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
22namespace clang {
23namespace format {
24
25FormatTokenLexer::FormatTokenLexer(
26 const SourceManager &SourceMgr, FileID ID, unsigned Column,
27 const FormatStyle &Style, encoding::Encoding Encoding,
28 llvm::SpecificBumpPtrAllocator<FormatToken> &Allocator,
29 IdentifierTable &IdentTable)
30 : FormatTok(nullptr), IsFirstToken(true), StateStack({LexerState::NORMAL}),
31 Column(Column), TrailingWhitespace(0),
32 LangOpts(getFormattingLangOpts(Style)), SourceMgr(SourceMgr), ID(ID),
33 Style(Style), IdentTable(IdentTable), Keywords(IdentTable),
34 Encoding(Encoding), Allocator(Allocator), FirstInLineIndex(0),
35 FormattingDisabled(false), MacroBlockBeginRegex(Style.MacroBlockBegin),
36 MacroBlockEndRegex(Style.MacroBlockEnd) {
37 Lex.reset(p: new Lexer(ID, SourceMgr.getBufferOrFake(FID: ID), SourceMgr, LangOpts));
38 Lex->SetKeepWhitespaceMode(true);
39
40 for (const std::string &ForEachMacro : Style.ForEachMacros) {
41 auto Identifier = &IdentTable.get(Name: ForEachMacro);
42 Macros.insert(KV: {Identifier, TT_ForEachMacro});
43 }
44 for (const std::string &IfMacro : Style.IfMacros) {
45 auto Identifier = &IdentTable.get(Name: IfMacro);
46 Macros.insert(KV: {Identifier, TT_IfMacro});
47 }
48 for (const std::string &AttributeMacro : Style.AttributeMacros) {
49 auto Identifier = &IdentTable.get(Name: AttributeMacro);
50 Macros.insert(KV: {Identifier, TT_AttributeMacro});
51 }
52 for (const std::string &StatementMacro : Style.StatementMacros) {
53 auto Identifier = &IdentTable.get(Name: StatementMacro);
54 Macros.insert(KV: {Identifier, TT_StatementMacro});
55 }
56 for (const std::string &TypenameMacro : Style.TypenameMacros) {
57 auto Identifier = &IdentTable.get(Name: TypenameMacro);
58 Macros.insert(KV: {Identifier, TT_TypenameMacro});
59 }
60 for (const std::string &NamespaceMacro : Style.NamespaceMacros) {
61 auto Identifier = &IdentTable.get(Name: NamespaceMacro);
62 Macros.insert(KV: {Identifier, TT_NamespaceMacro});
63 }
64 for (const std::string &WhitespaceSensitiveMacro :
65 Style.WhitespaceSensitiveMacros) {
66 auto Identifier = &IdentTable.get(Name: WhitespaceSensitiveMacro);
67 Macros.insert(KV: {Identifier, TT_UntouchableMacroFunc});
68 }
69 for (const std::string &StatementAttributeLikeMacro :
70 Style.StatementAttributeLikeMacros) {
71 auto Identifier = &IdentTable.get(Name: StatementAttributeLikeMacro);
72 Macros.insert(KV: {Identifier, TT_StatementAttributeLikeMacro});
73 }
74
75 for (const auto &TypeName : Style.TypeNames)
76 TypeNames.insert(Ptr: &IdentTable.get(Name: TypeName));
77}
78
79ArrayRef<FormatToken *> FormatTokenLexer::lex() {
80 assert(Tokens.empty());
81 assert(FirstInLineIndex == 0);
82 do {
83 Tokens.push_back(Elt: getNextToken());
84 if (Style.isJavaScript()) {
85 tryParseJSRegexLiteral();
86 handleTemplateStrings();
87 }
88 if (Style.Language == FormatStyle::LK_TextProto)
89 tryParsePythonComment();
90 tryMergePreviousTokens();
91 if (Style.isCSharp()) {
92 // This needs to come after tokens have been merged so that C#
93 // string literals are correctly identified.
94 handleCSharpVerbatimAndInterpolatedStrings();
95 }
96 if (Style.isTableGen()) {
97 handleTableGenMultilineString();
98 handleTableGenNumericLikeIdentifier();
99 }
100 if (Tokens.back()->NewlinesBefore > 0 || Tokens.back()->IsMultiline)
101 FirstInLineIndex = Tokens.size() - 1;
102 } while (Tokens.back()->isNot(Kind: tok::eof));
103 return Tokens;
104}
105
106void FormatTokenLexer::tryMergePreviousTokens() {
107 if (tryMerge_TMacro())
108 return;
109 if (tryMergeConflictMarkers())
110 return;
111 if (tryMergeLessLess())
112 return;
113 if (tryMergeGreaterGreater())
114 return;
115 if (tryMergeForEach())
116 return;
117 if (Style.isCpp() && tryTransformTryUsageForC())
118 return;
119
120 if (Style.isJavaScript() || Style.isCSharp()) {
121 static const tok::TokenKind NullishCoalescingOperator[] = {tok::question,
122 tok::question};
123 static const tok::TokenKind NullPropagatingOperator[] = {tok::question,
124 tok::period};
125 static const tok::TokenKind FatArrow[] = {tok::equal, tok::greater};
126
127 if (tryMergeTokens(Kinds: FatArrow, NewType: TT_FatArrow))
128 return;
129 if (tryMergeTokens(Kinds: NullishCoalescingOperator, NewType: TT_NullCoalescingOperator)) {
130 // Treat like the "||" operator (as opposed to the ternary ?).
131 Tokens.back()->Tok.setKind(tok::pipepipe);
132 return;
133 }
134 if (tryMergeTokens(Kinds: NullPropagatingOperator, NewType: TT_NullPropagatingOperator)) {
135 // Treat like a regular "." access.
136 Tokens.back()->Tok.setKind(tok::period);
137 return;
138 }
139 if (tryMergeNullishCoalescingEqual())
140 return;
141 }
142
143 if (Style.isCSharp()) {
144 static const tok::TokenKind CSharpNullConditionalLSquare[] = {
145 tok::question, tok::l_square};
146
147 if (tryMergeCSharpKeywordVariables())
148 return;
149 if (tryMergeCSharpStringLiteral())
150 return;
151 if (tryTransformCSharpForEach())
152 return;
153 if (tryMergeTokens(Kinds: CSharpNullConditionalLSquare,
154 NewType: TT_CSharpNullConditionalLSquare)) {
155 // Treat like a regular "[" operator.
156 Tokens.back()->Tok.setKind(tok::l_square);
157 return;
158 }
159 }
160
161 if (tryMergeNSStringLiteral())
162 return;
163
164 if (Style.isJavaScript()) {
165 static const tok::TokenKind JSIdentity[] = {tok::equalequal, tok::equal};
166 static const tok::TokenKind JSNotIdentity[] = {tok::exclaimequal,
167 tok::equal};
168 static const tok::TokenKind JSShiftEqual[] = {tok::greater, tok::greater,
169 tok::greaterequal};
170 static const tok::TokenKind JSExponentiation[] = {tok::star, tok::star};
171 static const tok::TokenKind JSExponentiationEqual[] = {tok::star,
172 tok::starequal};
173 static const tok::TokenKind JSPipePipeEqual[] = {tok::pipepipe, tok::equal};
174 static const tok::TokenKind JSAndAndEqual[] = {tok::ampamp, tok::equal};
175
176 // FIXME: Investigate what token type gives the correct operator priority.
177 if (tryMergeTokens(Kinds: JSIdentity, NewType: TT_BinaryOperator))
178 return;
179 if (tryMergeTokens(Kinds: JSNotIdentity, NewType: TT_BinaryOperator))
180 return;
181 if (tryMergeTokens(Kinds: JSShiftEqual, NewType: TT_BinaryOperator))
182 return;
183 if (tryMergeTokens(Kinds: JSExponentiation, NewType: TT_JsExponentiation))
184 return;
185 if (tryMergeTokens(Kinds: JSExponentiationEqual, NewType: TT_JsExponentiationEqual)) {
186 Tokens.back()->Tok.setKind(tok::starequal);
187 return;
188 }
189 if (tryMergeTokens(Kinds: JSAndAndEqual, NewType: TT_JsAndAndEqual) ||
190 tryMergeTokens(Kinds: JSPipePipeEqual, NewType: TT_JsPipePipeEqual)) {
191 // Treat like the "=" assignment operator.
192 Tokens.back()->Tok.setKind(tok::equal);
193 return;
194 }
195 if (tryMergeJSPrivateIdentifier())
196 return;
197 }
198
199 if (Style.Language == FormatStyle::LK_Java) {
200 static const tok::TokenKind JavaRightLogicalShiftAssign[] = {
201 tok::greater, tok::greater, tok::greaterequal};
202 if (tryMergeTokens(Kinds: JavaRightLogicalShiftAssign, NewType: TT_BinaryOperator))
203 return;
204 }
205
206 if (Style.isVerilog()) {
207 // Merge the number following a base like `'h?a0`.
208 if (Tokens.size() >= 3 && Tokens.end()[-3]->is(TT: TT_VerilogNumberBase) &&
209 Tokens.end()[-2]->is(Kind: tok::numeric_constant) &&
210 Tokens.back()->isOneOf(K1: tok::numeric_constant, K2: tok::identifier,
211 Ks: tok::question) &&
212 tryMergeTokens(Count: 2, NewType: TT_Unknown)) {
213 return;
214 }
215 // Part select.
216 if (tryMergeTokensAny(Kinds: {{tok::minus, tok::colon}, {tok::plus, tok::colon}},
217 NewType: TT_BitFieldColon)) {
218 return;
219 }
220 // Xnor. The combined token is treated as a caret which can also be either a
221 // unary or binary operator. The actual type is determined in
222 // TokenAnnotator. We also check the token length so we know it is not
223 // already a merged token.
224 if (Tokens.back()->TokenText.size() == 1 &&
225 tryMergeTokensAny(Kinds: {{tok::caret, tok::tilde}, {tok::tilde, tok::caret}},
226 NewType: TT_BinaryOperator)) {
227 Tokens.back()->Tok.setKind(tok::caret);
228 return;
229 }
230 // Signed shift and distribution weight.
231 if (tryMergeTokens(Kinds: {tok::less, tok::less}, NewType: TT_BinaryOperator)) {
232 Tokens.back()->Tok.setKind(tok::lessless);
233 return;
234 }
235 if (tryMergeTokens(Kinds: {tok::greater, tok::greater}, NewType: TT_BinaryOperator)) {
236 Tokens.back()->Tok.setKind(tok::greatergreater);
237 return;
238 }
239 if (tryMergeTokensAny(Kinds: {{tok::lessless, tok::equal},
240 {tok::lessless, tok::lessequal},
241 {tok::greatergreater, tok::equal},
242 {tok::greatergreater, tok::greaterequal},
243 {tok::colon, tok::equal},
244 {tok::colon, tok::slash}},
245 NewType: TT_BinaryOperator)) {
246 Tokens.back()->ForcedPrecedence = prec::Assignment;
247 return;
248 }
249 // Exponentiation, signed shift, case equality, and wildcard equality.
250 if (tryMergeTokensAny(Kinds: {{tok::star, tok::star},
251 {tok::lessless, tok::less},
252 {tok::greatergreater, tok::greater},
253 {tok::exclaimequal, tok::equal},
254 {tok::exclaimequal, tok::question},
255 {tok::equalequal, tok::equal},
256 {tok::equalequal, tok::question}},
257 NewType: TT_BinaryOperator)) {
258 return;
259 }
260 // Module paths in specify blocks and the implication and boolean equality
261 // operators.
262 if (tryMergeTokensAny(Kinds: {{tok::plusequal, tok::greater},
263 {tok::plus, tok::star, tok::greater},
264 {tok::minusequal, tok::greater},
265 {tok::minus, tok::star, tok::greater},
266 {tok::less, tok::arrow},
267 {tok::equal, tok::greater},
268 {tok::star, tok::greater},
269 {tok::pipeequal, tok::greater},
270 {tok::pipe, tok::arrow},
271 {tok::hash, tok::minus, tok::hash},
272 {tok::hash, tok::equal, tok::hash}},
273 NewType: TT_BinaryOperator) ||
274 Tokens.back()->is(Kind: tok::arrow)) {
275 Tokens.back()->ForcedPrecedence = prec::Comma;
276 return;
277 }
278 }
279 if (Style.isTableGen()) {
280 // TableGen's Multi line string starts with [{
281 if (tryMergeTokens(Kinds: {tok::l_square, tok::l_brace},
282 NewType: TT_TableGenMultiLineString)) {
283 // Set again with finalizing. This must never be annotated as other types.
284 Tokens.back()->setFinalizedType(TT_TableGenMultiLineString);
285 Tokens.back()->Tok.setKind(tok::string_literal);
286 return;
287 }
288 // TableGen's bang operator is the form !<name>.
289 // !cond is a special case with specific syntax.
290 if (tryMergeTokens(Kinds: {tok::exclaim, tok::identifier},
291 NewType: TT_TableGenBangOperator)) {
292 Tokens.back()->Tok.setKind(tok::identifier);
293 Tokens.back()->Tok.setIdentifierInfo(nullptr);
294 if (Tokens.back()->TokenText == "!cond")
295 Tokens.back()->setFinalizedType(TT_TableGenCondOperator);
296 else
297 Tokens.back()->setFinalizedType(TT_TableGenBangOperator);
298 return;
299 }
300 if (tryMergeTokens(Kinds: {tok::exclaim, tok::kw_if}, NewType: TT_TableGenBangOperator)) {
301 // Here, "! if" becomes "!if". That is, ! captures if even when the space
302 // exists. That is only one possibility in TableGen's syntax.
303 Tokens.back()->Tok.setKind(tok::identifier);
304 Tokens.back()->Tok.setIdentifierInfo(nullptr);
305 Tokens.back()->setFinalizedType(TT_TableGenBangOperator);
306 return;
307 }
308 // +, - with numbers are literals. Not unary operators.
309 if (tryMergeTokens(Kinds: {tok::plus, tok::numeric_constant}, NewType: TT_Unknown)) {
310 Tokens.back()->Tok.setKind(tok::numeric_constant);
311 return;
312 }
313 if (tryMergeTokens(Kinds: {tok::minus, tok::numeric_constant}, NewType: TT_Unknown)) {
314 Tokens.back()->Tok.setKind(tok::numeric_constant);
315 return;
316 }
317 }
318}
319
320bool FormatTokenLexer::tryMergeNSStringLiteral() {
321 if (Tokens.size() < 2)
322 return false;
323 auto &At = *(Tokens.end() - 2);
324 auto &String = *(Tokens.end() - 1);
325 if (At->isNot(Kind: tok::at) || String->isNot(Kind: tok::string_literal))
326 return false;
327 At->Tok.setKind(tok::string_literal);
328 At->TokenText = StringRef(At->TokenText.begin(),
329 String->TokenText.end() - At->TokenText.begin());
330 At->ColumnWidth += String->ColumnWidth;
331 At->setType(TT_ObjCStringLiteral);
332 Tokens.erase(CI: Tokens.end() - 1);
333 return true;
334}
335
336bool FormatTokenLexer::tryMergeJSPrivateIdentifier() {
337 // Merges #idenfier into a single identifier with the text #identifier
338 // but the token tok::identifier.
339 if (Tokens.size() < 2)
340 return false;
341 auto &Hash = *(Tokens.end() - 2);
342 auto &Identifier = *(Tokens.end() - 1);
343 if (Hash->isNot(Kind: tok::hash) || Identifier->isNot(Kind: tok::identifier))
344 return false;
345 Hash->Tok.setKind(tok::identifier);
346 Hash->TokenText =
347 StringRef(Hash->TokenText.begin(),
348 Identifier->TokenText.end() - Hash->TokenText.begin());
349 Hash->ColumnWidth += Identifier->ColumnWidth;
350 Hash->setType(TT_JsPrivateIdentifier);
351 Tokens.erase(CI: Tokens.end() - 1);
352 return true;
353}
354
355// Search for verbatim or interpolated string literals @"ABC" or
356// $"aaaaa{abc}aaaaa" i and mark the token as TT_CSharpStringLiteral, and to
357// prevent splitting of @, $ and ".
358// Merging of multiline verbatim strings with embedded '"' is handled in
359// handleCSharpVerbatimAndInterpolatedStrings with lower-level lexing.
360bool FormatTokenLexer::tryMergeCSharpStringLiteral() {
361 if (Tokens.size() < 2)
362 return false;
363
364 // Look for @"aaaaaa" or $"aaaaaa".
365 const auto String = *(Tokens.end() - 1);
366 if (String->isNot(Kind: tok::string_literal))
367 return false;
368
369 auto Prefix = *(Tokens.end() - 2);
370 if (Prefix->isNot(Kind: tok::at) && Prefix->TokenText != "$")
371 return false;
372
373 if (Tokens.size() > 2) {
374 const auto Tok = *(Tokens.end() - 3);
375 if ((Tok->TokenText == "$" && Prefix->is(Kind: tok::at)) ||
376 (Tok->is(Kind: tok::at) && Prefix->TokenText == "$")) {
377 // This looks like $@"aaa" or @$"aaa" so we need to combine all 3 tokens.
378 Tok->ColumnWidth += Prefix->ColumnWidth;
379 Tokens.erase(CI: Tokens.end() - 2);
380 Prefix = Tok;
381 }
382 }
383
384 // Convert back into just a string_literal.
385 Prefix->Tok.setKind(tok::string_literal);
386 Prefix->TokenText =
387 StringRef(Prefix->TokenText.begin(),
388 String->TokenText.end() - Prefix->TokenText.begin());
389 Prefix->ColumnWidth += String->ColumnWidth;
390 Prefix->setType(TT_CSharpStringLiteral);
391 Tokens.erase(CI: Tokens.end() - 1);
392 return true;
393}
394
395// Valid C# attribute targets:
396// https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/concepts/attributes/#attribute-targets
397const llvm::StringSet<> FormatTokenLexer::CSharpAttributeTargets = {
398 "assembly", "module", "field", "event", "method",
399 "param", "property", "return", "type",
400};
401
402bool FormatTokenLexer::tryMergeNullishCoalescingEqual() {
403 if (Tokens.size() < 2)
404 return false;
405 auto &NullishCoalescing = *(Tokens.end() - 2);
406 auto &Equal = *(Tokens.end() - 1);
407 if (NullishCoalescing->isNot(Kind: TT_NullCoalescingOperator) ||
408 Equal->isNot(Kind: tok::equal)) {
409 return false;
410 }
411 NullishCoalescing->Tok.setKind(tok::equal); // no '??=' in clang tokens.
412 NullishCoalescing->TokenText =
413 StringRef(NullishCoalescing->TokenText.begin(),
414 Equal->TokenText.end() - NullishCoalescing->TokenText.begin());
415 NullishCoalescing->ColumnWidth += Equal->ColumnWidth;
416 NullishCoalescing->setType(TT_NullCoalescingEqual);
417 Tokens.erase(CI: Tokens.end() - 1);
418 return true;
419}
420
421bool FormatTokenLexer::tryMergeCSharpKeywordVariables() {
422 if (Tokens.size() < 2)
423 return false;
424 const auto At = *(Tokens.end() - 2);
425 if (At->isNot(Kind: tok::at))
426 return false;
427 const auto Keyword = *(Tokens.end() - 1);
428 if (Keyword->TokenText == "$")
429 return false;
430 if (!Keywords.isCSharpKeyword(Tok: *Keyword))
431 return false;
432
433 At->Tok.setKind(tok::identifier);
434 At->TokenText = StringRef(At->TokenText.begin(),
435 Keyword->TokenText.end() - At->TokenText.begin());
436 At->ColumnWidth += Keyword->ColumnWidth;
437 At->setType(Keyword->getType());
438 Tokens.erase(CI: Tokens.end() - 1);
439 return true;
440}
441
442// In C# transform identifier foreach into kw_foreach
443bool FormatTokenLexer::tryTransformCSharpForEach() {
444 if (Tokens.size() < 1)
445 return false;
446 auto &Identifier = *(Tokens.end() - 1);
447 if (Identifier->isNot(Kind: tok::identifier))
448 return false;
449 if (Identifier->TokenText != "foreach")
450 return false;
451
452 Identifier->setType(TT_ForEachMacro);
453 Identifier->Tok.setKind(tok::kw_for);
454 return true;
455}
456
457bool FormatTokenLexer::tryMergeForEach() {
458 if (Tokens.size() < 2)
459 return false;
460 auto &For = *(Tokens.end() - 2);
461 auto &Each = *(Tokens.end() - 1);
462 if (For->isNot(Kind: tok::kw_for))
463 return false;
464 if (Each->isNot(Kind: tok::identifier))
465 return false;
466 if (Each->TokenText != "each")
467 return false;
468
469 For->setType(TT_ForEachMacro);
470 For->Tok.setKind(tok::kw_for);
471
472 For->TokenText = StringRef(For->TokenText.begin(),
473 Each->TokenText.end() - For->TokenText.begin());
474 For->ColumnWidth += Each->ColumnWidth;
475 Tokens.erase(CI: Tokens.end() - 1);
476 return true;
477}
478
479bool FormatTokenLexer::tryTransformTryUsageForC() {
480 if (Tokens.size() < 2)
481 return false;
482 auto &Try = *(Tokens.end() - 2);
483 if (Try->isNot(Kind: tok::kw_try))
484 return false;
485 auto &Next = *(Tokens.end() - 1);
486 if (Next->isOneOf(K1: tok::l_brace, K2: tok::colon, Ks: tok::hash, Ks: tok::comment))
487 return false;
488
489 if (Tokens.size() > 2) {
490 auto &At = *(Tokens.end() - 3);
491 if (At->is(Kind: tok::at))
492 return false;
493 }
494
495 Try->Tok.setKind(tok::identifier);
496 return true;
497}
498
499bool FormatTokenLexer::tryMergeLessLess() {
500 // Merge X,less,less,Y into X,lessless,Y unless X or Y is less.
501 if (Tokens.size() < 3)
502 return false;
503
504 auto First = Tokens.end() - 3;
505 if (First[0]->isNot(Kind: tok::less) || First[1]->isNot(Kind: tok::less))
506 return false;
507
508 // Only merge if there currently is no whitespace between the two "<".
509 if (First[1]->hasWhitespaceBefore())
510 return false;
511
512 auto X = Tokens.size() > 3 ? First[-1] : nullptr;
513 if (X && X->is(Kind: tok::less))
514 return false;
515
516 auto Y = First[2];
517 if ((!X || X->isNot(Kind: tok::kw_operator)) && Y->is(Kind: tok::less))
518 return false;
519
520 First[0]->Tok.setKind(tok::lessless);
521 First[0]->TokenText = "<<";
522 First[0]->ColumnWidth += 1;
523 Tokens.erase(CI: Tokens.end() - 2);
524 return true;
525}
526
527bool FormatTokenLexer::tryMergeGreaterGreater() {
528 // Merge kw_operator,greater,greater into kw_operator,greatergreater.
529 if (Tokens.size() < 2)
530 return false;
531
532 auto First = Tokens.end() - 2;
533 if (First[0]->isNot(Kind: tok::greater) || First[1]->isNot(Kind: tok::greater))
534 return false;
535
536 // Only merge if there currently is no whitespace between the first two ">".
537 if (First[1]->hasWhitespaceBefore())
538 return false;
539
540 auto Tok = Tokens.size() > 2 ? First[-1] : nullptr;
541 if (Tok && Tok->isNot(Kind: tok::kw_operator))
542 return false;
543
544 First[0]->Tok.setKind(tok::greatergreater);
545 First[0]->TokenText = ">>";
546 First[0]->ColumnWidth += 1;
547 Tokens.erase(CI: Tokens.end() - 1);
548 return true;
549}
550
551bool FormatTokenLexer::tryMergeTokens(ArrayRef<tok::TokenKind> Kinds,
552 TokenType NewType) {
553 if (Tokens.size() < Kinds.size())
554 return false;
555
556 SmallVectorImpl<FormatToken *>::const_iterator First =
557 Tokens.end() - Kinds.size();
558 for (unsigned i = 0; i < Kinds.size(); ++i)
559 if (First[i]->isNot(Kind: Kinds[i]))
560 return false;
561
562 return tryMergeTokens(Count: Kinds.size(), NewType);
563}
564
565bool FormatTokenLexer::tryMergeTokens(size_t Count, TokenType NewType) {
566 if (Tokens.size() < Count)
567 return false;
568
569 SmallVectorImpl<FormatToken *>::const_iterator First = Tokens.end() - Count;
570 unsigned AddLength = 0;
571 for (size_t i = 1; i < Count; ++i) {
572 // If there is whitespace separating the token and the previous one,
573 // they should not be merged.
574 if (First[i]->hasWhitespaceBefore())
575 return false;
576 AddLength += First[i]->TokenText.size();
577 }
578
579 Tokens.resize(N: Tokens.size() - Count + 1);
580 First[0]->TokenText = StringRef(First[0]->TokenText.data(),
581 First[0]->TokenText.size() + AddLength);
582 First[0]->ColumnWidth += AddLength;
583 First[0]->setType(NewType);
584 return true;
585}
586
587bool FormatTokenLexer::tryMergeTokensAny(
588 ArrayRef<ArrayRef<tok::TokenKind>> Kinds, TokenType NewType) {
589 return llvm::any_of(Range&: Kinds, P: [this, NewType](ArrayRef<tok::TokenKind> Kinds) {
590 return tryMergeTokens(Kinds, NewType);
591 });
592}
593
594// Returns \c true if \p Tok can only be followed by an operand in JavaScript.
595bool FormatTokenLexer::precedesOperand(FormatToken *Tok) {
596 // NB: This is not entirely correct, as an r_paren can introduce an operand
597 // location in e.g. `if (foo) /bar/.exec(...);`. That is a rare enough
598 // corner case to not matter in practice, though.
599 return Tok->isOneOf(K1: tok::period, K2: tok::l_paren, Ks: tok::comma, Ks: tok::l_brace,
600 Ks: tok::r_brace, Ks: tok::l_square, Ks: tok::semi, Ks: tok::exclaim,
601 Ks: tok::colon, Ks: tok::question, Ks: tok::tilde) ||
602 Tok->isOneOf(K1: tok::kw_return, K2: tok::kw_do, Ks: tok::kw_case, Ks: tok::kw_throw,
603 Ks: tok::kw_else, Ks: tok::kw_new, Ks: tok::kw_delete, Ks: tok::kw_void,
604 Ks: tok::kw_typeof, Ks: Keywords.kw_instanceof, Ks: Keywords.kw_in) ||
605 Tok->isBinaryOperator();
606}
607
608bool FormatTokenLexer::canPrecedeRegexLiteral(FormatToken *Prev) {
609 if (!Prev)
610 return true;
611
612 // Regex literals can only follow after prefix unary operators, not after
613 // postfix unary operators. If the '++' is followed by a non-operand
614 // introducing token, the slash here is the operand and not the start of a
615 // regex.
616 // `!` is an unary prefix operator, but also a post-fix operator that casts
617 // away nullability, so the same check applies.
618 if (Prev->isOneOf(K1: tok::plusplus, K2: tok::minusminus, Ks: tok::exclaim))
619 return Tokens.size() < 3 || precedesOperand(Tok: Tokens[Tokens.size() - 3]);
620
621 // The previous token must introduce an operand location where regex
622 // literals can occur.
623 if (!precedesOperand(Tok: Prev))
624 return false;
625
626 return true;
627}
628
629// Tries to parse a JavaScript Regex literal starting at the current token,
630// if that begins with a slash and is in a location where JavaScript allows
631// regex literals. Changes the current token to a regex literal and updates
632// its text if successful.
633void FormatTokenLexer::tryParseJSRegexLiteral() {
634 FormatToken *RegexToken = Tokens.back();
635 if (!RegexToken->isOneOf(K1: tok::slash, K2: tok::slashequal))
636 return;
637
638 FormatToken *Prev = nullptr;
639 for (FormatToken *FT : llvm::drop_begin(RangeOrContainer: llvm::reverse(C&: Tokens))) {
640 // NB: Because previous pointers are not initialized yet, this cannot use
641 // Token.getPreviousNonComment.
642 if (FT->isNot(Kind: tok::comment)) {
643 Prev = FT;
644 break;
645 }
646 }
647
648 if (!canPrecedeRegexLiteral(Prev))
649 return;
650
651 // 'Manually' lex ahead in the current file buffer.
652 const char *Offset = Lex->getBufferLocation();
653 const char *RegexBegin = Offset - RegexToken->TokenText.size();
654 StringRef Buffer = Lex->getBuffer();
655 bool InCharacterClass = false;
656 bool HaveClosingSlash = false;
657 for (; !HaveClosingSlash && Offset != Buffer.end(); ++Offset) {
658 // Regular expressions are terminated with a '/', which can only be
659 // escaped using '\' or a character class between '[' and ']'.
660 // See http://www.ecma-international.org/ecma-262/5.1/#sec-7.8.5.
661 switch (*Offset) {
662 case '\\':
663 // Skip the escaped character.
664 ++Offset;
665 break;
666 case '[':
667 InCharacterClass = true;
668 break;
669 case ']':
670 InCharacterClass = false;
671 break;
672 case '/':
673 if (!InCharacterClass)
674 HaveClosingSlash = true;
675 break;
676 }
677 }
678
679 RegexToken->setType(TT_RegexLiteral);
680 // Treat regex literals like other string_literals.
681 RegexToken->Tok.setKind(tok::string_literal);
682 RegexToken->TokenText = StringRef(RegexBegin, Offset - RegexBegin);
683 RegexToken->ColumnWidth = RegexToken->TokenText.size();
684
685 resetLexer(Offset: SourceMgr.getFileOffset(SpellingLoc: Lex->getSourceLocation(Loc: Offset)));
686}
687
688static auto lexCSharpString(const char *Begin, const char *End, bool Verbatim,
689 bool Interpolated) {
690 auto Repeated = [&Begin, End]() {
691 return Begin + 1 < End && Begin[1] == Begin[0];
692 };
693
694 // Look for a terminating '"' in the current file buffer.
695 // Make no effort to format code within an interpolated or verbatim string.
696 //
697 // Interpolated strings could contain { } with " characters inside.
698 // $"{x ?? "null"}"
699 // should not be split into $"{x ?? ", null, "}" but should be treated as a
700 // single string-literal.
701 //
702 // We opt not to try and format expressions inside {} within a C#
703 // interpolated string. Formatting expressions within an interpolated string
704 // would require similar work as that done for JavaScript template strings
705 // in `handleTemplateStrings()`.
706 for (int UnmatchedOpeningBraceCount = 0; Begin < End; ++Begin) {
707 switch (*Begin) {
708 case '\\':
709 if (!Verbatim)
710 ++Begin;
711 break;
712 case '{':
713 if (Interpolated) {
714 // {{ inside an interpolated string is escaped, so skip it.
715 if (Repeated())
716 ++Begin;
717 else
718 ++UnmatchedOpeningBraceCount;
719 }
720 break;
721 case '}':
722 if (Interpolated) {
723 // }} inside an interpolated string is escaped, so skip it.
724 if (Repeated())
725 ++Begin;
726 else if (UnmatchedOpeningBraceCount > 0)
727 --UnmatchedOpeningBraceCount;
728 else
729 return End;
730 }
731 break;
732 case '"':
733 if (UnmatchedOpeningBraceCount > 0)
734 break;
735 // "" within a verbatim string is an escaped double quote: skip it.
736 if (Verbatim && Repeated()) {
737 ++Begin;
738 break;
739 }
740 return Begin;
741 }
742 }
743
744 return End;
745}
746
747void FormatTokenLexer::handleCSharpVerbatimAndInterpolatedStrings() {
748 FormatToken *CSharpStringLiteral = Tokens.back();
749
750 if (CSharpStringLiteral->isNot(Kind: TT_CSharpStringLiteral))
751 return;
752
753 auto &TokenText = CSharpStringLiteral->TokenText;
754
755 bool Verbatim = false;
756 bool Interpolated = false;
757 if (TokenText.starts_with(Prefix: R"($@")") || TokenText.starts_with(Prefix: R"(@$")")) {
758 Verbatim = true;
759 Interpolated = true;
760 } else if (TokenText.starts_with(Prefix: R"(@")")) {
761 Verbatim = true;
762 } else if (TokenText.starts_with(Prefix: R"($")")) {
763 Interpolated = true;
764 }
765
766 // Deal with multiline strings.
767 if (!Verbatim && !Interpolated)
768 return;
769
770 const char *StrBegin = Lex->getBufferLocation() - TokenText.size();
771 const char *Offset = StrBegin;
772 if (Verbatim && Interpolated)
773 Offset += 3;
774 else
775 Offset += 2;
776
777 const auto End = Lex->getBuffer().end();
778 Offset = lexCSharpString(Begin: Offset, End, Verbatim, Interpolated);
779
780 // Make no attempt to format code properly if a verbatim string is
781 // unterminated.
782 if (Offset >= End)
783 return;
784
785 StringRef LiteralText(StrBegin, Offset - StrBegin + 1);
786 TokenText = LiteralText;
787
788 // Adjust width for potentially multiline string literals.
789 size_t FirstBreak = LiteralText.find(C: '\n');
790 StringRef FirstLineText = FirstBreak == StringRef::npos
791 ? LiteralText
792 : LiteralText.substr(Start: 0, N: FirstBreak);
793 CSharpStringLiteral->ColumnWidth = encoding::columnWidthWithTabs(
794 Text: FirstLineText, StartColumn: CSharpStringLiteral->OriginalColumn, TabWidth: Style.TabWidth,
795 Encoding);
796 size_t LastBreak = LiteralText.rfind(C: '\n');
797 if (LastBreak != StringRef::npos) {
798 CSharpStringLiteral->IsMultiline = true;
799 unsigned StartColumn = 0;
800 CSharpStringLiteral->LastLineColumnWidth =
801 encoding::columnWidthWithTabs(Text: LiteralText.substr(Start: LastBreak + 1),
802 StartColumn, TabWidth: Style.TabWidth, Encoding);
803 }
804
805 assert(Offset < End);
806 resetLexer(Offset: SourceMgr.getFileOffset(SpellingLoc: Lex->getSourceLocation(Loc: Offset + 1)));
807}
808
809void FormatTokenLexer::handleTableGenMultilineString() {
810 FormatToken *MultiLineString = Tokens.back();
811 if (MultiLineString->isNot(Kind: TT_TableGenMultiLineString))
812 return;
813
814 auto OpenOffset = Lex->getCurrentBufferOffset() - 2 /* "[{" */;
815 // "}]" is the end of multi line string.
816 auto CloseOffset = Lex->getBuffer().find(Str: "}]", From: OpenOffset);
817 if (CloseOffset == StringRef::npos)
818 return;
819 auto Text = Lex->getBuffer().substr(Start: OpenOffset, N: CloseOffset - OpenOffset + 2);
820 MultiLineString->TokenText = Text;
821 resetLexer(Offset: SourceMgr.getFileOffset(
822 SpellingLoc: Lex->getSourceLocation(Loc: Lex->getBufferLocation() - 2 + Text.size())));
823 auto FirstLineText = Text;
824 auto FirstBreak = Text.find(C: '\n');
825 // Set ColumnWidth and LastLineColumnWidth when it has multiple lines.
826 if (FirstBreak != StringRef::npos) {
827 MultiLineString->IsMultiline = true;
828 FirstLineText = Text.substr(Start: 0, N: FirstBreak + 1);
829 // LastLineColumnWidth holds the width of the last line.
830 auto LastBreak = Text.rfind(C: '\n');
831 MultiLineString->LastLineColumnWidth = encoding::columnWidthWithTabs(
832 Text: Text.substr(Start: LastBreak + 1), StartColumn: MultiLineString->OriginalColumn,
833 TabWidth: Style.TabWidth, Encoding);
834 }
835 // ColumnWidth holds only the width of the first line.
836 MultiLineString->ColumnWidth = encoding::columnWidthWithTabs(
837 Text: FirstLineText, StartColumn: MultiLineString->OriginalColumn, TabWidth: Style.TabWidth, Encoding);
838}
839
840void FormatTokenLexer::handleTableGenNumericLikeIdentifier() {
841 FormatToken *Tok = Tokens.back();
842 // TableGen identifiers can begin with digits. Such tokens are lexed as
843 // numeric_constant now.
844 if (Tok->isNot(Kind: tok::numeric_constant))
845 return;
846 StringRef Text = Tok->TokenText;
847 // The following check is based on llvm::TGLexer::LexToken.
848 // That lexes the token as a number if any of the following holds:
849 // 1. It starts with '+', '-'.
850 // 2. All the characters are digits.
851 // 3. The first non-digit character is 'b', and the next is '0' or '1'.
852 // 4. The first non-digit character is 'x', and the next is a hex digit.
853 // Note that in the case 3 and 4, if the next character does not exists in
854 // this token, the token is an identifier.
855 if (Text.size() < 1 || Text[0] == '+' || Text[0] == '-')
856 return;
857 const auto NonDigitPos = Text.find_if(F: [](char C) { return !isdigit(C); });
858 // All the characters are digits
859 if (NonDigitPos == StringRef::npos)
860 return;
861 char FirstNonDigit = Text[NonDigitPos];
862 if (NonDigitPos < Text.size() - 1) {
863 char TheNext = Text[NonDigitPos + 1];
864 // Regarded as a binary number.
865 if (FirstNonDigit == 'b' && (TheNext == '0' || TheNext == '1'))
866 return;
867 // Regarded as hex number.
868 if (FirstNonDigit == 'x' && isxdigit(TheNext))
869 return;
870 }
871 if (isalpha(FirstNonDigit) || FirstNonDigit == '_') {
872 // This is actually an identifier in TableGen.
873 Tok->Tok.setKind(tok::identifier);
874 Tok->Tok.setIdentifierInfo(nullptr);
875 }
876}
877
878void FormatTokenLexer::handleTemplateStrings() {
879 FormatToken *BacktickToken = Tokens.back();
880
881 if (BacktickToken->is(Kind: tok::l_brace)) {
882 StateStack.push(x: LexerState::NORMAL);
883 return;
884 }
885 if (BacktickToken->is(Kind: tok::r_brace)) {
886 if (StateStack.size() == 1)
887 return;
888 StateStack.pop();
889 if (StateStack.top() != LexerState::TEMPLATE_STRING)
890 return;
891 // If back in TEMPLATE_STRING, fallthrough and continue parsing the
892 } else if (BacktickToken->is(Kind: tok::unknown) &&
893 BacktickToken->TokenText == "`") {
894 StateStack.push(x: LexerState::TEMPLATE_STRING);
895 } else {
896 return; // Not actually a template
897 }
898
899 // 'Manually' lex ahead in the current file buffer.
900 const char *Offset = Lex->getBufferLocation();
901 const char *TmplBegin = Offset - BacktickToken->TokenText.size(); // at "`"
902 for (; Offset != Lex->getBuffer().end(); ++Offset) {
903 if (Offset[0] == '`') {
904 StateStack.pop();
905 ++Offset;
906 break;
907 }
908 if (Offset[0] == '\\') {
909 ++Offset; // Skip the escaped character.
910 } else if (Offset + 1 < Lex->getBuffer().end() && Offset[0] == '$' &&
911 Offset[1] == '{') {
912 // '${' introduces an expression interpolation in the template string.
913 StateStack.push(x: LexerState::NORMAL);
914 Offset += 2;
915 break;
916 }
917 }
918
919 StringRef LiteralText(TmplBegin, Offset - TmplBegin);
920 BacktickToken->setType(TT_TemplateString);
921 BacktickToken->Tok.setKind(tok::string_literal);
922 BacktickToken->TokenText = LiteralText;
923
924 // Adjust width for potentially multiline string literals.
925 size_t FirstBreak = LiteralText.find(C: '\n');
926 StringRef FirstLineText = FirstBreak == StringRef::npos
927 ? LiteralText
928 : LiteralText.substr(Start: 0, N: FirstBreak);
929 BacktickToken->ColumnWidth = encoding::columnWidthWithTabs(
930 Text: FirstLineText, StartColumn: BacktickToken->OriginalColumn, TabWidth: Style.TabWidth, Encoding);
931 size_t LastBreak = LiteralText.rfind(C: '\n');
932 if (LastBreak != StringRef::npos) {
933 BacktickToken->IsMultiline = true;
934 unsigned StartColumn = 0; // The template tail spans the entire line.
935 BacktickToken->LastLineColumnWidth =
936 encoding::columnWidthWithTabs(Text: LiteralText.substr(Start: LastBreak + 1),
937 StartColumn, TabWidth: Style.TabWidth, Encoding);
938 }
939
940 SourceLocation loc = Lex->getSourceLocation(Loc: Offset);
941 resetLexer(Offset: SourceMgr.getFileOffset(SpellingLoc: loc));
942}
943
944void FormatTokenLexer::tryParsePythonComment() {
945 FormatToken *HashToken = Tokens.back();
946 if (!HashToken->isOneOf(K1: tok::hash, K2: tok::hashhash))
947 return;
948 // Turn the remainder of this line into a comment.
949 const char *CommentBegin =
950 Lex->getBufferLocation() - HashToken->TokenText.size(); // at "#"
951 size_t From = CommentBegin - Lex->getBuffer().begin();
952 size_t To = Lex->getBuffer().find_first_of(C: '\n', From);
953 if (To == StringRef::npos)
954 To = Lex->getBuffer().size();
955 size_t Len = To - From;
956 HashToken->setType(TT_LineComment);
957 HashToken->Tok.setKind(tok::comment);
958 HashToken->TokenText = Lex->getBuffer().substr(Start: From, N: Len);
959 SourceLocation Loc = To < Lex->getBuffer().size()
960 ? Lex->getSourceLocation(Loc: CommentBegin + Len)
961 : SourceMgr.getLocForEndOfFile(FID: ID);
962 resetLexer(Offset: SourceMgr.getFileOffset(SpellingLoc: Loc));
963}
964
965bool FormatTokenLexer::tryMerge_TMacro() {
966 if (Tokens.size() < 4)
967 return false;
968 FormatToken *Last = Tokens.back();
969 if (Last->isNot(Kind: tok::r_paren))
970 return false;
971
972 FormatToken *String = Tokens[Tokens.size() - 2];
973 if (String->isNot(Kind: tok::string_literal) || String->IsMultiline)
974 return false;
975
976 if (Tokens[Tokens.size() - 3]->isNot(Kind: tok::l_paren))
977 return false;
978
979 FormatToken *Macro = Tokens[Tokens.size() - 4];
980 if (Macro->TokenText != "_T")
981 return false;
982
983 const char *Start = Macro->TokenText.data();
984 const char *End = Last->TokenText.data() + Last->TokenText.size();
985 String->TokenText = StringRef(Start, End - Start);
986 String->IsFirst = Macro->IsFirst;
987 String->LastNewlineOffset = Macro->LastNewlineOffset;
988 String->WhitespaceRange = Macro->WhitespaceRange;
989 String->OriginalColumn = Macro->OriginalColumn;
990 String->ColumnWidth = encoding::columnWidthWithTabs(
991 Text: String->TokenText, StartColumn: String->OriginalColumn, TabWidth: Style.TabWidth, Encoding);
992 String->NewlinesBefore = Macro->NewlinesBefore;
993 String->HasUnescapedNewline = Macro->HasUnescapedNewline;
994
995 Tokens.pop_back();
996 Tokens.pop_back();
997 Tokens.pop_back();
998 Tokens.back() = String;
999 if (FirstInLineIndex >= Tokens.size())
1000 FirstInLineIndex = Tokens.size() - 1;
1001 return true;
1002}
1003
1004bool FormatTokenLexer::tryMergeConflictMarkers() {
1005 if (Tokens.back()->NewlinesBefore == 0 && Tokens.back()->isNot(Kind: tok::eof))
1006 return false;
1007
1008 // Conflict lines look like:
1009 // <marker> <text from the vcs>
1010 // For example:
1011 // >>>>>>> /file/in/file/system at revision 1234
1012 //
1013 // We merge all tokens in a line that starts with a conflict marker
1014 // into a single token with a special token type that the unwrapped line
1015 // parser will use to correctly rebuild the underlying code.
1016
1017 FileID ID;
1018 // Get the position of the first token in the line.
1019 unsigned FirstInLineOffset;
1020 std::tie(args&: ID, args&: FirstInLineOffset) = SourceMgr.getDecomposedLoc(
1021 Loc: Tokens[FirstInLineIndex]->getStartOfNonWhitespace());
1022 StringRef Buffer = SourceMgr.getBufferOrFake(FID: ID).getBuffer();
1023 // Calculate the offset of the start of the current line.
1024 auto LineOffset = Buffer.rfind(C: '\n', From: FirstInLineOffset);
1025 if (LineOffset == StringRef::npos)
1026 LineOffset = 0;
1027 else
1028 ++LineOffset;
1029
1030 auto FirstSpace = Buffer.find_first_of(Chars: " \n", From: LineOffset);
1031 StringRef LineStart;
1032 if (FirstSpace == StringRef::npos)
1033 LineStart = Buffer.substr(Start: LineOffset);
1034 else
1035 LineStart = Buffer.substr(Start: LineOffset, N: FirstSpace - LineOffset);
1036
1037 TokenType Type = TT_Unknown;
1038 if (LineStart == "<<<<<<<" || LineStart == ">>>>") {
1039 Type = TT_ConflictStart;
1040 } else if (LineStart == "|||||||" || LineStart == "=======" ||
1041 LineStart == "====") {
1042 Type = TT_ConflictAlternative;
1043 } else if (LineStart == ">>>>>>>" || LineStart == "<<<<") {
1044 Type = TT_ConflictEnd;
1045 }
1046
1047 if (Type != TT_Unknown) {
1048 FormatToken *Next = Tokens.back();
1049
1050 Tokens.resize(N: FirstInLineIndex + 1);
1051 // We do not need to build a complete token here, as we will skip it
1052 // during parsing anyway (as we must not touch whitespace around conflict
1053 // markers).
1054 Tokens.back()->setType(Type);
1055 Tokens.back()->Tok.setKind(tok::kw___unknown_anytype);
1056
1057 Tokens.push_back(Elt: Next);
1058 return true;
1059 }
1060
1061 return false;
1062}
1063
1064FormatToken *FormatTokenLexer::getStashedToken() {
1065 // Create a synthesized second '>' or '<' token.
1066 Token Tok = FormatTok->Tok;
1067 StringRef TokenText = FormatTok->TokenText;
1068
1069 unsigned OriginalColumn = FormatTok->OriginalColumn;
1070 FormatTok = new (Allocator.Allocate()) FormatToken;
1071 FormatTok->Tok = Tok;
1072 SourceLocation TokLocation =
1073 FormatTok->Tok.getLocation().getLocWithOffset(Offset: Tok.getLength() - 1);
1074 FormatTok->Tok.setLocation(TokLocation);
1075 FormatTok->WhitespaceRange = SourceRange(TokLocation, TokLocation);
1076 FormatTok->TokenText = TokenText;
1077 FormatTok->ColumnWidth = 1;
1078 FormatTok->OriginalColumn = OriginalColumn + 1;
1079
1080 return FormatTok;
1081}
1082
1083/// Truncate the current token to the new length and make the lexer continue
1084/// from the end of the truncated token. Used for other languages that have
1085/// different token boundaries, like JavaScript in which a comment ends at a
1086/// line break regardless of whether the line break follows a backslash. Also
1087/// used to set the lexer to the end of whitespace if the lexer regards
1088/// whitespace and an unrecognized symbol as one token.
1089void FormatTokenLexer::truncateToken(size_t NewLen) {
1090 assert(NewLen <= FormatTok->TokenText.size());
1091 resetLexer(Offset: SourceMgr.getFileOffset(SpellingLoc: Lex->getSourceLocation(
1092 Loc: Lex->getBufferLocation() - FormatTok->TokenText.size() + NewLen)));
1093 FormatTok->TokenText = FormatTok->TokenText.substr(Start: 0, N: NewLen);
1094 FormatTok->ColumnWidth = encoding::columnWidthWithTabs(
1095 Text: FormatTok->TokenText, StartColumn: FormatTok->OriginalColumn, TabWidth: Style.TabWidth,
1096 Encoding);
1097 FormatTok->Tok.setLength(NewLen);
1098}
1099
1100/// Count the length of leading whitespace in a token.
1101static size_t countLeadingWhitespace(StringRef Text) {
1102 // Basically counting the length matched by this regex.
1103 // "^([\n\r\f\v \t]|(\\\\|\\?\\?/)[\n\r])+"
1104 // Directly using the regex turned out to be slow. With the regex
1105 // version formatting all files in this directory took about 1.25
1106 // seconds. This version took about 0.5 seconds.
1107 const unsigned char *const Begin = Text.bytes_begin();
1108 const unsigned char *const End = Text.bytes_end();
1109 const unsigned char *Cur = Begin;
1110 while (Cur < End) {
1111 if (isspace(Cur[0])) {
1112 ++Cur;
1113 } else if (Cur[0] == '\\' && (Cur[1] == '\n' || Cur[1] == '\r')) {
1114 // A '\' followed by a newline always escapes the newline, regardless
1115 // of whether there is another '\' before it.
1116 // The source has a null byte at the end. So the end of the entire input
1117 // isn't reached yet. Also the lexer doesn't break apart an escaped
1118 // newline.
1119 assert(End - Cur >= 2);
1120 Cur += 2;
1121 } else if (Cur[0] == '?' && Cur[1] == '?' && Cur[2] == '/' &&
1122 (Cur[3] == '\n' || Cur[3] == '\r')) {
1123 // Newlines can also be escaped by a '?' '?' '/' trigraph. By the way, the
1124 // characters are quoted individually in this comment because if we write
1125 // them together some compilers warn that we have a trigraph in the code.
1126 assert(End - Cur >= 4);
1127 Cur += 4;
1128 } else {
1129 break;
1130 }
1131 }
1132 return Cur - Begin;
1133}
1134
1135FormatToken *FormatTokenLexer::getNextToken() {
1136 if (StateStack.top() == LexerState::TOKEN_STASHED) {
1137 StateStack.pop();
1138 return getStashedToken();
1139 }
1140
1141 FormatTok = new (Allocator.Allocate()) FormatToken;
1142 readRawToken(Tok&: *FormatTok);
1143 SourceLocation WhitespaceStart =
1144 FormatTok->Tok.getLocation().getLocWithOffset(Offset: -TrailingWhitespace);
1145 FormatTok->IsFirst = IsFirstToken;
1146 IsFirstToken = false;
1147
1148 // Consume and record whitespace until we find a significant token.
1149 // Some tok::unknown tokens are not just whitespace, e.g. whitespace
1150 // followed by a symbol such as backtick. Those symbols may be
1151 // significant in other languages.
1152 unsigned WhitespaceLength = TrailingWhitespace;
1153 while (FormatTok->isNot(Kind: tok::eof)) {
1154 auto LeadingWhitespace = countLeadingWhitespace(Text: FormatTok->TokenText);
1155 if (LeadingWhitespace == 0)
1156 break;
1157 if (LeadingWhitespace < FormatTok->TokenText.size())
1158 truncateToken(NewLen: LeadingWhitespace);
1159 StringRef Text = FormatTok->TokenText;
1160 bool InEscape = false;
1161 for (int i = 0, e = Text.size(); i != e; ++i) {
1162 switch (Text[i]) {
1163 case '\r':
1164 // If this is a CRLF sequence, break here and the LF will be handled on
1165 // the next loop iteration. Otherwise, this is a single Mac CR, treat it
1166 // the same as a single LF.
1167 if (i + 1 < e && Text[i + 1] == '\n')
1168 break;
1169 [[fallthrough]];
1170 case '\n':
1171 ++FormatTok->NewlinesBefore;
1172 if (!InEscape)
1173 FormatTok->HasUnescapedNewline = true;
1174 else
1175 InEscape = false;
1176 FormatTok->LastNewlineOffset = WhitespaceLength + i + 1;
1177 Column = 0;
1178 break;
1179 case '\f':
1180 case '\v':
1181 Column = 0;
1182 break;
1183 case ' ':
1184 ++Column;
1185 break;
1186 case '\t':
1187 Column +=
1188 Style.TabWidth - (Style.TabWidth ? Column % Style.TabWidth : 0);
1189 break;
1190 case '\\':
1191 case '?':
1192 case '/':
1193 // The text was entirely whitespace when this loop was entered. Thus
1194 // this has to be an escape sequence.
1195 assert(Text.substr(i, 2) == "\\\r" || Text.substr(i, 2) == "\\\n" ||
1196 Text.substr(i, 4) == "\?\?/\r" ||
1197 Text.substr(i, 4) == "\?\?/\n" ||
1198 (i >= 1 && (Text.substr(i - 1, 4) == "\?\?/\r" ||
1199 Text.substr(i - 1, 4) == "\?\?/\n")) ||
1200 (i >= 2 && (Text.substr(i - 2, 4) == "\?\?/\r" ||
1201 Text.substr(i - 2, 4) == "\?\?/\n")));
1202 InEscape = true;
1203 break;
1204 default:
1205 // This shouldn't happen.
1206 assert(false);
1207 break;
1208 }
1209 }
1210 WhitespaceLength += Text.size();
1211 readRawToken(Tok&: *FormatTok);
1212 }
1213
1214 if (FormatTok->is(Kind: tok::unknown))
1215 FormatTok->setType(TT_ImplicitStringLiteral);
1216
1217 // JavaScript and Java do not allow to escape the end of the line with a
1218 // backslash. Backslashes are syntax errors in plain source, but can occur in
1219 // comments. When a single line comment ends with a \, it'll cause the next
1220 // line of code to be lexed as a comment, breaking formatting. The code below
1221 // finds comments that contain a backslash followed by a line break, truncates
1222 // the comment token at the backslash, and resets the lexer to restart behind
1223 // the backslash.
1224 if ((Style.isJavaScript() || Style.Language == FormatStyle::LK_Java) &&
1225 FormatTok->is(Kind: tok::comment) && FormatTok->TokenText.starts_with(Prefix: "//")) {
1226 size_t BackslashPos = FormatTok->TokenText.find(C: '\\');
1227 while (BackslashPos != StringRef::npos) {
1228 if (BackslashPos + 1 < FormatTok->TokenText.size() &&
1229 FormatTok->TokenText[BackslashPos + 1] == '\n') {
1230 truncateToken(NewLen: BackslashPos + 1);
1231 break;
1232 }
1233 BackslashPos = FormatTok->TokenText.find(C: '\\', From: BackslashPos + 1);
1234 }
1235 }
1236
1237 if (Style.isVerilog()) {
1238 static const llvm::Regex NumberBase("^s?[bdho]", llvm::Regex::IgnoreCase);
1239 SmallVector<StringRef, 1> Matches;
1240 // Verilog uses the backtick instead of the hash for preprocessor stuff.
1241 // And it uses the hash for delays and parameter lists. In order to continue
1242 // using `tok::hash` in other places, the backtick gets marked as the hash
1243 // here. And in order to tell the backtick and hash apart for
1244 // Verilog-specific stuff, the hash becomes an identifier.
1245 if (FormatTok->is(Kind: tok::numeric_constant)) {
1246 // In Verilog the quote is not part of a number.
1247 auto Quote = FormatTok->TokenText.find(C: '\'');
1248 if (Quote != StringRef::npos)
1249 truncateToken(NewLen: Quote);
1250 } else if (FormatTok->isOneOf(K1: tok::hash, K2: tok::hashhash)) {
1251 FormatTok->Tok.setKind(tok::raw_identifier);
1252 } else if (FormatTok->is(Kind: tok::raw_identifier)) {
1253 if (FormatTok->TokenText == "`") {
1254 FormatTok->Tok.setIdentifierInfo(nullptr);
1255 FormatTok->Tok.setKind(tok::hash);
1256 } else if (FormatTok->TokenText == "``") {
1257 FormatTok->Tok.setIdentifierInfo(nullptr);
1258 FormatTok->Tok.setKind(tok::hashhash);
1259 } else if (Tokens.size() > 0 &&
1260 Tokens.back()->is(II: Keywords.kw_apostrophe) &&
1261 NumberBase.match(String: FormatTok->TokenText, Matches: &Matches)) {
1262 // In Verilog in a based number literal like `'b10`, there may be
1263 // whitespace between `'b` and `10`. Therefore we handle the base and
1264 // the rest of the number literal as two tokens. But if there is no
1265 // space in the input code, we need to manually separate the two parts.
1266 truncateToken(NewLen: Matches[0].size());
1267 FormatTok->setFinalizedType(TT_VerilogNumberBase);
1268 }
1269 }
1270 }
1271
1272 FormatTok->WhitespaceRange = SourceRange(
1273 WhitespaceStart, WhitespaceStart.getLocWithOffset(Offset: WhitespaceLength));
1274
1275 FormatTok->OriginalColumn = Column;
1276
1277 TrailingWhitespace = 0;
1278 if (FormatTok->is(Kind: tok::comment)) {
1279 // FIXME: Add the trimmed whitespace to Column.
1280 StringRef UntrimmedText = FormatTok->TokenText;
1281 FormatTok->TokenText = FormatTok->TokenText.rtrim(Chars: " \t\v\f");
1282 TrailingWhitespace = UntrimmedText.size() - FormatTok->TokenText.size();
1283 } else if (FormatTok->is(Kind: tok::raw_identifier)) {
1284 IdentifierInfo &Info = IdentTable.get(Name: FormatTok->TokenText);
1285 FormatTok->Tok.setIdentifierInfo(&Info);
1286 FormatTok->Tok.setKind(Info.getTokenID());
1287 if (Style.Language == FormatStyle::LK_Java &&
1288 FormatTok->isOneOf(K1: tok::kw_struct, K2: tok::kw_union, Ks: tok::kw_delete,
1289 Ks: tok::kw_operator)) {
1290 FormatTok->Tok.setKind(tok::identifier);
1291 FormatTok->Tok.setIdentifierInfo(nullptr);
1292 } else if (Style.isJavaScript() &&
1293 FormatTok->isOneOf(K1: tok::kw_struct, K2: tok::kw_union,
1294 Ks: tok::kw_operator)) {
1295 FormatTok->Tok.setKind(tok::identifier);
1296 FormatTok->Tok.setIdentifierInfo(nullptr);
1297 } else if (Style.isTableGen() && !Keywords.isTableGenKeyword(Tok: *FormatTok)) {
1298 FormatTok->Tok.setKind(tok::identifier);
1299 FormatTok->Tok.setIdentifierInfo(nullptr);
1300 }
1301 } else if (FormatTok->is(Kind: tok::greatergreater)) {
1302 FormatTok->Tok.setKind(tok::greater);
1303 FormatTok->TokenText = FormatTok->TokenText.substr(Start: 0, N: 1);
1304 ++Column;
1305 StateStack.push(x: LexerState::TOKEN_STASHED);
1306 } else if (FormatTok->is(Kind: tok::lessless)) {
1307 FormatTok->Tok.setKind(tok::less);
1308 FormatTok->TokenText = FormatTok->TokenText.substr(Start: 0, N: 1);
1309 ++Column;
1310 StateStack.push(x: LexerState::TOKEN_STASHED);
1311 }
1312
1313 if (Style.isVerilog() && Tokens.size() > 0 &&
1314 Tokens.back()->is(TT: TT_VerilogNumberBase) &&
1315 FormatTok->Tok.isOneOf(K1: tok::identifier, K2: tok::question)) {
1316 // Mark the number following a base like `'h?a0` as a number.
1317 FormatTok->Tok.setKind(tok::numeric_constant);
1318 }
1319
1320 // Now FormatTok is the next non-whitespace token.
1321
1322 StringRef Text = FormatTok->TokenText;
1323 size_t FirstNewlinePos = Text.find(C: '\n');
1324 if (FirstNewlinePos == StringRef::npos) {
1325 // FIXME: ColumnWidth actually depends on the start column, we need to
1326 // take this into account when the token is moved.
1327 FormatTok->ColumnWidth =
1328 encoding::columnWidthWithTabs(Text, StartColumn: Column, TabWidth: Style.TabWidth, Encoding);
1329 Column += FormatTok->ColumnWidth;
1330 } else {
1331 FormatTok->IsMultiline = true;
1332 // FIXME: ColumnWidth actually depends on the start column, we need to
1333 // take this into account when the token is moved.
1334 FormatTok->ColumnWidth = encoding::columnWidthWithTabs(
1335 Text: Text.substr(Start: 0, N: FirstNewlinePos), StartColumn: Column, TabWidth: Style.TabWidth, Encoding);
1336
1337 // The last line of the token always starts in column 0.
1338 // Thus, the length can be precomputed even in the presence of tabs.
1339 FormatTok->LastLineColumnWidth = encoding::columnWidthWithTabs(
1340 Text: Text.substr(Start: Text.find_last_of(C: '\n') + 1), StartColumn: 0, TabWidth: Style.TabWidth, Encoding);
1341 Column = FormatTok->LastLineColumnWidth;
1342 }
1343
1344 if (Style.isCpp()) {
1345 auto *Identifier = FormatTok->Tok.getIdentifierInfo();
1346 auto it = Macros.find(Key: Identifier);
1347 if (!(Tokens.size() > 0 && Tokens.back()->Tok.getIdentifierInfo() &&
1348 Tokens.back()->Tok.getIdentifierInfo()->getPPKeywordID() ==
1349 tok::pp_define) &&
1350 it != Macros.end()) {
1351 FormatTok->setType(it->second);
1352 if (it->second == TT_IfMacro) {
1353 // The lexer token currently has type tok::kw_unknown. However, for this
1354 // substitution to be treated correctly in the TokenAnnotator, faking
1355 // the tok value seems to be needed. Not sure if there's a more elegant
1356 // way.
1357 FormatTok->Tok.setKind(tok::kw_if);
1358 }
1359 } else if (FormatTok->is(Kind: tok::identifier)) {
1360 if (MacroBlockBeginRegex.match(String: Text))
1361 FormatTok->setType(TT_MacroBlockBegin);
1362 else if (MacroBlockEndRegex.match(String: Text))
1363 FormatTok->setType(TT_MacroBlockEnd);
1364 else if (TypeNames.contains(Ptr: Identifier))
1365 FormatTok->setFinalizedType(TT_TypeName);
1366 }
1367 }
1368
1369 return FormatTok;
1370}
1371
1372bool FormatTokenLexer::readRawTokenVerilogSpecific(Token &Tok) {
1373 // In Verilog the quote is not a character literal.
1374 //
1375 // Make the backtick and double backtick identifiers to match against them
1376 // more easily.
1377 //
1378 // In Verilog an escaped identifier starts with backslash and ends with
1379 // whitespace. Unless that whitespace is an escaped newline. A backslash can
1380 // also begin an escaped newline outside of an escaped identifier. We check
1381 // for that outside of the Regex since we can't use negative lookhead
1382 // assertions. Simply changing the '*' to '+' breaks stuff as the escaped
1383 // identifier may have a length of 0 according to Section A.9.3.
1384 // FIXME: If there is an escaped newline in the middle of an escaped
1385 // identifier, allow for pasting the two lines together, But escaped
1386 // identifiers usually occur only in generated code anyway.
1387 static const llvm::Regex VerilogToken(R"re(^('|``?|\\(\\)re"
1388 "(\r?\n|\r)|[^[:space:]])*)");
1389
1390 SmallVector<StringRef, 4> Matches;
1391 const char *Start = Lex->getBufferLocation();
1392 if (!VerilogToken.match(String: StringRef(Start, Lex->getBuffer().end() - Start),
1393 Matches: &Matches)) {
1394 return false;
1395 }
1396 // There is a null byte at the end of the buffer, so we don't have to check
1397 // Start[1] is within the buffer.
1398 if (Start[0] == '\\' && (Start[1] == '\r' || Start[1] == '\n'))
1399 return false;
1400 size_t Len = Matches[0].size();
1401
1402 // The kind has to be an identifier so we can match it against those defined
1403 // in Keywords. The kind has to be set before the length because the setLength
1404 // function checks that the kind is not an annotation.
1405 Tok.setKind(tok::raw_identifier);
1406 Tok.setLength(Len);
1407 Tok.setLocation(Lex->getSourceLocation(Loc: Start, TokLen: Len));
1408 Tok.setRawIdentifierData(Start);
1409 Lex->seek(Offset: Lex->getCurrentBufferOffset() + Len, /*IsAtStartofline=*/IsAtStartOfLine: false);
1410 return true;
1411}
1412
1413void FormatTokenLexer::readRawToken(FormatToken &Tok) {
1414 // For Verilog, first see if there is a special token, and fall back to the
1415 // normal lexer if there isn't one.
1416 if (!Style.isVerilog() || !readRawTokenVerilogSpecific(Tok&: Tok.Tok))
1417 Lex->LexFromRawLexer(Result&: Tok.Tok);
1418 Tok.TokenText = StringRef(SourceMgr.getCharacterData(SL: Tok.Tok.getLocation()),
1419 Tok.Tok.getLength());
1420 // For formatting, treat unterminated string literals like normal string
1421 // literals.
1422 if (Tok.is(Kind: tok::unknown)) {
1423 if (Tok.TokenText.starts_with(Prefix: "\"")) {
1424 Tok.Tok.setKind(tok::string_literal);
1425 Tok.IsUnterminatedLiteral = true;
1426 } else if (Style.isJavaScript() && Tok.TokenText == "''") {
1427 Tok.Tok.setKind(tok::string_literal);
1428 }
1429 }
1430
1431 if ((Style.isJavaScript() || Style.isProto()) && Tok.is(Kind: tok::char_constant))
1432 Tok.Tok.setKind(tok::string_literal);
1433
1434 if (Tok.is(Kind: tok::comment) && isClangFormatOn(Comment: Tok.TokenText))
1435 FormattingDisabled = false;
1436
1437 Tok.Finalized = FormattingDisabled;
1438
1439 if (Tok.is(Kind: tok::comment) && isClangFormatOff(Comment: Tok.TokenText))
1440 FormattingDisabled = true;
1441}
1442
1443void FormatTokenLexer::resetLexer(unsigned Offset) {
1444 StringRef Buffer = SourceMgr.getBufferData(FID: ID);
1445 Lex.reset(p: new Lexer(SourceMgr.getLocForStartOfFile(FID: ID), LangOpts,
1446 Buffer.begin(), Buffer.begin() + Offset, Buffer.end()));
1447 Lex->SetKeepWhitespaceMode(true);
1448 TrailingWhitespace = 0;
1449}
1450
1451} // namespace format
1452} // namespace clang
1453