1 //===--- FormatTokenLexer.cpp - Lex FormatTokens -------------*- C++ ----*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 ///
9 /// \file
10 /// This file implements FormatTokenLexer, which tokenizes a source file
11 /// into a FormatToken stream suitable for ClangFormat.
12 ///
13 //===----------------------------------------------------------------------===//
14 
15 #include "FormatTokenLexer.h"
16 #include "FormatToken.h"
17 #include "clang/Basic/SourceLocation.h"
18 #include "clang/Basic/SourceManager.h"
19 #include "clang/Format/Format.h"
20 #include "llvm/Support/Regex.h"
21 
22 namespace clang {
23 namespace format {
24 
25 FormatTokenLexer::FormatTokenLexer(const SourceManager &SourceMgr, FileID ID,
26                                    unsigned Column, const FormatStyle &Style,
27                                    encoding::Encoding Encoding)
28     : FormatTok(nullptr), IsFirstToken(true), StateStack({LexerState::NORMAL}),
29       Column(Column), TrailingWhitespace(0), SourceMgr(SourceMgr), ID(ID),
30       Style(Style), IdentTable(getFormattingLangOpts(Style)),
31       Keywords(IdentTable), Encoding(Encoding), FirstInLineIndex(0),
32       FormattingDisabled(false), MacroBlockBeginRegex(Style.MacroBlockBegin),
33       MacroBlockEndRegex(Style.MacroBlockEnd) {
34   Lex.reset(new Lexer(ID, SourceMgr.getBuffer(ID), SourceMgr,
35                       getFormattingLangOpts(Style)));
36   Lex->SetKeepWhitespaceMode(true);
37 
38   for (const std::string &ForEachMacro : Style.ForEachMacros)
39     Macros.insert({&IdentTable.get(ForEachMacro), TT_ForEachMacro});
40   for (const std::string &StatementMacro : Style.StatementMacros)
41     Macros.insert({&IdentTable.get(StatementMacro), TT_StatementMacro});
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 
70   if (Style.isCSharp()) {
71     if (tryMergeCSharpKeywordVariables())
72       return;
73     if (tryMergeCSharpVerbatimStringLiteral())
74       return;
75     if (tryMergeCSharpDoubleQuestion())
76       return;
77     if (tryMergeCSharpNullConditionals())
78       return;
79     static const tok::TokenKind JSRightArrow[] = {tok::equal, tok::greater};
80     if (tryMergeTokens(JSRightArrow, TT_JsFatArrow))
81       return;
82   }
83 
84   if (tryMergeNSStringLiteral())
85     return;
86 
87   if (Style.Language == FormatStyle::LK_JavaScript) {
88     static const tok::TokenKind JSIdentity[] = {tok::equalequal, tok::equal};
89     static const tok::TokenKind JSNotIdentity[] = {tok::exclaimequal,
90                                                    tok::equal};
91     static const tok::TokenKind JSShiftEqual[] = {tok::greater, tok::greater,
92                                                   tok::greaterequal};
93     static const tok::TokenKind JSRightArrow[] = {tok::equal, tok::greater};
94     static const tok::TokenKind JSExponentiation[] = {tok::star, tok::star};
95     static const tok::TokenKind JSExponentiationEqual[] = {tok::star,
96                                                            tok::starequal};
97 
98     // FIXME: Investigate what token type gives the correct operator priority.
99     if (tryMergeTokens(JSIdentity, TT_BinaryOperator))
100       return;
101     if (tryMergeTokens(JSNotIdentity, TT_BinaryOperator))
102       return;
103     if (tryMergeTokens(JSShiftEqual, TT_BinaryOperator))
104       return;
105     if (tryMergeTokens(JSRightArrow, TT_JsFatArrow))
106       return;
107     if (tryMergeTokens(JSExponentiation, TT_JsExponentiation))
108       return;
109     if (tryMergeTokens(JSExponentiationEqual, TT_JsExponentiationEqual)) {
110       Tokens.back()->Tok.setKind(tok::starequal);
111       return;
112     }
113     if (tryMergeJSPrivateIdentifier())
114       return;
115   }
116 
117   if (Style.Language == FormatStyle::LK_Java) {
118     static const tok::TokenKind JavaRightLogicalShiftAssign[] = {
119         tok::greater, tok::greater, tok::greaterequal};
120     if (tryMergeTokens(JavaRightLogicalShiftAssign, TT_BinaryOperator))
121       return;
122   }
123 }
124 
125 bool FormatTokenLexer::tryMergeNSStringLiteral() {
126   if (Tokens.size() < 2)
127     return false;
128   auto &At = *(Tokens.end() - 2);
129   auto &String = *(Tokens.end() - 1);
130   if (!At->is(tok::at) || !String->is(tok::string_literal))
131     return false;
132   At->Tok.setKind(tok::string_literal);
133   At->TokenText = StringRef(At->TokenText.begin(),
134                             String->TokenText.end() - At->TokenText.begin());
135   At->ColumnWidth += String->ColumnWidth;
136   At->Type = TT_ObjCStringLiteral;
137   Tokens.erase(Tokens.end() - 1);
138   return true;
139 }
140 
141 bool FormatTokenLexer::tryMergeJSPrivateIdentifier() {
142   // Merges #idenfier into a single identifier with the text #identifier
143   // but the token tok::identifier.
144   if (Tokens.size() < 2)
145     return false;
146   auto &Hash = *(Tokens.end() - 2);
147   auto &Identifier = *(Tokens.end() - 1);
148   if (!Hash->is(tok::hash) || !Identifier->is(tok::identifier))
149     return false;
150   Hash->Tok.setKind(tok::identifier);
151   Hash->TokenText =
152       StringRef(Hash->TokenText.begin(),
153                 Identifier->TokenText.end() - Hash->TokenText.begin());
154   Hash->ColumnWidth += Identifier->ColumnWidth;
155   Hash->Type = TT_JsPrivateIdentifier;
156   Tokens.erase(Tokens.end() - 1);
157   return true;
158 }
159 
160 // Search for verbatim or interpolated string literals @"ABC" or
161 // $"aaaaa{abc}aaaaa" i and mark the token as TT_CSharpStringLiteral, and to
162 // prevent splitting of @, $ and ".
163 bool FormatTokenLexer::tryMergeCSharpVerbatimStringLiteral() {
164   if (Tokens.size() < 2)
165     return false;
166   auto &At = *(Tokens.end() - 2);
167   auto &String = *(Tokens.end() - 1);
168 
169   // Look for $"aaaaaa" @"aaaaaa".
170   if (!(At->is(tok::at) || At->TokenText == "$") ||
171       !String->is(tok::string_literal))
172     return false;
173 
174   if (Tokens.size() >= 2 && At->is(tok::at)) {
175     auto &Dollar = *(Tokens.end() - 3);
176     if (Dollar->TokenText == "$") {
177       // This looks like $@"aaaaa" so we need to combine all 3 tokens.
178       Dollar->Tok.setKind(tok::string_literal);
179       Dollar->TokenText =
180           StringRef(Dollar->TokenText.begin(),
181                     String->TokenText.end() - Dollar->TokenText.begin());
182       Dollar->ColumnWidth += (At->ColumnWidth + String->ColumnWidth);
183       Dollar->Type = TT_CSharpStringLiteral;
184       Tokens.erase(Tokens.end() - 2);
185       Tokens.erase(Tokens.end() - 1);
186       return true;
187     }
188   }
189 
190   // Convert back into just a string_literal.
191   At->Tok.setKind(tok::string_literal);
192   At->TokenText = StringRef(At->TokenText.begin(),
193                             String->TokenText.end() - At->TokenText.begin());
194   At->ColumnWidth += String->ColumnWidth;
195   At->Type = TT_CSharpStringLiteral;
196   Tokens.erase(Tokens.end() - 1);
197   return true;
198 }
199 
200 bool FormatTokenLexer::tryMergeCSharpDoubleQuestion() {
201   if (Tokens.size() < 2)
202     return false;
203   auto &FirstQuestion = *(Tokens.end() - 2);
204   auto &SecondQuestion = *(Tokens.end() - 1);
205   if (!FirstQuestion->is(tok::question) || !SecondQuestion->is(tok::question))
206     return false;
207   FirstQuestion->Tok.setKind(tok::question);
208   FirstQuestion->TokenText = StringRef(FirstQuestion->TokenText.begin(),
209                                        SecondQuestion->TokenText.end() -
210                                            FirstQuestion->TokenText.begin());
211   FirstQuestion->ColumnWidth += SecondQuestion->ColumnWidth;
212   FirstQuestion->Type = TT_CSharpNullCoalescing;
213   Tokens.erase(Tokens.end() - 1);
214   return true;
215 }
216 
217 bool FormatTokenLexer::tryMergeCSharpKeywordVariables() {
218   if (Tokens.size() < 2)
219     return false;
220   auto &At = *(Tokens.end() - 2);
221   auto &Keyword = *(Tokens.end() - 1);
222   if (!At->is(tok::at))
223     return false;
224   if (!Keywords.isCSharpKeyword(*Keyword))
225     return false;
226 
227   At->Tok.setKind(tok::identifier);
228   At->TokenText = StringRef(At->TokenText.begin(),
229                             Keyword->TokenText.end() - At->TokenText.begin());
230   At->ColumnWidth += Keyword->ColumnWidth;
231   At->Type = Keyword->Type;
232   Tokens.erase(Tokens.end() - 1);
233   return true;
234 }
235 
236 // In C# merge the Identifier and the ? together e.g. arg?.
237 bool FormatTokenLexer::tryMergeCSharpNullConditionals() {
238   if (Tokens.size() < 2)
239     return false;
240   auto &Identifier = *(Tokens.end() - 2);
241   auto &Question = *(Tokens.end() - 1);
242   if (!Identifier->isOneOf(tok::r_square, tok::identifier) ||
243       !Question->is(tok::question))
244     return false;
245   Identifier->TokenText =
246       StringRef(Identifier->TokenText.begin(),
247                 Question->TokenText.end() - Identifier->TokenText.begin());
248   Identifier->ColumnWidth += Question->ColumnWidth;
249   Tokens.erase(Tokens.end() - 1);
250   return true;
251 }
252 
253 bool FormatTokenLexer::tryMergeLessLess() {
254   // Merge X,less,less,Y into X,lessless,Y unless X or Y is less.
255   if (Tokens.size() < 3)
256     return false;
257 
258   bool FourthTokenIsLess = false;
259   if (Tokens.size() > 3)
260     FourthTokenIsLess = (Tokens.end() - 4)[0]->is(tok::less);
261 
262   auto First = Tokens.end() - 3;
263   if (First[2]->is(tok::less) || First[1]->isNot(tok::less) ||
264       First[0]->isNot(tok::less) || FourthTokenIsLess)
265     return false;
266 
267   // Only merge if there currently is no whitespace between the two "<".
268   if (First[1]->WhitespaceRange.getBegin() !=
269       First[1]->WhitespaceRange.getEnd())
270     return false;
271 
272   First[0]->Tok.setKind(tok::lessless);
273   First[0]->TokenText = "<<";
274   First[0]->ColumnWidth += 1;
275   Tokens.erase(Tokens.end() - 2);
276   return true;
277 }
278 
279 bool FormatTokenLexer::tryMergeTokens(ArrayRef<tok::TokenKind> Kinds,
280                                       TokenType NewType) {
281   if (Tokens.size() < Kinds.size())
282     return false;
283 
284   SmallVectorImpl<FormatToken *>::const_iterator First =
285       Tokens.end() - Kinds.size();
286   if (!First[0]->is(Kinds[0]))
287     return false;
288   unsigned AddLength = 0;
289   for (unsigned i = 1; i < Kinds.size(); ++i) {
290     if (!First[i]->is(Kinds[i]) || First[i]->WhitespaceRange.getBegin() !=
291                                        First[i]->WhitespaceRange.getEnd())
292       return false;
293     AddLength += First[i]->TokenText.size();
294   }
295   Tokens.resize(Tokens.size() - Kinds.size() + 1);
296   First[0]->TokenText = StringRef(First[0]->TokenText.data(),
297                                   First[0]->TokenText.size() + AddLength);
298   First[0]->ColumnWidth += AddLength;
299   First[0]->Type = NewType;
300   return true;
301 }
302 
303 // Returns \c true if \p Tok can only be followed by an operand in JavaScript.
304 bool FormatTokenLexer::precedesOperand(FormatToken *Tok) {
305   // NB: This is not entirely correct, as an r_paren can introduce an operand
306   // location in e.g. `if (foo) /bar/.exec(...);`. That is a rare enough
307   // corner case to not matter in practice, though.
308   return Tok->isOneOf(tok::period, tok::l_paren, tok::comma, tok::l_brace,
309                       tok::r_brace, tok::l_square, tok::semi, tok::exclaim,
310                       tok::colon, tok::question, tok::tilde) ||
311          Tok->isOneOf(tok::kw_return, tok::kw_do, tok::kw_case, tok::kw_throw,
312                       tok::kw_else, tok::kw_new, tok::kw_delete, tok::kw_void,
313                       tok::kw_typeof, Keywords.kw_instanceof, Keywords.kw_in) ||
314          Tok->isBinaryOperator();
315 }
316 
317 bool FormatTokenLexer::canPrecedeRegexLiteral(FormatToken *Prev) {
318   if (!Prev)
319     return true;
320 
321   // Regex literals can only follow after prefix unary operators, not after
322   // postfix unary operators. If the '++' is followed by a non-operand
323   // introducing token, the slash here is the operand and not the start of a
324   // regex.
325   // `!` is an unary prefix operator, but also a post-fix operator that casts
326   // away nullability, so the same check applies.
327   if (Prev->isOneOf(tok::plusplus, tok::minusminus, tok::exclaim))
328     return (Tokens.size() < 3 || precedesOperand(Tokens[Tokens.size() - 3]));
329 
330   // The previous token must introduce an operand location where regex
331   // literals can occur.
332   if (!precedesOperand(Prev))
333     return false;
334 
335   return true;
336 }
337 
338 // Tries to parse a JavaScript Regex literal starting at the current token,
339 // if that begins with a slash and is in a location where JavaScript allows
340 // regex literals. Changes the current token to a regex literal and updates
341 // its text if successful.
342 void FormatTokenLexer::tryParseJSRegexLiteral() {
343   FormatToken *RegexToken = Tokens.back();
344   if (!RegexToken->isOneOf(tok::slash, tok::slashequal))
345     return;
346 
347   FormatToken *Prev = nullptr;
348   for (auto I = Tokens.rbegin() + 1, E = Tokens.rend(); I != E; ++I) {
349     // NB: Because previous pointers are not initialized yet, this cannot use
350     // Token.getPreviousNonComment.
351     if ((*I)->isNot(tok::comment)) {
352       Prev = *I;
353       break;
354     }
355   }
356 
357   if (!canPrecedeRegexLiteral(Prev))
358     return;
359 
360   // 'Manually' lex ahead in the current file buffer.
361   const char *Offset = Lex->getBufferLocation();
362   const char *RegexBegin = Offset - RegexToken->TokenText.size();
363   StringRef Buffer = Lex->getBuffer();
364   bool InCharacterClass = false;
365   bool HaveClosingSlash = false;
366   for (; !HaveClosingSlash && Offset != Buffer.end(); ++Offset) {
367     // Regular expressions are terminated with a '/', which can only be
368     // escaped using '\' or a character class between '[' and ']'.
369     // See http://www.ecma-international.org/ecma-262/5.1/#sec-7.8.5.
370     switch (*Offset) {
371     case '\\':
372       // Skip the escaped character.
373       ++Offset;
374       break;
375     case '[':
376       InCharacterClass = true;
377       break;
378     case ']':
379       InCharacterClass = false;
380       break;
381     case '/':
382       if (!InCharacterClass)
383         HaveClosingSlash = true;
384       break;
385     }
386   }
387 
388   RegexToken->Type = TT_RegexLiteral;
389   // Treat regex literals like other string_literals.
390   RegexToken->Tok.setKind(tok::string_literal);
391   RegexToken->TokenText = StringRef(RegexBegin, Offset - RegexBegin);
392   RegexToken->ColumnWidth = RegexToken->TokenText.size();
393 
394   resetLexer(SourceMgr.getFileOffset(Lex->getSourceLocation(Offset)));
395 }
396 
397 void FormatTokenLexer::handleTemplateStrings() {
398   FormatToken *BacktickToken = Tokens.back();
399 
400   if (BacktickToken->is(tok::l_brace)) {
401     StateStack.push(LexerState::NORMAL);
402     return;
403   }
404   if (BacktickToken->is(tok::r_brace)) {
405     if (StateStack.size() == 1)
406       return;
407     StateStack.pop();
408     if (StateStack.top() != LexerState::TEMPLATE_STRING)
409       return;
410     // If back in TEMPLATE_STRING, fallthrough and continue parsing the
411   } else if (BacktickToken->is(tok::unknown) &&
412              BacktickToken->TokenText == "`") {
413     StateStack.push(LexerState::TEMPLATE_STRING);
414   } else {
415     return; // Not actually a template
416   }
417 
418   // 'Manually' lex ahead in the current file buffer.
419   const char *Offset = Lex->getBufferLocation();
420   const char *TmplBegin = Offset - BacktickToken->TokenText.size(); // at "`"
421   for (; Offset != Lex->getBuffer().end(); ++Offset) {
422     if (Offset[0] == '`') {
423       StateStack.pop();
424       break;
425     }
426     if (Offset[0] == '\\') {
427       ++Offset; // Skip the escaped character.
428     } else if (Offset + 1 < Lex->getBuffer().end() && Offset[0] == '$' &&
429                Offset[1] == '{') {
430       // '${' introduces an expression interpolation in the template string.
431       StateStack.push(LexerState::NORMAL);
432       ++Offset;
433       break;
434     }
435   }
436 
437   StringRef LiteralText(TmplBegin, Offset - TmplBegin + 1);
438   BacktickToken->Type = TT_TemplateString;
439   BacktickToken->Tok.setKind(tok::string_literal);
440   BacktickToken->TokenText = LiteralText;
441 
442   // Adjust width for potentially multiline string literals.
443   size_t FirstBreak = LiteralText.find('\n');
444   StringRef FirstLineText = FirstBreak == StringRef::npos
445                                 ? LiteralText
446                                 : LiteralText.substr(0, FirstBreak);
447   BacktickToken->ColumnWidth = encoding::columnWidthWithTabs(
448       FirstLineText, BacktickToken->OriginalColumn, Style.TabWidth, Encoding);
449   size_t LastBreak = LiteralText.rfind('\n');
450   if (LastBreak != StringRef::npos) {
451     BacktickToken->IsMultiline = true;
452     unsigned StartColumn = 0; // The template tail spans the entire line.
453     BacktickToken->LastLineColumnWidth = encoding::columnWidthWithTabs(
454         LiteralText.substr(LastBreak + 1, LiteralText.size()), StartColumn,
455         Style.TabWidth, Encoding);
456   }
457 
458   SourceLocation loc = Offset < Lex->getBuffer().end()
459                            ? Lex->getSourceLocation(Offset + 1)
460                            : SourceMgr.getLocForEndOfFile(ID);
461   resetLexer(SourceMgr.getFileOffset(loc));
462 }
463 
464 void FormatTokenLexer::tryParsePythonComment() {
465   FormatToken *HashToken = Tokens.back();
466   if (!HashToken->isOneOf(tok::hash, tok::hashhash))
467     return;
468   // Turn the remainder of this line into a comment.
469   const char *CommentBegin =
470       Lex->getBufferLocation() - HashToken->TokenText.size(); // at "#"
471   size_t From = CommentBegin - Lex->getBuffer().begin();
472   size_t To = Lex->getBuffer().find_first_of('\n', From);
473   if (To == StringRef::npos)
474     To = Lex->getBuffer().size();
475   size_t Len = To - From;
476   HashToken->Type = TT_LineComment;
477   HashToken->Tok.setKind(tok::comment);
478   HashToken->TokenText = Lex->getBuffer().substr(From, Len);
479   SourceLocation Loc = To < Lex->getBuffer().size()
480                            ? Lex->getSourceLocation(CommentBegin + Len)
481                            : SourceMgr.getLocForEndOfFile(ID);
482   resetLexer(SourceMgr.getFileOffset(Loc));
483 }
484 
485 bool FormatTokenLexer::tryMerge_TMacro() {
486   if (Tokens.size() < 4)
487     return false;
488   FormatToken *Last = Tokens.back();
489   if (!Last->is(tok::r_paren))
490     return false;
491 
492   FormatToken *String = Tokens[Tokens.size() - 2];
493   if (!String->is(tok::string_literal) || String->IsMultiline)
494     return false;
495 
496   if (!Tokens[Tokens.size() - 3]->is(tok::l_paren))
497     return false;
498 
499   FormatToken *Macro = Tokens[Tokens.size() - 4];
500   if (Macro->TokenText != "_T")
501     return false;
502 
503   const char *Start = Macro->TokenText.data();
504   const char *End = Last->TokenText.data() + Last->TokenText.size();
505   String->TokenText = StringRef(Start, End - Start);
506   String->IsFirst = Macro->IsFirst;
507   String->LastNewlineOffset = Macro->LastNewlineOffset;
508   String->WhitespaceRange = Macro->WhitespaceRange;
509   String->OriginalColumn = Macro->OriginalColumn;
510   String->ColumnWidth = encoding::columnWidthWithTabs(
511       String->TokenText, String->OriginalColumn, Style.TabWidth, Encoding);
512   String->NewlinesBefore = Macro->NewlinesBefore;
513   String->HasUnescapedNewline = Macro->HasUnescapedNewline;
514 
515   Tokens.pop_back();
516   Tokens.pop_back();
517   Tokens.pop_back();
518   Tokens.back() = String;
519   return true;
520 }
521 
522 bool FormatTokenLexer::tryMergeConflictMarkers() {
523   if (Tokens.back()->NewlinesBefore == 0 && Tokens.back()->isNot(tok::eof))
524     return false;
525 
526   // Conflict lines look like:
527   // <marker> <text from the vcs>
528   // For example:
529   // >>>>>>> /file/in/file/system at revision 1234
530   //
531   // We merge all tokens in a line that starts with a conflict marker
532   // into a single token with a special token type that the unwrapped line
533   // parser will use to correctly rebuild the underlying code.
534 
535   FileID ID;
536   // Get the position of the first token in the line.
537   unsigned FirstInLineOffset;
538   std::tie(ID, FirstInLineOffset) = SourceMgr.getDecomposedLoc(
539       Tokens[FirstInLineIndex]->getStartOfNonWhitespace());
540   StringRef Buffer = SourceMgr.getBuffer(ID)->getBuffer();
541   // Calculate the offset of the start of the current line.
542   auto LineOffset = Buffer.rfind('\n', FirstInLineOffset);
543   if (LineOffset == StringRef::npos) {
544     LineOffset = 0;
545   } else {
546     ++LineOffset;
547   }
548 
549   auto FirstSpace = Buffer.find_first_of(" \n", LineOffset);
550   StringRef LineStart;
551   if (FirstSpace == StringRef::npos) {
552     LineStart = Buffer.substr(LineOffset);
553   } else {
554     LineStart = Buffer.substr(LineOffset, FirstSpace - LineOffset);
555   }
556 
557   TokenType Type = TT_Unknown;
558   if (LineStart == "<<<<<<<" || LineStart == ">>>>") {
559     Type = TT_ConflictStart;
560   } else if (LineStart == "|||||||" || LineStart == "=======" ||
561              LineStart == "====") {
562     Type = TT_ConflictAlternative;
563   } else if (LineStart == ">>>>>>>" || LineStart == "<<<<") {
564     Type = TT_ConflictEnd;
565   }
566 
567   if (Type != TT_Unknown) {
568     FormatToken *Next = Tokens.back();
569 
570     Tokens.resize(FirstInLineIndex + 1);
571     // We do not need to build a complete token here, as we will skip it
572     // during parsing anyway (as we must not touch whitespace around conflict
573     // markers).
574     Tokens.back()->Type = Type;
575     Tokens.back()->Tok.setKind(tok::kw___unknown_anytype);
576 
577     Tokens.push_back(Next);
578     return true;
579   }
580 
581   return false;
582 }
583 
584 FormatToken *FormatTokenLexer::getStashedToken() {
585   // Create a synthesized second '>' or '<' token.
586   Token Tok = FormatTok->Tok;
587   StringRef TokenText = FormatTok->TokenText;
588 
589   unsigned OriginalColumn = FormatTok->OriginalColumn;
590   FormatTok = new (Allocator.Allocate()) FormatToken;
591   FormatTok->Tok = Tok;
592   SourceLocation TokLocation =
593       FormatTok->Tok.getLocation().getLocWithOffset(Tok.getLength() - 1);
594   FormatTok->Tok.setLocation(TokLocation);
595   FormatTok->WhitespaceRange = SourceRange(TokLocation, TokLocation);
596   FormatTok->TokenText = TokenText;
597   FormatTok->ColumnWidth = 1;
598   FormatTok->OriginalColumn = OriginalColumn + 1;
599 
600   return FormatTok;
601 }
602 
603 FormatToken *FormatTokenLexer::getNextToken() {
604   if (StateStack.top() == LexerState::TOKEN_STASHED) {
605     StateStack.pop();
606     return getStashedToken();
607   }
608 
609   FormatTok = new (Allocator.Allocate()) FormatToken;
610   readRawToken(*FormatTok);
611   SourceLocation WhitespaceStart =
612       FormatTok->Tok.getLocation().getLocWithOffset(-TrailingWhitespace);
613   FormatTok->IsFirst = IsFirstToken;
614   IsFirstToken = false;
615 
616   // Consume and record whitespace until we find a significant token.
617   unsigned WhitespaceLength = TrailingWhitespace;
618   while (FormatTok->Tok.is(tok::unknown)) {
619     StringRef Text = FormatTok->TokenText;
620     auto EscapesNewline = [&](int pos) {
621       // A '\r' here is just part of '\r\n'. Skip it.
622       if (pos >= 0 && Text[pos] == '\r')
623         --pos;
624       // See whether there is an odd number of '\' before this.
625       // FIXME: This is wrong. A '\' followed by a newline is always removed,
626       // regardless of whether there is another '\' before it.
627       // FIXME: Newlines can also be escaped by a '?' '?' '/' trigraph.
628       unsigned count = 0;
629       for (; pos >= 0; --pos, ++count)
630         if (Text[pos] != '\\')
631           break;
632       return count & 1;
633     };
634     // FIXME: This miscounts tok:unknown tokens that are not just
635     // whitespace, e.g. a '`' character.
636     for (int i = 0, e = Text.size(); i != e; ++i) {
637       switch (Text[i]) {
638       case '\n':
639         ++FormatTok->NewlinesBefore;
640         FormatTok->HasUnescapedNewline = !EscapesNewline(i - 1);
641         FormatTok->LastNewlineOffset = WhitespaceLength + i + 1;
642         Column = 0;
643         break;
644       case '\r':
645         FormatTok->LastNewlineOffset = WhitespaceLength + i + 1;
646         Column = 0;
647         break;
648       case '\f':
649       case '\v':
650         Column = 0;
651         break;
652       case ' ':
653         ++Column;
654         break;
655       case '\t':
656         Column += Style.TabWidth - Column % Style.TabWidth;
657         break;
658       case '\\':
659         if (i + 1 == e || (Text[i + 1] != '\r' && Text[i + 1] != '\n'))
660           FormatTok->Type = TT_ImplicitStringLiteral;
661         break;
662       default:
663         FormatTok->Type = TT_ImplicitStringLiteral;
664         break;
665       }
666       if (FormatTok->Type == TT_ImplicitStringLiteral)
667         break;
668     }
669 
670     if (FormatTok->is(TT_ImplicitStringLiteral))
671       break;
672     WhitespaceLength += FormatTok->Tok.getLength();
673 
674     readRawToken(*FormatTok);
675   }
676 
677   // JavaScript and Java do not allow to escape the end of the line with a
678   // backslash. Backslashes are syntax errors in plain source, but can occur in
679   // comments. When a single line comment ends with a \, it'll cause the next
680   // line of code to be lexed as a comment, breaking formatting. The code below
681   // finds comments that contain a backslash followed by a line break, truncates
682   // the comment token at the backslash, and resets the lexer to restart behind
683   // the backslash.
684   if ((Style.Language == FormatStyle::LK_JavaScript ||
685        Style.Language == FormatStyle::LK_Java) &&
686       FormatTok->is(tok::comment) && FormatTok->TokenText.startswith("//")) {
687     size_t BackslashPos = FormatTok->TokenText.find('\\');
688     while (BackslashPos != StringRef::npos) {
689       if (BackslashPos + 1 < FormatTok->TokenText.size() &&
690           FormatTok->TokenText[BackslashPos + 1] == '\n') {
691         const char *Offset = Lex->getBufferLocation();
692         Offset -= FormatTok->TokenText.size();
693         Offset += BackslashPos + 1;
694         resetLexer(SourceMgr.getFileOffset(Lex->getSourceLocation(Offset)));
695         FormatTok->TokenText = FormatTok->TokenText.substr(0, BackslashPos + 1);
696         FormatTok->ColumnWidth = encoding::columnWidthWithTabs(
697             FormatTok->TokenText, FormatTok->OriginalColumn, Style.TabWidth,
698             Encoding);
699         break;
700       }
701       BackslashPos = FormatTok->TokenText.find('\\', BackslashPos + 1);
702     }
703   }
704 
705   // In case the token starts with escaped newlines, we want to
706   // take them into account as whitespace - this pattern is quite frequent
707   // in macro definitions.
708   // FIXME: Add a more explicit test.
709   while (FormatTok->TokenText.size() > 1 && FormatTok->TokenText[0] == '\\') {
710     unsigned SkippedWhitespace = 0;
711     if (FormatTok->TokenText.size() > 2 &&
712         (FormatTok->TokenText[1] == '\r' && FormatTok->TokenText[2] == '\n'))
713       SkippedWhitespace = 3;
714     else if (FormatTok->TokenText[1] == '\n')
715       SkippedWhitespace = 2;
716     else
717       break;
718 
719     ++FormatTok->NewlinesBefore;
720     WhitespaceLength += SkippedWhitespace;
721     FormatTok->LastNewlineOffset = SkippedWhitespace;
722     Column = 0;
723     FormatTok->TokenText = FormatTok->TokenText.substr(SkippedWhitespace);
724   }
725 
726   FormatTok->WhitespaceRange = SourceRange(
727       WhitespaceStart, WhitespaceStart.getLocWithOffset(WhitespaceLength));
728 
729   FormatTok->OriginalColumn = Column;
730 
731   TrailingWhitespace = 0;
732   if (FormatTok->Tok.is(tok::comment)) {
733     // FIXME: Add the trimmed whitespace to Column.
734     StringRef UntrimmedText = FormatTok->TokenText;
735     FormatTok->TokenText = FormatTok->TokenText.rtrim(" \t\v\f");
736     TrailingWhitespace = UntrimmedText.size() - FormatTok->TokenText.size();
737   } else if (FormatTok->Tok.is(tok::raw_identifier)) {
738     IdentifierInfo &Info = IdentTable.get(FormatTok->TokenText);
739     FormatTok->Tok.setIdentifierInfo(&Info);
740     FormatTok->Tok.setKind(Info.getTokenID());
741     if (Style.Language == FormatStyle::LK_Java &&
742         FormatTok->isOneOf(tok::kw_struct, tok::kw_union, tok::kw_delete,
743                            tok::kw_operator)) {
744       FormatTok->Tok.setKind(tok::identifier);
745       FormatTok->Tok.setIdentifierInfo(nullptr);
746     } else if (Style.Language == FormatStyle::LK_JavaScript &&
747                FormatTok->isOneOf(tok::kw_struct, tok::kw_union,
748                                   tok::kw_operator)) {
749       FormatTok->Tok.setKind(tok::identifier);
750       FormatTok->Tok.setIdentifierInfo(nullptr);
751     }
752   } else if (FormatTok->Tok.is(tok::greatergreater)) {
753     FormatTok->Tok.setKind(tok::greater);
754     FormatTok->TokenText = FormatTok->TokenText.substr(0, 1);
755     ++Column;
756     StateStack.push(LexerState::TOKEN_STASHED);
757   } else if (FormatTok->Tok.is(tok::lessless)) {
758     FormatTok->Tok.setKind(tok::less);
759     FormatTok->TokenText = FormatTok->TokenText.substr(0, 1);
760     ++Column;
761     StateStack.push(LexerState::TOKEN_STASHED);
762   }
763 
764   // Now FormatTok is the next non-whitespace token.
765 
766   StringRef Text = FormatTok->TokenText;
767   size_t FirstNewlinePos = Text.find('\n');
768   if (FirstNewlinePos == StringRef::npos) {
769     // FIXME: ColumnWidth actually depends on the start column, we need to
770     // take this into account when the token is moved.
771     FormatTok->ColumnWidth =
772         encoding::columnWidthWithTabs(Text, Column, Style.TabWidth, Encoding);
773     Column += FormatTok->ColumnWidth;
774   } else {
775     FormatTok->IsMultiline = true;
776     // FIXME: ColumnWidth actually depends on the start column, we need to
777     // take this into account when the token is moved.
778     FormatTok->ColumnWidth = encoding::columnWidthWithTabs(
779         Text.substr(0, FirstNewlinePos), Column, Style.TabWidth, Encoding);
780 
781     // The last line of the token always starts in column 0.
782     // Thus, the length can be precomputed even in the presence of tabs.
783     FormatTok->LastLineColumnWidth = encoding::columnWidthWithTabs(
784         Text.substr(Text.find_last_of('\n') + 1), 0, Style.TabWidth, Encoding);
785     Column = FormatTok->LastLineColumnWidth;
786   }
787 
788   if (Style.isCpp()) {
789     auto it = Macros.find(FormatTok->Tok.getIdentifierInfo());
790     if (!(Tokens.size() > 0 && Tokens.back()->Tok.getIdentifierInfo() &&
791           Tokens.back()->Tok.getIdentifierInfo()->getPPKeywordID() ==
792               tok::pp_define) &&
793         it != Macros.end()) {
794       FormatTok->Type = it->second;
795     } else if (FormatTok->is(tok::identifier)) {
796       if (MacroBlockBeginRegex.match(Text)) {
797         FormatTok->Type = TT_MacroBlockBegin;
798       } else if (MacroBlockEndRegex.match(Text)) {
799         FormatTok->Type = TT_MacroBlockEnd;
800       }
801     }
802   }
803 
804   return FormatTok;
805 }
806 
807 void FormatTokenLexer::readRawToken(FormatToken &Tok) {
808   Lex->LexFromRawLexer(Tok.Tok);
809   Tok.TokenText = StringRef(SourceMgr.getCharacterData(Tok.Tok.getLocation()),
810                             Tok.Tok.getLength());
811   // For formatting, treat unterminated string literals like normal string
812   // literals.
813   if (Tok.is(tok::unknown)) {
814     if (!Tok.TokenText.empty() && Tok.TokenText[0] == '"') {
815       Tok.Tok.setKind(tok::string_literal);
816       Tok.IsUnterminatedLiteral = true;
817     } else if (Style.Language == FormatStyle::LK_JavaScript &&
818                Tok.TokenText == "''") {
819       Tok.Tok.setKind(tok::string_literal);
820     }
821   }
822 
823   if ((Style.Language == FormatStyle::LK_JavaScript ||
824        Style.Language == FormatStyle::LK_Proto ||
825        Style.Language == FormatStyle::LK_TextProto) &&
826       Tok.is(tok::char_constant)) {
827     Tok.Tok.setKind(tok::string_literal);
828   }
829 
830   if (Tok.is(tok::comment) && (Tok.TokenText == "// clang-format on" ||
831                                Tok.TokenText == "/* clang-format on */")) {
832     FormattingDisabled = false;
833   }
834 
835   Tok.Finalized = FormattingDisabled;
836 
837   if (Tok.is(tok::comment) && (Tok.TokenText == "// clang-format off" ||
838                                Tok.TokenText == "/* clang-format off */")) {
839     FormattingDisabled = true;
840   }
841 }
842 
843 void FormatTokenLexer::resetLexer(unsigned Offset) {
844   StringRef Buffer = SourceMgr.getBufferData(ID);
845   Lex.reset(new Lexer(SourceMgr.getLocForStartOfFile(ID),
846                       getFormattingLangOpts(Style), Buffer.begin(),
847                       Buffer.begin() + Offset, Buffer.end()));
848   Lex->SetKeepWhitespaceMode(true);
849   TrailingWhitespace = 0;
850 }
851 
852 } // namespace format
853 } // namespace clang
854