1 //===- Tokens.cpp - collect tokens from preprocessing ---------------------===//
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 #include "clang/Tooling/Syntax/Tokens.h"
9 
10 #include "clang/Basic/Diagnostic.h"
11 #include "clang/Basic/IdentifierTable.h"
12 #include "clang/Basic/LLVM.h"
13 #include "clang/Basic/LangOptions.h"
14 #include "clang/Basic/SourceLocation.h"
15 #include "clang/Basic/SourceManager.h"
16 #include "clang/Basic/TokenKinds.h"
17 #include "clang/Lex/PPCallbacks.h"
18 #include "clang/Lex/Preprocessor.h"
19 #include "clang/Lex/Token.h"
20 #include "llvm/ADT/ArrayRef.h"
21 #include "llvm/ADT/None.h"
22 #include "llvm/ADT/Optional.h"
23 #include "llvm/ADT/STLExtras.h"
24 #include "llvm/Support/Debug.h"
25 #include "llvm/Support/ErrorHandling.h"
26 #include "llvm/Support/FormatVariadic.h"
27 #include "llvm/Support/raw_ostream.h"
28 #include <algorithm>
29 #include <cassert>
30 #include <iterator>
31 #include <string>
32 #include <utility>
33 #include <vector>
34 
35 using namespace clang;
36 using namespace clang::syntax;
37 
38 syntax::Token::Token(SourceLocation Location, unsigned Length,
39                      tok::TokenKind Kind)
40     : Location(Location), Length(Length), Kind(Kind) {
41   assert(Location.isValid());
42 }
43 
44 syntax::Token::Token(const clang::Token &T)
45     : Token(T.getLocation(), T.getLength(), T.getKind()) {
46   assert(!T.isAnnotation());
47 }
48 
49 llvm::StringRef syntax::Token::text(const SourceManager &SM) const {
50   bool Invalid = false;
51   const char *Start = SM.getCharacterData(location(), &Invalid);
52   assert(!Invalid);
53   return llvm::StringRef(Start, length());
54 }
55 
56 FileRange syntax::Token::range(const SourceManager &SM) const {
57   assert(location().isFileID() && "must be a spelled token");
58   FileID File;
59   unsigned StartOffset;
60   std::tie(File, StartOffset) = SM.getDecomposedLoc(location());
61   return FileRange(File, StartOffset, StartOffset + length());
62 }
63 
64 FileRange syntax::Token::range(const SourceManager &SM,
65                                const syntax::Token &First,
66                                const syntax::Token &Last) {
67   auto F = First.range(SM);
68   auto L = Last.range(SM);
69   assert(F.file() == L.file() && "tokens from different files");
70   assert((F == L || F.endOffset() <= L.beginOffset()) &&
71          "wrong order of tokens");
72   return FileRange(F.file(), F.beginOffset(), L.endOffset());
73 }
74 
75 llvm::raw_ostream &syntax::operator<<(llvm::raw_ostream &OS, const Token &T) {
76   return OS << T.str();
77 }
78 
79 FileRange::FileRange(FileID File, unsigned BeginOffset, unsigned EndOffset)
80     : File(File), Begin(BeginOffset), End(EndOffset) {
81   assert(File.isValid());
82   assert(BeginOffset <= EndOffset);
83 }
84 
85 FileRange::FileRange(const SourceManager &SM, SourceLocation BeginLoc,
86                      unsigned Length) {
87   assert(BeginLoc.isValid());
88   assert(BeginLoc.isFileID());
89 
90   std::tie(File, Begin) = SM.getDecomposedLoc(BeginLoc);
91   End = Begin + Length;
92 }
93 FileRange::FileRange(const SourceManager &SM, SourceLocation BeginLoc,
94                      SourceLocation EndLoc) {
95   assert(BeginLoc.isValid());
96   assert(BeginLoc.isFileID());
97   assert(EndLoc.isValid());
98   assert(EndLoc.isFileID());
99   assert(SM.getFileID(BeginLoc) == SM.getFileID(EndLoc));
100   assert(SM.getFileOffset(BeginLoc) <= SM.getFileOffset(EndLoc));
101 
102   std::tie(File, Begin) = SM.getDecomposedLoc(BeginLoc);
103   End = SM.getFileOffset(EndLoc);
104 }
105 
106 llvm::raw_ostream &syntax::operator<<(llvm::raw_ostream &OS,
107                                       const FileRange &R) {
108   return OS << llvm::formatv("FileRange(file = {0}, offsets = {1}-{2})",
109                              R.file().getHashValue(), R.beginOffset(),
110                              R.endOffset());
111 }
112 
113 llvm::StringRef FileRange::text(const SourceManager &SM) const {
114   bool Invalid = false;
115   StringRef Text = SM.getBufferData(File, &Invalid);
116   if (Invalid)
117     return "";
118   assert(Begin <= Text.size());
119   assert(End <= Text.size());
120   return Text.substr(Begin, length());
121 }
122 
123 llvm::ArrayRef<syntax::Token> TokenBuffer::expandedTokens(SourceRange R) const {
124   if (R.isInvalid())
125     return {};
126   const Token *Begin =
127       llvm::partition_point(expandedTokens(), [&](const syntax::Token &T) {
128         return SourceMgr->isBeforeInTranslationUnit(T.location(), R.getBegin());
129       });
130   const Token *End =
131       llvm::partition_point(expandedTokens(), [&](const syntax::Token &T) {
132         return !SourceMgr->isBeforeInTranslationUnit(R.getEnd(), T.location());
133       });
134   if (Begin > End)
135     return {};
136   return {Begin, End};
137 }
138 
139 CharSourceRange FileRange::toCharRange(const SourceManager &SM) const {
140   return CharSourceRange(
141       SourceRange(SM.getComposedLoc(File, Begin), SM.getComposedLoc(File, End)),
142       /*IsTokenRange=*/false);
143 }
144 
145 std::pair<const syntax::Token *, const TokenBuffer::Mapping *>
146 TokenBuffer::spelledForExpandedToken(const syntax::Token *Expanded) const {
147   assert(Expanded);
148   assert(ExpandedTokens.data() <= Expanded &&
149          Expanded < ExpandedTokens.data() + ExpandedTokens.size());
150 
151   auto FileIt = Files.find(
152       SourceMgr->getFileID(SourceMgr->getExpansionLoc(Expanded->location())));
153   assert(FileIt != Files.end() && "no file for an expanded token");
154 
155   const MarkedFile &File = FileIt->second;
156 
157   unsigned ExpandedIndex = Expanded - ExpandedTokens.data();
158   // Find the first mapping that produced tokens after \p Expanded.
159   auto It = llvm::partition_point(File.Mappings, [&](const Mapping &M) {
160     return M.BeginExpanded <= ExpandedIndex;
161   });
162   // Our token could only be produced by the previous mapping.
163   if (It == File.Mappings.begin()) {
164     // No previous mapping, no need to modify offsets.
165     return {&File.SpelledTokens[ExpandedIndex - File.BeginExpanded], nullptr};
166   }
167   --It; // 'It' now points to last mapping that started before our token.
168 
169   // Check if the token is part of the mapping.
170   if (ExpandedIndex < It->EndExpanded)
171     return {&File.SpelledTokens[It->BeginSpelled], /*Mapping*/ &*It};
172 
173   // Not part of the mapping, use the index from previous mapping to compute the
174   // corresponding spelled token.
175   return {
176       &File.SpelledTokens[It->EndSpelled + (ExpandedIndex - It->EndExpanded)],
177       /*Mapping*/ nullptr};
178 }
179 
180 llvm::ArrayRef<syntax::Token> TokenBuffer::spelledTokens(FileID FID) const {
181   auto It = Files.find(FID);
182   assert(It != Files.end());
183   return It->second.SpelledTokens;
184 }
185 
186 std::string TokenBuffer::Mapping::str() const {
187   return std::string(
188       llvm::formatv("spelled tokens: [{0},{1}), expanded tokens: [{2},{3})",
189                     BeginSpelled, EndSpelled, BeginExpanded, EndExpanded));
190 }
191 
192 llvm::Optional<llvm::ArrayRef<syntax::Token>>
193 TokenBuffer::spelledForExpanded(llvm::ArrayRef<syntax::Token> Expanded) const {
194   // Mapping an empty range is ambiguous in case of empty mappings at either end
195   // of the range, bail out in that case.
196   if (Expanded.empty())
197     return llvm::None;
198 
199   // FIXME: also allow changes uniquely mapping to macro arguments.
200 
201   const syntax::Token *BeginSpelled;
202   const Mapping *BeginMapping;
203   std::tie(BeginSpelled, BeginMapping) =
204       spelledForExpandedToken(&Expanded.front());
205 
206   const syntax::Token *LastSpelled;
207   const Mapping *LastMapping;
208   std::tie(LastSpelled, LastMapping) =
209       spelledForExpandedToken(&Expanded.back());
210 
211   FileID FID = SourceMgr->getFileID(BeginSpelled->location());
212   // FIXME: Handle multi-file changes by trying to map onto a common root.
213   if (FID != SourceMgr->getFileID(LastSpelled->location()))
214     return llvm::None;
215 
216   const MarkedFile &File = Files.find(FID)->second;
217 
218   // Do not allow changes that cross macro expansion boundaries.
219   unsigned BeginExpanded = Expanded.begin() - ExpandedTokens.data();
220   unsigned EndExpanded = Expanded.end() - ExpandedTokens.data();
221   if (BeginMapping && BeginMapping->BeginExpanded < BeginExpanded)
222     return llvm::None;
223   if (LastMapping && EndExpanded < LastMapping->EndExpanded)
224     return llvm::None;
225   // All is good, return the result.
226   return llvm::makeArrayRef(
227       BeginMapping ? File.SpelledTokens.data() + BeginMapping->BeginSpelled
228                    : BeginSpelled,
229       LastMapping ? File.SpelledTokens.data() + LastMapping->EndSpelled
230                   : LastSpelled + 1);
231 }
232 
233 llvm::Optional<TokenBuffer::Expansion>
234 TokenBuffer::expansionStartingAt(const syntax::Token *Spelled) const {
235   assert(Spelled);
236   assert(Spelled->location().isFileID() && "not a spelled token");
237   auto FileIt = Files.find(SourceMgr->getFileID(Spelled->location()));
238   assert(FileIt != Files.end() && "file not tracked by token buffer");
239 
240   auto &File = FileIt->second;
241   assert(File.SpelledTokens.data() <= Spelled &&
242          Spelled < (File.SpelledTokens.data() + File.SpelledTokens.size()));
243 
244   unsigned SpelledIndex = Spelled - File.SpelledTokens.data();
245   auto M = llvm::partition_point(File.Mappings, [&](const Mapping &M) {
246     return M.BeginSpelled < SpelledIndex;
247   });
248   if (M == File.Mappings.end() || M->BeginSpelled != SpelledIndex)
249     return llvm::None;
250 
251   Expansion E;
252   E.Spelled = llvm::makeArrayRef(File.SpelledTokens.data() + M->BeginSpelled,
253                                  File.SpelledTokens.data() + M->EndSpelled);
254   E.Expanded = llvm::makeArrayRef(ExpandedTokens.data() + M->BeginExpanded,
255                                   ExpandedTokens.data() + M->EndExpanded);
256   return E;
257 }
258 llvm::ArrayRef<syntax::Token>
259 syntax::spelledTokensTouching(SourceLocation Loc,
260                               llvm::ArrayRef<syntax::Token> Tokens) {
261   assert(Loc.isFileID());
262 
263   auto *Right = llvm::partition_point(
264       Tokens, [&](const syntax::Token &Tok) { return Tok.location() < Loc; });
265   bool AcceptRight = Right != Tokens.end() && Right->location() <= Loc;
266   bool AcceptLeft =
267       Right != Tokens.begin() && (Right - 1)->endLocation() >= Loc;
268   return llvm::makeArrayRef(Right - (AcceptLeft ? 1 : 0),
269                             Right + (AcceptRight ? 1 : 0));
270 }
271 
272 llvm::ArrayRef<syntax::Token>
273 syntax::spelledTokensTouching(SourceLocation Loc,
274                               const syntax::TokenBuffer &Tokens) {
275   return spelledTokensTouching(
276       Loc, Tokens.spelledTokens(Tokens.sourceManager().getFileID(Loc)));
277 }
278 
279 const syntax::Token *
280 syntax::spelledIdentifierTouching(SourceLocation Loc,
281                                   llvm::ArrayRef<syntax::Token> Tokens) {
282   for (const syntax::Token &Tok : spelledTokensTouching(Loc, Tokens)) {
283     if (Tok.kind() == tok::identifier)
284       return &Tok;
285   }
286   return nullptr;
287 }
288 
289 const syntax::Token *
290 syntax::spelledIdentifierTouching(SourceLocation Loc,
291                                   const syntax::TokenBuffer &Tokens) {
292   return spelledIdentifierTouching(
293       Loc, Tokens.spelledTokens(Tokens.sourceManager().getFileID(Loc)));
294 }
295 
296 std::vector<const syntax::Token *>
297 TokenBuffer::macroExpansions(FileID FID) const {
298   auto FileIt = Files.find(FID);
299   assert(FileIt != Files.end() && "file not tracked by token buffer");
300   auto &File = FileIt->second;
301   std::vector<const syntax::Token *> Expansions;
302   auto &Spelled = File.SpelledTokens;
303   for (auto Mapping : File.Mappings) {
304     const syntax::Token *Token = &Spelled[Mapping.BeginSpelled];
305     if (Token->kind() == tok::TokenKind::identifier)
306       Expansions.push_back(Token);
307   }
308   return Expansions;
309 }
310 
311 std::vector<syntax::Token> syntax::tokenize(const FileRange &FR,
312                                             const SourceManager &SM,
313                                             const LangOptions &LO) {
314   std::vector<syntax::Token> Tokens;
315   IdentifierTable Identifiers(LO);
316   auto AddToken = [&](clang::Token T) {
317     // Fill the proper token kind for keywords, etc.
318     if (T.getKind() == tok::raw_identifier && !T.needsCleaning() &&
319         !T.hasUCN()) { // FIXME: support needsCleaning and hasUCN cases.
320       clang::IdentifierInfo &II = Identifiers.get(T.getRawIdentifier());
321       T.setIdentifierInfo(&II);
322       T.setKind(II.getTokenID());
323     }
324     Tokens.push_back(syntax::Token(T));
325   };
326 
327   auto SrcBuffer = SM.getBufferData(FR.file());
328   Lexer L(SM.getLocForStartOfFile(FR.file()), LO, SrcBuffer.data(),
329           SrcBuffer.data() + FR.beginOffset(),
330           // We can't make BufEnd point to FR.endOffset, as Lexer requires a
331           // null terminated buffer.
332           SrcBuffer.data() + SrcBuffer.size());
333 
334   clang::Token T;
335   while (!L.LexFromRawLexer(T) && L.getCurrentBufferOffset() < FR.endOffset())
336     AddToken(T);
337   // LexFromRawLexer returns true when it parses the last token of the file, add
338   // it iff it starts within the range we are interested in.
339   if (SM.getFileOffset(T.getLocation()) < FR.endOffset())
340     AddToken(T);
341   return Tokens;
342 }
343 
344 std::vector<syntax::Token> syntax::tokenize(FileID FID, const SourceManager &SM,
345                                             const LangOptions &LO) {
346   return tokenize(syntax::FileRange(FID, 0, SM.getFileIDSize(FID)), SM, LO);
347 }
348 
349 /// Records information reqired to construct mappings for the token buffer that
350 /// we are collecting.
351 class TokenCollector::CollectPPExpansions : public PPCallbacks {
352 public:
353   CollectPPExpansions(TokenCollector &C) : Collector(&C) {}
354 
355   /// Disabled instance will stop reporting anything to TokenCollector.
356   /// This ensures that uses of the preprocessor after TokenCollector::consume()
357   /// is called do not access the (possibly invalid) collector instance.
358   void disable() { Collector = nullptr; }
359 
360   void MacroExpands(const clang::Token &MacroNameTok, const MacroDefinition &MD,
361                     SourceRange Range, const MacroArgs *Args) override {
362     if (!Collector)
363       return;
364     // Only record top-level expansions, not those where:
365     //   - the macro use is inside a macro body,
366     //   - the macro appears in an argument to another macro.
367     if (!MacroNameTok.getLocation().isFileID() ||
368         (LastExpansionEnd.isValid() &&
369          Collector->PP.getSourceManager().isBeforeInTranslationUnit(
370              Range.getBegin(), LastExpansionEnd)))
371       return;
372     Collector->Expansions[Range.getBegin().getRawEncoding()] = Range.getEnd();
373     LastExpansionEnd = Range.getEnd();
374   }
375   // FIXME: handle directives like #pragma, #include, etc.
376 private:
377   TokenCollector *Collector;
378   /// Used to detect recursive macro expansions.
379   SourceLocation LastExpansionEnd;
380 };
381 
382 /// Fills in the TokenBuffer by tracing the run of a preprocessor. The
383 /// implementation tracks the tokens, macro expansions and directives coming
384 /// from the preprocessor and:
385 /// - for each token, figures out if it is a part of an expanded token stream,
386 ///   spelled token stream or both. Stores the tokens appropriately.
387 /// - records mappings from the spelled to expanded token ranges, e.g. for macro
388 ///   expansions.
389 /// FIXME: also properly record:
390 ///          - #include directives,
391 ///          - #pragma, #line and other PP directives,
392 ///          - skipped pp regions,
393 ///          - ...
394 
395 TokenCollector::TokenCollector(Preprocessor &PP) : PP(PP) {
396   // Collect the expanded token stream during preprocessing.
397   PP.setTokenWatcher([this](const clang::Token &T) {
398     if (T.isAnnotation())
399       return;
400     DEBUG_WITH_TYPE("collect-tokens", llvm::dbgs()
401                                           << "Token: "
402                                           << syntax::Token(T).dumpForTests(
403                                                  this->PP.getSourceManager())
404                                           << "\n"
405 
406     );
407     Expanded.push_back(syntax::Token(T));
408   });
409   // And locations of macro calls, to properly recover boundaries of those in
410   // case of empty expansions.
411   auto CB = std::make_unique<CollectPPExpansions>(*this);
412   this->Collector = CB.get();
413   PP.addPPCallbacks(std::move(CB));
414 }
415 
416 /// Builds mappings and spelled tokens in the TokenBuffer based on the expanded
417 /// token stream.
418 class TokenCollector::Builder {
419 public:
420   Builder(std::vector<syntax::Token> Expanded, PPExpansions CollectedExpansions,
421           const SourceManager &SM, const LangOptions &LangOpts)
422       : Result(SM), CollectedExpansions(std::move(CollectedExpansions)), SM(SM),
423         LangOpts(LangOpts) {
424     Result.ExpandedTokens = std::move(Expanded);
425   }
426 
427   TokenBuffer build() && {
428     buildSpelledTokens();
429 
430     // Walk over expanded tokens and spelled tokens in parallel, building the
431     // mappings between those using source locations.
432     // To correctly recover empty macro expansions, we also take locations
433     // reported to PPCallbacks::MacroExpands into account as we do not have any
434     // expanded tokens with source locations to guide us.
435 
436     // The 'eof' token is special, it is not part of spelled token stream. We
437     // handle it separately at the end.
438     assert(!Result.ExpandedTokens.empty());
439     assert(Result.ExpandedTokens.back().kind() == tok::eof);
440     for (unsigned I = 0; I < Result.ExpandedTokens.size() - 1; ++I) {
441       // (!) I might be updated by the following call.
442       processExpandedToken(I);
443     }
444 
445     // 'eof' not handled in the loop, do it here.
446     assert(SM.getMainFileID() ==
447            SM.getFileID(Result.ExpandedTokens.back().location()));
448     fillGapUntil(Result.Files[SM.getMainFileID()],
449                  Result.ExpandedTokens.back().location(),
450                  Result.ExpandedTokens.size() - 1);
451     Result.Files[SM.getMainFileID()].EndExpanded = Result.ExpandedTokens.size();
452 
453     // Some files might have unaccounted spelled tokens at the end, add an empty
454     // mapping for those as they did not have expanded counterparts.
455     fillGapsAtEndOfFiles();
456 
457     return std::move(Result);
458   }
459 
460 private:
461   /// Process the next token in an expanded stream and move corresponding
462   /// spelled tokens, record any mapping if needed.
463   /// (!) \p I will be updated if this had to skip tokens, e.g. for macros.
464   void processExpandedToken(unsigned &I) {
465     auto L = Result.ExpandedTokens[I].location();
466     if (L.isMacroID()) {
467       processMacroExpansion(SM.getExpansionRange(L), I);
468       return;
469     }
470     if (L.isFileID()) {
471       auto FID = SM.getFileID(L);
472       TokenBuffer::MarkedFile &File = Result.Files[FID];
473 
474       fillGapUntil(File, L, I);
475 
476       // Skip the token.
477       assert(File.SpelledTokens[NextSpelled[FID]].location() == L &&
478              "no corresponding token in the spelled stream");
479       ++NextSpelled[FID];
480       return;
481     }
482   }
483 
484   /// Skipped expanded and spelled tokens of a macro expansion that covers \p
485   /// SpelledRange. Add a corresponding mapping.
486   /// (!) \p I will be the index of the last token in an expansion after this
487   /// function returns.
488   void processMacroExpansion(CharSourceRange SpelledRange, unsigned &I) {
489     auto FID = SM.getFileID(SpelledRange.getBegin());
490     assert(FID == SM.getFileID(SpelledRange.getEnd()));
491     TokenBuffer::MarkedFile &File = Result.Files[FID];
492 
493     fillGapUntil(File, SpelledRange.getBegin(), I);
494 
495     // Skip all expanded tokens from the same macro expansion.
496     unsigned BeginExpanded = I;
497     for (; I + 1 < Result.ExpandedTokens.size(); ++I) {
498       auto NextL = Result.ExpandedTokens[I + 1].location();
499       if (!NextL.isMacroID() ||
500           SM.getExpansionLoc(NextL) != SpelledRange.getBegin())
501         break;
502     }
503     unsigned EndExpanded = I + 1;
504     consumeMapping(File, SM.getFileOffset(SpelledRange.getEnd()), BeginExpanded,
505                    EndExpanded, NextSpelled[FID]);
506   }
507 
508   /// Initializes TokenBuffer::Files and fills spelled tokens and expanded
509   /// ranges for each of the files.
510   void buildSpelledTokens() {
511     for (unsigned I = 0; I < Result.ExpandedTokens.size(); ++I) {
512       auto FID =
513           SM.getFileID(SM.getExpansionLoc(Result.ExpandedTokens[I].location()));
514       auto It = Result.Files.try_emplace(FID);
515       TokenBuffer::MarkedFile &File = It.first->second;
516 
517       File.EndExpanded = I + 1;
518       if (!It.second)
519         continue; // we have seen this file before.
520 
521       // This is the first time we see this file.
522       File.BeginExpanded = I;
523       File.SpelledTokens = tokenize(FID, SM, LangOpts);
524     }
525   }
526 
527   void consumeEmptyMapping(TokenBuffer::MarkedFile &File, unsigned EndOffset,
528                            unsigned ExpandedIndex, unsigned &SpelledIndex) {
529     consumeMapping(File, EndOffset, ExpandedIndex, ExpandedIndex, SpelledIndex);
530   }
531 
532   /// Consumes spelled tokens that form a macro expansion and adds a entry to
533   /// the resulting token buffer.
534   /// (!) SpelledIndex is updated in-place.
535   void consumeMapping(TokenBuffer::MarkedFile &File, unsigned EndOffset,
536                       unsigned BeginExpanded, unsigned EndExpanded,
537                       unsigned &SpelledIndex) {
538     // We need to record this mapping before continuing.
539     unsigned MappingBegin = SpelledIndex;
540     ++SpelledIndex;
541 
542     bool HitMapping =
543         tryConsumeSpelledUntil(File, EndOffset + 1, SpelledIndex).hasValue();
544     (void)HitMapping;
545     assert(!HitMapping && "recursive macro expansion?");
546 
547     TokenBuffer::Mapping M;
548     M.BeginExpanded = BeginExpanded;
549     M.EndExpanded = EndExpanded;
550     M.BeginSpelled = MappingBegin;
551     M.EndSpelled = SpelledIndex;
552 
553     File.Mappings.push_back(M);
554   }
555 
556   /// Consumes spelled tokens until location \p L is reached and adds a mapping
557   /// covering the consumed tokens. The mapping will point to an empty expanded
558   /// range at position \p ExpandedIndex.
559   void fillGapUntil(TokenBuffer::MarkedFile &File, SourceLocation L,
560                     unsigned ExpandedIndex) {
561     assert(L.isFileID());
562     FileID FID;
563     unsigned Offset;
564     std::tie(FID, Offset) = SM.getDecomposedLoc(L);
565 
566     unsigned &SpelledIndex = NextSpelled[FID];
567     unsigned MappingBegin = SpelledIndex;
568     while (true) {
569       auto EndLoc = tryConsumeSpelledUntil(File, Offset, SpelledIndex);
570       if (SpelledIndex != MappingBegin) {
571         TokenBuffer::Mapping M;
572         M.BeginSpelled = MappingBegin;
573         M.EndSpelled = SpelledIndex;
574         M.BeginExpanded = M.EndExpanded = ExpandedIndex;
575         File.Mappings.push_back(M);
576       }
577       if (!EndLoc)
578         break;
579       consumeEmptyMapping(File, SM.getFileOffset(*EndLoc), ExpandedIndex,
580                           SpelledIndex);
581 
582       MappingBegin = SpelledIndex;
583     }
584   };
585 
586   /// Consumes spelled tokens until it reaches Offset or a mapping boundary,
587   /// i.e. a name of a macro expansion or the start '#' token of a PP directive.
588   /// (!) NextSpelled is updated in place.
589   ///
590   /// returns None if \p Offset was reached, otherwise returns the end location
591   /// of a mapping that starts at \p NextSpelled.
592   llvm::Optional<SourceLocation>
593   tryConsumeSpelledUntil(TokenBuffer::MarkedFile &File, unsigned Offset,
594                          unsigned &NextSpelled) {
595     for (; NextSpelled < File.SpelledTokens.size(); ++NextSpelled) {
596       auto L = File.SpelledTokens[NextSpelled].location();
597       if (Offset <= SM.getFileOffset(L))
598         return llvm::None; // reached the offset we are looking for.
599       auto Mapping = CollectedExpansions.find(L.getRawEncoding());
600       if (Mapping != CollectedExpansions.end())
601         return Mapping->second; // found a mapping before the offset.
602     }
603     return llvm::None; // no more tokens, we "reached" the offset.
604   }
605 
606   /// Adds empty mappings for unconsumed spelled tokens at the end of each file.
607   void fillGapsAtEndOfFiles() {
608     for (auto &F : Result.Files) {
609       if (F.second.SpelledTokens.empty())
610         continue;
611       fillGapUntil(F.second, F.second.SpelledTokens.back().endLocation(),
612                    F.second.EndExpanded);
613     }
614   }
615 
616   TokenBuffer Result;
617   /// For each file, a position of the next spelled token we will consume.
618   llvm::DenseMap<FileID, unsigned> NextSpelled;
619   PPExpansions CollectedExpansions;
620   const SourceManager &SM;
621   const LangOptions &LangOpts;
622 };
623 
624 TokenBuffer TokenCollector::consume() && {
625   PP.setTokenWatcher(nullptr);
626   Collector->disable();
627   return Builder(std::move(Expanded), std::move(Expansions),
628                  PP.getSourceManager(), PP.getLangOpts())
629       .build();
630 }
631 
632 std::string syntax::Token::str() const {
633   return std::string(llvm::formatv("Token({0}, length = {1})",
634                                    tok::getTokenName(kind()), length()));
635 }
636 
637 std::string syntax::Token::dumpForTests(const SourceManager &SM) const {
638   return std::string(
639       llvm::formatv("{0}   {1}", tok::getTokenName(kind()), text(SM)));
640 }
641 
642 std::string TokenBuffer::dumpForTests() const {
643   auto PrintToken = [this](const syntax::Token &T) -> std::string {
644     if (T.kind() == tok::eof)
645       return "<eof>";
646     return std::string(T.text(*SourceMgr));
647   };
648 
649   auto DumpTokens = [this, &PrintToken](llvm::raw_ostream &OS,
650                                         llvm::ArrayRef<syntax::Token> Tokens) {
651     if (Tokens.empty()) {
652       OS << "<empty>";
653       return;
654     }
655     OS << Tokens[0].text(*SourceMgr);
656     for (unsigned I = 1; I < Tokens.size(); ++I) {
657       if (Tokens[I].kind() == tok::eof)
658         continue;
659       OS << " " << PrintToken(Tokens[I]);
660     }
661   };
662 
663   std::string Dump;
664   llvm::raw_string_ostream OS(Dump);
665 
666   OS << "expanded tokens:\n"
667      << "  ";
668   // (!) we do not show '<eof>'.
669   DumpTokens(OS, llvm::makeArrayRef(ExpandedTokens).drop_back());
670   OS << "\n";
671 
672   std::vector<FileID> Keys;
673   for (auto F : Files)
674     Keys.push_back(F.first);
675   llvm::sort(Keys);
676 
677   for (FileID ID : Keys) {
678     const MarkedFile &File = Files.find(ID)->second;
679     auto *Entry = SourceMgr->getFileEntryForID(ID);
680     if (!Entry)
681       continue; // Skip builtin files.
682     OS << llvm::formatv("file '{0}'\n", Entry->getName())
683        << "  spelled tokens:\n"
684        << "    ";
685     DumpTokens(OS, File.SpelledTokens);
686     OS << "\n";
687 
688     if (File.Mappings.empty()) {
689       OS << "  no mappings.\n";
690       continue;
691     }
692     OS << "  mappings:\n";
693     for (auto &M : File.Mappings) {
694       OS << llvm::formatv(
695           "    ['{0}'_{1}, '{2}'_{3}) => ['{4}'_{5}, '{6}'_{7})\n",
696           PrintToken(File.SpelledTokens[M.BeginSpelled]), M.BeginSpelled,
697           M.EndSpelled == File.SpelledTokens.size()
698               ? "<eof>"
699               : PrintToken(File.SpelledTokens[M.EndSpelled]),
700           M.EndSpelled, PrintToken(ExpandedTokens[M.BeginExpanded]),
701           M.BeginExpanded, PrintToken(ExpandedTokens[M.EndExpanded]),
702           M.EndExpanded);
703     }
704   }
705   return OS.str();
706 }
707