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