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