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