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