1 //===--- UnwrappedLineParser.cpp - Format C++ code ------------------------===// 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 contains the implementation of the UnwrappedLineParser, 11 /// which turns a stream of tokens into UnwrappedLines. 12 /// 13 //===----------------------------------------------------------------------===// 14 15 #include "UnwrappedLineParser.h" 16 #include "FormatToken.h" 17 #include "TokenAnnotator.h" 18 #include "llvm/ADT/STLExtras.h" 19 #include "llvm/Support/Debug.h" 20 #include "llvm/Support/raw_ostream.h" 21 22 #include <algorithm> 23 #include <utility> 24 25 #define DEBUG_TYPE "format-parser" 26 27 namespace clang { 28 namespace format { 29 30 class FormatTokenSource { 31 public: 32 virtual ~FormatTokenSource() {} 33 34 // Returns the next token in the token stream. 35 virtual FormatToken *getNextToken() = 0; 36 37 // Returns the token preceding the token returned by the last call to 38 // getNextToken() in the token stream, or nullptr if no such token exists. 39 virtual FormatToken *getPreviousToken() = 0; 40 41 // Returns the token that would be returned by the next call to 42 // getNextToken(). 43 virtual FormatToken *peekNextToken() = 0; 44 45 // Returns the token that would be returned after the next N calls to 46 // getNextToken(). N needs to be greater than zero, and small enough that 47 // there are still tokens. Check for tok::eof with N-1 before calling it with 48 // N. 49 virtual FormatToken *peekNextToken(int N) = 0; 50 51 // Returns whether we are at the end of the file. 52 // This can be different from whether getNextToken() returned an eof token 53 // when the FormatTokenSource is a view on a part of the token stream. 54 virtual bool isEOF() = 0; 55 56 // Gets the current position in the token stream, to be used by setPosition(). 57 virtual unsigned getPosition() = 0; 58 59 // Resets the token stream to the state it was in when getPosition() returned 60 // Position, and return the token at that position in the stream. 61 virtual FormatToken *setPosition(unsigned Position) = 0; 62 }; 63 64 namespace { 65 66 class ScopedDeclarationState { 67 public: 68 ScopedDeclarationState(UnwrappedLine &Line, llvm::BitVector &Stack, 69 bool MustBeDeclaration) 70 : Line(Line), Stack(Stack) { 71 Line.MustBeDeclaration = MustBeDeclaration; 72 Stack.push_back(MustBeDeclaration); 73 } 74 ~ScopedDeclarationState() { 75 Stack.pop_back(); 76 if (!Stack.empty()) 77 Line.MustBeDeclaration = Stack.back(); 78 else 79 Line.MustBeDeclaration = true; 80 } 81 82 private: 83 UnwrappedLine &Line; 84 llvm::BitVector &Stack; 85 }; 86 87 static bool isLineComment(const FormatToken &FormatTok) { 88 return FormatTok.is(tok::comment) && !FormatTok.TokenText.startswith("/*"); 89 } 90 91 // Checks if \p FormatTok is a line comment that continues the line comment 92 // \p Previous. The original column of \p MinColumnToken is used to determine 93 // whether \p FormatTok is indented enough to the right to continue \p Previous. 94 static bool continuesLineComment(const FormatToken &FormatTok, 95 const FormatToken *Previous, 96 const FormatToken *MinColumnToken) { 97 if (!Previous || !MinColumnToken) 98 return false; 99 unsigned MinContinueColumn = 100 MinColumnToken->OriginalColumn + (isLineComment(*MinColumnToken) ? 0 : 1); 101 return isLineComment(FormatTok) && FormatTok.NewlinesBefore == 1 && 102 isLineComment(*Previous) && 103 FormatTok.OriginalColumn >= MinContinueColumn; 104 } 105 106 class ScopedMacroState : public FormatTokenSource { 107 public: 108 ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource, 109 FormatToken *&ResetToken) 110 : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken), 111 PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource), 112 Token(nullptr), PreviousToken(nullptr) { 113 FakeEOF.Tok.startToken(); 114 FakeEOF.Tok.setKind(tok::eof); 115 TokenSource = this; 116 Line.Level = 0; 117 Line.InPPDirective = true; 118 } 119 120 ~ScopedMacroState() override { 121 TokenSource = PreviousTokenSource; 122 ResetToken = Token; 123 Line.InPPDirective = false; 124 Line.Level = PreviousLineLevel; 125 } 126 127 FormatToken *getNextToken() override { 128 // The \c UnwrappedLineParser guards against this by never calling 129 // \c getNextToken() after it has encountered the first eof token. 130 assert(!eof()); 131 PreviousToken = Token; 132 Token = PreviousTokenSource->getNextToken(); 133 if (eof()) 134 return &FakeEOF; 135 return Token; 136 } 137 138 FormatToken *getPreviousToken() override { 139 return PreviousTokenSource->getPreviousToken(); 140 } 141 142 FormatToken *peekNextToken() override { 143 if (eof()) 144 return &FakeEOF; 145 return PreviousTokenSource->peekNextToken(); 146 } 147 148 FormatToken *peekNextToken(int N) override { 149 assert(N > 0); 150 if (eof()) 151 return &FakeEOF; 152 return PreviousTokenSource->peekNextToken(N); 153 } 154 155 bool isEOF() override { return PreviousTokenSource->isEOF(); } 156 157 unsigned getPosition() override { return PreviousTokenSource->getPosition(); } 158 159 FormatToken *setPosition(unsigned Position) override { 160 PreviousToken = nullptr; 161 Token = PreviousTokenSource->setPosition(Position); 162 return Token; 163 } 164 165 private: 166 bool eof() { 167 return Token && Token->HasUnescapedNewline && 168 !continuesLineComment(*Token, PreviousToken, 169 /*MinColumnToken=*/PreviousToken); 170 } 171 172 FormatToken FakeEOF; 173 UnwrappedLine &Line; 174 FormatTokenSource *&TokenSource; 175 FormatToken *&ResetToken; 176 unsigned PreviousLineLevel; 177 FormatTokenSource *PreviousTokenSource; 178 179 FormatToken *Token; 180 FormatToken *PreviousToken; 181 }; 182 183 } // end anonymous namespace 184 185 class ScopedLineState { 186 public: 187 ScopedLineState(UnwrappedLineParser &Parser, 188 bool SwitchToPreprocessorLines = false) 189 : Parser(Parser), OriginalLines(Parser.CurrentLines) { 190 if (SwitchToPreprocessorLines) 191 Parser.CurrentLines = &Parser.PreprocessorDirectives; 192 else if (!Parser.Line->Tokens.empty()) 193 Parser.CurrentLines = &Parser.Line->Tokens.back().Children; 194 PreBlockLine = std::move(Parser.Line); 195 Parser.Line = std::make_unique<UnwrappedLine>(); 196 Parser.Line->Level = PreBlockLine->Level; 197 Parser.Line->InPPDirective = PreBlockLine->InPPDirective; 198 } 199 200 ~ScopedLineState() { 201 if (!Parser.Line->Tokens.empty()) 202 Parser.addUnwrappedLine(); 203 assert(Parser.Line->Tokens.empty()); 204 Parser.Line = std::move(PreBlockLine); 205 if (Parser.CurrentLines == &Parser.PreprocessorDirectives) 206 Parser.MustBreakBeforeNextToken = true; 207 Parser.CurrentLines = OriginalLines; 208 } 209 210 private: 211 UnwrappedLineParser &Parser; 212 213 std::unique_ptr<UnwrappedLine> PreBlockLine; 214 SmallVectorImpl<UnwrappedLine> *OriginalLines; 215 }; 216 217 class CompoundStatementIndenter { 218 public: 219 CompoundStatementIndenter(UnwrappedLineParser *Parser, 220 const FormatStyle &Style, unsigned &LineLevel) 221 : CompoundStatementIndenter(Parser, LineLevel, 222 Style.BraceWrapping.AfterControlStatement, 223 Style.BraceWrapping.IndentBraces) {} 224 CompoundStatementIndenter(UnwrappedLineParser *Parser, unsigned &LineLevel, 225 bool WrapBrace, bool IndentBrace) 226 : LineLevel(LineLevel), OldLineLevel(LineLevel) { 227 if (WrapBrace) 228 Parser->addUnwrappedLine(); 229 if (IndentBrace) 230 ++LineLevel; 231 } 232 ~CompoundStatementIndenter() { LineLevel = OldLineLevel; } 233 234 private: 235 unsigned &LineLevel; 236 unsigned OldLineLevel; 237 }; 238 239 namespace { 240 241 class IndexedTokenSource : public FormatTokenSource { 242 public: 243 IndexedTokenSource(ArrayRef<FormatToken *> Tokens) 244 : Tokens(Tokens), Position(-1) {} 245 246 FormatToken *getNextToken() override { 247 if (Position >= 0 && Tokens[Position]->is(tok::eof)) { 248 LLVM_DEBUG({ 249 llvm::dbgs() << "Next "; 250 dbgToken(Position); 251 }); 252 return Tokens[Position]; 253 } 254 ++Position; 255 LLVM_DEBUG({ 256 llvm::dbgs() << "Next "; 257 dbgToken(Position); 258 }); 259 return Tokens[Position]; 260 } 261 262 FormatToken *getPreviousToken() override { 263 return Position > 0 ? Tokens[Position - 1] : nullptr; 264 } 265 266 FormatToken *peekNextToken() override { 267 int Next = Position + 1; 268 LLVM_DEBUG({ 269 llvm::dbgs() << "Peeking "; 270 dbgToken(Next); 271 }); 272 return Tokens[Next]; 273 } 274 275 FormatToken *peekNextToken(int N) override { 276 assert(N > 0); 277 int Next = Position + N; 278 LLVM_DEBUG({ 279 llvm::dbgs() << "Peeking (+" << (N - 1) << ") "; 280 dbgToken(Next); 281 }); 282 return Tokens[Next]; 283 } 284 285 bool isEOF() override { return Tokens[Position]->is(tok::eof); } 286 287 unsigned getPosition() override { 288 LLVM_DEBUG(llvm::dbgs() << "Getting Position: " << Position << "\n"); 289 assert(Position >= 0); 290 return Position; 291 } 292 293 FormatToken *setPosition(unsigned P) override { 294 LLVM_DEBUG(llvm::dbgs() << "Setting Position: " << P << "\n"); 295 Position = P; 296 return Tokens[Position]; 297 } 298 299 void reset() { Position = -1; } 300 301 private: 302 void dbgToken(int Position, llvm::StringRef Indent = "") { 303 FormatToken *Tok = Tokens[Position]; 304 llvm::dbgs() << Indent << "[" << Position 305 << "] Token: " << Tok->Tok.getName() << " / " << Tok->TokenText 306 << ", Macro: " << !!Tok->MacroCtx << "\n"; 307 } 308 309 ArrayRef<FormatToken *> Tokens; 310 int Position; 311 }; 312 313 } // end anonymous namespace 314 315 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style, 316 const AdditionalKeywords &Keywords, 317 unsigned FirstStartColumn, 318 ArrayRef<FormatToken *> Tokens, 319 UnwrappedLineConsumer &Callback) 320 : Line(new UnwrappedLine), MustBreakBeforeNextToken(false), 321 CurrentLines(&Lines), Style(Style), Keywords(Keywords), 322 CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr), 323 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1), 324 IncludeGuard(Style.IndentPPDirectives == FormatStyle::PPDIS_None 325 ? IG_Rejected 326 : IG_Inited), 327 IncludeGuardToken(nullptr), FirstStartColumn(FirstStartColumn) {} 328 329 void UnwrappedLineParser::reset() { 330 PPBranchLevel = -1; 331 IncludeGuard = Style.IndentPPDirectives == FormatStyle::PPDIS_None 332 ? IG_Rejected 333 : IG_Inited; 334 IncludeGuardToken = nullptr; 335 Line.reset(new UnwrappedLine); 336 CommentsBeforeNextToken.clear(); 337 FormatTok = nullptr; 338 MustBreakBeforeNextToken = false; 339 PreprocessorDirectives.clear(); 340 CurrentLines = &Lines; 341 DeclarationScopeStack.clear(); 342 NestedTooDeep.clear(); 343 PPStack.clear(); 344 Line->FirstStartColumn = FirstStartColumn; 345 } 346 347 void UnwrappedLineParser::parse() { 348 IndexedTokenSource TokenSource(AllTokens); 349 Line->FirstStartColumn = FirstStartColumn; 350 do { 351 LLVM_DEBUG(llvm::dbgs() << "----\n"); 352 reset(); 353 Tokens = &TokenSource; 354 TokenSource.reset(); 355 356 readToken(); 357 parseFile(); 358 359 // If we found an include guard then all preprocessor directives (other than 360 // the guard) are over-indented by one. 361 if (IncludeGuard == IG_Found) { 362 for (auto &Line : Lines) 363 if (Line.InPPDirective && Line.Level > 0) 364 --Line.Level; 365 } 366 367 // Create line with eof token. 368 pushToken(FormatTok); 369 addUnwrappedLine(); 370 371 for (const UnwrappedLine &Line : Lines) 372 Callback.consumeUnwrappedLine(Line); 373 374 Callback.finishRun(); 375 Lines.clear(); 376 while (!PPLevelBranchIndex.empty() && 377 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) { 378 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1); 379 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1); 380 } 381 if (!PPLevelBranchIndex.empty()) { 382 ++PPLevelBranchIndex.back(); 383 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size()); 384 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back()); 385 } 386 } while (!PPLevelBranchIndex.empty()); 387 } 388 389 void UnwrappedLineParser::parseFile() { 390 // The top-level context in a file always has declarations, except for pre- 391 // processor directives and JavaScript files. 392 bool MustBeDeclaration = !Line->InPPDirective && !Style.isJavaScript(); 393 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 394 MustBeDeclaration); 395 if (Style.Language == FormatStyle::LK_TextProto) 396 parseBracedList(); 397 else 398 parseLevel(); 399 // Make sure to format the remaining tokens. 400 // 401 // LK_TextProto is special since its top-level is parsed as the body of a 402 // braced list, which does not necessarily have natural line separators such 403 // as a semicolon. Comments after the last entry that have been determined to 404 // not belong to that line, as in: 405 // key: value 406 // // endfile comment 407 // do not have a chance to be put on a line of their own until this point. 408 // Here we add this newline before end-of-file comments. 409 if (Style.Language == FormatStyle::LK_TextProto && 410 !CommentsBeforeNextToken.empty()) { 411 addUnwrappedLine(); 412 } 413 flushComments(true); 414 addUnwrappedLine(); 415 } 416 417 void UnwrappedLineParser::parseCSharpGenericTypeConstraint() { 418 do { 419 switch (FormatTok->Tok.getKind()) { 420 case tok::l_brace: 421 return; 422 default: 423 if (FormatTok->is(Keywords.kw_where)) { 424 addUnwrappedLine(); 425 nextToken(); 426 parseCSharpGenericTypeConstraint(); 427 break; 428 } 429 nextToken(); 430 break; 431 } 432 } while (!eof()); 433 } 434 435 void UnwrappedLineParser::parseCSharpAttribute() { 436 int UnpairedSquareBrackets = 1; 437 do { 438 switch (FormatTok->Tok.getKind()) { 439 case tok::r_square: 440 nextToken(); 441 --UnpairedSquareBrackets; 442 if (UnpairedSquareBrackets == 0) { 443 addUnwrappedLine(); 444 return; 445 } 446 break; 447 case tok::l_square: 448 ++UnpairedSquareBrackets; 449 nextToken(); 450 break; 451 default: 452 nextToken(); 453 break; 454 } 455 } while (!eof()); 456 } 457 458 bool UnwrappedLineParser::precededByCommentOrPPDirective() const { 459 if (!Lines.empty() && Lines.back().InPPDirective) 460 return true; 461 462 const FormatToken *Previous = Tokens->getPreviousToken(); 463 return Previous && Previous->is(tok::comment) && 464 (Previous->IsMultiline || Previous->NewlinesBefore > 0); 465 } 466 467 /// \brief Parses a level, that is ???. 468 /// \param OpeningBrace Opening brace (\p nullptr if absent) of that level 469 /// \param CanContainBracedList If the content can contain (at any level) a 470 /// braced list. 471 /// \param NextLBracesType The type for left brace found in this level. 472 /// \param IfKind The \p if statement kind in the level. 473 /// \param IfLeftBrace The left brace of the \p if block in the level. 474 /// \returns true if a simple block of if/else/for/while, or false otherwise. 475 /// (A simple block has a single statement.) 476 bool UnwrappedLineParser::parseLevel(const FormatToken *OpeningBrace, 477 bool CanContainBracedList, 478 TokenType NextLBracesType, 479 IfStmtKind *IfKind, 480 FormatToken **IfLeftBrace) { 481 auto NextLevelLBracesType = NextLBracesType == TT_CompoundRequirementLBrace 482 ? TT_BracedListLBrace 483 : TT_Unknown; 484 const bool IsPrecededByCommentOrPPDirective = 485 !Style.RemoveBracesLLVM || precededByCommentOrPPDirective(); 486 FormatToken *IfLBrace = nullptr; 487 bool HasDoWhile = false; 488 bool HasLabel = false; 489 unsigned StatementCount = 0; 490 bool SwitchLabelEncountered = false; 491 492 do { 493 if (FormatTok->getType() == TT_AttributeMacro) { 494 nextToken(); 495 continue; 496 } 497 tok::TokenKind kind = FormatTok->Tok.getKind(); 498 if (FormatTok->getType() == TT_MacroBlockBegin) 499 kind = tok::l_brace; 500 else if (FormatTok->getType() == TT_MacroBlockEnd) 501 kind = tok::r_brace; 502 503 auto ParseDefault = [this, OpeningBrace, NextLevelLBracesType, IfKind, 504 &IfLBrace, &HasDoWhile, &HasLabel, &StatementCount] { 505 parseStructuralElement(!OpeningBrace, NextLevelLBracesType, IfKind, 506 &IfLBrace, HasDoWhile ? nullptr : &HasDoWhile, 507 HasLabel ? nullptr : &HasLabel); 508 ++StatementCount; 509 assert(StatementCount > 0 && "StatementCount overflow!"); 510 }; 511 512 switch (kind) { 513 case tok::comment: 514 nextToken(); 515 addUnwrappedLine(); 516 break; 517 case tok::l_brace: 518 if (NextLBracesType != TT_Unknown) { 519 FormatTok->setFinalizedType(NextLBracesType); 520 } else if (FormatTok->Previous && 521 FormatTok->Previous->ClosesRequiresClause) { 522 // We need the 'default' case here to correctly parse a function 523 // l_brace. 524 ParseDefault(); 525 continue; 526 } 527 if (CanContainBracedList && !FormatTok->is(TT_MacroBlockBegin) && 528 tryToParseBracedList()) { 529 continue; 530 } 531 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 532 /*MunchSemi=*/true, /*KeepBraces=*/true, /*IfKind=*/nullptr, 533 /*UnindentWhitesmithsBraces=*/false, CanContainBracedList, 534 NextLBracesType); 535 ++StatementCount; 536 assert(StatementCount > 0 && "StatementCount overflow!"); 537 addUnwrappedLine(); 538 break; 539 case tok::r_brace: 540 if (OpeningBrace) { 541 if (!Style.RemoveBracesLLVM || 542 !OpeningBrace->isOneOf(TT_ControlStatementLBrace, TT_ElseLBrace)) { 543 return false; 544 } 545 if (FormatTok->isNot(tok::r_brace) || StatementCount != 1 || HasLabel || 546 HasDoWhile || IsPrecededByCommentOrPPDirective || 547 precededByCommentOrPPDirective()) { 548 return false; 549 } 550 const FormatToken *Next = Tokens->peekNextToken(); 551 if (Next->is(tok::comment) && Next->NewlinesBefore == 0) 552 return false; 553 if (IfLeftBrace) 554 *IfLeftBrace = IfLBrace; 555 return true; 556 } 557 nextToken(); 558 addUnwrappedLine(); 559 break; 560 case tok::kw_default: { 561 unsigned StoredPosition = Tokens->getPosition(); 562 FormatToken *Next; 563 do { 564 Next = Tokens->getNextToken(); 565 assert(Next); 566 } while (Next->is(tok::comment)); 567 FormatTok = Tokens->setPosition(StoredPosition); 568 if (Next->isNot(tok::colon)) { 569 // default not followed by ':' is not a case label; treat it like 570 // an identifier. 571 parseStructuralElement(); 572 break; 573 } 574 // Else, if it is 'default:', fall through to the case handling. 575 LLVM_FALLTHROUGH; 576 } 577 case tok::kw_case: 578 if (Style.isJavaScript() && Line->MustBeDeclaration) { 579 // A 'case: string' style field declaration. 580 parseStructuralElement(); 581 break; 582 } 583 if (!SwitchLabelEncountered && 584 (Style.IndentCaseLabels || 585 (Line->InPPDirective && Line->Level == 1))) { 586 ++Line->Level; 587 } 588 SwitchLabelEncountered = true; 589 parseStructuralElement(); 590 break; 591 case tok::l_square: 592 if (Style.isCSharp()) { 593 nextToken(); 594 parseCSharpAttribute(); 595 break; 596 } 597 if (handleCppAttributes()) 598 break; 599 LLVM_FALLTHROUGH; 600 default: 601 ParseDefault(); 602 break; 603 } 604 } while (!eof()); 605 606 return false; 607 } 608 609 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) { 610 // We'll parse forward through the tokens until we hit 611 // a closing brace or eof - note that getNextToken() will 612 // parse macros, so this will magically work inside macro 613 // definitions, too. 614 unsigned StoredPosition = Tokens->getPosition(); 615 FormatToken *Tok = FormatTok; 616 const FormatToken *PrevTok = Tok->Previous; 617 // Keep a stack of positions of lbrace tokens. We will 618 // update information about whether an lbrace starts a 619 // braced init list or a different block during the loop. 620 SmallVector<FormatToken *, 8> LBraceStack; 621 assert(Tok->is(tok::l_brace)); 622 do { 623 // Get next non-comment token. 624 FormatToken *NextTok; 625 do { 626 NextTok = Tokens->getNextToken(); 627 } while (NextTok->is(tok::comment)); 628 629 switch (Tok->Tok.getKind()) { 630 case tok::l_brace: 631 if (Style.isJavaScript() && PrevTok) { 632 if (PrevTok->isOneOf(tok::colon, tok::less)) { 633 // A ':' indicates this code is in a type, or a braced list 634 // following a label in an object literal ({a: {b: 1}}). 635 // A '<' could be an object used in a comparison, but that is nonsense 636 // code (can never return true), so more likely it is a generic type 637 // argument (`X<{a: string; b: number}>`). 638 // The code below could be confused by semicolons between the 639 // individual members in a type member list, which would normally 640 // trigger BK_Block. In both cases, this must be parsed as an inline 641 // braced init. 642 Tok->setBlockKind(BK_BracedInit); 643 } else if (PrevTok->is(tok::r_paren)) { 644 // `) { }` can only occur in function or method declarations in JS. 645 Tok->setBlockKind(BK_Block); 646 } 647 } else { 648 Tok->setBlockKind(BK_Unknown); 649 } 650 LBraceStack.push_back(Tok); 651 break; 652 case tok::r_brace: 653 if (LBraceStack.empty()) 654 break; 655 if (LBraceStack.back()->is(BK_Unknown)) { 656 bool ProbablyBracedList = false; 657 if (Style.Language == FormatStyle::LK_Proto) { 658 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square); 659 } else { 660 // Skip NextTok over preprocessor lines, otherwise we may not 661 // properly diagnose the block as a braced intializer 662 // if the comma separator appears after the pp directive. 663 while (NextTok->is(tok::hash)) { 664 ScopedMacroState MacroState(*Line, Tokens, NextTok); 665 do { 666 NextTok = Tokens->getNextToken(); 667 } while (NextTok->isNot(tok::eof)); 668 } 669 670 // Using OriginalColumn to distinguish between ObjC methods and 671 // binary operators is a bit hacky. 672 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 673 NextTok->OriginalColumn == 0; 674 675 // Try to detect a braced list. Note that regardless how we mark inner 676 // braces here, we will overwrite the BlockKind later if we parse a 677 // braced list (where all blocks inside are by default braced lists), 678 // or when we explicitly detect blocks (for example while parsing 679 // lambdas). 680 681 // If we already marked the opening brace as braced list, the closing 682 // must also be part of it. 683 ProbablyBracedList = LBraceStack.back()->is(TT_BracedListLBrace); 684 685 ProbablyBracedList = ProbablyBracedList || 686 (Style.isJavaScript() && 687 NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in, 688 Keywords.kw_as)); 689 ProbablyBracedList = ProbablyBracedList || 690 (Style.isCpp() && NextTok->is(tok::l_paren)); 691 692 // If there is a comma, semicolon or right paren after the closing 693 // brace, we assume this is a braced initializer list. 694 // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a 695 // braced list in JS. 696 ProbablyBracedList = 697 ProbablyBracedList || 698 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 699 tok::r_paren, tok::r_square, tok::l_brace, 700 tok::ellipsis); 701 702 ProbablyBracedList = 703 ProbablyBracedList || 704 (NextTok->is(tok::identifier) && 705 !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace)); 706 707 ProbablyBracedList = ProbablyBracedList || 708 (NextTok->is(tok::semi) && 709 (!ExpectClassBody || LBraceStack.size() != 1)); 710 711 ProbablyBracedList = 712 ProbablyBracedList || 713 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 714 715 if (!Style.isCSharp() && NextTok->is(tok::l_square)) { 716 // We can have an array subscript after a braced init 717 // list, but C++11 attributes are expected after blocks. 718 NextTok = Tokens->getNextToken(); 719 ProbablyBracedList = NextTok->isNot(tok::l_square); 720 } 721 } 722 if (ProbablyBracedList) { 723 Tok->setBlockKind(BK_BracedInit); 724 LBraceStack.back()->setBlockKind(BK_BracedInit); 725 } else { 726 Tok->setBlockKind(BK_Block); 727 LBraceStack.back()->setBlockKind(BK_Block); 728 } 729 } 730 LBraceStack.pop_back(); 731 break; 732 case tok::identifier: 733 if (!Tok->is(TT_StatementMacro)) 734 break; 735 LLVM_FALLTHROUGH; 736 case tok::at: 737 case tok::semi: 738 case tok::kw_if: 739 case tok::kw_while: 740 case tok::kw_for: 741 case tok::kw_switch: 742 case tok::kw_try: 743 case tok::kw___try: 744 if (!LBraceStack.empty() && LBraceStack.back()->is(BK_Unknown)) 745 LBraceStack.back()->setBlockKind(BK_Block); 746 break; 747 default: 748 break; 749 } 750 PrevTok = Tok; 751 Tok = NextTok; 752 } while (Tok->isNot(tok::eof) && !LBraceStack.empty()); 753 754 // Assume other blocks for all unclosed opening braces. 755 for (FormatToken *LBrace : LBraceStack) 756 if (LBrace->is(BK_Unknown)) 757 LBrace->setBlockKind(BK_Block); 758 759 FormatTok = Tokens->setPosition(StoredPosition); 760 } 761 762 template <class T> 763 static inline void hash_combine(std::size_t &seed, const T &v) { 764 std::hash<T> hasher; 765 seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2); 766 } 767 768 size_t UnwrappedLineParser::computePPHash() const { 769 size_t h = 0; 770 for (const auto &i : PPStack) { 771 hash_combine(h, size_t(i.Kind)); 772 hash_combine(h, i.Line); 773 } 774 return h; 775 } 776 777 // Checks whether \p ParsedLine might fit on a single line. If \p OpeningBrace 778 // is not null, subtracts its length (plus the preceding space) when computing 779 // the length of \p ParsedLine. We must clone the tokens of \p ParsedLine before 780 // running the token annotator on it so that we can restore them afterward. 781 bool UnwrappedLineParser::mightFitOnOneLine( 782 UnwrappedLine &ParsedLine, const FormatToken *OpeningBrace) const { 783 const auto ColumnLimit = Style.ColumnLimit; 784 if (ColumnLimit == 0) 785 return true; 786 787 auto &Tokens = ParsedLine.Tokens; 788 assert(!Tokens.empty()); 789 790 const auto *LastToken = Tokens.back().Tok; 791 assert(LastToken); 792 793 SmallVector<UnwrappedLineNode> SavedTokens(Tokens.size()); 794 795 int Index = 0; 796 for (const auto &Token : Tokens) { 797 assert(Token.Tok); 798 auto &SavedToken = SavedTokens[Index++]; 799 SavedToken.Tok = new FormatToken; 800 SavedToken.Tok->copyFrom(*Token.Tok); 801 SavedToken.Children = std::move(Token.Children); 802 } 803 804 AnnotatedLine Line(ParsedLine); 805 assert(Line.Last == LastToken); 806 807 TokenAnnotator Annotator(Style, Keywords); 808 Annotator.annotate(Line); 809 Annotator.calculateFormattingInformation(Line); 810 811 auto Length = LastToken->TotalLength; 812 if (OpeningBrace) { 813 assert(OpeningBrace != Tokens.front().Tok); 814 Length -= OpeningBrace->TokenText.size() + 1; 815 } 816 817 Index = 0; 818 for (auto &Token : Tokens) { 819 const auto &SavedToken = SavedTokens[Index++]; 820 Token.Tok->copyFrom(*SavedToken.Tok); 821 Token.Children = std::move(SavedToken.Children); 822 delete SavedToken.Tok; 823 } 824 825 return Line.Level * Style.IndentWidth + Length <= ColumnLimit; 826 } 827 828 FormatToken *UnwrappedLineParser::parseBlock( 829 bool MustBeDeclaration, unsigned AddLevels, bool MunchSemi, bool KeepBraces, 830 IfStmtKind *IfKind, bool UnindentWhitesmithsBraces, 831 bool CanContainBracedList, TokenType NextLBracesType) { 832 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) && 833 "'{' or macro block token expected"); 834 FormatToken *Tok = FormatTok; 835 const bool FollowedByComment = Tokens->peekNextToken()->is(tok::comment); 836 auto Index = CurrentLines->size(); 837 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 838 FormatTok->setBlockKind(BK_Block); 839 840 // For Whitesmiths mode, jump to the next level prior to skipping over the 841 // braces. 842 if (AddLevels > 0 && Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) 843 ++Line->Level; 844 845 size_t PPStartHash = computePPHash(); 846 847 unsigned InitialLevel = Line->Level; 848 nextToken(/*LevelDifference=*/AddLevels); 849 850 // Bail out if there are too many levels. Otherwise, the stack might overflow. 851 if (Line->Level > 300) 852 return nullptr; 853 854 if (MacroBlock && FormatTok->is(tok::l_paren)) 855 parseParens(); 856 857 size_t NbPreprocessorDirectives = 858 CurrentLines == &Lines ? PreprocessorDirectives.size() : 0; 859 addUnwrappedLine(); 860 size_t OpeningLineIndex = 861 CurrentLines->empty() 862 ? (UnwrappedLine::kInvalidIndex) 863 : (CurrentLines->size() - 1 - NbPreprocessorDirectives); 864 865 // Whitesmiths is weird here. The brace needs to be indented for the namespace 866 // block, but the block itself may not be indented depending on the style 867 // settings. This allows the format to back up one level in those cases. 868 if (UnindentWhitesmithsBraces) 869 --Line->Level; 870 871 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 872 MustBeDeclaration); 873 if (AddLevels > 0u && Style.BreakBeforeBraces != FormatStyle::BS_Whitesmiths) 874 Line->Level += AddLevels; 875 876 FormatToken *IfLBrace = nullptr; 877 const bool SimpleBlock = 878 parseLevel(Tok, CanContainBracedList, NextLBracesType, IfKind, &IfLBrace); 879 880 if (eof()) 881 return IfLBrace; 882 883 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd) 884 : !FormatTok->is(tok::r_brace)) { 885 Line->Level = InitialLevel; 886 FormatTok->setBlockKind(BK_Block); 887 return IfLBrace; 888 } 889 890 auto RemoveBraces = [=]() mutable { 891 if (!SimpleBlock) 892 return false; 893 assert(Tok->isOneOf(TT_ControlStatementLBrace, TT_ElseLBrace)); 894 assert(FormatTok->is(tok::r_brace)); 895 const bool WrappedOpeningBrace = !Tok->Previous; 896 if (WrappedOpeningBrace && FollowedByComment) 897 return false; 898 const bool HasRequiredIfBraces = IfLBrace && !IfLBrace->Optional; 899 if (KeepBraces && !HasRequiredIfBraces) 900 return false; 901 if (Tok->isNot(TT_ElseLBrace) || !HasRequiredIfBraces) { 902 const FormatToken *Previous = Tokens->getPreviousToken(); 903 assert(Previous); 904 if (Previous->is(tok::r_brace) && !Previous->Optional) 905 return false; 906 } 907 assert(!CurrentLines->empty()); 908 if (!mightFitOnOneLine(CurrentLines->back())) 909 return false; 910 if (Tok->is(TT_ElseLBrace)) 911 return true; 912 if (WrappedOpeningBrace) { 913 assert(Index > 0); 914 --Index; // The line above the wrapped l_brace. 915 Tok = nullptr; 916 } 917 return mightFitOnOneLine((*CurrentLines)[Index], Tok); 918 }; 919 if (RemoveBraces()) { 920 Tok->MatchingParen = FormatTok; 921 FormatTok->MatchingParen = Tok; 922 } 923 924 size_t PPEndHash = computePPHash(); 925 926 // Munch the closing brace. 927 nextToken(/*LevelDifference=*/-AddLevels); 928 929 if (MacroBlock && FormatTok->is(tok::l_paren)) 930 parseParens(); 931 932 if (FormatTok->is(tok::kw_noexcept)) { 933 // A noexcept in a requires expression. 934 nextToken(); 935 } 936 937 if (FormatTok->is(tok::arrow)) { 938 // Following the } or noexcept we can find a trailing return type arrow 939 // as part of an implicit conversion constraint. 940 nextToken(); 941 parseStructuralElement(); 942 } 943 944 if (MunchSemi && FormatTok->is(tok::semi)) 945 nextToken(); 946 947 Line->Level = InitialLevel; 948 949 if (PPStartHash == PPEndHash) { 950 Line->MatchingOpeningBlockLineIndex = OpeningLineIndex; 951 if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) { 952 // Update the opening line to add the forward reference as well 953 (*CurrentLines)[OpeningLineIndex].MatchingClosingBlockLineIndex = 954 CurrentLines->size() - 1; 955 } 956 } 957 958 return IfLBrace; 959 } 960 961 static bool isGoogScope(const UnwrappedLine &Line) { 962 // FIXME: Closure-library specific stuff should not be hard-coded but be 963 // configurable. 964 if (Line.Tokens.size() < 4) 965 return false; 966 auto I = Line.Tokens.begin(); 967 if (I->Tok->TokenText != "goog") 968 return false; 969 ++I; 970 if (I->Tok->isNot(tok::period)) 971 return false; 972 ++I; 973 if (I->Tok->TokenText != "scope") 974 return false; 975 ++I; 976 return I->Tok->is(tok::l_paren); 977 } 978 979 static bool isIIFE(const UnwrappedLine &Line, 980 const AdditionalKeywords &Keywords) { 981 // Look for the start of an immediately invoked anonymous function. 982 // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression 983 // This is commonly done in JavaScript to create a new, anonymous scope. 984 // Example: (function() { ... })() 985 if (Line.Tokens.size() < 3) 986 return false; 987 auto I = Line.Tokens.begin(); 988 if (I->Tok->isNot(tok::l_paren)) 989 return false; 990 ++I; 991 if (I->Tok->isNot(Keywords.kw_function)) 992 return false; 993 ++I; 994 return I->Tok->is(tok::l_paren); 995 } 996 997 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 998 const FormatToken &InitialToken) { 999 tok::TokenKind Kind = InitialToken.Tok.getKind(); 1000 if (InitialToken.is(TT_NamespaceMacro)) 1001 Kind = tok::kw_namespace; 1002 1003 switch (Kind) { 1004 case tok::kw_namespace: 1005 return Style.BraceWrapping.AfterNamespace; 1006 case tok::kw_class: 1007 return Style.BraceWrapping.AfterClass; 1008 case tok::kw_union: 1009 return Style.BraceWrapping.AfterUnion; 1010 case tok::kw_struct: 1011 return Style.BraceWrapping.AfterStruct; 1012 case tok::kw_enum: 1013 return Style.BraceWrapping.AfterEnum; 1014 default: 1015 return false; 1016 } 1017 } 1018 1019 void UnwrappedLineParser::parseChildBlock( 1020 bool CanContainBracedList, clang::format::TokenType NextLBracesType) { 1021 assert(FormatTok->is(tok::l_brace)); 1022 FormatTok->setBlockKind(BK_Block); 1023 const FormatToken *OpeningBrace = FormatTok; 1024 nextToken(); 1025 { 1026 bool SkipIndent = (Style.isJavaScript() && 1027 (isGoogScope(*Line) || isIIFE(*Line, Keywords))); 1028 ScopedLineState LineState(*this); 1029 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 1030 /*MustBeDeclaration=*/false); 1031 Line->Level += SkipIndent ? 0 : 1; 1032 parseLevel(OpeningBrace, CanContainBracedList, NextLBracesType); 1033 flushComments(isOnNewLine(*FormatTok)); 1034 Line->Level -= SkipIndent ? 0 : 1; 1035 } 1036 nextToken(); 1037 } 1038 1039 void UnwrappedLineParser::parsePPDirective() { 1040 assert(FormatTok->is(tok::hash) && "'#' expected"); 1041 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 1042 1043 nextToken(); 1044 1045 if (!FormatTok->Tok.getIdentifierInfo()) { 1046 parsePPUnknown(); 1047 return; 1048 } 1049 1050 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 1051 case tok::pp_define: 1052 parsePPDefine(); 1053 return; 1054 case tok::pp_if: 1055 parsePPIf(/*IfDef=*/false); 1056 break; 1057 case tok::pp_ifdef: 1058 case tok::pp_ifndef: 1059 parsePPIf(/*IfDef=*/true); 1060 break; 1061 case tok::pp_else: 1062 parsePPElse(); 1063 break; 1064 case tok::pp_elifdef: 1065 case tok::pp_elifndef: 1066 case tok::pp_elif: 1067 parsePPElIf(); 1068 break; 1069 case tok::pp_endif: 1070 parsePPEndIf(); 1071 break; 1072 default: 1073 parsePPUnknown(); 1074 break; 1075 } 1076 } 1077 1078 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 1079 size_t Line = CurrentLines->size(); 1080 if (CurrentLines == &PreprocessorDirectives) 1081 Line += Lines.size(); 1082 1083 if (Unreachable || 1084 (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable)) { 1085 PPStack.push_back({PP_Unreachable, Line}); 1086 } else { 1087 PPStack.push_back({PP_Conditional, Line}); 1088 } 1089 } 1090 1091 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 1092 ++PPBranchLevel; 1093 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 1094 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 1095 PPLevelBranchIndex.push_back(0); 1096 PPLevelBranchCount.push_back(0); 1097 } 1098 PPChainBranchIndex.push(0); 1099 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 1100 conditionalCompilationCondition(Unreachable || Skip); 1101 } 1102 1103 void UnwrappedLineParser::conditionalCompilationAlternative() { 1104 if (!PPStack.empty()) 1105 PPStack.pop_back(); 1106 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 1107 if (!PPChainBranchIndex.empty()) 1108 ++PPChainBranchIndex.top(); 1109 conditionalCompilationCondition( 1110 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 1111 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 1112 } 1113 1114 void UnwrappedLineParser::conditionalCompilationEnd() { 1115 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 1116 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 1117 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) 1118 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 1119 } 1120 // Guard against #endif's without #if. 1121 if (PPBranchLevel > -1) 1122 --PPBranchLevel; 1123 if (!PPChainBranchIndex.empty()) 1124 PPChainBranchIndex.pop(); 1125 if (!PPStack.empty()) 1126 PPStack.pop_back(); 1127 } 1128 1129 void UnwrappedLineParser::parsePPIf(bool IfDef) { 1130 bool IfNDef = FormatTok->is(tok::pp_ifndef); 1131 nextToken(); 1132 bool Unreachable = false; 1133 if (!IfDef && (FormatTok->is(tok::kw_false) || FormatTok->TokenText == "0")) 1134 Unreachable = true; 1135 if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG") 1136 Unreachable = true; 1137 conditionalCompilationStart(Unreachable); 1138 FormatToken *IfCondition = FormatTok; 1139 // If there's a #ifndef on the first line, and the only lines before it are 1140 // comments, it could be an include guard. 1141 bool MaybeIncludeGuard = IfNDef; 1142 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) { 1143 for (auto &Line : Lines) { 1144 if (!Line.Tokens.front().Tok->is(tok::comment)) { 1145 MaybeIncludeGuard = false; 1146 IncludeGuard = IG_Rejected; 1147 break; 1148 } 1149 } 1150 } 1151 --PPBranchLevel; 1152 parsePPUnknown(); 1153 ++PPBranchLevel; 1154 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) { 1155 IncludeGuard = IG_IfNdefed; 1156 IncludeGuardToken = IfCondition; 1157 } 1158 } 1159 1160 void UnwrappedLineParser::parsePPElse() { 1161 // If a potential include guard has an #else, it's not an include guard. 1162 if (IncludeGuard == IG_Defined && PPBranchLevel == 0) 1163 IncludeGuard = IG_Rejected; 1164 conditionalCompilationAlternative(); 1165 if (PPBranchLevel > -1) 1166 --PPBranchLevel; 1167 parsePPUnknown(); 1168 ++PPBranchLevel; 1169 } 1170 1171 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 1172 1173 void UnwrappedLineParser::parsePPEndIf() { 1174 conditionalCompilationEnd(); 1175 parsePPUnknown(); 1176 // If the #endif of a potential include guard is the last thing in the file, 1177 // then we found an include guard. 1178 if (IncludeGuard == IG_Defined && PPBranchLevel == -1 && Tokens->isEOF() && 1179 Style.IndentPPDirectives != FormatStyle::PPDIS_None) { 1180 IncludeGuard = IG_Found; 1181 } 1182 } 1183 1184 void UnwrappedLineParser::parsePPDefine() { 1185 nextToken(); 1186 1187 if (!FormatTok->Tok.getIdentifierInfo()) { 1188 IncludeGuard = IG_Rejected; 1189 IncludeGuardToken = nullptr; 1190 parsePPUnknown(); 1191 return; 1192 } 1193 1194 if (IncludeGuard == IG_IfNdefed && 1195 IncludeGuardToken->TokenText == FormatTok->TokenText) { 1196 IncludeGuard = IG_Defined; 1197 IncludeGuardToken = nullptr; 1198 for (auto &Line : Lines) { 1199 if (!Line.Tokens.front().Tok->isOneOf(tok::comment, tok::hash)) { 1200 IncludeGuard = IG_Rejected; 1201 break; 1202 } 1203 } 1204 } 1205 1206 // In the context of a define, even keywords should be treated as normal 1207 // identifiers. Setting the kind to identifier is not enough, because we need 1208 // to treat additional keywords like __except as well, which are already 1209 // identifiers. Setting the identifier info to null interferes with include 1210 // guard processing above, and changes preprocessing nesting. 1211 FormatTok->Tok.setKind(tok::identifier); 1212 FormatTok->Tok.setIdentifierInfo(Keywords.kw_internal_ident_after_define); 1213 nextToken(); 1214 if (FormatTok->Tok.getKind() == tok::l_paren && 1215 !FormatTok->hasWhitespaceBefore()) { 1216 parseParens(); 1217 } 1218 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None) 1219 Line->Level += PPBranchLevel + 1; 1220 addUnwrappedLine(); 1221 ++Line->Level; 1222 1223 // Errors during a preprocessor directive can only affect the layout of the 1224 // preprocessor directive, and thus we ignore them. An alternative approach 1225 // would be to use the same approach we use on the file level (no 1226 // re-indentation if there was a structural error) within the macro 1227 // definition. 1228 parseFile(); 1229 } 1230 1231 void UnwrappedLineParser::parsePPUnknown() { 1232 do { 1233 nextToken(); 1234 } while (!eof()); 1235 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None) 1236 Line->Level += PPBranchLevel + 1; 1237 addUnwrappedLine(); 1238 } 1239 1240 // Here we exclude certain tokens that are not usually the first token in an 1241 // unwrapped line. This is used in attempt to distinguish macro calls without 1242 // trailing semicolons from other constructs split to several lines. 1243 static bool tokenCanStartNewLine(const FormatToken &Tok) { 1244 // Semicolon can be a null-statement, l_square can be a start of a macro or 1245 // a C++11 attribute, but this doesn't seem to be common. 1246 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 1247 Tok.isNot(TT_AttributeSquare) && 1248 // Tokens that can only be used as binary operators and a part of 1249 // overloaded operator names. 1250 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 1251 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 1252 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 1253 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 1254 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 1255 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 1256 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 1257 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 1258 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 1259 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 1260 Tok.isNot(tok::lesslessequal) && 1261 // Colon is used in labels, base class lists, initializer lists, 1262 // range-based for loops, ternary operator, but should never be the 1263 // first token in an unwrapped line. 1264 Tok.isNot(tok::colon) && 1265 // 'noexcept' is a trailing annotation. 1266 Tok.isNot(tok::kw_noexcept); 1267 } 1268 1269 static bool mustBeJSIdent(const AdditionalKeywords &Keywords, 1270 const FormatToken *FormatTok) { 1271 // FIXME: This returns true for C/C++ keywords like 'struct'. 1272 return FormatTok->is(tok::identifier) && 1273 (FormatTok->Tok.getIdentifierInfo() == nullptr || 1274 !FormatTok->isOneOf( 1275 Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async, 1276 Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally, 1277 Keywords.kw_function, Keywords.kw_import, Keywords.kw_is, 1278 Keywords.kw_let, Keywords.kw_var, tok::kw_const, 1279 Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements, 1280 Keywords.kw_instanceof, Keywords.kw_interface, 1281 Keywords.kw_override, Keywords.kw_throws, Keywords.kw_from)); 1282 } 1283 1284 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords, 1285 const FormatToken *FormatTok) { 1286 return FormatTok->Tok.isLiteral() || 1287 FormatTok->isOneOf(tok::kw_true, tok::kw_false) || 1288 mustBeJSIdent(Keywords, FormatTok); 1289 } 1290 1291 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement 1292 // when encountered after a value (see mustBeJSIdentOrValue). 1293 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords, 1294 const FormatToken *FormatTok) { 1295 return FormatTok->isOneOf( 1296 tok::kw_return, Keywords.kw_yield, 1297 // conditionals 1298 tok::kw_if, tok::kw_else, 1299 // loops 1300 tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break, 1301 // switch/case 1302 tok::kw_switch, tok::kw_case, 1303 // exceptions 1304 tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally, 1305 // declaration 1306 tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let, 1307 Keywords.kw_async, Keywords.kw_function, 1308 // import/export 1309 Keywords.kw_import, tok::kw_export); 1310 } 1311 1312 // Checks whether a token is a type in K&R C (aka C78). 1313 static bool isC78Type(const FormatToken &Tok) { 1314 return Tok.isOneOf(tok::kw_char, tok::kw_short, tok::kw_int, tok::kw_long, 1315 tok::kw_unsigned, tok::kw_float, tok::kw_double, 1316 tok::identifier); 1317 } 1318 1319 // This function checks whether a token starts the first parameter declaration 1320 // in a K&R C (aka C78) function definition, e.g.: 1321 // int f(a, b) 1322 // short a, b; 1323 // { 1324 // return a + b; 1325 // } 1326 static bool isC78ParameterDecl(const FormatToken *Tok, const FormatToken *Next, 1327 const FormatToken *FuncName) { 1328 assert(Tok); 1329 assert(Next); 1330 assert(FuncName); 1331 1332 if (FuncName->isNot(tok::identifier)) 1333 return false; 1334 1335 const FormatToken *Prev = FuncName->Previous; 1336 if (!Prev || (Prev->isNot(tok::star) && !isC78Type(*Prev))) 1337 return false; 1338 1339 if (!isC78Type(*Tok) && 1340 !Tok->isOneOf(tok::kw_register, tok::kw_struct, tok::kw_union)) { 1341 return false; 1342 } 1343 1344 if (Next->isNot(tok::star) && !Next->Tok.getIdentifierInfo()) 1345 return false; 1346 1347 Tok = Tok->Previous; 1348 if (!Tok || Tok->isNot(tok::r_paren)) 1349 return false; 1350 1351 Tok = Tok->Previous; 1352 if (!Tok || Tok->isNot(tok::identifier)) 1353 return false; 1354 1355 return Tok->Previous && Tok->Previous->isOneOf(tok::l_paren, tok::comma); 1356 } 1357 1358 void UnwrappedLineParser::parseModuleImport() { 1359 nextToken(); 1360 while (!eof()) { 1361 if (FormatTok->is(tok::colon)) { 1362 FormatTok->setFinalizedType(TT_ModulePartitionColon); 1363 } 1364 // Handle import <foo/bar.h> as we would an include statement. 1365 else if (FormatTok->is(tok::less)) { 1366 nextToken(); 1367 while (!FormatTok->isOneOf(tok::semi, tok::greater, tok::eof)) { 1368 // Mark tokens up to the trailing line comments as implicit string 1369 // literals. 1370 if (FormatTok->isNot(tok::comment) && 1371 !FormatTok->TokenText.startswith("//")) { 1372 FormatTok->setFinalizedType(TT_ImplicitStringLiteral); 1373 } 1374 nextToken(); 1375 } 1376 } 1377 if (FormatTok->is(tok::semi)) { 1378 nextToken(); 1379 break; 1380 } 1381 nextToken(); 1382 } 1383 1384 addUnwrappedLine(); 1385 } 1386 1387 // readTokenWithJavaScriptASI reads the next token and terminates the current 1388 // line if JavaScript Automatic Semicolon Insertion must 1389 // happen between the current token and the next token. 1390 // 1391 // This method is conservative - it cannot cover all edge cases of JavaScript, 1392 // but only aims to correctly handle certain well known cases. It *must not* 1393 // return true in speculative cases. 1394 void UnwrappedLineParser::readTokenWithJavaScriptASI() { 1395 FormatToken *Previous = FormatTok; 1396 readToken(); 1397 FormatToken *Next = FormatTok; 1398 1399 bool IsOnSameLine = 1400 CommentsBeforeNextToken.empty() 1401 ? Next->NewlinesBefore == 0 1402 : CommentsBeforeNextToken.front()->NewlinesBefore == 0; 1403 if (IsOnSameLine) 1404 return; 1405 1406 bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous); 1407 bool PreviousStartsTemplateExpr = 1408 Previous->is(TT_TemplateString) && Previous->TokenText.endswith("${"); 1409 if (PreviousMustBeValue || Previous->is(tok::r_paren)) { 1410 // If the line contains an '@' sign, the previous token might be an 1411 // annotation, which can precede another identifier/value. 1412 bool HasAt = llvm::any_of(Line->Tokens, [](UnwrappedLineNode &LineNode) { 1413 return LineNode.Tok->is(tok::at); 1414 }); 1415 if (HasAt) 1416 return; 1417 } 1418 if (Next->is(tok::exclaim) && PreviousMustBeValue) 1419 return addUnwrappedLine(); 1420 bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next); 1421 bool NextEndsTemplateExpr = 1422 Next->is(TT_TemplateString) && Next->TokenText.startswith("}"); 1423 if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr && 1424 (PreviousMustBeValue || 1425 Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus, 1426 tok::minusminus))) { 1427 return addUnwrappedLine(); 1428 } 1429 if ((PreviousMustBeValue || Previous->is(tok::r_paren)) && 1430 isJSDeclOrStmt(Keywords, Next)) { 1431 return addUnwrappedLine(); 1432 } 1433 } 1434 1435 void UnwrappedLineParser::parseStructuralElement( 1436 bool IsTopLevel, TokenType NextLBracesType, IfStmtKind *IfKind, 1437 FormatToken **IfLeftBrace, bool *HasDoWhile, bool *HasLabel) { 1438 if (Style.Language == FormatStyle::LK_TableGen && 1439 FormatTok->is(tok::pp_include)) { 1440 nextToken(); 1441 if (FormatTok->is(tok::string_literal)) 1442 nextToken(); 1443 addUnwrappedLine(); 1444 return; 1445 } 1446 switch (FormatTok->Tok.getKind()) { 1447 case tok::kw_asm: 1448 nextToken(); 1449 if (FormatTok->is(tok::l_brace)) { 1450 FormatTok->setFinalizedType(TT_InlineASMBrace); 1451 nextToken(); 1452 while (FormatTok && FormatTok->isNot(tok::eof)) { 1453 if (FormatTok->is(tok::r_brace)) { 1454 FormatTok->setFinalizedType(TT_InlineASMBrace); 1455 nextToken(); 1456 addUnwrappedLine(); 1457 break; 1458 } 1459 FormatTok->Finalized = true; 1460 nextToken(); 1461 } 1462 } 1463 break; 1464 case tok::kw_namespace: 1465 parseNamespace(); 1466 return; 1467 case tok::kw_public: 1468 case tok::kw_protected: 1469 case tok::kw_private: 1470 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 1471 Style.isCSharp()) { 1472 nextToken(); 1473 } else { 1474 parseAccessSpecifier(); 1475 } 1476 return; 1477 case tok::kw_if: { 1478 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1479 // field/method declaration. 1480 break; 1481 } 1482 FormatToken *Tok = parseIfThenElse(IfKind); 1483 if (IfLeftBrace) 1484 *IfLeftBrace = Tok; 1485 return; 1486 } 1487 case tok::kw_for: 1488 case tok::kw_while: 1489 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1490 // field/method declaration. 1491 break; 1492 } 1493 parseForOrWhileLoop(); 1494 return; 1495 case tok::kw_do: 1496 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1497 // field/method declaration. 1498 break; 1499 } 1500 parseDoWhile(); 1501 if (HasDoWhile) 1502 *HasDoWhile = true; 1503 return; 1504 case tok::kw_switch: 1505 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1506 // 'switch: string' field declaration. 1507 break; 1508 } 1509 parseSwitch(); 1510 return; 1511 case tok::kw_default: 1512 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1513 // 'default: string' field declaration. 1514 break; 1515 } 1516 nextToken(); 1517 if (FormatTok->is(tok::colon)) { 1518 parseLabel(); 1519 return; 1520 } 1521 // e.g. "default void f() {}" in a Java interface. 1522 break; 1523 case tok::kw_case: 1524 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1525 // 'case: string' field declaration. 1526 nextToken(); 1527 break; 1528 } 1529 parseCaseLabel(); 1530 return; 1531 case tok::kw_try: 1532 case tok::kw___try: 1533 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1534 // field/method declaration. 1535 break; 1536 } 1537 parseTryCatch(); 1538 return; 1539 case tok::kw_extern: 1540 nextToken(); 1541 if (FormatTok->is(tok::string_literal)) { 1542 nextToken(); 1543 if (FormatTok->is(tok::l_brace)) { 1544 if (Style.BraceWrapping.AfterExternBlock) 1545 addUnwrappedLine(); 1546 // Either we indent or for backwards compatibility we follow the 1547 // AfterExternBlock style. 1548 unsigned AddLevels = 1549 (Style.IndentExternBlock == FormatStyle::IEBS_Indent) || 1550 (Style.BraceWrapping.AfterExternBlock && 1551 Style.IndentExternBlock == 1552 FormatStyle::IEBS_AfterExternBlock) 1553 ? 1u 1554 : 0u; 1555 parseBlock(/*MustBeDeclaration=*/true, AddLevels); 1556 addUnwrappedLine(); 1557 return; 1558 } 1559 } 1560 break; 1561 case tok::kw_export: 1562 if (Style.isJavaScript()) { 1563 parseJavaScriptEs6ImportExport(); 1564 return; 1565 } 1566 if (!Style.isCpp()) 1567 break; 1568 // Handle C++ "(inline|export) namespace". 1569 LLVM_FALLTHROUGH; 1570 case tok::kw_inline: 1571 nextToken(); 1572 if (FormatTok->is(tok::kw_namespace)) { 1573 parseNamespace(); 1574 return; 1575 } 1576 break; 1577 case tok::identifier: 1578 if (FormatTok->is(TT_ForEachMacro)) { 1579 parseForOrWhileLoop(); 1580 return; 1581 } 1582 if (FormatTok->is(TT_MacroBlockBegin)) { 1583 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 1584 /*MunchSemi=*/false); 1585 return; 1586 } 1587 if (FormatTok->is(Keywords.kw_import)) { 1588 if (Style.isJavaScript()) { 1589 parseJavaScriptEs6ImportExport(); 1590 return; 1591 } 1592 if (Style.Language == FormatStyle::LK_Proto) { 1593 nextToken(); 1594 if (FormatTok->is(tok::kw_public)) 1595 nextToken(); 1596 if (!FormatTok->is(tok::string_literal)) 1597 return; 1598 nextToken(); 1599 if (FormatTok->is(tok::semi)) 1600 nextToken(); 1601 addUnwrappedLine(); 1602 return; 1603 } 1604 if (Style.isCpp()) { 1605 parseModuleImport(); 1606 return; 1607 } 1608 } 1609 if (Style.isCpp() && 1610 FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 1611 Keywords.kw_slots, Keywords.kw_qslots)) { 1612 nextToken(); 1613 if (FormatTok->is(tok::colon)) { 1614 nextToken(); 1615 addUnwrappedLine(); 1616 return; 1617 } 1618 } 1619 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1620 parseStatementMacro(); 1621 return; 1622 } 1623 if (Style.isCpp() && FormatTok->is(TT_NamespaceMacro)) { 1624 parseNamespace(); 1625 return; 1626 } 1627 // In all other cases, parse the declaration. 1628 break; 1629 default: 1630 break; 1631 } 1632 do { 1633 const FormatToken *Previous = FormatTok->Previous; 1634 switch (FormatTok->Tok.getKind()) { 1635 case tok::at: 1636 nextToken(); 1637 if (FormatTok->is(tok::l_brace)) { 1638 nextToken(); 1639 parseBracedList(); 1640 break; 1641 } else if (Style.Language == FormatStyle::LK_Java && 1642 FormatTok->is(Keywords.kw_interface)) { 1643 nextToken(); 1644 break; 1645 } 1646 switch (FormatTok->Tok.getObjCKeywordID()) { 1647 case tok::objc_public: 1648 case tok::objc_protected: 1649 case tok::objc_package: 1650 case tok::objc_private: 1651 return parseAccessSpecifier(); 1652 case tok::objc_interface: 1653 case tok::objc_implementation: 1654 return parseObjCInterfaceOrImplementation(); 1655 case tok::objc_protocol: 1656 if (parseObjCProtocol()) 1657 return; 1658 break; 1659 case tok::objc_end: 1660 return; // Handled by the caller. 1661 case tok::objc_optional: 1662 case tok::objc_required: 1663 nextToken(); 1664 addUnwrappedLine(); 1665 return; 1666 case tok::objc_autoreleasepool: 1667 nextToken(); 1668 if (FormatTok->is(tok::l_brace)) { 1669 if (Style.BraceWrapping.AfterControlStatement == 1670 FormatStyle::BWACS_Always) { 1671 addUnwrappedLine(); 1672 } 1673 parseBlock(); 1674 } 1675 addUnwrappedLine(); 1676 return; 1677 case tok::objc_synchronized: 1678 nextToken(); 1679 if (FormatTok->is(tok::l_paren)) { 1680 // Skip synchronization object 1681 parseParens(); 1682 } 1683 if (FormatTok->is(tok::l_brace)) { 1684 if (Style.BraceWrapping.AfterControlStatement == 1685 FormatStyle::BWACS_Always) { 1686 addUnwrappedLine(); 1687 } 1688 parseBlock(); 1689 } 1690 addUnwrappedLine(); 1691 return; 1692 case tok::objc_try: 1693 // This branch isn't strictly necessary (the kw_try case below would 1694 // do this too after the tok::at is parsed above). But be explicit. 1695 parseTryCatch(); 1696 return; 1697 default: 1698 break; 1699 } 1700 break; 1701 case tok::kw_concept: 1702 parseConcept(); 1703 return; 1704 case tok::kw_requires: { 1705 if (Style.isCpp()) { 1706 bool ParsedClause = parseRequires(); 1707 if (ParsedClause) 1708 return; 1709 } else { 1710 nextToken(); 1711 } 1712 break; 1713 } 1714 case tok::kw_enum: 1715 // Ignore if this is part of "template <enum ...". 1716 if (Previous && Previous->is(tok::less)) { 1717 nextToken(); 1718 break; 1719 } 1720 1721 // parseEnum falls through and does not yet add an unwrapped line as an 1722 // enum definition can start a structural element. 1723 if (!parseEnum()) 1724 break; 1725 // This only applies for C++. 1726 if (!Style.isCpp()) { 1727 addUnwrappedLine(); 1728 return; 1729 } 1730 break; 1731 case tok::kw_typedef: 1732 nextToken(); 1733 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 1734 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS, 1735 Keywords.kw_CF_CLOSED_ENUM, 1736 Keywords.kw_NS_CLOSED_ENUM)) { 1737 parseEnum(); 1738 } 1739 break; 1740 case tok::kw_struct: 1741 case tok::kw_union: 1742 case tok::kw_class: 1743 if (parseStructLike()) 1744 return; 1745 break; 1746 case tok::period: 1747 nextToken(); 1748 // In Java, classes have an implicit static member "class". 1749 if (Style.Language == FormatStyle::LK_Java && FormatTok && 1750 FormatTok->is(tok::kw_class)) { 1751 nextToken(); 1752 } 1753 if (Style.isJavaScript() && FormatTok && 1754 FormatTok->Tok.getIdentifierInfo()) { 1755 // JavaScript only has pseudo keywords, all keywords are allowed to 1756 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 1757 nextToken(); 1758 } 1759 break; 1760 case tok::semi: 1761 nextToken(); 1762 addUnwrappedLine(); 1763 return; 1764 case tok::r_brace: 1765 addUnwrappedLine(); 1766 return; 1767 case tok::l_paren: { 1768 parseParens(); 1769 // Break the unwrapped line if a K&R C function definition has a parameter 1770 // declaration. 1771 if (!IsTopLevel || !Style.isCpp() || !Previous || FormatTok->is(tok::eof)) 1772 break; 1773 if (isC78ParameterDecl(FormatTok, Tokens->peekNextToken(), Previous)) { 1774 addUnwrappedLine(); 1775 return; 1776 } 1777 break; 1778 } 1779 case tok::kw_operator: 1780 nextToken(); 1781 if (FormatTok->isBinaryOperator()) 1782 nextToken(); 1783 break; 1784 case tok::caret: 1785 nextToken(); 1786 if (FormatTok->Tok.isAnyIdentifier() || 1787 FormatTok->isSimpleTypeSpecifier()) { 1788 nextToken(); 1789 } 1790 if (FormatTok->is(tok::l_paren)) 1791 parseParens(); 1792 if (FormatTok->is(tok::l_brace)) 1793 parseChildBlock(); 1794 break; 1795 case tok::l_brace: 1796 if (NextLBracesType != TT_Unknown) 1797 FormatTok->setFinalizedType(NextLBracesType); 1798 if (!tryToParsePropertyAccessor() && !tryToParseBracedList()) { 1799 // A block outside of parentheses must be the last part of a 1800 // structural element. 1801 // FIXME: Figure out cases where this is not true, and add projections 1802 // for them (the one we know is missing are lambdas). 1803 if (Style.Language == FormatStyle::LK_Java && 1804 Line->Tokens.front().Tok->is(Keywords.kw_synchronized)) { 1805 // If necessary, we could set the type to something different than 1806 // TT_FunctionLBrace. 1807 if (Style.BraceWrapping.AfterControlStatement == 1808 FormatStyle::BWACS_Always) { 1809 addUnwrappedLine(); 1810 } 1811 } else if (Style.BraceWrapping.AfterFunction) { 1812 addUnwrappedLine(); 1813 } 1814 if (!Line->InPPDirective) 1815 FormatTok->setFinalizedType(TT_FunctionLBrace); 1816 parseBlock(); 1817 addUnwrappedLine(); 1818 return; 1819 } 1820 // Otherwise this was a braced init list, and the structural 1821 // element continues. 1822 break; 1823 case tok::kw_try: 1824 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1825 // field/method declaration. 1826 nextToken(); 1827 break; 1828 } 1829 // We arrive here when parsing function-try blocks. 1830 if (Style.BraceWrapping.AfterFunction) 1831 addUnwrappedLine(); 1832 parseTryCatch(); 1833 return; 1834 case tok::identifier: { 1835 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where) && 1836 Line->MustBeDeclaration) { 1837 addUnwrappedLine(); 1838 parseCSharpGenericTypeConstraint(); 1839 break; 1840 } 1841 if (FormatTok->is(TT_MacroBlockEnd)) { 1842 addUnwrappedLine(); 1843 return; 1844 } 1845 1846 // Function declarations (as opposed to function expressions) are parsed 1847 // on their own unwrapped line by continuing this loop. Function 1848 // expressions (functions that are not on their own line) must not create 1849 // a new unwrapped line, so they are special cased below. 1850 size_t TokenCount = Line->Tokens.size(); 1851 if (Style.isJavaScript() && FormatTok->is(Keywords.kw_function) && 1852 (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is( 1853 Keywords.kw_async)))) { 1854 tryToParseJSFunction(); 1855 break; 1856 } 1857 if ((Style.isJavaScript() || Style.Language == FormatStyle::LK_Java) && 1858 FormatTok->is(Keywords.kw_interface)) { 1859 if (Style.isJavaScript()) { 1860 // In JavaScript/TypeScript, "interface" can be used as a standalone 1861 // identifier, e.g. in `var interface = 1;`. If "interface" is 1862 // followed by another identifier, it is very like to be an actual 1863 // interface declaration. 1864 unsigned StoredPosition = Tokens->getPosition(); 1865 FormatToken *Next = Tokens->getNextToken(); 1866 FormatTok = Tokens->setPosition(StoredPosition); 1867 if (!mustBeJSIdent(Keywords, Next)) { 1868 nextToken(); 1869 break; 1870 } 1871 } 1872 parseRecord(); 1873 addUnwrappedLine(); 1874 return; 1875 } 1876 1877 if (FormatTok->is(Keywords.kw_interface)) { 1878 if (parseStructLike()) 1879 return; 1880 break; 1881 } 1882 1883 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1884 parseStatementMacro(); 1885 return; 1886 } 1887 1888 // See if the following token should start a new unwrapped line. 1889 StringRef Text = FormatTok->TokenText; 1890 1891 FormatToken *PreviousToken = FormatTok; 1892 nextToken(); 1893 1894 // JS doesn't have macros, and within classes colons indicate fields, not 1895 // labels. 1896 if (Style.isJavaScript()) 1897 break; 1898 1899 TokenCount = Line->Tokens.size(); 1900 if (TokenCount == 1 || 1901 (TokenCount == 2 && Line->Tokens.front().Tok->is(tok::comment))) { 1902 if (FormatTok->is(tok::colon) && !Line->MustBeDeclaration) { 1903 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1904 parseLabel(!Style.IndentGotoLabels); 1905 if (HasLabel) 1906 *HasLabel = true; 1907 return; 1908 } 1909 // Recognize function-like macro usages without trailing semicolon as 1910 // well as free-standing macros like Q_OBJECT. 1911 bool FunctionLike = FormatTok->is(tok::l_paren); 1912 if (FunctionLike) 1913 parseParens(); 1914 1915 bool FollowedByNewline = 1916 CommentsBeforeNextToken.empty() 1917 ? FormatTok->NewlinesBefore > 0 1918 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1919 1920 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1921 tokenCanStartNewLine(*FormatTok) && Text == Text.upper()) { 1922 PreviousToken->setFinalizedType(TT_FunctionLikeOrFreestandingMacro); 1923 addUnwrappedLine(); 1924 return; 1925 } 1926 } 1927 break; 1928 } 1929 case tok::equal: 1930 if ((Style.isJavaScript() || Style.isCSharp()) && 1931 FormatTok->is(TT_FatArrow)) { 1932 tryToParseChildBlock(); 1933 break; 1934 } 1935 1936 nextToken(); 1937 if (FormatTok->is(tok::l_brace)) { 1938 // Block kind should probably be set to BK_BracedInit for any language. 1939 // C# needs this change to ensure that array initialisers and object 1940 // initialisers are indented the same way. 1941 if (Style.isCSharp()) 1942 FormatTok->setBlockKind(BK_BracedInit); 1943 nextToken(); 1944 parseBracedList(); 1945 } else if (Style.Language == FormatStyle::LK_Proto && 1946 FormatTok->is(tok::less)) { 1947 nextToken(); 1948 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 1949 /*ClosingBraceKind=*/tok::greater); 1950 } 1951 break; 1952 case tok::l_square: 1953 parseSquare(); 1954 break; 1955 case tok::kw_new: 1956 parseNew(); 1957 break; 1958 case tok::kw_case: 1959 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1960 // 'case: string' field declaration. 1961 nextToken(); 1962 break; 1963 } 1964 parseCaseLabel(); 1965 break; 1966 default: 1967 nextToken(); 1968 break; 1969 } 1970 } while (!eof()); 1971 } 1972 1973 bool UnwrappedLineParser::tryToParsePropertyAccessor() { 1974 assert(FormatTok->is(tok::l_brace)); 1975 if (!Style.isCSharp()) 1976 return false; 1977 // See if it's a property accessor. 1978 if (FormatTok->Previous->isNot(tok::identifier)) 1979 return false; 1980 1981 // See if we are inside a property accessor. 1982 // 1983 // Record the current tokenPosition so that we can advance and 1984 // reset the current token. `Next` is not set yet so we need 1985 // another way to advance along the token stream. 1986 unsigned int StoredPosition = Tokens->getPosition(); 1987 FormatToken *Tok = Tokens->getNextToken(); 1988 1989 // A trivial property accessor is of the form: 1990 // { [ACCESS_SPECIFIER] [get]; [ACCESS_SPECIFIER] [set|init] } 1991 // Track these as they do not require line breaks to be introduced. 1992 bool HasSpecialAccessor = false; 1993 bool IsTrivialPropertyAccessor = true; 1994 while (!eof()) { 1995 if (Tok->isOneOf(tok::semi, tok::kw_public, tok::kw_private, 1996 tok::kw_protected, Keywords.kw_internal, Keywords.kw_get, 1997 Keywords.kw_init, Keywords.kw_set)) { 1998 if (Tok->isOneOf(Keywords.kw_get, Keywords.kw_init, Keywords.kw_set)) 1999 HasSpecialAccessor = true; 2000 Tok = Tokens->getNextToken(); 2001 continue; 2002 } 2003 if (Tok->isNot(tok::r_brace)) 2004 IsTrivialPropertyAccessor = false; 2005 break; 2006 } 2007 2008 if (!HasSpecialAccessor) { 2009 Tokens->setPosition(StoredPosition); 2010 return false; 2011 } 2012 2013 // Try to parse the property accessor: 2014 // https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties 2015 Tokens->setPosition(StoredPosition); 2016 if (!IsTrivialPropertyAccessor && Style.BraceWrapping.AfterFunction) 2017 addUnwrappedLine(); 2018 nextToken(); 2019 do { 2020 switch (FormatTok->Tok.getKind()) { 2021 case tok::r_brace: 2022 nextToken(); 2023 if (FormatTok->is(tok::equal)) { 2024 while (!eof() && FormatTok->isNot(tok::semi)) 2025 nextToken(); 2026 nextToken(); 2027 } 2028 addUnwrappedLine(); 2029 return true; 2030 case tok::l_brace: 2031 ++Line->Level; 2032 parseBlock(/*MustBeDeclaration=*/true); 2033 addUnwrappedLine(); 2034 --Line->Level; 2035 break; 2036 case tok::equal: 2037 if (FormatTok->is(TT_FatArrow)) { 2038 ++Line->Level; 2039 do { 2040 nextToken(); 2041 } while (!eof() && FormatTok->isNot(tok::semi)); 2042 nextToken(); 2043 addUnwrappedLine(); 2044 --Line->Level; 2045 break; 2046 } 2047 nextToken(); 2048 break; 2049 default: 2050 if (FormatTok->isOneOf(Keywords.kw_get, Keywords.kw_init, 2051 Keywords.kw_set) && 2052 !IsTrivialPropertyAccessor) { 2053 // Non-trivial get/set needs to be on its own line. 2054 addUnwrappedLine(); 2055 } 2056 nextToken(); 2057 } 2058 } while (!eof()); 2059 2060 // Unreachable for well-formed code (paired '{' and '}'). 2061 return true; 2062 } 2063 2064 bool UnwrappedLineParser::tryToParseLambda() { 2065 assert(FormatTok->is(tok::l_square)); 2066 if (!Style.isCpp()) { 2067 nextToken(); 2068 return false; 2069 } 2070 FormatToken &LSquare = *FormatTok; 2071 if (!tryToParseLambdaIntroducer()) 2072 return false; 2073 2074 bool SeenArrow = false; 2075 bool InTemplateParameterList = false; 2076 2077 while (FormatTok->isNot(tok::l_brace)) { 2078 if (FormatTok->isSimpleTypeSpecifier()) { 2079 nextToken(); 2080 continue; 2081 } 2082 switch (FormatTok->Tok.getKind()) { 2083 case tok::l_brace: 2084 break; 2085 case tok::l_paren: 2086 parseParens(); 2087 break; 2088 case tok::l_square: 2089 parseSquare(); 2090 break; 2091 case tok::kw_class: 2092 case tok::kw_template: 2093 case tok::kw_typename: 2094 assert(FormatTok->Previous); 2095 if (FormatTok->Previous->is(tok::less)) 2096 InTemplateParameterList = true; 2097 nextToken(); 2098 break; 2099 case tok::amp: 2100 case tok::star: 2101 case tok::kw_const: 2102 case tok::comma: 2103 case tok::less: 2104 case tok::greater: 2105 case tok::identifier: 2106 case tok::numeric_constant: 2107 case tok::coloncolon: 2108 case tok::kw_mutable: 2109 case tok::kw_noexcept: 2110 nextToken(); 2111 break; 2112 // Specialization of a template with an integer parameter can contain 2113 // arithmetic, logical, comparison and ternary operators. 2114 // 2115 // FIXME: This also accepts sequences of operators that are not in the scope 2116 // of a template argument list. 2117 // 2118 // In a C++ lambda a template type can only occur after an arrow. We use 2119 // this as an heuristic to distinguish between Objective-C expressions 2120 // followed by an `a->b` expression, such as: 2121 // ([obj func:arg] + a->b) 2122 // Otherwise the code below would parse as a lambda. 2123 // 2124 // FIXME: This heuristic is incorrect for C++20 generic lambdas with 2125 // explicit template lists: []<bool b = true && false>(U &&u){} 2126 case tok::plus: 2127 case tok::minus: 2128 case tok::exclaim: 2129 case tok::tilde: 2130 case tok::slash: 2131 case tok::percent: 2132 case tok::lessless: 2133 case tok::pipe: 2134 case tok::pipepipe: 2135 case tok::ampamp: 2136 case tok::caret: 2137 case tok::equalequal: 2138 case tok::exclaimequal: 2139 case tok::greaterequal: 2140 case tok::lessequal: 2141 case tok::question: 2142 case tok::colon: 2143 case tok::ellipsis: 2144 case tok::kw_true: 2145 case tok::kw_false: 2146 if (SeenArrow || InTemplateParameterList) { 2147 nextToken(); 2148 break; 2149 } 2150 return true; 2151 case tok::arrow: 2152 // This might or might not actually be a lambda arrow (this could be an 2153 // ObjC method invocation followed by a dereferencing arrow). We might 2154 // reset this back to TT_Unknown in TokenAnnotator. 2155 FormatTok->setFinalizedType(TT_LambdaArrow); 2156 SeenArrow = true; 2157 nextToken(); 2158 break; 2159 default: 2160 return true; 2161 } 2162 } 2163 FormatTok->setFinalizedType(TT_LambdaLBrace); 2164 LSquare.setFinalizedType(TT_LambdaLSquare); 2165 parseChildBlock(); 2166 return true; 2167 } 2168 2169 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 2170 const FormatToken *Previous = FormatTok->Previous; 2171 const FormatToken *LeftSquare = FormatTok; 2172 nextToken(); 2173 if (Previous && 2174 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 2175 tok::kw_delete, tok::l_square) || 2176 LeftSquare->isCppStructuredBinding(Style) || Previous->closesScope() || 2177 Previous->isSimpleTypeSpecifier())) { 2178 return false; 2179 } 2180 if (FormatTok->is(tok::l_square)) 2181 return false; 2182 if (FormatTok->is(tok::r_square)) { 2183 const FormatToken *Next = Tokens->peekNextToken(); 2184 if (Next->is(tok::greater)) 2185 return false; 2186 } 2187 parseSquare(/*LambdaIntroducer=*/true); 2188 return true; 2189 } 2190 2191 void UnwrappedLineParser::tryToParseJSFunction() { 2192 assert(FormatTok->is(Keywords.kw_function) || 2193 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)); 2194 if (FormatTok->is(Keywords.kw_async)) 2195 nextToken(); 2196 // Consume "function". 2197 nextToken(); 2198 2199 // Consume * (generator function). Treat it like C++'s overloaded operators. 2200 if (FormatTok->is(tok::star)) { 2201 FormatTok->setFinalizedType(TT_OverloadedOperator); 2202 nextToken(); 2203 } 2204 2205 // Consume function name. 2206 if (FormatTok->is(tok::identifier)) 2207 nextToken(); 2208 2209 if (FormatTok->isNot(tok::l_paren)) 2210 return; 2211 2212 // Parse formal parameter list. 2213 parseParens(); 2214 2215 if (FormatTok->is(tok::colon)) { 2216 // Parse a type definition. 2217 nextToken(); 2218 2219 // Eat the type declaration. For braced inline object types, balance braces, 2220 // otherwise just parse until finding an l_brace for the function body. 2221 if (FormatTok->is(tok::l_brace)) 2222 tryToParseBracedList(); 2223 else 2224 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 2225 nextToken(); 2226 } 2227 2228 if (FormatTok->is(tok::semi)) 2229 return; 2230 2231 parseChildBlock(); 2232 } 2233 2234 bool UnwrappedLineParser::tryToParseBracedList() { 2235 if (FormatTok->is(BK_Unknown)) 2236 calculateBraceTypes(); 2237 assert(FormatTok->isNot(BK_Unknown)); 2238 if (FormatTok->is(BK_Block)) 2239 return false; 2240 nextToken(); 2241 parseBracedList(); 2242 return true; 2243 } 2244 2245 bool UnwrappedLineParser::tryToParseChildBlock() { 2246 assert(Style.isJavaScript() || Style.isCSharp()); 2247 assert(FormatTok->is(TT_FatArrow)); 2248 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType TT_FatArrow. 2249 // They always start an expression or a child block if followed by a curly 2250 // brace. 2251 nextToken(); 2252 if (FormatTok->isNot(tok::l_brace)) 2253 return false; 2254 parseChildBlock(); 2255 return true; 2256 } 2257 2258 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons, 2259 bool IsEnum, 2260 tok::TokenKind ClosingBraceKind) { 2261 bool HasError = false; 2262 2263 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 2264 // replace this by using parseAssignmentExpression() inside. 2265 do { 2266 if (Style.isCSharp() && FormatTok->is(TT_FatArrow) && 2267 tryToParseChildBlock()) { 2268 continue; 2269 } 2270 if (Style.isJavaScript()) { 2271 if (FormatTok->is(Keywords.kw_function) || 2272 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) { 2273 tryToParseJSFunction(); 2274 continue; 2275 } 2276 if (FormatTok->is(tok::l_brace)) { 2277 // Could be a method inside of a braced list `{a() { return 1; }}`. 2278 if (tryToParseBracedList()) 2279 continue; 2280 parseChildBlock(); 2281 } 2282 } 2283 if (FormatTok->Tok.getKind() == ClosingBraceKind) { 2284 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 2285 addUnwrappedLine(); 2286 nextToken(); 2287 return !HasError; 2288 } 2289 switch (FormatTok->Tok.getKind()) { 2290 case tok::l_square: 2291 if (Style.isCSharp()) 2292 parseSquare(); 2293 else 2294 tryToParseLambda(); 2295 break; 2296 case tok::l_paren: 2297 parseParens(); 2298 // JavaScript can just have free standing methods and getters/setters in 2299 // object literals. Detect them by a "{" following ")". 2300 if (Style.isJavaScript()) { 2301 if (FormatTok->is(tok::l_brace)) 2302 parseChildBlock(); 2303 break; 2304 } 2305 break; 2306 case tok::l_brace: 2307 // Assume there are no blocks inside a braced init list apart 2308 // from the ones we explicitly parse out (like lambdas). 2309 FormatTok->setBlockKind(BK_BracedInit); 2310 nextToken(); 2311 parseBracedList(); 2312 break; 2313 case tok::less: 2314 if (Style.Language == FormatStyle::LK_Proto || 2315 ClosingBraceKind == tok::greater) { 2316 nextToken(); 2317 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 2318 /*ClosingBraceKind=*/tok::greater); 2319 } else { 2320 nextToken(); 2321 } 2322 break; 2323 case tok::semi: 2324 // JavaScript (or more precisely TypeScript) can have semicolons in braced 2325 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 2326 // used for error recovery if we have otherwise determined that this is 2327 // a braced list. 2328 if (Style.isJavaScript()) { 2329 nextToken(); 2330 break; 2331 } 2332 HasError = true; 2333 if (!ContinueOnSemicolons) 2334 return !HasError; 2335 nextToken(); 2336 break; 2337 case tok::comma: 2338 nextToken(); 2339 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 2340 addUnwrappedLine(); 2341 break; 2342 default: 2343 nextToken(); 2344 break; 2345 } 2346 } while (!eof()); 2347 return false; 2348 } 2349 2350 /// \brief Parses a pair of parentheses (and everything between them). 2351 /// \param AmpAmpTokenType If different than TT_Unknown sets this type for all 2352 /// double ampersands. This only counts for the current parens scope. 2353 void UnwrappedLineParser::parseParens(TokenType AmpAmpTokenType) { 2354 assert(FormatTok->is(tok::l_paren) && "'(' expected."); 2355 nextToken(); 2356 do { 2357 switch (FormatTok->Tok.getKind()) { 2358 case tok::l_paren: 2359 parseParens(); 2360 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 2361 parseChildBlock(); 2362 break; 2363 case tok::r_paren: 2364 nextToken(); 2365 return; 2366 case tok::r_brace: 2367 // A "}" inside parenthesis is an error if there wasn't a matching "{". 2368 return; 2369 case tok::l_square: 2370 tryToParseLambda(); 2371 break; 2372 case tok::l_brace: 2373 if (!tryToParseBracedList()) 2374 parseChildBlock(); 2375 break; 2376 case tok::at: 2377 nextToken(); 2378 if (FormatTok->is(tok::l_brace)) { 2379 nextToken(); 2380 parseBracedList(); 2381 } 2382 break; 2383 case tok::equal: 2384 if (Style.isCSharp() && FormatTok->is(TT_FatArrow)) 2385 tryToParseChildBlock(); 2386 else 2387 nextToken(); 2388 break; 2389 case tok::kw_class: 2390 if (Style.isJavaScript()) 2391 parseRecord(/*ParseAsExpr=*/true); 2392 else 2393 nextToken(); 2394 break; 2395 case tok::identifier: 2396 if (Style.isJavaScript() && 2397 (FormatTok->is(Keywords.kw_function) || 2398 FormatTok->startsSequence(Keywords.kw_async, 2399 Keywords.kw_function))) { 2400 tryToParseJSFunction(); 2401 } else { 2402 nextToken(); 2403 } 2404 break; 2405 case tok::kw_requires: { 2406 auto RequiresToken = FormatTok; 2407 nextToken(); 2408 parseRequiresExpression(RequiresToken); 2409 break; 2410 } 2411 case tok::ampamp: 2412 if (AmpAmpTokenType != TT_Unknown) 2413 FormatTok->setFinalizedType(AmpAmpTokenType); 2414 LLVM_FALLTHROUGH; 2415 default: 2416 nextToken(); 2417 break; 2418 } 2419 } while (!eof()); 2420 } 2421 2422 void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) { 2423 if (!LambdaIntroducer) { 2424 assert(FormatTok->is(tok::l_square) && "'[' expected."); 2425 if (tryToParseLambda()) 2426 return; 2427 } 2428 do { 2429 switch (FormatTok->Tok.getKind()) { 2430 case tok::l_paren: 2431 parseParens(); 2432 break; 2433 case tok::r_square: 2434 nextToken(); 2435 return; 2436 case tok::r_brace: 2437 // A "}" inside parenthesis is an error if there wasn't a matching "{". 2438 return; 2439 case tok::l_square: 2440 parseSquare(); 2441 break; 2442 case tok::l_brace: { 2443 if (!tryToParseBracedList()) 2444 parseChildBlock(); 2445 break; 2446 } 2447 case tok::at: 2448 nextToken(); 2449 if (FormatTok->is(tok::l_brace)) { 2450 nextToken(); 2451 parseBracedList(); 2452 } 2453 break; 2454 default: 2455 nextToken(); 2456 break; 2457 } 2458 } while (!eof()); 2459 } 2460 2461 void UnwrappedLineParser::keepAncestorBraces() { 2462 if (!Style.RemoveBracesLLVM) 2463 return; 2464 2465 const int MaxNestingLevels = 2; 2466 const int Size = NestedTooDeep.size(); 2467 if (Size >= MaxNestingLevels) 2468 NestedTooDeep[Size - MaxNestingLevels] = true; 2469 NestedTooDeep.push_back(false); 2470 } 2471 2472 static FormatToken *getLastNonComment(const UnwrappedLine &Line) { 2473 for (const auto &Token : llvm::reverse(Line.Tokens)) 2474 if (Token.Tok->isNot(tok::comment)) 2475 return Token.Tok; 2476 2477 return nullptr; 2478 } 2479 2480 void UnwrappedLineParser::parseUnbracedBody(bool CheckEOF) { 2481 FormatToken *Tok = nullptr; 2482 2483 if (Style.InsertBraces && !Line->InPPDirective && !Line->Tokens.empty() && 2484 PreprocessorDirectives.empty()) { 2485 Tok = getLastNonComment(*Line); 2486 assert(Tok); 2487 if (Tok->BraceCount < 0) { 2488 assert(Tok->BraceCount == -1); 2489 Tok = nullptr; 2490 } else { 2491 Tok->BraceCount = -1; 2492 } 2493 } 2494 2495 addUnwrappedLine(); 2496 ++Line->Level; 2497 parseStructuralElement(); 2498 2499 if (Tok) { 2500 assert(!Line->InPPDirective); 2501 Tok = nullptr; 2502 for (const auto &L : llvm::reverse(*CurrentLines)) { 2503 if (!L.InPPDirective && getLastNonComment(L)) { 2504 Tok = L.Tokens.back().Tok; 2505 break; 2506 } 2507 } 2508 assert(Tok); 2509 ++Tok->BraceCount; 2510 } 2511 2512 if (CheckEOF && FormatTok->is(tok::eof)) 2513 addUnwrappedLine(); 2514 2515 --Line->Level; 2516 } 2517 2518 static void markOptionalBraces(FormatToken *LeftBrace) { 2519 if (!LeftBrace) 2520 return; 2521 2522 assert(LeftBrace->is(tok::l_brace)); 2523 2524 FormatToken *RightBrace = LeftBrace->MatchingParen; 2525 if (!RightBrace) { 2526 assert(!LeftBrace->Optional); 2527 return; 2528 } 2529 2530 assert(RightBrace->is(tok::r_brace)); 2531 assert(RightBrace->MatchingParen == LeftBrace); 2532 assert(LeftBrace->Optional == RightBrace->Optional); 2533 2534 LeftBrace->Optional = true; 2535 RightBrace->Optional = true; 2536 } 2537 2538 void UnwrappedLineParser::handleAttributes() { 2539 // Handle AttributeMacro, e.g. `if (x) UNLIKELY`. 2540 if (FormatTok->is(TT_AttributeMacro)) 2541 nextToken(); 2542 handleCppAttributes(); 2543 } 2544 2545 bool UnwrappedLineParser::handleCppAttributes() { 2546 // Handle [[likely]] / [[unlikely]] attributes. 2547 if (FormatTok->is(tok::l_square) && tryToParseSimpleAttribute()) { 2548 parseSquare(); 2549 return true; 2550 } 2551 return false; 2552 } 2553 2554 FormatToken *UnwrappedLineParser::parseIfThenElse(IfStmtKind *IfKind, 2555 bool KeepBraces) { 2556 assert(FormatTok->is(tok::kw_if) && "'if' expected"); 2557 nextToken(); 2558 if (FormatTok->is(tok::exclaim)) 2559 nextToken(); 2560 2561 bool KeepIfBraces = true; 2562 if (FormatTok->is(tok::kw_consteval)) { 2563 nextToken(); 2564 } else { 2565 KeepIfBraces = !Style.RemoveBracesLLVM || KeepBraces; 2566 if (FormatTok->isOneOf(tok::kw_constexpr, tok::identifier)) 2567 nextToken(); 2568 if (FormatTok->is(tok::l_paren)) 2569 parseParens(); 2570 } 2571 handleAttributes(); 2572 2573 bool NeedsUnwrappedLine = false; 2574 keepAncestorBraces(); 2575 2576 FormatToken *IfLeftBrace = nullptr; 2577 IfStmtKind IfBlockKind = IfStmtKind::NotIf; 2578 2579 if (FormatTok->is(tok::l_brace)) { 2580 FormatTok->setFinalizedType(TT_ControlStatementLBrace); 2581 IfLeftBrace = FormatTok; 2582 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2583 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 2584 /*MunchSemi=*/true, KeepIfBraces, &IfBlockKind); 2585 if (Style.BraceWrapping.BeforeElse) 2586 addUnwrappedLine(); 2587 else 2588 NeedsUnwrappedLine = true; 2589 } else { 2590 parseUnbracedBody(); 2591 } 2592 2593 if (Style.RemoveBracesLLVM) { 2594 assert(!NestedTooDeep.empty()); 2595 KeepIfBraces = KeepIfBraces || 2596 (IfLeftBrace && !IfLeftBrace->MatchingParen) || 2597 NestedTooDeep.back() || IfBlockKind == IfStmtKind::IfOnly || 2598 IfBlockKind == IfStmtKind::IfElseIf; 2599 } 2600 2601 bool KeepElseBraces = KeepIfBraces; 2602 FormatToken *ElseLeftBrace = nullptr; 2603 IfStmtKind Kind = IfStmtKind::IfOnly; 2604 2605 if (FormatTok->is(tok::kw_else)) { 2606 if (Style.RemoveBracesLLVM) { 2607 NestedTooDeep.back() = false; 2608 Kind = IfStmtKind::IfElse; 2609 } 2610 nextToken(); 2611 handleAttributes(); 2612 if (FormatTok->is(tok::l_brace)) { 2613 FormatTok->setFinalizedType(TT_ElseLBrace); 2614 ElseLeftBrace = FormatTok; 2615 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2616 IfStmtKind ElseBlockKind = IfStmtKind::NotIf; 2617 FormatToken *IfLBrace = 2618 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 2619 /*MunchSemi=*/true, KeepElseBraces, &ElseBlockKind); 2620 if (FormatTok->is(tok::kw_else)) { 2621 KeepElseBraces = KeepElseBraces || 2622 ElseBlockKind == IfStmtKind::IfOnly || 2623 ElseBlockKind == IfStmtKind::IfElseIf; 2624 } else if (IfLBrace && !IfLBrace->Optional) { 2625 KeepElseBraces = true; 2626 assert(ElseLeftBrace->MatchingParen); 2627 markOptionalBraces(ElseLeftBrace); 2628 } 2629 addUnwrappedLine(); 2630 } else if (FormatTok->is(tok::kw_if)) { 2631 const FormatToken *Previous = Tokens->getPreviousToken(); 2632 assert(Previous); 2633 const bool IsPrecededByComment = Previous->is(tok::comment); 2634 if (IsPrecededByComment) { 2635 addUnwrappedLine(); 2636 ++Line->Level; 2637 } 2638 bool TooDeep = true; 2639 if (Style.RemoveBracesLLVM) { 2640 Kind = IfStmtKind::IfElseIf; 2641 TooDeep = NestedTooDeep.pop_back_val(); 2642 } 2643 ElseLeftBrace = parseIfThenElse(/*IfKind=*/nullptr, KeepIfBraces); 2644 if (Style.RemoveBracesLLVM) 2645 NestedTooDeep.push_back(TooDeep); 2646 if (IsPrecededByComment) 2647 --Line->Level; 2648 } else { 2649 parseUnbracedBody(/*CheckEOF=*/true); 2650 } 2651 } else { 2652 KeepIfBraces = KeepIfBraces || IfBlockKind == IfStmtKind::IfElse; 2653 if (NeedsUnwrappedLine) 2654 addUnwrappedLine(); 2655 } 2656 2657 if (!Style.RemoveBracesLLVM) 2658 return nullptr; 2659 2660 assert(!NestedTooDeep.empty()); 2661 KeepElseBraces = KeepElseBraces || 2662 (ElseLeftBrace && !ElseLeftBrace->MatchingParen) || 2663 NestedTooDeep.back(); 2664 2665 NestedTooDeep.pop_back(); 2666 2667 if (!KeepIfBraces && !KeepElseBraces) { 2668 markOptionalBraces(IfLeftBrace); 2669 markOptionalBraces(ElseLeftBrace); 2670 } else if (IfLeftBrace) { 2671 FormatToken *IfRightBrace = IfLeftBrace->MatchingParen; 2672 if (IfRightBrace) { 2673 assert(IfRightBrace->MatchingParen == IfLeftBrace); 2674 assert(!IfLeftBrace->Optional); 2675 assert(!IfRightBrace->Optional); 2676 IfLeftBrace->MatchingParen = nullptr; 2677 IfRightBrace->MatchingParen = nullptr; 2678 } 2679 } 2680 2681 if (IfKind) 2682 *IfKind = Kind; 2683 2684 return IfLeftBrace; 2685 } 2686 2687 void UnwrappedLineParser::parseTryCatch() { 2688 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 2689 nextToken(); 2690 bool NeedsUnwrappedLine = false; 2691 if (FormatTok->is(tok::colon)) { 2692 // We are in a function try block, what comes is an initializer list. 2693 nextToken(); 2694 2695 // In case identifiers were removed by clang-tidy, what might follow is 2696 // multiple commas in sequence - before the first identifier. 2697 while (FormatTok->is(tok::comma)) 2698 nextToken(); 2699 2700 while (FormatTok->is(tok::identifier)) { 2701 nextToken(); 2702 if (FormatTok->is(tok::l_paren)) 2703 parseParens(); 2704 if (FormatTok->Previous && FormatTok->Previous->is(tok::identifier) && 2705 FormatTok->is(tok::l_brace)) { 2706 do { 2707 nextToken(); 2708 } while (!FormatTok->is(tok::r_brace)); 2709 nextToken(); 2710 } 2711 2712 // In case identifiers were removed by clang-tidy, what might follow is 2713 // multiple commas in sequence - after the first identifier. 2714 while (FormatTok->is(tok::comma)) 2715 nextToken(); 2716 } 2717 } 2718 // Parse try with resource. 2719 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) 2720 parseParens(); 2721 2722 keepAncestorBraces(); 2723 2724 if (FormatTok->is(tok::l_brace)) { 2725 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2726 parseBlock(); 2727 if (Style.BraceWrapping.BeforeCatch) 2728 addUnwrappedLine(); 2729 else 2730 NeedsUnwrappedLine = true; 2731 } else if (!FormatTok->is(tok::kw_catch)) { 2732 // The C++ standard requires a compound-statement after a try. 2733 // If there's none, we try to assume there's a structuralElement 2734 // and try to continue. 2735 addUnwrappedLine(); 2736 ++Line->Level; 2737 parseStructuralElement(); 2738 --Line->Level; 2739 } 2740 while (true) { 2741 if (FormatTok->is(tok::at)) 2742 nextToken(); 2743 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 2744 tok::kw___finally) || 2745 ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2746 FormatTok->is(Keywords.kw_finally)) || 2747 (FormatTok->isObjCAtKeyword(tok::objc_catch) || 2748 FormatTok->isObjCAtKeyword(tok::objc_finally)))) { 2749 break; 2750 } 2751 nextToken(); 2752 while (FormatTok->isNot(tok::l_brace)) { 2753 if (FormatTok->is(tok::l_paren)) { 2754 parseParens(); 2755 continue; 2756 } 2757 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) { 2758 if (Style.RemoveBracesLLVM) 2759 NestedTooDeep.pop_back(); 2760 return; 2761 } 2762 nextToken(); 2763 } 2764 NeedsUnwrappedLine = false; 2765 Line->MustBeDeclaration = false; 2766 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2767 parseBlock(); 2768 if (Style.BraceWrapping.BeforeCatch) 2769 addUnwrappedLine(); 2770 else 2771 NeedsUnwrappedLine = true; 2772 } 2773 2774 if (Style.RemoveBracesLLVM) 2775 NestedTooDeep.pop_back(); 2776 2777 if (NeedsUnwrappedLine) 2778 addUnwrappedLine(); 2779 } 2780 2781 void UnwrappedLineParser::parseNamespace() { 2782 assert(FormatTok->isOneOf(tok::kw_namespace, TT_NamespaceMacro) && 2783 "'namespace' expected"); 2784 2785 const FormatToken &InitialToken = *FormatTok; 2786 nextToken(); 2787 if (InitialToken.is(TT_NamespaceMacro)) { 2788 parseParens(); 2789 } else { 2790 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::kw_inline, 2791 tok::l_square, tok::period, tok::l_paren) || 2792 (Style.isCSharp() && FormatTok->is(tok::kw_union))) { 2793 if (FormatTok->is(tok::l_square)) 2794 parseSquare(); 2795 else if (FormatTok->is(tok::l_paren)) 2796 parseParens(); 2797 else 2798 nextToken(); 2799 } 2800 } 2801 if (FormatTok->is(tok::l_brace)) { 2802 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2803 addUnwrappedLine(); 2804 2805 unsigned AddLevels = 2806 Style.NamespaceIndentation == FormatStyle::NI_All || 2807 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 2808 DeclarationScopeStack.size() > 1) 2809 ? 1u 2810 : 0u; 2811 bool ManageWhitesmithsBraces = 2812 AddLevels == 0u && 2813 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths; 2814 2815 // If we're in Whitesmiths mode, indent the brace if we're not indenting 2816 // the whole block. 2817 if (ManageWhitesmithsBraces) 2818 ++Line->Level; 2819 2820 parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/true, 2821 /*KeepBraces=*/true, /*IfKind=*/nullptr, 2822 ManageWhitesmithsBraces); 2823 2824 // Munch the semicolon after a namespace. This is more common than one would 2825 // think. Putting the semicolon into its own line is very ugly. 2826 if (FormatTok->is(tok::semi)) 2827 nextToken(); 2828 2829 addUnwrappedLine(AddLevels > 0 ? LineLevel::Remove : LineLevel::Keep); 2830 2831 if (ManageWhitesmithsBraces) 2832 --Line->Level; 2833 } 2834 // FIXME: Add error handling. 2835 } 2836 2837 void UnwrappedLineParser::parseNew() { 2838 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 2839 nextToken(); 2840 2841 if (Style.isCSharp()) { 2842 do { 2843 if (FormatTok->is(tok::l_brace)) 2844 parseBracedList(); 2845 2846 if (FormatTok->isOneOf(tok::semi, tok::comma)) 2847 return; 2848 2849 nextToken(); 2850 } while (!eof()); 2851 } 2852 2853 if (Style.Language != FormatStyle::LK_Java) 2854 return; 2855 2856 // In Java, we can parse everything up to the parens, which aren't optional. 2857 do { 2858 // There should not be a ;, { or } before the new's open paren. 2859 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 2860 return; 2861 2862 // Consume the parens. 2863 if (FormatTok->is(tok::l_paren)) { 2864 parseParens(); 2865 2866 // If there is a class body of an anonymous class, consume that as child. 2867 if (FormatTok->is(tok::l_brace)) 2868 parseChildBlock(); 2869 return; 2870 } 2871 nextToken(); 2872 } while (!eof()); 2873 } 2874 2875 void UnwrappedLineParser::parseLoopBody(bool KeepBraces, bool WrapRightBrace) { 2876 keepAncestorBraces(); 2877 2878 if (FormatTok->is(tok::l_brace)) { 2879 if (!KeepBraces) 2880 FormatTok->setFinalizedType(TT_ControlStatementLBrace); 2881 FormatToken *LeftBrace = FormatTok; 2882 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2883 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 2884 /*MunchSemi=*/true, KeepBraces); 2885 if (!KeepBraces) { 2886 assert(!NestedTooDeep.empty()); 2887 if (!NestedTooDeep.back()) 2888 markOptionalBraces(LeftBrace); 2889 } 2890 if (WrapRightBrace) 2891 addUnwrappedLine(); 2892 } else { 2893 parseUnbracedBody(); 2894 } 2895 2896 if (!KeepBraces) 2897 NestedTooDeep.pop_back(); 2898 } 2899 2900 void UnwrappedLineParser::parseForOrWhileLoop() { 2901 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 2902 "'for', 'while' or foreach macro expected"); 2903 const bool KeepBraces = !Style.RemoveBracesLLVM || 2904 !FormatTok->isOneOf(tok::kw_for, tok::kw_while); 2905 2906 nextToken(); 2907 // JS' for await ( ... 2908 if (Style.isJavaScript() && FormatTok->is(Keywords.kw_await)) 2909 nextToken(); 2910 if (Style.isCpp() && FormatTok->is(tok::kw_co_await)) 2911 nextToken(); 2912 if (FormatTok->is(tok::l_paren)) 2913 parseParens(); 2914 2915 handleAttributes(); 2916 parseLoopBody(KeepBraces, /*WrapRightBrace=*/true); 2917 } 2918 2919 void UnwrappedLineParser::parseDoWhile() { 2920 assert(FormatTok->is(tok::kw_do) && "'do' expected"); 2921 nextToken(); 2922 2923 parseLoopBody(/*KeepBraces=*/true, Style.BraceWrapping.BeforeWhile); 2924 2925 // FIXME: Add error handling. 2926 if (!FormatTok->is(tok::kw_while)) { 2927 addUnwrappedLine(); 2928 return; 2929 } 2930 2931 // If in Whitesmiths mode, the line with the while() needs to be indented 2932 // to the same level as the block. 2933 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) 2934 ++Line->Level; 2935 2936 nextToken(); 2937 parseStructuralElement(); 2938 } 2939 2940 void UnwrappedLineParser::parseLabel(bool LeftAlignLabel) { 2941 nextToken(); 2942 unsigned OldLineLevel = Line->Level; 2943 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 2944 --Line->Level; 2945 if (LeftAlignLabel) 2946 Line->Level = 0; 2947 2948 if (!Style.IndentCaseBlocks && CommentsBeforeNextToken.empty() && 2949 FormatTok->is(tok::l_brace)) { 2950 2951 CompoundStatementIndenter Indenter(this, Line->Level, 2952 Style.BraceWrapping.AfterCaseLabel, 2953 Style.BraceWrapping.IndentBraces); 2954 parseBlock(); 2955 if (FormatTok->is(tok::kw_break)) { 2956 if (Style.BraceWrapping.AfterControlStatement == 2957 FormatStyle::BWACS_Always) { 2958 addUnwrappedLine(); 2959 if (!Style.IndentCaseBlocks && 2960 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) { 2961 ++Line->Level; 2962 } 2963 } 2964 parseStructuralElement(); 2965 } 2966 addUnwrappedLine(); 2967 } else { 2968 if (FormatTok->is(tok::semi)) 2969 nextToken(); 2970 addUnwrappedLine(); 2971 } 2972 Line->Level = OldLineLevel; 2973 if (FormatTok->isNot(tok::l_brace)) { 2974 parseStructuralElement(); 2975 addUnwrappedLine(); 2976 } 2977 } 2978 2979 void UnwrappedLineParser::parseCaseLabel() { 2980 assert(FormatTok->is(tok::kw_case) && "'case' expected"); 2981 2982 // FIXME: fix handling of complex expressions here. 2983 do { 2984 nextToken(); 2985 } while (!eof() && !FormatTok->is(tok::colon)); 2986 parseLabel(); 2987 } 2988 2989 void UnwrappedLineParser::parseSwitch() { 2990 assert(FormatTok->is(tok::kw_switch) && "'switch' expected"); 2991 nextToken(); 2992 if (FormatTok->is(tok::l_paren)) 2993 parseParens(); 2994 2995 keepAncestorBraces(); 2996 2997 if (FormatTok->is(tok::l_brace)) { 2998 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2999 parseBlock(); 3000 addUnwrappedLine(); 3001 } else { 3002 addUnwrappedLine(); 3003 ++Line->Level; 3004 parseStructuralElement(); 3005 --Line->Level; 3006 } 3007 3008 if (Style.RemoveBracesLLVM) 3009 NestedTooDeep.pop_back(); 3010 } 3011 3012 // Operators that can follow a C variable. 3013 static bool isCOperatorFollowingVar(tok::TokenKind kind) { 3014 switch (kind) { 3015 case tok::ampamp: 3016 case tok::ampequal: 3017 case tok::arrow: 3018 case tok::caret: 3019 case tok::caretequal: 3020 case tok::comma: 3021 case tok::ellipsis: 3022 case tok::equal: 3023 case tok::equalequal: 3024 case tok::exclaim: 3025 case tok::exclaimequal: 3026 case tok::greater: 3027 case tok::greaterequal: 3028 case tok::greatergreater: 3029 case tok::greatergreaterequal: 3030 case tok::l_paren: 3031 case tok::l_square: 3032 case tok::less: 3033 case tok::lessequal: 3034 case tok::lessless: 3035 case tok::lesslessequal: 3036 case tok::minus: 3037 case tok::minusequal: 3038 case tok::minusminus: 3039 case tok::percent: 3040 case tok::percentequal: 3041 case tok::period: 3042 case tok::pipe: 3043 case tok::pipeequal: 3044 case tok::pipepipe: 3045 case tok::plus: 3046 case tok::plusequal: 3047 case tok::plusplus: 3048 case tok::question: 3049 case tok::r_brace: 3050 case tok::r_paren: 3051 case tok::r_square: 3052 case tok::semi: 3053 case tok::slash: 3054 case tok::slashequal: 3055 case tok::star: 3056 case tok::starequal: 3057 return true; 3058 default: 3059 return false; 3060 } 3061 } 3062 3063 void UnwrappedLineParser::parseAccessSpecifier() { 3064 FormatToken *AccessSpecifierCandidate = FormatTok; 3065 nextToken(); 3066 // Understand Qt's slots. 3067 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 3068 nextToken(); 3069 // Otherwise, we don't know what it is, and we'd better keep the next token. 3070 if (FormatTok->is(tok::colon)) { 3071 nextToken(); 3072 addUnwrappedLine(); 3073 } else if (!FormatTok->is(tok::coloncolon) && 3074 !isCOperatorFollowingVar(FormatTok->Tok.getKind())) { 3075 // Not a variable name nor namespace name. 3076 addUnwrappedLine(); 3077 } else if (AccessSpecifierCandidate) { 3078 // Consider the access specifier to be a C identifier. 3079 AccessSpecifierCandidate->Tok.setKind(tok::identifier); 3080 } 3081 } 3082 3083 /// \brief Parses a concept definition. 3084 /// \pre The current token has to be the concept keyword. 3085 /// 3086 /// Returns if either the concept has been completely parsed, or if it detects 3087 /// that the concept definition is incorrect. 3088 void UnwrappedLineParser::parseConcept() { 3089 assert(FormatTok->is(tok::kw_concept) && "'concept' expected"); 3090 nextToken(); 3091 if (!FormatTok->is(tok::identifier)) 3092 return; 3093 nextToken(); 3094 if (!FormatTok->is(tok::equal)) 3095 return; 3096 nextToken(); 3097 parseConstraintExpression(); 3098 if (FormatTok->is(tok::semi)) 3099 nextToken(); 3100 addUnwrappedLine(); 3101 } 3102 3103 /// \brief Parses a requires, decides if it is a clause or an expression. 3104 /// \pre The current token has to be the requires keyword. 3105 /// \returns true if it parsed a clause. 3106 bool clang::format::UnwrappedLineParser::parseRequires() { 3107 assert(FormatTok->is(tok::kw_requires) && "'requires' expected"); 3108 auto RequiresToken = FormatTok; 3109 3110 // We try to guess if it is a requires clause, or a requires expression. For 3111 // that we first consume the keyword and check the next token. 3112 nextToken(); 3113 3114 switch (FormatTok->Tok.getKind()) { 3115 case tok::l_brace: 3116 // This can only be an expression, never a clause. 3117 parseRequiresExpression(RequiresToken); 3118 return false; 3119 case tok::l_paren: 3120 // Clauses and expression can start with a paren, it's unclear what we have. 3121 break; 3122 default: 3123 // All other tokens can only be a clause. 3124 parseRequiresClause(RequiresToken); 3125 return true; 3126 } 3127 3128 // Looking forward we would have to decide if there are function declaration 3129 // like arguments to the requires expression: 3130 // requires (T t) { 3131 // Or there is a constraint expression for the requires clause: 3132 // requires (C<T> && ... 3133 3134 // But first let's look behind. 3135 auto *PreviousNonComment = RequiresToken->getPreviousNonComment(); 3136 3137 if (!PreviousNonComment || 3138 PreviousNonComment->is(TT_RequiresExpressionLBrace)) { 3139 // If there is no token, or an expression left brace, we are a requires 3140 // clause within a requires expression. 3141 parseRequiresClause(RequiresToken); 3142 return true; 3143 } 3144 3145 switch (PreviousNonComment->Tok.getKind()) { 3146 case tok::greater: 3147 case tok::r_paren: 3148 case tok::kw_noexcept: 3149 case tok::kw_const: 3150 // This is a requires clause. 3151 parseRequiresClause(RequiresToken); 3152 return true; 3153 case tok::amp: 3154 case tok::ampamp: { 3155 // This can be either: 3156 // if (... && requires (T t) ...) 3157 // Or 3158 // void member(...) && requires (C<T> ... 3159 // We check the one token before that for a const: 3160 // void member(...) const && requires (C<T> ... 3161 auto PrevPrev = PreviousNonComment->getPreviousNonComment(); 3162 if (PrevPrev && PrevPrev->is(tok::kw_const)) { 3163 parseRequiresClause(RequiresToken); 3164 return true; 3165 } 3166 break; 3167 } 3168 default: 3169 // It's an expression. 3170 parseRequiresExpression(RequiresToken); 3171 return false; 3172 } 3173 3174 // Now we look forward and try to check if the paren content is a parameter 3175 // list. The parameters can be cv-qualified and contain references or 3176 // pointers. 3177 // So we want basically to check for TYPE NAME, but TYPE can contain all kinds 3178 // of stuff: typename, const, *, &, &&, ::, identifiers. 3179 3180 int NextTokenOffset = 1; 3181 auto NextToken = Tokens->peekNextToken(NextTokenOffset); 3182 auto PeekNext = [&NextTokenOffset, &NextToken, this] { 3183 ++NextTokenOffset; 3184 NextToken = Tokens->peekNextToken(NextTokenOffset); 3185 }; 3186 3187 bool FoundType = false; 3188 bool LastWasColonColon = false; 3189 int OpenAngles = 0; 3190 3191 for (; NextTokenOffset < 50; PeekNext()) { 3192 switch (NextToken->Tok.getKind()) { 3193 case tok::kw_volatile: 3194 case tok::kw_const: 3195 case tok::comma: 3196 parseRequiresExpression(RequiresToken); 3197 return false; 3198 case tok::r_paren: 3199 case tok::pipepipe: 3200 parseRequiresClause(RequiresToken); 3201 return true; 3202 case tok::eof: 3203 // Break out of the loop. 3204 NextTokenOffset = 50; 3205 break; 3206 case tok::coloncolon: 3207 LastWasColonColon = true; 3208 break; 3209 case tok::identifier: 3210 if (FoundType && !LastWasColonColon && OpenAngles == 0) { 3211 parseRequiresExpression(RequiresToken); 3212 return false; 3213 } 3214 FoundType = true; 3215 LastWasColonColon = false; 3216 break; 3217 case tok::less: 3218 ++OpenAngles; 3219 break; 3220 case tok::greater: 3221 --OpenAngles; 3222 break; 3223 default: 3224 if (NextToken->isSimpleTypeSpecifier()) { 3225 parseRequiresExpression(RequiresToken); 3226 return false; 3227 } 3228 break; 3229 } 3230 } 3231 3232 // This seems to be a complicated expression, just assume it's a clause. 3233 parseRequiresClause(RequiresToken); 3234 return true; 3235 } 3236 3237 /// \brief Parses a requires clause. 3238 /// \param RequiresToken The requires keyword token, which starts this clause. 3239 /// \pre We need to be on the next token after the requires keyword. 3240 /// \sa parseRequiresExpression 3241 /// 3242 /// Returns if it either has finished parsing the clause, or it detects, that 3243 /// the clause is incorrect. 3244 void UnwrappedLineParser::parseRequiresClause(FormatToken *RequiresToken) { 3245 assert(FormatTok->getPreviousNonComment() == RequiresToken); 3246 assert(RequiresToken->is(tok::kw_requires) && "'requires' expected"); 3247 3248 // If there is no previous token, we are within a requires expression, 3249 // otherwise we will always have the template or function declaration in front 3250 // of it. 3251 bool InRequiresExpression = 3252 !RequiresToken->Previous || 3253 RequiresToken->Previous->is(TT_RequiresExpressionLBrace); 3254 3255 RequiresToken->setFinalizedType(InRequiresExpression 3256 ? TT_RequiresClauseInARequiresExpression 3257 : TT_RequiresClause); 3258 3259 parseConstraintExpression(); 3260 3261 if (!InRequiresExpression) 3262 FormatTok->Previous->ClosesRequiresClause = true; 3263 } 3264 3265 /// \brief Parses a requires expression. 3266 /// \param RequiresToken The requires keyword token, which starts this clause. 3267 /// \pre We need to be on the next token after the requires keyword. 3268 /// \sa parseRequiresClause 3269 /// 3270 /// Returns if it either has finished parsing the expression, or it detects, 3271 /// that the expression is incorrect. 3272 void UnwrappedLineParser::parseRequiresExpression(FormatToken *RequiresToken) { 3273 assert(FormatTok->getPreviousNonComment() == RequiresToken); 3274 assert(RequiresToken->is(tok::kw_requires) && "'requires' expected"); 3275 3276 RequiresToken->setFinalizedType(TT_RequiresExpression); 3277 3278 if (FormatTok->is(tok::l_paren)) { 3279 FormatTok->setFinalizedType(TT_RequiresExpressionLParen); 3280 parseParens(); 3281 } 3282 3283 if (FormatTok->is(tok::l_brace)) { 3284 FormatTok->setFinalizedType(TT_RequiresExpressionLBrace); 3285 parseChildBlock(/*CanContainBracedList=*/false, 3286 /*NextLBracesType=*/TT_CompoundRequirementLBrace); 3287 } 3288 } 3289 3290 /// \brief Parses a constraint expression. 3291 /// 3292 /// This is either the definition of a concept, or the body of a requires 3293 /// clause. It returns, when the parsing is complete, or the expression is 3294 /// incorrect. 3295 void UnwrappedLineParser::parseConstraintExpression() { 3296 // The special handling for lambdas is needed since tryToParseLambda() eats a 3297 // token and if a requires expression is the last part of a requires clause 3298 // and followed by an attribute like [[nodiscard]] the ClosesRequiresClause is 3299 // not set on the correct token. Thus we need to be aware if we even expect a 3300 // lambda to be possible. 3301 // template <typename T> requires requires { ... } [[nodiscard]] ...; 3302 bool LambdaNextTimeAllowed = true; 3303 do { 3304 bool LambdaThisTimeAllowed = std::exchange(LambdaNextTimeAllowed, false); 3305 3306 switch (FormatTok->Tok.getKind()) { 3307 case tok::kw_requires: { 3308 auto RequiresToken = FormatTok; 3309 nextToken(); 3310 parseRequiresExpression(RequiresToken); 3311 break; 3312 } 3313 3314 case tok::l_paren: 3315 parseParens(/*AmpAmpTokenType=*/TT_BinaryOperator); 3316 break; 3317 3318 case tok::l_square: 3319 if (!LambdaThisTimeAllowed || !tryToParseLambda()) 3320 return; 3321 break; 3322 3323 case tok::kw_const: 3324 case tok::semi: 3325 case tok::kw_class: 3326 case tok::kw_struct: 3327 case tok::kw_union: 3328 return; 3329 3330 case tok::l_brace: 3331 // Potential function body. 3332 return; 3333 3334 case tok::ampamp: 3335 case tok::pipepipe: 3336 FormatTok->setFinalizedType(TT_BinaryOperator); 3337 nextToken(); 3338 LambdaNextTimeAllowed = true; 3339 break; 3340 3341 case tok::comma: 3342 case tok::comment: 3343 LambdaNextTimeAllowed = LambdaThisTimeAllowed; 3344 nextToken(); 3345 break; 3346 3347 case tok::kw_sizeof: 3348 case tok::greater: 3349 case tok::greaterequal: 3350 case tok::greatergreater: 3351 case tok::less: 3352 case tok::lessequal: 3353 case tok::lessless: 3354 case tok::equalequal: 3355 case tok::exclaim: 3356 case tok::exclaimequal: 3357 case tok::plus: 3358 case tok::minus: 3359 case tok::star: 3360 case tok::slash: 3361 case tok::kw_decltype: 3362 LambdaNextTimeAllowed = true; 3363 // Just eat them. 3364 nextToken(); 3365 break; 3366 3367 case tok::numeric_constant: 3368 case tok::coloncolon: 3369 case tok::kw_true: 3370 case tok::kw_false: 3371 // Just eat them. 3372 nextToken(); 3373 break; 3374 3375 case tok::kw_static_cast: 3376 case tok::kw_const_cast: 3377 case tok::kw_reinterpret_cast: 3378 case tok::kw_dynamic_cast: 3379 nextToken(); 3380 if (!FormatTok->is(tok::less)) 3381 return; 3382 3383 nextToken(); 3384 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 3385 /*ClosingBraceKind=*/tok::greater); 3386 break; 3387 3388 case tok::kw_bool: 3389 // bool is only allowed if it is directly followed by a paren for a cast: 3390 // concept C = bool(...); 3391 // and bool is the only type, all other types as cast must be inside a 3392 // cast to bool an thus are handled by the other cases. 3393 nextToken(); 3394 if (FormatTok->isNot(tok::l_paren)) 3395 return; 3396 parseParens(); 3397 break; 3398 3399 default: 3400 if (!FormatTok->Tok.getIdentifierInfo()) { 3401 // Identifiers are part of the default case, we check for more then 3402 // tok::identifier to handle builtin type traits. 3403 return; 3404 } 3405 3406 // We need to differentiate identifiers for a template deduction guide, 3407 // variables, or function return types (the constraint expression has 3408 // ended before that), and basically all other cases. But it's easier to 3409 // check the other way around. 3410 assert(FormatTok->Previous); 3411 switch (FormatTok->Previous->Tok.getKind()) { 3412 case tok::coloncolon: // Nested identifier. 3413 case tok::ampamp: // Start of a function or variable for the 3414 case tok::pipepipe: // constraint expression. 3415 case tok::kw_requires: // Initial identifier of a requires clause. 3416 case tok::equal: // Initial identifier of a concept declaration. 3417 break; 3418 default: 3419 return; 3420 } 3421 3422 // Read identifier with optional template declaration. 3423 nextToken(); 3424 if (FormatTok->is(tok::less)) { 3425 nextToken(); 3426 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 3427 /*ClosingBraceKind=*/tok::greater); 3428 } 3429 break; 3430 } 3431 } while (!eof()); 3432 } 3433 3434 bool UnwrappedLineParser::parseEnum() { 3435 const FormatToken &InitialToken = *FormatTok; 3436 3437 // Won't be 'enum' for NS_ENUMs. 3438 if (FormatTok->is(tok::kw_enum)) 3439 nextToken(); 3440 3441 // In TypeScript, "enum" can also be used as property name, e.g. in interface 3442 // declarations. An "enum" keyword followed by a colon would be a syntax 3443 // error and thus assume it is just an identifier. 3444 if (Style.isJavaScript() && FormatTok->isOneOf(tok::colon, tok::question)) 3445 return false; 3446 3447 // In protobuf, "enum" can be used as a field name. 3448 if (Style.Language == FormatStyle::LK_Proto && FormatTok->is(tok::equal)) 3449 return false; 3450 3451 // Eat up enum class ... 3452 if (FormatTok->isOneOf(tok::kw_class, tok::kw_struct)) 3453 nextToken(); 3454 3455 while (FormatTok->Tok.getIdentifierInfo() || 3456 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 3457 tok::greater, tok::comma, tok::question, 3458 tok::l_square, tok::r_square)) { 3459 nextToken(); 3460 // We can have macros or attributes in between 'enum' and the enum name. 3461 if (FormatTok->is(tok::l_paren)) 3462 parseParens(); 3463 if (FormatTok->is(TT_AttributeSquare)) { 3464 parseSquare(); 3465 // Consume the closing TT_AttributeSquare. 3466 if (FormatTok->Next && FormatTok->is(TT_AttributeSquare)) 3467 nextToken(); 3468 } 3469 if (FormatTok->is(tok::identifier)) { 3470 nextToken(); 3471 // If there are two identifiers in a row, this is likely an elaborate 3472 // return type. In Java, this can be "implements", etc. 3473 if (Style.isCpp() && FormatTok->is(tok::identifier)) 3474 return false; 3475 } 3476 } 3477 3478 // Just a declaration or something is wrong. 3479 if (FormatTok->isNot(tok::l_brace)) 3480 return true; 3481 FormatTok->setFinalizedType(TT_EnumLBrace); 3482 FormatTok->setBlockKind(BK_Block); 3483 3484 if (Style.Language == FormatStyle::LK_Java) { 3485 // Java enums are different. 3486 parseJavaEnumBody(); 3487 return true; 3488 } 3489 if (Style.Language == FormatStyle::LK_Proto) { 3490 parseBlock(/*MustBeDeclaration=*/true); 3491 return true; 3492 } 3493 3494 if (!Style.AllowShortEnumsOnASingleLine && 3495 ShouldBreakBeforeBrace(Style, InitialToken)) { 3496 addUnwrappedLine(); 3497 } 3498 // Parse enum body. 3499 nextToken(); 3500 if (!Style.AllowShortEnumsOnASingleLine) { 3501 addUnwrappedLine(); 3502 Line->Level += 1; 3503 } 3504 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true, 3505 /*IsEnum=*/true); 3506 if (!Style.AllowShortEnumsOnASingleLine) 3507 Line->Level -= 1; 3508 if (HasError) { 3509 if (FormatTok->is(tok::semi)) 3510 nextToken(); 3511 addUnwrappedLine(); 3512 } 3513 return true; 3514 3515 // There is no addUnwrappedLine() here so that we fall through to parsing a 3516 // structural element afterwards. Thus, in "enum A {} n, m;", 3517 // "} n, m;" will end up in one unwrapped line. 3518 } 3519 3520 bool UnwrappedLineParser::parseStructLike() { 3521 // parseRecord falls through and does not yet add an unwrapped line as a 3522 // record declaration or definition can start a structural element. 3523 parseRecord(); 3524 // This does not apply to Java, JavaScript and C#. 3525 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 3526 Style.isCSharp()) { 3527 if (FormatTok->is(tok::semi)) 3528 nextToken(); 3529 addUnwrappedLine(); 3530 return true; 3531 } 3532 return false; 3533 } 3534 3535 namespace { 3536 // A class used to set and restore the Token position when peeking 3537 // ahead in the token source. 3538 class ScopedTokenPosition { 3539 unsigned StoredPosition; 3540 FormatTokenSource *Tokens; 3541 3542 public: 3543 ScopedTokenPosition(FormatTokenSource *Tokens) : Tokens(Tokens) { 3544 assert(Tokens && "Tokens expected to not be null"); 3545 StoredPosition = Tokens->getPosition(); 3546 } 3547 3548 ~ScopedTokenPosition() { Tokens->setPosition(StoredPosition); } 3549 }; 3550 } // namespace 3551 3552 // Look to see if we have [[ by looking ahead, if 3553 // its not then rewind to the original position. 3554 bool UnwrappedLineParser::tryToParseSimpleAttribute() { 3555 ScopedTokenPosition AutoPosition(Tokens); 3556 FormatToken *Tok = Tokens->getNextToken(); 3557 // We already read the first [ check for the second. 3558 if (!Tok->is(tok::l_square)) 3559 return false; 3560 // Double check that the attribute is just something 3561 // fairly simple. 3562 while (Tok->isNot(tok::eof)) { 3563 if (Tok->is(tok::r_square)) 3564 break; 3565 Tok = Tokens->getNextToken(); 3566 } 3567 if (Tok->is(tok::eof)) 3568 return false; 3569 Tok = Tokens->getNextToken(); 3570 if (!Tok->is(tok::r_square)) 3571 return false; 3572 Tok = Tokens->getNextToken(); 3573 if (Tok->is(tok::semi)) 3574 return false; 3575 return true; 3576 } 3577 3578 void UnwrappedLineParser::parseJavaEnumBody() { 3579 assert(FormatTok->is(tok::l_brace)); 3580 const FormatToken *OpeningBrace = FormatTok; 3581 3582 // Determine whether the enum is simple, i.e. does not have a semicolon or 3583 // constants with class bodies. Simple enums can be formatted like braced 3584 // lists, contracted to a single line, etc. 3585 unsigned StoredPosition = Tokens->getPosition(); 3586 bool IsSimple = true; 3587 FormatToken *Tok = Tokens->getNextToken(); 3588 while (!Tok->is(tok::eof)) { 3589 if (Tok->is(tok::r_brace)) 3590 break; 3591 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 3592 IsSimple = false; 3593 break; 3594 } 3595 // FIXME: This will also mark enums with braces in the arguments to enum 3596 // constants as "not simple". This is probably fine in practice, though. 3597 Tok = Tokens->getNextToken(); 3598 } 3599 FormatTok = Tokens->setPosition(StoredPosition); 3600 3601 if (IsSimple) { 3602 nextToken(); 3603 parseBracedList(); 3604 addUnwrappedLine(); 3605 return; 3606 } 3607 3608 // Parse the body of a more complex enum. 3609 // First add a line for everything up to the "{". 3610 nextToken(); 3611 addUnwrappedLine(); 3612 ++Line->Level; 3613 3614 // Parse the enum constants. 3615 while (FormatTok->isNot(tok::eof)) { 3616 if (FormatTok->is(tok::l_brace)) { 3617 // Parse the constant's class body. 3618 parseBlock(/*MustBeDeclaration=*/true, /*AddLevels=*/1u, 3619 /*MunchSemi=*/false); 3620 } else if (FormatTok->is(tok::l_paren)) { 3621 parseParens(); 3622 } else if (FormatTok->is(tok::comma)) { 3623 nextToken(); 3624 addUnwrappedLine(); 3625 } else if (FormatTok->is(tok::semi)) { 3626 nextToken(); 3627 addUnwrappedLine(); 3628 break; 3629 } else if (FormatTok->is(tok::r_brace)) { 3630 addUnwrappedLine(); 3631 break; 3632 } else { 3633 nextToken(); 3634 } 3635 } 3636 3637 // Parse the class body after the enum's ";" if any. 3638 parseLevel(OpeningBrace); 3639 nextToken(); 3640 --Line->Level; 3641 addUnwrappedLine(); 3642 } 3643 3644 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) { 3645 const FormatToken &InitialToken = *FormatTok; 3646 nextToken(); 3647 3648 // The actual identifier can be a nested name specifier, and in macros 3649 // it is often token-pasted. 3650 // An [[attribute]] can be before the identifier. 3651 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 3652 tok::kw___attribute, tok::kw___declspec, 3653 tok::kw_alignas, tok::l_square, tok::r_square) || 3654 ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 3655 FormatTok->isOneOf(tok::period, tok::comma))) { 3656 if (Style.isJavaScript() && 3657 FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) { 3658 // JavaScript/TypeScript supports inline object types in 3659 // extends/implements positions: 3660 // class Foo implements {bar: number} { } 3661 nextToken(); 3662 if (FormatTok->is(tok::l_brace)) { 3663 tryToParseBracedList(); 3664 continue; 3665 } 3666 } 3667 bool IsNonMacroIdentifier = 3668 FormatTok->is(tok::identifier) && 3669 FormatTok->TokenText != FormatTok->TokenText.upper(); 3670 nextToken(); 3671 // We can have macros or attributes in between 'class' and the class name. 3672 if (!IsNonMacroIdentifier) { 3673 if (FormatTok->is(tok::l_paren)) { 3674 parseParens(); 3675 } else if (FormatTok->is(TT_AttributeSquare)) { 3676 parseSquare(); 3677 // Consume the closing TT_AttributeSquare. 3678 if (FormatTok->Next && FormatTok->is(TT_AttributeSquare)) 3679 nextToken(); 3680 } 3681 } 3682 } 3683 3684 // Note that parsing away template declarations here leads to incorrectly 3685 // accepting function declarations as record declarations. 3686 // In general, we cannot solve this problem. Consider: 3687 // class A<int> B() {} 3688 // which can be a function definition or a class definition when B() is a 3689 // macro. If we find enough real-world cases where this is a problem, we 3690 // can parse for the 'template' keyword in the beginning of the statement, 3691 // and thus rule out the record production in case there is no template 3692 // (this would still leave us with an ambiguity between template function 3693 // and class declarations). 3694 if (FormatTok->isOneOf(tok::colon, tok::less)) { 3695 do { 3696 if (FormatTok->is(tok::l_brace)) { 3697 calculateBraceTypes(/*ExpectClassBody=*/true); 3698 if (!tryToParseBracedList()) 3699 break; 3700 } 3701 if (FormatTok->is(tok::l_square)) { 3702 FormatToken *Previous = FormatTok->Previous; 3703 if (!Previous || 3704 !(Previous->is(tok::r_paren) || Previous->isTypeOrIdentifier())) { 3705 // Don't try parsing a lambda if we had a closing parenthesis before, 3706 // it was probably a pointer to an array: int (*)[]. 3707 if (!tryToParseLambda()) 3708 break; 3709 } else { 3710 parseSquare(); 3711 continue; 3712 } 3713 } 3714 if (FormatTok->is(tok::semi)) 3715 return; 3716 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where)) { 3717 addUnwrappedLine(); 3718 nextToken(); 3719 parseCSharpGenericTypeConstraint(); 3720 break; 3721 } 3722 nextToken(); 3723 } while (!eof()); 3724 } 3725 3726 auto GetBraceType = [](const FormatToken &RecordTok) { 3727 switch (RecordTok.Tok.getKind()) { 3728 case tok::kw_class: 3729 return TT_ClassLBrace; 3730 case tok::kw_struct: 3731 return TT_StructLBrace; 3732 case tok::kw_union: 3733 return TT_UnionLBrace; 3734 default: 3735 // Useful for e.g. interface. 3736 return TT_RecordLBrace; 3737 } 3738 }; 3739 if (FormatTok->is(tok::l_brace)) { 3740 FormatTok->setFinalizedType(GetBraceType(InitialToken)); 3741 if (ParseAsExpr) { 3742 parseChildBlock(); 3743 } else { 3744 if (ShouldBreakBeforeBrace(Style, InitialToken)) 3745 addUnwrappedLine(); 3746 3747 unsigned AddLevels = Style.IndentAccessModifiers ? 2u : 1u; 3748 parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/false); 3749 } 3750 } 3751 // There is no addUnwrappedLine() here so that we fall through to parsing a 3752 // structural element afterwards. Thus, in "class A {} n, m;", 3753 // "} n, m;" will end up in one unwrapped line. 3754 } 3755 3756 void UnwrappedLineParser::parseObjCMethod() { 3757 assert(FormatTok->isOneOf(tok::l_paren, tok::identifier) && 3758 "'(' or identifier expected."); 3759 do { 3760 if (FormatTok->is(tok::semi)) { 3761 nextToken(); 3762 addUnwrappedLine(); 3763 return; 3764 } else if (FormatTok->is(tok::l_brace)) { 3765 if (Style.BraceWrapping.AfterFunction) 3766 addUnwrappedLine(); 3767 parseBlock(); 3768 addUnwrappedLine(); 3769 return; 3770 } else { 3771 nextToken(); 3772 } 3773 } while (!eof()); 3774 } 3775 3776 void UnwrappedLineParser::parseObjCProtocolList() { 3777 assert(FormatTok->is(tok::less) && "'<' expected."); 3778 do { 3779 nextToken(); 3780 // Early exit in case someone forgot a close angle. 3781 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 3782 FormatTok->isObjCAtKeyword(tok::objc_end)) { 3783 return; 3784 } 3785 } while (!eof() && FormatTok->isNot(tok::greater)); 3786 nextToken(); // Skip '>'. 3787 } 3788 3789 void UnwrappedLineParser::parseObjCUntilAtEnd() { 3790 do { 3791 if (FormatTok->isObjCAtKeyword(tok::objc_end)) { 3792 nextToken(); 3793 addUnwrappedLine(); 3794 break; 3795 } 3796 if (FormatTok->is(tok::l_brace)) { 3797 parseBlock(); 3798 // In ObjC interfaces, nothing should be following the "}". 3799 addUnwrappedLine(); 3800 } else if (FormatTok->is(tok::r_brace)) { 3801 // Ignore stray "}". parseStructuralElement doesn't consume them. 3802 nextToken(); 3803 addUnwrappedLine(); 3804 } else if (FormatTok->isOneOf(tok::minus, tok::plus)) { 3805 nextToken(); 3806 parseObjCMethod(); 3807 } else { 3808 parseStructuralElement(); 3809 } 3810 } while (!eof()); 3811 } 3812 3813 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 3814 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_interface || 3815 FormatTok->Tok.getObjCKeywordID() == tok::objc_implementation); 3816 nextToken(); 3817 nextToken(); // interface name 3818 3819 // @interface can be followed by a lightweight generic 3820 // specialization list, then either a base class or a category. 3821 if (FormatTok->is(tok::less)) 3822 parseObjCLightweightGenerics(); 3823 if (FormatTok->is(tok::colon)) { 3824 nextToken(); 3825 nextToken(); // base class name 3826 // The base class can also have lightweight generics applied to it. 3827 if (FormatTok->is(tok::less)) 3828 parseObjCLightweightGenerics(); 3829 } else if (FormatTok->is(tok::l_paren)) { 3830 // Skip category, if present. 3831 parseParens(); 3832 } 3833 3834 if (FormatTok->is(tok::less)) 3835 parseObjCProtocolList(); 3836 3837 if (FormatTok->is(tok::l_brace)) { 3838 if (Style.BraceWrapping.AfterObjCDeclaration) 3839 addUnwrappedLine(); 3840 parseBlock(/*MustBeDeclaration=*/true); 3841 } 3842 3843 // With instance variables, this puts '}' on its own line. Without instance 3844 // variables, this ends the @interface line. 3845 addUnwrappedLine(); 3846 3847 parseObjCUntilAtEnd(); 3848 } 3849 3850 void UnwrappedLineParser::parseObjCLightweightGenerics() { 3851 assert(FormatTok->is(tok::less)); 3852 // Unlike protocol lists, generic parameterizations support 3853 // nested angles: 3854 // 3855 // @interface Foo<ValueType : id <NSCopying, NSSecureCoding>> : 3856 // NSObject <NSCopying, NSSecureCoding> 3857 // 3858 // so we need to count how many open angles we have left. 3859 unsigned NumOpenAngles = 1; 3860 do { 3861 nextToken(); 3862 // Early exit in case someone forgot a close angle. 3863 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 3864 FormatTok->isObjCAtKeyword(tok::objc_end)) { 3865 break; 3866 } 3867 if (FormatTok->is(tok::less)) { 3868 ++NumOpenAngles; 3869 } else if (FormatTok->is(tok::greater)) { 3870 assert(NumOpenAngles > 0 && "'>' makes NumOpenAngles negative"); 3871 --NumOpenAngles; 3872 } 3873 } while (!eof() && NumOpenAngles != 0); 3874 nextToken(); // Skip '>'. 3875 } 3876 3877 // Returns true for the declaration/definition form of @protocol, 3878 // false for the expression form. 3879 bool UnwrappedLineParser::parseObjCProtocol() { 3880 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_protocol); 3881 nextToken(); 3882 3883 if (FormatTok->is(tok::l_paren)) { 3884 // The expression form of @protocol, e.g. "Protocol* p = @protocol(foo);". 3885 return false; 3886 } 3887 3888 // The definition/declaration form, 3889 // @protocol Foo 3890 // - (int)someMethod; 3891 // @end 3892 3893 nextToken(); // protocol name 3894 3895 if (FormatTok->is(tok::less)) 3896 parseObjCProtocolList(); 3897 3898 // Check for protocol declaration. 3899 if (FormatTok->is(tok::semi)) { 3900 nextToken(); 3901 addUnwrappedLine(); 3902 return true; 3903 } 3904 3905 addUnwrappedLine(); 3906 parseObjCUntilAtEnd(); 3907 return true; 3908 } 3909 3910 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 3911 bool IsImport = FormatTok->is(Keywords.kw_import); 3912 assert(IsImport || FormatTok->is(tok::kw_export)); 3913 nextToken(); 3914 3915 // Consume the "default" in "export default class/function". 3916 if (FormatTok->is(tok::kw_default)) 3917 nextToken(); 3918 3919 // Consume "async function", "function" and "default function", so that these 3920 // get parsed as free-standing JS functions, i.e. do not require a trailing 3921 // semicolon. 3922 if (FormatTok->is(Keywords.kw_async)) 3923 nextToken(); 3924 if (FormatTok->is(Keywords.kw_function)) { 3925 nextToken(); 3926 return; 3927 } 3928 3929 // For imports, `export *`, `export {...}`, consume the rest of the line up 3930 // to the terminating `;`. For everything else, just return and continue 3931 // parsing the structural element, i.e. the declaration or expression for 3932 // `export default`. 3933 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 3934 !FormatTok->isStringLiteral()) { 3935 return; 3936 } 3937 3938 while (!eof()) { 3939 if (FormatTok->is(tok::semi)) 3940 return; 3941 if (Line->Tokens.empty()) { 3942 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 3943 // import statement should terminate. 3944 return; 3945 } 3946 if (FormatTok->is(tok::l_brace)) { 3947 FormatTok->setBlockKind(BK_Block); 3948 nextToken(); 3949 parseBracedList(); 3950 } else { 3951 nextToken(); 3952 } 3953 } 3954 } 3955 3956 void UnwrappedLineParser::parseStatementMacro() { 3957 nextToken(); 3958 if (FormatTok->is(tok::l_paren)) 3959 parseParens(); 3960 if (FormatTok->is(tok::semi)) 3961 nextToken(); 3962 addUnwrappedLine(); 3963 } 3964 3965 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 3966 StringRef Prefix = "") { 3967 llvm::dbgs() << Prefix << "Line(" << Line.Level 3968 << ", FSC=" << Line.FirstStartColumn << ")" 3969 << (Line.InPPDirective ? " MACRO" : "") << ": "; 3970 for (const auto &Node : Line.Tokens) { 3971 llvm::dbgs() << Node.Tok->Tok.getName() << "[" 3972 << "T=" << static_cast<unsigned>(Node.Tok->getType()) 3973 << ", OC=" << Node.Tok->OriginalColumn << "] "; 3974 } 3975 for (const auto &Node : Line.Tokens) 3976 for (const auto &ChildNode : Node.Children) 3977 printDebugInfo(ChildNode, "\nChild: "); 3978 3979 llvm::dbgs() << "\n"; 3980 } 3981 3982 void UnwrappedLineParser::addUnwrappedLine(LineLevel AdjustLevel) { 3983 if (Line->Tokens.empty()) 3984 return; 3985 LLVM_DEBUG({ 3986 if (CurrentLines == &Lines) 3987 printDebugInfo(*Line); 3988 }); 3989 3990 // If this line closes a block when in Whitesmiths mode, remember that 3991 // information so that the level can be decreased after the line is added. 3992 // This has to happen after the addition of the line since the line itself 3993 // needs to be indented. 3994 bool ClosesWhitesmithsBlock = 3995 Line->MatchingOpeningBlockLineIndex != UnwrappedLine::kInvalidIndex && 3996 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths; 3997 3998 CurrentLines->push_back(std::move(*Line)); 3999 Line->Tokens.clear(); 4000 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex; 4001 Line->FirstStartColumn = 0; 4002 4003 if (ClosesWhitesmithsBlock && AdjustLevel == LineLevel::Remove) 4004 --Line->Level; 4005 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 4006 CurrentLines->append( 4007 std::make_move_iterator(PreprocessorDirectives.begin()), 4008 std::make_move_iterator(PreprocessorDirectives.end())); 4009 PreprocessorDirectives.clear(); 4010 } 4011 // Disconnect the current token from the last token on the previous line. 4012 FormatTok->Previous = nullptr; 4013 } 4014 4015 bool UnwrappedLineParser::eof() const { return FormatTok->is(tok::eof); } 4016 4017 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 4018 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 4019 FormatTok.NewlinesBefore > 0; 4020 } 4021 4022 // Checks if \p FormatTok is a line comment that continues the line comment 4023 // section on \p Line. 4024 static bool 4025 continuesLineCommentSection(const FormatToken &FormatTok, 4026 const UnwrappedLine &Line, 4027 const llvm::Regex &CommentPragmasRegex) { 4028 if (Line.Tokens.empty()) 4029 return false; 4030 4031 StringRef IndentContent = FormatTok.TokenText; 4032 if (FormatTok.TokenText.startswith("//") || 4033 FormatTok.TokenText.startswith("/*")) { 4034 IndentContent = FormatTok.TokenText.substr(2); 4035 } 4036 if (CommentPragmasRegex.match(IndentContent)) 4037 return false; 4038 4039 // If Line starts with a line comment, then FormatTok continues the comment 4040 // section if its original column is greater or equal to the original start 4041 // column of the line. 4042 // 4043 // Define the min column token of a line as follows: if a line ends in '{' or 4044 // contains a '{' followed by a line comment, then the min column token is 4045 // that '{'. Otherwise, the min column token of the line is the first token of 4046 // the line. 4047 // 4048 // If Line starts with a token other than a line comment, then FormatTok 4049 // continues the comment section if its original column is greater than the 4050 // original start column of the min column token of the line. 4051 // 4052 // For example, the second line comment continues the first in these cases: 4053 // 4054 // // first line 4055 // // second line 4056 // 4057 // and: 4058 // 4059 // // first line 4060 // // second line 4061 // 4062 // and: 4063 // 4064 // int i; // first line 4065 // // second line 4066 // 4067 // and: 4068 // 4069 // do { // first line 4070 // // second line 4071 // int i; 4072 // } while (true); 4073 // 4074 // and: 4075 // 4076 // enum { 4077 // a, // first line 4078 // // second line 4079 // b 4080 // }; 4081 // 4082 // The second line comment doesn't continue the first in these cases: 4083 // 4084 // // first line 4085 // // second line 4086 // 4087 // and: 4088 // 4089 // int i; // first line 4090 // // second line 4091 // 4092 // and: 4093 // 4094 // do { // first line 4095 // // second line 4096 // int i; 4097 // } while (true); 4098 // 4099 // and: 4100 // 4101 // enum { 4102 // a, // first line 4103 // // second line 4104 // }; 4105 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 4106 4107 // Scan for '{//'. If found, use the column of '{' as a min column for line 4108 // comment section continuation. 4109 const FormatToken *PreviousToken = nullptr; 4110 for (const UnwrappedLineNode &Node : Line.Tokens) { 4111 if (PreviousToken && PreviousToken->is(tok::l_brace) && 4112 isLineComment(*Node.Tok)) { 4113 MinColumnToken = PreviousToken; 4114 break; 4115 } 4116 PreviousToken = Node.Tok; 4117 4118 // Grab the last newline preceding a token in this unwrapped line. 4119 if (Node.Tok->NewlinesBefore > 0) 4120 MinColumnToken = Node.Tok; 4121 } 4122 if (PreviousToken && PreviousToken->is(tok::l_brace)) 4123 MinColumnToken = PreviousToken; 4124 4125 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok, 4126 MinColumnToken); 4127 } 4128 4129 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 4130 bool JustComments = Line->Tokens.empty(); 4131 for (FormatToken *Tok : CommentsBeforeNextToken) { 4132 // Line comments that belong to the same line comment section are put on the 4133 // same line since later we might want to reflow content between them. 4134 // Additional fine-grained breaking of line comment sections is controlled 4135 // by the class BreakableLineCommentSection in case it is desirable to keep 4136 // several line comment sections in the same unwrapped line. 4137 // 4138 // FIXME: Consider putting separate line comment sections as children to the 4139 // unwrapped line instead. 4140 Tok->ContinuesLineCommentSection = 4141 continuesLineCommentSection(*Tok, *Line, CommentPragmasRegex); 4142 if (isOnNewLine(*Tok) && JustComments && !Tok->ContinuesLineCommentSection) 4143 addUnwrappedLine(); 4144 pushToken(Tok); 4145 } 4146 if (NewlineBeforeNext && JustComments) 4147 addUnwrappedLine(); 4148 CommentsBeforeNextToken.clear(); 4149 } 4150 4151 void UnwrappedLineParser::nextToken(int LevelDifference) { 4152 if (eof()) 4153 return; 4154 flushComments(isOnNewLine(*FormatTok)); 4155 pushToken(FormatTok); 4156 FormatToken *Previous = FormatTok; 4157 if (!Style.isJavaScript()) 4158 readToken(LevelDifference); 4159 else 4160 readTokenWithJavaScriptASI(); 4161 FormatTok->Previous = Previous; 4162 } 4163 4164 void UnwrappedLineParser::distributeComments( 4165 const SmallVectorImpl<FormatToken *> &Comments, 4166 const FormatToken *NextTok) { 4167 // Whether or not a line comment token continues a line is controlled by 4168 // the method continuesLineCommentSection, with the following caveat: 4169 // 4170 // Define a trail of Comments to be a nonempty proper postfix of Comments such 4171 // that each comment line from the trail is aligned with the next token, if 4172 // the next token exists. If a trail exists, the beginning of the maximal 4173 // trail is marked as a start of a new comment section. 4174 // 4175 // For example in this code: 4176 // 4177 // int a; // line about a 4178 // // line 1 about b 4179 // // line 2 about b 4180 // int b; 4181 // 4182 // the two lines about b form a maximal trail, so there are two sections, the 4183 // first one consisting of the single comment "// line about a" and the 4184 // second one consisting of the next two comments. 4185 if (Comments.empty()) 4186 return; 4187 bool ShouldPushCommentsInCurrentLine = true; 4188 bool HasTrailAlignedWithNextToken = false; 4189 unsigned StartOfTrailAlignedWithNextToken = 0; 4190 if (NextTok) { 4191 // We are skipping the first element intentionally. 4192 for (unsigned i = Comments.size() - 1; i > 0; --i) { 4193 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) { 4194 HasTrailAlignedWithNextToken = true; 4195 StartOfTrailAlignedWithNextToken = i; 4196 } 4197 } 4198 } 4199 for (unsigned i = 0, e = Comments.size(); i < e; ++i) { 4200 FormatToken *FormatTok = Comments[i]; 4201 if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) { 4202 FormatTok->ContinuesLineCommentSection = false; 4203 } else { 4204 FormatTok->ContinuesLineCommentSection = 4205 continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex); 4206 } 4207 if (!FormatTok->ContinuesLineCommentSection && 4208 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) { 4209 ShouldPushCommentsInCurrentLine = false; 4210 } 4211 if (ShouldPushCommentsInCurrentLine) 4212 pushToken(FormatTok); 4213 else 4214 CommentsBeforeNextToken.push_back(FormatTok); 4215 } 4216 } 4217 4218 void UnwrappedLineParser::readToken(int LevelDifference) { 4219 SmallVector<FormatToken *, 1> Comments; 4220 bool PreviousWasComment = false; 4221 bool FirstNonCommentOnLine = false; 4222 do { 4223 FormatTok = Tokens->getNextToken(); 4224 assert(FormatTok); 4225 while (FormatTok->getType() == TT_ConflictStart || 4226 FormatTok->getType() == TT_ConflictEnd || 4227 FormatTok->getType() == TT_ConflictAlternative) { 4228 if (FormatTok->getType() == TT_ConflictStart) 4229 conditionalCompilationStart(/*Unreachable=*/false); 4230 else if (FormatTok->getType() == TT_ConflictAlternative) 4231 conditionalCompilationAlternative(); 4232 else if (FormatTok->getType() == TT_ConflictEnd) 4233 conditionalCompilationEnd(); 4234 FormatTok = Tokens->getNextToken(); 4235 FormatTok->MustBreakBefore = true; 4236 } 4237 4238 auto IsFirstNonCommentOnLine = [](bool FirstNonCommentOnLine, 4239 const FormatToken &Tok, 4240 bool PreviousWasComment) { 4241 auto IsFirstOnLine = [](const FormatToken &Tok) { 4242 return Tok.HasUnescapedNewline || Tok.IsFirst; 4243 }; 4244 4245 // Consider preprocessor directives preceded by block comments as first 4246 // on line. 4247 if (PreviousWasComment) 4248 return FirstNonCommentOnLine || IsFirstOnLine(Tok); 4249 return IsFirstOnLine(Tok); 4250 }; 4251 4252 FirstNonCommentOnLine = IsFirstNonCommentOnLine( 4253 FirstNonCommentOnLine, *FormatTok, PreviousWasComment); 4254 PreviousWasComment = FormatTok->is(tok::comment); 4255 4256 while (!Line->InPPDirective && FormatTok->is(tok::hash) && 4257 FirstNonCommentOnLine) { 4258 distributeComments(Comments, FormatTok); 4259 Comments.clear(); 4260 // If there is an unfinished unwrapped line, we flush the preprocessor 4261 // directives only after that unwrapped line was finished later. 4262 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 4263 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 4264 assert((LevelDifference >= 0 || 4265 static_cast<unsigned>(-LevelDifference) <= Line->Level) && 4266 "LevelDifference makes Line->Level negative"); 4267 Line->Level += LevelDifference; 4268 // Comments stored before the preprocessor directive need to be output 4269 // before the preprocessor directive, at the same level as the 4270 // preprocessor directive, as we consider them to apply to the directive. 4271 if (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash && 4272 PPBranchLevel > 0) { 4273 Line->Level += PPBranchLevel; 4274 } 4275 flushComments(isOnNewLine(*FormatTok)); 4276 parsePPDirective(); 4277 PreviousWasComment = FormatTok->is(tok::comment); 4278 FirstNonCommentOnLine = IsFirstNonCommentOnLine( 4279 FirstNonCommentOnLine, *FormatTok, PreviousWasComment); 4280 } 4281 4282 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) && 4283 !Line->InPPDirective) { 4284 continue; 4285 } 4286 4287 if (!FormatTok->is(tok::comment)) { 4288 distributeComments(Comments, FormatTok); 4289 Comments.clear(); 4290 return; 4291 } 4292 4293 Comments.push_back(FormatTok); 4294 } while (!eof()); 4295 4296 distributeComments(Comments, nullptr); 4297 Comments.clear(); 4298 } 4299 4300 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 4301 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 4302 if (MustBreakBeforeNextToken) { 4303 Line->Tokens.back().Tok->MustBreakBefore = true; 4304 MustBreakBeforeNextToken = false; 4305 } 4306 } 4307 4308 } // end namespace format 4309 } // end namespace clang 4310