1 //===--- UnwrappedLineParser.cpp - Format C++ code ------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// 9 /// \file 10 /// This file contains the implementation of the UnwrappedLineParser, 11 /// which turns a stream of tokens into UnwrappedLines. 12 /// 13 //===----------------------------------------------------------------------===// 14 15 #include "UnwrappedLineParser.h" 16 #include "FormatToken.h" 17 #include "llvm/ADT/STLExtras.h" 18 #include "llvm/Support/Debug.h" 19 #include "llvm/Support/raw_ostream.h" 20 21 #include <algorithm> 22 #include <utility> 23 24 #define DEBUG_TYPE "format-parser" 25 26 namespace clang { 27 namespace format { 28 29 class FormatTokenSource { 30 public: 31 virtual ~FormatTokenSource() {} 32 33 // Returns the next token in the token stream. 34 virtual FormatToken *getNextToken() = 0; 35 36 // Returns the token preceding the token returned by the last call to 37 // getNextToken() in the token stream, or nullptr if no such token exists. 38 virtual FormatToken *getPreviousToken() = 0; 39 40 // Returns the token that would be returned by the next call to 41 // getNextToken(). 42 virtual FormatToken *peekNextToken() = 0; 43 44 // Returns the token that would be returned after the next N calls to 45 // getNextToken(). N needs to be greater than zero, and small enough that 46 // there are still tokens. Check for tok::eof with N-1 before calling it with 47 // N. 48 virtual FormatToken *peekNextToken(int N) = 0; 49 50 // Returns whether we are at the end of the file. 51 // This can be different from whether getNextToken() returned an eof token 52 // when the FormatTokenSource is a view on a part of the token stream. 53 virtual bool isEOF() = 0; 54 55 // Gets the current position in the token stream, to be used by setPosition(). 56 virtual unsigned getPosition() = 0; 57 58 // Resets the token stream to the state it was in when getPosition() returned 59 // Position, and return the token at that position in the stream. 60 virtual FormatToken *setPosition(unsigned Position) = 0; 61 }; 62 63 namespace { 64 65 class ScopedDeclarationState { 66 public: 67 ScopedDeclarationState(UnwrappedLine &Line, llvm::BitVector &Stack, 68 bool MustBeDeclaration) 69 : Line(Line), Stack(Stack) { 70 Line.MustBeDeclaration = MustBeDeclaration; 71 Stack.push_back(MustBeDeclaration); 72 } 73 ~ScopedDeclarationState() { 74 Stack.pop_back(); 75 if (!Stack.empty()) 76 Line.MustBeDeclaration = Stack.back(); 77 else 78 Line.MustBeDeclaration = true; 79 } 80 81 private: 82 UnwrappedLine &Line; 83 llvm::BitVector &Stack; 84 }; 85 86 static bool isLineComment(const FormatToken &FormatTok) { 87 return FormatTok.is(tok::comment) && !FormatTok.TokenText.startswith("/*"); 88 } 89 90 // Checks if \p FormatTok is a line comment that continues the line comment 91 // \p Previous. The original column of \p MinColumnToken is used to determine 92 // whether \p FormatTok is indented enough to the right to continue \p Previous. 93 static bool continuesLineComment(const FormatToken &FormatTok, 94 const FormatToken *Previous, 95 const FormatToken *MinColumnToken) { 96 if (!Previous || !MinColumnToken) 97 return false; 98 unsigned MinContinueColumn = 99 MinColumnToken->OriginalColumn + (isLineComment(*MinColumnToken) ? 0 : 1); 100 return isLineComment(FormatTok) && FormatTok.NewlinesBefore == 1 && 101 isLineComment(*Previous) && 102 FormatTok.OriginalColumn >= MinContinueColumn; 103 } 104 105 class ScopedMacroState : public FormatTokenSource { 106 public: 107 ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource, 108 FormatToken *&ResetToken) 109 : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken), 110 PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource), 111 Token(nullptr), PreviousToken(nullptr) { 112 FakeEOF.Tok.startToken(); 113 FakeEOF.Tok.setKind(tok::eof); 114 TokenSource = this; 115 Line.Level = 0; 116 Line.InPPDirective = true; 117 } 118 119 ~ScopedMacroState() override { 120 TokenSource = PreviousTokenSource; 121 ResetToken = Token; 122 Line.InPPDirective = false; 123 Line.Level = PreviousLineLevel; 124 } 125 126 FormatToken *getNextToken() override { 127 // The \c UnwrappedLineParser guards against this by never calling 128 // \c getNextToken() after it has encountered the first eof token. 129 assert(!eof()); 130 PreviousToken = Token; 131 Token = PreviousTokenSource->getNextToken(); 132 if (eof()) 133 return &FakeEOF; 134 return Token; 135 } 136 137 FormatToken *getPreviousToken() override { 138 return PreviousTokenSource->getPreviousToken(); 139 } 140 141 FormatToken *peekNextToken() override { 142 if (eof()) 143 return &FakeEOF; 144 return PreviousTokenSource->peekNextToken(); 145 } 146 147 FormatToken *peekNextToken(int N) override { 148 assert(N > 0); 149 if (eof()) 150 return &FakeEOF; 151 return PreviousTokenSource->peekNextToken(N); 152 } 153 154 bool isEOF() override { return PreviousTokenSource->isEOF(); } 155 156 unsigned getPosition() override { return PreviousTokenSource->getPosition(); } 157 158 FormatToken *setPosition(unsigned Position) override { 159 PreviousToken = nullptr; 160 Token = PreviousTokenSource->setPosition(Position); 161 return Token; 162 } 163 164 private: 165 bool eof() { 166 return Token && Token->HasUnescapedNewline && 167 !continuesLineComment(*Token, PreviousToken, 168 /*MinColumnToken=*/PreviousToken); 169 } 170 171 FormatToken FakeEOF; 172 UnwrappedLine &Line; 173 FormatTokenSource *&TokenSource; 174 FormatToken *&ResetToken; 175 unsigned PreviousLineLevel; 176 FormatTokenSource *PreviousTokenSource; 177 178 FormatToken *Token; 179 FormatToken *PreviousToken; 180 }; 181 182 } // end anonymous namespace 183 184 class ScopedLineState { 185 public: 186 ScopedLineState(UnwrappedLineParser &Parser, 187 bool SwitchToPreprocessorLines = false) 188 : Parser(Parser), OriginalLines(Parser.CurrentLines) { 189 if (SwitchToPreprocessorLines) 190 Parser.CurrentLines = &Parser.PreprocessorDirectives; 191 else if (!Parser.Line->Tokens.empty()) 192 Parser.CurrentLines = &Parser.Line->Tokens.back().Children; 193 PreBlockLine = std::move(Parser.Line); 194 Parser.Line = std::make_unique<UnwrappedLine>(); 195 Parser.Line->Level = PreBlockLine->Level; 196 Parser.Line->InPPDirective = PreBlockLine->InPPDirective; 197 } 198 199 ~ScopedLineState() { 200 if (!Parser.Line->Tokens.empty()) 201 Parser.addUnwrappedLine(); 202 assert(Parser.Line->Tokens.empty()); 203 Parser.Line = std::move(PreBlockLine); 204 if (Parser.CurrentLines == &Parser.PreprocessorDirectives) 205 Parser.MustBreakBeforeNextToken = true; 206 Parser.CurrentLines = OriginalLines; 207 } 208 209 private: 210 UnwrappedLineParser &Parser; 211 212 std::unique_ptr<UnwrappedLine> PreBlockLine; 213 SmallVectorImpl<UnwrappedLine> *OriginalLines; 214 }; 215 216 class CompoundStatementIndenter { 217 public: 218 CompoundStatementIndenter(UnwrappedLineParser *Parser, 219 const FormatStyle &Style, unsigned &LineLevel) 220 : CompoundStatementIndenter(Parser, LineLevel, 221 Style.BraceWrapping.AfterControlStatement, 222 Style.BraceWrapping.IndentBraces) {} 223 CompoundStatementIndenter(UnwrappedLineParser *Parser, unsigned &LineLevel, 224 bool WrapBrace, bool IndentBrace) 225 : LineLevel(LineLevel), OldLineLevel(LineLevel) { 226 if (WrapBrace) 227 Parser->addUnwrappedLine(); 228 if (IndentBrace) 229 ++LineLevel; 230 } 231 ~CompoundStatementIndenter() { LineLevel = OldLineLevel; } 232 233 private: 234 unsigned &LineLevel; 235 unsigned OldLineLevel; 236 }; 237 238 namespace { 239 240 class IndexedTokenSource : public FormatTokenSource { 241 public: 242 IndexedTokenSource(ArrayRef<FormatToken *> Tokens) 243 : Tokens(Tokens), Position(-1) {} 244 245 FormatToken *getNextToken() override { 246 if (Position >= 0 && Tokens[Position]->is(tok::eof)) { 247 LLVM_DEBUG({ 248 llvm::dbgs() << "Next "; 249 dbgToken(Position); 250 }); 251 return Tokens[Position]; 252 } 253 ++Position; 254 LLVM_DEBUG({ 255 llvm::dbgs() << "Next "; 256 dbgToken(Position); 257 }); 258 return Tokens[Position]; 259 } 260 261 FormatToken *getPreviousToken() override { 262 return Position > 0 ? Tokens[Position - 1] : nullptr; 263 } 264 265 FormatToken *peekNextToken() override { 266 int Next = Position + 1; 267 LLVM_DEBUG({ 268 llvm::dbgs() << "Peeking "; 269 dbgToken(Next); 270 }); 271 return Tokens[Next]; 272 } 273 274 FormatToken *peekNextToken(int N) override { 275 assert(N > 0); 276 int Next = Position + N; 277 LLVM_DEBUG({ 278 llvm::dbgs() << "Peeking (+" << (N - 1) << ") "; 279 dbgToken(Next); 280 }); 281 return Tokens[Next]; 282 } 283 284 bool isEOF() override { return Tokens[Position]->is(tok::eof); } 285 286 unsigned getPosition() override { 287 LLVM_DEBUG(llvm::dbgs() << "Getting Position: " << Position << "\n"); 288 assert(Position >= 0); 289 return Position; 290 } 291 292 FormatToken *setPosition(unsigned P) override { 293 LLVM_DEBUG(llvm::dbgs() << "Setting Position: " << P << "\n"); 294 Position = P; 295 return Tokens[Position]; 296 } 297 298 void reset() { Position = -1; } 299 300 private: 301 void dbgToken(int Position, llvm::StringRef Indent = "") { 302 FormatToken *Tok = Tokens[Position]; 303 llvm::dbgs() << Indent << "[" << Position 304 << "] Token: " << Tok->Tok.getName() << " / " << Tok->TokenText 305 << ", Macro: " << !!Tok->MacroCtx << "\n"; 306 } 307 308 ArrayRef<FormatToken *> Tokens; 309 int Position; 310 }; 311 312 } // end anonymous namespace 313 314 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style, 315 const AdditionalKeywords &Keywords, 316 unsigned FirstStartColumn, 317 ArrayRef<FormatToken *> Tokens, 318 UnwrappedLineConsumer &Callback) 319 : Line(new UnwrappedLine), MustBreakBeforeNextToken(false), 320 CurrentLines(&Lines), Style(Style), Keywords(Keywords), 321 CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr), 322 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1), 323 IncludeGuard(Style.IndentPPDirectives == FormatStyle::PPDIS_None 324 ? IG_Rejected 325 : IG_Inited), 326 IncludeGuardToken(nullptr), FirstStartColumn(FirstStartColumn) {} 327 328 void UnwrappedLineParser::reset() { 329 PPBranchLevel = -1; 330 IncludeGuard = Style.IndentPPDirectives == FormatStyle::PPDIS_None 331 ? IG_Rejected 332 : IG_Inited; 333 IncludeGuardToken = nullptr; 334 Line.reset(new UnwrappedLine); 335 CommentsBeforeNextToken.clear(); 336 FormatTok = nullptr; 337 MustBreakBeforeNextToken = false; 338 PreprocessorDirectives.clear(); 339 CurrentLines = &Lines; 340 DeclarationScopeStack.clear(); 341 NestedTooDeep.clear(); 342 PPStack.clear(); 343 Line->FirstStartColumn = FirstStartColumn; 344 } 345 346 void UnwrappedLineParser::parse() { 347 IndexedTokenSource TokenSource(AllTokens); 348 Line->FirstStartColumn = FirstStartColumn; 349 do { 350 LLVM_DEBUG(llvm::dbgs() << "----\n"); 351 reset(); 352 Tokens = &TokenSource; 353 TokenSource.reset(); 354 355 readToken(); 356 parseFile(); 357 358 // If we found an include guard then all preprocessor directives (other than 359 // the guard) are over-indented by one. 360 if (IncludeGuard == IG_Found) 361 for (auto &Line : Lines) 362 if (Line.InPPDirective && Line.Level > 0) 363 --Line.Level; 364 365 // Create line with eof token. 366 pushToken(FormatTok); 367 addUnwrappedLine(); 368 369 for (const UnwrappedLine &Line : Lines) 370 Callback.consumeUnwrappedLine(Line); 371 372 Callback.finishRun(); 373 Lines.clear(); 374 while (!PPLevelBranchIndex.empty() && 375 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) { 376 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1); 377 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1); 378 } 379 if (!PPLevelBranchIndex.empty()) { 380 ++PPLevelBranchIndex.back(); 381 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size()); 382 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back()); 383 } 384 } while (!PPLevelBranchIndex.empty()); 385 } 386 387 void UnwrappedLineParser::parseFile() { 388 // The top-level context in a file always has declarations, except for pre- 389 // processor directives and JavaScript files. 390 bool MustBeDeclaration = !Line->InPPDirective && !Style.isJavaScript(); 391 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 392 MustBeDeclaration); 393 if (Style.Language == FormatStyle::LK_TextProto) 394 parseBracedList(); 395 else 396 parseLevel(/*HasOpeningBrace=*/false, /*CanContainBracedList=*/true); 397 // Make sure to format the remaining tokens. 398 // 399 // LK_TextProto is special since its top-level is parsed as the body of a 400 // braced list, which does not necessarily have natural line separators such 401 // as a semicolon. Comments after the last entry that have been determined to 402 // not belong to that line, as in: 403 // key: value 404 // // endfile comment 405 // do not have a chance to be put on a line of their own until this point. 406 // Here we add this newline before end-of-file comments. 407 if (Style.Language == FormatStyle::LK_TextProto && 408 !CommentsBeforeNextToken.empty()) 409 addUnwrappedLine(); 410 flushComments(true); 411 addUnwrappedLine(); 412 } 413 414 void UnwrappedLineParser::parseCSharpGenericTypeConstraint() { 415 do { 416 switch (FormatTok->Tok.getKind()) { 417 case tok::l_brace: 418 return; 419 default: 420 if (FormatTok->is(Keywords.kw_where)) { 421 addUnwrappedLine(); 422 nextToken(); 423 parseCSharpGenericTypeConstraint(); 424 break; 425 } 426 nextToken(); 427 break; 428 } 429 } while (!eof()); 430 } 431 432 void UnwrappedLineParser::parseCSharpAttribute() { 433 int UnpairedSquareBrackets = 1; 434 do { 435 switch (FormatTok->Tok.getKind()) { 436 case tok::r_square: 437 nextToken(); 438 --UnpairedSquareBrackets; 439 if (UnpairedSquareBrackets == 0) { 440 addUnwrappedLine(); 441 return; 442 } 443 break; 444 case tok::l_square: 445 ++UnpairedSquareBrackets; 446 nextToken(); 447 break; 448 default: 449 nextToken(); 450 break; 451 } 452 } while (!eof()); 453 } 454 455 bool UnwrappedLineParser::precededByCommentOrPPDirective() const { 456 if (!Lines.empty() && Lines.back().InPPDirective) 457 return true; 458 459 const FormatToken *Previous = Tokens->getPreviousToken(); 460 return Previous && Previous->is(tok::comment) && 461 (Previous->IsMultiline || Previous->NewlinesBefore > 0); 462 } 463 /// \brief Parses a level, that is ???. 464 /// \param HasOpeningBrace If that level is started by an opening brace. 465 /// \param CanContainBracedList If the content can contain (at any level) a 466 /// braced list. 467 /// \param NextLBracesType The type for left brace found in this level. 468 /// \returns true if a simple block, or false otherwise. (A simple block has a 469 /// single statement.) 470 bool UnwrappedLineParser::parseLevel(bool HasOpeningBrace, 471 bool CanContainBracedList, 472 IfStmtKind *IfKind, 473 TokenType NextLBracesType) { 474 auto NextLevelLBracesType = NextLBracesType == TT_CompoundRequirementLBrace 475 ? TT_BracedListLBrace 476 : TT_Unknown; 477 const bool IsPrecededByCommentOrPPDirective = 478 !Style.RemoveBracesLLVM || precededByCommentOrPPDirective(); 479 bool HasLabel = false; 480 unsigned StatementCount = 0; 481 bool SwitchLabelEncountered = false; 482 do { 483 if (FormatTok->getType() == TT_AttributeMacro) { 484 nextToken(); 485 continue; 486 } 487 tok::TokenKind kind = FormatTok->Tok.getKind(); 488 if (FormatTok->getType() == TT_MacroBlockBegin) 489 kind = tok::l_brace; 490 else if (FormatTok->getType() == TT_MacroBlockEnd) 491 kind = tok::r_brace; 492 493 auto ParseDefault = [this, HasOpeningBrace, IfKind, NextLevelLBracesType, 494 &HasLabel, &StatementCount] { 495 parseStructuralElement(IfKind, !HasOpeningBrace, NextLevelLBracesType, 496 HasLabel ? nullptr : &HasLabel); 497 ++StatementCount; 498 assert(StatementCount > 0 && "StatementCount overflow!"); 499 }; 500 501 switch (kind) { 502 case tok::comment: 503 nextToken(); 504 addUnwrappedLine(); 505 break; 506 case tok::l_brace: 507 if (NextLBracesType != TT_Unknown) 508 FormatTok->setFinalizedType(NextLBracesType); 509 else if (FormatTok->Previous && 510 FormatTok->Previous->ClosesRequiresClause) { 511 // We need the 'default' case here to correctly parse a function 512 // l_brace. 513 ParseDefault(); 514 continue; 515 } 516 if (CanContainBracedList && !FormatTok->is(TT_MacroBlockBegin) && 517 tryToParseBracedList()) 518 continue; 519 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 520 /*MunchSemi=*/true, /*UnindentWhitesmithBraces=*/false, 521 CanContainBracedList, 522 /*NextLBracesType=*/NextLBracesType); 523 ++StatementCount; 524 assert(StatementCount > 0 && "StatementCount overflow!"); 525 addUnwrappedLine(); 526 break; 527 case tok::r_brace: 528 if (HasOpeningBrace) { 529 if (!Style.RemoveBracesLLVM) 530 return false; 531 if (FormatTok->isNot(tok::r_brace) || StatementCount != 1 || HasLabel || 532 IsPrecededByCommentOrPPDirective || 533 precededByCommentOrPPDirective()) 534 return false; 535 const FormatToken *Next = Tokens->peekNextToken(); 536 return Next->isNot(tok::comment) || Next->NewlinesBefore > 0; 537 } 538 nextToken(); 539 addUnwrappedLine(); 540 break; 541 case tok::kw_default: { 542 unsigned StoredPosition = Tokens->getPosition(); 543 FormatToken *Next; 544 do { 545 Next = Tokens->getNextToken(); 546 assert(Next); 547 } while (Next->is(tok::comment)); 548 FormatTok = Tokens->setPosition(StoredPosition); 549 if (Next->isNot(tok::colon)) { 550 // default not followed by ':' is not a case label; treat it like 551 // an identifier. 552 parseStructuralElement(); 553 break; 554 } 555 // Else, if it is 'default:', fall through to the case handling. 556 LLVM_FALLTHROUGH; 557 } 558 case tok::kw_case: 559 if (Style.isJavaScript() && Line->MustBeDeclaration) { 560 // A 'case: string' style field declaration. 561 parseStructuralElement(); 562 break; 563 } 564 if (!SwitchLabelEncountered && 565 (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1))) 566 ++Line->Level; 567 SwitchLabelEncountered = true; 568 parseStructuralElement(); 569 break; 570 case tok::l_square: 571 if (Style.isCSharp()) { 572 nextToken(); 573 parseCSharpAttribute(); 574 break; 575 } 576 if (handleCppAttributes()) 577 break; 578 LLVM_FALLTHROUGH; 579 default: 580 ParseDefault(); 581 break; 582 } 583 } while (!eof()); 584 return false; 585 } 586 587 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) { 588 // We'll parse forward through the tokens until we hit 589 // a closing brace or eof - note that getNextToken() will 590 // parse macros, so this will magically work inside macro 591 // definitions, too. 592 unsigned StoredPosition = Tokens->getPosition(); 593 FormatToken *Tok = FormatTok; 594 const FormatToken *PrevTok = Tok->Previous; 595 // Keep a stack of positions of lbrace tokens. We will 596 // update information about whether an lbrace starts a 597 // braced init list or a different block during the loop. 598 SmallVector<FormatToken *, 8> LBraceStack; 599 assert(Tok->is(tok::l_brace)); 600 do { 601 // Get next non-comment token. 602 FormatToken *NextTok; 603 unsigned ReadTokens = 0; 604 do { 605 NextTok = Tokens->getNextToken(); 606 ++ReadTokens; 607 } while (NextTok->is(tok::comment)); 608 609 switch (Tok->Tok.getKind()) { 610 case tok::l_brace: 611 if (Style.isJavaScript() && PrevTok) { 612 if (PrevTok->isOneOf(tok::colon, tok::less)) 613 // A ':' indicates this code is in a type, or a braced list 614 // following a label in an object literal ({a: {b: 1}}). 615 // A '<' could be an object used in a comparison, but that is nonsense 616 // code (can never return true), so more likely it is a generic type 617 // argument (`X<{a: string; b: number}>`). 618 // The code below could be confused by semicolons between the 619 // individual members in a type member list, which would normally 620 // trigger BK_Block. In both cases, this must be parsed as an inline 621 // braced init. 622 Tok->setBlockKind(BK_BracedInit); 623 else if (PrevTok->is(tok::r_paren)) 624 // `) { }` can only occur in function or method declarations in JS. 625 Tok->setBlockKind(BK_Block); 626 } else { 627 Tok->setBlockKind(BK_Unknown); 628 } 629 LBraceStack.push_back(Tok); 630 break; 631 case tok::r_brace: 632 if (LBraceStack.empty()) 633 break; 634 if (LBraceStack.back()->is(BK_Unknown)) { 635 bool ProbablyBracedList = false; 636 if (Style.Language == FormatStyle::LK_Proto) { 637 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square); 638 } else { 639 // Skip NextTok over preprocessor lines, otherwise we may not 640 // properly diagnose the block as a braced intializer 641 // if the comma separator appears after the pp directive. 642 while (NextTok->is(tok::hash)) { 643 ScopedMacroState MacroState(*Line, Tokens, NextTok); 644 do { 645 NextTok = Tokens->getNextToken(); 646 ++ReadTokens; 647 } while (NextTok->isNot(tok::eof)); 648 } 649 650 // Using OriginalColumn to distinguish between ObjC methods and 651 // binary operators is a bit hacky. 652 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 653 NextTok->OriginalColumn == 0; 654 655 // Try to detect a braced list. Note that regardless how we mark inner 656 // braces here, we will overwrite the BlockKind later if we parse a 657 // braced list (where all blocks inside are by default braced lists), 658 // or when we explicitly detect blocks (for example while parsing 659 // lambdas). 660 661 // If we already marked the opening brace as braced list, the closing 662 // must also be part of it. 663 ProbablyBracedList = LBraceStack.back()->is(TT_BracedListLBrace); 664 665 ProbablyBracedList = ProbablyBracedList || 666 (Style.isJavaScript() && 667 NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in, 668 Keywords.kw_as)); 669 ProbablyBracedList = ProbablyBracedList || 670 (Style.isCpp() && NextTok->is(tok::l_paren)); 671 672 // If there is a comma, semicolon or right paren after the closing 673 // brace, we assume this is a braced initializer list. 674 // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a 675 // braced list in JS. 676 ProbablyBracedList = 677 ProbablyBracedList || 678 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 679 tok::r_paren, tok::r_square, tok::l_brace, 680 tok::ellipsis); 681 682 ProbablyBracedList = 683 ProbablyBracedList || 684 (NextTok->is(tok::identifier) && 685 !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace)); 686 687 ProbablyBracedList = ProbablyBracedList || 688 (NextTok->is(tok::semi) && 689 (!ExpectClassBody || LBraceStack.size() != 1)); 690 691 ProbablyBracedList = 692 ProbablyBracedList || 693 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 694 695 if (!Style.isCSharp() && NextTok->is(tok::l_square)) { 696 // We can have an array subscript after a braced init 697 // list, but C++11 attributes are expected after blocks. 698 NextTok = Tokens->getNextToken(); 699 ++ReadTokens; 700 ProbablyBracedList = NextTok->isNot(tok::l_square); 701 } 702 } 703 if (ProbablyBracedList) { 704 Tok->setBlockKind(BK_BracedInit); 705 LBraceStack.back()->setBlockKind(BK_BracedInit); 706 } else { 707 Tok->setBlockKind(BK_Block); 708 LBraceStack.back()->setBlockKind(BK_Block); 709 } 710 } 711 LBraceStack.pop_back(); 712 break; 713 case tok::identifier: 714 if (!Tok->is(TT_StatementMacro)) 715 break; 716 LLVM_FALLTHROUGH; 717 case tok::at: 718 case tok::semi: 719 case tok::kw_if: 720 case tok::kw_while: 721 case tok::kw_for: 722 case tok::kw_switch: 723 case tok::kw_try: 724 case tok::kw___try: 725 if (!LBraceStack.empty() && LBraceStack.back()->is(BK_Unknown)) 726 LBraceStack.back()->setBlockKind(BK_Block); 727 break; 728 default: 729 break; 730 } 731 PrevTok = Tok; 732 Tok = NextTok; 733 } while (Tok->isNot(tok::eof) && !LBraceStack.empty()); 734 735 // Assume other blocks for all unclosed opening braces. 736 for (FormatToken *LBrace : LBraceStack) 737 if (LBrace->is(BK_Unknown)) 738 LBrace->setBlockKind(BK_Block); 739 740 FormatTok = Tokens->setPosition(StoredPosition); 741 } 742 743 template <class T> 744 static inline void hash_combine(std::size_t &seed, const T &v) { 745 std::hash<T> hasher; 746 seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2); 747 } 748 749 size_t UnwrappedLineParser::computePPHash() const { 750 size_t h = 0; 751 for (const auto &i : PPStack) { 752 hash_combine(h, size_t(i.Kind)); 753 hash_combine(h, i.Line); 754 } 755 return h; 756 } 757 758 UnwrappedLineParser::IfStmtKind 759 UnwrappedLineParser::parseBlock(bool MustBeDeclaration, unsigned AddLevels, 760 bool MunchSemi, bool UnindentWhitesmithsBraces, 761 bool CanContainBracedList, 762 TokenType NextLBracesType) { 763 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) && 764 "'{' or macro block token expected"); 765 FormatToken *Tok = FormatTok; 766 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 767 FormatTok->setBlockKind(BK_Block); 768 769 // For Whitesmiths mode, jump to the next level prior to skipping over the 770 // braces. 771 if (AddLevels > 0 && Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) 772 ++Line->Level; 773 774 size_t PPStartHash = computePPHash(); 775 776 unsigned InitialLevel = Line->Level; 777 nextToken(/*LevelDifference=*/AddLevels); 778 779 if (MacroBlock && FormatTok->is(tok::l_paren)) 780 parseParens(); 781 782 size_t NbPreprocessorDirectives = 783 CurrentLines == &Lines ? PreprocessorDirectives.size() : 0; 784 addUnwrappedLine(); 785 size_t OpeningLineIndex = 786 CurrentLines->empty() 787 ? (UnwrappedLine::kInvalidIndex) 788 : (CurrentLines->size() - 1 - NbPreprocessorDirectives); 789 790 // Whitesmiths is weird here. The brace needs to be indented for the namespace 791 // block, but the block itself may not be indented depending on the style 792 // settings. This allows the format to back up one level in those cases. 793 if (UnindentWhitesmithsBraces) 794 --Line->Level; 795 796 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 797 MustBeDeclaration); 798 if (AddLevels > 0u && Style.BreakBeforeBraces != FormatStyle::BS_Whitesmiths) 799 Line->Level += AddLevels; 800 801 IfStmtKind IfKind = IfStmtKind::NotIf; 802 const bool SimpleBlock = parseLevel( 803 /*HasOpeningBrace=*/true, CanContainBracedList, &IfKind, NextLBracesType); 804 805 if (eof()) 806 return IfKind; 807 808 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd) 809 : !FormatTok->is(tok::r_brace)) { 810 Line->Level = InitialLevel; 811 FormatTok->setBlockKind(BK_Block); 812 return IfKind; 813 } 814 815 if (SimpleBlock && Tok->is(tok::l_brace)) { 816 assert(FormatTok->is(tok::r_brace)); 817 const FormatToken *Previous = Tokens->getPreviousToken(); 818 assert(Previous); 819 if (Previous->isNot(tok::r_brace) || Previous->Optional) { 820 Tok->MatchingParen = FormatTok; 821 FormatTok->MatchingParen = Tok; 822 } 823 } 824 825 size_t PPEndHash = computePPHash(); 826 827 // Munch the closing brace. 828 nextToken(/*LevelDifference=*/-AddLevels); 829 830 if (MacroBlock && FormatTok->is(tok::l_paren)) 831 parseParens(); 832 833 if (FormatTok->is(tok::kw_noexcept)) { 834 // A noexcept in a requires expression. 835 nextToken(); 836 } 837 838 if (FormatTok->is(tok::arrow)) { 839 // Following the } or noexcept we can find a trailing return type arrow 840 // as part of an implicit conversion constraint. 841 nextToken(); 842 parseStructuralElement(); 843 } 844 845 if (MunchSemi && FormatTok->is(tok::semi)) 846 nextToken(); 847 848 Line->Level = InitialLevel; 849 850 if (PPStartHash == PPEndHash) { 851 Line->MatchingOpeningBlockLineIndex = OpeningLineIndex; 852 if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) { 853 // Update the opening line to add the forward reference as well 854 (*CurrentLines)[OpeningLineIndex].MatchingClosingBlockLineIndex = 855 CurrentLines->size() - 1; 856 } 857 } 858 859 return IfKind; 860 } 861 862 static bool isGoogScope(const UnwrappedLine &Line) { 863 // FIXME: Closure-library specific stuff should not be hard-coded but be 864 // configurable. 865 if (Line.Tokens.size() < 4) 866 return false; 867 auto I = Line.Tokens.begin(); 868 if (I->Tok->TokenText != "goog") 869 return false; 870 ++I; 871 if (I->Tok->isNot(tok::period)) 872 return false; 873 ++I; 874 if (I->Tok->TokenText != "scope") 875 return false; 876 ++I; 877 return I->Tok->is(tok::l_paren); 878 } 879 880 static bool isIIFE(const UnwrappedLine &Line, 881 const AdditionalKeywords &Keywords) { 882 // Look for the start of an immediately invoked anonymous function. 883 // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression 884 // This is commonly done in JavaScript to create a new, anonymous scope. 885 // Example: (function() { ... })() 886 if (Line.Tokens.size() < 3) 887 return false; 888 auto I = Line.Tokens.begin(); 889 if (I->Tok->isNot(tok::l_paren)) 890 return false; 891 ++I; 892 if (I->Tok->isNot(Keywords.kw_function)) 893 return false; 894 ++I; 895 return I->Tok->is(tok::l_paren); 896 } 897 898 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 899 const FormatToken &InitialToken) { 900 tok::TokenKind Kind = InitialToken.Tok.getKind(); 901 if (InitialToken.is(TT_NamespaceMacro)) 902 Kind = tok::kw_namespace; 903 904 switch (Kind) { 905 case tok::kw_namespace: 906 return Style.BraceWrapping.AfterNamespace; 907 case tok::kw_class: 908 return Style.BraceWrapping.AfterClass; 909 case tok::kw_union: 910 return Style.BraceWrapping.AfterUnion; 911 case tok::kw_struct: 912 return Style.BraceWrapping.AfterStruct; 913 case tok::kw_enum: 914 return Style.BraceWrapping.AfterEnum; 915 default: 916 return false; 917 } 918 } 919 920 void UnwrappedLineParser::parseChildBlock( 921 bool CanContainBracedList, clang::format::TokenType NextLBracesType) { 922 FormatTok->setBlockKind(BK_Block); 923 nextToken(); 924 { 925 bool SkipIndent = (Style.isJavaScript() && 926 (isGoogScope(*Line) || isIIFE(*Line, Keywords))); 927 ScopedLineState LineState(*this); 928 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 929 /*MustBeDeclaration=*/false); 930 Line->Level += SkipIndent ? 0 : 1; 931 parseLevel(/*HasOpeningBrace=*/true, CanContainBracedList, 932 /*IfKind=*/nullptr, NextLBracesType); 933 flushComments(isOnNewLine(*FormatTok)); 934 Line->Level -= SkipIndent ? 0 : 1; 935 } 936 nextToken(); 937 } 938 939 void UnwrappedLineParser::parsePPDirective() { 940 assert(FormatTok->is(tok::hash) && "'#' expected"); 941 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 942 943 nextToken(); 944 945 if (!FormatTok->Tok.getIdentifierInfo()) { 946 parsePPUnknown(); 947 return; 948 } 949 950 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 951 case tok::pp_define: 952 parsePPDefine(); 953 return; 954 case tok::pp_if: 955 parsePPIf(/*IfDef=*/false); 956 break; 957 case tok::pp_ifdef: 958 case tok::pp_ifndef: 959 parsePPIf(/*IfDef=*/true); 960 break; 961 case tok::pp_else: 962 parsePPElse(); 963 break; 964 case tok::pp_elifdef: 965 case tok::pp_elifndef: 966 case tok::pp_elif: 967 parsePPElIf(); 968 break; 969 case tok::pp_endif: 970 parsePPEndIf(); 971 break; 972 default: 973 parsePPUnknown(); 974 break; 975 } 976 } 977 978 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 979 size_t Line = CurrentLines->size(); 980 if (CurrentLines == &PreprocessorDirectives) 981 Line += Lines.size(); 982 983 if (Unreachable || 984 (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable)) 985 PPStack.push_back({PP_Unreachable, Line}); 986 else 987 PPStack.push_back({PP_Conditional, Line}); 988 } 989 990 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 991 ++PPBranchLevel; 992 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 993 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 994 PPLevelBranchIndex.push_back(0); 995 PPLevelBranchCount.push_back(0); 996 } 997 PPChainBranchIndex.push(0); 998 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 999 conditionalCompilationCondition(Unreachable || Skip); 1000 } 1001 1002 void UnwrappedLineParser::conditionalCompilationAlternative() { 1003 if (!PPStack.empty()) 1004 PPStack.pop_back(); 1005 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 1006 if (!PPChainBranchIndex.empty()) 1007 ++PPChainBranchIndex.top(); 1008 conditionalCompilationCondition( 1009 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 1010 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 1011 } 1012 1013 void UnwrappedLineParser::conditionalCompilationEnd() { 1014 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 1015 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 1016 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) 1017 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 1018 } 1019 // Guard against #endif's without #if. 1020 if (PPBranchLevel > -1) 1021 --PPBranchLevel; 1022 if (!PPChainBranchIndex.empty()) 1023 PPChainBranchIndex.pop(); 1024 if (!PPStack.empty()) 1025 PPStack.pop_back(); 1026 } 1027 1028 void UnwrappedLineParser::parsePPIf(bool IfDef) { 1029 bool IfNDef = FormatTok->is(tok::pp_ifndef); 1030 nextToken(); 1031 bool Unreachable = false; 1032 if (!IfDef && (FormatTok->is(tok::kw_false) || FormatTok->TokenText == "0")) 1033 Unreachable = true; 1034 if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG") 1035 Unreachable = true; 1036 conditionalCompilationStart(Unreachable); 1037 FormatToken *IfCondition = FormatTok; 1038 // If there's a #ifndef on the first line, and the only lines before it are 1039 // comments, it could be an include guard. 1040 bool MaybeIncludeGuard = IfNDef; 1041 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) 1042 for (auto &Line : Lines) { 1043 if (!Line.Tokens.front().Tok->is(tok::comment)) { 1044 MaybeIncludeGuard = false; 1045 IncludeGuard = IG_Rejected; 1046 break; 1047 } 1048 } 1049 --PPBranchLevel; 1050 parsePPUnknown(); 1051 ++PPBranchLevel; 1052 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) { 1053 IncludeGuard = IG_IfNdefed; 1054 IncludeGuardToken = IfCondition; 1055 } 1056 } 1057 1058 void UnwrappedLineParser::parsePPElse() { 1059 // If a potential include guard has an #else, it's not an include guard. 1060 if (IncludeGuard == IG_Defined && PPBranchLevel == 0) 1061 IncludeGuard = IG_Rejected; 1062 conditionalCompilationAlternative(); 1063 if (PPBranchLevel > -1) 1064 --PPBranchLevel; 1065 parsePPUnknown(); 1066 ++PPBranchLevel; 1067 } 1068 1069 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 1070 1071 void UnwrappedLineParser::parsePPEndIf() { 1072 conditionalCompilationEnd(); 1073 parsePPUnknown(); 1074 // If the #endif of a potential include guard is the last thing in the file, 1075 // then we found an include guard. 1076 if (IncludeGuard == IG_Defined && PPBranchLevel == -1 && Tokens->isEOF() && 1077 Style.IndentPPDirectives != FormatStyle::PPDIS_None) 1078 IncludeGuard = IG_Found; 1079 } 1080 1081 void UnwrappedLineParser::parsePPDefine() { 1082 nextToken(); 1083 1084 if (!FormatTok->Tok.getIdentifierInfo()) { 1085 IncludeGuard = IG_Rejected; 1086 IncludeGuardToken = nullptr; 1087 parsePPUnknown(); 1088 return; 1089 } 1090 1091 if (IncludeGuard == IG_IfNdefed && 1092 IncludeGuardToken->TokenText == FormatTok->TokenText) { 1093 IncludeGuard = IG_Defined; 1094 IncludeGuardToken = nullptr; 1095 for (auto &Line : Lines) { 1096 if (!Line.Tokens.front().Tok->isOneOf(tok::comment, tok::hash)) { 1097 IncludeGuard = IG_Rejected; 1098 break; 1099 } 1100 } 1101 } 1102 1103 // In the context of a define, even keywords should be treated as normal 1104 // identifiers. Setting the kind to identifier is not enough, because we need 1105 // to treat additional keywords like __except as well, which are already 1106 // identifiers. Setting the identifier info to null interferes with include 1107 // guard processing above, and changes preprocessing nesting. 1108 FormatTok->Tok.setKind(tok::identifier); 1109 FormatTok->Tok.setIdentifierInfo(Keywords.kw_internal_ident_after_define); 1110 nextToken(); 1111 if (FormatTok->Tok.getKind() == tok::l_paren && 1112 !FormatTok->hasWhitespaceBefore()) 1113 parseParens(); 1114 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None) 1115 Line->Level += PPBranchLevel + 1; 1116 addUnwrappedLine(); 1117 ++Line->Level; 1118 1119 // Errors during a preprocessor directive can only affect the layout of the 1120 // preprocessor directive, and thus we ignore them. An alternative approach 1121 // would be to use the same approach we use on the file level (no 1122 // re-indentation if there was a structural error) within the macro 1123 // definition. 1124 parseFile(); 1125 } 1126 1127 void UnwrappedLineParser::parsePPUnknown() { 1128 do { 1129 nextToken(); 1130 } while (!eof()); 1131 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None) 1132 Line->Level += PPBranchLevel + 1; 1133 addUnwrappedLine(); 1134 } 1135 1136 // Here we exclude certain tokens that are not usually the first token in an 1137 // unwrapped line. This is used in attempt to distinguish macro calls without 1138 // trailing semicolons from other constructs split to several lines. 1139 static bool tokenCanStartNewLine(const FormatToken &Tok) { 1140 // Semicolon can be a null-statement, l_square can be a start of a macro or 1141 // a C++11 attribute, but this doesn't seem to be common. 1142 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 1143 Tok.isNot(TT_AttributeSquare) && 1144 // Tokens that can only be used as binary operators and a part of 1145 // overloaded operator names. 1146 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 1147 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 1148 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 1149 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 1150 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 1151 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 1152 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 1153 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 1154 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 1155 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 1156 Tok.isNot(tok::lesslessequal) && 1157 // Colon is used in labels, base class lists, initializer lists, 1158 // range-based for loops, ternary operator, but should never be the 1159 // first token in an unwrapped line. 1160 Tok.isNot(tok::colon) && 1161 // 'noexcept' is a trailing annotation. 1162 Tok.isNot(tok::kw_noexcept); 1163 } 1164 1165 static bool mustBeJSIdent(const AdditionalKeywords &Keywords, 1166 const FormatToken *FormatTok) { 1167 // FIXME: This returns true for C/C++ keywords like 'struct'. 1168 return FormatTok->is(tok::identifier) && 1169 (FormatTok->Tok.getIdentifierInfo() == nullptr || 1170 !FormatTok->isOneOf( 1171 Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async, 1172 Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally, 1173 Keywords.kw_function, Keywords.kw_import, Keywords.kw_is, 1174 Keywords.kw_let, Keywords.kw_var, tok::kw_const, 1175 Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements, 1176 Keywords.kw_instanceof, Keywords.kw_interface, 1177 Keywords.kw_override, Keywords.kw_throws, Keywords.kw_from)); 1178 } 1179 1180 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords, 1181 const FormatToken *FormatTok) { 1182 return FormatTok->Tok.isLiteral() || 1183 FormatTok->isOneOf(tok::kw_true, tok::kw_false) || 1184 mustBeJSIdent(Keywords, FormatTok); 1185 } 1186 1187 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement 1188 // when encountered after a value (see mustBeJSIdentOrValue). 1189 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords, 1190 const FormatToken *FormatTok) { 1191 return FormatTok->isOneOf( 1192 tok::kw_return, Keywords.kw_yield, 1193 // conditionals 1194 tok::kw_if, tok::kw_else, 1195 // loops 1196 tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break, 1197 // switch/case 1198 tok::kw_switch, tok::kw_case, 1199 // exceptions 1200 tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally, 1201 // declaration 1202 tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let, 1203 Keywords.kw_async, Keywords.kw_function, 1204 // import/export 1205 Keywords.kw_import, tok::kw_export); 1206 } 1207 1208 // Checks whether a token is a type in K&R C (aka C78). 1209 static bool isC78Type(const FormatToken &Tok) { 1210 return Tok.isOneOf(tok::kw_char, tok::kw_short, tok::kw_int, tok::kw_long, 1211 tok::kw_unsigned, tok::kw_float, tok::kw_double, 1212 tok::identifier); 1213 } 1214 1215 // This function checks whether a token starts the first parameter declaration 1216 // in a K&R C (aka C78) function definition, e.g.: 1217 // int f(a, b) 1218 // short a, b; 1219 // { 1220 // return a + b; 1221 // } 1222 static bool isC78ParameterDecl(const FormatToken *Tok, const FormatToken *Next, 1223 const FormatToken *FuncName) { 1224 assert(Tok); 1225 assert(Next); 1226 assert(FuncName); 1227 1228 if (FuncName->isNot(tok::identifier)) 1229 return false; 1230 1231 const FormatToken *Prev = FuncName->Previous; 1232 if (!Prev || (Prev->isNot(tok::star) && !isC78Type(*Prev))) 1233 return false; 1234 1235 if (!isC78Type(*Tok) && 1236 !Tok->isOneOf(tok::kw_register, tok::kw_struct, tok::kw_union)) 1237 return false; 1238 1239 if (Next->isNot(tok::star) && !Next->Tok.getIdentifierInfo()) 1240 return false; 1241 1242 Tok = Tok->Previous; 1243 if (!Tok || Tok->isNot(tok::r_paren)) 1244 return false; 1245 1246 Tok = Tok->Previous; 1247 if (!Tok || Tok->isNot(tok::identifier)) 1248 return false; 1249 1250 return Tok->Previous && Tok->Previous->isOneOf(tok::l_paren, tok::comma); 1251 } 1252 1253 void UnwrappedLineParser::parseModuleImport() { 1254 nextToken(); 1255 while (!eof()) { 1256 if (FormatTok->is(tok::colon)) { 1257 FormatTok->setFinalizedType(TT_ModulePartitionColon); 1258 } 1259 // Handle import <foo/bar.h> as we would an include statement. 1260 else if (FormatTok->is(tok::less)) { 1261 nextToken(); 1262 while (!FormatTok->isOneOf(tok::semi, tok::greater, tok::eof)) { 1263 // Mark tokens up to the trailing line comments as implicit string 1264 // literals. 1265 if (FormatTok->isNot(tok::comment) && 1266 !FormatTok->TokenText.startswith("//")) 1267 FormatTok->setFinalizedType(TT_ImplicitStringLiteral); 1268 nextToken(); 1269 } 1270 } 1271 if (FormatTok->is(tok::semi)) { 1272 nextToken(); 1273 break; 1274 } 1275 nextToken(); 1276 } 1277 1278 addUnwrappedLine(); 1279 } 1280 1281 // readTokenWithJavaScriptASI reads the next token and terminates the current 1282 // line if JavaScript Automatic Semicolon Insertion must 1283 // happen between the current token and the next token. 1284 // 1285 // This method is conservative - it cannot cover all edge cases of JavaScript, 1286 // but only aims to correctly handle certain well known cases. It *must not* 1287 // return true in speculative cases. 1288 void UnwrappedLineParser::readTokenWithJavaScriptASI() { 1289 FormatToken *Previous = FormatTok; 1290 readToken(); 1291 FormatToken *Next = FormatTok; 1292 1293 bool IsOnSameLine = 1294 CommentsBeforeNextToken.empty() 1295 ? Next->NewlinesBefore == 0 1296 : CommentsBeforeNextToken.front()->NewlinesBefore == 0; 1297 if (IsOnSameLine) 1298 return; 1299 1300 bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous); 1301 bool PreviousStartsTemplateExpr = 1302 Previous->is(TT_TemplateString) && Previous->TokenText.endswith("${"); 1303 if (PreviousMustBeValue || Previous->is(tok::r_paren)) { 1304 // If the line contains an '@' sign, the previous token might be an 1305 // annotation, which can precede another identifier/value. 1306 bool HasAt = llvm::any_of(Line->Tokens, [](UnwrappedLineNode &LineNode) { 1307 return LineNode.Tok->is(tok::at); 1308 }); 1309 if (HasAt) 1310 return; 1311 } 1312 if (Next->is(tok::exclaim) && PreviousMustBeValue) 1313 return addUnwrappedLine(); 1314 bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next); 1315 bool NextEndsTemplateExpr = 1316 Next->is(TT_TemplateString) && Next->TokenText.startswith("}"); 1317 if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr && 1318 (PreviousMustBeValue || 1319 Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus, 1320 tok::minusminus))) 1321 return addUnwrappedLine(); 1322 if ((PreviousMustBeValue || Previous->is(tok::r_paren)) && 1323 isJSDeclOrStmt(Keywords, Next)) 1324 return addUnwrappedLine(); 1325 } 1326 1327 void UnwrappedLineParser::parseStructuralElement(IfStmtKind *IfKind, 1328 bool IsTopLevel, 1329 TokenType NextLBracesType, 1330 bool *HasLabel) { 1331 if (Style.Language == FormatStyle::LK_TableGen && 1332 FormatTok->is(tok::pp_include)) { 1333 nextToken(); 1334 if (FormatTok->is(tok::string_literal)) 1335 nextToken(); 1336 addUnwrappedLine(); 1337 return; 1338 } 1339 switch (FormatTok->Tok.getKind()) { 1340 case tok::kw_asm: 1341 nextToken(); 1342 if (FormatTok->is(tok::l_brace)) { 1343 FormatTok->setFinalizedType(TT_InlineASMBrace); 1344 nextToken(); 1345 while (FormatTok && FormatTok->isNot(tok::eof)) { 1346 if (FormatTok->is(tok::r_brace)) { 1347 FormatTok->setFinalizedType(TT_InlineASMBrace); 1348 nextToken(); 1349 addUnwrappedLine(); 1350 break; 1351 } 1352 FormatTok->Finalized = true; 1353 nextToken(); 1354 } 1355 } 1356 break; 1357 case tok::kw_namespace: 1358 parseNamespace(); 1359 return; 1360 case tok::kw_public: 1361 case tok::kw_protected: 1362 case tok::kw_private: 1363 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 1364 Style.isCSharp()) 1365 nextToken(); 1366 else 1367 parseAccessSpecifier(); 1368 return; 1369 case tok::kw_if: 1370 if (Style.isJavaScript() && Line->MustBeDeclaration) 1371 // field/method declaration. 1372 break; 1373 parseIfThenElse(IfKind); 1374 return; 1375 case tok::kw_for: 1376 case tok::kw_while: 1377 if (Style.isJavaScript() && Line->MustBeDeclaration) 1378 // field/method declaration. 1379 break; 1380 parseForOrWhileLoop(); 1381 return; 1382 case tok::kw_do: 1383 if (Style.isJavaScript() && Line->MustBeDeclaration) 1384 // field/method declaration. 1385 break; 1386 parseDoWhile(); 1387 return; 1388 case tok::kw_switch: 1389 if (Style.isJavaScript() && Line->MustBeDeclaration) 1390 // 'switch: string' field declaration. 1391 break; 1392 parseSwitch(); 1393 return; 1394 case tok::kw_default: 1395 if (Style.isJavaScript() && Line->MustBeDeclaration) 1396 // 'default: string' field declaration. 1397 break; 1398 nextToken(); 1399 if (FormatTok->is(tok::colon)) { 1400 parseLabel(); 1401 return; 1402 } 1403 // e.g. "default void f() {}" in a Java interface. 1404 break; 1405 case tok::kw_case: 1406 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1407 // 'case: string' field declaration. 1408 nextToken(); 1409 break; 1410 } 1411 parseCaseLabel(); 1412 return; 1413 case tok::kw_try: 1414 case tok::kw___try: 1415 if (Style.isJavaScript() && Line->MustBeDeclaration) 1416 // field/method declaration. 1417 break; 1418 parseTryCatch(); 1419 return; 1420 case tok::kw_extern: 1421 nextToken(); 1422 if (FormatTok->is(tok::string_literal)) { 1423 nextToken(); 1424 if (FormatTok->is(tok::l_brace)) { 1425 if (Style.BraceWrapping.AfterExternBlock) 1426 addUnwrappedLine(); 1427 // Either we indent or for backwards compatibility we follow the 1428 // AfterExternBlock style. 1429 unsigned AddLevels = 1430 (Style.IndentExternBlock == FormatStyle::IEBS_Indent) || 1431 (Style.BraceWrapping.AfterExternBlock && 1432 Style.IndentExternBlock == 1433 FormatStyle::IEBS_AfterExternBlock) 1434 ? 1u 1435 : 0u; 1436 parseBlock(/*MustBeDeclaration=*/true, AddLevels); 1437 addUnwrappedLine(); 1438 return; 1439 } 1440 } 1441 break; 1442 case tok::kw_export: 1443 if (Style.isJavaScript()) { 1444 parseJavaScriptEs6ImportExport(); 1445 return; 1446 } 1447 if (!Style.isCpp()) 1448 break; 1449 // Handle C++ "(inline|export) namespace". 1450 LLVM_FALLTHROUGH; 1451 case tok::kw_inline: 1452 nextToken(); 1453 if (FormatTok->is(tok::kw_namespace)) { 1454 parseNamespace(); 1455 return; 1456 } 1457 break; 1458 case tok::identifier: 1459 if (FormatTok->is(TT_ForEachMacro)) { 1460 parseForOrWhileLoop(); 1461 return; 1462 } 1463 if (FormatTok->is(TT_MacroBlockBegin)) { 1464 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 1465 /*MunchSemi=*/false); 1466 return; 1467 } 1468 if (FormatTok->is(Keywords.kw_import)) { 1469 if (Style.isJavaScript()) { 1470 parseJavaScriptEs6ImportExport(); 1471 return; 1472 } 1473 if (Style.Language == FormatStyle::LK_Proto) { 1474 nextToken(); 1475 if (FormatTok->is(tok::kw_public)) 1476 nextToken(); 1477 if (!FormatTok->is(tok::string_literal)) 1478 return; 1479 nextToken(); 1480 if (FormatTok->is(tok::semi)) 1481 nextToken(); 1482 addUnwrappedLine(); 1483 return; 1484 } 1485 if (Style.isCpp()) { 1486 parseModuleImport(); 1487 return; 1488 } 1489 } 1490 if (Style.isCpp() && 1491 FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 1492 Keywords.kw_slots, Keywords.kw_qslots)) { 1493 nextToken(); 1494 if (FormatTok->is(tok::colon)) { 1495 nextToken(); 1496 addUnwrappedLine(); 1497 return; 1498 } 1499 } 1500 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1501 parseStatementMacro(); 1502 return; 1503 } 1504 if (Style.isCpp() && FormatTok->is(TT_NamespaceMacro)) { 1505 parseNamespace(); 1506 return; 1507 } 1508 // In all other cases, parse the declaration. 1509 break; 1510 default: 1511 break; 1512 } 1513 do { 1514 const FormatToken *Previous = FormatTok->Previous; 1515 switch (FormatTok->Tok.getKind()) { 1516 case tok::at: 1517 nextToken(); 1518 if (FormatTok->is(tok::l_brace)) { 1519 nextToken(); 1520 parseBracedList(); 1521 break; 1522 } else if (Style.Language == FormatStyle::LK_Java && 1523 FormatTok->is(Keywords.kw_interface)) { 1524 nextToken(); 1525 break; 1526 } 1527 switch (FormatTok->Tok.getObjCKeywordID()) { 1528 case tok::objc_public: 1529 case tok::objc_protected: 1530 case tok::objc_package: 1531 case tok::objc_private: 1532 return parseAccessSpecifier(); 1533 case tok::objc_interface: 1534 case tok::objc_implementation: 1535 return parseObjCInterfaceOrImplementation(); 1536 case tok::objc_protocol: 1537 if (parseObjCProtocol()) 1538 return; 1539 break; 1540 case tok::objc_end: 1541 return; // Handled by the caller. 1542 case tok::objc_optional: 1543 case tok::objc_required: 1544 nextToken(); 1545 addUnwrappedLine(); 1546 return; 1547 case tok::objc_autoreleasepool: 1548 nextToken(); 1549 if (FormatTok->is(tok::l_brace)) { 1550 if (Style.BraceWrapping.AfterControlStatement == 1551 FormatStyle::BWACS_Always) 1552 addUnwrappedLine(); 1553 parseBlock(); 1554 } 1555 addUnwrappedLine(); 1556 return; 1557 case tok::objc_synchronized: 1558 nextToken(); 1559 if (FormatTok->is(tok::l_paren)) 1560 // Skip synchronization object 1561 parseParens(); 1562 if (FormatTok->is(tok::l_brace)) { 1563 if (Style.BraceWrapping.AfterControlStatement == 1564 FormatStyle::BWACS_Always) 1565 addUnwrappedLine(); 1566 parseBlock(); 1567 } 1568 addUnwrappedLine(); 1569 return; 1570 case tok::objc_try: 1571 // This branch isn't strictly necessary (the kw_try case below would 1572 // do this too after the tok::at is parsed above). But be explicit. 1573 parseTryCatch(); 1574 return; 1575 default: 1576 break; 1577 } 1578 break; 1579 case tok::kw_concept: 1580 parseConcept(); 1581 return; 1582 case tok::kw_requires: { 1583 if (Style.isCpp()) { 1584 bool ParsedClause = parseRequires(); 1585 if (ParsedClause) 1586 return; 1587 } else { 1588 nextToken(); 1589 } 1590 break; 1591 } 1592 case tok::kw_enum: 1593 // Ignore if this is part of "template <enum ...". 1594 if (Previous && Previous->is(tok::less)) { 1595 nextToken(); 1596 break; 1597 } 1598 1599 // parseEnum falls through and does not yet add an unwrapped line as an 1600 // enum definition can start a structural element. 1601 if (!parseEnum()) 1602 break; 1603 // This only applies for C++. 1604 if (!Style.isCpp()) { 1605 addUnwrappedLine(); 1606 return; 1607 } 1608 break; 1609 case tok::kw_typedef: 1610 nextToken(); 1611 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 1612 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS, 1613 Keywords.kw_CF_CLOSED_ENUM, 1614 Keywords.kw_NS_CLOSED_ENUM)) 1615 parseEnum(); 1616 break; 1617 case tok::kw_struct: 1618 case tok::kw_union: 1619 case tok::kw_class: 1620 if (parseStructLike()) 1621 return; 1622 break; 1623 case tok::period: 1624 nextToken(); 1625 // In Java, classes have an implicit static member "class". 1626 if (Style.Language == FormatStyle::LK_Java && FormatTok && 1627 FormatTok->is(tok::kw_class)) 1628 nextToken(); 1629 if (Style.isJavaScript() && FormatTok && 1630 FormatTok->Tok.getIdentifierInfo()) 1631 // JavaScript only has pseudo keywords, all keywords are allowed to 1632 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 1633 nextToken(); 1634 break; 1635 case tok::semi: 1636 nextToken(); 1637 addUnwrappedLine(); 1638 return; 1639 case tok::r_brace: 1640 addUnwrappedLine(); 1641 return; 1642 case tok::l_paren: { 1643 parseParens(); 1644 // Break the unwrapped line if a K&R C function definition has a parameter 1645 // declaration. 1646 if (!IsTopLevel || !Style.isCpp() || !Previous || FormatTok->is(tok::eof)) 1647 break; 1648 if (isC78ParameterDecl(FormatTok, Tokens->peekNextToken(), Previous)) { 1649 addUnwrappedLine(); 1650 return; 1651 } 1652 break; 1653 } 1654 case tok::kw_operator: 1655 nextToken(); 1656 if (FormatTok->isBinaryOperator()) 1657 nextToken(); 1658 break; 1659 case tok::caret: 1660 nextToken(); 1661 if (FormatTok->Tok.isAnyIdentifier() || 1662 FormatTok->isSimpleTypeSpecifier()) 1663 nextToken(); 1664 if (FormatTok->is(tok::l_paren)) 1665 parseParens(); 1666 if (FormatTok->is(tok::l_brace)) 1667 parseChildBlock(); 1668 break; 1669 case tok::l_brace: 1670 if (NextLBracesType != TT_Unknown) 1671 FormatTok->setFinalizedType(NextLBracesType); 1672 if (!tryToParsePropertyAccessor() && !tryToParseBracedList()) { 1673 // A block outside of parentheses must be the last part of a 1674 // structural element. 1675 // FIXME: Figure out cases where this is not true, and add projections 1676 // for them (the one we know is missing are lambdas). 1677 if (Style.Language == FormatStyle::LK_Java && 1678 Line->Tokens.front().Tok->is(Keywords.kw_synchronized)) { 1679 // If necessary, we could set the type to something different than 1680 // TT_FunctionLBrace. 1681 if (Style.BraceWrapping.AfterControlStatement == 1682 FormatStyle::BWACS_Always) 1683 addUnwrappedLine(); 1684 } else if (Style.BraceWrapping.AfterFunction) { 1685 addUnwrappedLine(); 1686 } 1687 if (!Line->InPPDirective) 1688 FormatTok->setFinalizedType(TT_FunctionLBrace); 1689 parseBlock(); 1690 addUnwrappedLine(); 1691 return; 1692 } 1693 // Otherwise this was a braced init list, and the structural 1694 // element continues. 1695 break; 1696 case tok::kw_try: 1697 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1698 // field/method declaration. 1699 nextToken(); 1700 break; 1701 } 1702 // We arrive here when parsing function-try blocks. 1703 if (Style.BraceWrapping.AfterFunction) 1704 addUnwrappedLine(); 1705 parseTryCatch(); 1706 return; 1707 case tok::identifier: { 1708 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where) && 1709 Line->MustBeDeclaration) { 1710 addUnwrappedLine(); 1711 parseCSharpGenericTypeConstraint(); 1712 break; 1713 } 1714 if (FormatTok->is(TT_MacroBlockEnd)) { 1715 addUnwrappedLine(); 1716 return; 1717 } 1718 1719 // Function declarations (as opposed to function expressions) are parsed 1720 // on their own unwrapped line by continuing this loop. Function 1721 // expressions (functions that are not on their own line) must not create 1722 // a new unwrapped line, so they are special cased below. 1723 size_t TokenCount = Line->Tokens.size(); 1724 if (Style.isJavaScript() && FormatTok->is(Keywords.kw_function) && 1725 (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is( 1726 Keywords.kw_async)))) { 1727 tryToParseJSFunction(); 1728 break; 1729 } 1730 if ((Style.isJavaScript() || Style.Language == FormatStyle::LK_Java) && 1731 FormatTok->is(Keywords.kw_interface)) { 1732 if (Style.isJavaScript()) { 1733 // In JavaScript/TypeScript, "interface" can be used as a standalone 1734 // identifier, e.g. in `var interface = 1;`. If "interface" is 1735 // followed by another identifier, it is very like to be an actual 1736 // interface declaration. 1737 unsigned StoredPosition = Tokens->getPosition(); 1738 FormatToken *Next = Tokens->getNextToken(); 1739 FormatTok = Tokens->setPosition(StoredPosition); 1740 if (!mustBeJSIdent(Keywords, Next)) { 1741 nextToken(); 1742 break; 1743 } 1744 } 1745 parseRecord(); 1746 addUnwrappedLine(); 1747 return; 1748 } 1749 1750 if (FormatTok->is(Keywords.kw_interface)) { 1751 if (parseStructLike()) 1752 return; 1753 break; 1754 } 1755 1756 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1757 parseStatementMacro(); 1758 return; 1759 } 1760 1761 // See if the following token should start a new unwrapped line. 1762 StringRef Text = FormatTok->TokenText; 1763 1764 FormatToken *PreviousToken = FormatTok; 1765 nextToken(); 1766 1767 // JS doesn't have macros, and within classes colons indicate fields, not 1768 // labels. 1769 if (Style.isJavaScript()) 1770 break; 1771 1772 TokenCount = Line->Tokens.size(); 1773 if (TokenCount == 1 || 1774 (TokenCount == 2 && Line->Tokens.front().Tok->is(tok::comment))) { 1775 if (FormatTok->is(tok::colon) && !Line->MustBeDeclaration) { 1776 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1777 parseLabel(!Style.IndentGotoLabels); 1778 if (HasLabel) 1779 *HasLabel = true; 1780 return; 1781 } 1782 // Recognize function-like macro usages without trailing semicolon as 1783 // well as free-standing macros like Q_OBJECT. 1784 bool FunctionLike = FormatTok->is(tok::l_paren); 1785 if (FunctionLike) 1786 parseParens(); 1787 1788 bool FollowedByNewline = 1789 CommentsBeforeNextToken.empty() 1790 ? FormatTok->NewlinesBefore > 0 1791 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1792 1793 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1794 tokenCanStartNewLine(*FormatTok) && Text == Text.upper()) { 1795 PreviousToken->setFinalizedType(TT_FunctionLikeOrFreestandingMacro); 1796 addUnwrappedLine(); 1797 return; 1798 } 1799 } 1800 break; 1801 } 1802 case tok::equal: 1803 if ((Style.isJavaScript() || Style.isCSharp()) && 1804 FormatTok->is(TT_FatArrow)) { 1805 tryToParseChildBlock(); 1806 break; 1807 } 1808 1809 nextToken(); 1810 if (FormatTok->is(tok::l_brace)) { 1811 // Block kind should probably be set to BK_BracedInit for any language. 1812 // C# needs this change to ensure that array initialisers and object 1813 // initialisers are indented the same way. 1814 if (Style.isCSharp()) 1815 FormatTok->setBlockKind(BK_BracedInit); 1816 nextToken(); 1817 parseBracedList(); 1818 } else if (Style.Language == FormatStyle::LK_Proto && 1819 FormatTok->is(tok::less)) { 1820 nextToken(); 1821 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 1822 /*ClosingBraceKind=*/tok::greater); 1823 } 1824 break; 1825 case tok::l_square: 1826 parseSquare(); 1827 break; 1828 case tok::kw_new: 1829 parseNew(); 1830 break; 1831 case tok::kw_case: 1832 if (Style.isJavaScript() && Line->MustBeDeclaration) 1833 // 'case: string' field declaration. 1834 break; 1835 parseCaseLabel(); 1836 break; 1837 default: 1838 nextToken(); 1839 break; 1840 } 1841 } while (!eof()); 1842 } 1843 1844 bool UnwrappedLineParser::tryToParsePropertyAccessor() { 1845 assert(FormatTok->is(tok::l_brace)); 1846 if (!Style.isCSharp()) 1847 return false; 1848 // See if it's a property accessor. 1849 if (FormatTok->Previous->isNot(tok::identifier)) 1850 return false; 1851 1852 // See if we are inside a property accessor. 1853 // 1854 // Record the current tokenPosition so that we can advance and 1855 // reset the current token. `Next` is not set yet so we need 1856 // another way to advance along the token stream. 1857 unsigned int StoredPosition = Tokens->getPosition(); 1858 FormatToken *Tok = Tokens->getNextToken(); 1859 1860 // A trivial property accessor is of the form: 1861 // { [ACCESS_SPECIFIER] [get]; [ACCESS_SPECIFIER] [set|init] } 1862 // Track these as they do not require line breaks to be introduced. 1863 bool HasSpecialAccessor = false; 1864 bool IsTrivialPropertyAccessor = true; 1865 while (!eof()) { 1866 if (Tok->isOneOf(tok::semi, tok::kw_public, tok::kw_private, 1867 tok::kw_protected, Keywords.kw_internal, Keywords.kw_get, 1868 Keywords.kw_init, Keywords.kw_set)) { 1869 if (Tok->isOneOf(Keywords.kw_get, Keywords.kw_init, Keywords.kw_set)) 1870 HasSpecialAccessor = true; 1871 Tok = Tokens->getNextToken(); 1872 continue; 1873 } 1874 if (Tok->isNot(tok::r_brace)) 1875 IsTrivialPropertyAccessor = false; 1876 break; 1877 } 1878 1879 if (!HasSpecialAccessor) { 1880 Tokens->setPosition(StoredPosition); 1881 return false; 1882 } 1883 1884 // Try to parse the property accessor: 1885 // https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties 1886 Tokens->setPosition(StoredPosition); 1887 if (!IsTrivialPropertyAccessor && Style.BraceWrapping.AfterFunction) 1888 addUnwrappedLine(); 1889 nextToken(); 1890 do { 1891 switch (FormatTok->Tok.getKind()) { 1892 case tok::r_brace: 1893 nextToken(); 1894 if (FormatTok->is(tok::equal)) { 1895 while (!eof() && FormatTok->isNot(tok::semi)) 1896 nextToken(); 1897 nextToken(); 1898 } 1899 addUnwrappedLine(); 1900 return true; 1901 case tok::l_brace: 1902 ++Line->Level; 1903 parseBlock(/*MustBeDeclaration=*/true); 1904 addUnwrappedLine(); 1905 --Line->Level; 1906 break; 1907 case tok::equal: 1908 if (FormatTok->is(TT_FatArrow)) { 1909 ++Line->Level; 1910 do { 1911 nextToken(); 1912 } while (!eof() && FormatTok->isNot(tok::semi)); 1913 nextToken(); 1914 addUnwrappedLine(); 1915 --Line->Level; 1916 break; 1917 } 1918 nextToken(); 1919 break; 1920 default: 1921 if (FormatTok->isOneOf(Keywords.kw_get, Keywords.kw_init, 1922 Keywords.kw_set) && 1923 !IsTrivialPropertyAccessor) { 1924 // Non-trivial get/set needs to be on its own line. 1925 addUnwrappedLine(); 1926 } 1927 nextToken(); 1928 } 1929 } while (!eof()); 1930 1931 // Unreachable for well-formed code (paired '{' and '}'). 1932 return true; 1933 } 1934 1935 bool UnwrappedLineParser::tryToParseLambda() { 1936 if (!Style.isCpp()) { 1937 nextToken(); 1938 return false; 1939 } 1940 assert(FormatTok->is(tok::l_square)); 1941 FormatToken &LSquare = *FormatTok; 1942 if (!tryToParseLambdaIntroducer()) 1943 return false; 1944 1945 // `[something] >` is not a lambda, but an array type in a template parameter 1946 // list. 1947 if (FormatTok->is(tok::greater)) 1948 return false; 1949 1950 bool SeenArrow = false; 1951 bool InTemplateParameterList = false; 1952 1953 while (FormatTok->isNot(tok::l_brace)) { 1954 if (FormatTok->isSimpleTypeSpecifier()) { 1955 nextToken(); 1956 continue; 1957 } 1958 switch (FormatTok->Tok.getKind()) { 1959 case tok::l_brace: 1960 break; 1961 case tok::l_paren: 1962 parseParens(); 1963 break; 1964 case tok::l_square: 1965 parseSquare(); 1966 break; 1967 case tok::kw_class: 1968 case tok::kw_template: 1969 case tok::kw_typename: 1970 assert(FormatTok->Previous); 1971 if (FormatTok->Previous->is(tok::less)) 1972 InTemplateParameterList = true; 1973 nextToken(); 1974 break; 1975 case tok::amp: 1976 case tok::star: 1977 case tok::kw_const: 1978 case tok::comma: 1979 case tok::less: 1980 case tok::greater: 1981 case tok::identifier: 1982 case tok::numeric_constant: 1983 case tok::coloncolon: 1984 case tok::kw_mutable: 1985 case tok::kw_noexcept: 1986 nextToken(); 1987 break; 1988 // Specialization of a template with an integer parameter can contain 1989 // arithmetic, logical, comparison and ternary operators. 1990 // 1991 // FIXME: This also accepts sequences of operators that are not in the scope 1992 // of a template argument list. 1993 // 1994 // In a C++ lambda a template type can only occur after an arrow. We use 1995 // this as an heuristic to distinguish between Objective-C expressions 1996 // followed by an `a->b` expression, such as: 1997 // ([obj func:arg] + a->b) 1998 // Otherwise the code below would parse as a lambda. 1999 // 2000 // FIXME: This heuristic is incorrect for C++20 generic lambdas with 2001 // explicit template lists: []<bool b = true && false>(U &&u){} 2002 case tok::plus: 2003 case tok::minus: 2004 case tok::exclaim: 2005 case tok::tilde: 2006 case tok::slash: 2007 case tok::percent: 2008 case tok::lessless: 2009 case tok::pipe: 2010 case tok::pipepipe: 2011 case tok::ampamp: 2012 case tok::caret: 2013 case tok::equalequal: 2014 case tok::exclaimequal: 2015 case tok::greaterequal: 2016 case tok::lessequal: 2017 case tok::question: 2018 case tok::colon: 2019 case tok::ellipsis: 2020 case tok::kw_true: 2021 case tok::kw_false: 2022 if (SeenArrow || InTemplateParameterList) { 2023 nextToken(); 2024 break; 2025 } 2026 return true; 2027 case tok::arrow: 2028 // This might or might not actually be a lambda arrow (this could be an 2029 // ObjC method invocation followed by a dereferencing arrow). We might 2030 // reset this back to TT_Unknown in TokenAnnotator. 2031 FormatTok->setFinalizedType(TT_LambdaArrow); 2032 SeenArrow = true; 2033 nextToken(); 2034 break; 2035 default: 2036 return true; 2037 } 2038 } 2039 FormatTok->setFinalizedType(TT_LambdaLBrace); 2040 LSquare.setFinalizedType(TT_LambdaLSquare); 2041 parseChildBlock(); 2042 return true; 2043 } 2044 2045 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 2046 const FormatToken *Previous = FormatTok->Previous; 2047 if (Previous && 2048 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 2049 tok::kw_delete, tok::l_square) || 2050 FormatTok->isCppStructuredBinding(Style) || Previous->closesScope() || 2051 Previous->isSimpleTypeSpecifier())) { 2052 nextToken(); 2053 return false; 2054 } 2055 nextToken(); 2056 if (FormatTok->is(tok::l_square)) 2057 return false; 2058 parseSquare(/*LambdaIntroducer=*/true); 2059 return true; 2060 } 2061 2062 void UnwrappedLineParser::tryToParseJSFunction() { 2063 assert(FormatTok->is(Keywords.kw_function) || 2064 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)); 2065 if (FormatTok->is(Keywords.kw_async)) 2066 nextToken(); 2067 // Consume "function". 2068 nextToken(); 2069 2070 // Consume * (generator function). Treat it like C++'s overloaded operators. 2071 if (FormatTok->is(tok::star)) { 2072 FormatTok->setFinalizedType(TT_OverloadedOperator); 2073 nextToken(); 2074 } 2075 2076 // Consume function name. 2077 if (FormatTok->is(tok::identifier)) 2078 nextToken(); 2079 2080 if (FormatTok->isNot(tok::l_paren)) 2081 return; 2082 2083 // Parse formal parameter list. 2084 parseParens(); 2085 2086 if (FormatTok->is(tok::colon)) { 2087 // Parse a type definition. 2088 nextToken(); 2089 2090 // Eat the type declaration. For braced inline object types, balance braces, 2091 // otherwise just parse until finding an l_brace for the function body. 2092 if (FormatTok->is(tok::l_brace)) 2093 tryToParseBracedList(); 2094 else 2095 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 2096 nextToken(); 2097 } 2098 2099 if (FormatTok->is(tok::semi)) 2100 return; 2101 2102 parseChildBlock(); 2103 } 2104 2105 bool UnwrappedLineParser::tryToParseBracedList() { 2106 if (FormatTok->is(BK_Unknown)) 2107 calculateBraceTypes(); 2108 assert(FormatTok->isNot(BK_Unknown)); 2109 if (FormatTok->is(BK_Block)) 2110 return false; 2111 nextToken(); 2112 parseBracedList(); 2113 return true; 2114 } 2115 2116 bool UnwrappedLineParser::tryToParseChildBlock() { 2117 assert(Style.isJavaScript() || Style.isCSharp()); 2118 assert(FormatTok->is(TT_FatArrow)); 2119 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType TT_FatArrow. 2120 // They always start an expression or a child block if followed by a curly 2121 // brace. 2122 nextToken(); 2123 if (FormatTok->isNot(tok::l_brace)) 2124 return false; 2125 parseChildBlock(); 2126 return true; 2127 } 2128 2129 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons, 2130 bool IsEnum, 2131 tok::TokenKind ClosingBraceKind) { 2132 bool HasError = false; 2133 2134 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 2135 // replace this by using parseAssignmentExpression() inside. 2136 do { 2137 if (Style.isCSharp() && FormatTok->is(TT_FatArrow) && 2138 tryToParseChildBlock()) 2139 continue; 2140 if (Style.isJavaScript()) { 2141 if (FormatTok->is(Keywords.kw_function) || 2142 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) { 2143 tryToParseJSFunction(); 2144 continue; 2145 } 2146 if (FormatTok->is(tok::l_brace)) { 2147 // Could be a method inside of a braced list `{a() { return 1; }}`. 2148 if (tryToParseBracedList()) 2149 continue; 2150 parseChildBlock(); 2151 } 2152 } 2153 if (FormatTok->Tok.getKind() == ClosingBraceKind) { 2154 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 2155 addUnwrappedLine(); 2156 nextToken(); 2157 return !HasError; 2158 } 2159 switch (FormatTok->Tok.getKind()) { 2160 case tok::l_square: 2161 if (Style.isCSharp()) 2162 parseSquare(); 2163 else 2164 tryToParseLambda(); 2165 break; 2166 case tok::l_paren: 2167 parseParens(); 2168 // JavaScript can just have free standing methods and getters/setters in 2169 // object literals. Detect them by a "{" following ")". 2170 if (Style.isJavaScript()) { 2171 if (FormatTok->is(tok::l_brace)) 2172 parseChildBlock(); 2173 break; 2174 } 2175 break; 2176 case tok::l_brace: 2177 // Assume there are no blocks inside a braced init list apart 2178 // from the ones we explicitly parse out (like lambdas). 2179 FormatTok->setBlockKind(BK_BracedInit); 2180 nextToken(); 2181 parseBracedList(); 2182 break; 2183 case tok::less: 2184 if (Style.Language == FormatStyle::LK_Proto) { 2185 nextToken(); 2186 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 2187 /*ClosingBraceKind=*/tok::greater); 2188 } else { 2189 nextToken(); 2190 } 2191 break; 2192 case tok::semi: 2193 // JavaScript (or more precisely TypeScript) can have semicolons in braced 2194 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 2195 // used for error recovery if we have otherwise determined that this is 2196 // a braced list. 2197 if (Style.isJavaScript()) { 2198 nextToken(); 2199 break; 2200 } 2201 HasError = true; 2202 if (!ContinueOnSemicolons) 2203 return !HasError; 2204 nextToken(); 2205 break; 2206 case tok::comma: 2207 nextToken(); 2208 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 2209 addUnwrappedLine(); 2210 break; 2211 default: 2212 nextToken(); 2213 break; 2214 } 2215 } while (!eof()); 2216 return false; 2217 } 2218 2219 /// \brief Parses a pair of parentheses (and everything between them). 2220 /// \param AmpAmpTokenType If different than TT_Unknown sets this type for all 2221 /// double ampersands. This only counts for the current parens scope. 2222 void UnwrappedLineParser::parseParens(TokenType AmpAmpTokenType) { 2223 assert(FormatTok->is(tok::l_paren) && "'(' expected."); 2224 nextToken(); 2225 do { 2226 switch (FormatTok->Tok.getKind()) { 2227 case tok::l_paren: 2228 parseParens(); 2229 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 2230 parseChildBlock(); 2231 break; 2232 case tok::r_paren: 2233 nextToken(); 2234 return; 2235 case tok::r_brace: 2236 // A "}" inside parenthesis is an error if there wasn't a matching "{". 2237 return; 2238 case tok::l_square: 2239 tryToParseLambda(); 2240 break; 2241 case tok::l_brace: 2242 if (!tryToParseBracedList()) 2243 parseChildBlock(); 2244 break; 2245 case tok::at: 2246 nextToken(); 2247 if (FormatTok->is(tok::l_brace)) { 2248 nextToken(); 2249 parseBracedList(); 2250 } 2251 break; 2252 case tok::equal: 2253 if (Style.isCSharp() && FormatTok->is(TT_FatArrow)) 2254 tryToParseChildBlock(); 2255 else 2256 nextToken(); 2257 break; 2258 case tok::kw_class: 2259 if (Style.isJavaScript()) 2260 parseRecord(/*ParseAsExpr=*/true); 2261 else 2262 nextToken(); 2263 break; 2264 case tok::identifier: 2265 if (Style.isJavaScript() && 2266 (FormatTok->is(Keywords.kw_function) || 2267 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function))) 2268 tryToParseJSFunction(); 2269 else 2270 nextToken(); 2271 break; 2272 case tok::kw_requires: { 2273 auto RequiresToken = FormatTok; 2274 nextToken(); 2275 parseRequiresExpression(RequiresToken); 2276 break; 2277 } 2278 case tok::ampamp: 2279 if (AmpAmpTokenType != TT_Unknown) 2280 FormatTok->setFinalizedType(AmpAmpTokenType); 2281 LLVM_FALLTHROUGH; 2282 default: 2283 nextToken(); 2284 break; 2285 } 2286 } while (!eof()); 2287 } 2288 2289 void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) { 2290 if (!LambdaIntroducer) { 2291 assert(FormatTok->is(tok::l_square) && "'[' expected."); 2292 if (tryToParseLambda()) 2293 return; 2294 } 2295 do { 2296 switch (FormatTok->Tok.getKind()) { 2297 case tok::l_paren: 2298 parseParens(); 2299 break; 2300 case tok::r_square: 2301 nextToken(); 2302 return; 2303 case tok::r_brace: 2304 // A "}" inside parenthesis is an error if there wasn't a matching "{". 2305 return; 2306 case tok::l_square: 2307 parseSquare(); 2308 break; 2309 case tok::l_brace: { 2310 if (!tryToParseBracedList()) 2311 parseChildBlock(); 2312 break; 2313 } 2314 case tok::at: 2315 nextToken(); 2316 if (FormatTok->is(tok::l_brace)) { 2317 nextToken(); 2318 parseBracedList(); 2319 } 2320 break; 2321 default: 2322 nextToken(); 2323 break; 2324 } 2325 } while (!eof()); 2326 } 2327 2328 void UnwrappedLineParser::keepAncestorBraces() { 2329 if (!Style.RemoveBracesLLVM) 2330 return; 2331 2332 const int MaxNestingLevels = 2; 2333 const int Size = NestedTooDeep.size(); 2334 if (Size >= MaxNestingLevels) 2335 NestedTooDeep[Size - MaxNestingLevels] = true; 2336 NestedTooDeep.push_back(false); 2337 } 2338 2339 static FormatToken *getLastNonComment(const UnwrappedLine &Line) { 2340 for (const auto &Token : llvm::reverse(Line.Tokens)) 2341 if (Token.Tok->isNot(tok::comment)) 2342 return Token.Tok; 2343 2344 return nullptr; 2345 } 2346 2347 void UnwrappedLineParser::parseUnbracedBody(bool CheckEOF) { 2348 FormatToken *Tok = nullptr; 2349 2350 if (Style.InsertBraces && !Line->InPPDirective && !Line->Tokens.empty() && 2351 PreprocessorDirectives.empty()) { 2352 Tok = getLastNonComment(*Line); 2353 assert(Tok); 2354 if (Tok->BraceCount < 0) { 2355 assert(Tok->BraceCount == -1); 2356 Tok = nullptr; 2357 } else { 2358 Tok->BraceCount = -1; 2359 } 2360 } 2361 2362 addUnwrappedLine(); 2363 ++Line->Level; 2364 parseStructuralElement(); 2365 2366 if (Tok) { 2367 assert(!Line->InPPDirective); 2368 Tok = nullptr; 2369 for (const auto &L : llvm::reverse(*CurrentLines)) { 2370 if (!L.InPPDirective && getLastNonComment(L)) { 2371 Tok = L.Tokens.back().Tok; 2372 break; 2373 } 2374 } 2375 assert(Tok); 2376 ++Tok->BraceCount; 2377 } 2378 2379 if (CheckEOF && FormatTok->is(tok::eof)) 2380 addUnwrappedLine(); 2381 2382 --Line->Level; 2383 } 2384 2385 static void markOptionalBraces(FormatToken *LeftBrace) { 2386 if (!LeftBrace) 2387 return; 2388 2389 assert(LeftBrace->is(tok::l_brace)); 2390 2391 FormatToken *RightBrace = LeftBrace->MatchingParen; 2392 if (!RightBrace) { 2393 assert(!LeftBrace->Optional); 2394 return; 2395 } 2396 2397 assert(RightBrace->is(tok::r_brace)); 2398 assert(RightBrace->MatchingParen == LeftBrace); 2399 assert(LeftBrace->Optional == RightBrace->Optional); 2400 2401 LeftBrace->Optional = true; 2402 RightBrace->Optional = true; 2403 } 2404 2405 void UnwrappedLineParser::handleAttributes() { 2406 // Handle AttributeMacro, e.g. `if (x) UNLIKELY`. 2407 if (FormatTok->is(TT_AttributeMacro)) 2408 nextToken(); 2409 handleCppAttributes(); 2410 } 2411 2412 bool UnwrappedLineParser::handleCppAttributes() { 2413 // Handle [[likely]] / [[unlikely]] attributes. 2414 if (FormatTok->is(tok::l_square) && tryToParseSimpleAttribute()) { 2415 parseSquare(); 2416 return true; 2417 } 2418 return false; 2419 } 2420 2421 FormatToken *UnwrappedLineParser::parseIfThenElse(IfStmtKind *IfKind, 2422 bool KeepBraces) { 2423 assert(FormatTok->is(tok::kw_if) && "'if' expected"); 2424 nextToken(); 2425 if (FormatTok->is(tok::exclaim)) 2426 nextToken(); 2427 if (FormatTok->is(tok::kw_consteval)) { 2428 nextToken(); 2429 } else { 2430 if (FormatTok->isOneOf(tok::kw_constexpr, tok::identifier)) 2431 nextToken(); 2432 if (FormatTok->is(tok::l_paren)) 2433 parseParens(); 2434 } 2435 handleAttributes(); 2436 2437 bool NeedsUnwrappedLine = false; 2438 keepAncestorBraces(); 2439 2440 FormatToken *IfLeftBrace = nullptr; 2441 IfStmtKind IfBlockKind = IfStmtKind::NotIf; 2442 2443 if (FormatTok->is(tok::l_brace)) { 2444 IfLeftBrace = FormatTok; 2445 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2446 IfBlockKind = parseBlock(); 2447 if (Style.BraceWrapping.BeforeElse) 2448 addUnwrappedLine(); 2449 else 2450 NeedsUnwrappedLine = true; 2451 } else { 2452 parseUnbracedBody(); 2453 } 2454 2455 bool KeepIfBraces = false; 2456 if (Style.RemoveBracesLLVM) { 2457 assert(!NestedTooDeep.empty()); 2458 KeepIfBraces = (IfLeftBrace && !IfLeftBrace->MatchingParen) || 2459 NestedTooDeep.back() || IfBlockKind == IfStmtKind::IfOnly || 2460 IfBlockKind == IfStmtKind::IfElseIf; 2461 } 2462 2463 FormatToken *ElseLeftBrace = nullptr; 2464 IfStmtKind Kind = IfStmtKind::IfOnly; 2465 2466 if (FormatTok->is(tok::kw_else)) { 2467 if (Style.RemoveBracesLLVM) { 2468 NestedTooDeep.back() = false; 2469 Kind = IfStmtKind::IfElse; 2470 } 2471 nextToken(); 2472 handleAttributes(); 2473 if (FormatTok->is(tok::l_brace)) { 2474 ElseLeftBrace = FormatTok; 2475 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2476 if (parseBlock() == IfStmtKind::IfOnly) 2477 Kind = IfStmtKind::IfElseIf; 2478 addUnwrappedLine(); 2479 } else if (FormatTok->is(tok::kw_if)) { 2480 FormatToken *Previous = Tokens->getPreviousToken(); 2481 const bool IsPrecededByComment = Previous && Previous->is(tok::comment); 2482 if (IsPrecededByComment) { 2483 addUnwrappedLine(); 2484 ++Line->Level; 2485 } 2486 bool TooDeep = true; 2487 if (Style.RemoveBracesLLVM) { 2488 Kind = IfStmtKind::IfElseIf; 2489 TooDeep = NestedTooDeep.pop_back_val(); 2490 } 2491 ElseLeftBrace = 2492 parseIfThenElse(/*IfKind=*/nullptr, KeepBraces || KeepIfBraces); 2493 if (Style.RemoveBracesLLVM) 2494 NestedTooDeep.push_back(TooDeep); 2495 if (IsPrecededByComment) 2496 --Line->Level; 2497 } else { 2498 parseUnbracedBody(/*CheckEOF=*/true); 2499 } 2500 } else { 2501 if (Style.RemoveBracesLLVM) 2502 KeepIfBraces = KeepIfBraces || IfBlockKind == IfStmtKind::IfElse; 2503 if (NeedsUnwrappedLine) 2504 addUnwrappedLine(); 2505 } 2506 2507 if (!Style.RemoveBracesLLVM) 2508 return nullptr; 2509 2510 assert(!NestedTooDeep.empty()); 2511 const bool KeepElseBraces = 2512 (ElseLeftBrace && !ElseLeftBrace->MatchingParen) || NestedTooDeep.back(); 2513 2514 NestedTooDeep.pop_back(); 2515 2516 if (!KeepBraces && !KeepIfBraces && !KeepElseBraces) { 2517 markOptionalBraces(IfLeftBrace); 2518 markOptionalBraces(ElseLeftBrace); 2519 } else if (IfLeftBrace) { 2520 FormatToken *IfRightBrace = IfLeftBrace->MatchingParen; 2521 if (IfRightBrace) { 2522 assert(IfRightBrace->MatchingParen == IfLeftBrace); 2523 assert(!IfLeftBrace->Optional); 2524 assert(!IfRightBrace->Optional); 2525 IfLeftBrace->MatchingParen = nullptr; 2526 IfRightBrace->MatchingParen = nullptr; 2527 } 2528 } 2529 2530 if (IfKind) 2531 *IfKind = Kind; 2532 2533 return IfLeftBrace; 2534 } 2535 2536 void UnwrappedLineParser::parseTryCatch() { 2537 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 2538 nextToken(); 2539 bool NeedsUnwrappedLine = false; 2540 if (FormatTok->is(tok::colon)) { 2541 // We are in a function try block, what comes is an initializer list. 2542 nextToken(); 2543 2544 // In case identifiers were removed by clang-tidy, what might follow is 2545 // multiple commas in sequence - before the first identifier. 2546 while (FormatTok->is(tok::comma)) 2547 nextToken(); 2548 2549 while (FormatTok->is(tok::identifier)) { 2550 nextToken(); 2551 if (FormatTok->is(tok::l_paren)) 2552 parseParens(); 2553 if (FormatTok->Previous && FormatTok->Previous->is(tok::identifier) && 2554 FormatTok->is(tok::l_brace)) { 2555 do { 2556 nextToken(); 2557 } while (!FormatTok->is(tok::r_brace)); 2558 nextToken(); 2559 } 2560 2561 // In case identifiers were removed by clang-tidy, what might follow is 2562 // multiple commas in sequence - after the first identifier. 2563 while (FormatTok->is(tok::comma)) 2564 nextToken(); 2565 } 2566 } 2567 // Parse try with resource. 2568 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) 2569 parseParens(); 2570 2571 keepAncestorBraces(); 2572 2573 if (FormatTok->is(tok::l_brace)) { 2574 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2575 parseBlock(); 2576 if (Style.BraceWrapping.BeforeCatch) 2577 addUnwrappedLine(); 2578 else 2579 NeedsUnwrappedLine = true; 2580 } else if (!FormatTok->is(tok::kw_catch)) { 2581 // The C++ standard requires a compound-statement after a try. 2582 // If there's none, we try to assume there's a structuralElement 2583 // and try to continue. 2584 addUnwrappedLine(); 2585 ++Line->Level; 2586 parseStructuralElement(); 2587 --Line->Level; 2588 } 2589 while (true) { 2590 if (FormatTok->is(tok::at)) 2591 nextToken(); 2592 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 2593 tok::kw___finally) || 2594 ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2595 FormatTok->is(Keywords.kw_finally)) || 2596 (FormatTok->isObjCAtKeyword(tok::objc_catch) || 2597 FormatTok->isObjCAtKeyword(tok::objc_finally)))) 2598 break; 2599 nextToken(); 2600 while (FormatTok->isNot(tok::l_brace)) { 2601 if (FormatTok->is(tok::l_paren)) { 2602 parseParens(); 2603 continue; 2604 } 2605 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) { 2606 if (Style.RemoveBracesLLVM) 2607 NestedTooDeep.pop_back(); 2608 return; 2609 } 2610 nextToken(); 2611 } 2612 NeedsUnwrappedLine = false; 2613 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2614 parseBlock(); 2615 if (Style.BraceWrapping.BeforeCatch) 2616 addUnwrappedLine(); 2617 else 2618 NeedsUnwrappedLine = true; 2619 } 2620 2621 if (Style.RemoveBracesLLVM) 2622 NestedTooDeep.pop_back(); 2623 2624 if (NeedsUnwrappedLine) 2625 addUnwrappedLine(); 2626 } 2627 2628 void UnwrappedLineParser::parseNamespace() { 2629 assert(FormatTok->isOneOf(tok::kw_namespace, TT_NamespaceMacro) && 2630 "'namespace' expected"); 2631 2632 const FormatToken &InitialToken = *FormatTok; 2633 nextToken(); 2634 if (InitialToken.is(TT_NamespaceMacro)) { 2635 parseParens(); 2636 } else { 2637 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::kw_inline, 2638 tok::l_square, tok::period, tok::l_paren) || 2639 (Style.isCSharp() && FormatTok->is(tok::kw_union))) 2640 if (FormatTok->is(tok::l_square)) 2641 parseSquare(); 2642 else if (FormatTok->is(tok::l_paren)) 2643 parseParens(); 2644 else 2645 nextToken(); 2646 } 2647 if (FormatTok->is(tok::l_brace)) { 2648 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2649 addUnwrappedLine(); 2650 2651 unsigned AddLevels = 2652 Style.NamespaceIndentation == FormatStyle::NI_All || 2653 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 2654 DeclarationScopeStack.size() > 1) 2655 ? 1u 2656 : 0u; 2657 bool ManageWhitesmithsBraces = 2658 AddLevels == 0u && 2659 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths; 2660 2661 // If we're in Whitesmiths mode, indent the brace if we're not indenting 2662 // the whole block. 2663 if (ManageWhitesmithsBraces) 2664 ++Line->Level; 2665 2666 parseBlock(/*MustBeDeclaration=*/true, AddLevels, 2667 /*MunchSemi=*/true, 2668 /*UnindentWhitesmithsBraces=*/ManageWhitesmithsBraces); 2669 2670 // Munch the semicolon after a namespace. This is more common than one would 2671 // think. Putting the semicolon into its own line is very ugly. 2672 if (FormatTok->is(tok::semi)) 2673 nextToken(); 2674 2675 addUnwrappedLine(AddLevels > 0 ? LineLevel::Remove : LineLevel::Keep); 2676 2677 if (ManageWhitesmithsBraces) 2678 --Line->Level; 2679 } 2680 // FIXME: Add error handling. 2681 } 2682 2683 void UnwrappedLineParser::parseNew() { 2684 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 2685 nextToken(); 2686 2687 if (Style.isCSharp()) { 2688 do { 2689 if (FormatTok->is(tok::l_brace)) 2690 parseBracedList(); 2691 2692 if (FormatTok->isOneOf(tok::semi, tok::comma)) 2693 return; 2694 2695 nextToken(); 2696 } while (!eof()); 2697 } 2698 2699 if (Style.Language != FormatStyle::LK_Java) 2700 return; 2701 2702 // In Java, we can parse everything up to the parens, which aren't optional. 2703 do { 2704 // There should not be a ;, { or } before the new's open paren. 2705 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 2706 return; 2707 2708 // Consume the parens. 2709 if (FormatTok->is(tok::l_paren)) { 2710 parseParens(); 2711 2712 // If there is a class body of an anonymous class, consume that as child. 2713 if (FormatTok->is(tok::l_brace)) 2714 parseChildBlock(); 2715 return; 2716 } 2717 nextToken(); 2718 } while (!eof()); 2719 } 2720 2721 void UnwrappedLineParser::parseForOrWhileLoop() { 2722 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 2723 "'for', 'while' or foreach macro expected"); 2724 nextToken(); 2725 // JS' for await ( ... 2726 if (Style.isJavaScript() && FormatTok->is(Keywords.kw_await)) 2727 nextToken(); 2728 if (Style.isCpp() && FormatTok->is(tok::kw_co_await)) 2729 nextToken(); 2730 if (FormatTok->is(tok::l_paren)) 2731 parseParens(); 2732 2733 keepAncestorBraces(); 2734 2735 if (FormatTok->is(tok::l_brace)) { 2736 FormatToken *LeftBrace = FormatTok; 2737 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2738 parseBlock(); 2739 if (Style.RemoveBracesLLVM) { 2740 assert(!NestedTooDeep.empty()); 2741 if (!NestedTooDeep.back()) 2742 markOptionalBraces(LeftBrace); 2743 } 2744 addUnwrappedLine(); 2745 } else { 2746 parseUnbracedBody(); 2747 } 2748 2749 if (Style.RemoveBracesLLVM) 2750 NestedTooDeep.pop_back(); 2751 } 2752 2753 void UnwrappedLineParser::parseDoWhile() { 2754 assert(FormatTok->is(tok::kw_do) && "'do' expected"); 2755 nextToken(); 2756 2757 keepAncestorBraces(); 2758 2759 if (FormatTok->is(tok::l_brace)) { 2760 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2761 parseBlock(); 2762 if (Style.BraceWrapping.BeforeWhile) 2763 addUnwrappedLine(); 2764 } else { 2765 parseUnbracedBody(); 2766 } 2767 2768 if (Style.RemoveBracesLLVM) 2769 NestedTooDeep.pop_back(); 2770 2771 // FIXME: Add error handling. 2772 if (!FormatTok->is(tok::kw_while)) { 2773 addUnwrappedLine(); 2774 return; 2775 } 2776 2777 // If in Whitesmiths mode, the line with the while() needs to be indented 2778 // to the same level as the block. 2779 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) 2780 ++Line->Level; 2781 2782 nextToken(); 2783 parseStructuralElement(); 2784 } 2785 2786 void UnwrappedLineParser::parseLabel(bool LeftAlignLabel) { 2787 nextToken(); 2788 unsigned OldLineLevel = Line->Level; 2789 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 2790 --Line->Level; 2791 if (LeftAlignLabel) 2792 Line->Level = 0; 2793 2794 if (!Style.IndentCaseBlocks && CommentsBeforeNextToken.empty() && 2795 FormatTok->is(tok::l_brace)) { 2796 2797 CompoundStatementIndenter Indenter(this, Line->Level, 2798 Style.BraceWrapping.AfterCaseLabel, 2799 Style.BraceWrapping.IndentBraces); 2800 parseBlock(); 2801 if (FormatTok->is(tok::kw_break)) { 2802 if (Style.BraceWrapping.AfterControlStatement == 2803 FormatStyle::BWACS_Always) { 2804 addUnwrappedLine(); 2805 if (!Style.IndentCaseBlocks && 2806 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) 2807 ++Line->Level; 2808 } 2809 parseStructuralElement(); 2810 } 2811 addUnwrappedLine(); 2812 } else { 2813 if (FormatTok->is(tok::semi)) 2814 nextToken(); 2815 addUnwrappedLine(); 2816 } 2817 Line->Level = OldLineLevel; 2818 if (FormatTok->isNot(tok::l_brace)) { 2819 parseStructuralElement(); 2820 addUnwrappedLine(); 2821 } 2822 } 2823 2824 void UnwrappedLineParser::parseCaseLabel() { 2825 assert(FormatTok->is(tok::kw_case) && "'case' expected"); 2826 2827 // FIXME: fix handling of complex expressions here. 2828 do { 2829 nextToken(); 2830 } while (!eof() && !FormatTok->is(tok::colon)); 2831 parseLabel(); 2832 } 2833 2834 void UnwrappedLineParser::parseSwitch() { 2835 assert(FormatTok->is(tok::kw_switch) && "'switch' expected"); 2836 nextToken(); 2837 if (FormatTok->is(tok::l_paren)) 2838 parseParens(); 2839 2840 keepAncestorBraces(); 2841 2842 if (FormatTok->is(tok::l_brace)) { 2843 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2844 parseBlock(); 2845 addUnwrappedLine(); 2846 } else { 2847 addUnwrappedLine(); 2848 ++Line->Level; 2849 parseStructuralElement(); 2850 --Line->Level; 2851 } 2852 2853 if (Style.RemoveBracesLLVM) 2854 NestedTooDeep.pop_back(); 2855 } 2856 2857 // Operators that can follow a C variable. 2858 static bool isCOperatorFollowingVar(tok::TokenKind kind) { 2859 switch (kind) { 2860 case tok::ampamp: 2861 case tok::ampequal: 2862 case tok::arrow: 2863 case tok::caret: 2864 case tok::caretequal: 2865 case tok::comma: 2866 case tok::ellipsis: 2867 case tok::equal: 2868 case tok::equalequal: 2869 case tok::exclaim: 2870 case tok::exclaimequal: 2871 case tok::greater: 2872 case tok::greaterequal: 2873 case tok::greatergreater: 2874 case tok::greatergreaterequal: 2875 case tok::l_paren: 2876 case tok::l_square: 2877 case tok::less: 2878 case tok::lessequal: 2879 case tok::lessless: 2880 case tok::lesslessequal: 2881 case tok::minus: 2882 case tok::minusequal: 2883 case tok::minusminus: 2884 case tok::percent: 2885 case tok::percentequal: 2886 case tok::period: 2887 case tok::pipe: 2888 case tok::pipeequal: 2889 case tok::pipepipe: 2890 case tok::plus: 2891 case tok::plusequal: 2892 case tok::plusplus: 2893 case tok::question: 2894 case tok::r_brace: 2895 case tok::r_paren: 2896 case tok::r_square: 2897 case tok::semi: 2898 case tok::slash: 2899 case tok::slashequal: 2900 case tok::star: 2901 case tok::starequal: 2902 return true; 2903 default: 2904 return false; 2905 } 2906 } 2907 2908 void UnwrappedLineParser::parseAccessSpecifier() { 2909 FormatToken *AccessSpecifierCandidate = FormatTok; 2910 nextToken(); 2911 // Understand Qt's slots. 2912 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 2913 nextToken(); 2914 // Otherwise, we don't know what it is, and we'd better keep the next token. 2915 if (FormatTok->is(tok::colon)) { 2916 nextToken(); 2917 addUnwrappedLine(); 2918 } else if (!FormatTok->is(tok::coloncolon) && 2919 !isCOperatorFollowingVar(FormatTok->Tok.getKind())) { 2920 // Not a variable name nor namespace name. 2921 addUnwrappedLine(); 2922 } else if (AccessSpecifierCandidate) { 2923 // Consider the access specifier to be a C identifier. 2924 AccessSpecifierCandidate->Tok.setKind(tok::identifier); 2925 } 2926 } 2927 2928 /// \brief Parses a concept definition. 2929 /// \pre The current token has to be the concept keyword. 2930 /// 2931 /// Returns if either the concept has been completely parsed, or if it detects 2932 /// that the concept definition is incorrect. 2933 void UnwrappedLineParser::parseConcept() { 2934 assert(FormatTok->is(tok::kw_concept) && "'concept' expected"); 2935 nextToken(); 2936 if (!FormatTok->is(tok::identifier)) 2937 return; 2938 nextToken(); 2939 if (!FormatTok->is(tok::equal)) 2940 return; 2941 nextToken(); 2942 parseConstraintExpression(); 2943 if (FormatTok->is(tok::semi)) 2944 nextToken(); 2945 addUnwrappedLine(); 2946 } 2947 2948 /// \brief Parses a requires, decides if it is a clause or an expression. 2949 /// \pre The current token has to be the requires keyword. 2950 /// \returns true if it parsed a clause. 2951 bool clang::format::UnwrappedLineParser::parseRequires() { 2952 assert(FormatTok->is(tok::kw_requires) && "'requires' expected"); 2953 auto RequiresToken = FormatTok; 2954 2955 // We try to guess if it is a requires clause, or a requires expression. For 2956 // that we first consume the keyword and check the next token. 2957 nextToken(); 2958 2959 switch (FormatTok->Tok.getKind()) { 2960 case tok::l_brace: 2961 // This can only be an expression, never a clause. 2962 parseRequiresExpression(RequiresToken); 2963 return false; 2964 case tok::l_paren: 2965 // Clauses and expression can start with a paren, it's unclear what we have. 2966 break; 2967 default: 2968 // All other tokens can only be a clause. 2969 parseRequiresClause(RequiresToken); 2970 return true; 2971 } 2972 2973 // Looking forward we would have to decide if there are function declaration 2974 // like arguments to the requires expression: 2975 // requires (T t) { 2976 // Or there is a constraint expression for the requires clause: 2977 // requires (C<T> && ... 2978 2979 // But first let's look behind. 2980 auto *PreviousNonComment = RequiresToken->getPreviousNonComment(); 2981 2982 if (!PreviousNonComment || 2983 PreviousNonComment->is(TT_RequiresExpressionLBrace)) { 2984 // If there is no token, or an expression left brace, we are a requires 2985 // clause within a requires expression. 2986 parseRequiresClause(RequiresToken); 2987 return true; 2988 } 2989 2990 switch (PreviousNonComment->Tok.getKind()) { 2991 case tok::greater: 2992 case tok::r_paren: 2993 case tok::kw_noexcept: 2994 case tok::kw_const: 2995 // This is a requires clause. 2996 parseRequiresClause(RequiresToken); 2997 return true; 2998 case tok::amp: 2999 case tok::ampamp: { 3000 // This can be either: 3001 // if (... && requires (T t) ...) 3002 // Or 3003 // void member(...) && requires (C<T> ... 3004 // We check the one token before that for a const: 3005 // void member(...) const && requires (C<T> ... 3006 auto PrevPrev = PreviousNonComment->getPreviousNonComment(); 3007 if (PrevPrev && PrevPrev->is(tok::kw_const)) { 3008 parseRequiresClause(RequiresToken); 3009 return true; 3010 } 3011 break; 3012 } 3013 default: 3014 // It's an expression. 3015 parseRequiresExpression(RequiresToken); 3016 return false; 3017 } 3018 3019 // Now we look forward and try to check if the paren content is a parameter 3020 // list. The parameters can be cv-qualified and contain references or 3021 // pointers. 3022 // So we want basically to check for TYPE NAME, but TYPE can contain all kinds 3023 // of stuff: typename, const, *, &, &&, ::, identifiers. 3024 3025 int NextTokenOffset = 1; 3026 auto NextToken = Tokens->peekNextToken(NextTokenOffset); 3027 auto PeekNext = [&NextTokenOffset, &NextToken, this] { 3028 ++NextTokenOffset; 3029 NextToken = Tokens->peekNextToken(NextTokenOffset); 3030 }; 3031 3032 bool FoundType = false; 3033 bool LastWasColonColon = false; 3034 int OpenAngles = 0; 3035 3036 for (; NextTokenOffset < 50; PeekNext()) { 3037 switch (NextToken->Tok.getKind()) { 3038 case tok::kw_volatile: 3039 case tok::kw_const: 3040 case tok::comma: 3041 parseRequiresExpression(RequiresToken); 3042 return false; 3043 case tok::r_paren: 3044 case tok::pipepipe: 3045 parseRequiresClause(RequiresToken); 3046 return true; 3047 case tok::eof: 3048 // Break out of the loop. 3049 NextTokenOffset = 50; 3050 break; 3051 case tok::coloncolon: 3052 LastWasColonColon = true; 3053 break; 3054 case tok::identifier: 3055 if (FoundType && !LastWasColonColon && OpenAngles == 0) { 3056 parseRequiresExpression(RequiresToken); 3057 return false; 3058 } 3059 FoundType = true; 3060 LastWasColonColon = false; 3061 break; 3062 case tok::less: 3063 ++OpenAngles; 3064 break; 3065 case tok::greater: 3066 --OpenAngles; 3067 break; 3068 default: 3069 if (NextToken->isSimpleTypeSpecifier()) { 3070 parseRequiresExpression(RequiresToken); 3071 return false; 3072 } 3073 break; 3074 } 3075 } 3076 3077 // This seems to be a complicated expression, just assume it's a clause. 3078 parseRequiresClause(RequiresToken); 3079 return true; 3080 } 3081 3082 /// \brief Parses a requires clause. 3083 /// \param RequiresToken The requires keyword token, which starts this clause. 3084 /// \pre We need to be on the next token after the requires keyword. 3085 /// \sa parseRequiresExpression 3086 /// 3087 /// Returns if it either has finished parsing the clause, or it detects, that 3088 /// the clause is incorrect. 3089 void UnwrappedLineParser::parseRequiresClause(FormatToken *RequiresToken) { 3090 assert(FormatTok->getPreviousNonComment() == RequiresToken); 3091 assert(RequiresToken->is(tok::kw_requires) && "'requires' expected"); 3092 3093 // If there is no previous token, we are within a requires expression, 3094 // otherwise we will always have the template or function declaration in front 3095 // of it. 3096 bool InRequiresExpression = 3097 !RequiresToken->Previous || 3098 RequiresToken->Previous->is(TT_RequiresExpressionLBrace); 3099 3100 RequiresToken->setFinalizedType(InRequiresExpression 3101 ? TT_RequiresClauseInARequiresExpression 3102 : TT_RequiresClause); 3103 3104 parseConstraintExpression(); 3105 3106 if (!InRequiresExpression) 3107 FormatTok->Previous->ClosesRequiresClause = true; 3108 } 3109 3110 /// \brief Parses a requires expression. 3111 /// \param RequiresToken The requires keyword token, which starts this clause. 3112 /// \pre We need to be on the next token after the requires keyword. 3113 /// \sa parseRequiresClause 3114 /// 3115 /// Returns if it either has finished parsing the expression, or it detects, 3116 /// that the expression is incorrect. 3117 void UnwrappedLineParser::parseRequiresExpression(FormatToken *RequiresToken) { 3118 assert(FormatTok->getPreviousNonComment() == RequiresToken); 3119 assert(RequiresToken->is(tok::kw_requires) && "'requires' expected"); 3120 3121 RequiresToken->setFinalizedType(TT_RequiresExpression); 3122 3123 if (FormatTok->is(tok::l_paren)) { 3124 FormatTok->setFinalizedType(TT_RequiresExpressionLParen); 3125 parseParens(); 3126 } 3127 3128 if (FormatTok->is(tok::l_brace)) { 3129 FormatTok->setFinalizedType(TT_RequiresExpressionLBrace); 3130 parseChildBlock(/*CanContainBracedList=*/false, 3131 /*NextLBracesType=*/TT_CompoundRequirementLBrace); 3132 } 3133 } 3134 3135 /// \brief Parses a constraint expression. 3136 /// 3137 /// This is either the definition of a concept, or the body of a requires 3138 /// clause. It returns, when the parsing is complete, or the expression is 3139 /// incorrect. 3140 void UnwrappedLineParser::parseConstraintExpression() { 3141 // The special handling for lambdas is needed since tryToParseLambda() eats a 3142 // token and if a requires expression is the last part of a requires clause 3143 // and followed by an attribute like [[nodiscard]] the ClosesRequiresClause is 3144 // not set on the correct token. Thus we need to be aware if we even expect a 3145 // lambda to be possible. 3146 // template <typename T> requires requires { ... } [[nodiscard]] ...; 3147 bool LambdaNextTimeAllowed = true; 3148 do { 3149 bool LambdaThisTimeAllowed = std::exchange(LambdaNextTimeAllowed, false); 3150 3151 switch (FormatTok->Tok.getKind()) { 3152 case tok::kw_requires: { 3153 auto RequiresToken = FormatTok; 3154 nextToken(); 3155 parseRequiresExpression(RequiresToken); 3156 break; 3157 } 3158 3159 case tok::l_paren: 3160 parseParens(/*AmpAmpTokenType=*/TT_BinaryOperator); 3161 break; 3162 3163 case tok::l_square: 3164 if (!LambdaThisTimeAllowed || !tryToParseLambda()) 3165 return; 3166 break; 3167 3168 case tok::kw_const: 3169 case tok::semi: 3170 case tok::kw_class: 3171 case tok::kw_struct: 3172 case tok::kw_union: 3173 return; 3174 3175 case tok::l_brace: 3176 // Potential function body. 3177 return; 3178 3179 case tok::ampamp: 3180 case tok::pipepipe: 3181 FormatTok->setFinalizedType(TT_BinaryOperator); 3182 nextToken(); 3183 LambdaNextTimeAllowed = true; 3184 break; 3185 3186 case tok::comma: 3187 case tok::comment: 3188 LambdaNextTimeAllowed = LambdaThisTimeAllowed; 3189 nextToken(); 3190 break; 3191 3192 case tok::kw_sizeof: 3193 case tok::greater: 3194 case tok::greaterequal: 3195 case tok::greatergreater: 3196 case tok::less: 3197 case tok::lessequal: 3198 case tok::lessless: 3199 case tok::equalequal: 3200 case tok::exclaim: 3201 case tok::exclaimequal: 3202 case tok::plus: 3203 case tok::minus: 3204 case tok::star: 3205 case tok::slash: 3206 case tok::kw_decltype: 3207 LambdaNextTimeAllowed = true; 3208 // Just eat them. 3209 nextToken(); 3210 break; 3211 3212 case tok::numeric_constant: 3213 case tok::coloncolon: 3214 case tok::kw_true: 3215 case tok::kw_false: 3216 // Just eat them. 3217 nextToken(); 3218 break; 3219 3220 case tok::kw_static_cast: 3221 case tok::kw_const_cast: 3222 case tok::kw_reinterpret_cast: 3223 case tok::kw_dynamic_cast: 3224 nextToken(); 3225 if (!FormatTok->is(tok::less)) 3226 return; 3227 3228 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 3229 /*ClosingBraceKind=*/tok::greater); 3230 break; 3231 3232 case tok::kw_bool: 3233 // bool is only allowed if it is directly followed by a paren for a cast: 3234 // concept C = bool(...); 3235 // and bool is the only type, all other types as cast must be inside a 3236 // cast to bool an thus are handled by the other cases. 3237 nextToken(); 3238 if (FormatTok->isNot(tok::l_paren)) 3239 return; 3240 parseParens(); 3241 break; 3242 3243 default: 3244 if (!FormatTok->Tok.getIdentifierInfo()) { 3245 // Identifiers are part of the default case, we check for more then 3246 // tok::identifier to handle builtin type traits. 3247 return; 3248 } 3249 3250 // We need to differentiate identifiers for a template deduction guide, 3251 // variables, or function return types (the constraint expression has 3252 // ended before that), and basically all other cases. But it's easier to 3253 // check the other way around. 3254 assert(FormatTok->Previous); 3255 switch (FormatTok->Previous->Tok.getKind()) { 3256 case tok::coloncolon: // Nested identifier. 3257 case tok::ampamp: // Start of a function or variable for the 3258 case tok::pipepipe: // constraint expression. 3259 case tok::kw_requires: // Initial identifier of a requires clause. 3260 case tok::equal: // Initial identifier of a concept declaration. 3261 break; 3262 default: 3263 return; 3264 } 3265 3266 // Read identifier with optional template declaration. 3267 nextToken(); 3268 if (FormatTok->is(tok::less)) 3269 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 3270 /*ClosingBraceKind=*/tok::greater); 3271 break; 3272 } 3273 } while (!eof()); 3274 } 3275 3276 bool UnwrappedLineParser::parseEnum() { 3277 const FormatToken &InitialToken = *FormatTok; 3278 3279 // Won't be 'enum' for NS_ENUMs. 3280 if (FormatTok->is(tok::kw_enum)) 3281 nextToken(); 3282 3283 // In TypeScript, "enum" can also be used as property name, e.g. in interface 3284 // declarations. An "enum" keyword followed by a colon would be a syntax 3285 // error and thus assume it is just an identifier. 3286 if (Style.isJavaScript() && FormatTok->isOneOf(tok::colon, tok::question)) 3287 return false; 3288 3289 // In protobuf, "enum" can be used as a field name. 3290 if (Style.Language == FormatStyle::LK_Proto && FormatTok->is(tok::equal)) 3291 return false; 3292 3293 // Eat up enum class ... 3294 if (FormatTok->isOneOf(tok::kw_class, tok::kw_struct)) 3295 nextToken(); 3296 3297 while (FormatTok->Tok.getIdentifierInfo() || 3298 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 3299 tok::greater, tok::comma, tok::question)) { 3300 nextToken(); 3301 // We can have macros or attributes in between 'enum' and the enum name. 3302 if (FormatTok->is(tok::l_paren)) 3303 parseParens(); 3304 if (FormatTok->is(tok::identifier)) { 3305 nextToken(); 3306 // If there are two identifiers in a row, this is likely an elaborate 3307 // return type. In Java, this can be "implements", etc. 3308 if (Style.isCpp() && FormatTok->is(tok::identifier)) 3309 return false; 3310 } 3311 } 3312 3313 // Just a declaration or something is wrong. 3314 if (FormatTok->isNot(tok::l_brace)) 3315 return true; 3316 FormatTok->setFinalizedType(TT_EnumLBrace); 3317 FormatTok->setBlockKind(BK_Block); 3318 3319 if (Style.Language == FormatStyle::LK_Java) { 3320 // Java enums are different. 3321 parseJavaEnumBody(); 3322 return true; 3323 } 3324 if (Style.Language == FormatStyle::LK_Proto) { 3325 parseBlock(/*MustBeDeclaration=*/true); 3326 return true; 3327 } 3328 3329 if (!Style.AllowShortEnumsOnASingleLine && 3330 ShouldBreakBeforeBrace(Style, InitialToken)) 3331 addUnwrappedLine(); 3332 // Parse enum body. 3333 nextToken(); 3334 if (!Style.AllowShortEnumsOnASingleLine) { 3335 addUnwrappedLine(); 3336 Line->Level += 1; 3337 } 3338 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true, 3339 /*IsEnum=*/true); 3340 if (!Style.AllowShortEnumsOnASingleLine) 3341 Line->Level -= 1; 3342 if (HasError) { 3343 if (FormatTok->is(tok::semi)) 3344 nextToken(); 3345 addUnwrappedLine(); 3346 } 3347 return true; 3348 3349 // There is no addUnwrappedLine() here so that we fall through to parsing a 3350 // structural element afterwards. Thus, in "enum A {} n, m;", 3351 // "} n, m;" will end up in one unwrapped line. 3352 } 3353 3354 bool UnwrappedLineParser::parseStructLike() { 3355 // parseRecord falls through and does not yet add an unwrapped line as a 3356 // record declaration or definition can start a structural element. 3357 parseRecord(); 3358 // This does not apply to Java, JavaScript and C#. 3359 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 3360 Style.isCSharp()) { 3361 if (FormatTok->is(tok::semi)) 3362 nextToken(); 3363 addUnwrappedLine(); 3364 return true; 3365 } 3366 return false; 3367 } 3368 3369 namespace { 3370 // A class used to set and restore the Token position when peeking 3371 // ahead in the token source. 3372 class ScopedTokenPosition { 3373 unsigned StoredPosition; 3374 FormatTokenSource *Tokens; 3375 3376 public: 3377 ScopedTokenPosition(FormatTokenSource *Tokens) : Tokens(Tokens) { 3378 assert(Tokens && "Tokens expected to not be null"); 3379 StoredPosition = Tokens->getPosition(); 3380 } 3381 3382 ~ScopedTokenPosition() { Tokens->setPosition(StoredPosition); } 3383 }; 3384 } // namespace 3385 3386 // Look to see if we have [[ by looking ahead, if 3387 // its not then rewind to the original position. 3388 bool UnwrappedLineParser::tryToParseSimpleAttribute() { 3389 ScopedTokenPosition AutoPosition(Tokens); 3390 FormatToken *Tok = Tokens->getNextToken(); 3391 // We already read the first [ check for the second. 3392 if (!Tok->is(tok::l_square)) 3393 return false; 3394 // Double check that the attribute is just something 3395 // fairly simple. 3396 while (Tok->isNot(tok::eof)) { 3397 if (Tok->is(tok::r_square)) 3398 break; 3399 Tok = Tokens->getNextToken(); 3400 } 3401 if (Tok->is(tok::eof)) 3402 return false; 3403 Tok = Tokens->getNextToken(); 3404 if (!Tok->is(tok::r_square)) 3405 return false; 3406 Tok = Tokens->getNextToken(); 3407 if (Tok->is(tok::semi)) 3408 return false; 3409 return true; 3410 } 3411 3412 void UnwrappedLineParser::parseJavaEnumBody() { 3413 // Determine whether the enum is simple, i.e. does not have a semicolon or 3414 // constants with class bodies. Simple enums can be formatted like braced 3415 // lists, contracted to a single line, etc. 3416 unsigned StoredPosition = Tokens->getPosition(); 3417 bool IsSimple = true; 3418 FormatToken *Tok = Tokens->getNextToken(); 3419 while (!Tok->is(tok::eof)) { 3420 if (Tok->is(tok::r_brace)) 3421 break; 3422 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 3423 IsSimple = false; 3424 break; 3425 } 3426 // FIXME: This will also mark enums with braces in the arguments to enum 3427 // constants as "not simple". This is probably fine in practice, though. 3428 Tok = Tokens->getNextToken(); 3429 } 3430 FormatTok = Tokens->setPosition(StoredPosition); 3431 3432 if (IsSimple) { 3433 nextToken(); 3434 parseBracedList(); 3435 addUnwrappedLine(); 3436 return; 3437 } 3438 3439 // Parse the body of a more complex enum. 3440 // First add a line for everything up to the "{". 3441 nextToken(); 3442 addUnwrappedLine(); 3443 ++Line->Level; 3444 3445 // Parse the enum constants. 3446 while (FormatTok) { 3447 if (FormatTok->is(tok::l_brace)) { 3448 // Parse the constant's class body. 3449 parseBlock(/*MustBeDeclaration=*/true, /*AddLevels=*/1u, 3450 /*MunchSemi=*/false); 3451 } else if (FormatTok->is(tok::l_paren)) { 3452 parseParens(); 3453 } else if (FormatTok->is(tok::comma)) { 3454 nextToken(); 3455 addUnwrappedLine(); 3456 } else if (FormatTok->is(tok::semi)) { 3457 nextToken(); 3458 addUnwrappedLine(); 3459 break; 3460 } else if (FormatTok->is(tok::r_brace)) { 3461 addUnwrappedLine(); 3462 break; 3463 } else { 3464 nextToken(); 3465 } 3466 } 3467 3468 // Parse the class body after the enum's ";" if any. 3469 parseLevel(/*HasOpeningBrace=*/true, /*CanContainBracedList=*/true); 3470 nextToken(); 3471 --Line->Level; 3472 addUnwrappedLine(); 3473 } 3474 3475 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) { 3476 const FormatToken &InitialToken = *FormatTok; 3477 nextToken(); 3478 3479 // The actual identifier can be a nested name specifier, and in macros 3480 // it is often token-pasted. 3481 // An [[attribute]] can be before the identifier. 3482 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 3483 tok::kw___attribute, tok::kw___declspec, 3484 tok::kw_alignas, tok::l_square, tok::r_square) || 3485 ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 3486 FormatTok->isOneOf(tok::period, tok::comma))) { 3487 if (Style.isJavaScript() && 3488 FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) { 3489 // JavaScript/TypeScript supports inline object types in 3490 // extends/implements positions: 3491 // class Foo implements {bar: number} { } 3492 nextToken(); 3493 if (FormatTok->is(tok::l_brace)) { 3494 tryToParseBracedList(); 3495 continue; 3496 } 3497 } 3498 bool IsNonMacroIdentifier = 3499 FormatTok->is(tok::identifier) && 3500 FormatTok->TokenText != FormatTok->TokenText.upper(); 3501 nextToken(); 3502 // We can have macros or attributes in between 'class' and the class name. 3503 if (!IsNonMacroIdentifier) { 3504 if (FormatTok->is(tok::l_paren)) { 3505 parseParens(); 3506 } else if (FormatTok->is(TT_AttributeSquare)) { 3507 parseSquare(); 3508 // Consume the closing TT_AttributeSquare. 3509 if (FormatTok->Next && FormatTok->is(TT_AttributeSquare)) 3510 nextToken(); 3511 } 3512 } 3513 } 3514 3515 // Note that parsing away template declarations here leads to incorrectly 3516 // accepting function declarations as record declarations. 3517 // In general, we cannot solve this problem. Consider: 3518 // class A<int> B() {} 3519 // which can be a function definition or a class definition when B() is a 3520 // macro. If we find enough real-world cases where this is a problem, we 3521 // can parse for the 'template' keyword in the beginning of the statement, 3522 // and thus rule out the record production in case there is no template 3523 // (this would still leave us with an ambiguity between template function 3524 // and class declarations). 3525 if (FormatTok->isOneOf(tok::colon, tok::less)) { 3526 do { 3527 if (FormatTok->is(tok::l_brace)) { 3528 calculateBraceTypes(/*ExpectClassBody=*/true); 3529 if (!tryToParseBracedList()) 3530 break; 3531 } 3532 if (FormatTok->is(tok::l_square)) { 3533 FormatToken *Previous = FormatTok->Previous; 3534 if (!Previous || 3535 !(Previous->is(tok::r_paren) || Previous->isTypeOrIdentifier())) { 3536 // Don't try parsing a lambda if we had a closing parenthesis before, 3537 // it was probably a pointer to an array: int (*)[]. 3538 if (!tryToParseLambda()) 3539 continue; 3540 } else { 3541 parseSquare(); 3542 continue; 3543 } 3544 } 3545 if (FormatTok->is(tok::semi)) 3546 return; 3547 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where)) { 3548 addUnwrappedLine(); 3549 nextToken(); 3550 parseCSharpGenericTypeConstraint(); 3551 break; 3552 } 3553 nextToken(); 3554 } while (!eof()); 3555 } 3556 3557 auto GetBraceType = [](const FormatToken &RecordTok) { 3558 switch (RecordTok.Tok.getKind()) { 3559 case tok::kw_class: 3560 return TT_ClassLBrace; 3561 case tok::kw_struct: 3562 return TT_StructLBrace; 3563 case tok::kw_union: 3564 return TT_UnionLBrace; 3565 default: 3566 // Useful for e.g. interface. 3567 return TT_RecordLBrace; 3568 } 3569 }; 3570 if (FormatTok->is(tok::l_brace)) { 3571 FormatTok->setFinalizedType(GetBraceType(InitialToken)); 3572 if (ParseAsExpr) { 3573 parseChildBlock(); 3574 } else { 3575 if (ShouldBreakBeforeBrace(Style, InitialToken)) 3576 addUnwrappedLine(); 3577 3578 unsigned AddLevels = Style.IndentAccessModifiers ? 2u : 1u; 3579 parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/false); 3580 } 3581 } 3582 // There is no addUnwrappedLine() here so that we fall through to parsing a 3583 // structural element afterwards. Thus, in "class A {} n, m;", 3584 // "} n, m;" will end up in one unwrapped line. 3585 } 3586 3587 void UnwrappedLineParser::parseObjCMethod() { 3588 assert(FormatTok->isOneOf(tok::l_paren, tok::identifier) && 3589 "'(' or identifier expected."); 3590 do { 3591 if (FormatTok->is(tok::semi)) { 3592 nextToken(); 3593 addUnwrappedLine(); 3594 return; 3595 } else if (FormatTok->is(tok::l_brace)) { 3596 if (Style.BraceWrapping.AfterFunction) 3597 addUnwrappedLine(); 3598 parseBlock(); 3599 addUnwrappedLine(); 3600 return; 3601 } else { 3602 nextToken(); 3603 } 3604 } while (!eof()); 3605 } 3606 3607 void UnwrappedLineParser::parseObjCProtocolList() { 3608 assert(FormatTok->is(tok::less) && "'<' expected."); 3609 do { 3610 nextToken(); 3611 // Early exit in case someone forgot a close angle. 3612 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 3613 FormatTok->isObjCAtKeyword(tok::objc_end)) 3614 return; 3615 } while (!eof() && FormatTok->isNot(tok::greater)); 3616 nextToken(); // Skip '>'. 3617 } 3618 3619 void UnwrappedLineParser::parseObjCUntilAtEnd() { 3620 do { 3621 if (FormatTok->isObjCAtKeyword(tok::objc_end)) { 3622 nextToken(); 3623 addUnwrappedLine(); 3624 break; 3625 } 3626 if (FormatTok->is(tok::l_brace)) { 3627 parseBlock(); 3628 // In ObjC interfaces, nothing should be following the "}". 3629 addUnwrappedLine(); 3630 } else if (FormatTok->is(tok::r_brace)) { 3631 // Ignore stray "}". parseStructuralElement doesn't consume them. 3632 nextToken(); 3633 addUnwrappedLine(); 3634 } else if (FormatTok->isOneOf(tok::minus, tok::plus)) { 3635 nextToken(); 3636 parseObjCMethod(); 3637 } else { 3638 parseStructuralElement(); 3639 } 3640 } while (!eof()); 3641 } 3642 3643 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 3644 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_interface || 3645 FormatTok->Tok.getObjCKeywordID() == tok::objc_implementation); 3646 nextToken(); 3647 nextToken(); // interface name 3648 3649 // @interface can be followed by a lightweight generic 3650 // specialization list, then either a base class or a category. 3651 if (FormatTok->is(tok::less)) 3652 parseObjCLightweightGenerics(); 3653 if (FormatTok->is(tok::colon)) { 3654 nextToken(); 3655 nextToken(); // base class name 3656 // The base class can also have lightweight generics applied to it. 3657 if (FormatTok->is(tok::less)) 3658 parseObjCLightweightGenerics(); 3659 } else if (FormatTok->is(tok::l_paren)) 3660 // Skip category, if present. 3661 parseParens(); 3662 3663 if (FormatTok->is(tok::less)) 3664 parseObjCProtocolList(); 3665 3666 if (FormatTok->is(tok::l_brace)) { 3667 if (Style.BraceWrapping.AfterObjCDeclaration) 3668 addUnwrappedLine(); 3669 parseBlock(/*MustBeDeclaration=*/true); 3670 } 3671 3672 // With instance variables, this puts '}' on its own line. Without instance 3673 // variables, this ends the @interface line. 3674 addUnwrappedLine(); 3675 3676 parseObjCUntilAtEnd(); 3677 } 3678 3679 void UnwrappedLineParser::parseObjCLightweightGenerics() { 3680 assert(FormatTok->is(tok::less)); 3681 // Unlike protocol lists, generic parameterizations support 3682 // nested angles: 3683 // 3684 // @interface Foo<ValueType : id <NSCopying, NSSecureCoding>> : 3685 // NSObject <NSCopying, NSSecureCoding> 3686 // 3687 // so we need to count how many open angles we have left. 3688 unsigned NumOpenAngles = 1; 3689 do { 3690 nextToken(); 3691 // Early exit in case someone forgot a close angle. 3692 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 3693 FormatTok->isObjCAtKeyword(tok::objc_end)) 3694 break; 3695 if (FormatTok->is(tok::less)) 3696 ++NumOpenAngles; 3697 else if (FormatTok->is(tok::greater)) { 3698 assert(NumOpenAngles > 0 && "'>' makes NumOpenAngles negative"); 3699 --NumOpenAngles; 3700 } 3701 } while (!eof() && NumOpenAngles != 0); 3702 nextToken(); // Skip '>'. 3703 } 3704 3705 // Returns true for the declaration/definition form of @protocol, 3706 // false for the expression form. 3707 bool UnwrappedLineParser::parseObjCProtocol() { 3708 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_protocol); 3709 nextToken(); 3710 3711 if (FormatTok->is(tok::l_paren)) 3712 // The expression form of @protocol, e.g. "Protocol* p = @protocol(foo);". 3713 return false; 3714 3715 // The definition/declaration form, 3716 // @protocol Foo 3717 // - (int)someMethod; 3718 // @end 3719 3720 nextToken(); // protocol name 3721 3722 if (FormatTok->is(tok::less)) 3723 parseObjCProtocolList(); 3724 3725 // Check for protocol declaration. 3726 if (FormatTok->is(tok::semi)) { 3727 nextToken(); 3728 addUnwrappedLine(); 3729 return true; 3730 } 3731 3732 addUnwrappedLine(); 3733 parseObjCUntilAtEnd(); 3734 return true; 3735 } 3736 3737 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 3738 bool IsImport = FormatTok->is(Keywords.kw_import); 3739 assert(IsImport || FormatTok->is(tok::kw_export)); 3740 nextToken(); 3741 3742 // Consume the "default" in "export default class/function". 3743 if (FormatTok->is(tok::kw_default)) 3744 nextToken(); 3745 3746 // Consume "async function", "function" and "default function", so that these 3747 // get parsed as free-standing JS functions, i.e. do not require a trailing 3748 // semicolon. 3749 if (FormatTok->is(Keywords.kw_async)) 3750 nextToken(); 3751 if (FormatTok->is(Keywords.kw_function)) { 3752 nextToken(); 3753 return; 3754 } 3755 3756 // For imports, `export *`, `export {...}`, consume the rest of the line up 3757 // to the terminating `;`. For everything else, just return and continue 3758 // parsing the structural element, i.e. the declaration or expression for 3759 // `export default`. 3760 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 3761 !FormatTok->isStringLiteral()) 3762 return; 3763 3764 while (!eof()) { 3765 if (FormatTok->is(tok::semi)) 3766 return; 3767 if (Line->Tokens.empty()) { 3768 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 3769 // import statement should terminate. 3770 return; 3771 } 3772 if (FormatTok->is(tok::l_brace)) { 3773 FormatTok->setBlockKind(BK_Block); 3774 nextToken(); 3775 parseBracedList(); 3776 } else { 3777 nextToken(); 3778 } 3779 } 3780 } 3781 3782 void UnwrappedLineParser::parseStatementMacro() { 3783 nextToken(); 3784 if (FormatTok->is(tok::l_paren)) 3785 parseParens(); 3786 if (FormatTok->is(tok::semi)) 3787 nextToken(); 3788 addUnwrappedLine(); 3789 } 3790 3791 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 3792 StringRef Prefix = "") { 3793 llvm::dbgs() << Prefix << "Line(" << Line.Level 3794 << ", FSC=" << Line.FirstStartColumn << ")" 3795 << (Line.InPPDirective ? " MACRO" : "") << ": "; 3796 for (const auto &Node : Line.Tokens) { 3797 llvm::dbgs() << Node.Tok->Tok.getName() << "[" 3798 << "T=" << static_cast<unsigned>(Node.Tok->getType()) 3799 << ", OC=" << Node.Tok->OriginalColumn << "] "; 3800 } 3801 for (const auto &Node : Line.Tokens) 3802 for (const auto &ChildNode : Node.Children) 3803 printDebugInfo(ChildNode, "\nChild: "); 3804 3805 llvm::dbgs() << "\n"; 3806 } 3807 3808 void UnwrappedLineParser::addUnwrappedLine(LineLevel AdjustLevel) { 3809 if (Line->Tokens.empty()) 3810 return; 3811 LLVM_DEBUG({ 3812 if (CurrentLines == &Lines) 3813 printDebugInfo(*Line); 3814 }); 3815 3816 // If this line closes a block when in Whitesmiths mode, remember that 3817 // information so that the level can be decreased after the line is added. 3818 // This has to happen after the addition of the line since the line itself 3819 // needs to be indented. 3820 bool ClosesWhitesmithsBlock = 3821 Line->MatchingOpeningBlockLineIndex != UnwrappedLine::kInvalidIndex && 3822 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths; 3823 3824 CurrentLines->push_back(std::move(*Line)); 3825 Line->Tokens.clear(); 3826 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex; 3827 Line->FirstStartColumn = 0; 3828 3829 if (ClosesWhitesmithsBlock && AdjustLevel == LineLevel::Remove) 3830 --Line->Level; 3831 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 3832 CurrentLines->append( 3833 std::make_move_iterator(PreprocessorDirectives.begin()), 3834 std::make_move_iterator(PreprocessorDirectives.end())); 3835 PreprocessorDirectives.clear(); 3836 } 3837 // Disconnect the current token from the last token on the previous line. 3838 FormatTok->Previous = nullptr; 3839 } 3840 3841 bool UnwrappedLineParser::eof() const { return FormatTok->is(tok::eof); } 3842 3843 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 3844 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 3845 FormatTok.NewlinesBefore > 0; 3846 } 3847 3848 // Checks if \p FormatTok is a line comment that continues the line comment 3849 // section on \p Line. 3850 static bool 3851 continuesLineCommentSection(const FormatToken &FormatTok, 3852 const UnwrappedLine &Line, 3853 const llvm::Regex &CommentPragmasRegex) { 3854 if (Line.Tokens.empty()) 3855 return false; 3856 3857 StringRef IndentContent = FormatTok.TokenText; 3858 if (FormatTok.TokenText.startswith("//") || 3859 FormatTok.TokenText.startswith("/*")) 3860 IndentContent = FormatTok.TokenText.substr(2); 3861 if (CommentPragmasRegex.match(IndentContent)) 3862 return false; 3863 3864 // If Line starts with a line comment, then FormatTok continues the comment 3865 // section if its original column is greater or equal to the original start 3866 // column of the line. 3867 // 3868 // Define the min column token of a line as follows: if a line ends in '{' or 3869 // contains a '{' followed by a line comment, then the min column token is 3870 // that '{'. Otherwise, the min column token of the line is the first token of 3871 // the line. 3872 // 3873 // If Line starts with a token other than a line comment, then FormatTok 3874 // continues the comment section if its original column is greater than the 3875 // original start column of the min column token of the line. 3876 // 3877 // For example, the second line comment continues the first in these cases: 3878 // 3879 // // first line 3880 // // second line 3881 // 3882 // and: 3883 // 3884 // // first line 3885 // // second line 3886 // 3887 // and: 3888 // 3889 // int i; // first line 3890 // // second line 3891 // 3892 // and: 3893 // 3894 // do { // first line 3895 // // second line 3896 // int i; 3897 // } while (true); 3898 // 3899 // and: 3900 // 3901 // enum { 3902 // a, // first line 3903 // // second line 3904 // b 3905 // }; 3906 // 3907 // The second line comment doesn't continue the first in these cases: 3908 // 3909 // // first line 3910 // // second line 3911 // 3912 // and: 3913 // 3914 // int i; // first line 3915 // // second line 3916 // 3917 // and: 3918 // 3919 // do { // first line 3920 // // second line 3921 // int i; 3922 // } while (true); 3923 // 3924 // and: 3925 // 3926 // enum { 3927 // a, // first line 3928 // // second line 3929 // }; 3930 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 3931 3932 // Scan for '{//'. If found, use the column of '{' as a min column for line 3933 // comment section continuation. 3934 const FormatToken *PreviousToken = nullptr; 3935 for (const UnwrappedLineNode &Node : Line.Tokens) { 3936 if (PreviousToken && PreviousToken->is(tok::l_brace) && 3937 isLineComment(*Node.Tok)) { 3938 MinColumnToken = PreviousToken; 3939 break; 3940 } 3941 PreviousToken = Node.Tok; 3942 3943 // Grab the last newline preceding a token in this unwrapped line. 3944 if (Node.Tok->NewlinesBefore > 0) 3945 MinColumnToken = Node.Tok; 3946 } 3947 if (PreviousToken && PreviousToken->is(tok::l_brace)) 3948 MinColumnToken = PreviousToken; 3949 3950 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok, 3951 MinColumnToken); 3952 } 3953 3954 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 3955 bool JustComments = Line->Tokens.empty(); 3956 for (FormatToken *Tok : CommentsBeforeNextToken) { 3957 // Line comments that belong to the same line comment section are put on the 3958 // same line since later we might want to reflow content between them. 3959 // Additional fine-grained breaking of line comment sections is controlled 3960 // by the class BreakableLineCommentSection in case it is desirable to keep 3961 // several line comment sections in the same unwrapped line. 3962 // 3963 // FIXME: Consider putting separate line comment sections as children to the 3964 // unwrapped line instead. 3965 Tok->ContinuesLineCommentSection = 3966 continuesLineCommentSection(*Tok, *Line, CommentPragmasRegex); 3967 if (isOnNewLine(*Tok) && JustComments && !Tok->ContinuesLineCommentSection) 3968 addUnwrappedLine(); 3969 pushToken(Tok); 3970 } 3971 if (NewlineBeforeNext && JustComments) 3972 addUnwrappedLine(); 3973 CommentsBeforeNextToken.clear(); 3974 } 3975 3976 void UnwrappedLineParser::nextToken(int LevelDifference) { 3977 if (eof()) 3978 return; 3979 flushComments(isOnNewLine(*FormatTok)); 3980 pushToken(FormatTok); 3981 FormatToken *Previous = FormatTok; 3982 if (!Style.isJavaScript()) 3983 readToken(LevelDifference); 3984 else 3985 readTokenWithJavaScriptASI(); 3986 FormatTok->Previous = Previous; 3987 } 3988 3989 void UnwrappedLineParser::distributeComments( 3990 const SmallVectorImpl<FormatToken *> &Comments, 3991 const FormatToken *NextTok) { 3992 // Whether or not a line comment token continues a line is controlled by 3993 // the method continuesLineCommentSection, with the following caveat: 3994 // 3995 // Define a trail of Comments to be a nonempty proper postfix of Comments such 3996 // that each comment line from the trail is aligned with the next token, if 3997 // the next token exists. If a trail exists, the beginning of the maximal 3998 // trail is marked as a start of a new comment section. 3999 // 4000 // For example in this code: 4001 // 4002 // int a; // line about a 4003 // // line 1 about b 4004 // // line 2 about b 4005 // int b; 4006 // 4007 // the two lines about b form a maximal trail, so there are two sections, the 4008 // first one consisting of the single comment "// line about a" and the 4009 // second one consisting of the next two comments. 4010 if (Comments.empty()) 4011 return; 4012 bool ShouldPushCommentsInCurrentLine = true; 4013 bool HasTrailAlignedWithNextToken = false; 4014 unsigned StartOfTrailAlignedWithNextToken = 0; 4015 if (NextTok) { 4016 // We are skipping the first element intentionally. 4017 for (unsigned i = Comments.size() - 1; i > 0; --i) { 4018 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) { 4019 HasTrailAlignedWithNextToken = true; 4020 StartOfTrailAlignedWithNextToken = i; 4021 } 4022 } 4023 } 4024 for (unsigned i = 0, e = Comments.size(); i < e; ++i) { 4025 FormatToken *FormatTok = Comments[i]; 4026 if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) { 4027 FormatTok->ContinuesLineCommentSection = false; 4028 } else { 4029 FormatTok->ContinuesLineCommentSection = 4030 continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex); 4031 } 4032 if (!FormatTok->ContinuesLineCommentSection && 4033 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) 4034 ShouldPushCommentsInCurrentLine = false; 4035 if (ShouldPushCommentsInCurrentLine) 4036 pushToken(FormatTok); 4037 else 4038 CommentsBeforeNextToken.push_back(FormatTok); 4039 } 4040 } 4041 4042 void UnwrappedLineParser::readToken(int LevelDifference) { 4043 SmallVector<FormatToken *, 1> Comments; 4044 bool PreviousWasComment = false; 4045 bool FirstNonCommentOnLine = false; 4046 do { 4047 FormatTok = Tokens->getNextToken(); 4048 assert(FormatTok); 4049 while (FormatTok->getType() == TT_ConflictStart || 4050 FormatTok->getType() == TT_ConflictEnd || 4051 FormatTok->getType() == TT_ConflictAlternative) { 4052 if (FormatTok->getType() == TT_ConflictStart) 4053 conditionalCompilationStart(/*Unreachable=*/false); 4054 else if (FormatTok->getType() == TT_ConflictAlternative) 4055 conditionalCompilationAlternative(); 4056 else if (FormatTok->getType() == TT_ConflictEnd) 4057 conditionalCompilationEnd(); 4058 FormatTok = Tokens->getNextToken(); 4059 FormatTok->MustBreakBefore = true; 4060 } 4061 4062 auto IsFirstNonCommentOnLine = [](bool FirstNonCommentOnLine, 4063 const FormatToken &Tok, 4064 bool PreviousWasComment) { 4065 auto IsFirstOnLine = [](const FormatToken &Tok) { 4066 return Tok.HasUnescapedNewline || Tok.IsFirst; 4067 }; 4068 4069 // Consider preprocessor directives preceded by block comments as first 4070 // on line. 4071 if (PreviousWasComment) 4072 return FirstNonCommentOnLine || IsFirstOnLine(Tok); 4073 return IsFirstOnLine(Tok); 4074 }; 4075 4076 FirstNonCommentOnLine = IsFirstNonCommentOnLine( 4077 FirstNonCommentOnLine, *FormatTok, PreviousWasComment); 4078 PreviousWasComment = FormatTok->is(tok::comment); 4079 4080 while (!Line->InPPDirective && FormatTok->is(tok::hash) && 4081 FirstNonCommentOnLine) { 4082 distributeComments(Comments, FormatTok); 4083 Comments.clear(); 4084 // If there is an unfinished unwrapped line, we flush the preprocessor 4085 // directives only after that unwrapped line was finished later. 4086 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 4087 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 4088 assert((LevelDifference >= 0 || 4089 static_cast<unsigned>(-LevelDifference) <= Line->Level) && 4090 "LevelDifference makes Line->Level negative"); 4091 Line->Level += LevelDifference; 4092 // Comments stored before the preprocessor directive need to be output 4093 // before the preprocessor directive, at the same level as the 4094 // preprocessor directive, as we consider them to apply to the directive. 4095 if (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash && 4096 PPBranchLevel > 0) 4097 Line->Level += PPBranchLevel; 4098 flushComments(isOnNewLine(*FormatTok)); 4099 parsePPDirective(); 4100 PreviousWasComment = FormatTok->is(tok::comment); 4101 FirstNonCommentOnLine = IsFirstNonCommentOnLine( 4102 FirstNonCommentOnLine, *FormatTok, PreviousWasComment); 4103 } 4104 4105 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) && 4106 !Line->InPPDirective) 4107 continue; 4108 4109 if (!FormatTok->is(tok::comment)) { 4110 distributeComments(Comments, FormatTok); 4111 Comments.clear(); 4112 return; 4113 } 4114 4115 Comments.push_back(FormatTok); 4116 } while (!eof()); 4117 4118 distributeComments(Comments, nullptr); 4119 Comments.clear(); 4120 } 4121 4122 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 4123 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 4124 if (MustBreakBeforeNextToken) { 4125 Line->Tokens.back().Tok->MustBreakBefore = true; 4126 MustBreakBeforeNextToken = false; 4127 } 4128 } 4129 4130 } // end namespace format 4131 } // end namespace clang 4132