1 #include "clang/AST/CommentLexer.h"
2 #include "clang/AST/CommentCommandTraits.h"
3 #include "clang/AST/CommentDiagnostic.h"
4 #include "clang/Basic/CharInfo.h"
5 #include "llvm/ADT/StringExtras.h"
6 #include "llvm/ADT/StringSwitch.h"
7 #include "llvm/Support/ConvertUTF.h"
8 #include "llvm/Support/ErrorHandling.h"
13 void Token::dump(const Lexer &L, const SourceManager &SM) const {
14 llvm::errs() << "comments::Token Kind=" << Kind << " ";
16 llvm::errs() << " " << Length << " \"" << L.getSpelling(*this, SM) << "\"\n";
19 static inline bool isHTMLNamedCharacterReferenceCharacter(char C) {
23 static inline bool isHTMLDecimalCharacterReferenceCharacter(char C) {
27 static inline bool isHTMLHexCharacterReferenceCharacter(char C) {
31 static inline StringRef convertCodePointToUTF8(
32 llvm::BumpPtrAllocator &Allocator,
34 char *Resolved = Allocator.Allocate<char>(UNI_MAX_UTF8_BYTES_PER_CODE_POINT);
35 char *ResolvedPtr = Resolved;
36 if (llvm::ConvertCodePointToUTF8(CodePoint, ResolvedPtr))
37 return StringRef(Resolved, ResolvedPtr - Resolved);
44 #include "clang/AST/CommentHTMLTags.inc"
45 #include "clang/AST/CommentHTMLNamedCharacterReferences.inc"
47 } // unnamed namespace
49 StringRef Lexer::resolveHTMLNamedCharacterReference(StringRef Name) const {
50 // Fast path, first check a few most widely used named character references.
51 return llvm::StringSwitch<StringRef>(Name)
58 .Default(translateHTMLNamedCharacterReferenceToUTF8(Name));
61 StringRef Lexer::resolveHTMLDecimalCharacterReference(StringRef Name) const {
62 unsigned CodePoint = 0;
63 for (unsigned i = 0, e = Name.size(); i != e; ++i) {
64 assert(isHTMLDecimalCharacterReferenceCharacter(Name[i]));
66 CodePoint += Name[i] - '0';
68 return convertCodePointToUTF8(Allocator, CodePoint);
71 StringRef Lexer::resolveHTMLHexCharacterReference(StringRef Name) const {
72 unsigned CodePoint = 0;
73 for (unsigned i = 0, e = Name.size(); i != e; ++i) {
75 const char C = Name[i];
76 assert(isHTMLHexCharacterReferenceCharacter(C));
77 CodePoint += llvm::hexDigitValue(C);
79 return convertCodePointToUTF8(Allocator, CodePoint);
82 void Lexer::skipLineStartingDecorations() {
83 // This function should be called only for C comments
84 assert(CommentState == LCS_InsideCComment);
86 if (BufferPtr == CommentEnd)
94 const char *NewBufferPtr = BufferPtr;
96 if (NewBufferPtr == CommentEnd)
99 char C = *NewBufferPtr;
100 while (isHorizontalWhitespace(C)) {
102 if (NewBufferPtr == CommentEnd)
107 BufferPtr = NewBufferPtr + 1;
117 /// Returns pointer to the first newline character in the string.
118 const char *findNewline(const char *BufferPtr, const char *BufferEnd) {
119 for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
120 if (isVerticalWhitespace(*BufferPtr))
126 const char *skipNewline(const char *BufferPtr, const char *BufferEnd) {
127 if (BufferPtr == BufferEnd)
130 if (*BufferPtr == '\n')
133 assert(*BufferPtr == '\r');
135 if (BufferPtr != BufferEnd && *BufferPtr == '\n')
141 const char *skipNamedCharacterReference(const char *BufferPtr,
142 const char *BufferEnd) {
143 for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
144 if (!isHTMLNamedCharacterReferenceCharacter(*BufferPtr))
150 const char *skipDecimalCharacterReference(const char *BufferPtr,
151 const char *BufferEnd) {
152 for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
153 if (!isHTMLDecimalCharacterReferenceCharacter(*BufferPtr))
159 const char *skipHexCharacterReference(const char *BufferPtr,
160 const char *BufferEnd) {
161 for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
162 if (!isHTMLHexCharacterReferenceCharacter(*BufferPtr))
168 bool isHTMLIdentifierStartingCharacter(char C) {
172 bool isHTMLIdentifierCharacter(char C) {
173 return isAlphanumeric(C);
176 const char *skipHTMLIdentifier(const char *BufferPtr, const char *BufferEnd) {
177 for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
178 if (!isHTMLIdentifierCharacter(*BufferPtr))
184 /// Skip HTML string quoted in single or double quotes. Escaping quotes inside
187 /// Returns pointer to closing quote.
188 const char *skipHTMLQuotedString(const char *BufferPtr, const char *BufferEnd)
190 const char Quote = *BufferPtr;
191 assert(Quote == '\"' || Quote == '\'');
194 for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
195 const char C = *BufferPtr;
196 if (C == Quote && BufferPtr[-1] != '\\')
202 const char *skipWhitespace(const char *BufferPtr, const char *BufferEnd) {
203 for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
204 if (!isWhitespace(*BufferPtr))
210 bool isWhitespace(const char *BufferPtr, const char *BufferEnd) {
211 return skipWhitespace(BufferPtr, BufferEnd) == BufferEnd;
214 bool isCommandNameStartCharacter(char C) {
218 bool isCommandNameCharacter(char C) {
219 return isAlphanumeric(C);
222 const char *skipCommandName(const char *BufferPtr, const char *BufferEnd) {
223 for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
224 if (!isCommandNameCharacter(*BufferPtr))
230 /// Return the one past end pointer for BCPL comments.
231 /// Handles newlines escaped with backslash or trigraph for backslahs.
232 const char *findBCPLCommentEnd(const char *BufferPtr, const char *BufferEnd) {
233 const char *CurPtr = BufferPtr;
234 while (CurPtr != BufferEnd) {
235 while (!isVerticalWhitespace(*CurPtr)) {
237 if (CurPtr == BufferEnd)
240 // We found a newline, check if it is escaped.
241 const char *EscapePtr = CurPtr - 1;
242 while(isHorizontalWhitespace(*EscapePtr))
245 if (*EscapePtr == '\\' ||
246 (EscapePtr - 2 >= BufferPtr && EscapePtr[0] == '/' &&
247 EscapePtr[-1] == '?' && EscapePtr[-2] == '?')) {
248 // We found an escaped newline.
249 CurPtr = skipNewline(CurPtr, BufferEnd);
251 return CurPtr; // Not an escaped newline.
256 /// Return the one past end pointer for C comments.
257 /// Very dumb, does not handle escaped newlines or trigraphs.
258 const char *findCCommentEnd(const char *BufferPtr, const char *BufferEnd) {
259 for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
260 if (*BufferPtr == '*') {
261 assert(BufferPtr + 1 != BufferEnd);
262 if (*(BufferPtr + 1) == '/')
266 llvm_unreachable("buffer end hit before '*/' was seen");
269 } // unnamed namespace
271 void Lexer::formTokenWithChars(Token &Result, const char *TokEnd,
272 tok::TokenKind Kind) {
273 const unsigned TokLen = TokEnd - BufferPtr;
274 Result.setLocation(getSourceLocation(BufferPtr));
275 Result.setKind(Kind);
276 Result.setLength(TokLen);
278 Result.TextPtr = "<UNSET>";
284 void Lexer::lexCommentText(Token &T) {
285 assert(CommentState == LCS_InsideBCPLComment ||
286 CommentState == LCS_InsideCComment);
291 case LS_VerbatimBlockFirstLine:
292 lexVerbatimBlockFirstLine(T);
294 case LS_VerbatimBlockBody:
295 lexVerbatimBlockBody(T);
297 case LS_VerbatimLineText:
298 lexVerbatimLineText(T);
300 case LS_HTMLStartTag:
308 assert(State == LS_Normal);
310 const char *TokenPtr = BufferPtr;
311 assert(TokenPtr < CommentEnd);
312 while (TokenPtr != CommentEnd) {
316 // Commands that start with a backslash and commands that start with
317 // 'at' have equivalent semantics. But we keep information about the
318 // exact syntax in AST for comments.
319 tok::TokenKind CommandKind =
320 (*TokenPtr == '@') ? tok::at_command : tok::backslash_command;
322 if (TokenPtr == CommentEnd) {
323 formTextToken(T, TokenPtr);
331 case '\\': case '@': case '&': case '$':
332 case '#': case '<': case '>': case '%':
333 case '\"': case '.': case ':':
334 // This is one of \\ \@ \& \$ etc escape sequences.
336 if (C == ':' && TokenPtr != CommentEnd && *TokenPtr == ':') {
337 // This is the \:: escape sequence.
340 StringRef UnescapedText(BufferPtr + 1, TokenPtr - (BufferPtr + 1));
341 formTokenWithChars(T, TokenPtr, tok::text);
342 T.setText(UnescapedText);
346 // Don't make zero-length commands.
347 if (!isCommandNameStartCharacter(*TokenPtr)) {
348 formTextToken(T, TokenPtr);
352 TokenPtr = skipCommandName(TokenPtr, CommentEnd);
353 unsigned Length = TokenPtr - (BufferPtr + 1);
355 // Hardcoded support for lexing LaTeX formula commands
356 // \f$ \f[ \f] \f{ \f} as a single command.
357 if (Length == 1 && TokenPtr[-1] == 'f' && TokenPtr != CommentEnd) {
359 if (C == '$' || C == '[' || C == ']' || C == '{' || C == '}') {
365 StringRef CommandName(BufferPtr + 1, Length);
367 const CommandInfo *Info = Traits.getCommandInfoOrNULL(CommandName);
369 if ((Info = Traits.getTypoCorrectCommandInfo(CommandName))) {
370 StringRef CorrectedName = Info->Name;
371 SourceLocation Loc = getSourceLocation(BufferPtr);
372 SourceRange CommandRange(Loc.getLocWithOffset(1),
373 getSourceLocation(TokenPtr));
374 Diag(Loc, diag::warn_correct_comment_command_name)
375 << CommandName << CorrectedName
376 << FixItHint::CreateReplacement(CommandRange, CorrectedName);
378 formTokenWithChars(T, TokenPtr, tok::unknown_command);
379 T.setUnknownCommandName(CommandName);
380 Diag(T.getLocation(), diag::warn_unknown_comment_command_name);
384 if (Info->IsVerbatimBlockCommand) {
385 setupAndLexVerbatimBlock(T, TokenPtr, *BufferPtr, Info);
388 if (Info->IsVerbatimLineCommand) {
389 setupAndLexVerbatimLine(T, TokenPtr, Info);
392 formTokenWithChars(T, TokenPtr, CommandKind);
393 T.setCommandID(Info->getID());
398 lexHTMLCharacterReference(T);
403 if (TokenPtr == CommentEnd) {
404 formTextToken(T, TokenPtr);
407 const char C = *TokenPtr;
408 if (isHTMLIdentifierStartingCharacter(C))
409 setupAndLexHTMLStartTag(T);
411 setupAndLexHTMLEndTag(T);
413 formTextToken(T, TokenPtr);
420 TokenPtr = skipNewline(TokenPtr, CommentEnd);
421 formTokenWithChars(T, TokenPtr, tok::newline);
423 if (CommentState == LCS_InsideCComment)
424 skipLineStartingDecorations();
428 size_t End = StringRef(TokenPtr, CommentEnd - TokenPtr).
429 find_first_of("\n\r\\@&<");
430 if (End != StringRef::npos)
433 TokenPtr = CommentEnd;
434 formTextToken(T, TokenPtr);
441 void Lexer::setupAndLexVerbatimBlock(Token &T,
442 const char *TextBegin,
443 char Marker, const CommandInfo *Info) {
444 assert(Info->IsVerbatimBlockCommand);
446 VerbatimBlockEndCommandName.clear();
447 VerbatimBlockEndCommandName.append(Marker == '\\' ? "\\" : "@");
448 VerbatimBlockEndCommandName.append(Info->EndCommandName);
450 formTokenWithChars(T, TextBegin, tok::verbatim_block_begin);
451 T.setVerbatimBlockID(Info->getID());
453 // If there is a newline following the verbatim opening command, skip the
454 // newline so that we don't create an tok::verbatim_block_line with empty
456 if (BufferPtr != CommentEnd &&
457 isVerticalWhitespace(*BufferPtr)) {
458 BufferPtr = skipNewline(BufferPtr, CommentEnd);
459 State = LS_VerbatimBlockBody;
463 State = LS_VerbatimBlockFirstLine;
466 void Lexer::lexVerbatimBlockFirstLine(Token &T) {
468 assert(BufferPtr < CommentEnd);
470 // FIXME: It would be better to scan the text once, finding either the block
471 // end command or newline.
473 // Extract current line.
474 const char *Newline = findNewline(BufferPtr, CommentEnd);
475 StringRef Line(BufferPtr, Newline - BufferPtr);
477 // Look for end command in current line.
478 size_t Pos = Line.find(VerbatimBlockEndCommandName);
480 const char *NextLine;
481 if (Pos == StringRef::npos) {
482 // Current line is completely verbatim.
484 NextLine = skipNewline(Newline, CommentEnd);
485 } else if (Pos == 0) {
486 // Current line contains just an end command.
487 const char *End = BufferPtr + VerbatimBlockEndCommandName.size();
488 StringRef Name(BufferPtr + 1, End - (BufferPtr + 1));
489 formTokenWithChars(T, End, tok::verbatim_block_end);
490 T.setVerbatimBlockID(Traits.getCommandInfo(Name)->getID());
494 // There is some text, followed by end command. Extract text first.
495 TextEnd = BufferPtr + Pos;
497 // If there is only whitespace before end command, skip whitespace.
498 if (isWhitespace(BufferPtr, TextEnd)) {
504 StringRef Text(BufferPtr, TextEnd - BufferPtr);
505 formTokenWithChars(T, NextLine, tok::verbatim_block_line);
506 T.setVerbatimBlockText(Text);
508 State = LS_VerbatimBlockBody;
511 void Lexer::lexVerbatimBlockBody(Token &T) {
512 assert(State == LS_VerbatimBlockBody);
514 if (CommentState == LCS_InsideCComment)
515 skipLineStartingDecorations();
517 lexVerbatimBlockFirstLine(T);
520 void Lexer::setupAndLexVerbatimLine(Token &T, const char *TextBegin,
521 const CommandInfo *Info) {
522 assert(Info->IsVerbatimLineCommand);
523 formTokenWithChars(T, TextBegin, tok::verbatim_line_name);
524 T.setVerbatimLineID(Info->getID());
526 State = LS_VerbatimLineText;
529 void Lexer::lexVerbatimLineText(Token &T) {
530 assert(State == LS_VerbatimLineText);
532 // Extract current line.
533 const char *Newline = findNewline(BufferPtr, CommentEnd);
534 StringRef Text(BufferPtr, Newline - BufferPtr);
535 formTokenWithChars(T, Newline, tok::verbatim_line_text);
536 T.setVerbatimLineText(Text);
541 void Lexer::lexHTMLCharacterReference(Token &T) {
542 const char *TokenPtr = BufferPtr;
543 assert(*TokenPtr == '&');
545 if (TokenPtr == CommentEnd) {
546 formTextToken(T, TokenPtr);
550 bool isNamed = false;
551 bool isDecimal = false;
553 if (isHTMLNamedCharacterReferenceCharacter(C)) {
555 TokenPtr = skipNamedCharacterReference(TokenPtr, CommentEnd);
557 } else if (C == '#') {
559 if (TokenPtr == CommentEnd) {
560 formTextToken(T, TokenPtr);
564 if (isHTMLDecimalCharacterReferenceCharacter(C)) {
566 TokenPtr = skipDecimalCharacterReference(TokenPtr, CommentEnd);
568 } else if (C == 'x' || C == 'X') {
571 TokenPtr = skipHexCharacterReference(TokenPtr, CommentEnd);
573 formTextToken(T, TokenPtr);
577 formTextToken(T, TokenPtr);
580 if (NamePtr == TokenPtr || TokenPtr == CommentEnd ||
582 formTextToken(T, TokenPtr);
585 StringRef Name(NamePtr, TokenPtr - NamePtr);
586 TokenPtr++; // Skip semicolon.
589 Resolved = resolveHTMLNamedCharacterReference(Name);
591 Resolved = resolveHTMLDecimalCharacterReference(Name);
593 Resolved = resolveHTMLHexCharacterReference(Name);
595 if (Resolved.empty()) {
596 formTextToken(T, TokenPtr);
599 formTokenWithChars(T, TokenPtr, tok::text);
604 void Lexer::setupAndLexHTMLStartTag(Token &T) {
605 assert(BufferPtr[0] == '<' &&
606 isHTMLIdentifierStartingCharacter(BufferPtr[1]));
607 const char *TagNameEnd = skipHTMLIdentifier(BufferPtr + 2, CommentEnd);
608 StringRef Name(BufferPtr + 1, TagNameEnd - (BufferPtr + 1));
609 if (!isHTMLTagName(Name)) {
610 formTextToken(T, TagNameEnd);
614 formTokenWithChars(T, TagNameEnd, tok::html_start_tag);
615 T.setHTMLTagStartName(Name);
617 BufferPtr = skipWhitespace(BufferPtr, CommentEnd);
619 const char C = *BufferPtr;
620 if (BufferPtr != CommentEnd &&
621 (C == '>' || C == '/' || isHTMLIdentifierStartingCharacter(C)))
622 State = LS_HTMLStartTag;
625 void Lexer::lexHTMLStartTag(Token &T) {
626 assert(State == LS_HTMLStartTag);
628 const char *TokenPtr = BufferPtr;
630 if (isHTMLIdentifierCharacter(C)) {
631 TokenPtr = skipHTMLIdentifier(TokenPtr, CommentEnd);
632 StringRef Ident(BufferPtr, TokenPtr - BufferPtr);
633 formTokenWithChars(T, TokenPtr, tok::html_ident);
634 T.setHTMLIdent(Ident);
639 formTokenWithChars(T, TokenPtr, tok::html_equals);
643 const char *OpenQuote = TokenPtr;
644 TokenPtr = skipHTMLQuotedString(TokenPtr, CommentEnd);
645 const char *ClosingQuote = TokenPtr;
646 if (TokenPtr != CommentEnd) // Skip closing quote.
648 formTokenWithChars(T, TokenPtr, tok::html_quoted_string);
649 T.setHTMLQuotedString(StringRef(OpenQuote + 1,
650 ClosingQuote - (OpenQuote + 1)));
655 formTokenWithChars(T, TokenPtr, tok::html_greater);
660 if (TokenPtr != CommentEnd && *TokenPtr == '>') {
662 formTokenWithChars(T, TokenPtr, tok::html_slash_greater);
664 formTextToken(T, TokenPtr);
671 // Now look ahead and return to normal state if we don't see any HTML tokens
673 BufferPtr = skipWhitespace(BufferPtr, CommentEnd);
674 if (BufferPtr == CommentEnd) {
680 if (!isHTMLIdentifierStartingCharacter(C) &&
681 C != '=' && C != '\"' && C != '\'' && C != '>') {
687 void Lexer::setupAndLexHTMLEndTag(Token &T) {
688 assert(BufferPtr[0] == '<' && BufferPtr[1] == '/');
690 const char *TagNameBegin = skipWhitespace(BufferPtr + 2, CommentEnd);
691 const char *TagNameEnd = skipHTMLIdentifier(TagNameBegin, CommentEnd);
692 StringRef Name(TagNameBegin, TagNameEnd - TagNameBegin);
693 if (!isHTMLTagName(Name)) {
694 formTextToken(T, TagNameEnd);
698 const char *End = skipWhitespace(TagNameEnd, CommentEnd);
700 formTokenWithChars(T, End, tok::html_end_tag);
701 T.setHTMLTagEndName(Name);
703 if (BufferPtr != CommentEnd && *BufferPtr == '>')
704 State = LS_HTMLEndTag;
707 void Lexer::lexHTMLEndTag(Token &T) {
708 assert(BufferPtr != CommentEnd && *BufferPtr == '>');
710 formTokenWithChars(T, BufferPtr + 1, tok::html_greater);
714 Lexer::Lexer(llvm::BumpPtrAllocator &Allocator, DiagnosticsEngine &Diags,
715 const CommandTraits &Traits,
716 SourceLocation FileLoc,
717 const char *BufferStart, const char *BufferEnd):
718 Allocator(Allocator), Diags(Diags), Traits(Traits),
719 BufferStart(BufferStart), BufferEnd(BufferEnd),
720 FileLoc(FileLoc), BufferPtr(BufferStart),
721 CommentState(LCS_BeforeComment), State(LS_Normal) {
724 void Lexer::lex(Token &T) {
726 switch (CommentState) {
727 case LCS_BeforeComment:
728 if (BufferPtr == BufferEnd) {
729 formTokenWithChars(T, BufferPtr, tok::eof);
733 assert(*BufferPtr == '/');
734 BufferPtr++; // Skip first slash.
736 case '/': { // BCPL comment.
737 BufferPtr++; // Skip second slash.
739 if (BufferPtr != BufferEnd) {
740 // Skip Doxygen magic marker, if it is present.
741 // It might be missing because of a typo //< or /*<, or because we
742 // merged this non-Doxygen comment into a bunch of Doxygen comments
743 // around it: /** ... */ /* ... */ /** ... */
744 const char C = *BufferPtr;
745 if (C == '/' || C == '!')
749 // Skip less-than symbol that marks trailing comments.
750 // Skip it even if the comment is not a Doxygen one, because //< and /*<
751 // are frequent typos.
752 if (BufferPtr != BufferEnd && *BufferPtr == '<')
755 CommentState = LCS_InsideBCPLComment;
756 if (State != LS_VerbatimBlockBody && State != LS_VerbatimBlockFirstLine)
758 CommentEnd = findBCPLCommentEnd(BufferPtr, BufferEnd);
761 case '*': { // C comment.
762 BufferPtr++; // Skip star.
764 // Skip Doxygen magic marker.
765 const char C = *BufferPtr;
766 if ((C == '*' && *(BufferPtr + 1) != '/') || C == '!')
769 // Skip less-than symbol that marks trailing comments.
770 if (BufferPtr != BufferEnd && *BufferPtr == '<')
773 CommentState = LCS_InsideCComment;
775 CommentEnd = findCCommentEnd(BufferPtr, BufferEnd);
779 llvm_unreachable("second character of comment should be '/' or '*'");
782 case LCS_BetweenComments: {
783 // Consecutive comments are extracted only if there is only whitespace
784 // between them. So we can search for the start of the next comment.
785 const char *EndWhitespace = BufferPtr;
786 while(EndWhitespace != BufferEnd && *EndWhitespace != '/')
789 // Turn any whitespace between comments (and there is only whitespace
790 // between them -- guaranteed by comment extraction) into a newline. We
791 // have two newlines between C comments in total (first one was synthesized
793 formTokenWithChars(T, EndWhitespace, tok::newline);
795 CommentState = LCS_BeforeComment;
799 case LCS_InsideBCPLComment:
800 case LCS_InsideCComment:
801 if (BufferPtr != CommentEnd) {
805 // Skip C comment closing sequence.
806 if (CommentState == LCS_InsideCComment) {
807 assert(BufferPtr[0] == '*' && BufferPtr[1] == '/');
809 assert(BufferPtr <= BufferEnd);
811 // Synthenize newline just after the C comment, regardless if there is
812 // actually a newline.
813 formTokenWithChars(T, BufferPtr, tok::newline);
815 CommentState = LCS_BetweenComments;
818 // Don't synthesized a newline after BCPL comment.
819 CommentState = LCS_BetweenComments;
826 StringRef Lexer::getSpelling(const Token &Tok,
827 const SourceManager &SourceMgr,
828 bool *Invalid) const {
829 SourceLocation Loc = Tok.getLocation();
830 std::pair<FileID, unsigned> LocInfo = SourceMgr.getDecomposedLoc(Loc);
832 bool InvalidTemp = false;
833 StringRef File = SourceMgr.getBufferData(LocInfo.first, &InvalidTemp);
839 const char *Begin = File.data() + LocInfo.second;
840 return StringRef(Begin, Tok.getLength());
843 } // end namespace comments
844 } // end namespace clang