tok2->linkAt(4)->next();
|
|
else if (Token::Match(tok2->next(), "* ( * %type% ("))
|
|
tok2 = tok2->linkAt(5)->next();
|
|
else if (Token::simpleMatch(tok2->next(), "* [") &&
|
|
Token::simpleMatch(tok2->linkAt(2), "] ;"))
|
|
tok2 = tok2->next();
|
|
else {
|
|
if (tok2->next()->str() == "(")
|
|
tok2 = tok2->next()->link();
|
|
else if (!inOperator && !Token::Match(tok2->next(), "[|>|;")) {
|
|
tok2 = tok2->next();
|
|
|
|
while (Token::Match(tok2, "*|&") &&
|
|
!Token::Match(tok2->next(), ")|>"))
|
|
tok2 = tok2->next();
|
|
|
|
// skip over namespace
|
|
while (Token::Match(tok2, "%name% ::"))
|
|
tok2 = tok2->tokAt(2);
|
|
|
|
if (!tok2)
|
|
return nullptr;
|
|
|
|
if (tok2->str() == "(" &&
|
|
tok2->link()->next() &&
|
|
tok2->link()->next()->str() == "(") {
|
|
tok2 = tok2->link();
|
|
|
|
if (tok2->next()->str() == "(")
|
|
tok2 = tok2->next()->link();
|
|
}
|
|
|
|
// skip over typedef parameter
|
|
if (tok2->next() && tok2->next()->str() == "(") {
|
|
tok2 = tok2->next()->link();
|
|
if (!tok2->next())
|
|
syntaxError(tok2);
|
|
|
|
if (tok2->next()->str() == "(")
|
|
tok2 = tok2->next()->link();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return tok2;
|
|
}
|
|
|
|
void Tokenizer::simplifyUsingToTypedef()
|
|
{
|
|
if (!isCPP() || mSettings->standards.cpp < Standards::CPP11)
|
|
return;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
// using a::b; => typedef a::b b;
|
|
if ((Token::Match(tok, "[;{}] using %name% :: %name% ::|;") && !tok->tokAt(2)->isKeyword()) ||
|
|
(Token::Match(tok, "[;{}] using :: %name% :: %name% ::|;") && !tok->tokAt(3)->isKeyword())) {
|
|
Token *endtok = tok->tokAt(5);
|
|
if (Token::Match(endtok, "%name%"))
|
|
endtok = endtok->next();
|
|
while (Token::Match(endtok, ":: %name%"))
|
|
endtok = endtok->tokAt(2);
|
|
if (endtok && endtok->str() == ";") {
|
|
tok->next()->str("typedef");
|
|
endtok = endtok->previous();
|
|
endtok->insertToken(endtok->str());
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyTypedef()
|
|
{
|
|
std::vector<Space> spaceInfo;
|
|
bool isNamespace = false;
|
|
std::string className;
|
|
std::string fullClassName;
|
|
bool hasClass = false;
|
|
bool goback = false;
|
|
|
|
// add global namespace
|
|
spaceInfo.emplace_back(Space{});
|
|
|
|
// Convert "using a::b;" to corresponding typedef statements
|
|
simplifyUsingToTypedef();
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (mErrorLogger && !list.getFiles().empty())
|
|
mErrorLogger->reportProgress(list.getFiles()[0], "Tokenize (typedef)", tok->progressValue());
|
|
|
|
if (Settings::terminated())
|
|
return;
|
|
|
|
if (isMaxTime())
|
|
return;
|
|
|
|
if (goback) {
|
|
//jump back once, see the comment at the end of the function
|
|
goback = false;
|
|
tok = tok->previous();
|
|
}
|
|
|
|
if (tok->str() != "typedef") {
|
|
if (Token::simpleMatch(tok, "( typedef")) {
|
|
// Skip typedefs inside parentheses (#2453 and #4002)
|
|
tok = tok->next();
|
|
} else if (Token::Match(tok, "class|struct|namespace %any%") &&
|
|
(!tok->previous() || tok->previous()->str() != "enum")) {
|
|
isNamespace = (tok->str() == "namespace");
|
|
hasClass = true;
|
|
className = tok->next()->str();
|
|
const Token *tok1 = tok->next();
|
|
fullClassName = className;
|
|
while (Token::Match(tok1, "%name% :: %name%")) {
|
|
tok1 = tok1->tokAt(2);
|
|
fullClassName += " :: " + tok1->str();
|
|
}
|
|
} else if (hasClass && tok->str() == ";") {
|
|
hasClass = false;
|
|
} else if (hasClass && tok->str() == "{") {
|
|
if (!isNamespace)
|
|
spaceInfo.back().recordTypes.insert(fullClassName);
|
|
|
|
Space info;
|
|
info.isNamespace = isNamespace;
|
|
info.className = className;
|
|
info.bodyEnd = tok->link();
|
|
info.bodyEnd2 = tok->link();
|
|
spaceInfo.push_back(info);
|
|
|
|
hasClass = false;
|
|
} else if (spaceInfo.size() > 1 && tok->str() == "}" && spaceInfo.back().bodyEnd == tok) {
|
|
spaceInfo.pop_back();
|
|
}
|
|
continue;
|
|
}
|
|
|
|
// pull struct, union, enum or class definition out of typedef
|
|
// use typedef name for unnamed struct, union, enum or class
|
|
if (Token::Match(tok->next(), "const| struct|enum|union|class %type%| {|:")) {
|
|
Token *tok1 = splitDefinitionFromTypedef(tok, &mUnnamedCount);
|
|
if (!tok1)
|
|
continue;
|
|
tok = tok1;
|
|
}
|
|
|
|
/** @todo add support for union */
|
|
if (Token::Match(tok->next(), "enum %type% %type% ;") && tok->strAt(2) == tok->strAt(3)) {
|
|
tok->deleteNext(3);
|
|
tok->deleteThis();
|
|
if (tok->next())
|
|
tok->deleteThis();
|
|
//now the next token to process is 'tok', not 'tok->next()';
|
|
goback = true;
|
|
continue;
|
|
}
|
|
|
|
Token *typeName;
|
|
Token *typeStart = nullptr;
|
|
Token *typeEnd = nullptr;
|
|
Token *argStart = nullptr;
|
|
Token *argEnd = nullptr;
|
|
Token *arrayStart = nullptr;
|
|
Token *arrayEnd = nullptr;
|
|
Token *specStart = nullptr;
|
|
Token *specEnd = nullptr;
|
|
Token *typeDef = tok;
|
|
Token *argFuncRetStart = nullptr;
|
|
Token *argFuncRetEnd = nullptr;
|
|
Token *funcStart = nullptr;
|
|
Token *funcEnd = nullptr;
|
|
Token *tokOffset = tok->next();
|
|
bool function = false;
|
|
bool functionPtr = false;
|
|
bool functionRetFuncPtr = false;
|
|
bool functionPtrRetFuncPtr = false;
|
|
bool ptrToArray = false;
|
|
bool refToArray = false;
|
|
bool ptrMember = false;
|
|
bool typeOf = false;
|
|
Token *namespaceStart = nullptr;
|
|
Token *namespaceEnd = nullptr;
|
|
|
|
// check for invalid input
|
|
if (!tokOffset)
|
|
syntaxError(tok);
|
|
|
|
|
|
if (tokOffset->str() == "::") {
|
|
typeStart = tokOffset;
|
|
tokOffset = tokOffset->next();
|
|
|
|
while (Token::Match(tokOffset, "%type% ::"))
|
|
tokOffset = tokOffset->tokAt(2);
|
|
|
|
typeEnd = tokOffset;
|
|
|
|
if (Token::Match(tokOffset, "%type%"))
|
|
tokOffset = tokOffset->next();
|
|
} else if (Token::Match(tokOffset, "%type% ::")) {
|
|
typeStart = tokOffset;
|
|
|
|
do {
|
|
tokOffset = tokOffset->tokAt(2);
|
|
} while (Token::Match(tokOffset, "%type% ::"));
|
|
|
|
typeEnd = tokOffset;
|
|
|
|
if (Token::Match(tokOffset, "%type%"))
|
|
tokOffset = tokOffset->next();
|
|
} else if (Token::Match(tokOffset, "%type%")) {
|
|
typeStart = tokOffset;
|
|
|
|
while (Token::Match(tokOffset, "const|struct|enum %type%") ||
|
|
(tokOffset->next() && tokOffset->next()->isStandardType()))
|
|
tokOffset = tokOffset->next();
|
|
|
|
typeEnd = tokOffset;
|
|
tokOffset = tokOffset->next();
|
|
|
|
while (Token::Match(tokOffset, "%type%") &&
|
|
(tokOffset->isStandardType() || Token::Match(tokOffset, "unsigned|signed"))) {
|
|
typeEnd = tokOffset;
|
|
tokOffset = tokOffset->next();
|
|
}
|
|
|
|
bool atEnd = false;
|
|
while (!atEnd) {
|
|
if (tokOffset && tokOffset->str() == "::") {
|
|
typeEnd = tokOffset;
|
|
tokOffset = tokOffset->next();
|
|
}
|
|
|
|
if (Token::Match(tokOffset, "%type%") &&
|
|
tokOffset->next() && !Token::Match(tokOffset->next(), "[|;|,|(")) {
|
|
typeEnd = tokOffset;
|
|
tokOffset = tokOffset->next();
|
|
} else if (Token::simpleMatch(tokOffset, "const (")) {
|
|
typeEnd = tokOffset;
|
|
tokOffset = tokOffset->next();
|
|
atEnd = true;
|
|
} else
|
|
atEnd = true;
|
|
}
|
|
} else
|
|
continue; // invalid input
|
|
|
|
// check for invalid input
|
|
if (!tokOffset)
|
|
syntaxError(tok);
|
|
|
|
// check for template
|
|
if (!isC() && tokOffset->str() == "<") {
|
|
typeEnd = tokOffset->findClosingBracket();
|
|
|
|
while (typeEnd && Token::Match(typeEnd->next(), ":: %type%"))
|
|
typeEnd = typeEnd->tokAt(2);
|
|
|
|
if (!typeEnd) {
|
|
// internal error
|
|
return;
|
|
}
|
|
|
|
while (Token::Match(typeEnd->next(), "const|volatile"))
|
|
typeEnd = typeEnd->next();
|
|
|
|
tok = typeEnd;
|
|
tokOffset = tok->next();
|
|
}
|
|
|
|
std::list<std::string> pointers;
|
|
// check for pointers and references
|
|
while (Token::Match(tokOffset, "*|&|&&|const")) {
|
|
pointers.push_back(tokOffset->str());
|
|
tokOffset = tokOffset->next();
|
|
}
|
|
|
|
// check for invalid input
|
|
if (!tokOffset)
|
|
syntaxError(tok);
|
|
|
|
if (tokOffset->isName() && !tokOffset->isKeyword()) {
|
|
// found the type name
|
|
typeName = tokOffset;
|
|
tokOffset = tokOffset->next();
|
|
|
|
// check for array
|
|
while (tokOffset && tokOffset->str() == "[") {
|
|
if (!arrayStart)
|
|
arrayStart = tokOffset;
|
|
arrayEnd = tokOffset->link();
|
|
tokOffset = arrayEnd->next();
|
|
}
|
|
|
|
// check for end or another
|
|
if (Token::Match(tokOffset, ";|,"))
|
|
tok = tokOffset;
|
|
|
|
// or a function typedef
|
|
else if (tokOffset && tokOffset->str() == "(") {
|
|
Token *tokOffset2 = nullptr;
|
|
if (Token::Match(tokOffset, "( *|%name%")) {
|
|
tokOffset2 = tokOffset->next();
|
|
if (tokOffset2->str() == "typename")
|
|
tokOffset2 = tokOffset2->next();
|
|
while (Token::Match(tokOffset2, "%type% ::"))
|
|
tokOffset2 = tokOffset2->tokAt(2);
|
|
}
|
|
|
|
// unhandled typedef, skip it and continue
|
|
if (typeName->str() == "void") {
|
|
unsupportedTypedef(typeDef);
|
|
tok = deleteInvalidTypedef(typeDef);
|
|
if (tok == list.front())
|
|
//now the next token to process is 'tok', not 'tok->next()';
|
|
goback = true;
|
|
continue;
|
|
}
|
|
|
|
// function pointer
|
|
else if (Token::Match(tokOffset2, "* %name% ) (")) {
|
|
// name token wasn't a name, it was part of the type
|
|
typeEnd = typeEnd->next();
|
|
functionPtr = true;
|
|
funcStart = funcEnd = tokOffset2; // *
|
|
tokOffset = tokOffset2->tokAt(3); // (
|
|
typeName = tokOffset->tokAt(-2);
|
|
argStart = tokOffset;
|
|
argEnd = tokOffset->link();
|
|
tok = argEnd->next();
|
|
}
|
|
|
|
// function
|
|
else if (isFunctionHead(tokOffset->link(), ";,")) {
|
|
function = true;
|
|
if (tokOffset->link()->next()->str() == "const") {
|
|
specStart = tokOffset->link()->next();
|
|
specEnd = specStart;
|
|
}
|
|
argStart = tokOffset;
|
|
argEnd = tokOffset->link();
|
|
tok = argEnd->next();
|
|
if (specStart)
|
|
tok = tok->next();
|
|
}
|
|
|
|
// syntax error
|
|
else
|
|
syntaxError(tok);
|
|
}
|
|
|
|
// unhandled typedef, skip it and continue
|
|
else {
|
|
unsupportedTypedef(typeDef);
|
|
tok = deleteInvalidTypedef(typeDef);
|
|
if (tok == list.front())
|
|
//now the next token to process is 'tok', not 'tok->next()';
|
|
goback = true;
|
|
continue;
|
|
}
|
|
}
|
|
|
|
// typeof: typedef typeof ( ... ) type;
|
|
else if (Token::simpleMatch(tokOffset->previous(), "typeof (") &&
|
|
Token::Match(tokOffset->link(), ") %type% ;")) {
|
|
argStart = tokOffset;
|
|
argEnd = tokOffset->link();
|
|
typeName = tokOffset->link()->next();
|
|
tok = typeName->next();
|
|
typeOf = true;
|
|
}
|
|
|
|
// function: typedef ... ( ... type )( ... );
|
|
// typedef ... (( ... type )( ... ));
|
|
// typedef ... ( * ( ... type )( ... ));
|
|
else if (tokOffset->str() == "(" && (
|
|
(tokOffset->link() && Token::Match(tokOffset->link()->previous(), "%type% ) (") &&
|
|
Token::Match(tokOffset->link()->next()->link(), ") const|volatile|;")) ||
|
|
(Token::simpleMatch(tokOffset, "( (") &&
|
|
tokOffset->next() && Token::Match(tokOffset->next()->link()->previous(), "%type% ) (") &&
|
|
Token::Match(tokOffset->next()->link()->next()->link(), ") const|volatile| ) ;|,")) ||
|
|
(Token::simpleMatch(tokOffset, "( * (") &&
|
|
tokOffset->linkAt(2) && Token::Match(tokOffset->linkAt(2)->previous(), "%type% ) (") &&
|
|
Token::Match(tokOffset->linkAt(2)->next()->link(), ") const|volatile| ) ;|,")))) {
|
|
if (tokOffset->next()->str() == "(")
|
|
tokOffset = tokOffset->next();
|
|
else if (Token::simpleMatch(tokOffset, "( * (")) {
|
|
pointers.emplace_back("*");
|
|
tokOffset = tokOffset->tokAt(2);
|
|
}
|
|
|
|
if (tokOffset->link()->strAt(-2) == "*")
|
|
functionPtr = true;
|
|
else
|
|
function = true;
|
|
funcStart = tokOffset->next();
|
|
tokOffset = tokOffset->link();
|
|
funcEnd = tokOffset->tokAt(-2);
|
|
typeName = tokOffset->previous();
|
|
argStart = tokOffset->next();
|
|
argEnd = tokOffset->next()->link();
|
|
if (!argEnd)
|
|
syntaxError(argStart);
|
|
|
|
tok = argEnd->next();
|
|
Token *spec = tok;
|
|
if (Token::Match(spec, "const|volatile")) {
|
|
specStart = spec;
|
|
specEnd = spec;
|
|
while (Token::Match(spec->next(), "const|volatile")) {
|
|
specEnd = spec->next();
|
|
spec = specEnd;
|
|
}
|
|
tok = specEnd->next();
|
|
}
|
|
if (!tok)
|
|
syntaxError(specEnd);
|
|
|
|
if (tok->str() == ")")
|
|
tok = tok->next();
|
|
}
|
|
|
|
else if (Token::Match(tokOffset, "( %type% (")) {
|
|
function = true;
|
|
if (tokOffset->link()->next()) {
|
|
tok = tokOffset->link()->next();
|
|
tokOffset = tokOffset->tokAt(2);
|
|
typeName = tokOffset->previous();
|
|
argStart = tokOffset;
|
|
argEnd = tokOffset->link();
|
|
} else {
|
|
// internal error
|
|
continue;
|
|
}
|
|
}
|
|
|
|
// pointer to function returning pointer to function
|
|
else if (Token::Match(tokOffset, "( * ( * %type% ) (") &&
|
|
Token::simpleMatch(tokOffset->linkAt(6), ") ) (") &&
|
|
Token::Match(tokOffset->linkAt(6)->linkAt(2), ") ;|,")) {
|
|
functionPtrRetFuncPtr = true;
|
|
|
|
tokOffset = tokOffset->tokAt(6);
|
|
typeName = tokOffset->tokAt(-2);
|
|
argStart = tokOffset;
|
|
argEnd = tokOffset->link();
|
|
if (!argEnd)
|
|
syntaxError(arrayStart);
|
|
|
|
argFuncRetStart = argEnd->tokAt(2);
|
|
argFuncRetEnd = argFuncRetStart->link();
|
|
if (!argFuncRetEnd)
|
|
syntaxError(argFuncRetStart);
|
|
|
|
tok = argFuncRetEnd->next();
|
|
}
|
|
|
|
// function returning pointer to function
|
|
else if (Token::Match(tokOffset, "( * %type% (") &&
|
|
Token::simpleMatch(tokOffset->linkAt(3), ") ) (") &&
|
|
Token::Match(tokOffset->linkAt(3)->linkAt(2), ") ;|,")) {
|
|
functionRetFuncPtr = true;
|
|
|
|
tokOffset = tokOffset->tokAt(3);
|
|
typeName = tokOffset->previous();
|
|
argStart = tokOffset;
|
|
argEnd = tokOffset->link();
|
|
|
|
argFuncRetStart = argEnd->tokAt(2);
|
|
if (!argFuncRetStart)
|
|
syntaxError(tokOffset);
|
|
|
|
argFuncRetEnd = argFuncRetStart->link();
|
|
if (!argFuncRetEnd)
|
|
syntaxError(tokOffset);
|
|
|
|
tok = argFuncRetEnd->next();
|
|
} else if (Token::Match(tokOffset, "( * ( %type% ) (")) {
|
|
functionRetFuncPtr = true;
|
|
|
|
tokOffset = tokOffset->tokAt(5);
|
|
typeName = tokOffset->tokAt(-2);
|
|
argStart = tokOffset;
|
|
argEnd = tokOffset->link();
|
|
if (!argEnd)
|
|
syntaxError(arrayStart);
|
|
|
|
argFuncRetStart = argEnd->tokAt(2);
|
|
if (!argFuncRetStart)
|
|
syntaxError(tokOffset);
|
|
|
|
argFuncRetEnd = argFuncRetStart->link();
|
|
if (!argFuncRetEnd)
|
|
syntaxError(tokOffset);
|
|
|
|
tok = argFuncRetEnd->next();
|
|
}
|
|
|
|
// pointer/reference to array
|
|
else if (Token::Match(tokOffset, "( *|& %type% ) [")) {
|
|
ptrToArray = (tokOffset->next()->str() == "*");
|
|
refToArray = !ptrToArray;
|
|
tokOffset = tokOffset->tokAt(2);
|
|
typeName = tokOffset;
|
|
arrayStart = tokOffset->tokAt(2);
|
|
arrayEnd = arrayStart->link();
|
|
if (!arrayEnd)
|
|
syntaxError(arrayStart);
|
|
|
|
tok = arrayEnd->next();
|
|
}
|
|
|
|
// pointer to class member
|
|
else if (Token::Match(tokOffset, "( %type% :: * %type% ) ;")) {
|
|
tokOffset = tokOffset->tokAt(2);
|
|
namespaceStart = tokOffset->previous();
|
|
namespaceEnd = tokOffset;
|
|
ptrMember = true;
|
|
tokOffset = tokOffset->tokAt(2);
|
|
typeName = tokOffset;
|
|
tok = tokOffset->tokAt(2);
|
|
}
|
|
|
|
// unhandled typedef, skip it and continue
|
|
else {
|
|
unsupportedTypedef(typeDef);
|
|
tok = deleteInvalidTypedef(typeDef);
|
|
if (tok == list.front())
|
|
//now the next token to process is 'tok', not 'tok->next()';
|
|
goback = true;
|
|
continue;
|
|
}
|
|
|
|
bool done = false;
|
|
bool ok = true;
|
|
|
|
TypedefInfo typedefInfo;
|
|
typedefInfo.name = typeName->str();
|
|
typedefInfo.filename = list.file(typeName);
|
|
typedefInfo.lineNumber = typeName->linenr();
|
|
typedefInfo.column = typeName->column();
|
|
typedefInfo.used = false;
|
|
mTypedefInfo.push_back(typedefInfo);
|
|
|
|
while (!done) {
|
|
std::string pattern = typeName->str();
|
|
int scope = 0;
|
|
bool simplifyType = false;
|
|
bool inMemberFunc = false;
|
|
int memberScope = 0;
|
|
bool globalScope = false;
|
|
int classLevel = spaceInfo.size();
|
|
bool inTypeDef = false;
|
|
bool inEnumClass = false;
|
|
std::string removed;
|
|
std::string classPath;
|
|
for (size_t i = 1; i < spaceInfo.size(); ++i) {
|
|
if (!classPath.empty())
|
|
classPath += " :: ";
|
|
classPath += spaceInfo[i].className;
|
|
}
|
|
|
|
for (Token *tok2 = tok; tok2; tok2 = tok2->next()) {
|
|
if (Settings::terminated())
|
|
return;
|
|
|
|
removed.clear();
|
|
|
|
if (Token::simpleMatch(tok2, "typedef"))
|
|
inTypeDef = true;
|
|
|
|
if (inTypeDef && Token::simpleMatch(tok2, ";"))
|
|
inTypeDef = false;
|
|
|
|
// Check for variable declared with the same name
|
|
if (!inTypeDef && spaceInfo.size() == 1 && Token::Match(tok2->previous(), "%name%") &&
|
|
!tok2->previous()->isKeyword()) {
|
|
Token* varDecl = tok2;
|
|
while (Token::Match(varDecl, "*|&|&&|const"))
|
|
varDecl = varDecl->next();
|
|
if (Token::Match(varDecl, "%name% ;|,|)|=") && varDecl->str() == typeName->str()) {
|
|
// Skip to the next closing brace
|
|
if (Token::Match(varDecl, "%name% ) {")) { // is argument variable
|
|
tok2 = varDecl->linkAt(2)->next();
|
|
} else {
|
|
tok2 = varDecl;
|
|
while (tok2 && !Token::simpleMatch(tok2, "}")) {
|
|
if (Token::Match(tok2, "(|{|["))
|
|
tok2 = tok2->link();
|
|
tok2 = tok2->next();
|
|
}
|
|
}
|
|
if (!tok2)
|
|
break;
|
|
continue;
|
|
}
|
|
}
|
|
|
|
if (tok2->link()) { // Pre-check for performance
|
|
// check for end of scope
|
|
if (tok2->str() == "}") {
|
|
// check for end of member function
|
|
if (inMemberFunc) {
|
|
--memberScope;
|
|
if (memberScope == 0)
|
|
inMemberFunc = false;
|
|
}
|
|
inEnumClass = false;
|
|
|
|
if (classLevel > 1 && tok2 == spaceInfo[classLevel - 1].bodyEnd2) {
|
|
--classLevel;
|
|
pattern.clear();
|
|
|
|
for (int i = classLevel; i < spaceInfo.size(); ++i)
|
|
pattern += (spaceInfo[i].className + " :: ");
|
|
|
|
pattern += typeName->str();
|
|
} else {
|
|
if (scope == 0)
|
|
break;
|
|
--scope;
|
|
}
|
|
}
|
|
|
|
// check for member functions
|
|
else if (isCPP() && tok2->str() == "(" && isFunctionHead(tok2, "{")) {
|
|
const Token *func = tok2->previous();
|
|
|
|
/** @todo add support for multi-token operators */
|
|
if (func->previous()->str() == "operator")
|
|
func = func->previous();
|
|
|
|
if (!func->previous())
|
|
syntaxError(func);
|
|
|
|
// check for qualifier
|
|
if (Token::Match(func->tokAt(-2), "%name% ::")) {
|
|
int offset = -2;
|
|
while (Token::Match(func->tokAt(offset - 2), "%name% ::"))
|
|
offset -= 2;
|
|
// check for available and matching class name
|
|
if (spaceInfo.size() > 1 && classLevel < spaceInfo.size() &&
|
|
func->strAt(offset) == spaceInfo[classLevel].className) {
|
|
memberScope = 0;
|
|
inMemberFunc = true;
|
|
}
|
|
}
|
|
}
|
|
|
|
// check for entering a new scope
|
|
else if (tok2->str() == "{") {
|
|
// check for entering a new namespace
|
|
if (isCPP()) {
|
|
if (tok2->strAt(-2) == "namespace") {
|
|
if (classLevel < spaceInfo.size() &&
|
|
spaceInfo[classLevel].isNamespace &&
|
|
spaceInfo[classLevel].className == tok2->previous()->str()) {
|
|
spaceInfo[classLevel].bodyEnd2 = tok2->link();
|
|
++classLevel;
|
|
pattern.clear();
|
|
for (int i = classLevel; i < spaceInfo.size(); ++i)
|
|
pattern += spaceInfo[i].className + " :: ";
|
|
|
|
pattern += typeName->str();
|
|
}
|
|
++scope;
|
|
}
|
|
if (Token::Match(tok2->tokAt(-3), "enum class %name%"))
|
|
inEnumClass = true;
|
|
}
|
|
|
|
// keep track of scopes within member function
|
|
if (inMemberFunc)
|
|
++memberScope;
|
|
|
|
++scope;
|
|
}
|
|
}
|
|
|
|
// check for operator typedef
|
|
/** @todo add support for multi-token operators */
|
|
else if (isCPP() &&
|
|
tok2->str() == "operator" &&
|
|
tok2->next() &&
|
|
tok2->next()->str() == typeName->str() &&
|
|
tok2->linkAt(2) &&
|
|
tok2->strAt(2) == "(" &&
|
|
Token::Match(tok2->linkAt(2), ") const| {")) {
|
|
// check for qualifier
|
|
if (tok2->previous()->str() == "::") {
|
|
// check for available and matching class name
|
|
if (spaceInfo.size() > 1 && classLevel < spaceInfo.size() &&
|
|
tok2->strAt(-2) == spaceInfo[classLevel].className) {
|
|
tok2 = tok2->next();
|
|
simplifyType = true;
|
|
}
|
|
}
|
|
}
|
|
|
|
else if (Token::Match(tok2->previous(), "class|struct %name% [:{]")) {
|
|
// don't replace names in struct/class definition
|
|
}
|
|
|
|
// check for typedef that can be substituted
|
|
else if ((tok2->isNameOnly() || (tok2->isName() && tok2->isExpandedMacro())) &&
|
|
(Token::simpleMatch(tok2, pattern.c_str(), pattern.size()) ||
|
|
(inMemberFunc && tok2->str() == typeName->str()))) {
|
|
// member function class variables don't need qualification
|
|
if (!(inMemberFunc && tok2->str() == typeName->str()) && pattern.find("::") != std::string::npos) { // has a "something ::"
|
|
Token *start = tok2;
|
|
int count = 0;
|
|
int back = classLevel - 1;
|
|
bool good = true;
|
|
// check for extra qualification
|
|
while (back >= 1) {
|
|
Token *qualificationTok = start->tokAt(-2);
|
|
if (!Token::Match(qualificationTok, "%type% ::"))
|
|
break;
|
|
if (qualificationTok->str() == spaceInfo[back].className) {
|
|
start = qualificationTok;
|
|
back--;
|
|
count++;
|
|
} else {
|
|
good = false;
|
|
break;
|
|
}
|
|
}
|
|
// check global namespace
|
|
if (good && back == 1 && start->strAt(-1) == "::")
|
|
good = false;
|
|
|
|
if (good) {
|
|
// remove any extra qualification if present
|
|
while (count) {
|
|
if (!removed.empty())
|
|
removed.insert(0, " ");
|
|
removed.insert(0, tok2->strAt(-2) + " " + tok2->strAt(-1));
|
|
tok2->tokAt(-3)->deleteNext(2);
|
|
--count;
|
|
}
|
|
|
|
// remove global namespace if present
|
|
if (tok2->strAt(-1) == "::") {
|
|
removed.insert(0, ":: ");
|
|
tok2->tokAt(-2)->deleteNext();
|
|
globalScope = true;
|
|
}
|
|
|
|
// remove qualification if present
|
|
for (int i = classLevel; i < spaceInfo.size(); ++i) {
|
|
if (!removed.empty())
|
|
removed += " ";
|
|
removed += (tok2->str() + " " + tok2->strAt(1));
|
|
tok2->deleteThis();
|
|
tok2->deleteThis();
|
|
}
|
|
simplifyType = true;
|
|
}
|
|
} else {
|
|
if (tok2->strAt(-1) == "::") {
|
|
int relativeSpaceInfoSize = spaceInfo.size();
|
|
Token * tokBeforeType = tok2->previous();
|
|
while (relativeSpaceInfoSize > 1 &&
|
|
tokBeforeType && tokBeforeType->str() == "::" &&
|
|
tokBeforeType->strAt(-1) == spaceInfo[relativeSpaceInfoSize-1].className) {
|
|
tokBeforeType = tokBeforeType->tokAt(-2);
|
|
--relativeSpaceInfoSize;
|
|
}
|
|
if (tokBeforeType && tokBeforeType->str() != "::") {
|
|
Token::eraseTokens(tokBeforeType, tok2);
|
|
simplifyType = true;
|
|
}
|
|
} else if (Token::Match(tok2->previous(), "case|;|{|} %type% :")) {
|
|
tok2 = tok2->next();
|
|
} else if (duplicateTypedef(&tok2, typeName, typeDef)) {
|
|
// skip to end of scope if not already there
|
|
if (tok2->str() != "}") {
|
|
while (tok2->next()) {
|
|
if (tok2->next()->str() == "{")
|
|
tok2 = tok2->linkAt(1)->previous();
|
|
else if (tok2->next()->str() == "}")
|
|
break;
|
|
|
|
tok2 = tok2->next();
|
|
}
|
|
}
|
|
} else if (Token::Match(tok2->tokAt(-2), "%type% *|&")) {
|
|
// Ticket #5868: Don't substitute variable names
|
|
} else if (tok2->previous()->str() != ".") {
|
|
simplifyType = true;
|
|
}
|
|
}
|
|
}
|
|
|
|
simplifyType = simplifyType && !inEnumClass;
|
|
|
|
if (simplifyType) {
|
|
mTypedefInfo.back().used = true;
|
|
|
|
// can't simplify 'operator functionPtr ()' and 'functionPtr operator ... ()'
|
|
if (functionPtr && (tok2->previous()->str() == "operator" ||
|
|
(tok2->next() && tok2->next()->str() == "operator"))) {
|
|
simplifyType = false;
|
|
tok2 = tok2->next();
|
|
continue;
|
|
}
|
|
|
|
// There are 2 categories of typedef substitutions:
|
|
// 1. variable declarations that preserve the variable name like
|
|
// global, local, and function parameters
|
|
// 2. not variable declarations that have no name like derived
|
|
// classes, casts, operators, and template parameters
|
|
|
|
// try to determine which category this substitution is
|
|
bool inCast = false;
|
|
bool inTemplate = false;
|
|
bool inOperator = false;
|
|
bool inSizeof = false;
|
|
|
|
const bool sameStartEnd = (typeStart == typeEnd);
|
|
|
|
// check for derived class: class A : some_typedef {
|
|
const bool isDerived = Token::Match(tok2->previous(), "public|protected|private %type% {|,");
|
|
|
|
// check for cast: (some_typedef) A or static_cast<some_typedef>(A)
|
|
// todo: check for more complicated casts like: (const some_typedef *)A
|
|
if ((tok2->previous()->str() == "(" && tok2->next()->str() == ")" && tok2->strAt(-2) != "sizeof") ||
|
|
(tok2->previous()->str() == "<" && Token::simpleMatch(tok2->next(), "> (")) ||
|
|
Token::Match(tok2->tokAt(-2), "( const %name% )"))
|
|
inCast = true;
|
|
|
|
// check for template parameters: t<some_typedef> t1
|
|
else if (Token::Match(tok2->previous(), "<|,") &&
|
|
Token::Match(tok2->next(), "&|*| &|*| >|,"))
|
|
inTemplate = true;
|
|
|
|
else if (Token::Match(tok2->tokAt(-2), "sizeof ( %type% )"))
|
|
inSizeof = true;
|
|
|
|
// check for operator
|
|
if (tok2->strAt(-1) == "operator" ||
|
|
Token::simpleMatch(tok2->tokAt(-2), "operator const"))
|
|
inOperator = true;
|
|
|
|
if (typeStart->str() == "typename" && tok2->strAt(-1)=="typename") {
|
|
// Remove one typename if it is already contained in the goal
|
|
typeStart = typeStart->next();
|
|
}
|
|
|
|
// skip over class or struct in derived class declaration
|
|
bool structRemoved = false;
|
|
if (isDerived && Token::Match(typeStart, "class|struct")) {
|
|
if (typeStart->str() == "struct")
|
|
structRemoved = true;
|
|
typeStart = typeStart->next();
|
|
}
|
|
if (Token::Match(typeStart, "struct|class|union") && Token::Match(tok2, "%name% ::"))
|
|
typeStart = typeStart->next();
|
|
|
|
if (sameStartEnd)
|
|
typeEnd = typeStart;
|
|
|
|
// start substituting at the typedef name by replacing it with the type
|
|
tok2->str(typeStart->str());
|
|
|
|
// restore qualification if it was removed
|
|
if (typeStart->str() == "struct" || structRemoved) {
|
|
if (structRemoved)
|
|
tok2 = tok2->previous();
|
|
|
|
if (globalScope) {
|
|
tok2->insertToken("::");
|
|
tok2 = tok2->next();
|
|
}
|
|
|
|
for (int i = classLevel; i < spaceInfo.size(); ++i) {
|
|
tok2->insertToken(spaceInfo[i].className);
|
|
tok2 = tok2->next();
|
|
tok2->insertToken("::");
|
|
tok2 = tok2->next();
|
|
}
|
|
}
|
|
|
|
// add some qualification back if needed
|
|
Token *start = tok2;
|
|
std::string removed1 = removed;
|
|
std::string::size_type idx = removed1.rfind(" ::");
|
|
|
|
if (idx != std::string::npos)
|
|
removed1.resize(idx);
|
|
if (removed1 == classPath && !removed1.empty()) {
|
|
for (std::vector<Space>::const_reverse_iterator it = spaceInfo.crbegin(); it != spaceInfo.crend(); ++it) {
|
|
if (it->recordTypes.find(start->str()) != it->recordTypes.end()) {
|
|
std::string::size_type spaceIdx = 0;
|
|
std::string::size_type startIdx = 0;
|
|
while ((spaceIdx = removed1.find(" ", startIdx)) != std::string::npos) {
|
|
tok2->previous()->insertToken(removed1.substr(startIdx, spaceIdx - startIdx));
|
|
startIdx = spaceIdx + 1;
|
|
}
|
|
tok2->previous()->insertToken(removed1.substr(startIdx));
|
|
tok2->previous()->insertToken("::");
|
|
break;
|
|
}
|
|
idx = removed1.rfind(" ::");
|
|
if (idx == std::string::npos)
|
|
break;
|
|
|
|
removed1.resize(idx);
|
|
}
|
|
}
|
|
// add remainder of type
|
|
tok2 = TokenList::copyTokens(tok2, typeStart->next(), typeEnd);
|
|
|
|
if (!pointers.empty()) {
|
|
for (const std::string &p : pointers) {
|
|
tok2->insertToken(p);
|
|
tok2 = tok2->next();
|
|
}
|
|
}
|
|
|
|
if (funcStart && funcEnd) {
|
|
tok2->insertToken("(");
|
|
tok2 = tok2->next();
|
|
Token *paren = tok2;
|
|
tok2 = TokenList::copyTokens(tok2, funcStart, funcEnd);
|
|
|
|
if (!inCast)
|
|
tok2 = processFunc(tok2, inOperator);
|
|
|
|
if (!tok2)
|
|
break;
|
|
|
|
while (Token::Match(tok2, "%name%|] ["))
|
|
tok2 = tok2->linkAt(1);
|
|
|
|
tok2->insertToken(")");
|
|
tok2 = tok2->next();
|
|
Token::createMutualLinks(tok2, paren);
|
|
|
|
tok2 = TokenList::copyTokens(tok2, argStart, argEnd);
|
|
|
|
if (specStart) {
|
|
Token *spec = specStart;
|
|
tok2->insertToken(spec->str());
|
|
tok2 = tok2->next();
|
|
while (spec != specEnd) {
|
|
spec = spec->next();
|
|
tok2->insertToken(spec->str());
|
|
tok2 = tok2->next();
|
|
}
|
|
}
|
|
}
|
|
|
|
else if (functionPtr || function) {
|
|
// don't add parentheses around function names because it
|
|
// confuses other simplifications
|
|
bool needParen = true;
|
|
if (!inTemplate && function && tok2->next() && tok2->next()->str() != "*")
|
|
needParen = false;
|
|
if (needParen) {
|
|
tok2->insertToken("(");
|
|
tok2 = tok2->next();
|
|
}
|
|
Token *tok3 = tok2;
|
|
if (namespaceStart) {
|
|
const Token *tok4 = namespaceStart;
|
|
|
|
while (tok4 != namespaceEnd) {
|
|
tok2->insertToken(tok4->str());
|
|
tok2 = tok2->next();
|
|
tok4 = tok4->next();
|
|
}
|
|
tok2->insertToken(namespaceEnd->str());
|
|
tok2 = tok2->next();
|
|
}
|
|
if (functionPtr) {
|
|
tok2->insertToken("*");
|
|
tok2 = tok2->next();
|
|
}
|
|
|
|
if (!inCast)
|
|
tok2 = processFunc(tok2, inOperator);
|
|
|
|
if (needParen) {
|
|
if (!tok2)
|
|
syntaxError(nullptr);
|
|
|
|
tok2->insertToken(")");
|
|
tok2 = tok2->next();
|
|
Token::createMutualLinks(tok2, tok3);
|
|
}
|
|
if (!tok2)
|
|
syntaxError(nullptr);
|
|
|
|
tok2 = TokenList::copyTokens(tok2, argStart, argEnd);
|
|
if (inTemplate) {
|
|
if (!tok2)
|
|
syntaxError(nullptr);
|
|
|
|
tok2 = tok2->next();
|
|
}
|
|
|
|
if (specStart) {
|
|
Token *spec = specStart;
|
|
tok2->insertToken(spec->str());
|
|
tok2 = tok2->next();
|
|
while (spec != specEnd) {
|
|
spec = spec->next();
|
|
tok2->insertToken(spec->str());
|
|
tok2 = tok2->next();
|
|
}
|
|
}
|
|
} else if (functionRetFuncPtr || functionPtrRetFuncPtr) {
|
|
tok2->insertToken("(");
|
|
tok2 = tok2->next();
|
|
Token *tok3 = tok2;
|
|
tok2->insertToken("*");
|
|
tok2 = tok2->next();
|
|
|
|
Token * tok4 = nullptr;
|
|
if (functionPtrRetFuncPtr) {
|
|
tok2->insertToken("(");
|
|
tok2 = tok2->next();
|
|
tok4 = tok2;
|
|
tok2->insertToken("*");
|
|
tok2 = tok2->next();
|
|
}
|
|
|
|
// skip over variable name if there
|
|
if (!inCast) {
|
|
if (!tok2 || !tok2->next())
|
|
syntaxError(nullptr);
|
|
|
|
if (tok2->next()->str() != ")")
|
|
tok2 = tok2->next();
|
|
}
|
|
|
|
if (tok4 && functionPtrRetFuncPtr) {
|
|
tok2->insertToken(")");
|
|
tok2 = tok2->next();
|
|
Token::createMutualLinks(tok2, tok4);
|
|
}
|
|
|
|
tok2 = TokenList::copyTokens(tok2, argStart, argEnd);
|
|
|
|
tok2->insertToken(")");
|
|
tok2 = tok2->next();
|
|
Token::createMutualLinks(tok2, tok3);
|
|
|
|
tok2 = TokenList::copyTokens(tok2, argFuncRetStart, argFuncRetEnd);
|
|
} else if (ptrToArray || refToArray) {
|
|
tok2->insertToken("(");
|
|
tok2 = tok2->next();
|
|
Token *tok3 = tok2;
|
|
|
|
if (ptrToArray)
|
|
tok2->insertToken("*");
|
|
else
|
|
tok2->insertToken("&");
|
|
tok2 = tok2->next();
|
|
|
|
bool hasName = false;
|
|
// skip over name
|
|
if (tok2->next() && tok2->next()->str() != ")" && tok2->next()->str() != "," &&
|
|
tok2->next()->str() != ">") {
|
|
hasName = true;
|
|
if (tok2->next()->str() != "(")
|
|
tok2 = tok2->next();
|
|
|
|
// check for function and skip over args
|
|
if (tok2 && tok2->next() && tok2->next()->str() == "(")
|
|
tok2 = tok2->next()->link();
|
|
|
|
// check for array
|
|
if (tok2 && tok2->next() && tok2->next()->str() == "[")
|
|
tok2 = tok2->next()->link();
|
|
}
|
|
|
|
tok2->insertToken(")");
|
|
Token::createMutualLinks(tok2->next(), tok3);
|
|
|
|
if (!hasName)
|
|
tok2 = tok2->next();
|
|
} else if (ptrMember) {
|
|
if (Token::simpleMatch(tok2, "* (")) {
|
|
tok2->insertToken("*");
|
|
tok2 = tok2->next();
|
|
} else {
|
|
// This is the case of casting operator.
|
|
// Name is not available, and () should not be
|
|
// inserted
|
|
const bool castOperator = inOperator && Token::Match(tok2, "%type% (");
|
|
Token *openParenthesis = nullptr;
|
|
|
|
if (!castOperator) {
|
|
tok2->insertToken("(");
|
|
tok2 = tok2->next();
|
|
|
|
openParenthesis = tok2;
|
|
}
|
|
|
|
const Token *tok4 = namespaceStart;
|
|
|
|
while (tok4 != namespaceEnd) {
|
|
tok2->insertToken(tok4->str());
|
|
tok2 = tok2->next();
|
|
tok4 = tok4->next();
|
|
}
|
|
tok2->insertToken(namespaceEnd->str());
|
|
tok2 = tok2->next();
|
|
|
|
tok2->insertToken("*");
|
|
tok2 = tok2->next();
|
|
|
|
if (openParenthesis) {
|
|
// Skip over name, if any
|
|
if (Token::Match(tok2->next(), "%name%"))
|
|
tok2 = tok2->next();
|
|
|
|
tok2->insertToken(")");
|
|
tok2 = tok2->next();
|
|
|
|
Token::createMutualLinks(tok2, openParenthesis);
|
|
}
|
|
}
|
|
} else if (typeOf) {
|
|
tok2 = TokenList::copyTokens(tok2, argStart, argEnd);
|
|
} else if (Token::Match(tok2, "%name% [")) {
|
|
while (Token::Match(tok2, "%name%|] [")) {
|
|
tok2 = tok2->linkAt(1);
|
|
}
|
|
tok2 = tok2->previous();
|
|
}
|
|
|
|
if (arrayStart && arrayEnd) {
|
|
do {
|
|
if (!tok2->next())
|
|
syntaxError(tok2); // can't recover so quit
|
|
|
|
if (!inCast && !inSizeof && !inTemplate)
|
|
tok2 = tok2->next();
|
|
|
|
if (tok2->str() == "const")
|
|
tok2 = tok2->next();
|
|
|
|
// reference or pointer to array?
|
|
if (Token::Match(tok2, "&|*|&&")) {
|
|
tok2 = tok2->previous();
|
|
tok2->insertToken("(");
|
|
Token *tok3 = tok2->next();
|
|
|
|
// handle missing variable name
|
|
if (Token::Match(tok3, "( *|&|&& *|&|&& %name%"))
|
|
tok2 = tok3->tokAt(3);
|
|
else if (Token::Match(tok2->tokAt(3), "[(),;]"))
|
|
tok2 = tok2->tokAt(2);
|
|
else
|
|
tok2 = tok2->tokAt(3);
|
|
if (!tok2)
|
|
syntaxError(nullptr);
|
|
|
|
while (tok2->strAt(1) == "::")
|
|
tok2 = tok2->tokAt(2);
|
|
|
|
// skip over function parameters
|
|
if (tok2->str() == "(")
|
|
tok2 = tok2->link();
|
|
|
|
if (tok2->strAt(1) == "(")
|
|
tok2 = tok2->linkAt(1);
|
|
|
|
// skip over const/noexcept
|
|
while (Token::Match(tok2->next(), "const|noexcept"))
|
|
tok2 = tok2->next();
|
|
|
|
tok2->insertToken(")");
|
|
tok2 = tok2->next();
|
|
Token::createMutualLinks(tok2, tok3);
|
|
}
|
|
|
|
if (!tok2->next())
|
|
syntaxError(tok2); // can't recover so quit
|
|
|
|
// skip over array dimensions
|
|
while (tok2->next()->str() == "[")
|
|
tok2 = tok2->linkAt(1);
|
|
|
|
tok2 = TokenList::copyTokens(tok2, arrayStart, arrayEnd);
|
|
if (!tok2->next())
|
|
syntaxError(tok2);
|
|
|
|
if (tok2->str() == "=") {
|
|
if (!tok2->next())
|
|
syntaxError(tok2);
|
|
if (tok2->next()->str() == "{")
|
|
tok2 = tok2->next()->link()->next();
|
|
else if (tok2->next()->str().at(0) == '\"')
|
|
tok2 = tok2->tokAt(2);
|
|
}
|
|
} while (Token::Match(tok2, ", %name% ;|=|,"));
|
|
}
|
|
|
|
simplifyType = false;
|
|
}
|
|
if (!tok2)
|
|
break;
|
|
}
|
|
|
|
if (!tok)
|
|
syntaxError(nullptr);
|
|
|
|
if (tok->str() == ";")
|
|
done = true;
|
|
else if (tok->str() == ",") {
|
|
arrayStart = nullptr;
|
|
arrayEnd = nullptr;
|
|
tokOffset = tok->next();
|
|
pointers.clear();
|
|
|
|
while (Token::Match(tokOffset, "*|&")) {
|
|
pointers.push_back(tokOffset->str());
|
|
tokOffset = tokOffset->next();
|
|
}
|
|
|
|
if (Token::Match(tokOffset, "%type%")) {
|
|
typeName = tokOffset;
|
|
tokOffset = tokOffset->next();
|
|
|
|
if (tokOffset && tokOffset->str() == "[") {
|
|
arrayStart = tokOffset;
|
|
|
|
for (;;) {
|
|
while (tokOffset->next() && !Token::Match(tokOffset->next(), ";|,"))
|
|
tokOffset = tokOffset->next();
|
|
|
|
if (!tokOffset->next())
|
|
return; // invalid input
|
|
else if (tokOffset->next()->str() == ";")
|
|
break;
|
|
else if (tokOffset->str() == "]")
|
|
break;
|
|
else
|
|
tokOffset = tokOffset->next();
|
|
}
|
|
|
|
arrayEnd = tokOffset;
|
|
tokOffset = tokOffset->next();
|
|
}
|
|
|
|
if (Token::Match(tokOffset, ";|,"))
|
|
tok = tokOffset;
|
|
else {
|
|
// we encountered a typedef we don't support yet so just continue
|
|
done = true;
|
|
ok = false;
|
|
}
|
|
} else {
|
|
// we encountered a typedef we don't support yet so just continue
|
|
done = true;
|
|
ok = false;
|
|
}
|
|
} else {
|
|
// something is really wrong (internal error)
|
|
done = true;
|
|
ok = false;
|
|
}
|
|
}
|
|
|
|
if (ok) {
|
|
// remove typedef
|
|
Token::eraseTokens(typeDef, tok);
|
|
|
|
if (typeDef != list.front()) {
|
|
tok = typeDef->previous();
|
|
tok->deleteNext();
|
|
//no need to remove last token in the list
|
|
if (tok->tokAt(2))
|
|
tok->deleteNext();
|
|
} else {
|
|
list.front()->deleteThis();
|
|
//no need to remove last token in the list
|
|
if (list.front()->next())
|
|
list.front()->deleteThis();
|
|
tok = list.front();
|
|
//now the next token to process is 'tok', not 'tok->next()';
|
|
goback = true;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
namespace {
|
|
struct ScopeInfo3 {
|
|
enum Type { Global, Namespace, Record, MemberFunction, Other };
|
|
ScopeInfo3() : parent(nullptr), type(Global), bodyStart(nullptr), bodyEnd(nullptr) {}
|
|
ScopeInfo3(ScopeInfo3 *parent_, Type type_, const std::string &name_, const Token *bodyStart_, const Token *bodyEnd_)
|
|
: parent(parent_), type(type_), name(name_), bodyStart(bodyStart_), bodyEnd(bodyEnd_) {
|
|
if (name.empty())
|
|
return;
|
|
fullName = name;
|
|
ScopeInfo3 *scope = parent;
|
|
while (scope && scope->parent) {
|
|
if (scope->name.empty())
|
|
break;
|
|
fullName = scope->name + " :: " + fullName;
|
|
scope = scope->parent;
|
|
}
|
|
}
|
|
ScopeInfo3 *parent;
|
|
std::list<ScopeInfo3> children;
|
|
Type type;
|
|
std::string fullName;
|
|
std::string name;
|
|
const Token * bodyStart;
|
|
const Token * bodyEnd;
|
|
std::set<std::string> usingNamespaces;
|
|
std::set<std::string> recordTypes;
|
|
std::set<std::string> baseTypes;
|
|
|
|
ScopeInfo3 *addChild(Type scopeType, const std::string &scopeName, const Token *bodyStartToken, const Token *bodyEndToken) {
|
|
children.emplace_back(this, scopeType, scopeName, bodyStartToken, bodyEndToken);
|
|
return &children.back();
|
|
}
|
|
|
|
bool hasChild(const std::string &childName) const {
|
|
for (const auto & child : children) {
|
|
if (child.name == childName)
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
const ScopeInfo3 * findInChildren(const std::string & scope) const {
|
|
for (const auto & child : children) {
|
|
if (child.type == Record && (child.name == scope || child.fullName == scope))
|
|
return &child;
|
|
else {
|
|
const ScopeInfo3 * temp = child.findInChildren(scope);
|
|
if (temp)
|
|
return temp;
|
|
}
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
const ScopeInfo3 * findScope(const std::string & scope) const {
|
|
const ScopeInfo3 * tempScope = this;
|
|
while (tempScope) {
|
|
// check children
|
|
for (const auto & child : tempScope->children) {
|
|
if (&child != this && child.type == Record && (child.name == scope || child.fullName == scope))
|
|
return &child;
|
|
}
|
|
// check siblings for same name
|
|
if (tempScope->parent) {
|
|
for (const auto &sibling : tempScope->parent->children) {
|
|
if (sibling.name == tempScope->name && &sibling != this) {
|
|
const ScopeInfo3 * temp = sibling.findInChildren(scope);
|
|
if (temp)
|
|
return temp;
|
|
}
|
|
}
|
|
}
|
|
tempScope = tempScope->parent;
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
bool findTypeInBase(const std::string &scope) const {
|
|
// check in base types first
|
|
if (baseTypes.find(scope) != baseTypes.end())
|
|
return true;
|
|
// check in base types base types
|
|
for (const std::string & base : baseTypes) {
|
|
const ScopeInfo3 * baseScope = findScope(base);
|
|
// bail on uninstantiated recursive template
|
|
if (baseScope == this)
|
|
return false;
|
|
if (baseScope && baseScope->fullName == scope)
|
|
return true;
|
|
if (baseScope && baseScope->findTypeInBase(scope))
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
ScopeInfo3 * findScope(const ScopeInfo3 * scope) {
|
|
if (scope->bodyStart == bodyStart)
|
|
return this;
|
|
for (auto & child : children) {
|
|
ScopeInfo3 * temp = child.findScope(scope);
|
|
if (temp)
|
|
return temp;
|
|
}
|
|
return nullptr;
|
|
}
|
|
};
|
|
|
|
void setScopeInfo(Token *tok, ScopeInfo3 **scopeInfo, bool debug=false)
|
|
{
|
|
if (!tok)
|
|
return;
|
|
if (tok->str() == "{" && (*scopeInfo)->parent && tok == (*scopeInfo)->bodyStart)
|
|
return;
|
|
if (tok->str() == "}") {
|
|
if ((*scopeInfo)->parent && tok == (*scopeInfo)->bodyEnd)
|
|
*scopeInfo = (*scopeInfo)->parent;
|
|
else {
|
|
// Try to find parent scope
|
|
ScopeInfo3 *parent = (*scopeInfo)->parent;
|
|
while (parent && parent->bodyEnd != tok)
|
|
parent = parent->parent;
|
|
if (parent) {
|
|
*scopeInfo = parent;
|
|
if (debug)
|
|
throw std::runtime_error("Internal error: unmatched }");
|
|
}
|
|
}
|
|
return;
|
|
}
|
|
if (!Token::Match(tok, "namespace|class|struct|union %name% {|:|::|<")) {
|
|
// check for using namespace
|
|
if (Token::Match(tok, "using namespace %name% ;|::")) {
|
|
const Token * tok1 = tok->tokAt(2);
|
|
std::string nameSpace;
|
|
while (tok1 && tok1->str() != ";") {
|
|
if (!nameSpace.empty())
|
|
nameSpace += " ";
|
|
nameSpace += tok1->str();
|
|
tok1 = tok1->next();
|
|
}
|
|
(*scopeInfo)->usingNamespaces.insert(nameSpace);
|
|
}
|
|
// check for member function
|
|
else if (tok->str() == "{") {
|
|
bool added = false;
|
|
Token *tok1 = tok;
|
|
while (Token::Match(tok1->previous(), "const|volatile|final|override|&|&&|noexcept"))
|
|
tok1 = tok1->previous();
|
|
if (tok1->previous() && (tok1->strAt(-1) == ")" || tok->strAt(-1) == "}")) {
|
|
tok1 = tok1->linkAt(-1);
|
|
if (Token::Match(tok1->previous(), "throw|noexcept (")) {
|
|
tok1 = tok1->previous();
|
|
while (Token::Match(tok1->previous(), "const|volatile|final|override|&|&&|noexcept"))
|
|
tok1 = tok1->previous();
|
|
if (tok1->strAt(-1) != ")")
|
|
return;
|
|
tok1 = tok1->linkAt(-1);
|
|
} else {
|
|
while (Token::Match(tok1->tokAt(-2), ":|, %name%")) {
|
|
tok1 = tok1->tokAt(-2);
|
|
if (tok1->strAt(-1) != ")" && tok1->strAt(-1) != "}")
|
|
return;
|
|
tok1 = tok1->linkAt(-1);
|
|
}
|
|
}
|
|
if (tok1->strAt(-1) == ">")
|
|
tok1 = tok1->previous()->findOpeningBracket();
|
|
if (tok1 && (Token::Match(tok1->tokAt(-3), "%name% :: %name%") ||
|
|
Token::Match(tok1->tokAt(-4), "%name% :: ~ %name%"))) {
|
|
tok1 = tok1->tokAt(-2);
|
|
if (tok1->str() == "~")
|
|
tok1 = tok1->previous();
|
|
std::string scope = tok1->strAt(-1);
|
|
while (Token::Match(tok1->tokAt(-2), ":: %name%")) {
|
|
scope = tok1->strAt(-3) + " :: " + scope;
|
|
tok1 = tok1->tokAt(-2);
|
|
}
|
|
*scopeInfo = (*scopeInfo)->addChild(ScopeInfo3::MemberFunction, scope, tok, tok->link());
|
|
added = true;
|
|
}
|
|
}
|
|
|
|
if (!added)
|
|
*scopeInfo = (*scopeInfo)->addChild(ScopeInfo3::Other, "", tok, tok->link());
|
|
}
|
|
return;
|
|
}
|
|
|
|
const bool record = Token::Match(tok, "class|struct|union %name%");
|
|
tok = tok->next();
|
|
std::string classname = tok->str();
|
|
while (Token::Match(tok, "%name% :: %name%")) {
|
|
tok = tok->tokAt(2);
|
|
classname += " :: " + tok->str();
|
|
}
|
|
|
|
// add record type to scope info
|
|
if (record)
|
|
(*scopeInfo)->recordTypes.insert(classname);
|
|
tok = tok->next();
|
|
|
|
// skip template parameters
|
|
if (tok && tok->str() == "<") {
|
|
tok = tok->findClosingBracket();
|
|
if (tok)
|
|
tok = tok->next();
|
|
}
|
|
|
|
// get base class types
|
|
std::set<std::string> baseTypes;
|
|
if (tok && tok->str() == ":") {
|
|
do {
|
|
tok = tok->next();
|
|
while (Token::Match(tok, "public|protected|private|virtual"))
|
|
tok = tok->next();
|
|
std::string base;
|
|
while (tok && !Token::Match(tok, ";|,|{")) {
|
|
if (!base.empty())
|
|
base += ' ';
|
|
base += tok->str();
|
|
tok = tok->next();
|
|
// skip template parameters
|
|
if (tok && tok->str() == "<") {
|
|
tok = tok->findClosingBracket();
|
|
if (tok)
|
|
tok = tok->next();
|
|
}
|
|
}
|
|
baseTypes.insert(base);
|
|
} while (tok && !Token::Match(tok, ";|{"));
|
|
}
|
|
|
|
if (tok && tok->str() == "{") {
|
|
*scopeInfo = (*scopeInfo)->addChild(record ? ScopeInfo3::Record : ScopeInfo3::Namespace, classname, tok, tok->link());
|
|
(*scopeInfo)->baseTypes = baseTypes;
|
|
}
|
|
}
|
|
|
|
Token *findSemicolon(Token *tok)
|
|
{
|
|
int level = 0;
|
|
|
|
for (; tok && (level > 0 || tok->str() != ";"); tok = tok->next()) {
|
|
if (tok->str() == "{")
|
|
++level;
|
|
else if (level > 0 && tok->str() == "}")
|
|
--level;
|
|
}
|
|
|
|
return tok;
|
|
}
|
|
|
|
bool usingMatch(
|
|
const Token *nameToken,
|
|
const std::string &scope,
|
|
Token **tok,
|
|
const std::string &scope1,
|
|
const ScopeInfo3 *currentScope,
|
|
const ScopeInfo3 *memberClassScope)
|
|
{
|
|
Token *tok1 = *tok;
|
|
|
|
if (tok1 && tok1->str() != nameToken->str())
|
|
return false;
|
|
|
|
// skip this using
|
|
if (tok1 == nameToken) {
|
|
*tok = findSemicolon(tok1);
|
|
return false;
|
|
}
|
|
|
|
// skip other using with this name
|
|
if (tok1->strAt(-1) == "using") {
|
|
// fixme: this is wrong
|
|
// skip to end of scope
|
|
if (currentScope->bodyEnd)
|
|
*tok = currentScope->bodyEnd->previous();
|
|
return false;
|
|
}
|
|
|
|
if (Token::Match(tok1->tokAt(-1), "class|struct|union|enum|namespace")) {
|
|
// fixme
|
|
return false;
|
|
}
|
|
|
|
// get qualification
|
|
std::string qualification;
|
|
const Token* tok2 = tok1;
|
|
std::string::size_type index = scope.size();
|
|
std::string::size_type new_index = std::string::npos;
|
|
bool match = true;
|
|
while (Token::Match(tok2->tokAt(-2), "%name% ::") && !tok2->tokAt(-2)->isKeyword()) {
|
|
std::string last;
|
|
if (match && !scope1.empty()) {
|
|
new_index = scope1.rfind(' ', index - 1);
|
|
if (new_index != std::string::npos)
|
|
last = scope1.substr(new_index, index - new_index);
|
|
else if (!qualification.empty())
|
|
last.clear();
|
|
else
|
|
last = scope1;
|
|
} else
|
|
match = false;
|
|
if (match && tok2->strAt(-2) == last)
|
|
index = new_index;
|
|
else {
|
|
if (!qualification.empty())
|
|
qualification = " :: " + qualification;
|
|
qualification = tok2->strAt(-2) + qualification;
|
|
}
|
|
tok2 = tok2->tokAt(-2);
|
|
}
|
|
|
|
std::string fullScope1 = scope1;
|
|
if (!scope1.empty() && !qualification.empty())
|
|
fullScope1 += " :: ";
|
|
fullScope1 += qualification;
|
|
|
|
if (scope == fullScope1)
|
|
return true;
|
|
|
|
const ScopeInfo3 *scopeInfo = memberClassScope ? memberClassScope : currentScope;
|
|
|
|
// check in base types
|
|
if (scopeInfo->findTypeInBase(scope))
|
|
return true;
|
|
|
|
// check using namespace
|
|
const ScopeInfo3 * tempScope = scopeInfo;
|
|
while (tempScope) {
|
|
//if (!tempScope->parent->usingNamespaces.empty()) {
|
|
if (!tempScope->usingNamespaces.empty()) {
|
|
if (qualification.empty()) {
|
|
if (tempScope->usingNamespaces.find(scope) != tempScope->usingNamespaces.end())
|
|
return true;
|
|
} else {
|
|
for (const auto &ns : tempScope->usingNamespaces) {
|
|
if (scope == ns + " :: " + qualification)
|
|
return true;
|
|
}
|
|
}
|
|
}
|
|
tempScope = tempScope->parent;
|
|
}
|
|
|
|
std::string newScope1 = scope1;
|
|
|
|
// scopes didn't match so try higher scopes
|
|
index = newScope1.size();
|
|
while (!newScope1.empty()) {
|
|
std::string::size_type separator = newScope1.rfind(" :: ", index - 1);
|
|
if (separator != std::string::npos)
|
|
newScope1 = newScope1.substr(0, separator);
|
|
else
|
|
newScope1.clear();
|
|
|
|
std::string newFullScope1 = newScope1;
|
|
if (!newScope1.empty() && !qualification.empty())
|
|
newFullScope1 += " :: ";
|
|
newFullScope1 += qualification;
|
|
|
|
if (scope == newFullScope1)
|
|
return true;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
std::string memberFunctionScope(const Token *tok)
|
|
{
|
|
std::string qualification;
|
|
const Token *qualTok = tok->strAt(-2) == "~" ? tok->tokAt(-4) : tok->tokAt(-3);
|
|
while (Token::Match(qualTok, "%type% ::")) {
|
|
if (!qualification.empty())
|
|
qualification = " :: " + qualification;
|
|
qualification = qualTok->str() + qualification;
|
|
qualTok = qualTok->tokAt(-2);
|
|
}
|
|
return qualification;
|
|
}
|
|
|
|
const Token * memberFunctionEnd(const Token *tok)
|
|
{
|
|
if (tok->str() != "(")
|
|
return nullptr;
|
|
const Token *end = tok->link()->next();
|
|
while (end) {
|
|
if (end->str() == "{" && !Token::Match(end->tokAt(-2), ":|, %name%"))
|
|
return end;
|
|
else if (end->str() == ";")
|
|
break;
|
|
end = end->next();
|
|
}
|
|
return nullptr;
|
|
}
|
|
} // namespace
|
|
|
|
bool Tokenizer::isMemberFunction(const Token *openParen) const
|
|
{
|
|
return (Token::Match(openParen->tokAt(-2), ":: %name% (") ||
|
|
Token::Match(openParen->tokAt(-3), ":: ~ %name% (")) &&
|
|
isFunctionHead(openParen, "{|:");
|
|
}
|
|
|
|
static bool scopesMatch(const std::string &scope1, const std::string &scope2, const ScopeInfo3 *globalScope)
|
|
{
|
|
if (scope1.empty() || scope2.empty())
|
|
return false;
|
|
|
|
// check if scopes match
|
|
if (scope1 == scope2)
|
|
return true;
|
|
|
|
// check if scopes only differ by global qualification
|
|
if (scope1 == (":: " + scope2)) {
|
|
std::string::size_type end = scope2.find_first_of(' ');
|
|
if (end == std::string::npos)
|
|
end = scope2.size();
|
|
if (globalScope->hasChild(scope2.substr(0, end)))
|
|
return true;
|
|
} else if (scope2 == (":: " + scope1)) {
|
|
std::string::size_type end = scope1.find_first_of(' ');
|
|
if (end == std::string::npos)
|
|
end = scope1.size();
|
|
if (globalScope->hasChild(scope1.substr(0, end)))
|
|
return true;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
bool Tokenizer::simplifyUsing()
|
|
{
|
|
if (!isCPP() || mSettings->standards.cpp < Standards::CPP11)
|
|
return false;
|
|
|
|
bool substitute = false;
|
|
ScopeInfo3 scopeInfo;
|
|
ScopeInfo3 *currentScope = &scopeInfo;
|
|
struct Using {
|
|
Using(Token *start, Token *end) : startTok(start), endTok(end) {}
|
|
Token *startTok;
|
|
Token *endTok;
|
|
};
|
|
std::list<Using> usingList;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (mErrorLogger && !list.getFiles().empty())
|
|
mErrorLogger->reportProgress(list.getFiles()[0], "Tokenize (using)", tok->progressValue());
|
|
|
|
if (Settings::terminated())
|
|
return substitute;
|
|
|
|
if (Token::Match(tok, "enum class|struct")) {
|
|
Token *bodyStart = tok;
|
|
while (Token::Match(bodyStart, "%name%|:|::|<")) {
|
|
if (bodyStart->str() == "<")
|
|
bodyStart = bodyStart->findClosingBracket();
|
|
bodyStart = bodyStart ? bodyStart->next() : nullptr;
|
|
}
|
|
if (Token::simpleMatch(bodyStart, "{"))
|
|
tok = bodyStart->link();
|
|
continue;
|
|
}
|
|
|
|
if (Token::Match(tok, "{|}|namespace|class|struct|union") ||
|
|
Token::Match(tok, "using namespace %name% ;|::")) {
|
|
try {
|
|
setScopeInfo(tok, ¤tScope, mSettings->debugwarnings);
|
|
} catch (const std::runtime_error &) {
|
|
reportError(tok, Severity::debug, "simplifyUsingUnmatchedBodyEnd",
|
|
"simplifyUsing: unmatched body end");
|
|
}
|
|
continue;
|
|
}
|
|
|
|
// skip template declarations
|
|
if (Token::Match(tok, "template < !!>")) {
|
|
// add template record type to scope info
|
|
const Token *end = tok->next()->findClosingBracket();
|
|
if (end && Token::Match(end->next(), "class|struct|union %name%"))
|
|
currentScope->recordTypes.insert(end->strAt(2));
|
|
|
|
Token *declEndToken = TemplateSimplifier::findTemplateDeclarationEnd(tok);
|
|
if (declEndToken)
|
|
tok = declEndToken;
|
|
continue;
|
|
}
|
|
|
|
// look for non-template type aliases
|
|
if (!(tok->strAt(-1) != ">" &&
|
|
(Token::Match(tok, "using %name% = ::| %name%") ||
|
|
(Token::Match(tok, "using %name% [ [") &&
|
|
Token::Match(tok->linkAt(2), "] ] = ::| %name%")))))
|
|
continue;
|
|
|
|
std::string name = tok->strAt(1);
|
|
const Token *nameToken = tok->next();
|
|
std::string scope = currentScope->fullName;
|
|
Token *usingStart = tok;
|
|
Token *start;
|
|
if (tok->strAt(2) == "=")
|
|
start = tok->tokAt(3);
|
|
else
|
|
start = tok->linkAt(2)->tokAt(3);
|
|
Token *usingEnd = findSemicolon(start);
|
|
if (!usingEnd)
|
|
continue;
|
|
|
|
// Move struct defined in using out of using.
|
|
// using T = struct t { }; => struct t { }; using T = struct t;
|
|
// fixme: this doesn't handle attributes
|
|
if (Token::Match(start, "class|struct|union|enum %name%| {|:")) {
|
|
Token *structEnd = start->tokAt(1);
|
|
const bool hasName = Token::Match(structEnd, "%name%");
|
|
|
|
// skip over name if present
|
|
if (hasName)
|
|
structEnd = structEnd->next();
|
|
|
|
// skip over base class information
|
|
if (structEnd->str() == ":") {
|
|
structEnd = structEnd->next(); // skip over ":"
|
|
while (structEnd && structEnd->str() != "{")
|
|
structEnd = structEnd->next();
|
|
if (!structEnd)
|
|
continue;
|
|
}
|
|
|
|
// use link to go to end
|
|
structEnd = structEnd->link();
|
|
|
|
// add ';' after end of struct
|
|
structEnd->insertToken(";", "");
|
|
|
|
// add name for anonymous struct
|
|
if (!hasName) {
|
|
std::string newName;
|
|
if (structEnd->strAt(2) == ";")
|
|
newName = name;
|
|
else
|
|
newName = "Unnamed" + MathLib::toString(mUnnamedCount++);
|
|
TokenList::copyTokens(structEnd->next(), tok, start);
|
|
structEnd->tokAt(5)->insertToken(newName, "");
|
|
start->insertToken(newName, "");
|
|
} else
|
|
TokenList::copyTokens(structEnd->next(), tok, start->next());
|
|
|
|
// add using after end of struct
|
|
usingStart = structEnd->tokAt(2);
|
|
nameToken = usingStart->next();
|
|
if (usingStart->strAt(2) == "=")
|
|
start = usingStart->tokAt(3);
|
|
else
|
|
start = usingStart->linkAt(2)->tokAt(3);
|
|
usingEnd = findSemicolon(start);
|
|
|
|
// delete original using before struct
|
|
tok->deleteThis();
|
|
tok->deleteThis();
|
|
tok->deleteThis();
|
|
tok = usingStart;
|
|
}
|
|
|
|
// remove 'typename' and 'template'
|
|
else if (start->str() == "typename") {
|
|
start->deleteThis();
|
|
Token *temp = start;
|
|
while (Token::Match(temp, "%name% ::"))
|
|
temp = temp->tokAt(2);
|
|
if (Token::Match(temp, "template %name%"))
|
|
temp->deleteThis();
|
|
}
|
|
|
|
if (usingEnd)
|
|
tok = usingEnd;
|
|
|
|
// Unfortunately we have to start searching from the beginning
|
|
// of the token stream because templates are instantiated at
|
|
// the end of the token stream and it may be used before then.
|
|
ScopeInfo3 scopeInfo1;
|
|
ScopeInfo3 *currentScope1 = &scopeInfo1;
|
|
Token *startToken = list.front();
|
|
Token *endToken = nullptr;
|
|
bool inMemberFunc = false;
|
|
const ScopeInfo3 * memberFuncScope = nullptr;
|
|
const Token * memberFuncEnd = nullptr;
|
|
|
|
// We can limit the search to the current function when the type alias
|
|
// is defined in that function.
|
|
if (currentScope->type == ScopeInfo3::Other ||
|
|
currentScope->type == ScopeInfo3::MemberFunction) {
|
|
scopeInfo1 = scopeInfo;
|
|
currentScope1 = scopeInfo1.findScope(currentScope);
|
|
if (!currentScope1)
|
|
return substitute; // something bad happened
|
|
startToken = usingEnd->next();
|
|
endToken = currentScope->bodyEnd->next();
|
|
if (currentScope->type == ScopeInfo3::MemberFunction) {
|
|
const ScopeInfo3 * temp = currentScope->findScope(currentScope->fullName);
|
|
if (temp) {
|
|
inMemberFunc = true;
|
|
memberFuncScope = temp;
|
|
memberFuncEnd = endToken;
|
|
}
|
|
}
|
|
}
|
|
|
|
std::string scope1 = currentScope1->fullName;
|
|
bool skip = false; // don't erase type aliases we can't parse
|
|
Token *enumOpenBrace = nullptr;
|
|
for (Token* tok1 = startToken; !skip && tok1 && tok1 != endToken; tok1 = tok1->next()) {
|
|
// skip enum body
|
|
if (tok1 && tok1 == enumOpenBrace) {
|
|
tok1 = tok1->link();
|
|
enumOpenBrace = nullptr;
|
|
continue;
|
|
}
|
|
|
|
if ((Token::Match(tok1, "{|}|namespace|class|struct|union") && tok1->strAt(-1) != "using") ||
|
|
Token::Match(tok1, "using namespace %name% ;|::")) {
|
|
try {
|
|
setScopeInfo(tok1, ¤tScope1, mSettings->debugwarnings);
|
|
} catch (const std::runtime_error &) {
|
|
reportError(tok1, Severity::debug, "simplifyUsingUnmatchedBodyEnd",
|
|
"simplifyUsing: unmatched body end");
|
|
}
|
|
scope1 = currentScope1->fullName;
|
|
if (inMemberFunc && memberFuncEnd && tok1 == memberFuncEnd) {
|
|
inMemberFunc = false;
|
|
memberFuncScope = nullptr;
|
|
memberFuncEnd = nullptr;
|
|
}
|
|
continue;
|
|
}
|
|
|
|
// skip template definitions
|
|
if (Token::Match(tok1, "template < !!>")) {
|
|
Token *declEndToken = TemplateSimplifier::findTemplateDeclarationEnd(tok1);
|
|
if (declEndToken)
|
|
tok1 = declEndToken;
|
|
continue;
|
|
}
|
|
|
|
// check for enum with body
|
|
if (tok1->str() == "enum") {
|
|
if (Token::Match(tok1, "enum class|struct"))
|
|
tok1 = tok1->next();
|
|
Token *defStart = tok1;
|
|
while (Token::Match(defStart, "%name%|::|:"))
|
|
defStart = defStart->next();
|
|
if (Token::simpleMatch(defStart, "{"))
|
|
enumOpenBrace = defStart;
|
|
continue;
|
|
}
|
|
|
|
// check for member function and adjust scope
|
|
if (isMemberFunction(tok1)) {
|
|
if (!scope1.empty())
|
|
scope1 += " :: ";
|
|
scope1 += memberFunctionScope(tok1);
|
|
const ScopeInfo3 * temp = currentScope1->findScope(scope1);
|
|
if (temp) {
|
|
const Token *end = memberFunctionEnd(tok1);
|
|
if (end) {
|
|
inMemberFunc = true;
|
|
memberFuncScope = temp;
|
|
memberFuncEnd = end;
|
|
}
|
|
}
|
|
continue;
|
|
} else if (inMemberFunc && memberFuncScope) {
|
|
if (!usingMatch(nameToken, scope, &tok1, scope1, currentScope1, memberFuncScope))
|
|
continue;
|
|
} else if (!usingMatch(nameToken, scope, &tok1, scope1, currentScope1, nullptr))
|
|
continue;
|
|
|
|
// remove the qualification
|
|
std::string fullScope = scope;
|
|
std::string removed;
|
|
while (Token::Match(tok1->tokAt(-2), "%name% ::") && !tok1->tokAt(-2)->isKeyword()) {
|
|
removed = (tok1->strAt(-2) + " :: ") + removed;
|
|
if (fullScope == tok1->strAt(-2)) {
|
|
tok1->deletePrevious();
|
|
tok1->deletePrevious();
|
|
break;
|
|
} else {
|
|
const std::string::size_type idx = fullScope.rfind(" ");
|
|
|
|
if (idx == std::string::npos)
|
|
break;
|
|
|
|
if (tok1->strAt(-2) == fullScope.substr(idx + 1)) {
|
|
tok1->deletePrevious();
|
|
tok1->deletePrevious();
|
|
fullScope.resize(idx - 3);
|
|
} else
|
|
break;
|
|
}
|
|
}
|
|
|
|
// remove global namespace if present
|
|
if (tok1->strAt(-1) == "::") {
|
|
removed.insert(0, ":: ");
|
|
tok1->deletePrevious();
|
|
}
|
|
|
|
Token * arrayStart = nullptr;
|
|
|
|
// parse the type
|
|
Token *type = start;
|
|
if (type->str() == "::") {
|
|
type = type->next();
|
|
while (Token::Match(type, "%type% ::"))
|
|
type = type->tokAt(2);
|
|
if (Token::Match(type, "%type%"))
|
|
type = type->next();
|
|
} else if (Token::Match(type, "%type% ::")) {
|
|
do {
|
|
type = type->tokAt(2);
|
|
} while (Token::Match(type, "%type% ::"));
|
|
if (Token::Match(type, "%type%"))
|
|
type = type->next();
|
|
} else if (Token::Match(type, "%type%")) {
|
|
while (Token::Match(type, "const|class|struct|union|enum %type%") ||
|
|
(type->next() && type->next()->isStandardType()))
|
|
type = type->next();
|
|
|
|
type = type->next();
|
|
|
|
while (Token::Match(type, "%type%") &&
|
|
(type->isStandardType() || Token::Match(type, "unsigned|signed"))) {
|
|
type = type->next();
|
|
}
|
|
|
|
bool atEnd = false;
|
|
while (!atEnd) {
|
|
if (type && type->str() == "::") {
|
|
type = type->next();
|
|
}
|
|
|
|
if (Token::Match(type, "%type%") &&
|
|
type->next() && !Token::Match(type->next(), "[|,|(")) {
|
|
type = type->next();
|
|
} else if (Token::simpleMatch(type, "const (")) {
|
|
type = type->next();
|
|
atEnd = true;
|
|
} else
|
|
atEnd = true;
|
|
}
|
|
} else
|
|
syntaxError(type);
|
|
|
|
// check for invalid input
|
|
if (!type)
|
|
syntaxError(tok1);
|
|
|
|
// check for template
|
|
if (type->str() == "<") {
|
|
type = type->findClosingBracket();
|
|
|
|
while (type && Token::Match(type->next(), ":: %type%"))
|
|
type = type->tokAt(2);
|
|
|
|
if (!type) {
|
|
syntaxError(tok1);
|
|
}
|
|
|
|
while (Token::Match(type->next(), "const|volatile"))
|
|
type = type->next();
|
|
|
|
type = type->next();
|
|
}
|
|
|
|
// check for pointers and references
|
|
std::list<std::string> pointers;
|
|
while (Token::Match(type, "*|&|&&|const")) {
|
|
pointers.push_back(type->str());
|
|
type = type->next();
|
|
}
|
|
|
|
// check for array
|
|
if (type && type->str() == "[") {
|
|
do {
|
|
if (!arrayStart)
|
|
arrayStart = type;
|
|
|
|
bool atEnd = false;
|
|
while (!atEnd) {
|
|
while (type->next() && !Token::Match(type->next(), ";|,")) {
|
|
type = type->next();
|
|
}
|
|
|
|
if (!type->next())
|
|
syntaxError(type); // invalid input
|
|
else if (type->next()->str() == ";")
|
|
atEnd = true;
|
|
else if (type->str() == "]")
|
|
atEnd = true;
|
|
else
|
|
type = type->next();
|
|
}
|
|
|
|
type = type->next();
|
|
} while (type && type->str() == "[");
|
|
}
|
|
|
|
// make sure we are in a good state
|
|
if (!tok1 || !tok1->next())
|
|
break; // bail
|
|
|
|
Token* after = tok1->next();
|
|
// check if type was parsed
|
|
if (type && type == usingEnd) {
|
|
// check for array syntax and add type around variable
|
|
if (arrayStart) {
|
|
if (Token::Match(tok1->next(), "%name%")) {
|
|
TokenList::copyTokens(tok1->next(), arrayStart, usingEnd->previous());
|
|
TokenList::copyTokens(tok1, start, arrayStart->previous());
|
|
tok1->deleteThis();
|
|
substitute = true;
|
|
}
|
|
} else {
|
|
// add some qualification back if needed
|
|
std::string removed1 = removed;
|
|
std::string::size_type idx = removed1.rfind(" ::");
|
|
if (idx != std::string::npos)
|
|
removed1.resize(idx);
|
|
if (scopesMatch(removed1, scope, &scopeInfo1)) {
|
|
ScopeInfo3 * tempScope = currentScope;
|
|
while (tempScope->parent) {
|
|
if (tempScope->recordTypes.find(start->str()) != tempScope->recordTypes.end()) {
|
|
std::string::size_type spaceIdx = 0;
|
|
std::string::size_type startIdx = 0;
|
|
while ((spaceIdx = removed1.find(" ", startIdx)) != std::string::npos) {
|
|
tok1->previous()->insertToken(removed1.substr(startIdx, spaceIdx - startIdx));
|
|
startIdx = spaceIdx + 1;
|
|
}
|
|
tok1->previous()->insertToken(removed1.substr(startIdx));
|
|
tok1->previous()->insertToken("::");
|
|
break;
|
|
}
|
|
idx = removed1.rfind(" ::");
|
|
if (idx == std::string::npos)
|
|
break;
|
|
|
|
removed1.resize(idx);
|
|
tempScope = tempScope->parent;
|
|
}
|
|
}
|
|
|
|
// just replace simple type aliases
|
|
TokenList::copyTokens(tok1, start, usingEnd->previous());
|
|
tok1->deleteThis();
|
|
substitute = true;
|
|
}
|
|
} else {
|
|
skip = true;
|
|
if (mSettings->debugwarnings && mErrorLogger) {
|
|
std::string str;
|
|
for (Token *tok3 = usingStart; tok3 && tok3 != usingEnd; tok3 = tok3->next()) {
|
|
if (!str.empty())
|
|
str += ' ';
|
|
str += tok3->str();
|
|
}
|
|
str += " ;";
|
|
std::list<const Token *> callstack(1, usingStart);
|
|
mErrorLogger->reportErr(ErrorMessage(callstack, &list, Severity::debug, "simplifyUsing",
|
|
"Failed to parse \'" + str + "\'. The checking continues anyway.", Certainty::normal));
|
|
}
|
|
}
|
|
tok1 = after;
|
|
}
|
|
|
|
if (!skip)
|
|
usingList.emplace_back(usingStart, usingEnd);
|
|
}
|
|
|
|
// delete all used type alias definitions
|
|
for (std::list<Using>::reverse_iterator it = usingList.rbegin(); it != usingList.rend(); ++it) {
|
|
Token *usingStart = it->startTok;
|
|
Token *usingEnd = it->endTok;
|
|
if (usingStart->previous()) {
|
|
if (usingEnd->next())
|
|
Token::eraseTokens(usingStart->previous(), usingEnd->next());
|
|
else {
|
|
Token::eraseTokens(usingStart->previous(), usingEnd);
|
|
usingEnd->deleteThis();
|
|
}
|
|
} else {
|
|
if (usingEnd->next()) {
|
|
Token::eraseTokens(usingStart, usingEnd->next());
|
|
usingStart->deleteThis();
|
|
} else {
|
|
// this is the only code being checked so leave ';'
|
|
Token::eraseTokens(usingStart, usingEnd);
|
|
usingStart->deleteThis();
|
|
}
|
|
}
|
|
}
|
|
|
|
return substitute;
|
|
}
|
|
void Tokenizer::simplifyMulAndParens()
|
|
{
|
|
if (!list.front())
|
|
return;
|
|
for (Token *tok = list.front()->tokAt(3); tok; tok = tok->next()) {
|
|
if (!tok->isName())
|
|
continue;
|
|
//fix ticket #2784 - improved by ticket #3184
|
|
int closedPars = 0;
|
|
Token *tokend = tok->next();
|
|
Token *tokbegin = tok->previous();
|
|
while (tokend && tokend->str() == ")") {
|
|
++closedPars;
|
|
tokend = tokend->next();
|
|
}
|
|
if (!tokend || !(tokend->isAssignmentOp()))
|
|
continue;
|
|
while (Token::Match(tokbegin, "&|(")) {
|
|
if (tokbegin->str() == "&") {
|
|
if (Token::Match(tokbegin->tokAt(-2), "[;{}&(] *")) {
|
|
//remove '* &'
|
|
tokbegin = tokbegin->tokAt(-2);
|
|
tokbegin->deleteNext(2);
|
|
} else if (Token::Match(tokbegin->tokAt(-3), "[;{}&(] * (")) {
|
|
if (closedPars == 0)
|
|
break;
|
|
--closedPars;
|
|
//remove ')'
|
|
tok->deleteNext();
|
|
//remove '* ( &'
|
|
tokbegin = tokbegin->tokAt(-3);
|
|
tokbegin->deleteNext(3);
|
|
} else
|
|
break;
|
|
} else if (tokbegin->str() == "(") {
|
|
if (closedPars == 0)
|
|
break;
|
|
|
|
//find consecutive opening parentheses
|
|
int openPars = 0;
|
|
while (tokbegin && tokbegin->str() == "(" && openPars <= closedPars) {
|
|
++openPars;
|
|
tokbegin = tokbegin->previous();
|
|
}
|
|
if (!tokbegin || openPars > closedPars)
|
|
break;
|
|
|
|
if ((openPars == closedPars && Token::Match(tokbegin, "[;{}]")) ||
|
|
Token::Match(tokbegin->tokAt(-2), "[;{}&(] * &") ||
|
|
Token::Match(tokbegin->tokAt(-3), "[;{}&(] * ( &")) {
|
|
//remove the excessive parentheses around the variable
|
|
while (openPars > 0) {
|
|
tok->deleteNext();
|
|
tokbegin->deleteNext();
|
|
--closedPars;
|
|
--openPars;
|
|
}
|
|
} else
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
bool Tokenizer::createTokens(std::istream &code,
|
|
const std::string& FileName)
|
|
{
|
|
// make sure settings specified
|
|
assert(mSettings);
|
|
|
|
return list.createTokens(code, FileName);
|
|
}
|
|
|
|
void Tokenizer::createTokens(simplecpp::TokenList&& tokenList)
|
|
{
|
|
// make sure settings specified
|
|
assert(mSettings);
|
|
list.createTokens(std::move(tokenList));
|
|
}
|
|
|
|
bool Tokenizer::simplifyTokens1(const std::string &configuration)
|
|
{
|
|
// Fill the map mTypeSize..
|
|
fillTypeSizes();
|
|
|
|
mConfiguration = configuration;
|
|
|
|
if (!simplifyTokenList1(list.getFiles().front().c_str()))
|
|
return false;
|
|
|
|
if (mTimerResults) {
|
|
Timer t("Tokenizer::simplifyTokens1::createAst", mSettings->showtime, mTimerResults);
|
|
list.createAst();
|
|
list.validateAst();
|
|
} else {
|
|
list.createAst();
|
|
list.validateAst();
|
|
}
|
|
|
|
if (mTimerResults) {
|
|
Timer t("Tokenizer::simplifyTokens1::createSymbolDatabase", mSettings->showtime, mTimerResults);
|
|
createSymbolDatabase();
|
|
} else {
|
|
createSymbolDatabase();
|
|
}
|
|
|
|
if (mTimerResults) {
|
|
Timer t("Tokenizer::simplifyTokens1::setValueType", mSettings->showtime, mTimerResults);
|
|
mSymbolDatabase->setValueTypeInTokenList(true);
|
|
} else {
|
|
mSymbolDatabase->setValueTypeInTokenList(true);
|
|
}
|
|
|
|
if (!mSettings->buildDir.empty())
|
|
Summaries::create(this, configuration);
|
|
|
|
// TODO: do not run valueflow if no checks are being performed at all - e.g. unusedFunctions only
|
|
const char* disableValueflowEnv = std::getenv("DISABLE_VALUEFLOW");
|
|
const bool doValueFlow = !disableValueflowEnv || (std::strcmp(disableValueflowEnv, "1") != 0);
|
|
|
|
if (doValueFlow) {
|
|
if (mTimerResults) {
|
|
Timer t("Tokenizer::simplifyTokens1::ValueFlow", mSettings->showtime, mTimerResults);
|
|
ValueFlow::setValues(&list, mSymbolDatabase, mErrorLogger, mSettings);
|
|
} else {
|
|
ValueFlow::setValues(&list, mSymbolDatabase, mErrorLogger, mSettings);
|
|
}
|
|
}
|
|
|
|
// Warn about unhandled character literals
|
|
if (mSettings->severity.isEnabled(Severity::portability)) {
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (tok->tokType() == Token::eChar && tok->values().empty()) {
|
|
try {
|
|
simplecpp::characterLiteralToLL(tok->str());
|
|
} catch (const std::exception &e) {
|
|
unhandledCharLiteral(tok, e.what());
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if (doValueFlow) {
|
|
mSymbolDatabase->setArrayDimensionsUsingValueFlow();
|
|
}
|
|
|
|
printDebugOutput(1);
|
|
|
|
return true;
|
|
}
|
|
|
|
bool Tokenizer::tokenize(std::istream &code,
|
|
const char FileName[],
|
|
const std::string &configuration)
|
|
{
|
|
if (!createTokens(code, FileName))
|
|
return false;
|
|
|
|
return simplifyTokens1(configuration);
|
|
}
|
|
//---------------------------------------------------------------------------
|
|
|
|
void Tokenizer::findComplicatedSyntaxErrorsInTemplates()
|
|
{
|
|
validate();
|
|
mTemplateSimplifier->checkComplicatedSyntaxErrorsInTemplates();
|
|
}
|
|
|
|
void Tokenizer::checkForEnumsWithTypedef()
|
|
{
|
|
for (const Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "enum %name% {")) {
|
|
tok = tok->tokAt(2);
|
|
const Token *tok2 = Token::findsimplematch(tok, "typedef", tok->link());
|
|
if (tok2)
|
|
syntaxError(tok2);
|
|
tok = tok->link();
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::fillTypeSizes()
|
|
{
|
|
mTypeSize.clear();
|
|
mTypeSize["char"] = 1;
|
|
mTypeSize["_Bool"] = mSettings->sizeof_bool;
|
|
mTypeSize["bool"] = mSettings->sizeof_bool;
|
|
mTypeSize["short"] = mSettings->sizeof_short;
|
|
mTypeSize["int"] = mSettings->sizeof_int;
|
|
mTypeSize["long"] = mSettings->sizeof_long;
|
|
mTypeSize["float"] = mSettings->sizeof_float;
|
|
mTypeSize["double"] = mSettings->sizeof_double;
|
|
mTypeSize["wchar_t"] = mSettings->sizeof_wchar_t;
|
|
mTypeSize["size_t"] = mSettings->sizeof_size_t;
|
|
mTypeSize["*"] = mSettings->sizeof_pointer;
|
|
}
|
|
|
|
void Tokenizer::combineOperators()
|
|
{
|
|
const bool cpp = isCPP();
|
|
|
|
// Combine tokens..
|
|
for (Token *tok = list.front(); tok && tok->next(); tok = tok->next()) {
|
|
const char c1 = tok->str()[0];
|
|
|
|
if (tok->str().length() == 1 && tok->next()->str().length() == 1) {
|
|
const char c2 = tok->next()->str()[0];
|
|
|
|
// combine +-*/ and =
|
|
if (c2 == '=' && (std::strchr("+-*/%|^=!<>", c1)) && !Token::Match(tok->previous(), "%type% *")) {
|
|
// skip templates
|
|
if (cpp && (tok->str() == ">" || Token::simpleMatch(tok->previous(), "> *"))) {
|
|
const Token* opening =
|
|
tok->str() == ">" ? tok->findOpeningBracket() : tok->previous()->findOpeningBracket();
|
|
if (opening && Token::Match(opening->previous(), "%name%"))
|
|
continue;
|
|
}
|
|
tok->str(tok->str() + c2);
|
|
tok->deleteNext();
|
|
continue;
|
|
}
|
|
} else if (tok->next()->str() == "=") {
|
|
if (tok->str() == ">>") {
|
|
tok->str(">>=");
|
|
tok->deleteNext();
|
|
} else if (tok->str() == "<<") {
|
|
tok->str("<<=");
|
|
tok->deleteNext();
|
|
}
|
|
} else if (cpp && (c1 == 'p' || c1 == '_') &&
|
|
Token::Match(tok, "private|protected|public|__published : !!:")) {
|
|
bool simplify = false;
|
|
int par = 0;
|
|
for (const Token *prev = tok->previous(); prev; prev = prev->previous()) {
|
|
if (prev->str() == ")") {
|
|
++par;
|
|
} else if (prev->str() == "(") {
|
|
if (par == 0U)
|
|
break;
|
|
--par;
|
|
}
|
|
if (par != 0U || prev->str() == "(")
|
|
continue;
|
|
if (Token::Match(prev, "[;{}]")) {
|
|
simplify = true;
|
|
break;
|
|
}
|
|
if (prev->isName() && prev->isUpperCaseName())
|
|
continue;
|
|
if (prev->isName() && endsWith(prev->str(), ':'))
|
|
simplify = true;
|
|
break;
|
|
}
|
|
if (simplify) {
|
|
tok->str(tok->str() + ":");
|
|
tok->deleteNext();
|
|
}
|
|
} else if (tok->str() == "->") {
|
|
// If the preceding sequence is "( & %name% )", replace it by "%name%"
|
|
Token *t = tok->tokAt(-4);
|
|
if (Token::Match(t, "( & %name% )") && !Token::simpleMatch(t->previous(), ">")) {
|
|
t->deleteThis();
|
|
t->deleteThis();
|
|
t->deleteNext();
|
|
tok->str(".");
|
|
} else {
|
|
tok->str(".");
|
|
tok->originalName("->");
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::combineStringAndCharLiterals()
|
|
{
|
|
// Combine strings
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!isStringLiteral(tok->str()))
|
|
continue;
|
|
|
|
tok->str(simplifyString(tok->str()));
|
|
|
|
while (Token::Match(tok->next(), "%str%") || Token::Match(tok->next(), "_T|_TEXT|TEXT ( %str% )")) {
|
|
if (tok->next()->isName()) {
|
|
if (!mSettings->isWindowsPlatform())
|
|
break;
|
|
tok->deleteNext(2);
|
|
tok->next()->deleteNext();
|
|
}
|
|
// Two strings after each other, combine them
|
|
tok->concatStr(simplifyString(tok->next()->str()));
|
|
tok->deleteNext();
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::concatenateNegativeNumberAndAnyPositive()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "?|:|,|(|[|{|return|case|sizeof|%op% +|-") || tok->tokType() == Token::eIncDecOp)
|
|
continue;
|
|
|
|
while (tok->str() != ">" && tok->next() && tok->next()->str() == "+")
|
|
tok->deleteNext();
|
|
|
|
if (Token::Match(tok->next(), "- %num%")) {
|
|
tok->deleteNext();
|
|
tok->next()->str("-" + tok->next()->str());
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyExternC()
|
|
{
|
|
if (isC())
|
|
return;
|
|
|
|
// Add attributes to all tokens within `extern "C"` inlines and blocks, and remove the `extern "C"` tokens.
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::simpleMatch(tok, "extern \"C\"")) {
|
|
Token *tok2 = tok->next();
|
|
if (tok->strAt(2) == "{") {
|
|
tok2 = tok2->next(); // skip {
|
|
while ((tok2 = tok2->next()) && tok2 != tok->linkAt(2))
|
|
tok2->isExternC(true);
|
|
tok->linkAt(2)->deleteThis(); // }
|
|
tok->deleteNext(2); // "C" {
|
|
} else {
|
|
while ((tok2 = tok2->next()) && !Token::simpleMatch(tok2, ";"))
|
|
tok2->isExternC(true);
|
|
tok->deleteNext(); // "C"
|
|
}
|
|
tok->deleteThis(); // extern
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyRoundCurlyParentheses()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
while (Token::Match(tok, "[;{}:] ( {") &&
|
|
Token::simpleMatch(tok->linkAt(2), "} ) ;")) {
|
|
if (tok->str() == ":" && !Token::Match(tok->tokAt(-2),"[;{}] %type% :"))
|
|
break;
|
|
Token *end = tok->linkAt(2)->tokAt(-3);
|
|
if (Token::Match(end, "[;{}] %num%|%str% ;"))
|
|
end->deleteNext(2);
|
|
tok->linkAt(2)->previous()->deleteNext(3);
|
|
tok->deleteNext(2);
|
|
}
|
|
if (Token::Match(tok, "( { %bool%|%char%|%num%|%str%|%name% ; } )")) {
|
|
tok->deleteNext();
|
|
tok->deleteThis();
|
|
tok->deleteNext(3);
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifySQL()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::simpleMatch(tok, "__CPPCHECK_EMBEDDED_SQL_EXEC__ SQL"))
|
|
continue;
|
|
|
|
const Token *end = findSQLBlockEnd(tok);
|
|
if (end == nullptr)
|
|
syntaxError(nullptr);
|
|
|
|
const std::string instruction = tok->stringifyList(end);
|
|
// delete all tokens until the embedded SQL block end
|
|
Token::eraseTokens(tok, end);
|
|
|
|
// insert "asm ( "instruction" ) ;"
|
|
tok->str("asm");
|
|
// it can happen that 'end' is NULL when wrong code is inserted
|
|
if (!tok->next())
|
|
tok->insertToken(";");
|
|
tok->insertToken(")");
|
|
tok->insertToken("\"" + instruction + "\"");
|
|
tok->insertToken("(");
|
|
// jump to ';' and continue
|
|
tok = tok->tokAt(3);
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyArrayAccessSyntax()
|
|
{
|
|
// 0[a] -> a[0]
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->isNumber() && Token::Match(tok, "%num% [ %name% ]")) {
|
|
const std::string number(tok->str());
|
|
Token* indexTok = tok->tokAt(2);
|
|
tok->str(indexTok->str());
|
|
tok->varId(indexTok->varId());
|
|
indexTok->str(number);
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyRedundantConsecutiveBraces()
|
|
{
|
|
// Remove redundant consecutive braces, i.e. '.. { { .. } } ..' -> '.. { .. } ..'.
|
|
for (Token *tok = list.front(); tok;) {
|
|
if (Token::simpleMatch(tok, "= {")) {
|
|
tok = tok->linkAt(1);
|
|
} else if (Token::simpleMatch(tok, "{ {") && Token::simpleMatch(tok->next()->link(), "} }")) {
|
|
//remove internal parentheses
|
|
tok->next()->link()->deleteThis();
|
|
tok->deleteNext();
|
|
} else
|
|
tok = tok->next();
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyDoublePlusAndDoubleMinus()
|
|
{
|
|
// Convert - - into + and + - into -
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
while (tok->next()) {
|
|
if (tok->str() == "+") {
|
|
if (tok->next()->str()[0] == '-') {
|
|
tok = tok->next();
|
|
if (tok->str().size() == 1) {
|
|
tok = tok->previous();
|
|
tok->str("-");
|
|
tok->deleteNext();
|
|
} else if (tok->isNumber()) {
|
|
tok->str(tok->str().substr(1));
|
|
tok = tok->previous();
|
|
tok->str("-");
|
|
}
|
|
continue;
|
|
}
|
|
} else if (tok->str() == "-") {
|
|
if (tok->next()->str()[0] == '-') {
|
|
tok = tok->next();
|
|
if (tok->str().size() == 1) {
|
|
tok = tok->previous();
|
|
tok->str("+");
|
|
tok->deleteNext();
|
|
} else if (tok->isNumber()) {
|
|
tok->str(tok->str().substr(1));
|
|
tok = tok->previous();
|
|
tok->str("+");
|
|
}
|
|
continue;
|
|
}
|
|
}
|
|
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
/** Specify array size if it hasn't been given */
|
|
|
|
void Tokenizer::arraySize()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!tok->isName() || !Token::Match(tok, "%var% [ ] ="))
|
|
continue;
|
|
bool addlength = false;
|
|
if (Token::Match(tok, "%var% [ ] = { %str% } ;")) {
|
|
Token *t = tok->tokAt(3);
|
|
t->deleteNext();
|
|
t->next()->deleteNext();
|
|
addlength = true;
|
|
}
|
|
|
|
if (addlength || Token::Match(tok, "%var% [ ] = %str% ;")) {
|
|
tok = tok->next();
|
|
const int sz = Token::getStrArraySize(tok->tokAt(3));
|
|
tok->insertToken(MathLib::toString(sz));
|
|
tok = tok->tokAt(5);
|
|
}
|
|
|
|
else if (Token::Match(tok, "%var% [ ] = {")) {
|
|
MathLib::biguint sz = 1;
|
|
tok = tok->next();
|
|
Token *end = tok->linkAt(3);
|
|
for (Token *tok2 = tok->tokAt(4); tok2 && tok2 != end; tok2 = tok2->next()) {
|
|
if (tok2->link() && Token::Match(tok2, "{|(|[|<")) {
|
|
if (tok2->str() == "[" && tok2->link()->strAt(1) == "=") { // designated initializer
|
|
if (Token::Match(tok2, "[ %num% ]"))
|
|
sz = std::max(sz, MathLib::toULongNumber(tok2->strAt(1)) + 1U);
|
|
else {
|
|
sz = 0;
|
|
break;
|
|
}
|
|
}
|
|
tok2 = tok2->link();
|
|
} else if (tok2->str() == ",") {
|
|
if (!Token::Match(tok2->next(), "[},]"))
|
|
++sz;
|
|
else {
|
|
tok2 = tok2->previous();
|
|
tok2->deleteNext();
|
|
}
|
|
}
|
|
}
|
|
|
|
if (sz != 0)
|
|
tok->insertToken(MathLib::toString(sz));
|
|
|
|
tok = end->next() ? end->next() : end;
|
|
}
|
|
}
|
|
}
|
|
|
|
static Token *skipTernaryOp(Token *tok)
|
|
{
|
|
int colonLevel = 1;
|
|
while (nullptr != (tok = tok->next())) {
|
|
if (tok->str() == "?") {
|
|
++colonLevel;
|
|
} else if (tok->str() == ":") {
|
|
--colonLevel;
|
|
if (colonLevel == 0) {
|
|
tok = tok->next();
|
|
break;
|
|
}
|
|
}
|
|
if (tok->link() && Token::Match(tok, "[(<]"))
|
|
tok = tok->link();
|
|
else if (Token::Match(tok->next(), "[{};)]"))
|
|
break;
|
|
}
|
|
if (colonLevel > 0) // Ticket #5214: Make sure the ':' matches the proper '?'
|
|
return nullptr;
|
|
return tok;
|
|
}
|
|
|
|
// Skips until the colon at the end of the case label, the argument must point to the "case" token.
|
|
// In case of success returns the colon token.
|
|
// In case of failure returns the token that caused the error.
|
|
static Token *skipCaseLabel(Token *tok)
|
|
{
|
|
assert(tok->str() == "case");
|
|
while (nullptr != (tok = tok->next())) {
|
|
if (Token::Match(tok, "(|["))
|
|
tok = tok->link();
|
|
else if (tok->str() == "?") {
|
|
Token * tok1 = skipTernaryOp(tok);
|
|
if (!tok1)
|
|
return tok;
|
|
tok = tok1;
|
|
}
|
|
if (Token::Match(tok, "[:{};]"))
|
|
return tok;
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
const Token * Tokenizer::startOfExecutableScope(const Token * tok)
|
|
{
|
|
if (tok->str() != ")")
|
|
return nullptr;
|
|
|
|
tok = isFunctionHead(tok, ":{", true);
|
|
|
|
if (Token::Match(tok, ": %name% [({]")) {
|
|
while (Token::Match(tok, "[:,] %name% [({]"))
|
|
tok = tok->linkAt(2)->next();
|
|
}
|
|
|
|
return (tok && tok->str() == "{") ? tok : nullptr;
|
|
}
|
|
|
|
|
|
/** simplify labels and case|default in the code: add a ";" if not already in.*/
|
|
|
|
void Tokenizer::simplifyLabelsCaseDefault()
|
|
{
|
|
const bool cpp = isCPP();
|
|
bool executablescope = false;
|
|
int indentLevel = 0;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
// Simplify labels in the executable scope..
|
|
Token *start = const_cast<Token *>(startOfExecutableScope(tok));
|
|
if (start) {
|
|
tok = start;
|
|
executablescope = true;
|
|
}
|
|
|
|
if (!executablescope)
|
|
continue;
|
|
|
|
if (tok->str() == "{") {
|
|
if (tok->previous()->str() == "=")
|
|
tok = tok->link();
|
|
else
|
|
++indentLevel;
|
|
} else if (tok->str() == "}") {
|
|
--indentLevel;
|
|
if (indentLevel == 0) {
|
|
executablescope = false;
|
|
continue;
|
|
}
|
|
} else if (Token::Match(tok, "(|["))
|
|
tok = tok->link();
|
|
|
|
if (Token::Match(tok, "[;{}:] case")) {
|
|
tok = skipCaseLabel(tok->next());
|
|
if (!tok)
|
|
break;
|
|
if (tok->str() != ":" || tok->strAt(-1) == "case" || !tok->next())
|
|
syntaxError(tok);
|
|
if (tok->next()->str() != ";" && tok->next()->str() != "case")
|
|
tok->insertToken(";");
|
|
else
|
|
tok = tok->previous();
|
|
} else if (Token::Match(tok, "[;{}] %name% : !!;")) {
|
|
if (!cpp || !Token::Match(tok->next(), "class|struct|enum")) {
|
|
tok = tok->tokAt(2);
|
|
tok->insertToken(";");
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
void Tokenizer::simplifyCaseRange()
|
|
{
|
|
for (Token* tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "case %num%|%char% ... %num%|%char% :")) {
|
|
const MathLib::bigint start = MathLib::toLongNumber(tok->strAt(1));
|
|
MathLib::bigint end = MathLib::toLongNumber(tok->strAt(3));
|
|
end = std::min(start + 50, end); // Simplify it 50 times at maximum
|
|
if (start < end) {
|
|
tok = tok->tokAt(2);
|
|
tok->str(":");
|
|
tok->insertToken("case");
|
|
for (MathLib::bigint i = end-1; i > start; i--) {
|
|
tok->insertToken(":");
|
|
tok->insertToken(MathLib::toString(i));
|
|
tok->insertToken("case");
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::calculateScopes()
|
|
{
|
|
for (auto *tok = list.front(); tok; tok = tok->next())
|
|
tok->scopeInfo(nullptr);
|
|
|
|
std::string nextScopeNameAddition;
|
|
std::shared_ptr<ScopeInfo2> primaryScope = std::make_shared<ScopeInfo2>("", nullptr);
|
|
list.front()->scopeInfo(primaryScope);
|
|
|
|
for (Token* tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok == list.front() || !tok->scopeInfo()) {
|
|
if (tok != list.front())
|
|
tok->scopeInfo(tok->previous()->scopeInfo());
|
|
|
|
if (Token::Match(tok, "using namespace %name% ::|<|;")) {
|
|
std::string usingNamespaceName;
|
|
for (const Token* namespaceNameToken = tok->tokAt(2);
|
|
namespaceNameToken && namespaceNameToken->str() != ";";
|
|
namespaceNameToken = namespaceNameToken->next()) {
|
|
usingNamespaceName += namespaceNameToken->str();
|
|
usingNamespaceName += " ";
|
|
}
|
|
if (usingNamespaceName.length() > 0)
|
|
usingNamespaceName = usingNamespaceName.substr(0, usingNamespaceName.length() - 1);
|
|
tok->scopeInfo()->usingNamespaces.insert(usingNamespaceName);
|
|
} else if (Token::Match(tok, "namespace|class|struct|union %name% {|::|:|<")) {
|
|
for (Token* nameTok = tok->next(); nameTok && !Token::Match(nameTok, "{|:"); nameTok = nameTok->next()) {
|
|
if (Token::Match(nameTok, ";|<")) {
|
|
nextScopeNameAddition = "";
|
|
break;
|
|
}
|
|
nextScopeNameAddition.append(nameTok->str());
|
|
nextScopeNameAddition.append(" ");
|
|
}
|
|
if (nextScopeNameAddition.length() > 0)
|
|
nextScopeNameAddition = nextScopeNameAddition.substr(0, nextScopeNameAddition.length() - 1);
|
|
}
|
|
|
|
if (Token::simpleMatch(tok, "{")) {
|
|
// This might be the opening of a member function
|
|
Token *tok1 = tok;
|
|
while (Token::Match(tok1->previous(), "const|volatile|final|override|&|&&|noexcept"))
|
|
tok1 = tok1->previous();
|
|
if (tok1->previous() && tok1->strAt(-1) == ")") {
|
|
bool member = true;
|
|
tok1 = tok1->linkAt(-1);
|
|
if (Token::Match(tok1->previous(), "throw|noexcept")) {
|
|
tok1 = tok1->previous();
|
|
while (Token::Match(tok1->previous(), "const|volatile|final|override|&|&&|noexcept"))
|
|
tok1 = tok1->previous();
|
|
if (tok1->strAt(-1) != ")")
|
|
member = false;
|
|
} else if (Token::Match(tok->tokAt(-2), ":|, %name%")) {
|
|
tok1 = tok1->tokAt(-2);
|
|
if (tok1->strAt(-1) != ")")
|
|
member = false;
|
|
}
|
|
if (member) {
|
|
if (tok1->strAt(-1) == ">")
|
|
tok1 = tok1->previous()->findOpeningBracket();
|
|
if (tok1 && Token::Match(tok1->tokAt(-3), "%name% :: %name%")) {
|
|
tok1 = tok1->tokAt(-2);
|
|
std::string scope = tok1->strAt(-1);
|
|
while (Token::Match(tok1->tokAt(-2), ":: %name%")) {
|
|
scope = tok1->strAt(-3) + " :: " + scope;
|
|
tok1 = tok1->tokAt(-2);
|
|
}
|
|
|
|
if (!nextScopeNameAddition.empty() && !scope.empty())
|
|
nextScopeNameAddition += " :: ";
|
|
nextScopeNameAddition += scope;
|
|
}
|
|
}
|
|
}
|
|
|
|
// New scope is opening, record it here
|
|
std::shared_ptr<ScopeInfo2> newScopeInfo = std::make_shared<ScopeInfo2>(tok->scopeInfo()->name, tok->link(), tok->scopeInfo()->usingNamespaces);
|
|
|
|
if (newScopeInfo->name != "" && nextScopeNameAddition != "")
|
|
newScopeInfo->name.append(" :: ");
|
|
newScopeInfo->name.append(nextScopeNameAddition);
|
|
nextScopeNameAddition = "";
|
|
|
|
if (tok->link())
|
|
tok->link()->scopeInfo(tok->scopeInfo());
|
|
tok->scopeInfo(newScopeInfo);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyTemplates()
|
|
{
|
|
if (isC())
|
|
return;
|
|
|
|
mTemplateSimplifier->simplifyTemplates(
|
|
#ifdef MAXTIME
|
|
mMaxTime,
|
|
#else
|
|
0, // ignored
|
|
#endif
|
|
mCodeWithTemplates);
|
|
}
|
|
//---------------------------------------------------------------------------
|
|
|
|
|
|
static bool setVarIdParseDeclaration(const Token **tok, const std::map<std::string,int> &variableId, bool executableScope, bool cpp, bool c)
|
|
{
|
|
const Token *tok2 = *tok;
|
|
if (!tok2->isName())
|
|
return false;
|
|
|
|
int typeCount = 0;
|
|
int singleNameCount = 0;
|
|
bool hasstruct = false; // Is there a "struct" or "class"?
|
|
bool bracket = false;
|
|
bool ref = false;
|
|
while (tok2) {
|
|
if (tok2->isName()) {
|
|
if (cpp && Token::Match(tok2, "namespace|public|private|protected"))
|
|
return false;
|
|
if (cpp && Token::simpleMatch(tok2, "decltype (")) {
|
|
typeCount = 1;
|
|
tok2 = tok2->linkAt(1)->next();
|
|
continue;
|
|
}
|
|
if (Token::Match(tok2, "struct|union|enum") || (!c && Token::Match(tok2, "class|typename"))) {
|
|
hasstruct = true;
|
|
typeCount = 0;
|
|
singleNameCount = 0;
|
|
} else if (tok2->str() == "const") {
|
|
// just skip "const"
|
|
} else if (!hasstruct && variableId.find(tok2->str()) != variableId.end() && tok2->previous()->str() != "::") {
|
|
++typeCount;
|
|
tok2 = tok2->next();
|
|
if (!tok2 || tok2->str() != "::")
|
|
break;
|
|
} else {
|
|
if (tok2->str() != "void" || Token::Match(tok2, "void const| *|(")) // just "void" cannot be a variable type
|
|
++typeCount;
|
|
++singleNameCount;
|
|
}
|
|
} else if (!c && ((TemplateSimplifier::templateParameters(tok2) > 0) ||
|
|
Token::simpleMatch(tok2, "< >") /* Ticket #4764 */)) {
|
|
const Token *start = *tok;
|
|
if (Token::Match(start->previous(), "%or%|%oror%|&&|&|^|+|-|*|/"))
|
|
return false;
|
|
const Token * const closingBracket = tok2->findClosingBracket();
|
|
if (closingBracket == nullptr) { /* Ticket #8151 */
|
|
throw tok2;
|
|
}
|
|
tok2 = closingBracket;
|
|
if (tok2->str() != ">")
|
|
break;
|
|
singleNameCount = 1;
|
|
if (Token::Match(tok2, "> %name% %or%|%oror%|&&|&|^|+|-|*|/") && !Token::Match(tok2, "> const [*&]"))
|
|
return false;
|
|
if (Token::Match(tok2, "> %name% )")) {
|
|
if (Token::Match(tok2->linkAt(2)->previous(), "if|for|while ("))
|
|
return false;
|
|
if (!Token::Match(tok2->linkAt(2)->previous(), "%name%|] ("))
|
|
return false;
|
|
}
|
|
} else if (Token::Match(tok2, "&|&&")) {
|
|
ref = !bracket;
|
|
} else if (singleNameCount >= 1 && Token::Match(tok2, "( [*&]") && Token::Match(tok2->link()->next(), "(|[")) {
|
|
bracket = true; // Skip: Seems to be valid pointer to array or function pointer
|
|
} else if (singleNameCount >= 1 && Token::Match(tok2, "( * %name% [") && Token::Match(tok2->linkAt(3), "] ) [;,]")) {
|
|
bracket = true;
|
|
} else if (tok2->str() == "::") {
|
|
singleNameCount = 0;
|
|
} else if (tok2->str() != "*" && tok2->str() != "::" && tok2->str() != "...") {
|
|
break;
|
|
}
|
|
tok2 = tok2->next();
|
|
}
|
|
|
|
if (tok2) {
|
|
bool isLambdaArg = false;
|
|
{
|
|
const Token *tok3 = (*tok)->previous();
|
|
if (tok3 && tok3->str() == ",") {
|
|
while (tok3 && !Token::Match(tok3,";|(|[|{")) {
|
|
if (Token::Match(tok3, ")|]"))
|
|
tok3 = tok3->link();
|
|
tok3 = tok3->previous();
|
|
}
|
|
|
|
if (tok3 && executableScope && Token::Match(tok3->previous(), "%name% (")) {
|
|
const Token *fdecl = tok3->previous();
|
|
int count = 0;
|
|
while (Token::Match(fdecl, "%name%|*")) {
|
|
fdecl = fdecl->previous();
|
|
count++;
|
|
}
|
|
if (!Token::Match(fdecl, "[;{}] %name%") || count <= 1)
|
|
return false;
|
|
}
|
|
}
|
|
|
|
if (cpp && tok3 && Token::simpleMatch(tok3->previous(), "] (") &&
|
|
(Token::simpleMatch(tok3->link(), ") {") || Token::Match(tok3->link(), ") . %name%")))
|
|
isLambdaArg = true;
|
|
}
|
|
|
|
|
|
*tok = tok2;
|
|
|
|
// In executable scopes, references must be assigned
|
|
// Catching by reference is an exception
|
|
if (executableScope && ref && !isLambdaArg) {
|
|
if (Token::Match(tok2, "(|=|{|:"))
|
|
; // reference is assigned => ok
|
|
else if (tok2->str() != ")" || tok2->link()->strAt(-1) != "catch")
|
|
return false; // not catching by reference => not declaration
|
|
}
|
|
}
|
|
|
|
// Check if array declaration is valid (#2638)
|
|
// invalid declaration: AAA a[4] = 0;
|
|
if (typeCount >= 2 && executableScope && tok2 && tok2->str() == "[") {
|
|
const Token *tok3 = tok2->link()->next();
|
|
while (tok3 && tok3->str() == "[") {
|
|
tok3 = tok3->link()->next();
|
|
}
|
|
if (Token::Match(tok3, "= %num%"))
|
|
return false;
|
|
}
|
|
|
|
return (typeCount >= 2 && tok2 && Token::Match(tok2->tokAt(-2), "!!:: %type%"));
|
|
}
|
|
|
|
|
|
void Tokenizer::setVarIdStructMembers(Token **tok1,
|
|
std::map<int, std::map<std::string, int>>& structMembers,
|
|
nonneg int *varId) const
|
|
{
|
|
Token *tok = *tok1;
|
|
|
|
if (Token::Match(tok, "%name% = { . %name% =|{")) {
|
|
const int struct_varid = tok->varId();
|
|
if (struct_varid == 0)
|
|
return;
|
|
|
|
std::map<std::string, int>& members = structMembers[struct_varid];
|
|
|
|
tok = tok->tokAt(3);
|
|
while (tok->str() != "}") {
|
|
if (Token::Match(tok, "{|[|("))
|
|
tok = tok->link();
|
|
if (Token::Match(tok->previous(), "[,{] . %name% =|{")) {
|
|
tok = tok->next();
|
|
const std::map<std::string, int>::iterator it = members.find(tok->str());
|
|
if (it == members.end()) {
|
|
members[tok->str()] = ++(*varId);
|
|
tok->varId(*varId);
|
|
} else {
|
|
tok->varId(it->second);
|
|
}
|
|
}
|
|
tok = tok->next();
|
|
}
|
|
|
|
return;
|
|
}
|
|
|
|
while (Token::Match(tok->next(), ")| . %name% !!(")) {
|
|
// Don't set varid for trailing return type
|
|
if (tok->strAt(1) == ")" && (tok->linkAt(1)->previous()->isName() || tok->linkAt(1)->strAt(-1) == "]") &&
|
|
isFunctionHead(tok->linkAt(1), "{|;")) {
|
|
tok = tok->tokAt(3);
|
|
continue;
|
|
}
|
|
const int struct_varid = tok->varId();
|
|
tok = tok->tokAt(2);
|
|
if (struct_varid == 0)
|
|
continue;
|
|
|
|
if (tok->str() == ".")
|
|
tok = tok->next();
|
|
|
|
// Don't set varid for template function
|
|
if (TemplateSimplifier::templateParameters(tok->next()) > 0)
|
|
break;
|
|
|
|
std::map<std::string, int>& members = structMembers[struct_varid];
|
|
const std::map<std::string, int>::iterator it = members.find(tok->str());
|
|
if (it == members.end()) {
|
|
members[tok->str()] = ++(*varId);
|
|
tok->varId(*varId);
|
|
} else {
|
|
tok->varId(it->second);
|
|
}
|
|
}
|
|
// tok can't be null
|
|
*tok1 = tok;
|
|
}
|
|
|
|
|
|
void Tokenizer::setVarIdClassDeclaration(const Token * const startToken,
|
|
const VariableMap &variableMap,
|
|
const nonneg int scopeStartVarId,
|
|
std::map<int, std::map<std::string,int>>& structMembers)
|
|
{
|
|
// end of scope
|
|
const Token * const endToken = startToken->link();
|
|
|
|
// determine class name
|
|
std::string className;
|
|
for (const Token *tok = startToken->previous(); tok; tok = tok->previous()) {
|
|
if (!tok->isName() && tok->str() != ":")
|
|
break;
|
|
if (Token::Match(tok, "class|struct|enum %type% [:{]")) {
|
|
className = tok->next()->str();
|
|
break;
|
|
}
|
|
}
|
|
|
|
// replace varids..
|
|
int indentlevel = 0;
|
|
bool initList = false;
|
|
bool inEnum = false;
|
|
const Token *initListArgLastToken = nullptr;
|
|
for (Token *tok = startToken->next(); tok != endToken; tok = tok->next()) {
|
|
if (!tok)
|
|
syntaxError(nullptr);
|
|
if (initList) {
|
|
if (tok == initListArgLastToken)
|
|
initListArgLastToken = nullptr;
|
|
else if (!initListArgLastToken &&
|
|
Token::Match(tok->previous(), "%name%|>|>> {|(") &&
|
|
Token::Match(tok->link(), "}|) ,|{"))
|
|
initListArgLastToken = tok->link();
|
|
}
|
|
if (tok->str() == "{") {
|
|
inEnum = isEnumStart(tok);
|
|
if (initList && !initListArgLastToken)
|
|
initList = false;
|
|
++indentlevel;
|
|
} else if (tok->str() == "}") {
|
|
--indentlevel;
|
|
inEnum = false;
|
|
} else if (initList && indentlevel == 0 && Token::Match(tok->previous(), "[,:] %name% [({]")) {
|
|
const std::map<std::string, int>::const_iterator it = variableMap.find(tok->str());
|
|
if (it != variableMap.end()) {
|
|
tok->varId(it->second);
|
|
}
|
|
} else if (tok->isName() && tok->varId() <= scopeStartVarId) {
|
|
if (indentlevel > 0 || initList) {
|
|
if (Token::Match(tok->previous(), "::|.") && tok->strAt(-2) != "this" && !Token::simpleMatch(tok->tokAt(-5), "( * this ) ."))
|
|
continue;
|
|
if (!tok->next())
|
|
syntaxError(nullptr);
|
|
if (tok->next()->str() == "::") {
|
|
if (tok->str() == className)
|
|
tok = tok->tokAt(2);
|
|
else
|
|
continue;
|
|
}
|
|
|
|
if (!inEnum) {
|
|
const std::map<std::string, int>::const_iterator it = variableMap.find(tok->str());
|
|
if (it != variableMap.end()) {
|
|
tok->varId(it->second);
|
|
setVarIdStructMembers(&tok, structMembers, variableMap.getVarId());
|
|
}
|
|
}
|
|
}
|
|
} else if (indentlevel == 0 && tok->str() == ":" && !initListArgLastToken)
|
|
initList = true;
|
|
}
|
|
}
|
|
|
|
|
|
|
|
// Update the variable ids..
|
|
// Parse each function..
|
|
void Tokenizer::setVarIdClassFunction(const std::string &classname,
|
|
Token * const startToken,
|
|
const Token * const endToken,
|
|
const std::map<std::string, int> &varlist,
|
|
std::map<int, std::map<std::string, int>>& structMembers,
|
|
nonneg int *varId_)
|
|
{
|
|
for (Token *tok2 = startToken; tok2 && tok2 != endToken; tok2 = tok2->next()) {
|
|
if (tok2->varId() != 0 || !tok2->isName())
|
|
continue;
|
|
if (Token::Match(tok2->tokAt(-2), ("!!" + classname + " ::").c_str()))
|
|
continue;
|
|
if (Token::Match(tok2->tokAt(-4), "%name% :: %name% ::")) // Currently unsupported
|
|
continue;
|
|
if (Token::Match(tok2->tokAt(-2), "!!this .") && !Token::simpleMatch(tok2->tokAt(-5), "( * this ) ."))
|
|
continue;
|
|
if (Token::Match(tok2, "%name% ::"))
|
|
continue;
|
|
|
|
const std::map<std::string,int>::const_iterator it = varlist.find(tok2->str());
|
|
if (it != varlist.end()) {
|
|
tok2->varId(it->second);
|
|
setVarIdStructMembers(&tok2, structMembers, varId_);
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
|
|
void Tokenizer::setVarId()
|
|
{
|
|
// Clear all variable ids
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->isName())
|
|
tok->varId(0);
|
|
}
|
|
|
|
setPodTypes();
|
|
|
|
setVarIdPass1();
|
|
|
|
setVarIdPass2();
|
|
}
|
|
|
|
|
|
// Variable declarations can't start with "return" etc.
|
|
#define NOTSTART_C "NOT", "case", "default", "goto", "not", "return", "sizeof", "typedef"
|
|
static const std::unordered_set<std::string> notstart_c = { NOTSTART_C };
|
|
static const std::unordered_set<std::string> notstart_cpp = { NOTSTART_C,
|
|
"delete", "friend", "new", "throw", "using", "virtual", "explicit", "const_cast", "dynamic_cast", "reinterpret_cast", "static_cast", "template"
|
|
};
|
|
|
|
void Tokenizer::setVarIdPass1()
|
|
{
|
|
// Variable declarations can't start with "return" etc.
|
|
const std::unordered_set<std::string>& notstart = (isC()) ? notstart_c : notstart_cpp;
|
|
|
|
VariableMap variableMap;
|
|
std::map<int, std::map<std::string, int>> structMembers;
|
|
|
|
std::stack<VarIdScopeInfo> scopeStack;
|
|
|
|
scopeStack.push(VarIdScopeInfo());
|
|
std::stack<const Token *> functionDeclEndStack;
|
|
const Token *functionDeclEndToken = nullptr;
|
|
bool initlist = false;
|
|
bool inlineFunction = false;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->isOp())
|
|
continue;
|
|
if (tok == functionDeclEndToken) {
|
|
functionDeclEndStack.pop();
|
|
functionDeclEndToken = functionDeclEndStack.empty() ? nullptr : functionDeclEndStack.top();
|
|
if (tok->str() == ":")
|
|
initlist = true;
|
|
else if (tok->str() == ";") {
|
|
if (!variableMap.leaveScope())
|
|
cppcheckError(tok);
|
|
} else if (tok->str() == "{") {
|
|
scopeStack.push(VarIdScopeInfo(true, scopeStack.top().isStructInit || tok->strAt(-1) == "=", /*isEnum=*/ false, *variableMap.getVarId()));
|
|
|
|
// check if this '{' is a start of an "if" body
|
|
const Token * ifToken = tok->previous();
|
|
if (ifToken && ifToken->str() == ")")
|
|
ifToken = ifToken->link();
|
|
else
|
|
ifToken = nullptr;
|
|
if (ifToken)
|
|
ifToken = ifToken->previous();
|
|
if (ifToken && ifToken->str() == "if") {
|
|
// open another scope to differentiate between variables declared in the "if" condition and in the "if" body
|
|
variableMap.enterScope();
|
|
}
|
|
}
|
|
} else if (!initlist && tok->str()=="(") {
|
|
const Token * newFunctionDeclEnd = nullptr;
|
|
if (!scopeStack.top().isExecutable)
|
|
newFunctionDeclEnd = isFunctionHead(tok, "{:;");
|
|
else {
|
|
Token const * const tokenLinkNext = tok->link()->next();
|
|
if (tokenLinkNext && tokenLinkNext->str() == "{") // might be for- or while-loop or if-statement
|
|
newFunctionDeclEnd = tokenLinkNext;
|
|
}
|
|
if (newFunctionDeclEnd && newFunctionDeclEnd != functionDeclEndToken) {
|
|
functionDeclEndStack.push(newFunctionDeclEnd);
|
|
functionDeclEndToken = newFunctionDeclEnd;
|
|
variableMap.enterScope();
|
|
}
|
|
} else if (Token::Match(tok, "{|}")) {
|
|
inlineFunction = false;
|
|
|
|
const Token * const startToken = (tok->str() == "{") ? tok : tok->link();
|
|
|
|
// parse anonymous unions as part of the current scope
|
|
if (!Token::Match(startToken->previous(), "union|struct|enum {") &&
|
|
!(initlist && Token::Match(startToken->previous(), "%name%|>|>>") && Token::Match(startToken->link(), "} ,|{"))) {
|
|
|
|
if (tok->str() == "{") {
|
|
bool isExecutable;
|
|
const Token *prev = tok->previous();
|
|
while (Token::Match(prev, "%name%|."))
|
|
prev = prev->previous();
|
|
const bool isLambda = prev && prev->str() == ")" && Token::simpleMatch(prev->link()->previous(), "] (");
|
|
if ((!isLambda && (tok->strAt(-1) == ")" || Token::Match(tok->tokAt(-2), ") %type%"))) ||
|
|
(initlist && tok->strAt(-1) == "}")) {
|
|
isExecutable = true;
|
|
} else {
|
|
isExecutable = ((scopeStack.top().isExecutable || initlist || tok->strAt(-1) == "else") &&
|
|
!isClassStructUnionEnumStart(tok));
|
|
if (!(scopeStack.top().isStructInit || tok->strAt(-1) == "="))
|
|
variableMap.enterScope();
|
|
}
|
|
initlist = false;
|
|
scopeStack.push(VarIdScopeInfo(isExecutable, scopeStack.top().isStructInit || tok->strAt(-1) == "=", isEnumStart(tok), *variableMap.getVarId()));
|
|
} else { /* if (tok->str() == "}") */
|
|
bool isNamespace = false;
|
|
for (const Token *tok1 = tok->link()->previous(); tok1 && tok1->isName(); tok1 = tok1->previous()) {
|
|
if (tok1->str() == "namespace") {
|
|
isNamespace = true;
|
|
break;
|
|
}
|
|
}
|
|
// Set variable ids in class declaration..
|
|
if (!initlist && !isC() && !scopeStack.top().isExecutable && tok->link() && !isNamespace) {
|
|
setVarIdClassDeclaration(tok->link(),
|
|
variableMap,
|
|
scopeStack.top().startVarid,
|
|
structMembers);
|
|
}
|
|
|
|
if (!scopeStack.top().isStructInit) {
|
|
variableMap.leaveScope();
|
|
|
|
// check if this '}' is an end of an "else" body or an "if" body without an "else" part
|
|
const Token * ifToken = startToken->previous();
|
|
if (ifToken && ifToken->str() == ")")
|
|
ifToken = ifToken->link()->previous();
|
|
else
|
|
ifToken = nullptr;
|
|
if (startToken->strAt(-1) == "else" || (ifToken && ifToken->str() == "if" && tok->strAt(1) != "else")) {
|
|
// leave the extra scope used to differentiate between variables declared in the "if" condition and in the "if" body
|
|
variableMap.leaveScope();
|
|
}
|
|
}
|
|
|
|
scopeStack.pop();
|
|
if (scopeStack.empty()) { // should be impossible
|
|
scopeStack.push(VarIdScopeInfo());
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if (!scopeStack.top().isStructInit &&
|
|
(tok == list.front() ||
|
|
Token::Match(tok, "[;{}]") ||
|
|
(tok->str() == "(" && isFunctionHead(tok,"{")) ||
|
|
(tok->str() == "(" && !scopeStack.top().isExecutable && isFunctionHead(tok,";:")) ||
|
|
(tok->str() == "," && (!scopeStack.top().isExecutable || inlineFunction)) ||
|
|
(tok->isName() && endsWith(tok->str(), ':')))) {
|
|
|
|
// No variable declarations in sizeof
|
|
if (Token::simpleMatch(tok->previous(), "sizeof (")) {
|
|
continue;
|
|
}
|
|
|
|
if (Settings::terminated())
|
|
return;
|
|
|
|
// locate the variable name..
|
|
const Token *tok2 = (tok->isName()) ? tok : tok->next();
|
|
|
|
// private: protected: public: etc
|
|
while (tok2 && endsWith(tok2->str(), ':')) {
|
|
tok2 = tok2->next();
|
|
}
|
|
if (!tok2)
|
|
break;
|
|
|
|
// Variable declaration can't start with "return", etc
|
|
if (notstart.find(tok2->str()) != notstart.end())
|
|
continue;
|
|
|
|
if (!isC() && Token::simpleMatch(tok2, "const new"))
|
|
continue;
|
|
|
|
bool decl;
|
|
if (isCPP() && mSettings->standards.cpp >= Standards::CPP17 && Token::Match(tok, "[(;{}] const| auto &|&&| [")) {
|
|
// Structured bindings
|
|
tok2 = Token::findsimplematch(tok, "[");
|
|
if ((Token::simpleMatch(tok->previous(), "for (") && Token::simpleMatch(tok2->link(), "] :")) ||
|
|
Token::simpleMatch(tok2->link(), "] =")) {
|
|
while (tok2 && tok2->str() != "]") {
|
|
if (Token::Match(tok2, "%name% [,]]"))
|
|
variableMap.addVariable(tok2->str());
|
|
tok2 = tok2->next();
|
|
}
|
|
continue;
|
|
}
|
|
}
|
|
|
|
try { /* Ticket #8151 */
|
|
decl = setVarIdParseDeclaration(&tok2, variableMap.map(), scopeStack.top().isExecutable, isCPP(), isC());
|
|
} catch (const Token * errTok) {
|
|
syntaxError(errTok);
|
|
}
|
|
if (decl) {
|
|
if (isCPP()) {
|
|
if (Token *declTypeTok = Token::findsimplematch(tok, "decltype (", tok2)) {
|
|
for (Token *declTok = declTypeTok->linkAt(1); declTok != declTypeTok; declTok = declTok->previous()) {
|
|
if (declTok->isName() && !Token::Match(declTok->previous(), "::|.") && variableMap.hasVariable(declTok->str()))
|
|
declTok->varId(variableMap.find(declTok->str())->second);
|
|
}
|
|
}
|
|
}
|
|
|
|
if (tok->str() == "(" && isFunctionHead(tok,"{") && scopeStack.top().isExecutable)
|
|
inlineFunction = true;
|
|
|
|
const Token* prev2 = tok2->previous();
|
|
if (Token::Match(prev2, "%type% [;[=,)]") && tok2->previous()->str() != "const")
|
|
;
|
|
else if (Token::Match(prev2, "%type% :") && tok->strAt(-1) == "for")
|
|
;
|
|
else if (Token::Match(prev2, "%type% ( !!)") && Token::simpleMatch(tok2->link(), ") ;")) {
|
|
// In C++ , a variable can't be called operator+ or something like that.
|
|
if (isCPP() &&
|
|
prev2->isOperatorKeyword())
|
|
continue;
|
|
|
|
const Token *tok3 = tok2->next();
|
|
if (!tok3->isStandardType() && tok3->str() != "void" && !Token::Match(tok3, "struct|union|class %type%") && tok3->str() != "." && !Token::Match(tok2->link()->previous(), "[&*]")) {
|
|
if (!scopeStack.top().isExecutable) {
|
|
// Detecting initializations with () in non-executable scope is hard and often impossible to be done safely. Thus, only treat code as a variable that definitely is one.
|
|
decl = false;
|
|
bool rhs = false;
|
|
for (; tok3; tok3 = tok3->nextArgumentBeforeCreateLinks2()) {
|
|
if (tok3->str() == "=") {
|
|
rhs = true;
|
|
continue;
|
|
}
|
|
|
|
if (tok3->str() == ",") {
|
|
rhs = false;
|
|
continue;
|
|
}
|
|
|
|
if (rhs)
|
|
continue;
|
|
|
|
if (tok3->isLiteral() ||
|
|
(tok3->isName() && variableMap.hasVariable(tok3->str())) ||
|
|
tok3->isOp() ||
|
|
tok3->str() == "(" ||
|
|
notstart.find(tok3->str()) != notstart.end()) {
|
|
decl = true;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
} else
|
|
decl = false;
|
|
} else if (isCPP() && Token::Match(prev2, "%type% {") && Token::simpleMatch(tok2->link(), "} ;")) { // C++11 initialization style
|
|
if (tok2->link() != tok2->next() && // add value-initialized variable T x{};
|
|
(Token::Match(prev2, "do|try|else") || Token::Match(prev2->tokAt(-2), "struct|class|:")))
|
|
continue;
|
|
} else
|
|
decl = false;
|
|
|
|
if (decl) {
|
|
variableMap.addVariable(prev2->str());
|
|
|
|
if (Token::simpleMatch(tok->previous(), "for (") && Token::Match(prev2, "%name% [=,]")) {
|
|
for (const Token *tok3 = prev2->next(); tok3 && tok3->str() != ";"; tok3 = tok3->next()) {
|
|
if (Token::Match(tok3, "[([]"))
|
|
tok3 = tok3->link();
|
|
if (Token::Match(tok3, ", %name% [,=;]"))
|
|
variableMap.addVariable(tok3->next()->str());
|
|
}
|
|
}
|
|
|
|
// set varid for template parameters..
|
|
tok = tok->next();
|
|
while (Token::Match(tok, "%name%|::"))
|
|
tok = tok->next();
|
|
if (tok && tok->str() == "<") {
|
|
const Token *end = tok->findClosingBracket();
|
|
while (tok != end) {
|
|
if (tok->isName() && !(Token::simpleMatch(tok->next(), "<") &&
|
|
Token::Match(tok->tokAt(-2), "std :: %name%"))) {
|
|
const std::map<std::string, int>::const_iterator it = variableMap.find(tok->str());
|
|
if (it != variableMap.end())
|
|
tok->varId(it->second);
|
|
}
|
|
tok = tok->next();
|
|
}
|
|
}
|
|
|
|
tok = tok2->previous();
|
|
}
|
|
}
|
|
}
|
|
|
|
if (tok->isName()) {
|
|
// don't set variable id after a struct|enum|union
|
|
if (Token::Match(tok->previous(), "struct|enum|union") || (isCPP() && tok->strAt(-1) == "class"))
|
|
continue;
|
|
|
|
if (!isC()) {
|
|
if (tok->previous() && tok->previous()->str() == "::")
|
|
continue;
|
|
if (tok->next() && tok->next()->str() == "::")
|
|
continue;
|
|
if (Token::simpleMatch(tok->tokAt(-2), ":: template"))
|
|
continue;
|
|
}
|
|
|
|
// function declaration inside executable scope? Function declaration is of form: type name "(" args ")"
|
|
if (scopeStack.top().isExecutable && Token::Match(tok, "%name% [,)]")) {
|
|
bool par = false;
|
|
const Token *start, *end;
|
|
|
|
// search begin of function declaration
|
|
for (start = tok; Token::Match(start, "%name%|*|&|,|("); start = start->previous()) {
|
|
if (start->str() == "(") {
|
|
if (par)
|
|
break;
|
|
par = true;
|
|
}
|
|
if (Token::Match(start, "[(,]")) {
|
|
if (!Token::Match(start, "[(,] %type% %name%|*|&"))
|
|
break;
|
|
}
|
|
if (start->varId() > 0)
|
|
break;
|
|
}
|
|
|
|
// search end of function declaration
|
|
for (end = tok->next(); Token::Match(end, "%name%|*|&|,"); end = end->next()) {}
|
|
|
|
// there are tokens which can't appear at the begin of a function declaration such as "return"
|
|
const bool isNotstartKeyword = start->next() && notstart.find(start->next()->str()) != notstart.end();
|
|
|
|
// now check if it is a function declaration
|
|
if (Token::Match(start, "[;{}] %type% %name%|*") && par && Token::simpleMatch(end, ") ;") && !isNotstartKeyword)
|
|
// function declaration => don't set varid
|
|
continue;
|
|
}
|
|
|
|
if (!scopeStack.top().isEnum || !(Token::Match(tok->previous(), "{|,") && Token::Match(tok->next(), ",|=|}"))) {
|
|
const std::map<std::string, int>::const_iterator it = variableMap.find(tok->str());
|
|
if (it != variableMap.end()) {
|
|
tok->varId(it->second);
|
|
setVarIdStructMembers(&tok, structMembers, variableMap.getVarId());
|
|
}
|
|
}
|
|
} else if (Token::Match(tok, "::|. %name%")) {
|
|
// Don't set varid after a :: or . token
|
|
tok = tok->next();
|
|
} else if (tok->str() == ":" && Token::Match(tok->tokAt(-2), "class %type%")) {
|
|
do {
|
|
tok = tok->next();
|
|
} while (tok && (tok->isName() || tok->str() == ","));
|
|
if (!tok)
|
|
break;
|
|
tok = tok->previous();
|
|
}
|
|
}
|
|
|
|
mVarId = *variableMap.getVarId();
|
|
}
|
|
|
|
namespace {
|
|
struct Member {
|
|
Member(const std::list<std::string> &s, const std::list<const Token *> &ns, Token *t) : usingnamespaces(ns), scope(s), tok(t) {}
|
|
std::list<const Token *> usingnamespaces;
|
|
std::list<std::string> scope;
|
|
Token *tok;
|
|
};
|
|
}
|
|
|
|
static std::string getScopeName(const std::list<ScopeInfo2> &scopeInfo)
|
|
{
|
|
std::string ret;
|
|
for (const ScopeInfo2 &si : scopeInfo)
|
|
ret += (ret.empty() ? "" : " :: ") + (si.name);
|
|
return ret;
|
|
}
|
|
|
|
static Token * matchMemberName(const std::list<std::string> &scope, const Token *nsToken, Token *memberToken, const std::list<ScopeInfo2> &scopeInfo)
|
|
{
|
|
std::list<ScopeInfo2>::const_iterator scopeIt = scopeInfo.begin();
|
|
|
|
// Current scope..
|
|
for (std::list<std::string>::const_iterator it = scope.begin(); it != scope.end(); ++it) {
|
|
if (scopeIt == scopeInfo.end() || scopeIt->name != *it)
|
|
return nullptr;
|
|
++scopeIt;
|
|
}
|
|
|
|
// using namespace..
|
|
if (nsToken) {
|
|
while (Token::Match(nsToken, "%name% ::")) {
|
|
if (scopeIt != scopeInfo.end() && nsToken->str() == scopeIt->name) {
|
|
nsToken = nsToken->tokAt(2);
|
|
++scopeIt;
|
|
} else {
|
|
return nullptr;
|
|
}
|
|
}
|
|
if (!Token::Match(nsToken, "%name% ;"))
|
|
return nullptr;
|
|
if (scopeIt == scopeInfo.end() || nsToken->str() != scopeIt->name)
|
|
return nullptr;
|
|
++scopeIt;
|
|
}
|
|
|
|
// Parse member tokens..
|
|
while (scopeIt != scopeInfo.end()) {
|
|
if (!Token::Match(memberToken, "%name% ::|<"))
|
|
return nullptr;
|
|
if (memberToken->str() != scopeIt->name)
|
|
return nullptr;
|
|
if (memberToken->next()->str() == "<") {
|
|
memberToken = memberToken->next()->findClosingBracket();
|
|
if (!Token::simpleMatch(memberToken, "> ::"))
|
|
return nullptr;
|
|
}
|
|
memberToken = memberToken->tokAt(2);
|
|
++scopeIt;
|
|
}
|
|
|
|
return Token::Match(memberToken, "~| %name%") ? memberToken : nullptr;
|
|
}
|
|
|
|
static Token * matchMemberName(const Member &member, const std::list<ScopeInfo2> &scopeInfo)
|
|
{
|
|
if (scopeInfo.empty())
|
|
return nullptr;
|
|
|
|
// Does this member match without "using namespace"..
|
|
Token *ret = matchMemberName(member.scope, nullptr, member.tok, scopeInfo);
|
|
if (ret)
|
|
return ret;
|
|
|
|
// Try to match member using the "using namespace ..." namespaces..
|
|
for (const Token *ns : member.usingnamespaces) {
|
|
ret = matchMemberName(member.scope, ns, member.tok, scopeInfo);
|
|
if (ret)
|
|
return ret;
|
|
}
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
static Token * matchMemberVarName(const Member &var, const std::list<ScopeInfo2> &scopeInfo)
|
|
{
|
|
Token *tok = matchMemberName(var, scopeInfo);
|
|
return Token::Match(tok, "%name% !!(") ? tok : nullptr;
|
|
}
|
|
|
|
static Token * matchMemberFunctionName(const Member &func, const std::list<ScopeInfo2> &scopeInfo)
|
|
{
|
|
Token *tok = matchMemberName(func, scopeInfo);
|
|
return Token::Match(tok, "~| %name% (") ? tok : nullptr;
|
|
}
|
|
|
|
void Tokenizer::setVarIdPass2()
|
|
{
|
|
std::map<int, std::map<std::string, int>> structMembers;
|
|
|
|
// Member functions and variables in this source
|
|
std::list<Member> allMemberFunctions;
|
|
std::list<Member> allMemberVars;
|
|
if (!isC()) {
|
|
std::map<const Token *, std::string> endOfScope;
|
|
std::list<std::string> scope;
|
|
std::list<const Token *> usingnamespaces;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!tok->previous() || Token::Match(tok->previous(), "[;{}]")) {
|
|
if (Token::Match(tok, "using namespace %name% ::|;")) {
|
|
Token *endtok = tok->tokAt(2);
|
|
while (Token::Match(endtok, "%name% ::"))
|
|
endtok = endtok->tokAt(2);
|
|
if (Token::Match(endtok, "%name% ;"))
|
|
usingnamespaces.push_back(tok->tokAt(2));
|
|
tok = endtok;
|
|
continue;
|
|
} else if (Token::Match(tok, "namespace %name% {")) {
|
|
scope.push_back(tok->strAt(1));
|
|
endOfScope[tok->linkAt(2)] = tok->strAt(1);
|
|
}
|
|
}
|
|
|
|
if (tok->str() == "}") {
|
|
const std::map<const Token *, std::string>::iterator it = endOfScope.find(tok);
|
|
if (it != endOfScope.end())
|
|
scope.remove(it->second);
|
|
}
|
|
|
|
Token* const tok1 = tok;
|
|
if (Token::Match(tok->previous(), "!!:: %name% :: ~| %name%"))
|
|
tok = tok->next();
|
|
else if (Token::Match(tok->previous(), "!!:: %name% <") && Token::Match(tok->next()->findClosingBracket(),"> :: ~| %name%"))
|
|
tok = tok->next()->findClosingBracket()->next();
|
|
else if (usingnamespaces.empty() || tok->varId() || !tok->isName() || tok->isStandardType() || tok->tokType() == Token::eKeyword || tok->tokType() == Token::eBoolean ||
|
|
Token::Match(tok->previous(), ".|namespace|class|struct|&|&&|*|> %name%") || Token::Match(tok->previous(), "%type%| %name% ( %type%|)") || Token::Match(tok, "public:|private:|protected:"))
|
|
continue;
|
|
|
|
while (Token::Match(tok, ":: ~| %name%")) {
|
|
tok = tok->next();
|
|
if (tok->str() == "~")
|
|
tok = tok->next();
|
|
else if (Token::Match(tok, "%name% <") && Token::Match(tok->next()->findClosingBracket(),"> :: ~| %name%"))
|
|
tok = tok->next()->findClosingBracket()->next();
|
|
else if (Token::Match(tok, "%name% ::"))
|
|
tok = tok->next();
|
|
else
|
|
break;
|
|
}
|
|
if (!tok->next())
|
|
syntaxError(tok);
|
|
if (Token::Match(tok, "%name% ("))
|
|
allMemberFunctions.emplace_back(scope, usingnamespaces, tok1);
|
|
else
|
|
allMemberVars.emplace_back(scope, usingnamespaces, tok1);
|
|
}
|
|
}
|
|
|
|
std::list<ScopeInfo2> scopeInfo;
|
|
|
|
// class members..
|
|
std::map<std::string, std::map<std::string, int>> varsByClass;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
while (tok->str() == "}" && !scopeInfo.empty() && tok == scopeInfo.back().bodyEnd)
|
|
scopeInfo.pop_back();
|
|
|
|
if (!Token::Match(tok, "namespace|class|struct %name% {|:|::|<"))
|
|
continue;
|
|
|
|
const std::string &scopeName(getScopeName(scopeInfo));
|
|
const std::string scopeName2(scopeName.empty() ? std::string() : (scopeName + " :: "));
|
|
|
|
std::list<const Token *> classnameTokens;
|
|
classnameTokens.push_back(tok->next());
|
|
const Token* tokStart = tok->tokAt(2);
|
|
while (Token::Match(tokStart, ":: %name%") || tokStart->str() == "<") {
|
|
if (tokStart->str() == "<") {
|
|
// skip the template part
|
|
const Token* closeTok = tokStart->findClosingBracket();
|
|
if (!closeTok)
|
|
syntaxError(tok);
|
|
tokStart = closeTok->next();
|
|
} else {
|
|
classnameTokens.push_back(tokStart->next());
|
|
tokStart = tokStart->tokAt(2);
|
|
}
|
|
}
|
|
|
|
std::string classname;
|
|
for (const Token *it : classnameTokens)
|
|
classname += (classname.empty() ? "" : " :: ") + it->str();
|
|
|
|
std::map<std::string, int> &thisClassVars = varsByClass[scopeName2 + classname];
|
|
while (Token::Match(tokStart, ":|::|,|%name%")) {
|
|
if (Token::Match(tokStart, "%name% <")) {
|
|
tokStart = tokStart->next()->findClosingBracket();
|
|
if (tokStart)
|
|
tokStart = tokStart->next();
|
|
continue;
|
|
}
|
|
if (Token::Match(tokStart, "%name% ,|{")) {
|
|
std::string baseClassName = tokStart->str();
|
|
std::string scopeName3(scopeName2);
|
|
while (!scopeName3.empty()) {
|
|
const std::string name = scopeName3 + baseClassName;
|
|
if (varsByClass.find(name) != varsByClass.end()) {
|
|
baseClassName = name;
|
|
break;
|
|
}
|
|
// Remove last scope name
|
|
if (scopeName3.size() <= 8)
|
|
break;
|
|
scopeName3.erase(scopeName3.size() - 4);
|
|
const std::string::size_type pos = scopeName3.rfind(" :: ");
|
|
if (pos == std::string::npos)
|
|
break;
|
|
scopeName3.erase(pos + 4);
|
|
}
|
|
const std::map<std::string, int>& baseClassVars = varsByClass[baseClassName];
|
|
thisClassVars.insert(baseClassVars.begin(), baseClassVars.end());
|
|
}
|
|
tokStart = tokStart->next();
|
|
}
|
|
if (!Token::simpleMatch(tokStart, "{"))
|
|
continue;
|
|
|
|
// What member variables are there in this class?
|
|
for (const Token *it : classnameTokens)
|
|
scopeInfo.emplace_back(it->str(), tokStart->link());
|
|
|
|
for (Token *tok2 = tokStart->next(); tok2 && tok2 != tokStart->link(); tok2 = tok2->next()) {
|
|
// skip parentheses..
|
|
if (tok2->link()) {
|
|
if (tok2->str() == "(") {
|
|
Token *funcstart = const_cast<Token*>(isFunctionHead(tok2, "{"));
|
|
if (funcstart) {
|
|
setVarIdClassFunction(scopeName2 + classname, funcstart, funcstart->link(), thisClassVars, structMembers, &mVarId);
|
|
tok2 = funcstart->link();
|
|
continue;
|
|
}
|
|
}
|
|
if (tok2->str() == "{") {
|
|
if (tok2->strAt(-1) == ")")
|
|
setVarIdClassFunction(scopeName2 + classname, tok2, tok2->link(), thisClassVars, structMembers, &mVarId);
|
|
tok2 = tok2->link();
|
|
} else if (Token::Match(tok2, "( %name%|)") && !Token::Match(tok2->link(), "(|[")) {
|
|
tok2 = tok2->link();
|
|
|
|
// Skip initialization list
|
|
while (Token::Match(tok2, ") [:,] %name% ("))
|
|
tok2 = tok2->linkAt(3);
|
|
}
|
|
}
|
|
|
|
// Found a member variable..
|
|
else if (tok2->varId() > 0)
|
|
thisClassVars[tok2->str()] = tok2->varId();
|
|
}
|
|
|
|
// Are there any member variables in this class?
|
|
if (thisClassVars.empty())
|
|
continue;
|
|
|
|
// Member variables
|
|
for (const Member &var : allMemberVars) {
|
|
Token *tok2 = matchMemberVarName(var, scopeInfo);
|
|
if (!tok2)
|
|
continue;
|
|
if (tok2->varId() == 0)
|
|
tok2->varId(thisClassVars[tok2->str()]);
|
|
}
|
|
|
|
if (isC() || tok->str() == "namespace")
|
|
continue;
|
|
|
|
// Set variable ids in member functions for this class..
|
|
for (const Member &func : allMemberFunctions) {
|
|
Token *tok2 = matchMemberFunctionName(func, scopeInfo);
|
|
if (!tok2)
|
|
continue;
|
|
|
|
if (tok2->str() == "~")
|
|
tok2 = tok2->linkAt(2);
|
|
else
|
|
tok2 = tok2->linkAt(1);
|
|
|
|
// If this is a function implementation.. add it to funclist
|
|
Token * start = const_cast<Token *>(isFunctionHead(tok2, "{"));
|
|
if (start) {
|
|
setVarIdClassFunction(classname, start, start->link(), thisClassVars, structMembers, &mVarId);
|
|
}
|
|
|
|
if (Token::Match(tok2, ") %name% ("))
|
|
tok2 = tok2->linkAt(2);
|
|
|
|
// constructor with initializer list
|
|
if (!Token::Match(tok2, ") : ::| %name%"))
|
|
continue;
|
|
|
|
Token *tok3 = tok2;
|
|
while (Token::Match(tok3, "[)}] [,:]")) {
|
|
tok3 = tok3->tokAt(2);
|
|
if (Token::Match(tok3, ":: %name%"))
|
|
tok3 = tok3->next();
|
|
while (Token::Match(tok3, "%name% :: %name%"))
|
|
tok3 = tok3->tokAt(2);
|
|
if (!Token::Match(tok3, "%name% (|{|<"))
|
|
break;
|
|
|
|
// set varid
|
|
const std::map<std::string, int>::const_iterator varpos = thisClassVars.find(tok3->str());
|
|
if (varpos != thisClassVars.end())
|
|
tok3->varId(varpos->second);
|
|
|
|
// goto end of var
|
|
if (tok3->strAt(1) == "<") {
|
|
tok3 = tok3->next()->findClosingBracket();
|
|
if (tok3 && tok3->next() && tok3->next()->link())
|
|
tok3 = tok3->next()->link();
|
|
} else
|
|
tok3 = tok3->linkAt(1);
|
|
}
|
|
if (Token::Match(tok3, ")|} {")) {
|
|
setVarIdClassFunction(classname, tok2, tok3->next()->link(), thisClassVars, structMembers, &mVarId);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
static void linkBrackets(const Tokenizer * const tokenizer, std::stack<const Token*>& type, std::stack<Token*>& links, Token * const token, const char open, const char close)
|
|
{
|
|
if (token->str()[0] == open) {
|
|
links.push(token);
|
|
type.push(token);
|
|
} else if (token->str()[0] == close) {
|
|
if (links.empty()) {
|
|
// Error, { and } don't match.
|
|
tokenizer->unmatchedToken(token);
|
|
}
|
|
if (type.top()->str()[0] != open) {
|
|
tokenizer->unmatchedToken(type.top());
|
|
}
|
|
type.pop();
|
|
|
|
Token::createMutualLinks(links.top(), token);
|
|
links.pop();
|
|
}
|
|
}
|
|
|
|
void Tokenizer::createLinks()
|
|
{
|
|
std::stack<const Token*> type;
|
|
std::stack<Token*> links1;
|
|
std::stack<Token*> links2;
|
|
std::stack<Token*> links3;
|
|
for (Token *token = list.front(); token; token = token->next()) {
|
|
if (token->link()) {
|
|
token->link(nullptr);
|
|
}
|
|
|
|
linkBrackets(this, type, links1, token, '{', '}');
|
|
|
|
linkBrackets(this, type, links2, token, '(', ')');
|
|
|
|
linkBrackets(this, type, links3, token, '[', ']');
|
|
}
|
|
|
|
if (!links1.empty()) {
|
|
// Error, { and } don't match.
|
|
unmatchedToken(links1.top());
|
|
}
|
|
|
|
if (!links2.empty()) {
|
|
// Error, ( and ) don't match.
|
|
unmatchedToken(links2.top());
|
|
}
|
|
|
|
if (!links3.empty()) {
|
|
// Error, [ and ] don't match.
|
|
unmatchedToken(links3.top());
|
|
}
|
|
}
|
|
|
|
void Tokenizer::createLinks2()
|
|
{
|
|
if (isC())
|
|
return;
|
|
|
|
bool isStruct = false;
|
|
|
|
std::stack<Token*> type;
|
|
std::stack<Token*> templateTokens;
|
|
for (Token *token = list.front(); token; token = token->next()) {
|
|
if (Token::Match(token, "%name%|> %name% [:<]"))
|
|
isStruct = true;
|
|
else if (Token::Match(token, "[;{}]"))
|
|
isStruct = false;
|
|
|
|
if (token->link()) {
|
|
if (Token::Match(token, "{|[|("))
|
|
type.push(token);
|
|
else if (!type.empty() && Token::Match(token, "}|]|)")) {
|
|
while (type.top()->str() == "<") {
|
|
if (!templateTokens.empty() && templateTokens.top()->next() == type.top())
|
|
templateTokens.pop();
|
|
type.pop();
|
|
}
|
|
type.pop();
|
|
}
|
|
} else if (templateTokens.empty() && !isStruct && Token::Match(token, "%oror%|&&|;")) {
|
|
if (Token::Match(token, "&& [,>]"))
|
|
continue;
|
|
// If there is some such code: A<B||C>..
|
|
// Then this is probably a template instantiation if either "B" or "C" has comparisons
|
|
if (token->tokType() == Token::eLogicalOp && !type.empty() && type.top()->str() == "<") {
|
|
const Token *prev = token->previous();
|
|
bool foundComparison = false;
|
|
while (Token::Match(prev, "%name%|%num%|%str%|%cop%|)|]") && prev != type.top()) {
|
|
if (prev->str() == ")" || prev->str() == "]")
|
|
prev = prev->link();
|
|
else if (prev->tokType() == Token::eLogicalOp)
|
|
break;
|
|
else if (prev->isComparisonOp())
|
|
foundComparison = true;
|
|
prev = prev->previous();
|
|
}
|
|
if (prev == type.top() && foundComparison)
|
|
continue;
|
|
const Token *next = token->next();
|
|
foundComparison = false;
|
|
while (Token::Match(next, "%name%|%num%|%str%|%cop%|(|[") && next->str() != ">") {
|
|
if (next->str() == "(" || next->str() == "[")
|
|
next = next->link();
|
|
else if (next->tokType() == Token::eLogicalOp)
|
|
break;
|
|
else if (next->isComparisonOp())
|
|
foundComparison = true;
|
|
next = next->next();
|
|
}
|
|
if (next && next->str() == ">" && foundComparison)
|
|
continue;
|
|
}
|
|
|
|
while (!type.empty() && type.top()->str() == "<") {
|
|
const Token* end = type.top()->findClosingBracket();
|
|
if (Token::Match(end, "> %comp%|;|.|=|{|::"))
|
|
break;
|
|
// Variable declaration
|
|
if (Token::Match(end, "> %var% ;") && (type.top()->tokAt(-2) == nullptr || Token::Match(type.top()->tokAt(-2), ";|}|{")))
|
|
break;
|
|
type.pop();
|
|
}
|
|
} else if (token->str() == "<" &&
|
|
((token->previous() && (token->previous()->isTemplate() ||
|
|
(token->previous()->isName() && !token->previous()->varId()))) ||
|
|
Token::Match(token->next(), ">|>>"))) {
|
|
type.push(token);
|
|
if (token->previous()->str() == "template")
|
|
templateTokens.push(token);
|
|
} else if (token->str() == ">" || token->str() == ">>") {
|
|
if (type.empty() || type.top()->str() != "<") // < and > don't match.
|
|
continue;
|
|
Token * const top1 = type.top();
|
|
type.pop();
|
|
Token * const top2 = type.empty() ? nullptr : type.top();
|
|
type.push(top1);
|
|
if (!top2 || top2->str() != "<") {
|
|
if (token->str() == ">>")
|
|
continue;
|
|
if (!Token::Match(token->next(), "%name%|%cop%|%assign%|::|,|(|)|{|}|;|[|:|.|=|...") &&
|
|
!Token::Match(token->next(), "&& %name% ="))
|
|
continue;
|
|
}
|
|
// if > is followed by [ .. "new a<b>[" is expected
|
|
// unless this is from varidiac expansion
|
|
if (token->strAt(1) == "[" && !Token::simpleMatch(token->tokAt(-1), "... >") &&
|
|
!Token::Match(token->tokAt(1), "[ ]")) {
|
|
Token *prev = type.top()->previous();
|
|
while (prev && Token::Match(prev->previous(), ":: %name%"))
|
|
prev = prev->tokAt(-2);
|
|
if (prev && prev->str() != "new")
|
|
prev = prev->previous();
|
|
if (!prev || prev->str() != "new")
|
|
continue;
|
|
}
|
|
|
|
if (token->str() == ">>" && top1 && top2) {
|
|
type.pop();
|
|
type.pop();
|
|
// Split the angle brackets
|
|
token->str(">");
|
|
Token::createMutualLinks(top1, token->insertTokenBefore(">"));
|
|
Token::createMutualLinks(top2, token);
|
|
if (templateTokens.size() == 2 && (top1 == templateTokens.top() || top2 == templateTokens.top())) {
|
|
templateTokens.pop();
|
|
templateTokens.pop();
|
|
}
|
|
} else {
|
|
type.pop();
|
|
if (Token::Match(token, "> %name%") && !token->next()->isKeyword() &&
|
|
Token::Match(top1->tokAt(-2), "%op% %name% <") &&
|
|
(templateTokens.empty() || top1 != templateTokens.top()))
|
|
continue;
|
|
Token::createMutualLinks(top1, token);
|
|
if (!templateTokens.empty() && top1 == templateTokens.top())
|
|
templateTokens.pop();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::sizeofAddParentheses()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "sizeof !!("))
|
|
continue;
|
|
if (tok->next()->isLiteral() || Token::Match(tok->next(), "%name%|*|~|!|&")) {
|
|
Token *endToken = tok->next();
|
|
while (Token::simpleMatch(endToken, "* *"))
|
|
endToken = endToken->next();
|
|
while (Token::Match(endToken->next(), "%name%|%num%|%str%|[|(|.|::|++|--|!|~") || (Token::Match(endToken, "%type% * %op%|?|:|const|;|,"))) {
|
|
if (Token::Match(endToken->next(), "(|["))
|
|
endToken = endToken->linkAt(1);
|
|
else
|
|
endToken = endToken->next();
|
|
}
|
|
|
|
// Add ( after sizeof and ) behind endToken
|
|
tok->insertToken("(");
|
|
endToken->insertToken(")");
|
|
Token::createMutualLinks(tok->next(), endToken->next());
|
|
}
|
|
}
|
|
}
|
|
|
|
bool Tokenizer::simplifySizeof()
|
|
{
|
|
// Locate variable declarations and calculate the size
|
|
std::map<int, int> sizeOfVar;
|
|
std::map<int, const Token *> declTokOfVar;
|
|
for (const Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->varId() != 0 && sizeOfVar.find(tok->varId()) == sizeOfVar.end()) {
|
|
const int varId = tok->varId();
|
|
if (Token::Match(tok->tokAt(-3), "[;{}(,] %type% * %name% [;,)]") ||
|
|
Token::Match(tok->tokAt(-4), "[;{}(,] const %type% * %name% [;),]") ||
|
|
Token::Match(tok->tokAt(-2), "[;{}(,] %type% %name% [;),]") ||
|
|
Token::Match(tok->tokAt(-3), "[;{}(,] const %type% %name% [;),]")) {
|
|
const int size = sizeOfType(tok->previous());
|
|
if (size == 0) {
|
|
continue;
|
|
}
|
|
|
|
sizeOfVar[varId] = size;
|
|
declTokOfVar[varId] = tok;
|
|
}
|
|
|
|
else if (Token::Match(tok->previous(), "%type% %name% [ %num% ] [[;=]") ||
|
|
Token::Match(tok->tokAt(-2), "%type% * %name% [ %num% ] [[;=]")) {
|
|
int size = sizeOfType(tok->previous());
|
|
if (size == 0)
|
|
continue;
|
|
|
|
const Token* tok2 = tok->next();
|
|
do {
|
|
const MathLib::bigint num = MathLib::toLongNumber(tok2->strAt(1));
|
|
if (num<0)
|
|
break;
|
|
size *= num;
|
|
tok2 = tok2->tokAt(3);
|
|
} while (Token::Match(tok2, "[ %num% ]"));
|
|
if (Token::Match(tok2, "[;=]")) {
|
|
sizeOfVar[varId] = size;
|
|
declTokOfVar[varId] = tok;
|
|
}
|
|
if (!tok2) {
|
|
syntaxError(tok);
|
|
}
|
|
tok = tok2;
|
|
}
|
|
|
|
else if (Token::Match(tok->previous(), "%type% %name% [ %num% ] [,)]") ||
|
|
Token::Match(tok->tokAt(-2), "%type% * %name% [ %num% ] [,)]")) {
|
|
Token tempTok;
|
|
tempTok.str("*");
|
|
sizeOfVar[varId] = sizeOfType(&tempTok);
|
|
declTokOfVar[varId] = tok;
|
|
}
|
|
}
|
|
}
|
|
|
|
bool ret = false;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() != "sizeof")
|
|
continue;
|
|
|
|
if (Token::simpleMatch(tok->next(), "...")) {
|
|
//tok->deleteNext(3);
|
|
tok->deleteNext();
|
|
}
|
|
|
|
// sizeof('x')
|
|
if (Token::Match(tok->next(), "( %char% )")) {
|
|
tok->deleteNext();
|
|
tok->deleteThis();
|
|
tok->deleteNext();
|
|
std::ostringstream sz;
|
|
sz << ((isC()) ? mSettings->sizeof_int : 1);
|
|
tok->str(sz.str());
|
|
ret = true;
|
|
continue;
|
|
}
|
|
|
|
// sizeof ("text")
|
|
if (Token::Match(tok->next(), "( %str% )")) {
|
|
tok->deleteNext();
|
|
tok->deleteThis();
|
|
tok->deleteNext();
|
|
std::ostringstream ostr;
|
|
ostr << (Token::getStrLength(tok) + 1);
|
|
tok->str(ostr.str());
|
|
ret = true;
|
|
continue;
|
|
}
|
|
|
|
// sizeof(type *) => sizeof(*)
|
|
if (Token::Match(tok->next(), "( %type% * )")) {
|
|
tok->next()->deleteNext();
|
|
}
|
|
|
|
if (Token::simpleMatch(tok->next(), "( * )")) {
|
|
tok->str(MathLib::toString(sizeOfType(tok->tokAt(2))));
|
|
tok->deleteNext(3);
|
|
ret = true;
|
|
}
|
|
|
|
// sizeof( a )
|
|
else if (Token::Match(tok->next(), "( %var% )")) {
|
|
const std::map<int, int>::const_iterator sizeOfVarPos = sizeOfVar.find(tok->tokAt(2)->varId());
|
|
if (sizeOfVarPos != sizeOfVar.end()) {
|
|
tok->deleteNext();
|
|
tok->deleteThis();
|
|
tok->deleteNext();
|
|
tok->str(MathLib::toString(sizeOfVarPos->second));
|
|
ret = true;
|
|
} else {
|
|
// don't try to replace size of variable if variable has
|
|
// similar name with type (#329)
|
|
}
|
|
}
|
|
|
|
else if (Token::Match(tok->next(), "( %type% )")) {
|
|
const int size = sizeOfType(tok->tokAt(2));
|
|
if (size > 0) {
|
|
tok->str(MathLib::toString(size));
|
|
tok->deleteNext(3);
|
|
ret = true;
|
|
}
|
|
}
|
|
|
|
else if (Token::simpleMatch(tok->next(), "( *") || Token::Match(tok->next(), "( %name% [")) {
|
|
int derefs = 0;
|
|
|
|
const Token* nametok = tok->tokAt(2);
|
|
if (nametok->str() == "*") {
|
|
do {
|
|
nametok = nametok->next();
|
|
derefs++;
|
|
} while (nametok && nametok->str() == "*");
|
|
|
|
if (!Token::Match(nametok, "%name% )"))
|
|
continue;
|
|
} else {
|
|
const Token* tok2 = nametok->next();
|
|
do {
|
|
tok2 = tok2->link()->next();
|
|
derefs++;
|
|
} while (tok2 && tok2->str() == "[");
|
|
|
|
if (!tok2 || tok2->str() != ")")
|
|
continue;
|
|
}
|
|
|
|
// Some default value
|
|
MathLib::biguint size = 0;
|
|
|
|
const int varid = nametok->varId();
|
|
if (derefs != 0 && varid != 0 && declTokOfVar.find(varid) != declTokOfVar.end()) {
|
|
// Try to locate variable declaration..
|
|
const Token *decltok = declTokOfVar[varid];
|
|
if (Token::Match(decltok->previous(), "%type%|* %name% [")) {
|
|
size = sizeOfType(decltok->previous());
|
|
} else if (Token::Match(decltok->tokAt(-2), "%type% * %name%")) {
|
|
size = sizeOfType(decltok->tokAt(-2));
|
|
}
|
|
// Multi-dimensional array..
|
|
if (Token::Match(decltok, "%name% [") && Token::simpleMatch(decltok->linkAt(1), "] [")) {
|
|
const Token *tok2 = decltok;
|
|
for (int i = 0; i < derefs; i++)
|
|
tok2 = tok2->linkAt(1); // Skip all dimensions that are dereferenced before the sizeof call
|
|
while (Token::Match(tok2, "] [ %num% ]")) {
|
|
size *= MathLib::toULongNumber(tok2->strAt(2));
|
|
tok2 = tok2->linkAt(1);
|
|
}
|
|
if (Token::simpleMatch(tok2, "] ["))
|
|
continue;
|
|
}
|
|
} else if (nametok->strAt(1) == "[" && nametok->isStandardType()) {
|
|
size = sizeOfType(nametok);
|
|
if (size == 0)
|
|
continue;
|
|
const Token *tok2 = nametok->next();
|
|
while (Token::Match(tok2, "[ %num% ]")) {
|
|
size *= MathLib::toULongNumber(tok2->strAt(1));
|
|
tok2 = tok2->link()->next();
|
|
}
|
|
if (!tok2 || tok2->str() != ")")
|
|
continue;
|
|
}
|
|
|
|
if (size > 0) {
|
|
tok->str(MathLib::toString(size));
|
|
Token::eraseTokens(tok, tok->next()->link()->next());
|
|
ret = true;
|
|
}
|
|
}
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
bool Tokenizer::simplifyTokenList1(const char FileName[])
|
|
{
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
// if MACRO
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "if|for|while|BOOST_FOREACH %name% (")) {
|
|
if (Token::simpleMatch(tok, "for each")) {
|
|
// 'for each ( )' -> 'asm ( )'
|
|
tok->str("asm");
|
|
tok->deleteNext();
|
|
} else if (tok->strAt(1) == "constexpr") {
|
|
tok->deleteNext();
|
|
tok->isConstexpr(true);
|
|
} else {
|
|
syntaxError(tok);
|
|
}
|
|
}
|
|
}
|
|
|
|
// Is there C++ code in C file?
|
|
validateC();
|
|
|
|
// remove MACRO in variable declaration: MACRO int x;
|
|
removeMacroInVarDecl();
|
|
|
|
// Combine strings and character literals, e.g. L"string", L'c', "string1" "string2"
|
|
combineStringAndCharLiterals();
|
|
|
|
// replace inline SQL with "asm()" (Oracle PRO*C). Ticket: #1959
|
|
simplifySQL();
|
|
|
|
createLinks();
|
|
|
|
removePragma();
|
|
|
|
// Simplify the C alternative tokens (and, or, etc.)
|
|
simplifyCAlternativeTokens();
|
|
|
|
simplifyFunctionTryCatch();
|
|
|
|
simplifyHeadersAndUnusedTemplates();
|
|
|
|
// Remove __asm..
|
|
simplifyAsm();
|
|
|
|
// foo < bar < >> => foo < bar < > >
|
|
if (isCPP() || mSettings->daca)
|
|
splitTemplateRightAngleBrackets(!isCPP());
|
|
|
|
// Remove extra "template" tokens that are not used by cppcheck
|
|
removeExtraTemplateKeywords();
|
|
|
|
removeAlignas();
|
|
|
|
simplifySpaceshipOperator();
|
|
|
|
// Bail out if code is garbage
|
|
if (mTimerResults) {
|
|
Timer t("Tokenizer::tokenize::findGarbageCode", mSettings->showtime, mTimerResults);
|
|
findGarbageCode();
|
|
} else {
|
|
findGarbageCode();
|
|
}
|
|
|
|
checkConfiguration();
|
|
|
|
// if (x) MACRO() ..
|
|
for (const Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::simpleMatch(tok, "if (")) {
|
|
tok = tok->next()->link();
|
|
if (Token::Match(tok, ") %name% (") &&
|
|
tok->next()->isUpperCaseName() &&
|
|
Token::Match(tok->linkAt(2), ") {|else")) {
|
|
syntaxError(tok->next());
|
|
}
|
|
}
|
|
}
|
|
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
// convert C++17 style nested namespaces to old style namespaces
|
|
simplifyNestedNamespace();
|
|
|
|
// convert c++20 coroutines
|
|
simplifyCoroutines();
|
|
|
|
// simplify namespace aliases
|
|
simplifyNamespaceAliases();
|
|
|
|
// Remove [[attribute]] and alignas(?)
|
|
simplifyCPPAttribute();
|
|
|
|
// remove __attribute__((?))
|
|
simplifyAttribute();
|
|
|
|
// simplify cppcheck attributes __cppcheck_?__(?)
|
|
simplifyCppcheckAttribute();
|
|
|
|
// Combine tokens..
|
|
combineOperators();
|
|
|
|
// replace 'sin(0)' to '0' and other similar math expressions
|
|
simplifyMathExpressions();
|
|
|
|
// combine "- %num%"
|
|
concatenateNegativeNumberAndAnyPositive();
|
|
|
|
// remove extern "C" and extern "C" {}
|
|
if (isCPP())
|
|
simplifyExternC();
|
|
|
|
// simplify weird but legal code: "[;{}] ( { code; } ) ;"->"[;{}] code;"
|
|
simplifyRoundCurlyParentheses();
|
|
|
|
// check for simple syntax errors..
|
|
for (const Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::simpleMatch(tok, "> struct {") &&
|
|
Token::simpleMatch(tok->linkAt(2), "} ;")) {
|
|
syntaxError(tok);
|
|
}
|
|
}
|
|
|
|
if (!simplifyAddBraces())
|
|
return false;
|
|
|
|
sizeofAddParentheses();
|
|
|
|
// Simplify: 0[foo] -> *(foo)
|
|
for (Token* tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::simpleMatch(tok, "0 [") && tok->linkAt(1)) {
|
|
tok->str("*");
|
|
tok->next()->str("(");
|
|
tok->linkAt(1)->str(")");
|
|
}
|
|
}
|
|
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
// Remove __declspec()
|
|
simplifyDeclspec();
|
|
validate();
|
|
|
|
// Remove "inline", "register", and "restrict"
|
|
simplifyKeyword();
|
|
|
|
// simplify simple calculations inside <..>
|
|
if (isCPP()) {
|
|
Token *lt = nullptr;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "[;{}]"))
|
|
lt = nullptr;
|
|
else if (Token::Match(tok, "%type% <"))
|
|
lt = tok->next();
|
|
else if (lt && Token::Match(tok, ">|>> %name%|::|(")) {
|
|
const Token * const end = tok;
|
|
for (tok = lt; tok != end; tok = tok->next()) {
|
|
if (tok->isNumber())
|
|
TemplateSimplifier::simplifyNumericCalculations(tok);
|
|
}
|
|
lt = tok->next();
|
|
}
|
|
}
|
|
}
|
|
|
|
// Convert K&R function declarations to modern C
|
|
simplifyVarDecl(true);
|
|
simplifyFunctionParameters();
|
|
|
|
// simplify case ranges (gcc extension)
|
|
simplifyCaseRange();
|
|
|
|
// simplify labels and 'case|default'-like syntaxes
|
|
simplifyLabelsCaseDefault();
|
|
|
|
// simplify '[;{}] * & ( %any% ) =' to '%any% ='
|
|
simplifyMulAndParens();
|
|
|
|
if (!isC() && !mSettings->library.markupFile(FileName)) {
|
|
findComplicatedSyntaxErrorsInTemplates();
|
|
}
|
|
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
// remove calling conventions __cdecl, __stdcall..
|
|
simplifyCallingConvention();
|
|
|
|
addSemicolonAfterUnknownMacro();
|
|
|
|
// remove some unhandled macros in global scope
|
|
removeMacrosInGlobalScope();
|
|
|
|
// remove undefined macro in class definition:
|
|
// class DLLEXPORT Fred { };
|
|
// class Fred FINAL : Base { };
|
|
removeMacroInClassDef();
|
|
|
|
// That call here fixes #7190
|
|
validate();
|
|
|
|
// remove unnecessary member qualification..
|
|
removeUnnecessaryQualification();
|
|
|
|
// convert Microsoft memory functions
|
|
simplifyMicrosoftMemoryFunctions();
|
|
|
|
// convert Microsoft string functions
|
|
simplifyMicrosoftStringFunctions();
|
|
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
// Remove Qt signals and slots
|
|
simplifyQtSignalsSlots();
|
|
|
|
// remove Borland stuff..
|
|
simplifyBorland();
|
|
|
|
// syntax error: enum with typedef in it
|
|
checkForEnumsWithTypedef();
|
|
|
|
// Add parentheses to ternary operator where necessary
|
|
prepareTernaryOpForAST();
|
|
|
|
// Change initialisation of variable to assignment
|
|
simplifyInitVar();
|
|
|
|
// Split up variable declarations.
|
|
simplifyVarDecl(false);
|
|
|
|
reportUnknownMacros();
|
|
|
|
// typedef..
|
|
if (mTimerResults) {
|
|
Timer t("Tokenizer::tokenize::simplifyTypedef", mSettings->showtime, mTimerResults);
|
|
simplifyTypedef();
|
|
} else {
|
|
simplifyTypedef();
|
|
}
|
|
|
|
// using A = B;
|
|
while (simplifyUsing())
|
|
;
|
|
|
|
// Add parentheses to ternary operator where necessary
|
|
// TODO: this is only necessary if one typedef simplification had a comma and was used within ?:
|
|
// If typedef handling is refactored and moved to symboldatabase someday we can remove this
|
|
prepareTernaryOpForAST();
|
|
|
|
for (Token* tok = list.front(); tok;) {
|
|
if (Token::Match(tok, "union|struct|class union|struct|class"))
|
|
tok->deleteNext();
|
|
else
|
|
tok = tok->next();
|
|
}
|
|
|
|
// class x y {
|
|
if (isCPP() && mSettings->severity.isEnabled(Severity::information)) {
|
|
for (const Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "class %type% %type% [:{]")) {
|
|
unhandled_macro_class_x_y(tok);
|
|
}
|
|
}
|
|
}
|
|
|
|
// catch bad typedef canonicalization
|
|
//
|
|
// to reproduce bad typedef, download upx-ucl from:
|
|
// http://packages.debian.org/sid/upx-ucl
|
|
// analyse the file src/stub/src/i386-linux.elf.interp-main.c
|
|
validate();
|
|
|
|
// The simplify enum have inner loops
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
// Put ^{} statements in asm()
|
|
simplifyAsm2();
|
|
|
|
// @..
|
|
simplifyAt();
|
|
|
|
// When the assembly code has been cleaned up, no @ is allowed
|
|
for (const Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() == "(") {
|
|
const Token *tok1 = tok;
|
|
tok = tok->link();
|
|
if (!tok)
|
|
syntaxError(tok1);
|
|
} else if (tok->str() == "@") {
|
|
syntaxError(tok);
|
|
}
|
|
}
|
|
|
|
// Order keywords "static" and "const"
|
|
simplifyStaticConst();
|
|
|
|
// convert platform dependent types to standard types
|
|
// 32 bits: size_t -> unsigned long
|
|
// 64 bits: size_t -> unsigned long long
|
|
list.simplifyPlatformTypes();
|
|
|
|
// collapse compound standard types into a single token
|
|
// unsigned long long int => long (with _isUnsigned=true,_isLong=true)
|
|
list.simplifyStdType();
|
|
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
// simplify bit fields..
|
|
simplifyBitfields();
|
|
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
// struct simplification "struct S {} s; => struct S { } ; S s ;
|
|
simplifyStructDecl();
|
|
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
// x = ({ 123; }); => { x = 123; }
|
|
simplifyAssignmentBlock();
|
|
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
simplifyVariableMultipleAssign();
|
|
|
|
// Collapse operator name tokens into single token
|
|
// operator = => operator=
|
|
simplifyOperatorName();
|
|
|
|
// Remove redundant parentheses
|
|
simplifyRedundantParentheses();
|
|
|
|
if (isCPP())
|
|
simplifyTypeIntrinsics();
|
|
|
|
if (!isC()) {
|
|
// Handle templates..
|
|
if (mTimerResults) {
|
|
Timer t("Tokenizer::tokenize::simplifyTemplates", mSettings->showtime, mTimerResults);
|
|
simplifyTemplates();
|
|
} else {
|
|
simplifyTemplates();
|
|
}
|
|
|
|
// The simplifyTemplates have inner loops
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
validate(); // #6847 - invalid code
|
|
}
|
|
|
|
// Simplify pointer to standard types (C only)
|
|
simplifyPointerToStandardType();
|
|
|
|
// simplify function pointers
|
|
simplifyFunctionPointers();
|
|
|
|
// Change initialisation of variable to assignment
|
|
simplifyInitVar();
|
|
|
|
// Split up variable declarations.
|
|
simplifyVarDecl(false);
|
|
|
|
elseif();
|
|
|
|
validate(); // #6772 "segmentation fault (invalid code) in Tokenizer::setVarId"
|
|
|
|
if (mTimerResults) {
|
|
Timer t("Tokenizer::tokenize::setVarId", mSettings->showtime, mTimerResults);
|
|
setVarId();
|
|
} else {
|
|
setVarId();
|
|
}
|
|
|
|
// Link < with >
|
|
createLinks2();
|
|
|
|
if (mTimerResults) {
|
|
Timer t("Tokenizer::tokenize::setVarId (2)", mSettings->showtime, mTimerResults);
|
|
setVarId();
|
|
}
|
|
else {
|
|
setVarId();
|
|
}
|
|
|
|
// Mark C++ casts
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "const_cast|dynamic_cast|reinterpret_cast|static_cast <") && Token::simpleMatch(tok->linkAt(1), "> (")) {
|
|
tok = tok->linkAt(1)->next();
|
|
tok->isCast(true);
|
|
}
|
|
}
|
|
|
|
// specify array size
|
|
arraySize();
|
|
|
|
// The simplify enum might have inner loops
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
// Add std:: in front of std classes, when using namespace std; was given
|
|
simplifyNamespaceStd();
|
|
|
|
// Change initialisation of variable to assignment
|
|
simplifyInitVar();
|
|
|
|
simplifyDoublePlusAndDoubleMinus();
|
|
|
|
simplifyArrayAccessSyntax();
|
|
|
|
Token::assignProgressValues(list.front());
|
|
|
|
removeRedundantSemicolons();
|
|
|
|
simplifyRedundantConsecutiveBraces();
|
|
|
|
simplifyEmptyNamespaces();
|
|
|
|
simplifyIfSwitchForInit();
|
|
|
|
simplifyOverloadedOperators();
|
|
|
|
validate();
|
|
|
|
list.front()->assignIndexes();
|
|
|
|
return true;
|
|
}
|
|
|
|
bool Tokenizer::simplifyTokenList2()
|
|
{
|
|
// clear the _functionList so it can't contain dead pointers
|
|
deleteSymbolDatabase();
|
|
|
|
// Clear AST,ValueFlow. These will be created again at the end of this function.
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
tok->clearAst();
|
|
tok->clearValueFlow();
|
|
}
|
|
|
|
// Convert e.g. atol("0") into 0
|
|
simplifyMathFunctions();
|
|
|
|
// f(x=g()) => x=g(); f(x)
|
|
simplifyAssignmentInFunctionCall();
|
|
|
|
// ";a+=b;" => ";a=a+b;"
|
|
simplifyCompoundAssignment();
|
|
|
|
simplifyCharAt();
|
|
|
|
// simplify references
|
|
simplifyReference();
|
|
|
|
simplifyStd();
|
|
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
simplifySizeof();
|
|
|
|
simplifyUndefinedSizeArray();
|
|
|
|
simplifyCasts();
|
|
|
|
// Simplify simple calculations before replace constants, this allows the replacement of constants that are calculated
|
|
// e.g. const static int value = sizeof(X)/sizeof(Y);
|
|
simplifyCalculations();
|
|
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
// Replace "*(ptr + num)" => "ptr[num]"
|
|
simplifyOffsetPointerDereference();
|
|
|
|
// Replace "&str[num]" => "(str + num)"
|
|
simplifyOffsetPointerReference();
|
|
|
|
removeRedundantAssignment();
|
|
|
|
simplifyRealloc();
|
|
|
|
// Change initialisation of variable to assignment
|
|
simplifyInitVar();
|
|
|
|
// Simplify variable declarations
|
|
simplifyVarDecl(false);
|
|
|
|
simplifyErrNoInWhile();
|
|
simplifyIfAndWhileAssign();
|
|
simplifyRedundantParentheses();
|
|
simplifyNestedStrcat();
|
|
simplifyFuncInWhile();
|
|
|
|
simplifyIfAndWhileAssign();
|
|
|
|
// replace strlen(str)
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "strlen ( %str% )")) {
|
|
tok->str(MathLib::toString(Token::getStrLength(tok->tokAt(2))));
|
|
tok->deleteNext(3);
|
|
}
|
|
}
|
|
|
|
bool modified = true;
|
|
while (modified) {
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
modified = false;
|
|
modified |= simplifyConditions();
|
|
modified |= simplifyFunctionReturn();
|
|
modified |= simplifyKnownVariables();
|
|
modified |= simplifyStrlen();
|
|
|
|
modified |= removeRedundantConditions();
|
|
modified |= simplifyRedundantParentheses();
|
|
modified |= simplifyConstTernaryOp();
|
|
modified |= simplifyCalculations();
|
|
validate();
|
|
}
|
|
|
|
// simplify redundant loops
|
|
simplifyWhile0();
|
|
removeRedundantFor();
|
|
|
|
// Remove redundant parentheses in return..
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
while (Token::simpleMatch(tok, "return (")) {
|
|
Token *tok2 = tok->next()->link();
|
|
if (Token::simpleMatch(tok2, ") ;")) {
|
|
tok->deleteNext();
|
|
tok2->deleteThis();
|
|
} else {
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
simplifyReturnStrncat();
|
|
|
|
removeRedundantAssignment();
|
|
|
|
simplifyComma();
|
|
|
|
removeRedundantSemicolons();
|
|
|
|
simplifyFlowControl();
|
|
|
|
simplifyRedundantConsecutiveBraces();
|
|
|
|
simplifyEmptyNamespaces();
|
|
|
|
simplifyMathFunctions();
|
|
|
|
validate();
|
|
|
|
Token::assignProgressValues(list.front());
|
|
|
|
list.front()->assignIndexes();
|
|
|
|
list.createAst();
|
|
// needed for #7208 (garbage code) and #7724 (ast max depth limit)
|
|
list.validateAst();
|
|
|
|
// Create symbol database and then remove const keywords
|
|
createSymbolDatabase();
|
|
mSymbolDatabase->setValueTypeInTokenList(true);
|
|
|
|
ValueFlow::setValues(&list, mSymbolDatabase, mErrorLogger, mSettings);
|
|
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
printDebugOutput(2);
|
|
|
|
return true;
|
|
}
|
|
//---------------------------------------------------------------------------
|
|
|
|
void Tokenizer::printDebugOutput(int simplification) const
|
|
{
|
|
const bool debug = (simplification != 1U && mSettings->debugSimplified) ||
|
|
(simplification != 2U && mSettings->debugnormal);
|
|
|
|
if (debug && list.front()) {
|
|
list.front()->printOut(nullptr, list.getFiles());
|
|
|
|
if (mSettings->xml)
|
|
std::cout << "<debug>" << std::endl;
|
|
|
|
if (mSymbolDatabase) {
|
|
if (mSettings->xml)
|
|
mSymbolDatabase->printXml(std::cout);
|
|
else if (mSettings->verbose) {
|
|
mSymbolDatabase->printOut("Symbol database");
|
|
}
|
|
}
|
|
|
|
if (mSettings->verbose)
|
|
list.front()->printAst(mSettings->verbose, mSettings->xml, list.getFiles(), std::cout);
|
|
|
|
list.front()->printValueFlow(mSettings->xml, std::cout);
|
|
|
|
if (mSettings->xml)
|
|
std::cout << "</debug>" << std::endl;
|
|
}
|
|
|
|
if (mSymbolDatabase && simplification == 2U && mSettings->debugwarnings) {
|
|
printUnknownTypes();
|
|
|
|
// the typeStartToken() should come before typeEndToken()
|
|
for (const Variable *var : mSymbolDatabase->variableList()) {
|
|
if (!var)
|
|
continue;
|
|
|
|
const Token * typetok = var->typeStartToken();
|
|
while (typetok && typetok != var->typeEndToken())
|
|
typetok = typetok->next();
|
|
|
|
if (typetok != var->typeEndToken()) {
|
|
reportError(var->typeStartToken(),
|
|
Severity::debug,
|
|
"debug",
|
|
"Variable::typeStartToken() of variable '" + var->name() + "' is not located before Variable::typeEndToken(). The location of the typeStartToken() is '" + var->typeStartToken()->str() + "' at line " + MathLib::toString(var->typeStartToken()->linenr()));
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::dump(std::ostream &out) const
|
|
{
|
|
// Create a xml data dump.
|
|
// The idea is not that this will be readable for humans. It's a
|
|
// data dump that 3rd party tools could load and get useful info from.
|
|
|
|
// tokens..
|
|
out << " <tokenlist>" << std::endl;
|
|
for (const Token *tok = list.front(); tok; tok = tok->next()) {
|
|
out << " <token id=\"" << tok << "\" file=\"" << ErrorLogger::toxml(list.file(tok)) << "\" linenr=\"" << tok->linenr() << "\" column=\"" << tok->column() << "\"";
|
|
out << " str=\"" << ErrorLogger::toxml(tok->str()) << '\"';
|
|
out << " scope=\"" << tok->scope() << '\"';
|
|
if (tok->isName()) {
|
|
out << " type=\"name\"";
|
|
if (tok->isUnsigned())
|
|
out << " isUnsigned=\"true\"";
|
|
else if (tok->isSigned())
|
|
out << " isSigned=\"true\"";
|
|
} else if (tok->isNumber()) {
|
|
out << " type=\"number\"";
|
|
if (MathLib::isInt(tok->str()))
|
|
out << " isInt=\"true\"";
|
|
if (MathLib::isFloat(tok->str()))
|
|
out << " isFloat=\"true\"";
|
|
} else if (tok->tokType() == Token::eString)
|
|
out << " type=\"string\" strlen=\"" << Token::getStrLength(tok) << '\"';
|
|
else if (tok->tokType() == Token::eChar)
|
|
out << " type=\"char\"";
|
|
else if (tok->isBoolean())
|
|
out << " type=\"boolean\"";
|
|
else if (tok->isOp()) {
|
|
out << " type=\"op\"";
|
|
if (tok->isArithmeticalOp())
|
|
out << " isArithmeticalOp=\"true\"";
|
|
else if (tok->isAssignmentOp())
|
|
out << " isAssignmentOp=\"true\"";
|
|
else if (tok->isComparisonOp())
|
|
out << " isComparisonOp=\"true\"";
|
|
else if (tok->tokType() == Token::eLogicalOp)
|
|
out << " isLogicalOp=\"true\"";
|
|
}
|
|
if (tok->isExpandedMacro())
|
|
out << " isExpandedMacro=\"true\"";
|
|
if (tok->isSplittedVarDeclComma())
|
|
out << " isSplittedVarDeclComma=\"true\"";
|
|
if (tok->isSplittedVarDeclEq())
|
|
out << " isSplittedVarDeclEq=\"true\"";
|
|
if (tok->isImplicitInt())
|
|
out << " isImplicitInt=\"true\"";
|
|
if (tok->link())
|
|
out << " link=\"" << tok->link() << '\"';
|
|
if (tok->varId() > 0)
|
|
out << " varId=\"" << MathLib::toString(tok->varId()) << '\"';
|
|
if (tok->variable())
|
|
out << " variable=\"" << tok->variable() << '\"';
|
|
if (tok->function())
|
|
out << " function=\"" << tok->function() << '\"';
|
|
if (!tok->values().empty())
|
|
out << " values=\"" << &tok->values() << '\"';
|
|
if (tok->type())
|
|
out << " type-scope=\"" << tok->type()->classScope << '\"';
|
|
if (tok->astParent())
|
|
out << " astParent=\"" << tok->astParent() << '\"';
|
|
if (tok->astOperand1())
|
|
out << " astOperand1=\"" << tok->astOperand1() << '\"';
|
|
if (tok->astOperand2())
|
|
out << " astOperand2=\"" << tok->astOperand2() << '\"';
|
|
if (!tok->originalName().empty())
|
|
out << " originalName=\"" << tok->originalName() << '\"';
|
|
if (tok->valueType()) {
|
|
const std::string vt = tok->valueType()->dump();
|
|
if (!vt.empty())
|
|
out << ' ' << vt;
|
|
}
|
|
if (!tok->varId() && tok->scope()->isExecutable() && Token::Match(tok, "%name% (")) {
|
|
if (mSettings->library.isnoreturn(tok))
|
|
out << " noreturn=\"true\"";
|
|
}
|
|
|
|
out << "/>" << std::endl;
|
|
}
|
|
out << " </tokenlist>" << std::endl;
|
|
|
|
mSymbolDatabase->printXml(out);
|
|
if (list.front())
|
|
list.front()->printValueFlow(true, out);
|
|
|
|
if (!mTypedefInfo.empty()) {
|
|
out << " <typedef-info>" << std::endl;
|
|
for (const TypedefInfo &typedefInfo: mTypedefInfo) {
|
|
out << " <info"
|
|
<< " name=\"" << typedefInfo.name << "\""
|
|
<< " file=\"" << typedefInfo.filename << "\""
|
|
<< " line=\"" << typedefInfo.lineNumber << "\""
|
|
<< " column=\"" << typedefInfo.column << "\""
|
|
<< " used=\"" << (typedefInfo.used?1:0) << "\""
|
|
<< "/>" << std::endl;
|
|
}
|
|
out << " </typedef-info>" << std::endl;
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyHeadersAndUnusedTemplates()
|
|
{
|
|
if (mSettings->checkHeaders && mSettings->checkUnusedTemplates)
|
|
// Full analysis. All information in the headers are kept.
|
|
return;
|
|
|
|
const bool checkHeaders = mSettings->checkHeaders;
|
|
const bool removeUnusedIncludedFunctions = !mSettings->checkHeaders;
|
|
const bool removeUnusedIncludedClasses = !mSettings->checkHeaders;
|
|
const bool removeUnusedIncludedTemplates = !mSettings->checkUnusedTemplates || !mSettings->checkHeaders;
|
|
const bool removeUnusedTemplates = !mSettings->checkUnusedTemplates;
|
|
|
|
// checkHeaders:
|
|
//
|
|
// If it is true then keep all code in the headers. It's possible
|
|
// to remove unused types/variables if false positives / false
|
|
// negatives can be avoided.
|
|
//
|
|
// If it is false, then we want to remove selected stuff from the
|
|
// headers but not *everything*. The intention here is to not damage
|
|
// the analysis of the source file. You should get all warnings in
|
|
// the source file. You should not get false positives.
|
|
|
|
// functions and types to keep
|
|
std::set<std::string> keep;
|
|
for (const Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (isCPP() && Token::simpleMatch(tok, "template <")) {
|
|
const Token *closingBracket = tok->next()->findClosingBracket();
|
|
if (Token::Match(closingBracket, "> class|struct %name% {"))
|
|
tok = closingBracket->linkAt(3);
|
|
}
|
|
|
|
if (!tok->isName() || tok->isKeyword())
|
|
continue;
|
|
|
|
if (!checkHeaders && tok->fileIndex() != 0)
|
|
continue;
|
|
|
|
if (Token::Match(tok, "%name% (") && !Token::simpleMatch(tok->linkAt(1), ") {")) {
|
|
keep.insert(tok->str());
|
|
continue;
|
|
}
|
|
|
|
if (Token::Match(tok, "%name% %name%|::|*|&|<")) {
|
|
keep.insert(tok->str());
|
|
}
|
|
}
|
|
|
|
const std::set<std::string> functionStart{"static", "const", "unsigned", "signed", "void", "bool", "char", "short", "int", "long", "float", "*"};
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
const bool isIncluded = (tok->fileIndex() != 0);
|
|
|
|
// Remove executable code
|
|
if (isIncluded && !mSettings->checkHeaders && tok->str() == "{") {
|
|
// TODO: We probably need to keep the executable code if this function is called from the source file.
|
|
const Token *prev = tok->previous();
|
|
while (prev && prev->isName())
|
|
prev = prev->previous();
|
|
if (Token::simpleMatch(prev, ")")) {
|
|
// Replace all tokens from { to } with a ";".
|
|
Token::eraseTokens(tok,tok->link()->next());
|
|
tok->str(";");
|
|
tok->link(nullptr);
|
|
}
|
|
}
|
|
|
|
if (!tok->previous() || Token::Match(tok->previous(), "[;{}]")) {
|
|
// Remove unused function declarations
|
|
if (isIncluded && removeUnusedIncludedFunctions) {
|
|
while (true) {
|
|
Token *start = tok;
|
|
while (start && functionStart.find(start->str()) != functionStart.end())
|
|
start = start->next();
|
|
if (Token::Match(start, "%name% (") && Token::Match(start->linkAt(1), ") const| ;") && keep.find(start->str()) == keep.end()) {
|
|
Token::eraseTokens(tok, start->linkAt(1)->tokAt(2));
|
|
tok->deleteThis();
|
|
} else
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (isIncluded && removeUnusedIncludedClasses) {
|
|
if (Token::Match(tok, "class|struct %name% [:{]") && keep.find(tok->strAt(1)) == keep.end()) {
|
|
// Remove this class/struct
|
|
const Token *endToken = tok->tokAt(2);
|
|
if (endToken->str() == ":") {
|
|
endToken = endToken->next();
|
|
while (Token::Match(endToken, "%name%|,"))
|
|
endToken = endToken->next();
|
|
}
|
|
if (endToken && endToken->str() == "{" && Token::simpleMatch(endToken->link(), "} ;")) {
|
|
Token::eraseTokens(tok, endToken->link()->next());
|
|
tok->deleteThis();
|
|
}
|
|
}
|
|
}
|
|
|
|
if (removeUnusedTemplates || (isIncluded && removeUnusedIncludedTemplates)) {
|
|
if (Token::Match(tok, "template < %name%")) {
|
|
const Token *closingBracket = tok->next()->findClosingBracket();
|
|
if (Token::Match(closingBracket, "> class|struct %name% [;:{]") && keep.find(closingBracket->strAt(2)) == keep.end()) {
|
|
const Token *endToken = closingBracket->tokAt(3);
|
|
if (endToken->str() == ":") {
|
|
endToken = endToken->next();
|
|
while (Token::Match(endToken, "%name%|,"))
|
|
endToken = endToken->next();
|
|
}
|
|
if (endToken && endToken->str() == "{")
|
|
endToken = endToken->link()->next();
|
|
if (endToken && endToken->str() == ";") {
|
|
Token::eraseTokens(tok, endToken);
|
|
tok->deleteThis();
|
|
}
|
|
} else if (Token::Match(closingBracket, "> %type% %name% (") && Token::simpleMatch(closingBracket->linkAt(3), ") {") && keep.find(closingBracket->strAt(2)) == keep.end()) {
|
|
const Token *endToken = closingBracket->linkAt(3)->linkAt(1)->next();
|
|
Token::eraseTokens(tok, endToken);
|
|
tok->deleteThis();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::removeExtraTemplateKeywords()
|
|
{
|
|
if (isCPP()) {
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "%name%|>|) .|:: template %name%")) {
|
|
tok->next()->deleteNext();
|
|
Token* templateName = tok->tokAt(2);
|
|
while (Token::Match(templateName, "%name%|::")) {
|
|
templateName->isTemplate(true);
|
|
templateName = templateName->next();
|
|
}
|
|
if (Token::Match(templateName->previous(), "operator %op%|(")) {
|
|
templateName->isTemplate(true);
|
|
if (templateName->str() == "(" && templateName->link())
|
|
templateName->link()->isTemplate(true);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
static std::string getExpression(const Token *tok)
|
|
{
|
|
std::string line;
|
|
for (const Token *prev = tok->previous(); prev && !Token::Match(prev, "[;{}]"); prev = prev->previous())
|
|
line = prev->str() + " " + line;
|
|
line += "!!!" + tok->str() + "!!!";
|
|
for (const Token *next = tok->next(); next && !Token::Match(next, "[;{}]"); next = next->next())
|
|
line = line + " " + next->str();
|
|
return line;
|
|
}
|
|
|
|
void Tokenizer::splitTemplateRightAngleBrackets(bool check)
|
|
{
|
|
std::set<std::string> vars;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "[;{}] %type% %type% [;,=]") && tok->next()->isStandardType())
|
|
vars.insert(tok->strAt(2));
|
|
|
|
// Ticket #6181: normalize C++11 template parameter list closing syntax
|
|
if (tok->previous() && tok->str() == "<" && TemplateSimplifier::templateParameters(tok) && vars.find(tok->previous()->str()) == vars.end()) {
|
|
Token *endTok = tok->findClosingBracket();
|
|
if (check) {
|
|
if (Token::Match(endTok, ">>|>>="))
|
|
reportError(tok, Severity::debug, "dacaWrongSplitTemplateRightAngleBrackets", "bad closing bracket for !!!<!!!: " + getExpression(tok), false);
|
|
continue;
|
|
}
|
|
if (endTok && endTok->str() == ">>") {
|
|
endTok->str(">");
|
|
endTok->insertToken(">");
|
|
} else if (endTok && endTok->str() == ">>=") {
|
|
endTok->str(">");
|
|
endTok->insertToken("=");
|
|
endTok->insertToken(">");
|
|
}
|
|
} else if (Token::Match(tok, "class|struct|union|=|:|public|protected|private %name% <") && vars.find(tok->next()->str()) == vars.end()) {
|
|
Token *endTok = tok->tokAt(2)->findClosingBracket();
|
|
if (check) {
|
|
if (Token::simpleMatch(endTok, ">>"))
|
|
reportError(tok, Severity::debug, "dacaWrongSplitTemplateRightAngleBrackets", "bad closing bracket for !!!<!!!: " + getExpression(tok), false);
|
|
continue;
|
|
}
|
|
if (Token::Match(endTok, ">> ;|{|%type%")) {
|
|
endTok->str(">");
|
|
endTok->insertToken(">");
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::removeMacrosInGlobalScope()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() == "(") {
|
|
tok = tok->link();
|
|
if (Token::Match(tok, ") %type% {") &&
|
|
!Token::Match(tok->next(), "const|namespace|class|struct|union|noexcept|override|final|volatile|mutable"))
|
|
tok->deleteNext();
|
|
}
|
|
|
|
if (Token::Match(tok, "%type%") && tok->isUpperCaseName() &&
|
|
(!tok->previous() || Token::Match(tok->previous(), "[;{}]") || (tok->previous()->isName() && endsWith(tok->previous()->str(), ':')))) {
|
|
const Token *tok2 = tok->next();
|
|
if (tok2 && tok2->str() == "(")
|
|
tok2 = tok2->link()->next();
|
|
|
|
// Several unknown macros...
|
|
while (Token::Match(tok2, "%type% (") && tok2->isUpperCaseName())
|
|
tok2 = tok2->linkAt(1)->next();
|
|
|
|
if (Token::Match(tok, "%name% (") && Token::Match(tok2, "%name% *|&|::|<| %name%") && !Token::Match(tok2, "namespace|class|struct|union|private:|protected:|public:"))
|
|
unknownMacroError(tok);
|
|
|
|
if (Token::Match(tok, "%type% (") && Token::Match(tok2, "%type% (") && !Token::Match(tok2, "noexcept|throw") && isFunctionHead(tok2->next(), ":;{"))
|
|
unknownMacroError(tok);
|
|
|
|
// remove unknown macros before namespace|class|struct|union
|
|
if (Token::Match(tok2, "namespace|class|struct|union")) {
|
|
// is there a "{" for?
|
|
const Token *tok3 = tok2;
|
|
while (tok3 && !Token::Match(tok3,"[;{}()]"))
|
|
tok3 = tok3->next();
|
|
if (tok3 && tok3->str() == "{") {
|
|
Token::eraseTokens(tok, tok2);
|
|
tok->deleteThis();
|
|
}
|
|
continue;
|
|
}
|
|
|
|
// replace unknown macros before foo(
|
|
/*
|
|
if (Token::Match(tok2, "%type% (") && isFunctionHead(tok2->next(), "{")) {
|
|
std::string typeName;
|
|
for (const Token* tok3 = tok; tok3 != tok2; tok3 = tok3->next())
|
|
typeName += tok3->str();
|
|
Token::eraseTokens(tok, tok2);
|
|
tok->str(typeName);
|
|
}
|
|
*/
|
|
// remove unknown macros before foo::foo(
|
|
if (Token::Match(tok2, "%type% :: %type%")) {
|
|
const Token *tok3 = tok2;
|
|
while (Token::Match(tok3, "%type% :: %type% ::"))
|
|
tok3 = tok3->tokAt(2);
|
|
if (Token::Match(tok3, "%type% :: %type% (") && tok3->str() == tok3->strAt(2)) {
|
|
Token::eraseTokens(tok, tok2);
|
|
tok->deleteThis();
|
|
}
|
|
continue;
|
|
}
|
|
}
|
|
|
|
// Skip executable scopes
|
|
if (tok->str() == "{") {
|
|
const Token *prev = tok->previous();
|
|
while (prev && prev->isName())
|
|
prev = prev->previous();
|
|
if (prev && prev->str() == ")")
|
|
tok = tok->link();
|
|
}
|
|
}
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
|
|
void Tokenizer::removePragma()
|
|
{
|
|
if (isC() && mSettings->standards.c == Standards::C89)
|
|
return;
|
|
if (isCPP() && mSettings->standards.cpp == Standards::CPP03)
|
|
return;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
while (Token::simpleMatch(tok, "_Pragma (")) {
|
|
Token::eraseTokens(tok, tok->linkAt(1)->next());
|
|
tok->deleteThis();
|
|
}
|
|
}
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
|
|
void Tokenizer::removeMacroInClassDef()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "class|struct %name% %name% {|:"))
|
|
continue;
|
|
|
|
const bool nextIsUppercase = tok->next()->isUpperCaseName();
|
|
const bool afterNextIsUppercase = tok->tokAt(2)->isUpperCaseName();
|
|
if (nextIsUppercase && !afterNextIsUppercase)
|
|
tok->deleteNext();
|
|
else if (!nextIsUppercase && afterNextIsUppercase)
|
|
tok->next()->deleteNext();
|
|
}
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
|
|
void Tokenizer::removeMacroInVarDecl()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "[;{}] %name% (") && tok->next()->isUpperCaseName()) {
|
|
// goto ')' parentheses
|
|
const Token *tok2 = tok;
|
|
int parlevel = 0;
|
|
while (tok2) {
|
|
if (tok2->str() == "(")
|
|
++parlevel;
|
|
else if (tok2->str() == ")") {
|
|
if (--parlevel <= 0)
|
|
break;
|
|
}
|
|
tok2 = tok2->next();
|
|
}
|
|
tok2 = tok2 ? tok2->next() : nullptr;
|
|
|
|
// check if this is a variable declaration..
|
|
const Token *tok3 = tok2;
|
|
while (tok3 && tok3->isUpperCaseName())
|
|
tok3 = tok3->next();
|
|
if (tok3 && (tok3->isStandardType() || Token::Match(tok3,"const|static|struct|union|class")))
|
|
Token::eraseTokens(tok,tok2);
|
|
}
|
|
}
|
|
}
|
|
//---------------------------------------------------------------------------
|
|
|
|
void Tokenizer::addSemicolonAfterUnknownMacro()
|
|
{
|
|
if (!isCPP())
|
|
return;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() != ")")
|
|
continue;
|
|
const Token *macro = tok->link() ? tok->link()->previous() : nullptr;
|
|
if (!macro || !macro->isName())
|
|
continue;
|
|
if (Token::simpleMatch(tok, ") try") && !Token::Match(macro, "if|for|while"))
|
|
tok->insertToken(";");
|
|
else if (Token::simpleMatch(tok, ") using"))
|
|
tok->insertToken(";");
|
|
}
|
|
}
|
|
//---------------------------------------------------------------------------
|
|
|
|
void Tokenizer::removeRedundantAssignment()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() == "{")
|
|
tok = tok->link();
|
|
|
|
const Token * const start = const_cast<Token *>(startOfExecutableScope(tok));
|
|
if (start) {
|
|
tok = start->previous();
|
|
// parse in this function..
|
|
std::set<int> localvars;
|
|
const Token * const end = tok->next()->link();
|
|
for (Token * tok2 = tok->next(); tok2 && tok2 != end; tok2 = tok2->next()) {
|
|
// skip local class or struct
|
|
if (Token::Match(tok2, "class|struct %type% {|:")) {
|
|
// skip to '{'
|
|
tok2 = tok2->tokAt(2);
|
|
while (tok2 && tok2->str() != "{")
|
|
tok2 = tok2->next();
|
|
|
|
if (tok2)
|
|
tok2 = tok2->link(); // skip local class or struct
|
|
else
|
|
return;
|
|
} else if (Token::Match(tok2, "[;{}] %type% * %name% ;") && tok2->next()->str() != "return") {
|
|
tok2 = tok2->tokAt(3);
|
|
localvars.insert(tok2->varId());
|
|
} else if (Token::Match(tok2, "[;{}] %type% %name% ;") && tok2->next()->isStandardType()) {
|
|
tok2 = tok2->tokAt(2);
|
|
localvars.insert(tok2->varId());
|
|
} else if (tok2->varId() &&
|
|
!Token::Match(tok2->previous(), "[;{}] %name% = %char%|%num%|%name% ;")) {
|
|
localvars.erase(tok2->varId());
|
|
}
|
|
}
|
|
localvars.erase(0);
|
|
if (!localvars.empty()) {
|
|
for (Token *tok2 = tok->next(); tok2 && tok2 != end;) {
|
|
if (Token::Match(tok2, "[;{}] %type% %name% ;") && localvars.find(tok2->tokAt(2)->varId()) != localvars.end()) {
|
|
tok2->deleteNext(3);
|
|
} else if ((Token::Match(tok2, "[;{}] %type% * %name% ;") &&
|
|
localvars.find(tok2->tokAt(3)->varId()) != localvars.end()) ||
|
|
(Token::Match(tok2, "[;{}] %name% = %any% ;") &&
|
|
localvars.find(tok2->next()->varId()) != localvars.end())) {
|
|
tok2->deleteNext(4);
|
|
} else
|
|
tok2 = tok2->next();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyRealloc()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "(|[") ||
|
|
(tok->str() == "{" && tok->previous() && tok->previous()->str() == "="))
|
|
tok = tok->link();
|
|
else if (Token::Match(tok, "[;{}] %name% = realloc (")) {
|
|
tok = tok->tokAt(3);
|
|
if (Token::simpleMatch(tok->next(), "( 0 ,")) {
|
|
//no "x = realloc(0,);"
|
|
if (!Token::simpleMatch(tok->next()->link(), ") ;") || tok->next()->link()->previous() == tok->tokAt(3))
|
|
continue;
|
|
|
|
// delete "0 ,"
|
|
tok->next()->deleteNext(2);
|
|
|
|
// Change function name "realloc" to "malloc"
|
|
tok->str("malloc");
|
|
tok = tok->next()->link();
|
|
} else {
|
|
Token *tok2 = tok->next()->link()->tokAt(-2);
|
|
//no "x = realloc(,0);"
|
|
if (!Token::simpleMatch(tok2, ", 0 ) ;") || tok2 == tok->tokAt(2))
|
|
continue;
|
|
|
|
//remove ", 0"
|
|
tok2 = tok2->previous();
|
|
tok2->deleteNext(2);
|
|
//change "realloc" to "free"
|
|
tok->str("free");
|
|
//insert "0" after "var ="
|
|
tok = tok->previous();
|
|
tok->insertToken("0");
|
|
//move "var = 0" between "free(...)" and ";"
|
|
tok2 = tok2->next();
|
|
Token::move(tok->previous(), tok->next(), tok2);
|
|
//add missing ";" after "free(...)"
|
|
tok2->insertToken(";");
|
|
//goto before last ";" and continue
|
|
tok = tok->next();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyEmptyNamespaces()
|
|
{
|
|
if (isC())
|
|
return;
|
|
|
|
bool goback = false;
|
|
for (Token *tok = list.front(); tok; tok = tok ? tok->next() : nullptr) {
|
|
if (goback) {
|
|
tok = tok->previous();
|
|
goback = false;
|
|
}
|
|
if (Token::Match(tok, "(|[|{")) {
|
|
tok = tok->link();
|
|
continue;
|
|
}
|
|
if (!Token::Match(tok, "namespace %name%| {"))
|
|
continue;
|
|
bool isAnonymousNS = tok->strAt(1) == "{";
|
|
if (tok->strAt(3 - isAnonymousNS) == "}") {
|
|
tok->deleteNext(3 - isAnonymousNS); // remove '%name%| { }'
|
|
if (!tok->previous()) {
|
|
// remove 'namespace' or replace it with ';' if isolated
|
|
tok->deleteThis();
|
|
goback = true;
|
|
} else { // '%any% namespace %any%'
|
|
tok = tok->previous(); // goto previous token
|
|
tok->deleteNext(); // remove next token: 'namespace'
|
|
if (tok->str() == "{") {
|
|
// Go back in case we were within a namespace that's empty now
|
|
tok = tok->tokAt(-2) ? tok->tokAt(-2) : tok->previous();
|
|
goback = true;
|
|
}
|
|
}
|
|
} else {
|
|
tok = tok->tokAt(2 - isAnonymousNS);
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyFlowControl()
|
|
{
|
|
for (Token *begin = list.front(); begin; begin = begin->next()) {
|
|
|
|
if (Token::Match(begin, "(|[") ||
|
|
(begin->str() == "{" && begin->previous() && begin->strAt(-1) == "="))
|
|
begin = begin->link();
|
|
|
|
//function scope
|
|
if (!Token::simpleMatch(begin, ") {") && !Token::Match(begin, ") %name% {"))
|
|
continue;
|
|
|
|
Token* end = begin->linkAt(1+(begin->next()->str() == "{" ? 0 : 1));
|
|
int indentLevel = 0;
|
|
bool stilldead = false;
|
|
|
|
for (Token *tok = begin; tok && tok != end; tok = tok->next()) {
|
|
if (Token::Match(tok, "(|[")) {
|
|
tok = tok->link();
|
|
continue;
|
|
}
|
|
|
|
if (tok->str() == "{") {
|
|
if (tok->previous() && tok->previous()->str() == "=") {
|
|
tok = tok->link();
|
|
continue;
|
|
}
|
|
++indentLevel;
|
|
} else if (tok->str() == "}") {
|
|
if (indentLevel == 0)
|
|
break;
|
|
--indentLevel;
|
|
if (stilldead) {
|
|
eraseDeadCode(tok, nullptr);
|
|
if (indentLevel == 1 || tok->next()->str() != "}" || !Token::Match(tok->next()->link()->previous(), ";|{|}|do {"))
|
|
stilldead = false;
|
|
continue;
|
|
}
|
|
}
|
|
|
|
if (indentLevel == 0)
|
|
continue;
|
|
|
|
if (Token::Match(tok,"continue|break ;")) {
|
|
tok = tok->next();
|
|
eraseDeadCode(tok, nullptr);
|
|
|
|
} else if (Token::Match(tok,"return|goto") ||
|
|
(Token::Match(tok->previous(), "[;{}] %name% (") &&
|
|
mSettings->library.isnoreturn(tok)) ||
|
|
(isCPP() && tok->str() == "throw")) {
|
|
if (tok->next()->str() == "}")
|
|
syntaxError(tok->next()); // invalid code like in #6731
|
|
//TODO: ensure that we exclude user-defined 'exit|abort|throw', except for 'noreturn'
|
|
//catch the first ';'
|
|
for (Token *tok2 = tok->next(); tok2; tok2 = tok2->next()) {
|
|
if (Token::Match(tok2, "(|[")) {
|
|
tok2 = tok2->link();
|
|
} else if (tok2->str() == ";") {
|
|
tok = tok2;
|
|
eraseDeadCode(tok, nullptr);
|
|
break;
|
|
} else if (Token::Match(tok2, "[{}]"))
|
|
break;
|
|
}
|
|
//if everything is removed, then remove also the code after an inferior scope
|
|
//only if the actual scope is not special
|
|
if (indentLevel > 1 && tok->next()->str() == "}" && Token::Match(tok->next()->link()->previous(), ";|{|}|do {"))
|
|
stilldead = true;
|
|
}
|
|
}
|
|
begin = end;
|
|
}
|
|
}
|
|
|
|
|
|
bool Tokenizer::removeRedundantConditions()
|
|
{
|
|
// Return value for function. Set to true if there are any simplifications
|
|
bool ret = false;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "if ( %bool% ) {"))
|
|
continue;
|
|
|
|
// Find matching else
|
|
Token *elseTag = tok->linkAt(4)->next();
|
|
|
|
const bool boolValue = (tok->strAt(2) == "true");
|
|
|
|
// Handle if with else
|
|
if (Token::simpleMatch(elseTag, "else {")) {
|
|
// Handle else
|
|
if (!boolValue) {
|
|
// Convert "if( false ) {aaa;} else {bbb;}" => "{bbb;}"
|
|
|
|
//remove '(false)'
|
|
tok->deleteNext(3);
|
|
//delete dead code inside scope
|
|
eraseDeadCode(tok, elseTag);
|
|
//remove 'else'
|
|
elseTag->deleteThis();
|
|
//remove 'if'
|
|
tok->deleteThis();
|
|
} else {
|
|
// Convert "if( true ) {aaa;} else {bbb;}" => "{aaa;}"
|
|
const Token *end = elseTag->next()->link()->next();
|
|
|
|
// Remove "else { bbb; }"
|
|
elseTag = elseTag->previous();
|
|
eraseDeadCode(elseTag, end);
|
|
|
|
// Remove "if( true )"
|
|
tok->deleteNext(3);
|
|
tok->deleteThis();
|
|
}
|
|
|
|
ret = true;
|
|
}
|
|
|
|
// Handle if without else
|
|
else {
|
|
if (!boolValue) {
|
|
//remove '(false)'
|
|
tok->deleteNext(3);
|
|
//delete dead code inside scope
|
|
eraseDeadCode(tok, elseTag);
|
|
//remove 'if'
|
|
tok->deleteThis();
|
|
} else {
|
|
// convert "if( true ) {aaa;}" => "{aaa;}"
|
|
tok->deleteNext(3);
|
|
tok->deleteThis();
|
|
}
|
|
|
|
ret = true;
|
|
}
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
void Tokenizer::removeRedundantFor()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "[;{}] for ( %name% = %num% ; %name% < %num% ; ++| %name% ++| ) {") ||
|
|
Token::Match(tok, "[;{}] for ( %type% %name% = %num% ; %name% < %num% ; ++| %name% ++| ) {")) {
|
|
// Same variable name..
|
|
const Token* varTok = tok->tokAt(3);
|
|
const bool type = varTok->next()->isName();
|
|
if (type)
|
|
varTok = varTok->next();
|
|
const std::string varname(varTok->str());
|
|
const int varid(varTok->varId());
|
|
if (varname != varTok->strAt(4))
|
|
continue;
|
|
const Token *vartok2 = tok->linkAt(2)->previous();
|
|
if (vartok2->str() == "++")
|
|
vartok2 = vartok2->previous();
|
|
else if (vartok2->strAt(-1) != "++")
|
|
continue;
|
|
if (varname != vartok2->str())
|
|
continue;
|
|
|
|
// Check that the difference of the numeric values is 1
|
|
const MathLib::bigint num1(MathLib::toLongNumber(varTok->strAt(2)));
|
|
const MathLib::bigint num2(MathLib::toLongNumber(varTok->strAt(6)));
|
|
if (num1 + 1 != num2)
|
|
continue;
|
|
|
|
// check how loop variable is used in loop..
|
|
bool read = false;
|
|
bool write = false;
|
|
const Token* end = tok->linkAt(2)->next()->link();
|
|
for (const Token *tok2 = tok->linkAt(2); tok2 != end; tok2 = tok2->next()) {
|
|
if (tok2->str() == varname) {
|
|
if (tok2->previous()->isArithmeticalOp() &&
|
|
tok2->next() &&
|
|
(tok2->next()->isArithmeticalOp() || tok2->next()->str() == ";")) {
|
|
read = true;
|
|
} else {
|
|
read = write = true;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
// Simplify loop if loop variable isn't written
|
|
if (!write) {
|
|
Token* bodyBegin = tok->linkAt(2)->next();
|
|
// remove "for ("
|
|
tok->deleteNext(2);
|
|
|
|
// If loop variable is read then keep assignment before
|
|
// loop body..
|
|
if (type) {
|
|
tok->insertToken("{");
|
|
Token::createMutualLinks(tok->next(), bodyBegin->link());
|
|
bodyBegin->deleteThis();
|
|
tok = tok->tokAt(6);
|
|
} else if (read) {
|
|
// goto ";"
|
|
tok = tok->tokAt(4);
|
|
} else {
|
|
// remove "x = 0 ;"
|
|
tok->deleteNext(4);
|
|
}
|
|
|
|
// remove "x < 1 ; x ++ )"
|
|
tok->deleteNext(7);
|
|
|
|
if (!type) {
|
|
// Add assignment after the loop body so the loop variable
|
|
// get the correct end value
|
|
Token *tok2 = tok->next()->link();
|
|
tok2->insertToken(";");
|
|
tok2->insertToken(MathLib::toString(num2));
|
|
tok2->insertToken("=");
|
|
tok2->insertToken(varname);
|
|
tok2->next()->varId(varid);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
void Tokenizer::removeRedundantSemicolons()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->link() && tok->str() == "(") {
|
|
tok = tok->link();
|
|
continue;
|
|
}
|
|
for (;;) {
|
|
if (Token::simpleMatch(tok, "; ;")) {
|
|
tok->deleteNext();
|
|
} else if (Token::simpleMatch(tok, "; { ; }")) {
|
|
tok->deleteNext(3);
|
|
} else {
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
bool Tokenizer::simplifyAddBraces()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
Token const * tokRet=simplifyAddBracesToCommand(tok);
|
|
if (!tokRet)
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
Token *Tokenizer::simplifyAddBracesToCommand(Token *tok)
|
|
{
|
|
Token * tokEnd=tok;
|
|
if (Token::Match(tok,"for|switch|BOOST_FOREACH")) {
|
|
tokEnd=simplifyAddBracesPair(tok,true);
|
|
} else if (tok->str()=="while") {
|
|
Token *tokPossibleDo=tok->previous();
|
|
if (Token::simpleMatch(tok->previous(), "{"))
|
|
tokPossibleDo = nullptr;
|
|
else if (Token::simpleMatch(tokPossibleDo,"}"))
|
|
tokPossibleDo = tokPossibleDo->link();
|
|
if (!tokPossibleDo || tokPossibleDo->strAt(-1) != "do")
|
|
tokEnd=simplifyAddBracesPair(tok,true);
|
|
} else if (tok->str()=="do") {
|
|
tokEnd=simplifyAddBracesPair(tok,false);
|
|
if (tokEnd!=tok) {
|
|
// walk on to next token, i.e. "while"
|
|
// such that simplifyAddBracesPair does not close other braces
|
|
// before the "while"
|
|
if (tokEnd) {
|
|
tokEnd=tokEnd->next();
|
|
if (!tokEnd || tokEnd->str()!="while") // no while
|
|
syntaxError(tok);
|
|
}
|
|
}
|
|
} else if (tok->str()=="if" && !Token::simpleMatch(tok->tokAt(-2), "operator \"\"")) {
|
|
tokEnd=simplifyAddBracesPair(tok,true);
|
|
if (!tokEnd)
|
|
return nullptr;
|
|
if (tokEnd->strAt(1) == "else") {
|
|
Token * tokEndNextNext= tokEnd->tokAt(2);
|
|
if (!tokEndNextNext || tokEndNextNext->str() == "}")
|
|
syntaxError(tokEndNextNext);
|
|
if (tokEndNextNext->str() == "if")
|
|
// do not change "else if ..." to "else { if ... }"
|
|
tokEnd=simplifyAddBracesToCommand(tokEndNextNext);
|
|
else
|
|
tokEnd=simplifyAddBracesPair(tokEnd->next(),false);
|
|
}
|
|
}
|
|
|
|
return tokEnd;
|
|
}
|
|
|
|
Token *Tokenizer::simplifyAddBracesPair(Token *tok, bool commandWithCondition)
|
|
{
|
|
Token * tokCondition=tok->next();
|
|
if (!tokCondition) // Missing condition
|
|
return tok;
|
|
|
|
Token *tokAfterCondition=tokCondition;
|
|
if (commandWithCondition) {
|
|
if (tokCondition->str()=="(")
|
|
tokAfterCondition=tokCondition->link();
|
|
else
|
|
syntaxError(tok); // Bad condition
|
|
|
|
if (!tokAfterCondition || tokAfterCondition->strAt(1) == "]")
|
|
syntaxError(tok); // Bad condition
|
|
|
|
tokAfterCondition=tokAfterCondition->next();
|
|
if (!tokAfterCondition || Token::Match(tokAfterCondition, ")|}|,")) {
|
|
// No tokens left where to add braces around
|
|
return tok;
|
|
}
|
|
}
|
|
// Skip labels
|
|
Token * tokStatement = tokAfterCondition;
|
|
while (true) {
|
|
if (Token::Match(tokStatement, "%name% :"))
|
|
tokStatement = tokStatement->tokAt(2);
|
|
else if (tokStatement->str() == "case") {
|
|
tokStatement = skipCaseLabel(tokStatement);
|
|
if (!tokStatement)
|
|
return tok;
|
|
if (tokStatement->str() != ":")
|
|
syntaxError(tokStatement);
|
|
tokStatement = tokStatement->next();
|
|
} else
|
|
break;
|
|
if (!tokStatement)
|
|
return tok;
|
|
}
|
|
Token * tokBracesEnd=nullptr;
|
|
if (tokStatement->str() == "{") {
|
|
// already surrounded by braces
|
|
if (tokStatement != tokAfterCondition) {
|
|
// Move the opening brace before labels
|
|
Token::move(tokStatement, tokStatement, tokAfterCondition->previous());
|
|
}
|
|
tokBracesEnd = tokStatement->link();
|
|
} else if (Token::simpleMatch(tokStatement, "try {") &&
|
|
Token::simpleMatch(tokStatement->linkAt(1), "} catch (")) {
|
|
tokAfterCondition->previous()->insertToken("{");
|
|
Token * tokOpenBrace = tokAfterCondition->previous();
|
|
Token * tokEnd = tokStatement->linkAt(1)->linkAt(2)->linkAt(1);
|
|
if (!tokEnd) {
|
|
syntaxError(tokStatement);
|
|
}
|
|
tokEnd->insertToken("}");
|
|
Token * tokCloseBrace = tokEnd->next();
|
|
|
|
Token::createMutualLinks(tokOpenBrace, tokCloseBrace);
|
|
tokBracesEnd = tokCloseBrace;
|
|
} else {
|
|
Token * tokEnd = simplifyAddBracesToCommand(tokStatement);
|
|
if (!tokEnd) // Ticket #4887
|
|
return tok;
|
|
if (tokEnd->str()!="}") {
|
|
// Token does not end with brace
|
|
// Look for ; to add own closing brace after it
|
|
while (tokEnd && !Token::Match(tokEnd, ";|)|}")) {
|
|
if (tokEnd->tokType()==Token::eBracket || tokEnd->str() == "(") {
|
|
tokEnd = tokEnd->link();
|
|
if (!tokEnd) {
|
|
// Inner bracket does not close
|
|
return tok;
|
|
}
|
|
}
|
|
tokEnd=tokEnd->next();
|
|
}
|
|
if (!tokEnd || tokEnd->str() != ";") {
|
|
// No trailing ;
|
|
return tok;
|
|
}
|
|
}
|
|
|
|
tokAfterCondition->previous()->insertToken("{");
|
|
Token * tokOpenBrace=tokAfterCondition->previous();
|
|
|
|
tokEnd->insertToken("}");
|
|
Token * tokCloseBrace=tokEnd->next();
|
|
|
|
Token::createMutualLinks(tokOpenBrace,tokCloseBrace);
|
|
tokBracesEnd=tokCloseBrace;
|
|
}
|
|
|
|
return tokBracesEnd;
|
|
}
|
|
|
|
void Tokenizer::simplifyCompoundAssignment()
|
|
{
|
|
// Simplify compound assignments:
|
|
// "a+=b" => "a = a + b"
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "[;{}] (| *| (| %name%"))
|
|
continue;
|
|
if (tok->next()->str() == "return")
|
|
continue;
|
|
// backup current token..
|
|
Token * const tok1 = tok;
|
|
|
|
if (tok->next()->str() == "*")
|
|
tok = tok->next();
|
|
|
|
if (tok->next() && tok->next()->str() == "(") {
|
|
tok = tok->next()->link()->next();
|
|
} else {
|
|
// variable..
|
|
tok = tok->tokAt(2);
|
|
while (Token::Match(tok, ". %name%") ||
|
|
Token::Match(tok, "[|(")) {
|
|
if (tok->str() == ".")
|
|
tok = tok->tokAt(2);
|
|
else {
|
|
// goto "]" or ")"
|
|
tok = tok->link();
|
|
|
|
// goto next token..
|
|
tok = tok ? tok->next() : nullptr;
|
|
}
|
|
}
|
|
}
|
|
if (!tok)
|
|
break;
|
|
|
|
// Is current token at a compound assignment: +=|-=|.. ?
|
|
const std::string &str = tok->str();
|
|
std::string op; // operator used in assignment
|
|
if (tok->isAssignmentOp() && str.size() == 2)
|
|
op = str.substr(0, 1);
|
|
else if (tok->isAssignmentOp() && str.size() == 3)
|
|
op = str.substr(0, 2);
|
|
else {
|
|
tok = tok1;
|
|
continue;
|
|
}
|
|
|
|
// Remove the whole statement if it says: "+=0;", "-=0;", "*=1;" or "/=1;"
|
|
if (Token::Match(tok, "+=|-= 0 ;") ||
|
|
Token::simpleMatch(tok, "|= 0 ;") ||
|
|
Token::Match(tok, "*=|/= 1 ;")) {
|
|
tok = tok1;
|
|
while (tok->next()->str() != ";")
|
|
tok->deleteNext();
|
|
} else {
|
|
// Enclose the rhs in parentheses..
|
|
if (!Token::Match(tok->tokAt(2), "[;)]")) {
|
|
// Only enclose rhs in parentheses if there is some operator
|
|
bool someOperator = false;
|
|
for (Token *tok2 = tok->next(); tok2; tok2 = tok2->next()) {
|
|
if (tok2->link() && Token::Match(tok2, "{|[|("))
|
|
tok2 = tok2->link();
|
|
|
|
if (Token::Match(tok2->next(), "[;)]")) {
|
|
if (someOperator) {
|
|
tok->insertToken("(");
|
|
tok2->insertToken(")");
|
|
Token::createMutualLinks(tok->next(), tok2->next());
|
|
}
|
|
break;
|
|
}
|
|
|
|
someOperator |= (tok2->isOp() || tok2->str() == "?");
|
|
}
|
|
}
|
|
|
|
// simplify the compound assignment..
|
|
tok->str("=");
|
|
tok->insertToken(op);
|
|
|
|
std::stack<Token *> tokend;
|
|
for (Token *tok2 = tok->previous(); tok2 && tok2 != tok1; tok2 = tok2->previous()) {
|
|
// Don't duplicate ++ and --. Put preincrement in lhs. Put
|
|
// postincrement in rhs.
|
|
if (tok2->tokType() == Token::eIncDecOp) {
|
|
// pre increment/decrement => don't copy
|
|
if (tok2->next()->isName()) {
|
|
continue;
|
|
}
|
|
|
|
// post increment/decrement => move from lhs to rhs
|
|
tok->insertToken(tok2->str());
|
|
tok2->deleteThis();
|
|
continue;
|
|
}
|
|
|
|
// Copy token from lhs to rhs
|
|
tok->insertToken(tok2->str());
|
|
tok->next()->varId(tok2->varId());
|
|
if (Token::Match(tok->next(), "]|)|}"))
|
|
tokend.push(tok->next());
|
|
else if (Token::Match(tok->next(), "(|[|{")) {
|
|
Token::createMutualLinks(tok->next(), tokend.top());
|
|
tokend.pop();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
bool Tokenizer::simplifyConditions()
|
|
{
|
|
bool ret = false;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "! %bool%|%num%")) {
|
|
tok->deleteThis();
|
|
if (Token::Match(tok, "0|false"))
|
|
tok->str("true");
|
|
else
|
|
tok->str("false");
|
|
|
|
ret = true;
|
|
}
|
|
|
|
if (Token::simpleMatch(tok, "&& true &&")) {
|
|
tok->deleteNext(2);
|
|
ret = true;
|
|
}
|
|
|
|
else if (Token::simpleMatch(tok, "|| false ||")) {
|
|
tok->deleteNext(2);
|
|
ret = true;
|
|
}
|
|
|
|
else if (Token::Match(tok, "(|&& true && true &&|)")) {
|
|
tok->deleteNext(2);
|
|
ret = true;
|
|
}
|
|
|
|
else if (Token::Match(tok, "%oror%|( false %oror% false %oror%|)")) {
|
|
tok->deleteNext(2);
|
|
ret = true;
|
|
}
|
|
|
|
else if (Token::simpleMatch(tok, "( true ||") ||
|
|
Token::simpleMatch(tok, "( false &&")) {
|
|
Token::eraseTokens(tok->next(), tok->link());
|
|
ret = true;
|
|
}
|
|
|
|
else if (Token::simpleMatch(tok, "|| true )") ||
|
|
Token::simpleMatch(tok, "&& false )")) {
|
|
tok = tok->next();
|
|
Token::eraseTokens(tok->next()->link(), tok);
|
|
ret = true;
|
|
}
|
|
|
|
else if (Token::simpleMatch(tok, "&& false &&") ||
|
|
Token::simpleMatch(tok, "|| true ||")) {
|
|
//goto '('
|
|
Token *tok2 = tok;
|
|
while (tok2 && tok2->previous()) {
|
|
if (tok2->previous()->str() == ")")
|
|
tok2 = tok2->previous()->link();
|
|
else {
|
|
tok2 = tok2->previous();
|
|
if (tok2->str() == "(")
|
|
break;
|
|
}
|
|
}
|
|
if (!tok2)
|
|
continue;
|
|
//move tok to 'true|false' position
|
|
tok = tok->next();
|
|
//remove everything before 'true|false'
|
|
Token::eraseTokens(tok2, tok);
|
|
//remove everything after 'true|false'
|
|
Token::eraseTokens(tok, tok2->link());
|
|
ret = true;
|
|
}
|
|
|
|
// Change numeric constant in condition to "true" or "false"
|
|
if (Token::Match(tok, "if|while ( %num% )|%oror%|&&")) {
|
|
tok->tokAt(2)->str((tok->strAt(2) != "0") ? "true" : "false");
|
|
ret = true;
|
|
}
|
|
if (Token::Match(tok, "&&|%oror% %num% )|%oror%|&&")) {
|
|
tok->next()->str((tok->next()->str() != "0") ? "true" : "false");
|
|
ret = true;
|
|
}
|
|
|
|
// Reduce "(%num% == %num%)" => "(true)"/"(false)"
|
|
if (Token::Match(tok, "&&|%oror%|(") &&
|
|
(Token::Match(tok->next(), "%num% %any% %num%") ||
|
|
Token::Match(tok->next(), "%bool% %any% %bool%")) &&
|
|
Token::Match(tok->tokAt(4), "&&|%oror%|)|?")) {
|
|
std::string cmp = tok->strAt(2);
|
|
bool result = false;
|
|
if (tok->next()->isNumber()) {
|
|
// Compare numbers
|
|
|
|
if (cmp == "==" || cmp == "!=") {
|
|
const std::string& op1(tok->next()->str());
|
|
const std::string& op2(tok->strAt(3));
|
|
|
|
bool eq = false;
|
|
if (MathLib::isInt(op1) && MathLib::isInt(op2))
|
|
eq = (MathLib::toLongNumber(op1) == MathLib::toLongNumber(op2));
|
|
else {
|
|
eq = (op1 == op2);
|
|
|
|
// It is inconclusive whether two unequal float representations are numerically equal
|
|
if (!eq && MathLib::isFloat(op1))
|
|
cmp.clear();
|
|
}
|
|
|
|
if (cmp == "==")
|
|
result = eq;
|
|
else
|
|
result = !eq;
|
|
} else {
|
|
const double op1 = MathLib::toDoubleNumber(tok->next()->str());
|
|
const double op2 = MathLib::toDoubleNumber(tok->strAt(3));
|
|
if (cmp == ">=")
|
|
result = (op1 >= op2);
|
|
else if (cmp == ">")
|
|
result = (op1 > op2);
|
|
else if (cmp == "<=")
|
|
result = (op1 <= op2);
|
|
else if (cmp == "<")
|
|
result = (op1 < op2);
|
|
else
|
|
cmp.clear();
|
|
}
|
|
} else {
|
|
// Compare boolean
|
|
const bool op1 = (tok->next()->str() == std::string("true"));
|
|
const bool op2 = (tok->strAt(3) == std::string("true"));
|
|
|
|
if (cmp == "==")
|
|
result = (op1 == op2);
|
|
else if (cmp == "!=")
|
|
result = (op1 != op2);
|
|
else if (cmp == ">=")
|
|
result = (op1 >= op2);
|
|
else if (cmp == ">")
|
|
result = (op1 > op2);
|
|
else if (cmp == "<=")
|
|
result = (op1 <= op2);
|
|
else if (cmp == "<")
|
|
result = (op1 < op2);
|
|
else
|
|
cmp.clear();
|
|
}
|
|
|
|
if (!cmp.empty()) {
|
|
tok = tok->next();
|
|
tok->deleteNext(2);
|
|
|
|
tok->str(result ? "true" : "false");
|
|
ret = true;
|
|
}
|
|
}
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
bool Tokenizer::simplifyConstTernaryOp()
|
|
{
|
|
bool ret = false;
|
|
const Token *templateParameterEnd = nullptr; // The end of the current template parameter list, if any
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() == "<" && TemplateSimplifier::templateParameters(tok))
|
|
templateParameterEnd = tok->findClosingBracket();
|
|
if (tok == templateParameterEnd)
|
|
templateParameterEnd = nullptr; // End of the current template parameter list
|
|
if (tok->str() != "?")
|
|
continue;
|
|
|
|
if (!Token::Match(tok->tokAt(-2), "<|=|,|(|[|{|}|;|case|return %bool%|%num%") &&
|
|
!Token::Match(tok->tokAt(-4), "<|=|,|(|[|{|}|;|case|return ( %bool%|%num% )"))
|
|
continue;
|
|
|
|
const int offset = (tok->previous()->str() == ")") ? 2 : 1;
|
|
|
|
if (tok->strAt(-2*offset) == "<") {
|
|
if (isC() || !TemplateSimplifier::templateParameters(tok->tokAt(-2*offset)))
|
|
continue; // '<' is less than; the condition is not a constant
|
|
}
|
|
|
|
// Find the token ":" then go to the next token
|
|
Token *colon = skipTernaryOp(tok);
|
|
if (!colon || colon->previous()->str() != ":" || !colon->next())
|
|
continue;
|
|
|
|
//handle the GNU extension: "x ? : y" <-> "x ? x : y"
|
|
if (colon->previous() == tok->next())
|
|
tok->insertToken(tok->strAt(-offset));
|
|
|
|
// go back before the condition, if possible
|
|
tok = tok->tokAt(-2);
|
|
if (offset == 2) {
|
|
// go further back before the "("
|
|
tok = tok->tokAt(-2);
|
|
//simplify the parentheses
|
|
tok->deleteNext();
|
|
tok->next()->deleteNext();
|
|
}
|
|
|
|
if (Token::Match(tok->next(), "false|0")) {
|
|
// Use code after colon, remove code before it.
|
|
Token::eraseTokens(tok, colon);
|
|
|
|
tok = tok->next();
|
|
ret = true;
|
|
}
|
|
|
|
// The condition is true. Delete the operator after the ":"..
|
|
else {
|
|
// delete the condition token and the "?"
|
|
tok->deleteNext(2);
|
|
|
|
int ternaryOplevel = 0;
|
|
for (const Token *endTok = colon; endTok; endTok = endTok->next()) {
|
|
if (Token::Match(endTok, "(|[|{"))
|
|
endTok = endTok->link();
|
|
else if (endTok->str() == "<" && (endTok->strAt(1) == ">" || TemplateSimplifier::templateParameters(endTok)))
|
|
endTok = endTok->findClosingBracket();
|
|
else if (endTok->str() == "?")
|
|
++ternaryOplevel;
|
|
else if (Token::Match(endTok, ")|}|]|;|,|:|>")) {
|
|
if (endTok->str() == ":" && ternaryOplevel)
|
|
--ternaryOplevel;
|
|
else if (endTok->str() == ">" && !templateParameterEnd)
|
|
;
|
|
else {
|
|
Token::eraseTokens(colon->tokAt(-2), endTok);
|
|
ret = true;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
void Tokenizer::simplifyUndefinedSizeArray()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "%type%")) {
|
|
Token *tok2 = tok->next();
|
|
while (tok2 && tok2->str() == "*")
|
|
tok2 = tok2->next();
|
|
if (!Token::Match(tok2, "%name% [ ] ;|["))
|
|
continue;
|
|
|
|
tok = tok2->previous();
|
|
Token *end = tok2->next();
|
|
int count = 0;
|
|
do {
|
|
end = end->tokAt(2);
|
|
++count;
|
|
} while (Token::Match(end, "[ ] [;=[]"));
|
|
if (Token::Match(end, "[;=]")) {
|
|
do {
|
|
tok2->deleteNext(2);
|
|
tok->insertToken("*");
|
|
} while (--count);
|
|
tok = end;
|
|
} else
|
|
tok = tok->tokAt(3);
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyCasts()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
// Don't remove cast in such cases:
|
|
// *((char *)a + 1) = 0;
|
|
// Remove cast when casting a function pointer:
|
|
// (*(void (*)(char *))fp)(x);
|
|
if (!tok->isName() &&
|
|
Token::simpleMatch(tok->next(), "* (") &&
|
|
!Token::Match(tok->linkAt(2), ") %name%|&")) {
|
|
tok = tok->linkAt(2);
|
|
continue;
|
|
}
|
|
// #3935 : don't remove cast in such cases:
|
|
// ((char *)a)[1] = 0;
|
|
if (tok->str() == "(" && Token::simpleMatch(tok->link(), ") [")) {
|
|
tok = tok->link();
|
|
continue;
|
|
}
|
|
// #4164 : ((unsigned char)1) => (1)
|
|
if (Token::Match(tok->next(), "( %type% ) %num%") && tok->next()->link()->previous()->isStandardType()) {
|
|
const MathLib::bigint value = MathLib::toLongNumber(tok->next()->link()->next()->str());
|
|
int bits = mSettings->char_bit * mTypeSize[tok->next()->link()->previous()->str()];
|
|
if (!tok->tokAt(2)->isUnsigned() && bits > 0)
|
|
bits--;
|
|
if (bits < 31 && value >= 0 && value < (1LL << bits)) {
|
|
tok->linkAt(1)->next()->isCast(true);
|
|
Token::eraseTokens(tok, tok->next()->link()->next());
|
|
}
|
|
continue;
|
|
}
|
|
|
|
while ((Token::Match(tok->next(), "( %type% *| *| *|&| ) *|&| %name%") && (tok->str() != ")" || tok->tokAt(2)->isStandardType())) ||
|
|
Token::Match(tok->next(), "( const| %type% * *| *|&| ) *|&| %name%") ||
|
|
Token::Match(tok->next(), "( const| %type% %type% *| *| *|&| ) *|&| %name%") ||
|
|
(!tok->isName() && (Token::Match(tok->next(), "( %type% * *| *|&| ) (") ||
|
|
Token::Match(tok->next(), "( const| %type% %type% * *| *|&| ) (")))) {
|
|
if (tok->isName() && tok->str() != "return")
|
|
break;
|
|
|
|
if (isCPP() && tok->strAt(-1) == "operator")
|
|
break;
|
|
|
|
// Remove cast..
|
|
Token::eraseTokens(tok, tok->next()->link()->next());
|
|
|
|
// Set isCasted flag.
|
|
Token *tok2 = tok->next();
|
|
if (!Token::Match(tok2, "%name% [|."))
|
|
tok2->isCast(true);
|
|
else {
|
|
// TODO: handle more complex expressions
|
|
tok2->next()->isCast(true);
|
|
}
|
|
|
|
// Remove '* &'
|
|
if (Token::simpleMatch(tok, "* &")) {
|
|
tok->deleteNext();
|
|
tok->deleteThis();
|
|
}
|
|
|
|
if (tok->str() == ")" && tok->link()->previous()) {
|
|
// If there was another cast before this, go back
|
|
// there to check it also. e.g. "(int)(char)x"
|
|
tok = tok->link()->previous();
|
|
}
|
|
}
|
|
|
|
// Replace pointer casts of 0.. "(char *)0" => "0"
|
|
while (Token::Match(tok->next(), "( %type% %type%| * *| ) 0")) {
|
|
tok->linkAt(1)->next()->isCast(true);
|
|
Token::eraseTokens(tok, tok->next()->link()->next());
|
|
if (tok->str() == ")" && tok->link()->previous()) {
|
|
// If there was another cast before this, go back
|
|
// there to check it also. e.g. "(char*)(char*)0"
|
|
tok = tok->link()->previous();
|
|
}
|
|
}
|
|
|
|
if (Token::Match(tok->next(), "dynamic_cast|reinterpret_cast|const_cast|static_cast <")) {
|
|
Token *tok2 = tok->linkAt(2);
|
|
if (!Token::simpleMatch(tok2, "> ("))
|
|
break;
|
|
|
|
tok2->tokAt(2)->isCast(true);
|
|
Token::eraseTokens(tok, tok2->next());
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
void Tokenizer::simplifyFunctionParameters()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->link() && Token::Match(tok, "{|[|(")) {
|
|
tok = tok->link();
|
|
}
|
|
|
|
// Find the function e.g. foo( x ) or foo( x, y )
|
|
else if (Token::Match(tok, "%name% ( %name% [,)]") &&
|
|
!(tok->strAt(-1) == ":" || tok->strAt(-1) == "," || tok->strAt(-1) == "::")) {
|
|
// We have found old style function, now we need to change it
|
|
|
|
// First step: Get list of argument names in parentheses
|
|
std::map<std::string, Token *> argumentNames;
|
|
bool bailOut = false;
|
|
Token * tokparam = nullptr;
|
|
|
|
//take count of the function name..
|
|
const std::string& funcName(tok->str());
|
|
|
|
//floating token used to check for parameters
|
|
Token *tok1 = tok;
|
|
|
|
while (nullptr != (tok1 = tok1->tokAt(2))) {
|
|
if (!Token::Match(tok1, "%name% [,)]")) {
|
|
bailOut = true;
|
|
break;
|
|
}
|
|
|
|
//same parameters: take note of the parameter
|
|
if (argumentNames.find(tok1->str()) != argumentNames.end())
|
|
tokparam = tok1;
|
|
else if (tok1->str() != funcName)
|
|
argumentNames[tok1->str()] = tok1;
|
|
else {
|
|
if (tok1->next()->str() == ")") {
|
|
if (tok1->previous()->str() == ",") {
|
|
tok1 = tok1->tokAt(-2);
|
|
tok1->deleteNext(2);
|
|
} else {
|
|
tok1 = tok1->previous();
|
|
tok1->deleteNext();
|
|
bailOut = true;
|
|
break;
|
|
}
|
|
} else {
|
|
tok1 = tok1->tokAt(-2);
|
|
tok1->next()->deleteNext(2);
|
|
}
|
|
}
|
|
|
|
if (tok1->next()->str() == ")") {
|
|
tok1 = tok1->tokAt(2);
|
|
//expect at least a type name after round brace..
|
|
if (!tok1 || !tok1->isName())
|
|
bailOut = true;
|
|
break;
|
|
}
|
|
}
|
|
|
|
//goto '('
|
|
tok = tok->next();
|
|
|
|
if (bailOut) {
|
|
tok = tok->link();
|
|
continue;
|
|
}
|
|
|
|
tok1 = tok->link()->next();
|
|
|
|
// there should be the sequence '; {' after the round parentheses
|
|
for (const Token* tok2 = tok1; tok2; tok2 = tok2->next()) {
|
|
if (Token::simpleMatch(tok2, "; {"))
|
|
break;
|
|
else if (tok2->str() == "{") {
|
|
bailOut = true;
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (bailOut) {
|
|
tok = tok->link();
|
|
continue;
|
|
}
|
|
|
|
// Last step: check out if the declarations between ')' and '{' match the parameters list
|
|
std::map<std::string, Token *> argumentNames2;
|
|
|
|
while (tok1 && tok1->str() != "{") {
|
|
if (Token::Match(tok1, "(|)")) {
|
|
bailOut = true;
|
|
break;
|
|
}
|
|
if (tok1->str() == ";") {
|
|
if (tokparam) {
|
|
syntaxError(tokparam);
|
|
}
|
|
Token *tok2 = tok1->previous();
|
|
while (tok2->str() == "]")
|
|
tok2 = tok2->link()->previous();
|
|
|
|
//it should be a name..
|
|
if (!tok2->isName()) {
|
|
bailOut = true;
|
|
break;
|
|
}
|
|
|
|
if (argumentNames2.find(tok2->str()) != argumentNames2.end()) {
|
|
//same parameter names...
|
|
syntaxError(tok1);
|
|
} else
|
|
argumentNames2[tok2->str()] = tok2;
|
|
|
|
if (argumentNames.find(tok2->str()) == argumentNames.end()) {
|
|
//non-matching parameter... bailout
|
|
bailOut = true;
|
|
break;
|
|
}
|
|
}
|
|
tok1 = tok1->next();
|
|
}
|
|
|
|
if (bailOut || !tok1) {
|
|
tok = tok->link();
|
|
continue;
|
|
}
|
|
|
|
//the two containers may not hold the same size...
|
|
//in that case, the missing parameters are defined as 'int'
|
|
if (argumentNames.size() != argumentNames2.size()) {
|
|
//move back 'tok1' to the last ';'
|
|
tok1 = tok1->previous();
|
|
for (std::pair<const std::string, Token *>& argumentName : argumentNames) {
|
|
if (argumentNames2.find(argumentName.first) == argumentNames2.end()) {
|
|
//add the missing parameter argument declaration
|
|
tok1->insertToken(";");
|
|
tok1->insertToken(argumentName.first);
|
|
//register the change inside argumentNames2
|
|
argumentNames2[argumentName.first] = tok1->next();
|
|
tok1->insertToken("int");
|
|
}
|
|
}
|
|
}
|
|
|
|
while (tok->str() != ")") {
|
|
//initialize start and end tokens to be moved
|
|
Token *declStart = argumentNames2[tok->next()->str()];
|
|
Token *declEnd = declStart;
|
|
while (declStart->previous()->str() != ";" && declStart->previous()->str() != ")")
|
|
declStart = declStart->previous();
|
|
while (declEnd->next()->str() != ";" && declEnd->next()->str() != "{")
|
|
declEnd = declEnd->next();
|
|
|
|
//remove ';' after declaration
|
|
declEnd->deleteNext();
|
|
|
|
//replace the parameter name in the parentheses with all the declaration
|
|
Token::replace(tok->next(), declStart, declEnd);
|
|
|
|
//since there are changes to tokens, put tok where tok1 is
|
|
tok = declEnd->next();
|
|
|
|
//fix up line number
|
|
if (tok->str() == ",")
|
|
tok->linenr(tok->previous()->linenr());
|
|
}
|
|
//goto forward and continue
|
|
tok = tok->next()->link();
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyPointerToStandardType()
|
|
{
|
|
if (!isC())
|
|
return;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "& %name% [ 0 ] !!["))
|
|
continue;
|
|
|
|
if (!Token::Match(tok->previous(), "[,(=]"))
|
|
continue;
|
|
|
|
// Remove '[ 0 ]' suffix
|
|
Token::eraseTokens(tok->next(), tok->tokAt(5));
|
|
// Remove '&' prefix
|
|
tok = tok->previous();
|
|
if (!tok)
|
|
break;
|
|
tok->deleteNext();
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyFunctionPointers()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
// #2873 - do not simplify function pointer usage here:
|
|
// (void)(xy(*p)(0));
|
|
if (Token::simpleMatch(tok, ") (")) {
|
|
tok = tok->next()->link();
|
|
continue;
|
|
}
|
|
|
|
// check for function pointer cast
|
|
if (Token::Match(tok, "( %type% %type%| *| *| ( * ) (") ||
|
|
Token::Match(tok, "static_cast < %type% %type%| *| *| ( * ) (")) {
|
|
Token *tok1 = tok;
|
|
|
|
if (isCPP() && tok1->str() == "static_cast")
|
|
tok1 = tok1->next();
|
|
|
|
tok1 = tok1->next();
|
|
|
|
if (Token::Match(tok1->next(), "%type%"))
|
|
tok1 = tok1->next();
|
|
|
|
while (tok1->next()->str() == "*")
|
|
tok1 = tok1->next();
|
|
|
|
// check that the cast ends
|
|
if (!Token::Match(tok1->linkAt(4), ") )|>"))
|
|
continue;
|
|
|
|
// ok simplify this function pointer cast to an ordinary pointer cast
|
|
tok1->deleteNext();
|
|
tok1->next()->deleteNext();
|
|
Token::eraseTokens(tok1->next(), tok1->linkAt(2)->next());
|
|
continue;
|
|
}
|
|
|
|
// check for start of statement
|
|
else if (tok->previous() && !Token::Match(tok->previous(), "{|}|;|,|(|public:|protected:|private:"))
|
|
continue;
|
|
|
|
if (Token::Match(tok, "delete|else|return|throw|typedef"))
|
|
continue;
|
|
|
|
while (Token::Match(tok, "%type%|:: %type%|::"))
|
|
tok = tok->next();
|
|
|
|
Token *tok2 = (tok && tok->isName()) ? tok->next() : nullptr;
|
|
while (Token::Match(tok2, "*|&"))
|
|
tok2 = tok2->next();
|
|
if (!tok2 || tok2->str() != "(")
|
|
continue;
|
|
while (Token::Match(tok2, "(|:: %type%"))
|
|
tok2 = tok2->tokAt(2);
|
|
if (!Token::Match(tok2, "(|:: * *| %name%"))
|
|
continue;
|
|
tok2 = tok2->tokAt(2);
|
|
if (tok2->str() == "*")
|
|
tok2 = tok2->next();
|
|
while (Token::Match(tok2, "%type%|:: %type%|::"))
|
|
tok2 = tok2->next();
|
|
|
|
if (!Token::Match(tok2, "%name% ) (") &&
|
|
!Token::Match(tok2, "%name% [ ] ) (") &&
|
|
!(Token::Match(tok2, "%name% (") && Token::simpleMatch(tok2->linkAt(1), ") ) (")))
|
|
continue;
|
|
|
|
while (tok && tok->str() != "(")
|
|
tok = tok->next();
|
|
|
|
// check that the declaration ends
|
|
if (!tok || !tok->link() || !tok->link()->next()) {
|
|
syntaxError(nullptr);
|
|
}
|
|
Token *endTok = tok->link()->next()->link();
|
|
if (Token::simpleMatch(endTok, ") throw ("))
|
|
endTok = endTok->linkAt(2);
|
|
if (!Token::Match(endTok, ") const|volatile| const|volatile| ;|,|)|=|[|{"))
|
|
continue;
|
|
|
|
while (Token::Match(endTok->next(), "const|volatile"))
|
|
endTok->deleteNext();
|
|
|
|
// ok simplify this function pointer to an ordinary pointer
|
|
Token::eraseTokens(tok->link(), endTok->next());
|
|
if (Token::simpleMatch(tok->link()->previous(), ") )")) {
|
|
// Function returning function pointer
|
|
// void (*dostuff(void))(void) {}
|
|
tok->link()->deleteThis();
|
|
tok->deleteThis();
|
|
} else {
|
|
// Function pointer variable
|
|
// void (*p)(void) {}
|
|
tok->link()->insertToken("(");
|
|
Token *par1 = tok->link()->next();
|
|
par1->insertToken(")");
|
|
par1->link(par1->next());
|
|
par1->next()->link(par1);
|
|
while (Token::Match(tok, "( %type% ::"))
|
|
tok->deleteNext(2);
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
bool Tokenizer::simplifyFunctionReturn()
|
|
{
|
|
std::map<std::string, const Token*> functions;
|
|
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (tok->str() == "{")
|
|
tok = tok->link();
|
|
|
|
else if (Token::Match(tok, "%name% ( ) { return %bool%|%char%|%num%|%str% ; }") && tok->strAt(-1) != "::") {
|
|
const Token* const any = tok->tokAt(5);
|
|
functions[tok->str()] = any;
|
|
tok = any;
|
|
}
|
|
}
|
|
|
|
if (functions.empty())
|
|
return false;
|
|
|
|
bool ret = false;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "(|[|=|return|%op% %name% ( ) ;|]|)|%cop%")) {
|
|
tok = tok->next();
|
|
auto it = functions.find(tok->str());
|
|
if (it != functions.cend()) {
|
|
tok->str(it->second->str());
|
|
tok->deleteNext(2);
|
|
ret = true;
|
|
}
|
|
}
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
void Tokenizer::simplifyVarDecl(const bool only_k_r_fpar)
|
|
{
|
|
simplifyVarDecl(list.front(), nullptr, only_k_r_fpar);
|
|
}
|
|
|
|
void Tokenizer::simplifyVarDecl(Token * tokBegin, const Token * const tokEnd, const bool only_k_r_fpar)
|
|
{
|
|
const bool isCPP11 = mSettings->standards.cpp >= Standards::CPP11;
|
|
|
|
// Split up variable declarations..
|
|
// "int a=4;" => "int a; a=4;"
|
|
bool finishedwithkr = true;
|
|
bool scopeDecl = false;
|
|
for (Token *tok = tokBegin; tok != tokEnd; tok = tok->next()) {
|
|
if (Token::Match(tok, "{|;"))
|
|
scopeDecl = false;
|
|
if (isCPP()) {
|
|
if (Token::Match(tok, "class|struct|namespace|union"))
|
|
scopeDecl = true;
|
|
if (Token::Match(tok, "decltype|noexcept (")) {
|
|
tok = tok->next()->link();
|
|
// skip decltype(...){...}
|
|
if (tok && Token::simpleMatch(tok->previous(), ") {"))
|
|
tok = tok->link();
|
|
} else if (Token::simpleMatch(tok, "= {") ||
|
|
(!scopeDecl && Token::Match(tok, "%name%|> {") &&
|
|
!Token::Match(tok, "else|try|do|const|constexpr|override|volatile|noexcept"))) {
|
|
if (!tok->next()->link())
|
|
syntaxError(tokBegin);
|
|
// Check for lambdas before skipping
|
|
for (Token* tok2 = tok->next(); tok2 != tok->next()->link(); tok2 = tok2->next()) {
|
|
Token* lambdaEnd = findLambdaEndScope(tok2);
|
|
if (!lambdaEnd)
|
|
continue;
|
|
simplifyVarDecl(lambdaEnd->link()->next(), lambdaEnd, only_k_r_fpar);
|
|
}
|
|
tok = tok->next()->link();
|
|
}
|
|
|
|
} else if (Token::simpleMatch(tok, "= {")) {
|
|
tok = tok->next()->link();
|
|
}
|
|
if (!tok) {
|
|
syntaxError(tokBegin);
|
|
}
|
|
if (only_k_r_fpar && finishedwithkr) {
|
|
if (Token::Match(tok, "(|[|{")) {
|
|
tok = tok->link();
|
|
if (tok->next() && Token::Match(tok, ") !!{"))
|
|
tok = tok->next();
|
|
else
|
|
continue;
|
|
} else
|
|
continue;
|
|
} else if (tok->str() == "(") {
|
|
if (isCPP()) {
|
|
for (Token * tok2 = tok; tok2 && tok2 != tok->link(); tok2 = tok2->next()) {
|
|
if (Token::Match(tok2, "[(,] [")) {
|
|
// lambda function at tok2->next()
|
|
// find start of lambda body
|
|
Token * lambdaBody = tok2;
|
|
while (lambdaBody && lambdaBody != tok2->link() && lambdaBody->str() != "{")
|
|
lambdaBody = lambdaBody->next();
|
|
if (lambdaBody && lambdaBody != tok2->link() && lambdaBody->link())
|
|
simplifyVarDecl(lambdaBody, lambdaBody->link()->next(), only_k_r_fpar);
|
|
}
|
|
}
|
|
}
|
|
tok = tok->link();
|
|
}
|
|
|
|
if (!tok)
|
|
syntaxError(nullptr); // #7043 invalid code
|
|
if (tok->previous() && !Token::Match(tok->previous(), "{|}|;|)|public:|protected:|private:"))
|
|
continue;
|
|
if (Token::simpleMatch(tok, "template <"))
|
|
continue;
|
|
|
|
Token *type0 = tok;
|
|
if (!Token::Match(type0, "::|extern| %type%"))
|
|
continue;
|
|
if (Token::Match(type0, "else|return|public:|protected:|private:"))
|
|
continue;
|
|
if (isCPP11 && type0->str() == "using")
|
|
continue;
|
|
if (isCPP() && type0->str() == "namespace")
|
|
continue;
|
|
|
|
bool isconst = false;
|
|
bool isstatic = false;
|
|
Token *tok2 = type0;
|
|
int typelen = 1;
|
|
|
|
if (Token::Match(tok2, "::|extern")) {
|
|
tok2 = tok2->next();
|
|
typelen++;
|
|
}
|
|
|
|
//check if variable is declared 'const' or 'static' or both
|
|
while (tok2) {
|
|
if (!Token::Match(tok2, "const|static|constexpr") && Token::Match(tok2, "%type% const|static")) {
|
|
tok2 = tok2->next();
|
|
++typelen;
|
|
}
|
|
|
|
if (Token::Match(tok2, "const|constexpr"))
|
|
isconst = true;
|
|
|
|
else if (Token::Match(tok2, "static|constexpr"))
|
|
isstatic = true;
|
|
|
|
else if (Token::Match(tok2, "%type% :: %type%")) {
|
|
tok2 = tok2->next();
|
|
++typelen;
|
|
}
|
|
|
|
else
|
|
break;
|
|
|
|
if (tok2->strAt(1) == "*")
|
|
break;
|
|
|
|
if (Token::Match(tok2->next(), "& %name% ,"))
|
|
break;
|
|
|
|
tok2 = tok2->next();
|
|
++typelen;
|
|
}
|
|
|
|
// strange looking variable declaration => don't split up.
|
|
if (Token::Match(tok2, "%type% *|&| %name% , %type% *|&| %name%"))
|
|
continue;
|
|
|
|
if (Token::Match(tok2, "struct|union|class %type%")) {
|
|
tok2 = tok2->next();
|
|
++typelen;
|
|
}
|
|
|
|
// check for qualification..
|
|
if (Token::Match(tok2, ":: %type%")) {
|
|
++typelen;
|
|
tok2 = tok2->next();
|
|
}
|
|
|
|
//skip combinations of templates and namespaces
|
|
while (!isC() && (Token::Match(tok2, "%type% <") || Token::Match(tok2, "%type% ::"))) {
|
|
if (tok2->next()->str() == "<" && !TemplateSimplifier::templateParameters(tok2->next())) {
|
|
tok2 = nullptr;
|
|
break;
|
|
}
|
|
typelen += 2;
|
|
tok2 = tok2->tokAt(2);
|
|
if (tok2 && tok2->previous()->str() == "::")
|
|
continue;
|
|
int indentlevel = 0;
|
|
int parens = 0;
|
|
|
|
for (Token *tok3 = tok2; tok3; tok3 = tok3->next()) {
|
|
++typelen;
|
|
|
|
if (!parens && tok3->str() == "<") {
|
|
++indentlevel;
|
|
} else if (!parens && tok3->str() == ">") {
|
|
if (indentlevel == 0) {
|
|
tok2 = tok3->next();
|
|
break;
|
|
}
|
|
--indentlevel;
|
|
} else if (!parens && tok3->str() == ">>") {
|
|
if (indentlevel <= 1) {
|
|
tok2 = tok3->next();
|
|
break;
|
|
}
|
|
indentlevel -= 2;
|
|
} else if (tok3->str() == "(") {
|
|
++parens;
|
|
} else if (tok3->str() == ")") {
|
|
if (!parens) {
|
|
tok2 = nullptr;
|
|
break;
|
|
}
|
|
--parens;
|
|
} else if (tok3->str() == ";") {
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (Token::Match(tok2, ":: %type%")) {
|
|
++typelen;
|
|
tok2 = tok2->next();
|
|
}
|
|
}
|
|
|
|
//pattern: "%type% *| ... *| const| %name% ,|="
|
|
if (Token::Match(tok2, "%type%") ||
|
|
(tok2 && tok2->previous() && tok2->previous()->str() == ">")) {
|
|
Token *varName = tok2;
|
|
if (!tok2->previous() || tok2->previous()->str() != ">")
|
|
varName = varName->next();
|
|
else
|
|
--typelen;
|
|
//skip all the pointer part
|
|
bool isPointerOrRef = false;
|
|
while (Token::simpleMatch(varName, "*") || Token::Match(varName, "& %name% ,")) {
|
|
isPointerOrRef = true;
|
|
varName = varName->next();
|
|
}
|
|
|
|
while (Token::Match(varName, "%type% %type%")) {
|
|
if (varName->str() != "const") {
|
|
++typelen;
|
|
}
|
|
varName = varName->next();
|
|
}
|
|
// Function pointer
|
|
if (Token::simpleMatch(varName, "( *") && Token::Match(varName->link()->previous(), "%name% ) ( ) =")) {
|
|
Token *endDecl = varName->link()->tokAt(2);
|
|
varName = varName->link()->previous();
|
|
endDecl->insertToken(";");
|
|
endDecl = endDecl->next();
|
|
endDecl->next()->isSplittedVarDeclEq(true);
|
|
endDecl->insertToken(varName->str());
|
|
continue;
|
|
}
|
|
//non-VLA case
|
|
else if (Token::Match(varName, "%name% ,|=")) {
|
|
if (varName->str() != "operator") {
|
|
tok2 = varName->next(); // The ',' or '=' token
|
|
|
|
if (tok2->str() == "=" && (isstatic || (isconst && !isPointerOrRef))) {
|
|
//do not split const non-pointer variables..
|
|
while (tok2 && tok2->str() != "," && tok2->str() != ";") {
|
|
if (Token::Match(tok2, "{|(|["))
|
|
tok2 = tok2->link();
|
|
const Token *tok3 = tok2;
|
|
if (!isC() && tok2->str() == "<" && TemplateSimplifier::templateParameters(tok2) > 0) {
|
|
tok2 = tok2->findClosingBracket();
|
|
}
|
|
if (!tok2)
|
|
syntaxError(tok3); // #6881 invalid code
|
|
tok2 = tok2->next();
|
|
}
|
|
if (tok2 && tok2->str() == ";")
|
|
tok2 = nullptr;
|
|
}
|
|
} else
|
|
tok2 = nullptr;
|
|
}
|
|
|
|
//VLA case
|
|
else if (Token::Match(varName, "%name% [")) {
|
|
tok2 = varName->next();
|
|
|
|
while (Token::Match(tok2->link(), "] ,|=|["))
|
|
tok2 = tok2->link()->next();
|
|
if (!Token::Match(tok2, "=|,"))
|
|
tok2 = nullptr;
|
|
if (tok2 && tok2->str() == "=") {
|
|
while (tok2 && tok2->str() != "," && tok2->str() != ";") {
|
|
if (Token::Match(tok2, "{|(|["))
|
|
tok2 = tok2->link();
|
|
tok2 = tok2->next();
|
|
}
|
|
if (tok2 && tok2->str() == ";")
|
|
tok2 = nullptr;
|
|
}
|
|
}
|
|
|
|
// brace initialization
|
|
else if (Token::Match(varName, "%name% {")) {
|
|
tok2 = varName->next();
|
|
tok2 = tok2->link();
|
|
if (tok2)
|
|
tok2 = tok2->next();
|
|
if (tok2 && tok2->str() != ",")
|
|
tok2 = nullptr;
|
|
}
|
|
|
|
// parenthesis, functions can't be declared like:
|
|
// int f1(a,b), f2(c,d);
|
|
// so if there is a comma assume this is a variable declaration
|
|
else if (Token::Match(varName, "%name% (") && Token::simpleMatch(varName->linkAt(1), ") ,")) {
|
|
tok2 = varName->linkAt(1)->next();
|
|
}
|
|
|
|
else
|
|
tok2 = nullptr;
|
|
} else {
|
|
tok2 = nullptr;
|
|
}
|
|
|
|
if (!tok2) {
|
|
if (only_k_r_fpar)
|
|
finishedwithkr = false;
|
|
continue;
|
|
}
|
|
|
|
if (tok2->str() == ",") {
|
|
tok2->str(";");
|
|
tok2->isSplittedVarDeclComma(true);
|
|
//TODO: should we have to add also template '<>' links?
|
|
TokenList::insertTokens(tok2, type0, typelen);
|
|
}
|
|
|
|
else {
|
|
Token *eq = tok2;
|
|
|
|
while (tok2) {
|
|
if (Token::Match(tok2, "{|(|["))
|
|
tok2 = tok2->link();
|
|
|
|
else if (!isC() && tok2->str() == "<" && tok2->previous()->isName() && !tok2->previous()->varId())
|
|
tok2 = tok2->findClosingBracket();
|
|
|
|
else if (std::strchr(";,", tok2->str()[0])) {
|
|
// "type var =" => "type var; var ="
|
|
const Token *varTok = type0->tokAt(typelen);
|
|
while (Token::Match(varTok, "%name%|*|& %name%|*|&"))
|
|
varTok = varTok->next();
|
|
if (!varTok)
|
|
syntaxError(tok2); // invalid code
|
|
TokenList::insertTokens(eq, varTok, 2);
|
|
eq->str(";");
|
|
eq->isSplittedVarDeclEq(true);
|
|
|
|
// "= x, " => "= x; type "
|
|
if (tok2->str() == ",") {
|
|
tok2->str(";");
|
|
tok2->isSplittedVarDeclComma(true);
|
|
TokenList::insertTokens(tok2, type0, typelen);
|
|
}
|
|
break;
|
|
}
|
|
if (tok2)
|
|
tok2 = tok2->next();
|
|
}
|
|
}
|
|
finishedwithkr = (only_k_r_fpar && tok2 && tok2->strAt(1) == "{");
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyStaticConst()
|
|
{
|
|
// This function will simplify the token list so that the qualifiers "extern", "static"
|
|
// and "const" appear in the same order as in the array below.
|
|
const std::string qualifiers[] = {"extern", "static", "const"};
|
|
|
|
// Move 'const' before all other qualifiers and types and then
|
|
// move 'static' before all other qualifiers and types, ...
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
bool continue2 = false;
|
|
for (int i = 0; i < sizeof(qualifiers)/sizeof(qualifiers[0]); i++) {
|
|
|
|
// Keep searching for a qualifier
|
|
if (!tok->next() || tok->next()->str() != qualifiers[i])
|
|
continue;
|
|
|
|
// Look backwards to find the beginning of the declaration
|
|
Token* leftTok = tok;
|
|
bool behindOther = false;
|
|
for (; leftTok; leftTok = leftTok->previous()) {
|
|
for (int j = 0; j <= i; j++) {
|
|
if (leftTok->str() == qualifiers[j]) {
|
|
behindOther = true;
|
|
break;
|
|
}
|
|
}
|
|
if (behindOther)
|
|
break;
|
|
if (!Token::Match(leftTok, "%type%|struct|::") ||
|
|
(isCPP() && Token::Match(leftTok, "private:|protected:|public:|operator|template"))) {
|
|
break;
|
|
}
|
|
}
|
|
|
|
// The token preceding the declaration should indicate the start of a declaration
|
|
if (leftTok == tok)
|
|
continue;
|
|
|
|
if (leftTok && !behindOther && !Token::Match(leftTok, ";|{|}|(|,|private:|protected:|public:")) {
|
|
continue2 = true;
|
|
break;
|
|
}
|
|
|
|
// Move the qualifier to the left-most position in the declaration
|
|
tok->deleteNext();
|
|
if (!leftTok) {
|
|
list.front()->insertToken(qualifiers[i], emptyString, false);
|
|
list.front()->swapWithNext();
|
|
tok = list.front();
|
|
} else if (leftTok->next()) {
|
|
leftTok->next()->insertToken(qualifiers[i], emptyString, true);
|
|
tok = leftTok->next();
|
|
} else {
|
|
leftTok->insertToken(qualifiers[i]);
|
|
tok = leftTok;
|
|
}
|
|
}
|
|
if (continue2)
|
|
continue;
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyIfAndWhileAssign()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok->next(), "if|while ("))
|
|
continue;
|
|
|
|
const Token* tokAt3 = tok->tokAt(3);
|
|
if (!Token::Match(tokAt3, "!| (| %name% =") &&
|
|
!Token::Match(tokAt3, "!| (| %name% . %name% =") &&
|
|
!Token::Match(tokAt3, "0 == (| %name% =") &&
|
|
!Token::Match(tokAt3, "0 == (| %name% . %name% ="))
|
|
continue;
|
|
|
|
// simplifying a "while(cond) { }" condition ?
|
|
const bool iswhile(tok->next()->str() == "while");
|
|
|
|
// simplifying a "do { } while(cond);" condition ?
|
|
const bool isDoWhile = iswhile && Token::simpleMatch(tok, "}") && Token::simpleMatch(tok->link()->previous(), "do");
|
|
Token* openBraceTok = tok->link();
|
|
|
|
// delete the "if|while"
|
|
tok->deleteNext();
|
|
|
|
// Remember if there is a "!" or not. And delete it if there are.
|
|
const bool isNot(Token::Match(tok->tokAt(2), "!|0"));
|
|
if (isNot)
|
|
tok->next()->deleteNext((tok->strAt(2) == "0") ? 2 : 1);
|
|
|
|
// Delete parentheses.. and remember how many there are with
|
|
// their links.
|
|
std::stack<Token *> braces;
|
|
while (tok->next()->str() == "(") {
|
|
braces.push(tok->next()->link());
|
|
tok->deleteNext();
|
|
}
|
|
|
|
// Skip the "%name% = ..."
|
|
Token *tok2;
|
|
for (tok2 = tok->next(); tok2; tok2 = tok2->next()) {
|
|
if (tok2->str() == "(")
|
|
tok2 = tok2->link();
|
|
else if (tok2->str() == ")")
|
|
break;
|
|
}
|
|
|
|
// Insert "; if|while ( .."
|
|
tok2 = tok2->previous();
|
|
if (tok->strAt(2) == ".") {
|
|
tok2->insertToken(tok->strAt(3));
|
|
tok2->next()->varId(tok->tokAt(3)->varId());
|
|
tok2->insertToken(".");
|
|
}
|
|
tok2->insertToken(tok->next()->str());
|
|
tok2->next()->varId(tok->next()->varId());
|
|
|
|
while (!braces.empty()) {
|
|
tok2->insertToken("(");
|
|
Token::createMutualLinks(tok2->next(), braces.top());
|
|
braces.pop();
|
|
}
|
|
|
|
if (isNot)
|
|
tok2->next()->insertToken("!");
|
|
tok2->insertToken(iswhile ? "while" : "if");
|
|
if (isDoWhile) {
|
|
tok2->insertToken("}");
|
|
Token::createMutualLinks(openBraceTok, tok2->next());
|
|
}
|
|
|
|
tok2->insertToken(";");
|
|
|
|
// delete the extra "}"
|
|
if (isDoWhile)
|
|
tok->deleteThis();
|
|
|
|
// If it's a while loop, insert the assignment in the loop
|
|
if (iswhile && !isDoWhile) {
|
|
int indentlevel = 0;
|
|
Token *tok3 = tok2;
|
|
|
|
for (; tok3; tok3 = tok3->next()) {
|
|
if (tok3->str() == "{")
|
|
++indentlevel;
|
|
else if (tok3->str() == "}") {
|
|
if (indentlevel <= 1)
|
|
break;
|
|
--indentlevel;
|
|
}
|
|
}
|
|
|
|
if (tok3 && indentlevel == 1) {
|
|
tok3 = tok3->previous();
|
|
std::stack<Token *> braces2;
|
|
|
|
for (tok2 = tok2->next(); tok2 && tok2 != tok; tok2 = tok2->previous()) {
|
|
tok3->insertToken(tok2->str());
|
|
Token *newTok = tok3->next();
|
|
|
|
newTok->varId(tok2->varId());
|
|
newTok->fileIndex(tok2->fileIndex());
|
|
newTok->linenr(tok2->linenr());
|
|
|
|
// link() new tokens manually
|
|
if (tok2->link()) {
|
|
if (Token::Match(newTok, "}|)|]|>")) {
|
|
braces2.push(newTok);
|
|
} else {
|
|
Token::createMutualLinks(newTok, braces2.top());
|
|
braces2.pop();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyVariableMultipleAssign()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "%name% = %name% = %num%|%name% ;")) {
|
|
// skip intermediate assignments
|
|
Token *tok2 = tok->previous();
|
|
while (tok2 &&
|
|
tok2->str() == "=" &&
|
|
Token::Match(tok2->previous(), "%name%")) {
|
|
tok2 = tok2->tokAt(-2);
|
|
}
|
|
|
|
if (!tok2 || tok2->str() != ";") {
|
|
continue;
|
|
}
|
|
|
|
Token *stopAt = tok->tokAt(2);
|
|
const Token *valueTok = stopAt->tokAt(2);
|
|
const std::string& value(valueTok->str());
|
|
tok2 = tok2->next();
|
|
|
|
while (tok2 != stopAt) {
|
|
tok2->next()->insertToken(";");
|
|
tok2->next()->insertToken(value);
|
|
tok2 = tok2->tokAt(4);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// Binary operators simplification map
|
|
static const std::unordered_map<std::string, std::string> cAlternativeTokens = {
|
|
std::make_pair("and", "&&")
|
|
, std::make_pair("and_eq", "&=")
|
|
, std::make_pair("bitand", "&")
|
|
, std::make_pair("bitor", "|")
|
|
, std::make_pair("not_eq", "!=")
|
|
, std::make_pair("or", "||")
|
|
, std::make_pair("or_eq", "|=")
|
|
, std::make_pair("xor", "^")
|
|
, std::make_pair("xor_eq", "^=")
|
|
};
|
|
|
|
// Simplify the C alternative tokens:
|
|
// and => &&
|
|
// and_eq => &=
|
|
// bitand => &
|
|
// bitor => |
|
|
// compl => ~
|
|
// not => !
|
|
// not_eq => !=
|
|
// or => ||
|
|
// or_eq => |=
|
|
// xor => ^
|
|
// xor_eq => ^=
|
|
bool Tokenizer::simplifyCAlternativeTokens()
|
|
{
|
|
/* executable scope level */
|
|
int executableScopeLevel = 0;
|
|
|
|
std::vector<Token *> alt;
|
|
bool replaceAll = false; // replace all or none
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() == ")") {
|
|
if (const Token *end = isFunctionHead(tok, "{")) {
|
|
++executableScopeLevel;
|
|
tok = const_cast<Token *>(end);
|
|
continue;
|
|
}
|
|
}
|
|
|
|
if (tok->str() == "{") {
|
|
if (executableScopeLevel > 0)
|
|
++executableScopeLevel;
|
|
continue;
|
|
}
|
|
|
|
if (tok->str() == "}") {
|
|
if (executableScopeLevel > 0)
|
|
--executableScopeLevel;
|
|
continue;
|
|
}
|
|
|
|
if (!tok->isName())
|
|
continue;
|
|
|
|
const std::unordered_map<std::string, std::string>::const_iterator cOpIt = cAlternativeTokens.find(tok->str());
|
|
if (cOpIt != cAlternativeTokens.end()) {
|
|
alt.push_back(tok);
|
|
|
|
// Is this a variable declaration..
|
|
if (isC() && Token::Match(tok->previous(), "%type%|* %name% [;,=]"))
|
|
return false;
|
|
|
|
if (!Token::Match(tok->previous(), "%name%|%num%|%char%|)|]|> %name% %name%|%num%|%char%|%op%|("))
|
|
continue;
|
|
if (Token::Match(tok->next(), "%assign%|%or%|%oror%|&&|*|/|%|^") && !Token::Match(tok->previous(), "%num%|%char%|) %name% *"))
|
|
continue;
|
|
if (executableScopeLevel == 0 && Token::Match(tok, "%name% (")) {
|
|
const Token *start = tok;
|
|
while (Token::Match(start, "%name%|*"))
|
|
start = start->previous();
|
|
if (!start || Token::Match(start, "[;}]"))
|
|
continue;
|
|
}
|
|
replaceAll = true;
|
|
} else if (Token::Match(tok, "not|compl")) {
|
|
alt.push_back(tok);
|
|
|
|
if (Token::Match(tok->previous(), "%assign%") || Token::Match(tok->next(), "%num%")) {
|
|
replaceAll = true;
|
|
continue;
|
|
}
|
|
|
|
// Don't simplify 'not p;' (in case 'not' is a type)
|
|
if (!Token::Match(tok->next(), "%name%|(") ||
|
|
Token::Match(tok->previous(), "[;{}]") ||
|
|
(executableScopeLevel == 0U && tok->strAt(-1) == "("))
|
|
continue;
|
|
|
|
replaceAll = true;
|
|
}
|
|
}
|
|
|
|
if (!replaceAll)
|
|
return false;
|
|
|
|
for (Token *tok: alt) {
|
|
const std::unordered_map<std::string, std::string>::const_iterator cOpIt = cAlternativeTokens.find(tok->str());
|
|
if (cOpIt != cAlternativeTokens.end())
|
|
tok->str(cOpIt->second);
|
|
else if (tok->str() == "not")
|
|
tok->str("!");
|
|
else
|
|
tok->str("~");
|
|
}
|
|
|
|
return !alt.empty();
|
|
}
|
|
|
|
// int i(0); => int i; i = 0;
|
|
// int i(0), j; => int i; i = 0; int j;
|
|
void Tokenizer::simplifyInitVar()
|
|
{
|
|
if (isC())
|
|
return;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!tok->isName() || (tok->previous() && !Token::Match(tok->previous(), "[;{}]")))
|
|
continue;
|
|
|
|
if (tok->str() == "return")
|
|
continue;
|
|
|
|
if (Token::Match(tok, "class|struct|union| %type% *| %name% ( &| %any% ) ;")) {
|
|
tok = initVar(tok);
|
|
} else if (Token::Match(tok, "%type% *| %name% ( %type% (")) {
|
|
const Token* tok2 = tok->tokAt(2);
|
|
if (!tok2->link())
|
|
tok2 = tok2->next();
|
|
if (!tok2->link() || (tok2->link()->strAt(1) == ";" && !Token::simpleMatch(tok2->linkAt(2), ") (")))
|
|
tok = initVar(tok);
|
|
} else if (Token::Match(tok, "class|struct|union| %type% *| %name% ( &| %any% ) ,")) {
|
|
Token *tok1 = tok->tokAt(5);
|
|
while (tok1->str() != ",")
|
|
tok1 = tok1->next();
|
|
tok1->str(";");
|
|
|
|
const int numTokens = (Token::Match(tok, "class|struct|union")) ? 2U : 1U;
|
|
TokenList::insertTokens(tok1, tok, numTokens);
|
|
tok = initVar(tok);
|
|
}
|
|
}
|
|
}
|
|
|
|
Token * Tokenizer::initVar(Token * tok)
|
|
{
|
|
// call constructor of class => no simplification
|
|
if (Token::Match(tok, "class|struct|union")) {
|
|
if (tok->strAt(2) != "*")
|
|
return tok;
|
|
|
|
tok = tok->next();
|
|
} else if (!tok->isStandardType() && tok->str() != "auto" && tok->next()->str() != "*")
|
|
return tok;
|
|
|
|
// goto variable name..
|
|
tok = tok->next();
|
|
if (tok->str() == "*")
|
|
tok = tok->next();
|
|
|
|
// sizeof is not a variable name..
|
|
if (tok->str() == "sizeof")
|
|
return tok;
|
|
|
|
// check initializer..
|
|
if (tok->tokAt(2)->isStandardType() || tok->strAt(2) == "void")
|
|
return tok;
|
|
else if (!tok->tokAt(2)->isNumber() && !Token::Match(tok->tokAt(2), "%type% (") && tok->strAt(2) != "&" && tok->tokAt(2)->varId() == 0)
|
|
return tok;
|
|
|
|
// insert '; var ='
|
|
tok->insertToken(";");
|
|
tok->next()->insertToken(tok->str());
|
|
tok->tokAt(2)->varId(tok->varId());
|
|
tok = tok->tokAt(2);
|
|
tok->insertToken("=");
|
|
|
|
// goto '('..
|
|
tok = tok->tokAt(2);
|
|
|
|
// delete ')'
|
|
tok->link()->deleteThis();
|
|
|
|
// delete this
|
|
tok->deleteThis();
|
|
|
|
return tok;
|
|
}
|
|
|
|
|
|
bool Tokenizer::simplifyKnownVariables()
|
|
{
|
|
// return value for function. Set to true if any simplifications are made
|
|
bool ret = false;
|
|
|
|
// constants..
|
|
{
|
|
std::unordered_map<int, std::string> constantValues;
|
|
std::map<int, Token*> constantVars;
|
|
std::unordered_map<int, std::list<Token*>> constantValueUsages;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
// Reference to variable
|
|
if (Token::Match(tok, "%type%|* & %name% = %name% ;")) {
|
|
Token *start = tok->previous();
|
|
while (Token::Match(start,"%type%|*|&"))
|
|
start = start->previous();
|
|
if (!Token::Match(start,"[;{}]"))
|
|
continue;
|
|
const Token *reftok = tok->tokAt(2);
|
|
const Token *vartok = reftok->tokAt(2);
|
|
int level = 0;
|
|
for (Token *tok2 = tok->tokAt(6); tok2; tok2 = tok2->next()) {
|
|
if (tok2->str() == "{") {
|
|
++level;
|
|
} else if (tok2->str() == "}") {
|
|
if (level <= 0)
|
|
break;
|
|
--level;
|
|
} else if (tok2->varId() == reftok->varId()) {
|
|
tok2->str(vartok->str());
|
|
tok2->varId(vartok->varId());
|
|
}
|
|
}
|
|
Token::eraseTokens(start, tok->tokAt(6));
|
|
tok = start;
|
|
}
|
|
|
|
if (tok->isName() && (Token::Match(tok, "static| const| static| %type% const| %name% = %any% ;") ||
|
|
Token::Match(tok, "static| const| static| %type% const| %name% ( %any% ) ;"))) {
|
|
bool isconst = false;
|
|
for (const Token *tok2 = tok; (tok2->str() != "=") && (tok2->str() != "("); tok2 = tok2->next()) {
|
|
if (tok2->str() == "const") {
|
|
isconst = true;
|
|
break;
|
|
}
|
|
}
|
|
if (!isconst)
|
|
continue;
|
|
|
|
Token *tok1 = tok;
|
|
|
|
// start of statement
|
|
if (tok != list.front() && !Token::Match(tok->previous(),";|{|}|private:|protected:|public:"))
|
|
continue;
|
|
// skip "const" and "static"
|
|
while (Token::Match(tok, "const|static"))
|
|
tok = tok->next();
|
|
// pod type
|
|
if (!tok->isStandardType())
|
|
continue;
|
|
|
|
Token * const vartok = (tok->next() && tok->next()->str() == "const") ? tok->tokAt(2) : tok->next();
|
|
const Token * const valuetok = vartok->tokAt(2);
|
|
if (Token::Match(valuetok, "%bool%|%char%|%num%|%str% )| ;")) {
|
|
// record a constant value for this variable
|
|
constantValues[vartok->varId()] = valuetok->str();
|
|
constantVars[vartok->varId()] = tok1;
|
|
}
|
|
} else if (tok->varId()) {
|
|
// find the entry for the known variable, if any. Exclude the location where the variable is assigned with next == "="
|
|
if (constantValues.find(tok->varId()) != constantValues.end() && tok->next()->str() != "=") {
|
|
constantValueUsages[tok->varId()].push_back(tok);
|
|
}
|
|
}
|
|
}
|
|
|
|
for (auto constantVar = constantVars.rbegin(); constantVar != constantVars.rend(); constantVar++) {
|
|
bool referenceFound = false;
|
|
std::list<Token*> usageList = constantValueUsages[constantVar->first];
|
|
for (Token* usage : usageList) {
|
|
// check if any usages of each known variable are a reference
|
|
if (Token::Match(usage->tokAt(-2), "(|[|,|{|return|%op% & %varid%", constantVar->first)) {
|
|
referenceFound = true;
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (!referenceFound) {
|
|
// replace all usages of non-referenced known variables with their value
|
|
for (Token* usage : usageList) {
|
|
usage->str(constantValues[constantVar->first]);
|
|
}
|
|
|
|
Token* startTok = constantVar->second;
|
|
// remove variable assignment statement
|
|
while (startTok->next()->str() != ";")
|
|
startTok->deleteNext();
|
|
startTok->deleteNext();
|
|
|
|
// #8579 if we can we want another token to delete startTok. if we can't it doesn't matter
|
|
if (startTok->previous()) {
|
|
startTok->previous()->deleteNext();
|
|
} else if (startTok->next()) {
|
|
startTok->next()->deletePrevious();
|
|
} else {
|
|
startTok->deleteThis();
|
|
}
|
|
startTok = nullptr;
|
|
|
|
constantVar->second = nullptr;
|
|
ret = true;
|
|
}
|
|
}
|
|
}
|
|
|
|
// variable id for local, float/double, array variables
|
|
std::set<int> localvars;
|
|
std::set<int> floatvars;
|
|
std::set<int> arrays;
|
|
|
|
// auto variables..
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
// Search for a block of code
|
|
Token * const start = const_cast<Token *>(startOfExecutableScope(tok));
|
|
if (!start)
|
|
continue;
|
|
|
|
for (const Token *tok2 = start->previous(); tok2 && !Token::Match(tok2, "[;{}]"); tok2 = tok2->previous()) {
|
|
if (tok2->varId() != 0)
|
|
localvars.insert(tok2->varId());
|
|
}
|
|
|
|
tok = start;
|
|
// parse the block of code..
|
|
int indentlevel = 0;
|
|
Token *tok2 = tok;
|
|
for (; tok2; tok2 = tok2->next()) {
|
|
if (Token::Match(tok2, "[;{}] %type% %name%|*")) {
|
|
bool isfloat = false;
|
|
bool ispointer = false;
|
|
const Token *vartok = tok2->next();
|
|
while (Token::Match(vartok, "%name%|* %name%|*")) {
|
|
if (Token::Match(vartok, "float|double"))
|
|
isfloat = true;
|
|
if (vartok->str() == "*")
|
|
ispointer = true;
|
|
vartok = vartok->next();
|
|
}
|
|
if (Token::Match(vartok, "%var% ;|["))
|
|
localvars.insert(vartok->varId());
|
|
if (isfloat && !ispointer && Token::Match(vartok, "%var% ;"))
|
|
floatvars.insert(vartok->varId());
|
|
if (Token::Match(vartok, "%var% ["))
|
|
arrays.insert(vartok->varId());
|
|
}
|
|
|
|
if (tok2->str() == "{")
|
|
++indentlevel;
|
|
|
|
else if (tok2->str() == "}") {
|
|
--indentlevel;
|
|
if (indentlevel <= 0)
|
|
break;
|
|
}
|
|
|
|
else if (Token::simpleMatch(tok2, "for ("))
|
|
tok2 = tok2->next()->link();
|
|
|
|
else if (tok2->previous()->str() != "*" && !Token::Match(tok2->tokAt(-2), "* --|++") &&
|
|
(Token::Match(tok2, "%name% = %bool%|%char%|%num%|%str%|%name% ;") ||
|
|
Token::Match(tok2, "%name% [ %num%| ] = %str% ;") ||
|
|
Token::Match(tok2, "%name% = & %name% ;") ||
|
|
(Token::Match(tok2, "%name% = & %name% [ 0 ] ;") && arrays.find(tok2->tokAt(3)->varId()) != arrays.end()))) {
|
|
const int varid = tok2->varId();
|
|
if (varid == 0)
|
|
continue;
|
|
|
|
if (Token::Match(tok2->previous(), "[;{}]") && localvars.find(varid) == localvars.end())
|
|
continue;
|
|
|
|
// initialization of static variable => the value is not *known*
|
|
{
|
|
bool isstatic = false;
|
|
const Token *decl = tok2->previous();
|
|
while (decl && (decl->isName() || decl->str() == "*")) {
|
|
if (decl->str() == "static") {
|
|
isstatic = true;
|
|
break;
|
|
}
|
|
decl = decl->previous();
|
|
}
|
|
if (isstatic)
|
|
continue;
|
|
}
|
|
|
|
// skip loop variable
|
|
if (Token::Match(tok2->tokAt(-2), "(|:: %type%")) {
|
|
const Token *tok3 = tok2->previous();
|
|
do {
|
|
tok3 = tok3->tokAt(-2);
|
|
} while (Token::Match(tok3->previous(), ":: %type%"));
|
|
if (Token::Match(tok3->tokAt(-2), "for ( %type%"))
|
|
continue;
|
|
}
|
|
|
|
// struct name..
|
|
if (Token::Match(tok2, "%varid% = &| %varid%", tok2->varId()))
|
|
continue;
|
|
|
|
const std::string structname = Token::Match(tok2->tokAt(-3), "[;{}] %name% .") ?
|
|
std::string(tok2->strAt(-2) + " .") :
|
|
std::string();
|
|
|
|
const Token * const valueToken = tok2->tokAt(2);
|
|
|
|
std::string value;
|
|
nonneg int valueVarId = 0;
|
|
|
|
Token *tok3 = nullptr;
|
|
bool valueIsPointer = false;
|
|
|
|
// there could be a hang here if tok2 is moved back by the function calls below for some reason
|
|
if (Settings::terminated())
|
|
return false;
|
|
|
|
if (!simplifyKnownVariablesGetData(varid, &tok2, &tok3, value, valueVarId, valueIsPointer, floatvars.find(tok2->varId()) != floatvars.end()))
|
|
continue;
|
|
|
|
if (valueVarId > 0 && arrays.find(valueVarId) != arrays.end())
|
|
continue;
|
|
|
|
ret |= simplifyKnownVariablesSimplify(&tok2, tok3, varid, structname, value, valueVarId, valueIsPointer, valueToken, indentlevel);
|
|
}
|
|
|
|
else if (Token::Match(tok2, "strcpy|sprintf ( %name% , %str% ) ;")) {
|
|
const int varid(tok2->tokAt(2)->varId());
|
|
if (varid == 0)
|
|
continue;
|
|
|
|
const Token * const valueToken = tok2->tokAt(4);
|
|
std::string value(valueToken->str());
|
|
if (tok2->str() == "sprintf") {
|
|
std::string::size_type n = 0;
|
|
while ((n = value.find("%%", n)) != std::string::npos) {
|
|
// Replace "%%" with "%" - erase the first '%' and continue past the second '%'
|
|
value.erase(n, 1);
|
|
++n;
|
|
}
|
|
}
|
|
const int valueVarId(0);
|
|
const bool valueIsPointer(false);
|
|
Token *tok3 = tok2->tokAt(6);
|
|
ret |= simplifyKnownVariablesSimplify(&tok2, tok3, varid, emptyString, value, valueVarId, valueIsPointer, valueToken, indentlevel);
|
|
|
|
// there could be a hang here if tok2 was moved back by the function call above for some reason
|
|
if (Settings::terminated())
|
|
return false;
|
|
}
|
|
}
|
|
|
|
if (tok2)
|
|
tok = tok2->previous();
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
bool Tokenizer::simplifyKnownVariablesGetData(nonneg int varid, Token **_tok2, Token **_tok3, std::string &value, nonneg int &valueVarId, bool &valueIsPointer, bool floatvar)
|
|
{
|
|
Token *tok2 = *_tok2;
|
|
Token *tok3 = nullptr;
|
|
|
|
if (Token::simpleMatch(tok2->tokAt(-2), "for (")) {
|
|
// only specific for loops is handled
|
|
if (!Token::Match(tok2, "%varid% = %num% ; %varid% <|<= %num% ; ++| %varid% ++| ) {", varid))
|
|
return false;
|
|
|
|
// is there a "break" in the for loop?
|
|
bool hasbreak = false;
|
|
const Token* end4 = tok2->linkAt(-1)->linkAt(1);
|
|
for (const Token *tok4 = tok2->previous()->link(); tok4 != end4; tok4 = tok4->next()) {
|
|
if (tok4->str() == "break") {
|
|
hasbreak = true;
|
|
break;
|
|
}
|
|
}
|
|
if (hasbreak)
|
|
return false;
|
|
|
|
// no break => the value of the counter value is known after the for loop..
|
|
const Token* compareTok = tok2->tokAt(5);
|
|
if (compareTok->str() == "<") {
|
|
value = compareTok->next()->str();
|
|
valueVarId = compareTok->next()->varId();
|
|
} else
|
|
value = MathLib::toString(MathLib::toLongNumber(compareTok->next()->str()) + 1);
|
|
|
|
// Skip for-body..
|
|
tok3 = tok2->previous()->link()->next()->link()->next();
|
|
} else {
|
|
value = tok2->strAt(2);
|
|
valueVarId = tok2->tokAt(2)->varId();
|
|
if (tok2->strAt(1) == "[") {
|
|
value = tok2->next()->link()->strAt(2);
|
|
valueVarId = 0;
|
|
} else if (value == "&") {
|
|
value = tok2->strAt(3);
|
|
valueVarId = tok2->tokAt(3)->varId();
|
|
|
|
// *ptr = &var; *ptr = 5;
|
|
// equals
|
|
// var = 5; not *var = 5;
|
|
if (tok2->strAt(4) == ";")
|
|
valueIsPointer = true;
|
|
}
|
|
|
|
// Add a '.0' to a decimal value and therefore convert it to an floating point number.
|
|
else if (MathLib::isDec(tok2->strAt(2)) && floatvar) {
|
|
value += ".0";
|
|
}
|
|
|
|
// float variable: convert true/false to 1.0 / 0.0
|
|
else if (tok2->tokAt(2)->isBoolean() && floatvar) {
|
|
value = (value == "true") ? "1.0" : "0.0";
|
|
}
|
|
|
|
if (Token::simpleMatch(tok2->next(), "= &"))
|
|
tok2 = tok2->tokAt(3);
|
|
|
|
tok3 = tok2->next();
|
|
}
|
|
*_tok2 = tok2;
|
|
*_tok3 = tok3;
|
|
return true;
|
|
}
|
|
|
|
bool Tokenizer::simplifyKnownVariablesSimplify(Token **tok2, Token *tok3, nonneg int varid, const std::string &structname, std::string &value, nonneg int valueVarId, bool valueIsPointer, const Token * const valueToken, int indentlevel) const
|
|
{
|
|
const bool pointeralias(valueToken->isName() || Token::Match(valueToken, "& %name% ["));
|
|
const bool varIsGlobal = (indentlevel == 0);
|
|
const bool printDebug = mSettings->debugwarnings;
|
|
|
|
if (mErrorLogger && !list.getFiles().empty())
|
|
mErrorLogger->reportProgress(list.getFiles()[0], "Tokenize (simplifyKnownVariables)", tok3->progressValue());
|
|
|
|
if (isMaxTime())
|
|
return false;
|
|
|
|
bool ret = false;
|
|
|
|
Token* bailOutFromLoop = nullptr;
|
|
int indentlevel3 = indentlevel;
|
|
bool ret3 = false;
|
|
for (; tok3; tok3 = tok3->next()) {
|
|
if (tok3->str() == "{") {
|
|
++indentlevel3;
|
|
} else if (tok3->str() == "}") {
|
|
--indentlevel3;
|
|
if (indentlevel3 < indentlevel) {
|
|
if (Token::Match((*tok2)->tokAt(-7), "%type% * %name% ; %name% = & %name% ;") &&
|
|
(*tok2)->strAt(-5) == (*tok2)->strAt(-3)) {
|
|
(*tok2) = (*tok2)->tokAt(-4);
|
|
Token::eraseTokens((*tok2), (*tok2)->tokAt(6));
|
|
}
|
|
break;
|
|
}
|
|
}
|
|
|
|
// Stop if there is a pointer alias and a shadow variable is
|
|
// declared in an inner scope (#3058)
|
|
if (valueIsPointer && tok3->varId() > 0 &&
|
|
tok3->previous() && (tok3->previous()->isName() || tok3->previous()->str() == "*") &&
|
|
valueToken->str() == "&" &&
|
|
valueToken->next() &&
|
|
valueToken->next()->isName() &&
|
|
tok3->str() == valueToken->next()->str() &&
|
|
tok3->varId() > valueToken->next()->varId()) {
|
|
// more checking if this is a variable declaration
|
|
bool decl = true;
|
|
for (const Token *tok4 = tok3->previous(); tok4; tok4 = tok4->previous()) {
|
|
if (Token::Match(tok4, "[;{}]"))
|
|
break;
|
|
|
|
else if (tok4->isName()) {
|
|
if (tok4->varId() > 0) {
|
|
decl = false;
|
|
break;
|
|
}
|
|
}
|
|
|
|
else if (!Token::Match(tok4, "[&*]")) {
|
|
decl = false;
|
|
break;
|
|
}
|
|
}
|
|
if (decl)
|
|
break;
|
|
}
|
|
|
|
// Stop if label is found
|
|
if (Token::Match(tok3, "; %type% : ;"))
|
|
break;
|
|
|
|
// Stop if break/continue is found ..
|
|
if (Token::Match(tok3, "break|continue"))
|
|
break;
|
|
if ((indentlevel3 > 1 || !Token::simpleMatch(Token::findsimplematch(tok3,";"), "; }")) && tok3->str() == "return")
|
|
ret3 = true;
|
|
if (ret3 && tok3->str() == ";")
|
|
break;
|
|
|
|
if (pointeralias && Token::Match(tok3, ("!!= " + value).c_str()))
|
|
break;
|
|
|
|
// Stop if a loop is found
|
|
if (pointeralias && Token::Match(tok3, "do|for|while"))
|
|
break;
|
|
|
|
// Stop if unknown function call is seen and the variable is global: it might be
|
|
// changed by the function call
|
|
if (varIsGlobal && tok3->str() == ")" && tok3->link() &&
|
|
Token::Match(tok3->link()->tokAt(-2), "[;{}] %name% (") &&
|
|
!Token::Match(tok3->link()->previous(), "if|for|while|switch|BOOST_FOREACH"))
|
|
break;
|
|
|
|
// Stop if something like 'while (--var)' is found
|
|
if (Token::Match(tok3, "for|while|do")) {
|
|
const Token *endpar = tok3->next()->link();
|
|
if (Token::simpleMatch(endpar, ") {"))
|
|
endpar = endpar->next()->link();
|
|
bool bailout = false;
|
|
for (const Token *tok4 = tok3; tok4 && tok4 != endpar; tok4 = tok4->next()) {
|
|
if (Token::Match(tok4, "++|-- %varid%", varid) ||
|
|
Token::Match(tok4, "%varid% ++|--|=", varid)) {
|
|
bailout = true;
|
|
break;
|
|
}
|
|
}
|
|
if (bailout)
|
|
break;
|
|
}
|
|
|
|
if (bailOutFromLoop) {
|
|
// This could be a loop, skip it, but only if it doesn't contain
|
|
// the variable we are checking for. If it contains the variable
|
|
// we will bail out.
|
|
if (tok3->varId() == varid) {
|
|
// Continue
|
|
//tok2 = bailOutFromLoop;
|
|
break;
|
|
} else if (tok3 == bailOutFromLoop) {
|
|
// We have skipped the loop
|
|
bailOutFromLoop = nullptr;
|
|
continue;
|
|
}
|
|
|
|
continue;
|
|
} else if (tok3->str() == "{" && tok3->previous()->str() == ")") {
|
|
// There is a possible loop after the assignment. Try to skip it.
|
|
if (tok3->previous()->link() &&
|
|
tok3->previous()->link()->strAt(-1) != "if")
|
|
bailOutFromLoop = tok3->link();
|
|
continue;
|
|
}
|
|
|
|
// Variable used in realloc (see Ticket #1649)
|
|
if (Token::Match(tok3, "%name% = realloc ( %name% ,") &&
|
|
tok3->varId() == varid &&
|
|
tok3->tokAt(4)->varId() == varid) {
|
|
tok3->tokAt(4)->str(value);
|
|
ret = true;
|
|
}
|
|
|
|
// condition "(|&&|%OROR% %varid% )|&&|%OROR%|;
|
|
if (!Token::Match(tok3->previous(), "( %name% )") &&
|
|
Token::Match(tok3->previous(), "&&|(|%oror% %varid% &&|%oror%|)|;", varid)) {
|
|
tok3->str(value);
|
|
tok3->varId(valueVarId);
|
|
ret = true;
|
|
}
|
|
|
|
// parameter in function call..
|
|
if (tok3->varId() == varid && Token::Match(tok3->previous(), "[(,] %name% [,)]")) {
|
|
// If the parameter is passed by value then simplify it
|
|
if (isFunctionParameterPassedByValue(tok3)) {
|
|
tok3->str(value);
|
|
tok3->varId(valueVarId);
|
|
ret = true;
|
|
}
|
|
}
|
|
|
|
// Variable is used somehow in a non-defined pattern => bail out
|
|
if (tok3->varId() == varid) {
|
|
// This is a really generic bailout so let's try to avoid this.
|
|
// There might be lots of false negatives.
|
|
if (printDebug) {
|
|
// FIXME: Fix all the debug warnings for values and then
|
|
// remove this bailout
|
|
if (pointeralias)
|
|
break;
|
|
|
|
// suppress debug-warning when calling member function
|
|
if (Token::Match(tok3->next(), ". %name% ("))
|
|
break;
|
|
|
|
// suppress debug-warning when assignment
|
|
if (tok3->strAt(1) == "=")
|
|
break;
|
|
|
|
// taking address of variable..
|
|
if (Token::Match(tok3->tokAt(-2), "return|= & %name% ;"))
|
|
break;
|
|
|
|
// parameter in function call..
|
|
if (Token::Match(tok3->tokAt(-2), "%name% ( %name% ,|)") ||
|
|
Token::Match(tok3->previous(), ", %name% ,|)"))
|
|
break;
|
|
|
|
// conditional increment
|
|
if (Token::Match(tok3->tokAt(-3), ") { ++|--") ||
|
|
Token::Match(tok3->tokAt(-2), ") { %name% ++|--"))
|
|
break;
|
|
|
|
reportError(tok3, Severity::debug, "debug",
|
|
"simplifyKnownVariables: bailing out (variable="+tok3->str()+", value="+value+")");
|
|
}
|
|
|
|
break;
|
|
}
|
|
|
|
// Using the variable in condition..
|
|
if (Token::Match(tok3->previous(), ("if ( " + structname + " %varid% %cop%|)").c_str(), varid) ||
|
|
Token::Match(tok3, ("( " + structname + " %varid% %comp%").c_str(), varid) ||
|
|
Token::Match(tok3, ("%comp%|!|= " + structname + " %varid% %cop%|)|;").c_str(), varid) ||
|
|
Token::Match(tok3->previous(), "strlen|free ( %varid% )", varid)) {
|
|
if (value[0] == '\"' && tok3->previous()->str() != "strlen") {
|
|
// bail out if value is a string unless if it's just given
|
|
// as parameter to strlen
|
|
break;
|
|
}
|
|
if (!structname.empty()) {
|
|
tok3->deleteNext(2);
|
|
}
|
|
if (Token::Match(valueToken, "& %name% ;")) {
|
|
tok3->insertToken("&");
|
|
tok3 = tok3->next();
|
|
}
|
|
tok3 = tok3->next();
|
|
tok3->str(value);
|
|
tok3->varId(valueVarId);
|
|
ret = true;
|
|
}
|
|
|
|
// pointer alias used in condition..
|
|
if (Token::Match(valueToken,"& %name% ;") && Token::Match(tok3, ("( * " + structname + " %varid% %cop%").c_str(), varid)) {
|
|
tok3->deleteNext();
|
|
if (!structname.empty())
|
|
tok3->deleteNext(2);
|
|
tok3 = tok3->next();
|
|
tok3->str(value);
|
|
tok3->varId(valueVarId);
|
|
ret = true;
|
|
}
|
|
|
|
// Delete pointer alias
|
|
if (isCPP() && pointeralias && (tok3->str() == "delete") && tok3->next() &&
|
|
(Token::Match(tok3->next(), "%varid% ;", varid) ||
|
|
Token::Match(tok3->next(), "[ ] %varid%", varid))) {
|
|
tok3 = (tok3->next()->str() == "[") ? tok3->tokAt(3) : tok3->next();
|
|
tok3->str(value);
|
|
tok3->varId(valueVarId);
|
|
ret = true;
|
|
}
|
|
|
|
// Variable is used in function call..
|
|
if (Token::Match(tok3, ("%name% ( " + structname + " %varid% ,").c_str(), varid)) {
|
|
static const char * const functionName[] = {
|
|
// always simplify
|
|
"strcmp", "strdup",
|
|
// don't simplify buffer value
|
|
"memcmp","memcpy","memmove","memset","strcpy","strncmp","strncpy"
|
|
};
|
|
for (int i = 0; i < (sizeof(functionName) / sizeof(*functionName)); ++i) {
|
|
if (valueVarId == 0U && i >= 2)
|
|
break;
|
|
if (tok3->str() == functionName[i]) {
|
|
Token *par1 = tok3->tokAt(2);
|
|
if (!structname.empty()) {
|
|
par1->deleteNext();
|
|
par1->deleteThis();
|
|
}
|
|
par1->str(value);
|
|
par1->varId(valueVarId);
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
// Variable is used as 2nd parameter in function call..
|
|
if (Token::Match(tok3, ("%name% ( %any% , " + structname + " %varid% ,|)").c_str(), varid)) {
|
|
static const char * const functionName[] = {
|
|
// always simplify
|
|
"strcmp","strcpy","strncmp","strncpy",
|
|
// don't simplify buffer value
|
|
"memcmp","memcpy","memmove"
|
|
};
|
|
for (int i = 0; i < (sizeof(functionName) / sizeof(*functionName)); ++i) {
|
|
if (valueVarId == 0U && i >= 4)
|
|
break;
|
|
if (tok3->str() == functionName[i]) {
|
|
Token *par = tok3->tokAt(4);
|
|
if (!structname.empty()) {
|
|
par->deleteNext();
|
|
par->deleteThis();
|
|
}
|
|
par->str(value);
|
|
par->varId(valueVarId);
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
// array usage
|
|
if (value[0] != '\"' && Token::Match(tok3, ("[(,] " + structname + " %varid% [|%cop%").c_str(), varid)) {
|
|
if (!structname.empty()) {
|
|
tok3->deleteNext(2);
|
|
}
|
|
tok3 = tok3->next();
|
|
tok3->str(value);
|
|
tok3->varId(valueVarId);
|
|
ret = true;
|
|
}
|
|
|
|
// The >> operator is sometimes used to assign a variable in C++
|
|
if (isCPP() && Token::Match(tok3, (">> " + structname + " %varid%").c_str(), varid)) {
|
|
// bailout for such code: ; std :: cin >> i ;
|
|
const Token *prev = tok3->previous();
|
|
while (prev && prev->str() != "return" && Token::Match(prev, "%name%|::|*"))
|
|
prev = prev->previous();
|
|
if (Token::Match(prev, ";|{|}|>>"))
|
|
break;
|
|
}
|
|
|
|
// Variable is used in calculation..
|
|
if (((tok3->previous()->varId() > 0) && Token::Match(tok3, ("& " + structname + " %varid%").c_str(), varid)) ||
|
|
(Token::Match(tok3, ("[=+-*/%^|[] " + structname + " %varid% [=?+-*/%^|;])]").c_str(), varid) && !Token::Match(tok3, ("= " + structname + " %name% =").c_str())) ||
|
|
Token::Match(tok3, ("[(=+-*/%^|[] " + structname + " %varid% <<|>>").c_str(), varid) ||
|
|
Token::Match(tok3, ("<<|>> " + structname + " %varid% %cop%|;|]|)").c_str(), varid) ||
|
|
Token::Match(tok3->previous(), ("[=+-*/%^|[] ( " + structname + " %varid% !!=").c_str(), varid)) {
|
|
if (value[0] == '\"')
|
|
break;
|
|
if (!structname.empty()) {
|
|
tok3->deleteNext(2);
|
|
ret = true;
|
|
}
|
|
tok3 = tok3->next();
|
|
if (tok3->str() != value)
|
|
ret = true;
|
|
tok3->str(value);
|
|
tok3->varId(valueVarId);
|
|
if (tok3->previous()->str() == "*" && (valueIsPointer || Token::Match(valueToken, "& %name% ;"))) {
|
|
tok3 = tok3->previous();
|
|
tok3->deleteThis();
|
|
ret = true;
|
|
} else if (Token::Match(valueToken, "& %name% ;"))
|
|
tok3->insertToken("&", emptyString, true);
|
|
}
|
|
|
|
if (Token::simpleMatch(tok3, "= {")) {
|
|
const Token* const end4 = tok3->linkAt(1);
|
|
for (const Token *tok4 = tok3; tok4 != end4; tok4 = tok4->next()) {
|
|
if (Token::Match(tok4, "{|, %varid% ,|}", varid)) {
|
|
tok4->next()->str(value);
|
|
tok4->next()->varId(valueVarId);
|
|
ret = true;
|
|
}
|
|
}
|
|
}
|
|
|
|
// Using the variable in for-condition..
|
|
if (Token::simpleMatch(tok3, "for (")) {
|
|
for (Token *tok4 = tok3->tokAt(2); tok4; tok4 = tok4->next()) {
|
|
if (Token::Match(tok4, "(|)"))
|
|
break;
|
|
|
|
// Replace variable used in condition..
|
|
if (Token::Match(tok4, "; %name% <|<=|!= %name% ; ++| %name% ++| )")) {
|
|
const Token *inctok = tok4->tokAt(5);
|
|
if (inctok->str() == "++")
|
|
inctok = inctok->next();
|
|
if (inctok->varId() == varid)
|
|
break;
|
|
|
|
if (tok4->next()->varId() == varid) {
|
|
tok4->next()->str(value);
|
|
tok4->next()->varId(valueVarId);
|
|
ret = true;
|
|
}
|
|
if (tok4->tokAt(3)->varId() == varid) {
|
|
tok4->tokAt(3)->str(value);
|
|
tok4->tokAt(3)->varId(valueVarId);
|
|
ret = true;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if (indentlevel == indentlevel3 && Token::Match(tok3->next(), "%varid% ++|--", varid) && MathLib::isInt(value)) {
|
|
const std::string op(tok3->strAt(2));
|
|
if (Token::Match(tok3, "[{};] %any% %any% ;")) {
|
|
tok3->deleteNext(3);
|
|
} else {
|
|
tok3 = tok3->next();
|
|
tok3->str(value);
|
|
tok3->varId(valueVarId);
|
|
tok3->deleteNext();
|
|
}
|
|
value = MathLib::incdec(value, op);
|
|
if (!Token::simpleMatch((*tok2)->tokAt(-2), "for (")) {
|
|
(*tok2)->tokAt(2)->str(value);
|
|
(*tok2)->tokAt(2)->varId(valueVarId);
|
|
}
|
|
ret = true;
|
|
}
|
|
|
|
if (indentlevel == indentlevel3 && Token::Match(tok3->next(), "++|-- %varid%", varid) && MathLib::isInt(value) &&
|
|
!Token::Match(tok3->tokAt(3), "[.[]")) {
|
|
value = MathLib::incdec(value, tok3->next()->str());
|
|
(*tok2)->tokAt(2)->str(value);
|
|
(*tok2)->tokAt(2)->varId(valueVarId);
|
|
if (Token::Match(tok3, "[;{}] %any% %any% ;")) {
|
|
tok3->deleteNext(3);
|
|
} else {
|
|
tok3->deleteNext();
|
|
tok3->next()->str(value);
|
|
tok3->next()->varId(valueVarId);
|
|
}
|
|
tok3 = tok3->next();
|
|
ret = true;
|
|
}
|
|
|
|
// return variable..
|
|
if (Token::Match(tok3, "return %varid% %any%", varid) &&
|
|
valueToken->str() != "&" &&
|
|
(tok3->tokAt(2)->isExtendedOp() || tok3->strAt(2) == ";") &&
|
|
value[0] != '\"') {
|
|
tok3->next()->str(value);
|
|
tok3->next()->varId(valueVarId);
|
|
}
|
|
|
|
else if (pointeralias && Token::Match(tok3, "return * %varid% ;", varid) && value[0] != '\"') {
|
|
tok3->deleteNext();
|
|
tok3->next()->str(value);
|
|
tok3->next()->varId(valueVarId);
|
|
}
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
|
|
void Tokenizer::elseif()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::simpleMatch(tok, "else if"))
|
|
continue;
|
|
|
|
for (Token *tok2 = tok; tok2; tok2 = tok2->next()) {
|
|
if (Token::Match(tok2, "(|{|["))
|
|
tok2 = tok2->link();
|
|
|
|
if (Token::Match(tok2, "}|;")) {
|
|
if (tok2->next() && tok2->next()->str() != "else") {
|
|
tok->insertToken("{");
|
|
tok2->insertToken("}");
|
|
Token::createMutualLinks(tok->next(), tok2->next());
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
void Tokenizer::simplifyIfSwitchForInit()
|
|
{
|
|
if (!isCPP() || mSettings->standards.cpp < Standards::CPP17)
|
|
return;
|
|
|
|
const bool forInit = (mSettings->standards.cpp >= Standards::CPP20);
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "if|switch|for ("))
|
|
continue;
|
|
|
|
Token *semicolon = tok->tokAt(2);
|
|
while (!Token::Match(semicolon, "[;)]")) {
|
|
if (Token::Match(semicolon, "(|{|[") && semicolon->link())
|
|
semicolon = semicolon->link();
|
|
semicolon = semicolon->next();
|
|
}
|
|
if (semicolon->str() != ";")
|
|
continue;
|
|
|
|
if (tok->str() == "for") {
|
|
if (!forInit)
|
|
continue;
|
|
|
|
// Is it a for range..
|
|
const Token *tok2 = semicolon->next();
|
|
bool rangeFor = false;
|
|
while (!Token::Match(tok2, "[;)]")) {
|
|
if (tok2->str() == "(")
|
|
tok2 = tok2->link();
|
|
else if (!rangeFor && tok2->str() == "?")
|
|
break;
|
|
else if (tok2->str() == ":")
|
|
rangeFor = true;
|
|
tok2 = tok2->next();
|
|
}
|
|
if (!rangeFor || tok2->str() != ")")
|
|
continue;
|
|
}
|
|
|
|
Token *endpar = tok->linkAt(1);
|
|
if (!Token::simpleMatch(endpar, ") {"))
|
|
continue;
|
|
|
|
Token *endscope = endpar->linkAt(1);
|
|
if (Token::simpleMatch(endscope, "} else {"))
|
|
endscope = endscope->linkAt(2);
|
|
|
|
// Simplify, the initialization expression is broken out..
|
|
semicolon->insertToken(tok->str());
|
|
semicolon->next()->insertToken("(");
|
|
Token::createMutualLinks(semicolon->next()->next(), endpar);
|
|
tok->deleteNext();
|
|
tok->str("{");
|
|
endscope->insertToken("}");
|
|
Token::createMutualLinks(tok, endscope->next());
|
|
tok->isSimplifiedScope(true);
|
|
}
|
|
}
|
|
|
|
|
|
bool Tokenizer::simplifyRedundantParentheses()
|
|
{
|
|
bool ret = false;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() != "(")
|
|
continue;
|
|
|
|
if (isCPP() && Token::simpleMatch(tok->previous(), "} (")) {
|
|
const Token* plp = tok->previous()->link()->previous();
|
|
if (Token::Match(plp, "%name%|>|] {") || (Token::simpleMatch(plp, ")") && Token::simpleMatch(plp->link()->previous(), "]")))
|
|
continue;
|
|
}
|
|
|
|
if (Token::simpleMatch(tok, "( {"))
|
|
continue;
|
|
|
|
if (Token::Match(tok->link(), ") %num%")) {
|
|
tok = tok->link();
|
|
continue;
|
|
}
|
|
|
|
// Do not simplify if there is comma inside parentheses..
|
|
if (Token::Match(tok->previous(), "%op% (") || Token::Match(tok->link(), ") %op%")) {
|
|
bool innerComma = false;
|
|
for (const Token *inner = tok->link()->previous(); inner != tok; inner = inner->previous()) {
|
|
if (inner->str() == ")")
|
|
inner = inner->link();
|
|
if (inner->str() == ",") {
|
|
innerComma = true;
|
|
break;
|
|
}
|
|
}
|
|
if (innerComma)
|
|
continue;
|
|
}
|
|
|
|
// !!operator = ( x ) ;
|
|
if (tok->strAt(-2) != "operator" &&
|
|
tok->previous() && tok->previous()->str() == "=" &&
|
|
tok->next() && tok->next()->str() != "{" &&
|
|
Token::simpleMatch(tok->link(), ") ;")) {
|
|
tok->link()->deleteThis();
|
|
tok->deleteThis();
|
|
continue;
|
|
}
|
|
|
|
while (Token::simpleMatch(tok, "( (") &&
|
|
tok->link() && tok->link()->previous() == tok->next()->link()) {
|
|
// We have "(( *something* ))", remove the inner
|
|
// parentheses
|
|
tok->deleteNext();
|
|
tok->link()->tokAt(-2)->deleteNext();
|
|
ret = true;
|
|
}
|
|
|
|
if (isCPP() && Token::Match(tok->tokAt(-2), "[;{}=(] new (") && Token::Match(tok->link(), ") [;,{}[]")) {
|
|
// Remove the parentheses in "new (type)" constructs
|
|
tok->link()->deleteThis();
|
|
tok->deleteThis();
|
|
ret = true;
|
|
}
|
|
|
|
if (Token::Match(tok->previous(), "! ( %name% )")) {
|
|
// Remove the parentheses
|
|
tok->deleteThis();
|
|
tok->deleteNext();
|
|
ret = true;
|
|
}
|
|
|
|
if (Token::Match(tok->previous(), "[(,;{}] ( %name% ) .")) {
|
|
// Remove the parentheses
|
|
tok->deleteThis();
|
|
tok->deleteNext();
|
|
ret = true;
|
|
}
|
|
|
|
if (Token::Match(tok->previous(), "[(,;{}] ( %name% (") &&
|
|
tok->link()->previous() == tok->linkAt(2)) {
|
|
// We have "( func ( *something* ))", remove the outer
|
|
// parentheses
|
|
tok->link()->deleteThis();
|
|
tok->deleteThis();
|
|
ret = true;
|
|
}
|
|
|
|
if (Token::Match(tok->previous(), "[,;{}] ( delete [| ]| %name% ) ;")) {
|
|
// We have "( delete [| ]| var )", remove the outer
|
|
// parentheses
|
|
tok->link()->deleteThis();
|
|
tok->deleteThis();
|
|
ret = true;
|
|
}
|
|
|
|
if (!Token::simpleMatch(tok->tokAt(-2), "operator delete") &&
|
|
Token::Match(tok->previous(), "delete|; (") &&
|
|
(tok->previous()->str() != "delete" || tok->next()->varId() > 0) &&
|
|
Token::Match(tok->link(), ") ;|,")) {
|
|
tok->link()->deleteThis();
|
|
tok->deleteThis();
|
|
ret = true;
|
|
}
|
|
|
|
if (Token::Match(tok->previous(), "[(!*;{}] ( %name% )") &&
|
|
(tok->next()->varId() != 0 || Token::Match(tok->tokAt(3), "[+-/=]")) && !tok->next()->isStandardType()) {
|
|
// We have "( var )", remove the parentheses
|
|
tok->deleteThis();
|
|
tok->deleteNext();
|
|
ret = true;
|
|
}
|
|
|
|
while (Token::Match(tok->previous(), "[;{}[(,!*] ( %name% .")) {
|
|
Token *tok2 = tok->tokAt(2);
|
|
while (Token::Match(tok2, ". %name%")) {
|
|
tok2 = tok2->tokAt(2);
|
|
}
|
|
if (tok2 != tok->link())
|
|
break;
|
|
// We have "( var . var . ... . var )", remove the parentheses
|
|
tok = tok->previous();
|
|
tok->deleteNext();
|
|
tok2->deleteThis();
|
|
ret = true;
|
|
}
|
|
|
|
if (Token::simpleMatch(tok->previous(), "? (") && Token::simpleMatch(tok->link(), ") :")) {
|
|
const Token *tok2 = tok->next();
|
|
while (tok2 && (Token::Match(tok2,"%bool%|%num%|%name%") || tok2->isArithmeticalOp()))
|
|
tok2 = tok2->next();
|
|
if (tok2 && tok2->str() == ")") {
|
|
tok->link()->deleteThis();
|
|
tok->deleteThis();
|
|
ret = true;
|
|
continue;
|
|
}
|
|
}
|
|
|
|
while (Token::Match(tok->previous(), "[{([,] ( !!{") &&
|
|
Token::Match(tok->link(), ") [;,])]") &&
|
|
!Token::simpleMatch(tok->tokAt(-2), "operator ,") && // Ticket #5709
|
|
!Token::findsimplematch(tok, ",", tok->link())) {
|
|
// We have "( ... )", remove the parentheses
|
|
tok->link()->deleteThis();
|
|
tok->deleteThis();
|
|
ret = true;
|
|
}
|
|
|
|
if (Token::simpleMatch(tok->previous(), ", (") &&
|
|
Token::simpleMatch(tok->link(), ") =")) {
|
|
tok->link()->deleteThis();
|
|
tok->deleteThis();
|
|
ret = true;
|
|
}
|
|
|
|
// Simplify "!!operator !!%name%|)|]|>|>> ( %num%|%bool% ) %op%|;|,|)"
|
|
if (Token::Match(tok, "( %bool%|%num% ) %cop%|;|,|)") &&
|
|
tok->strAt(-2) != "operator" &&
|
|
tok->previous() &&
|
|
!Token::Match(tok->previous(), "%name%|)|]") &&
|
|
(!(isCPP() && Token::Match(tok->previous(),">|>>")))) {
|
|
tok->link()->deleteThis();
|
|
tok->deleteThis();
|
|
ret = true;
|
|
}
|
|
|
|
if (Token::Match(tok->previous(), "*|& ( %name% )")) {
|
|
// We may have a variable declaration looking like "type_name *(var_name)"
|
|
Token *tok2 = tok->tokAt(-2);
|
|
while (Token::Match(tok2, "%type%|static|const|extern") && tok2->str() != "operator") {
|
|
tok2 = tok2->previous();
|
|
}
|
|
if (tok2 && !Token::Match(tok2, "[;,{]")) {
|
|
// Not a variable declaration
|
|
} else {
|
|
tok->deleteThis();
|
|
tok->deleteNext();
|
|
}
|
|
}
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
void Tokenizer::simplifyTypeIntrinsics()
|
|
{
|
|
static const std::unordered_map<std::string, std::string> intrinsics = {
|
|
{ "__has_nothrow_assign", "has_nothrow_assign" },
|
|
{ "__has_nothrow_constructor", "has_nothrow_constructor" },
|
|
{ "__has_nothrow_copy", "has_nothrow_copy" },
|
|
{ "__has_trivial_assign", "has_trivial_assign" },
|
|
{ "__has_trivial_constructor", "has_trivial_constructor" },
|
|
{ "__has_trivial_copy", "has_trivial_copy" },
|
|
{ "__has_trivial_destructor", "has_trivial_destructor" },
|
|
{ "__has_virtual_destructor", "has_virtual_destructor" },
|
|
{ "__is_abstract", "is_abstract" },
|
|
{ "__is_aggregate", "is_aggregate" },
|
|
{ "__is_assignable", "is_assignable" },
|
|
{ "__is_base_of", "is_base_of" },
|
|
{ "__is_class", "is_class" },
|
|
{ "__is_constructible", "is_constructible" },
|
|
{ "__is_convertible_to", "is_convertible_to" },
|
|
{ "__is_destructible", "is_destructible" },
|
|
{ "__is_empty", "is_empty" },
|
|
{ "__is_enum", "is_enum" },
|
|
{ "__is_final", "is_final" },
|
|
{ "__is_nothrow_assignable", "is_nothrow_assignable" },
|
|
{ "__is_nothrow_constructible", "is_nothrow_constructible" },
|
|
{ "__is_nothrow_destructible", "is_nothrow_destructible" },
|
|
{ "__is_pod", "is_pod" },
|
|
{ "__is_polymorphic", "is_polymorphic" },
|
|
{ "__is_trivially_assignable", "is_trivially_assignable" },
|
|
{ "__is_trivially_constructible", "is_trivially_constructible" },
|
|
{ "__is_union", "is_union" },
|
|
};
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "%name% ("))
|
|
continue;
|
|
auto p = intrinsics.find(tok->str());
|
|
if (p == intrinsics.end())
|
|
continue;
|
|
Token * end = tok->next()->link();
|
|
Token * prev = tok->previous();
|
|
tok->str(p->second);
|
|
prev->insertToken("::");
|
|
prev->insertToken("std");
|
|
tok->next()->str("<");
|
|
end->str(">");
|
|
end->insertToken("}");
|
|
end->insertToken("{");
|
|
Token::createMutualLinks(end->tokAt(1), end->tokAt(2));
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyCharAt()
|
|
{
|
|
// Replace "string"[0] with 's'
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "%str% [ %num% ]")) {
|
|
const MathLib::bigint index = MathLib::toLongNumber(tok->strAt(2));
|
|
// Check within range
|
|
if (index >= 0 && index <= Token::getStrLength(tok)) {
|
|
tok->str("'" + Token::getCharAt(tok, index) + "'");
|
|
tok->deleteNext(3);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyReference()
|
|
{
|
|
if (isC())
|
|
return;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
// starting executable scope..
|
|
Token *start = const_cast<Token *>(startOfExecutableScope(tok));
|
|
if (start) {
|
|
tok = start;
|
|
// replace references in this scope..
|
|
Token * const end = tok->link();
|
|
for (Token *tok2 = tok; tok2 && tok2 != end; tok2 = tok2->next()) {
|
|
// found a reference..
|
|
if (Token::Match(tok2, "[;{}] %type% & %name% (|= %name% )| ;")) {
|
|
const int refId = tok2->tokAt(3)->varId();
|
|
if (!refId)
|
|
continue;
|
|
|
|
// replace reference in the code..
|
|
for (Token *tok3 = tok2->tokAt(7); tok3 && tok3 != end; tok3 = tok3->next()) {
|
|
if (tok3->varId() == refId) {
|
|
tok3->str(tok2->strAt(5));
|
|
tok3->varId(tok2->tokAt(5)->varId());
|
|
}
|
|
}
|
|
|
|
tok2->deleteNext(6+(tok2->strAt(6)==")" ? 1 : 0));
|
|
}
|
|
}
|
|
tok = end;
|
|
}
|
|
}
|
|
}
|
|
|
|
bool Tokenizer::simplifyCalculations()
|
|
{
|
|
return mTemplateSimplifier->simplifyCalculations(nullptr, nullptr, false);
|
|
}
|
|
|
|
void Tokenizer::simplifyOffsetPointerDereference()
|
|
{
|
|
// Replace "*(str + num)" => "str[num]" and
|
|
// Replace "*(str - num)" => "str[-num]"
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!tok->isName() && !tok->isLiteral()
|
|
&& !Token::Match(tok, "]|)|++|--")
|
|
&& Token::Match(tok->next(), "* ( %name% +|- %num%|%name% )")) {
|
|
|
|
// remove '* ('
|
|
tok->deleteNext(2);
|
|
|
|
// '+'->'['
|
|
tok = tok->tokAt(2);
|
|
Token* const openBraceTok = tok;
|
|
const bool isNegativeIndex = (tok->str() == "-");
|
|
tok->str("[");
|
|
|
|
// Insert a "-" in front of the number or variable
|
|
if (isNegativeIndex) {
|
|
if (tok->next()->isName()) {
|
|
tok->insertToken("-");
|
|
tok = tok->next();
|
|
} else
|
|
tok->next()->str(std::string("-") + tok->next()->str());
|
|
}
|
|
|
|
tok = tok->tokAt(2);
|
|
tok->str("]");
|
|
Token::createMutualLinks(openBraceTok, tok);
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyOffsetPointerReference()
|
|
{
|
|
std::set<int> pod;
|
|
for (const Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->isStandardType()) {
|
|
tok = tok->next();
|
|
while (tok && (tok->str() == "*" || tok->isName())) {
|
|
if (tok->varId() > 0) {
|
|
pod.insert(tok->varId());
|
|
break;
|
|
}
|
|
tok = tok->next();
|
|
}
|
|
if (!tok)
|
|
break;
|
|
}
|
|
}
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "%num%|%name%|]|)") &&
|
|
(Token::Match(tok->next(), "& %name% [ %num%|%name% ] !!["))) {
|
|
tok = tok->next();
|
|
|
|
if (tok->next()->varId()) {
|
|
if (pod.find(tok->next()->varId()) == pod.end()) {
|
|
tok = tok->tokAt(5);
|
|
if (!tok)
|
|
syntaxError(tok);
|
|
continue;
|
|
}
|
|
}
|
|
|
|
// '&' => '('
|
|
tok->str("(");
|
|
|
|
tok = tok->next();
|
|
// '[' => '+'
|
|
tok->deleteNext();
|
|
tok->insertToken("+");
|
|
|
|
tok = tok->tokAt(3);
|
|
//remove ']'
|
|
tok->str(")");
|
|
Token::createMutualLinks(tok->tokAt(-4), tok);
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyNestedStrcat()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "[;{}] strcat ( strcat (")) {
|
|
continue;
|
|
}
|
|
|
|
// find inner strcat call
|
|
Token *tok2 = tok->tokAt(3);
|
|
while (Token::simpleMatch(tok2, "strcat ( strcat"))
|
|
tok2 = tok2->tokAt(2);
|
|
|
|
if (tok2->strAt(3) != ",")
|
|
continue;
|
|
|
|
// If we have this code:
|
|
// strcat(strcat(dst, foo), bar);
|
|
// We move this part of code before all strcat() calls: strcat(dst, foo)
|
|
// And place "dst" token where the code was.
|
|
Token *prevTok = tok2->previous();
|
|
|
|
// Move tokens to new place
|
|
Token::move(tok2, tok2->next()->link(), tok);
|
|
tok = tok2->next()->link();
|
|
|
|
// Insert the "dst" token
|
|
prevTok->insertToken(tok2->strAt(2));
|
|
prevTok->next()->varId(tok2->tokAt(2)->varId());
|
|
|
|
// Insert semicolon after the moved strcat()
|
|
tok->insertToken(";");
|
|
}
|
|
}
|
|
|
|
static const std::set<std::string> stdFunctionsPresentInC = {
|
|
"strcat",
|
|
"strcpy",
|
|
"strncat",
|
|
"strncpy",
|
|
"free",
|
|
"malloc",
|
|
"strdup"
|
|
};
|
|
|
|
void Tokenizer::simplifyStd()
|
|
{
|
|
if (isC())
|
|
return;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() != "std")
|
|
continue;
|
|
|
|
if (Token::Match(tok->previous(), "[(,{};] std :: %name% (") &&
|
|
stdFunctionsPresentInC.find(tok->strAt(2)) != stdFunctionsPresentInC.end()) {
|
|
tok->deleteNext();
|
|
tok->deleteThis();
|
|
}
|
|
}
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
// Helper functions for handling the tokens list
|
|
//---------------------------------------------------------------------------
|
|
|
|
//---------------------------------------------------------------------------
|
|
|
|
bool Tokenizer::isScopeNoReturn(const Token *endScopeToken, bool *unknown) const
|
|
{
|
|
std::string unknownFunc;
|
|
const bool ret = mSettings->library.isScopeNoReturn(endScopeToken,&unknownFunc);
|
|
if (!unknownFunc.empty() && mSettings->summaryReturn.find(unknownFunc) != mSettings->summaryReturn.end()) {
|
|
return false;
|
|
}
|
|
if (unknown)
|
|
*unknown = !unknownFunc.empty();
|
|
if (!unknownFunc.empty() && mSettings->checkLibrary && mSettings->severity.isEnabled(Severity::information)) {
|
|
// Is function global?
|
|
bool globalFunction = true;
|
|
if (Token::simpleMatch(endScopeToken->tokAt(-2), ") ; }")) {
|
|
const Token * const ftok = endScopeToken->linkAt(-2)->previous();
|
|
if (ftok &&
|
|
ftok->isName() &&
|
|
ftok->function() &&
|
|
ftok->function()->nestedIn &&
|
|
ftok->function()->nestedIn->type != Scope::eGlobal) {
|
|
globalFunction = false;
|
|
}
|
|
}
|
|
|
|
// don't warn for nonglobal functions (class methods, functions hidden in namespaces) since they can't be configured yet
|
|
// FIXME: when methods and namespaces can be configured properly, remove the "globalFunction" check
|
|
if (globalFunction) {
|
|
reportError(endScopeToken->previous(),
|
|
Severity::information,
|
|
"checkLibraryNoReturn",
|
|
"--check-library: Function " + unknownFunc + "() should have <noreturn> configuration");
|
|
}
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
|
|
bool Tokenizer::isFunctionParameterPassedByValue(const Token *fpar) const
|
|
{
|
|
// TODO: If symbol database is available, use it.
|
|
const Token *ftok;
|
|
|
|
// Look at function call, what parameter number is it?
|
|
int parameter = 1;
|
|
for (ftok = fpar->previous(); ftok; ftok = ftok->previous()) {
|
|
if (ftok->str() == "(")
|
|
break;
|
|
else if (ftok->str() == ")")
|
|
ftok = ftok->link();
|
|
else if (ftok->str() == ",")
|
|
++parameter;
|
|
else if (Token::Match(ftok, "[;{}]"))
|
|
break;
|
|
}
|
|
|
|
// Is this a function call?
|
|
if (ftok && Token::Match(ftok->tokAt(-2), "[;{}=] %name% (")) {
|
|
const std::string& functionName(ftok->previous()->str());
|
|
|
|
if (functionName == "return")
|
|
return true;
|
|
|
|
// Locate function declaration..
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (tok->str() == "{")
|
|
tok = tok->link();
|
|
else if (Token::Match(tok, "%type% (") && tok->str() == functionName) {
|
|
// Goto parameter
|
|
tok = tok->tokAt(2);
|
|
int par = 1;
|
|
while (tok && par < parameter) {
|
|
if (tok->str() == ")")
|
|
break;
|
|
if (tok->str() == ",")
|
|
++par;
|
|
tok = tok->next();
|
|
}
|
|
if (!tok)
|
|
return false;
|
|
|
|
// If parameter was found, determine if it's passed by value
|
|
if (par == parameter) {
|
|
bool knowntype = false;
|
|
while (tok && tok->isName()) {
|
|
knowntype |= tok->isStandardType();
|
|
knowntype |= (tok->str() == "struct");
|
|
tok = tok->next();
|
|
}
|
|
if (!tok || !knowntype)
|
|
return false;
|
|
if (tok->str() != "," && tok->str() != ")")
|
|
return false;
|
|
return true;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return false;
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
|
|
void Tokenizer::eraseDeadCode(Token *begin, const Token *end)
|
|
{
|
|
if (!begin)
|
|
return;
|
|
const bool isgoto = Token::Match(begin->tokAt(-2), "goto %name% ;");
|
|
int indentlevel = 1;
|
|
int indentcase = 0;
|
|
int indentswitch = 0;
|
|
int indentlabel = 0;
|
|
int roundbraces = 0;
|
|
int indentcheck = 0;
|
|
std::vector<int> switchindents;
|
|
bool checklabel = false;
|
|
Token *tok = begin;
|
|
Token *tokcheck = nullptr;
|
|
while (tok->next() && tok->next() != end) {
|
|
if (tok->next()->str() == "(") {
|
|
++roundbraces;
|
|
tok->deleteNext();
|
|
continue;
|
|
} else if (tok->next()->str() == ")") {
|
|
if (!roundbraces)
|
|
break; //too many ending round parentheses
|
|
--roundbraces;
|
|
tok->deleteNext();
|
|
continue;
|
|
}
|
|
|
|
if (roundbraces) {
|
|
tok->deleteNext();
|
|
continue;
|
|
}
|
|
|
|
if (Token::Match(tok, "[{};] switch (")) {
|
|
if (!checklabel) {
|
|
if (!indentlabel) {
|
|
//remove 'switch ( ... )'
|
|
Token::eraseTokens(tok, tok->linkAt(2)->next());
|
|
} else {
|
|
tok = tok->linkAt(2);
|
|
}
|
|
if (tok->next()->str() == "{") {
|
|
++indentswitch;
|
|
indentcase = indentlevel + 1;
|
|
switchindents.push_back(indentcase);
|
|
}
|
|
} else {
|
|
tok = tok->linkAt(2);
|
|
if (Token::simpleMatch(tok, ") {")) {
|
|
++indentswitch;
|
|
indentcase = indentlevel + 1;
|
|
switchindents.push_back(indentcase);
|
|
}
|
|
}
|
|
} else if (tok->next()->str() == "{") {
|
|
++indentlevel;
|
|
if (!checklabel) {
|
|
checklabel = true;
|
|
tokcheck = tok;
|
|
indentcheck = indentlevel;
|
|
indentlabel = 0;
|
|
}
|
|
tok = tok->next();
|
|
} else if (tok->next()->str() == "}") {
|
|
--indentlevel;
|
|
if (!indentlevel)
|
|
break;
|
|
|
|
if (!checklabel) {
|
|
tok->deleteNext();
|
|
} else {
|
|
if (indentswitch && indentlevel == indentcase)
|
|
--indentlevel;
|
|
if (indentlevel < indentcheck) {
|
|
const Token *end2 = tok->next();
|
|
tok = end2->link()->previous(); //return to initial '{'
|
|
if (indentswitch && Token::simpleMatch(tok, ") {") && Token::Match(tok->link()->tokAt(-2), "[{};] switch ("))
|
|
tok = tok->link()->tokAt(-2); //remove also 'switch ( ... )'
|
|
Token::eraseTokens(tok, end2->next());
|
|
checklabel = false;
|
|
tokcheck = nullptr;
|
|
indentcheck = 0;
|
|
} else {
|
|
tok = tok->next();
|
|
}
|
|
}
|
|
if (indentswitch && indentlevel <= indentcase) {
|
|
--indentswitch;
|
|
switchindents.pop_back();
|
|
if (!indentswitch)
|
|
indentcase = 0;
|
|
else
|
|
indentcase = switchindents[indentswitch-1];
|
|
}
|
|
} else if (Token::Match(tok, "[{};:] case")) {
|
|
const Token *tok2 = Token::findsimplematch(tok->next(), ": ;", end);
|
|
if (!tok2) {
|
|
tok->deleteNext();
|
|
continue;
|
|
}
|
|
if (indentlevel == 1)
|
|
break; //it seems like the function was called inside a case-default block.
|
|
if (indentlevel == indentcase)
|
|
++indentlevel;
|
|
tok2 = tok2->next();
|
|
if (!checklabel || !indentswitch) {
|
|
Token::eraseTokens(tok, tok2->next());
|
|
} else {
|
|
tok = const_cast<Token *>(tok2);
|
|
}
|
|
} else if (Token::Match(tok, "[{};] default : ;")) {
|
|
if (indentlevel == 1)
|
|
break; //it seems like the function was called inside a case-default block.
|
|
if (indentlevel == indentcase)
|
|
++indentlevel;
|
|
if (!checklabel || !indentswitch) {
|
|
tok->deleteNext(3);
|
|
} else {
|
|
tok = tok->tokAt(3);
|
|
}
|
|
} else if (Token::Match(tok, "[{};] %name% : ;") && tok->next()->str() != "default") {
|
|
if (checklabel) {
|
|
indentlabel = indentlevel;
|
|
tok = tokcheck->next();
|
|
checklabel = false;
|
|
indentlevel = indentcheck;
|
|
} else {
|
|
if (indentswitch) {
|
|
//Before stopping the function, since the 'switch()'
|
|
//instruction is removed, there's no sense to keep the
|
|
//case instructions. Remove them, if there are any.
|
|
Token *tok2 = tok->tokAt(3);
|
|
int indentlevel2 = indentlevel;
|
|
while (tok2->next() && tok2->next() != end) {
|
|
if (Token::Match(tok2->next(), "{|[|(")) {
|
|
tok2 = tok2->next()->link();
|
|
} else if (Token::Match(tok2, "[{};:] case")) {
|
|
const Token *tok3 = Token::findsimplematch(tok2->next(), ": ;", end);
|
|
if (!tok3) {
|
|
tok2 = tok2->next();
|
|
continue;
|
|
}
|
|
Token::eraseTokens(tok2, tok3->next());
|
|
} else if (Token::Match(tok2, "[{};] default : ;")) {
|
|
tok2->deleteNext(3);
|
|
} else if (tok2->next()->str() == "}") {
|
|
--indentlevel2;
|
|
if (indentlevel2 <= indentcase)
|
|
break;
|
|
tok2 = tok2->next();
|
|
} else {
|
|
tok2 = tok2->next();
|
|
}
|
|
}
|
|
}
|
|
break; //stop removing tokens, we arrived to the label.
|
|
}
|
|
} else if (isgoto && Token::Match(tok, "[{};] do|while|for|BOOST_FOREACH")) {
|
|
//it's possible that code inside loop is not dead,
|
|
//because of the possible presence of the label pointed by 'goto'
|
|
const Token *start = tok->tokAt(2);
|
|
if (start && start->str() == "(")
|
|
start = start->link()->next();
|
|
if (start && start->str() == "{") {
|
|
std::string labelpattern = "[{};] " + begin->previous()->str() + " : ;";
|
|
bool simplify = true;
|
|
for (Token *tok2 = start->next(); tok2 != start->link(); tok2 = tok2->next()) {
|
|
if (Token::Match(tok2, labelpattern.c_str())) {
|
|
simplify = false;
|
|
break;
|
|
}
|
|
}
|
|
//bailout for now
|
|
if (!simplify)
|
|
break;
|
|
}
|
|
tok->deleteNext();
|
|
} else {
|
|
// no need to keep the other strings, remove them.
|
|
if (tok->strAt(1) == "while") {
|
|
if (tok->str() == "}" && tok->link()->strAt(-1) == "do")
|
|
tok->link()->previous()->deleteThis();
|
|
}
|
|
tok->deleteNext();
|
|
}
|
|
}
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
|
|
void Tokenizer::syntaxError(const Token *tok, const std::string &code) const
|
|
{
|
|
printDebugOutput(0);
|
|
throw InternalError(tok, code.empty() ? "syntax error" : "syntax error: " + code, InternalError::SYNTAX);
|
|
}
|
|
|
|
void Tokenizer::unmatchedToken(const Token *tok) const
|
|
{
|
|
printDebugOutput(0);
|
|
throw InternalError(tok,
|
|
"Unmatched '" + tok->str() + "'. Configuration: '" + mConfiguration + "'.",
|
|
InternalError::SYNTAX);
|
|
}
|
|
|
|
void Tokenizer::syntaxErrorC(const Token *tok, const std::string &what) const
|
|
{
|
|
printDebugOutput(0);
|
|
throw InternalError(tok, "Code '"+what+"' is invalid C code. Use --std or --language to configure the language.", InternalError::SYNTAX);
|
|
}
|
|
|
|
void Tokenizer::unknownMacroError(const Token *tok1) const
|
|
{
|
|
printDebugOutput(0);
|
|
throw InternalError(tok1, "There is an unknown macro here somewhere. Configuration is required. If " + tok1->str() + " is a macro then please configure it.", InternalError::UNKNOWN_MACRO);
|
|
}
|
|
|
|
void Tokenizer::unhandled_macro_class_x_y(const Token *tok) const
|
|
{
|
|
reportError(tok,
|
|
Severity::information,
|
|
"class_X_Y",
|
|
"The code '" +
|
|
tok->str() + " " +
|
|
tok->strAt(1) + " " +
|
|
tok->strAt(2) + " " +
|
|
tok->strAt(3) + "' is not handled. You can use -I or --include to add handling of this code.");
|
|
}
|
|
|
|
void Tokenizer::macroWithSemicolonError(const Token *tok, const std::string ¯oName) const
|
|
{
|
|
reportError(tok,
|
|
Severity::information,
|
|
"macroWithSemicolon",
|
|
"Ensure that '" + macroName + "' is defined either using -I, --include or -D.");
|
|
}
|
|
|
|
void Tokenizer::cppcheckError(const Token *tok) const
|
|
{
|
|
printDebugOutput(0);
|
|
throw InternalError(tok, "Analysis failed. If the code is valid then please report this failure.", InternalError::INTERNAL);
|
|
}
|
|
|
|
void Tokenizer::unhandledCharLiteral(const Token *tok, const std::string& msg) const
|
|
{
|
|
std::string s = tok ? (" " + tok->str()) : "";
|
|
for (int i = 0; i < s.size(); ++i) {
|
|
if ((unsigned char)s[i] >= 0x80)
|
|
s.clear();
|
|
}
|
|
|
|
reportError(tok,
|
|
Severity::portability,
|
|
"nonStandardCharLiteral",
|
|
"Non-standard character literal" + s + ". " + msg);
|
|
}
|
|
|
|
/**
|
|
* Helper function to check whether number is equal to integer constant X
|
|
* or floating point pattern X.0
|
|
* @param s the string to check
|
|
* @param intConstant the integer constant to check against
|
|
* @param floatConstant the string with stringified float constant to check against
|
|
* @return true in case s is equal to X or X.0 and false otherwise.
|
|
*/
|
|
static bool isNumberOneOf(const std::string &s, const MathLib::bigint& intConstant, const char* floatConstant)
|
|
{
|
|
if (MathLib::isInt(s)) {
|
|
if (MathLib::toLongNumber(s) == intConstant)
|
|
return true;
|
|
} else if (MathLib::isFloat(s)) {
|
|
if (MathLib::toString(MathLib::toDoubleNumber(s)) == floatConstant)
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
// ------------------------------------------------------------------------
|
|
// Helper function to check whether number is zero (0 or 0.0 or 0E+0) or not?
|
|
// @param s the string to check
|
|
// @return true in case s is zero and false otherwise.
|
|
// ------------------------------------------------------------------------
|
|
bool Tokenizer::isZeroNumber(const std::string &s)
|
|
{
|
|
return isNumberOneOf(s, 0L, "0.0");
|
|
}
|
|
|
|
// ------------------------------------------------------------------------
|
|
// Helper function to check whether number is one (1 or 0.1E+1 or 1E+0) or not?
|
|
// @param s the string to check
|
|
// @return true in case s is one and false otherwise.
|
|
// ------------------------------------------------------------------------
|
|
bool Tokenizer::isOneNumber(const std::string &s)
|
|
{
|
|
if (!MathLib::isPositive(s))
|
|
return false;
|
|
return isNumberOneOf(s, 1L, "1.0");
|
|
}
|
|
|
|
// ------------------------------------------------------------------------
|
|
// Helper function to check whether number is two (2 or 0.2E+1 or 2E+0) or not?
|
|
// @param s the string to check
|
|
// @return true in case s is two and false otherwise.
|
|
// ------------------------------------------------------------------------
|
|
bool Tokenizer::isTwoNumber(const std::string &s)
|
|
{
|
|
if (!MathLib::isPositive(s))
|
|
return false;
|
|
return isNumberOneOf(s, 2L, "2.0");
|
|
}
|
|
|
|
// ------------------------------------------------------
|
|
// Simplify math functions.
|
|
// It simplifies the following functions: atol(), fmin(),
|
|
// fminl(), fminf(), fmax(), fmaxl(), fmaxf(), pow(),
|
|
// powf(), powl(), cbrt(), cbrtl(), cbtrf(), sqrt(),
|
|
// sqrtf(), sqrtl(), exp(), expf(), expl(), exp2(),
|
|
// exp2f(), exp2l(), log2(), log2f(), log2l(), log1p(),
|
|
// log1pf(), log1pl(), log10(), log10l(), log10f(),
|
|
// log(), logf(), logl(), logb(), logbf(), logbl(), acosh()
|
|
// acoshf(), acoshl(), acos(), acosf(), acosl(), cosh()
|
|
// coshf(), coshf(), cos(), cosf(), cosl(), erfc(),
|
|
// erfcf(), erfcl(), ilogb(), ilogbf(), ilogbf(), erf(),
|
|
// erfl(), erff(), asin(), asinf(), asinf(), asinh(),
|
|
// asinhf(), asinhl(), tan(), tanf(), tanl(), tanh(),
|
|
// tanhf(), tanhl(), atan(), atanf(), atanl(), atanh(),
|
|
// atanhf(), atanhl(), expm1(), expm1l(), expm1f(), sin(),
|
|
// sinf(), sinl(), sinh(), sinhf(), sinhl()
|
|
// in the tokenlist.
|
|
//
|
|
// Reference:
|
|
// - http://www.cplusplus.com/reference/cmath/
|
|
// ------------------------------------------------------
|
|
void Tokenizer::simplifyMathFunctions()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->isName() && !tok->varId() && tok->strAt(1) == "(") { // precondition for function
|
|
bool simplifcationMade = false;
|
|
if (Token::Match(tok, "atol ( %str% )")) { //@todo Add support for atoll()
|
|
if (Token::simpleMatch(tok->tokAt(-2), "std ::")) {
|
|
tok = tok->tokAt(-2);// set token index two steps back
|
|
tok->deleteNext(2); // delete "std ::"
|
|
}
|
|
const std::string& strNumber = tok->tokAt(2)->strValue(); // get number
|
|
const bool isNotAnInteger = (!MathLib::isInt(strNumber));// check: is not an integer
|
|
if (strNumber.empty() || isNotAnInteger) {
|
|
// Ignore strings which we can't convert
|
|
continue;
|
|
}
|
|
// Convert string into a number and insert into token list
|
|
tok->str(MathLib::toString(MathLib::toLongNumber(strNumber)));
|
|
// remove ( %num% )
|
|
tok->deleteNext(3);
|
|
simplifcationMade = true;
|
|
} else if (Token::Match(tok, "sqrt|sqrtf|sqrtl|cbrt|cbrtf|cbrtl ( %num% )")) {
|
|
// Simplify: sqrt(0) = 0 and cbrt(0) == 0
|
|
// sqrt(1) = 1 and cbrt(1) == 1
|
|
// get number string
|
|
const std::string& parameter(tok->strAt(2));
|
|
// is parameter 0 ?
|
|
if (isZeroNumber(parameter)) {
|
|
tok->deleteNext(3); // delete tokens
|
|
tok->str("0"); // insert result into token list
|
|
simplifcationMade = true;
|
|
} else if (isOneNumber(parameter)) {
|
|
tok->deleteNext(3); // delete tokens
|
|
tok->str("1"); // insert result into token list
|
|
simplifcationMade = true;
|
|
}
|
|
} else if (Token::Match(tok, "exp|expf|expl|exp2|exp2f|exp2l|cos|cosf|cosl|cosh|coshf|coshl|erfc|erfcf|erfcl ( %num% )")) {
|
|
// Simplify: exp[f|l](0) = 1 and exp2[f|l](0) = 1
|
|
// cosh[f|l](0) = 1 and cos[f|l](0) = 1
|
|
// erfc[f|l](0) = 1
|
|
// get number string
|
|
const std::string& parameter(tok->strAt(2));
|
|
// is parameter 0 ?
|
|
if (isZeroNumber(parameter)) {
|
|
tok->deleteNext(3); // delete tokens
|
|
tok->str("1"); // insert result into token list
|
|
simplifcationMade = true;
|
|
}
|
|
} else if (Token::Match(tok, "log1p|log1pf|log1pl|sin|sinf|sinl|sinh|sinhf|sinhl|erf|erff|erfl|asin|asinf|asinl|asinh|asinhf|asinhl|tan|tanf|tanl|tanh|tanhf|tanhl|atan|atanf|atanl|atanh|atanhf|atanhl|expm1|expm1f|expm1l ( %num% )")) {
|
|
// Simplify: log1p[f|l](0) = 0 and sin[f|l](0) = 0
|
|
// sinh[f|l](0) = 0 and erf[f|l](0) = 0
|
|
// asin[f|l](0) = 0 and sinh[f|l](0) = 0
|
|
// tan[f|l](0) = 0 and tanh[f|l](0) = 0
|
|
// atan[f|l](0) = 0 and atanh[f|l](0)= 0
|
|
// expm1[f|l](0) = 0
|
|
// get number string
|
|
const std::string& parameter(tok->strAt(2));
|
|
// is parameter 0 ?
|
|
if (isZeroNumber(parameter)) {
|
|
tok->deleteNext(3); // delete tokens
|
|
tok->str("0"); // insert result into token list
|
|
simplifcationMade = true;
|
|
}
|
|
} else if (Token::Match(tok, "log2|log2f|log2l|log|logf|logl|log10|log10f|log10l|logb|logbf|logbl|acosh|acoshf|acoshl|acos|acosf|acosl|ilogb|ilogbf|ilogbl ( %num% )")) {
|
|
// Simplify: log2[f|l](1) = 0 , log10[f|l](1) = 0
|
|
// log[f|l](1) = 0 , logb10[f|l](1) = 0
|
|
// acosh[f|l](1) = 0 , acos[f|l](1) = 0
|
|
// ilogb[f|l](1) = 0
|
|
// get number string
|
|
const std::string& parameter(tok->strAt(2));
|
|
// is parameter 1 ?
|
|
if (isOneNumber(parameter)) {
|
|
tok->deleteNext(3); // delete tokens
|
|
tok->str("0"); // insert result into token list
|
|
simplifcationMade = true;
|
|
}
|
|
} else if (Token::Match(tok, "fmin|fminl|fminf ( %num% , %num% )")) {
|
|
// @todo if one of the parameters is NaN the other is returned
|
|
// e.g. printf ("fmin (NaN, -1.0) = %f\n", fmin(NaN,-1.0));
|
|
// e.g. printf ("fmin (-1.0, NaN) = %f\n", fmin(-1.0,NaN));
|
|
const std::string& strLeftNumber(tok->strAt(2));
|
|
const std::string& strRightNumber(tok->strAt(4));
|
|
const bool isLessEqual = MathLib::isLessEqual(strLeftNumber, strRightNumber);
|
|
// case: left <= right ==> insert left
|
|
if (isLessEqual) {
|
|
tok->str(strLeftNumber); // insert e.g. -1.0
|
|
tok->deleteNext(5); // delete e.g. fmin ( -1.0, 1.0 )
|
|
simplifcationMade = true;
|
|
} else { // case left > right ==> insert right
|
|
tok->str(strRightNumber); // insert e.g. 0.0
|
|
tok->deleteNext(5); // delete e.g. fmin ( 1.0, 0.0 )
|
|
simplifcationMade = true;
|
|
}
|
|
} else if (Token::Match(tok, "fmax|fmaxl|fmaxf ( %num% , %num% )")) {
|
|
// @todo if one of the parameters is NaN the other is returned
|
|
// e.g. printf ("fmax (NaN, -1.0) = %f\n", fmax(NaN,-1.0));
|
|
// e.g. printf ("fmax (-1.0, NaN) = %f\n", fmax(-1.0,NaN));
|
|
const std::string& strLeftNumber(tok->strAt(2));
|
|
const std::string& strRightNumber(tok->strAt(4));
|
|
const bool isLessEqual = MathLib::isLessEqual(strLeftNumber, strRightNumber);
|
|
// case: left <= right ==> insert right
|
|
if (isLessEqual) {
|
|
tok->str(strRightNumber);// insert e.g. 1.0
|
|
tok->deleteNext(5); // delete e.g. fmax ( -1.0, 1.0 )
|
|
simplifcationMade = true;
|
|
} else { // case left > right ==> insert left
|
|
tok->str(strLeftNumber); // insert e.g. 1.0
|
|
tok->deleteNext(5); // delete e.g. fmax ( 1.0, 0.0 )
|
|
simplifcationMade = true;
|
|
}
|
|
} else if (Token::Match(tok, "pow|powf|powl (")) {
|
|
if (Token::Match(tok->tokAt(2), "%num% , %num% )")) {
|
|
// In case of pow ( 0 , anyNumber > 0): It can be simplified to 0
|
|
// In case of pow ( 0 , 0 ): It simplified to 1
|
|
// In case of pow ( 1 , anyNumber ): It simplified to 1
|
|
const std::string& leftNumber(tok->strAt(2)); // get the left parameter
|
|
const std::string& rightNumber(tok->strAt(4)); // get the right parameter
|
|
const bool isLeftNumberZero = isZeroNumber(leftNumber);
|
|
const bool isLeftNumberOne = isOneNumber(leftNumber);
|
|
const bool isRightNumberZero = isZeroNumber(rightNumber);
|
|
if (isLeftNumberZero && !isRightNumberZero && MathLib::isPositive(rightNumber)) { // case: 0^(y) = 0 and y > 0
|
|
tok->deleteNext(5); // delete tokens
|
|
tok->str("0"); // insert simplified result
|
|
simplifcationMade = true;
|
|
} else if (isLeftNumberZero && isRightNumberZero) { // case: 0^0 = 1
|
|
tok->deleteNext(5); // delete tokens
|
|
tok->str("1"); // insert simplified result
|
|
simplifcationMade = true;
|
|
} else if (isLeftNumberOne) { // case 1^(y) = 1
|
|
tok->deleteNext(5); // delete tokens
|
|
tok->str("1"); // insert simplified result
|
|
simplifcationMade = true;
|
|
}
|
|
}
|
|
if (Token::Match(tok->tokAt(2), "%any% , %num% )")) {
|
|
// In case of pow( x , 1 ): It can be simplified to x.
|
|
const std::string& leftParameter(tok->strAt(2)); // get the left parameter
|
|
const std::string& rightNumber(tok->strAt(4)); // get right number
|
|
if (isOneNumber(rightNumber)) { // case: x^(1) = x
|
|
tok->str(leftParameter); // insert simplified result
|
|
tok->deleteNext(5); // delete tokens
|
|
simplifcationMade = true;
|
|
} else if (isZeroNumber(rightNumber)) { // case: x^(0) = 1
|
|
tok->deleteNext(5); // delete tokens
|
|
tok->str("1"); // insert simplified result
|
|
simplifcationMade = true;
|
|
}
|
|
}
|
|
}
|
|
// Jump back to begin of statement if a simplification was performed
|
|
if (simplifcationMade) {
|
|
while (tok->previous() && tok->str() != ";") {
|
|
tok = tok->previous();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyComma()
|
|
{
|
|
bool inReturn = false;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
|
|
// skip enums
|
|
if (Token::Match(tok, "enum class|struct| %name%| :|{")) {
|
|
skipEnumBody(&tok);
|
|
}
|
|
if (!tok)
|
|
syntaxError(nullptr); // invalid code like in #4195
|
|
|
|
if (Token::Match(tok, "(|[") || Token::Match(tok->previous(), "%name%|= {")) {
|
|
tok = tok->link();
|
|
continue;
|
|
}
|
|
|
|
if (Token::simpleMatch(tok, "= (") && Token::simpleMatch(tok->linkAt(1), ") {")) {
|
|
tok = tok->linkAt(1)->linkAt(1);
|
|
continue;
|
|
}
|
|
|
|
// Skip unhandled template specifiers..
|
|
if (tok->link() && tok->str() == "<")
|
|
tok = tok->link();
|
|
|
|
if (tok->str() == "return" && Token::Match(tok->previous(), "[;{}]"))
|
|
inReturn = true;
|
|
|
|
if (inReturn && Token::Match(tok, "[;{}?:]"))
|
|
inReturn = false;
|
|
|
|
if (!tok->next() || tok->str() != ",")
|
|
continue;
|
|
|
|
// We must not accept just any keyword, e.g. accepting int
|
|
// would cause function parameters to corrupt.
|
|
if (isCPP() && tok->strAt(1) == "delete") {
|
|
// Handle "delete a, delete b;"
|
|
tok->str(";");
|
|
}
|
|
|
|
if (isCPP() && Token::Match(tok->tokAt(-2), "delete %name% , %name% ;") &&
|
|
tok->next()->varId() != 0) {
|
|
// Handle "delete a, b;" - convert to delete a; b;
|
|
tok->str(";");
|
|
} else if (!inReturn && tok->tokAt(-2)) {
|
|
bool replace = false;
|
|
for (Token *tok2 = tok->previous(); tok2; tok2 = tok2->previous()) {
|
|
if (tok2->str() == "=") {
|
|
// Handle "a = 0, b = 0;"
|
|
replace = true;
|
|
} else if (isCPP() && (Token::Match(tok2, "delete %name%") ||
|
|
Token::Match(tok2, "delete [ ] %name%"))) {
|
|
// Handle "delete a, a = 0;"
|
|
replace = true;
|
|
} else if (Token::Match(tok2, "[?:;,{}()]")) {
|
|
if (replace && Token::Match(tok2, "[;{}]"))
|
|
tok->str(";");
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
// find token where return ends and also count commas
|
|
if (inReturn) {
|
|
Token *startFrom = nullptr; // "[;{}]" token before "return"
|
|
Token *endAt = nullptr; // first ";" token after "[;{}] return"
|
|
|
|
// find "; return" pattern before comma
|
|
for (Token *tok2 = tok->previous(); tok2; tok2 = tok2->previous()) {
|
|
if (tok2->str() == "return") {
|
|
startFrom = tok2->previous();
|
|
break;
|
|
}
|
|
}
|
|
if (!startFrom)
|
|
// to be very sure...
|
|
return;
|
|
int commaCounter = 0;
|
|
for (Token *tok2 = startFrom->next(); tok2; tok2 = tok2->next()) {
|
|
if (tok2->str() == ";") {
|
|
endAt = tok2;
|
|
break;
|
|
|
|
} else if (Token::Match(tok2, "(|[") ||
|
|
(tok2->str() == "{" && tok2->previous() && tok2->previous()->str() == "=")) {
|
|
tok2 = tok2->link();
|
|
|
|
} else if (tok2->str() == ",") {
|
|
++commaCounter;
|
|
}
|
|
}
|
|
|
|
if (!endAt)
|
|
//probably a syntax error
|
|
return;
|
|
|
|
if (commaCounter) {
|
|
// change tokens:
|
|
// "; return a ( ) , b ( ) , c ;"
|
|
// to
|
|
// "; a ( ) ; b ( ) ; return c ;"
|
|
|
|
// remove "return"
|
|
startFrom->deleteNext();
|
|
for (Token *tok2 = startFrom->next(); tok2 != endAt; tok2 = tok2->next()) {
|
|
if (Token::Match(tok2, "(|[") ||
|
|
(tok2->str() == "{" && tok2->previous() && tok2->previous()->str() == "=")) {
|
|
tok2 = tok2->link();
|
|
|
|
} else if (tok2->str() == ",") {
|
|
tok2->str(";");
|
|
--commaCounter;
|
|
if (commaCounter == 0) {
|
|
tok2->insertToken("return");
|
|
}
|
|
}
|
|
}
|
|
tok = endAt;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::checkConfiguration() const
|
|
{
|
|
if (!mSettings->checkConfiguration)
|
|
return;
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "%name% ("))
|
|
continue;
|
|
if (tok->isControlFlowKeyword())
|
|
continue;
|
|
for (const Token *tok2 = tok->tokAt(2); tok2 && tok2->str() != ")"; tok2 = tok2->next()) {
|
|
if (tok2->str() == ";") {
|
|
macroWithSemicolonError(tok, tok->str());
|
|
break;
|
|
}
|
|
if (Token::Match(tok2, "(|{"))
|
|
tok2 = tok2->link();
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::validateC() const
|
|
{
|
|
if (isCPP())
|
|
return;
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
// That might trigger false positives, but it's much faster to have this truncated pattern
|
|
if (Token::Match(tok, "const_cast|dynamic_cast|reinterpret_cast|static_cast <"))
|
|
syntaxErrorC(tok, "C++ cast <...");
|
|
// Template function..
|
|
if (Token::Match(tok, "%name% < %name% > (")) {
|
|
const Token *tok2 = tok->tokAt(5);
|
|
while (tok2 && !Token::Match(tok2, "[()]"))
|
|
tok2 = tok2->next();
|
|
if (Token::simpleMatch(tok2, ") {"))
|
|
syntaxErrorC(tok, tok->str() + '<' + tok->strAt(2) + ">() {}");
|
|
}
|
|
if (tok->previous() && !Token::Match(tok->previous(), "[;{}]"))
|
|
continue;
|
|
if (Token::Match(tok, "using namespace %name% ;"))
|
|
syntaxErrorC(tok, "using namespace " + tok->strAt(2));
|
|
if (Token::Match(tok, "template < class|typename %name% [,>]"))
|
|
syntaxErrorC(tok, "template<...");
|
|
if (Token::Match(tok, "%name% :: %name%"))
|
|
syntaxErrorC(tok, tok->str() + tok->strAt(1) + tok->strAt(2));
|
|
if (Token::Match(tok, "class|namespace %name% [:{]"))
|
|
syntaxErrorC(tok, tok->str() + tok->strAt(1) + tok->strAt(2));
|
|
}
|
|
}
|
|
|
|
void Tokenizer::validate() const
|
|
{
|
|
std::stack<const Token *> linkTokens;
|
|
const Token *lastTok = nullptr;
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
lastTok = tok;
|
|
if (Token::Match(tok, "[{([]") || (tok->str() == "<" && tok->link())) {
|
|
if (tok->link() == nullptr)
|
|
cppcheckError(tok);
|
|
|
|
linkTokens.push(tok);
|
|
}
|
|
|
|
else if (Token::Match(tok, "[})]]") || (Token::Match(tok, ">|>>") && tok->link())) {
|
|
if (tok->link() == nullptr)
|
|
cppcheckError(tok);
|
|
|
|
if (linkTokens.empty() == true)
|
|
cppcheckError(tok);
|
|
|
|
if (tok->link() != linkTokens.top())
|
|
cppcheckError(tok);
|
|
|
|
if (tok != tok->link()->link())
|
|
cppcheckError(tok);
|
|
|
|
linkTokens.pop();
|
|
}
|
|
|
|
else if (tok->link() != nullptr)
|
|
cppcheckError(tok);
|
|
}
|
|
|
|
if (!linkTokens.empty())
|
|
cppcheckError(linkTokens.top());
|
|
|
|
// Validate that the Tokenizer::list.back() is updated correctly during simplifications
|
|
if (lastTok != list.back())
|
|
cppcheckError(lastTok);
|
|
}
|
|
|
|
static const Token *findUnmatchedTernaryOp(const Token * const begin, const Token * const end, int depth = 0)
|
|
{
|
|
std::stack<const Token *> ternaryOp;
|
|
for (const Token *tok = begin; tok != end && tok->str() != ";"; tok = tok->next()) {
|
|
if (tok->str() == "?")
|
|
ternaryOp.push(tok);
|
|
else if (!ternaryOp.empty() && tok->str() == ":")
|
|
ternaryOp.pop();
|
|
else if (depth < 100 && Token::Match(tok,"(|[")) {
|
|
const Token *inner = findUnmatchedTernaryOp(tok->next(), tok->link(), depth+1);
|
|
if (inner)
|
|
return inner;
|
|
tok = tok->link();
|
|
}
|
|
}
|
|
return ternaryOp.empty() ? nullptr : ternaryOp.top();
|
|
}
|
|
|
|
static bool isCPPAttribute(const Token * tok)
|
|
{
|
|
return Token::simpleMatch(tok, "[ [") && tok->link() && tok->link()->previous() == tok->linkAt(1);
|
|
}
|
|
|
|
static bool isAlignAttribute(const Token * tok)
|
|
{
|
|
return Token::simpleMatch(tok, "alignas (") && tok->next()->link();
|
|
}
|
|
|
|
static const Token* skipCPPOrAlignAttribute(const Token * tok)
|
|
{
|
|
if (isCPPAttribute(tok)) {
|
|
return tok->link();
|
|
} else if (isAlignAttribute(tok)) {
|
|
return tok->next()->link();
|
|
}
|
|
return tok;
|
|
}
|
|
|
|
static bool isNonMacro(const Token* tok)
|
|
{
|
|
if (tok->isKeyword())
|
|
return true;
|
|
if (cAlternativeTokens.count(tok->str()) > 0)
|
|
return true;
|
|
if (tok->str().compare(0, 2, "__") == 0) // attribute/annotation
|
|
return true;
|
|
return false;
|
|
}
|
|
|
|
void Tokenizer::reportUnknownMacros() const
|
|
{
|
|
// Report unknown macros used in expressions "%name% %num%"
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "%name% %num%")) {
|
|
// A keyword is not an unknown macro
|
|
if (tok->isKeyword())
|
|
continue;
|
|
|
|
if (Token::Match(tok->previous(), "%op%|("))
|
|
unknownMacroError(tok);
|
|
}
|
|
}
|
|
|
|
// Report unknown macros that contain several statements "MACRO(a;b;c)"
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "%name% ("))
|
|
continue;
|
|
if (!tok->isUpperCaseName())
|
|
continue;
|
|
const Token *endTok = tok->linkAt(1);
|
|
for (const Token *inner = tok->tokAt(2); inner != endTok; inner = inner->next()) {
|
|
if (Token::Match(inner, "[[({]"))
|
|
inner = inner->link();
|
|
else if (inner->str() == ";")
|
|
unknownMacroError(inner);
|
|
}
|
|
}
|
|
|
|
// Report unknown macros that contain struct initialization "MACRO(a, .b=3)"
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "%name% ("))
|
|
continue;
|
|
const Token *endTok = tok->linkAt(1);
|
|
for (const Token *inner = tok->tokAt(2); inner != endTok; inner = inner->next()) {
|
|
if (Token::Match(inner, "[[({]"))
|
|
inner = inner->link();
|
|
else if (Token::Match(inner->previous(), "[,(] . %name% =|{"))
|
|
unknownMacroError(tok);
|
|
}
|
|
}
|
|
|
|
// Report unknown macros in non-executable scopes..
|
|
std::set<std::string> possible;
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
// Skip executable scopes..
|
|
if (tok->str() == "{") {
|
|
const Token *prev = tok->previous();
|
|
while (prev && prev->isName())
|
|
prev = prev->previous();
|
|
if (prev && prev->str() == ")")
|
|
tok = tok->link();
|
|
else
|
|
possible.clear();
|
|
} else if (tok->str() == "}")
|
|
possible.clear();
|
|
|
|
if (Token::Match(tok, "%name% (") && tok->isUpperCaseName() && Token::simpleMatch(tok->linkAt(1), ") (") && Token::simpleMatch(tok->linkAt(1)->linkAt(1), ") {")) {
|
|
// A keyword is not an unknown macro
|
|
if (tok->isKeyword())
|
|
continue;
|
|
|
|
const Token *bodyStart = tok->linkAt(1)->linkAt(1)->tokAt(2);
|
|
const Token *bodyEnd = tok->link();
|
|
for (const Token *tok2 = bodyStart; tok2 && tok2 != bodyEnd; tok2 = tok2->next()) {
|
|
if (Token::Match(tok2, "if|switch|for|while|return"))
|
|
unknownMacroError(tok);
|
|
}
|
|
} else if (Token::Match(tok, "%name% (") && tok->isUpperCaseName() && Token::Match(tok->linkAt(1), ") %name% (") && Token::Match(tok->linkAt(1)->linkAt(2), ") [;{]")) {
|
|
if (possible.count(tok->str()) == 0)
|
|
possible.insert(tok->str());
|
|
else
|
|
unknownMacroError(tok);
|
|
}
|
|
}
|
|
|
|
// String concatenation with unknown macros
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "%str% %name% (") && Token::Match(tok->linkAt(2), ") %str%")) {
|
|
if (tok->next()->isKeyword())
|
|
continue;
|
|
unknownMacroError(tok->next());
|
|
}
|
|
if (Token::Match(tok, "[(,] %name% (") && Token::Match(tok->linkAt(2), ") %name% %name%|,|)")) {
|
|
if (tok->next()->isKeyword() || tok->linkAt(2)->next()->isKeyword())
|
|
continue;
|
|
if (cAlternativeTokens.count(tok->linkAt(2)->next()->str()) > 0)
|
|
continue;
|
|
if (tok->next()->str().compare(0, 2, "__") == 0) // attribute/annotation
|
|
continue;
|
|
unknownMacroError(tok->next());
|
|
}
|
|
}
|
|
|
|
// Report unknown macros without commas or operators inbetween statements: MACRO1() MACRO2()
|
|
for (const Token* tok = tokens(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "%name% ("))
|
|
continue;
|
|
if (isNonMacro(tok))
|
|
continue;
|
|
|
|
const Token* endTok = tok->linkAt(1);
|
|
if (!Token::Match(endTok, ") %name% (|."))
|
|
continue;
|
|
|
|
const Token* tok2 = endTok->next();
|
|
if (isNonMacro(tok2))
|
|
continue;
|
|
|
|
if (tok2->next()->str() == "(") {
|
|
if (Token::Match(tok->previous(), "%name%|::|>"))
|
|
continue;
|
|
}
|
|
|
|
unknownMacroError(tok);
|
|
}
|
|
}
|
|
|
|
void Tokenizer::findGarbageCode() const
|
|
{
|
|
const bool isCPP11 = isCPP() && mSettings->standards.cpp >= Standards::CPP11;
|
|
|
|
static const std::unordered_set<std::string> nonConsecutiveKeywords{ "break",
|
|
"continue",
|
|
"for",
|
|
"goto",
|
|
"if",
|
|
"return",
|
|
"switch",
|
|
"throw",
|
|
"typedef",
|
|
"while" };
|
|
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
// initialization: = {
|
|
if (Token::simpleMatch(tok, "= {") && Token::simpleMatch(tok->linkAt(1), "} ("))
|
|
syntaxError(tok->linkAt(1));
|
|
|
|
// Inside [] there can't be ; or various keywords
|
|
else if (tok->str() == "[") {
|
|
for (const Token* inner = tok->next(); inner != tok->link(); inner = inner->next()) {
|
|
if (Token::Match(inner, "(|[|{"))
|
|
inner = inner->link();
|
|
else if (Token::Match(inner, ";|goto|return|typedef"))
|
|
syntaxError(inner);
|
|
}
|
|
}
|
|
|
|
// array assignment
|
|
else if (Token::Match(tok, "%assign% [") && Token::simpleMatch(tok->linkAt(1), "] ;"))
|
|
syntaxError(tok, tok->str() + "[...];");
|
|
|
|
// UNKNOWN_MACRO(return)
|
|
if (tok->isKeyword() && Token::Match(tok, "throw|return )") && Token::Match(tok->linkAt(1)->previous(), "%name% ("))
|
|
unknownMacroError(tok->linkAt(1)->previous());
|
|
|
|
// UNKNOWN_MACRO(return)
|
|
else if (Token::Match(tok, "%name% throw|return") && std::isupper(tok->str()[0]))
|
|
unknownMacroError(tok);
|
|
|
|
// Assign/increment/decrement literal
|
|
else if (Token::Match(tok, "!!) %num%|%str%|%char% %assign%|++|--")) {
|
|
if (!isCPP() || mSettings->standards.cpp < Standards::CPP20 || !Token::Match(tok->previous(), "%name% : %num% ="))
|
|
syntaxError(tok, tok->next()->str() + " " + tok->strAt(2));
|
|
}
|
|
|
|
if (tok->isControlFlowKeyword() && Token::Match(tok, "if|while|for|switch")) { // if|while|for|switch (EXPR) { ... }
|
|
if (tok->previous() && !Token::Match(tok->previous(), "%name%|:|;|{|}|)")) {
|
|
if (Token::Match(tok->previous(), "[,(]")) {
|
|
const Token *prev = tok->previous();
|
|
while (prev && prev->str() != "(") {
|
|
if (prev->str() == ")")
|
|
prev = prev->link();
|
|
prev = prev->previous();
|
|
}
|
|
if (prev && Token::Match(prev->previous(), "%name% ("))
|
|
unknownMacroError(prev->previous());
|
|
}
|
|
if (!Token::simpleMatch(tok->tokAt(-2), "operator \"\" if"))
|
|
syntaxError(tok);
|
|
}
|
|
if (!Token::Match(tok->next(), "( !!)"))
|
|
syntaxError(tok);
|
|
if (tok->str() != "for") {
|
|
if (isGarbageExpr(tok->next(), tok->linkAt(1), mSettings->standards.cpp>=Standards::cppstd_t::CPP17))
|
|
syntaxError(tok);
|
|
}
|
|
}
|
|
|
|
// keyword keyword
|
|
if (tok->isKeyword() && nonConsecutiveKeywords.count(tok->str()) != 0) {
|
|
if (Token::Match(tok, "%name% %name%") && nonConsecutiveKeywords.count(tok->next()->str()) == 1)
|
|
syntaxError(tok);
|
|
const Token* prev = tok;
|
|
while (prev && prev->isName())
|
|
prev = prev->previous();
|
|
if (Token::Match(prev, "%op%|%num%|%str%|%char%")) {
|
|
if (!Token::simpleMatch(tok->tokAt(-2), "operator \"\" if") &&
|
|
!Token::simpleMatch(tok->tokAt(-2), "extern \"C\""))
|
|
syntaxError(tok, prev == tok->previous() ? (prev->str() + " " + tok->str()) : (prev->str() + " .. " + tok->str()));
|
|
}
|
|
}
|
|
}
|
|
|
|
// invalid struct declaration
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "struct|class|enum %name%| {") && (!tok->previous() || Token::Match(tok->previous(), "[;{}]"))) {
|
|
const Token *tok2 = tok->linkAt(tok->next()->isName() ? 2 : 1);
|
|
if (Token::Match(tok2, "} %op%")) {
|
|
tok2 = tok2->next();
|
|
if (!Token::Match(tok2, "*|&|&&"))
|
|
syntaxError(tok2, "Unexpected token '" + tok2->str() + "'");
|
|
while (Token::Match(tok2, "*|&|&&"))
|
|
tok2 = tok2->next();
|
|
if (!Token::Match(tok2, "%name%"))
|
|
syntaxError(tok2, "Unexpected token '" + tok2->str() + "'");
|
|
}
|
|
}
|
|
}
|
|
|
|
// Keywords in global scope
|
|
static const std::unordered_set<std::string> nonGlobalKeywords{"break",
|
|
"continue",
|
|
"for",
|
|
"goto",
|
|
"if",
|
|
"return",
|
|
"switch",
|
|
"while",
|
|
"try",
|
|
"catch"};
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (tok->str() == "{")
|
|
tok = tok->link();
|
|
else if (tok->isKeyword() && nonGlobalKeywords.count(tok->str()) && !Token::Match(tok->tokAt(-2), "operator %str%"))
|
|
syntaxError(tok, "keyword '" + tok->str() + "' is not allowed in global scope");
|
|
}
|
|
|
|
// case keyword must be inside switch
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (Token::simpleMatch(tok, "switch (")) {
|
|
if (Token::simpleMatch(tok->linkAt(1), ") {")) {
|
|
tok = tok->linkAt(1)->linkAt(1);
|
|
continue;
|
|
}
|
|
const Token *switchToken = tok;
|
|
tok = tok->linkAt(1);
|
|
if (!tok)
|
|
syntaxError(switchToken);
|
|
// Look for the end of the switch statement, i.e. the first semi-colon or '}'
|
|
for (; tok; tok = tok->next()) {
|
|
if (tok->str() == "{") {
|
|
tok = tok->link();
|
|
}
|
|
if (Token::Match(tok, ";|}")) {
|
|
// We're at the end of the switch block
|
|
if (tok->str() == "}" && tok->strAt(-1) == ":") // Invalid case
|
|
syntaxError(switchToken);
|
|
break;
|
|
}
|
|
}
|
|
if (!tok)
|
|
break;
|
|
} else if (tok->str() == "(") {
|
|
tok = tok->link();
|
|
} else if (tok->str() == "case") {
|
|
syntaxError(tok);
|
|
}
|
|
}
|
|
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (!Token::simpleMatch(tok, "for (")) // find for loops
|
|
continue;
|
|
// count number of semicolons
|
|
int semicolons = 0;
|
|
const Token* const startTok = tok;
|
|
tok = tok->next()->link()->previous(); // find ")" of the for-loop
|
|
// walk backwards until we find the beginning (startTok) of the for() again
|
|
for (; tok != startTok; tok = tok->previous()) {
|
|
if (tok->str() == ";") { // do the counting
|
|
semicolons++;
|
|
} else if (tok->str() == ")") { // skip pairs of ( )
|
|
tok = tok->link();
|
|
}
|
|
}
|
|
// if we have an invalid number of semicolons inside for( ), assume syntax error
|
|
if (semicolons > 2)
|
|
syntaxError(tok);
|
|
if (semicolons == 1 && !(isCPP() && mSettings->standards.cpp >= Standards::CPP20))
|
|
syntaxError(tok);
|
|
}
|
|
|
|
// Operators without operands..
|
|
const Token *templateEndToken = nullptr;
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (!templateEndToken) {
|
|
if (tok->str() == "<" && isCPP())
|
|
templateEndToken = tok->findClosingBracket();
|
|
} else {
|
|
if (templateEndToken == tok)
|
|
templateEndToken = nullptr;
|
|
if (Token::Match(tok, "> %cop%"))
|
|
continue;
|
|
}
|
|
// skip C++ attributes [[...]]
|
|
if (isCPP11 && (isCPPAttribute(tok) || isAlignAttribute(tok))) {
|
|
tok = skipCPPOrAlignAttribute(tok);
|
|
continue;
|
|
}
|
|
{
|
|
bool match1 = Token::Match(tok, "%or%|%oror%|==|!=|+|-|/|!|>=|<=|~|^|++|--|::|sizeof");
|
|
bool match2 = Token::Match(tok->next(), "{|if|else|while|do|for|return|switch|break");
|
|
if (isCPP()) {
|
|
match1 = match1 || Token::Match(tok, "::|throw|decltype|typeof");
|
|
match2 = match2 || Token::Match(tok->next(), "try|catch|namespace");
|
|
}
|
|
if (match1 && match2)
|
|
syntaxError(tok);
|
|
}
|
|
if (Token::Match(tok, "%or%|%oror%|~|^|!|%comp%|+|-|/|%")) {
|
|
std::string code;
|
|
if (Token::Match(tok->next(), ")|]|}"))
|
|
code = tok->str() + tok->next()->str();
|
|
if (Token::simpleMatch(tok->next(), "( )"))
|
|
code = tok->str() + "()";
|
|
if (!code.empty()) {
|
|
if (isC() || (tok->str() != ">" && !Token::simpleMatch(tok->previous(), "operator")))
|
|
syntaxError(tok, code);
|
|
}
|
|
}
|
|
if (Token::Match(tok, "%num%|%bool%|%char%|%str% %num%|%bool%|%char%|%str%") && !Token::Match(tok, "%str% %str%"))
|
|
syntaxError(tok);
|
|
if (Token::Match(tok, "%assign% typename|class %assign%"))
|
|
syntaxError(tok);
|
|
if (Token::Match(tok, "%cop%|=|,|[ %or%|%oror%|/|%"))
|
|
syntaxError(tok);
|
|
if (Token::Match(tok, ";|(|[ %comp%"))
|
|
syntaxError(tok);
|
|
if (Token::Match(tok, "%cop%|= ]") && !(isCPP() && Token::Match(tok->previous(), "[|,|%num% &|=|> ]")))
|
|
syntaxError(tok);
|
|
if (Token::Match(tok, "[+-] [;,)]}]") && !(isCPP() && Token::Match(tok->previous(), "operator [+-] ;")))
|
|
syntaxError(tok);
|
|
if (Token::simpleMatch(tok, ",") &&
|
|
!Token::Match(tok->tokAt(-2), "[ = , &|%name%")) {
|
|
if (Token::Match(tok->previous(), "(|[|{|<|%assign%|%or%|%oror%|==|!=|+|-|/|!|>=|<=|~|^|::|sizeof"))
|
|
syntaxError(tok);
|
|
if (isCPP() && Token::Match(tok->previous(), "throw|decltype|typeof"))
|
|
syntaxError(tok);
|
|
if (Token::Match(tok->next(), ")|]|>|%assign%|%or%|%oror%|==|!=|/|>=|<=|&&"))
|
|
syntaxError(tok);
|
|
}
|
|
if (Token::simpleMatch(tok, ".") &&
|
|
!Token::simpleMatch(tok->previous(), ".") &&
|
|
!Token::simpleMatch(tok->next(), ".") &&
|
|
!Token::Match(tok->previous(), "{|, . %name% =|.|[|{") &&
|
|
!Token::Match(tok->previous(), ", . %name%")) {
|
|
if (!Token::Match(tok->previous(), "%name%|)|]|>|}"))
|
|
syntaxError(tok, tok->strAt(-1) + " " + tok->str() + " " + tok->strAt(1));
|
|
if (!Token::Match(tok->next(), "%name%|*|~"))
|
|
syntaxError(tok, tok->strAt(-1) + " " + tok->str() + " " + tok->strAt(1));
|
|
}
|
|
if (Token::Match(tok, "[!|+-/%^~] )|]"))
|
|
syntaxError(tok);
|
|
if (Token::Match(tok, "==|!=|<=|>= %comp%") && tok->strAt(-1) != "operator")
|
|
syntaxError(tok, tok->str() + " " + tok->strAt(1));
|
|
}
|
|
|
|
// ternary operator without :
|
|
if (const Token *ternaryOp = findUnmatchedTernaryOp(tokens(), nullptr))
|
|
syntaxError(ternaryOp);
|
|
|
|
// Code must not start with an arithmetical operand
|
|
if (Token::Match(list.front(), "%cop%"))
|
|
syntaxError(list.front());
|
|
|
|
// Code must end with } ; ) NAME
|
|
if (!Token::Match(list.back(), "%name%|;|}|)"))
|
|
syntaxError(list.back());
|
|
if (list.back()->str() == ")" && !Token::Match(list.back()->link()->previous(), "%name%|> ("))
|
|
syntaxError(list.back());
|
|
for (const Token *end = list.back(); end && end->isName(); end = end->previous()) {
|
|
if (Token::Match(end, "void|char|short|int|long|float|double|const|volatile|static|inline|struct|class|enum|union|template|sizeof|case|break|continue|typedef"))
|
|
syntaxError(list.back());
|
|
}
|
|
if ((list.back()->str()==")" || list.back()->str()=="}") && list.back()->previous() && list.back()->previous()->isControlFlowKeyword())
|
|
syntaxError(list.back()->previous());
|
|
|
|
// Garbage templates..
|
|
if (isCPP()) {
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (!Token::simpleMatch(tok, "template <"))
|
|
continue;
|
|
if (tok->previous() && !Token::Match(tok->previous(), ":|;|{|}|)|>|\"C++\"")) {
|
|
if (tok->previous()->isUpperCaseName())
|
|
unknownMacroError(tok->previous());
|
|
else
|
|
syntaxError(tok);
|
|
}
|
|
const Token * const tok1 = tok;
|
|
tok = tok->next()->findClosingBracket();
|
|
if (!tok)
|
|
syntaxError(tok1);
|
|
if (!Token::Match(tok, ">|>> ::|...| %name%") &&
|
|
!Token::Match(tok, ">|>> [ [ %name%") &&
|
|
!Token::Match(tok, "> >|*"))
|
|
syntaxError(tok->next() ? tok->next() : tok1);
|
|
}
|
|
}
|
|
|
|
// Objective C/C++
|
|
for (const Token *tok = tokens(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "[;{}] [ %name% %name% ] ;"))
|
|
syntaxError(tok->next());
|
|
}
|
|
}
|
|
|
|
|
|
bool Tokenizer::isGarbageExpr(const Token *start, const Token *end, bool allowSemicolon)
|
|
{
|
|
for (const Token *tok = start; tok != end; tok = tok->next()) {
|
|
if (tok->isControlFlowKeyword())
|
|
return true;
|
|
if (!allowSemicolon && tok->str() == ";")
|
|
return true;
|
|
if (tok->str() == "{")
|
|
tok = tok->link();
|
|
}
|
|
return false;
|
|
}
|
|
|
|
std::string Tokenizer::simplifyString(const std::string &source)
|
|
{
|
|
std::string str = source;
|
|
|
|
for (std::string::size_type i = 0; i + 1U < str.size(); ++i) {
|
|
if (str[i] != '\\')
|
|
continue;
|
|
|
|
int c = 'a'; // char
|
|
int sz = 0; // size of stringdata
|
|
if (str[i+1] == 'x') {
|
|
sz = 2;
|
|
while (sz < 4 && std::isxdigit((unsigned char)str[i+sz]))
|
|
sz++;
|
|
if (sz > 2) {
|
|
std::istringstream istr(str.substr(i+2, sz-2));
|
|
istr >> std::hex >> c;
|
|
}
|
|
} else if (MathLib::isOctalDigit(str[i+1])) {
|
|
sz = 2;
|
|
while (sz < 4 && MathLib::isOctalDigit(str[i+sz]))
|
|
sz++;
|
|
std::istringstream istr(str.substr(i+1, sz-1));
|
|
istr >> std::oct >> c;
|
|
str = str.substr(0,i) + (char)c + str.substr(i+sz);
|
|
continue;
|
|
}
|
|
|
|
if (sz <= 2)
|
|
i++;
|
|
else if (i+sz < str.size())
|
|
str.replace(i, sz, std::string(1U, (char)c));
|
|
else
|
|
str.replace(i, str.size() - i - 1U, "a");
|
|
}
|
|
|
|
return str;
|
|
}
|
|
|
|
void Tokenizer::simplifyWhile0()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
// while (0)
|
|
const bool while0(Token::Match(tok->previous(), "[{};] while ( 0|false )"));
|
|
|
|
// for (0) - not banal, ticket #3140
|
|
const bool for0((Token::Match(tok->previous(), "[{};] for ( %name% = %num% ; %name% < %num% ;") &&
|
|
tok->strAt(2) == tok->strAt(6) && tok->strAt(4) == tok->strAt(8)) ||
|
|
(Token::Match(tok->previous(), "[{};] for ( %type% %name% = %num% ; %name% < %num% ;") &&
|
|
tok->strAt(3) == tok->strAt(7) && tok->strAt(5) == tok->strAt(9)));
|
|
|
|
if (!while0 && !for0)
|
|
continue;
|
|
|
|
if (while0 && tok->previous()->str() == "}") {
|
|
// find "do"
|
|
Token *tok2 = tok->previous()->link();
|
|
tok2 = tok2->previous();
|
|
if (tok2 && tok2->str() == "do") {
|
|
const bool flowmatch = Token::findmatch(tok2, "continue|break", tok) != nullptr;
|
|
// delete "do ({)"
|
|
tok2->deleteThis();
|
|
if (!flowmatch)
|
|
tok2->deleteThis();
|
|
|
|
// delete "(}) while ( 0 ) (;)"
|
|
tok = tok->previous();
|
|
tok->deleteNext(4); // while ( 0 )
|
|
if (tok->next() && tok->next()->str() == ";")
|
|
tok->deleteNext(); // ;
|
|
if (!flowmatch)
|
|
tok->deleteThis(); // }
|
|
|
|
continue;
|
|
}
|
|
}
|
|
|
|
// remove "while (0) { .. }"
|
|
if (Token::simpleMatch(tok->next()->link(), ") {")) {
|
|
Token *end = tok->next()->link(), *old_prev = tok->previous();
|
|
end = end->next()->link();
|
|
if (Token::Match(tok, "for ( %name% ="))
|
|
old_prev = end->link();
|
|
eraseDeadCode(old_prev, end->next());
|
|
if (old_prev && old_prev->next())
|
|
tok = old_prev->next();
|
|
else
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyFunctionTryCatch()
|
|
{
|
|
if (!isCPP())
|
|
return;
|
|
|
|
for (Token * tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::simpleMatch(tok, "try {"))
|
|
continue;
|
|
if (!isFunctionHead(tok->previous(), "try"))
|
|
continue;
|
|
|
|
// find the end of the last catch block
|
|
Token * const tryEndToken = tok->linkAt(1);
|
|
Token * endToken = tryEndToken;
|
|
while (Token::simpleMatch(endToken, "} catch (")) {
|
|
endToken = endToken->linkAt(2)->next();
|
|
if (!endToken)
|
|
break;
|
|
if (endToken->str() != "{") {
|
|
endToken = nullptr;
|
|
break;
|
|
}
|
|
endToken = endToken->link();
|
|
}
|
|
if (!endToken || endToken == tryEndToken)
|
|
continue;
|
|
|
|
tok->previous()->insertToken("{");
|
|
endToken->insertToken("}");
|
|
Token::createMutualLinks(tok->previous(), endToken->next());
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyErrNoInWhile()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() != "errno")
|
|
continue;
|
|
|
|
Token *endpar = nullptr;
|
|
if (Token::Match(tok->previous(), "&& errno == EINTR ) { ;| }"))
|
|
endpar = tok->tokAt(3);
|
|
else if (Token::Match(tok->tokAt(-2), "&& ( errno == EINTR ) ) { ;| }"))
|
|
endpar = tok->tokAt(4);
|
|
else
|
|
continue;
|
|
|
|
if (Token::simpleMatch(endpar->link()->previous(), "while (")) {
|
|
Token *tok1 = tok->previous();
|
|
if (tok1->str() == "(")
|
|
tok1 = tok1->previous();
|
|
|
|
// erase "&& errno == EINTR"
|
|
tok1 = tok1->previous();
|
|
Token::eraseTokens(tok1, endpar);
|
|
|
|
// tok is invalid.. move to endpar
|
|
tok = endpar;
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
void Tokenizer::simplifyFuncInWhile()
|
|
{
|
|
int count = 0;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!Token::Match(tok, "while ( %name% ( %name% ) ) {"))
|
|
continue;
|
|
|
|
Token *func = tok->tokAt(2);
|
|
const Token * const var = tok->tokAt(4);
|
|
Token * const end = tok->next()->link()->next()->link();
|
|
|
|
const int varid = ++mVarId; // Create new variable
|
|
const std::string varname("cppcheck:r" + MathLib::toString(++count));
|
|
tok->str("int");
|
|
tok->next()->insertToken(varname);
|
|
tok->tokAt(2)->varId(varid);
|
|
tok->insertToken("while");
|
|
tok->insertToken(";");
|
|
tok->insertToken(")");
|
|
tok->insertToken(var->str());
|
|
tok->next()->varId(var->varId());
|
|
tok->insertToken("(");
|
|
tok->insertToken(func->str());
|
|
tok->insertToken("=");
|
|
tok->insertToken(varname);
|
|
tok->next()->varId(varid);
|
|
Token::createMutualLinks(tok->tokAt(4), tok->tokAt(6));
|
|
end->previous()->insertToken(varname);
|
|
end->previous()->varId(varid);
|
|
end->previous()->insertToken("=");
|
|
Token::move(func, func->tokAt(3), end->previous());
|
|
end->previous()->insertToken(";");
|
|
|
|
tok = end;
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyStructDecl()
|
|
{
|
|
const bool cpp = isCPP();
|
|
|
|
// A counter that is used when giving unique names for anonymous structs.
|
|
int count = 0;
|
|
|
|
// Skip simplification of unions in class definition
|
|
std::stack<bool> skip; // true = in function, false = not in function
|
|
skip.push(false);
|
|
|
|
// Add names for anonymous structs
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!tok->isName())
|
|
continue;
|
|
// check for anonymous struct/union
|
|
if (Token::Match(tok, "struct|union {")) {
|
|
if (Token::Match(tok->next()->link(), "} const| *|&| const| %type% ,|;|[|(|{|=")) {
|
|
tok->insertToken("Anonymous" + MathLib::toString(count++));
|
|
}
|
|
}
|
|
// check for derived anonymous class/struct
|
|
else if (cpp && Token::Match(tok, "class|struct :")) {
|
|
const Token *tok1 = Token::findsimplematch(tok, "{");
|
|
if (tok1 && Token::Match(tok1->link(), "} const| *|&| const| %type% ,|;|[|(|{")) {
|
|
tok->insertToken("Anonymous" + MathLib::toString(count++));
|
|
}
|
|
}
|
|
// check for anonymous enum
|
|
else if ((Token::simpleMatch(tok, "enum {") &&
|
|
!Token::Match(tok->tokAt(-3), "using %name% =") &&
|
|
Token::Match(tok->next()->link(), "} (| %type%| )| ,|;|[|(|{")) ||
|
|
(Token::Match(tok, "enum : %type% {") && Token::Match(tok->linkAt(3), "} (| %type%| )| ,|;|[|(|{"))) {
|
|
Token *start = tok->strAt(1) == ":" ? tok->linkAt(3) : tok->linkAt(1);
|
|
if (start && Token::Match(start->next(), "( %type% )")) {
|
|
start->next()->link()->deleteThis();
|
|
start->next()->deleteThis();
|
|
}
|
|
tok->insertToken("Anonymous" + MathLib::toString(count++));
|
|
}
|
|
}
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
|
|
// check for start of scope and determine if it is in a function
|
|
if (tok->str() == "{")
|
|
skip.push(Token::Match(tok->previous(), "const|)"));
|
|
|
|
// end of scope
|
|
else if (tok->str() == "}" && !skip.empty())
|
|
skip.pop();
|
|
|
|
// check for named struct/union
|
|
else if (Token::Match(tok, "class|struct|union|enum %type% :|{")) {
|
|
Token *start = tok;
|
|
while (Token::Match(start->previous(), "%type%"))
|
|
start = start->previous();
|
|
const Token * const type = tok->next();
|
|
Token *next = tok->tokAt(2);
|
|
|
|
while (next && next->str() != "{")
|
|
next = next->next();
|
|
if (!next)
|
|
continue;
|
|
skip.push(false);
|
|
tok = next->link();
|
|
if (!tok)
|
|
break; // see #4869 segmentation fault in Tokenizer::simplifyStructDecl (invalid code)
|
|
Token *restart = next;
|
|
|
|
// check for named type
|
|
if (Token::Match(tok->next(), "const| *|&| const| (| %type% )| ,|;|[|=|(|{")) {
|
|
tok->insertToken(";");
|
|
tok = tok->next();
|
|
while (!Token::Match(start, "struct|class|union|enum")) {
|
|
tok->insertToken(start->str());
|
|
tok = tok->next();
|
|
start->deleteThis();
|
|
}
|
|
if (!tok)
|
|
break; // see #4869 segmentation fault in Tokenizer::simplifyStructDecl (invalid code)
|
|
tok->insertToken(type->str());
|
|
if (start->str() != "class") {
|
|
tok->insertToken(start->str());
|
|
tok = tok->next();
|
|
}
|
|
|
|
tok = tok->tokAt(2);
|
|
|
|
if (Token::Match(tok, "( %type% )")) {
|
|
tok->link()->deleteThis();
|
|
tok->deleteThis();
|
|
}
|
|
|
|
// check for initialization
|
|
if (tok && (tok->next()->str() == "(" || tok->next()->str() == "{")) {
|
|
tok->insertToken("=");
|
|
tok = tok->next();
|
|
|
|
if (start->str() == "enum") {
|
|
if (tok->next()->str() == "{") {
|
|
tok->next()->str("(");
|
|
tok->linkAt(1)->str(")");
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
tok = restart;
|
|
}
|
|
|
|
// check for anonymous struct/union
|
|
else if (Token::Match(tok, "struct|union {")) {
|
|
const bool inFunction = skip.top();
|
|
skip.push(false);
|
|
Token *tok1 = tok;
|
|
|
|
Token *restart = tok->next();
|
|
tok = tok->next()->link();
|
|
|
|
// unnamed anonymous struct/union so possibly remove it
|
|
if (tok && tok->next() && tok->next()->str() == ";") {
|
|
if (inFunction && tok1->str() == "union") {
|
|
// Try to create references in the union..
|
|
Token *tok2 = tok1->tokAt(2);
|
|
while (tok2) {
|
|
if (Token::Match(tok2, "%type% %name% ;"))
|
|
tok2 = tok2->tokAt(3);
|
|
else
|
|
break;
|
|
}
|
|
if (!Token::simpleMatch(tok2, "} ;"))
|
|
continue;
|
|
Token *vartok = nullptr;
|
|
tok2 = tok1->tokAt(2);
|
|
while (Token::Match(tok2, "%type% %name% ;")) {
|
|
if (!vartok) {
|
|
vartok = tok2->next();
|
|
tok2 = tok2->tokAt(3);
|
|
} else {
|
|
tok2->insertToken("&");
|
|
tok2 = tok2->tokAt(2);
|
|
tok2->insertToken(vartok->str());
|
|
tok2->next()->varId(vartok->varId());
|
|
tok2->insertToken("=");
|
|
tok2 = tok2->tokAt(4);
|
|
}
|
|
}
|
|
}
|
|
|
|
// don't remove unnamed anonymous unions from a class, struct or union
|
|
if (!(!inFunction && tok1->str() == "union") && !Token::Match(tok1->tokAt(-3), "using %name% =")) {
|
|
skip.pop();
|
|
tok1->deleteThis();
|
|
if (tok1->next() == tok) {
|
|
tok1->deleteThis();
|
|
tok = tok1;
|
|
} else
|
|
tok1->deleteThis();
|
|
restart = tok1->previous();
|
|
tok->deleteThis();
|
|
if (tok->next())
|
|
tok->deleteThis();
|
|
}
|
|
}
|
|
|
|
if (!restart) {
|
|
simplifyStructDecl();
|
|
return;
|
|
} else if (!restart->next())
|
|
return;
|
|
|
|
tok = restart;
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyCallingConvention()
|
|
{
|
|
const bool windows = mSettings->isWindowsPlatform();
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
while (Token::Match(tok, "__cdecl|__stdcall|__fastcall|__thiscall|__clrcall|__syscall|__pascal|__fortran|__far|__near") || (windows && Token::Match(tok, "WINAPI|APIENTRY|CALLBACK"))) {
|
|
tok->deleteThis();
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyDeclspec()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
while (Token::Match(tok, "__declspec|_declspec (") && tok->next()->link() && tok->next()->link()->next()) {
|
|
if (Token::Match(tok->tokAt(2), "noreturn|nothrow")) {
|
|
Token *tok1 = tok->next()->link()->next();
|
|
while (tok1 && !Token::Match(tok1, "%name%")) {
|
|
tok1 = tok1->next();
|
|
}
|
|
if (tok1) {
|
|
if (tok->strAt(2) == "noreturn")
|
|
tok1->isAttributeNoreturn(true);
|
|
else
|
|
tok1->isAttributeNothrow(true);
|
|
}
|
|
} else if (tok->strAt(2) == "property")
|
|
tok->next()->link()->insertToken("__property");
|
|
|
|
Token::eraseTokens(tok, tok->next()->link()->next());
|
|
tok->deleteThis();
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyAttribute()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "%type% (") && !mSettings->library.isNotLibraryFunction(tok)) {
|
|
if (mSettings->library.isFunctionConst(tok->str(), true))
|
|
tok->isAttributePure(true);
|
|
if (mSettings->library.isFunctionConst(tok->str(), false))
|
|
tok->isAttributeConst(true);
|
|
}
|
|
while (Token::Match(tok, "__attribute__|__attribute (")) {
|
|
Token *after = tok;
|
|
while (Token::Match(after, "__attribute__|__attribute ("))
|
|
after = after->linkAt(1)->next();
|
|
if (!after)
|
|
syntaxError(tok);
|
|
|
|
Token *functok = nullptr;
|
|
if (Token::Match(after, "%name%|*")) {
|
|
Token *ftok = after;
|
|
while (Token::Match(ftok, "%name%|::|<|* !!(")) {
|
|
if (ftok->str() == "<") {
|
|
ftok = ftok->findClosingBracket();
|
|
if (!ftok)
|
|
break;
|
|
}
|
|
ftok = ftok->next();
|
|
}
|
|
if (Token::Match(ftok, "%name% ("))
|
|
functok = ftok;
|
|
} else if (Token::Match(after, "[;{=:]")) {
|
|
Token *prev = tok->previous();
|
|
while (Token::Match(prev, "%name%"))
|
|
prev = prev->previous();
|
|
if (Token::simpleMatch(prev, ")") && Token::Match(prev->link()->previous(), "%name% ("))
|
|
functok = prev->link()->previous();
|
|
else if ((!prev || Token::Match(prev, "[;{}*]")) && Token::Match(tok->previous(), "%name%"))
|
|
functok = tok->previous();
|
|
}
|
|
|
|
for (Token *attr = tok->tokAt(2); attr->str() != ")"; attr = attr->next()) {
|
|
if (Token::Match(attr, "%name% ("))
|
|
attr = attr->linkAt(1);
|
|
|
|
if (Token::Match(attr, "[(,] constructor|__constructor__ [,()]")) {
|
|
if (!functok)
|
|
syntaxError(tok);
|
|
functok->isAttributeConstructor(true);
|
|
}
|
|
|
|
else if (Token::Match(attr, "[(,] destructor|__destructor__ [,()]")) {
|
|
if (!functok)
|
|
syntaxError(tok);
|
|
functok->isAttributeDestructor(true);
|
|
}
|
|
|
|
else if (Token::Match(attr, "[(,] unused|__unused__|used|__used__ [,)]")) {
|
|
Token *vartok = nullptr;
|
|
|
|
// check if after variable name
|
|
if (Token::Match(after, ";|=")) {
|
|
if (Token::Match(tok->previous(), "%type%"))
|
|
vartok = tok->previous();
|
|
}
|
|
|
|
// check if before variable name
|
|
else if (Token::Match(after, "%type%"))
|
|
vartok = after;
|
|
|
|
if (vartok) {
|
|
const std::string &attribute(attr->next()->str());
|
|
if (attribute.find("unused") != std::string::npos)
|
|
vartok->isAttributeUnused(true);
|
|
else
|
|
vartok->isAttributeUsed(true);
|
|
}
|
|
}
|
|
|
|
else if (Token::Match(attr, "[(,] pure|__pure__|const|__const__|noreturn|__noreturn__|nothrow|__nothrow__|warn_unused_result [,)]")) {
|
|
if (!functok)
|
|
syntaxError(tok);
|
|
|
|
const std::string &attribute(attr->next()->str());
|
|
if (attribute.find("pure") != std::string::npos)
|
|
functok->isAttributePure(true);
|
|
else if (attribute.find("const") != std::string::npos)
|
|
functok->isAttributeConst(true);
|
|
else if (attribute.find("noreturn") != std::string::npos)
|
|
functok->isAttributeNoreturn(true);
|
|
else if (attribute.find("nothrow") != std::string::npos)
|
|
functok->isAttributeNothrow(true);
|
|
else if (attribute.find("warn_unused_result") != std::string::npos)
|
|
functok->isAttributeNodiscard(true);
|
|
}
|
|
|
|
else if (Token::Match(attr, "[(,] packed [,)]") && Token::simpleMatch(tok->previous(), "}"))
|
|
tok->previous()->isAttributePacked(true);
|
|
}
|
|
|
|
Token::eraseTokens(tok, tok->linkAt(1)->next());
|
|
tok->deleteThis();
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyCppcheckAttribute()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() != "(")
|
|
continue;
|
|
if (!tok->previous())
|
|
continue;
|
|
const std::string &attr = tok->previous()->str();
|
|
if (attr.compare(0, 11, "__cppcheck_") != 0) // TODO: starts_with("__cppcheck_")
|
|
continue;
|
|
if (attr.compare(attr.size()-2, 2, "__") != 0) // TODO: ends_with("__")
|
|
continue;
|
|
|
|
Token *vartok = tok->link();
|
|
while (Token::Match(vartok->next(), "%name%|*|&|::")) {
|
|
vartok = vartok->next();
|
|
if (Token::Match(vartok, "%name% (") && vartok->str().compare(0,11,"__cppcheck_") == 0)
|
|
vartok = vartok->linkAt(1);
|
|
}
|
|
|
|
if (vartok->isName()) {
|
|
if (Token::Match(tok->previous(), "__cppcheck_low__ ( %num% )"))
|
|
vartok->setCppcheckAttribute(TokenImpl::CppcheckAttributes::Type::LOW, MathLib::toLongNumber(tok->next()->str()));
|
|
else if (Token::Match(tok->previous(), "__cppcheck_high__ ( %num% )"))
|
|
vartok->setCppcheckAttribute(TokenImpl::CppcheckAttributes::Type::HIGH, MathLib::toLongNumber(tok->next()->str()));
|
|
}
|
|
|
|
// Delete cppcheck attribute..
|
|
if (tok->tokAt(-2)) {
|
|
tok = tok->tokAt(-2);
|
|
Token::eraseTokens(tok, tok->linkAt(2)->next());
|
|
} else {
|
|
tok = tok->previous();
|
|
Token::eraseTokens(tok, tok->linkAt(1)->next());
|
|
tok->str(";");
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyCPPAttribute()
|
|
{
|
|
if (mSettings->standards.cpp < Standards::CPP11 || isC())
|
|
return;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!isCPPAttribute(tok) && !isAlignAttribute(tok)) {
|
|
continue;
|
|
}
|
|
if (isCPPAttribute(tok)) {
|
|
if (Token::findsimplematch(tok->tokAt(2), "noreturn", tok->link())) {
|
|
const Token * head = skipCPPOrAlignAttribute(tok);
|
|
while (isCPPAttribute(head) || isAlignAttribute(head))
|
|
head = skipCPPOrAlignAttribute(head);
|
|
head = head->next();
|
|
while (Token::Match(head, "%name%|::|*|&|<|>|,")) // skip return type
|
|
head = head->next();
|
|
if (head && head->str() == "(" && isFunctionHead(head, "{|;")) {
|
|
head->previous()->isAttributeNoreturn(true);
|
|
}
|
|
} else if (Token::findsimplematch(tok->tokAt(2), "nodiscard", tok->link())) {
|
|
const Token * head = skipCPPOrAlignAttribute(tok);
|
|
while (isCPPAttribute(head) || isAlignAttribute(head))
|
|
head = skipCPPOrAlignAttribute(head);
|
|
head = head->next();
|
|
while (Token::Match(head, "%name%|::|*|&|<|>|,"))
|
|
head = head->next();
|
|
if (head && head->str() == "(" && isFunctionHead(head, "{|;")) {
|
|
head->previous()->isAttributeNodiscard(true);
|
|
}
|
|
} else if (Token::findsimplematch(tok->tokAt(2), "maybe_unused", tok->link())) {
|
|
const Token* head = skipCPPOrAlignAttribute(tok);
|
|
while (isCPPAttribute(head) || isAlignAttribute(head))
|
|
head = skipCPPOrAlignAttribute(head);
|
|
head->next()->isAttributeMaybeUnused(true);
|
|
} else if (Token::Match(tok->previous(), ") [ [ expects|ensures|assert default|audit|axiom| : %name% <|<=|>|>= %num% ] ]")) {
|
|
const Token *vartok = tok->tokAt(4);
|
|
if (vartok->str() == ":")
|
|
vartok = vartok->next();
|
|
Token *argtok = tok->tokAt(-2);
|
|
while (argtok && argtok->str() != "(") {
|
|
if (argtok->str() == vartok->str())
|
|
break;
|
|
if (argtok->str() == ")")
|
|
argtok = argtok->link();
|
|
argtok = argtok->previous();
|
|
}
|
|
if (argtok && argtok->str() == vartok->str()) {
|
|
if (vartok->next()->str() == ">=")
|
|
argtok->setCppcheckAttribute(TokenImpl::CppcheckAttributes::Type::LOW, MathLib::toLongNumber(vartok->strAt(2)));
|
|
else if (vartok->next()->str() == ">")
|
|
argtok->setCppcheckAttribute(TokenImpl::CppcheckAttributes::Type::LOW, MathLib::toLongNumber(vartok->strAt(2))+1);
|
|
else if (vartok->next()->str() == "<=")
|
|
argtok->setCppcheckAttribute(TokenImpl::CppcheckAttributes::Type::HIGH, MathLib::toLongNumber(vartok->strAt(2)));
|
|
else if (vartok->next()->str() == "<")
|
|
argtok->setCppcheckAttribute(TokenImpl::CppcheckAttributes::Type::HIGH, MathLib::toLongNumber(vartok->strAt(2))-1);
|
|
}
|
|
}
|
|
} else {
|
|
if (Token::simpleMatch(tok, "alignas (")) {
|
|
// alignment requirements could be checked here
|
|
}
|
|
}
|
|
Token::eraseTokens(tok, skipCPPOrAlignAttribute(tok)->next());
|
|
// fix iterator after removing
|
|
if (tok->previous()) {
|
|
tok = tok->previous();
|
|
tok->next()->deleteThis();
|
|
} else {
|
|
tok->deleteThis();
|
|
tok = list.front();
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::removeAlignas()
|
|
{
|
|
if (!isCPP() || mSettings->standards.cpp < Standards::CPP11)
|
|
return;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "[;{}] alignas (") && Token::Match(tok->linkAt(2), ") %name%"))
|
|
Token::eraseTokens(tok, tok->linkAt(2)->next());
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifySpaceshipOperator()
|
|
{
|
|
if (isCPP() && mSettings->standards.cpp >= Standards::CPP20) {
|
|
for (Token *tok = list.front(); tok && tok->next(); tok = tok->next()) {
|
|
if (Token::simpleMatch(tok, "<= >")) {
|
|
tok->str("<=>");
|
|
tok->deleteNext();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
static const std::unordered_set<std::string> keywords = {
|
|
"inline"
|
|
, "_inline"
|
|
, "__inline"
|
|
, "__forceinline"
|
|
, "register"
|
|
, "__restrict"
|
|
, "__restrict__"
|
|
, "__thread"
|
|
};
|
|
// Remove "inline", "register", "restrict", "override", "static" and "constexpr"
|
|
// "restrict" keyword
|
|
// - New to 1999 ANSI/ISO C standard
|
|
// - Not in C++ standard yet
|
|
void Tokenizer::simplifyKeyword()
|
|
{
|
|
// FIXME: There is a risk that "keywords" are removed by mistake. This
|
|
// code should be fixed so it doesn't remove variables etc. Nonstandard
|
|
// keywords should be defined with a library instead. For instance the
|
|
// linux kernel code at least uses "_inline" as struct member name at some
|
|
// places.
|
|
|
|
const bool c99 = isC() && mSettings->standards.c >= Standards::C99;
|
|
const bool cpp11 = isCPP() && mSettings->standards.cpp >= Standards::CPP11;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (keywords.find(tok->str()) != keywords.end()) {
|
|
// Don't remove struct members
|
|
if (!Token::simpleMatch(tok->previous(), ".")) {
|
|
if (tok->str().find("inline") != std::string::npos) {
|
|
Token *temp = tok->next();
|
|
while (temp != nullptr && Token::Match(temp, "%name%")) {
|
|
temp->isInline(true);
|
|
temp = temp->next();
|
|
}
|
|
}
|
|
tok->deleteThis(); // Simplify..
|
|
}
|
|
}
|
|
|
|
if (isC() || mSettings->standards.cpp == Standards::CPP03) {
|
|
if (tok->str() == "auto")
|
|
tok->deleteThis();
|
|
}
|
|
|
|
// simplify static keyword:
|
|
// void foo( int [ static 5 ] ); ==> void foo( int [ 5 ] );
|
|
if (Token::Match(tok, "[ static %num%"))
|
|
tok->deleteNext();
|
|
|
|
if (c99) {
|
|
while (tok->str() == "restrict")
|
|
tok->deleteThis();
|
|
|
|
if (mSettings->standards.c >= Standards::C11) {
|
|
while (tok->str() == "_Atomic")
|
|
tok->deleteThis();
|
|
}
|
|
}
|
|
|
|
else if (cpp11) {
|
|
// final:
|
|
// 1) struct name final { }; <- struct is final
|
|
if (Token::Match(tok->previous(), "struct|class|union %type% final [:{]")) {
|
|
tok->deleteNext();
|
|
}
|
|
|
|
// noexcept -> noexcept(true)
|
|
// 2) void f() noexcept; -> void f() noexcept(true);
|
|
else if (Token::Match(tok, ") noexcept :|{|;|const|override|final")) {
|
|
// Insertion is done in inverse order
|
|
// The brackets are linked together accordingly afterwards
|
|
Token * tokNoExcept = tok->next();
|
|
tokNoExcept->insertToken(")");
|
|
Token * braceEnd = tokNoExcept->next();
|
|
tokNoExcept->insertToken("true");
|
|
tokNoExcept->insertToken("(");
|
|
Token * braceStart = tokNoExcept->next();
|
|
tok = tok->tokAt(3);
|
|
Token::createMutualLinks(braceStart, braceEnd);
|
|
}
|
|
|
|
// 3) thread_local -> static
|
|
// on single thread thread_local has the effect of static
|
|
else if (tok->str() == "thread_local") {
|
|
tok->originalName(tok->str());
|
|
tok->str("static");
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyAssignmentInFunctionCall()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() == "(")
|
|
tok = tok->link();
|
|
|
|
// Find 'foo(var='. Exclude 'assert(var=' to allow tests to check that assert(...) does not contain side-effects
|
|
else if (Token::Match(tok, "[;{}] %name% ( %name% =") &&
|
|
Token::simpleMatch(tok->linkAt(2), ") ;") &&
|
|
!Token::Match(tok->next(), "assert|while")) {
|
|
const std::string& funcname(tok->next()->str());
|
|
Token* const vartok = tok->tokAt(3);
|
|
|
|
// Goto ',' or ')'..
|
|
for (Token *tok2 = vartok->tokAt(2); tok2; tok2 = tok2->next()) {
|
|
if (tok2->link() && Token::Match(tok2, "(|[|{"))
|
|
tok2 = tok2->link();
|
|
else if (tok2->str() == ";")
|
|
break;
|
|
else if (Token::Match(tok2, ")|,")) {
|
|
tok2 = tok2->previous();
|
|
|
|
tok2->insertToken(vartok->str());
|
|
tok2->next()->varId(vartok->varId());
|
|
|
|
tok2->insertToken("(");
|
|
Token::createMutualLinks(tok2->next(), tok->linkAt(2));
|
|
|
|
tok2->insertToken(funcname);
|
|
tok2->insertToken(";");
|
|
|
|
Token::eraseTokens(tok, vartok);
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyAssignmentBlock()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "[;{}] %name% = ( {")) {
|
|
const std::string &varname = tok->next()->str();
|
|
|
|
// goto the "} )"
|
|
int indentlevel = 0;
|
|
Token *tok2 = tok;
|
|
while (nullptr != (tok2 = tok2->next())) {
|
|
if (Token::Match(tok2, "(|{"))
|
|
++indentlevel;
|
|
else if (Token::Match(tok2, ")|}")) {
|
|
if (indentlevel <= 2)
|
|
break;
|
|
--indentlevel;
|
|
} else if (indentlevel == 2 && tok2->str() == varname && Token::Match(tok2->previous(), "%type%|*"))
|
|
// declaring variable in inner scope with same name as lhs variable
|
|
break;
|
|
}
|
|
if (indentlevel == 2 && Token::simpleMatch(tok2, "} )")) {
|
|
tok2 = tok2->tokAt(-3);
|
|
if (Token::Match(tok2, "[;{}] %num%|%name% ;")) {
|
|
tok2->insertToken("=");
|
|
tok2->insertToken(tok->next()->str());
|
|
tok2->next()->varId(tok->next()->varId());
|
|
tok->deleteNext(3);
|
|
tok2->tokAt(5)->deleteNext();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// Remove __asm..
|
|
void Tokenizer::simplifyAsm()
|
|
{
|
|
std::string instruction;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "__asm|_asm|asm {") &&
|
|
tok->next()->link()->next()) {
|
|
instruction = tok->tokAt(2)->stringifyList(tok->next()->link());
|
|
Token::eraseTokens(tok, tok->next()->link()->next());
|
|
}
|
|
|
|
else if (Token::Match(tok, "asm|__asm|__asm__ volatile|__volatile|__volatile__| (")) {
|
|
// Goto "("
|
|
Token *partok = tok->next();
|
|
if (partok->str() != "(")
|
|
partok = partok->next();
|
|
instruction = partok->next()->stringifyList(partok->link());
|
|
Token::eraseTokens(tok, partok->link()->next());
|
|
}
|
|
|
|
else if (Token::Match(tok, "_asm|__asm")) {
|
|
Token *endasm = tok->next();
|
|
const Token *firstSemiColon = nullptr;
|
|
int comment = 0;
|
|
while (Token::Match(endasm, "%num%|%name%|,|:|;") || (endasm && endasm->linenr() == comment)) {
|
|
if (Token::Match(endasm, "_asm|__asm|__endasm"))
|
|
break;
|
|
if (endasm->str() == ";") {
|
|
comment = endasm->linenr();
|
|
if (!firstSemiColon)
|
|
firstSemiColon = endasm;
|
|
}
|
|
endasm = endasm->next();
|
|
}
|
|
if (Token::simpleMatch(endasm, "__endasm")) {
|
|
instruction = tok->next()->stringifyList(endasm);
|
|
Token::eraseTokens(tok, endasm->next());
|
|
if (!Token::simpleMatch(tok->next(), ";"))
|
|
tok->insertToken(";");
|
|
} else if (firstSemiColon) {
|
|
instruction = tok->next()->stringifyList(firstSemiColon);
|
|
Token::eraseTokens(tok, firstSemiColon);
|
|
} else if (!endasm) {
|
|
instruction = tok->next()->stringifyList(endasm);
|
|
Token::eraseTokens(tok, endasm);
|
|
tok->insertToken(";");
|
|
} else
|
|
continue;
|
|
}
|
|
|
|
else
|
|
continue;
|
|
|
|
// insert "asm ( "instruction" )"
|
|
tok->str("asm");
|
|
if (tok->strAt(1) != ";" && tok->strAt(1) != "{")
|
|
tok->insertToken(";");
|
|
tok->insertToken(")");
|
|
tok->insertToken("\"" + instruction + "\"");
|
|
tok->insertToken("(");
|
|
|
|
tok = tok->next();
|
|
Token::createMutualLinks(tok, tok->tokAt(2));
|
|
|
|
//move the new tokens in the same line as ";" if available
|
|
tok = tok->tokAt(2);
|
|
if (tok->next() && tok->next()->str() == ";" &&
|
|
tok->next()->linenr() != tok->linenr()) {
|
|
const int endposition = tok->next()->linenr();
|
|
tok = tok->tokAt(-3);
|
|
for (int i = 0; i < 4; ++i) {
|
|
tok = tok->next();
|
|
tok->linenr(endposition);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyAsm2()
|
|
{
|
|
// Block declarations: ^{}
|
|
// A C extension used to create lambda like closures.
|
|
|
|
// Put ^{} statements in asm()
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() != "^")
|
|
continue;
|
|
|
|
if (Token::simpleMatch(tok, "^ {") || (Token::simpleMatch(tok->linkAt(1), ") {") && tok->strAt(-1) != "operator")) {
|
|
Token * start = tok;
|
|
while (start && !Token::Match(start, "[,(;{}=]")) {
|
|
if (start->link() && Token::Match(start, ")|]|>"))
|
|
start = start->link();
|
|
start = start->previous();
|
|
}
|
|
|
|
const Token *last = tok->next()->link();
|
|
if (Token::simpleMatch(last, ") {"))
|
|
last = last->linkAt(1);
|
|
last = last->next();
|
|
while (last && !Token::Match(last, "%cop%|,|;|{|}|)")) {
|
|
if (Token::Match(last, "(|["))
|
|
last = last->link();
|
|
last = last->next();
|
|
}
|
|
|
|
if (start && last) {
|
|
std::string asmcode;
|
|
while (start->next() != last) {
|
|
asmcode += start->next()->str();
|
|
start->deleteNext();
|
|
}
|
|
if (last->str() == "}")
|
|
start->insertToken(";");
|
|
start->insertToken(")");
|
|
start->insertToken("\"" + asmcode + "\"");
|
|
start->insertToken("(");
|
|
start->insertToken("asm");
|
|
start->tokAt(2)->link(start->tokAt(4));
|
|
start->tokAt(4)->link(start->tokAt(2));
|
|
tok = start->tokAt(4);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyAt()
|
|
{
|
|
std::set<std::string> var;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "%name%|] @ %num%|%name%|(")) {
|
|
const Token *end = tok->tokAt(2);
|
|
if (end->isNumber())
|
|
end = end->next();
|
|
else if (end->str() == "(") {
|
|
int par = 0;
|
|
while ((end = end->next()) != nullptr) {
|
|
if (end->str() == "(")
|
|
par++;
|
|
else if (end->str() == ")") {
|
|
if (--par < 0)
|
|
break;
|
|
}
|
|
}
|
|
end = end ? end->next() : nullptr;
|
|
} else if (var.find(end->str()) != var.end())
|
|
end = end->next();
|
|
else
|
|
continue;
|
|
|
|
if (Token::Match(end, ": %num% ;"))
|
|
end = end->tokAt(2);
|
|
|
|
if (end && end->str() == ";") {
|
|
if (tok->isName())
|
|
var.insert(tok->str());
|
|
tok->isAtAddress(true);
|
|
Token::eraseTokens(tok, end);
|
|
}
|
|
}
|
|
|
|
// keywords in compiler from cosmic software for STM8
|
|
// TODO: Should use platform configuration.
|
|
if (Token::Match(tok, "@ builtin|eeprom|far|inline|interrupt|near|noprd|nostack|nosvf|packed|stack|svlreg|tiny|vector")) {
|
|
tok->str(tok->next()->str() + "@");
|
|
tok->deleteNext();
|
|
}
|
|
}
|
|
}
|
|
|
|
// Simplify bitfields
|
|
void Tokenizer::simplifyBitfields()
|
|
{
|
|
bool goback = false;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (goback) {
|
|
goback = false;
|
|
tok = tok->previous();
|
|
}
|
|
Token *last = nullptr;
|
|
|
|
if (Token::simpleMatch(tok, "for ("))
|
|
tok = tok->linkAt(1);
|
|
|
|
if (!Token::Match(tok, ";|{|}|public:|protected:|private:"))
|
|
continue;
|
|
|
|
bool isEnum = false;
|
|
if (tok->str() == "}") {
|
|
const Token *type = tok->link()->previous();
|
|
while (type && type->isName()) {
|
|
if (type->str() == "enum") {
|
|
isEnum = true;
|
|
break;
|
|
}
|
|
type = type->previous();
|
|
}
|
|
}
|
|
|
|
if (Token::Match(tok->next(), "const| %type% %name% :") &&
|
|
!Token::Match(tok->next(), "case|public|protected|private|class|struct") &&
|
|
!Token::simpleMatch(tok->tokAt(2), "default :")) {
|
|
Token *tok1 = (tok->next()->str() == "const") ? tok->tokAt(3) : tok->tokAt(2);
|
|
if (Token::Match(tok1, "%name% : %num% [;=]"))
|
|
tok1->setBits(MathLib::toLongNumber(tok1->strAt(2)));
|
|
if (tok1 && tok1->tokAt(2) &&
|
|
(Token::Match(tok1->tokAt(2), "%bool%|%num%") ||
|
|
!Token::Match(tok1->tokAt(2), "public|protected|private| %type% ::|<|,|{|;"))) {
|
|
while (tok1->next() && !Token::Match(tok1->next(), "[;,)]{}=]")) {
|
|
if (Token::Match(tok1->next(), "[([]"))
|
|
Token::eraseTokens(tok1, tok1->next()->link());
|
|
tok1->deleteNext();
|
|
}
|
|
|
|
last = tok1->next();
|
|
}
|
|
} else if (isEnum && Token::Match(tok, "} %name%| : %num% ;")) {
|
|
if (tok->next()->str() == ":") {
|
|
tok->deleteNext(2);
|
|
tok->insertToken("Anonymous");
|
|
} else {
|
|
tok->next()->deleteNext(2);
|
|
}
|
|
} else if (Token::Match(tok->next(), "const| %type% : %num%|%bool% ;") &&
|
|
tok->next()->str() != "default") {
|
|
const int offset = (tok->next()->str() == "const") ? 1 : 0;
|
|
if (!Token::Match(tok->tokAt(3 + offset), "[{};()]")) {
|
|
tok->deleteNext(4 + offset);
|
|
goback = true;
|
|
}
|
|
}
|
|
|
|
if (last && last->str() == ",") {
|
|
Token * tok1 = last;
|
|
tok1->str(";");
|
|
|
|
const Token *const tok2 = tok->next();
|
|
tok1->insertToken(tok2->str());
|
|
tok1 = tok1->next();
|
|
tok1->isSigned(tok2->isSigned());
|
|
tok1->isUnsigned(tok2->isUnsigned());
|
|
tok1->isLong(tok2->isLong());
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
// Types and objects in std namespace that are neither functions nor templates
|
|
static const std::set<std::string> stdTypes = {
|
|
"string", "wstring", "u16string", "u32string",
|
|
"iostream", "ostream", "ofstream", "ostringstream",
|
|
"istream", "ifstream", "istringstream", "fstream", "stringstream",
|
|
"wstringstream", "wistringstream", "wostringstream", "wstringbuf",
|
|
"stringbuf", "streambuf", "ios", "filebuf", "ios_base",
|
|
"exception", "bad_exception", "bad_alloc",
|
|
"logic_error", "domain_error", "invalid_argument_", "length_error",
|
|
"out_of_range", "runtime_error", "range_error", "overflow_error", "underflow_error",
|
|
"locale",
|
|
"cout", "cerr", "clog", "cin",
|
|
"wcerr", "wcin", "wclog", "wcout",
|
|
"endl", "ends", "flush",
|
|
"boolalpha", "noboolalpha", "showbase", "noshowbase",
|
|
"showpoint", "noshowpoint", "showpos", "noshowpos",
|
|
"skipws", "noskipws", "unitbuf", "nounitbuf", "uppercase", "nouppercase",
|
|
"dec", "hex", "oct",
|
|
"fixed", "scientific",
|
|
"internal", "left", "right",
|
|
"fpos", "streamoff", "streampos", "streamsize"
|
|
};
|
|
|
|
static const std::set<std::string> stdTemplates = {
|
|
"array", "basic_string", "bitset", "deque", "list", "map", "multimap",
|
|
"priority_queue", "queue", "set", "multiset", "stack", "vector", "pair",
|
|
"iterator", "iterator_traits",
|
|
"unordered_map", "unordered_multimap", "unordered_set", "unordered_multiset",
|
|
"tuple", "function"
|
|
};
|
|
static const std::set<std::string> stdFunctions = {
|
|
"getline",
|
|
"for_each", "find", "find_if", "find_end", "find_first_of",
|
|
"adjacent_find", "count", "count_if", "mismatch", "equal", "search", "search_n",
|
|
"copy", "copy_backward", "swap", "swap_ranges", "iter_swap", "transform", "replace",
|
|
"replace_if", "replace_copy", "replace_copy_if", "fill", "fill_n", "generate", "generate_n", "remove",
|
|
"remove_if", "remove_copy", "remove_copy_if",
|
|
"unique", "unique_copy", "reverse", "reverse_copy",
|
|
"rotate", "rotate_copy", "random_shuffle", "partition", "stable_partition",
|
|
"sort", "stable_sort", "partial_sort", "partial_sort_copy", "nth_element",
|
|
"lower_bound", "upper_bound", "equal_range", "binary_search", "merge", "inplace_merge", "includes",
|
|
"set_union", "set_intersection", "set_difference",
|
|
"set_symmetric_difference", "push_heap", "pop_heap", "make_heap", "sort_heap",
|
|
"min", "max", "min_element", "max_element", "lexicographical_compare", "next_permutation", "prev_permutation",
|
|
"advance", "back_inserter", "distance", "front_inserter", "inserter",
|
|
"make_pair", "make_shared", "make_tuple"
|
|
};
|
|
|
|
|
|
// Add std:: in front of std classes, when using namespace std; was given
|
|
void Tokenizer::simplifyNamespaceStd()
|
|
{
|
|
if (!isCPP())
|
|
return;
|
|
|
|
const bool isCPP11 = mSettings->standards.cpp == Standards::CPP11;
|
|
|
|
std::set<std::string> userFunctions;
|
|
|
|
for (const Token* tok = Token::findsimplematch(list.front(), "using namespace std ;"); tok; tok = tok->next()) {
|
|
bool insert = false;
|
|
if (Token::Match(tok, "enum class|struct| %name%| :|{")) { // Don't replace within enum definitions
|
|
skipEnumBody(&tok);
|
|
}
|
|
if (!Token::Match(tok->previous(), ".|::")) {
|
|
if (Token::Match(tok, "%name% (")) {
|
|
if (isFunctionHead(tok->next(), "{"))
|
|
userFunctions.insert(tok->str());
|
|
else if (isFunctionHead(tok->next(), ";")) {
|
|
const Token *start = tok;
|
|
while (Token::Match(start->previous(), "%type%|*|&"))
|
|
start = start->previous();
|
|
if (start != tok && start->isName() && (!start->previous() || Token::Match(start->previous(), "[;{}]")))
|
|
userFunctions.insert(tok->str());
|
|
}
|
|
if (userFunctions.find(tok->str()) == userFunctions.end() && stdFunctions.find(tok->str()) != stdFunctions.end())
|
|
insert = true;
|
|
} else if (Token::Match(tok, "%name% <") && stdTemplates.find(tok->str()) != stdTemplates.end())
|
|
insert = true;
|
|
else if (tok->isName() && !tok->varId() && !Token::Match(tok->next(), "(|<") && stdTypes.find(tok->str()) != stdTypes.end())
|
|
insert = true;
|
|
}
|
|
|
|
if (insert) {
|
|
tok->previous()->insertToken("std");
|
|
tok->previous()->linenr(tok->linenr()); // For stylistic reasons we put the std:: in the same line as the following token
|
|
tok->previous()->fileIndex(tok->fileIndex());
|
|
tok->previous()->insertToken("::");
|
|
} else if (isCPP11 && Token::Match(tok, "!!:: tr1 ::"))
|
|
tok->next()->str("std");
|
|
}
|
|
|
|
for (Token* tok = list.front(); tok; tok = tok->next()) {
|
|
if (isCPP11 && Token::simpleMatch(tok, "std :: tr1 ::"))
|
|
Token::eraseTokens(tok, tok->tokAt(3));
|
|
|
|
else if (Token::simpleMatch(tok, "using namespace std ;")) {
|
|
Token::eraseTokens(tok, tok->tokAt(4));
|
|
tok->deleteThis();
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
void Tokenizer::simplifyMicrosoftMemoryFunctions()
|
|
{
|
|
// skip if not Windows
|
|
if (!mSettings->isWindowsPlatform())
|
|
return;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->strAt(1) != "(")
|
|
continue;
|
|
|
|
if (Token::Match(tok, "CopyMemory|RtlCopyMemory|RtlCopyBytes")) {
|
|
tok->str("memcpy");
|
|
} else if (Token::Match(tok, "MoveMemory|RtlMoveMemory")) {
|
|
tok->str("memmove");
|
|
} else if (Token::Match(tok, "FillMemory|RtlFillMemory|RtlFillBytes")) {
|
|
// FillMemory(dst, len, val) -> memset(dst, val, len)
|
|
tok->str("memset");
|
|
|
|
Token *tok1 = tok->tokAt(2);
|
|
if (tok1)
|
|
tok1 = tok1->nextArgument(); // Second argument
|
|
if (tok1) {
|
|
Token *tok2 = tok1->nextArgument(); // Third argument
|
|
|
|
if (tok2)
|
|
Token::move(tok1->previous(), tok2->tokAt(-2), tok->next()->link()->previous()); // Swap third with second argument
|
|
}
|
|
} else if (Token::Match(tok, "ZeroMemory|RtlZeroMemory|RtlZeroBytes|RtlSecureZeroMemory")) {
|
|
// ZeroMemory(dst, len) -> memset(dst, 0, len)
|
|
tok->str("memset");
|
|
|
|
Token *tok1 = tok->tokAt(2);
|
|
if (tok1)
|
|
tok1 = tok1->nextArgument(); // Second argument
|
|
|
|
if (tok1) {
|
|
tok1 = tok1->previous();
|
|
tok1->insertToken("0");
|
|
tok1 = tok1->next();
|
|
tok1->insertToken(",");
|
|
}
|
|
} else if (Token::simpleMatch(tok, "RtlCompareMemory")) {
|
|
// RtlCompareMemory(src1, src2, len) -> memcmp(src1, src2, len)
|
|
tok->str("memcmp");
|
|
// For the record, when memcmp returns 0, both strings are equal.
|
|
// When RtlCompareMemory returns len, both strings are equal.
|
|
// It might be needed to improve this replacement by something
|
|
// like ((len - memcmp(src1, src2, len)) % (len + 1)) to
|
|
// respect execution path (if required)
|
|
}
|
|
}
|
|
}
|
|
|
|
namespace {
|
|
struct triplet {
|
|
triplet(const char* m, const char* u) : mbcs(m), unicode(u) {}
|
|
std::string mbcs, unicode;
|
|
};
|
|
|
|
const std::map<std::string, triplet> apis = {
|
|
std::make_pair("_topen", triplet("open", "_wopen")),
|
|
std::make_pair("_tsopen_s", triplet("_sopen_s", "_wsopen_s")),
|
|
std::make_pair("_tfopen", triplet("fopen", "_wfopen")),
|
|
std::make_pair("_tfopen_s", triplet("fopen_s", "_wfopen_s")),
|
|
std::make_pair("_tfreopen", triplet("freopen", "_wfreopen")),
|
|
std::make_pair("_tfreopen_s", triplet("freopen_s", "_wfreopen_s")),
|
|
std::make_pair("_tcscat", triplet("strcat", "wcscat")),
|
|
std::make_pair("_tcschr", triplet("strchr", "wcschr")),
|
|
std::make_pair("_tcscmp", triplet("strcmp", "wcscmp")),
|
|
std::make_pair("_tcsdup", triplet("strdup", "wcsdup")),
|
|
std::make_pair("_tcscpy", triplet("strcpy", "wcscpy")),
|
|
std::make_pair("_tcslen", triplet("strlen", "wcslen")),
|
|
std::make_pair("_tcsncat", triplet("strncat", "wcsncat")),
|
|
std::make_pair("_tcsncpy", triplet("strncpy", "wcsncpy")),
|
|
std::make_pair("_tcsnlen", triplet("strnlen", "wcsnlen")),
|
|
std::make_pair("_tcsrchr", triplet("strrchr", "wcsrchr")),
|
|
std::make_pair("_tcsstr", triplet("strstr", "wcsstr")),
|
|
std::make_pair("_tcstok", triplet("strtok", "wcstok")),
|
|
std::make_pair("_ftprintf", triplet("fprintf", "fwprintf")),
|
|
std::make_pair("_tprintf", triplet("printf", "wprintf")),
|
|
std::make_pair("_stprintf", triplet("sprintf", "swprintf")),
|
|
std::make_pair("_sntprintf", triplet("_snprintf", "_snwprintf")),
|
|
std::make_pair("_ftscanf", triplet("fscanf", "fwscanf")),
|
|
std::make_pair("_tscanf", triplet("scanf", "wscanf")),
|
|
std::make_pair("_stscanf", triplet("sscanf", "swscanf")),
|
|
std::make_pair("_ftprintf_s", triplet("fprintf_s", "fwprintf_s")),
|
|
std::make_pair("_tprintf_s", triplet("printf_s", "wprintf_s")),
|
|
std::make_pair("_stprintf_s", triplet("sprintf_s", "swprintf_s")),
|
|
std::make_pair("_sntprintf_s", triplet("_snprintf_s", "_snwprintf_s")),
|
|
std::make_pair("_ftscanf_s", triplet("fscanf_s", "fwscanf_s")),
|
|
std::make_pair("_tscanf_s", triplet("scanf_s", "wscanf_s")),
|
|
std::make_pair("_stscanf_s", triplet("sscanf_s", "swscanf_s"))
|
|
};
|
|
}
|
|
|
|
void Tokenizer::simplifyMicrosoftStringFunctions()
|
|
{
|
|
// skip if not Windows
|
|
if (!mSettings->isWindowsPlatform())
|
|
return;
|
|
|
|
const bool ansi = mSettings->platformType == Settings::Win32A;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->strAt(1) != "(")
|
|
continue;
|
|
|
|
const std::map<std::string, triplet>::const_iterator match = apis.find(tok->str());
|
|
if (match!=apis.end()) {
|
|
tok->str(ansi ? match->second.mbcs : match->second.unicode);
|
|
tok->originalName(match->first);
|
|
} else if (Token::Match(tok, "_T|_TEXT|TEXT ( %char%|%str% )")) {
|
|
tok->deleteNext();
|
|
tok->deleteThis();
|
|
tok->deleteNext();
|
|
if (!ansi) {
|
|
tok->isLong(true);
|
|
if (tok->str()[0] != 'L')
|
|
tok->str("L" + tok->str());
|
|
}
|
|
while (Token::Match(tok->next(), "_T|_TEXT|TEXT ( %char%|%str% )")) {
|
|
tok->next()->deleteNext();
|
|
tok->next()->deleteThis();
|
|
tok->next()->deleteNext();
|
|
tok->concatStr(tok->next()->str());
|
|
tok->deleteNext();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// Remove Borland code
|
|
void Tokenizer::simplifyBorland()
|
|
{
|
|
// skip if not Windows
|
|
if (!mSettings->isWindowsPlatform())
|
|
return;
|
|
if (isC())
|
|
return;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "( __closure * %name% )")) {
|
|
tok->deleteNext();
|
|
}
|
|
}
|
|
|
|
// I think that these classes are always declared at the outer scope
|
|
// I save some time by ignoring inner classes.
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() == "{" && !Token::Match(tok->tokAt(-2), "namespace %type%")) {
|
|
tok = tok->link();
|
|
if (!tok)
|
|
break;
|
|
} else if (Token::Match(tok, "class %name% :|{")) {
|
|
while (tok && tok->str() != "{" && tok->str() != ";")
|
|
tok = tok->next();
|
|
if (!tok)
|
|
break;
|
|
if (tok->str() == ";")
|
|
continue;
|
|
|
|
const Token* end = tok->link()->next();
|
|
for (Token *tok2 = tok->next(); tok2 != end; tok2 = tok2->next()) {
|
|
if (tok2->str() == "__property" &&
|
|
Token::Match(tok2->previous(), ";|{|}|protected:|public:|__published:")) {
|
|
while (tok2->next() && !Token::Match(tok2->next(), "{|;"))
|
|
tok2->deleteNext();
|
|
tok2->deleteThis();
|
|
if (tok2->str() == "{") {
|
|
Token::eraseTokens(tok2, tok2->link());
|
|
tok2->deleteNext();
|
|
tok2->deleteThis();
|
|
|
|
// insert "; __property ;"
|
|
tok2->previous()->insertToken(";");
|
|
tok2->previous()->insertToken("__property");
|
|
tok2->previous()->insertToken(";");
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// Remove Qt signals and slots
|
|
void Tokenizer::simplifyQtSignalsSlots()
|
|
{
|
|
if (isC())
|
|
return;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
// check for emit which can be outside of class
|
|
if (Token::Match(tok, "emit|Q_EMIT %name% (") &&
|
|
Token::simpleMatch(tok->linkAt(2), ") ;")) {
|
|
tok->deleteThis();
|
|
} else if (!Token::Match(tok, "class %name% :|::|{"))
|
|
continue;
|
|
|
|
if (tok->previous() && tok->previous()->str() == "enum") {
|
|
tok = tok->tokAt(2);
|
|
continue;
|
|
}
|
|
|
|
// count { and } for tok2
|
|
int indentlevel = 0;
|
|
for (Token *tok2 = tok; tok2; tok2 = tok2->next()) {
|
|
if (tok2->str() == "{") {
|
|
++indentlevel;
|
|
if (indentlevel == 1)
|
|
tok = tok2;
|
|
else
|
|
tok2 = tok2->link();
|
|
} else if (tok2->str() == "}") {
|
|
if (indentlevel<2)
|
|
break;
|
|
else
|
|
--indentlevel;
|
|
} else if (tok2->str() == ";" && indentlevel == 0)
|
|
break;
|
|
|
|
if (tok2->strAt(1) == "Q_OBJECT")
|
|
tok2->deleteNext();
|
|
|
|
if (Token::Match(tok2->next(), "public|protected|private slots|Q_SLOTS :")) {
|
|
tok2 = tok2->next();
|
|
tok2->str(tok2->str() + ":");
|
|
tok2->deleteNext(2);
|
|
tok2 = tok2->previous();
|
|
} else if (Token::Match(tok2->next(), "signals|Q_SIGNALS :")) {
|
|
tok2 = tok2->next();
|
|
tok2->str("protected:");
|
|
tok2->deleteNext();
|
|
} else if (Token::Match(tok2->next(), "emit|Q_EMIT %name% (") &&
|
|
Token::simpleMatch(tok2->linkAt(3), ") ;")) {
|
|
tok2->deleteNext();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::createSymbolDatabase()
|
|
{
|
|
if (!mSymbolDatabase)
|
|
mSymbolDatabase = new SymbolDatabase(this, mSettings, mErrorLogger);
|
|
mSymbolDatabase->validate();
|
|
}
|
|
|
|
void Tokenizer::deleteSymbolDatabase()
|
|
{
|
|
delete mSymbolDatabase;
|
|
mSymbolDatabase = nullptr;
|
|
}
|
|
|
|
bool Tokenizer::operatorEnd(const Token * tok) const
|
|
{
|
|
if (tok && tok->str() == ")") {
|
|
if (isFunctionHead(tok, "{|;|?|:|["))
|
|
return true;
|
|
|
|
tok = tok->next();
|
|
while (tok && !Token::Match(tok, "[=;{),]")) {
|
|
if (Token::Match(tok, "const|volatile|override")) {
|
|
tok = tok->next();
|
|
} else if (tok->str() == "noexcept") {
|
|
tok = tok->next();
|
|
if (tok && tok->str() == "(") {
|
|
tok = tok->link()->next();
|
|
}
|
|
} else if (tok->str() == "throw" && tok->next() && tok->next()->str() == "(") {
|
|
tok = tok->next()->link()->next();
|
|
}
|
|
// unknown macros ") MACRO {" and ") MACRO(...) {"
|
|
else if (tok->isUpperCaseName()) {
|
|
tok = tok->next();
|
|
if (tok && tok->str() == "(") {
|
|
tok = tok->link()->next();
|
|
}
|
|
} else if (Token::Match(tok, "%op% !!(") ||
|
|
(Token::Match(tok, "%op% (") && !isFunctionHead(tok->next(), "{")))
|
|
break;
|
|
else
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
void Tokenizer::simplifyOperatorName()
|
|
{
|
|
if (isC())
|
|
return;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "using|:: operator %op%|%name% ;")) {
|
|
tok->next()->str("operator" + tok->strAt(2));
|
|
tok->next()->deleteNext();
|
|
continue;
|
|
}
|
|
|
|
if (tok->str() != "operator")
|
|
continue;
|
|
// operator op
|
|
if (Token::Match(tok, "operator %op% (") && !operatorEnd(tok->linkAt(2))) {
|
|
tok->str(tok->str() + tok->next()->str());
|
|
tok->deleteNext();
|
|
continue;
|
|
}
|
|
std::string op;
|
|
Token *par = tok->next();
|
|
bool done = false;
|
|
while (!done && par) {
|
|
done = true;
|
|
if (par->isName()) {
|
|
op += par->str();
|
|
par = par->next();
|
|
// merge namespaces eg. 'operator std :: string () const {'
|
|
if (Token::Match(par, ":: %name%|%op%|.")) {
|
|
op += par->str();
|
|
par = par->next();
|
|
}
|
|
done = false;
|
|
} else if (Token::Match(par, ".|%op%|,")) {
|
|
// check for operator in template
|
|
if (par->str() == "," && !op.empty())
|
|
break;
|
|
if (!(Token::Match(par, "<|>") && !op.empty())) {
|
|
op += par->str();
|
|
par = par->next();
|
|
done = false;
|
|
}
|
|
} else if (Token::simpleMatch(par, "[ ]")) {
|
|
op += "[]";
|
|
par = par->tokAt(2);
|
|
done = false;
|
|
} else if (Token::Match(par, "( *| )")) {
|
|
// break out and simplify..
|
|
if (operatorEnd(par->next()))
|
|
break;
|
|
|
|
while (par->str() != ")") {
|
|
op += par->str();
|
|
par = par->next();
|
|
}
|
|
op += ")";
|
|
par = par->next();
|
|
if (Token::simpleMatch(par, "...")) {
|
|
op.clear();
|
|
par = nullptr;
|
|
break;
|
|
}
|
|
done = false;
|
|
} else if (Token::Match(par, "\"\" %name% (|;|<")) {
|
|
op += "\"\"";
|
|
op += par->strAt(1);
|
|
par = par->tokAt(2);
|
|
done = true;
|
|
} else if (par->str() == "::") {
|
|
op += par->str();
|
|
par = par->next();
|
|
done = false;
|
|
} else if (par->str() == ";" || par->str() == ")") {
|
|
done = true;
|
|
} else if (par->str() != "(") {
|
|
syntaxError(par, "operator");
|
|
}
|
|
}
|
|
|
|
if (par && !op.empty()) {
|
|
tok->str("operator" + op);
|
|
Token::eraseTokens(tok, par);
|
|
}
|
|
|
|
if (!op.empty())
|
|
tok->isOperatorKeyword(true);
|
|
}
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "%op% %str% %name%")) {
|
|
std::string name = tok->strAt(2);
|
|
Token * const str = tok->next();
|
|
str->deleteNext();
|
|
tok->insertToken("operator\"\"" + name);
|
|
tok = tok->next();
|
|
tok->isOperatorKeyword(true);
|
|
tok->insertToken("(");
|
|
str->insertToken(")");
|
|
Token::createMutualLinks(tok->next(), str->next());
|
|
str->insertToken(MathLib::toString(Token::getStrLength(str)));
|
|
str->insertToken(",");
|
|
}
|
|
}
|
|
|
|
if (mSettings->debugwarnings) {
|
|
const Token *tok = list.front();
|
|
|
|
while ((tok = Token::findsimplematch(tok, "operator")) != nullptr) {
|
|
reportError(tok, Severity::debug, "debug",
|
|
"simplifyOperatorName: found unsimplified operator name");
|
|
tok = tok->next();
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyOverloadedOperators()
|
|
{
|
|
if (isC())
|
|
return;
|
|
std::set<std::string> classNames;
|
|
std::set<nonneg int> classVars;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!tok->isName())
|
|
continue;
|
|
|
|
if (Token::simpleMatch(tok, "this ) (") && Token::simpleMatch(tok->tokAt(-2), "( *")) {
|
|
tok = tok->next();
|
|
tok->insertToken("operator()");
|
|
tok->insertToken(".");
|
|
continue;
|
|
}
|
|
|
|
// Get classes that have operator() member
|
|
if (Token::Match(tok, "class|struct %name% [:{]")) {
|
|
int indent = 0;
|
|
for (const Token *tok2 = tok->next(); tok2; tok2 = tok2->next()) {
|
|
if (tok2->str() == "}")
|
|
break;
|
|
else if (indent == 0 && tok2->str() == ";")
|
|
break;
|
|
else if (tok2->str() == "{") {
|
|
if (indent == 0)
|
|
++indent;
|
|
else
|
|
tok2 = tok2->link();
|
|
} else if (indent == 1 && Token::simpleMatch(tok2, "operator() (") && isFunctionHead(tok2->next(), ";{")) {
|
|
classNames.insert(tok->strAt(1));
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
// Get variables that have operator() member
|
|
if (Token::Match(tok, "%type% &| %var%") && classNames.find(tok->str()) != classNames.end()) {
|
|
tok = tok->next();
|
|
while (!tok->isName())
|
|
tok = tok->next();
|
|
classVars.insert(tok->varId());
|
|
}
|
|
|
|
// Simplify operator() calls
|
|
if (Token::Match(tok, "%var% (") && classVars.find(tok->varId()) != classVars.end()) {
|
|
// constructor init list..
|
|
if (Token::Match(tok->previous(), "[:,]")) {
|
|
const Token *start = tok->previous();
|
|
while (Token::simpleMatch(start, ",")) {
|
|
if (Token::simpleMatch(start->previous(), ")"))
|
|
start = start->linkAt(-1);
|
|
else
|
|
break;
|
|
if (Token::Match(start->previous(), "%name%"))
|
|
start = start->tokAt(-2);
|
|
else
|
|
break;
|
|
}
|
|
const Token *after = tok->linkAt(1);
|
|
while (Token::Match(after, ")|} , %name% (|{"))
|
|
after = after->linkAt(3);
|
|
|
|
// Do not simplify initlist
|
|
if (Token::simpleMatch(start, ":") && Token::simpleMatch(after, ") {"))
|
|
continue;
|
|
}
|
|
|
|
tok->insertToken("operator()");
|
|
tok->insertToken(".");
|
|
}
|
|
}
|
|
}
|
|
|
|
// remove unnecessary member qualification..
|
|
void Tokenizer::removeUnnecessaryQualification()
|
|
{
|
|
if (isC())
|
|
return;
|
|
|
|
std::vector<Space> classInfo;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "class|struct|namespace %type% :|{") &&
|
|
(!tok->previous() || tok->previous()->str() != "enum")) {
|
|
Space info;
|
|
info.isNamespace = tok->str() == "namespace";
|
|
tok = tok->next();
|
|
info.className = tok->str();
|
|
tok = tok->next();
|
|
while (tok && tok->str() != "{")
|
|
tok = tok->next();
|
|
if (!tok)
|
|
return;
|
|
info.bodyEnd = tok->link();
|
|
classInfo.push_back(info);
|
|
} else if (!classInfo.empty()) {
|
|
if (tok == classInfo.back().bodyEnd)
|
|
classInfo.pop_back();
|
|
else if (tok->str() == classInfo.back().className &&
|
|
!classInfo.back().isNamespace && tok->previous()->str() != ":" &&
|
|
(Token::Match(tok, "%type% :: ~| %type% (") ||
|
|
Token::Match(tok, "%type% :: operator"))) {
|
|
const Token *tok1 = tok->tokAt(3);
|
|
if (tok->strAt(2) == "operator") {
|
|
// check for operator ()
|
|
if (tok1->str() == "(")
|
|
tok1 = tok1->next();
|
|
|
|
while (tok1 && tok1->str() != "(") {
|
|
if (tok1->str() == ";")
|
|
break;
|
|
tok1 = tok1->next();
|
|
}
|
|
if (!tok1 || tok1->str() != "(")
|
|
continue;
|
|
} else if (tok->strAt(2) == "~")
|
|
tok1 = tok1->next();
|
|
|
|
if (!tok1 || !Token::Match(tok1->link(), ") const| {|;|:")) {
|
|
continue;
|
|
}
|
|
|
|
const bool isConstructorOrDestructor =
|
|
Token::Match(tok, "%type% :: ~| %type%") && (tok->strAt(2) == tok->str() || (tok->strAt(2) == "~" && tok->strAt(3) == tok->str()));
|
|
if (!isConstructorOrDestructor) {
|
|
bool isPrependedByType = Token::Match(tok->previous(), "%type%");
|
|
if (!isPrependedByType) {
|
|
const Token* tok2 = tok->tokAt(-2);
|
|
isPrependedByType = Token::Match(tok2, "%type% *|&");
|
|
}
|
|
if (!isPrependedByType) {
|
|
const Token* tok3 = tok->tokAt(-3);
|
|
isPrependedByType = Token::Match(tok3, "%type% * *|&");
|
|
}
|
|
if (!isPrependedByType) {
|
|
// It's not a constructor declaration and it's not a function declaration so
|
|
// this is a function call which can have all the qualifiers just fine - skip.
|
|
continue;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyReturnStrncat()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::simpleMatch(tok, "return strncat (") &&
|
|
Token::simpleMatch(tok->linkAt(2), ") ;") &&
|
|
tok->strAt(3) != ")" && tok->strAt(3) != ",") {
|
|
|
|
//first argument
|
|
Token *tok2 = tok->tokAt(3);
|
|
|
|
//check if there are at least three arguments
|
|
for (int i = 0; i < 2; ++i) {
|
|
tok2 = tok2->nextArgument();
|
|
if (!tok2) {
|
|
tok = tok->linkAt(2)->next();
|
|
break;
|
|
}
|
|
}
|
|
if (!tok2)
|
|
continue;
|
|
|
|
tok2 = tok2->nextArgument();
|
|
//we want only three arguments
|
|
if (tok2) {
|
|
tok = tok->linkAt(2)->next();
|
|
continue;
|
|
}
|
|
|
|
// Remove 'return'
|
|
tok->deleteThis();
|
|
|
|
// Add 'return arg1 ;' after 'strncat(arg1, arg2, arg3);'
|
|
tok = tok->next();
|
|
|
|
tok2 = tok->link()->next();
|
|
tok2->insertToken(";");
|
|
|
|
//the last token of the first argument before ','
|
|
const Token * const end = tok->next()->nextArgument()->tokAt(-2);
|
|
|
|
//all the first argument is copied
|
|
TokenList::copyTokens(tok2, tok->next(), end);
|
|
tok2->insertToken("return");
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::printUnknownTypes() const
|
|
{
|
|
if (!mSymbolDatabase)
|
|
return;
|
|
|
|
std::multimap<std::string, const Token *> unknowns;
|
|
|
|
for (int i = 1; i <= mVarId; ++i) {
|
|
const Variable *var = mSymbolDatabase->getVariableFromVarId(i);
|
|
if (!var)
|
|
continue;
|
|
// is unknown type?
|
|
if (var->type() || var->typeStartToken()->isStandardType())
|
|
continue;
|
|
|
|
std::string name;
|
|
const Token * nameTok;
|
|
|
|
// single token type?
|
|
if (var->typeStartToken() == var->typeEndToken()) {
|
|
nameTok = var->typeStartToken();
|
|
name = nameTok->str();
|
|
}
|
|
|
|
// complicated type
|
|
else {
|
|
const Token *tok = var->typeStartToken();
|
|
int level = 0;
|
|
|
|
nameTok = tok;
|
|
|
|
while (tok) {
|
|
// skip pointer and reference part of type
|
|
if (level == 0 && Token::Match(tok, "*|&"))
|
|
break;
|
|
|
|
name += tok->str();
|
|
|
|
if (Token::Match(tok, "struct|union|enum"))
|
|
name += " ";
|
|
|
|
// pointers and references are OK in template
|
|
else if (tok->str() == "<")
|
|
++level;
|
|
else if (tok->str() == ">")
|
|
--level;
|
|
|
|
if (tok == var->typeEndToken())
|
|
break;
|
|
|
|
tok = tok->next();
|
|
}
|
|
}
|
|
|
|
unknowns.insert(std::pair<std::string, const Token *>(name, nameTok));
|
|
}
|
|
|
|
if (!unknowns.empty()) {
|
|
std::string last;
|
|
int count = 0;
|
|
|
|
for (std::multimap<std::string, const Token *>::const_iterator it = unknowns.begin(); it != unknowns.end(); ++it) {
|
|
// skip types is std namespace because they are not interesting
|
|
if (it->first.find("std::") != 0) {
|
|
if (it->first != last) {
|
|
last = it->first;
|
|
count = 1;
|
|
reportError(it->second, Severity::debug, "debug", "Unknown type \'" + it->first + "\'.");
|
|
} else {
|
|
if (count < 3) // limit same type to 3
|
|
reportError(it->second, Severity::debug, "debug", "Unknown type \'" + it->first + "\'.");
|
|
count++;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyMathExpressions()
|
|
{
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
|
|
//simplify Pythagorean trigonometric identity: pow(sin(x),2)+pow(cos(x),2) = 1
|
|
// pow(cos(x),2)+pow(sin(x),2) = 1
|
|
// @todo: sin(x) * sin(x) + cos(x) * cos(x) = 1
|
|
// cos(x) * cos(x) + sin(x) * sin(x) = 1
|
|
//simplify Hyperbolic identity: pow(sinh(x),2)-pow(cosh(x),2) = -1
|
|
// pow(cosh(x),2)-pow(sinh(x),2) = -1
|
|
// @todo: sinh(x) * sinh(x) - cosh(x) * cosh(x) = -1
|
|
// cosh(x) * cosh(x) - sinh(x) * sinh(x) = -1
|
|
if (Token::Match(tok, "pow|powf|powl (")) {
|
|
if (Token::Match(tok->tokAt(2), "sin|sinf|sinl (")) {
|
|
Token * const tok2 = tok->linkAt(3);
|
|
if (!Token::Match(tok2, ") , %num% ) + pow|powf|powl ( cos|cosf|cosl ("))
|
|
continue;
|
|
const std::string& leftExponent = tok2->strAt(2);
|
|
if (!isTwoNumber(leftExponent))
|
|
continue; // left exponent is not 2
|
|
const Token * const tok3 = tok2->tokAt(8);
|
|
Token * const tok4 = tok3->link();
|
|
if (!Token::Match(tok4, ") , %num% )"))
|
|
continue;
|
|
const std::string& rightExponent = tok4->strAt(2);
|
|
if (!isTwoNumber(rightExponent))
|
|
continue; // right exponent is not 2
|
|
if (tok->tokAt(3)->stringifyList(tok2->next()) == tok3->stringifyList(tok4->next())) {
|
|
Token::eraseTokens(tok, tok4->tokAt(4));
|
|
tok->str("1");
|
|
}
|
|
} else if (Token::Match(tok->tokAt(2), "cos|cosf|cosl (")) {
|
|
Token * const tok2 = tok->linkAt(3);
|
|
if (!Token::Match(tok2, ") , %num% ) + pow|powf|powl ( sin|sinf|sinl ("))
|
|
continue;
|
|
const std::string& leftExponent = tok2->strAt(2);
|
|
if (!isTwoNumber(leftExponent))
|
|
continue; // left exponent is not 2
|
|
const Token * const tok3 = tok2->tokAt(8);
|
|
Token * const tok4 = tok3->link();
|
|
if (!Token::Match(tok4, ") , %num% )"))
|
|
continue;
|
|
const std::string& rightExponent = tok4->strAt(2);
|
|
if (!isTwoNumber(rightExponent))
|
|
continue; // right exponent is not 2
|
|
if (tok->tokAt(3)->stringifyList(tok2->next()) == tok3->stringifyList(tok4->next())) {
|
|
Token::eraseTokens(tok, tok4->tokAt(4));
|
|
tok->str("1");
|
|
}
|
|
} else if (Token::Match(tok->tokAt(2), "sinh|sinhf|sinhl (")) {
|
|
Token * const tok2 = tok->linkAt(3);
|
|
if (!Token::Match(tok2, ") , %num% ) - pow|powf|powl ( cosh|coshf|coshl ("))
|
|
continue;
|
|
const std::string& leftExponent = tok2->strAt(2);
|
|
if (!isTwoNumber(leftExponent))
|
|
continue; // left exponent is not 2
|
|
const Token * const tok3 = tok2->tokAt(8);
|
|
Token * const tok4 = tok3->link();
|
|
if (!Token::Match(tok4, ") , %num% )"))
|
|
continue;
|
|
const std::string& rightExponent = tok4->strAt(2);
|
|
if (!isTwoNumber(rightExponent))
|
|
continue; // right exponent is not 2
|
|
if (tok->tokAt(3)->stringifyList(tok2->next()) == tok3->stringifyList(tok4->next())) {
|
|
Token::eraseTokens(tok, tok4->tokAt(4));
|
|
tok->str("-1");
|
|
}
|
|
} else if (Token::Match(tok->tokAt(2), "cosh|coshf|coshl (")) {
|
|
Token * const tok2 = tok->linkAt(3);
|
|
if (!Token::Match(tok2, ") , %num% ) - pow|powf|powl ( sinh|sinhf|sinhl ("))
|
|
continue;
|
|
const std::string& leftExponent = tok2->strAt(2);
|
|
if (!isTwoNumber(leftExponent))
|
|
continue; // left exponent is not 2
|
|
const Token * const tok3 = tok2->tokAt(8);
|
|
Token * const tok4 = tok3->link();
|
|
if (!Token::Match(tok4, ") , %num% )"))
|
|
continue;
|
|
const std::string& rightExponent = tok4->strAt(2);
|
|
if (!isTwoNumber(rightExponent))
|
|
continue; // right exponent is not 2
|
|
if (tok->tokAt(3)->stringifyList(tok2->next()) == tok3->stringifyList(tok4->next())) {
|
|
Token::eraseTokens(tok, tok4->tokAt(4));
|
|
tok->str("-1");
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
bool Tokenizer::simplifyStrlen()
|
|
{
|
|
// replace strlen(str)
|
|
bool modified=false;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "strlen ( %str% )")) {
|
|
tok->str(MathLib::toString(Token::getStrLength(tok->tokAt(2))));
|
|
tok->deleteNext(3);
|
|
modified=true;
|
|
}
|
|
}
|
|
return modified;
|
|
}
|
|
|
|
void Tokenizer::prepareTernaryOpForAST()
|
|
{
|
|
// http://en.cppreference.com/w/cpp/language/operator_precedence says about ternary operator:
|
|
// "The expression in the middle of the conditional operator (between ? and :) is parsed as if parenthesized: its precedence relative to ?: is ignored."
|
|
// The AST parser relies on this function to add such parentheses where necessary.
|
|
for (Token* tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() == "?") {
|
|
bool parenthesesNeeded = false;
|
|
int depth = 0;
|
|
Token* tok2 = tok->next();
|
|
for (; tok2; tok2 = tok2->next()) {
|
|
if (tok2->link() && Token::Match(tok2, "[|(|<"))
|
|
tok2 = tok2->link();
|
|
else if (tok2->str() == ":") {
|
|
if (depth == 0)
|
|
break;
|
|
depth--;
|
|
} else if (tok2->str() == ";" || (tok2->link() && tok2->str() != "{" && tok2->str() != "}"))
|
|
break;
|
|
else if (tok2->str() == ",")
|
|
parenthesesNeeded = true;
|
|
else if (tok2->str() == "<")
|
|
parenthesesNeeded = true;
|
|
else if (tok2->str() == "?") {
|
|
depth++;
|
|
parenthesesNeeded = true;
|
|
}
|
|
}
|
|
if (parenthesesNeeded && tok2 && tok2->str() == ":") {
|
|
tok->insertToken("(");
|
|
tok2->insertToken(")", emptyString, true);
|
|
Token::createMutualLinks(tok->next(), tok2->previous());
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::reportError(const Token* tok, const Severity::SeverityType severity, const std::string& id, const std::string& msg, bool inconclusive) const
|
|
{
|
|
const std::list<const Token*> callstack(1, tok);
|
|
reportError(callstack, severity, id, msg, inconclusive);
|
|
}
|
|
|
|
void Tokenizer::reportError(const std::list<const Token*>& callstack, Severity::SeverityType severity, const std::string& id, const std::string& msg, bool inconclusive) const
|
|
{
|
|
const ErrorMessage errmsg(callstack, &list, severity, id, msg, inconclusive ? Certainty::inconclusive : Certainty::normal);
|
|
if (mErrorLogger)
|
|
mErrorLogger->reportErr(errmsg);
|
|
else
|
|
Check::reportError(errmsg);
|
|
}
|
|
|
|
void Tokenizer::setPodTypes()
|
|
{
|
|
if (!mSettings)
|
|
return;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!tok->isName())
|
|
continue;
|
|
|
|
// pod type
|
|
const struct Library::PodType *podType = mSettings->library.podtype(tok->str());
|
|
if (podType) {
|
|
const Token *prev = tok->previous();
|
|
while (prev && prev->isName())
|
|
prev = prev->previous();
|
|
if (prev && !Token::Match(prev, ";|{|}|,|("))
|
|
continue;
|
|
tok->isStandardType(true);
|
|
}
|
|
}
|
|
}
|
|
|
|
const Token *Tokenizer::findSQLBlockEnd(const Token *tokSQLStart)
|
|
{
|
|
const Token *tokLastEnd = nullptr;
|
|
for (const Token *tok = tokSQLStart->tokAt(2); tok != nullptr; tok = tok->next()) {
|
|
if (tokLastEnd == nullptr && tok->str() == ";")
|
|
tokLastEnd = tok;
|
|
else if (tok->str() == "__CPPCHECK_EMBEDDED_SQL_EXEC__") {
|
|
if (Token::simpleMatch(tok->tokAt(-2), "END - __CPPCHECK_EMBEDDED_SQL_EXEC__ ;"))
|
|
return tok->next();
|
|
return tokLastEnd;
|
|
} else if (Token::Match(tok, "{|}|==|&&|!|^|<<|>>|++|+=|-=|/=|*=|>>=|<<=|~"))
|
|
break; // We are obviously outside the SQL block
|
|
}
|
|
|
|
return tokLastEnd;
|
|
}
|
|
|
|
void Tokenizer::simplifyNestedNamespace()
|
|
{
|
|
if (!isCPP())
|
|
return;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (Token::Match(tok, "namespace %name% ::") && tok->strAt(-1) != "using") {
|
|
Token * tok2 = tok->tokAt(2);
|
|
|
|
// validate syntax
|
|
while (Token::Match(tok2, ":: %name%"))
|
|
tok2 = tok2->tokAt(2);
|
|
|
|
if (!tok2 || tok2->str() != "{")
|
|
return; // syntax error
|
|
|
|
std::stack<Token *> links;
|
|
tok2 = tok->tokAt(2);
|
|
|
|
while (tok2->str() == "::") {
|
|
links.push(tok2);
|
|
tok2->str("{");
|
|
tok2->insertToken("namespace");
|
|
tok2 = tok2->tokAt(3);
|
|
}
|
|
|
|
tok = tok2;
|
|
|
|
if (!links.empty() && tok2->str() == "{") {
|
|
tok2 = tok2->link();
|
|
while (!links.empty()) {
|
|
tok2->insertToken("}");
|
|
tok2 = tok2->next();
|
|
Token::createMutualLinks(links.top(), tok2);
|
|
links.pop();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void Tokenizer::simplifyCoroutines()
|
|
{
|
|
if (!isCPP() || mSettings->standards.cpp < Standards::CPP20)
|
|
return;
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (!tok->isName() || !Token::Match(tok, "co_return|co_yield|co_await"))
|
|
continue;
|
|
Token *end = tok->next();
|
|
while (end && end->str() != ";") {
|
|
if (Token::Match(end, "[({[]"))
|
|
end = end->link();
|
|
else if (Token::Match(end, "[)]}]"))
|
|
break;
|
|
end = end->next();
|
|
}
|
|
if (Token::simpleMatch(end, ";")) {
|
|
tok->insertToken("(");
|
|
end->previous()->insertToken(")");
|
|
Token::createMutualLinks(tok->next(), end->previous());
|
|
}
|
|
}
|
|
}
|
|
|
|
static bool sameTokens(const Token *first, const Token *last, const Token *other)
|
|
{
|
|
while (other && first->str() == other->str()) {
|
|
if (first == last)
|
|
return true;
|
|
first = first->next();
|
|
other = other->next();
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
static bool alreadyHasNamespace(const Token *first, const Token *last, const Token *end)
|
|
{
|
|
while (end && last->str() == end->str()) {
|
|
if (first == last)
|
|
return true;
|
|
last = last->previous();
|
|
end = end->previous();
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
static Token * deleteAlias(Token * tok)
|
|
{
|
|
Token::eraseTokens(tok, Token::findsimplematch(tok, ";"));
|
|
|
|
// delete first token
|
|
tok->deleteThis();
|
|
|
|
// delete ';' if not last token
|
|
tok->deleteThis();
|
|
|
|
return tok;
|
|
}
|
|
|
|
void Tokenizer::simplifyNamespaceAliases()
|
|
{
|
|
if (!isCPP())
|
|
return;
|
|
|
|
int scope = 0;
|
|
|
|
for (Token *tok = list.front(); tok; tok = tok->next()) {
|
|
if (tok->str() == "{")
|
|
scope++;
|
|
else if (tok->str() == "}")
|
|
scope--;
|
|
else if (Token::Match(tok, "namespace %name% =")) {
|
|
const std::string name(tok->next()->str());
|
|
Token * tokNameStart = tok->tokAt(3);
|
|
Token * tokNameEnd = tokNameStart;
|
|
|
|
while (tokNameEnd && tokNameEnd->next() && tokNameEnd->next()->str() != ";")
|
|
tokNameEnd = tokNameEnd->next();
|
|
|
|
if (!tokNameEnd)
|
|
return; // syntax error
|
|
|
|
int endScope = scope;
|
|
Token * tokLast = tokNameEnd->next();
|
|
Token * tokNext = tokLast->next();
|
|
Token * tok2 = tokNext;
|
|
|
|
while (tok2 && endScope >= scope) {
|
|
if (Token::simpleMatch(tok2, "{"))
|
|
endScope++;
|
|
else if (Token::simpleMatch(tok2, "}"))
|
|
endScope--;
|
|
else if (tok2->str() == name) {
|
|
if (Token::Match(tok2->previous(), "namespace %name% =")) {
|
|
// check for possible duplicate aliases
|
|
if (sameTokens(tokNameStart, tokNameEnd, tok2->tokAt(2))) {
|
|
// delete duplicate
|
|
tok2 = deleteAlias(tok2->previous());
|
|
continue;
|
|
} else {
|
|
// conflicting declaration (syntax error)
|
|
// cppcheck-suppress duplicateBranch - remove when TODO below is addressed
|
|
if (endScope == scope) {
|
|
// delete conflicting declaration
|
|
tok2 = deleteAlias(tok2->previous());
|
|
}
|
|
|
|
// new declaration
|
|
else {
|
|
// TODO: use the new alias in this scope
|
|
tok2 = deleteAlias(tok2->previous());
|
|
}
|
|
continue;
|
|
}
|
|
}
|
|
|
|
if (tok2->strAt(1) == "::" && !alreadyHasNamespace(tokNameStart, tokNameEnd, tok2)) {
|
|
tok2->str(tokNameStart->str());
|
|
Token * tok3 = tokNameStart;
|
|
while (tok3 != tokNameEnd) {
|
|
tok2->insertToken(tok3->next()->str());
|
|
tok2 = tok2->next();
|
|
tok3 = tok3->next();
|
|
}
|
|
}
|
|
}
|
|
tok2 = tok2->next();
|
|
}
|
|
|
|
if (tok->previous() && tokNext) {
|
|
Token::eraseTokens(tok->previous(), tokNext);
|
|
tok = tokNext->previous();
|
|
} else if (tok->previous()) {
|
|
Token::eraseTokens(tok->previous(), tokLast);
|
|
tok = tokLast;
|
|
} else if (tokNext) {
|
|
Token::eraseTokens(tok, tokNext);
|
|
tok->deleteThis();
|
|
} else {
|
|
Token::eraseTokens(tok, tokLast);
|
|
tok->deleteThis();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
Tokenizer::VariableMap::VariableMap() : mVarId(0) {}
|
|
|
|
void Tokenizer::VariableMap::enterScope()
|
|
{
|
|
mScopeInfo.push(std::list<std::pair<std::string, int>>());
|
|
}
|
|
|
|
bool Tokenizer::VariableMap::leaveScope()
|
|
{
|
|
if (mScopeInfo.empty())
|
|
return false;
|
|
|
|
for (const std::pair<std::string, int> &outerVariable : mScopeInfo.top()) {
|
|
if (outerVariable.second != 0)
|
|
mVariableId[outerVariable.first] = outerVariable.second;
|
|
else
|
|
mVariableId.erase(outerVariable.first);
|
|
}
|
|
mScopeInfo.pop();
|
|
return true;
|
|
}
|
|
|
|
void Tokenizer::VariableMap::addVariable(const std::string &varname)
|
|
{
|
|
if (mScopeInfo.empty()) {
|
|
mVariableId[varname] = ++mVarId;
|
|
return;
|
|
}
|
|
std::map<std::string, int>::iterator it = mVariableId.find(varname);
|
|
if (it == mVariableId.end()) {
|
|
mScopeInfo.top().push_back(std::pair<std::string, int>(varname, 0));
|
|
mVariableId[varname] = ++mVarId;
|
|
return;
|
|
}
|
|
mScopeInfo.top().push_back(std::pair<std::string, int>(varname, it->second));
|
|
it->second = ++mVarId;
|
|
}
|
|
|
|
bool Tokenizer::VariableMap::hasVariable(const std::string &varname) const
|
|
{
|
|
return mVariableId.find(varname) != mVariableId.end();
|
|
}
|
|
|
|
bool Tokenizer::hasIfdef(const Token *start, const Token *end) const
|
|
{
|
|
if (!mPreprocessor)
|
|
return false;
|
|
for (const Directive &d: mPreprocessor->getDirectives()) {
|
|
if (d.str.compare(0,3,"#if") == 0 &&
|
|
d.linenr >= start->linenr() &&
|
|
d.linenr <= end->linenr() &&
|
|
start->fileIndex() < list.getFiles().size() &&
|
|
d.file == list.getFiles()[start->fileIndex()])
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|