1 //===--- UnwrappedLineParser.cpp - Format C++ code ------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// 9 /// \file 10 /// This file contains the implementation of the UnwrappedLineParser, 11 /// which turns a stream of tokens into UnwrappedLines. 12 /// 13 //===----------------------------------------------------------------------===// 14 15 #include "UnwrappedLineParser.h" 16 #include "FormatToken.h" 17 #include "TokenAnnotator.h" 18 #include "llvm/ADT/STLExtras.h" 19 #include "llvm/Support/Debug.h" 20 #include "llvm/Support/raw_ostream.h" 21 22 #include <algorithm> 23 #include <utility> 24 25 #define DEBUG_TYPE "format-parser" 26 27 namespace clang { 28 namespace format { 29 30 class FormatTokenSource { 31 public: 32 virtual ~FormatTokenSource() {} 33 34 // Returns the next token in the token stream. 35 virtual FormatToken *getNextToken() = 0; 36 37 // Returns the token preceding the token returned by the last call to 38 // getNextToken() in the token stream, or nullptr if no such token exists. 39 virtual FormatToken *getPreviousToken() = 0; 40 41 // Returns the token that would be returned by the next call to 42 // getNextToken(). 43 virtual FormatToken *peekNextToken() = 0; 44 45 // Returns the token that would be returned after the next N calls to 46 // getNextToken(). N needs to be greater than zero, and small enough that 47 // there are still tokens. Check for tok::eof with N-1 before calling it with 48 // N. 49 virtual FormatToken *peekNextToken(int N) = 0; 50 51 // Returns whether we are at the end of the file. 52 // This can be different from whether getNextToken() returned an eof token 53 // when the FormatTokenSource is a view on a part of the token stream. 54 virtual bool isEOF() = 0; 55 56 // Gets the current position in the token stream, to be used by setPosition(). 57 virtual unsigned getPosition() = 0; 58 59 // Resets the token stream to the state it was in when getPosition() returned 60 // Position, and return the token at that position in the stream. 61 virtual FormatToken *setPosition(unsigned Position) = 0; 62 }; 63 64 namespace { 65 66 class ScopedDeclarationState { 67 public: 68 ScopedDeclarationState(UnwrappedLine &Line, llvm::BitVector &Stack, 69 bool MustBeDeclaration) 70 : Line(Line), Stack(Stack) { 71 Line.MustBeDeclaration = MustBeDeclaration; 72 Stack.push_back(MustBeDeclaration); 73 } 74 ~ScopedDeclarationState() { 75 Stack.pop_back(); 76 if (!Stack.empty()) 77 Line.MustBeDeclaration = Stack.back(); 78 else 79 Line.MustBeDeclaration = true; 80 } 81 82 private: 83 UnwrappedLine &Line; 84 llvm::BitVector &Stack; 85 }; 86 87 static bool isLineComment(const FormatToken &FormatTok) { 88 return FormatTok.is(tok::comment) && !FormatTok.TokenText.startswith("/*"); 89 } 90 91 // Checks if \p FormatTok is a line comment that continues the line comment 92 // \p Previous. The original column of \p MinColumnToken is used to determine 93 // whether \p FormatTok is indented enough to the right to continue \p Previous. 94 static bool continuesLineComment(const FormatToken &FormatTok, 95 const FormatToken *Previous, 96 const FormatToken *MinColumnToken) { 97 if (!Previous || !MinColumnToken) 98 return false; 99 unsigned MinContinueColumn = 100 MinColumnToken->OriginalColumn + (isLineComment(*MinColumnToken) ? 0 : 1); 101 return isLineComment(FormatTok) && FormatTok.NewlinesBefore == 1 && 102 isLineComment(*Previous) && 103 FormatTok.OriginalColumn >= MinContinueColumn; 104 } 105 106 class ScopedMacroState : public FormatTokenSource { 107 public: 108 ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource, 109 FormatToken *&ResetToken) 110 : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken), 111 PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource), 112 Token(nullptr), PreviousToken(nullptr) { 113 FakeEOF.Tok.startToken(); 114 FakeEOF.Tok.setKind(tok::eof); 115 TokenSource = this; 116 Line.Level = 0; 117 Line.InPPDirective = true; 118 } 119 120 ~ScopedMacroState() override { 121 TokenSource = PreviousTokenSource; 122 ResetToken = Token; 123 Line.InPPDirective = false; 124 Line.Level = PreviousLineLevel; 125 } 126 127 FormatToken *getNextToken() override { 128 // The \c UnwrappedLineParser guards against this by never calling 129 // \c getNextToken() after it has encountered the first eof token. 130 assert(!eof()); 131 PreviousToken = Token; 132 Token = PreviousTokenSource->getNextToken(); 133 if (eof()) 134 return &FakeEOF; 135 return Token; 136 } 137 138 FormatToken *getPreviousToken() override { 139 return PreviousTokenSource->getPreviousToken(); 140 } 141 142 FormatToken *peekNextToken() override { 143 if (eof()) 144 return &FakeEOF; 145 return PreviousTokenSource->peekNextToken(); 146 } 147 148 FormatToken *peekNextToken(int N) override { 149 assert(N > 0); 150 if (eof()) 151 return &FakeEOF; 152 return PreviousTokenSource->peekNextToken(N); 153 } 154 155 bool isEOF() override { return PreviousTokenSource->isEOF(); } 156 157 unsigned getPosition() override { return PreviousTokenSource->getPosition(); } 158 159 FormatToken *setPosition(unsigned Position) override { 160 PreviousToken = nullptr; 161 Token = PreviousTokenSource->setPosition(Position); 162 return Token; 163 } 164 165 private: 166 bool eof() { 167 return Token && Token->HasUnescapedNewline && 168 !continuesLineComment(*Token, PreviousToken, 169 /*MinColumnToken=*/PreviousToken); 170 } 171 172 FormatToken FakeEOF; 173 UnwrappedLine &Line; 174 FormatTokenSource *&TokenSource; 175 FormatToken *&ResetToken; 176 unsigned PreviousLineLevel; 177 FormatTokenSource *PreviousTokenSource; 178 179 FormatToken *Token; 180 FormatToken *PreviousToken; 181 }; 182 183 } // end anonymous namespace 184 185 class ScopedLineState { 186 public: 187 ScopedLineState(UnwrappedLineParser &Parser, 188 bool SwitchToPreprocessorLines = false) 189 : Parser(Parser), OriginalLines(Parser.CurrentLines) { 190 if (SwitchToPreprocessorLines) 191 Parser.CurrentLines = &Parser.PreprocessorDirectives; 192 else if (!Parser.Line->Tokens.empty()) 193 Parser.CurrentLines = &Parser.Line->Tokens.back().Children; 194 PreBlockLine = std::move(Parser.Line); 195 Parser.Line = std::make_unique<UnwrappedLine>(); 196 Parser.Line->Level = PreBlockLine->Level; 197 Parser.Line->InPPDirective = PreBlockLine->InPPDirective; 198 } 199 200 ~ScopedLineState() { 201 if (!Parser.Line->Tokens.empty()) 202 Parser.addUnwrappedLine(); 203 assert(Parser.Line->Tokens.empty()); 204 Parser.Line = std::move(PreBlockLine); 205 if (Parser.CurrentLines == &Parser.PreprocessorDirectives) 206 Parser.MustBreakBeforeNextToken = true; 207 Parser.CurrentLines = OriginalLines; 208 } 209 210 private: 211 UnwrappedLineParser &Parser; 212 213 std::unique_ptr<UnwrappedLine> PreBlockLine; 214 SmallVectorImpl<UnwrappedLine> *OriginalLines; 215 }; 216 217 class CompoundStatementIndenter { 218 public: 219 CompoundStatementIndenter(UnwrappedLineParser *Parser, 220 const FormatStyle &Style, unsigned &LineLevel) 221 : CompoundStatementIndenter(Parser, LineLevel, 222 Style.BraceWrapping.AfterControlStatement, 223 Style.BraceWrapping.IndentBraces) {} 224 CompoundStatementIndenter(UnwrappedLineParser *Parser, unsigned &LineLevel, 225 bool WrapBrace, bool IndentBrace) 226 : LineLevel(LineLevel), OldLineLevel(LineLevel) { 227 if (WrapBrace) 228 Parser->addUnwrappedLine(); 229 if (IndentBrace) 230 ++LineLevel; 231 } 232 ~CompoundStatementIndenter() { LineLevel = OldLineLevel; } 233 234 private: 235 unsigned &LineLevel; 236 unsigned OldLineLevel; 237 }; 238 239 namespace { 240 241 class IndexedTokenSource : public FormatTokenSource { 242 public: 243 IndexedTokenSource(ArrayRef<FormatToken *> Tokens) 244 : Tokens(Tokens), Position(-1) {} 245 246 FormatToken *getNextToken() override { 247 if (Position >= 0 && Tokens[Position]->is(tok::eof)) { 248 LLVM_DEBUG({ 249 llvm::dbgs() << "Next "; 250 dbgToken(Position); 251 }); 252 return Tokens[Position]; 253 } 254 ++Position; 255 LLVM_DEBUG({ 256 llvm::dbgs() << "Next "; 257 dbgToken(Position); 258 }); 259 return Tokens[Position]; 260 } 261 262 FormatToken *getPreviousToken() override { 263 return Position > 0 ? Tokens[Position - 1] : nullptr; 264 } 265 266 FormatToken *peekNextToken() override { 267 int Next = Position + 1; 268 LLVM_DEBUG({ 269 llvm::dbgs() << "Peeking "; 270 dbgToken(Next); 271 }); 272 return Tokens[Next]; 273 } 274 275 FormatToken *peekNextToken(int N) override { 276 assert(N > 0); 277 int Next = Position + N; 278 LLVM_DEBUG({ 279 llvm::dbgs() << "Peeking (+" << (N - 1) << ") "; 280 dbgToken(Next); 281 }); 282 return Tokens[Next]; 283 } 284 285 bool isEOF() override { return Tokens[Position]->is(tok::eof); } 286 287 unsigned getPosition() override { 288 LLVM_DEBUG(llvm::dbgs() << "Getting Position: " << Position << "\n"); 289 assert(Position >= 0); 290 return Position; 291 } 292 293 FormatToken *setPosition(unsigned P) override { 294 LLVM_DEBUG(llvm::dbgs() << "Setting Position: " << P << "\n"); 295 Position = P; 296 return Tokens[Position]; 297 } 298 299 void reset() { Position = -1; } 300 301 private: 302 void dbgToken(int Position, llvm::StringRef Indent = "") { 303 FormatToken *Tok = Tokens[Position]; 304 llvm::dbgs() << Indent << "[" << Position 305 << "] Token: " << Tok->Tok.getName() << " / " << Tok->TokenText 306 << ", Macro: " << !!Tok->MacroCtx << "\n"; 307 } 308 309 ArrayRef<FormatToken *> Tokens; 310 int Position; 311 }; 312 313 } // end anonymous namespace 314 315 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style, 316 const AdditionalKeywords &Keywords, 317 unsigned FirstStartColumn, 318 ArrayRef<FormatToken *> Tokens, 319 UnwrappedLineConsumer &Callback) 320 : Line(new UnwrappedLine), MustBreakBeforeNextToken(false), 321 CurrentLines(&Lines), Style(Style), Keywords(Keywords), 322 CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr), 323 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1), 324 IncludeGuard(Style.IndentPPDirectives == FormatStyle::PPDIS_None 325 ? IG_Rejected 326 : IG_Inited), 327 IncludeGuardToken(nullptr), FirstStartColumn(FirstStartColumn) {} 328 329 void UnwrappedLineParser::reset() { 330 PPBranchLevel = -1; 331 IncludeGuard = Style.IndentPPDirectives == FormatStyle::PPDIS_None 332 ? IG_Rejected 333 : IG_Inited; 334 IncludeGuardToken = nullptr; 335 Line.reset(new UnwrappedLine); 336 CommentsBeforeNextToken.clear(); 337 FormatTok = nullptr; 338 MustBreakBeforeNextToken = false; 339 PreprocessorDirectives.clear(); 340 CurrentLines = &Lines; 341 DeclarationScopeStack.clear(); 342 NestedTooDeep.clear(); 343 PPStack.clear(); 344 Line->FirstStartColumn = FirstStartColumn; 345 } 346 347 void UnwrappedLineParser::parse() { 348 IndexedTokenSource TokenSource(AllTokens); 349 Line->FirstStartColumn = FirstStartColumn; 350 do { 351 LLVM_DEBUG(llvm::dbgs() << "----\n"); 352 reset(); 353 Tokens = &TokenSource; 354 TokenSource.reset(); 355 356 readToken(); 357 parseFile(); 358 359 // If we found an include guard then all preprocessor directives (other than 360 // the guard) are over-indented by one. 361 if (IncludeGuard == IG_Found) 362 for (auto &Line : Lines) 363 if (Line.InPPDirective && Line.Level > 0) 364 --Line.Level; 365 366 // Create line with eof token. 367 pushToken(FormatTok); 368 addUnwrappedLine(); 369 370 for (const UnwrappedLine &Line : Lines) 371 Callback.consumeUnwrappedLine(Line); 372 373 Callback.finishRun(); 374 Lines.clear(); 375 while (!PPLevelBranchIndex.empty() && 376 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) { 377 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1); 378 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1); 379 } 380 if (!PPLevelBranchIndex.empty()) { 381 ++PPLevelBranchIndex.back(); 382 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size()); 383 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back()); 384 } 385 } while (!PPLevelBranchIndex.empty()); 386 } 387 388 void UnwrappedLineParser::parseFile() { 389 // The top-level context in a file always has declarations, except for pre- 390 // processor directives and JavaScript files. 391 bool MustBeDeclaration = !Line->InPPDirective && !Style.isJavaScript(); 392 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 393 MustBeDeclaration); 394 if (Style.Language == FormatStyle::LK_TextProto) 395 parseBracedList(); 396 else 397 parseLevel(/*HasOpeningBrace=*/false, /*CanContainBracedList=*/true); 398 // Make sure to format the remaining tokens. 399 // 400 // LK_TextProto is special since its top-level is parsed as the body of a 401 // braced list, which does not necessarily have natural line separators such 402 // as a semicolon. Comments after the last entry that have been determined to 403 // not belong to that line, as in: 404 // key: value 405 // // endfile comment 406 // do not have a chance to be put on a line of their own until this point. 407 // Here we add this newline before end-of-file comments. 408 if (Style.Language == FormatStyle::LK_TextProto && 409 !CommentsBeforeNextToken.empty()) 410 addUnwrappedLine(); 411 flushComments(true); 412 addUnwrappedLine(); 413 } 414 415 void UnwrappedLineParser::parseCSharpGenericTypeConstraint() { 416 do { 417 switch (FormatTok->Tok.getKind()) { 418 case tok::l_brace: 419 return; 420 default: 421 if (FormatTok->is(Keywords.kw_where)) { 422 addUnwrappedLine(); 423 nextToken(); 424 parseCSharpGenericTypeConstraint(); 425 break; 426 } 427 nextToken(); 428 break; 429 } 430 } while (!eof()); 431 } 432 433 void UnwrappedLineParser::parseCSharpAttribute() { 434 int UnpairedSquareBrackets = 1; 435 do { 436 switch (FormatTok->Tok.getKind()) { 437 case tok::r_square: 438 nextToken(); 439 --UnpairedSquareBrackets; 440 if (UnpairedSquareBrackets == 0) { 441 addUnwrappedLine(); 442 return; 443 } 444 break; 445 case tok::l_square: 446 ++UnpairedSquareBrackets; 447 nextToken(); 448 break; 449 default: 450 nextToken(); 451 break; 452 } 453 } while (!eof()); 454 } 455 456 bool UnwrappedLineParser::precededByCommentOrPPDirective() const { 457 if (!Lines.empty() && Lines.back().InPPDirective) 458 return true; 459 460 const FormatToken *Previous = Tokens->getPreviousToken(); 461 return Previous && Previous->is(tok::comment) && 462 (Previous->IsMultiline || Previous->NewlinesBefore > 0); 463 } 464 465 /// \brief Parses a level, that is ???. 466 /// \param HasOpeningBrace If that level is started by an opening brace. 467 /// \param CanContainBracedList If the content can contain (at any level) a 468 /// braced list. 469 /// \param NextLBracesType The type for left brace found in this level. 470 /// \returns true if a simple block, or false otherwise. (A simple block has a 471 /// single statement.) 472 bool UnwrappedLineParser::parseLevel(bool HasOpeningBrace, 473 bool CanContainBracedList, 474 IfStmtKind *IfKind, 475 TokenType NextLBracesType) { 476 auto NextLevelLBracesType = NextLBracesType == TT_CompoundRequirementLBrace 477 ? TT_BracedListLBrace 478 : TT_Unknown; 479 const bool IsPrecededByCommentOrPPDirective = 480 !Style.RemoveBracesLLVM || precededByCommentOrPPDirective(); 481 bool HasLabel = false; 482 unsigned StatementCount = 0; 483 bool SwitchLabelEncountered = false; 484 do { 485 if (FormatTok->getType() == TT_AttributeMacro) { 486 nextToken(); 487 continue; 488 } 489 tok::TokenKind kind = FormatTok->Tok.getKind(); 490 if (FormatTok->getType() == TT_MacroBlockBegin) 491 kind = tok::l_brace; 492 else if (FormatTok->getType() == TT_MacroBlockEnd) 493 kind = tok::r_brace; 494 495 auto ParseDefault = [this, HasOpeningBrace, IfKind, NextLevelLBracesType, 496 &HasLabel, &StatementCount] { 497 parseStructuralElement(IfKind, !HasOpeningBrace, NextLevelLBracesType, 498 HasLabel ? nullptr : &HasLabel); 499 ++StatementCount; 500 assert(StatementCount > 0 && "StatementCount overflow!"); 501 }; 502 503 switch (kind) { 504 case tok::comment: 505 nextToken(); 506 addUnwrappedLine(); 507 break; 508 case tok::l_brace: 509 if (NextLBracesType != TT_Unknown) 510 FormatTok->setFinalizedType(NextLBracesType); 511 else if (FormatTok->Previous && 512 FormatTok->Previous->ClosesRequiresClause) { 513 // We need the 'default' case here to correctly parse a function 514 // l_brace. 515 ParseDefault(); 516 continue; 517 } 518 if (CanContainBracedList && !FormatTok->is(TT_MacroBlockBegin) && 519 tryToParseBracedList()) 520 continue; 521 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 522 /*MunchSemi=*/true, /*UnindentWhitesmithBraces=*/false, 523 CanContainBracedList, 524 /*NextLBracesType=*/NextLBracesType); 525 ++StatementCount; 526 assert(StatementCount > 0 && "StatementCount overflow!"); 527 addUnwrappedLine(); 528 break; 529 case tok::r_brace: 530 if (HasOpeningBrace) { 531 if (!Style.RemoveBracesLLVM) 532 return false; 533 if (FormatTok->isNot(tok::r_brace) || StatementCount != 1 || HasLabel || 534 IsPrecededByCommentOrPPDirective || 535 precededByCommentOrPPDirective()) 536 return false; 537 const FormatToken *Next = Tokens->peekNextToken(); 538 return Next->isNot(tok::comment) || Next->NewlinesBefore > 0; 539 } 540 nextToken(); 541 addUnwrappedLine(); 542 break; 543 case tok::kw_default: { 544 unsigned StoredPosition = Tokens->getPosition(); 545 FormatToken *Next; 546 do { 547 Next = Tokens->getNextToken(); 548 assert(Next); 549 } while (Next->is(tok::comment)); 550 FormatTok = Tokens->setPosition(StoredPosition); 551 if (Next->isNot(tok::colon)) { 552 // default not followed by ':' is not a case label; treat it like 553 // an identifier. 554 parseStructuralElement(); 555 break; 556 } 557 // Else, if it is 'default:', fall through to the case handling. 558 LLVM_FALLTHROUGH; 559 } 560 case tok::kw_case: 561 if (Style.isJavaScript() && Line->MustBeDeclaration) { 562 // A 'case: string' style field declaration. 563 parseStructuralElement(); 564 break; 565 } 566 if (!SwitchLabelEncountered && 567 (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1))) 568 ++Line->Level; 569 SwitchLabelEncountered = true; 570 parseStructuralElement(); 571 break; 572 case tok::l_square: 573 if (Style.isCSharp()) { 574 nextToken(); 575 parseCSharpAttribute(); 576 break; 577 } 578 if (handleCppAttributes()) 579 break; 580 LLVM_FALLTHROUGH; 581 default: 582 ParseDefault(); 583 break; 584 } 585 } while (!eof()); 586 return false; 587 } 588 589 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) { 590 // We'll parse forward through the tokens until we hit 591 // a closing brace or eof - note that getNextToken() will 592 // parse macros, so this will magically work inside macro 593 // definitions, too. 594 unsigned StoredPosition = Tokens->getPosition(); 595 FormatToken *Tok = FormatTok; 596 const FormatToken *PrevTok = Tok->Previous; 597 // Keep a stack of positions of lbrace tokens. We will 598 // update information about whether an lbrace starts a 599 // braced init list or a different block during the loop. 600 SmallVector<FormatToken *, 8> LBraceStack; 601 assert(Tok->is(tok::l_brace)); 602 do { 603 // Get next non-comment token. 604 FormatToken *NextTok; 605 do { 606 NextTok = Tokens->getNextToken(); 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 } while (NextTok->isNot(tok::eof)); 647 } 648 649 // Using OriginalColumn to distinguish between ObjC methods and 650 // binary operators is a bit hacky. 651 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 652 NextTok->OriginalColumn == 0; 653 654 // Try to detect a braced list. Note that regardless how we mark inner 655 // braces here, we will overwrite the BlockKind later if we parse a 656 // braced list (where all blocks inside are by default braced lists), 657 // or when we explicitly detect blocks (for example while parsing 658 // lambdas). 659 660 // If we already marked the opening brace as braced list, the closing 661 // must also be part of it. 662 ProbablyBracedList = LBraceStack.back()->is(TT_BracedListLBrace); 663 664 ProbablyBracedList = ProbablyBracedList || 665 (Style.isJavaScript() && 666 NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in, 667 Keywords.kw_as)); 668 ProbablyBracedList = ProbablyBracedList || 669 (Style.isCpp() && NextTok->is(tok::l_paren)); 670 671 // If there is a comma, semicolon or right paren after the closing 672 // brace, we assume this is a braced initializer list. 673 // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a 674 // braced list in JS. 675 ProbablyBracedList = 676 ProbablyBracedList || 677 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 678 tok::r_paren, tok::r_square, tok::l_brace, 679 tok::ellipsis); 680 681 ProbablyBracedList = 682 ProbablyBracedList || 683 (NextTok->is(tok::identifier) && 684 !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace)); 685 686 ProbablyBracedList = ProbablyBracedList || 687 (NextTok->is(tok::semi) && 688 (!ExpectClassBody || LBraceStack.size() != 1)); 689 690 ProbablyBracedList = 691 ProbablyBracedList || 692 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 693 694 if (!Style.isCSharp() && NextTok->is(tok::l_square)) { 695 // We can have an array subscript after a braced init 696 // list, but C++11 attributes are expected after blocks. 697 NextTok = Tokens->getNextToken(); 698 ProbablyBracedList = NextTok->isNot(tok::l_square); 699 } 700 } 701 if (ProbablyBracedList) { 702 Tok->setBlockKind(BK_BracedInit); 703 LBraceStack.back()->setBlockKind(BK_BracedInit); 704 } else { 705 Tok->setBlockKind(BK_Block); 706 LBraceStack.back()->setBlockKind(BK_Block); 707 } 708 } 709 LBraceStack.pop_back(); 710 break; 711 case tok::identifier: 712 if (!Tok->is(TT_StatementMacro)) 713 break; 714 LLVM_FALLTHROUGH; 715 case tok::at: 716 case tok::semi: 717 case tok::kw_if: 718 case tok::kw_while: 719 case tok::kw_for: 720 case tok::kw_switch: 721 case tok::kw_try: 722 case tok::kw___try: 723 if (!LBraceStack.empty() && LBraceStack.back()->is(BK_Unknown)) 724 LBraceStack.back()->setBlockKind(BK_Block); 725 break; 726 default: 727 break; 728 } 729 PrevTok = Tok; 730 Tok = NextTok; 731 } while (Tok->isNot(tok::eof) && !LBraceStack.empty()); 732 733 // Assume other blocks for all unclosed opening braces. 734 for (FormatToken *LBrace : LBraceStack) 735 if (LBrace->is(BK_Unknown)) 736 LBrace->setBlockKind(BK_Block); 737 738 FormatTok = Tokens->setPosition(StoredPosition); 739 } 740 741 template <class T> 742 static inline void hash_combine(std::size_t &seed, const T &v) { 743 std::hash<T> hasher; 744 seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2); 745 } 746 747 size_t UnwrappedLineParser::computePPHash() const { 748 size_t h = 0; 749 for (const auto &i : PPStack) { 750 hash_combine(h, size_t(i.Kind)); 751 hash_combine(h, i.Line); 752 } 753 return h; 754 } 755 756 // Checks whether \p ParsedLine might fit on a single line. We must clone the 757 // tokens of \p ParsedLine before running the token annotator on it so that we 758 // can restore them afterward. 759 bool UnwrappedLineParser::mightFitOnOneLine(UnwrappedLine &ParsedLine) const { 760 const auto ColumnLimit = Style.ColumnLimit; 761 if (ColumnLimit == 0) 762 return true; 763 764 auto &Tokens = ParsedLine.Tokens; 765 assert(!Tokens.empty()); 766 const auto *LastToken = Tokens.back().Tok; 767 assert(LastToken); 768 769 SmallVector<UnwrappedLineNode> SavedTokens(Tokens.size()); 770 771 int Index = 0; 772 for (const auto &Token : Tokens) { 773 assert(Token.Tok); 774 auto &SavedToken = SavedTokens[Index++]; 775 SavedToken.Tok = new FormatToken; 776 SavedToken.Tok->copyFrom(*Token.Tok); 777 SavedToken.Children = std::move(Token.Children); 778 } 779 780 AnnotatedLine Line(ParsedLine); 781 assert(Line.Last == LastToken); 782 783 TokenAnnotator Annotator(Style, Keywords); 784 Annotator.annotate(Line); 785 Annotator.calculateFormattingInformation(Line); 786 787 const int Length = LastToken->TotalLength; 788 789 Index = 0; 790 for (auto &Token : Tokens) { 791 const auto &SavedToken = SavedTokens[Index++]; 792 Token.Tok->copyFrom(*SavedToken.Tok); 793 Token.Children = std::move(SavedToken.Children); 794 delete SavedToken.Tok; 795 } 796 797 return Line.Level * Style.IndentWidth + Length <= ColumnLimit; 798 } 799 800 UnwrappedLineParser::IfStmtKind 801 UnwrappedLineParser::parseBlock(bool MustBeDeclaration, unsigned AddLevels, 802 bool MunchSemi, bool UnindentWhitesmithsBraces, 803 bool CanContainBracedList, 804 TokenType NextLBracesType) { 805 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) && 806 "'{' or macro block token expected"); 807 FormatToken *Tok = FormatTok; 808 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 809 FormatTok->setBlockKind(BK_Block); 810 811 // For Whitesmiths mode, jump to the next level prior to skipping over the 812 // braces. 813 if (AddLevels > 0 && Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) 814 ++Line->Level; 815 816 size_t PPStartHash = computePPHash(); 817 818 unsigned InitialLevel = Line->Level; 819 nextToken(/*LevelDifference=*/AddLevels); 820 821 if (MacroBlock && FormatTok->is(tok::l_paren)) 822 parseParens(); 823 824 size_t NbPreprocessorDirectives = 825 CurrentLines == &Lines ? PreprocessorDirectives.size() : 0; 826 addUnwrappedLine(); 827 size_t OpeningLineIndex = 828 CurrentLines->empty() 829 ? (UnwrappedLine::kInvalidIndex) 830 : (CurrentLines->size() - 1 - NbPreprocessorDirectives); 831 832 // Whitesmiths is weird here. The brace needs to be indented for the namespace 833 // block, but the block itself may not be indented depending on the style 834 // settings. This allows the format to back up one level in those cases. 835 if (UnindentWhitesmithsBraces) 836 --Line->Level; 837 838 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 839 MustBeDeclaration); 840 if (AddLevels > 0u && Style.BreakBeforeBraces != FormatStyle::BS_Whitesmiths) 841 Line->Level += AddLevels; 842 843 IfStmtKind IfKind = IfStmtKind::NotIf; 844 const bool SimpleBlock = parseLevel( 845 /*HasOpeningBrace=*/true, CanContainBracedList, &IfKind, NextLBracesType); 846 847 if (eof()) 848 return IfKind; 849 850 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd) 851 : !FormatTok->is(tok::r_brace)) { 852 Line->Level = InitialLevel; 853 FormatTok->setBlockKind(BK_Block); 854 return IfKind; 855 } 856 857 if (SimpleBlock && Tok->is(tok::l_brace)) { 858 assert(FormatTok->is(tok::r_brace)); 859 const FormatToken *Previous = Tokens->getPreviousToken(); 860 assert(Previous); 861 if (Previous->isNot(tok::r_brace) || Previous->Optional) { 862 assert(!CurrentLines->empty()); 863 if (mightFitOnOneLine(CurrentLines->back())) { 864 Tok->MatchingParen = FormatTok; 865 FormatTok->MatchingParen = Tok; 866 } 867 } 868 } 869 870 size_t PPEndHash = computePPHash(); 871 872 // Munch the closing brace. 873 nextToken(/*LevelDifference=*/-AddLevels); 874 875 if (MacroBlock && FormatTok->is(tok::l_paren)) 876 parseParens(); 877 878 if (FormatTok->is(tok::kw_noexcept)) { 879 // A noexcept in a requires expression. 880 nextToken(); 881 } 882 883 if (FormatTok->is(tok::arrow)) { 884 // Following the } or noexcept we can find a trailing return type arrow 885 // as part of an implicit conversion constraint. 886 nextToken(); 887 parseStructuralElement(); 888 } 889 890 if (MunchSemi && FormatTok->is(tok::semi)) 891 nextToken(); 892 893 Line->Level = InitialLevel; 894 895 if (PPStartHash == PPEndHash) { 896 Line->MatchingOpeningBlockLineIndex = OpeningLineIndex; 897 if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) { 898 // Update the opening line to add the forward reference as well 899 (*CurrentLines)[OpeningLineIndex].MatchingClosingBlockLineIndex = 900 CurrentLines->size() - 1; 901 } 902 } 903 904 return IfKind; 905 } 906 907 static bool isGoogScope(const UnwrappedLine &Line) { 908 // FIXME: Closure-library specific stuff should not be hard-coded but be 909 // configurable. 910 if (Line.Tokens.size() < 4) 911 return false; 912 auto I = Line.Tokens.begin(); 913 if (I->Tok->TokenText != "goog") 914 return false; 915 ++I; 916 if (I->Tok->isNot(tok::period)) 917 return false; 918 ++I; 919 if (I->Tok->TokenText != "scope") 920 return false; 921 ++I; 922 return I->Tok->is(tok::l_paren); 923 } 924 925 static bool isIIFE(const UnwrappedLine &Line, 926 const AdditionalKeywords &Keywords) { 927 // Look for the start of an immediately invoked anonymous function. 928 // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression 929 // This is commonly done in JavaScript to create a new, anonymous scope. 930 // Example: (function() { ... })() 931 if (Line.Tokens.size() < 3) 932 return false; 933 auto I = Line.Tokens.begin(); 934 if (I->Tok->isNot(tok::l_paren)) 935 return false; 936 ++I; 937 if (I->Tok->isNot(Keywords.kw_function)) 938 return false; 939 ++I; 940 return I->Tok->is(tok::l_paren); 941 } 942 943 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 944 const FormatToken &InitialToken) { 945 tok::TokenKind Kind = InitialToken.Tok.getKind(); 946 if (InitialToken.is(TT_NamespaceMacro)) 947 Kind = tok::kw_namespace; 948 949 switch (Kind) { 950 case tok::kw_namespace: 951 return Style.BraceWrapping.AfterNamespace; 952 case tok::kw_class: 953 return Style.BraceWrapping.AfterClass; 954 case tok::kw_union: 955 return Style.BraceWrapping.AfterUnion; 956 case tok::kw_struct: 957 return Style.BraceWrapping.AfterStruct; 958 case tok::kw_enum: 959 return Style.BraceWrapping.AfterEnum; 960 default: 961 return false; 962 } 963 } 964 965 void UnwrappedLineParser::parseChildBlock( 966 bool CanContainBracedList, clang::format::TokenType NextLBracesType) { 967 FormatTok->setBlockKind(BK_Block); 968 nextToken(); 969 { 970 bool SkipIndent = (Style.isJavaScript() && 971 (isGoogScope(*Line) || isIIFE(*Line, Keywords))); 972 ScopedLineState LineState(*this); 973 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 974 /*MustBeDeclaration=*/false); 975 Line->Level += SkipIndent ? 0 : 1; 976 parseLevel(/*HasOpeningBrace=*/true, CanContainBracedList, 977 /*IfKind=*/nullptr, NextLBracesType); 978 flushComments(isOnNewLine(*FormatTok)); 979 Line->Level -= SkipIndent ? 0 : 1; 980 } 981 nextToken(); 982 } 983 984 void UnwrappedLineParser::parsePPDirective() { 985 assert(FormatTok->is(tok::hash) && "'#' expected"); 986 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 987 988 nextToken(); 989 990 if (!FormatTok->Tok.getIdentifierInfo()) { 991 parsePPUnknown(); 992 return; 993 } 994 995 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 996 case tok::pp_define: 997 parsePPDefine(); 998 return; 999 case tok::pp_if: 1000 parsePPIf(/*IfDef=*/false); 1001 break; 1002 case tok::pp_ifdef: 1003 case tok::pp_ifndef: 1004 parsePPIf(/*IfDef=*/true); 1005 break; 1006 case tok::pp_else: 1007 parsePPElse(); 1008 break; 1009 case tok::pp_elifdef: 1010 case tok::pp_elifndef: 1011 case tok::pp_elif: 1012 parsePPElIf(); 1013 break; 1014 case tok::pp_endif: 1015 parsePPEndIf(); 1016 break; 1017 default: 1018 parsePPUnknown(); 1019 break; 1020 } 1021 } 1022 1023 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 1024 size_t Line = CurrentLines->size(); 1025 if (CurrentLines == &PreprocessorDirectives) 1026 Line += Lines.size(); 1027 1028 if (Unreachable || 1029 (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable)) 1030 PPStack.push_back({PP_Unreachable, Line}); 1031 else 1032 PPStack.push_back({PP_Conditional, Line}); 1033 } 1034 1035 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 1036 ++PPBranchLevel; 1037 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 1038 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 1039 PPLevelBranchIndex.push_back(0); 1040 PPLevelBranchCount.push_back(0); 1041 } 1042 PPChainBranchIndex.push(0); 1043 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 1044 conditionalCompilationCondition(Unreachable || Skip); 1045 } 1046 1047 void UnwrappedLineParser::conditionalCompilationAlternative() { 1048 if (!PPStack.empty()) 1049 PPStack.pop_back(); 1050 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 1051 if (!PPChainBranchIndex.empty()) 1052 ++PPChainBranchIndex.top(); 1053 conditionalCompilationCondition( 1054 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 1055 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 1056 } 1057 1058 void UnwrappedLineParser::conditionalCompilationEnd() { 1059 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 1060 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 1061 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) 1062 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 1063 } 1064 // Guard against #endif's without #if. 1065 if (PPBranchLevel > -1) 1066 --PPBranchLevel; 1067 if (!PPChainBranchIndex.empty()) 1068 PPChainBranchIndex.pop(); 1069 if (!PPStack.empty()) 1070 PPStack.pop_back(); 1071 } 1072 1073 void UnwrappedLineParser::parsePPIf(bool IfDef) { 1074 bool IfNDef = FormatTok->is(tok::pp_ifndef); 1075 nextToken(); 1076 bool Unreachable = false; 1077 if (!IfDef && (FormatTok->is(tok::kw_false) || FormatTok->TokenText == "0")) 1078 Unreachable = true; 1079 if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG") 1080 Unreachable = true; 1081 conditionalCompilationStart(Unreachable); 1082 FormatToken *IfCondition = FormatTok; 1083 // If there's a #ifndef on the first line, and the only lines before it are 1084 // comments, it could be an include guard. 1085 bool MaybeIncludeGuard = IfNDef; 1086 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) 1087 for (auto &Line : Lines) { 1088 if (!Line.Tokens.front().Tok->is(tok::comment)) { 1089 MaybeIncludeGuard = false; 1090 IncludeGuard = IG_Rejected; 1091 break; 1092 } 1093 } 1094 --PPBranchLevel; 1095 parsePPUnknown(); 1096 ++PPBranchLevel; 1097 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) { 1098 IncludeGuard = IG_IfNdefed; 1099 IncludeGuardToken = IfCondition; 1100 } 1101 } 1102 1103 void UnwrappedLineParser::parsePPElse() { 1104 // If a potential include guard has an #else, it's not an include guard. 1105 if (IncludeGuard == IG_Defined && PPBranchLevel == 0) 1106 IncludeGuard = IG_Rejected; 1107 conditionalCompilationAlternative(); 1108 if (PPBranchLevel > -1) 1109 --PPBranchLevel; 1110 parsePPUnknown(); 1111 ++PPBranchLevel; 1112 } 1113 1114 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 1115 1116 void UnwrappedLineParser::parsePPEndIf() { 1117 conditionalCompilationEnd(); 1118 parsePPUnknown(); 1119 // If the #endif of a potential include guard is the last thing in the file, 1120 // then we found an include guard. 1121 if (IncludeGuard == IG_Defined && PPBranchLevel == -1 && Tokens->isEOF() && 1122 Style.IndentPPDirectives != FormatStyle::PPDIS_None) 1123 IncludeGuard = IG_Found; 1124 } 1125 1126 void UnwrappedLineParser::parsePPDefine() { 1127 nextToken(); 1128 1129 if (!FormatTok->Tok.getIdentifierInfo()) { 1130 IncludeGuard = IG_Rejected; 1131 IncludeGuardToken = nullptr; 1132 parsePPUnknown(); 1133 return; 1134 } 1135 1136 if (IncludeGuard == IG_IfNdefed && 1137 IncludeGuardToken->TokenText == FormatTok->TokenText) { 1138 IncludeGuard = IG_Defined; 1139 IncludeGuardToken = nullptr; 1140 for (auto &Line : Lines) { 1141 if (!Line.Tokens.front().Tok->isOneOf(tok::comment, tok::hash)) { 1142 IncludeGuard = IG_Rejected; 1143 break; 1144 } 1145 } 1146 } 1147 1148 // In the context of a define, even keywords should be treated as normal 1149 // identifiers. Setting the kind to identifier is not enough, because we need 1150 // to treat additional keywords like __except as well, which are already 1151 // identifiers. Setting the identifier info to null interferes with include 1152 // guard processing above, and changes preprocessing nesting. 1153 FormatTok->Tok.setKind(tok::identifier); 1154 FormatTok->Tok.setIdentifierInfo(Keywords.kw_internal_ident_after_define); 1155 nextToken(); 1156 if (FormatTok->Tok.getKind() == tok::l_paren && 1157 !FormatTok->hasWhitespaceBefore()) 1158 parseParens(); 1159 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None) 1160 Line->Level += PPBranchLevel + 1; 1161 addUnwrappedLine(); 1162 ++Line->Level; 1163 1164 // Errors during a preprocessor directive can only affect the layout of the 1165 // preprocessor directive, and thus we ignore them. An alternative approach 1166 // would be to use the same approach we use on the file level (no 1167 // re-indentation if there was a structural error) within the macro 1168 // definition. 1169 parseFile(); 1170 } 1171 1172 void UnwrappedLineParser::parsePPUnknown() { 1173 do { 1174 nextToken(); 1175 } while (!eof()); 1176 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None) 1177 Line->Level += PPBranchLevel + 1; 1178 addUnwrappedLine(); 1179 } 1180 1181 // Here we exclude certain tokens that are not usually the first token in an 1182 // unwrapped line. This is used in attempt to distinguish macro calls without 1183 // trailing semicolons from other constructs split to several lines. 1184 static bool tokenCanStartNewLine(const FormatToken &Tok) { 1185 // Semicolon can be a null-statement, l_square can be a start of a macro or 1186 // a C++11 attribute, but this doesn't seem to be common. 1187 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 1188 Tok.isNot(TT_AttributeSquare) && 1189 // Tokens that can only be used as binary operators and a part of 1190 // overloaded operator names. 1191 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 1192 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 1193 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 1194 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 1195 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 1196 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 1197 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 1198 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 1199 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 1200 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 1201 Tok.isNot(tok::lesslessequal) && 1202 // Colon is used in labels, base class lists, initializer lists, 1203 // range-based for loops, ternary operator, but should never be the 1204 // first token in an unwrapped line. 1205 Tok.isNot(tok::colon) && 1206 // 'noexcept' is a trailing annotation. 1207 Tok.isNot(tok::kw_noexcept); 1208 } 1209 1210 static bool mustBeJSIdent(const AdditionalKeywords &Keywords, 1211 const FormatToken *FormatTok) { 1212 // FIXME: This returns true for C/C++ keywords like 'struct'. 1213 return FormatTok->is(tok::identifier) && 1214 (FormatTok->Tok.getIdentifierInfo() == nullptr || 1215 !FormatTok->isOneOf( 1216 Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async, 1217 Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally, 1218 Keywords.kw_function, Keywords.kw_import, Keywords.kw_is, 1219 Keywords.kw_let, Keywords.kw_var, tok::kw_const, 1220 Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements, 1221 Keywords.kw_instanceof, Keywords.kw_interface, 1222 Keywords.kw_override, Keywords.kw_throws, Keywords.kw_from)); 1223 } 1224 1225 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords, 1226 const FormatToken *FormatTok) { 1227 return FormatTok->Tok.isLiteral() || 1228 FormatTok->isOneOf(tok::kw_true, tok::kw_false) || 1229 mustBeJSIdent(Keywords, FormatTok); 1230 } 1231 1232 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement 1233 // when encountered after a value (see mustBeJSIdentOrValue). 1234 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords, 1235 const FormatToken *FormatTok) { 1236 return FormatTok->isOneOf( 1237 tok::kw_return, Keywords.kw_yield, 1238 // conditionals 1239 tok::kw_if, tok::kw_else, 1240 // loops 1241 tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break, 1242 // switch/case 1243 tok::kw_switch, tok::kw_case, 1244 // exceptions 1245 tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally, 1246 // declaration 1247 tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let, 1248 Keywords.kw_async, Keywords.kw_function, 1249 // import/export 1250 Keywords.kw_import, tok::kw_export); 1251 } 1252 1253 // Checks whether a token is a type in K&R C (aka C78). 1254 static bool isC78Type(const FormatToken &Tok) { 1255 return Tok.isOneOf(tok::kw_char, tok::kw_short, tok::kw_int, tok::kw_long, 1256 tok::kw_unsigned, tok::kw_float, tok::kw_double, 1257 tok::identifier); 1258 } 1259 1260 // This function checks whether a token starts the first parameter declaration 1261 // in a K&R C (aka C78) function definition, e.g.: 1262 // int f(a, b) 1263 // short a, b; 1264 // { 1265 // return a + b; 1266 // } 1267 static bool isC78ParameterDecl(const FormatToken *Tok, const FormatToken *Next, 1268 const FormatToken *FuncName) { 1269 assert(Tok); 1270 assert(Next); 1271 assert(FuncName); 1272 1273 if (FuncName->isNot(tok::identifier)) 1274 return false; 1275 1276 const FormatToken *Prev = FuncName->Previous; 1277 if (!Prev || (Prev->isNot(tok::star) && !isC78Type(*Prev))) 1278 return false; 1279 1280 if (!isC78Type(*Tok) && 1281 !Tok->isOneOf(tok::kw_register, tok::kw_struct, tok::kw_union)) 1282 return false; 1283 1284 if (Next->isNot(tok::star) && !Next->Tok.getIdentifierInfo()) 1285 return false; 1286 1287 Tok = Tok->Previous; 1288 if (!Tok || Tok->isNot(tok::r_paren)) 1289 return false; 1290 1291 Tok = Tok->Previous; 1292 if (!Tok || Tok->isNot(tok::identifier)) 1293 return false; 1294 1295 return Tok->Previous && Tok->Previous->isOneOf(tok::l_paren, tok::comma); 1296 } 1297 1298 void UnwrappedLineParser::parseModuleImport() { 1299 nextToken(); 1300 while (!eof()) { 1301 if (FormatTok->is(tok::colon)) { 1302 FormatTok->setFinalizedType(TT_ModulePartitionColon); 1303 } 1304 // Handle import <foo/bar.h> as we would an include statement. 1305 else if (FormatTok->is(tok::less)) { 1306 nextToken(); 1307 while (!FormatTok->isOneOf(tok::semi, tok::greater, tok::eof)) { 1308 // Mark tokens up to the trailing line comments as implicit string 1309 // literals. 1310 if (FormatTok->isNot(tok::comment) && 1311 !FormatTok->TokenText.startswith("//")) 1312 FormatTok->setFinalizedType(TT_ImplicitStringLiteral); 1313 nextToken(); 1314 } 1315 } 1316 if (FormatTok->is(tok::semi)) { 1317 nextToken(); 1318 break; 1319 } 1320 nextToken(); 1321 } 1322 1323 addUnwrappedLine(); 1324 } 1325 1326 // readTokenWithJavaScriptASI reads the next token and terminates the current 1327 // line if JavaScript Automatic Semicolon Insertion must 1328 // happen between the current token and the next token. 1329 // 1330 // This method is conservative - it cannot cover all edge cases of JavaScript, 1331 // but only aims to correctly handle certain well known cases. It *must not* 1332 // return true in speculative cases. 1333 void UnwrappedLineParser::readTokenWithJavaScriptASI() { 1334 FormatToken *Previous = FormatTok; 1335 readToken(); 1336 FormatToken *Next = FormatTok; 1337 1338 bool IsOnSameLine = 1339 CommentsBeforeNextToken.empty() 1340 ? Next->NewlinesBefore == 0 1341 : CommentsBeforeNextToken.front()->NewlinesBefore == 0; 1342 if (IsOnSameLine) 1343 return; 1344 1345 bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous); 1346 bool PreviousStartsTemplateExpr = 1347 Previous->is(TT_TemplateString) && Previous->TokenText.endswith("${"); 1348 if (PreviousMustBeValue || Previous->is(tok::r_paren)) { 1349 // If the line contains an '@' sign, the previous token might be an 1350 // annotation, which can precede another identifier/value. 1351 bool HasAt = llvm::any_of(Line->Tokens, [](UnwrappedLineNode &LineNode) { 1352 return LineNode.Tok->is(tok::at); 1353 }); 1354 if (HasAt) 1355 return; 1356 } 1357 if (Next->is(tok::exclaim) && PreviousMustBeValue) 1358 return addUnwrappedLine(); 1359 bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next); 1360 bool NextEndsTemplateExpr = 1361 Next->is(TT_TemplateString) && Next->TokenText.startswith("}"); 1362 if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr && 1363 (PreviousMustBeValue || 1364 Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus, 1365 tok::minusminus))) 1366 return addUnwrappedLine(); 1367 if ((PreviousMustBeValue || Previous->is(tok::r_paren)) && 1368 isJSDeclOrStmt(Keywords, Next)) 1369 return addUnwrappedLine(); 1370 } 1371 1372 void UnwrappedLineParser::parseStructuralElement(IfStmtKind *IfKind, 1373 bool IsTopLevel, 1374 TokenType NextLBracesType, 1375 bool *HasLabel) { 1376 if (Style.Language == FormatStyle::LK_TableGen && 1377 FormatTok->is(tok::pp_include)) { 1378 nextToken(); 1379 if (FormatTok->is(tok::string_literal)) 1380 nextToken(); 1381 addUnwrappedLine(); 1382 return; 1383 } 1384 switch (FormatTok->Tok.getKind()) { 1385 case tok::kw_asm: 1386 nextToken(); 1387 if (FormatTok->is(tok::l_brace)) { 1388 FormatTok->setFinalizedType(TT_InlineASMBrace); 1389 nextToken(); 1390 while (FormatTok && FormatTok->isNot(tok::eof)) { 1391 if (FormatTok->is(tok::r_brace)) { 1392 FormatTok->setFinalizedType(TT_InlineASMBrace); 1393 nextToken(); 1394 addUnwrappedLine(); 1395 break; 1396 } 1397 FormatTok->Finalized = true; 1398 nextToken(); 1399 } 1400 } 1401 break; 1402 case tok::kw_namespace: 1403 parseNamespace(); 1404 return; 1405 case tok::kw_public: 1406 case tok::kw_protected: 1407 case tok::kw_private: 1408 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 1409 Style.isCSharp()) 1410 nextToken(); 1411 else 1412 parseAccessSpecifier(); 1413 return; 1414 case tok::kw_if: 1415 if (Style.isJavaScript() && Line->MustBeDeclaration) 1416 // field/method declaration. 1417 break; 1418 parseIfThenElse(IfKind); 1419 return; 1420 case tok::kw_for: 1421 case tok::kw_while: 1422 if (Style.isJavaScript() && Line->MustBeDeclaration) 1423 // field/method declaration. 1424 break; 1425 parseForOrWhileLoop(); 1426 return; 1427 case tok::kw_do: 1428 if (Style.isJavaScript() && Line->MustBeDeclaration) 1429 // field/method declaration. 1430 break; 1431 parseDoWhile(); 1432 return; 1433 case tok::kw_switch: 1434 if (Style.isJavaScript() && Line->MustBeDeclaration) 1435 // 'switch: string' field declaration. 1436 break; 1437 parseSwitch(); 1438 return; 1439 case tok::kw_default: 1440 if (Style.isJavaScript() && Line->MustBeDeclaration) 1441 // 'default: string' field declaration. 1442 break; 1443 nextToken(); 1444 if (FormatTok->is(tok::colon)) { 1445 parseLabel(); 1446 return; 1447 } 1448 // e.g. "default void f() {}" in a Java interface. 1449 break; 1450 case tok::kw_case: 1451 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1452 // 'case: string' field declaration. 1453 nextToken(); 1454 break; 1455 } 1456 parseCaseLabel(); 1457 return; 1458 case tok::kw_try: 1459 case tok::kw___try: 1460 if (Style.isJavaScript() && Line->MustBeDeclaration) 1461 // field/method declaration. 1462 break; 1463 parseTryCatch(); 1464 return; 1465 case tok::kw_extern: 1466 nextToken(); 1467 if (FormatTok->is(tok::string_literal)) { 1468 nextToken(); 1469 if (FormatTok->is(tok::l_brace)) { 1470 if (Style.BraceWrapping.AfterExternBlock) 1471 addUnwrappedLine(); 1472 // Either we indent or for backwards compatibility we follow the 1473 // AfterExternBlock style. 1474 unsigned AddLevels = 1475 (Style.IndentExternBlock == FormatStyle::IEBS_Indent) || 1476 (Style.BraceWrapping.AfterExternBlock && 1477 Style.IndentExternBlock == 1478 FormatStyle::IEBS_AfterExternBlock) 1479 ? 1u 1480 : 0u; 1481 parseBlock(/*MustBeDeclaration=*/true, AddLevels); 1482 addUnwrappedLine(); 1483 return; 1484 } 1485 } 1486 break; 1487 case tok::kw_export: 1488 if (Style.isJavaScript()) { 1489 parseJavaScriptEs6ImportExport(); 1490 return; 1491 } 1492 if (!Style.isCpp()) 1493 break; 1494 // Handle C++ "(inline|export) namespace". 1495 LLVM_FALLTHROUGH; 1496 case tok::kw_inline: 1497 nextToken(); 1498 if (FormatTok->is(tok::kw_namespace)) { 1499 parseNamespace(); 1500 return; 1501 } 1502 break; 1503 case tok::identifier: 1504 if (FormatTok->is(TT_ForEachMacro)) { 1505 parseForOrWhileLoop(); 1506 return; 1507 } 1508 if (FormatTok->is(TT_MacroBlockBegin)) { 1509 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 1510 /*MunchSemi=*/false); 1511 return; 1512 } 1513 if (FormatTok->is(Keywords.kw_import)) { 1514 if (Style.isJavaScript()) { 1515 parseJavaScriptEs6ImportExport(); 1516 return; 1517 } 1518 if (Style.Language == FormatStyle::LK_Proto) { 1519 nextToken(); 1520 if (FormatTok->is(tok::kw_public)) 1521 nextToken(); 1522 if (!FormatTok->is(tok::string_literal)) 1523 return; 1524 nextToken(); 1525 if (FormatTok->is(tok::semi)) 1526 nextToken(); 1527 addUnwrappedLine(); 1528 return; 1529 } 1530 if (Style.isCpp()) { 1531 parseModuleImport(); 1532 return; 1533 } 1534 } 1535 if (Style.isCpp() && 1536 FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 1537 Keywords.kw_slots, Keywords.kw_qslots)) { 1538 nextToken(); 1539 if (FormatTok->is(tok::colon)) { 1540 nextToken(); 1541 addUnwrappedLine(); 1542 return; 1543 } 1544 } 1545 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1546 parseStatementMacro(); 1547 return; 1548 } 1549 if (Style.isCpp() && FormatTok->is(TT_NamespaceMacro)) { 1550 parseNamespace(); 1551 return; 1552 } 1553 // In all other cases, parse the declaration. 1554 break; 1555 default: 1556 break; 1557 } 1558 do { 1559 const FormatToken *Previous = FormatTok->Previous; 1560 switch (FormatTok->Tok.getKind()) { 1561 case tok::at: 1562 nextToken(); 1563 if (FormatTok->is(tok::l_brace)) { 1564 nextToken(); 1565 parseBracedList(); 1566 break; 1567 } else if (Style.Language == FormatStyle::LK_Java && 1568 FormatTok->is(Keywords.kw_interface)) { 1569 nextToken(); 1570 break; 1571 } 1572 switch (FormatTok->Tok.getObjCKeywordID()) { 1573 case tok::objc_public: 1574 case tok::objc_protected: 1575 case tok::objc_package: 1576 case tok::objc_private: 1577 return parseAccessSpecifier(); 1578 case tok::objc_interface: 1579 case tok::objc_implementation: 1580 return parseObjCInterfaceOrImplementation(); 1581 case tok::objc_protocol: 1582 if (parseObjCProtocol()) 1583 return; 1584 break; 1585 case tok::objc_end: 1586 return; // Handled by the caller. 1587 case tok::objc_optional: 1588 case tok::objc_required: 1589 nextToken(); 1590 addUnwrappedLine(); 1591 return; 1592 case tok::objc_autoreleasepool: 1593 nextToken(); 1594 if (FormatTok->is(tok::l_brace)) { 1595 if (Style.BraceWrapping.AfterControlStatement == 1596 FormatStyle::BWACS_Always) 1597 addUnwrappedLine(); 1598 parseBlock(); 1599 } 1600 addUnwrappedLine(); 1601 return; 1602 case tok::objc_synchronized: 1603 nextToken(); 1604 if (FormatTok->is(tok::l_paren)) 1605 // Skip synchronization object 1606 parseParens(); 1607 if (FormatTok->is(tok::l_brace)) { 1608 if (Style.BraceWrapping.AfterControlStatement == 1609 FormatStyle::BWACS_Always) 1610 addUnwrappedLine(); 1611 parseBlock(); 1612 } 1613 addUnwrappedLine(); 1614 return; 1615 case tok::objc_try: 1616 // This branch isn't strictly necessary (the kw_try case below would 1617 // do this too after the tok::at is parsed above). But be explicit. 1618 parseTryCatch(); 1619 return; 1620 default: 1621 break; 1622 } 1623 break; 1624 case tok::kw_concept: 1625 parseConcept(); 1626 return; 1627 case tok::kw_requires: { 1628 if (Style.isCpp()) { 1629 bool ParsedClause = parseRequires(); 1630 if (ParsedClause) 1631 return; 1632 } else { 1633 nextToken(); 1634 } 1635 break; 1636 } 1637 case tok::kw_enum: 1638 // Ignore if this is part of "template <enum ...". 1639 if (Previous && Previous->is(tok::less)) { 1640 nextToken(); 1641 break; 1642 } 1643 1644 // parseEnum falls through and does not yet add an unwrapped line as an 1645 // enum definition can start a structural element. 1646 if (!parseEnum()) 1647 break; 1648 // This only applies for C++. 1649 if (!Style.isCpp()) { 1650 addUnwrappedLine(); 1651 return; 1652 } 1653 break; 1654 case tok::kw_typedef: 1655 nextToken(); 1656 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 1657 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS, 1658 Keywords.kw_CF_CLOSED_ENUM, 1659 Keywords.kw_NS_CLOSED_ENUM)) 1660 parseEnum(); 1661 break; 1662 case tok::kw_struct: 1663 case tok::kw_union: 1664 case tok::kw_class: 1665 if (parseStructLike()) 1666 return; 1667 break; 1668 case tok::period: 1669 nextToken(); 1670 // In Java, classes have an implicit static member "class". 1671 if (Style.Language == FormatStyle::LK_Java && FormatTok && 1672 FormatTok->is(tok::kw_class)) 1673 nextToken(); 1674 if (Style.isJavaScript() && FormatTok && 1675 FormatTok->Tok.getIdentifierInfo()) 1676 // JavaScript only has pseudo keywords, all keywords are allowed to 1677 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 1678 nextToken(); 1679 break; 1680 case tok::semi: 1681 nextToken(); 1682 addUnwrappedLine(); 1683 return; 1684 case tok::r_brace: 1685 addUnwrappedLine(); 1686 return; 1687 case tok::l_paren: { 1688 parseParens(); 1689 // Break the unwrapped line if a K&R C function definition has a parameter 1690 // declaration. 1691 if (!IsTopLevel || !Style.isCpp() || !Previous || FormatTok->is(tok::eof)) 1692 break; 1693 if (isC78ParameterDecl(FormatTok, Tokens->peekNextToken(), Previous)) { 1694 addUnwrappedLine(); 1695 return; 1696 } 1697 break; 1698 } 1699 case tok::kw_operator: 1700 nextToken(); 1701 if (FormatTok->isBinaryOperator()) 1702 nextToken(); 1703 break; 1704 case tok::caret: 1705 nextToken(); 1706 if (FormatTok->Tok.isAnyIdentifier() || 1707 FormatTok->isSimpleTypeSpecifier()) 1708 nextToken(); 1709 if (FormatTok->is(tok::l_paren)) 1710 parseParens(); 1711 if (FormatTok->is(tok::l_brace)) 1712 parseChildBlock(); 1713 break; 1714 case tok::l_brace: 1715 if (NextLBracesType != TT_Unknown) 1716 FormatTok->setFinalizedType(NextLBracesType); 1717 if (!tryToParsePropertyAccessor() && !tryToParseBracedList()) { 1718 // A block outside of parentheses must be the last part of a 1719 // structural element. 1720 // FIXME: Figure out cases where this is not true, and add projections 1721 // for them (the one we know is missing are lambdas). 1722 if (Style.Language == FormatStyle::LK_Java && 1723 Line->Tokens.front().Tok->is(Keywords.kw_synchronized)) { 1724 // If necessary, we could set the type to something different than 1725 // TT_FunctionLBrace. 1726 if (Style.BraceWrapping.AfterControlStatement == 1727 FormatStyle::BWACS_Always) 1728 addUnwrappedLine(); 1729 } else if (Style.BraceWrapping.AfterFunction) { 1730 addUnwrappedLine(); 1731 } 1732 if (!Line->InPPDirective) 1733 FormatTok->setFinalizedType(TT_FunctionLBrace); 1734 parseBlock(); 1735 addUnwrappedLine(); 1736 return; 1737 } 1738 // Otherwise this was a braced init list, and the structural 1739 // element continues. 1740 break; 1741 case tok::kw_try: 1742 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1743 // field/method declaration. 1744 nextToken(); 1745 break; 1746 } 1747 // We arrive here when parsing function-try blocks. 1748 if (Style.BraceWrapping.AfterFunction) 1749 addUnwrappedLine(); 1750 parseTryCatch(); 1751 return; 1752 case tok::identifier: { 1753 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where) && 1754 Line->MustBeDeclaration) { 1755 addUnwrappedLine(); 1756 parseCSharpGenericTypeConstraint(); 1757 break; 1758 } 1759 if (FormatTok->is(TT_MacroBlockEnd)) { 1760 addUnwrappedLine(); 1761 return; 1762 } 1763 1764 // Function declarations (as opposed to function expressions) are parsed 1765 // on their own unwrapped line by continuing this loop. Function 1766 // expressions (functions that are not on their own line) must not create 1767 // a new unwrapped line, so they are special cased below. 1768 size_t TokenCount = Line->Tokens.size(); 1769 if (Style.isJavaScript() && FormatTok->is(Keywords.kw_function) && 1770 (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is( 1771 Keywords.kw_async)))) { 1772 tryToParseJSFunction(); 1773 break; 1774 } 1775 if ((Style.isJavaScript() || Style.Language == FormatStyle::LK_Java) && 1776 FormatTok->is(Keywords.kw_interface)) { 1777 if (Style.isJavaScript()) { 1778 // In JavaScript/TypeScript, "interface" can be used as a standalone 1779 // identifier, e.g. in `var interface = 1;`. If "interface" is 1780 // followed by another identifier, it is very like to be an actual 1781 // interface declaration. 1782 unsigned StoredPosition = Tokens->getPosition(); 1783 FormatToken *Next = Tokens->getNextToken(); 1784 FormatTok = Tokens->setPosition(StoredPosition); 1785 if (!mustBeJSIdent(Keywords, Next)) { 1786 nextToken(); 1787 break; 1788 } 1789 } 1790 parseRecord(); 1791 addUnwrappedLine(); 1792 return; 1793 } 1794 1795 if (FormatTok->is(Keywords.kw_interface)) { 1796 if (parseStructLike()) 1797 return; 1798 break; 1799 } 1800 1801 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1802 parseStatementMacro(); 1803 return; 1804 } 1805 1806 // See if the following token should start a new unwrapped line. 1807 StringRef Text = FormatTok->TokenText; 1808 1809 FormatToken *PreviousToken = FormatTok; 1810 nextToken(); 1811 1812 // JS doesn't have macros, and within classes colons indicate fields, not 1813 // labels. 1814 if (Style.isJavaScript()) 1815 break; 1816 1817 TokenCount = Line->Tokens.size(); 1818 if (TokenCount == 1 || 1819 (TokenCount == 2 && Line->Tokens.front().Tok->is(tok::comment))) { 1820 if (FormatTok->is(tok::colon) && !Line->MustBeDeclaration) { 1821 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1822 parseLabel(!Style.IndentGotoLabels); 1823 if (HasLabel) 1824 *HasLabel = true; 1825 return; 1826 } 1827 // Recognize function-like macro usages without trailing semicolon as 1828 // well as free-standing macros like Q_OBJECT. 1829 bool FunctionLike = FormatTok->is(tok::l_paren); 1830 if (FunctionLike) 1831 parseParens(); 1832 1833 bool FollowedByNewline = 1834 CommentsBeforeNextToken.empty() 1835 ? FormatTok->NewlinesBefore > 0 1836 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1837 1838 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1839 tokenCanStartNewLine(*FormatTok) && Text == Text.upper() && 1840 !PreviousToken->isTypeFinalized()) { 1841 PreviousToken->setFinalizedType(TT_FunctionLikeOrFreestandingMacro); 1842 addUnwrappedLine(); 1843 return; 1844 } 1845 } 1846 break; 1847 } 1848 case tok::equal: 1849 if ((Style.isJavaScript() || Style.isCSharp()) && 1850 FormatTok->is(TT_FatArrow)) { 1851 tryToParseChildBlock(); 1852 break; 1853 } 1854 1855 nextToken(); 1856 if (FormatTok->is(tok::l_brace)) { 1857 // Block kind should probably be set to BK_BracedInit for any language. 1858 // C# needs this change to ensure that array initialisers and object 1859 // initialisers are indented the same way. 1860 if (Style.isCSharp()) 1861 FormatTok->setBlockKind(BK_BracedInit); 1862 nextToken(); 1863 parseBracedList(); 1864 } else if (Style.Language == FormatStyle::LK_Proto && 1865 FormatTok->is(tok::less)) { 1866 nextToken(); 1867 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 1868 /*ClosingBraceKind=*/tok::greater); 1869 } 1870 break; 1871 case tok::l_square: 1872 parseSquare(); 1873 break; 1874 case tok::kw_new: 1875 parseNew(); 1876 break; 1877 case tok::kw_case: 1878 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1879 // 'case: string' field declaration. 1880 nextToken(); 1881 break; 1882 } 1883 parseCaseLabel(); 1884 break; 1885 default: 1886 nextToken(); 1887 break; 1888 } 1889 } while (!eof()); 1890 } 1891 1892 bool UnwrappedLineParser::tryToParsePropertyAccessor() { 1893 assert(FormatTok->is(tok::l_brace)); 1894 if (!Style.isCSharp()) 1895 return false; 1896 // See if it's a property accessor. 1897 if (FormatTok->Previous->isNot(tok::identifier)) 1898 return false; 1899 1900 // See if we are inside a property accessor. 1901 // 1902 // Record the current tokenPosition so that we can advance and 1903 // reset the current token. `Next` is not set yet so we need 1904 // another way to advance along the token stream. 1905 unsigned int StoredPosition = Tokens->getPosition(); 1906 FormatToken *Tok = Tokens->getNextToken(); 1907 1908 // A trivial property accessor is of the form: 1909 // { [ACCESS_SPECIFIER] [get]; [ACCESS_SPECIFIER] [set|init] } 1910 // Track these as they do not require line breaks to be introduced. 1911 bool HasSpecialAccessor = false; 1912 bool IsTrivialPropertyAccessor = true; 1913 while (!eof()) { 1914 if (Tok->isOneOf(tok::semi, tok::kw_public, tok::kw_private, 1915 tok::kw_protected, Keywords.kw_internal, Keywords.kw_get, 1916 Keywords.kw_init, Keywords.kw_set)) { 1917 if (Tok->isOneOf(Keywords.kw_get, Keywords.kw_init, Keywords.kw_set)) 1918 HasSpecialAccessor = true; 1919 Tok = Tokens->getNextToken(); 1920 continue; 1921 } 1922 if (Tok->isNot(tok::r_brace)) 1923 IsTrivialPropertyAccessor = false; 1924 break; 1925 } 1926 1927 if (!HasSpecialAccessor) { 1928 Tokens->setPosition(StoredPosition); 1929 return false; 1930 } 1931 1932 // Try to parse the property accessor: 1933 // https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties 1934 Tokens->setPosition(StoredPosition); 1935 if (!IsTrivialPropertyAccessor && Style.BraceWrapping.AfterFunction) 1936 addUnwrappedLine(); 1937 nextToken(); 1938 do { 1939 switch (FormatTok->Tok.getKind()) { 1940 case tok::r_brace: 1941 nextToken(); 1942 if (FormatTok->is(tok::equal)) { 1943 while (!eof() && FormatTok->isNot(tok::semi)) 1944 nextToken(); 1945 nextToken(); 1946 } 1947 addUnwrappedLine(); 1948 return true; 1949 case tok::l_brace: 1950 ++Line->Level; 1951 parseBlock(/*MustBeDeclaration=*/true); 1952 addUnwrappedLine(); 1953 --Line->Level; 1954 break; 1955 case tok::equal: 1956 if (FormatTok->is(TT_FatArrow)) { 1957 ++Line->Level; 1958 do { 1959 nextToken(); 1960 } while (!eof() && FormatTok->isNot(tok::semi)); 1961 nextToken(); 1962 addUnwrappedLine(); 1963 --Line->Level; 1964 break; 1965 } 1966 nextToken(); 1967 break; 1968 default: 1969 if (FormatTok->isOneOf(Keywords.kw_get, Keywords.kw_init, 1970 Keywords.kw_set) && 1971 !IsTrivialPropertyAccessor) { 1972 // Non-trivial get/set needs to be on its own line. 1973 addUnwrappedLine(); 1974 } 1975 nextToken(); 1976 } 1977 } while (!eof()); 1978 1979 // Unreachable for well-formed code (paired '{' and '}'). 1980 return true; 1981 } 1982 1983 bool UnwrappedLineParser::tryToParseLambda() { 1984 assert(FormatTok->is(tok::l_square)); 1985 if (!Style.isCpp()) { 1986 nextToken(); 1987 return false; 1988 } 1989 FormatToken &LSquare = *FormatTok; 1990 if (!tryToParseLambdaIntroducer()) 1991 return false; 1992 1993 bool SeenArrow = false; 1994 bool InTemplateParameterList = false; 1995 1996 while (FormatTok->isNot(tok::l_brace)) { 1997 if (FormatTok->isSimpleTypeSpecifier()) { 1998 nextToken(); 1999 continue; 2000 } 2001 switch (FormatTok->Tok.getKind()) { 2002 case tok::l_brace: 2003 break; 2004 case tok::l_paren: 2005 parseParens(); 2006 break; 2007 case tok::l_square: 2008 parseSquare(); 2009 break; 2010 case tok::kw_class: 2011 case tok::kw_template: 2012 case tok::kw_typename: 2013 assert(FormatTok->Previous); 2014 if (FormatTok->Previous->is(tok::less)) 2015 InTemplateParameterList = true; 2016 nextToken(); 2017 break; 2018 case tok::amp: 2019 case tok::star: 2020 case tok::kw_const: 2021 case tok::comma: 2022 case tok::less: 2023 case tok::greater: 2024 case tok::identifier: 2025 case tok::numeric_constant: 2026 case tok::coloncolon: 2027 case tok::kw_mutable: 2028 case tok::kw_noexcept: 2029 nextToken(); 2030 break; 2031 // Specialization of a template with an integer parameter can contain 2032 // arithmetic, logical, comparison and ternary operators. 2033 // 2034 // FIXME: This also accepts sequences of operators that are not in the scope 2035 // of a template argument list. 2036 // 2037 // In a C++ lambda a template type can only occur after an arrow. We use 2038 // this as an heuristic to distinguish between Objective-C expressions 2039 // followed by an `a->b` expression, such as: 2040 // ([obj func:arg] + a->b) 2041 // Otherwise the code below would parse as a lambda. 2042 // 2043 // FIXME: This heuristic is incorrect for C++20 generic lambdas with 2044 // explicit template lists: []<bool b = true && false>(U &&u){} 2045 case tok::plus: 2046 case tok::minus: 2047 case tok::exclaim: 2048 case tok::tilde: 2049 case tok::slash: 2050 case tok::percent: 2051 case tok::lessless: 2052 case tok::pipe: 2053 case tok::pipepipe: 2054 case tok::ampamp: 2055 case tok::caret: 2056 case tok::equalequal: 2057 case tok::exclaimequal: 2058 case tok::greaterequal: 2059 case tok::lessequal: 2060 case tok::question: 2061 case tok::colon: 2062 case tok::ellipsis: 2063 case tok::kw_true: 2064 case tok::kw_false: 2065 if (SeenArrow || InTemplateParameterList) { 2066 nextToken(); 2067 break; 2068 } 2069 return true; 2070 case tok::arrow: 2071 // This might or might not actually be a lambda arrow (this could be an 2072 // ObjC method invocation followed by a dereferencing arrow). We might 2073 // reset this back to TT_Unknown in TokenAnnotator. 2074 FormatTok->setFinalizedType(TT_LambdaArrow); 2075 SeenArrow = true; 2076 nextToken(); 2077 break; 2078 default: 2079 return true; 2080 } 2081 } 2082 FormatTok->setFinalizedType(TT_LambdaLBrace); 2083 LSquare.setFinalizedType(TT_LambdaLSquare); 2084 parseChildBlock(); 2085 return true; 2086 } 2087 2088 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 2089 const FormatToken *Previous = FormatTok->Previous; 2090 const FormatToken *LeftSquare = FormatTok; 2091 nextToken(); 2092 if (Previous && 2093 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 2094 tok::kw_delete, tok::l_square) || 2095 LeftSquare->isCppStructuredBinding(Style) || Previous->closesScope() || 2096 Previous->isSimpleTypeSpecifier())) { 2097 return false; 2098 } 2099 if (FormatTok->is(tok::l_square)) 2100 return false; 2101 if (FormatTok->is(tok::r_square)) { 2102 const FormatToken *Next = Tokens->peekNextToken(); 2103 if (Next->is(tok::greater)) 2104 return false; 2105 } 2106 parseSquare(/*LambdaIntroducer=*/true); 2107 return true; 2108 } 2109 2110 void UnwrappedLineParser::tryToParseJSFunction() { 2111 assert(FormatTok->is(Keywords.kw_function) || 2112 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)); 2113 if (FormatTok->is(Keywords.kw_async)) 2114 nextToken(); 2115 // Consume "function". 2116 nextToken(); 2117 2118 // Consume * (generator function). Treat it like C++'s overloaded operators. 2119 if (FormatTok->is(tok::star)) { 2120 FormatTok->setFinalizedType(TT_OverloadedOperator); 2121 nextToken(); 2122 } 2123 2124 // Consume function name. 2125 if (FormatTok->is(tok::identifier)) 2126 nextToken(); 2127 2128 if (FormatTok->isNot(tok::l_paren)) 2129 return; 2130 2131 // Parse formal parameter list. 2132 parseParens(); 2133 2134 if (FormatTok->is(tok::colon)) { 2135 // Parse a type definition. 2136 nextToken(); 2137 2138 // Eat the type declaration. For braced inline object types, balance braces, 2139 // otherwise just parse until finding an l_brace for the function body. 2140 if (FormatTok->is(tok::l_brace)) 2141 tryToParseBracedList(); 2142 else 2143 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 2144 nextToken(); 2145 } 2146 2147 if (FormatTok->is(tok::semi)) 2148 return; 2149 2150 parseChildBlock(); 2151 } 2152 2153 bool UnwrappedLineParser::tryToParseBracedList() { 2154 if (FormatTok->is(BK_Unknown)) 2155 calculateBraceTypes(); 2156 assert(FormatTok->isNot(BK_Unknown)); 2157 if (FormatTok->is(BK_Block)) 2158 return false; 2159 nextToken(); 2160 parseBracedList(); 2161 return true; 2162 } 2163 2164 bool UnwrappedLineParser::tryToParseChildBlock() { 2165 assert(Style.isJavaScript() || Style.isCSharp()); 2166 assert(FormatTok->is(TT_FatArrow)); 2167 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType TT_FatArrow. 2168 // They always start an expression or a child block if followed by a curly 2169 // brace. 2170 nextToken(); 2171 if (FormatTok->isNot(tok::l_brace)) 2172 return false; 2173 parseChildBlock(); 2174 return true; 2175 } 2176 2177 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons, 2178 bool IsEnum, 2179 tok::TokenKind ClosingBraceKind) { 2180 bool HasError = false; 2181 2182 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 2183 // replace this by using parseAssignmentExpression() inside. 2184 do { 2185 if (Style.isCSharp() && FormatTok->is(TT_FatArrow) && 2186 tryToParseChildBlock()) 2187 continue; 2188 if (Style.isJavaScript()) { 2189 if (FormatTok->is(Keywords.kw_function) || 2190 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) { 2191 tryToParseJSFunction(); 2192 continue; 2193 } 2194 if (FormatTok->is(tok::l_brace)) { 2195 // Could be a method inside of a braced list `{a() { return 1; }}`. 2196 if (tryToParseBracedList()) 2197 continue; 2198 parseChildBlock(); 2199 } 2200 } 2201 if (FormatTok->Tok.getKind() == ClosingBraceKind) { 2202 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 2203 addUnwrappedLine(); 2204 nextToken(); 2205 return !HasError; 2206 } 2207 switch (FormatTok->Tok.getKind()) { 2208 case tok::l_square: 2209 if (Style.isCSharp()) 2210 parseSquare(); 2211 else 2212 tryToParseLambda(); 2213 break; 2214 case tok::l_paren: 2215 parseParens(); 2216 // JavaScript can just have free standing methods and getters/setters in 2217 // object literals. Detect them by a "{" following ")". 2218 if (Style.isJavaScript()) { 2219 if (FormatTok->is(tok::l_brace)) 2220 parseChildBlock(); 2221 break; 2222 } 2223 break; 2224 case tok::l_brace: 2225 // Assume there are no blocks inside a braced init list apart 2226 // from the ones we explicitly parse out (like lambdas). 2227 FormatTok->setBlockKind(BK_BracedInit); 2228 nextToken(); 2229 parseBracedList(); 2230 break; 2231 case tok::less: 2232 if (Style.Language == FormatStyle::LK_Proto || 2233 ClosingBraceKind == tok::greater) { 2234 nextToken(); 2235 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 2236 /*ClosingBraceKind=*/tok::greater); 2237 } else { 2238 nextToken(); 2239 } 2240 break; 2241 case tok::semi: 2242 // JavaScript (or more precisely TypeScript) can have semicolons in braced 2243 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 2244 // used for error recovery if we have otherwise determined that this is 2245 // a braced list. 2246 if (Style.isJavaScript()) { 2247 nextToken(); 2248 break; 2249 } 2250 HasError = true; 2251 if (!ContinueOnSemicolons) 2252 return !HasError; 2253 nextToken(); 2254 break; 2255 case tok::comma: 2256 nextToken(); 2257 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 2258 addUnwrappedLine(); 2259 break; 2260 default: 2261 nextToken(); 2262 break; 2263 } 2264 } while (!eof()); 2265 return false; 2266 } 2267 2268 /// \brief Parses a pair of parentheses (and everything between them). 2269 /// \param AmpAmpTokenType If different than TT_Unknown sets this type for all 2270 /// double ampersands. This only counts for the current parens scope. 2271 void UnwrappedLineParser::parseParens(TokenType AmpAmpTokenType) { 2272 assert(FormatTok->is(tok::l_paren) && "'(' expected."); 2273 nextToken(); 2274 do { 2275 switch (FormatTok->Tok.getKind()) { 2276 case tok::l_paren: 2277 parseParens(); 2278 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 2279 parseChildBlock(); 2280 break; 2281 case tok::r_paren: 2282 nextToken(); 2283 return; 2284 case tok::r_brace: 2285 // A "}" inside parenthesis is an error if there wasn't a matching "{". 2286 return; 2287 case tok::l_square: 2288 tryToParseLambda(); 2289 break; 2290 case tok::l_brace: 2291 if (!tryToParseBracedList()) 2292 parseChildBlock(); 2293 break; 2294 case tok::at: 2295 nextToken(); 2296 if (FormatTok->is(tok::l_brace)) { 2297 nextToken(); 2298 parseBracedList(); 2299 } 2300 break; 2301 case tok::equal: 2302 if (Style.isCSharp() && FormatTok->is(TT_FatArrow)) 2303 tryToParseChildBlock(); 2304 else 2305 nextToken(); 2306 break; 2307 case tok::kw_class: 2308 if (Style.isJavaScript()) 2309 parseRecord(/*ParseAsExpr=*/true); 2310 else 2311 nextToken(); 2312 break; 2313 case tok::identifier: 2314 if (Style.isJavaScript() && 2315 (FormatTok->is(Keywords.kw_function) || 2316 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function))) 2317 tryToParseJSFunction(); 2318 else 2319 nextToken(); 2320 break; 2321 case tok::kw_requires: { 2322 auto RequiresToken = FormatTok; 2323 nextToken(); 2324 parseRequiresExpression(RequiresToken); 2325 break; 2326 } 2327 case tok::ampamp: 2328 if (AmpAmpTokenType != TT_Unknown) 2329 FormatTok->setFinalizedType(AmpAmpTokenType); 2330 LLVM_FALLTHROUGH; 2331 default: 2332 nextToken(); 2333 break; 2334 } 2335 } while (!eof()); 2336 } 2337 2338 void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) { 2339 if (!LambdaIntroducer) { 2340 assert(FormatTok->is(tok::l_square) && "'[' expected."); 2341 if (tryToParseLambda()) 2342 return; 2343 } 2344 do { 2345 switch (FormatTok->Tok.getKind()) { 2346 case tok::l_paren: 2347 parseParens(); 2348 break; 2349 case tok::r_square: 2350 nextToken(); 2351 return; 2352 case tok::r_brace: 2353 // A "}" inside parenthesis is an error if there wasn't a matching "{". 2354 return; 2355 case tok::l_square: 2356 parseSquare(); 2357 break; 2358 case tok::l_brace: { 2359 if (!tryToParseBracedList()) 2360 parseChildBlock(); 2361 break; 2362 } 2363 case tok::at: 2364 nextToken(); 2365 if (FormatTok->is(tok::l_brace)) { 2366 nextToken(); 2367 parseBracedList(); 2368 } 2369 break; 2370 default: 2371 nextToken(); 2372 break; 2373 } 2374 } while (!eof()); 2375 } 2376 2377 void UnwrappedLineParser::keepAncestorBraces() { 2378 if (!Style.RemoveBracesLLVM) 2379 return; 2380 2381 const int MaxNestingLevels = 2; 2382 const int Size = NestedTooDeep.size(); 2383 if (Size >= MaxNestingLevels) 2384 NestedTooDeep[Size - MaxNestingLevels] = true; 2385 NestedTooDeep.push_back(false); 2386 } 2387 2388 static FormatToken *getLastNonComment(const UnwrappedLine &Line) { 2389 for (const auto &Token : llvm::reverse(Line.Tokens)) 2390 if (Token.Tok->isNot(tok::comment)) 2391 return Token.Tok; 2392 2393 return nullptr; 2394 } 2395 2396 void UnwrappedLineParser::parseUnbracedBody(bool CheckEOF) { 2397 FormatToken *Tok = nullptr; 2398 2399 if (Style.InsertBraces && !Line->InPPDirective && !Line->Tokens.empty() && 2400 PreprocessorDirectives.empty()) { 2401 Tok = getLastNonComment(*Line); 2402 assert(Tok); 2403 if (Tok->BraceCount < 0) { 2404 assert(Tok->BraceCount == -1); 2405 Tok = nullptr; 2406 } else { 2407 Tok->BraceCount = -1; 2408 } 2409 } 2410 2411 addUnwrappedLine(); 2412 ++Line->Level; 2413 parseStructuralElement(); 2414 2415 if (Tok) { 2416 assert(!Line->InPPDirective); 2417 Tok = nullptr; 2418 for (const auto &L : llvm::reverse(*CurrentLines)) { 2419 if (!L.InPPDirective && getLastNonComment(L)) { 2420 Tok = L.Tokens.back().Tok; 2421 break; 2422 } 2423 } 2424 assert(Tok); 2425 ++Tok->BraceCount; 2426 } 2427 2428 if (CheckEOF && FormatTok->is(tok::eof)) 2429 addUnwrappedLine(); 2430 2431 --Line->Level; 2432 } 2433 2434 static void markOptionalBraces(FormatToken *LeftBrace) { 2435 if (!LeftBrace) 2436 return; 2437 2438 assert(LeftBrace->is(tok::l_brace)); 2439 2440 FormatToken *RightBrace = LeftBrace->MatchingParen; 2441 if (!RightBrace) { 2442 assert(!LeftBrace->Optional); 2443 return; 2444 } 2445 2446 assert(RightBrace->is(tok::r_brace)); 2447 assert(RightBrace->MatchingParen == LeftBrace); 2448 assert(LeftBrace->Optional == RightBrace->Optional); 2449 2450 LeftBrace->Optional = true; 2451 RightBrace->Optional = true; 2452 } 2453 2454 void UnwrappedLineParser::handleAttributes() { 2455 // Handle AttributeMacro, e.g. `if (x) UNLIKELY`. 2456 if (FormatTok->is(TT_AttributeMacro)) 2457 nextToken(); 2458 handleCppAttributes(); 2459 } 2460 2461 bool UnwrappedLineParser::handleCppAttributes() { 2462 // Handle [[likely]] / [[unlikely]] attributes. 2463 if (FormatTok->is(tok::l_square) && tryToParseSimpleAttribute()) { 2464 parseSquare(); 2465 return true; 2466 } 2467 return false; 2468 } 2469 2470 FormatToken *UnwrappedLineParser::parseIfThenElse(IfStmtKind *IfKind, 2471 bool KeepBraces) { 2472 assert(FormatTok->is(tok::kw_if) && "'if' expected"); 2473 nextToken(); 2474 if (FormatTok->is(tok::exclaim)) 2475 nextToken(); 2476 2477 bool KeepIfBraces = false; 2478 bool KeepElseBraces = false; 2479 if (FormatTok->is(tok::kw_consteval)) { 2480 KeepIfBraces = true; 2481 KeepElseBraces = true; 2482 nextToken(); 2483 } else { 2484 if (FormatTok->isOneOf(tok::kw_constexpr, tok::identifier)) 2485 nextToken(); 2486 if (FormatTok->is(tok::l_paren)) 2487 parseParens(); 2488 } 2489 handleAttributes(); 2490 2491 bool NeedsUnwrappedLine = false; 2492 keepAncestorBraces(); 2493 2494 FormatToken *IfLeftBrace = nullptr; 2495 IfStmtKind IfBlockKind = IfStmtKind::NotIf; 2496 2497 if (FormatTok->is(tok::l_brace)) { 2498 IfLeftBrace = FormatTok; 2499 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2500 IfBlockKind = parseBlock(); 2501 if (Style.BraceWrapping.BeforeElse) 2502 addUnwrappedLine(); 2503 else 2504 NeedsUnwrappedLine = true; 2505 } else { 2506 parseUnbracedBody(); 2507 } 2508 2509 if (Style.RemoveBracesLLVM) { 2510 assert(!NestedTooDeep.empty()); 2511 KeepIfBraces = KeepIfBraces || 2512 (IfLeftBrace && !IfLeftBrace->MatchingParen) || 2513 NestedTooDeep.back() || IfBlockKind == IfStmtKind::IfOnly || 2514 IfBlockKind == IfStmtKind::IfElseIf; 2515 } 2516 2517 FormatToken *ElseLeftBrace = nullptr; 2518 IfStmtKind Kind = IfStmtKind::IfOnly; 2519 2520 if (FormatTok->is(tok::kw_else)) { 2521 if (Style.RemoveBracesLLVM) { 2522 NestedTooDeep.back() = false; 2523 Kind = IfStmtKind::IfElse; 2524 } 2525 nextToken(); 2526 handleAttributes(); 2527 if (FormatTok->is(tok::l_brace)) { 2528 ElseLeftBrace = FormatTok; 2529 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2530 if (parseBlock() == IfStmtKind::IfOnly) 2531 Kind = IfStmtKind::IfElseIf; 2532 addUnwrappedLine(); 2533 } else if (FormatTok->is(tok::kw_if)) { 2534 const FormatToken *Previous = Tokens->getPreviousToken(); 2535 assert(Previous); 2536 const bool IsPrecededByComment = Previous->is(tok::comment); 2537 if (IsPrecededByComment) { 2538 addUnwrappedLine(); 2539 ++Line->Level; 2540 } 2541 bool TooDeep = true; 2542 if (Style.RemoveBracesLLVM) { 2543 Kind = IfStmtKind::IfElseIf; 2544 TooDeep = NestedTooDeep.pop_back_val(); 2545 } 2546 ElseLeftBrace = 2547 parseIfThenElse(/*IfKind=*/nullptr, KeepBraces || KeepIfBraces); 2548 if (Style.RemoveBracesLLVM) 2549 NestedTooDeep.push_back(TooDeep); 2550 if (IsPrecededByComment) 2551 --Line->Level; 2552 } else { 2553 parseUnbracedBody(/*CheckEOF=*/true); 2554 } 2555 } else { 2556 if (Style.RemoveBracesLLVM) 2557 KeepIfBraces = KeepIfBraces || IfBlockKind == IfStmtKind::IfElse; 2558 if (NeedsUnwrappedLine) 2559 addUnwrappedLine(); 2560 } 2561 2562 if (!Style.RemoveBracesLLVM) 2563 return nullptr; 2564 2565 assert(!NestedTooDeep.empty()); 2566 KeepElseBraces = KeepElseBraces || 2567 (ElseLeftBrace && !ElseLeftBrace->MatchingParen) || 2568 NestedTooDeep.back(); 2569 2570 NestedTooDeep.pop_back(); 2571 2572 if (!KeepBraces && !KeepIfBraces && !KeepElseBraces) { 2573 markOptionalBraces(IfLeftBrace); 2574 markOptionalBraces(ElseLeftBrace); 2575 } else if (IfLeftBrace) { 2576 FormatToken *IfRightBrace = IfLeftBrace->MatchingParen; 2577 if (IfRightBrace) { 2578 assert(IfRightBrace->MatchingParen == IfLeftBrace); 2579 assert(!IfLeftBrace->Optional); 2580 assert(!IfRightBrace->Optional); 2581 IfLeftBrace->MatchingParen = nullptr; 2582 IfRightBrace->MatchingParen = nullptr; 2583 } 2584 } 2585 2586 if (IfKind) 2587 *IfKind = Kind; 2588 2589 return IfLeftBrace; 2590 } 2591 2592 void UnwrappedLineParser::parseTryCatch() { 2593 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 2594 nextToken(); 2595 bool NeedsUnwrappedLine = false; 2596 if (FormatTok->is(tok::colon)) { 2597 // We are in a function try block, what comes is an initializer list. 2598 nextToken(); 2599 2600 // In case identifiers were removed by clang-tidy, what might follow is 2601 // multiple commas in sequence - before the first identifier. 2602 while (FormatTok->is(tok::comma)) 2603 nextToken(); 2604 2605 while (FormatTok->is(tok::identifier)) { 2606 nextToken(); 2607 if (FormatTok->is(tok::l_paren)) 2608 parseParens(); 2609 if (FormatTok->Previous && FormatTok->Previous->is(tok::identifier) && 2610 FormatTok->is(tok::l_brace)) { 2611 do { 2612 nextToken(); 2613 } while (!FormatTok->is(tok::r_brace)); 2614 nextToken(); 2615 } 2616 2617 // In case identifiers were removed by clang-tidy, what might follow is 2618 // multiple commas in sequence - after the first identifier. 2619 while (FormatTok->is(tok::comma)) 2620 nextToken(); 2621 } 2622 } 2623 // Parse try with resource. 2624 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) 2625 parseParens(); 2626 2627 keepAncestorBraces(); 2628 2629 if (FormatTok->is(tok::l_brace)) { 2630 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2631 parseBlock(); 2632 if (Style.BraceWrapping.BeforeCatch) 2633 addUnwrappedLine(); 2634 else 2635 NeedsUnwrappedLine = true; 2636 } else if (!FormatTok->is(tok::kw_catch)) { 2637 // The C++ standard requires a compound-statement after a try. 2638 // If there's none, we try to assume there's a structuralElement 2639 // and try to continue. 2640 addUnwrappedLine(); 2641 ++Line->Level; 2642 parseStructuralElement(); 2643 --Line->Level; 2644 } 2645 while (true) { 2646 if (FormatTok->is(tok::at)) 2647 nextToken(); 2648 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 2649 tok::kw___finally) || 2650 ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2651 FormatTok->is(Keywords.kw_finally)) || 2652 (FormatTok->isObjCAtKeyword(tok::objc_catch) || 2653 FormatTok->isObjCAtKeyword(tok::objc_finally)))) 2654 break; 2655 nextToken(); 2656 while (FormatTok->isNot(tok::l_brace)) { 2657 if (FormatTok->is(tok::l_paren)) { 2658 parseParens(); 2659 continue; 2660 } 2661 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) { 2662 if (Style.RemoveBracesLLVM) 2663 NestedTooDeep.pop_back(); 2664 return; 2665 } 2666 nextToken(); 2667 } 2668 NeedsUnwrappedLine = false; 2669 Line->MustBeDeclaration = false; 2670 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2671 parseBlock(); 2672 if (Style.BraceWrapping.BeforeCatch) 2673 addUnwrappedLine(); 2674 else 2675 NeedsUnwrappedLine = true; 2676 } 2677 2678 if (Style.RemoveBracesLLVM) 2679 NestedTooDeep.pop_back(); 2680 2681 if (NeedsUnwrappedLine) 2682 addUnwrappedLine(); 2683 } 2684 2685 void UnwrappedLineParser::parseNamespace() { 2686 assert(FormatTok->isOneOf(tok::kw_namespace, TT_NamespaceMacro) && 2687 "'namespace' expected"); 2688 2689 const FormatToken &InitialToken = *FormatTok; 2690 nextToken(); 2691 if (InitialToken.is(TT_NamespaceMacro)) { 2692 parseParens(); 2693 } else { 2694 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::kw_inline, 2695 tok::l_square, tok::period, tok::l_paren) || 2696 (Style.isCSharp() && FormatTok->is(tok::kw_union))) 2697 if (FormatTok->is(tok::l_square)) 2698 parseSquare(); 2699 else if (FormatTok->is(tok::l_paren)) 2700 parseParens(); 2701 else 2702 nextToken(); 2703 } 2704 if (FormatTok->is(tok::l_brace)) { 2705 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2706 addUnwrappedLine(); 2707 2708 unsigned AddLevels = 2709 Style.NamespaceIndentation == FormatStyle::NI_All || 2710 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 2711 DeclarationScopeStack.size() > 1) 2712 ? 1u 2713 : 0u; 2714 bool ManageWhitesmithsBraces = 2715 AddLevels == 0u && 2716 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths; 2717 2718 // If we're in Whitesmiths mode, indent the brace if we're not indenting 2719 // the whole block. 2720 if (ManageWhitesmithsBraces) 2721 ++Line->Level; 2722 2723 parseBlock(/*MustBeDeclaration=*/true, AddLevels, 2724 /*MunchSemi=*/true, 2725 /*UnindentWhitesmithsBraces=*/ManageWhitesmithsBraces); 2726 2727 // Munch the semicolon after a namespace. This is more common than one would 2728 // think. Putting the semicolon into its own line is very ugly. 2729 if (FormatTok->is(tok::semi)) 2730 nextToken(); 2731 2732 addUnwrappedLine(AddLevels > 0 ? LineLevel::Remove : LineLevel::Keep); 2733 2734 if (ManageWhitesmithsBraces) 2735 --Line->Level; 2736 } 2737 // FIXME: Add error handling. 2738 } 2739 2740 void UnwrappedLineParser::parseNew() { 2741 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 2742 nextToken(); 2743 2744 if (Style.isCSharp()) { 2745 do { 2746 if (FormatTok->is(tok::l_brace)) 2747 parseBracedList(); 2748 2749 if (FormatTok->isOneOf(tok::semi, tok::comma)) 2750 return; 2751 2752 nextToken(); 2753 } while (!eof()); 2754 } 2755 2756 if (Style.Language != FormatStyle::LK_Java) 2757 return; 2758 2759 // In Java, we can parse everything up to the parens, which aren't optional. 2760 do { 2761 // There should not be a ;, { or } before the new's open paren. 2762 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 2763 return; 2764 2765 // Consume the parens. 2766 if (FormatTok->is(tok::l_paren)) { 2767 parseParens(); 2768 2769 // If there is a class body of an anonymous class, consume that as child. 2770 if (FormatTok->is(tok::l_brace)) 2771 parseChildBlock(); 2772 return; 2773 } 2774 nextToken(); 2775 } while (!eof()); 2776 } 2777 2778 void UnwrappedLineParser::parseLoopBody(bool TryRemoveBraces, 2779 bool WrapRightBrace) { 2780 keepAncestorBraces(); 2781 2782 if (FormatTok->is(tok::l_brace)) { 2783 FormatToken *LeftBrace = FormatTok; 2784 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2785 parseBlock(); 2786 if (TryRemoveBraces) { 2787 assert(!NestedTooDeep.empty()); 2788 if (!NestedTooDeep.back()) 2789 markOptionalBraces(LeftBrace); 2790 } 2791 if (WrapRightBrace) 2792 addUnwrappedLine(); 2793 } else { 2794 parseUnbracedBody(); 2795 } 2796 2797 if (TryRemoveBraces) 2798 NestedTooDeep.pop_back(); 2799 } 2800 2801 void UnwrappedLineParser::parseForOrWhileLoop() { 2802 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 2803 "'for', 'while' or foreach macro expected"); 2804 nextToken(); 2805 // JS' for await ( ... 2806 if (Style.isJavaScript() && FormatTok->is(Keywords.kw_await)) 2807 nextToken(); 2808 if (Style.isCpp() && FormatTok->is(tok::kw_co_await)) 2809 nextToken(); 2810 if (FormatTok->is(tok::l_paren)) 2811 parseParens(); 2812 2813 parseLoopBody(Style.RemoveBracesLLVM, true); 2814 } 2815 2816 void UnwrappedLineParser::parseDoWhile() { 2817 assert(FormatTok->is(tok::kw_do) && "'do' expected"); 2818 nextToken(); 2819 2820 parseLoopBody(false, Style.BraceWrapping.BeforeWhile); 2821 2822 // FIXME: Add error handling. 2823 if (!FormatTok->is(tok::kw_while)) { 2824 addUnwrappedLine(); 2825 return; 2826 } 2827 2828 // If in Whitesmiths mode, the line with the while() needs to be indented 2829 // to the same level as the block. 2830 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) 2831 ++Line->Level; 2832 2833 nextToken(); 2834 parseStructuralElement(); 2835 } 2836 2837 void UnwrappedLineParser::parseLabel(bool LeftAlignLabel) { 2838 nextToken(); 2839 unsigned OldLineLevel = Line->Level; 2840 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 2841 --Line->Level; 2842 if (LeftAlignLabel) 2843 Line->Level = 0; 2844 2845 if (!Style.IndentCaseBlocks && CommentsBeforeNextToken.empty() && 2846 FormatTok->is(tok::l_brace)) { 2847 2848 CompoundStatementIndenter Indenter(this, Line->Level, 2849 Style.BraceWrapping.AfterCaseLabel, 2850 Style.BraceWrapping.IndentBraces); 2851 parseBlock(); 2852 if (FormatTok->is(tok::kw_break)) { 2853 if (Style.BraceWrapping.AfterControlStatement == 2854 FormatStyle::BWACS_Always) { 2855 addUnwrappedLine(); 2856 if (!Style.IndentCaseBlocks && 2857 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) 2858 ++Line->Level; 2859 } 2860 parseStructuralElement(); 2861 } 2862 addUnwrappedLine(); 2863 } else { 2864 if (FormatTok->is(tok::semi)) 2865 nextToken(); 2866 addUnwrappedLine(); 2867 } 2868 Line->Level = OldLineLevel; 2869 if (FormatTok->isNot(tok::l_brace)) { 2870 parseStructuralElement(); 2871 addUnwrappedLine(); 2872 } 2873 } 2874 2875 void UnwrappedLineParser::parseCaseLabel() { 2876 assert(FormatTok->is(tok::kw_case) && "'case' expected"); 2877 2878 // FIXME: fix handling of complex expressions here. 2879 do { 2880 nextToken(); 2881 } while (!eof() && !FormatTok->is(tok::colon)); 2882 parseLabel(); 2883 } 2884 2885 void UnwrappedLineParser::parseSwitch() { 2886 assert(FormatTok->is(tok::kw_switch) && "'switch' expected"); 2887 nextToken(); 2888 if (FormatTok->is(tok::l_paren)) 2889 parseParens(); 2890 2891 keepAncestorBraces(); 2892 2893 if (FormatTok->is(tok::l_brace)) { 2894 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2895 parseBlock(); 2896 addUnwrappedLine(); 2897 } else { 2898 addUnwrappedLine(); 2899 ++Line->Level; 2900 parseStructuralElement(); 2901 --Line->Level; 2902 } 2903 2904 if (Style.RemoveBracesLLVM) 2905 NestedTooDeep.pop_back(); 2906 } 2907 2908 // Operators that can follow a C variable. 2909 static bool isCOperatorFollowingVar(tok::TokenKind kind) { 2910 switch (kind) { 2911 case tok::ampamp: 2912 case tok::ampequal: 2913 case tok::arrow: 2914 case tok::caret: 2915 case tok::caretequal: 2916 case tok::comma: 2917 case tok::ellipsis: 2918 case tok::equal: 2919 case tok::equalequal: 2920 case tok::exclaim: 2921 case tok::exclaimequal: 2922 case tok::greater: 2923 case tok::greaterequal: 2924 case tok::greatergreater: 2925 case tok::greatergreaterequal: 2926 case tok::l_paren: 2927 case tok::l_square: 2928 case tok::less: 2929 case tok::lessequal: 2930 case tok::lessless: 2931 case tok::lesslessequal: 2932 case tok::minus: 2933 case tok::minusequal: 2934 case tok::minusminus: 2935 case tok::percent: 2936 case tok::percentequal: 2937 case tok::period: 2938 case tok::pipe: 2939 case tok::pipeequal: 2940 case tok::pipepipe: 2941 case tok::plus: 2942 case tok::plusequal: 2943 case tok::plusplus: 2944 case tok::question: 2945 case tok::r_brace: 2946 case tok::r_paren: 2947 case tok::r_square: 2948 case tok::semi: 2949 case tok::slash: 2950 case tok::slashequal: 2951 case tok::star: 2952 case tok::starequal: 2953 return true; 2954 default: 2955 return false; 2956 } 2957 } 2958 2959 void UnwrappedLineParser::parseAccessSpecifier() { 2960 FormatToken *AccessSpecifierCandidate = FormatTok; 2961 nextToken(); 2962 // Understand Qt's slots. 2963 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 2964 nextToken(); 2965 // Otherwise, we don't know what it is, and we'd better keep the next token. 2966 if (FormatTok->is(tok::colon)) { 2967 nextToken(); 2968 addUnwrappedLine(); 2969 } else if (!FormatTok->is(tok::coloncolon) && 2970 !isCOperatorFollowingVar(FormatTok->Tok.getKind())) { 2971 // Not a variable name nor namespace name. 2972 addUnwrappedLine(); 2973 } else if (AccessSpecifierCandidate) { 2974 // Consider the access specifier to be a C identifier. 2975 AccessSpecifierCandidate->Tok.setKind(tok::identifier); 2976 } 2977 } 2978 2979 /// \brief Parses a concept definition. 2980 /// \pre The current token has to be the concept keyword. 2981 /// 2982 /// Returns if either the concept has been completely parsed, or if it detects 2983 /// that the concept definition is incorrect. 2984 void UnwrappedLineParser::parseConcept() { 2985 assert(FormatTok->is(tok::kw_concept) && "'concept' expected"); 2986 nextToken(); 2987 if (!FormatTok->is(tok::identifier)) 2988 return; 2989 nextToken(); 2990 if (!FormatTok->is(tok::equal)) 2991 return; 2992 nextToken(); 2993 parseConstraintExpression(); 2994 if (FormatTok->is(tok::semi)) 2995 nextToken(); 2996 addUnwrappedLine(); 2997 } 2998 2999 /// \brief Parses a requires, decides if it is a clause or an expression. 3000 /// \pre The current token has to be the requires keyword. 3001 /// \returns true if it parsed a clause. 3002 bool clang::format::UnwrappedLineParser::parseRequires() { 3003 assert(FormatTok->is(tok::kw_requires) && "'requires' expected"); 3004 auto RequiresToken = FormatTok; 3005 3006 // We try to guess if it is a requires clause, or a requires expression. For 3007 // that we first consume the keyword and check the next token. 3008 nextToken(); 3009 3010 switch (FormatTok->Tok.getKind()) { 3011 case tok::l_brace: 3012 // This can only be an expression, never a clause. 3013 parseRequiresExpression(RequiresToken); 3014 return false; 3015 case tok::l_paren: 3016 // Clauses and expression can start with a paren, it's unclear what we have. 3017 break; 3018 default: 3019 // All other tokens can only be a clause. 3020 parseRequiresClause(RequiresToken); 3021 return true; 3022 } 3023 3024 // Looking forward we would have to decide if there are function declaration 3025 // like arguments to the requires expression: 3026 // requires (T t) { 3027 // Or there is a constraint expression for the requires clause: 3028 // requires (C<T> && ... 3029 3030 // But first let's look behind. 3031 auto *PreviousNonComment = RequiresToken->getPreviousNonComment(); 3032 3033 if (!PreviousNonComment || 3034 PreviousNonComment->is(TT_RequiresExpressionLBrace)) { 3035 // If there is no token, or an expression left brace, we are a requires 3036 // clause within a requires expression. 3037 parseRequiresClause(RequiresToken); 3038 return true; 3039 } 3040 3041 switch (PreviousNonComment->Tok.getKind()) { 3042 case tok::greater: 3043 case tok::r_paren: 3044 case tok::kw_noexcept: 3045 case tok::kw_const: 3046 // This is a requires clause. 3047 parseRequiresClause(RequiresToken); 3048 return true; 3049 case tok::amp: 3050 case tok::ampamp: { 3051 // This can be either: 3052 // if (... && requires (T t) ...) 3053 // Or 3054 // void member(...) && requires (C<T> ... 3055 // We check the one token before that for a const: 3056 // void member(...) const && requires (C<T> ... 3057 auto PrevPrev = PreviousNonComment->getPreviousNonComment(); 3058 if (PrevPrev && PrevPrev->is(tok::kw_const)) { 3059 parseRequiresClause(RequiresToken); 3060 return true; 3061 } 3062 break; 3063 } 3064 default: 3065 // It's an expression. 3066 parseRequiresExpression(RequiresToken); 3067 return false; 3068 } 3069 3070 // Now we look forward and try to check if the paren content is a parameter 3071 // list. The parameters can be cv-qualified and contain references or 3072 // pointers. 3073 // So we want basically to check for TYPE NAME, but TYPE can contain all kinds 3074 // of stuff: typename, const, *, &, &&, ::, identifiers. 3075 3076 int NextTokenOffset = 1; 3077 auto NextToken = Tokens->peekNextToken(NextTokenOffset); 3078 auto PeekNext = [&NextTokenOffset, &NextToken, this] { 3079 ++NextTokenOffset; 3080 NextToken = Tokens->peekNextToken(NextTokenOffset); 3081 }; 3082 3083 bool FoundType = false; 3084 bool LastWasColonColon = false; 3085 int OpenAngles = 0; 3086 3087 for (; NextTokenOffset < 50; PeekNext()) { 3088 switch (NextToken->Tok.getKind()) { 3089 case tok::kw_volatile: 3090 case tok::kw_const: 3091 case tok::comma: 3092 parseRequiresExpression(RequiresToken); 3093 return false; 3094 case tok::r_paren: 3095 case tok::pipepipe: 3096 parseRequiresClause(RequiresToken); 3097 return true; 3098 case tok::eof: 3099 // Break out of the loop. 3100 NextTokenOffset = 50; 3101 break; 3102 case tok::coloncolon: 3103 LastWasColonColon = true; 3104 break; 3105 case tok::identifier: 3106 if (FoundType && !LastWasColonColon && OpenAngles == 0) { 3107 parseRequiresExpression(RequiresToken); 3108 return false; 3109 } 3110 FoundType = true; 3111 LastWasColonColon = false; 3112 break; 3113 case tok::less: 3114 ++OpenAngles; 3115 break; 3116 case tok::greater: 3117 --OpenAngles; 3118 break; 3119 default: 3120 if (NextToken->isSimpleTypeSpecifier()) { 3121 parseRequiresExpression(RequiresToken); 3122 return false; 3123 } 3124 break; 3125 } 3126 } 3127 3128 // This seems to be a complicated expression, just assume it's a clause. 3129 parseRequiresClause(RequiresToken); 3130 return true; 3131 } 3132 3133 /// \brief Parses a requires clause. 3134 /// \param RequiresToken The requires keyword token, which starts this clause. 3135 /// \pre We need to be on the next token after the requires keyword. 3136 /// \sa parseRequiresExpression 3137 /// 3138 /// Returns if it either has finished parsing the clause, or it detects, that 3139 /// the clause is incorrect. 3140 void UnwrappedLineParser::parseRequiresClause(FormatToken *RequiresToken) { 3141 assert(FormatTok->getPreviousNonComment() == RequiresToken); 3142 assert(RequiresToken->is(tok::kw_requires) && "'requires' expected"); 3143 3144 // If there is no previous token, we are within a requires expression, 3145 // otherwise we will always have the template or function declaration in front 3146 // of it. 3147 bool InRequiresExpression = 3148 !RequiresToken->Previous || 3149 RequiresToken->Previous->is(TT_RequiresExpressionLBrace); 3150 3151 RequiresToken->setFinalizedType(InRequiresExpression 3152 ? TT_RequiresClauseInARequiresExpression 3153 : TT_RequiresClause); 3154 3155 parseConstraintExpression(); 3156 3157 if (!InRequiresExpression) 3158 FormatTok->Previous->ClosesRequiresClause = true; 3159 } 3160 3161 /// \brief Parses a requires expression. 3162 /// \param RequiresToken The requires keyword token, which starts this clause. 3163 /// \pre We need to be on the next token after the requires keyword. 3164 /// \sa parseRequiresClause 3165 /// 3166 /// Returns if it either has finished parsing the expression, or it detects, 3167 /// that the expression is incorrect. 3168 void UnwrappedLineParser::parseRequiresExpression(FormatToken *RequiresToken) { 3169 assert(FormatTok->getPreviousNonComment() == RequiresToken); 3170 assert(RequiresToken->is(tok::kw_requires) && "'requires' expected"); 3171 3172 RequiresToken->setFinalizedType(TT_RequiresExpression); 3173 3174 if (FormatTok->is(tok::l_paren)) { 3175 FormatTok->setFinalizedType(TT_RequiresExpressionLParen); 3176 parseParens(); 3177 } 3178 3179 if (FormatTok->is(tok::l_brace)) { 3180 FormatTok->setFinalizedType(TT_RequiresExpressionLBrace); 3181 parseChildBlock(/*CanContainBracedList=*/false, 3182 /*NextLBracesType=*/TT_CompoundRequirementLBrace); 3183 } 3184 } 3185 3186 /// \brief Parses a constraint expression. 3187 /// 3188 /// This is either the definition of a concept, or the body of a requires 3189 /// clause. It returns, when the parsing is complete, or the expression is 3190 /// incorrect. 3191 void UnwrappedLineParser::parseConstraintExpression() { 3192 // The special handling for lambdas is needed since tryToParseLambda() eats a 3193 // token and if a requires expression is the last part of a requires clause 3194 // and followed by an attribute like [[nodiscard]] the ClosesRequiresClause is 3195 // not set on the correct token. Thus we need to be aware if we even expect a 3196 // lambda to be possible. 3197 // template <typename T> requires requires { ... } [[nodiscard]] ...; 3198 bool LambdaNextTimeAllowed = true; 3199 do { 3200 bool LambdaThisTimeAllowed = std::exchange(LambdaNextTimeAllowed, false); 3201 3202 switch (FormatTok->Tok.getKind()) { 3203 case tok::kw_requires: { 3204 auto RequiresToken = FormatTok; 3205 nextToken(); 3206 parseRequiresExpression(RequiresToken); 3207 break; 3208 } 3209 3210 case tok::l_paren: 3211 parseParens(/*AmpAmpTokenType=*/TT_BinaryOperator); 3212 break; 3213 3214 case tok::l_square: 3215 if (!LambdaThisTimeAllowed || !tryToParseLambda()) 3216 return; 3217 break; 3218 3219 case tok::kw_const: 3220 case tok::semi: 3221 case tok::kw_class: 3222 case tok::kw_struct: 3223 case tok::kw_union: 3224 return; 3225 3226 case tok::l_brace: 3227 // Potential function body. 3228 return; 3229 3230 case tok::ampamp: 3231 case tok::pipepipe: 3232 FormatTok->setFinalizedType(TT_BinaryOperator); 3233 nextToken(); 3234 LambdaNextTimeAllowed = true; 3235 break; 3236 3237 case tok::comma: 3238 case tok::comment: 3239 LambdaNextTimeAllowed = LambdaThisTimeAllowed; 3240 nextToken(); 3241 break; 3242 3243 case tok::kw_sizeof: 3244 case tok::greater: 3245 case tok::greaterequal: 3246 case tok::greatergreater: 3247 case tok::less: 3248 case tok::lessequal: 3249 case tok::lessless: 3250 case tok::equalequal: 3251 case tok::exclaim: 3252 case tok::exclaimequal: 3253 case tok::plus: 3254 case tok::minus: 3255 case tok::star: 3256 case tok::slash: 3257 case tok::kw_decltype: 3258 LambdaNextTimeAllowed = true; 3259 // Just eat them. 3260 nextToken(); 3261 break; 3262 3263 case tok::numeric_constant: 3264 case tok::coloncolon: 3265 case tok::kw_true: 3266 case tok::kw_false: 3267 // Just eat them. 3268 nextToken(); 3269 break; 3270 3271 case tok::kw_static_cast: 3272 case tok::kw_const_cast: 3273 case tok::kw_reinterpret_cast: 3274 case tok::kw_dynamic_cast: 3275 nextToken(); 3276 if (!FormatTok->is(tok::less)) 3277 return; 3278 3279 nextToken(); 3280 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 3281 /*ClosingBraceKind=*/tok::greater); 3282 break; 3283 3284 case tok::kw_bool: 3285 // bool is only allowed if it is directly followed by a paren for a cast: 3286 // concept C = bool(...); 3287 // and bool is the only type, all other types as cast must be inside a 3288 // cast to bool an thus are handled by the other cases. 3289 nextToken(); 3290 if (FormatTok->isNot(tok::l_paren)) 3291 return; 3292 parseParens(); 3293 break; 3294 3295 default: 3296 if (!FormatTok->Tok.getIdentifierInfo()) { 3297 // Identifiers are part of the default case, we check for more then 3298 // tok::identifier to handle builtin type traits. 3299 return; 3300 } 3301 3302 // We need to differentiate identifiers for a template deduction guide, 3303 // variables, or function return types (the constraint expression has 3304 // ended before that), and basically all other cases. But it's easier to 3305 // check the other way around. 3306 assert(FormatTok->Previous); 3307 switch (FormatTok->Previous->Tok.getKind()) { 3308 case tok::coloncolon: // Nested identifier. 3309 case tok::ampamp: // Start of a function or variable for the 3310 case tok::pipepipe: // constraint expression. 3311 case tok::kw_requires: // Initial identifier of a requires clause. 3312 case tok::equal: // Initial identifier of a concept declaration. 3313 break; 3314 default: 3315 return; 3316 } 3317 3318 // Read identifier with optional template declaration. 3319 nextToken(); 3320 if (FormatTok->is(tok::less)) { 3321 nextToken(); 3322 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 3323 /*ClosingBraceKind=*/tok::greater); 3324 } 3325 break; 3326 } 3327 } while (!eof()); 3328 } 3329 3330 bool UnwrappedLineParser::parseEnum() { 3331 const FormatToken &InitialToken = *FormatTok; 3332 3333 // Won't be 'enum' for NS_ENUMs. 3334 if (FormatTok->is(tok::kw_enum)) 3335 nextToken(); 3336 3337 // In TypeScript, "enum" can also be used as property name, e.g. in interface 3338 // declarations. An "enum" keyword followed by a colon would be a syntax 3339 // error and thus assume it is just an identifier. 3340 if (Style.isJavaScript() && FormatTok->isOneOf(tok::colon, tok::question)) 3341 return false; 3342 3343 // In protobuf, "enum" can be used as a field name. 3344 if (Style.Language == FormatStyle::LK_Proto && FormatTok->is(tok::equal)) 3345 return false; 3346 3347 // Eat up enum class ... 3348 if (FormatTok->isOneOf(tok::kw_class, tok::kw_struct)) 3349 nextToken(); 3350 3351 while (FormatTok->Tok.getIdentifierInfo() || 3352 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 3353 tok::greater, tok::comma, tok::question)) { 3354 nextToken(); 3355 // We can have macros or attributes in between 'enum' and the enum name. 3356 if (FormatTok->is(tok::l_paren)) 3357 parseParens(); 3358 if (FormatTok->is(tok::identifier)) { 3359 nextToken(); 3360 // If there are two identifiers in a row, this is likely an elaborate 3361 // return type. In Java, this can be "implements", etc. 3362 if (Style.isCpp() && FormatTok->is(tok::identifier)) 3363 return false; 3364 } 3365 } 3366 3367 // Just a declaration or something is wrong. 3368 if (FormatTok->isNot(tok::l_brace)) 3369 return true; 3370 FormatTok->setFinalizedType(TT_EnumLBrace); 3371 FormatTok->setBlockKind(BK_Block); 3372 3373 if (Style.Language == FormatStyle::LK_Java) { 3374 // Java enums are different. 3375 parseJavaEnumBody(); 3376 return true; 3377 } 3378 if (Style.Language == FormatStyle::LK_Proto) { 3379 parseBlock(/*MustBeDeclaration=*/true); 3380 return true; 3381 } 3382 3383 if (!Style.AllowShortEnumsOnASingleLine && 3384 ShouldBreakBeforeBrace(Style, InitialToken)) 3385 addUnwrappedLine(); 3386 // Parse enum body. 3387 nextToken(); 3388 if (!Style.AllowShortEnumsOnASingleLine) { 3389 addUnwrappedLine(); 3390 Line->Level += 1; 3391 } 3392 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true, 3393 /*IsEnum=*/true); 3394 if (!Style.AllowShortEnumsOnASingleLine) 3395 Line->Level -= 1; 3396 if (HasError) { 3397 if (FormatTok->is(tok::semi)) 3398 nextToken(); 3399 addUnwrappedLine(); 3400 } 3401 return true; 3402 3403 // There is no addUnwrappedLine() here so that we fall through to parsing a 3404 // structural element afterwards. Thus, in "enum A {} n, m;", 3405 // "} n, m;" will end up in one unwrapped line. 3406 } 3407 3408 bool UnwrappedLineParser::parseStructLike() { 3409 // parseRecord falls through and does not yet add an unwrapped line as a 3410 // record declaration or definition can start a structural element. 3411 parseRecord(); 3412 // This does not apply to Java, JavaScript and C#. 3413 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 3414 Style.isCSharp()) { 3415 if (FormatTok->is(tok::semi)) 3416 nextToken(); 3417 addUnwrappedLine(); 3418 return true; 3419 } 3420 return false; 3421 } 3422 3423 namespace { 3424 // A class used to set and restore the Token position when peeking 3425 // ahead in the token source. 3426 class ScopedTokenPosition { 3427 unsigned StoredPosition; 3428 FormatTokenSource *Tokens; 3429 3430 public: 3431 ScopedTokenPosition(FormatTokenSource *Tokens) : Tokens(Tokens) { 3432 assert(Tokens && "Tokens expected to not be null"); 3433 StoredPosition = Tokens->getPosition(); 3434 } 3435 3436 ~ScopedTokenPosition() { Tokens->setPosition(StoredPosition); } 3437 }; 3438 } // namespace 3439 3440 // Look to see if we have [[ by looking ahead, if 3441 // its not then rewind to the original position. 3442 bool UnwrappedLineParser::tryToParseSimpleAttribute() { 3443 ScopedTokenPosition AutoPosition(Tokens); 3444 FormatToken *Tok = Tokens->getNextToken(); 3445 // We already read the first [ check for the second. 3446 if (!Tok->is(tok::l_square)) 3447 return false; 3448 // Double check that the attribute is just something 3449 // fairly simple. 3450 while (Tok->isNot(tok::eof)) { 3451 if (Tok->is(tok::r_square)) 3452 break; 3453 Tok = Tokens->getNextToken(); 3454 } 3455 if (Tok->is(tok::eof)) 3456 return false; 3457 Tok = Tokens->getNextToken(); 3458 if (!Tok->is(tok::r_square)) 3459 return false; 3460 Tok = Tokens->getNextToken(); 3461 if (Tok->is(tok::semi)) 3462 return false; 3463 return true; 3464 } 3465 3466 void UnwrappedLineParser::parseJavaEnumBody() { 3467 // Determine whether the enum is simple, i.e. does not have a semicolon or 3468 // constants with class bodies. Simple enums can be formatted like braced 3469 // lists, contracted to a single line, etc. 3470 unsigned StoredPosition = Tokens->getPosition(); 3471 bool IsSimple = true; 3472 FormatToken *Tok = Tokens->getNextToken(); 3473 while (!Tok->is(tok::eof)) { 3474 if (Tok->is(tok::r_brace)) 3475 break; 3476 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 3477 IsSimple = false; 3478 break; 3479 } 3480 // FIXME: This will also mark enums with braces in the arguments to enum 3481 // constants as "not simple". This is probably fine in practice, though. 3482 Tok = Tokens->getNextToken(); 3483 } 3484 FormatTok = Tokens->setPosition(StoredPosition); 3485 3486 if (IsSimple) { 3487 nextToken(); 3488 parseBracedList(); 3489 addUnwrappedLine(); 3490 return; 3491 } 3492 3493 // Parse the body of a more complex enum. 3494 // First add a line for everything up to the "{". 3495 nextToken(); 3496 addUnwrappedLine(); 3497 ++Line->Level; 3498 3499 // Parse the enum constants. 3500 while (FormatTok) { 3501 if (FormatTok->is(tok::l_brace)) { 3502 // Parse the constant's class body. 3503 parseBlock(/*MustBeDeclaration=*/true, /*AddLevels=*/1u, 3504 /*MunchSemi=*/false); 3505 } else if (FormatTok->is(tok::l_paren)) { 3506 parseParens(); 3507 } else if (FormatTok->is(tok::comma)) { 3508 nextToken(); 3509 addUnwrappedLine(); 3510 } else if (FormatTok->is(tok::semi)) { 3511 nextToken(); 3512 addUnwrappedLine(); 3513 break; 3514 } else if (FormatTok->is(tok::r_brace)) { 3515 addUnwrappedLine(); 3516 break; 3517 } else { 3518 nextToken(); 3519 } 3520 } 3521 3522 // Parse the class body after the enum's ";" if any. 3523 parseLevel(/*HasOpeningBrace=*/true, /*CanContainBracedList=*/true); 3524 nextToken(); 3525 --Line->Level; 3526 addUnwrappedLine(); 3527 } 3528 3529 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) { 3530 const FormatToken &InitialToken = *FormatTok; 3531 nextToken(); 3532 3533 // The actual identifier can be a nested name specifier, and in macros 3534 // it is often token-pasted. 3535 // An [[attribute]] can be before the identifier. 3536 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 3537 tok::kw___attribute, tok::kw___declspec, 3538 tok::kw_alignas, tok::l_square, tok::r_square) || 3539 ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 3540 FormatTok->isOneOf(tok::period, tok::comma))) { 3541 if (Style.isJavaScript() && 3542 FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) { 3543 // JavaScript/TypeScript supports inline object types in 3544 // extends/implements positions: 3545 // class Foo implements {bar: number} { } 3546 nextToken(); 3547 if (FormatTok->is(tok::l_brace)) { 3548 tryToParseBracedList(); 3549 continue; 3550 } 3551 } 3552 bool IsNonMacroIdentifier = 3553 FormatTok->is(tok::identifier) && 3554 FormatTok->TokenText != FormatTok->TokenText.upper(); 3555 nextToken(); 3556 // We can have macros or attributes in between 'class' and the class name. 3557 if (!IsNonMacroIdentifier) { 3558 if (FormatTok->is(tok::l_paren)) { 3559 parseParens(); 3560 } else if (FormatTok->is(TT_AttributeSquare)) { 3561 parseSquare(); 3562 // Consume the closing TT_AttributeSquare. 3563 if (FormatTok->Next && FormatTok->is(TT_AttributeSquare)) 3564 nextToken(); 3565 } 3566 } 3567 } 3568 3569 // Note that parsing away template declarations here leads to incorrectly 3570 // accepting function declarations as record declarations. 3571 // In general, we cannot solve this problem. Consider: 3572 // class A<int> B() {} 3573 // which can be a function definition or a class definition when B() is a 3574 // macro. If we find enough real-world cases where this is a problem, we 3575 // can parse for the 'template' keyword in the beginning of the statement, 3576 // and thus rule out the record production in case there is no template 3577 // (this would still leave us with an ambiguity between template function 3578 // and class declarations). 3579 if (FormatTok->isOneOf(tok::colon, tok::less)) { 3580 do { 3581 if (FormatTok->is(tok::l_brace)) { 3582 calculateBraceTypes(/*ExpectClassBody=*/true); 3583 if (!tryToParseBracedList()) 3584 break; 3585 } 3586 if (FormatTok->is(tok::l_square)) { 3587 FormatToken *Previous = FormatTok->Previous; 3588 if (!Previous || 3589 !(Previous->is(tok::r_paren) || Previous->isTypeOrIdentifier())) { 3590 // Don't try parsing a lambda if we had a closing parenthesis before, 3591 // it was probably a pointer to an array: int (*)[]. 3592 if (!tryToParseLambda()) 3593 break; 3594 } else { 3595 parseSquare(); 3596 continue; 3597 } 3598 } 3599 if (FormatTok->is(tok::semi)) 3600 return; 3601 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where)) { 3602 addUnwrappedLine(); 3603 nextToken(); 3604 parseCSharpGenericTypeConstraint(); 3605 break; 3606 } 3607 nextToken(); 3608 } while (!eof()); 3609 } 3610 3611 auto GetBraceType = [](const FormatToken &RecordTok) { 3612 switch (RecordTok.Tok.getKind()) { 3613 case tok::kw_class: 3614 return TT_ClassLBrace; 3615 case tok::kw_struct: 3616 return TT_StructLBrace; 3617 case tok::kw_union: 3618 return TT_UnionLBrace; 3619 default: 3620 // Useful for e.g. interface. 3621 return TT_RecordLBrace; 3622 } 3623 }; 3624 if (FormatTok->is(tok::l_brace)) { 3625 FormatTok->setFinalizedType(GetBraceType(InitialToken)); 3626 if (ParseAsExpr) { 3627 parseChildBlock(); 3628 } else { 3629 if (ShouldBreakBeforeBrace(Style, InitialToken)) 3630 addUnwrappedLine(); 3631 3632 unsigned AddLevels = Style.IndentAccessModifiers ? 2u : 1u; 3633 parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/false); 3634 } 3635 } 3636 // There is no addUnwrappedLine() here so that we fall through to parsing a 3637 // structural element afterwards. Thus, in "class A {} n, m;", 3638 // "} n, m;" will end up in one unwrapped line. 3639 } 3640 3641 void UnwrappedLineParser::parseObjCMethod() { 3642 assert(FormatTok->isOneOf(tok::l_paren, tok::identifier) && 3643 "'(' or identifier expected."); 3644 do { 3645 if (FormatTok->is(tok::semi)) { 3646 nextToken(); 3647 addUnwrappedLine(); 3648 return; 3649 } else if (FormatTok->is(tok::l_brace)) { 3650 if (Style.BraceWrapping.AfterFunction) 3651 addUnwrappedLine(); 3652 parseBlock(); 3653 addUnwrappedLine(); 3654 return; 3655 } else { 3656 nextToken(); 3657 } 3658 } while (!eof()); 3659 } 3660 3661 void UnwrappedLineParser::parseObjCProtocolList() { 3662 assert(FormatTok->is(tok::less) && "'<' expected."); 3663 do { 3664 nextToken(); 3665 // Early exit in case someone forgot a close angle. 3666 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 3667 FormatTok->isObjCAtKeyword(tok::objc_end)) 3668 return; 3669 } while (!eof() && FormatTok->isNot(tok::greater)); 3670 nextToken(); // Skip '>'. 3671 } 3672 3673 void UnwrappedLineParser::parseObjCUntilAtEnd() { 3674 do { 3675 if (FormatTok->isObjCAtKeyword(tok::objc_end)) { 3676 nextToken(); 3677 addUnwrappedLine(); 3678 break; 3679 } 3680 if (FormatTok->is(tok::l_brace)) { 3681 parseBlock(); 3682 // In ObjC interfaces, nothing should be following the "}". 3683 addUnwrappedLine(); 3684 } else if (FormatTok->is(tok::r_brace)) { 3685 // Ignore stray "}". parseStructuralElement doesn't consume them. 3686 nextToken(); 3687 addUnwrappedLine(); 3688 } else if (FormatTok->isOneOf(tok::minus, tok::plus)) { 3689 nextToken(); 3690 parseObjCMethod(); 3691 } else { 3692 parseStructuralElement(); 3693 } 3694 } while (!eof()); 3695 } 3696 3697 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 3698 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_interface || 3699 FormatTok->Tok.getObjCKeywordID() == tok::objc_implementation); 3700 nextToken(); 3701 nextToken(); // interface name 3702 3703 // @interface can be followed by a lightweight generic 3704 // specialization list, then either a base class or a category. 3705 if (FormatTok->is(tok::less)) 3706 parseObjCLightweightGenerics(); 3707 if (FormatTok->is(tok::colon)) { 3708 nextToken(); 3709 nextToken(); // base class name 3710 // The base class can also have lightweight generics applied to it. 3711 if (FormatTok->is(tok::less)) 3712 parseObjCLightweightGenerics(); 3713 } else if (FormatTok->is(tok::l_paren)) 3714 // Skip category, if present. 3715 parseParens(); 3716 3717 if (FormatTok->is(tok::less)) 3718 parseObjCProtocolList(); 3719 3720 if (FormatTok->is(tok::l_brace)) { 3721 if (Style.BraceWrapping.AfterObjCDeclaration) 3722 addUnwrappedLine(); 3723 parseBlock(/*MustBeDeclaration=*/true); 3724 } 3725 3726 // With instance variables, this puts '}' on its own line. Without instance 3727 // variables, this ends the @interface line. 3728 addUnwrappedLine(); 3729 3730 parseObjCUntilAtEnd(); 3731 } 3732 3733 void UnwrappedLineParser::parseObjCLightweightGenerics() { 3734 assert(FormatTok->is(tok::less)); 3735 // Unlike protocol lists, generic parameterizations support 3736 // nested angles: 3737 // 3738 // @interface Foo<ValueType : id <NSCopying, NSSecureCoding>> : 3739 // NSObject <NSCopying, NSSecureCoding> 3740 // 3741 // so we need to count how many open angles we have left. 3742 unsigned NumOpenAngles = 1; 3743 do { 3744 nextToken(); 3745 // Early exit in case someone forgot a close angle. 3746 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 3747 FormatTok->isObjCAtKeyword(tok::objc_end)) 3748 break; 3749 if (FormatTok->is(tok::less)) 3750 ++NumOpenAngles; 3751 else if (FormatTok->is(tok::greater)) { 3752 assert(NumOpenAngles > 0 && "'>' makes NumOpenAngles negative"); 3753 --NumOpenAngles; 3754 } 3755 } while (!eof() && NumOpenAngles != 0); 3756 nextToken(); // Skip '>'. 3757 } 3758 3759 // Returns true for the declaration/definition form of @protocol, 3760 // false for the expression form. 3761 bool UnwrappedLineParser::parseObjCProtocol() { 3762 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_protocol); 3763 nextToken(); 3764 3765 if (FormatTok->is(tok::l_paren)) 3766 // The expression form of @protocol, e.g. "Protocol* p = @protocol(foo);". 3767 return false; 3768 3769 // The definition/declaration form, 3770 // @protocol Foo 3771 // - (int)someMethod; 3772 // @end 3773 3774 nextToken(); // protocol name 3775 3776 if (FormatTok->is(tok::less)) 3777 parseObjCProtocolList(); 3778 3779 // Check for protocol declaration. 3780 if (FormatTok->is(tok::semi)) { 3781 nextToken(); 3782 addUnwrappedLine(); 3783 return true; 3784 } 3785 3786 addUnwrappedLine(); 3787 parseObjCUntilAtEnd(); 3788 return true; 3789 } 3790 3791 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 3792 bool IsImport = FormatTok->is(Keywords.kw_import); 3793 assert(IsImport || FormatTok->is(tok::kw_export)); 3794 nextToken(); 3795 3796 // Consume the "default" in "export default class/function". 3797 if (FormatTok->is(tok::kw_default)) 3798 nextToken(); 3799 3800 // Consume "async function", "function" and "default function", so that these 3801 // get parsed as free-standing JS functions, i.e. do not require a trailing 3802 // semicolon. 3803 if (FormatTok->is(Keywords.kw_async)) 3804 nextToken(); 3805 if (FormatTok->is(Keywords.kw_function)) { 3806 nextToken(); 3807 return; 3808 } 3809 3810 // For imports, `export *`, `export {...}`, consume the rest of the line up 3811 // to the terminating `;`. For everything else, just return and continue 3812 // parsing the structural element, i.e. the declaration or expression for 3813 // `export default`. 3814 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 3815 !FormatTok->isStringLiteral()) 3816 return; 3817 3818 while (!eof()) { 3819 if (FormatTok->is(tok::semi)) 3820 return; 3821 if (Line->Tokens.empty()) { 3822 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 3823 // import statement should terminate. 3824 return; 3825 } 3826 if (FormatTok->is(tok::l_brace)) { 3827 FormatTok->setBlockKind(BK_Block); 3828 nextToken(); 3829 parseBracedList(); 3830 } else { 3831 nextToken(); 3832 } 3833 } 3834 } 3835 3836 void UnwrappedLineParser::parseStatementMacro() { 3837 nextToken(); 3838 if (FormatTok->is(tok::l_paren)) 3839 parseParens(); 3840 if (FormatTok->is(tok::semi)) 3841 nextToken(); 3842 addUnwrappedLine(); 3843 } 3844 3845 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 3846 StringRef Prefix = "") { 3847 llvm::dbgs() << Prefix << "Line(" << Line.Level 3848 << ", FSC=" << Line.FirstStartColumn << ")" 3849 << (Line.InPPDirective ? " MACRO" : "") << ": "; 3850 for (const auto &Node : Line.Tokens) { 3851 llvm::dbgs() << Node.Tok->Tok.getName() << "[" 3852 << "T=" << static_cast<unsigned>(Node.Tok->getType()) 3853 << ", OC=" << Node.Tok->OriginalColumn << "] "; 3854 } 3855 for (const auto &Node : Line.Tokens) 3856 for (const auto &ChildNode : Node.Children) 3857 printDebugInfo(ChildNode, "\nChild: "); 3858 3859 llvm::dbgs() << "\n"; 3860 } 3861 3862 void UnwrappedLineParser::addUnwrappedLine(LineLevel AdjustLevel) { 3863 if (Line->Tokens.empty()) 3864 return; 3865 LLVM_DEBUG({ 3866 if (CurrentLines == &Lines) 3867 printDebugInfo(*Line); 3868 }); 3869 3870 // If this line closes a block when in Whitesmiths mode, remember that 3871 // information so that the level can be decreased after the line is added. 3872 // This has to happen after the addition of the line since the line itself 3873 // needs to be indented. 3874 bool ClosesWhitesmithsBlock = 3875 Line->MatchingOpeningBlockLineIndex != UnwrappedLine::kInvalidIndex && 3876 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths; 3877 3878 CurrentLines->push_back(std::move(*Line)); 3879 Line->Tokens.clear(); 3880 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex; 3881 Line->FirstStartColumn = 0; 3882 3883 if (ClosesWhitesmithsBlock && AdjustLevel == LineLevel::Remove) 3884 --Line->Level; 3885 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 3886 CurrentLines->append( 3887 std::make_move_iterator(PreprocessorDirectives.begin()), 3888 std::make_move_iterator(PreprocessorDirectives.end())); 3889 PreprocessorDirectives.clear(); 3890 } 3891 // Disconnect the current token from the last token on the previous line. 3892 FormatTok->Previous = nullptr; 3893 } 3894 3895 bool UnwrappedLineParser::eof() const { return FormatTok->is(tok::eof); } 3896 3897 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 3898 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 3899 FormatTok.NewlinesBefore > 0; 3900 } 3901 3902 // Checks if \p FormatTok is a line comment that continues the line comment 3903 // section on \p Line. 3904 static bool 3905 continuesLineCommentSection(const FormatToken &FormatTok, 3906 const UnwrappedLine &Line, 3907 const llvm::Regex &CommentPragmasRegex) { 3908 if (Line.Tokens.empty()) 3909 return false; 3910 3911 StringRef IndentContent = FormatTok.TokenText; 3912 if (FormatTok.TokenText.startswith("//") || 3913 FormatTok.TokenText.startswith("/*")) 3914 IndentContent = FormatTok.TokenText.substr(2); 3915 if (CommentPragmasRegex.match(IndentContent)) 3916 return false; 3917 3918 // If Line starts with a line comment, then FormatTok continues the comment 3919 // section if its original column is greater or equal to the original start 3920 // column of the line. 3921 // 3922 // Define the min column token of a line as follows: if a line ends in '{' or 3923 // contains a '{' followed by a line comment, then the min column token is 3924 // that '{'. Otherwise, the min column token of the line is the first token of 3925 // the line. 3926 // 3927 // If Line starts with a token other than a line comment, then FormatTok 3928 // continues the comment section if its original column is greater than the 3929 // original start column of the min column token of the line. 3930 // 3931 // For example, the second line comment continues the first in these cases: 3932 // 3933 // // first line 3934 // // second line 3935 // 3936 // and: 3937 // 3938 // // first line 3939 // // second line 3940 // 3941 // and: 3942 // 3943 // int i; // first line 3944 // // second line 3945 // 3946 // and: 3947 // 3948 // do { // first line 3949 // // second line 3950 // int i; 3951 // } while (true); 3952 // 3953 // and: 3954 // 3955 // enum { 3956 // a, // first line 3957 // // second line 3958 // b 3959 // }; 3960 // 3961 // The second line comment doesn't continue the first in these cases: 3962 // 3963 // // first line 3964 // // second line 3965 // 3966 // and: 3967 // 3968 // int i; // first line 3969 // // second line 3970 // 3971 // and: 3972 // 3973 // do { // first line 3974 // // second line 3975 // int i; 3976 // } while (true); 3977 // 3978 // and: 3979 // 3980 // enum { 3981 // a, // first line 3982 // // second line 3983 // }; 3984 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 3985 3986 // Scan for '{//'. If found, use the column of '{' as a min column for line 3987 // comment section continuation. 3988 const FormatToken *PreviousToken = nullptr; 3989 for (const UnwrappedLineNode &Node : Line.Tokens) { 3990 if (PreviousToken && PreviousToken->is(tok::l_brace) && 3991 isLineComment(*Node.Tok)) { 3992 MinColumnToken = PreviousToken; 3993 break; 3994 } 3995 PreviousToken = Node.Tok; 3996 3997 // Grab the last newline preceding a token in this unwrapped line. 3998 if (Node.Tok->NewlinesBefore > 0) 3999 MinColumnToken = Node.Tok; 4000 } 4001 if (PreviousToken && PreviousToken->is(tok::l_brace)) 4002 MinColumnToken = PreviousToken; 4003 4004 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok, 4005 MinColumnToken); 4006 } 4007 4008 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 4009 bool JustComments = Line->Tokens.empty(); 4010 for (FormatToken *Tok : CommentsBeforeNextToken) { 4011 // Line comments that belong to the same line comment section are put on the 4012 // same line since later we might want to reflow content between them. 4013 // Additional fine-grained breaking of line comment sections is controlled 4014 // by the class BreakableLineCommentSection in case it is desirable to keep 4015 // several line comment sections in the same unwrapped line. 4016 // 4017 // FIXME: Consider putting separate line comment sections as children to the 4018 // unwrapped line instead. 4019 Tok->ContinuesLineCommentSection = 4020 continuesLineCommentSection(*Tok, *Line, CommentPragmasRegex); 4021 if (isOnNewLine(*Tok) && JustComments && !Tok->ContinuesLineCommentSection) 4022 addUnwrappedLine(); 4023 pushToken(Tok); 4024 } 4025 if (NewlineBeforeNext && JustComments) 4026 addUnwrappedLine(); 4027 CommentsBeforeNextToken.clear(); 4028 } 4029 4030 void UnwrappedLineParser::nextToken(int LevelDifference) { 4031 if (eof()) 4032 return; 4033 flushComments(isOnNewLine(*FormatTok)); 4034 pushToken(FormatTok); 4035 FormatToken *Previous = FormatTok; 4036 if (!Style.isJavaScript()) 4037 readToken(LevelDifference); 4038 else 4039 readTokenWithJavaScriptASI(); 4040 FormatTok->Previous = Previous; 4041 } 4042 4043 void UnwrappedLineParser::distributeComments( 4044 const SmallVectorImpl<FormatToken *> &Comments, 4045 const FormatToken *NextTok) { 4046 // Whether or not a line comment token continues a line is controlled by 4047 // the method continuesLineCommentSection, with the following caveat: 4048 // 4049 // Define a trail of Comments to be a nonempty proper postfix of Comments such 4050 // that each comment line from the trail is aligned with the next token, if 4051 // the next token exists. If a trail exists, the beginning of the maximal 4052 // trail is marked as a start of a new comment section. 4053 // 4054 // For example in this code: 4055 // 4056 // int a; // line about a 4057 // // line 1 about b 4058 // // line 2 about b 4059 // int b; 4060 // 4061 // the two lines about b form a maximal trail, so there are two sections, the 4062 // first one consisting of the single comment "// line about a" and the 4063 // second one consisting of the next two comments. 4064 if (Comments.empty()) 4065 return; 4066 bool ShouldPushCommentsInCurrentLine = true; 4067 bool HasTrailAlignedWithNextToken = false; 4068 unsigned StartOfTrailAlignedWithNextToken = 0; 4069 if (NextTok) { 4070 // We are skipping the first element intentionally. 4071 for (unsigned i = Comments.size() - 1; i > 0; --i) { 4072 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) { 4073 HasTrailAlignedWithNextToken = true; 4074 StartOfTrailAlignedWithNextToken = i; 4075 } 4076 } 4077 } 4078 for (unsigned i = 0, e = Comments.size(); i < e; ++i) { 4079 FormatToken *FormatTok = Comments[i]; 4080 if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) { 4081 FormatTok->ContinuesLineCommentSection = false; 4082 } else { 4083 FormatTok->ContinuesLineCommentSection = 4084 continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex); 4085 } 4086 if (!FormatTok->ContinuesLineCommentSection && 4087 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) 4088 ShouldPushCommentsInCurrentLine = false; 4089 if (ShouldPushCommentsInCurrentLine) 4090 pushToken(FormatTok); 4091 else 4092 CommentsBeforeNextToken.push_back(FormatTok); 4093 } 4094 } 4095 4096 void UnwrappedLineParser::readToken(int LevelDifference) { 4097 SmallVector<FormatToken *, 1> Comments; 4098 bool PreviousWasComment = false; 4099 bool FirstNonCommentOnLine = false; 4100 do { 4101 FormatTok = Tokens->getNextToken(); 4102 assert(FormatTok); 4103 while (FormatTok->getType() == TT_ConflictStart || 4104 FormatTok->getType() == TT_ConflictEnd || 4105 FormatTok->getType() == TT_ConflictAlternative) { 4106 if (FormatTok->getType() == TT_ConflictStart) 4107 conditionalCompilationStart(/*Unreachable=*/false); 4108 else if (FormatTok->getType() == TT_ConflictAlternative) 4109 conditionalCompilationAlternative(); 4110 else if (FormatTok->getType() == TT_ConflictEnd) 4111 conditionalCompilationEnd(); 4112 FormatTok = Tokens->getNextToken(); 4113 FormatTok->MustBreakBefore = true; 4114 } 4115 4116 auto IsFirstNonCommentOnLine = [](bool FirstNonCommentOnLine, 4117 const FormatToken &Tok, 4118 bool PreviousWasComment) { 4119 auto IsFirstOnLine = [](const FormatToken &Tok) { 4120 return Tok.HasUnescapedNewline || Tok.IsFirst; 4121 }; 4122 4123 // Consider preprocessor directives preceded by block comments as first 4124 // on line. 4125 if (PreviousWasComment) 4126 return FirstNonCommentOnLine || IsFirstOnLine(Tok); 4127 return IsFirstOnLine(Tok); 4128 }; 4129 4130 FirstNonCommentOnLine = IsFirstNonCommentOnLine( 4131 FirstNonCommentOnLine, *FormatTok, PreviousWasComment); 4132 PreviousWasComment = FormatTok->is(tok::comment); 4133 4134 while (!Line->InPPDirective && FormatTok->is(tok::hash) && 4135 FirstNonCommentOnLine) { 4136 distributeComments(Comments, FormatTok); 4137 Comments.clear(); 4138 // If there is an unfinished unwrapped line, we flush the preprocessor 4139 // directives only after that unwrapped line was finished later. 4140 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 4141 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 4142 assert((LevelDifference >= 0 || 4143 static_cast<unsigned>(-LevelDifference) <= Line->Level) && 4144 "LevelDifference makes Line->Level negative"); 4145 Line->Level += LevelDifference; 4146 // Comments stored before the preprocessor directive need to be output 4147 // before the preprocessor directive, at the same level as the 4148 // preprocessor directive, as we consider them to apply to the directive. 4149 if (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash && 4150 PPBranchLevel > 0) 4151 Line->Level += PPBranchLevel; 4152 flushComments(isOnNewLine(*FormatTok)); 4153 parsePPDirective(); 4154 PreviousWasComment = FormatTok->is(tok::comment); 4155 FirstNonCommentOnLine = IsFirstNonCommentOnLine( 4156 FirstNonCommentOnLine, *FormatTok, PreviousWasComment); 4157 } 4158 4159 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) && 4160 !Line->InPPDirective) 4161 continue; 4162 4163 if (!FormatTok->is(tok::comment)) { 4164 distributeComments(Comments, FormatTok); 4165 Comments.clear(); 4166 return; 4167 } 4168 4169 Comments.push_back(FormatTok); 4170 } while (!eof()); 4171 4172 distributeComments(Comments, nullptr); 4173 Comments.clear(); 4174 } 4175 4176 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 4177 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 4178 if (MustBreakBeforeNextToken) { 4179 Line->Tokens.back().Tok->MustBreakBefore = true; 4180 MustBreakBeforeNextToken = false; 4181 } 4182 } 4183 4184 } // end namespace format 4185 } // end namespace clang 4186