cppcheck/lib/templatesimplifier.cpp

2294 lines
91 KiB
C++
Raw Normal View History

/*
* Cppcheck - A tool for static C/C++ code analysis
2018-01-14 15:37:52 +01:00
* Copyright (C) 2007-2018 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 "templatesimplifier.h"
2017-05-27 04:33:47 +02:00
#include "errorlogger.h"
#include "mathlib.h"
2017-05-27 04:33:47 +02:00
#include "settings.h"
#include "token.h"
#include "tokenize.h"
2017-05-27 04:33:47 +02:00
#include "tokenlist.h"
#include <algorithm>
#include <cassert>
#include <iostream>
2017-05-27 04:33:47 +02:00
#include <stack>
#include <utility>
namespace {
class FindToken {
public:
2018-06-17 19:24:40 +02:00
explicit FindToken(const Token *token) : mToken(token) {}
bool operator()(const TemplateSimplifier::TokenAndName &tokenAndName) const {
return tokenAndName.token == mToken;
}
private:
2018-06-17 19:24:40 +02:00
const Token * const mToken;
};
class FindName {
public:
2018-06-17 19:24:40 +02:00
explicit FindName(const std::string &name) : mName(name) {}
bool operator()(const TemplateSimplifier::TokenAndName &tokenAndName) const {
return tokenAndName.name == mName;
}
private:
2018-06-17 19:24:40 +02:00
const std::string mName;
};
}
TemplateSimplifier::TokenAndName::TokenAndName(Token *tok, const std::string &s, const std::string &n, const Token *nt, const Token *pe) :
token(tok), scope(s), name(n), nameToken(nt), paramEnd(pe)
{
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
if (token)
token->templateSimplifierPointer(this);
}
TemplateSimplifier::TokenAndName::TokenAndName(const TokenAndName& otherTok) :
token(otherTok.token), scope(otherTok.scope), name(otherTok.name), nameToken(otherTok.nameToken), paramEnd(otherTok.paramEnd)
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
{
if (token)
token->templateSimplifierPointer(this);
}
TemplateSimplifier::TokenAndName::~TokenAndName()
{
if (token)
token->templateSimplifierPointers().erase(this);
}
TemplateSimplifier::TemplateSimplifier(Tokenizer *tokenizer)
: mTokenList(tokenizer->list), mSettings(tokenizer->mSettings), mErrorLogger(tokenizer->mErrorLogger)
{
}
TemplateSimplifier::~TemplateSimplifier()
{
}
void TemplateSimplifier::cleanupAfterSimplify()
{
bool goback = false;
for (Token *tok = mTokenList.front(); tok; tok = tok->next()) {
if (goback) {
tok = tok->previous();
goback = false;
}
if (tok->str() == "(")
tok = tok->link();
else if (Token::Match(tok, "template < > %name%")) {
const Token *end = tok;
while (end) {
if (end->str() == ";")
break;
if (end->str() == "{") {
end = end->link()->next();
break;
}
if (!Token::Match(end, "%name%|::|<|>|,")) {
2014-02-16 10:32:10 +01:00
end = nullptr;
break;
}
end = end->next();
}
if (end) {
Token::eraseTokens(tok,end);
tok->deleteThis();
}
}
else if (Token::Match(tok, "%type% <") &&
(!tok->previous() || tok->previous()->str() == ";")) {
const Token *tok2 = tok->tokAt(2);
std::string type;
2016-08-24 14:10:39 +02:00
while (Token::Match(tok2, "%type%|%num% ,")) {
type += tok2->str() + ",";
tok2 = tok2->tokAt(2);
}
2016-08-24 14:10:39 +02:00
if (Token::Match(tok2, "%type%|%num% > (")) {
type += tok2->str();
tok->str(tok->str() + "<" + type + ">");
Token::eraseTokens(tok, tok2->tokAt(2));
if (tok == mTokenList.front())
goback = true;
}
}
}
}
void TemplateSimplifier::checkComplicatedSyntaxErrorsInTemplates()
{
// check for more complicated syntax errors when using templates..
for (const Token *tok = mTokenList.front(); tok; tok = tok->next()) {
// skip executing scopes (ticket #3183)..
if (Token::simpleMatch(tok, "( {")) {
tok = tok->link();
if (!tok)
syntaxError(nullptr);
}
// skip executing scopes..
const Token *start = Tokenizer::startOfExecutableScope(tok);
if (start) {
tok = start->link();
}
// skip executing scopes (ticket #1985)..
else if (Token::simpleMatch(tok, "try {")) {
tok = tok->next()->link();
while (Token::simpleMatch(tok, "} catch (")) {
tok = tok->linkAt(2);
if (Token::simpleMatch(tok, ") {"))
tok = tok->next()->link();
}
}
if (!tok)
syntaxError(nullptr);
// not start of statement?
if (tok->previous() && !Token::Match(tok, "[;{}]"))
continue;
// skip starting tokens.. ;;; typedef typename foo::bar::..
while (Token::simpleMatch(tok, ";"))
tok = tok->next();
while (Token::Match(tok, "typedef|typename"))
tok = tok->next();
while (Token::Match(tok, "%type% ::"))
tok = tok->tokAt(2);
if (!tok)
break;
// template variable or type..
if (Token::Match(tok, "%type% <")) {
// these are used types..
std::set<std::string> usedtypes;
// parse this statement and see if the '<' and '>' are matching
unsigned int level = 0;
for (const Token *tok2 = tok; tok2 && !Token::simpleMatch(tok2, ";"); tok2 = tok2->next()) {
2018-10-28 17:37:19 +01:00
if (Token::simpleMatch(tok2, "{") && (!Token::Match(tok2->previous(), ">|%type%") || Token::simpleMatch(tok2->link(), "} ;")))
break;
if (tok2->str() == "(")
tok2 = tok2->link();
else if (tok2->str() == "<") {
bool inclevel = false;
if (Token::simpleMatch(tok2->previous(), "operator <"))
;
else if (level == 0)
inclevel = true;
else if (tok2->next() && tok2->next()->isStandardType())
inclevel = true;
else if (Token::simpleMatch(tok2, "< typename"))
inclevel = true;
else if (Token::Match(tok2->tokAt(-2), "<|, %type% <") && usedtypes.find(tok2->previous()->str()) != usedtypes.end())
inclevel = true;
else if (Token::Match(tok2, "< %type%") && usedtypes.find(tok2->next()->str()) != usedtypes.end())
inclevel = true;
else if (Token::Match(tok2, "< %type%")) {
// is the next token a type and not a variable/constant?
// assume it's a type if there comes another "<"
const Token *tok3 = tok2->next();
while (Token::Match(tok3, "%type% ::"))
tok3 = tok3->tokAt(2);
if (Token::Match(tok3, "%type% <"))
inclevel = true;
}
if (inclevel) {
++level;
if (Token::Match(tok2->tokAt(-2), "<|, %type% <"))
usedtypes.insert(tok2->previous()->str());
}
} else if (tok2->str() == ">") {
if (level > 0)
--level;
} else if (tok2->str() == ">>") {
if (level > 0)
--level;
if (level > 0)
--level;
}
}
if (level > 0)
syntaxError(tok);
}
}
}
unsigned int TemplateSimplifier::templateParameters(const Token *tok)
{
unsigned int numberOfParameters = 1;
if (!tok)
return 0;
if (tok->str() != "<")
return 0;
tok = tok->next();
unsigned int level = 0;
while (tok) {
// skip const/volatile
if (Token::Match(tok, "const|volatile"))
tok = tok->next();
// skip struct/union
if (Token::Match(tok, "struct|union"))
tok = tok->next();
// Skip '&'
if (Token::Match(tok, "& ::| %name%"))
tok = tok->next();
// Skip variadic types (Ticket #5774, #6059, #6172)
if (Token::Match(tok, "%type% . . .")) {
tok = tok->tokAt(4);
continue;
}
// Skip '=', '?', ':'
if (Token::Match(tok, "=|?|:"))
tok = tok->next();
if (!tok)
return 0;
// Skip casts
if (tok->str() == "(") {
tok = tok->link();
if (tok)
tok = tok->next();
}
// skip std::
if (tok && tok->str() == "::")
tok = tok->next();
while (Token::Match(tok, "%name% ::")) {
tok = tok->tokAt(2);
if (tok && tok->str() == "*") // Ticket #5759: Class member pointer as a template argument; skip '*'
tok = tok->next();
}
if (!tok)
return 0;
// num/type ..
if (!tok->isNumber() && tok->tokType() != Token::eChar && !tok->isName() && !tok->isOp())
return 0;
tok = tok->next();
2012-04-26 10:58:35 +02:00
if (!tok)
return 0;
// * / const
while (Token::Match(tok, "*|&|&&|const"))
tok = tok->next();
if (!tok)
return 0;
// Function pointer or prototype..
while (Token::Match(tok, "(|[")) {
if (!tok->link())
syntaxError(tok);
tok = tok->link()->next();
while (Token::Match(tok, "const|volatile")) // Ticket #5786: Skip function cv-qualifiers
tok = tok->next();
}
if (!tok)
return 0;
// inner template
if (tok->str() == "<") {
++level;
tok = tok->next();
}
if (!tok)
return 0;
// ,/>
while (Token::Match(tok, ">|>>")) {
if (level == 0)
return ((tok->str() == ">") ? numberOfParameters : 0);
--level;
if (tok->str() == ">>") {
if (level == 0)
return numberOfParameters;
--level;
}
tok = tok->next();
if (Token::simpleMatch(tok,"("))
tok = tok->link()->next();
if (!tok)
return 0;
}
if (tok->str() != ",")
continue;
if (level == 0)
++numberOfParameters;
tok = tok->next();
}
return 0;
}
void TemplateSimplifier::eraseTokens(Token *begin, const Token *end)
{
if (!begin || begin == end)
return;
while (begin->next() && begin->next() != end) {
begin->deleteNext();
}
}
void TemplateSimplifier::deleteToken(Token *tok)
{
if (tok->next())
tok->next()->deletePrevious();
else
tok->deleteThis();
}
bool TemplateSimplifier::removeTemplate(Token *tok)
{
if (!Token::simpleMatch(tok, "template <"))
return false;
int indentlevel = 0;
unsigned int countgt = 0; // Counter for ">"
for (const Token *tok2 = tok->next(); tok2; tok2 = tok2->next()) {
if (tok2->str() == "(") {
tok2 = tok2->link();
} else if (tok2->str() == ")") { // garbage code! (#3504)
eraseTokens(tok,tok2);
deleteToken(tok);
return false;
}
else if (tok2->str() == "{") {
tok2 = tok2->link()->next();
if (tok2 && tok2->str() == ";" && tok2->next())
tok2 = tok2->next();
eraseTokens(tok, tok2);
deleteToken(tok);
return true;
} else if (tok2->str() == "}") { // garbage code! (#3449)
eraseTokens(tok,tok2);
deleteToken(tok);
return false;
}
// Count ">"
if (tok2->str() == ">")
countgt++;
// don't remove constructor
if (tok2->str() == "explicit" ||
(countgt == 1 && Token::Match(tok2->previous(), "> %type% (") &&
Tokenizer::startOfExecutableScope(tok2->linkAt(1)))) {
eraseTokens(tok, tok2);
deleteToken(tok);
return true;
}
if (tok2->str() == ";") {
tok2 = tok2->next();
eraseTokens(tok, tok2);
deleteToken(tok);
return true;
}
if (tok2->str() == "<")
++indentlevel;
else if (indentlevel >= 2 && tok2->str() == ">")
--indentlevel;
else if (Token::Match(tok2, "> class|struct|union %name% [,)]")) {
tok2 = tok2->next();
eraseTokens(tok, tok2);
deleteToken(tok);
return true;
}
}
return false;
}
/// TODO: This is copy pasted from Tokenizer. We should reuse this code.
namespace {
struct ScopeInfo2 {
ScopeInfo2(const std::string &name_, const Token *bodyEnd_) : name(name_), bodyEnd(bodyEnd_) {}
const std::string name;
const Token * const bodyEnd;
};
}
static std::string getScopeName(const std::list<ScopeInfo2> &scopeInfo)
{
std::string ret;
2018-07-15 14:45:15 +02:00
for (const ScopeInfo2 &i : scopeInfo)
ret += (ret.empty() ? "" : " :: ") + i.name;
return ret;
}
static void setScopeInfo(const Token *tok, std::list<ScopeInfo2> *scopeInfo)
{
while (tok->str() == "}" && !scopeInfo->empty() && tok == scopeInfo->back().bodyEnd)
scopeInfo->pop_back();
if (!Token::Match(tok, "namespace|class|struct|union %name% {|:|::"))
return;
tok = tok->next();
std::string classname = tok->str();
while (Token::Match(tok, "%name% :: %name%")) {
tok = tok->tokAt(2);
classname += " :: " + tok->str();
}
tok = tok->next();
if (tok && tok->str() == ":") {
// ...
}
if (tok && tok->str() == "{") {
scopeInfo->emplace_back(classname,tok->link());
}
}
bool TemplateSimplifier::getTemplateDeclarations()
{
bool codeWithTemplates = false;
std::list<ScopeInfo2> scopeInfo;
for (Token *tok = mTokenList.front(); tok; tok = tok->next()) {
if (Token::Match(tok, "}|namespace|class|struct|union")) {
setScopeInfo(tok, &scopeInfo);
continue;
}
2017-12-27 22:29:45 +01:00
if (!Token::simpleMatch(tok, "template <"))
continue;
// Some syntax checks, see #6865
if (!tok->tokAt(2))
syntaxError(tok->next());
if (tok->strAt(2)=="typename" &&
!Token::Match(tok->tokAt(3), "%name%|.|,|=|>"))
2017-12-27 22:29:45 +01:00
syntaxError(tok->next());
codeWithTemplates = true;
const Token * const parmEnd = tok->next()->findClosingBracket();
2017-12-27 22:29:45 +01:00
for (const Token *tok2 = parmEnd; tok2; tok2 = tok2->next()) {
if (tok2->str() == "(")
tok2 = tok2->link();
else if (tok2->str() == ")")
break;
// skip decltype(...)
else if (Token::Match(tok2, "decltype ("))
tok2 = tok2->linkAt(1);
// Declaration => add to mTemplateForwardDeclarations
else if (tok2->str() == ";") {
const int namepos = getTemplateNamePosition(parmEnd, true);
if (namepos > 0)
mTemplateForwardDeclarations.emplace_back(tok, getScopeName(scopeInfo), parmEnd->strAt(namepos), parmEnd->tokAt(namepos), parmEnd);
2017-12-27 22:29:45 +01:00
break;
}
// Implementation => add to mTemplateDeclarations
2017-12-27 22:29:45 +01:00
else if (tok2->str() == "{") {
const int namepos = getTemplateNamePosition(parmEnd, false);
if (namepos > 0)
mTemplateDeclarations.emplace_back(tok, getScopeName(scopeInfo), parmEnd->strAt(namepos), parmEnd->tokAt(namepos), parmEnd);
2017-12-27 22:29:45 +01:00
break;
}
}
}
return codeWithTemplates;
}
void TemplateSimplifier::getTemplateInstantiations()
{
std::list<ScopeInfo2> scopeList;
for (Token *tok = mTokenList.front(); tok; tok = tok->next()) {
if (Token::Match(tok, "}|namespace|class|struct|union")) {
setScopeInfo(tok, &scopeList);
continue;
}
// template definition.. skip it
if (Token::simpleMatch(tok, "template <")) {
2013-07-31 10:30:20 +02:00
tok = tok->next()->findClosingBracket();
if (!tok)
break;
// #7914
// Ignore template instantiations within template definitions: they will only be
// handled if the definition is actually instantiated
const Token *tok2 = Token::findmatch(tok, "{|;");
if (tok2 && tok2->str() == "{")
tok = tok2->link();
} else if (Token::Match(tok->previous(), "(|{|}|;|=|>|<<|:|.|*|& %name% ::|<") ||
Token::Match(tok->previous(), "%type% %name% ::|<") ||
Token::Match(tok->tokAt(-2), "[,:] private|protected|public %name% ::|<")) {
std::string scopeName = getScopeName(scopeList);
while (Token::Match(tok, "%name% :: %name%")) {
scopeName += (scopeName.empty() ? "" : " :: ") + tok->str();
tok = tok->tokAt(2);
}
if (!Token::Match(tok, "%name% <"))
continue;
// Add inner template instantiations first => go to the ">"
// and then parse backwards, adding all seen instantiations
2013-07-31 10:30:20 +02:00
const Token *tok2 = tok->next()->findClosingBracket();
// parse backwards and add template instantiations
// TODO
for (; tok2 && tok2 != tok; tok2 = tok2->previous()) {
if (Token::Match(tok2, ", %name% <") &&
templateParameters(tok2->tokAt(2))) {
mTemplateInstantiations.emplace_back(tok2->next(), getScopeName(scopeList), tok2->strAt(1), tok2->tokAt(1));
} else if (Token::Match(tok2->next(), "class|struct"))
const_cast<Token *>(tok2)->deleteNext();
}
// Add outer template..
if (templateParameters(tok->next())) {
2017-12-30 09:00:19 +01:00
const std::string scopeName1(scopeName);
while (true) {
2017-12-30 09:00:19 +01:00
const std::string fullName = scopeName + (scopeName.empty()?"":" :: ") + tok->str();
const std::list<TokenAndName>::const_iterator it = std::find_if(mTemplateDeclarations.begin(), mTemplateDeclarations.end(), FindName(fullName));
if (it != mTemplateDeclarations.end()) {
mTemplateInstantiations.emplace_back(tok, getScopeName(scopeList), fullName, tok);
break;
} else {
2017-12-30 09:00:19 +01:00
if (scopeName.empty()) {
mTemplateInstantiations.emplace_back(tok, getScopeName(scopeList), scopeName1 + (scopeName1.empty()?"":" :: ") + tok->str(), tok);
2017-12-30 09:00:19 +01:00
break;
}
const std::string::size_type pos = scopeName.rfind(" :: ");
scopeName = (pos == std::string::npos) ? std::string() : scopeName.substr(0,pos);
}
}
}
}
}
}
void TemplateSimplifier::useDefaultArgumentValues()
{
for (TokenAndName &template1 : mTemplateDeclarations) {
// template parameters with default value has syntax such as:
// x = y
// this list will contain all the '=' tokens for such arguments
std::list<Token *> eq;
// and this set the position of parameters with a default value
std::set<std::size_t> defaultedArgPos;
// parameter number. 1,2,3,..
std::size_t templatepar = 1;
// parameter depth
std::size_t templateParmDepth = 0;
// the template classname. This will be empty for template functions
std::string classname;
// Scan template declaration..
2018-07-15 14:45:15 +02:00
for (Token *tok = template1.token; tok; tok = tok->next()) {
if (Token::simpleMatch(tok, "template < >")) { // Ticket #5762: Skip specialization tokens
tok = tok->tokAt(2);
if (0 == templateParmDepth)
break;
continue;
}
if (tok->str() == "(") { // Ticket #6835
tok = tok->link();
continue;
}
if (tok->str() == "<" && templateParameters(tok))
++templateParmDepth;
// end of template parameters?
if (tok->str() == ">") {
if (Token::Match(tok, "> class|struct|union %name%"))
classname = tok->strAt(2);
if (templateParmDepth<2)
break;
else
--templateParmDepth;
}
// next template parameter
if (tok->str() == "," && (1 == templateParmDepth)) // Ticket #5823: Properly count parameters
++templatepar;
// default parameter value?
else if (Token::Match(tok, "= !!>")) {
if (defaultedArgPos.insert(templatepar).second) {
eq.push_back(tok);
} else {
// Ticket #5605: Syntax error (two equal signs for the same parameter), bail out
eq.clear();
break;
}
}
}
if (eq.empty() || classname.empty())
continue;
// iterate through all template instantiations
for (const TokenAndName &templateInst : mTemplateInstantiations) {
2018-07-15 14:45:15 +02:00
Token *tok = templateInst.token;
if (!Token::simpleMatch(tok, (classname + " <").c_str()))
continue;
// count the parameters..
tok = tok->next();
const unsigned int usedpar = templateParameters(tok);
tok = tok->findClosingBracket();
if (tok && tok->str() == ">") {
tok = tok->previous();
std::list<Token *>::const_iterator it = eq.begin();
for (std::size_t i = (templatepar - eq.size()); it != eq.end() && i < usedpar; ++i)
++it;
while (it != eq.end()) {
int indentlevel = 0;
tok->insertToken(",");
tok = tok->next();
const Token *from = (*it)->next();
std::stack<Token *> links;
while (from && (!links.empty() || indentlevel || !Token::Match(from, ",|>"))) {
if (from->str() == "<")
++indentlevel;
else if (from->str() == ">")
--indentlevel;
tok->insertToken(from->str(), from->originalName());
tok = tok->next();
if (Token::Match(tok, "(|["))
links.push(tok);
else if (!links.empty() && Token::Match(tok, ")|]")) {
Token::createMutualLinks(links.top(), tok);
links.pop();
}
from = from->next();
}
++it;
}
}
}
2018-07-15 14:45:15 +02:00
for (Token * const eqtok : eq) {
Token *tok2;
int indentlevel = 0;
for (tok2 = eqtok->next(); tok2; tok2 = tok2->next()) {
if (Token::Match(tok2, ";|)|}|]")) { // bail out #6607
tok2 = nullptr;
break;
}
if (Token::Match(tok2, "(|{|["))
tok2 = tok2->link();
else if (Token::Match(tok2, "%type% <") && templateParameters(tok2->next())) {
std::list<TokenAndName>::iterator ti = std::find_if(mTemplateInstantiations.begin(),
mTemplateInstantiations.end(),
FindToken(tok2));
if (ti != mTemplateInstantiations.end())
mTemplateInstantiations.erase(ti);
++indentlevel;
} else if (indentlevel > 0 && tok2->str() == ">")
--indentlevel;
else if (indentlevel == 0 && Token::Match(tok2, ",|>"))
break;
if (indentlevel < 0)
break;
}
// something went wrong, don't call eraseTokens()
// with a nullptr "end" parameter (=all remaining tokens).
if (!tok2)
continue;
// don't strip args from uninstantiated templates
std::list<TokenAndName>::iterator ti2 = std::find_if(mTemplateInstantiations.begin(),
mTemplateInstantiations.end(),
FindName(template1.name));
if (ti2 == mTemplateInstantiations.end())
continue;
eraseTokens(eqtok, tok2);
eqtok->deleteThis();
// update parameter end pointer
template1.paramEnd = template1.token->next()->findClosingBracket();
}
}
}
void TemplateSimplifier::simplifyTemplateAliases()
{
std::list<TokenAndName>::iterator it1, it2;
for (it1 = mTemplateInstantiations.begin(); it1 != mTemplateInstantiations.end();) {
TokenAndName &templateAlias = *it1;
++it1;
Token *startToken = templateAlias.token;
while (Token::Match(startToken->tokAt(-2), "%name% :: %name%"))
startToken = startToken->tokAt(-2);
if (!Token::Match(startToken->tokAt(-4), "> using %name% = %name% ::|<"))
continue;
const std::string aliasName(startToken->strAt(-2));
const Token * const aliasToken1 = startToken;
// Get start token for alias
startToken = startToken->tokAt(-5);
while (Token::Match(startToken, "%name%|<|>|>>|,"))
startToken = startToken->previous();
if (!Token::Match(startToken, "[;{}] template <"))
continue;
// alias parameters..
std::vector<const Token *> aliasParameters;
getTemplateParametersInDeclaration(startToken->tokAt(3), aliasParameters);
std::map<std::string, unsigned int> aliasParameterNames;
for (unsigned int argnr = 0; argnr < aliasParameters.size(); ++argnr)
aliasParameterNames[aliasParameters[argnr]->str()] = argnr;
// Look for alias usages..
const Token *endToken = nullptr;
for (it2 = it1; it2 != mTemplateInstantiations.end(); ++it2) {
TokenAndName &aliasUsage = *it2;
if (aliasUsage.name != aliasName)
continue;
std::vector<std::pair<Token *, Token *>> args;
Token *tok2 = aliasUsage.token->tokAt(2);
while (tok2) {
Token * const start = tok2;
while (tok2 && !Token::Match(tok2, "[,>;{}]")) {
if (tok2->link() && Token::Match(tok2, "(|<|["))
tok2 = tok2->link();
tok2 = tok2->next();
}
args.emplace_back(start, tok2);
if (tok2 && tok2->str() == ",") {
tok2 = tok2->next();
} else {
break;
}
}
if (!tok2 || tok2->str() != ">" || args.size() != aliasParameters.size())
continue;
// Replace template alias code..
aliasUsage.name = templateAlias.name;
if (aliasUsage.name.find(' ') == std::string::npos) {
const Token *temp = aliasToken1;
while (temp && temp != templateAlias.token) {
aliasUsage.token->insertToken(temp->str(), "", true);
temp = temp->next();
}
aliasUsage.token->str(templateAlias.token->str());
} else {
tok2 = TokenList::copyTokens(aliasUsage.token, aliasToken1, templateAlias.token, true);
deleteToken(aliasUsage.token);
aliasUsage.token = tok2;
}
tok2 = aliasUsage.token->next(); // the '<'
const Token * const endToken1 = templateAlias.token->next()->findClosingBracket();
const Token * const endToken2 = TokenList::copyTokens(tok2, templateAlias.token->tokAt(2), endToken1->previous(), false);
for (const Token *tok1 = templateAlias.token->next(); tok2 != endToken2; tok1 = tok1->next(), tok2 = tok2->next()) {
if (!tok2->isName())
continue;
if (aliasParameterNames.find(tok2->str()) == aliasParameterNames.end()) {
// Create template instance..
if (Token::Match(tok1, "%name% <")) {
const std::list<TokenAndName>::iterator it = std::find_if(mTemplateInstantiations.begin(),
mTemplateInstantiations.end(),
FindToken(tok1));
if (it != mTemplateInstantiations.end())
mTemplateInstantiations.emplace_back(tok2, it->scope, it->name, it->nameToken);
}
continue;
}
const unsigned int argnr = aliasParameterNames[tok2->str()];
const Token * const fromStart = args[argnr].first;
const Token * const fromEnd = args[argnr].second->previous();
Token * const destToken = tok2;
tok2 = TokenList::copyTokens(tok2, fromStart, fromEnd, true);
2018-01-05 17:06:02 +01:00
if (tok2 == destToken->next())
tok2 = destToken;
destToken->deleteThis();
}
endToken = endToken1->next();
// Remove alias usage code (parameters)
Token::eraseTokens(tok2, args.back().second);
}
if (endToken) {
// Remove all template instantiations in template alias
for (const Token *tok = startToken; tok != endToken; tok = tok->next()) {
if (!Token::Match(tok, "%name% <"))
continue;
std::list<TokenAndName>::iterator it = std::find_if(mTemplateInstantiations.begin(),
mTemplateInstantiations.end(),
FindToken(tok));
if (it == mTemplateInstantiations.end())
continue;
std::list<TokenAndName>::iterator next = it;
2018-01-05 17:29:40 +01:00
++next;
if (it == it1)
it1 = next;
mTemplateInstantiations.erase(it,next);
}
Token::eraseTokens(startToken, endToken);
}
}
}
bool TemplateSimplifier::instantiateMatch(const Token *instance, const std::size_t numberOfArguments, const char patternAfter[])
{
// if (!Token::simpleMatch(instance, (name + " <").c_str()))
// return false;
if (numberOfArguments != templateParameters(instance->next()))
return false;
if (patternAfter) {
const Token *tok = instance;
unsigned int indentlevel = 0;
for (tok = instance; tok && (tok->str() != ">" || indentlevel > 0); tok = tok->next()) {
if (Token::Match(tok, "<|,|(|:: %name% <") && templateParameters(tok->tokAt(2)) > 0)
++indentlevel;
if (indentlevel > 0 && tok->str() == ">")
--indentlevel;
}
if (!tok || !Token::Match(tok->next(), patternAfter))
return false;
}
// nothing mismatching was found..
return true;
}
// Utility function for TemplateSimplifier::getTemplateNamePosition, that works on template functions
bool TemplateSimplifier::getTemplateNamePositionTemplateFunction(const Token *tok, int &namepos)
{
namepos = 1;
while (tok && tok->next()) {
if (Token::Match(tok->next(), ";|{"))
return false;
// skip decltype(...)
else if (Token::Match(tok, "decltype (")) {
const Token * end = tok->linkAt(1);
while (tok && tok != end) {
tok = tok->next();
namepos++;
}
} else if (Token::Match(tok->next(), "%type% <")) {
const Token *closing = tok->tokAt(2)->findClosingBracket();
if (closing) {
if (closing->strAt(1) == "(" && Tokenizer::isFunctionHead(closing->next(), ";|{|:", true))
return true;
while (tok && tok->next() && tok->next() != closing) {
tok = tok->next();
namepos++;
}
}
} else if (Token::Match(tok->next(), "%type% (") && Tokenizer::isFunctionHead(tok->tokAt(2), ";|{|:", true)) {
return true;
}
tok = tok->next();
namepos++;
}
return false;
}
int TemplateSimplifier::getTemplateNamePosition(const Token *tok, bool forward)
{
// get the position of the template name
int namepos = 0;
if ((forward && Token::Match(tok, "> class|struct|union %type% :|<|;")) ||
(!forward && Token::Match(tok, "> class|struct|union %type% {|:|<")))
namepos = 2;
else if (!getTemplateNamePositionTemplateFunction(tok, namepos))
return -1; // Name not found
return namepos;
}
void TemplateSimplifier::addNamespace(const TokenAndName &templateDeclaration, const Token *tok)
{
// find start of qualification
const Token * tokStart = tok;
int offset = 0;
while (Token::Match(tokStart->tokAt(-2), "%name% ::")) {
tokStart = tokStart->tokAt(-2);
offset -= 2;
}
// decide if namespace needs to be inserted in or appended to token list
const bool insert = tokStart != tok;
std::string::size_type start = 0;
std::string::size_type end = 0;
while ((end = templateDeclaration.scope.find(" ", start)) != std::string::npos) {
std::string token = templateDeclaration.scope.substr(start, end - start);
if (insert)
mTokenList.back()->tokAt(offset)->insertToken(token, "");
else
mTokenList.addtoken(token, tok->linenr(), tok->fileIndex());
start = end + 1;
}
if (insert) {
mTokenList.back()->tokAt(offset)->insertToken(templateDeclaration.scope.substr(start), "");
mTokenList.back()->tokAt(offset)->insertToken("::", "");
} else {
mTokenList.addtoken(templateDeclaration.scope.substr(start), tok->linenr(), tok->fileIndex());
mTokenList.addtoken("::", tok->linenr(), tok->fileIndex());
}
}
bool TemplateSimplifier::alreadyHasNamespace(const TokenAndName &templateDeclaration, const Token *tok) const
{
std::string scope = templateDeclaration.scope;
// get the length in tokens of the namespace
std::string::size_type pos = 0;
int offset = -2;
while ((pos = scope.find("::", pos)) != std::string::npos) {
offset -= 2;
pos += 2;
}
return Token::simpleMatch(tok->tokAt(offset), scope.c_str()) ;
}
void TemplateSimplifier::expandTemplate(
const TokenAndName &templateDeclaration,
const TokenAndName &templateInstantiation,
const std::vector<const Token *> &typeParametersInDeclaration,
const std::string &newName,
bool copy)
{
std::list<ScopeInfo2> scopeInfo;
bool inTemplateDefinition = false;
const Token *startOfTemplateDeclaration = nullptr;
const Token *endOfTemplateDefinition = nullptr;
const Token * const templateDeclarationNameToken = templateDeclaration.nameToken;
const Token * const templateDeclarationToken = templateDeclaration.paramEnd;
const bool isClass = Token::Match(templateDeclaration.paramEnd->next(), "class|struct|union %name% <|{|:");
const bool isFunction = templateDeclarationNameToken->strAt(1) == "(" ||
(templateDeclarationNameToken->strAt(1) == "<" && templateDeclarationNameToken->next()->findClosingBracket()->strAt(1) == "(");
const bool isSpecialization = Token::Match(templateDeclaration.token, "template < >");
// add forward declarations
if (copy && isClass) {
templateDeclaration.token->insertToken(templateDeclarationToken->strAt(1), "", true);
templateDeclaration.token->insertToken(newName, "", true);
templateDeclaration.token->insertToken(";", "", true);
} else if (isFunction && (copy || (!copy && isSpecialization))) {
Token * dst = templateDeclaration.token;
bool isStatic = false;
std::string scope;
Token * start;
Token * end;
auto it = mTemplateForwardDeclarationsMap.find(dst);
if (it != mTemplateForwardDeclarationsMap.end()) {
dst = it->second;
const Token * temp1 = dst->tokAt(1)->findClosingBracket();
const Token * temp2 = temp1->tokAt(getTemplateNamePosition(temp1));
start = temp1->next();
end = temp2->linkAt(1)->next();
} else {
auto it2 = mTemplateUserSpecializationMap.find(dst);
if (it2 != mTemplateUserSpecializationMap.end()) {
dst = it2->second;
isStatic = dst->next()->findClosingBracket()->strAt(1) == "static";
const Token * temp = templateDeclarationNameToken;
while (Token::Match(temp->tokAt(-2), "%name% ::")) {
scope.insert(0, temp->strAt(-2) + " :: ");
temp = temp->tokAt(-2);
}
}
start = templateDeclarationToken->next();
if (templateDeclarationNameToken->strAt(1) == "(")
end = templateDeclarationNameToken->linkAt(1)->next();
else
end = templateDeclarationNameToken->next()->findClosingBracket()->linkAt(1)->next();
}
unsigned int typeindentlevel = 0;
while (!(typeindentlevel == 0 && Token::Match(end, ";|{|:"))) {
if (Token::Match(end, "<|(|{"))
++typeindentlevel;
else if (Token::Match(end, ">|)|}"))
--typeindentlevel;
end = end->next();
}
if (isStatic)
dst->insertToken("static", "", true);
std::map<const Token *, Token *> links;
while (start && start != end) {
unsigned int itype = 0;
while (itype < typeParametersInDeclaration.size() && typeParametersInDeclaration[itype]->str() != start->str())
++itype;
if (itype < typeParametersInDeclaration.size()) {
typeindentlevel = 0;
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
for (const Token *typetok = mTypesUsedInTemplateInstantiation[itype].token;
typetok && (typeindentlevel > 0 || !Token::Match(typetok, ",|>"));
typetok = typetok->next()) {
if (Token::simpleMatch(typetok, ". . .")) {
typetok = typetok->tokAt(2);
continue;
}
if (Token::Match(typetok, "%name% <") && templateParameters(typetok->next()) > 0)
++typeindentlevel;
else if (typeindentlevel > 0 && typetok->str() == ">")
--typeindentlevel;
dst->insertToken(typetok->str(), typetok->originalName(), true);
dst->previous()->isTemplateArg(true);
dst->previous()->isSigned(typetok->isSigned());
dst->previous()->isUnsigned(typetok->isUnsigned());
dst->previous()->isLong(typetok->isLong());
}
} else {
if (isSpecialization && !copy && !scope.empty() && Token::Match(start, (scope + templateDeclarationNameToken->str()).c_str())) {
// skip scope
while (start->strAt(1) != templateDeclarationNameToken->str())
start = start->next();
} else if (start->str() == templateDeclarationNameToken->str()) {
dst->insertToken(newName, "", true);
if (start->strAt(1) == "<")
start = start->next()->findClosingBracket();
} else {
// check if type is a template
if (start->strAt(1) == "<") {
// get the instantiated name
Token * closing = start->next()->findClosingBracket();
std::string name;
const Token * type = start;
while (type && type != closing->next()) {
if (!name.empty())
name += " ";
name += type->str();
type = type->next();
}
// check if type is instantiated
for (const auto & inst : mTemplateInstantiations) {
if (Token::simpleMatch(inst.nameToken, name.c_str())) {
// use the instantiated name
dst->insertToken(name, "", true);
start = closing;
break;
}
}
// just copy the token if it wasn't instantiated
if (start != closing)
dst->insertToken(start->str(), "", true);
} else
dst->insertToken(start->str(), "", true);
}
if (start->link()) {
if (Token::Match(start, "[|{|(")) {
links[start->link()] = dst->previous();
} else if (Token::Match(start, "]|}|)")) {
Token::createMutualLinks(links[start], dst->previous());
links.erase(start);
}
}
}
start = start->next();
}
dst->insertToken(";", "", true);
}
if (copy && (isClass || isFunction)) {
// check if this is an explicit instantiation
Token * start = templateInstantiation.token;
while (start && !Token::Match(start->previous(), "}|;|extern"))
start = start->previous();
if (Token::Match(start, "template !!<")) {
if (start->strAt(-1) == "extern")
start = start->previous();
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
mExplicitInstantiationsToDelete.emplace_back(start, "", "", nullptr);
}
}
for (Token *tok3 = mTokenList.front(); tok3; tok3 = tok3 ? tok3->next() : nullptr) {
if (Token::Match(tok3, "}|namespace|class|struct|union")) {
setScopeInfo(tok3, &scopeInfo);
continue;
}
if (inTemplateDefinition) {
if (!endOfTemplateDefinition && tok3->str() == "{")
endOfTemplateDefinition = tok3->link();
if (tok3 == endOfTemplateDefinition) {
inTemplateDefinition = false;
startOfTemplateDeclaration = nullptr;
}
}
if (tok3->str()=="template") {
if (tok3->next() && tok3->next()->str()=="<") {
std::vector<const Token *> localTypeParametersInDeclaration;
getTemplateParametersInDeclaration(tok3->tokAt(2), localTypeParametersInDeclaration);
if (localTypeParametersInDeclaration.size() != typeParametersInDeclaration.size())
inTemplateDefinition = false; // Partial specialization
else
inTemplateDefinition = true;
} else {
inTemplateDefinition = false; // Only template instantiation
}
startOfTemplateDeclaration = tok3;
}
if (Token::Match(tok3, "(|["))
tok3 = tok3->link();
// Start of template..
if (tok3 == templateDeclarationToken) {
tok3 = tok3->next();
if (tok3->str() == "static")
tok3 = tok3->next();
}
// member function implemented outside class definition
else if (inTemplateDefinition &&
Token::Match(tok3, "%name% <") &&
templateInstantiation.name == tok3->str() &&
instantiateMatch(tok3, typeParametersInDeclaration.size(), ":: ~| %name% (")) {
2018-01-06 15:16:03 +01:00
// there must be template..
bool istemplate = false;
Token * tok5 = nullptr; // start of function return type
for (Token *prev = tok3; prev && !Token::Match(prev, "[;{}]"); prev = prev->previous()) {
2018-01-06 15:16:03 +01:00
if (prev->str() == "template") {
istemplate = true;
tok5 = prev;
2018-01-06 15:16:03 +01:00
break;
}
}
if (!istemplate)
continue;
const Token *tok4 = tok3->next()->findClosingBracket();
while (tok4 && tok4->str() != "(")
tok4 = tok4->next();
if (!Tokenizer::isFunctionHead(tok4, ":{", true))
continue;
// find function return type start
tok5 = tok5->next()->findClosingBracket();
if (tok5)
tok5 = tok5->next();
// copy return type
while (tok5 && tok5 != tok3) {
// replace name if found
if (Token::Match(tok5, "%name% <") && tok5->str() == templateInstantiation.name) {
if (copy) {
if (!templateDeclaration.scope.empty() && tok5->strAt(-1) != "::")
addNamespace(templateDeclaration, tok5);
mTokenList.addtoken(newName, tok5->linenr(), tok5->fileIndex());
tok5 = tok5->next()->findClosingBracket();
} else {
tok5->str(newName);
eraseTokens(tok5, tok5->next()->findClosingBracket()->next());
}
} else if (copy)
mTokenList.addtoken(tok5, tok5->linenr(), tok5->fileIndex());
tok5 = tok5->next();
}
if (copy) {
if (!templateDeclaration.scope.empty() && tok3->strAt(-1) != "::")
addNamespace(templateDeclaration, tok3);
mTokenList.addtoken(newName, tok3->linenr(), tok3->fileIndex());
}
while (tok3 && tok3->str() != "::")
tok3 = tok3->next();
std::list<TokenAndName>::iterator it = std::find_if(mTemplateDeclarations.begin(),
mTemplateDeclarations.end(),
FindToken(startOfTemplateDeclaration));
if (it != mTemplateDeclarations.end())
mMemberFunctionsToDelete.push_back(*it);
}
// not part of template.. go on to next token
else
continue;
std::stack<Token *> brackets; // holds "(", "[" and "{" tokens
// FIXME use full name matching somehow
const std::string lastName = (templateInstantiation.name.find(' ') != std::string::npos) ? templateInstantiation.name.substr(templateInstantiation.name.rfind(' ')+1) : templateInstantiation.name;
for (; tok3; tok3 = tok3->next()) {
if (tok3->isName()) {
// search for this token in the type vector
unsigned int itype = 0;
while (itype < typeParametersInDeclaration.size() && typeParametersInDeclaration[itype]->str() != tok3->str())
++itype;
// replace type with given type..
if (itype < typeParametersInDeclaration.size()) {
unsigned int typeindentlevel = 0;
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
for (const Token *typetok = mTypesUsedInTemplateInstantiation[itype].token;
typetok && (typeindentlevel>0 || !Token::Match(typetok, ",|>"));
typetok = typetok->next()) {
if (Token::simpleMatch(typetok, ". . .")) {
typetok = typetok->tokAt(2);
continue;
}
if (Token::Match(typetok, "%name% <") && templateParameters(typetok->next()) > 0)
++typeindentlevel;
else if (typeindentlevel > 0 && typetok->str() == ">")
--typeindentlevel;
if (copy) {
mTokenList.addtoken(typetok, tok3->linenr(), tok3->fileIndex());
mTokenList.back()->isTemplateArg(true);
}
}
continue;
}
}
2017-12-23 15:41:32 +01:00
// replace name..
2018-01-15 22:03:23 +01:00
if (tok3->str() == lastName) {
if (Token::simpleMatch(tok3->next(), "<")) {
Token *closingBracket = tok3->next()->findClosingBracket();
if (closingBracket) {
// replace multi token name with single token name
if (tok3 == templateDeclarationNameToken ||
Token::Match(tok3, newName.c_str())) {
if (copy) {
mTokenList.addtoken(newName, tok3->linenr(), tok3->fileIndex());
tok3 = closingBracket;
} else {
tok3->str(newName);
eraseTokens(tok3, closingBracket->next());
}
continue;
} else if (!templateDeclaration.scope.empty() &&
!alreadyHasNamespace(templateDeclaration, tok3) &&
closingBracket->strAt(1) != "(") {
if (copy)
addNamespace(templateDeclaration, tok3);
}
}
} else {
if (copy) {
// add namespace if necessary
if (!templateDeclaration.scope.empty() &&
(isClass ? tok3->strAt(1) != "(" : true)) {
addNamespace(templateDeclaration, tok3);
}
mTokenList.addtoken(newName, tok3->linenr(), tok3->fileIndex());
} else if (!Token::Match(tok3->next(), ":|{"))
tok3->str(newName);
continue;
}
}
// copy
if (copy)
mTokenList.addtoken(tok3, tok3->linenr(), tok3->fileIndex());
if (Token::Match(tok3, "%type% <") && Token::Match(tok3->next()->findClosingBracket(), ">|>>")) {
const Token *closingBracket = tok3->next()->findClosingBracket();
if (Token::simpleMatch(closingBracket->next(), "&")) {
int num = 0;
const Token *par = tok3->next();
while (num < typeParametersInDeclaration.size() && par != closingBracket) {
const std::string pattern("[<,] " + typeParametersInDeclaration[num]->str() + " [,>]");
if (!Token::Match(par, pattern.c_str()))
break;
++num;
par = par->tokAt(2);
}
if (num < typeParametersInDeclaration.size() || par != closingBracket)
continue;
}
std::string name = tok3->str();
for (const Token *prev = tok3->tokAt(-2); Token::Match(prev, "%name% ::"); prev = prev->tokAt(-2))
name = prev->str() + " :: " + name;
mTemplateInstantiations.emplace_back(mTokenList.back(), getScopeName(scopeInfo), name, tok3);
}
// link() newly tokens manually
else if (copy) {
if (tok3->str() == "{") {
brackets.push(mTokenList.back());
} else if (tok3->str() == "(") {
brackets.push(mTokenList.back());
} else if (tok3->str() == "[") {
brackets.push(mTokenList.back());
} else if (tok3->str() == "}") {
assert(brackets.empty() == false && brackets.top()->str() == "{");
Token::createMutualLinks(brackets.top(), mTokenList.back());
if (tok3->strAt(1) == ";") {
const Token * tokSemicolon = tok3->next();
mTokenList.addtoken(tokSemicolon, tokSemicolon->linenr(), tokSemicolon->fileIndex());
}
brackets.pop();
if (brackets.empty()) {
inTemplateDefinition = false;
break;
}
} else if (tok3->str() == ")") {
assert(brackets.empty() == false && brackets.top()->str() == "(");
Token::createMutualLinks(brackets.top(), mTokenList.back());
brackets.pop();
} else if (tok3->str() == "]") {
assert(brackets.empty() == false && brackets.top()->str() == "[");
Token::createMutualLinks(brackets.top(), mTokenList.back());
brackets.pop();
}
}
}
assert(brackets.empty());
}
}
static bool isLowerThanLogicalAnd(const Token *lower)
{
return lower->isAssignmentOp() || Token::Match(lower, "}|;|(|[|]|)|,|?|:|%oror%|return|throw|case");
}
static bool isLowerThanOr(const Token* lower)
{
return isLowerThanLogicalAnd(lower) || lower->str() == "&&";
}
static bool isLowerThanXor(const Token* lower)
{
return isLowerThanOr(lower) || lower->str() == "|";
}
static bool isLowerThanAnd(const Token* lower)
{
return isLowerThanXor(lower) || lower->str() == "^";
}
static bool isLowerThanShift(const Token* lower)
{
2015-11-28 09:26:44 +01:00
return isLowerThanAnd(lower) || lower->str() == "&";
}
static bool isLowerThanPlusMinus(const Token* lower)
{
return isLowerThanShift(lower) || Token::Match(lower, "%comp%|<<|>>");
}
static bool isLowerThanMulDiv(const Token* lower)
{
return isLowerThanPlusMinus(lower) || Token::Match(lower, "+|-");
}
static bool isLowerEqualThanMulDiv(const Token* lower)
{
return isLowerThanMulDiv(lower) || Token::Match(lower, "[*/%]");
}
bool TemplateSimplifier::simplifyNumericCalculations(Token *tok)
{
bool ret = false;
// (1-2)
while (tok->tokAt(3) && tok->isNumber() && tok->tokAt(2)->isNumber()) { // %any% %num% %any% %num% %any%
const Token *before = tok->previous();
if (!before)
break;
const Token* op = tok->next();
const Token* after = tok->tokAt(3);
2018-05-29 22:41:37 +02:00
const std::string &num1 = op->previous()->str();
const std::string &num2 = op->next()->str();
if (Token::Match(before, "* %num% /") && (num2 != "0") && num1 == MathLib::multiply(num2, MathLib::divide(num1, num2))) {
// Division where result is a whole number
2018-05-29 22:41:37 +02:00
} else if (!((op->str() == "*" && (isLowerThanMulDiv(before) || before->str() == "*") && isLowerEqualThanMulDiv(after)) || // associative
(Token::Match(op, "[/%]") && isLowerThanMulDiv(before) && isLowerEqualThanMulDiv(after)) || // NOT associative
(Token::Match(op, "[+-]") && isLowerThanMulDiv(before) && isLowerThanMulDiv(after)) || // Only partially (+) associative, but handled later
(Token::Match(op, ">>|<<") && isLowerThanShift(before) && isLowerThanPlusMinus(after)) || // NOT associative
(op->str() == "&" && isLowerThanShift(before) && isLowerThanShift(after)) || // associative
(op->str() == "^" && isLowerThanAnd(before) && isLowerThanAnd(after)) || // associative
(op->str() == "|" && isLowerThanXor(before) && isLowerThanXor(after)) || // associative
(op->str() == "&&" && isLowerThanOr(before) && isLowerThanOr(after)) ||
(op->str() == "||" && isLowerThanLogicalAnd(before) && isLowerThanLogicalAnd(after))))
break;
// Don't simplify "%num% / 0"
if (Token::Match(op, "[/%] 0"))
break;
// Integer operations
if (Token::Match(op, ">>|<<|&|^|%or%")) {
// Don't simplify if operand is negative, shifting with negative
// operand is UB. Bitmasking with negative operand is implementation
// defined behaviour.
if (MathLib::isNegative(num1) || MathLib::isNegative(num2))
break;
2018-05-29 22:41:37 +02:00
const MathLib::value v1(num1);
const MathLib::value v2(num2);
if (!v1.isInt() || !v2.isInt())
break;
switch (op->str()[0]) {
case '<':
tok->str((v1 << v2).str());
break;
case '>':
tok->str((v1 >> v2).str());
break;
case '&':
tok->str((v1 & v2).str());
break;
case '|':
tok->str((v1 | v2).str());
break;
case '^':
tok->str((v1 ^ v2).str());
break;
};
}
// Logical operations
else if (Token::Match(op, "%oror%|&&")) {
2018-05-29 22:41:37 +02:00
const bool op1 = !MathLib::isNullValue(num1);
const bool op2 = !MathLib::isNullValue(num2);
2018-04-04 21:51:31 +02:00
const bool result = (op->str() == "||") ? (op1 || op2) : (op1 && op2);
tok->str(result ? "1" : "0");
}
else if (Token::Match(tok->previous(), "- %num% - %num%"))
2018-05-29 22:41:37 +02:00
tok->str(MathLib::add(num1, num2));
else if (Token::Match(tok->previous(), "- %num% + %num%"))
2018-05-29 22:41:37 +02:00
tok->str(MathLib::subtract(num1, num2));
else {
try {
2018-05-29 22:41:37 +02:00
tok->str(MathLib::calculate(num1, num2, op->str()[0]));
} catch (InternalError &e) {
e.token = tok;
throw;
}
}
tok->deleteNext(2);
ret = true;
}
return ret;
}
// TODO: This is not the correct class for simplifyCalculations(), so it
// should be moved away.
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
bool TemplateSimplifier::simplifyCalculations(Token* frontToken)
{
bool ret = false;
2018-12-21 13:54:59 +01:00
if (!frontToken) {
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
frontToken = mTokenList.front();
}
for (Token *tok = frontToken; tok; tok = tok->next()) {
// Remove parentheses around variable..
// keep parentheses here: dynamic_cast<Fred *>(p);
// keep parentheses here: A operator * (int);
// keep parentheses here: int ( * ( * f ) ( ... ) ) (int) ;
// keep parentheses here: int ( * * ( * compilerHookVector ) (void) ) ( ) ;
// keep parentheses here: operator new [] (size_t);
// keep parentheses here: Functor()(a ... )
// keep parentheses here: ) ( var ) ;
if ((Token::Match(tok->next(), "( %name% ) ;|)|,|]") ||
(Token::Match(tok->next(), "( %name% ) %cop%") && (tok->tokAt(2)->varId()>0 || !Token::Match(tok->tokAt(4), "[*&+-~]")))) &&
!tok->isName() &&
tok->str() != ">" &&
tok->str() != ")" &&
tok->str() != "]") {
tok->deleteNext();
tok = tok->next();
tok->deleteNext();
ret = true;
}
if (Token::Match(tok->previous(), "(|&&|%oror% %char% %comp% %num% &&|%oror%|)")) {
tok->str(MathLib::toString(MathLib::toLongNumber(tok->str())));
}
if (tok->isNumber()) {
if (simplifyNumericCalculations(tok)) {
ret = true;
Token *prev = tok->tokAt(-2);
while (prev && simplifyNumericCalculations(prev)) {
tok = prev;
prev = prev->tokAt(-2);
}
}
// Remove redundant conditions (0&&x) (1||x)
if (Token::Match(tok->previous(), "[(=,] 0 &&") ||
Token::Match(tok->previous(), "[(=,] 1 %oror%")) {
unsigned int par = 0;
const Token *tok2 = tok;
2018-04-04 21:51:31 +02:00
const bool andAnd = (tok->next()->str() == "&&");
for (; tok2; tok2 = tok2->next()) {
if (tok2->str() == "(" || tok2->str() == "[")
++par;
else if (tok2->str() == ")" || tok2->str() == "]") {
if (par == 0)
break;
--par;
} else if (par == 0 && isLowerThanLogicalAnd(tok2) && (andAnd || tok2->str() != "||"))
break;
}
if (tok2) {
eraseTokens(tok, tok2);
ret = true;
}
continue;
}
if (tok->str() == "0") {
if ((Token::Match(tok->previous(), "[+-] 0 %cop%|;") && isLowerThanMulDiv(tok->next())) ||
(Token::Match(tok->previous(), "%or% 0 %cop%|;") && isLowerThanXor(tok->next()))) {
tok = tok->previous();
if (Token::Match(tok->tokAt(-4), "[;{}] %name% = %name% [+-|] 0 ;") &&
tok->strAt(-3) == tok->previous()->str()) {
tok = tok->tokAt(-4);
tok->deleteNext(5);
} else {
tok = tok->previous();
tok->deleteNext(2);
}
ret = true;
} else if (Token::Match(tok->previous(), "[=([,] 0 [+|]") ||
Token::Match(tok->previous(), "return|case 0 [+|]")) {
tok = tok->previous();
tok->deleteNext(2);
ret = true;
} else if (Token::Match(tok->previous(), "[=[(,] 0 * %name%|%num% ,|]|)|;|=|%cop%") ||
Token::Match(tok->previous(), "[=[(,] 0 * (") ||
Token::Match(tok->previous(), "return|case 0 *|&& %name%|%num% ,|:|;|=|%cop%") ||
Token::Match(tok->previous(), "return|case 0 *|&& (")) {
tok->deleteNext();
if (tok->next()->str() == "(")
eraseTokens(tok, tok->next()->link());
tok->deleteNext();
ret = true;
} else if (Token::Match(tok->previous(), "[=[(,] 0 && *|& %any% ,|]|)|;|=|%cop%") ||
Token::Match(tok->previous(), "return|case 0 && *|& %any% ,|:|;|=|%cop%")) {
tok->deleteNext();
tok->deleteNext();
if (tok->next()->str() == "(")
eraseTokens(tok, tok->next()->link());
tok->deleteNext();
ret = true;
}
}
if (tok->str() == "1") {
if (Token::Match(tok->previous(), "[=[(,] 1 %oror% %any% ,|]|)|;|=|%cop%") ||
Token::Match(tok->previous(), "return|case 1 %oror% %any% ,|:|;|=|%cop%")) {
tok->deleteNext();
if (tok->next()->str() == "(")
eraseTokens(tok, tok->next()->link());
tok->deleteNext();
ret = true;
} else if (Token::Match(tok->previous(), "[=[(,] 1 %oror% *|& %any% ,|]|)|;|=|%cop%") ||
Token::Match(tok->previous(), "return|case 1 %oror% *|& %any% ,|:|;|=|%cop%")) {
tok->deleteNext();
tok->deleteNext();
if (tok->next()->str() == "(")
eraseTokens(tok, tok->next()->link());
tok->deleteNext();
ret = true;
}
}
if (Token::Match(tok->tokAt(-2), "%any% * 1") || Token::Match(tok->previous(), "%any% 1 *")) {
tok = tok->previous();
if (tok->str() == "*")
tok = tok->previous();
tok->deleteNext(2);
ret = true;
}
// Remove parentheses around number..
if (Token::Match(tok->tokAt(-2), "%op%|< ( %num% )") && tok->strAt(-2) != ">") {
tok = tok->previous();
tok->deleteThis();
tok->deleteNext();
ret = true;
}
if (Token::Match(tok->previous(), "( 0 [|+]") ||
Token::Match(tok->previous(), "[|+-] 0 )")) {
tok = tok->previous();
if (Token::Match(tok, "[|+-]"))
tok = tok->previous();
tok->deleteNext(2);
ret = true;
}
if (Token::Match(tok, "%num% %comp% %num%") &&
MathLib::isInt(tok->str()) &&
MathLib::isInt(tok->strAt(2))) {
if (Token::Match(tok->previous(), "(|&&|%oror%") && Token::Match(tok->tokAt(3), ")|&&|%oror%|?")) {
const MathLib::bigint op1(MathLib::toLongNumber(tok->str()));
const std::string &cmp(tok->next()->str());
const MathLib::bigint op2(MathLib::toLongNumber(tok->strAt(2)));
std::string result;
if (cmp == "==")
result = (op1 == op2) ? "1" : "0";
else if (cmp == "!=")
result = (op1 != op2) ? "1" : "0";
else if (cmp == "<=")
result = (op1 <= op2) ? "1" : "0";
else if (cmp == ">=")
result = (op1 >= op2) ? "1" : "0";
else if (cmp == "<")
result = (op1 < op2) ? "1" : "0";
2015-08-15 19:46:31 +02:00
else
result = (op1 > op2) ? "1" : "0";
tok->str(result);
tok->deleteNext(2);
ret = true;
tok = tok->previous();
}
}
}
}
return ret;
}
const Token * TemplateSimplifier::getTemplateParametersInDeclaration(
const Token * tok,
std::vector<const Token *> & typeParametersInDeclaration)
{
typeParametersInDeclaration.clear();
for (; tok && tok->str() != ">"; tok = tok->next()) {
if (Token::Match(tok, "%name% ,|>|="))
typeParametersInDeclaration.push_back(tok);
}
return tok;
}
bool TemplateSimplifier::matchSpecialization(
const Token *templateDeclarationNameToken,
const Token *templateInstantiationNameToken,
const std::list<const Token *> & specializations)
{
// Is there a matching specialization?
for (std::list<const Token *>::const_iterator it = specializations.begin(); it != specializations.end(); ++it) {
if (!Token::Match(*it, "%name% <"))
continue;
const Token *startToken = (*it);
while (startToken->previous() && !Token::Match(startToken->previous(), "[;{}]"))
startToken = startToken->previous();
if (!Token::simpleMatch(startToken, "template <"))
continue;
std::vector<const Token *> templateParameters;
getTemplateParametersInDeclaration(startToken->tokAt(2), templateParameters);
const Token *instToken = templateInstantiationNameToken->tokAt(2);
const Token *declToken = (*it)->tokAt(2);
const Token * const endToken = (*it)->next()->findClosingBracket();
while (declToken != endToken) {
if (declToken->str() != instToken->str()) {
int nr = 0;
while (nr < templateParameters.size() && templateParameters[nr]->str() != declToken->str())
++nr;
if (nr == templateParameters.size())
break;
}
declToken = declToken->next();
instToken = instToken->next();
}
if (declToken == endToken && instToken->str() == ">") {
// specialization matches.
return templateDeclarationNameToken == *it;
}
}
// No specialization matches. Return true if the declaration is not a specialization.
return Token::Match(templateDeclarationNameToken, "%name% !!<");
}
std::string TemplateSimplifier::getNewName(
Token *tok2,
std::list<std::string> &typeStringsUsedInTemplateInstantiation)
{
std::string typeForNewName;
unsigned int indentlevel = 0;
for (Token *tok3 = tok2->tokAt(2); tok3 && (indentlevel > 0 || tok3->str() != ">"); tok3 = tok3->next()) {
// #2648 - unhandled parentheses => bail out
// #2721 - unhandled [ => bail out
if (Token::Match(tok3, "(|[")) {
typeForNewName.clear();
break;
}
if (!tok3->next()) {
typeForNewName.clear();
break;
}
if (Token::Match(tok3->tokAt(-2), "<|,|:: %name% <") && templateParameters(tok3) > 0)
++indentlevel;
else if (indentlevel > 0 && Token::Match(tok3, "> [,>]"))
--indentlevel;
if (indentlevel == 0 && Token::Match(tok3->previous(), "[<,]")) {
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
mTypesUsedInTemplateInstantiation.emplace_back(tok3, "", "", nullptr);
}
const bool constconst = tok3->str() == "const" && tok3->strAt(1) == "const";
if (!constconst) {
typeStringsUsedInTemplateInstantiation.push_back(tok3->str());
}
// add additional type information
if (!constconst && !Token::Match(tok3, "class|struct|enum")) {
if (tok3->isUnsigned())
typeForNewName += "unsigned";
else if (tok3->isSigned())
typeForNewName += "signed";
if (tok3->isLong())
typeForNewName += "long";
if (!typeForNewName.empty())
typeForNewName += ' ';
typeForNewName += tok3->str();
}
}
return typeForNewName;
}
bool TemplateSimplifier::simplifyTemplateInstantiations(
const TokenAndName &templateDeclaration,
const std::list<const Token *> &specializations,
2016-02-12 12:05:32 +01:00
const std::time_t maxtime,
std::set<std::string> &expandedtemplates)
{
// this variable is not used at the moment. The intention was to
// allow continuous instantiations until all templates has been expanded
//bool done = false;
// Contains tokens such as "T"
std::vector<const Token *> typeParametersInDeclaration;
getTemplateParametersInDeclaration(templateDeclaration.token->tokAt(2), typeParametersInDeclaration);
const bool printDebug = mSettings->debugwarnings;
const bool specialized = Token::simpleMatch(templateDeclaration.token, "template < >");
bool isfunc = false;
if (templateDeclaration.nameToken->strAt(1) == "(")
isfunc = true;
else if (templateDeclaration.nameToken->strAt(1) == "<") {
const Token *tok1 = templateDeclaration.nameToken->tokAt(1)->findClosingBracket();
if (tok1 && tok1->strAt(1) == "(")
isfunc = true;
}
// locate template usage..
std::string::size_type numberOfTemplateInstantiations = mTemplateInstantiations.size();
unsigned int recursiveCount = 0;
bool instantiated = false;
for (const TokenAndName &instantiation : mTemplateInstantiations) {
if (numberOfTemplateInstantiations != mTemplateInstantiations.size()) {
numberOfTemplateInstantiations = mTemplateInstantiations.size();
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
simplifyCalculations(instantiation.token);
++recursiveCount;
if (recursiveCount > 100) {
// bail out..
break;
}
}
// already simplified
if (!Token::Match(instantiation.token, "%name% <"))
continue;
if (instantiation.name != templateDeclaration.name)
continue;
if (!matchSpecialization(templateDeclaration.nameToken, instantiation.token, specializations))
continue;
Token * const tok2 = instantiation.token;
if (mErrorLogger && !mTokenList.getFiles().empty())
mErrorLogger->reportProgress(mTokenList.getFiles()[0], "TemplateSimplifier::simplifyTemplateInstantiations()", tok2->progressValue());
2016-02-12 12:05:32 +01:00
#ifdef MAXTIME
if (std::time(0) > maxtime)
return false;
#else
(void)maxtime;
#endif
assert(mTokenList.validateToken(tok2)); // that assertion fails on examples from #6021
const Token *startToken = tok2;
while (Token::Match(startToken->tokAt(-2), "%name% :: %name%") ||
Token::Match(startToken->tokAt(-2), "> :: %name%")) {
if (startToken->strAt(-2) == ">") {
const Token * tok3 = startToken->tokAt(-2)->findOpeningBracket();
if (tok3)
startToken = tok3->previous();
else
break;
} else
startToken = startToken->tokAt(-2);
}
if (Token::Match(startToken->previous(), ";|{|}|=|const") &&
(!specialized && !instantiateMatch(tok2, typeParametersInDeclaration.size(), isfunc ? "(" : "*|&| %name%")))
continue;
// New type..
mTypesUsedInTemplateInstantiation.clear();
std::list<std::string> typeStringsUsedInTemplateInstantiation;
std::string typeForNewName = getNewName(tok2, typeStringsUsedInTemplateInstantiation);
if (typeForNewName.empty() || (!typeParametersInDeclaration.empty() && typeParametersInDeclaration.size() != mTypesUsedInTemplateInstantiation.size())) {
if (printDebug && mErrorLogger) {
std::list<const Token *> callstack(1, tok2);
mErrorLogger->reportErr(ErrorLogger::ErrorMessage(callstack, &mTokenList, Severity::debug, "debug",
"Failed to instantiate template \"" + instantiation.name + "\". The checking continues anyway.", false));
}
if (typeForNewName.empty())
continue;
break;
}
// New classname/funcname..
const std::string newName(templateDeclaration.name + " < " + typeForNewName + " >");
if (expandedtemplates.find(newName) == expandedtemplates.end()) {
expandedtemplates.insert(newName);
expandTemplate(templateDeclaration, instantiation, typeParametersInDeclaration, newName, !specialized);
instantiated = true;
}
// Replace all these template usages..
replaceTemplateUsage(tok2, instantiation.name, typeStringsUsedInTemplateInstantiation, newName);
}
// process uninstantiated templates
// TODO: remove the specialized check and handle all uninstantiated templates someday.
if (!instantiated && specialized) {
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
simplifyCalculations(templateDeclaration.token);
Token * tok2 = const_cast<Token *>(templateDeclaration.nameToken);
if (mErrorLogger && !mTokenList.getFiles().empty())
mErrorLogger->reportProgress(mTokenList.getFiles()[0], "TemplateSimplifier::simplifyTemplateInstantiations()", tok2->progressValue());
#ifdef MAXTIME
if (std::time(0) > maxtime)
return false;
#else
(void)maxtime;
#endif
assert(mTokenList.validateToken(tok2)); // that assertion fails on examples from #6021
Token *startToken = tok2;
while (Token::Match(startToken->tokAt(-2), "%name% :: %name%") ||
Token::Match(startToken->tokAt(-2), "> :: %name%")) {
if (startToken->strAt(-2) == ">") {
const Token * tok3 = startToken->tokAt(-2)->findOpeningBracket();
if (tok3)
startToken = tok3->previous();
else
break;
} else
startToken = startToken->tokAt(-2);
}
if (Token::Match(startToken->previous(), ";|{|}|=|const") &&
(!specialized && !instantiateMatch(tok2, typeParametersInDeclaration.size(), isfunc ? "(" : "*|&| %name%")))
return false;
// already simplified
if (!Token::Match(tok2, "%name% <"))
return false;
if (!matchSpecialization(templateDeclaration.nameToken, tok2, specializations))
return false;
// New type..
mTypesUsedInTemplateInstantiation.clear();
std::list<std::string> typeStringsUsedInTemplateInstantiation;
std::string typeForNewName = getNewName(tok2, typeStringsUsedInTemplateInstantiation);
if (typeForNewName.empty()) {
if (printDebug && mErrorLogger) {
std::list<const Token *> callstack(1, tok2);
mErrorLogger->reportErr(ErrorLogger::ErrorMessage(callstack, &mTokenList, Severity::debug, "debug",
"Failed to instantiate template \"" + templateDeclaration.name + "\". The checking continues anyway.", false));
}
return false;
}
// New classname/funcname..
const std::string newName(templateDeclaration.name + " < " + typeForNewName + " >");
if (expandedtemplates.find(newName) == expandedtemplates.end()) {
expandedtemplates.insert(newName);
expandTemplate(templateDeclaration, templateDeclaration, typeParametersInDeclaration, newName, false);
instantiated = true;
}
// Replace all these template usages..
replaceTemplateUsage(startToken, templateDeclaration.name, typeStringsUsedInTemplateInstantiation, newName);
}
// Template has been instantiated .. then remove the template declaration
return instantiated;
}
static bool matchTemplateParameters(const Token *nameTok, const std::list<std::string> &strings)
{
std::list<std::string>::const_iterator it = strings.begin();
const Token *tok = nameTok->tokAt(2);
while (tok && it != strings.end() && *it == tok->str()) {
tok = tok->next();
++it;
}
return it == strings.end() && tok && tok->str() == ">";
}
void TemplateSimplifier::replaceTemplateUsage(
Token * const instantiationToken,
const std::string &templateName,
const std::list<std::string> &typeStringsUsedInTemplateInstantiation,
const std::string &newName)
{
std::list<ScopeInfo2> scopeInfo;
std::list< std::pair<Token *, Token *> > removeTokens;
for (Token *nameTok = instantiationToken; nameTok; nameTok = nameTok->next()) {
if (Token::Match(nameTok, "}|namespace|class|struct|union")) {
setScopeInfo(nameTok, &scopeInfo);
continue;
}
if (!Token::Match(nameTok, "%name% <"))
continue;
if (!matchTemplateParameters(nameTok, typeStringsUsedInTemplateInstantiation))
continue;
// FIXME Proper name matching
const std::string lastName(templateName.find(' ') == std::string::npos ? templateName : templateName.substr(templateName.rfind(' ') + 1));
if (lastName != nameTok->str())
continue;
// match parameters
Token * tok2 = nameTok->tokAt(2);
unsigned int typeCountInInstantiation = 1U; // There is always at least one type
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
const Token *typetok = (!mTypesUsedInTemplateInstantiation.empty()) ? mTypesUsedInTemplateInstantiation[0].token : nullptr;
unsigned int indentlevel2 = 0; // indentlevel for tokgt
while (tok2 && (indentlevel2 > 0 || tok2->str() != ">")) {
if (tok2->str() == "<" && templateParameters(tok2) > 0)
++indentlevel2;
else if (indentlevel2 > 0 && Token::Match(tok2, "> [,>]"))
--indentlevel2;
else if (indentlevel2 == 0) {
if (tok2->str() != ",") {
if (!typetok ||
tok2->isUnsigned() != typetok->isUnsigned() ||
tok2->isSigned() != typetok->isSigned() ||
tok2->isLong() != typetok->isLong()) {
break;
}
typetok = typetok->next();
} else {
if (typeCountInInstantiation < mTypesUsedInTemplateInstantiation.size())
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
typetok = mTypesUsedInTemplateInstantiation[typeCountInInstantiation++].token;
else
typetok = nullptr;
}
}
tok2 = tok2->next();
}
if (!tok2)
break;
// matching template usage => replace tokens..
// Foo < int > => Foo<int>
if (tok2->str() == ">" && typeCountInInstantiation == mTypesUsedInTemplateInstantiation.size()) {
const Token * const nameTok1 = nameTok;
nameTok->str(newName);
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
for (Token *tok = nameTok1->next(); tok != tok2; tok = tok->next()) {
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
if (tok->isName() && !tok->templateSimplifierPointers().empty()) {
std::list<TokenAndName>::iterator ti;
for (ti = mTemplateInstantiations.begin(); ti != mTemplateInstantiations.end();) {
2018-12-21 13:54:59 +01:00
if (ti->token == tok) {
mTemplateInstantiations.erase(ti++);
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
break;
} else {
++ti;
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
}
}
}
}
removeTokens.emplace_back(nameTok, tok2->next());
}
nameTok = tok2;
}
while (!removeTokens.empty()) {
eraseTokens(removeTokens.back().first, removeTokens.back().second);
removeTokens.pop_back();
}
}
static std::string getPathName(const TemplateSimplifier::TokenAndName & decl)
{
std::string name = decl.scope;
if (!name.empty())
name += " :: ";
if (decl.nameToken->strAt(-1) == "::") {
const Token * start = decl.nameToken;
while (Token::Match(start->tokAt(-2), "%name% ::"))
start = start->tokAt(-2);
while (start != decl.nameToken) {
name += (start->str() + " ");
start = start->next();
}
}
name += decl.name;
return name;
}
void TemplateSimplifier::getUserDefinedSpecializations()
{
// try to locate a matching declaration for each user defined specialization
for (auto & spec : mTemplateDeclarations) {
if (Token::Match(spec.token, "template < >")) {
std::string specName = getPathName(spec);
bool found = false;
for (auto & decl : mTemplateDeclarations) {
if (Token::Match(decl.token, "template < >"))
continue;
std::string declName = getPathName(decl);
// make sure the scopes and names match
if (specName == declName) {
// @todo make sure function parameters also match
mTemplateUserSpecializationMap[spec.token] = decl.token;
}
}
if (!found) {
for (auto & decl : mTemplateForwardDeclarations) {
std::string declName = getPathName(decl);
// make sure the scopes and names match
if (specName == declName) {
// @todo make sure function parameters also match
mTemplateUserSpecializationMap[spec.token] = decl.token;
}
}
}
}
}
}
void TemplateSimplifier::fixForwardDeclaredDefaultArgumentValues()
{
// try to locate a matching declaration for each forward declaration
for (const auto & forwardDecl : mTemplateForwardDeclarations) {
std::vector<const Token *> params1;
getTemplateParametersInDeclaration(forwardDecl.token->tokAt(2), params1);
for (auto & decl : mTemplateDeclarations) {
std::vector<const Token *> params2;
getTemplateParametersInDeclaration(decl.token->tokAt(2), params2);
// make sure the number of arguments match
if (params1.size() == params2.size()) {
std::string declName = getPathName(decl);
std::string forwardDeclName = getPathName(forwardDecl);
// make sure the scopes and names match
if (forwardDeclName == declName) {
// save forward declaration for lookup later
if ((decl.nameToken->strAt(1) == "(" && forwardDecl.nameToken->strAt(1) == "(") ||
(decl.nameToken->strAt(1) == "{" && forwardDecl.nameToken->strAt(1) == ";")) {
mTemplateForwardDeclarationsMap[decl.token] = forwardDecl.token;
}
for (size_t k = 0; k < params1.size(); k++) {
// copy default value to declaration if not present
if (params1[k]->strAt(1) == "=" && params2[k]->strAt(1) != "=") {
const_cast<Token *>(params2[k])->insertToken(params1[k]->strAt(2));
const_cast<Token *>(params2[k])->insertToken(params1[k]->strAt(1));
}
}
// update parameter end pointer
decl.paramEnd = decl.token->next()->findClosingBracket();
}
}
}
}
}
void TemplateSimplifier::simplifyTemplates(
2016-02-12 12:05:32 +01:00
const std::time_t maxtime,
bool &codeWithTemplates)
{
// TODO: 2 is not the ideal number of loops.
// We should loop until the number of declarations is 0 but we can't
// do that until we instantiate unintstantiated templates with their symbolic types.
// That will allow the uninstantiated template code to be removed from the symbol database.
// Unfortunately the template simplifier doesn't handle namespaces properly so
// the uninstantiated template code in the symbol database can't be removed until #8768
// is fixed.
for (int i = 0; i < 2; ++i) {
if (i) {
mTemplateDeclarations.clear();
mTemplateForwardDeclarations.clear();
mTemplateForwardDeclarationsMap.clear();
mTemplateUserSpecializationMap.clear();
mTemplateInstantiations.clear();
mInstantiatedTemplates.clear();
mExplicitInstantiationsToDelete.clear();
}
bool hasTemplates = getTemplateDeclarations();
if (i == 0) {
codeWithTemplates = hasTemplates;
if (hasTemplates) {
// There are templates..
// Remove "typename" unless used in template arguments..
for (Token *tok = mTokenList.front(); tok; tok = tok->next()) {
if (tok->str() == "typename")
tok->deleteThis();
if (Token::simpleMatch(tok, "template <")) {
while (tok && tok->str() != ">")
tok = tok->next();
if (!tok)
break;
}
}
}
}
// Make sure there is something to simplify.
if (mTemplateDeclarations.empty())
return;
// Copy default argument values from forward declaration to declaration
fixForwardDeclaredDefaultArgumentValues();
// Locate user defined specializations.
getUserDefinedSpecializations();
// Locate possible instantiations of templates..
getTemplateInstantiations();
// Template arguments with default values
useDefaultArgumentValues();
simplifyTemplateAliases();
std::set<std::string> expandedtemplates;
for (std::list<TokenAndName>::reverse_iterator iter1 = mTemplateDeclarations.rbegin(); iter1 != mTemplateDeclarations.rend(); ++iter1) {
// get specializations..
std::list<const Token *> specializations;
for (std::list<TokenAndName>::const_iterator iter2 = mTemplateDeclarations.begin(); iter2 != mTemplateDeclarations.end(); ++iter2) {
if (iter1->name == iter2->name)
specializations.push_back(iter2->nameToken);
}
const bool instantiated = simplifyTemplateInstantiations(
*iter1,
specializations,
maxtime,
expandedtemplates);
if (instantiated)
mInstantiatedTemplates.push_back(*iter1);
}
for (std::list<TokenAndName>::const_iterator it = mInstantiatedTemplates.begin(); it != mInstantiatedTemplates.end(); ++it) {
2018-05-30 14:33:55 +02:00
std::list<TokenAndName>::iterator decl;
for (decl = mTemplateDeclarations.begin(); decl != mTemplateDeclarations.end(); ++decl) {
2018-05-30 14:33:55 +02:00
if (decl->token == it->token)
break;
}
if (decl != mTemplateDeclarations.end()) {
if (Token::simpleMatch(it->token, "template < >")) {
// delete the "template < >"
Token * tok = it->token;
tok->deleteNext(2);
tok->deleteThis();
} else {
// remove forward declaration if found
auto it1 = mTemplateForwardDeclarationsMap.find(it->token);
if (it1 != mTemplateForwardDeclarationsMap.end())
removeTemplate(it1->second);
removeTemplate(it->token);
}
mTemplateDeclarations.erase(decl);
}
}
// remove out of line member functions
while (!mMemberFunctionsToDelete.empty()) {
const std::list<TokenAndName>::iterator it = std::find_if(mTemplateDeclarations.begin(),
mTemplateDeclarations.end(),
FindToken(mMemberFunctionsToDelete.begin()->token));
// multiple functions can share the same declaration so make sure it hasn't already been deleted
if (it != mTemplateDeclarations.end()) {
removeTemplate(it->token);
mTemplateDeclarations.erase(it);
}
mMemberFunctionsToDelete.erase(mMemberFunctionsToDelete.begin());
}
// remove explicit instantiations
for (size_t j = 0; j < mExplicitInstantiationsToDelete.size(); ++j) {
Several fairly significant optimisations (#1518) * Code changes for Token::mImpl optimisation * Added new TokenImpl optimisation Moving members to the TokenImpl struct reduces the size of the Token class, which is a fairly significant optimisation. In my testing on Windows with 32-bit Release-PCRE, this change reduced the size of the Token class from 108 bits to 52 bits and reduced run-time of my test case by around 20%. * Several optimisations Deleted some code that ran very slowly and did nothing, as there is no need to change a Token's string to null if you are about to delete it. Added a frontToken to simplifyCalculations to reduce the amount of work it has to do on already-simplified calculations. Moved template removal to the end of the list as this reduces redundant iteration and saves time. * Added tok argument to simplifyCalculations This means callers can avoid unnecessary work if they know which tokens have already been simplified. Passing nullptr indicates the original behaviour (starting from the front of the list). * Removed mention of member from another change * Re-added and optimised some code deleted in error Changing mTemplateInstantiations to a vector avoids the high cost of doing repeated linear searches. Changing how the code iterates through the array was necessary because the vector can be resized at several points during the loop, which breaks existing references and iterators. * Changed mTemplateInstantiations to a vector This is an optimisation that makes repeated linear searches of this collection significantly faster. Also added a copy constructor to TokenAndName so code can make copies of these objects to keep a reference if a vector gets resized. * A cleaner optimisation to removing template tokens This reverts the previous change to made mInstantiatedTemplates a vector and the iterator changes to support this, and makes mTypesUsedInTemplateInstantiation so the eraseTokens logic can be unified. * Reverted vector to list Also made mTypesUsedInTemplateInstantiation a vector of TokenAndName objects so it can share the same logic as the other members. * Added member for template simplifier pointer This can be used more efficiently than marking Tokens with a flag and then searching through all templates to find the one that matches. * Turned loop inside out This means we only have to iterate through the std::list once. std::list is very expensive to iterate through. * Latest code from danmar and fixed optimisations In particular I have optimised simplifying template instantiation names as this was incredibly slow because of the number of times it had to iterate through the template instantiation list. Previous optimisations to this weren't very effective and broke some edge cases. * Added changes from danmar Made mExplicitInstantiationsToDelete a vector of TokenAndName to be consistent with the rest of the members, which are cleaned up very efficiently. * Tokens can have many templateSimplifierPointers * templateSimplifierPointers must be kept in sync
2018-12-21 13:51:45 +01:00
Token * start = mExplicitInstantiationsToDelete[j].token;
if (start) {
Token * end = start->next();
while (end && end->str() != ";")
end = end->next();
if (start->previous())
start = start->previous();
if (end->next())
end = end->next();
eraseTokens(start, end);
}
}
}
}
void TemplateSimplifier::syntaxError(const Token *tok)
{
throw InternalError(tok, "syntax error", InternalError::SYNTAX);
}