1 //===--- CommentLexer.cpp -------------------------------------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 
10 #include "clang/AST/CommentLexer.h"
11 #include "clang/AST/CommentCommandTraits.h"
12 #include "clang/AST/CommentDiagnostic.h"
13 #include "clang/Basic/CharInfo.h"
14 #include "llvm/ADT/StringExtras.h"
15 #include "llvm/ADT/StringSwitch.h"
16 #include "llvm/Support/ConvertUTF.h"
17 #include "llvm/Support/ErrorHandling.h"
18 
19 namespace clang {
20 namespace comments {
21 
22 void Token::dump(const Lexer &L, const SourceManager &SM) const {
23   llvm::errs() << "comments::Token Kind=" << Kind << " ";
24   Loc.dump(SM);
25   llvm::errs() << " " << Length << " \"" << L.getSpelling(*this, SM) << "\"\n";
26 }
27 
28 static inline bool isHTMLNamedCharacterReferenceCharacter(char C) {
29   return isLetter(C);
30 }
31 
32 static inline bool isHTMLDecimalCharacterReferenceCharacter(char C) {
33   return isDigit(C);
34 }
35 
36 static inline bool isHTMLHexCharacterReferenceCharacter(char C) {
37   return isHexDigit(C);
38 }
39 
40 static inline StringRef convertCodePointToUTF8(
41                                       llvm::BumpPtrAllocator &Allocator,
42                                       unsigned CodePoint) {
43   char *Resolved = Allocator.Allocate<char>(UNI_MAX_UTF8_BYTES_PER_CODE_POINT);
44   char *ResolvedPtr = Resolved;
45   if (llvm::ConvertCodePointToUTF8(CodePoint, ResolvedPtr))
46     return StringRef(Resolved, ResolvedPtr - Resolved);
47   else
48     return StringRef();
49 }
50 
51 namespace {
52 
53 #include "clang/AST/CommentHTMLTags.inc"
54 #include "clang/AST/CommentHTMLNamedCharacterReferences.inc"
55 
56 } // end anonymous namespace
57 
58 StringRef Lexer::resolveHTMLNamedCharacterReference(StringRef Name) const {
59   // Fast path, first check a few most widely used named character references.
60   return llvm::StringSwitch<StringRef>(Name)
61       .Case("amp", "&")
62       .Case("lt", "<")
63       .Case("gt", ">")
64       .Case("quot", "\"")
65       .Case("apos", "\'")
66       // Slow path.
67       .Default(translateHTMLNamedCharacterReferenceToUTF8(Name));
68 }
69 
70 StringRef Lexer::resolveHTMLDecimalCharacterReference(StringRef Name) const {
71   unsigned CodePoint = 0;
72   for (unsigned i = 0, e = Name.size(); i != e; ++i) {
73     assert(isHTMLDecimalCharacterReferenceCharacter(Name[i]));
74     CodePoint *= 10;
75     CodePoint += Name[i] - '0';
76   }
77   return convertCodePointToUTF8(Allocator, CodePoint);
78 }
79 
80 StringRef Lexer::resolveHTMLHexCharacterReference(StringRef Name) const {
81   unsigned CodePoint = 0;
82   for (unsigned i = 0, e = Name.size(); i != e; ++i) {
83     CodePoint *= 16;
84     const char C = Name[i];
85     assert(isHTMLHexCharacterReferenceCharacter(C));
86     CodePoint += llvm::hexDigitValue(C);
87   }
88   return convertCodePointToUTF8(Allocator, CodePoint);
89 }
90 
91 void Lexer::skipLineStartingDecorations() {
92   // This function should be called only for C comments
93   assert(CommentState == LCS_InsideCComment);
94 
95   if (BufferPtr == CommentEnd)
96     return;
97 
98   switch (*BufferPtr) {
99   case ' ':
100   case '\t':
101   case '\f':
102   case '\v': {
103     const char *NewBufferPtr = BufferPtr;
104     NewBufferPtr++;
105     if (NewBufferPtr == CommentEnd)
106       return;
107 
108     char C = *NewBufferPtr;
109     while (isHorizontalWhitespace(C)) {
110       NewBufferPtr++;
111       if (NewBufferPtr == CommentEnd)
112         return;
113       C = *NewBufferPtr;
114     }
115     if (C == '*')
116       BufferPtr = NewBufferPtr + 1;
117     break;
118   }
119   case '*':
120     BufferPtr++;
121     break;
122   }
123 }
124 
125 namespace {
126 /// Returns pointer to the first newline character in the string.
127 const char *findNewline(const char *BufferPtr, const char *BufferEnd) {
128   for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
129     if (isVerticalWhitespace(*BufferPtr))
130       return BufferPtr;
131   }
132   return BufferEnd;
133 }
134 
135 const char *skipNewline(const char *BufferPtr, const char *BufferEnd) {
136   if (BufferPtr == BufferEnd)
137     return BufferPtr;
138 
139   if (*BufferPtr == '\n')
140     BufferPtr++;
141   else {
142     assert(*BufferPtr == '\r');
143     BufferPtr++;
144     if (BufferPtr != BufferEnd && *BufferPtr == '\n')
145       BufferPtr++;
146   }
147   return BufferPtr;
148 }
149 
150 const char *skipNamedCharacterReference(const char *BufferPtr,
151                                         const char *BufferEnd) {
152   for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
153     if (!isHTMLNamedCharacterReferenceCharacter(*BufferPtr))
154       return BufferPtr;
155   }
156   return BufferEnd;
157 }
158 
159 const char *skipDecimalCharacterReference(const char *BufferPtr,
160                                           const char *BufferEnd) {
161   for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
162     if (!isHTMLDecimalCharacterReferenceCharacter(*BufferPtr))
163       return BufferPtr;
164   }
165   return BufferEnd;
166 }
167 
168 const char *skipHexCharacterReference(const char *BufferPtr,
169                                       const char *BufferEnd) {
170   for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
171     if (!isHTMLHexCharacterReferenceCharacter(*BufferPtr))
172       return BufferPtr;
173   }
174   return BufferEnd;
175 }
176 
177 bool isHTMLIdentifierStartingCharacter(char C) {
178   return isLetter(C);
179 }
180 
181 bool isHTMLIdentifierCharacter(char C) {
182   return isAlphanumeric(C);
183 }
184 
185 const char *skipHTMLIdentifier(const char *BufferPtr, const char *BufferEnd) {
186   for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
187     if (!isHTMLIdentifierCharacter(*BufferPtr))
188       return BufferPtr;
189   }
190   return BufferEnd;
191 }
192 
193 /// Skip HTML string quoted in single or double quotes.  Escaping quotes inside
194 /// string allowed.
195 ///
196 /// Returns pointer to closing quote.
197 const char *skipHTMLQuotedString(const char *BufferPtr, const char *BufferEnd)
198 {
199   const char Quote = *BufferPtr;
200   assert(Quote == '\"' || Quote == '\'');
201 
202   BufferPtr++;
203   for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
204     const char C = *BufferPtr;
205     if (C == Quote && BufferPtr[-1] != '\\')
206       return BufferPtr;
207   }
208   return BufferEnd;
209 }
210 
211 const char *skipWhitespace(const char *BufferPtr, const char *BufferEnd) {
212   for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
213     if (!isWhitespace(*BufferPtr))
214       return BufferPtr;
215   }
216   return BufferEnd;
217 }
218 
219 bool isWhitespace(const char *BufferPtr, const char *BufferEnd) {
220   return skipWhitespace(BufferPtr, BufferEnd) == BufferEnd;
221 }
222 
223 bool isCommandNameStartCharacter(char C) {
224   return isLetter(C);
225 }
226 
227 bool isCommandNameCharacter(char C) {
228   return isAlphanumeric(C);
229 }
230 
231 const char *skipCommandName(const char *BufferPtr, const char *BufferEnd) {
232   for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
233     if (!isCommandNameCharacter(*BufferPtr))
234       return BufferPtr;
235   }
236   return BufferEnd;
237 }
238 
239 /// Return the one past end pointer for BCPL comments.
240 /// Handles newlines escaped with backslash or trigraph for backslahs.
241 const char *findBCPLCommentEnd(const char *BufferPtr, const char *BufferEnd) {
242   const char *CurPtr = BufferPtr;
243   while (CurPtr != BufferEnd) {
244     while (!isVerticalWhitespace(*CurPtr)) {
245       CurPtr++;
246       if (CurPtr == BufferEnd)
247         return BufferEnd;
248     }
249     // We found a newline, check if it is escaped.
250     const char *EscapePtr = CurPtr - 1;
251     while(isHorizontalWhitespace(*EscapePtr))
252       EscapePtr--;
253 
254     if (*EscapePtr == '\\' ||
255         (EscapePtr - 2 >= BufferPtr && EscapePtr[0] == '/' &&
256          EscapePtr[-1] == '?' && EscapePtr[-2] == '?')) {
257       // We found an escaped newline.
258       CurPtr = skipNewline(CurPtr, BufferEnd);
259     } else
260       return CurPtr; // Not an escaped newline.
261   }
262   return BufferEnd;
263 }
264 
265 /// Return the one past end pointer for C comments.
266 /// Very dumb, does not handle escaped newlines or trigraphs.
267 const char *findCCommentEnd(const char *BufferPtr, const char *BufferEnd) {
268   for ( ; BufferPtr != BufferEnd; ++BufferPtr) {
269     if (*BufferPtr == '*') {
270       assert(BufferPtr + 1 != BufferEnd);
271       if (*(BufferPtr + 1) == '/')
272         return BufferPtr;
273     }
274   }
275   llvm_unreachable("buffer end hit before '*/' was seen");
276 }
277 
278 } // end anonymous namespace
279 
280 void Lexer::formTokenWithChars(Token &Result, const char *TokEnd,
281                                tok::TokenKind Kind) {
282   const unsigned TokLen = TokEnd - BufferPtr;
283   Result.setLocation(getSourceLocation(BufferPtr));
284   Result.setKind(Kind);
285   Result.setLength(TokLen);
286 #ifndef NDEBUG
287   Result.TextPtr = "<UNSET>";
288   Result.IntVal = 7;
289 #endif
290   BufferPtr = TokEnd;
291 }
292 
293 void Lexer::lexCommentText(Token &T) {
294   assert(CommentState == LCS_InsideBCPLComment ||
295          CommentState == LCS_InsideCComment);
296 
297   switch (State) {
298   case LS_Normal:
299     break;
300   case LS_VerbatimBlockFirstLine:
301     lexVerbatimBlockFirstLine(T);
302     return;
303   case LS_VerbatimBlockBody:
304     lexVerbatimBlockBody(T);
305     return;
306   case LS_VerbatimLineText:
307     lexVerbatimLineText(T);
308     return;
309   case LS_HTMLStartTag:
310     lexHTMLStartTag(T);
311     return;
312   case LS_HTMLEndTag:
313     lexHTMLEndTag(T);
314     return;
315   }
316 
317   assert(State == LS_Normal);
318 
319   const char *TokenPtr = BufferPtr;
320   assert(TokenPtr < CommentEnd);
321   while (TokenPtr != CommentEnd) {
322     switch(*TokenPtr) {
323       case '\\':
324       case '@': {
325         // Commands that start with a backslash and commands that start with
326         // 'at' have equivalent semantics.  But we keep information about the
327         // exact syntax in AST for comments.
328         tok::TokenKind CommandKind =
329             (*TokenPtr == '@') ? tok::at_command : tok::backslash_command;
330         TokenPtr++;
331         if (TokenPtr == CommentEnd) {
332           formTextToken(T, TokenPtr);
333           return;
334         }
335         char C = *TokenPtr;
336         switch (C) {
337         default:
338           break;
339 
340         case '\\': case '@': case '&': case '$':
341         case '#':  case '<': case '>': case '%':
342         case '\"': case '.': case ':':
343           // This is one of \\ \@ \& \$ etc escape sequences.
344           TokenPtr++;
345           if (C == ':' && TokenPtr != CommentEnd && *TokenPtr == ':') {
346             // This is the \:: escape sequence.
347             TokenPtr++;
348           }
349           StringRef UnescapedText(BufferPtr + 1, TokenPtr - (BufferPtr + 1));
350           formTokenWithChars(T, TokenPtr, tok::text);
351           T.setText(UnescapedText);
352           return;
353         }
354 
355         // Don't make zero-length commands.
356         if (!isCommandNameStartCharacter(*TokenPtr)) {
357           formTextToken(T, TokenPtr);
358           return;
359         }
360 
361         TokenPtr = skipCommandName(TokenPtr, CommentEnd);
362         unsigned Length = TokenPtr - (BufferPtr + 1);
363 
364         // Hardcoded support for lexing LaTeX formula commands
365         // \f$ \f[ \f] \f{ \f} as a single command.
366         if (Length == 1 && TokenPtr[-1] == 'f' && TokenPtr != CommentEnd) {
367           C = *TokenPtr;
368           if (C == '$' || C == '[' || C == ']' || C == '{' || C == '}') {
369             TokenPtr++;
370             Length++;
371           }
372         }
373 
374         StringRef CommandName(BufferPtr + 1, Length);
375 
376         const CommandInfo *Info = Traits.getCommandInfoOrNULL(CommandName);
377         if (!Info) {
378           if ((Info = Traits.getTypoCorrectCommandInfo(CommandName))) {
379             StringRef CorrectedName = Info->Name;
380             SourceLocation Loc = getSourceLocation(BufferPtr);
381             SourceLocation EndLoc = getSourceLocation(TokenPtr);
382             SourceRange FullRange = SourceRange(Loc, EndLoc);
383             SourceRange CommandRange(Loc.getLocWithOffset(1), EndLoc);
384             Diag(Loc, diag::warn_correct_comment_command_name)
385               << FullRange << CommandName << CorrectedName
386               << FixItHint::CreateReplacement(CommandRange, CorrectedName);
387           } else {
388             formTokenWithChars(T, TokenPtr, tok::unknown_command);
389             T.setUnknownCommandName(CommandName);
390             Diag(T.getLocation(), diag::warn_unknown_comment_command_name)
391                 << SourceRange(T.getLocation(), T.getEndLocation());
392             return;
393           }
394         }
395         if (Info->IsVerbatimBlockCommand) {
396           setupAndLexVerbatimBlock(T, TokenPtr, *BufferPtr, Info);
397           return;
398         }
399         if (Info->IsVerbatimLineCommand) {
400           setupAndLexVerbatimLine(T, TokenPtr, Info);
401           return;
402         }
403         formTokenWithChars(T, TokenPtr, CommandKind);
404         T.setCommandID(Info->getID());
405         return;
406       }
407 
408       case '&':
409         lexHTMLCharacterReference(T);
410         return;
411 
412       case '<': {
413         TokenPtr++;
414         if (TokenPtr == CommentEnd) {
415           formTextToken(T, TokenPtr);
416           return;
417         }
418         const char C = *TokenPtr;
419         if (isHTMLIdentifierStartingCharacter(C))
420           setupAndLexHTMLStartTag(T);
421         else if (C == '/')
422           setupAndLexHTMLEndTag(T);
423         else
424           formTextToken(T, TokenPtr);
425         return;
426       }
427 
428       case '\n':
429       case '\r':
430         TokenPtr = skipNewline(TokenPtr, CommentEnd);
431         formTokenWithChars(T, TokenPtr, tok::newline);
432 
433         if (CommentState == LCS_InsideCComment)
434           skipLineStartingDecorations();
435         return;
436 
437       default: {
438         size_t End = StringRef(TokenPtr, CommentEnd - TokenPtr).
439                          find_first_of("\n\r\\@&<");
440         if (End != StringRef::npos)
441           TokenPtr += End;
442         else
443           TokenPtr = CommentEnd;
444         formTextToken(T, TokenPtr);
445         return;
446       }
447     }
448   }
449 }
450 
451 void Lexer::setupAndLexVerbatimBlock(Token &T,
452                                      const char *TextBegin,
453                                      char Marker, const CommandInfo *Info) {
454   assert(Info->IsVerbatimBlockCommand);
455 
456   VerbatimBlockEndCommandName.clear();
457   VerbatimBlockEndCommandName.append(Marker == '\\' ? "\\" : "@");
458   VerbatimBlockEndCommandName.append(Info->EndCommandName);
459 
460   formTokenWithChars(T, TextBegin, tok::verbatim_block_begin);
461   T.setVerbatimBlockID(Info->getID());
462 
463   // If there is a newline following the verbatim opening command, skip the
464   // newline so that we don't create an tok::verbatim_block_line with empty
465   // text content.
466   if (BufferPtr != CommentEnd &&
467       isVerticalWhitespace(*BufferPtr)) {
468     BufferPtr = skipNewline(BufferPtr, CommentEnd);
469     State = LS_VerbatimBlockBody;
470     return;
471   }
472 
473   State = LS_VerbatimBlockFirstLine;
474 }
475 
476 void Lexer::lexVerbatimBlockFirstLine(Token &T) {
477 again:
478   assert(BufferPtr < CommentEnd);
479 
480   // FIXME: It would be better to scan the text once, finding either the block
481   // end command or newline.
482   //
483   // Extract current line.
484   const char *Newline = findNewline(BufferPtr, CommentEnd);
485   StringRef Line(BufferPtr, Newline - BufferPtr);
486 
487   // Look for end command in current line.
488   size_t Pos = Line.find(VerbatimBlockEndCommandName);
489   const char *TextEnd;
490   const char *NextLine;
491   if (Pos == StringRef::npos) {
492     // Current line is completely verbatim.
493     TextEnd = Newline;
494     NextLine = skipNewline(Newline, CommentEnd);
495   } else if (Pos == 0) {
496     // Current line contains just an end command.
497     const char *End = BufferPtr + VerbatimBlockEndCommandName.size();
498     StringRef Name(BufferPtr + 1, End - (BufferPtr + 1));
499     formTokenWithChars(T, End, tok::verbatim_block_end);
500     T.setVerbatimBlockID(Traits.getCommandInfo(Name)->getID());
501     State = LS_Normal;
502     return;
503   } else {
504     // There is some text, followed by end command.  Extract text first.
505     TextEnd = BufferPtr + Pos;
506     NextLine = TextEnd;
507     // If there is only whitespace before end command, skip whitespace.
508     if (isWhitespace(BufferPtr, TextEnd)) {
509       BufferPtr = TextEnd;
510       goto again;
511     }
512   }
513 
514   StringRef Text(BufferPtr, TextEnd - BufferPtr);
515   formTokenWithChars(T, NextLine, tok::verbatim_block_line);
516   T.setVerbatimBlockText(Text);
517 
518   State = LS_VerbatimBlockBody;
519 }
520 
521 void Lexer::lexVerbatimBlockBody(Token &T) {
522   assert(State == LS_VerbatimBlockBody);
523 
524   if (CommentState == LCS_InsideCComment)
525     skipLineStartingDecorations();
526 
527   if (BufferPtr == CommentEnd) {
528     formTokenWithChars(T, BufferPtr, tok::verbatim_block_line);
529     T.setVerbatimBlockText("");
530     return;
531   }
532 
533   lexVerbatimBlockFirstLine(T);
534 }
535 
536 void Lexer::setupAndLexVerbatimLine(Token &T, const char *TextBegin,
537                                     const CommandInfo *Info) {
538   assert(Info->IsVerbatimLineCommand);
539   formTokenWithChars(T, TextBegin, tok::verbatim_line_name);
540   T.setVerbatimLineID(Info->getID());
541 
542   State = LS_VerbatimLineText;
543 }
544 
545 void Lexer::lexVerbatimLineText(Token &T) {
546   assert(State == LS_VerbatimLineText);
547 
548   // Extract current line.
549   const char *Newline = findNewline(BufferPtr, CommentEnd);
550   StringRef Text(BufferPtr, Newline - BufferPtr);
551   formTokenWithChars(T, Newline, tok::verbatim_line_text);
552   T.setVerbatimLineText(Text);
553 
554   State = LS_Normal;
555 }
556 
557 void Lexer::lexHTMLCharacterReference(Token &T) {
558   const char *TokenPtr = BufferPtr;
559   assert(*TokenPtr == '&');
560   TokenPtr++;
561   if (TokenPtr == CommentEnd) {
562     formTextToken(T, TokenPtr);
563     return;
564   }
565   const char *NamePtr;
566   bool isNamed = false;
567   bool isDecimal = false;
568   char C = *TokenPtr;
569   if (isHTMLNamedCharacterReferenceCharacter(C)) {
570     NamePtr = TokenPtr;
571     TokenPtr = skipNamedCharacterReference(TokenPtr, CommentEnd);
572     isNamed = true;
573   } else if (C == '#') {
574     TokenPtr++;
575     if (TokenPtr == CommentEnd) {
576       formTextToken(T, TokenPtr);
577       return;
578     }
579     C = *TokenPtr;
580     if (isHTMLDecimalCharacterReferenceCharacter(C)) {
581       NamePtr = TokenPtr;
582       TokenPtr = skipDecimalCharacterReference(TokenPtr, CommentEnd);
583       isDecimal = true;
584     } else if (C == 'x' || C == 'X') {
585       TokenPtr++;
586       NamePtr = TokenPtr;
587       TokenPtr = skipHexCharacterReference(TokenPtr, CommentEnd);
588     } else {
589       formTextToken(T, TokenPtr);
590       return;
591     }
592   } else {
593     formTextToken(T, TokenPtr);
594     return;
595   }
596   if (NamePtr == TokenPtr || TokenPtr == CommentEnd ||
597       *TokenPtr != ';') {
598     formTextToken(T, TokenPtr);
599     return;
600   }
601   StringRef Name(NamePtr, TokenPtr - NamePtr);
602   TokenPtr++; // Skip semicolon.
603   StringRef Resolved;
604   if (isNamed)
605     Resolved = resolveHTMLNamedCharacterReference(Name);
606   else if (isDecimal)
607     Resolved = resolveHTMLDecimalCharacterReference(Name);
608   else
609     Resolved = resolveHTMLHexCharacterReference(Name);
610 
611   if (Resolved.empty()) {
612     formTextToken(T, TokenPtr);
613     return;
614   }
615   formTokenWithChars(T, TokenPtr, tok::text);
616   T.setText(Resolved);
617 }
618 
619 void Lexer::setupAndLexHTMLStartTag(Token &T) {
620   assert(BufferPtr[0] == '<' &&
621          isHTMLIdentifierStartingCharacter(BufferPtr[1]));
622   const char *TagNameEnd = skipHTMLIdentifier(BufferPtr + 2, CommentEnd);
623   StringRef Name(BufferPtr + 1, TagNameEnd - (BufferPtr + 1));
624   if (!isHTMLTagName(Name)) {
625     formTextToken(T, TagNameEnd);
626     return;
627   }
628 
629   formTokenWithChars(T, TagNameEnd, tok::html_start_tag);
630   T.setHTMLTagStartName(Name);
631 
632   BufferPtr = skipWhitespace(BufferPtr, CommentEnd);
633 
634   const char C = *BufferPtr;
635   if (BufferPtr != CommentEnd &&
636       (C == '>' || C == '/' || isHTMLIdentifierStartingCharacter(C)))
637     State = LS_HTMLStartTag;
638 }
639 
640 void Lexer::lexHTMLStartTag(Token &T) {
641   assert(State == LS_HTMLStartTag);
642 
643   const char *TokenPtr = BufferPtr;
644   char C = *TokenPtr;
645   if (isHTMLIdentifierCharacter(C)) {
646     TokenPtr = skipHTMLIdentifier(TokenPtr, CommentEnd);
647     StringRef Ident(BufferPtr, TokenPtr - BufferPtr);
648     formTokenWithChars(T, TokenPtr, tok::html_ident);
649     T.setHTMLIdent(Ident);
650   } else {
651     switch (C) {
652     case '=':
653       TokenPtr++;
654       formTokenWithChars(T, TokenPtr, tok::html_equals);
655       break;
656     case '\"':
657     case '\'': {
658       const char *OpenQuote = TokenPtr;
659       TokenPtr = skipHTMLQuotedString(TokenPtr, CommentEnd);
660       const char *ClosingQuote = TokenPtr;
661       if (TokenPtr != CommentEnd) // Skip closing quote.
662         TokenPtr++;
663       formTokenWithChars(T, TokenPtr, tok::html_quoted_string);
664       T.setHTMLQuotedString(StringRef(OpenQuote + 1,
665                                       ClosingQuote - (OpenQuote + 1)));
666       break;
667     }
668     case '>':
669       TokenPtr++;
670       formTokenWithChars(T, TokenPtr, tok::html_greater);
671       State = LS_Normal;
672       return;
673     case '/':
674       TokenPtr++;
675       if (TokenPtr != CommentEnd && *TokenPtr == '>') {
676         TokenPtr++;
677         formTokenWithChars(T, TokenPtr, tok::html_slash_greater);
678       } else
679         formTextToken(T, TokenPtr);
680 
681       State = LS_Normal;
682       return;
683     }
684   }
685 
686   // Now look ahead and return to normal state if we don't see any HTML tokens
687   // ahead.
688   BufferPtr = skipWhitespace(BufferPtr, CommentEnd);
689   if (BufferPtr == CommentEnd) {
690     State = LS_Normal;
691     return;
692   }
693 
694   C = *BufferPtr;
695   if (!isHTMLIdentifierStartingCharacter(C) &&
696       C != '=' && C != '\"' && C != '\'' && C != '>') {
697     State = LS_Normal;
698     return;
699   }
700 }
701 
702 void Lexer::setupAndLexHTMLEndTag(Token &T) {
703   assert(BufferPtr[0] == '<' && BufferPtr[1] == '/');
704 
705   const char *TagNameBegin = skipWhitespace(BufferPtr + 2, CommentEnd);
706   const char *TagNameEnd = skipHTMLIdentifier(TagNameBegin, CommentEnd);
707   StringRef Name(TagNameBegin, TagNameEnd - TagNameBegin);
708   if (!isHTMLTagName(Name)) {
709     formTextToken(T, TagNameEnd);
710     return;
711   }
712 
713   const char *End = skipWhitespace(TagNameEnd, CommentEnd);
714 
715   formTokenWithChars(T, End, tok::html_end_tag);
716   T.setHTMLTagEndName(Name);
717 
718   if (BufferPtr != CommentEnd && *BufferPtr == '>')
719     State = LS_HTMLEndTag;
720 }
721 
722 void Lexer::lexHTMLEndTag(Token &T) {
723   assert(BufferPtr != CommentEnd && *BufferPtr == '>');
724 
725   formTokenWithChars(T, BufferPtr + 1, tok::html_greater);
726   State = LS_Normal;
727 }
728 
729 Lexer::Lexer(llvm::BumpPtrAllocator &Allocator, DiagnosticsEngine &Diags,
730              const CommandTraits &Traits,
731              SourceLocation FileLoc,
732              const char *BufferStart, const char *BufferEnd):
733     Allocator(Allocator), Diags(Diags), Traits(Traits),
734     BufferStart(BufferStart), BufferEnd(BufferEnd),
735     FileLoc(FileLoc), BufferPtr(BufferStart),
736     CommentState(LCS_BeforeComment), State(LS_Normal) {
737 }
738 
739 void Lexer::lex(Token &T) {
740 again:
741   switch (CommentState) {
742   case LCS_BeforeComment:
743     if (BufferPtr == BufferEnd) {
744       formTokenWithChars(T, BufferPtr, tok::eof);
745       return;
746     }
747 
748     assert(*BufferPtr == '/');
749     BufferPtr++; // Skip first slash.
750     switch(*BufferPtr) {
751     case '/': { // BCPL comment.
752       BufferPtr++; // Skip second slash.
753 
754       if (BufferPtr != BufferEnd) {
755         // Skip Doxygen magic marker, if it is present.
756         // It might be missing because of a typo //< or /*<, or because we
757         // merged this non-Doxygen comment into a bunch of Doxygen comments
758         // around it: /** ... */ /* ... */ /** ... */
759         const char C = *BufferPtr;
760         if (C == '/' || C == '!')
761           BufferPtr++;
762       }
763 
764       // Skip less-than symbol that marks trailing comments.
765       // Skip it even if the comment is not a Doxygen one, because //< and /*<
766       // are frequent typos.
767       if (BufferPtr != BufferEnd && *BufferPtr == '<')
768         BufferPtr++;
769 
770       CommentState = LCS_InsideBCPLComment;
771       if (State != LS_VerbatimBlockBody && State != LS_VerbatimBlockFirstLine)
772         State = LS_Normal;
773       CommentEnd = findBCPLCommentEnd(BufferPtr, BufferEnd);
774       goto again;
775     }
776     case '*': { // C comment.
777       BufferPtr++; // Skip star.
778 
779       // Skip Doxygen magic marker.
780       const char C = *BufferPtr;
781       if ((C == '*' && *(BufferPtr + 1) != '/') || C == '!')
782         BufferPtr++;
783 
784       // Skip less-than symbol that marks trailing comments.
785       if (BufferPtr != BufferEnd && *BufferPtr == '<')
786         BufferPtr++;
787 
788       CommentState = LCS_InsideCComment;
789       State = LS_Normal;
790       CommentEnd = findCCommentEnd(BufferPtr, BufferEnd);
791       goto again;
792     }
793     default:
794       llvm_unreachable("second character of comment should be '/' or '*'");
795     }
796 
797   case LCS_BetweenComments: {
798     // Consecutive comments are extracted only if there is only whitespace
799     // between them.  So we can search for the start of the next comment.
800     const char *EndWhitespace = BufferPtr;
801     while(EndWhitespace != BufferEnd && *EndWhitespace != '/')
802       EndWhitespace++;
803 
804     // Turn any whitespace between comments (and there is only whitespace
805     // between them -- guaranteed by comment extraction) into a newline.  We
806     // have two newlines between C comments in total (first one was synthesized
807     // after a comment).
808     formTokenWithChars(T, EndWhitespace, tok::newline);
809 
810     CommentState = LCS_BeforeComment;
811     break;
812   }
813 
814   case LCS_InsideBCPLComment:
815   case LCS_InsideCComment:
816     if (BufferPtr != CommentEnd) {
817       lexCommentText(T);
818       break;
819     } else {
820       // Skip C comment closing sequence.
821       if (CommentState == LCS_InsideCComment) {
822         assert(BufferPtr[0] == '*' && BufferPtr[1] == '/');
823         BufferPtr += 2;
824         assert(BufferPtr <= BufferEnd);
825 
826         // Synthenize newline just after the C comment, regardless if there is
827         // actually a newline.
828         formTokenWithChars(T, BufferPtr, tok::newline);
829 
830         CommentState = LCS_BetweenComments;
831         break;
832       } else {
833         // Don't synthesized a newline after BCPL comment.
834         CommentState = LCS_BetweenComments;
835         goto again;
836       }
837     }
838   }
839 }
840 
841 StringRef Lexer::getSpelling(const Token &Tok,
842                              const SourceManager &SourceMgr,
843                              bool *Invalid) const {
844   SourceLocation Loc = Tok.getLocation();
845   std::pair<FileID, unsigned> LocInfo = SourceMgr.getDecomposedLoc(Loc);
846 
847   bool InvalidTemp = false;
848   StringRef File = SourceMgr.getBufferData(LocInfo.first, &InvalidTemp);
849   if (InvalidTemp) {
850     *Invalid = true;
851     return StringRef();
852   }
853 
854   const char *Begin = File.data() + LocInfo.second;
855   return StringRef(Begin, Tok.getLength());
856 }
857 
858 } // end namespace comments
859 } // end namespace clang
860