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/STLExtras.h"
22#include "llvm/Support/Debug.h"
23#include "llvm/Support/ErrorHandling.h"
24#include "llvm/Support/FormatVariadic.h"
25#include "llvm/Support/raw_ostream.h"
26#include <algorithm>
27#include <cassert>
28#include <iterator>
29#include <optional>
30#include <string>
31#include <utility>
32#include <vector>
33
34using namespace clang;
35using namespace clang::syntax;
36
37namespace {
38// Finds the smallest consecutive subsuquence of Toks that covers R.
39llvm::ArrayRef<syntax::Token>
40getTokensCovering(llvm::ArrayRef<syntax::Token> Toks, SourceRange R,
41 const SourceManager &SM) {
42 if (R.isInvalid())
43 return {};
44 const syntax::Token *Begin =
45 llvm::partition_point(Range&: Toks, P: [&](const syntax::Token &T) {
46 return SM.isBeforeInTranslationUnit(LHS: T.location(), RHS: R.getBegin());
47 });
48 const syntax::Token *End =
49 llvm::partition_point(Range&: Toks, P: [&](const syntax::Token &T) {
50 return !SM.isBeforeInTranslationUnit(LHS: R.getEnd(), RHS: T.location());
51 });
52 if (Begin > End)
53 return {};
54 return {Begin, End};
55}
56
57// Finds the range within FID corresponding to expanded tokens [First, Last].
58// Prev precedes First and Next follows Last, these must *not* be included.
59// If no range satisfies the criteria, returns an invalid range.
60//
61// #define ID(x) x
62// ID(ID(ID(a1) a2))
63// ~~ -> a1
64// ~~ -> a2
65// ~~~~~~~~~ -> a1 a2
66SourceRange spelledForExpandedSlow(SourceLocation First, SourceLocation Last,
67 SourceLocation Prev, SourceLocation Next,
68 FileID TargetFile,
69 const SourceManager &SM) {
70 // There are two main parts to this algorithm:
71 // - identifying which spelled range covers the expanded tokens
72 // - validating that this range doesn't cover any extra tokens (First/Last)
73 //
74 // We do these in order. However as we transform the expanded range into the
75 // spelled one, we adjust First/Last so the validation remains simple.
76
77 assert(SM.getSLocEntry(TargetFile).isFile());
78 // In most cases, to select First and Last we must return their expansion
79 // range, i.e. the whole of any macros they are included in.
80 //
81 // When First and Last are part of the *same macro arg* of a macro written
82 // in TargetFile, we that slice of the arg, i.e. their spelling range.
83 //
84 // Unwrap such macro calls. If the target file has A(B(C)), the
85 // SourceLocation stack of a token inside C shows us the expansion of A first,
86 // then B, then any macros inside C's body, then C itself.
87 // (This is the reverse of the order the PP applies the expansions in).
88 while (First.isMacroID() && Last.isMacroID()) {
89 auto DecFirst = SM.getDecomposedLoc(Loc: First);
90 auto DecLast = SM.getDecomposedLoc(Loc: Last);
91 auto &ExpFirst = SM.getSLocEntry(FID: DecFirst.first).getExpansion();
92 auto &ExpLast = SM.getSLocEntry(FID: DecLast.first).getExpansion();
93
94 if (!ExpFirst.isMacroArgExpansion() || !ExpLast.isMacroArgExpansion())
95 break;
96 // Locations are in the same macro arg if they expand to the same place.
97 // (They may still have different FileIDs - an arg can have >1 chunks!)
98 if (ExpFirst.getExpansionLocStart() != ExpLast.getExpansionLocStart())
99 break;
100 // Careful, given:
101 // #define HIDE ID(ID(a))
102 // ID(ID(HIDE))
103 // The token `a` is wrapped in 4 arg-expansions, we only want to unwrap 2.
104 // We distinguish them by whether the macro expands into the target file.
105 // Fortunately, the target file ones will always appear first.
106 auto ExpFileID = SM.getFileID(SpellingLoc: ExpFirst.getExpansionLocStart());
107 if (ExpFileID == TargetFile)
108 break;
109 // Replace each endpoint with its spelling inside the macro arg.
110 // (This is getImmediateSpellingLoc without repeating lookups).
111 First = ExpFirst.getSpellingLoc().getLocWithOffset(Offset: DecFirst.second);
112 Last = ExpLast.getSpellingLoc().getLocWithOffset(Offset: DecLast.second);
113 }
114
115 // In all remaining cases we need the full containing macros.
116 // If this overlaps Prev or Next, then no range is possible.
117 SourceRange Candidate =
118 SM.getExpansionRange(Range: SourceRange(First, Last)).getAsRange();
119 auto DecFirst = SM.getDecomposedExpansionLoc(Loc: Candidate.getBegin());
120 auto DecLast = SM.getDecomposedExpansionLoc(Loc: Candidate.getEnd());
121 // Can end up in the wrong file due to bad input or token-pasting shenanigans.
122 if (Candidate.isInvalid() || DecFirst.first != TargetFile ||
123 DecLast.first != TargetFile)
124 return SourceRange();
125 // Check bounds, which may still be inside macros.
126 if (Prev.isValid()) {
127 auto Dec = SM.getDecomposedLoc(Loc: SM.getExpansionRange(Loc: Prev).getBegin());
128 if (Dec.first != DecFirst.first || Dec.second >= DecFirst.second)
129 return SourceRange();
130 }
131 if (Next.isValid()) {
132 auto Dec = SM.getDecomposedLoc(Loc: SM.getExpansionRange(Loc: Next).getEnd());
133 if (Dec.first != DecLast.first || Dec.second <= DecLast.second)
134 return SourceRange();
135 }
136 // Now we know that Candidate is a file range that covers [First, Last]
137 // without encroaching on {Prev, Next}. Ship it!
138 return Candidate;
139}
140
141} // namespace
142
143syntax::Token::Token(SourceLocation Location, unsigned Length,
144 tok::TokenKind Kind)
145 : Location(Location), Length(Length), Kind(Kind) {
146 assert(Location.isValid());
147}
148
149syntax::Token::Token(const clang::Token &T)
150 : Token(T.getLocation(), T.getLength(), T.getKind()) {
151 assert(!T.isAnnotation());
152}
153
154llvm::StringRef syntax::Token::text(const SourceManager &SM) const {
155 bool Invalid = false;
156 const char *Start = SM.getCharacterData(SL: location(), Invalid: &Invalid);
157 assert(!Invalid);
158 return llvm::StringRef(Start, length());
159}
160
161FileRange syntax::Token::range(const SourceManager &SM) const {
162 assert(location().isFileID() && "must be a spelled token");
163 FileID File;
164 unsigned StartOffset;
165 std::tie(args&: File, args&: StartOffset) = SM.getDecomposedLoc(Loc: location());
166 return FileRange(File, StartOffset, StartOffset + length());
167}
168
169FileRange syntax::Token::range(const SourceManager &SM,
170 const syntax::Token &First,
171 const syntax::Token &Last) {
172 auto F = First.range(SM);
173 auto L = Last.range(SM);
174 assert(F.file() == L.file() && "tokens from different files");
175 assert((F == L || F.endOffset() <= L.beginOffset()) &&
176 "wrong order of tokens");
177 return FileRange(F.file(), F.beginOffset(), L.endOffset());
178}
179
180llvm::raw_ostream &syntax::operator<<(llvm::raw_ostream &OS, const Token &T) {
181 return OS << T.str();
182}
183
184FileRange::FileRange(FileID File, unsigned BeginOffset, unsigned EndOffset)
185 : File(File), Begin(BeginOffset), End(EndOffset) {
186 assert(File.isValid());
187 assert(BeginOffset <= EndOffset);
188}
189
190FileRange::FileRange(const SourceManager &SM, SourceLocation BeginLoc,
191 unsigned Length) {
192 assert(BeginLoc.isValid());
193 assert(BeginLoc.isFileID());
194
195 std::tie(args&: File, args&: Begin) = SM.getDecomposedLoc(Loc: BeginLoc);
196 End = Begin + Length;
197}
198FileRange::FileRange(const SourceManager &SM, SourceLocation BeginLoc,
199 SourceLocation EndLoc) {
200 assert(BeginLoc.isValid());
201 assert(BeginLoc.isFileID());
202 assert(EndLoc.isValid());
203 assert(EndLoc.isFileID());
204 assert(SM.getFileID(BeginLoc) == SM.getFileID(EndLoc));
205 assert(SM.getFileOffset(BeginLoc) <= SM.getFileOffset(EndLoc));
206
207 std::tie(args&: File, args&: Begin) = SM.getDecomposedLoc(Loc: BeginLoc);
208 End = SM.getFileOffset(SpellingLoc: EndLoc);
209}
210
211llvm::raw_ostream &syntax::operator<<(llvm::raw_ostream &OS,
212 const FileRange &R) {
213 return OS << llvm::formatv(Fmt: "FileRange(file = {0}, offsets = {1}-{2})",
214 Vals: R.file().getHashValue(), Vals: R.beginOffset(),
215 Vals: R.endOffset());
216}
217
218llvm::StringRef FileRange::text(const SourceManager &SM) const {
219 bool Invalid = false;
220 StringRef Text = SM.getBufferData(FID: File, Invalid: &Invalid);
221 if (Invalid)
222 return "";
223 assert(Begin <= Text.size());
224 assert(End <= Text.size());
225 return Text.substr(Start: Begin, N: length());
226}
227
228void TokenBuffer::indexExpandedTokens() {
229 // No-op if the index is already created.
230 if (!ExpandedTokIndex.empty())
231 return;
232 ExpandedTokIndex.reserve(NumEntries: ExpandedTokens.size());
233 // Index ExpandedTokens for faster lookups by SourceLocation.
234 for (size_t I = 0, E = ExpandedTokens.size(); I != E; ++I) {
235 SourceLocation Loc = ExpandedTokens[I].location();
236 if (Loc.isValid())
237 ExpandedTokIndex[Loc] = I;
238 }
239}
240
241llvm::ArrayRef<syntax::Token> TokenBuffer::expandedTokens(SourceRange R) const {
242 if (R.isInvalid())
243 return {};
244 if (!ExpandedTokIndex.empty()) {
245 // Quick lookup if `R` is a token range.
246 // This is a huge win since majority of the users use ranges provided by an
247 // AST. Ranges in AST are token ranges from expanded token stream.
248 const auto B = ExpandedTokIndex.find(Val: R.getBegin());
249 const auto E = ExpandedTokIndex.find(Val: R.getEnd());
250 if (B != ExpandedTokIndex.end() && E != ExpandedTokIndex.end()) {
251 const Token *L = ExpandedTokens.data() + B->getSecond();
252 // Add 1 to End to make a half-open range.
253 const Token *R = ExpandedTokens.data() + E->getSecond() + 1;
254 if (L > R)
255 return {};
256 return {L, R};
257 }
258 }
259 // Slow case. Use `isBeforeInTranslationUnit` to binary search for the
260 // required range.
261 return getTokensCovering(Toks: expandedTokens(), R, SM: *SourceMgr);
262}
263
264CharSourceRange FileRange::toCharRange(const SourceManager &SM) const {
265 return CharSourceRange(
266 SourceRange(SM.getComposedLoc(FID: File, Offset: Begin), SM.getComposedLoc(FID: File, Offset: End)),
267 /*IsTokenRange=*/false);
268}
269
270std::pair<const syntax::Token *, const TokenBuffer::Mapping *>
271TokenBuffer::spelledForExpandedToken(const syntax::Token *Expanded) const {
272 assert(Expanded);
273 assert(ExpandedTokens.data() <= Expanded &&
274 Expanded < ExpandedTokens.data() + ExpandedTokens.size());
275
276 auto FileIt = Files.find(
277 Val: SourceMgr->getFileID(SpellingLoc: SourceMgr->getExpansionLoc(Loc: Expanded->location())));
278 assert(FileIt != Files.end() && "no file for an expanded token");
279
280 const MarkedFile &File = FileIt->second;
281
282 unsigned ExpandedIndex = Expanded - ExpandedTokens.data();
283 // Find the first mapping that produced tokens after \p Expanded.
284 auto It = llvm::partition_point(Range: File.Mappings, P: [&](const Mapping &M) {
285 return M.BeginExpanded <= ExpandedIndex;
286 });
287 // Our token could only be produced by the previous mapping.
288 if (It == File.Mappings.begin()) {
289 // No previous mapping, no need to modify offsets.
290 return {&File.SpelledTokens[ExpandedIndex - File.BeginExpanded],
291 /*Mapping=*/nullptr};
292 }
293 --It; // 'It' now points to last mapping that started before our token.
294
295 // Check if the token is part of the mapping.
296 if (ExpandedIndex < It->EndExpanded)
297 return {&File.SpelledTokens[It->BeginSpelled], /*Mapping=*/&*It};
298
299 // Not part of the mapping, use the index from previous mapping to compute the
300 // corresponding spelled token.
301 return {
302 &File.SpelledTokens[It->EndSpelled + (ExpandedIndex - It->EndExpanded)],
303 /*Mapping=*/nullptr};
304}
305
306const TokenBuffer::Mapping *
307TokenBuffer::mappingStartingBeforeSpelled(const MarkedFile &F,
308 const syntax::Token *Spelled) {
309 assert(F.SpelledTokens.data() <= Spelled);
310 unsigned SpelledI = Spelled - F.SpelledTokens.data();
311 assert(SpelledI < F.SpelledTokens.size());
312
313 auto It = llvm::partition_point(Range: F.Mappings, P: [SpelledI](const Mapping &M) {
314 return M.BeginSpelled <= SpelledI;
315 });
316 if (It == F.Mappings.begin())
317 return nullptr;
318 --It;
319 return &*It;
320}
321
322llvm::SmallVector<llvm::ArrayRef<syntax::Token>, 1>
323TokenBuffer::expandedForSpelled(llvm::ArrayRef<syntax::Token> Spelled) const {
324 if (Spelled.empty())
325 return {};
326 const auto &File = fileForSpelled(Spelled);
327
328 auto *FrontMapping = mappingStartingBeforeSpelled(F: File, Spelled: &Spelled.front());
329 unsigned SpelledFrontI = &Spelled.front() - File.SpelledTokens.data();
330 assert(SpelledFrontI < File.SpelledTokens.size());
331 unsigned ExpandedBegin;
332 if (!FrontMapping) {
333 // No mapping that starts before the first token of Spelled, we don't have
334 // to modify offsets.
335 ExpandedBegin = File.BeginExpanded + SpelledFrontI;
336 } else if (SpelledFrontI < FrontMapping->EndSpelled) {
337 // This mapping applies to Spelled tokens.
338 if (SpelledFrontI != FrontMapping->BeginSpelled) {
339 // Spelled tokens don't cover the entire mapping, returning empty result.
340 return {}; // FIXME: support macro arguments.
341 }
342 // Spelled tokens start at the beginning of this mapping.
343 ExpandedBegin = FrontMapping->BeginExpanded;
344 } else {
345 // Spelled tokens start after the mapping ends (they start in the hole
346 // between 2 mappings, or between a mapping and end of the file).
347 ExpandedBegin =
348 FrontMapping->EndExpanded + (SpelledFrontI - FrontMapping->EndSpelled);
349 }
350
351 auto *BackMapping = mappingStartingBeforeSpelled(F: File, Spelled: &Spelled.back());
352 unsigned SpelledBackI = &Spelled.back() - File.SpelledTokens.data();
353 unsigned ExpandedEnd;
354 if (!BackMapping) {
355 // No mapping that starts before the last token of Spelled, we don't have to
356 // modify offsets.
357 ExpandedEnd = File.BeginExpanded + SpelledBackI + 1;
358 } else if (SpelledBackI < BackMapping->EndSpelled) {
359 // This mapping applies to Spelled tokens.
360 if (SpelledBackI + 1 != BackMapping->EndSpelled) {
361 // Spelled tokens don't cover the entire mapping, returning empty result.
362 return {}; // FIXME: support macro arguments.
363 }
364 ExpandedEnd = BackMapping->EndExpanded;
365 } else {
366 // Spelled tokens end after the mapping ends.
367 ExpandedEnd =
368 BackMapping->EndExpanded + (SpelledBackI - BackMapping->EndSpelled) + 1;
369 }
370
371 assert(ExpandedBegin < ExpandedTokens.size());
372 assert(ExpandedEnd < ExpandedTokens.size());
373 // Avoid returning empty ranges.
374 if (ExpandedBegin == ExpandedEnd)
375 return {};
376 return {llvm::ArrayRef(ExpandedTokens.data() + ExpandedBegin,
377 ExpandedTokens.data() + ExpandedEnd)};
378}
379
380llvm::ArrayRef<syntax::Token> TokenBuffer::spelledTokens(FileID FID) const {
381 auto It = Files.find(Val: FID);
382 assert(It != Files.end());
383 return It->second.SpelledTokens;
384}
385
386const syntax::Token *TokenBuffer::spelledTokenAt(SourceLocation Loc) const {
387 assert(Loc.isFileID());
388 const auto *Tok = llvm::partition_point(
389 Range: spelledTokens(FID: SourceMgr->getFileID(SpellingLoc: Loc)),
390 P: [&](const syntax::Token &Tok) { return Tok.location() < Loc; });
391 if (!Tok || Tok->location() != Loc)
392 return nullptr;
393 return Tok;
394}
395
396std::string TokenBuffer::Mapping::str() const {
397 return std::string(
398 llvm::formatv(Fmt: "spelled tokens: [{0},{1}), expanded tokens: [{2},{3})",
399 Vals: BeginSpelled, Vals: EndSpelled, Vals: BeginExpanded, Vals: EndExpanded));
400}
401
402std::optional<llvm::ArrayRef<syntax::Token>>
403TokenBuffer::spelledForExpanded(llvm::ArrayRef<syntax::Token> Expanded) const {
404 // In cases of invalid code, AST nodes can have source ranges that include
405 // the `eof` token. As there's no spelling for this token, exclude it from
406 // the range.
407 if (!Expanded.empty() && Expanded.back().kind() == tok::eof) {
408 Expanded = Expanded.drop_back();
409 }
410 // Mapping an empty range is ambiguous in case of empty mappings at either end
411 // of the range, bail out in that case.
412 if (Expanded.empty())
413 return std::nullopt;
414 const syntax::Token *First = &Expanded.front();
415 const syntax::Token *Last = &Expanded.back();
416 auto [FirstSpelled, FirstMapping] = spelledForExpandedToken(Expanded: First);
417 auto [LastSpelled, LastMapping] = spelledForExpandedToken(Expanded: Last);
418
419 FileID FID = SourceMgr->getFileID(SpellingLoc: FirstSpelled->location());
420 // FIXME: Handle multi-file changes by trying to map onto a common root.
421 if (FID != SourceMgr->getFileID(SpellingLoc: LastSpelled->location()))
422 return std::nullopt;
423
424 const MarkedFile &File = Files.find(Val: FID)->second;
425
426 // If the range is within one macro argument, the result may be only part of a
427 // Mapping. We must use the general (SourceManager-based) algorithm.
428 if (FirstMapping && FirstMapping == LastMapping &&
429 SourceMgr->isMacroArgExpansion(Loc: First->location()) &&
430 SourceMgr->isMacroArgExpansion(Loc: Last->location())) {
431 // We use excluded Prev/Next token for bounds checking.
432 SourceLocation Prev = (First == &ExpandedTokens.front())
433 ? SourceLocation()
434 : (First - 1)->location();
435 SourceLocation Next = (Last == &ExpandedTokens.back())
436 ? SourceLocation()
437 : (Last + 1)->location();
438 SourceRange Range = spelledForExpandedSlow(
439 First: First->location(), Last: Last->location(), Prev, Next, TargetFile: FID, SM: *SourceMgr);
440 if (Range.isInvalid())
441 return std::nullopt;
442 return getTokensCovering(Toks: File.SpelledTokens, R: Range, SM: *SourceMgr);
443 }
444
445 // Otherwise, use the fast version based on Mappings.
446 // Do not allow changes that doesn't cover full expansion.
447 unsigned FirstExpanded = Expanded.begin() - ExpandedTokens.data();
448 unsigned LastExpanded = Expanded.end() - ExpandedTokens.data();
449 if (FirstMapping && FirstExpanded != FirstMapping->BeginExpanded)
450 return std::nullopt;
451 if (LastMapping && LastMapping->EndExpanded != LastExpanded)
452 return std::nullopt;
453 return llvm::ArrayRef(
454 FirstMapping ? File.SpelledTokens.data() + FirstMapping->BeginSpelled
455 : FirstSpelled,
456 LastMapping ? File.SpelledTokens.data() + LastMapping->EndSpelled
457 : LastSpelled + 1);
458}
459
460TokenBuffer::Expansion TokenBuffer::makeExpansion(const MarkedFile &F,
461 const Mapping &M) const {
462 Expansion E;
463 E.Spelled = llvm::ArrayRef(F.SpelledTokens.data() + M.BeginSpelled,
464 F.SpelledTokens.data() + M.EndSpelled);
465 E.Expanded = llvm::ArrayRef(ExpandedTokens.data() + M.BeginExpanded,
466 ExpandedTokens.data() + M.EndExpanded);
467 return E;
468}
469
470const TokenBuffer::MarkedFile &
471TokenBuffer::fileForSpelled(llvm::ArrayRef<syntax::Token> Spelled) const {
472 assert(!Spelled.empty());
473 assert(Spelled.front().location().isFileID() && "not a spelled token");
474 auto FileIt = Files.find(Val: SourceMgr->getFileID(SpellingLoc: Spelled.front().location()));
475 assert(FileIt != Files.end() && "file not tracked by token buffer");
476 const auto &File = FileIt->second;
477 assert(File.SpelledTokens.data() <= Spelled.data() &&
478 Spelled.end() <=
479 (File.SpelledTokens.data() + File.SpelledTokens.size()) &&
480 "Tokens not in spelled range");
481#ifndef NDEBUG
482 auto T1 = Spelled.back().location();
483 auto T2 = File.SpelledTokens.back().location();
484 assert(T1 == T2 || sourceManager().isBeforeInTranslationUnit(T1, T2));
485#endif
486 return File;
487}
488
489std::optional<TokenBuffer::Expansion>
490TokenBuffer::expansionStartingAt(const syntax::Token *Spelled) const {
491 assert(Spelled);
492 const auto &File = fileForSpelled(Spelled: *Spelled);
493
494 unsigned SpelledIndex = Spelled - File.SpelledTokens.data();
495 auto M = llvm::partition_point(Range: File.Mappings, P: [&](const Mapping &M) {
496 return M.BeginSpelled < SpelledIndex;
497 });
498 if (M == File.Mappings.end() || M->BeginSpelled != SpelledIndex)
499 return std::nullopt;
500 return makeExpansion(F: File, M: *M);
501}
502
503std::vector<TokenBuffer::Expansion> TokenBuffer::expansionsOverlapping(
504 llvm::ArrayRef<syntax::Token> Spelled) const {
505 if (Spelled.empty())
506 return {};
507 const auto &File = fileForSpelled(Spelled);
508
509 // Find the first overlapping range, and then copy until we stop overlapping.
510 unsigned SpelledBeginIndex = Spelled.begin() - File.SpelledTokens.data();
511 unsigned SpelledEndIndex = Spelled.end() - File.SpelledTokens.data();
512 auto M = llvm::partition_point(Range: File.Mappings, P: [&](const Mapping &M) {
513 return M.EndSpelled <= SpelledBeginIndex;
514 });
515 std::vector<TokenBuffer::Expansion> Expansions;
516 for (; M != File.Mappings.end() && M->BeginSpelled < SpelledEndIndex; ++M)
517 Expansions.push_back(x: makeExpansion(F: File, M: *M));
518 return Expansions;
519}
520
521llvm::ArrayRef<syntax::Token>
522syntax::spelledTokensTouching(SourceLocation Loc,
523 llvm::ArrayRef<syntax::Token> Tokens) {
524 assert(Loc.isFileID());
525
526 auto *Right = llvm::partition_point(
527 Range&: Tokens, P: [&](const syntax::Token &Tok) { return Tok.location() < Loc; });
528 bool AcceptRight = Right != Tokens.end() && Right->location() <= Loc;
529 bool AcceptLeft =
530 Right != Tokens.begin() && (Right - 1)->endLocation() >= Loc;
531 return llvm::ArrayRef(Right - (AcceptLeft ? 1 : 0),
532 Right + (AcceptRight ? 1 : 0));
533}
534
535llvm::ArrayRef<syntax::Token>
536syntax::spelledTokensTouching(SourceLocation Loc,
537 const syntax::TokenBuffer &Tokens) {
538 return spelledTokensTouching(
539 Loc, Tokens: Tokens.spelledTokens(FID: Tokens.sourceManager().getFileID(SpellingLoc: Loc)));
540}
541
542const syntax::Token *
543syntax::spelledIdentifierTouching(SourceLocation Loc,
544 llvm::ArrayRef<syntax::Token> Tokens) {
545 for (const syntax::Token &Tok : spelledTokensTouching(Loc, Tokens)) {
546 if (Tok.kind() == tok::identifier)
547 return &Tok;
548 }
549 return nullptr;
550}
551
552const syntax::Token *
553syntax::spelledIdentifierTouching(SourceLocation Loc,
554 const syntax::TokenBuffer &Tokens) {
555 return spelledIdentifierTouching(
556 Loc, Tokens: Tokens.spelledTokens(FID: Tokens.sourceManager().getFileID(SpellingLoc: Loc)));
557}
558
559std::vector<const syntax::Token *>
560TokenBuffer::macroExpansions(FileID FID) const {
561 auto FileIt = Files.find(Val: FID);
562 assert(FileIt != Files.end() && "file not tracked by token buffer");
563 auto &File = FileIt->second;
564 std::vector<const syntax::Token *> Expansions;
565 auto &Spelled = File.SpelledTokens;
566 for (auto Mapping : File.Mappings) {
567 const syntax::Token *Token = &Spelled[Mapping.BeginSpelled];
568 if (Token->kind() == tok::TokenKind::identifier)
569 Expansions.push_back(x: Token);
570 }
571 return Expansions;
572}
573
574std::vector<syntax::Token> syntax::tokenize(const FileRange &FR,
575 const SourceManager &SM,
576 const LangOptions &LO) {
577 std::vector<syntax::Token> Tokens;
578 IdentifierTable Identifiers(LO);
579 auto AddToken = [&](clang::Token T) {
580 // Fill the proper token kind for keywords, etc.
581 if (T.getKind() == tok::raw_identifier && !T.needsCleaning() &&
582 !T.hasUCN()) { // FIXME: support needsCleaning and hasUCN cases.
583 clang::IdentifierInfo &II = Identifiers.get(Name: T.getRawIdentifier());
584 T.setIdentifierInfo(&II);
585 T.setKind(II.getTokenID());
586 }
587 Tokens.push_back(x: syntax::Token(T));
588 };
589
590 auto SrcBuffer = SM.getBufferData(FID: FR.file());
591 Lexer L(SM.getLocForStartOfFile(FID: FR.file()), LO, SrcBuffer.data(),
592 SrcBuffer.data() + FR.beginOffset(),
593 // We can't make BufEnd point to FR.endOffset, as Lexer requires a
594 // null terminated buffer.
595 SrcBuffer.data() + SrcBuffer.size());
596
597 clang::Token T;
598 while (!L.LexFromRawLexer(Result&: T) && L.getCurrentBufferOffset() < FR.endOffset())
599 AddToken(T);
600 // LexFromRawLexer returns true when it parses the last token of the file, add
601 // it iff it starts within the range we are interested in.
602 if (SM.getFileOffset(SpellingLoc: T.getLocation()) < FR.endOffset())
603 AddToken(T);
604 return Tokens;
605}
606
607std::vector<syntax::Token> syntax::tokenize(FileID FID, const SourceManager &SM,
608 const LangOptions &LO) {
609 return tokenize(FR: syntax::FileRange(FID, 0, SM.getFileIDSize(FID)), SM, LO);
610}
611
612/// Records information reqired to construct mappings for the token buffer that
613/// we are collecting.
614class TokenCollector::CollectPPExpansions : public PPCallbacks {
615public:
616 CollectPPExpansions(TokenCollector &C) : Collector(&C) {}
617
618 /// Disabled instance will stop reporting anything to TokenCollector.
619 /// This ensures that uses of the preprocessor after TokenCollector::consume()
620 /// is called do not access the (possibly invalid) collector instance.
621 void disable() { Collector = nullptr; }
622
623 void MacroExpands(const clang::Token &MacroNameTok, const MacroDefinition &MD,
624 SourceRange Range, const MacroArgs *Args) override {
625 if (!Collector)
626 return;
627 const auto &SM = Collector->PP.getSourceManager();
628 // Only record top-level expansions that directly produce expanded tokens.
629 // This excludes those where:
630 // - the macro use is inside a macro body,
631 // - the macro appears in an argument to another macro.
632 // However macro expansion isn't really a tree, it's token rewrite rules,
633 // so there are other cases, e.g.
634 // #define B(X) X
635 // #define A 1 + B
636 // A(2)
637 // Both A and B produce expanded tokens, though the macro name 'B' comes
638 // from an expansion. The best we can do is merge the mappings for both.
639
640 // The *last* token of any top-level macro expansion must be in a file.
641 // (In the example above, see the closing paren of the expansion of B).
642 if (!Range.getEnd().isFileID())
643 return;
644 // If there's a current expansion that encloses this one, this one can't be
645 // top-level.
646 if (LastExpansionEnd.isValid() &&
647 !SM.isBeforeInTranslationUnit(LHS: LastExpansionEnd, RHS: Range.getEnd()))
648 return;
649
650 // If the macro invocation (B) starts in a macro (A) but ends in a file,
651 // we'll create a merged mapping for A + B by overwriting the endpoint for
652 // A's startpoint.
653 if (!Range.getBegin().isFileID()) {
654 Range.setBegin(SM.getExpansionLoc(Loc: Range.getBegin()));
655 assert(Collector->Expansions.count(Range.getBegin()) &&
656 "Overlapping macros should have same expansion location");
657 }
658
659 Collector->Expansions[Range.getBegin()] = Range.getEnd();
660 LastExpansionEnd = Range.getEnd();
661 }
662 // FIXME: handle directives like #pragma, #include, etc.
663private:
664 TokenCollector *Collector;
665 /// Used to detect recursive macro expansions.
666 SourceLocation LastExpansionEnd;
667};
668
669/// Fills in the TokenBuffer by tracing the run of a preprocessor. The
670/// implementation tracks the tokens, macro expansions and directives coming
671/// from the preprocessor and:
672/// - for each token, figures out if it is a part of an expanded token stream,
673/// spelled token stream or both. Stores the tokens appropriately.
674/// - records mappings from the spelled to expanded token ranges, e.g. for macro
675/// expansions.
676/// FIXME: also properly record:
677/// - #include directives,
678/// - #pragma, #line and other PP directives,
679/// - skipped pp regions,
680/// - ...
681
682TokenCollector::TokenCollector(Preprocessor &PP) : PP(PP) {
683 // Collect the expanded token stream during preprocessing.
684 PP.setTokenWatcher([this](const clang::Token &T) {
685 if (T.isAnnotation())
686 return;
687 DEBUG_WITH_TYPE("collect-tokens", llvm::dbgs()
688 << "Token: "
689 << syntax::Token(T).dumpForTests(
690 this->PP.getSourceManager())
691 << "\n"
692
693 );
694 Expanded.push_back(x: syntax::Token(T));
695 });
696 // And locations of macro calls, to properly recover boundaries of those in
697 // case of empty expansions.
698 auto CB = std::make_unique<CollectPPExpansions>(args&: *this);
699 this->Collector = CB.get();
700 PP.addPPCallbacks(C: std::move(CB));
701}
702
703/// Builds mappings and spelled tokens in the TokenBuffer based on the expanded
704/// token stream.
705class TokenCollector::Builder {
706public:
707 Builder(std::vector<syntax::Token> Expanded, PPExpansions CollectedExpansions,
708 const SourceManager &SM, const LangOptions &LangOpts)
709 : Result(SM), CollectedExpansions(std::move(CollectedExpansions)), SM(SM),
710 LangOpts(LangOpts) {
711 Result.ExpandedTokens = std::move(Expanded);
712 }
713
714 TokenBuffer build() && {
715 assert(!Result.ExpandedTokens.empty());
716 assert(Result.ExpandedTokens.back().kind() == tok::eof);
717
718 // Tokenize every file that contributed tokens to the expanded stream.
719 buildSpelledTokens();
720
721 // The expanded token stream consists of runs of tokens that came from
722 // the same source (a macro expansion, part of a file etc).
723 // Between these runs are the logical positions of spelled tokens that
724 // didn't expand to anything.
725 while (NextExpanded < Result.ExpandedTokens.size() - 1 /* eof */) {
726 // Create empty mappings for spelled tokens that expanded to nothing here.
727 // May advance NextSpelled, but NextExpanded is unchanged.
728 discard();
729 // Create mapping for a contiguous run of expanded tokens.
730 // Advances NextExpanded past the run, and NextSpelled accordingly.
731 unsigned OldPosition = NextExpanded;
732 advance();
733 if (NextExpanded == OldPosition)
734 diagnoseAdvanceFailure();
735 }
736 // If any tokens remain in any of the files, they didn't expand to anything.
737 // Create empty mappings up until the end of the file.
738 for (const auto &File : Result.Files)
739 discard(Drain: File.first);
740
741#ifndef NDEBUG
742 for (auto &pair : Result.Files) {
743 auto &mappings = pair.second.Mappings;
744 assert(llvm::is_sorted(mappings, [](const TokenBuffer::Mapping &M1,
745 const TokenBuffer::Mapping &M2) {
746 return M1.BeginSpelled < M2.BeginSpelled &&
747 M1.EndSpelled < M2.EndSpelled &&
748 M1.BeginExpanded < M2.BeginExpanded &&
749 M1.EndExpanded < M2.EndExpanded;
750 }));
751 }
752#endif
753
754 return std::move(Result);
755 }
756
757private:
758 // Consume a sequence of spelled tokens that didn't expand to anything.
759 // In the simplest case, skips spelled tokens until finding one that produced
760 // the NextExpanded token, and creates an empty mapping for them.
761 // If Drain is provided, skips remaining tokens from that file instead.
762 void discard(std::optional<FileID> Drain = std::nullopt) {
763 SourceLocation Target =
764 Drain ? SM.getLocForEndOfFile(FID: *Drain)
765 : SM.getExpansionLoc(
766 Loc: Result.ExpandedTokens[NextExpanded].location());
767 FileID File = SM.getFileID(SpellingLoc: Target);
768 const auto &SpelledTokens = Result.Files[File].SpelledTokens;
769 auto &NextSpelled = this->NextSpelled[File];
770
771 TokenBuffer::Mapping Mapping;
772 Mapping.BeginSpelled = NextSpelled;
773 // When dropping trailing tokens from a file, the empty mapping should
774 // be positioned within the file's expanded-token range (at the end).
775 Mapping.BeginExpanded = Mapping.EndExpanded =
776 Drain ? Result.Files[*Drain].EndExpanded : NextExpanded;
777 // We may want to split into several adjacent empty mappings.
778 // FlushMapping() emits the current mapping and starts a new one.
779 auto FlushMapping = [&, this] {
780 Mapping.EndSpelled = NextSpelled;
781 if (Mapping.BeginSpelled != Mapping.EndSpelled)
782 Result.Files[File].Mappings.push_back(x: Mapping);
783 Mapping.BeginSpelled = NextSpelled;
784 };
785
786 while (NextSpelled < SpelledTokens.size() &&
787 SpelledTokens[NextSpelled].location() < Target) {
788 // If we know mapping bounds at [NextSpelled, KnownEnd] (macro expansion)
789 // then we want to partition our (empty) mapping.
790 // [Start, NextSpelled) [NextSpelled, KnownEnd] (KnownEnd, Target)
791 SourceLocation KnownEnd =
792 CollectedExpansions.lookup(Val: SpelledTokens[NextSpelled].location());
793 if (KnownEnd.isValid()) {
794 FlushMapping(); // Emits [Start, NextSpelled)
795 while (NextSpelled < SpelledTokens.size() &&
796 SpelledTokens[NextSpelled].location() <= KnownEnd)
797 ++NextSpelled;
798 FlushMapping(); // Emits [NextSpelled, KnownEnd]
799 // Now the loop continues and will emit (KnownEnd, Target).
800 } else {
801 ++NextSpelled;
802 }
803 }
804 FlushMapping();
805 }
806
807 // Consumes the NextExpanded token and others that are part of the same run.
808 // Increases NextExpanded and NextSpelled by at least one, and adds a mapping
809 // (unless this is a run of file tokens, which we represent with no mapping).
810 void advance() {
811 const syntax::Token &Tok = Result.ExpandedTokens[NextExpanded];
812 SourceLocation Expansion = SM.getExpansionLoc(Loc: Tok.location());
813 FileID File = SM.getFileID(SpellingLoc: Expansion);
814 const auto &SpelledTokens = Result.Files[File].SpelledTokens;
815 auto &NextSpelled = this->NextSpelled[File];
816
817 if (Tok.location().isFileID()) {
818 // A run of file tokens continues while the expanded/spelled tokens match.
819 while (NextSpelled < SpelledTokens.size() &&
820 NextExpanded < Result.ExpandedTokens.size() &&
821 SpelledTokens[NextSpelled].location() ==
822 Result.ExpandedTokens[NextExpanded].location()) {
823 ++NextSpelled;
824 ++NextExpanded;
825 }
826 // We need no mapping for file tokens copied to the expanded stream.
827 } else {
828 // We found a new macro expansion. We should have its spelling bounds.
829 auto End = CollectedExpansions.lookup(Val: Expansion);
830 assert(End.isValid() && "Macro expansion wasn't captured?");
831
832 // Mapping starts here...
833 TokenBuffer::Mapping Mapping;
834 Mapping.BeginExpanded = NextExpanded;
835 Mapping.BeginSpelled = NextSpelled;
836 // ... consumes spelled tokens within bounds we captured ...
837 while (NextSpelled < SpelledTokens.size() &&
838 SpelledTokens[NextSpelled].location() <= End)
839 ++NextSpelled;
840 // ... consumes expanded tokens rooted at the same expansion ...
841 while (NextExpanded < Result.ExpandedTokens.size() &&
842 SM.getExpansionLoc(
843 Loc: Result.ExpandedTokens[NextExpanded].location()) == Expansion)
844 ++NextExpanded;
845 // ... and ends here.
846 Mapping.EndExpanded = NextExpanded;
847 Mapping.EndSpelled = NextSpelled;
848 Result.Files[File].Mappings.push_back(x: Mapping);
849 }
850 }
851
852 // advance() is supposed to consume at least one token - if not, we crash.
853 void diagnoseAdvanceFailure() {
854#ifndef NDEBUG
855 // Show the failed-to-map token in context.
856 for (unsigned I = (NextExpanded < 10) ? 0 : NextExpanded - 10;
857 I < NextExpanded + 5 && I < Result.ExpandedTokens.size(); ++I) {
858 const char *L =
859 (I == NextExpanded) ? "!! " : (I < NextExpanded) ? "ok " : " ";
860 llvm::errs() << L << Result.ExpandedTokens[I].dumpForTests(SM) << "\n";
861 }
862#endif
863 llvm_unreachable("Couldn't map expanded token to spelled tokens!");
864 }
865
866 /// Initializes TokenBuffer::Files and fills spelled tokens and expanded
867 /// ranges for each of the files.
868 void buildSpelledTokens() {
869 for (unsigned I = 0; I < Result.ExpandedTokens.size(); ++I) {
870 const auto &Tok = Result.ExpandedTokens[I];
871 auto FID = SM.getFileID(SpellingLoc: SM.getExpansionLoc(Loc: Tok.location()));
872 auto It = Result.Files.try_emplace(Key: FID);
873 TokenBuffer::MarkedFile &File = It.first->second;
874
875 // The eof token should not be considered part of the main-file's range.
876 File.EndExpanded = Tok.kind() == tok::eof ? I : I + 1;
877
878 if (!It.second)
879 continue; // we have seen this file before.
880 // This is the first time we see this file.
881 File.BeginExpanded = I;
882 File.SpelledTokens = tokenize(FID, SM, LO: LangOpts);
883 }
884 }
885
886 TokenBuffer Result;
887 unsigned NextExpanded = 0; // cursor in ExpandedTokens
888 llvm::DenseMap<FileID, unsigned> NextSpelled; // cursor in SpelledTokens
889 PPExpansions CollectedExpansions;
890 const SourceManager &SM;
891 const LangOptions &LangOpts;
892};
893
894TokenBuffer TokenCollector::consume() && {
895 PP.setTokenWatcher(nullptr);
896 Collector->disable();
897 return Builder(std::move(Expanded), std::move(Expansions),
898 PP.getSourceManager(), PP.getLangOpts())
899 .build();
900}
901
902std::string syntax::Token::str() const {
903 return std::string(llvm::formatv(Fmt: "Token({0}, length = {1})",
904 Vals: tok::getTokenName(Kind: kind()), Vals: length()));
905}
906
907std::string syntax::Token::dumpForTests(const SourceManager &SM) const {
908 return std::string(llvm::formatv(Fmt: "Token(`{0}`, {1}, length = {2})", Vals: text(SM),
909 Vals: tok::getTokenName(Kind: kind()), Vals: length()));
910}
911
912std::string TokenBuffer::dumpForTests() const {
913 auto PrintToken = [this](const syntax::Token &T) -> std::string {
914 if (T.kind() == tok::eof)
915 return "<eof>";
916 return std::string(T.text(SM: *SourceMgr));
917 };
918
919 auto DumpTokens = [this, &PrintToken](llvm::raw_ostream &OS,
920 llvm::ArrayRef<syntax::Token> Tokens) {
921 if (Tokens.empty()) {
922 OS << "<empty>";
923 return;
924 }
925 OS << Tokens[0].text(SM: *SourceMgr);
926 for (unsigned I = 1; I < Tokens.size(); ++I) {
927 if (Tokens[I].kind() == tok::eof)
928 continue;
929 OS << " " << PrintToken(Tokens[I]);
930 }
931 };
932
933 std::string Dump;
934 llvm::raw_string_ostream OS(Dump);
935
936 OS << "expanded tokens:\n"
937 << " ";
938 // (!) we do not show '<eof>'.
939 DumpTokens(OS, llvm::ArrayRef(ExpandedTokens).drop_back());
940 OS << "\n";
941
942 std::vector<FileID> Keys;
943 for (const auto &F : Files)
944 Keys.push_back(x: F.first);
945 llvm::sort(C&: Keys);
946
947 for (FileID ID : Keys) {
948 const MarkedFile &File = Files.find(Val: ID)->second;
949 auto Entry = SourceMgr->getFileEntryRefForID(FID: ID);
950 if (!Entry)
951 continue; // Skip builtin files.
952 std::string Path = llvm::sys::path::convert_to_slash(path: Entry->getName());
953 OS << llvm::formatv(Fmt: "file '{0}'\n", Vals&: Path) << " spelled tokens:\n"
954 << " ";
955 DumpTokens(OS, File.SpelledTokens);
956 OS << "\n";
957
958 if (File.Mappings.empty()) {
959 OS << " no mappings.\n";
960 continue;
961 }
962 OS << " mappings:\n";
963 for (auto &M : File.Mappings) {
964 OS << llvm::formatv(
965 Fmt: " ['{0}'_{1}, '{2}'_{3}) => ['{4}'_{5}, '{6}'_{7})\n",
966 Vals: PrintToken(File.SpelledTokens[M.BeginSpelled]), Vals: M.BeginSpelled,
967 Vals: M.EndSpelled == File.SpelledTokens.size()
968 ? "<eof>"
969 : PrintToken(File.SpelledTokens[M.EndSpelled]),
970 Vals: M.EndSpelled, Vals: PrintToken(ExpandedTokens[M.BeginExpanded]),
971 Vals: M.BeginExpanded, Vals: PrintToken(ExpandedTokens[M.EndExpanded]),
972 Vals: M.EndExpanded);
973 }
974 }
975 return Dump;
976}
977

source code of clang/lib/Tooling/Syntax/Tokens.cpp