1 //===--- FormatTokenLexer.cpp - Lex FormatTokens -------------*- C++ ----*-===//
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 /// \file
11 /// \brief This file implements FormatTokenLexer, which tokenizes a source file
12 /// into a FormatToken stream suitable for ClangFormat.
13 ///
14 //===----------------------------------------------------------------------===//
15 
16 #include "FormatTokenLexer.h"
17 #include "FormatToken.h"
18 #include "clang/Basic/SourceLocation.h"
19 #include "clang/Basic/SourceManager.h"
20 #include "clang/Format/Format.h"
21 #include "llvm/Support/Regex.h"
22 
23 namespace clang {
24 namespace format {
25 
26 FormatTokenLexer::FormatTokenLexer(const SourceManager &SourceMgr, FileID ID,
27                                    unsigned Column, const FormatStyle &Style,
28                                    encoding::Encoding Encoding)
29     : FormatTok(nullptr), IsFirstToken(true), StateStack({LexerState::NORMAL}),
30       Column(Column), TrailingWhitespace(0), SourceMgr(SourceMgr), ID(ID),
31       Style(Style), IdentTable(getFormattingLangOpts(Style)),
32       Keywords(IdentTable), Encoding(Encoding), FirstInLineIndex(0),
33       FormattingDisabled(false), MacroBlockBeginRegex(Style.MacroBlockBegin),
34       MacroBlockEndRegex(Style.MacroBlockEnd) {
35   Lex.reset(new Lexer(ID, SourceMgr.getBuffer(ID), SourceMgr,
36                       getFormattingLangOpts(Style)));
37   Lex->SetKeepWhitespaceMode(true);
38 
39   for (const std::string &ForEachMacro : Style.ForEachMacros)
40     ForEachMacros.push_back(&IdentTable.get(ForEachMacro));
41   std::sort(ForEachMacros.begin(), ForEachMacros.end());
42 }
43 
44 ArrayRef<FormatToken *> FormatTokenLexer::lex() {
45   assert(Tokens.empty());
46   assert(FirstInLineIndex == 0);
47   do {
48     Tokens.push_back(getNextToken());
49     if (Style.Language == FormatStyle::LK_JavaScript) {
50       tryParseJSRegexLiteral();
51       handleTemplateStrings();
52     }
53     if (Style.Language == FormatStyle::LK_TextProto)
54       tryParsePythonComment();
55     tryMergePreviousTokens();
56     if (Tokens.back()->NewlinesBefore > 0 || Tokens.back()->IsMultiline)
57       FirstInLineIndex = Tokens.size() - 1;
58   } while (Tokens.back()->Tok.isNot(tok::eof));
59   return Tokens;
60 }
61 
62 void FormatTokenLexer::tryMergePreviousTokens() {
63   if (tryMerge_TMacro())
64     return;
65   if (tryMergeConflictMarkers())
66     return;
67   if (tryMergeLessLess())
68     return;
69   if (tryMergeNSStringLiteral())
70     return;
71 
72   if (Style.Language == FormatStyle::LK_JavaScript) {
73     static const tok::TokenKind JSIdentity[] = {tok::equalequal, tok::equal};
74     static const tok::TokenKind JSNotIdentity[] = {tok::exclaimequal,
75                                                    tok::equal};
76     static const tok::TokenKind JSShiftEqual[] = {tok::greater, tok::greater,
77                                                   tok::greaterequal};
78     static const tok::TokenKind JSRightArrow[] = {tok::equal, tok::greater};
79     static const tok::TokenKind JSExponentiation[] = {tok::star, tok::star};
80     static const tok::TokenKind JSExponentiationEqual[] = {tok::star,
81                                                            tok::starequal};
82 
83     // FIXME: Investigate what token type gives the correct operator priority.
84     if (tryMergeTokens(JSIdentity, TT_BinaryOperator))
85       return;
86     if (tryMergeTokens(JSNotIdentity, TT_BinaryOperator))
87       return;
88     if (tryMergeTokens(JSShiftEqual, TT_BinaryOperator))
89       return;
90     if (tryMergeTokens(JSRightArrow, TT_JsFatArrow))
91       return;
92     if (tryMergeTokens(JSExponentiation, TT_JsExponentiation))
93       return;
94     if (tryMergeTokens(JSExponentiationEqual, TT_JsExponentiationEqual)) {
95       Tokens.back()->Tok.setKind(tok::starequal);
96       return;
97     }
98   }
99 
100   if (Style.Language == FormatStyle::LK_Java) {
101     static const tok::TokenKind JavaRightLogicalShiftAssign[] = {
102         tok::greater, tok::greater, tok::greaterequal};
103     if (tryMergeTokens(JavaRightLogicalShiftAssign, TT_BinaryOperator))
104       return;
105   }
106 }
107 
108 bool FormatTokenLexer::tryMergeNSStringLiteral() {
109   if (Tokens.size() < 2)
110     return false;
111   auto &At = *(Tokens.end() - 2);
112   auto &String = *(Tokens.end() - 1);
113   if (!At->is(tok::at) || !String->is(tok::string_literal))
114     return false;
115   At->Tok.setKind(tok::string_literal);
116   At->TokenText = StringRef(At->TokenText.begin(),
117                             String->TokenText.end() - At->TokenText.begin());
118   At->ColumnWidth += String->ColumnWidth;
119   At->Type = TT_ObjCStringLiteral;
120   Tokens.erase(Tokens.end() - 1);
121   return true;
122 }
123 
124 bool FormatTokenLexer::tryMergeLessLess() {
125   // Merge X,less,less,Y into X,lessless,Y unless X or Y is less.
126   if (Tokens.size() < 3)
127     return false;
128 
129   bool FourthTokenIsLess = false;
130   if (Tokens.size() > 3)
131     FourthTokenIsLess = (Tokens.end() - 4)[0]->is(tok::less);
132 
133   auto First = Tokens.end() - 3;
134   if (First[2]->is(tok::less) || First[1]->isNot(tok::less) ||
135       First[0]->isNot(tok::less) || FourthTokenIsLess)
136     return false;
137 
138   // Only merge if there currently is no whitespace between the two "<".
139   if (First[1]->WhitespaceRange.getBegin() !=
140       First[1]->WhitespaceRange.getEnd())
141     return false;
142 
143   First[0]->Tok.setKind(tok::lessless);
144   First[0]->TokenText = "<<";
145   First[0]->ColumnWidth += 1;
146   Tokens.erase(Tokens.end() - 2);
147   return true;
148 }
149 
150 bool FormatTokenLexer::tryMergeTokens(ArrayRef<tok::TokenKind> Kinds,
151                                       TokenType NewType) {
152   if (Tokens.size() < Kinds.size())
153     return false;
154 
155   SmallVectorImpl<FormatToken *>::const_iterator First =
156       Tokens.end() - Kinds.size();
157   if (!First[0]->is(Kinds[0]))
158     return false;
159   unsigned AddLength = 0;
160   for (unsigned i = 1; i < Kinds.size(); ++i) {
161     if (!First[i]->is(Kinds[i]) || First[i]->WhitespaceRange.getBegin() !=
162                                        First[i]->WhitespaceRange.getEnd())
163       return false;
164     AddLength += First[i]->TokenText.size();
165   }
166   Tokens.resize(Tokens.size() - Kinds.size() + 1);
167   First[0]->TokenText = StringRef(First[0]->TokenText.data(),
168                                   First[0]->TokenText.size() + AddLength);
169   First[0]->ColumnWidth += AddLength;
170   First[0]->Type = NewType;
171   return true;
172 }
173 
174 // Returns \c true if \p Tok can only be followed by an operand in JavaScript.
175 bool FormatTokenLexer::precedesOperand(FormatToken *Tok) {
176   // NB: This is not entirely correct, as an r_paren can introduce an operand
177   // location in e.g. `if (foo) /bar/.exec(...);`. That is a rare enough
178   // corner case to not matter in practice, though.
179   return Tok->isOneOf(tok::period, tok::l_paren, tok::comma, tok::l_brace,
180                       tok::r_brace, tok::l_square, tok::semi, tok::exclaim,
181                       tok::colon, tok::question, tok::tilde) ||
182          Tok->isOneOf(tok::kw_return, tok::kw_do, tok::kw_case, tok::kw_throw,
183                       tok::kw_else, tok::kw_new, tok::kw_delete, tok::kw_void,
184                       tok::kw_typeof, Keywords.kw_instanceof, Keywords.kw_in) ||
185          Tok->isBinaryOperator();
186 }
187 
188 bool FormatTokenLexer::canPrecedeRegexLiteral(FormatToken *Prev) {
189   if (!Prev)
190     return true;
191 
192   // Regex literals can only follow after prefix unary operators, not after
193   // postfix unary operators. If the '++' is followed by a non-operand
194   // introducing token, the slash here is the operand and not the start of a
195   // regex.
196   // `!` is an unary prefix operator, but also a post-fix operator that casts
197   // away nullability, so the same check applies.
198   if (Prev->isOneOf(tok::plusplus, tok::minusminus, tok::exclaim))
199     return (Tokens.size() < 3 || precedesOperand(Tokens[Tokens.size() - 3]));
200 
201   // The previous token must introduce an operand location where regex
202   // literals can occur.
203   if (!precedesOperand(Prev))
204     return false;
205 
206   return true;
207 }
208 
209 // Tries to parse a JavaScript Regex literal starting at the current token,
210 // if that begins with a slash and is in a location where JavaScript allows
211 // regex literals. Changes the current token to a regex literal and updates
212 // its text if successful.
213 void FormatTokenLexer::tryParseJSRegexLiteral() {
214   FormatToken *RegexToken = Tokens.back();
215   if (!RegexToken->isOneOf(tok::slash, tok::slashequal))
216     return;
217 
218   FormatToken *Prev = nullptr;
219   for (auto I = Tokens.rbegin() + 1, E = Tokens.rend(); I != E; ++I) {
220     // NB: Because previous pointers are not initialized yet, this cannot use
221     // Token.getPreviousNonComment.
222     if ((*I)->isNot(tok::comment)) {
223       Prev = *I;
224       break;
225     }
226   }
227 
228   if (!canPrecedeRegexLiteral(Prev))
229     return;
230 
231   // 'Manually' lex ahead in the current file buffer.
232   const char *Offset = Lex->getBufferLocation();
233   const char *RegexBegin = Offset - RegexToken->TokenText.size();
234   StringRef Buffer = Lex->getBuffer();
235   bool InCharacterClass = false;
236   bool HaveClosingSlash = false;
237   for (; !HaveClosingSlash && Offset != Buffer.end(); ++Offset) {
238     // Regular expressions are terminated with a '/', which can only be
239     // escaped using '\' or a character class between '[' and ']'.
240     // See http://www.ecma-international.org/ecma-262/5.1/#sec-7.8.5.
241     switch (*Offset) {
242     case '\\':
243       // Skip the escaped character.
244       ++Offset;
245       break;
246     case '[':
247       InCharacterClass = true;
248       break;
249     case ']':
250       InCharacterClass = false;
251       break;
252     case '/':
253       if (!InCharacterClass)
254         HaveClosingSlash = true;
255       break;
256     }
257   }
258 
259   RegexToken->Type = TT_RegexLiteral;
260   // Treat regex literals like other string_literals.
261   RegexToken->Tok.setKind(tok::string_literal);
262   RegexToken->TokenText = StringRef(RegexBegin, Offset - RegexBegin);
263   RegexToken->ColumnWidth = RegexToken->TokenText.size();
264 
265   resetLexer(SourceMgr.getFileOffset(Lex->getSourceLocation(Offset)));
266 }
267 
268 void FormatTokenLexer::handleTemplateStrings() {
269   FormatToken *BacktickToken = Tokens.back();
270 
271   if (BacktickToken->is(tok::l_brace)) {
272     StateStack.push(LexerState::NORMAL);
273     return;
274   }
275   if (BacktickToken->is(tok::r_brace)) {
276     if (StateStack.size() == 1)
277       return;
278     StateStack.pop();
279     if (StateStack.top() != LexerState::TEMPLATE_STRING)
280       return;
281     // If back in TEMPLATE_STRING, fallthrough and continue parsing the
282   } else if (BacktickToken->is(tok::unknown) &&
283              BacktickToken->TokenText == "`") {
284     StateStack.push(LexerState::TEMPLATE_STRING);
285   } else {
286     return; // Not actually a template
287   }
288 
289   // 'Manually' lex ahead in the current file buffer.
290   const char *Offset = Lex->getBufferLocation();
291   const char *TmplBegin = Offset - BacktickToken->TokenText.size(); // at "`"
292   for (; Offset != Lex->getBuffer().end(); ++Offset) {
293     if (Offset[0] == '`') {
294       StateStack.pop();
295       break;
296     }
297     if (Offset[0] == '\\') {
298       ++Offset; // Skip the escaped character.
299     } else if (Offset + 1 < Lex->getBuffer().end() && Offset[0] == '$' &&
300                Offset[1] == '{') {
301       // '${' introduces an expression interpolation in the template string.
302       StateStack.push(LexerState::NORMAL);
303       ++Offset;
304       break;
305     }
306   }
307 
308   StringRef LiteralText(TmplBegin, Offset - TmplBegin + 1);
309   BacktickToken->Type = TT_TemplateString;
310   BacktickToken->Tok.setKind(tok::string_literal);
311   BacktickToken->TokenText = LiteralText;
312 
313   // Adjust width for potentially multiline string literals.
314   size_t FirstBreak = LiteralText.find('\n');
315   StringRef FirstLineText = FirstBreak == StringRef::npos
316                                 ? LiteralText
317                                 : LiteralText.substr(0, FirstBreak);
318   BacktickToken->ColumnWidth = encoding::columnWidthWithTabs(
319       FirstLineText, BacktickToken->OriginalColumn, Style.TabWidth, Encoding);
320   size_t LastBreak = LiteralText.rfind('\n');
321   if (LastBreak != StringRef::npos) {
322     BacktickToken->IsMultiline = true;
323     unsigned StartColumn = 0; // The template tail spans the entire line.
324     BacktickToken->LastLineColumnWidth = encoding::columnWidthWithTabs(
325         LiteralText.substr(LastBreak + 1, LiteralText.size()), StartColumn,
326         Style.TabWidth, Encoding);
327   }
328 
329   SourceLocation loc = Offset < Lex->getBuffer().end()
330                            ? Lex->getSourceLocation(Offset + 1)
331                            : SourceMgr.getLocForEndOfFile(ID);
332   resetLexer(SourceMgr.getFileOffset(loc));
333 }
334 
335 void FormatTokenLexer::tryParsePythonComment() {
336   FormatToken *HashToken = Tokens.back();
337   if (HashToken->isNot(tok::hash))
338     return;
339   // Turn the remainder of this line into a comment.
340   const char *CommentBegin =
341       Lex->getBufferLocation() - HashToken->TokenText.size(); // at "#"
342   size_t From = CommentBegin - Lex->getBuffer().begin();
343   size_t To = Lex->getBuffer().find_first_of('\n', From);
344   if (To == StringRef::npos)
345     To = Lex->getBuffer().size();
346   size_t Len = To - From;
347   HashToken->Type = TT_LineComment;
348   HashToken->Tok.setKind(tok::comment);
349   HashToken->TokenText = Lex->getBuffer().substr(From, Len);
350   SourceLocation Loc = To < Lex->getBuffer().size()
351                            ? Lex->getSourceLocation(CommentBegin + Len)
352                            : SourceMgr.getLocForEndOfFile(ID);
353   resetLexer(SourceMgr.getFileOffset(Loc));
354 }
355 
356 bool FormatTokenLexer::tryMerge_TMacro() {
357   if (Tokens.size() < 4)
358     return false;
359   FormatToken *Last = Tokens.back();
360   if (!Last->is(tok::r_paren))
361     return false;
362 
363   FormatToken *String = Tokens[Tokens.size() - 2];
364   if (!String->is(tok::string_literal) || String->IsMultiline)
365     return false;
366 
367   if (!Tokens[Tokens.size() - 3]->is(tok::l_paren))
368     return false;
369 
370   FormatToken *Macro = Tokens[Tokens.size() - 4];
371   if (Macro->TokenText != "_T")
372     return false;
373 
374   const char *Start = Macro->TokenText.data();
375   const char *End = Last->TokenText.data() + Last->TokenText.size();
376   String->TokenText = StringRef(Start, End - Start);
377   String->IsFirst = Macro->IsFirst;
378   String->LastNewlineOffset = Macro->LastNewlineOffset;
379   String->WhitespaceRange = Macro->WhitespaceRange;
380   String->OriginalColumn = Macro->OriginalColumn;
381   String->ColumnWidth = encoding::columnWidthWithTabs(
382       String->TokenText, String->OriginalColumn, Style.TabWidth, Encoding);
383   String->NewlinesBefore = Macro->NewlinesBefore;
384   String->HasUnescapedNewline = Macro->HasUnescapedNewline;
385 
386   Tokens.pop_back();
387   Tokens.pop_back();
388   Tokens.pop_back();
389   Tokens.back() = String;
390   return true;
391 }
392 
393 bool FormatTokenLexer::tryMergeConflictMarkers() {
394   if (Tokens.back()->NewlinesBefore == 0 && Tokens.back()->isNot(tok::eof))
395     return false;
396 
397   // Conflict lines look like:
398   // <marker> <text from the vcs>
399   // For example:
400   // >>>>>>> /file/in/file/system at revision 1234
401   //
402   // We merge all tokens in a line that starts with a conflict marker
403   // into a single token with a special token type that the unwrapped line
404   // parser will use to correctly rebuild the underlying code.
405 
406   FileID ID;
407   // Get the position of the first token in the line.
408   unsigned FirstInLineOffset;
409   std::tie(ID, FirstInLineOffset) = SourceMgr.getDecomposedLoc(
410       Tokens[FirstInLineIndex]->getStartOfNonWhitespace());
411   StringRef Buffer = SourceMgr.getBuffer(ID)->getBuffer();
412   // Calculate the offset of the start of the current line.
413   auto LineOffset = Buffer.rfind('\n', FirstInLineOffset);
414   if (LineOffset == StringRef::npos) {
415     LineOffset = 0;
416   } else {
417     ++LineOffset;
418   }
419 
420   auto FirstSpace = Buffer.find_first_of(" \n", LineOffset);
421   StringRef LineStart;
422   if (FirstSpace == StringRef::npos) {
423     LineStart = Buffer.substr(LineOffset);
424   } else {
425     LineStart = Buffer.substr(LineOffset, FirstSpace - LineOffset);
426   }
427 
428   TokenType Type = TT_Unknown;
429   if (LineStart == "<<<<<<<" || LineStart == ">>>>") {
430     Type = TT_ConflictStart;
431   } else if (LineStart == "|||||||" || LineStart == "=======" ||
432              LineStart == "====") {
433     Type = TT_ConflictAlternative;
434   } else if (LineStart == ">>>>>>>" || LineStart == "<<<<") {
435     Type = TT_ConflictEnd;
436   }
437 
438   if (Type != TT_Unknown) {
439     FormatToken *Next = Tokens.back();
440 
441     Tokens.resize(FirstInLineIndex + 1);
442     // We do not need to build a complete token here, as we will skip it
443     // during parsing anyway (as we must not touch whitespace around conflict
444     // markers).
445     Tokens.back()->Type = Type;
446     Tokens.back()->Tok.setKind(tok::kw___unknown_anytype);
447 
448     Tokens.push_back(Next);
449     return true;
450   }
451 
452   return false;
453 }
454 
455 FormatToken *FormatTokenLexer::getStashedToken() {
456   // Create a synthesized second '>' or '<' token.
457   Token Tok = FormatTok->Tok;
458   StringRef TokenText = FormatTok->TokenText;
459 
460   unsigned OriginalColumn = FormatTok->OriginalColumn;
461   FormatTok = new (Allocator.Allocate()) FormatToken;
462   FormatTok->Tok = Tok;
463   SourceLocation TokLocation =
464       FormatTok->Tok.getLocation().getLocWithOffset(Tok.getLength() - 1);
465   FormatTok->Tok.setLocation(TokLocation);
466   FormatTok->WhitespaceRange = SourceRange(TokLocation, TokLocation);
467   FormatTok->TokenText = TokenText;
468   FormatTok->ColumnWidth = 1;
469   FormatTok->OriginalColumn = OriginalColumn + 1;
470 
471   return FormatTok;
472 }
473 
474 FormatToken *FormatTokenLexer::getNextToken() {
475   if (StateStack.top() == LexerState::TOKEN_STASHED) {
476     StateStack.pop();
477     return getStashedToken();
478   }
479 
480   FormatTok = new (Allocator.Allocate()) FormatToken;
481   readRawToken(*FormatTok);
482   SourceLocation WhitespaceStart =
483       FormatTok->Tok.getLocation().getLocWithOffset(-TrailingWhitespace);
484   FormatTok->IsFirst = IsFirstToken;
485   IsFirstToken = false;
486 
487   // Consume and record whitespace until we find a significant token.
488   unsigned WhitespaceLength = TrailingWhitespace;
489   while (FormatTok->Tok.is(tok::unknown)) {
490     StringRef Text = FormatTok->TokenText;
491     auto EscapesNewline = [&](int pos) {
492       // A '\r' here is just part of '\r\n'. Skip it.
493       if (pos >= 0 && Text[pos] == '\r')
494         --pos;
495       // See whether there is an odd number of '\' before this.
496       // FIXME: This is wrong. A '\' followed by a newline is always removed,
497       // regardless of whether there is another '\' before it.
498       // FIXME: Newlines can also be escaped by a '?' '?' '/' trigraph.
499       unsigned count = 0;
500       for (; pos >= 0; --pos, ++count)
501         if (Text[pos] != '\\')
502           break;
503       return count & 1;
504     };
505     // FIXME: This miscounts tok:unknown tokens that are not just
506     // whitespace, e.g. a '`' character.
507     for (int i = 0, e = Text.size(); i != e; ++i) {
508       switch (Text[i]) {
509       case '\n':
510         ++FormatTok->NewlinesBefore;
511         FormatTok->HasUnescapedNewline = !EscapesNewline(i - 1);
512         FormatTok->LastNewlineOffset = WhitespaceLength + i + 1;
513         Column = 0;
514         break;
515       case '\r':
516         FormatTok->LastNewlineOffset = WhitespaceLength + i + 1;
517         Column = 0;
518         break;
519       case '\f':
520       case '\v':
521         Column = 0;
522         break;
523       case ' ':
524         ++Column;
525         break;
526       case '\t':
527         Column += Style.TabWidth - Column % Style.TabWidth;
528         break;
529       case '\\':
530         if (i + 1 == e || (Text[i + 1] != '\r' && Text[i + 1] != '\n'))
531           FormatTok->Type = TT_ImplicitStringLiteral;
532         break;
533       default:
534         FormatTok->Type = TT_ImplicitStringLiteral;
535         break;
536       }
537       if (FormatTok->Type == TT_ImplicitStringLiteral)
538         break;
539     }
540 
541     if (FormatTok->is(TT_ImplicitStringLiteral))
542       break;
543     WhitespaceLength += FormatTok->Tok.getLength();
544 
545     readRawToken(*FormatTok);
546   }
547 
548   // JavaScript and Java do not allow to escape the end of the line with a
549   // backslash. Backslashes are syntax errors in plain source, but can occur in
550   // comments. When a single line comment ends with a \, it'll cause the next
551   // line of code to be lexed as a comment, breaking formatting. The code below
552   // finds comments that contain a backslash followed by a line break, truncates
553   // the comment token at the backslash, and resets the lexer to restart behind
554   // the backslash.
555   if ((Style.Language == FormatStyle::LK_JavaScript ||
556        Style.Language == FormatStyle::LK_Java) &&
557       FormatTok->is(tok::comment) && FormatTok->TokenText.startswith("//")) {
558     size_t BackslashPos = FormatTok->TokenText.find('\\');
559     while (BackslashPos != StringRef::npos) {
560       if (BackslashPos + 1 < FormatTok->TokenText.size() &&
561           FormatTok->TokenText[BackslashPos + 1] == '\n') {
562         const char *Offset = Lex->getBufferLocation();
563         Offset -= FormatTok->TokenText.size();
564         Offset += BackslashPos + 1;
565         resetLexer(SourceMgr.getFileOffset(Lex->getSourceLocation(Offset)));
566         FormatTok->TokenText = FormatTok->TokenText.substr(0, BackslashPos + 1);
567         FormatTok->ColumnWidth = encoding::columnWidthWithTabs(
568             FormatTok->TokenText, FormatTok->OriginalColumn, Style.TabWidth,
569             Encoding);
570         break;
571       }
572       BackslashPos = FormatTok->TokenText.find('\\', BackslashPos + 1);
573     }
574   }
575 
576   // In case the token starts with escaped newlines, we want to
577   // take them into account as whitespace - this pattern is quite frequent
578   // in macro definitions.
579   // FIXME: Add a more explicit test.
580   while (FormatTok->TokenText.size() > 1 && FormatTok->TokenText[0] == '\\') {
581     unsigned SkippedWhitespace = 0;
582     if (FormatTok->TokenText.size() > 2 &&
583         (FormatTok->TokenText[1] == '\r' && FormatTok->TokenText[2] == '\n'))
584       SkippedWhitespace = 3;
585     else if (FormatTok->TokenText[1] == '\n')
586       SkippedWhitespace = 2;
587     else
588       break;
589 
590     ++FormatTok->NewlinesBefore;
591     WhitespaceLength += SkippedWhitespace;
592     FormatTok->LastNewlineOffset = SkippedWhitespace;
593     Column = 0;
594     FormatTok->TokenText = FormatTok->TokenText.substr(SkippedWhitespace);
595   }
596 
597   FormatTok->WhitespaceRange = SourceRange(
598       WhitespaceStart, WhitespaceStart.getLocWithOffset(WhitespaceLength));
599 
600   FormatTok->OriginalColumn = Column;
601 
602   TrailingWhitespace = 0;
603   if (FormatTok->Tok.is(tok::comment)) {
604     // FIXME: Add the trimmed whitespace to Column.
605     StringRef UntrimmedText = FormatTok->TokenText;
606     FormatTok->TokenText = FormatTok->TokenText.rtrim(" \t\v\f");
607     TrailingWhitespace = UntrimmedText.size() - FormatTok->TokenText.size();
608   } else if (FormatTok->Tok.is(tok::raw_identifier)) {
609     IdentifierInfo &Info = IdentTable.get(FormatTok->TokenText);
610     FormatTok->Tok.setIdentifierInfo(&Info);
611     FormatTok->Tok.setKind(Info.getTokenID());
612     if (Style.Language == FormatStyle::LK_Java &&
613         FormatTok->isOneOf(tok::kw_struct, tok::kw_union, tok::kw_delete,
614                            tok::kw_operator)) {
615       FormatTok->Tok.setKind(tok::identifier);
616       FormatTok->Tok.setIdentifierInfo(nullptr);
617     } else if (Style.Language == FormatStyle::LK_JavaScript &&
618                FormatTok->isOneOf(tok::kw_struct, tok::kw_union,
619                                   tok::kw_operator)) {
620       FormatTok->Tok.setKind(tok::identifier);
621       FormatTok->Tok.setIdentifierInfo(nullptr);
622     }
623   } else if (FormatTok->Tok.is(tok::greatergreater)) {
624     FormatTok->Tok.setKind(tok::greater);
625     FormatTok->TokenText = FormatTok->TokenText.substr(0, 1);
626     ++Column;
627     StateStack.push(LexerState::TOKEN_STASHED);
628   } else if (FormatTok->Tok.is(tok::lessless)) {
629     FormatTok->Tok.setKind(tok::less);
630     FormatTok->TokenText = FormatTok->TokenText.substr(0, 1);
631     ++Column;
632     StateStack.push(LexerState::TOKEN_STASHED);
633   }
634 
635   // Now FormatTok is the next non-whitespace token.
636 
637   StringRef Text = FormatTok->TokenText;
638   size_t FirstNewlinePos = Text.find('\n');
639   if (FirstNewlinePos == StringRef::npos) {
640     // FIXME: ColumnWidth actually depends on the start column, we need to
641     // take this into account when the token is moved.
642     FormatTok->ColumnWidth =
643         encoding::columnWidthWithTabs(Text, Column, Style.TabWidth, Encoding);
644     Column += FormatTok->ColumnWidth;
645   } else {
646     FormatTok->IsMultiline = true;
647     // FIXME: ColumnWidth actually depends on the start column, we need to
648     // take this into account when the token is moved.
649     FormatTok->ColumnWidth = encoding::columnWidthWithTabs(
650         Text.substr(0, FirstNewlinePos), Column, Style.TabWidth, Encoding);
651 
652     // The last line of the token always starts in column 0.
653     // Thus, the length can be precomputed even in the presence of tabs.
654     FormatTok->LastLineColumnWidth = encoding::columnWidthWithTabs(
655         Text.substr(Text.find_last_of('\n') + 1), 0, Style.TabWidth, Encoding);
656     Column = FormatTok->LastLineColumnWidth;
657   }
658 
659   if (Style.isCpp()) {
660     if (!(Tokens.size() > 0 && Tokens.back()->Tok.getIdentifierInfo() &&
661           Tokens.back()->Tok.getIdentifierInfo()->getPPKeywordID() ==
662               tok::pp_define) &&
663         std::find(ForEachMacros.begin(), ForEachMacros.end(),
664                   FormatTok->Tok.getIdentifierInfo()) != ForEachMacros.end()) {
665       FormatTok->Type = TT_ForEachMacro;
666     } else if (FormatTok->is(tok::identifier)) {
667       if (MacroBlockBeginRegex.match(Text)) {
668         FormatTok->Type = TT_MacroBlockBegin;
669       } else if (MacroBlockEndRegex.match(Text)) {
670         FormatTok->Type = TT_MacroBlockEnd;
671       }
672     }
673   }
674 
675   return FormatTok;
676 }
677 
678 void FormatTokenLexer::readRawToken(FormatToken &Tok) {
679   Lex->LexFromRawLexer(Tok.Tok);
680   Tok.TokenText = StringRef(SourceMgr.getCharacterData(Tok.Tok.getLocation()),
681                             Tok.Tok.getLength());
682   // For formatting, treat unterminated string literals like normal string
683   // literals.
684   if (Tok.is(tok::unknown)) {
685     if (!Tok.TokenText.empty() && Tok.TokenText[0] == '"') {
686       Tok.Tok.setKind(tok::string_literal);
687       Tok.IsUnterminatedLiteral = true;
688     } else if (Style.Language == FormatStyle::LK_JavaScript &&
689                Tok.TokenText == "''") {
690       Tok.Tok.setKind(tok::string_literal);
691     }
692   }
693 
694   if (Style.Language == FormatStyle::LK_JavaScript &&
695       Tok.is(tok::char_constant)) {
696     Tok.Tok.setKind(tok::string_literal);
697   }
698 
699   if (Tok.is(tok::comment) && (Tok.TokenText == "// clang-format on" ||
700                                Tok.TokenText == "/* clang-format on */")) {
701     FormattingDisabled = false;
702   }
703 
704   Tok.Finalized = FormattingDisabled;
705 
706   if (Tok.is(tok::comment) && (Tok.TokenText == "// clang-format off" ||
707                                Tok.TokenText == "/* clang-format off */")) {
708     FormattingDisabled = true;
709   }
710 }
711 
712 void FormatTokenLexer::resetLexer(unsigned Offset) {
713   StringRef Buffer = SourceMgr.getBufferData(ID);
714   Lex.reset(new Lexer(SourceMgr.getLocForStartOfFile(ID),
715                       getFormattingLangOpts(Style), Buffer.begin(),
716                       Buffer.begin() + Offset, Buffer.end()));
717   Lex->SetKeepWhitespaceMode(true);
718   TrailingWhitespace = 0;
719 }
720 
721 } // namespace format
722 } // namespace clang
723