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