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