1 //===--- UnwrappedLineParser.cpp - Format C++ code ------------------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 /// 10 /// \file 11 /// \brief This file contains the implementation of the UnwrappedLineParser, 12 /// which turns a stream of tokens into UnwrappedLines. 13 /// 14 //===----------------------------------------------------------------------===// 15 16 #include "UnwrappedLineParser.h" 17 #include "llvm/ADT/STLExtras.h" 18 #include "llvm/Support/Debug.h" 19 #include "llvm/Support/raw_ostream.h" 20 21 #define DEBUG_TYPE "format-parser" 22 23 namespace clang { 24 namespace format { 25 26 class FormatTokenSource { 27 public: 28 virtual ~FormatTokenSource() {} 29 virtual FormatToken *getNextToken() = 0; 30 31 virtual unsigned getPosition() = 0; 32 virtual FormatToken *setPosition(unsigned Position) = 0; 33 }; 34 35 namespace { 36 37 class ScopedDeclarationState { 38 public: 39 ScopedDeclarationState(UnwrappedLine &Line, std::vector<bool> &Stack, 40 bool MustBeDeclaration) 41 : Line(Line), Stack(Stack) { 42 Line.MustBeDeclaration = MustBeDeclaration; 43 Stack.push_back(MustBeDeclaration); 44 } 45 ~ScopedDeclarationState() { 46 Stack.pop_back(); 47 if (!Stack.empty()) 48 Line.MustBeDeclaration = Stack.back(); 49 else 50 Line.MustBeDeclaration = true; 51 } 52 53 private: 54 UnwrappedLine &Line; 55 std::vector<bool> &Stack; 56 }; 57 58 static bool isLineComment(const FormatToken &FormatTok) { 59 return FormatTok.is(tok::comment) && 60 FormatTok.TokenText.startswith("//"); 61 } 62 63 // Checks if \p FormatTok is a line comment that continues the line comment 64 // \p Previous. The original column of \p MinColumnToken is used to determine 65 // whether \p FormatTok is indented enough to the right to continue \p Previous. 66 static bool continuesLineComment(const FormatToken &FormatTok, 67 const FormatToken *Previous, 68 const FormatToken *MinColumnToken) { 69 if (!Previous || !MinColumnToken) 70 return false; 71 unsigned MinContinueColumn = 72 MinColumnToken->OriginalColumn + (isLineComment(*MinColumnToken) ? 0 : 1); 73 return isLineComment(FormatTok) && FormatTok.NewlinesBefore == 1 && 74 isLineComment(*Previous) && 75 FormatTok.OriginalColumn >= MinContinueColumn; 76 } 77 78 class ScopedMacroState : public FormatTokenSource { 79 public: 80 ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource, 81 FormatToken *&ResetToken) 82 : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken), 83 PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource), 84 Token(nullptr), PreviousToken(nullptr) { 85 TokenSource = this; 86 Line.Level = 0; 87 Line.InPPDirective = true; 88 } 89 90 ~ScopedMacroState() override { 91 TokenSource = PreviousTokenSource; 92 ResetToken = Token; 93 Line.InPPDirective = false; 94 Line.Level = PreviousLineLevel; 95 } 96 97 FormatToken *getNextToken() override { 98 // The \c UnwrappedLineParser guards against this by never calling 99 // \c getNextToken() after it has encountered the first eof token. 100 assert(!eof()); 101 PreviousToken = Token; 102 Token = PreviousTokenSource->getNextToken(); 103 if (eof()) 104 return getFakeEOF(); 105 return Token; 106 } 107 108 unsigned getPosition() override { return PreviousTokenSource->getPosition(); } 109 110 FormatToken *setPosition(unsigned Position) override { 111 PreviousToken = nullptr; 112 Token = PreviousTokenSource->setPosition(Position); 113 return Token; 114 } 115 116 private: 117 bool eof() { 118 return Token && Token->HasUnescapedNewline && 119 !continuesLineComment(*Token, PreviousToken, 120 /*MinColumnToken=*/PreviousToken); 121 } 122 123 FormatToken *getFakeEOF() { 124 static bool EOFInitialized = false; 125 static FormatToken FormatTok; 126 if (!EOFInitialized) { 127 FormatTok.Tok.startToken(); 128 FormatTok.Tok.setKind(tok::eof); 129 EOFInitialized = true; 130 } 131 return &FormatTok; 132 } 133 134 UnwrappedLine &Line; 135 FormatTokenSource *&TokenSource; 136 FormatToken *&ResetToken; 137 unsigned PreviousLineLevel; 138 FormatTokenSource *PreviousTokenSource; 139 140 FormatToken *Token; 141 FormatToken *PreviousToken; 142 }; 143 144 } // end anonymous namespace 145 146 class ScopedLineState { 147 public: 148 ScopedLineState(UnwrappedLineParser &Parser, 149 bool SwitchToPreprocessorLines = false) 150 : Parser(Parser), OriginalLines(Parser.CurrentLines) { 151 if (SwitchToPreprocessorLines) 152 Parser.CurrentLines = &Parser.PreprocessorDirectives; 153 else if (!Parser.Line->Tokens.empty()) 154 Parser.CurrentLines = &Parser.Line->Tokens.back().Children; 155 PreBlockLine = std::move(Parser.Line); 156 Parser.Line = llvm::make_unique<UnwrappedLine>(); 157 Parser.Line->Level = PreBlockLine->Level; 158 Parser.Line->InPPDirective = PreBlockLine->InPPDirective; 159 } 160 161 ~ScopedLineState() { 162 if (!Parser.Line->Tokens.empty()) { 163 Parser.addUnwrappedLine(); 164 } 165 assert(Parser.Line->Tokens.empty()); 166 Parser.Line = std::move(PreBlockLine); 167 if (Parser.CurrentLines == &Parser.PreprocessorDirectives) 168 Parser.MustBreakBeforeNextToken = true; 169 Parser.CurrentLines = OriginalLines; 170 } 171 172 private: 173 UnwrappedLineParser &Parser; 174 175 std::unique_ptr<UnwrappedLine> PreBlockLine; 176 SmallVectorImpl<UnwrappedLine> *OriginalLines; 177 }; 178 179 class CompoundStatementIndenter { 180 public: 181 CompoundStatementIndenter(UnwrappedLineParser *Parser, 182 const FormatStyle &Style, unsigned &LineLevel) 183 : LineLevel(LineLevel), OldLineLevel(LineLevel) { 184 if (Style.BraceWrapping.AfterControlStatement) 185 Parser->addUnwrappedLine(); 186 if (Style.BraceWrapping.IndentBraces) 187 ++LineLevel; 188 } 189 ~CompoundStatementIndenter() { LineLevel = OldLineLevel; } 190 191 private: 192 unsigned &LineLevel; 193 unsigned OldLineLevel; 194 }; 195 196 namespace { 197 198 class IndexedTokenSource : public FormatTokenSource { 199 public: 200 IndexedTokenSource(ArrayRef<FormatToken *> Tokens) 201 : Tokens(Tokens), Position(-1) {} 202 203 FormatToken *getNextToken() override { 204 ++Position; 205 return Tokens[Position]; 206 } 207 208 unsigned getPosition() override { 209 assert(Position >= 0); 210 return Position; 211 } 212 213 FormatToken *setPosition(unsigned P) override { 214 Position = P; 215 return Tokens[Position]; 216 } 217 218 void reset() { Position = -1; } 219 220 private: 221 ArrayRef<FormatToken *> Tokens; 222 int Position; 223 }; 224 225 } // end anonymous namespace 226 227 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style, 228 const AdditionalKeywords &Keywords, 229 ArrayRef<FormatToken *> Tokens, 230 UnwrappedLineConsumer &Callback) 231 : Line(new UnwrappedLine), MustBreakBeforeNextToken(false), 232 CurrentLines(&Lines), Style(Style), Keywords(Keywords), 233 CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr), 234 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1), 235 IfNdefCondition(nullptr), FoundIncludeGuardStart(false), 236 IncludeGuardRejected(false) {} 237 238 void UnwrappedLineParser::reset() { 239 PPBranchLevel = -1; 240 IfNdefCondition = nullptr; 241 FoundIncludeGuardStart = false; 242 IncludeGuardRejected = false; 243 Line.reset(new UnwrappedLine); 244 CommentsBeforeNextToken.clear(); 245 FormatTok = nullptr; 246 MustBreakBeforeNextToken = false; 247 PreprocessorDirectives.clear(); 248 CurrentLines = &Lines; 249 DeclarationScopeStack.clear(); 250 PPStack.clear(); 251 } 252 253 void UnwrappedLineParser::parse() { 254 IndexedTokenSource TokenSource(AllTokens); 255 do { 256 DEBUG(llvm::dbgs() << "----\n"); 257 reset(); 258 Tokens = &TokenSource; 259 TokenSource.reset(); 260 261 readToken(); 262 parseFile(); 263 // Create line with eof token. 264 pushToken(FormatTok); 265 addUnwrappedLine(); 266 267 for (SmallVectorImpl<UnwrappedLine>::iterator I = Lines.begin(), 268 E = Lines.end(); 269 I != E; ++I) { 270 Callback.consumeUnwrappedLine(*I); 271 } 272 Callback.finishRun(); 273 Lines.clear(); 274 while (!PPLevelBranchIndex.empty() && 275 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) { 276 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1); 277 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1); 278 } 279 if (!PPLevelBranchIndex.empty()) { 280 ++PPLevelBranchIndex.back(); 281 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size()); 282 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back()); 283 } 284 } while (!PPLevelBranchIndex.empty()); 285 } 286 287 void UnwrappedLineParser::parseFile() { 288 // The top-level context in a file always has declarations, except for pre- 289 // processor directives and JavaScript files. 290 bool MustBeDeclaration = 291 !Line->InPPDirective && Style.Language != FormatStyle::LK_JavaScript; 292 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 293 MustBeDeclaration); 294 if (Style.Language == FormatStyle::LK_TextProto) 295 parseBracedList(); 296 else 297 parseLevel(/*HasOpeningBrace=*/false); 298 // Make sure to format the remaining tokens. 299 flushComments(true); 300 addUnwrappedLine(); 301 } 302 303 void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) { 304 bool SwitchLabelEncountered = false; 305 do { 306 tok::TokenKind kind = FormatTok->Tok.getKind(); 307 if (FormatTok->Type == TT_MacroBlockBegin) { 308 kind = tok::l_brace; 309 } else if (FormatTok->Type == TT_MacroBlockEnd) { 310 kind = tok::r_brace; 311 } 312 313 switch (kind) { 314 case tok::comment: 315 nextToken(); 316 addUnwrappedLine(); 317 break; 318 case tok::l_brace: 319 // FIXME: Add parameter whether this can happen - if this happens, we must 320 // be in a non-declaration context. 321 if (!FormatTok->is(TT_MacroBlockBegin) && tryToParseBracedList()) 322 continue; 323 parseBlock(/*MustBeDeclaration=*/false); 324 addUnwrappedLine(); 325 break; 326 case tok::r_brace: 327 if (HasOpeningBrace) 328 return; 329 nextToken(); 330 addUnwrappedLine(); 331 break; 332 case tok::kw_default: 333 case tok::kw_case: 334 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) { 335 // A 'case: string' style field declaration. 336 parseStructuralElement(); 337 break; 338 } 339 if (!SwitchLabelEncountered && 340 (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1))) 341 ++Line->Level; 342 SwitchLabelEncountered = true; 343 parseStructuralElement(); 344 break; 345 default: 346 parseStructuralElement(); 347 break; 348 } 349 } while (!eof()); 350 } 351 352 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) { 353 // We'll parse forward through the tokens until we hit 354 // a closing brace or eof - note that getNextToken() will 355 // parse macros, so this will magically work inside macro 356 // definitions, too. 357 unsigned StoredPosition = Tokens->getPosition(); 358 FormatToken *Tok = FormatTok; 359 const FormatToken *PrevTok = getPreviousToken(); 360 // Keep a stack of positions of lbrace tokens. We will 361 // update information about whether an lbrace starts a 362 // braced init list or a different block during the loop. 363 SmallVector<FormatToken *, 8> LBraceStack; 364 assert(Tok->Tok.is(tok::l_brace)); 365 do { 366 // Get next non-comment token. 367 FormatToken *NextTok; 368 unsigned ReadTokens = 0; 369 do { 370 NextTok = Tokens->getNextToken(); 371 ++ReadTokens; 372 } while (NextTok->is(tok::comment)); 373 374 switch (Tok->Tok.getKind()) { 375 case tok::l_brace: 376 if (Style.Language == FormatStyle::LK_JavaScript && PrevTok) { 377 if (PrevTok->is(tok::colon)) 378 // A colon indicates this code is in a type, or a braced list 379 // following a label in an object literal ({a: {b: 1}}). The code 380 // below could be confused by semicolons between the individual 381 // members in a type member list, which would normally trigger 382 // BK_Block. In both cases, this must be parsed as an inline braced 383 // init. 384 Tok->BlockKind = BK_BracedInit; 385 else if (PrevTok->is(tok::r_paren)) 386 // `) { }` can only occur in function or method declarations in JS. 387 Tok->BlockKind = BK_Block; 388 } else { 389 Tok->BlockKind = BK_Unknown; 390 } 391 LBraceStack.push_back(Tok); 392 break; 393 case tok::r_brace: 394 if (LBraceStack.empty()) 395 break; 396 if (LBraceStack.back()->BlockKind == BK_Unknown) { 397 bool ProbablyBracedList = false; 398 if (Style.Language == FormatStyle::LK_Proto) { 399 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square); 400 } else { 401 // Using OriginalColumn to distinguish between ObjC methods and 402 // binary operators is a bit hacky. 403 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 404 NextTok->OriginalColumn == 0; 405 406 // If there is a comma, semicolon or right paren after the closing 407 // brace, we assume this is a braced initializer list. Note that 408 // regardless how we mark inner braces here, we will overwrite the 409 // BlockKind later if we parse a braced list (where all blocks 410 // inside are by default braced lists), or when we explicitly detect 411 // blocks (for example while parsing lambdas). 412 // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a 413 // braced list in JS. 414 ProbablyBracedList = 415 (Style.Language == FormatStyle::LK_JavaScript && 416 NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in, 417 Keywords.kw_as)) || 418 (Style.isCpp() && NextTok->is(tok::l_paren)) || 419 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 420 tok::r_paren, tok::r_square, tok::l_brace, 421 tok::l_square, tok::ellipsis) || 422 (NextTok->is(tok::identifier) && 423 !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace)) || 424 (NextTok->is(tok::semi) && 425 (!ExpectClassBody || LBraceStack.size() != 1)) || 426 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 427 } 428 if (ProbablyBracedList) { 429 Tok->BlockKind = BK_BracedInit; 430 LBraceStack.back()->BlockKind = BK_BracedInit; 431 } else { 432 Tok->BlockKind = BK_Block; 433 LBraceStack.back()->BlockKind = BK_Block; 434 } 435 } 436 LBraceStack.pop_back(); 437 break; 438 case tok::at: 439 case tok::semi: 440 case tok::kw_if: 441 case tok::kw_while: 442 case tok::kw_for: 443 case tok::kw_switch: 444 case tok::kw_try: 445 case tok::kw___try: 446 if (!LBraceStack.empty() && LBraceStack.back()->BlockKind == BK_Unknown) 447 LBraceStack.back()->BlockKind = BK_Block; 448 break; 449 default: 450 break; 451 } 452 PrevTok = Tok; 453 Tok = NextTok; 454 } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty()); 455 456 // Assume other blocks for all unclosed opening braces. 457 for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) { 458 if (LBraceStack[i]->BlockKind == BK_Unknown) 459 LBraceStack[i]->BlockKind = BK_Block; 460 } 461 462 FormatTok = Tokens->setPosition(StoredPosition); 463 } 464 465 template <class T> 466 static inline void hash_combine(std::size_t &seed, const T &v) { 467 std::hash<T> hasher; 468 seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2); 469 } 470 471 size_t UnwrappedLineParser::computePPHash() const { 472 size_t h = 0; 473 for (const auto &i : PPStack) { 474 hash_combine(h, size_t(i.Kind)); 475 hash_combine(h, i.Line); 476 } 477 return h; 478 } 479 480 void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, bool AddLevel, 481 bool MunchSemi) { 482 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) && 483 "'{' or macro block token expected"); 484 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 485 FormatTok->BlockKind = BK_Block; 486 487 size_t PPStartHash = computePPHash(); 488 489 unsigned InitialLevel = Line->Level; 490 nextToken(/*LevelDifference=*/AddLevel ? 1 : 0); 491 492 if (MacroBlock && FormatTok->is(tok::l_paren)) 493 parseParens(); 494 495 size_t NbPreprocessorDirectives = 496 CurrentLines == &Lines ? PreprocessorDirectives.size() : 0; 497 addUnwrappedLine(); 498 size_t OpeningLineIndex = 499 CurrentLines->empty() 500 ? (UnwrappedLine::kInvalidIndex) 501 : (CurrentLines->size() - 1 - NbPreprocessorDirectives); 502 503 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 504 MustBeDeclaration); 505 if (AddLevel) 506 ++Line->Level; 507 parseLevel(/*HasOpeningBrace=*/true); 508 509 if (eof()) 510 return; 511 512 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd) 513 : !FormatTok->is(tok::r_brace)) { 514 Line->Level = InitialLevel; 515 FormatTok->BlockKind = BK_Block; 516 return; 517 } 518 519 size_t PPEndHash = computePPHash(); 520 521 // Munch the closing brace. 522 nextToken(/*LevelDifference=*/AddLevel ? -1 : 0); 523 524 if (MacroBlock && FormatTok->is(tok::l_paren)) 525 parseParens(); 526 527 if (MunchSemi && FormatTok->Tok.is(tok::semi)) 528 nextToken(); 529 Line->Level = InitialLevel; 530 531 if (PPStartHash == PPEndHash) { 532 Line->MatchingOpeningBlockLineIndex = OpeningLineIndex; 533 if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) { 534 // Update the opening line to add the forward reference as well 535 (*CurrentLines)[OpeningLineIndex].MatchingOpeningBlockLineIndex = 536 CurrentLines->size() - 1; 537 } 538 } 539 } 540 541 static bool isGoogScope(const UnwrappedLine &Line) { 542 // FIXME: Closure-library specific stuff should not be hard-coded but be 543 // configurable. 544 if (Line.Tokens.size() < 4) 545 return false; 546 auto I = Line.Tokens.begin(); 547 if (I->Tok->TokenText != "goog") 548 return false; 549 ++I; 550 if (I->Tok->isNot(tok::period)) 551 return false; 552 ++I; 553 if (I->Tok->TokenText != "scope") 554 return false; 555 ++I; 556 return I->Tok->is(tok::l_paren); 557 } 558 559 static bool isIIFE(const UnwrappedLine &Line, 560 const AdditionalKeywords &Keywords) { 561 // Look for the start of an immediately invoked anonymous function. 562 // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression 563 // This is commonly done in JavaScript to create a new, anonymous scope. 564 // Example: (function() { ... })() 565 if (Line.Tokens.size() < 3) 566 return false; 567 auto I = Line.Tokens.begin(); 568 if (I->Tok->isNot(tok::l_paren)) 569 return false; 570 ++I; 571 if (I->Tok->isNot(Keywords.kw_function)) 572 return false; 573 ++I; 574 return I->Tok->is(tok::l_paren); 575 } 576 577 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 578 const FormatToken &InitialToken) { 579 if (InitialToken.is(tok::kw_namespace)) 580 return Style.BraceWrapping.AfterNamespace; 581 if (InitialToken.is(tok::kw_class)) 582 return Style.BraceWrapping.AfterClass; 583 if (InitialToken.is(tok::kw_union)) 584 return Style.BraceWrapping.AfterUnion; 585 if (InitialToken.is(tok::kw_struct)) 586 return Style.BraceWrapping.AfterStruct; 587 return false; 588 } 589 590 void UnwrappedLineParser::parseChildBlock() { 591 FormatTok->BlockKind = BK_Block; 592 nextToken(); 593 { 594 bool SkipIndent = 595 (Style.Language == FormatStyle::LK_JavaScript && 596 (isGoogScope(*Line) || isIIFE(*Line, Keywords))); 597 ScopedLineState LineState(*this); 598 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 599 /*MustBeDeclaration=*/false); 600 Line->Level += SkipIndent ? 0 : 1; 601 parseLevel(/*HasOpeningBrace=*/true); 602 flushComments(isOnNewLine(*FormatTok)); 603 Line->Level -= SkipIndent ? 0 : 1; 604 } 605 nextToken(); 606 } 607 608 void UnwrappedLineParser::parsePPDirective() { 609 assert(FormatTok->Tok.is(tok::hash) && "'#' expected"); 610 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 611 nextToken(); 612 613 if (!FormatTok->Tok.getIdentifierInfo()) { 614 parsePPUnknown(); 615 return; 616 } 617 618 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 619 case tok::pp_define: 620 parsePPDefine(); 621 return; 622 case tok::pp_if: 623 parsePPIf(/*IfDef=*/false); 624 break; 625 case tok::pp_ifdef: 626 case tok::pp_ifndef: 627 parsePPIf(/*IfDef=*/true); 628 break; 629 case tok::pp_else: 630 parsePPElse(); 631 break; 632 case tok::pp_elif: 633 parsePPElIf(); 634 break; 635 case tok::pp_endif: 636 parsePPEndIf(); 637 break; 638 default: 639 parsePPUnknown(); 640 break; 641 } 642 } 643 644 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 645 size_t Line = CurrentLines->size(); 646 if (CurrentLines == &PreprocessorDirectives) 647 Line += Lines.size(); 648 649 if (Unreachable || 650 (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable)) 651 PPStack.push_back({PP_Unreachable, Line}); 652 else 653 PPStack.push_back({PP_Conditional, Line}); 654 } 655 656 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 657 ++PPBranchLevel; 658 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 659 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 660 PPLevelBranchIndex.push_back(0); 661 PPLevelBranchCount.push_back(0); 662 } 663 PPChainBranchIndex.push(0); 664 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 665 conditionalCompilationCondition(Unreachable || Skip); 666 } 667 668 void UnwrappedLineParser::conditionalCompilationAlternative() { 669 if (!PPStack.empty()) 670 PPStack.pop_back(); 671 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 672 if (!PPChainBranchIndex.empty()) 673 ++PPChainBranchIndex.top(); 674 conditionalCompilationCondition( 675 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 676 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 677 } 678 679 void UnwrappedLineParser::conditionalCompilationEnd() { 680 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 681 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 682 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) { 683 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 684 } 685 } 686 // Guard against #endif's without #if. 687 if (PPBranchLevel > -1) 688 --PPBranchLevel; 689 if (!PPChainBranchIndex.empty()) 690 PPChainBranchIndex.pop(); 691 if (!PPStack.empty()) 692 PPStack.pop_back(); 693 } 694 695 void UnwrappedLineParser::parsePPIf(bool IfDef) { 696 bool IfNDef = FormatTok->is(tok::pp_ifndef); 697 nextToken(); 698 bool Unreachable = false; 699 if (!IfDef && (FormatTok->is(tok::kw_false) || FormatTok->TokenText == "0")) 700 Unreachable = true; 701 if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG") 702 Unreachable = true; 703 conditionalCompilationStart(Unreachable); 704 FormatToken *IfCondition = FormatTok; 705 // If there's a #ifndef on the first line, and the only lines before it are 706 // comments, it could be an include guard. 707 bool MaybeIncludeGuard = IfNDef; 708 if (!IncludeGuardRejected && !FoundIncludeGuardStart && MaybeIncludeGuard) { 709 for (auto &Line : Lines) { 710 if (!Line.Tokens.front().Tok->is(tok::comment)) { 711 MaybeIncludeGuard = false; 712 IncludeGuardRejected = true; 713 break; 714 } 715 } 716 } 717 --PPBranchLevel; 718 parsePPUnknown(); 719 ++PPBranchLevel; 720 if (!IncludeGuardRejected && !FoundIncludeGuardStart && MaybeIncludeGuard) 721 IfNdefCondition = IfCondition; 722 } 723 724 void UnwrappedLineParser::parsePPElse() { 725 // If a potential include guard has an #else, it's not an include guard. 726 if (FoundIncludeGuardStart && PPBranchLevel == 0) 727 FoundIncludeGuardStart = false; 728 conditionalCompilationAlternative(); 729 if (PPBranchLevel > -1) 730 --PPBranchLevel; 731 parsePPUnknown(); 732 ++PPBranchLevel; 733 } 734 735 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 736 737 void UnwrappedLineParser::parsePPEndIf() { 738 conditionalCompilationEnd(); 739 parsePPUnknown(); 740 // If the #endif of a potential include guard is the last thing in the file, 741 // then we count it as a real include guard and subtract one from every 742 // preprocessor indent. 743 unsigned TokenPosition = Tokens->getPosition(); 744 FormatToken *PeekNext = AllTokens[TokenPosition]; 745 if (FoundIncludeGuardStart && PPBranchLevel == -1 && PeekNext->is(tok::eof) && 746 Style.IndentPPDirectives != FormatStyle::PPDIS_None) 747 for (auto &Line : Lines) 748 if (Line.InPPDirective && Line.Level > 0) 749 --Line.Level; 750 } 751 752 void UnwrappedLineParser::parsePPDefine() { 753 nextToken(); 754 755 if (FormatTok->Tok.getKind() != tok::identifier) { 756 parsePPUnknown(); 757 return; 758 } 759 if (IfNdefCondition && IfNdefCondition->TokenText == FormatTok->TokenText) { 760 FoundIncludeGuardStart = true; 761 for (auto &Line : Lines) { 762 if (!Line.Tokens.front().Tok->isOneOf(tok::comment, tok::hash)) { 763 FoundIncludeGuardStart = false; 764 break; 765 } 766 } 767 } 768 IfNdefCondition = nullptr; 769 nextToken(); 770 if (FormatTok->Tok.getKind() == tok::l_paren && 771 FormatTok->WhitespaceRange.getBegin() == 772 FormatTok->WhitespaceRange.getEnd()) { 773 parseParens(); 774 } 775 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash) 776 Line->Level += PPBranchLevel + 1; 777 addUnwrappedLine(); 778 ++Line->Level; 779 780 // Errors during a preprocessor directive can only affect the layout of the 781 // preprocessor directive, and thus we ignore them. An alternative approach 782 // would be to use the same approach we use on the file level (no 783 // re-indentation if there was a structural error) within the macro 784 // definition. 785 parseFile(); 786 } 787 788 void UnwrappedLineParser::parsePPUnknown() { 789 do { 790 nextToken(); 791 } while (!eof()); 792 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash) 793 Line->Level += PPBranchLevel + 1; 794 addUnwrappedLine(); 795 IfNdefCondition = nullptr; 796 } 797 798 // Here we blacklist certain tokens that are not usually the first token in an 799 // unwrapped line. This is used in attempt to distinguish macro calls without 800 // trailing semicolons from other constructs split to several lines. 801 static bool tokenCanStartNewLine(const clang::Token &Tok) { 802 // Semicolon can be a null-statement, l_square can be a start of a macro or 803 // a C++11 attribute, but this doesn't seem to be common. 804 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 805 Tok.isNot(tok::l_square) && 806 // Tokens that can only be used as binary operators and a part of 807 // overloaded operator names. 808 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 809 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 810 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 811 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 812 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 813 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 814 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 815 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 816 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 817 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 818 Tok.isNot(tok::lesslessequal) && 819 // Colon is used in labels, base class lists, initializer lists, 820 // range-based for loops, ternary operator, but should never be the 821 // first token in an unwrapped line. 822 Tok.isNot(tok::colon) && 823 // 'noexcept' is a trailing annotation. 824 Tok.isNot(tok::kw_noexcept); 825 } 826 827 static bool mustBeJSIdent(const AdditionalKeywords &Keywords, 828 const FormatToken *FormatTok) { 829 // FIXME: This returns true for C/C++ keywords like 'struct'. 830 return FormatTok->is(tok::identifier) && 831 (FormatTok->Tok.getIdentifierInfo() == nullptr || 832 !FormatTok->isOneOf( 833 Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async, 834 Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally, 835 Keywords.kw_function, Keywords.kw_import, Keywords.kw_is, 836 Keywords.kw_let, Keywords.kw_var, tok::kw_const, 837 Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements, 838 Keywords.kw_instanceof, Keywords.kw_interface, 839 Keywords.kw_throws, Keywords.kw_from)); 840 } 841 842 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords, 843 const FormatToken *FormatTok) { 844 return FormatTok->Tok.isLiteral() || 845 FormatTok->isOneOf(tok::kw_true, tok::kw_false) || 846 mustBeJSIdent(Keywords, FormatTok); 847 } 848 849 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement 850 // when encountered after a value (see mustBeJSIdentOrValue). 851 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords, 852 const FormatToken *FormatTok) { 853 return FormatTok->isOneOf( 854 tok::kw_return, Keywords.kw_yield, 855 // conditionals 856 tok::kw_if, tok::kw_else, 857 // loops 858 tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break, 859 // switch/case 860 tok::kw_switch, tok::kw_case, 861 // exceptions 862 tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally, 863 // declaration 864 tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let, 865 Keywords.kw_async, Keywords.kw_function, 866 // import/export 867 Keywords.kw_import, tok::kw_export); 868 } 869 870 // readTokenWithJavaScriptASI reads the next token and terminates the current 871 // line if JavaScript Automatic Semicolon Insertion must 872 // happen between the current token and the next token. 873 // 874 // This method is conservative - it cannot cover all edge cases of JavaScript, 875 // but only aims to correctly handle certain well known cases. It *must not* 876 // return true in speculative cases. 877 void UnwrappedLineParser::readTokenWithJavaScriptASI() { 878 FormatToken *Previous = FormatTok; 879 readToken(); 880 FormatToken *Next = FormatTok; 881 882 bool IsOnSameLine = 883 CommentsBeforeNextToken.empty() 884 ? Next->NewlinesBefore == 0 885 : CommentsBeforeNextToken.front()->NewlinesBefore == 0; 886 if (IsOnSameLine) 887 return; 888 889 bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous); 890 bool PreviousStartsTemplateExpr = 891 Previous->is(TT_TemplateString) && Previous->TokenText.endswith("${"); 892 if (PreviousMustBeValue && Line && Line->Tokens.size() > 1) { 893 // If the token before the previous one is an '@', the previous token is an 894 // annotation and can precede another identifier/value. 895 const FormatToken *PrePrevious = std::prev(Line->Tokens.end(), 2)->Tok; 896 if (PrePrevious->is(tok::at)) 897 return; 898 } 899 if (Next->is(tok::exclaim) && PreviousMustBeValue) 900 return addUnwrappedLine(); 901 bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next); 902 bool NextEndsTemplateExpr = 903 Next->is(TT_TemplateString) && Next->TokenText.startswith("}"); 904 if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr && 905 (PreviousMustBeValue || 906 Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus, 907 tok::minusminus))) 908 return addUnwrappedLine(); 909 if ((PreviousMustBeValue || Previous->is(tok::r_paren)) && 910 isJSDeclOrStmt(Keywords, Next)) 911 return addUnwrappedLine(); 912 } 913 914 void UnwrappedLineParser::parseStructuralElement() { 915 assert(!FormatTok->is(tok::l_brace)); 916 if (Style.Language == FormatStyle::LK_TableGen && 917 FormatTok->is(tok::pp_include)) { 918 nextToken(); 919 if (FormatTok->is(tok::string_literal)) 920 nextToken(); 921 addUnwrappedLine(); 922 return; 923 } 924 switch (FormatTok->Tok.getKind()) { 925 case tok::at: 926 nextToken(); 927 if (FormatTok->Tok.is(tok::l_brace)) { 928 nextToken(); 929 parseBracedList(); 930 break; 931 } 932 switch (FormatTok->Tok.getObjCKeywordID()) { 933 case tok::objc_public: 934 case tok::objc_protected: 935 case tok::objc_package: 936 case tok::objc_private: 937 return parseAccessSpecifier(); 938 case tok::objc_interface: 939 case tok::objc_implementation: 940 return parseObjCInterfaceOrImplementation(); 941 case tok::objc_protocol: 942 return parseObjCProtocol(); 943 case tok::objc_end: 944 return; // Handled by the caller. 945 case tok::objc_optional: 946 case tok::objc_required: 947 nextToken(); 948 addUnwrappedLine(); 949 return; 950 case tok::objc_autoreleasepool: 951 nextToken(); 952 if (FormatTok->Tok.is(tok::l_brace)) { 953 if (Style.BraceWrapping.AfterObjCDeclaration) 954 addUnwrappedLine(); 955 parseBlock(/*MustBeDeclaration=*/false); 956 } 957 addUnwrappedLine(); 958 return; 959 case tok::objc_try: 960 // This branch isn't strictly necessary (the kw_try case below would 961 // do this too after the tok::at is parsed above). But be explicit. 962 parseTryCatch(); 963 return; 964 default: 965 break; 966 } 967 break; 968 case tok::kw_asm: 969 nextToken(); 970 if (FormatTok->is(tok::l_brace)) { 971 FormatTok->Type = TT_InlineASMBrace; 972 nextToken(); 973 while (FormatTok && FormatTok->isNot(tok::eof)) { 974 if (FormatTok->is(tok::r_brace)) { 975 FormatTok->Type = TT_InlineASMBrace; 976 nextToken(); 977 addUnwrappedLine(); 978 break; 979 } 980 FormatTok->Finalized = true; 981 nextToken(); 982 } 983 } 984 break; 985 case tok::kw_namespace: 986 parseNamespace(); 987 return; 988 case tok::kw_inline: 989 nextToken(); 990 if (FormatTok->Tok.is(tok::kw_namespace)) { 991 parseNamespace(); 992 return; 993 } 994 break; 995 case tok::kw_public: 996 case tok::kw_protected: 997 case tok::kw_private: 998 if (Style.Language == FormatStyle::LK_Java || 999 Style.Language == FormatStyle::LK_JavaScript) 1000 nextToken(); 1001 else 1002 parseAccessSpecifier(); 1003 return; 1004 case tok::kw_if: 1005 parseIfThenElse(); 1006 return; 1007 case tok::kw_for: 1008 case tok::kw_while: 1009 parseForOrWhileLoop(); 1010 return; 1011 case tok::kw_do: 1012 parseDoWhile(); 1013 return; 1014 case tok::kw_switch: 1015 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1016 // 'switch: string' field declaration. 1017 break; 1018 parseSwitch(); 1019 return; 1020 case tok::kw_default: 1021 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1022 // 'default: string' field declaration. 1023 break; 1024 nextToken(); 1025 parseLabel(); 1026 return; 1027 case tok::kw_case: 1028 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1029 // 'case: string' field declaration. 1030 break; 1031 parseCaseLabel(); 1032 return; 1033 case tok::kw_try: 1034 case tok::kw___try: 1035 parseTryCatch(); 1036 return; 1037 case tok::kw_extern: 1038 nextToken(); 1039 if (FormatTok->Tok.is(tok::string_literal)) { 1040 nextToken(); 1041 if (FormatTok->Tok.is(tok::l_brace)) { 1042 if (Style.BraceWrapping.AfterExternBlock) { 1043 addUnwrappedLine(); 1044 parseBlock(/*MustBeDeclaration=*/true); 1045 } else { 1046 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/false); 1047 } 1048 addUnwrappedLine(); 1049 return; 1050 } 1051 } 1052 break; 1053 case tok::kw_export: 1054 if (Style.Language == FormatStyle::LK_JavaScript) { 1055 parseJavaScriptEs6ImportExport(); 1056 return; 1057 } 1058 break; 1059 case tok::identifier: 1060 if (FormatTok->is(TT_ForEachMacro)) { 1061 parseForOrWhileLoop(); 1062 return; 1063 } 1064 if (FormatTok->is(TT_MacroBlockBegin)) { 1065 parseBlock(/*MustBeDeclaration=*/false, /*AddLevel=*/true, 1066 /*MunchSemi=*/false); 1067 return; 1068 } 1069 if (FormatTok->is(Keywords.kw_import)) { 1070 if (Style.Language == FormatStyle::LK_JavaScript) { 1071 parseJavaScriptEs6ImportExport(); 1072 return; 1073 } 1074 if (Style.Language == FormatStyle::LK_Proto) { 1075 nextToken(); 1076 if (FormatTok->is(tok::kw_public)) 1077 nextToken(); 1078 if (!FormatTok->is(tok::string_literal)) 1079 return; 1080 nextToken(); 1081 if (FormatTok->is(tok::semi)) 1082 nextToken(); 1083 addUnwrappedLine(); 1084 return; 1085 } 1086 } 1087 if (Style.isCpp() && 1088 FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 1089 Keywords.kw_slots, Keywords.kw_qslots)) { 1090 nextToken(); 1091 if (FormatTok->is(tok::colon)) { 1092 nextToken(); 1093 addUnwrappedLine(); 1094 return; 1095 } 1096 } 1097 // In all other cases, parse the declaration. 1098 break; 1099 default: 1100 break; 1101 } 1102 do { 1103 const FormatToken *Previous = getPreviousToken(); 1104 switch (FormatTok->Tok.getKind()) { 1105 case tok::at: 1106 nextToken(); 1107 if (FormatTok->Tok.is(tok::l_brace)) { 1108 nextToken(); 1109 parseBracedList(); 1110 } 1111 break; 1112 case tok::kw_enum: 1113 // Ignore if this is part of "template <enum ...". 1114 if (Previous && Previous->is(tok::less)) { 1115 nextToken(); 1116 break; 1117 } 1118 1119 // parseEnum falls through and does not yet add an unwrapped line as an 1120 // enum definition can start a structural element. 1121 if (!parseEnum()) 1122 break; 1123 // This only applies for C++. 1124 if (!Style.isCpp()) { 1125 addUnwrappedLine(); 1126 return; 1127 } 1128 break; 1129 case tok::kw_typedef: 1130 nextToken(); 1131 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 1132 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS)) 1133 parseEnum(); 1134 break; 1135 case tok::kw_struct: 1136 case tok::kw_union: 1137 case tok::kw_class: 1138 // parseRecord falls through and does not yet add an unwrapped line as a 1139 // record declaration or definition can start a structural element. 1140 parseRecord(); 1141 // This does not apply for Java and JavaScript. 1142 if (Style.Language == FormatStyle::LK_Java || 1143 Style.Language == FormatStyle::LK_JavaScript) { 1144 if (FormatTok->is(tok::semi)) 1145 nextToken(); 1146 addUnwrappedLine(); 1147 return; 1148 } 1149 break; 1150 case tok::period: 1151 nextToken(); 1152 // In Java, classes have an implicit static member "class". 1153 if (Style.Language == FormatStyle::LK_Java && FormatTok && 1154 FormatTok->is(tok::kw_class)) 1155 nextToken(); 1156 if (Style.Language == FormatStyle::LK_JavaScript && FormatTok && 1157 FormatTok->Tok.getIdentifierInfo()) 1158 // JavaScript only has pseudo keywords, all keywords are allowed to 1159 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 1160 nextToken(); 1161 break; 1162 case tok::semi: 1163 nextToken(); 1164 addUnwrappedLine(); 1165 return; 1166 case tok::r_brace: 1167 addUnwrappedLine(); 1168 return; 1169 case tok::l_paren: 1170 parseParens(); 1171 break; 1172 case tok::kw_operator: 1173 nextToken(); 1174 if (FormatTok->isBinaryOperator()) 1175 nextToken(); 1176 break; 1177 case tok::caret: 1178 nextToken(); 1179 if (FormatTok->Tok.isAnyIdentifier() || 1180 FormatTok->isSimpleTypeSpecifier()) 1181 nextToken(); 1182 if (FormatTok->is(tok::l_paren)) 1183 parseParens(); 1184 if (FormatTok->is(tok::l_brace)) 1185 parseChildBlock(); 1186 break; 1187 case tok::l_brace: 1188 if (!tryToParseBracedList()) { 1189 // A block outside of parentheses must be the last part of a 1190 // structural element. 1191 // FIXME: Figure out cases where this is not true, and add projections 1192 // for them (the one we know is missing are lambdas). 1193 if (Style.BraceWrapping.AfterFunction) 1194 addUnwrappedLine(); 1195 FormatTok->Type = TT_FunctionLBrace; 1196 parseBlock(/*MustBeDeclaration=*/false); 1197 addUnwrappedLine(); 1198 return; 1199 } 1200 // Otherwise this was a braced init list, and the structural 1201 // element continues. 1202 break; 1203 case tok::kw_try: 1204 // We arrive here when parsing function-try blocks. 1205 parseTryCatch(); 1206 return; 1207 case tok::identifier: { 1208 if (FormatTok->is(TT_MacroBlockEnd)) { 1209 addUnwrappedLine(); 1210 return; 1211 } 1212 1213 // Function declarations (as opposed to function expressions) are parsed 1214 // on their own unwrapped line by continuing this loop. Function 1215 // expressions (functions that are not on their own line) must not create 1216 // a new unwrapped line, so they are special cased below. 1217 size_t TokenCount = Line->Tokens.size(); 1218 if (Style.Language == FormatStyle::LK_JavaScript && 1219 FormatTok->is(Keywords.kw_function) && 1220 (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is( 1221 Keywords.kw_async)))) { 1222 tryToParseJSFunction(); 1223 break; 1224 } 1225 if ((Style.Language == FormatStyle::LK_JavaScript || 1226 Style.Language == FormatStyle::LK_Java) && 1227 FormatTok->is(Keywords.kw_interface)) { 1228 if (Style.Language == FormatStyle::LK_JavaScript) { 1229 // In JavaScript/TypeScript, "interface" can be used as a standalone 1230 // identifier, e.g. in `var interface = 1;`. If "interface" is 1231 // followed by another identifier, it is very like to be an actual 1232 // interface declaration. 1233 unsigned StoredPosition = Tokens->getPosition(); 1234 FormatToken *Next = Tokens->getNextToken(); 1235 FormatTok = Tokens->setPosition(StoredPosition); 1236 if (Next && !mustBeJSIdent(Keywords, Next)) { 1237 nextToken(); 1238 break; 1239 } 1240 } 1241 parseRecord(); 1242 addUnwrappedLine(); 1243 return; 1244 } 1245 1246 // See if the following token should start a new unwrapped line. 1247 StringRef Text = FormatTok->TokenText; 1248 nextToken(); 1249 if (Line->Tokens.size() == 1 && 1250 // JS doesn't have macros, and within classes colons indicate fields, 1251 // not labels. 1252 Style.Language != FormatStyle::LK_JavaScript) { 1253 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) { 1254 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1255 parseLabel(); 1256 return; 1257 } 1258 // Recognize function-like macro usages without trailing semicolon as 1259 // well as free-standing macros like Q_OBJECT. 1260 bool FunctionLike = FormatTok->is(tok::l_paren); 1261 if (FunctionLike) 1262 parseParens(); 1263 1264 bool FollowedByNewline = 1265 CommentsBeforeNextToken.empty() 1266 ? FormatTok->NewlinesBefore > 0 1267 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1268 1269 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1270 tokenCanStartNewLine(FormatTok->Tok) && Text == Text.upper()) { 1271 addUnwrappedLine(); 1272 return; 1273 } 1274 } 1275 break; 1276 } 1277 case tok::equal: 1278 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType 1279 // TT_JsFatArrow. The always start an expression or a child block if 1280 // followed by a curly. 1281 if (FormatTok->is(TT_JsFatArrow)) { 1282 nextToken(); 1283 if (FormatTok->is(tok::l_brace)) 1284 parseChildBlock(); 1285 break; 1286 } 1287 1288 nextToken(); 1289 if (FormatTok->Tok.is(tok::l_brace)) { 1290 nextToken(); 1291 parseBracedList(); 1292 } else if (Style.Language == FormatStyle::LK_Proto && 1293 FormatTok->Tok.is(tok::less)) { 1294 nextToken(); 1295 parseBracedList(/*ContinueOnSemicolons=*/false, 1296 /*ClosingBraceKind=*/tok::greater); 1297 } 1298 break; 1299 case tok::l_square: 1300 parseSquare(); 1301 break; 1302 case tok::kw_new: 1303 parseNew(); 1304 break; 1305 default: 1306 nextToken(); 1307 break; 1308 } 1309 } while (!eof()); 1310 } 1311 1312 bool UnwrappedLineParser::tryToParseLambda() { 1313 if (!Style.isCpp()) { 1314 nextToken(); 1315 return false; 1316 } 1317 const FormatToken* Previous = getPreviousToken(); 1318 if (Previous && 1319 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 1320 tok::kw_delete) || 1321 Previous->closesScope() || Previous->isSimpleTypeSpecifier())) { 1322 nextToken(); 1323 return false; 1324 } 1325 assert(FormatTok->is(tok::l_square)); 1326 FormatToken &LSquare = *FormatTok; 1327 if (!tryToParseLambdaIntroducer()) 1328 return false; 1329 1330 while (FormatTok->isNot(tok::l_brace)) { 1331 if (FormatTok->isSimpleTypeSpecifier()) { 1332 nextToken(); 1333 continue; 1334 } 1335 switch (FormatTok->Tok.getKind()) { 1336 case tok::l_brace: 1337 break; 1338 case tok::l_paren: 1339 parseParens(); 1340 break; 1341 case tok::amp: 1342 case tok::star: 1343 case tok::kw_const: 1344 case tok::comma: 1345 case tok::less: 1346 case tok::greater: 1347 case tok::identifier: 1348 case tok::numeric_constant: 1349 case tok::coloncolon: 1350 case tok::kw_mutable: 1351 nextToken(); 1352 break; 1353 case tok::arrow: 1354 FormatTok->Type = TT_LambdaArrow; 1355 nextToken(); 1356 break; 1357 default: 1358 return true; 1359 } 1360 } 1361 LSquare.Type = TT_LambdaLSquare; 1362 parseChildBlock(); 1363 return true; 1364 } 1365 1366 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 1367 nextToken(); 1368 if (FormatTok->is(tok::equal)) { 1369 nextToken(); 1370 if (FormatTok->is(tok::r_square)) { 1371 nextToken(); 1372 return true; 1373 } 1374 if (FormatTok->isNot(tok::comma)) 1375 return false; 1376 nextToken(); 1377 } else if (FormatTok->is(tok::amp)) { 1378 nextToken(); 1379 if (FormatTok->is(tok::r_square)) { 1380 nextToken(); 1381 return true; 1382 } 1383 if (!FormatTok->isOneOf(tok::comma, tok::identifier)) { 1384 return false; 1385 } 1386 if (FormatTok->is(tok::comma)) 1387 nextToken(); 1388 } else if (FormatTok->is(tok::r_square)) { 1389 nextToken(); 1390 return true; 1391 } 1392 do { 1393 if (FormatTok->is(tok::amp)) 1394 nextToken(); 1395 if (!FormatTok->isOneOf(tok::identifier, tok::kw_this)) 1396 return false; 1397 nextToken(); 1398 if (FormatTok->is(tok::ellipsis)) 1399 nextToken(); 1400 if (FormatTok->is(tok::comma)) { 1401 nextToken(); 1402 } else if (FormatTok->is(tok::r_square)) { 1403 nextToken(); 1404 return true; 1405 } else { 1406 return false; 1407 } 1408 } while (!eof()); 1409 return false; 1410 } 1411 1412 void UnwrappedLineParser::tryToParseJSFunction() { 1413 assert(FormatTok->is(Keywords.kw_function) || 1414 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)); 1415 if (FormatTok->is(Keywords.kw_async)) 1416 nextToken(); 1417 // Consume "function". 1418 nextToken(); 1419 1420 // Consume * (generator function). Treat it like C++'s overloaded operators. 1421 if (FormatTok->is(tok::star)) { 1422 FormatTok->Type = TT_OverloadedOperator; 1423 nextToken(); 1424 } 1425 1426 // Consume function name. 1427 if (FormatTok->is(tok::identifier)) 1428 nextToken(); 1429 1430 if (FormatTok->isNot(tok::l_paren)) 1431 return; 1432 1433 // Parse formal parameter list. 1434 parseParens(); 1435 1436 if (FormatTok->is(tok::colon)) { 1437 // Parse a type definition. 1438 nextToken(); 1439 1440 // Eat the type declaration. For braced inline object types, balance braces, 1441 // otherwise just parse until finding an l_brace for the function body. 1442 if (FormatTok->is(tok::l_brace)) 1443 tryToParseBracedList(); 1444 else 1445 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 1446 nextToken(); 1447 } 1448 1449 if (FormatTok->is(tok::semi)) 1450 return; 1451 1452 parseChildBlock(); 1453 } 1454 1455 bool UnwrappedLineParser::tryToParseBracedList() { 1456 if (FormatTok->BlockKind == BK_Unknown) 1457 calculateBraceTypes(); 1458 assert(FormatTok->BlockKind != BK_Unknown); 1459 if (FormatTok->BlockKind == BK_Block) 1460 return false; 1461 nextToken(); 1462 parseBracedList(); 1463 return true; 1464 } 1465 1466 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons, 1467 tok::TokenKind ClosingBraceKind) { 1468 bool HasError = false; 1469 1470 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 1471 // replace this by using parseAssigmentExpression() inside. 1472 do { 1473 if (Style.Language == FormatStyle::LK_JavaScript) { 1474 if (FormatTok->is(Keywords.kw_function) || 1475 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) { 1476 tryToParseJSFunction(); 1477 continue; 1478 } 1479 if (FormatTok->is(TT_JsFatArrow)) { 1480 nextToken(); 1481 // Fat arrows can be followed by simple expressions or by child blocks 1482 // in curly braces. 1483 if (FormatTok->is(tok::l_brace)) { 1484 parseChildBlock(); 1485 continue; 1486 } 1487 } 1488 if (FormatTok->is(tok::l_brace)) { 1489 // Could be a method inside of a braced list `{a() { return 1; }}`. 1490 if (tryToParseBracedList()) 1491 continue; 1492 parseChildBlock(); 1493 } 1494 } 1495 if (FormatTok->Tok.getKind() == ClosingBraceKind) { 1496 nextToken(); 1497 return !HasError; 1498 } 1499 switch (FormatTok->Tok.getKind()) { 1500 case tok::caret: 1501 nextToken(); 1502 if (FormatTok->is(tok::l_brace)) { 1503 parseChildBlock(); 1504 } 1505 break; 1506 case tok::l_square: 1507 tryToParseLambda(); 1508 break; 1509 case tok::l_paren: 1510 parseParens(); 1511 // JavaScript can just have free standing methods and getters/setters in 1512 // object literals. Detect them by a "{" following ")". 1513 if (Style.Language == FormatStyle::LK_JavaScript) { 1514 if (FormatTok->is(tok::l_brace)) 1515 parseChildBlock(); 1516 break; 1517 } 1518 break; 1519 case tok::l_brace: 1520 // Assume there are no blocks inside a braced init list apart 1521 // from the ones we explicitly parse out (like lambdas). 1522 FormatTok->BlockKind = BK_BracedInit; 1523 nextToken(); 1524 parseBracedList(); 1525 break; 1526 case tok::less: 1527 if (Style.Language == FormatStyle::LK_Proto) { 1528 nextToken(); 1529 parseBracedList(/*ContinueOnSemicolons=*/false, 1530 /*ClosingBraceKind=*/tok::greater); 1531 } else { 1532 nextToken(); 1533 } 1534 break; 1535 case tok::semi: 1536 // JavaScript (or more precisely TypeScript) can have semicolons in braced 1537 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 1538 // used for error recovery if we have otherwise determined that this is 1539 // a braced list. 1540 if (Style.Language == FormatStyle::LK_JavaScript) { 1541 nextToken(); 1542 break; 1543 } 1544 HasError = true; 1545 if (!ContinueOnSemicolons) 1546 return !HasError; 1547 nextToken(); 1548 break; 1549 case tok::comma: 1550 nextToken(); 1551 break; 1552 default: 1553 nextToken(); 1554 break; 1555 } 1556 } while (!eof()); 1557 return false; 1558 } 1559 1560 void UnwrappedLineParser::parseParens() { 1561 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 1562 nextToken(); 1563 do { 1564 switch (FormatTok->Tok.getKind()) { 1565 case tok::l_paren: 1566 parseParens(); 1567 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 1568 parseChildBlock(); 1569 break; 1570 case tok::r_paren: 1571 nextToken(); 1572 return; 1573 case tok::r_brace: 1574 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1575 return; 1576 case tok::l_square: 1577 tryToParseLambda(); 1578 break; 1579 case tok::l_brace: 1580 if (!tryToParseBracedList()) 1581 parseChildBlock(); 1582 break; 1583 case tok::at: 1584 nextToken(); 1585 if (FormatTok->Tok.is(tok::l_brace)) { 1586 nextToken(); 1587 parseBracedList(); 1588 } 1589 break; 1590 case tok::kw_class: 1591 if (Style.Language == FormatStyle::LK_JavaScript) 1592 parseRecord(/*ParseAsExpr=*/true); 1593 else 1594 nextToken(); 1595 break; 1596 case tok::identifier: 1597 if (Style.Language == FormatStyle::LK_JavaScript && 1598 (FormatTok->is(Keywords.kw_function) || 1599 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function))) 1600 tryToParseJSFunction(); 1601 else 1602 nextToken(); 1603 break; 1604 default: 1605 nextToken(); 1606 break; 1607 } 1608 } while (!eof()); 1609 } 1610 1611 void UnwrappedLineParser::parseSquare() { 1612 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected."); 1613 if (tryToParseLambda()) 1614 return; 1615 do { 1616 switch (FormatTok->Tok.getKind()) { 1617 case tok::l_paren: 1618 parseParens(); 1619 break; 1620 case tok::r_square: 1621 nextToken(); 1622 return; 1623 case tok::r_brace: 1624 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1625 return; 1626 case tok::l_square: 1627 parseSquare(); 1628 break; 1629 case tok::l_brace: { 1630 if (!tryToParseBracedList()) 1631 parseChildBlock(); 1632 break; 1633 } 1634 case tok::at: 1635 nextToken(); 1636 if (FormatTok->Tok.is(tok::l_brace)) { 1637 nextToken(); 1638 parseBracedList(); 1639 } 1640 break; 1641 default: 1642 nextToken(); 1643 break; 1644 } 1645 } while (!eof()); 1646 } 1647 1648 void UnwrappedLineParser::parseIfThenElse() { 1649 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 1650 nextToken(); 1651 if (FormatTok->Tok.is(tok::kw_constexpr)) 1652 nextToken(); 1653 if (FormatTok->Tok.is(tok::l_paren)) 1654 parseParens(); 1655 bool NeedsUnwrappedLine = false; 1656 if (FormatTok->Tok.is(tok::l_brace)) { 1657 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1658 parseBlock(/*MustBeDeclaration=*/false); 1659 if (Style.BraceWrapping.BeforeElse) 1660 addUnwrappedLine(); 1661 else 1662 NeedsUnwrappedLine = true; 1663 } else { 1664 addUnwrappedLine(); 1665 ++Line->Level; 1666 parseStructuralElement(); 1667 --Line->Level; 1668 } 1669 if (FormatTok->Tok.is(tok::kw_else)) { 1670 nextToken(); 1671 if (FormatTok->Tok.is(tok::l_brace)) { 1672 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1673 parseBlock(/*MustBeDeclaration=*/false); 1674 addUnwrappedLine(); 1675 } else if (FormatTok->Tok.is(tok::kw_if)) { 1676 parseIfThenElse(); 1677 } else { 1678 addUnwrappedLine(); 1679 ++Line->Level; 1680 parseStructuralElement(); 1681 if (FormatTok->is(tok::eof)) 1682 addUnwrappedLine(); 1683 --Line->Level; 1684 } 1685 } else if (NeedsUnwrappedLine) { 1686 addUnwrappedLine(); 1687 } 1688 } 1689 1690 void UnwrappedLineParser::parseTryCatch() { 1691 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 1692 nextToken(); 1693 bool NeedsUnwrappedLine = false; 1694 if (FormatTok->is(tok::colon)) { 1695 // We are in a function try block, what comes is an initializer list. 1696 nextToken(); 1697 while (FormatTok->is(tok::identifier)) { 1698 nextToken(); 1699 if (FormatTok->is(tok::l_paren)) 1700 parseParens(); 1701 if (FormatTok->is(tok::comma)) 1702 nextToken(); 1703 } 1704 } 1705 // Parse try with resource. 1706 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) { 1707 parseParens(); 1708 } 1709 if (FormatTok->is(tok::l_brace)) { 1710 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1711 parseBlock(/*MustBeDeclaration=*/false); 1712 if (Style.BraceWrapping.BeforeCatch) { 1713 addUnwrappedLine(); 1714 } else { 1715 NeedsUnwrappedLine = true; 1716 } 1717 } else if (!FormatTok->is(tok::kw_catch)) { 1718 // The C++ standard requires a compound-statement after a try. 1719 // If there's none, we try to assume there's a structuralElement 1720 // and try to continue. 1721 addUnwrappedLine(); 1722 ++Line->Level; 1723 parseStructuralElement(); 1724 --Line->Level; 1725 } 1726 while (1) { 1727 if (FormatTok->is(tok::at)) 1728 nextToken(); 1729 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 1730 tok::kw___finally) || 1731 ((Style.Language == FormatStyle::LK_Java || 1732 Style.Language == FormatStyle::LK_JavaScript) && 1733 FormatTok->is(Keywords.kw_finally)) || 1734 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) || 1735 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally)))) 1736 break; 1737 nextToken(); 1738 while (FormatTok->isNot(tok::l_brace)) { 1739 if (FormatTok->is(tok::l_paren)) { 1740 parseParens(); 1741 continue; 1742 } 1743 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) 1744 return; 1745 nextToken(); 1746 } 1747 NeedsUnwrappedLine = false; 1748 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1749 parseBlock(/*MustBeDeclaration=*/false); 1750 if (Style.BraceWrapping.BeforeCatch) 1751 addUnwrappedLine(); 1752 else 1753 NeedsUnwrappedLine = true; 1754 } 1755 if (NeedsUnwrappedLine) 1756 addUnwrappedLine(); 1757 } 1758 1759 void UnwrappedLineParser::parseNamespace() { 1760 assert(FormatTok->Tok.is(tok::kw_namespace) && "'namespace' expected"); 1761 1762 const FormatToken &InitialToken = *FormatTok; 1763 nextToken(); 1764 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon)) 1765 nextToken(); 1766 if (FormatTok->Tok.is(tok::l_brace)) { 1767 if (ShouldBreakBeforeBrace(Style, InitialToken)) 1768 addUnwrappedLine(); 1769 1770 bool AddLevel = Style.NamespaceIndentation == FormatStyle::NI_All || 1771 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 1772 DeclarationScopeStack.size() > 1); 1773 parseBlock(/*MustBeDeclaration=*/true, AddLevel); 1774 // Munch the semicolon after a namespace. This is more common than one would 1775 // think. Puttin the semicolon into its own line is very ugly. 1776 if (FormatTok->Tok.is(tok::semi)) 1777 nextToken(); 1778 addUnwrappedLine(); 1779 } 1780 // FIXME: Add error handling. 1781 } 1782 1783 void UnwrappedLineParser::parseNew() { 1784 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 1785 nextToken(); 1786 if (Style.Language != FormatStyle::LK_Java) 1787 return; 1788 1789 // In Java, we can parse everything up to the parens, which aren't optional. 1790 do { 1791 // There should not be a ;, { or } before the new's open paren. 1792 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 1793 return; 1794 1795 // Consume the parens. 1796 if (FormatTok->is(tok::l_paren)) { 1797 parseParens(); 1798 1799 // If there is a class body of an anonymous class, consume that as child. 1800 if (FormatTok->is(tok::l_brace)) 1801 parseChildBlock(); 1802 return; 1803 } 1804 nextToken(); 1805 } while (!eof()); 1806 } 1807 1808 void UnwrappedLineParser::parseForOrWhileLoop() { 1809 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 1810 "'for', 'while' or foreach macro expected"); 1811 nextToken(); 1812 // JS' for await ( ... 1813 if (Style.Language == FormatStyle::LK_JavaScript && 1814 FormatTok->is(Keywords.kw_await)) 1815 nextToken(); 1816 if (FormatTok->Tok.is(tok::l_paren)) 1817 parseParens(); 1818 if (FormatTok->Tok.is(tok::l_brace)) { 1819 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1820 parseBlock(/*MustBeDeclaration=*/false); 1821 addUnwrappedLine(); 1822 } else { 1823 addUnwrappedLine(); 1824 ++Line->Level; 1825 parseStructuralElement(); 1826 --Line->Level; 1827 } 1828 } 1829 1830 void UnwrappedLineParser::parseDoWhile() { 1831 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 1832 nextToken(); 1833 if (FormatTok->Tok.is(tok::l_brace)) { 1834 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1835 parseBlock(/*MustBeDeclaration=*/false); 1836 if (Style.BraceWrapping.IndentBraces) 1837 addUnwrappedLine(); 1838 } else { 1839 addUnwrappedLine(); 1840 ++Line->Level; 1841 parseStructuralElement(); 1842 --Line->Level; 1843 } 1844 1845 // FIXME: Add error handling. 1846 if (!FormatTok->Tok.is(tok::kw_while)) { 1847 addUnwrappedLine(); 1848 return; 1849 } 1850 1851 nextToken(); 1852 parseStructuralElement(); 1853 } 1854 1855 void UnwrappedLineParser::parseLabel() { 1856 nextToken(); 1857 unsigned OldLineLevel = Line->Level; 1858 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 1859 --Line->Level; 1860 if (CommentsBeforeNextToken.empty() && FormatTok->Tok.is(tok::l_brace)) { 1861 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1862 parseBlock(/*MustBeDeclaration=*/false); 1863 if (FormatTok->Tok.is(tok::kw_break)) { 1864 if (Style.BraceWrapping.AfterControlStatement) 1865 addUnwrappedLine(); 1866 parseStructuralElement(); 1867 } 1868 addUnwrappedLine(); 1869 } else { 1870 if (FormatTok->is(tok::semi)) 1871 nextToken(); 1872 addUnwrappedLine(); 1873 } 1874 Line->Level = OldLineLevel; 1875 if (FormatTok->isNot(tok::l_brace)) { 1876 parseStructuralElement(); 1877 addUnwrappedLine(); 1878 } 1879 } 1880 1881 void UnwrappedLineParser::parseCaseLabel() { 1882 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 1883 // FIXME: fix handling of complex expressions here. 1884 do { 1885 nextToken(); 1886 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 1887 parseLabel(); 1888 } 1889 1890 void UnwrappedLineParser::parseSwitch() { 1891 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 1892 nextToken(); 1893 if (FormatTok->Tok.is(tok::l_paren)) 1894 parseParens(); 1895 if (FormatTok->Tok.is(tok::l_brace)) { 1896 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1897 parseBlock(/*MustBeDeclaration=*/false); 1898 addUnwrappedLine(); 1899 } else { 1900 addUnwrappedLine(); 1901 ++Line->Level; 1902 parseStructuralElement(); 1903 --Line->Level; 1904 } 1905 } 1906 1907 void UnwrappedLineParser::parseAccessSpecifier() { 1908 nextToken(); 1909 // Understand Qt's slots. 1910 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 1911 nextToken(); 1912 // Otherwise, we don't know what it is, and we'd better keep the next token. 1913 if (FormatTok->Tok.is(tok::colon)) 1914 nextToken(); 1915 addUnwrappedLine(); 1916 } 1917 1918 bool UnwrappedLineParser::parseEnum() { 1919 // Won't be 'enum' for NS_ENUMs. 1920 if (FormatTok->Tok.is(tok::kw_enum)) 1921 nextToken(); 1922 1923 // In TypeScript, "enum" can also be used as property name, e.g. in interface 1924 // declarations. An "enum" keyword followed by a colon would be a syntax 1925 // error and thus assume it is just an identifier. 1926 if (Style.Language == FormatStyle::LK_JavaScript && 1927 FormatTok->isOneOf(tok::colon, tok::question)) 1928 return false; 1929 1930 // Eat up enum class ... 1931 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct)) 1932 nextToken(); 1933 1934 while (FormatTok->Tok.getIdentifierInfo() || 1935 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 1936 tok::greater, tok::comma, tok::question)) { 1937 nextToken(); 1938 // We can have macros or attributes in between 'enum' and the enum name. 1939 if (FormatTok->is(tok::l_paren)) 1940 parseParens(); 1941 if (FormatTok->is(tok::identifier)) { 1942 nextToken(); 1943 // If there are two identifiers in a row, this is likely an elaborate 1944 // return type. In Java, this can be "implements", etc. 1945 if (Style.isCpp() && FormatTok->is(tok::identifier)) 1946 return false; 1947 } 1948 } 1949 1950 // Just a declaration or something is wrong. 1951 if (FormatTok->isNot(tok::l_brace)) 1952 return true; 1953 FormatTok->BlockKind = BK_Block; 1954 1955 if (Style.Language == FormatStyle::LK_Java) { 1956 // Java enums are different. 1957 parseJavaEnumBody(); 1958 return true; 1959 } 1960 if (Style.Language == FormatStyle::LK_Proto) { 1961 parseBlock(/*MustBeDeclaration=*/true); 1962 return true; 1963 } 1964 1965 // Parse enum body. 1966 nextToken(); 1967 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true); 1968 if (HasError) { 1969 if (FormatTok->is(tok::semi)) 1970 nextToken(); 1971 addUnwrappedLine(); 1972 } 1973 return true; 1974 1975 // There is no addUnwrappedLine() here so that we fall through to parsing a 1976 // structural element afterwards. Thus, in "enum A {} n, m;", 1977 // "} n, m;" will end up in one unwrapped line. 1978 } 1979 1980 void UnwrappedLineParser::parseJavaEnumBody() { 1981 // Determine whether the enum is simple, i.e. does not have a semicolon or 1982 // constants with class bodies. Simple enums can be formatted like braced 1983 // lists, contracted to a single line, etc. 1984 unsigned StoredPosition = Tokens->getPosition(); 1985 bool IsSimple = true; 1986 FormatToken *Tok = Tokens->getNextToken(); 1987 while (Tok) { 1988 if (Tok->is(tok::r_brace)) 1989 break; 1990 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 1991 IsSimple = false; 1992 break; 1993 } 1994 // FIXME: This will also mark enums with braces in the arguments to enum 1995 // constants as "not simple". This is probably fine in practice, though. 1996 Tok = Tokens->getNextToken(); 1997 } 1998 FormatTok = Tokens->setPosition(StoredPosition); 1999 2000 if (IsSimple) { 2001 nextToken(); 2002 parseBracedList(); 2003 addUnwrappedLine(); 2004 return; 2005 } 2006 2007 // Parse the body of a more complex enum. 2008 // First add a line for everything up to the "{". 2009 nextToken(); 2010 addUnwrappedLine(); 2011 ++Line->Level; 2012 2013 // Parse the enum constants. 2014 while (FormatTok) { 2015 if (FormatTok->is(tok::l_brace)) { 2016 // Parse the constant's class body. 2017 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 2018 /*MunchSemi=*/false); 2019 } else if (FormatTok->is(tok::l_paren)) { 2020 parseParens(); 2021 } else if (FormatTok->is(tok::comma)) { 2022 nextToken(); 2023 addUnwrappedLine(); 2024 } else if (FormatTok->is(tok::semi)) { 2025 nextToken(); 2026 addUnwrappedLine(); 2027 break; 2028 } else if (FormatTok->is(tok::r_brace)) { 2029 addUnwrappedLine(); 2030 break; 2031 } else { 2032 nextToken(); 2033 } 2034 } 2035 2036 // Parse the class body after the enum's ";" if any. 2037 parseLevel(/*HasOpeningBrace=*/true); 2038 nextToken(); 2039 --Line->Level; 2040 addUnwrappedLine(); 2041 } 2042 2043 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) { 2044 const FormatToken &InitialToken = *FormatTok; 2045 nextToken(); 2046 2047 // The actual identifier can be a nested name specifier, and in macros 2048 // it is often token-pasted. 2049 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 2050 tok::kw___attribute, tok::kw___declspec, 2051 tok::kw_alignas) || 2052 ((Style.Language == FormatStyle::LK_Java || 2053 Style.Language == FormatStyle::LK_JavaScript) && 2054 FormatTok->isOneOf(tok::period, tok::comma))) { 2055 if (Style.Language == FormatStyle::LK_JavaScript && 2056 FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) { 2057 // JavaScript/TypeScript supports inline object types in 2058 // extends/implements positions: 2059 // class Foo implements {bar: number} { } 2060 nextToken(); 2061 if (FormatTok->is(tok::l_brace)) { 2062 tryToParseBracedList(); 2063 continue; 2064 } 2065 } 2066 bool IsNonMacroIdentifier = 2067 FormatTok->is(tok::identifier) && 2068 FormatTok->TokenText != FormatTok->TokenText.upper(); 2069 nextToken(); 2070 // We can have macros or attributes in between 'class' and the class name. 2071 if (!IsNonMacroIdentifier && FormatTok->Tok.is(tok::l_paren)) 2072 parseParens(); 2073 } 2074 2075 // Note that parsing away template declarations here leads to incorrectly 2076 // accepting function declarations as record declarations. 2077 // In general, we cannot solve this problem. Consider: 2078 // class A<int> B() {} 2079 // which can be a function definition or a class definition when B() is a 2080 // macro. If we find enough real-world cases where this is a problem, we 2081 // can parse for the 'template' keyword in the beginning of the statement, 2082 // and thus rule out the record production in case there is no template 2083 // (this would still leave us with an ambiguity between template function 2084 // and class declarations). 2085 if (FormatTok->isOneOf(tok::colon, tok::less)) { 2086 while (!eof()) { 2087 if (FormatTok->is(tok::l_brace)) { 2088 calculateBraceTypes(/*ExpectClassBody=*/true); 2089 if (!tryToParseBracedList()) 2090 break; 2091 } 2092 if (FormatTok->Tok.is(tok::semi)) 2093 return; 2094 nextToken(); 2095 } 2096 } 2097 if (FormatTok->Tok.is(tok::l_brace)) { 2098 if (ParseAsExpr) { 2099 parseChildBlock(); 2100 } else { 2101 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2102 addUnwrappedLine(); 2103 2104 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 2105 /*MunchSemi=*/false); 2106 } 2107 } 2108 // There is no addUnwrappedLine() here so that we fall through to parsing a 2109 // structural element afterwards. Thus, in "class A {} n, m;", 2110 // "} n, m;" will end up in one unwrapped line. 2111 } 2112 2113 void UnwrappedLineParser::parseObjCProtocolList() { 2114 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 2115 do 2116 nextToken(); 2117 while (!eof() && FormatTok->Tok.isNot(tok::greater)); 2118 nextToken(); // Skip '>'. 2119 } 2120 2121 void UnwrappedLineParser::parseObjCUntilAtEnd() { 2122 do { 2123 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 2124 nextToken(); 2125 addUnwrappedLine(); 2126 break; 2127 } 2128 if (FormatTok->is(tok::l_brace)) { 2129 parseBlock(/*MustBeDeclaration=*/false); 2130 // In ObjC interfaces, nothing should be following the "}". 2131 addUnwrappedLine(); 2132 } else if (FormatTok->is(tok::r_brace)) { 2133 // Ignore stray "}". parseStructuralElement doesn't consume them. 2134 nextToken(); 2135 addUnwrappedLine(); 2136 } else { 2137 parseStructuralElement(); 2138 } 2139 } while (!eof()); 2140 } 2141 2142 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 2143 nextToken(); 2144 nextToken(); // interface name 2145 2146 // @interface can be followed by either a base class, or a category. 2147 if (FormatTok->Tok.is(tok::colon)) { 2148 nextToken(); 2149 nextToken(); // base class name 2150 } else if (FormatTok->Tok.is(tok::l_paren)) 2151 // Skip category, if present. 2152 parseParens(); 2153 2154 if (FormatTok->Tok.is(tok::less)) 2155 parseObjCProtocolList(); 2156 2157 if (FormatTok->Tok.is(tok::l_brace)) { 2158 if (Style.BraceWrapping.AfterObjCDeclaration) 2159 addUnwrappedLine(); 2160 parseBlock(/*MustBeDeclaration=*/true); 2161 } 2162 2163 // With instance variables, this puts '}' on its own line. Without instance 2164 // variables, this ends the @interface line. 2165 addUnwrappedLine(); 2166 2167 parseObjCUntilAtEnd(); 2168 } 2169 2170 void UnwrappedLineParser::parseObjCProtocol() { 2171 nextToken(); 2172 nextToken(); // protocol name 2173 2174 if (FormatTok->Tok.is(tok::less)) 2175 parseObjCProtocolList(); 2176 2177 // Check for protocol declaration. 2178 if (FormatTok->Tok.is(tok::semi)) { 2179 nextToken(); 2180 return addUnwrappedLine(); 2181 } 2182 2183 addUnwrappedLine(); 2184 parseObjCUntilAtEnd(); 2185 } 2186 2187 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 2188 bool IsImport = FormatTok->is(Keywords.kw_import); 2189 assert(IsImport || FormatTok->is(tok::kw_export)); 2190 nextToken(); 2191 2192 // Consume the "default" in "export default class/function". 2193 if (FormatTok->is(tok::kw_default)) 2194 nextToken(); 2195 2196 // Consume "async function", "function" and "default function", so that these 2197 // get parsed as free-standing JS functions, i.e. do not require a trailing 2198 // semicolon. 2199 if (FormatTok->is(Keywords.kw_async)) 2200 nextToken(); 2201 if (FormatTok->is(Keywords.kw_function)) { 2202 nextToken(); 2203 return; 2204 } 2205 2206 // For imports, `export *`, `export {...}`, consume the rest of the line up 2207 // to the terminating `;`. For everything else, just return and continue 2208 // parsing the structural element, i.e. the declaration or expression for 2209 // `export default`. 2210 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 2211 !FormatTok->isStringLiteral()) 2212 return; 2213 2214 while (!eof()) { 2215 if (FormatTok->is(tok::semi)) 2216 return; 2217 if (Line->Tokens.size() == 0) { 2218 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 2219 // import statement should terminate. 2220 return; 2221 } 2222 if (FormatTok->is(tok::l_brace)) { 2223 FormatTok->BlockKind = BK_Block; 2224 nextToken(); 2225 parseBracedList(); 2226 } else { 2227 nextToken(); 2228 } 2229 } 2230 } 2231 2232 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 2233 StringRef Prefix = "") { 2234 llvm::dbgs() << Prefix << "Line(" << Line.Level << ")" 2235 << (Line.InPPDirective ? " MACRO" : "") << ": "; 2236 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2237 E = Line.Tokens.end(); 2238 I != E; ++I) { 2239 llvm::dbgs() << I->Tok->Tok.getName() << "[" 2240 << "T=" << I->Tok->Type 2241 << ", OC=" << I->Tok->OriginalColumn << "] "; 2242 } 2243 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2244 E = Line.Tokens.end(); 2245 I != E; ++I) { 2246 const UnwrappedLineNode &Node = *I; 2247 for (SmallVectorImpl<UnwrappedLine>::const_iterator 2248 I = Node.Children.begin(), 2249 E = Node.Children.end(); 2250 I != E; ++I) { 2251 printDebugInfo(*I, "\nChild: "); 2252 } 2253 } 2254 llvm::dbgs() << "\n"; 2255 } 2256 2257 void UnwrappedLineParser::addUnwrappedLine() { 2258 if (Line->Tokens.empty()) 2259 return; 2260 DEBUG({ 2261 if (CurrentLines == &Lines) 2262 printDebugInfo(*Line); 2263 }); 2264 CurrentLines->push_back(std::move(*Line)); 2265 Line->Tokens.clear(); 2266 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex; 2267 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 2268 CurrentLines->append( 2269 std::make_move_iterator(PreprocessorDirectives.begin()), 2270 std::make_move_iterator(PreprocessorDirectives.end())); 2271 PreprocessorDirectives.clear(); 2272 } 2273 } 2274 2275 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 2276 2277 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 2278 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 2279 FormatTok.NewlinesBefore > 0; 2280 } 2281 2282 // Checks if \p FormatTok is a line comment that continues the line comment 2283 // section on \p Line. 2284 static bool continuesLineCommentSection(const FormatToken &FormatTok, 2285 const UnwrappedLine &Line, 2286 llvm::Regex &CommentPragmasRegex) { 2287 if (Line.Tokens.empty()) 2288 return false; 2289 2290 StringRef IndentContent = FormatTok.TokenText; 2291 if (FormatTok.TokenText.startswith("//") || 2292 FormatTok.TokenText.startswith("/*")) 2293 IndentContent = FormatTok.TokenText.substr(2); 2294 if (CommentPragmasRegex.match(IndentContent)) 2295 return false; 2296 2297 // If Line starts with a line comment, then FormatTok continues the comment 2298 // section if its original column is greater or equal to the original start 2299 // column of the line. 2300 // 2301 // Define the min column token of a line as follows: if a line ends in '{' or 2302 // contains a '{' followed by a line comment, then the min column token is 2303 // that '{'. Otherwise, the min column token of the line is the first token of 2304 // the line. 2305 // 2306 // If Line starts with a token other than a line comment, then FormatTok 2307 // continues the comment section if its original column is greater than the 2308 // original start column of the min column token of the line. 2309 // 2310 // For example, the second line comment continues the first in these cases: 2311 // 2312 // // first line 2313 // // second line 2314 // 2315 // and: 2316 // 2317 // // first line 2318 // // second line 2319 // 2320 // and: 2321 // 2322 // int i; // first line 2323 // // second line 2324 // 2325 // and: 2326 // 2327 // do { // first line 2328 // // second line 2329 // int i; 2330 // } while (true); 2331 // 2332 // and: 2333 // 2334 // enum { 2335 // a, // first line 2336 // // second line 2337 // b 2338 // }; 2339 // 2340 // The second line comment doesn't continue the first in these cases: 2341 // 2342 // // first line 2343 // // second line 2344 // 2345 // and: 2346 // 2347 // int i; // first line 2348 // // second line 2349 // 2350 // and: 2351 // 2352 // do { // first line 2353 // // second line 2354 // int i; 2355 // } while (true); 2356 // 2357 // and: 2358 // 2359 // enum { 2360 // a, // first line 2361 // // second line 2362 // }; 2363 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 2364 2365 // Scan for '{//'. If found, use the column of '{' as a min column for line 2366 // comment section continuation. 2367 const FormatToken *PreviousToken = nullptr; 2368 for (const UnwrappedLineNode &Node : Line.Tokens) { 2369 if (PreviousToken && PreviousToken->is(tok::l_brace) && 2370 isLineComment(*Node.Tok)) { 2371 MinColumnToken = PreviousToken; 2372 break; 2373 } 2374 PreviousToken = Node.Tok; 2375 2376 // Grab the last newline preceding a token in this unwrapped line. 2377 if (Node.Tok->NewlinesBefore > 0) { 2378 MinColumnToken = Node.Tok; 2379 } 2380 } 2381 if (PreviousToken && PreviousToken->is(tok::l_brace)) { 2382 MinColumnToken = PreviousToken; 2383 } 2384 2385 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok, 2386 MinColumnToken); 2387 } 2388 2389 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 2390 bool JustComments = Line->Tokens.empty(); 2391 for (SmallVectorImpl<FormatToken *>::const_iterator 2392 I = CommentsBeforeNextToken.begin(), 2393 E = CommentsBeforeNextToken.end(); 2394 I != E; ++I) { 2395 // Line comments that belong to the same line comment section are put on the 2396 // same line since later we might want to reflow content between them. 2397 // Additional fine-grained breaking of line comment sections is controlled 2398 // by the class BreakableLineCommentSection in case it is desirable to keep 2399 // several line comment sections in the same unwrapped line. 2400 // 2401 // FIXME: Consider putting separate line comment sections as children to the 2402 // unwrapped line instead. 2403 (*I)->ContinuesLineCommentSection = 2404 continuesLineCommentSection(**I, *Line, CommentPragmasRegex); 2405 if (isOnNewLine(**I) && JustComments && !(*I)->ContinuesLineCommentSection) 2406 addUnwrappedLine(); 2407 pushToken(*I); 2408 } 2409 if (NewlineBeforeNext && JustComments) 2410 addUnwrappedLine(); 2411 CommentsBeforeNextToken.clear(); 2412 } 2413 2414 void UnwrappedLineParser::nextToken(int LevelDifference) { 2415 if (eof()) 2416 return; 2417 flushComments(isOnNewLine(*FormatTok)); 2418 pushToken(FormatTok); 2419 if (Style.Language != FormatStyle::LK_JavaScript) 2420 readToken(LevelDifference); 2421 else 2422 readTokenWithJavaScriptASI(); 2423 } 2424 2425 const FormatToken *UnwrappedLineParser::getPreviousToken() { 2426 // FIXME: This is a dirty way to access the previous token. Find a better 2427 // solution. 2428 if (!Line || Line->Tokens.empty()) 2429 return nullptr; 2430 return Line->Tokens.back().Tok; 2431 } 2432 2433 void UnwrappedLineParser::distributeComments( 2434 const SmallVectorImpl<FormatToken *> &Comments, 2435 const FormatToken *NextTok) { 2436 // Whether or not a line comment token continues a line is controlled by 2437 // the method continuesLineCommentSection, with the following caveat: 2438 // 2439 // Define a trail of Comments to be a nonempty proper postfix of Comments such 2440 // that each comment line from the trail is aligned with the next token, if 2441 // the next token exists. If a trail exists, the beginning of the maximal 2442 // trail is marked as a start of a new comment section. 2443 // 2444 // For example in this code: 2445 // 2446 // int a; // line about a 2447 // // line 1 about b 2448 // // line 2 about b 2449 // int b; 2450 // 2451 // the two lines about b form a maximal trail, so there are two sections, the 2452 // first one consisting of the single comment "// line about a" and the 2453 // second one consisting of the next two comments. 2454 if (Comments.empty()) 2455 return; 2456 bool ShouldPushCommentsInCurrentLine = true; 2457 bool HasTrailAlignedWithNextToken = false; 2458 unsigned StartOfTrailAlignedWithNextToken = 0; 2459 if (NextTok) { 2460 // We are skipping the first element intentionally. 2461 for (unsigned i = Comments.size() - 1; i > 0; --i) { 2462 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) { 2463 HasTrailAlignedWithNextToken = true; 2464 StartOfTrailAlignedWithNextToken = i; 2465 } 2466 } 2467 } 2468 for (unsigned i = 0, e = Comments.size(); i < e; ++i) { 2469 FormatToken *FormatTok = Comments[i]; 2470 if (HasTrailAlignedWithNextToken && 2471 i == StartOfTrailAlignedWithNextToken) { 2472 FormatTok->ContinuesLineCommentSection = false; 2473 } else { 2474 FormatTok->ContinuesLineCommentSection = 2475 continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex); 2476 } 2477 if (!FormatTok->ContinuesLineCommentSection && 2478 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) { 2479 ShouldPushCommentsInCurrentLine = false; 2480 } 2481 if (ShouldPushCommentsInCurrentLine) { 2482 pushToken(FormatTok); 2483 } else { 2484 CommentsBeforeNextToken.push_back(FormatTok); 2485 } 2486 } 2487 } 2488 2489 void UnwrappedLineParser::readToken(int LevelDifference) { 2490 SmallVector<FormatToken *, 1> Comments; 2491 do { 2492 FormatTok = Tokens->getNextToken(); 2493 assert(FormatTok); 2494 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 2495 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 2496 distributeComments(Comments, FormatTok); 2497 Comments.clear(); 2498 // If there is an unfinished unwrapped line, we flush the preprocessor 2499 // directives only after that unwrapped line was finished later. 2500 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 2501 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 2502 assert((LevelDifference >= 0 || 2503 static_cast<unsigned>(-LevelDifference) <= Line->Level) && 2504 "LevelDifference makes Line->Level negative"); 2505 Line->Level += LevelDifference; 2506 // Comments stored before the preprocessor directive need to be output 2507 // before the preprocessor directive, at the same level as the 2508 // preprocessor directive, as we consider them to apply to the directive. 2509 flushComments(isOnNewLine(*FormatTok)); 2510 parsePPDirective(); 2511 } 2512 while (FormatTok->Type == TT_ConflictStart || 2513 FormatTok->Type == TT_ConflictEnd || 2514 FormatTok->Type == TT_ConflictAlternative) { 2515 if (FormatTok->Type == TT_ConflictStart) { 2516 conditionalCompilationStart(/*Unreachable=*/false); 2517 } else if (FormatTok->Type == TT_ConflictAlternative) { 2518 conditionalCompilationAlternative(); 2519 } else if (FormatTok->Type == TT_ConflictEnd) { 2520 conditionalCompilationEnd(); 2521 } 2522 FormatTok = Tokens->getNextToken(); 2523 FormatTok->MustBreakBefore = true; 2524 } 2525 2526 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) && 2527 !Line->InPPDirective) { 2528 continue; 2529 } 2530 2531 if (!FormatTok->Tok.is(tok::comment)) { 2532 distributeComments(Comments, FormatTok); 2533 Comments.clear(); 2534 return; 2535 } 2536 2537 Comments.push_back(FormatTok); 2538 } while (!eof()); 2539 2540 distributeComments(Comments, nullptr); 2541 Comments.clear(); 2542 } 2543 2544 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 2545 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 2546 if (MustBreakBeforeNextToken) { 2547 Line->Tokens.back().Tok->MustBreakBefore = true; 2548 MustBreakBeforeNextToken = false; 2549 } 2550 } 2551 2552 } // end namespace format 2553 } // end namespace clang 2554