bgfx/3rdparty/scintilla/lexers/LexCPP.cxx

1627 lines
52 KiB
C++
Raw Normal View History

2015-06-08 20:07:35 -04:00
// Scintilla source code edit control
/** @file LexCPP.cxx
** Lexer for C++, C, Java, and JavaScript.
** Further folding features and configuration properties added by "Udo Lechner" <dlchnr(at)gmx(dot)net>
**/
// Copyright 1998-2005 by Neil Hodgson <neilh@scintilla.org>
// The License.txt file describes the conditions under which this software may be distributed.
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#include <stdarg.h>
#include <assert.h>
#include <ctype.h>
#include <string>
#include <vector>
#include <map>
#include <algorithm>
#include "ILexer.h"
#include "Scintilla.h"
#include "SciLexer.h"
#include "WordList.h"
#include "LexAccessor.h"
#include "Accessor.h"
#include "StyleContext.h"
#include "CharacterSet.h"
#include "LexerModule.h"
#include "OptionSet.h"
#include "SparseState.h"
#include "SubStyles.h"
#ifdef SCI_NAMESPACE
using namespace Scintilla;
#endif
namespace {
// Use an unnamed namespace to protect the functions and classes from name conflicts
bool IsSpaceEquiv(int state) {
return (state <= SCE_C_COMMENTDOC) ||
// including SCE_C_DEFAULT, SCE_C_COMMENT, SCE_C_COMMENTLINE
(state == SCE_C_COMMENTLINEDOC) || (state == SCE_C_COMMENTDOCKEYWORD) ||
(state == SCE_C_COMMENTDOCKEYWORDERROR);
}
// Preconditions: sc.currentPos points to a character after '+' or '-'.
// The test for pos reaching 0 should be redundant,
// and is in only for safety measures.
// Limitation: this code will give the incorrect answer for code like
// a = b+++/ptn/...
// Putting a space between the '++' post-inc operator and the '+' binary op
// fixes this, and is highly recommended for readability anyway.
bool FollowsPostfixOperator(StyleContext &sc, LexAccessor &styler) {
int pos = (int) sc.currentPos;
while (--pos > 0) {
char ch = styler[pos];
if (ch == '+' || ch == '-') {
return styler[pos - 1] == ch;
}
}
return false;
}
bool followsReturnKeyword(StyleContext &sc, LexAccessor &styler) {
// Don't look at styles, so no need to flush.
int pos = (int) sc.currentPos;
int currentLine = styler.GetLine(pos);
int lineStartPos = styler.LineStart(currentLine);
while (--pos > lineStartPos) {
char ch = styler.SafeGetCharAt(pos);
if (ch != ' ' && ch != '\t') {
break;
}
}
const char *retBack = "nruter";
const char *s = retBack;
while (*s
&& pos >= lineStartPos
&& styler.SafeGetCharAt(pos) == *s) {
s++;
pos--;
}
return !*s;
}
bool IsSpaceOrTab(int ch) {
return ch == ' ' || ch == '\t';
}
bool OnlySpaceOrTab(const std::string &s) {
for (std::string::const_iterator it = s.begin(); it != s.end(); ++it) {
if (!IsSpaceOrTab(*it))
return false;
}
return true;
}
std::vector<std::string> StringSplit(const std::string &text, int separator) {
std::vector<std::string> vs(text.empty() ? 0 : 1);
for (std::string::const_iterator it = text.begin(); it != text.end(); ++it) {
if (*it == separator) {
vs.push_back(std::string());
} else {
vs.back() += *it;
}
}
return vs;
}
struct BracketPair {
std::vector<std::string>::iterator itBracket;
std::vector<std::string>::iterator itEndBracket;
};
BracketPair FindBracketPair(std::vector<std::string> &tokens) {
BracketPair bp;
std::vector<std::string>::iterator itTok = std::find(tokens.begin(), tokens.end(), "(");
bp.itBracket = tokens.end();
bp.itEndBracket = tokens.end();
if (itTok != tokens.end()) {
bp.itBracket = itTok;
size_t nest = 0;
while (itTok != tokens.end()) {
if (*itTok == "(") {
nest++;
} else if (*itTok == ")") {
nest--;
if (nest == 0) {
bp.itEndBracket = itTok;
return bp;
}
}
++itTok;
}
}
bp.itBracket = tokens.end();
return bp;
}
void highlightTaskMarker(StyleContext &sc, LexAccessor &styler,
int activity, WordList &markerList, bool caseSensitive){
if ((isoperator(sc.chPrev) || IsASpace(sc.chPrev)) && markerList.Length()) {
const int lengthMarker = 50;
char marker[lengthMarker+1];
int currPos = (int) sc.currentPos;
int i = 0;
while (i < lengthMarker) {
char ch = styler.SafeGetCharAt(currPos + i);
if (IsASpace(ch) || isoperator(ch)) {
break;
}
if (caseSensitive)
marker[i] = ch;
else
marker[i] = static_cast<char>(tolower(ch));
i++;
}
marker[i] = '\0';
if (markerList.InList(marker)) {
sc.SetState(SCE_C_TASKMARKER|activity);
}
}
}
struct EscapeSequence {
int digitsLeft;
CharacterSet setHexDigits;
CharacterSet setOctDigits;
CharacterSet setNoneNumeric;
CharacterSet *escapeSetValid;
EscapeSequence() {
digitsLeft = 0;
escapeSetValid = 0;
setHexDigits = CharacterSet(CharacterSet::setDigits, "ABCDEFabcdef");
setOctDigits = CharacterSet(CharacterSet::setNone, "01234567");
}
void resetEscapeState(int nextChar) {
digitsLeft = 0;
escapeSetValid = &setNoneNumeric;
if (nextChar == 'U') {
digitsLeft = 9;
escapeSetValid = &setHexDigits;
} else if (nextChar == 'u') {
digitsLeft = 5;
escapeSetValid = &setHexDigits;
} else if (nextChar == 'x') {
digitsLeft = 5;
escapeSetValid = &setHexDigits;
} else if (setOctDigits.Contains(nextChar)) {
digitsLeft = 3;
escapeSetValid = &setOctDigits;
}
}
bool atEscapeEnd(int currChar) const {
return (digitsLeft <= 0) || !escapeSetValid->Contains(currChar);
}
};
std::string GetRestOfLine(LexAccessor &styler, int start, bool allowSpace) {
std::string restOfLine;
int i =0;
char ch = styler.SafeGetCharAt(start, '\n');
int endLine = styler.LineEnd(styler.GetLine(start));
while (((start+i) < endLine) && (ch != '\r')) {
char chNext = styler.SafeGetCharAt(start + i + 1, '\n');
if (ch == '/' && (chNext == '/' || chNext == '*'))
break;
if (allowSpace || (ch != ' '))
restOfLine += ch;
i++;
ch = chNext;
}
return restOfLine;
}
bool IsStreamCommentStyle(int style) {
return style == SCE_C_COMMENT ||
style == SCE_C_COMMENTDOC ||
style == SCE_C_COMMENTDOCKEYWORD ||
style == SCE_C_COMMENTDOCKEYWORDERROR;
}
struct PPDefinition {
int line;
std::string key;
std::string value;
bool isUndef;
std::string arguments;
PPDefinition(int line_, const std::string &key_, const std::string &value_, bool isUndef_ = false, std::string arguments_="") :
line(line_), key(key_), value(value_), isUndef(isUndef_), arguments(arguments_) {
}
};
class LinePPState {
int state;
int ifTaken;
int level;
bool ValidLevel() const {
return level >= 0 && level < 32;
}
int maskLevel() const {
return 1 << level;
}
public:
LinePPState() : state(0), ifTaken(0), level(-1) {
}
bool IsInactive() const {
return state != 0;
}
bool CurrentIfTaken() const {
return (ifTaken & maskLevel()) != 0;
}
void StartSection(bool on) {
level++;
if (ValidLevel()) {
if (on) {
state &= ~maskLevel();
ifTaken |= maskLevel();
} else {
state |= maskLevel();
ifTaken &= ~maskLevel();
}
}
}
void EndSection() {
if (ValidLevel()) {
state &= ~maskLevel();
ifTaken &= ~maskLevel();
}
level--;
}
void InvertCurrentLevel() {
if (ValidLevel()) {
state ^= maskLevel();
ifTaken |= maskLevel();
}
}
};
// Hold the preprocessor state for each line seen.
// Currently one entry per line but could become sparse with just one entry per preprocessor line.
class PPStates {
std::vector<LinePPState> vlls;
public:
LinePPState ForLine(int line) const {
if ((line > 0) && (vlls.size() > static_cast<size_t>(line))) {
return vlls[line];
} else {
return LinePPState();
}
}
void Add(int line, LinePPState lls) {
vlls.resize(line+1);
vlls[line] = lls;
}
};
// An individual named option for use in an OptionSet
// Options used for LexerCPP
struct OptionsCPP {
bool stylingWithinPreprocessor;
bool identifiersAllowDollars;
bool trackPreprocessor;
bool updatePreprocessor;
bool verbatimStringsAllowEscapes;
bool triplequotedStrings;
bool hashquotedStrings;
bool backQuotedStrings;
bool escapeSequence;
bool fold;
bool foldSyntaxBased;
bool foldComment;
bool foldCommentMultiline;
bool foldCommentExplicit;
std::string foldExplicitStart;
std::string foldExplicitEnd;
bool foldExplicitAnywhere;
bool foldPreprocessor;
bool foldCompact;
bool foldAtElse;
OptionsCPP() {
stylingWithinPreprocessor = false;
identifiersAllowDollars = true;
trackPreprocessor = true;
updatePreprocessor = true;
verbatimStringsAllowEscapes = false;
triplequotedStrings = false;
hashquotedStrings = false;
backQuotedStrings = false;
escapeSequence = false;
fold = false;
foldSyntaxBased = true;
foldComment = false;
foldCommentMultiline = true;
foldCommentExplicit = true;
foldExplicitStart = "";
foldExplicitEnd = "";
foldExplicitAnywhere = false;
foldPreprocessor = false;
foldCompact = false;
foldAtElse = false;
}
};
const char *const cppWordLists[] = {
"Primary keywords and identifiers",
"Secondary keywords and identifiers",
"Documentation comment keywords",
"Global classes and typedefs",
"Preprocessor definitions",
"Task marker and error marker keywords",
0,
};
struct OptionSetCPP : public OptionSet<OptionsCPP> {
OptionSetCPP() {
DefineProperty("styling.within.preprocessor", &OptionsCPP::stylingWithinPreprocessor,
"For C++ code, determines whether all preprocessor code is styled in the "
"preprocessor style (0, the default) or only from the initial # to the end "
"of the command word(1).");
DefineProperty("lexer.cpp.allow.dollars", &OptionsCPP::identifiersAllowDollars,
"Set to 0 to disallow the '$' character in identifiers with the cpp lexer.");
DefineProperty("lexer.cpp.track.preprocessor", &OptionsCPP::trackPreprocessor,
"Set to 1 to interpret #if/#else/#endif to grey out code that is not active.");
DefineProperty("lexer.cpp.update.preprocessor", &OptionsCPP::updatePreprocessor,
"Set to 1 to update preprocessor definitions when #define found.");
DefineProperty("lexer.cpp.verbatim.strings.allow.escapes", &OptionsCPP::verbatimStringsAllowEscapes,
"Set to 1 to allow verbatim strings to contain escape sequences.");
DefineProperty("lexer.cpp.triplequoted.strings", &OptionsCPP::triplequotedStrings,
"Set to 1 to enable highlighting of triple-quoted strings.");
DefineProperty("lexer.cpp.hashquoted.strings", &OptionsCPP::hashquotedStrings,
"Set to 1 to enable highlighting of hash-quoted strings.");
DefineProperty("lexer.cpp.backquoted.strings", &OptionsCPP::backQuotedStrings,
"Set to 1 to enable highlighting of back-quoted raw strings .");
DefineProperty("lexer.cpp.escape.sequence", &OptionsCPP::escapeSequence,
"Set to 1 to enable highlighting of escape sequences in strings");
DefineProperty("fold", &OptionsCPP::fold);
DefineProperty("fold.cpp.syntax.based", &OptionsCPP::foldSyntaxBased,
"Set this property to 0 to disable syntax based folding.");
DefineProperty("fold.comment", &OptionsCPP::foldComment,
"This option enables folding multi-line comments and explicit fold points when using the C++ lexer. "
"Explicit fold points allows adding extra folding by placing a //{ comment at the start and a //} "
"at the end of a section that should fold.");
DefineProperty("fold.cpp.comment.multiline", &OptionsCPP::foldCommentMultiline,
"Set this property to 0 to disable folding multi-line comments when fold.comment=1.");
DefineProperty("fold.cpp.comment.explicit", &OptionsCPP::foldCommentExplicit,
"Set this property to 0 to disable folding explicit fold points when fold.comment=1.");
DefineProperty("fold.cpp.explicit.start", &OptionsCPP::foldExplicitStart,
"The string to use for explicit fold start points, replacing the standard //{.");
DefineProperty("fold.cpp.explicit.end", &OptionsCPP::foldExplicitEnd,
"The string to use for explicit fold end points, replacing the standard //}.");
DefineProperty("fold.cpp.explicit.anywhere", &OptionsCPP::foldExplicitAnywhere,
"Set this property to 1 to enable explicit fold points anywhere, not just in line comments.");
DefineProperty("fold.preprocessor", &OptionsCPP::foldPreprocessor,
"This option enables folding preprocessor directives when using the C++ lexer. "
"Includes C#'s explicit #region and #endregion folding directives.");
DefineProperty("fold.compact", &OptionsCPP::foldCompact);
DefineProperty("fold.at.else", &OptionsCPP::foldAtElse,
"This option enables C++ folding on a \"} else {\" line of an if statement.");
DefineWordListSets(cppWordLists);
}
};
const char styleSubable[] = {SCE_C_IDENTIFIER, SCE_C_COMMENTDOCKEYWORD, 0};
}
class LexerCPP : public ILexerWithSubStyles {
bool caseSensitive;
CharacterSet setWord;
CharacterSet setNegationOp;
CharacterSet setArithmethicOp;
CharacterSet setRelOp;
CharacterSet setLogicalOp;
CharacterSet setWordStart;
PPStates vlls;
std::vector<PPDefinition> ppDefineHistory;
WordList keywords;
WordList keywords2;
WordList keywords3;
WordList keywords4;
WordList ppDefinitions;
WordList markerList;
struct SymbolValue {
std::string value;
std::string arguments;
SymbolValue(const std::string &value_="", const std::string &arguments_="") : value(value_), arguments(arguments_) {
}
SymbolValue &operator = (const std::string &value_) {
value = value_;
arguments.clear();
return *this;
}
bool IsMacro() const {
return !arguments.empty();
}
};
typedef std::map<std::string, SymbolValue> SymbolTable;
SymbolTable preprocessorDefinitionsStart;
OptionsCPP options;
OptionSetCPP osCPP;
EscapeSequence escapeSeq;
SparseState<std::string> rawStringTerminators;
enum { activeFlag = 0x40 };
enum { ssIdentifier, ssDocKeyword };
SubStyles subStyles;
public:
explicit LexerCPP(bool caseSensitive_) :
caseSensitive(caseSensitive_),
setWord(CharacterSet::setAlphaNum, "._", 0x80, true),
setNegationOp(CharacterSet::setNone, "!"),
setArithmethicOp(CharacterSet::setNone, "+-/*%"),
setRelOp(CharacterSet::setNone, "=!<>"),
setLogicalOp(CharacterSet::setNone, "|&"),
subStyles(styleSubable, 0x80, 0x40, activeFlag) {
}
virtual ~LexerCPP() {
}
void SCI_METHOD Release() {
delete this;
}
int SCI_METHOD Version() const {
return lvSubStyles;
}
const char * SCI_METHOD PropertyNames() {
return osCPP.PropertyNames();
}
int SCI_METHOD PropertyType(const char *name) {
return osCPP.PropertyType(name);
}
const char * SCI_METHOD DescribeProperty(const char *name) {
return osCPP.DescribeProperty(name);
}
int SCI_METHOD PropertySet(const char *key, const char *val);
const char * SCI_METHOD DescribeWordListSets() {
return osCPP.DescribeWordListSets();
}
int SCI_METHOD WordListSet(int n, const char *wl);
void SCI_METHOD Lex(unsigned int startPos, int length, int initStyle, IDocument *pAccess);
void SCI_METHOD Fold(unsigned int startPos, int length, int initStyle, IDocument *pAccess);
void * SCI_METHOD PrivateCall(int, void *) {
return 0;
}
int SCI_METHOD LineEndTypesSupported() {
return SC_LINE_END_TYPE_UNICODE;
}
int SCI_METHOD AllocateSubStyles(int styleBase, int numberStyles) {
return subStyles.Allocate(styleBase, numberStyles);
}
int SCI_METHOD SubStylesStart(int styleBase) {
return subStyles.Start(styleBase);
}
int SCI_METHOD SubStylesLength(int styleBase) {
return subStyles.Length(styleBase);
}
int SCI_METHOD StyleFromSubStyle(int subStyle) {
int styleBase = subStyles.BaseStyle(MaskActive(subStyle));
int active = subStyle & activeFlag;
return styleBase | active;
}
int SCI_METHOD PrimaryStyleFromStyle(int style) {
return MaskActive(style);
}
void SCI_METHOD FreeSubStyles() {
subStyles.Free();
}
void SCI_METHOD SetIdentifiers(int style, const char *identifiers) {
subStyles.SetIdentifiers(style, identifiers);
}
int SCI_METHOD DistanceToSecondaryStyles() {
return activeFlag;
}
const char * SCI_METHOD GetSubStyleBases() {
return styleSubable;
}
static ILexer *LexerFactoryCPP() {
return new LexerCPP(true);
}
static ILexer *LexerFactoryCPPInsensitive() {
return new LexerCPP(false);
}
static int MaskActive(int style) {
return style & ~activeFlag;
}
void EvaluateTokens(std::vector<std::string> &tokens, const SymbolTable &preprocessorDefinitions);
std::vector<std::string> Tokenize(const std::string &expr) const;
bool EvaluateExpression(const std::string &expr, const SymbolTable &preprocessorDefinitions);
};
int SCI_METHOD LexerCPP::PropertySet(const char *key, const char *val) {
if (osCPP.PropertySet(&options, key, val)) {
if (strcmp(key, "lexer.cpp.allow.dollars") == 0) {
setWord = CharacterSet(CharacterSet::setAlphaNum, "._", 0x80, true);
if (options.identifiersAllowDollars) {
setWord.Add('$');
}
}
return 0;
}
return -1;
}
int SCI_METHOD LexerCPP::WordListSet(int n, const char *wl) {
WordList *wordListN = 0;
switch (n) {
case 0:
wordListN = &keywords;
break;
case 1:
wordListN = &keywords2;
break;
case 2:
wordListN = &keywords3;
break;
case 3:
wordListN = &keywords4;
break;
case 4:
wordListN = &ppDefinitions;
break;
case 5:
wordListN = &markerList;
break;
}
int firstModification = -1;
if (wordListN) {
WordList wlNew;
wlNew.Set(wl);
if (*wordListN != wlNew) {
wordListN->Set(wl);
firstModification = 0;
if (n == 4) {
// Rebuild preprocessorDefinitions
preprocessorDefinitionsStart.clear();
for (int nDefinition = 0; nDefinition < ppDefinitions.Length(); nDefinition++) {
const char *cpDefinition = ppDefinitions.WordAt(nDefinition);
const char *cpEquals = strchr(cpDefinition, '=');
if (cpEquals) {
std::string name(cpDefinition, cpEquals - cpDefinition);
std::string val(cpEquals+1);
size_t bracket = name.find('(');
size_t bracketEnd = name.find(')');
if ((bracket != std::string::npos) && (bracketEnd != std::string::npos)) {
// Macro
std::string args = name.substr(bracket + 1, bracketEnd - bracket - 1);
name = name.substr(0, bracket);
preprocessorDefinitionsStart[name] = SymbolValue(val, args);
} else {
preprocessorDefinitionsStart[name] = val;
}
} else {
std::string name(cpDefinition);
std::string val("1");
preprocessorDefinitionsStart[name] = val;
}
}
}
}
}
return firstModification;
}
// Functor used to truncate history
struct After {
int line;
explicit After(int line_) : line(line_) {}
bool operator()(PPDefinition &p) const {
return p.line > line;
}
};
void SCI_METHOD LexerCPP::Lex(unsigned int startPos, int length, int initStyle, IDocument *pAccess) {
LexAccessor styler(pAccess);
CharacterSet setOKBeforeRE(CharacterSet::setNone, "([{=,:;!%^&*|?~+-");
CharacterSet setCouldBePostOp(CharacterSet::setNone, "+-");
CharacterSet setDoxygen(CharacterSet::setAlpha, "$@\\&<>#{}[]");
setWordStart = CharacterSet(CharacterSet::setAlpha, "_", 0x80, true);
CharacterSet setInvalidRawFirst(CharacterSet::setNone, " )\\\t\v\f\n");
if (options.identifiersAllowDollars) {
setWordStart.Add('$');
}
int chPrevNonWhite = ' ';
int visibleChars = 0;
bool lastWordWasUUID = false;
int styleBeforeDCKeyword = SCE_C_DEFAULT;
int styleBeforeTaskMarker = SCE_C_DEFAULT;
bool continuationLine = false;
bool isIncludePreprocessor = false;
bool isStringInPreprocessor = false;
bool inRERange = false;
bool seenDocKeyBrace = false;
int lineCurrent = styler.GetLine(startPos);
if ((MaskActive(initStyle) == SCE_C_PREPROCESSOR) ||
(MaskActive(initStyle) == SCE_C_COMMENTLINE) ||
(MaskActive(initStyle) == SCE_C_COMMENTLINEDOC)) {
// Set continuationLine if last character of previous line is '\'
if (lineCurrent > 0) {
int endLinePrevious = styler.LineEnd(lineCurrent - 1);
if (endLinePrevious > 0) {
continuationLine = styler.SafeGetCharAt(endLinePrevious-1) == '\\';
}
}
}
// look back to set chPrevNonWhite properly for better regex colouring
if (startPos > 0) {
int back = startPos;
while (--back && IsSpaceEquiv(MaskActive(styler.StyleAt(back))))
;
if (MaskActive(styler.StyleAt(back)) == SCE_C_OPERATOR) {
chPrevNonWhite = styler.SafeGetCharAt(back);
}
}
StyleContext sc(startPos, length, initStyle, styler, static_cast<unsigned char>(0xff));
LinePPState preproc = vlls.ForLine(lineCurrent);
bool definitionsChanged = false;
// Truncate ppDefineHistory before current line
if (!options.updatePreprocessor)
ppDefineHistory.clear();
std::vector<PPDefinition>::iterator itInvalid = std::find_if(ppDefineHistory.begin(), ppDefineHistory.end(), After(lineCurrent-1));
if (itInvalid != ppDefineHistory.end()) {
ppDefineHistory.erase(itInvalid, ppDefineHistory.end());
definitionsChanged = true;
}
SymbolTable preprocessorDefinitions = preprocessorDefinitionsStart;
for (std::vector<PPDefinition>::iterator itDef = ppDefineHistory.begin(); itDef != ppDefineHistory.end(); ++itDef) {
if (itDef->isUndef)
preprocessorDefinitions.erase(itDef->key);
else
preprocessorDefinitions[itDef->key] = SymbolValue(itDef->value, itDef->arguments);
}
std::string rawStringTerminator = rawStringTerminators.ValueAt(lineCurrent-1);
SparseState<std::string> rawSTNew(lineCurrent);
int activitySet = preproc.IsInactive() ? activeFlag : 0;
const WordClassifier &classifierIdentifiers = subStyles.Classifier(SCE_C_IDENTIFIER);
const WordClassifier &classifierDocKeyWords = subStyles.Classifier(SCE_C_COMMENTDOCKEYWORD);
int lineEndNext = styler.LineEnd(lineCurrent);
for (; sc.More();) {
if (sc.atLineStart) {
// Using MaskActive() is not needed in the following statement.
// Inside inactive preprocessor declaration, state will be reset anyway at the end of this block.
if ((sc.state == SCE_C_STRING) || (sc.state == SCE_C_CHARACTER)) {
// Prevent SCE_C_STRINGEOL from leaking back to previous line which
// ends with a line continuation by locking in the state up to this position.
sc.SetState(sc.state);
}
if ((MaskActive(sc.state) == SCE_C_PREPROCESSOR) && (!continuationLine)) {
sc.SetState(SCE_C_DEFAULT|activitySet);
}
// Reset states to beginning of colourise so no surprises
// if different sets of lines lexed.
visibleChars = 0;
lastWordWasUUID = false;
isIncludePreprocessor = false;
inRERange = false;
if (preproc.IsInactive()) {
activitySet = activeFlag;
sc.SetState(sc.state | activitySet);
}
}
if (sc.atLineEnd) {
lineCurrent++;
lineEndNext = styler.LineEnd(lineCurrent);
vlls.Add(lineCurrent, preproc);
if (rawStringTerminator != "") {
rawSTNew.Set(lineCurrent-1, rawStringTerminator);
}
}
// Handle line continuation generically.
if (sc.ch == '\\') {
if (static_cast<int>((sc.currentPos+1)) >= lineEndNext) {
lineCurrent++;
lineEndNext = styler.LineEnd(lineCurrent);
vlls.Add(lineCurrent, preproc);
sc.Forward();
if (sc.ch == '\r' && sc.chNext == '\n') {
// Even in UTF-8, \r and \n are separate
sc.Forward();
}
continuationLine = true;
sc.Forward();
continue;
}
}
const bool atLineEndBeforeSwitch = sc.atLineEnd;
// Determine if the current state should terminate.
switch (MaskActive(sc.state)) {
case SCE_C_OPERATOR:
sc.SetState(SCE_C_DEFAULT|activitySet);
break;
case SCE_C_NUMBER:
// We accept almost anything because of hex. and number suffixes
if (sc.ch == '_') {
sc.ChangeState(SCE_C_USERLITERAL|activitySet);
} else if (!(setWord.Contains(sc.ch)
|| (sc.ch == '\'')
|| ((sc.ch == '+' || sc.ch == '-') && (sc.chPrev == 'e' || sc.chPrev == 'E' ||
sc.chPrev == 'p' || sc.chPrev == 'P')))) {
sc.SetState(SCE_C_DEFAULT|activitySet);
}
break;
case SCE_C_USERLITERAL:
if (!(setWord.Contains(sc.ch)))
sc.SetState(SCE_C_DEFAULT|activitySet);
break;
case SCE_C_IDENTIFIER:
if (sc.atLineStart || sc.atLineEnd || !setWord.Contains(sc.ch) || (sc.ch == '.')) {
char s[1000];
if (caseSensitive) {
sc.GetCurrent(s, sizeof(s));
} else {
sc.GetCurrentLowered(s, sizeof(s));
}
if (keywords.InList(s)) {
lastWordWasUUID = strcmp(s, "uuid") == 0;
sc.ChangeState(SCE_C_WORD|activitySet);
} else if (keywords2.InList(s)) {
sc.ChangeState(SCE_C_WORD2|activitySet);
} else if (keywords4.InList(s)) {
sc.ChangeState(SCE_C_GLOBALCLASS|activitySet);
} else {
int subStyle = classifierIdentifiers.ValueFor(s);
if (subStyle >= 0) {
sc.ChangeState(subStyle|activitySet);
}
}
const bool literalString = sc.ch == '\"';
if (literalString || sc.ch == '\'') {
size_t lenS = strlen(s);
const bool raw = literalString && sc.chPrev == 'R' && !setInvalidRawFirst.Contains(sc.chNext);
if (raw)
s[lenS--] = '\0';
bool valid =
(lenS == 0) ||
((lenS == 1) && ((s[0] == 'L') || (s[0] == 'u') || (s[0] == 'U'))) ||
((lenS == 2) && literalString && (s[0] == 'u') && (s[1] == '8'));
if (valid) {
if (literalString) {
if (raw) {
// Set the style of the string prefix to SCE_C_STRINGRAW but then change to
// SCE_C_DEFAULT as that allows the raw string start code to run.
sc.ChangeState(SCE_C_STRINGRAW|activitySet);
sc.SetState(SCE_C_DEFAULT|activitySet);
} else {
sc.ChangeState(SCE_C_STRING|activitySet);
}
} else {
sc.ChangeState(SCE_C_CHARACTER|activitySet);
}
} else {
sc.SetState(SCE_C_DEFAULT | activitySet);
}
} else {
sc.SetState(SCE_C_DEFAULT|activitySet);
}
}
break;
case SCE_C_PREPROCESSOR:
if (options.stylingWithinPreprocessor) {
if (IsASpace(sc.ch)) {
sc.SetState(SCE_C_DEFAULT|activitySet);
}
} else if (isStringInPreprocessor && (sc.Match('>') || sc.Match('\"') || sc.atLineEnd)) {
isStringInPreprocessor = false;
} else if (!isStringInPreprocessor) {
if ((isIncludePreprocessor && sc.Match('<')) || sc.Match('\"')) {
isStringInPreprocessor = true;
} else if (sc.Match('/', '*')) {
if (sc.Match("/**") || sc.Match("/*!")) {
sc.SetState(SCE_C_PREPROCESSORCOMMENTDOC|activitySet);
} else {
sc.SetState(SCE_C_PREPROCESSORCOMMENT|activitySet);
}
sc.Forward(); // Eat the *
} else if (sc.Match('/', '/')) {
sc.SetState(SCE_C_DEFAULT|activitySet);
}
}
break;
case SCE_C_PREPROCESSORCOMMENT:
case SCE_C_PREPROCESSORCOMMENTDOC:
if (sc.Match('*', '/')) {
sc.Forward();
sc.ForwardSetState(SCE_C_PREPROCESSOR|activitySet);
continue; // Without advancing in case of '\'.
}
break;
case SCE_C_COMMENT:
if (sc.Match('*', '/')) {
sc.Forward();
sc.ForwardSetState(SCE_C_DEFAULT|activitySet);
} else {
styleBeforeTaskMarker = SCE_C_COMMENT;
highlightTaskMarker(sc, styler, activitySet, markerList, caseSensitive);
}
break;
case SCE_C_COMMENTDOC:
if (sc.Match('*', '/')) {
sc.Forward();
sc.ForwardSetState(SCE_C_DEFAULT|activitySet);
} else if (sc.ch == '@' || sc.ch == '\\') { // JavaDoc and Doxygen support
// Verify that we have the conditions to mark a comment-doc-keyword
if ((IsASpace(sc.chPrev) || sc.chPrev == '*') && (!IsASpace(sc.chNext))) {
styleBeforeDCKeyword = SCE_C_COMMENTDOC;
sc.SetState(SCE_C_COMMENTDOCKEYWORD|activitySet);
}
}
break;
case SCE_C_COMMENTLINE:
if (sc.atLineStart && !continuationLine) {
sc.SetState(SCE_C_DEFAULT|activitySet);
} else {
styleBeforeTaskMarker = SCE_C_COMMENTLINE;
highlightTaskMarker(sc, styler, activitySet, markerList, caseSensitive);
}
break;
case SCE_C_COMMENTLINEDOC:
if (sc.atLineStart && !continuationLine) {
sc.SetState(SCE_C_DEFAULT|activitySet);
} else if (sc.ch == '@' || sc.ch == '\\') { // JavaDoc and Doxygen support
// Verify that we have the conditions to mark a comment-doc-keyword
if ((IsASpace(sc.chPrev) || sc.chPrev == '/' || sc.chPrev == '!') && (!IsASpace(sc.chNext))) {
styleBeforeDCKeyword = SCE_C_COMMENTLINEDOC;
sc.SetState(SCE_C_COMMENTDOCKEYWORD|activitySet);
}
}
break;
case SCE_C_COMMENTDOCKEYWORD:
if ((styleBeforeDCKeyword == SCE_C_COMMENTDOC) && sc.Match('*', '/')) {
sc.ChangeState(SCE_C_COMMENTDOCKEYWORDERROR);
sc.Forward();
sc.ForwardSetState(SCE_C_DEFAULT|activitySet);
seenDocKeyBrace = false;
} else if (sc.ch == '[' || sc.ch == '{') {
seenDocKeyBrace = true;
} else if (!setDoxygen.Contains(sc.ch)
&& !(seenDocKeyBrace && (sc.ch == ',' || sc.ch == '.'))) {
char s[100];
if (caseSensitive) {
sc.GetCurrent(s, sizeof(s));
} else {
sc.GetCurrentLowered(s, sizeof(s));
}
if (!(IsASpace(sc.ch) || (sc.ch == 0))) {
sc.ChangeState(SCE_C_COMMENTDOCKEYWORDERROR|activitySet);
} else if (!keywords3.InList(s + 1)) {
int subStyleCDKW = classifierDocKeyWords.ValueFor(s+1);
if (subStyleCDKW >= 0) {
sc.ChangeState(subStyleCDKW|activitySet);
} else {
sc.ChangeState(SCE_C_COMMENTDOCKEYWORDERROR|activitySet);
}
}
sc.SetState(styleBeforeDCKeyword|activitySet);
seenDocKeyBrace = false;
}
break;
case SCE_C_STRING:
if (sc.atLineEnd) {
sc.ChangeState(SCE_C_STRINGEOL|activitySet);
} else if (isIncludePreprocessor) {
if (sc.ch == '>') {
sc.ForwardSetState(SCE_C_DEFAULT|activitySet);
isIncludePreprocessor = false;
}
} else if (sc.ch == '\\') {
if (options.escapeSequence) {
sc.SetState(SCE_C_ESCAPESEQUENCE|activitySet);
escapeSeq.resetEscapeState(sc.chNext);
}
sc.Forward(); // Skip all characters after the backslash
} else if (sc.ch == '\"') {
if (sc.chNext == '_') {
sc.ChangeState(SCE_C_USERLITERAL|activitySet);
} else {
sc.ForwardSetState(SCE_C_DEFAULT|activitySet);
}
}
break;
case SCE_C_ESCAPESEQUENCE:
escapeSeq.digitsLeft--;
if (!escapeSeq.atEscapeEnd(sc.ch)) {
break;
}
if (sc.ch == '"') {
sc.SetState(SCE_C_STRING|activitySet);
sc.ForwardSetState(SCE_C_DEFAULT|activitySet);
} else if (sc.ch == '\\') {
escapeSeq.resetEscapeState(sc.chNext);
sc.Forward();
} else {
sc.SetState(SCE_C_STRING|activitySet);
if (sc.atLineEnd) {
sc.ChangeState(SCE_C_STRINGEOL|activitySet);
}
}
break;
case SCE_C_HASHQUOTEDSTRING:
if (sc.ch == '\\') {
if (sc.chNext == '\"' || sc.chNext == '\'' || sc.chNext == '\\') {
sc.Forward();
}
} else if (sc.ch == '\"') {
sc.ForwardSetState(SCE_C_DEFAULT|activitySet);
}
break;
case SCE_C_STRINGRAW:
if (sc.Match(rawStringTerminator.c_str())) {
for (size_t termPos=rawStringTerminator.size(); termPos; termPos--)
sc.Forward();
sc.SetState(SCE_C_DEFAULT|activitySet);
rawStringTerminator = "";
}
break;
case SCE_C_CHARACTER:
if (sc.atLineEnd) {
sc.ChangeState(SCE_C_STRINGEOL|activitySet);
} else if (sc.ch == '\\') {
if (sc.chNext == '\"' || sc.chNext == '\'' || sc.chNext == '\\') {
sc.Forward();
}
} else if (sc.ch == '\'') {
if (sc.chNext == '_') {
sc.ChangeState(SCE_C_USERLITERAL|activitySet);
} else {
sc.ForwardSetState(SCE_C_DEFAULT|activitySet);
}
}
break;
case SCE_C_REGEX:
if (sc.atLineStart) {
sc.SetState(SCE_C_DEFAULT|activitySet);
} else if (! inRERange && sc.ch == '/') {
sc.Forward();
while ((sc.ch < 0x80) && islower(sc.ch))
sc.Forward(); // gobble regex flags
sc.SetState(SCE_C_DEFAULT|activitySet);
} else if (sc.ch == '\\' && (static_cast<int>(sc.currentPos+1) < lineEndNext)) {
// Gobble up the escaped character
sc.Forward();
} else if (sc.ch == '[') {
inRERange = true;
} else if (sc.ch == ']') {
inRERange = false;
}
break;
case SCE_C_STRINGEOL:
if (sc.atLineStart) {
sc.SetState(SCE_C_DEFAULT|activitySet);
}
break;
case SCE_C_VERBATIM:
if (options.verbatimStringsAllowEscapes && (sc.ch == '\\')) {
sc.Forward(); // Skip all characters after the backslash
} else if (sc.ch == '\"') {
if (sc.chNext == '\"') {
sc.Forward();
} else {
sc.ForwardSetState(SCE_C_DEFAULT|activitySet);
}
}
break;
case SCE_C_TRIPLEVERBATIM:
if (sc.Match("\"\"\"")) {
while (sc.Match('"')) {
sc.Forward();
}
sc.SetState(SCE_C_DEFAULT|activitySet);
}
break;
case SCE_C_UUID:
if (sc.atLineEnd || sc.ch == ')') {
sc.SetState(SCE_C_DEFAULT|activitySet);
}
break;
case SCE_C_TASKMARKER:
if (isoperator(sc.ch) || IsASpace(sc.ch)) {
sc.SetState(styleBeforeTaskMarker|activitySet);
styleBeforeTaskMarker = SCE_C_DEFAULT;
}
}
if (sc.atLineEnd && !atLineEndBeforeSwitch) {
// State exit processing consumed characters up to end of line.
lineCurrent++;
lineEndNext = styler.LineEnd(lineCurrent);
vlls.Add(lineCurrent, preproc);
}
// Determine if a new state should be entered.
if (MaskActive(sc.state) == SCE_C_DEFAULT) {
if (sc.Match('@', '\"')) {
sc.SetState(SCE_C_VERBATIM|activitySet);
sc.Forward();
} else if (options.triplequotedStrings && sc.Match("\"\"\"")) {
sc.SetState(SCE_C_TRIPLEVERBATIM|activitySet);
sc.Forward(2);
} else if (options.hashquotedStrings && sc.Match('#', '\"')) {
sc.SetState(SCE_C_HASHQUOTEDSTRING|activitySet);
sc.Forward();
} else if (options.backQuotedStrings && sc.Match('`')) {
sc.SetState(SCE_C_STRINGRAW|activitySet);
rawStringTerminator = "`";
} else if (IsADigit(sc.ch) || (sc.ch == '.' && IsADigit(sc.chNext))) {
if (lastWordWasUUID) {
sc.SetState(SCE_C_UUID|activitySet);
lastWordWasUUID = false;
} else {
sc.SetState(SCE_C_NUMBER|activitySet);
}
} else if (!sc.atLineEnd && (setWordStart.Contains(sc.ch) || (sc.ch == '@'))) {
if (lastWordWasUUID) {
sc.SetState(SCE_C_UUID|activitySet);
lastWordWasUUID = false;
} else {
sc.SetState(SCE_C_IDENTIFIER|activitySet);
}
} else if (sc.Match('/', '*')) {
if (sc.Match("/**") || sc.Match("/*!")) { // Support of Qt/Doxygen doc. style
sc.SetState(SCE_C_COMMENTDOC|activitySet);
} else {
sc.SetState(SCE_C_COMMENT|activitySet);
}
sc.Forward(); // Eat the * so it isn't used for the end of the comment
} else if (sc.Match('/', '/')) {
if ((sc.Match("///") && !sc.Match("////")) || sc.Match("//!"))
// Support of Qt/Doxygen doc. style
sc.SetState(SCE_C_COMMENTLINEDOC|activitySet);
else
sc.SetState(SCE_C_COMMENTLINE|activitySet);
} else if (sc.ch == '/'
&& (setOKBeforeRE.Contains(chPrevNonWhite)
|| followsReturnKeyword(sc, styler))
&& (!setCouldBePostOp.Contains(chPrevNonWhite)
|| !FollowsPostfixOperator(sc, styler))) {
sc.SetState(SCE_C_REGEX|activitySet); // JavaScript's RegEx
inRERange = false;
} else if (sc.ch == '\"') {
if (sc.chPrev == 'R') {
styler.Flush();
if (MaskActive(styler.StyleAt(sc.currentPos - 1)) == SCE_C_STRINGRAW) {
sc.SetState(SCE_C_STRINGRAW|activitySet);
rawStringTerminator = ")";
for (int termPos = sc.currentPos + 1;; termPos++) {
char chTerminator = styler.SafeGetCharAt(termPos, '(');
if (chTerminator == '(')
break;
rawStringTerminator += chTerminator;
}
rawStringTerminator += '\"';
} else {
sc.SetState(SCE_C_STRING|activitySet);
}
} else {
sc.SetState(SCE_C_STRING|activitySet);
}
isIncludePreprocessor = false; // ensure that '>' won't end the string
} else if (isIncludePreprocessor && sc.ch == '<') {
sc.SetState(SCE_C_STRING|activitySet);
} else if (sc.ch == '\'') {
sc.SetState(SCE_C_CHARACTER|activitySet);
} else if (sc.ch == '#' && visibleChars == 0) {
// Preprocessor commands are alone on their line
sc.SetState(SCE_C_PREPROCESSOR|activitySet);
// Skip whitespace between # and preprocessor word
do {
sc.Forward();
} while ((sc.ch == ' ' || sc.ch == '\t') && sc.More());
if (sc.atLineEnd) {
sc.SetState(SCE_C_DEFAULT|activitySet);
} else if (sc.Match("include")) {
isIncludePreprocessor = true;
} else {
if (options.trackPreprocessor) {
if (sc.Match("ifdef") || sc.Match("ifndef")) {
bool isIfDef = sc.Match("ifdef");
int i = isIfDef ? 5 : 6;
std::string restOfLine = GetRestOfLine(styler, sc.currentPos + i + 1, false);
bool foundDef = preprocessorDefinitions.find(restOfLine) != preprocessorDefinitions.end();
preproc.StartSection(isIfDef == foundDef);
} else if (sc.Match("if")) {
std::string restOfLine = GetRestOfLine(styler, sc.currentPos + 2, true);
bool ifGood = EvaluateExpression(restOfLine, preprocessorDefinitions);
preproc.StartSection(ifGood);
} else if (sc.Match("else")) {
if (!preproc.CurrentIfTaken()) {
preproc.InvertCurrentLevel();
activitySet = preproc.IsInactive() ? activeFlag : 0;
if (!activitySet)
sc.ChangeState(SCE_C_PREPROCESSOR|activitySet);
} else if (!preproc.IsInactive()) {
preproc.InvertCurrentLevel();
activitySet = preproc.IsInactive() ? activeFlag : 0;
if (!activitySet)
sc.ChangeState(SCE_C_PREPROCESSOR|activitySet);
}
} else if (sc.Match("elif")) {
// Ensure only one chosen out of #if .. #elif .. #elif .. #else .. #endif
if (!preproc.CurrentIfTaken()) {
// Similar to #if
std::string restOfLine = GetRestOfLine(styler, sc.currentPos + 2, true);
bool ifGood = EvaluateExpression(restOfLine, preprocessorDefinitions);
if (ifGood) {
preproc.InvertCurrentLevel();
activitySet = preproc.IsInactive() ? activeFlag : 0;
if (!activitySet)
sc.ChangeState(SCE_C_PREPROCESSOR|activitySet);
}
} else if (!preproc.IsInactive()) {
preproc.InvertCurrentLevel();
activitySet = preproc.IsInactive() ? activeFlag : 0;
if (!activitySet)
sc.ChangeState(SCE_C_PREPROCESSOR|activitySet);
}
} else if (sc.Match("endif")) {
preproc.EndSection();
activitySet = preproc.IsInactive() ? activeFlag : 0;
sc.ChangeState(SCE_C_PREPROCESSOR|activitySet);
} else if (sc.Match("define")) {
if (options.updatePreprocessor && !preproc.IsInactive()) {
std::string restOfLine = GetRestOfLine(styler, sc.currentPos + 6, true);
size_t startName = 0;
while ((startName < restOfLine.length()) && IsSpaceOrTab(restOfLine[startName]))
startName++;
size_t endName = startName;
while ((endName < restOfLine.length()) && setWord.Contains(static_cast<unsigned char>(restOfLine[endName])))
endName++;
std::string key = restOfLine.substr(startName, endName-startName);
if ((endName < restOfLine.length()) && (restOfLine.at(endName) == '(')) {
// Macro
size_t endArgs = endName;
while ((endArgs < restOfLine.length()) && (restOfLine[endArgs] != ')'))
endArgs++;
std::string args = restOfLine.substr(endName + 1, endArgs - endName - 1);
size_t startValue = endArgs+1;
while ((startValue < restOfLine.length()) && IsSpaceOrTab(restOfLine[startValue]))
startValue++;
std::string value;
if (startValue < restOfLine.length())
value = restOfLine.substr(startValue);
preprocessorDefinitions[key] = SymbolValue(value, args);
ppDefineHistory.push_back(PPDefinition(lineCurrent, key, value, false, args));
definitionsChanged = true;
} else {
// Value
size_t startValue = endName;
while ((startValue < restOfLine.length()) && IsSpaceOrTab(restOfLine[startValue]))
startValue++;
std::string value = restOfLine.substr(startValue);
preprocessorDefinitions[key] = value;
ppDefineHistory.push_back(PPDefinition(lineCurrent, key, value));
definitionsChanged = true;
}
}
} else if (sc.Match("undef")) {
if (options.updatePreprocessor && !preproc.IsInactive()) {
const std::string restOfLine = GetRestOfLine(styler, sc.currentPos + 5, false);
std::vector<std::string> tokens = Tokenize(restOfLine);
if (tokens.size() >= 1) {
const std::string key = tokens[0];
preprocessorDefinitions.erase(key);
ppDefineHistory.push_back(PPDefinition(lineCurrent, key, "", true));
definitionsChanged = true;
}
}
}
}
}
} else if (isoperator(sc.ch)) {
sc.SetState(SCE_C_OPERATOR|activitySet);
}
}
if (!IsASpace(sc.ch) && !IsSpaceEquiv(MaskActive(sc.state))) {
chPrevNonWhite = sc.ch;
visibleChars++;
}
continuationLine = false;
sc.Forward();
}
const bool rawStringsChanged = rawStringTerminators.Merge(rawSTNew, lineCurrent);
if (definitionsChanged || rawStringsChanged)
styler.ChangeLexerState(startPos, startPos + length);
sc.Complete();
}
// Store both the current line's fold level and the next lines in the
// level store to make it easy to pick up with each increment
// and to make it possible to fiddle the current level for "} else {".
void SCI_METHOD LexerCPP::Fold(unsigned int startPos, int length, int initStyle, IDocument *pAccess) {
if (!options.fold)
return;
LexAccessor styler(pAccess);
unsigned int endPos = startPos + length;
int visibleChars = 0;
bool inLineComment = false;
int lineCurrent = styler.GetLine(startPos);
int levelCurrent = SC_FOLDLEVELBASE;
if (lineCurrent > 0)
levelCurrent = styler.LevelAt(lineCurrent-1) >> 16;
unsigned int lineStartNext = styler.LineStart(lineCurrent+1);
int levelMinCurrent = levelCurrent;
int levelNext = levelCurrent;
char chNext = styler[startPos];
int styleNext = MaskActive(styler.StyleAt(startPos));
int style = MaskActive(initStyle);
const bool userDefinedFoldMarkers = !options.foldExplicitStart.empty() && !options.foldExplicitEnd.empty();
for (unsigned int i = startPos; i < endPos; i++) {
char ch = chNext;
chNext = styler.SafeGetCharAt(i + 1);
int stylePrev = style;
style = styleNext;
styleNext = MaskActive(styler.StyleAt(i + 1));
bool atEOL = i == (lineStartNext-1);
if ((style == SCE_C_COMMENTLINE) || (style == SCE_C_COMMENTLINEDOC))
inLineComment = true;
if (options.foldComment && options.foldCommentMultiline && IsStreamCommentStyle(style) && !inLineComment) {
if (!IsStreamCommentStyle(stylePrev)) {
levelNext++;
} else if (!IsStreamCommentStyle(styleNext) && !atEOL) {
// Comments don't end at end of line and the next character may be unstyled.
levelNext--;
}
}
if (options.foldComment && options.foldCommentExplicit && ((style == SCE_C_COMMENTLINE) || options.foldExplicitAnywhere)) {
if (userDefinedFoldMarkers) {
if (styler.Match(i, options.foldExplicitStart.c_str())) {
levelNext++;
} else if (styler.Match(i, options.foldExplicitEnd.c_str())) {
levelNext--;
}
} else {
if ((ch == '/') && (chNext == '/')) {
char chNext2 = styler.SafeGetCharAt(i + 2);
if (chNext2 == '{') {
levelNext++;
} else if (chNext2 == '}') {
levelNext--;
}
}
}
}
if (options.foldPreprocessor && (style == SCE_C_PREPROCESSOR)) {
if (ch == '#') {
unsigned int j = i + 1;
while ((j < endPos) && IsASpaceOrTab(styler.SafeGetCharAt(j))) {
j++;
}
if (styler.Match(j, "region") || styler.Match(j, "if")) {
levelNext++;
} else if (styler.Match(j, "end")) {
levelNext--;
}
}
}
if (options.foldSyntaxBased && (style == SCE_C_OPERATOR)) {
if (ch == '{' || ch == '[') {
// Measure the minimum before a '{' to allow
// folding on "} else {"
if (levelMinCurrent > levelNext) {
levelMinCurrent = levelNext;
}
levelNext++;
} else if (ch == '}' || ch == ']') {
levelNext--;
}
}
if (!IsASpace(ch))
visibleChars++;
if (atEOL || (i == endPos-1)) {
int levelUse = levelCurrent;
if (options.foldSyntaxBased && options.foldAtElse) {
levelUse = levelMinCurrent;
}
int lev = levelUse | levelNext << 16;
if (visibleChars == 0 && options.foldCompact)
lev |= SC_FOLDLEVELWHITEFLAG;
if (levelUse < levelNext)
lev |= SC_FOLDLEVELHEADERFLAG;
if (lev != styler.LevelAt(lineCurrent)) {
styler.SetLevel(lineCurrent, lev);
}
lineCurrent++;
lineStartNext = styler.LineStart(lineCurrent+1);
levelCurrent = levelNext;
levelMinCurrent = levelCurrent;
if (atEOL && (i == static_cast<unsigned int>(styler.Length()-1))) {
// There is an empty line at end of file so give it same level and empty
styler.SetLevel(lineCurrent, (levelCurrent | levelCurrent << 16) | SC_FOLDLEVELWHITEFLAG);
}
visibleChars = 0;
inLineComment = false;
}
}
}
void LexerCPP::EvaluateTokens(std::vector<std::string> &tokens, const SymbolTable &preprocessorDefinitions) {
// Remove whitespace tokens
tokens.erase(std::remove_if(tokens.begin(), tokens.end(), OnlySpaceOrTab), tokens.end());
// Evaluate defined statements to either 0 or 1
for (size_t i=0; (i+1)<tokens.size();) {
if (tokens[i] == "defined") {
const char *val = "0";
if (tokens[i+1] == "(") {
if (((i + 2)<tokens.size()) && (tokens[i + 2] == ")")) {
// defined()
tokens.erase(tokens.begin() + i + 1, tokens.begin() + i + 3);
} else if (((i+3)<tokens.size()) && (tokens[i+3] == ")")) {
// defined(<identifier>)
SymbolTable::const_iterator it = preprocessorDefinitions.find(tokens[i+2]);
if (it != preprocessorDefinitions.end()) {
val = "1";
}
tokens.erase(tokens.begin() + i + 1, tokens.begin() + i + 4);
} else {
// Spurious '(' so erase as more likely to result in false
tokens.erase(tokens.begin() + i + 1, tokens.begin() + i + 2);
}
} else {
// defined <identifier>
SymbolTable::const_iterator it = preprocessorDefinitions.find(tokens[i+1]);
if (it != preprocessorDefinitions.end()) {
val = "1";
}
}
tokens[i] = val;
} else {
i++;
}
}
// Evaluate identifiers
const size_t maxIterations = 100;
size_t iterations = 0; // Limit number of iterations in case there is a recursive macro.
for (size_t i = 0; (i<tokens.size()) && (iterations < maxIterations);) {
iterations++;
if (setWordStart.Contains(static_cast<unsigned char>(tokens[i][0]))) {
SymbolTable::const_iterator it = preprocessorDefinitions.find(tokens[i]);
if (it != preprocessorDefinitions.end()) {
// Tokenize value
std::vector<std::string> macroTokens = Tokenize(it->second.value);
if (it->second.IsMacro()) {
if ((i + 1 < tokens.size()) && (tokens.at(i + 1) == "(")) {
// Create map of argument name to value
std::vector<std::string> argumentNames = StringSplit(it->second.arguments, ',');
std::map<std::string, std::string> arguments;
size_t arg = 0;
size_t tok = i+2;
while ((tok < tokens.size()) && (arg < argumentNames.size()) && (tokens.at(tok) != ")")) {
if (tokens.at(tok) != ",") {
arguments[argumentNames.at(arg)] = tokens.at(tok);
arg++;
}
tok++;
}
// Remove invocation
tokens.erase(tokens.begin() + i, tokens.begin() + tok + 1);
// Substitute values into macro
macroTokens.erase(std::remove_if(macroTokens.begin(), macroTokens.end(), OnlySpaceOrTab), macroTokens.end());
for (size_t iMacro = 0; iMacro < macroTokens.size();) {
if (setWordStart.Contains(static_cast<unsigned char>(macroTokens[iMacro][0]))) {
std::map<std::string, std::string>::const_iterator itFind = arguments.find(macroTokens[iMacro]);
if (itFind != arguments.end()) {
// TODO: Possible that value will be expression so should insert tokenized form
macroTokens[iMacro] = itFind->second;
}
}
iMacro++;
}
// Insert results back into tokens
tokens.insert(tokens.begin() + i, macroTokens.begin(), macroTokens.end());
} else {
i++;
}
} else {
// Remove invocation
tokens.erase(tokens.begin() + i);
// Insert results back into tokens
tokens.insert(tokens.begin() + i, macroTokens.begin(), macroTokens.end());
}
} else {
// Identifier not found
tokens.erase(tokens.begin() + i);
}
} else {
i++;
}
}
// Find bracketed subexpressions and recurse on them
BracketPair bracketPair = FindBracketPair(tokens);
while (bracketPair.itBracket != tokens.end()) {
std::vector<std::string> inBracket(bracketPair.itBracket + 1, bracketPair.itEndBracket);
EvaluateTokens(inBracket, preprocessorDefinitions);
// The insertion is done before the removal because there were failures with the opposite approach
tokens.insert(bracketPair.itBracket, inBracket.begin(), inBracket.end());
bracketPair = FindBracketPair(tokens);
tokens.erase(bracketPair.itBracket, bracketPair.itEndBracket + 1);
bracketPair = FindBracketPair(tokens);
}
// Evaluate logical negations
for (size_t j=0; (j+1)<tokens.size();) {
if (setNegationOp.Contains(tokens[j][0])) {
int isTrue = atoi(tokens[j+1].c_str());
if (tokens[j] == "!")
isTrue = !isTrue;
std::vector<std::string>::iterator itInsert =
tokens.erase(tokens.begin() + j, tokens.begin() + j + 2);
tokens.insert(itInsert, isTrue ? "1" : "0");
} else {
j++;
}
}
// Evaluate expressions in precedence order
enum precedence { precArithmetic, precRelative, precLogical };
for (int prec=precArithmetic; prec <= precLogical; prec++) {
// Looking at 3 tokens at a time so end at 2 before end
for (size_t k=0; (k+2)<tokens.size();) {
char chOp = tokens[k+1][0];
if (
((prec==precArithmetic) && setArithmethicOp.Contains(chOp)) ||
((prec==precRelative) && setRelOp.Contains(chOp)) ||
((prec==precLogical) && setLogicalOp.Contains(chOp))
) {
int valA = atoi(tokens[k].c_str());
int valB = atoi(tokens[k+2].c_str());
int result = 0;
if (tokens[k+1] == "+")
result = valA + valB;
else if (tokens[k+1] == "-")
result = valA - valB;
else if (tokens[k+1] == "*")
result = valA * valB;
else if (tokens[k+1] == "/")
result = valA / (valB ? valB : 1);
else if (tokens[k+1] == "%")
result = valA % (valB ? valB : 1);
else if (tokens[k+1] == "<")
result = valA < valB;
else if (tokens[k+1] == "<=")
result = valA <= valB;
else if (tokens[k+1] == ">")
result = valA > valB;
else if (tokens[k+1] == ">=")
result = valA >= valB;
else if (tokens[k+1] == "==")
result = valA == valB;
else if (tokens[k+1] == "!=")
result = valA != valB;
else if (tokens[k+1] == "||")
result = valA || valB;
else if (tokens[k+1] == "&&")
result = valA && valB;
char sResult[30];
sprintf(sResult, "%d", result);
std::vector<std::string>::iterator itInsert =
tokens.erase(tokens.begin() + k, tokens.begin() + k + 3);
tokens.insert(itInsert, sResult);
} else {
k++;
}
}
}
}
std::vector<std::string> LexerCPP::Tokenize(const std::string &expr) const {
// Break into tokens
std::vector<std::string> tokens;
const char *cp = expr.c_str();
while (*cp) {
std::string word;
if (setWord.Contains(static_cast<unsigned char>(*cp))) {
// Identifiers and numbers
while (setWord.Contains(static_cast<unsigned char>(*cp))) {
word += *cp;
cp++;
}
} else if (IsSpaceOrTab(*cp)) {
while (IsSpaceOrTab(*cp)) {
word += *cp;
cp++;
}
} else if (setRelOp.Contains(static_cast<unsigned char>(*cp))) {
word += *cp;
cp++;
if (setRelOp.Contains(static_cast<unsigned char>(*cp))) {
word += *cp;
cp++;
}
} else if (setLogicalOp.Contains(static_cast<unsigned char>(*cp))) {
word += *cp;
cp++;
if (setLogicalOp.Contains(static_cast<unsigned char>(*cp))) {
word += *cp;
cp++;
}
} else {
// Should handle strings, characters, and comments here
word += *cp;
cp++;
}
tokens.push_back(word);
}
return tokens;
}
bool LexerCPP::EvaluateExpression(const std::string &expr, const SymbolTable &preprocessorDefinitions) {
std::vector<std::string> tokens = Tokenize(expr);
EvaluateTokens(tokens, preprocessorDefinitions);
// "0" or "" -> false else true
bool isFalse = tokens.empty() ||
((tokens.size() == 1) && ((tokens[0] == "") || tokens[0] == "0"));
return !isFalse;
}
LexerModule lmCPP(SCLEX_CPP, LexerCPP::LexerFactoryCPP, "cpp", cppWordLists);
LexerModule lmCPPNoCase(SCLEX_CPPNOCASE, LexerCPP::LexerFactoryCPPInsensitive, "cppnocase", cppWordLists);