cppcheck/lib/token.h

1195 lines
38 KiB
C++

/*
* Cppcheck - A tool for static C/C++ code analysis
* Copyright (C) 2007-2019 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/>.
*/
//---------------------------------------------------------------------------
#ifndef tokenH
#define tokenH
//---------------------------------------------------------------------------
#include "config.h"
#include "mathlib.h"
#include "valueflow.h"
#include "templatesimplifier.h"
#include <algorithm>
#include <cstddef>
#include <functional>
#include <list>
#include <memory>
#include <ostream>
#include <string>
#include <vector>
class Enumerator;
class Function;
class Scope;
class Settings;
class Type;
class ValueType;
class Variable;
/**
* @brief This struct stores pointers to the front and back tokens of the list this token is in.
*/
struct TokensFrontBack {
Token *front;
Token *back;
};
struct ScopeInfo2 {
ScopeInfo2(const std::string &name_, const Token *bodyEnd_, const std::set<std::string> &usingNamespaces_ = std::set<std::string>())
: name(name_),
bodyEnd(bodyEnd_),
usingNamespaces(usingNamespaces_)
{}
std::string name;
const Token * const bodyEnd;
std::set<std::string> usingNamespaces;
};
struct TokenImpl {
unsigned int mVarId;
unsigned int mFileIndex;
unsigned int mLineNumber;
unsigned int mColumn;
// AST..
Token *mAstOperand1;
Token *mAstOperand2;
Token *mAstParent;
// symbol database information
const Scope *mScope;
union {
const Function *mFunction;
const Variable *mVariable;
const ::Type* mType;
const Enumerator *mEnumerator;
};
/**
* A value from 0-100 that provides a rough idea about where in the token
* list this token is located.
*/
unsigned int mProgressValue;
/**
* Token index. Position in token list
*/
unsigned int mIndex;
// original name like size_t
std::string* mOriginalName;
// ValueType
ValueType *mValueType;
// ValueFlow
std::list<ValueFlow::Value>* mValues;
static const std::list<ValueFlow::Value> mEmptyValueList;
/** Bitfield bit count. */
unsigned char mBits;
// Pointer to a template in the template simplifier
std::set<TemplateSimplifier::TokenAndName*> mTemplateSimplifierPointers;
std::shared_ptr<ScopeInfo2> mScopeInfo;
TokenImpl()
: mVarId(0)
, mFileIndex(0)
, mLineNumber(0)
, mColumn(0)
, mAstOperand1(nullptr)
, mAstOperand2(nullptr)
, mAstParent(nullptr)
, mScope(nullptr)
, mFunction(nullptr) // Initialize whole union
, mProgressValue(0)
, mIndex(0)
, mOriginalName(nullptr)
, mValueType(nullptr)
, mValues(nullptr)
, mBits(0)
, mTemplateSimplifierPointers()
, mScopeInfo(nullptr)
{}
~TokenImpl();
};
/// @addtogroup Core
/// @{
/**
* @brief The token list that the TokenList generates is a linked-list of this class.
*
* Tokens are stored as strings. The "if", "while", etc are stored in plain text.
* The reason the Token class is needed (instead of using the string class) is that some extra functionality is also needed for tokens:
* - location of the token is stored (fileIndex, linenr, column)
* - functions for classifying the token (isName, isNumber, isBoolean, isStandardType)
*
* The Token class also has other functions for management of token list, matching tokens, etc.
*/
class CPPCHECKLIB Token {
private:
TokensFrontBack* mTokensFrontBack;
// Not implemented..
Token(const Token &);
Token operator=(const Token &);
public:
enum Type {
eVariable, eType, eFunction, eKeyword, eName, // Names: Variable (varId), Type (typeId, later), Function (FuncId, later), Language keyword, Name (unknown identifier)
eNumber, eString, eChar, eBoolean, eLiteral, eEnumerator, // Literals: Number, String, Character, Boolean, User defined literal (C++11), Enumerator
eArithmeticalOp, eComparisonOp, eAssignmentOp, eLogicalOp, eBitOp, eIncDecOp, eExtendedOp, // Operators: Arithmetical, Comparison, Assignment, Logical, Bitwise, ++/--, Extended
eBracket, // {, }, <, >: < and > only if link() is set. Otherwise they are comparison operators.
eOther,
eNone
};
explicit Token(TokensFrontBack *tokensFrontBack = nullptr);
~Token();
template<typename T>
void str(T&& s) {
mStr = s;
mImpl->mVarId = 0;
update_property_info();
}
/**
* Concatenate two (quoted) strings. Automatically cuts of the last/first character.
* Example: "hello ""world" -> "hello world". Used by the token simplifier.
*/
void concatStr(std::string const& b);
const std::string &str() const {
return mStr;
}
/**
* Unlink and delete the next 'index' tokens.
*/
void deleteNext(unsigned long index = 1);
/**
* Unlink and delete the previous 'index' tokens.
*/
void deletePrevious(unsigned long index = 1);
/**
* Swap the contents of this token with the next token.
*/
void swapWithNext();
/**
* @return token in given index, related to this token.
* For example index 1 would return next token, and 2
* would return next from that one.
*/
const Token *tokAt(int index) const;
Token *tokAt(int index) {
return const_cast<Token *>(const_cast<const Token *>(this)->tokAt(index));
}
/**
* @return the link to the token in given index, related to this token.
* For example index 1 would return the link to next token.
*/
const Token *linkAt(int index) const;
Token *linkAt(int index) {
return const_cast<Token *>(const_cast<const Token *>(this)->linkAt(index));
}
/**
* @return String of the token in given index, related to this token.
* If that token does not exist, an empty string is being returned.
*/
const std::string &strAt(int index) const;
/**
* Match given token (or list of tokens) to a pattern list.
*
* Possible patterns
* "someRandomText" If token contains "someRandomText".
* @note Use Match() if you want to use flags in patterns
*
* The patterns can be also combined to compare to multiple tokens at once
* by separating tokens with a space, e.g.
* ") void {" will return true if first token is ')' next token
* is "void" and token after that is '{'. If even one of the tokens does
* not match its pattern, false is returned.
*
* @param tok List of tokens to be compared to the pattern
* @param pattern The pattern against which the tokens are compared,
* e.g. "const" or ") void {".
* @return true if given token matches with given pattern
* false if given token does not match with given pattern
*/
static bool simpleMatch(const Token *tok, const char pattern[]);
/**
* Match given token (or list of tokens) to a pattern list.
*
* Possible patterns
* - "%any%" any token
* - "%assign%" a assignment operand
* - "%bool%" true or false
* - "%char%" Any token enclosed in &apos;-character.
* - "%comp%" Any token such that isComparisonOp() returns true.
* - "%cop%" Any token such that isConstOp() returns true.
* - "%name%" any token which is a name, variable or type e.g. "hello" or "int"
* - "%num%" Any numeric token, e.g. "23"
* - "%op%" Any token such that isOp() returns true.
* - "%or%" A bitwise-or operator '|'
* - "%oror%" A logical-or operator '||'
* - "%type%" Anything that can be a variable type, e.g. "int", but not "delete".
* - "%str%" Any token starting with &quot;-character (C-string).
* - "%var%" Match with token with varId > 0
* - "%varid%" Match with parameter varid
* - "[abc]" Any of the characters 'a' or 'b' or 'c'
* - "int|void|char" Any of the strings, int, void or char
* - "int|void|char|" Any of the strings, int, void or char or empty string
* - "!!else" No tokens or any token that is not "else".
* - "someRandomText" If token contains "someRandomText".
*
* multi-compare patterns such as "int|void|char" can contain %%or%, %%oror% and %%op%
* it is recommended to put such an %%cmd% as the first pattern.
* For example: "%var%|%num%|)" means yes to a variable, a number or ')'.
*
* The patterns can be also combined to compare to multiple tokens at once
* by separating tokens with a space, e.g.
* ") const|void {" will return true if first token is ')' next token is either
* "const" or "void" and token after that is '{'. If even one of the tokens does not
* match its pattern, false is returned.
*
* @param tok List of tokens to be compared to the pattern
* @param pattern The pattern against which the tokens are compared,
* e.g. "const" or ") const|volatile| {".
* @param varid if %%varid% is given in the pattern the Token::varId
* will be matched against this argument
* @return true if given token matches with given pattern
* false if given token does not match with given pattern
*/
static bool Match(const Token *tok, const char pattern[], unsigned int varid = 0);
/**
* @return length of C-string.
*
* Should be called for %%str%% tokens only.
*
* @param tok token with C-string
**/
static std::size_t getStrLength(const Token *tok);
/**
* @return sizeof of C-string.
*
* Should be called for %%str%% tokens only.
*
* @param tok token with C-string
**/
static std::size_t getStrSize(const Token *tok);
/**
* @return char of C-string at index (possible escaped "\\n")
*
* Should be called for %%str%% tokens only.
*
* @param tok token with C-string
* @param index position of character
**/
static std::string getCharAt(const Token *tok, std::size_t index);
const ValueType *valueType() const {
return mImpl->mValueType;
}
void setValueType(ValueType *vt);
const ValueType *argumentType() const {
const Token *top = this;
while (top && !Token::Match(top->astParent(), ",|("))
top = top->astParent();
return top ? top->mImpl->mValueType : nullptr;
}
Token::Type tokType() const {
return mTokType;
}
void tokType(Token::Type t) {
mTokType = t;
const bool memoizedIsName = (mTokType == eName || mTokType == eType || mTokType == eVariable ||
mTokType == eFunction || mTokType == eKeyword || mTokType == eBoolean ||
mTokType == eEnumerator); // TODO: "true"/"false" aren't really a name...
setFlag(fIsName, memoizedIsName);
const bool memoizedIsLiteral = (mTokType == eNumber || mTokType == eString || mTokType == eChar ||
mTokType == eBoolean || mTokType == eLiteral || mTokType == eEnumerator);
setFlag(fIsLiteral, memoizedIsLiteral);
}
void isKeyword(const bool kwd) {
if (kwd)
tokType(eKeyword);
else if (mTokType == eKeyword)
tokType(eName);
}
bool isKeyword() const {
return mTokType == eKeyword;
}
bool isName() const {
return getFlag(fIsName);
}
bool isUpperCaseName() const;
bool isLiteral() const {
return getFlag(fIsLiteral);
}
bool isNumber() const {
return mTokType == eNumber;
}
bool isEnumerator() const {
return mTokType == eEnumerator;
}
bool isOp() const {
return (isConstOp() ||
isAssignmentOp() ||
mTokType == eIncDecOp);
}
bool isConstOp() const {
return (isArithmeticalOp() ||
mTokType == eLogicalOp ||
mTokType == eComparisonOp ||
mTokType == eBitOp);
}
bool isExtendedOp() const {
return isConstOp() ||
mTokType == eExtendedOp;
}
bool isArithmeticalOp() const {
return mTokType == eArithmeticalOp;
}
bool isComparisonOp() const {
return mTokType == eComparisonOp;
}
bool isAssignmentOp() const {
return mTokType == eAssignmentOp;
}
bool isBoolean() const {
return mTokType == eBoolean;
}
bool isBinaryOp() const {
return astOperand1() != nullptr && astOperand2() != nullptr;
}
bool isUnaryOp(const std::string &s) const {
return s == mStr && astOperand1() != nullptr && astOperand2() == nullptr;
}
bool isUnaryPreOp() const;
unsigned int flags() const {
return mFlags;
}
void flags(const unsigned int flags_) {
mFlags = flags_;
}
bool isUnsigned() const {
return getFlag(fIsUnsigned);
}
void isUnsigned(const bool sign) {
setFlag(fIsUnsigned, sign);
}
bool isSigned() const {
return getFlag(fIsSigned);
}
void isSigned(const bool sign) {
setFlag(fIsSigned, sign);
}
bool isPointerCompare() const {
return getFlag(fIsPointerCompare);
}
void isPointerCompare(const bool b) {
setFlag(fIsPointerCompare, b);
}
bool isLong() const {
return getFlag(fIsLong);
}
void isLong(bool size) {
setFlag(fIsLong, size);
}
bool isStandardType() const {
return getFlag(fIsStandardType);
}
void isStandardType(const bool b) {
setFlag(fIsStandardType, b);
}
bool isExpandedMacro() const {
return getFlag(fIsExpandedMacro);
}
void isExpandedMacro(const bool m) {
setFlag(fIsExpandedMacro, m);
}
bool isCast() const {
return getFlag(fIsCast);
}
void isCast(bool c) {
setFlag(fIsCast, c);
}
bool isAttributeConstructor() const {
return getFlag(fIsAttributeConstructor);
}
void isAttributeConstructor(const bool ac) {
setFlag(fIsAttributeConstructor, ac);
}
bool isAttributeDestructor() const {
return getFlag(fIsAttributeDestructor);
}
void isAttributeDestructor(const bool value) {
setFlag(fIsAttributeDestructor, value);
}
bool isAttributeUnused() const {
return getFlag(fIsAttributeUnused);
}
void isAttributeUnused(bool unused) {
setFlag(fIsAttributeUnused, unused);
}
bool isAttributeUsed() const {
return getFlag(fIsAttributeUsed);
}
void isAttributeUsed(const bool unused) {
setFlag(fIsAttributeUsed, unused);
}
bool isAttributePure() const {
return getFlag(fIsAttributePure);
}
void isAttributePure(const bool value) {
setFlag(fIsAttributePure, value);
}
bool isAttributeConst() const {
return getFlag(fIsAttributeConst);
}
void isAttributeConst(bool value) {
setFlag(fIsAttributeConst, value);
}
bool isAttributeNoreturn() const {
return getFlag(fIsAttributeNoreturn);
}
void isAttributeNoreturn(const bool value) {
setFlag(fIsAttributeNoreturn, value);
}
bool isAttributeNothrow() const {
return getFlag(fIsAttributeNothrow);
}
void isAttributeNothrow(const bool value) {
setFlag(fIsAttributeNothrow, value);
}
bool isAttributePacked() const {
return getFlag(fIsAttributePacked);
}
void isAttributePacked(const bool value) {
setFlag(fIsAttributePacked, value);
}
bool isAttributeNodiscard() const {
return getFlag(fIsAttributeNodiscard);
}
void isAttributeNodiscard(const bool value) {
setFlag(fIsAttributeNodiscard, value);
}
bool isControlFlowKeyword() const {
return getFlag(fIsControlFlowKeyword);
}
bool isOperatorKeyword() const {
return getFlag(fIsOperatorKeyword);
}
void isOperatorKeyword(const bool value) {
setFlag(fIsOperatorKeyword, value);
}
bool isComplex() const {
return getFlag(fIsComplex);
}
void isComplex(const bool value) {
setFlag(fIsComplex, value);
}
bool isEnumType() const {
return getFlag(fIsEnumType);
}
void isEnumType(const bool value) {
setFlag(fIsEnumType, value);
}
bool isAtAddress() const {
return getFlag(fAtAddress);
}
void isAtAddress(bool b) {
setFlag(fAtAddress, b);
}
bool isBitfield() const {
return mImpl->mBits > 0;
}
unsigned char bits() const {
return mImpl->mBits;
}
std::set<TemplateSimplifier::TokenAndName*> &templateSimplifierPointers() const {
return mImpl->mTemplateSimplifierPointers;
}
void templateSimplifierPointer(TemplateSimplifier::TokenAndName* tokenAndName) {
mImpl->mTemplateSimplifierPointers.insert(tokenAndName);
}
void setBits(const unsigned char b) {
mImpl->mBits = b;
}
/**
* @brief Is current token a template argument?
*
* Original code:
*
* template<class C> struct S {
* C x;
* };
* S<int> s;
*
* Resulting code:
*
* struct S<int> {
* int x ; // <- "int" is a template argument
* }
* S<int> s;
*/
bool isTemplateArg() const {
return getFlag(fIsTemplateArg);
}
void isTemplateArg(const bool value) {
setFlag(fIsTemplateArg, value);
}
static const Token *findsimplematch(const Token * const startTok, const char pattern[]);
static const Token *findsimplematch(const Token * const startTok, const char pattern[], const Token * const end);
static const Token *findmatch(const Token * const startTok, const char pattern[], const unsigned int varId = 0U);
static const Token *findmatch(const Token * const startTok, const char pattern[], const Token * const end, const unsigned int varId = 0U);
static Token *findsimplematch(Token * const startTok, const char pattern[]) {
return const_cast<Token *>(findsimplematch(const_cast<const Token *>(startTok), pattern));
}
static Token *findsimplematch(Token * const startTok, const char pattern[], const Token * const end) {
return const_cast<Token *>(findsimplematch(const_cast<const Token *>(startTok), pattern, end));
}
static Token *findmatch(Token * const startTok, const char pattern[], const unsigned int varId = 0U) {
return const_cast<Token *>(findmatch(const_cast<const Token *>(startTok), pattern, varId));
}
static Token *findmatch(Token * const startTok, const char pattern[], const Token * const end, const unsigned int varId = 0U) {
return const_cast<Token *>(findmatch(const_cast<const Token *>(startTok), pattern, end, varId));
}
/**
* Needle is build from multiple alternatives. If one of
* them is equal to haystack, return value is 1. If there
* are no matches, but one alternative to needle is empty
* string, return value is 0. If needle was not found, return
* value is -1.
*
* @param tok Current token (needle)
* @param haystack e.g. "one|two" or "|one|two"
* @param varid optional varid of token
* @return 1 if needle is found from the haystack
* 0 if needle was empty string
* -1 if needle was not found
*/
static int multiCompare(const Token *tok, const char *haystack, unsigned int varid);
unsigned int fileIndex() const {
return mImpl->mFileIndex;
}
void fileIndex(unsigned int indexOfFile) {
mImpl->mFileIndex = indexOfFile;
}
unsigned int linenr() const {
return mImpl->mLineNumber;
}
void linenr(unsigned int lineNumber) {
mImpl->mLineNumber = lineNumber;
}
unsigned int col() const {
return mImpl->mColumn;
}
void col(unsigned int c) {
mImpl->mColumn = c;
}
Token *next() const {
return mNext;
}
/**
* Delete tokens between begin and end. E.g. if begin = 1
* and end = 5, tokens 2,3 and 4 would be erased.
*
* @param begin Tokens after this will be erased.
* @param end Tokens before this will be erased.
*/
static void eraseTokens(Token *begin, const Token *end);
/**
* Insert new token after this token. This function will handle
* relations between next and previous token also.
* @param tokenStr String for the new token.
* @param originalNameStr String used for Token::originalName().
* @param prepend Insert the new token before this token when it's not
* the first one on the tokens list.
*/
void insertToken(const std::string &tokenStr, const std::string &originalNameStr=emptyString, bool prepend=false);
Token *previous() const {
return mPrevious;
}
unsigned int varId() const {
return mImpl->mVarId;
}
void varId(unsigned int id) {
mImpl->mVarId = id;
if (id != 0) {
tokType(eVariable);
isStandardType(false);
} else {
update_property_info();
}
}
/**
* For debugging purposes, prints token and all tokens
* followed by it.
* @param title Title for the printout or use default parameter or 0
* for no title.
*/
void printOut(const char *title = nullptr) const;
/**
* For debugging purposes, prints token and all tokens
* followed by it.
* @param title Title for the printout or use default parameter or 0
* for no title.
* @param fileNames Prints out file name instead of file index.
* File index should match the index of the string in this vector.
*/
void printOut(const char *title, const std::vector<std::string> &fileNames) const;
/**
* Replace token replaceThis with tokens between start and end,
* including start and end. The replaceThis token is deleted.
* @param replaceThis This token will be deleted.
* @param start This will be in the place of replaceThis
* @param end This is also in the place of replaceThis
*/
static void replace(Token *replaceThis, Token *start, Token *end);
/**
* Stringify a token
* @param os The result is shifted into that output stream
* @param varid Print varids. (Style: "varname@id")
* @param attributes Print attributes of tokens like "unsigned" in front of it.
* @param macro Prints $ in front of the token if it was expanded from a macro.
*/
void stringify(std::ostream& os, bool varid, bool attributes, bool macro) const;
/**
* Stringify a list of token, from current instance on.
* @param varid Print varids. (Style: "varname@id")
* @param attributes Print attributes of tokens like "unsigned" in front of it.
* @param linenumbers Print line number in front of each line
* @param linebreaks Insert "\\n" into string when line number changes
* @param files print Files as numbers or as names (if fileNames is given)
* @param fileNames Vector of filenames. Used (if given) to print filenames as strings instead of numbers.
* @param end Stringification ends before this token is reached. 0 to stringify until end of list.
* @return Stringified token list as a string
*/
std::string stringifyList(bool varid, bool attributes, bool linenumbers, bool linebreaks, bool files, const std::vector<std::string>* fileNames = nullptr, const Token* end = nullptr) const;
std::string stringifyList(const Token* end, bool attributes = true) const;
std::string stringifyList(bool varid = false) const;
/**
* Remove the contents for this token from the token list.
*
* The contents are replaced with the contents of the next token and
* the next token is unlinked and deleted from the token list.
*
* So this token will still be valid after the 'deleteThis()'.
*/
void deleteThis();
/**
* Create link to given token
* @param linkToToken The token where this token should link
* to.
*/
void link(Token *linkToToken) {
mLink = linkToToken;
if (mStr == "<" || mStr == ">")
update_property_info();
}
/**
* Return token where this token links to.
* Supported links are:
* "{" <-> "}"
* "(" <-> ")"
* "[" <-> "]"
*
* @return The token where this token links to.
*/
Token *link() const {
return mLink;
}
/**
* Associate this token with given scope
* @param s Scope to be associated
*/
void scope(const Scope *s) {
mImpl->mScope = s;
}
/**
* @return a pointer to the scope containing this token.
*/
const Scope *scope() const {
return mImpl->mScope;
}
/**
* Associate this token with given function
* @param f Function to be associated
*/
void function(const Function *f) {
mImpl->mFunction = f;
if (f)
tokType(eFunction);
else if (mTokType == eFunction)
tokType(eName);
}
/**
* @return a pointer to the Function associated with this token.
*/
const Function *function() const {
return mTokType == eFunction ? mImpl->mFunction : nullptr;
}
/**
* Associate this token with given variable
* @param v Variable to be associated
*/
void variable(const Variable *v) {
mImpl->mVariable = v;
if (v || mImpl->mVarId)
tokType(eVariable);
else if (mTokType == eVariable)
tokType(eName);
}
/**
* @return a pointer to the variable associated with this token.
*/
const Variable *variable() const {
return mTokType == eVariable ? mImpl->mVariable : nullptr;
}
/**
* Associate this token with given type
* @param t Type to be associated
*/
void type(const ::Type *t);
/**
* @return a pointer to the type associated with this token.
*/
const ::Type *type() const {
return mTokType == eType ? mImpl->mType : nullptr;
}
static const ::Type *typeOf(const Token *tok);
static std::pair<const Token*, const Token*> typeDecl(const Token * tok);
static std::string typeStr(const Token* tok);
/**
* @return a pointer to the Enumerator associated with this token.
*/
const Enumerator *enumerator() const {
return mTokType == eEnumerator ? mImpl->mEnumerator : nullptr;
}
/**
* Associate this token with given enumerator
* @param e Enumerator to be associated
*/
void enumerator(const Enumerator *e) {
mImpl->mEnumerator = e;
if (e)
tokType(eEnumerator);
else if (mTokType == eEnumerator)
tokType(eName);
}
/**
* Links two elements against each other.
**/
static void createMutualLinks(Token *begin, Token *end);
/**
* This can be called only for tokens that are strings, else
* the assert() is called. If Token is e.g. '"hello"', this will return
* 'hello' (removing the double quotes).
* @return String value
*/
std::string strValue() const;
/**
* Move srcStart and srcEnd tokens and all tokens between them
* into new a location. Only links between tokens are changed.
* @param srcStart This is the first token to be moved
* @param srcEnd The last token to be moved
* @param newLocation srcStart will be placed after this token.
*/
static void move(Token *srcStart, Token *srcEnd, Token *newLocation);
/** Get progressValue (0 - 100) */
unsigned int progressValue() const {
return mImpl->mProgressValue;
}
/** Calculate progress values for all tokens */
static void assignProgressValues(Token *tok);
/**
* @return the first token of the next argument. Does only work on argument
* lists. Requires that Tokenizer::createLinks2() has been called before.
* Returns 0, if there is no next argument.
*/
Token* nextArgument() const;
/**
* @return the first token of the next argument. Does only work on argument
* lists. Should be used only before Tokenizer::createLinks2() was called.
* Returns 0, if there is no next argument.
*/
Token* nextArgumentBeforeCreateLinks2() const;
/**
* @return the first token of the next template argument. Does only work on template argument
* lists. Requires that Tokenizer::createLinks2() has been called before.
* Returns 0, if there is no next argument.
*/
Token* nextTemplateArgument() const;
/**
* Returns the closing bracket of opening '<'. Should only be used if link()
* is unavailable.
* @return closing '>', ')', ']' or '}'. if no closing bracket is found, NULL is returned
*/
const Token* findClosingBracket() const;
Token* findClosingBracket();
const Token* findOpeningBracket() const;
Token* findOpeningBracket();
/**
* @return the original name.
*/
const std::string & originalName() const {
return mImpl->mOriginalName ? *mImpl->mOriginalName : emptyString;
}
const std::list<ValueFlow::Value>& values() const {
return mImpl->mValues ? *mImpl->mValues : mImpl->mEmptyValueList;
}
/**
* Sets the original name.
*/
template<typename T>
void originalName(T&& name) {
if (!mImpl->mOriginalName)
mImpl->mOriginalName = new std::string(name);
else
*mImpl->mOriginalName = name;
}
bool hasKnownIntValue() const {
return hasKnownValue() && std::any_of(mImpl->mValues->begin(), mImpl->mValues->end(), std::mem_fn(&ValueFlow::Value::isIntValue));
}
bool hasKnownValue() const {
return mImpl->mValues && std::any_of(mImpl->mValues->begin(), mImpl->mValues->end(), std::mem_fn(&ValueFlow::Value::isKnown));
}
MathLib::bigint getKnownIntValue() const {
return mImpl->mValues->front().intvalue;
}
const ValueFlow::Value * getValue(const MathLib::bigint val) const {
if (!mImpl->mValues)
return nullptr;
const auto it = std::find_if(mImpl->mValues->begin(), mImpl->mValues->end(), [=](const ValueFlow::Value &value) {
return value.isIntValue() && value.intvalue == val;
});
return it == mImpl->mValues->end() ? nullptr : &*it;;
}
const ValueFlow::Value * getMaxValue(bool condition) const {
if (!mImpl->mValues)
return nullptr;
const ValueFlow::Value *ret = nullptr;
for (const ValueFlow::Value &value : *mImpl->mValues) {
if (!value.isIntValue())
continue;
if ((!ret || value.intvalue > ret->intvalue) &&
((value.condition != nullptr) == condition))
ret = &value;
}
return ret;
}
const ValueFlow::Value * getMovedValue() const {
if (!mImpl->mValues)
return nullptr;
const auto it = std::find_if(mImpl->mValues->begin(), mImpl->mValues->end(), [](const ValueFlow::Value &value) {
return value.isMovedValue() && value.moveKind != ValueFlow::Value::NonMovedVariable;
});
return it == mImpl->mValues->end() ? nullptr : &*it;;
}
const ValueFlow::Value * getValueLE(const MathLib::bigint val, const Settings *settings) const;
const ValueFlow::Value * getValueGE(const MathLib::bigint val, const Settings *settings) const;
const ValueFlow::Value * getInvalidValue(const Token *ftok, unsigned int argnr, const Settings *settings) const;
const ValueFlow::Value * getContainerSizeValue(const MathLib::bigint val) const {
if (!mImpl->mValues)
return nullptr;
const auto it = std::find_if(mImpl->mValues->begin(), mImpl->mValues->end(), [=](const ValueFlow::Value &value) {
return value.isContainerSizeValue() && value.intvalue == val;
});
return it == mImpl->mValues->end() ? nullptr : &*it;;
}
const Token *getValueTokenMaxStrLength() const;
const Token *getValueTokenMinStrSize() const;
const Token *getValueTokenDeadPointer() const;
/** Add token value. Return true if value is added. */
bool addValue(const ValueFlow::Value &value);
template<class Predicate>
void removeValues(Predicate pred) {
if (mImpl->mValues)
mImpl->mValues->remove_if(pred);
}
unsigned int index() const {
return mImpl->mIndex;
}
void assignIndexes();
private:
void next(Token *nextToken) {
mNext = nextToken;
}
void previous(Token *previousToken) {
mPrevious = previousToken;
}
/** used by deleteThis() to take data from token to delete */
void takeData(Token *fromToken);
/**
* Works almost like strcmp() except returns only true or false and
* if str has empty space &apos; &apos; character, that character is handled
* as if it were &apos;\\0&apos;
*/
static bool firstWordEquals(const char *str, const char *word);
/**
* Works almost like strchr() except
* if str has empty space &apos; &apos; character, that character is handled
* as if it were &apos;\\0&apos;
*/
static const char *chrInFirstWord(const char *str, char c);
std::string mStr;
Token *mNext;
Token *mPrevious;
Token *mLink;
enum {
fIsUnsigned = (1 << 0),
fIsSigned = (1 << 1),
fIsPointerCompare = (1 << 2),
fIsLong = (1 << 3),
fIsStandardType = (1 << 4),
fIsExpandedMacro = (1 << 5),
fIsCast = (1 << 6),
fIsAttributeConstructor = (1 << 7), // __attribute__((constructor)) __attribute__((constructor(priority)))
fIsAttributeDestructor = (1 << 8), // __attribute__((destructor)) __attribute__((destructor(priority)))
fIsAttributeUnused = (1 << 9), // __attribute__((unused))
fIsAttributePure = (1 << 10), // __attribute__((pure))
fIsAttributeConst = (1 << 11), // __attribute__((const))
fIsAttributeNoreturn = (1 << 12), // __attribute__((noreturn)), __declspec(noreturn)
fIsAttributeNothrow = (1 << 13), // __attribute__((nothrow)), __declspec(nothrow)
fIsAttributeUsed = (1 << 14), // __attribute__((used))
fIsAttributePacked = (1 << 15), // __attribute__((packed))
fIsControlFlowKeyword = (1 << 16), // if/switch/while/...
fIsOperatorKeyword = (1 << 17), // operator=, etc
fIsComplex = (1 << 18), // complex/_Complex type
fIsEnumType = (1 << 19), // enumeration type
fIsName = (1 << 20),
fIsLiteral = (1 << 21),
fIsTemplateArg = (1 << 22),
fIsAttributeNodiscard = (1 << 23), // __attribute__ ((warn_unused_result)), [[nodiscard]]
fAtAddress = (1 << 24), // @ 0x4000
};
Token::Type mTokType;
unsigned int mFlags;
TokenImpl *mImpl;
/**
* Get specified flag state.
* @param flag_ flag to get state of
* @return true if flag set or false in flag not set
*/
bool getFlag(unsigned int flag_) const {
return ((mFlags & flag_) != 0);
}
/**
* Set specified flag state.
* @param flag_ flag to set state
* @param state_ new state of flag
*/
void setFlag(unsigned int flag_, bool state_) {
mFlags = state_ ? mFlags | flag_ : mFlags & ~flag_;
}
/** Updates internal property cache like _isName or _isBoolean.
Called after any mStr() modification. */
void update_property_info();
/** Update internal property cache about isStandardType() */
void update_property_isStandardType();
/** Update internal property cache about string and char literals */
void update_property_char_string_literal();
/** Internal helper function to avoid excessive string allocations */
void astStringVerboseRecursive(std::string& ret, const unsigned int indent1 = 0U, const unsigned int indent2 = 0U) const;
public:
void astOperand1(Token *tok);
void astOperand2(Token *tok);
Token * astOperand1() {
return mImpl->mAstOperand1;
}
const Token * astOperand1() const {
return mImpl->mAstOperand1;
}
Token * astOperand2() {
return mImpl->mAstOperand2;
}
const Token * astOperand2() const {
return mImpl->mAstOperand2;
}
Token * astParent() {
return mImpl->mAstParent;
}
const Token * astParent() const {
return mImpl->mAstParent;
}
Token *astTop() {
Token *ret = this;
while (ret->mImpl->mAstParent)
ret = ret->mImpl->mAstParent;
return ret;
}
const Token *astTop() const {
const Token *ret = this;
while (ret->mImpl->mAstParent)
ret = ret->mImpl->mAstParent;
return ret;
}
std::pair<const Token *, const Token *> findExpressionStartEndTokens() const;
/**
* Is current token a calculation? Only true for operands.
* For '*' and '&' tokens it is looked up if this is a
* dereference or address-of. A dereference or address-of is not
* counted as a calculation.
* @return returns true if current token is a calculation
*/
bool isCalculation() const;
void clearAst() {
mImpl->mAstOperand1 = mImpl->mAstOperand2 = mImpl->mAstParent = nullptr;
}
void clearValueFlow() {
delete mImpl->mValues;
mImpl->mValues = nullptr;
}
std::string astString(const char *sep = "") const {
std::string ret;
if (mImpl->mAstOperand1)
ret = mImpl->mAstOperand1->astString(sep);
if (mImpl->mAstOperand2)
ret += mImpl->mAstOperand2->astString(sep);
return ret + sep + mStr;
}
std::string astStringVerbose() const;
std::string expressionString() const;
void printAst(bool verbose, bool xml, std::ostream &out) const;
void printValueFlow(bool xml, std::ostream &out) const;
void scopeInfo(std::shared_ptr<ScopeInfo2> newScopeInfo);
std::shared_ptr<ScopeInfo2> scopeInfo() const;
};
/// @}
//---------------------------------------------------------------------------
#endif // tokenH