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