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 = !Line->InPPDirective && !Style.isJavaScript(); 370 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 371 MustBeDeclaration); 372 if (Style.Language == FormatStyle::LK_TextProto) 373 parseBracedList(); 374 else 375 parseLevel(/*HasOpeningBrace=*/false); 376 // Make sure to format the remaining tokens. 377 // 378 // LK_TextProto is special since its top-level is parsed as the body of a 379 // braced list, which does not necessarily have natural line separators such 380 // as a semicolon. Comments after the last entry that have been determined to 381 // not belong to that line, as in: 382 // key: value 383 // // endfile comment 384 // do not have a chance to be put on a line of their own until this point. 385 // Here we add this newline before end-of-file comments. 386 if (Style.Language == FormatStyle::LK_TextProto && 387 !CommentsBeforeNextToken.empty()) 388 addUnwrappedLine(); 389 flushComments(true); 390 addUnwrappedLine(); 391 } 392 393 void UnwrappedLineParser::parseCSharpGenericTypeConstraint() { 394 do { 395 switch (FormatTok->Tok.getKind()) { 396 case tok::l_brace: 397 return; 398 default: 399 if (FormatTok->is(Keywords.kw_where)) { 400 addUnwrappedLine(); 401 nextToken(); 402 parseCSharpGenericTypeConstraint(); 403 break; 404 } 405 nextToken(); 406 break; 407 } 408 } while (!eof()); 409 } 410 411 void UnwrappedLineParser::parseCSharpAttribute() { 412 int UnpairedSquareBrackets = 1; 413 do { 414 switch (FormatTok->Tok.getKind()) { 415 case tok::r_square: 416 nextToken(); 417 --UnpairedSquareBrackets; 418 if (UnpairedSquareBrackets == 0) { 419 addUnwrappedLine(); 420 return; 421 } 422 break; 423 case tok::l_square: 424 ++UnpairedSquareBrackets; 425 nextToken(); 426 break; 427 default: 428 nextToken(); 429 break; 430 } 431 } while (!eof()); 432 } 433 434 void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) { 435 bool SwitchLabelEncountered = false; 436 do { 437 tok::TokenKind kind = FormatTok->Tok.getKind(); 438 if (FormatTok->getType() == TT_MacroBlockBegin) { 439 kind = tok::l_brace; 440 } else if (FormatTok->getType() == TT_MacroBlockEnd) { 441 kind = tok::r_brace; 442 } 443 444 switch (kind) { 445 case tok::comment: 446 nextToken(); 447 addUnwrappedLine(); 448 break; 449 case tok::l_brace: 450 // FIXME: Add parameter whether this can happen - if this happens, we must 451 // be in a non-declaration context. 452 if (!FormatTok->is(TT_MacroBlockBegin) && tryToParseBracedList()) 453 continue; 454 parseBlock(); 455 addUnwrappedLine(); 456 break; 457 case tok::r_brace: 458 if (HasOpeningBrace) 459 return; 460 nextToken(); 461 addUnwrappedLine(); 462 break; 463 case tok::kw_default: { 464 unsigned StoredPosition = Tokens->getPosition(); 465 FormatToken *Next; 466 do { 467 Next = Tokens->getNextToken(); 468 } while (Next->is(tok::comment)); 469 FormatTok = Tokens->setPosition(StoredPosition); 470 if (Next && Next->isNot(tok::colon)) { 471 // default not followed by ':' is not a case label; treat it like 472 // an identifier. 473 parseStructuralElement(); 474 break; 475 } 476 // Else, if it is 'default:', fall through to the case handling. 477 LLVM_FALLTHROUGH; 478 } 479 case tok::kw_case: 480 if (Style.isJavaScript() && Line->MustBeDeclaration) { 481 // A 'case: string' style field declaration. 482 parseStructuralElement(); 483 break; 484 } 485 if (!SwitchLabelEncountered && 486 (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1))) 487 ++Line->Level; 488 SwitchLabelEncountered = true; 489 parseStructuralElement(); 490 break; 491 case tok::l_square: 492 if (Style.isCSharp()) { 493 nextToken(); 494 parseCSharpAttribute(); 495 break; 496 } 497 LLVM_FALLTHROUGH; 498 default: 499 parseStructuralElement(!HasOpeningBrace); 500 break; 501 } 502 } while (!eof()); 503 } 504 505 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) { 506 // We'll parse forward through the tokens until we hit 507 // a closing brace or eof - note that getNextToken() will 508 // parse macros, so this will magically work inside macro 509 // definitions, too. 510 unsigned StoredPosition = Tokens->getPosition(); 511 FormatToken *Tok = FormatTok; 512 const FormatToken *PrevTok = Tok->Previous; 513 // Keep a stack of positions of lbrace tokens. We will 514 // update information about whether an lbrace starts a 515 // braced init list or a different block during the loop. 516 SmallVector<FormatToken *, 8> LBraceStack; 517 assert(Tok->Tok.is(tok::l_brace)); 518 do { 519 // Get next non-comment token. 520 FormatToken *NextTok; 521 unsigned ReadTokens = 0; 522 do { 523 NextTok = Tokens->getNextToken(); 524 ++ReadTokens; 525 } while (NextTok->is(tok::comment)); 526 527 switch (Tok->Tok.getKind()) { 528 case tok::l_brace: 529 if (Style.isJavaScript() && PrevTok) { 530 if (PrevTok->isOneOf(tok::colon, tok::less)) 531 // A ':' indicates this code is in a type, or a braced list 532 // following a label in an object literal ({a: {b: 1}}). 533 // A '<' could be an object used in a comparison, but that is nonsense 534 // code (can never return true), so more likely it is a generic type 535 // argument (`X<{a: string; b: number}>`). 536 // The code below could be confused by semicolons between the 537 // individual members in a type member list, which would normally 538 // trigger BK_Block. In both cases, this must be parsed as an inline 539 // braced init. 540 Tok->setBlockKind(BK_BracedInit); 541 else if (PrevTok->is(tok::r_paren)) 542 // `) { }` can only occur in function or method declarations in JS. 543 Tok->setBlockKind(BK_Block); 544 } else { 545 Tok->setBlockKind(BK_Unknown); 546 } 547 LBraceStack.push_back(Tok); 548 break; 549 case tok::r_brace: 550 if (LBraceStack.empty()) 551 break; 552 if (LBraceStack.back()->is(BK_Unknown)) { 553 bool ProbablyBracedList = false; 554 if (Style.Language == FormatStyle::LK_Proto) { 555 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square); 556 } else { 557 // Skip NextTok over preprocessor lines, otherwise we may not 558 // properly diagnose the block as a braced intializer 559 // if the comma separator appears after the pp directive. 560 while (NextTok->is(tok::hash)) { 561 ScopedMacroState MacroState(*Line, Tokens, NextTok); 562 do { 563 NextTok = Tokens->getNextToken(); 564 ++ReadTokens; 565 } while (NextTok->isNot(tok::eof)); 566 } 567 568 // Using OriginalColumn to distinguish between ObjC methods and 569 // binary operators is a bit hacky. 570 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 571 NextTok->OriginalColumn == 0; 572 573 // If there is a comma, semicolon or right paren after the closing 574 // brace, we assume this is a braced initializer list. Note that 575 // regardless how we mark inner braces here, we will overwrite the 576 // BlockKind later if we parse a braced list (where all blocks 577 // inside are by default braced lists), or when we explicitly detect 578 // blocks (for example while parsing lambdas). 579 // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a 580 // braced list in JS. 581 ProbablyBracedList = 582 (Style.isJavaScript() && 583 NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in, 584 Keywords.kw_as)) || 585 (Style.isCpp() && NextTok->is(tok::l_paren)) || 586 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 587 tok::r_paren, tok::r_square, tok::l_brace, 588 tok::ellipsis) || 589 (NextTok->is(tok::identifier) && 590 !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace)) || 591 (NextTok->is(tok::semi) && 592 (!ExpectClassBody || LBraceStack.size() != 1)) || 593 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 594 if (!Style.isCSharp() && NextTok->is(tok::l_square)) { 595 // We can have an array subscript after a braced init 596 // list, but C++11 attributes are expected after blocks. 597 NextTok = Tokens->getNextToken(); 598 ++ReadTokens; 599 ProbablyBracedList = NextTok->isNot(tok::l_square); 600 } 601 } 602 if (ProbablyBracedList) { 603 Tok->setBlockKind(BK_BracedInit); 604 LBraceStack.back()->setBlockKind(BK_BracedInit); 605 } else { 606 Tok->setBlockKind(BK_Block); 607 LBraceStack.back()->setBlockKind(BK_Block); 608 } 609 } 610 LBraceStack.pop_back(); 611 break; 612 case tok::identifier: 613 if (!Tok->is(TT_StatementMacro)) 614 break; 615 LLVM_FALLTHROUGH; 616 case tok::at: 617 case tok::semi: 618 case tok::kw_if: 619 case tok::kw_while: 620 case tok::kw_for: 621 case tok::kw_switch: 622 case tok::kw_try: 623 case tok::kw___try: 624 if (!LBraceStack.empty() && LBraceStack.back()->is(BK_Unknown)) 625 LBraceStack.back()->setBlockKind(BK_Block); 626 break; 627 default: 628 break; 629 } 630 PrevTok = Tok; 631 Tok = NextTok; 632 } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty()); 633 634 // Assume other blocks for all unclosed opening braces. 635 for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) { 636 if (LBraceStack[i]->is(BK_Unknown)) 637 LBraceStack[i]->setBlockKind(BK_Block); 638 } 639 640 FormatTok = Tokens->setPosition(StoredPosition); 641 } 642 643 template <class T> 644 static inline void hash_combine(std::size_t &seed, const T &v) { 645 std::hash<T> hasher; 646 seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2); 647 } 648 649 size_t UnwrappedLineParser::computePPHash() const { 650 size_t h = 0; 651 for (const auto &i : PPStack) { 652 hash_combine(h, size_t(i.Kind)); 653 hash_combine(h, i.Line); 654 } 655 return h; 656 } 657 658 void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, unsigned AddLevels, 659 bool MunchSemi, 660 bool UnindentWhitesmithsBraces) { 661 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) && 662 "'{' or macro block token expected"); 663 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 664 FormatTok->setBlockKind(BK_Block); 665 666 // For Whitesmiths mode, jump to the next level prior to skipping over the 667 // braces. 668 if (AddLevels > 0 && Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) 669 ++Line->Level; 670 671 size_t PPStartHash = computePPHash(); 672 673 unsigned InitialLevel = Line->Level; 674 nextToken(/*LevelDifference=*/AddLevels); 675 676 if (MacroBlock && FormatTok->is(tok::l_paren)) 677 parseParens(); 678 679 size_t NbPreprocessorDirectives = 680 CurrentLines == &Lines ? PreprocessorDirectives.size() : 0; 681 addUnwrappedLine(); 682 size_t OpeningLineIndex = 683 CurrentLines->empty() 684 ? (UnwrappedLine::kInvalidIndex) 685 : (CurrentLines->size() - 1 - NbPreprocessorDirectives); 686 687 // Whitesmiths is weird here. The brace needs to be indented for the namespace 688 // block, but the block itself may not be indented depending on the style 689 // settings. This allows the format to back up one level in those cases. 690 if (UnindentWhitesmithsBraces) 691 --Line->Level; 692 693 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 694 MustBeDeclaration); 695 if (AddLevels > 0u && Style.BreakBeforeBraces != FormatStyle::BS_Whitesmiths) 696 Line->Level += AddLevels; 697 parseLevel(/*HasOpeningBrace=*/true); 698 699 if (eof()) 700 return; 701 702 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd) 703 : !FormatTok->is(tok::r_brace)) { 704 Line->Level = InitialLevel; 705 FormatTok->setBlockKind(BK_Block); 706 return; 707 } 708 709 size_t PPEndHash = computePPHash(); 710 711 // Munch the closing brace. 712 nextToken(/*LevelDifference=*/-AddLevels); 713 714 if (MacroBlock && FormatTok->is(tok::l_paren)) 715 parseParens(); 716 717 if (FormatTok->is(tok::arrow)) { 718 // Following the } we can find a trailing return type arrow 719 // as part of an implicit conversion constraint. 720 nextToken(); 721 parseStructuralElement(); 722 } 723 724 if (MunchSemi && FormatTok->Tok.is(tok::semi)) 725 nextToken(); 726 727 Line->Level = InitialLevel; 728 729 if (PPStartHash == PPEndHash) { 730 Line->MatchingOpeningBlockLineIndex = OpeningLineIndex; 731 if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) { 732 // Update the opening line to add the forward reference as well 733 (*CurrentLines)[OpeningLineIndex].MatchingClosingBlockLineIndex = 734 CurrentLines->size() - 1; 735 } 736 } 737 } 738 739 static bool isGoogScope(const UnwrappedLine &Line) { 740 // FIXME: Closure-library specific stuff should not be hard-coded but be 741 // configurable. 742 if (Line.Tokens.size() < 4) 743 return false; 744 auto I = Line.Tokens.begin(); 745 if (I->Tok->TokenText != "goog") 746 return false; 747 ++I; 748 if (I->Tok->isNot(tok::period)) 749 return false; 750 ++I; 751 if (I->Tok->TokenText != "scope") 752 return false; 753 ++I; 754 return I->Tok->is(tok::l_paren); 755 } 756 757 static bool isIIFE(const UnwrappedLine &Line, 758 const AdditionalKeywords &Keywords) { 759 // Look for the start of an immediately invoked anonymous function. 760 // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression 761 // This is commonly done in JavaScript to create a new, anonymous scope. 762 // Example: (function() { ... })() 763 if (Line.Tokens.size() < 3) 764 return false; 765 auto I = Line.Tokens.begin(); 766 if (I->Tok->isNot(tok::l_paren)) 767 return false; 768 ++I; 769 if (I->Tok->isNot(Keywords.kw_function)) 770 return false; 771 ++I; 772 return I->Tok->is(tok::l_paren); 773 } 774 775 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 776 const FormatToken &InitialToken) { 777 if (InitialToken.isOneOf(tok::kw_namespace, TT_NamespaceMacro)) 778 return Style.BraceWrapping.AfterNamespace; 779 if (InitialToken.is(tok::kw_class)) 780 return Style.BraceWrapping.AfterClass; 781 if (InitialToken.is(tok::kw_union)) 782 return Style.BraceWrapping.AfterUnion; 783 if (InitialToken.is(tok::kw_struct)) 784 return Style.BraceWrapping.AfterStruct; 785 if (InitialToken.is(tok::kw_enum)) 786 return Style.BraceWrapping.AfterEnum; 787 return false; 788 } 789 790 void UnwrappedLineParser::parseChildBlock() { 791 FormatTok->setBlockKind(BK_Block); 792 nextToken(); 793 { 794 bool SkipIndent = (Style.isJavaScript() && 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 } 1144 1145 // readTokenWithJavaScriptASI reads the next token and terminates the current 1146 // line if JavaScript Automatic Semicolon Insertion must 1147 // happen between the current token and the next token. 1148 // 1149 // This method is conservative - it cannot cover all edge cases of JavaScript, 1150 // but only aims to correctly handle certain well known cases. It *must not* 1151 // return true in speculative cases. 1152 void UnwrappedLineParser::readTokenWithJavaScriptASI() { 1153 FormatToken *Previous = FormatTok; 1154 readToken(); 1155 FormatToken *Next = FormatTok; 1156 1157 bool IsOnSameLine = 1158 CommentsBeforeNextToken.empty() 1159 ? Next->NewlinesBefore == 0 1160 : CommentsBeforeNextToken.front()->NewlinesBefore == 0; 1161 if (IsOnSameLine) 1162 return; 1163 1164 bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous); 1165 bool PreviousStartsTemplateExpr = 1166 Previous->is(TT_TemplateString) && Previous->TokenText.endswith("${"); 1167 if (PreviousMustBeValue || Previous->is(tok::r_paren)) { 1168 // If the line contains an '@' sign, the previous token might be an 1169 // annotation, which can precede another identifier/value. 1170 bool HasAt = llvm::any_of(Line->Tokens, [](UnwrappedLineNode &LineNode) { 1171 return LineNode.Tok->is(tok::at); 1172 }); 1173 if (HasAt) 1174 return; 1175 } 1176 if (Next->is(tok::exclaim) && PreviousMustBeValue) 1177 return addUnwrappedLine(); 1178 bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next); 1179 bool NextEndsTemplateExpr = 1180 Next->is(TT_TemplateString) && Next->TokenText.startswith("}"); 1181 if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr && 1182 (PreviousMustBeValue || 1183 Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus, 1184 tok::minusminus))) 1185 return addUnwrappedLine(); 1186 if ((PreviousMustBeValue || Previous->is(tok::r_paren)) && 1187 isJSDeclOrStmt(Keywords, Next)) 1188 return addUnwrappedLine(); 1189 } 1190 1191 void UnwrappedLineParser::parseStructuralElement(bool IsTopLevel) { 1192 if (Style.Language == FormatStyle::LK_TableGen && 1193 FormatTok->is(tok::pp_include)) { 1194 nextToken(); 1195 if (FormatTok->is(tok::string_literal)) 1196 nextToken(); 1197 addUnwrappedLine(); 1198 return; 1199 } 1200 switch (FormatTok->Tok.getKind()) { 1201 case tok::kw_asm: 1202 nextToken(); 1203 if (FormatTok->is(tok::l_brace)) { 1204 FormatTok->setType(TT_InlineASMBrace); 1205 nextToken(); 1206 while (FormatTok && FormatTok->isNot(tok::eof)) { 1207 if (FormatTok->is(tok::r_brace)) { 1208 FormatTok->setType(TT_InlineASMBrace); 1209 nextToken(); 1210 addUnwrappedLine(); 1211 break; 1212 } 1213 FormatTok->Finalized = true; 1214 nextToken(); 1215 } 1216 } 1217 break; 1218 case tok::kw_namespace: 1219 parseNamespace(); 1220 return; 1221 case tok::kw_public: 1222 case tok::kw_protected: 1223 case tok::kw_private: 1224 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 1225 Style.isCSharp()) 1226 nextToken(); 1227 else 1228 parseAccessSpecifier(); 1229 return; 1230 case tok::kw_if: 1231 if (Style.isJavaScript() && Line->MustBeDeclaration) 1232 // field/method declaration. 1233 break; 1234 parseIfThenElse(); 1235 return; 1236 case tok::kw_for: 1237 case tok::kw_while: 1238 if (Style.isJavaScript() && Line->MustBeDeclaration) 1239 // field/method declaration. 1240 break; 1241 parseForOrWhileLoop(); 1242 return; 1243 case tok::kw_do: 1244 if (Style.isJavaScript() && Line->MustBeDeclaration) 1245 // field/method declaration. 1246 break; 1247 parseDoWhile(); 1248 return; 1249 case tok::kw_switch: 1250 if (Style.isJavaScript() && Line->MustBeDeclaration) 1251 // 'switch: string' field declaration. 1252 break; 1253 parseSwitch(); 1254 return; 1255 case tok::kw_default: 1256 if (Style.isJavaScript() && Line->MustBeDeclaration) 1257 // 'default: string' field declaration. 1258 break; 1259 nextToken(); 1260 if (FormatTok->is(tok::colon)) { 1261 parseLabel(); 1262 return; 1263 } 1264 // e.g. "default void f() {}" in a Java interface. 1265 break; 1266 case tok::kw_case: 1267 if (Style.isJavaScript() && Line->MustBeDeclaration) 1268 // 'case: string' field declaration. 1269 break; 1270 parseCaseLabel(); 1271 return; 1272 case tok::kw_try: 1273 case tok::kw___try: 1274 if (Style.isJavaScript() && Line->MustBeDeclaration) 1275 // field/method declaration. 1276 break; 1277 parseTryCatch(); 1278 return; 1279 case tok::kw_extern: 1280 nextToken(); 1281 if (FormatTok->Tok.is(tok::string_literal)) { 1282 nextToken(); 1283 if (FormatTok->Tok.is(tok::l_brace)) { 1284 if (Style.BraceWrapping.AfterExternBlock) 1285 addUnwrappedLine(); 1286 // Either we indent or for backwards compatibility we follow the 1287 // AfterExternBlock style. 1288 unsigned AddLevels = 1289 (Style.IndentExternBlock == FormatStyle::IEBS_Indent) || 1290 (Style.BraceWrapping.AfterExternBlock && 1291 Style.IndentExternBlock == 1292 FormatStyle::IEBS_AfterExternBlock) 1293 ? 1u 1294 : 0u; 1295 parseBlock(/*MustBeDeclaration=*/true, AddLevels); 1296 addUnwrappedLine(); 1297 return; 1298 } 1299 } 1300 break; 1301 case tok::kw_export: 1302 if (Style.isJavaScript()) { 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.isJavaScript()) { 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 return; 1441 case tok::kw_requires: 1442 parseRequires(); 1443 return; 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.isJavaScript() && 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.isJavaScript() && Line->MustBeDeclaration) { 1540 // field/method declaration. 1541 nextToken(); 1542 break; 1543 } 1544 // We arrive here when parsing function-try blocks. 1545 if (Style.BraceWrapping.AfterFunction) 1546 addUnwrappedLine(); 1547 parseTryCatch(); 1548 return; 1549 case tok::identifier: { 1550 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where) && 1551 Line->MustBeDeclaration) { 1552 addUnwrappedLine(); 1553 parseCSharpGenericTypeConstraint(); 1554 break; 1555 } 1556 if (FormatTok->is(TT_MacroBlockEnd)) { 1557 addUnwrappedLine(); 1558 return; 1559 } 1560 1561 // Function declarations (as opposed to function expressions) are parsed 1562 // on their own unwrapped line by continuing this loop. Function 1563 // expressions (functions that are not on their own line) must not create 1564 // a new unwrapped line, so they are special cased below. 1565 size_t TokenCount = Line->Tokens.size(); 1566 if (Style.isJavaScript() && FormatTok->is(Keywords.kw_function) && 1567 (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is( 1568 Keywords.kw_async)))) { 1569 tryToParseJSFunction(); 1570 break; 1571 } 1572 if ((Style.isJavaScript() || Style.Language == FormatStyle::LK_Java) && 1573 FormatTok->is(Keywords.kw_interface)) { 1574 if (Style.isJavaScript()) { 1575 // In JavaScript/TypeScript, "interface" can be used as a standalone 1576 // identifier, e.g. in `var interface = 1;`. If "interface" is 1577 // followed by another identifier, it is very like to be an actual 1578 // interface declaration. 1579 unsigned StoredPosition = Tokens->getPosition(); 1580 FormatToken *Next = Tokens->getNextToken(); 1581 FormatTok = Tokens->setPosition(StoredPosition); 1582 if (!mustBeJSIdent(Keywords, Next)) { 1583 nextToken(); 1584 break; 1585 } 1586 } 1587 parseRecord(); 1588 addUnwrappedLine(); 1589 return; 1590 } 1591 1592 if (FormatTok->is(Keywords.kw_interface)) { 1593 if (parseStructLike()) { 1594 return; 1595 } 1596 break; 1597 } 1598 1599 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1600 parseStatementMacro(); 1601 return; 1602 } 1603 1604 // See if the following token should start a new unwrapped line. 1605 StringRef Text = FormatTok->TokenText; 1606 nextToken(); 1607 1608 // JS doesn't have macros, and within classes colons indicate fields, not 1609 // labels. 1610 if (Style.isJavaScript()) 1611 break; 1612 1613 TokenCount = Line->Tokens.size(); 1614 if (TokenCount == 1 || 1615 (TokenCount == 2 && Line->Tokens.front().Tok->is(tok::comment))) { 1616 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) { 1617 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1618 parseLabel(!Style.IndentGotoLabels); 1619 return; 1620 } 1621 // Recognize function-like macro usages without trailing semicolon as 1622 // well as free-standing macros like Q_OBJECT. 1623 bool FunctionLike = FormatTok->is(tok::l_paren); 1624 if (FunctionLike) 1625 parseParens(); 1626 1627 bool FollowedByNewline = 1628 CommentsBeforeNextToken.empty() 1629 ? FormatTok->NewlinesBefore > 0 1630 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1631 1632 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1633 tokenCanStartNewLine(*FormatTok) && Text == Text.upper()) { 1634 addUnwrappedLine(); 1635 return; 1636 } 1637 } 1638 break; 1639 } 1640 case tok::equal: 1641 if ((Style.isJavaScript() || Style.isCSharp()) && 1642 FormatTok->is(TT_FatArrow)) { 1643 tryToParseChildBlock(); 1644 break; 1645 } 1646 1647 nextToken(); 1648 if (FormatTok->Tok.is(tok::l_brace)) { 1649 // Block kind should probably be set to BK_BracedInit for any language. 1650 // C# needs this change to ensure that array initialisers and object 1651 // initialisers are indented the same way. 1652 if (Style.isCSharp()) 1653 FormatTok->setBlockKind(BK_BracedInit); 1654 nextToken(); 1655 parseBracedList(); 1656 } else if (Style.Language == FormatStyle::LK_Proto && 1657 FormatTok->Tok.is(tok::less)) { 1658 nextToken(); 1659 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 1660 /*ClosingBraceKind=*/tok::greater); 1661 } 1662 break; 1663 case tok::l_square: 1664 parseSquare(); 1665 break; 1666 case tok::kw_new: 1667 parseNew(); 1668 break; 1669 default: 1670 nextToken(); 1671 break; 1672 } 1673 } while (!eof()); 1674 } 1675 1676 bool UnwrappedLineParser::tryToParsePropertyAccessor() { 1677 assert(FormatTok->is(tok::l_brace)); 1678 if (!Style.isCSharp()) 1679 return false; 1680 // See if it's a property accessor. 1681 if (FormatTok->Previous->isNot(tok::identifier)) 1682 return false; 1683 1684 // See if we are inside a property accessor. 1685 // 1686 // Record the current tokenPosition so that we can advance and 1687 // reset the current token. `Next` is not set yet so we need 1688 // another way to advance along the token stream. 1689 unsigned int StoredPosition = Tokens->getPosition(); 1690 FormatToken *Tok = Tokens->getNextToken(); 1691 1692 // A trivial property accessor is of the form: 1693 // { [ACCESS_SPECIFIER] [get]; [ACCESS_SPECIFIER] [set] } 1694 // Track these as they do not require line breaks to be introduced. 1695 bool HasGetOrSet = false; 1696 bool IsTrivialPropertyAccessor = true; 1697 while (!eof()) { 1698 if (Tok->isOneOf(tok::semi, tok::kw_public, tok::kw_private, 1699 tok::kw_protected, Keywords.kw_internal, Keywords.kw_get, 1700 Keywords.kw_set)) { 1701 if (Tok->isOneOf(Keywords.kw_get, Keywords.kw_set)) 1702 HasGetOrSet = true; 1703 Tok = Tokens->getNextToken(); 1704 continue; 1705 } 1706 if (Tok->isNot(tok::r_brace)) 1707 IsTrivialPropertyAccessor = false; 1708 break; 1709 } 1710 1711 if (!HasGetOrSet) { 1712 Tokens->setPosition(StoredPosition); 1713 return false; 1714 } 1715 1716 // Try to parse the property accessor: 1717 // https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties 1718 Tokens->setPosition(StoredPosition); 1719 if (!IsTrivialPropertyAccessor && Style.BraceWrapping.AfterFunction) 1720 addUnwrappedLine(); 1721 nextToken(); 1722 do { 1723 switch (FormatTok->Tok.getKind()) { 1724 case tok::r_brace: 1725 nextToken(); 1726 if (FormatTok->is(tok::equal)) { 1727 while (!eof() && FormatTok->isNot(tok::semi)) 1728 nextToken(); 1729 nextToken(); 1730 } 1731 addUnwrappedLine(); 1732 return true; 1733 case tok::l_brace: 1734 ++Line->Level; 1735 parseBlock(/*MustBeDeclaration=*/true); 1736 addUnwrappedLine(); 1737 --Line->Level; 1738 break; 1739 case tok::equal: 1740 if (FormatTok->is(TT_FatArrow)) { 1741 ++Line->Level; 1742 do { 1743 nextToken(); 1744 } while (!eof() && FormatTok->isNot(tok::semi)); 1745 nextToken(); 1746 addUnwrappedLine(); 1747 --Line->Level; 1748 break; 1749 } 1750 nextToken(); 1751 break; 1752 default: 1753 if (FormatTok->isOneOf(Keywords.kw_get, Keywords.kw_set) && 1754 !IsTrivialPropertyAccessor) { 1755 // Non-trivial get/set needs to be on its own line. 1756 addUnwrappedLine(); 1757 } 1758 nextToken(); 1759 } 1760 } while (!eof()); 1761 1762 // Unreachable for well-formed code (paired '{' and '}'). 1763 return true; 1764 } 1765 1766 bool UnwrappedLineParser::tryToParseLambda() { 1767 if (!Style.isCpp()) { 1768 nextToken(); 1769 return false; 1770 } 1771 assert(FormatTok->is(tok::l_square)); 1772 FormatToken &LSquare = *FormatTok; 1773 if (!tryToParseLambdaIntroducer()) 1774 return false; 1775 1776 bool SeenArrow = false; 1777 1778 while (FormatTok->isNot(tok::l_brace)) { 1779 if (FormatTok->isSimpleTypeSpecifier()) { 1780 nextToken(); 1781 continue; 1782 } 1783 switch (FormatTok->Tok.getKind()) { 1784 case tok::l_brace: 1785 break; 1786 case tok::l_paren: 1787 parseParens(); 1788 break; 1789 case tok::l_square: 1790 parseSquare(); 1791 break; 1792 case tok::amp: 1793 case tok::star: 1794 case tok::kw_const: 1795 case tok::comma: 1796 case tok::less: 1797 case tok::greater: 1798 case tok::identifier: 1799 case tok::numeric_constant: 1800 case tok::coloncolon: 1801 case tok::kw_class: 1802 case tok::kw_mutable: 1803 case tok::kw_noexcept: 1804 case tok::kw_template: 1805 case tok::kw_typename: 1806 nextToken(); 1807 break; 1808 // Specialization of a template with an integer parameter can contain 1809 // arithmetic, logical, comparison and ternary operators. 1810 // 1811 // FIXME: This also accepts sequences of operators that are not in the scope 1812 // of a template argument list. 1813 // 1814 // In a C++ lambda a template type can only occur after an arrow. We use 1815 // this as an heuristic to distinguish between Objective-C expressions 1816 // followed by an `a->b` expression, such as: 1817 // ([obj func:arg] + a->b) 1818 // Otherwise the code below would parse as a lambda. 1819 // 1820 // FIXME: This heuristic is incorrect for C++20 generic lambdas with 1821 // explicit template lists: []<bool b = true && false>(U &&u){} 1822 case tok::plus: 1823 case tok::minus: 1824 case tok::exclaim: 1825 case tok::tilde: 1826 case tok::slash: 1827 case tok::percent: 1828 case tok::lessless: 1829 case tok::pipe: 1830 case tok::pipepipe: 1831 case tok::ampamp: 1832 case tok::caret: 1833 case tok::equalequal: 1834 case tok::exclaimequal: 1835 case tok::greaterequal: 1836 case tok::lessequal: 1837 case tok::question: 1838 case tok::colon: 1839 case tok::ellipsis: 1840 case tok::kw_true: 1841 case tok::kw_false: 1842 if (SeenArrow) { 1843 nextToken(); 1844 break; 1845 } 1846 return true; 1847 case tok::arrow: 1848 // This might or might not actually be a lambda arrow (this could be an 1849 // ObjC method invocation followed by a dereferencing arrow). We might 1850 // reset this back to TT_Unknown in TokenAnnotator. 1851 FormatTok->setType(TT_LambdaArrow); 1852 SeenArrow = true; 1853 nextToken(); 1854 break; 1855 default: 1856 return true; 1857 } 1858 } 1859 FormatTok->setType(TT_LambdaLBrace); 1860 LSquare.setType(TT_LambdaLSquare); 1861 parseChildBlock(); 1862 return true; 1863 } 1864 1865 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 1866 const FormatToken *Previous = FormatTok->Previous; 1867 if (Previous && 1868 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 1869 tok::kw_delete, tok::l_square) || 1870 FormatTok->isCppStructuredBinding(Style) || Previous->closesScope() || 1871 Previous->isSimpleTypeSpecifier())) { 1872 nextToken(); 1873 return false; 1874 } 1875 nextToken(); 1876 if (FormatTok->is(tok::l_square)) { 1877 return false; 1878 } 1879 parseSquare(/*LambdaIntroducer=*/true); 1880 return true; 1881 } 1882 1883 void UnwrappedLineParser::tryToParseJSFunction() { 1884 assert(FormatTok->is(Keywords.kw_function) || 1885 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)); 1886 if (FormatTok->is(Keywords.kw_async)) 1887 nextToken(); 1888 // Consume "function". 1889 nextToken(); 1890 1891 // Consume * (generator function). Treat it like C++'s overloaded operators. 1892 if (FormatTok->is(tok::star)) { 1893 FormatTok->setType(TT_OverloadedOperator); 1894 nextToken(); 1895 } 1896 1897 // Consume function name. 1898 if (FormatTok->is(tok::identifier)) 1899 nextToken(); 1900 1901 if (FormatTok->isNot(tok::l_paren)) 1902 return; 1903 1904 // Parse formal parameter list. 1905 parseParens(); 1906 1907 if (FormatTok->is(tok::colon)) { 1908 // Parse a type definition. 1909 nextToken(); 1910 1911 // Eat the type declaration. For braced inline object types, balance braces, 1912 // otherwise just parse until finding an l_brace for the function body. 1913 if (FormatTok->is(tok::l_brace)) 1914 tryToParseBracedList(); 1915 else 1916 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 1917 nextToken(); 1918 } 1919 1920 if (FormatTok->is(tok::semi)) 1921 return; 1922 1923 parseChildBlock(); 1924 } 1925 1926 bool UnwrappedLineParser::tryToParseBracedList() { 1927 if (FormatTok->is(BK_Unknown)) 1928 calculateBraceTypes(); 1929 assert(FormatTok->isNot(BK_Unknown)); 1930 if (FormatTok->is(BK_Block)) 1931 return false; 1932 nextToken(); 1933 parseBracedList(); 1934 return true; 1935 } 1936 1937 bool UnwrappedLineParser::tryToParseChildBlock() { 1938 assert(Style.isJavaScript() || Style.isCSharp()); 1939 assert(FormatTok->is(TT_FatArrow)); 1940 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType TT_FatArrow. 1941 // They always start an expression or a child block if followed by a curly 1942 // brace. 1943 nextToken(); 1944 if (FormatTok->isNot(tok::l_brace)) 1945 return false; 1946 parseChildBlock(); 1947 return true; 1948 } 1949 1950 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons, 1951 bool IsEnum, 1952 tok::TokenKind ClosingBraceKind) { 1953 bool HasError = false; 1954 1955 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 1956 // replace this by using parseAssignmentExpression() inside. 1957 do { 1958 if (Style.isCSharp() && FormatTok->is(TT_FatArrow) && 1959 tryToParseChildBlock()) 1960 continue; 1961 if (Style.isJavaScript()) { 1962 if (FormatTok->is(Keywords.kw_function) || 1963 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) { 1964 tryToParseJSFunction(); 1965 continue; 1966 } 1967 if (FormatTok->is(tok::l_brace)) { 1968 // Could be a method inside of a braced list `{a() { return 1; }}`. 1969 if (tryToParseBracedList()) 1970 continue; 1971 parseChildBlock(); 1972 } 1973 } 1974 if (FormatTok->Tok.getKind() == ClosingBraceKind) { 1975 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 1976 addUnwrappedLine(); 1977 nextToken(); 1978 return !HasError; 1979 } 1980 switch (FormatTok->Tok.getKind()) { 1981 case tok::l_square: 1982 if (Style.isCSharp()) 1983 parseSquare(); 1984 else 1985 tryToParseLambda(); 1986 break; 1987 case tok::l_paren: 1988 parseParens(); 1989 // JavaScript can just have free standing methods and getters/setters in 1990 // object literals. Detect them by a "{" following ")". 1991 if (Style.isJavaScript()) { 1992 if (FormatTok->is(tok::l_brace)) 1993 parseChildBlock(); 1994 break; 1995 } 1996 break; 1997 case tok::l_brace: 1998 // Assume there are no blocks inside a braced init list apart 1999 // from the ones we explicitly parse out (like lambdas). 2000 FormatTok->setBlockKind(BK_BracedInit); 2001 nextToken(); 2002 parseBracedList(); 2003 break; 2004 case tok::less: 2005 if (Style.Language == FormatStyle::LK_Proto) { 2006 nextToken(); 2007 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 2008 /*ClosingBraceKind=*/tok::greater); 2009 } else { 2010 nextToken(); 2011 } 2012 break; 2013 case tok::semi: 2014 // JavaScript (or more precisely TypeScript) can have semicolons in braced 2015 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 2016 // used for error recovery if we have otherwise determined that this is 2017 // a braced list. 2018 if (Style.isJavaScript()) { 2019 nextToken(); 2020 break; 2021 } 2022 HasError = true; 2023 if (!ContinueOnSemicolons) 2024 return !HasError; 2025 nextToken(); 2026 break; 2027 case tok::comma: 2028 nextToken(); 2029 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 2030 addUnwrappedLine(); 2031 break; 2032 default: 2033 nextToken(); 2034 break; 2035 } 2036 } while (!eof()); 2037 return false; 2038 } 2039 2040 void UnwrappedLineParser::parseParens() { 2041 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 2042 nextToken(); 2043 do { 2044 switch (FormatTok->Tok.getKind()) { 2045 case tok::l_paren: 2046 parseParens(); 2047 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 2048 parseChildBlock(); 2049 break; 2050 case tok::r_paren: 2051 nextToken(); 2052 return; 2053 case tok::r_brace: 2054 // A "}" inside parenthesis is an error if there wasn't a matching "{". 2055 return; 2056 case tok::l_square: 2057 tryToParseLambda(); 2058 break; 2059 case tok::l_brace: 2060 if (!tryToParseBracedList()) 2061 parseChildBlock(); 2062 break; 2063 case tok::at: 2064 nextToken(); 2065 if (FormatTok->Tok.is(tok::l_brace)) { 2066 nextToken(); 2067 parseBracedList(); 2068 } 2069 break; 2070 case tok::equal: 2071 if (Style.isCSharp() && FormatTok->is(TT_FatArrow)) 2072 tryToParseChildBlock(); 2073 else 2074 nextToken(); 2075 break; 2076 case tok::kw_class: 2077 if (Style.isJavaScript()) 2078 parseRecord(/*ParseAsExpr=*/true); 2079 else 2080 nextToken(); 2081 break; 2082 case tok::identifier: 2083 if (Style.isJavaScript() && 2084 (FormatTok->is(Keywords.kw_function) || 2085 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function))) 2086 tryToParseJSFunction(); 2087 else 2088 nextToken(); 2089 break; 2090 default: 2091 nextToken(); 2092 break; 2093 } 2094 } while (!eof()); 2095 } 2096 2097 void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) { 2098 if (!LambdaIntroducer) { 2099 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected."); 2100 if (tryToParseLambda()) 2101 return; 2102 } 2103 do { 2104 switch (FormatTok->Tok.getKind()) { 2105 case tok::l_paren: 2106 parseParens(); 2107 break; 2108 case tok::r_square: 2109 nextToken(); 2110 return; 2111 case tok::r_brace: 2112 // A "}" inside parenthesis is an error if there wasn't a matching "{". 2113 return; 2114 case tok::l_square: 2115 parseSquare(); 2116 break; 2117 case tok::l_brace: { 2118 if (!tryToParseBracedList()) 2119 parseChildBlock(); 2120 break; 2121 } 2122 case tok::at: 2123 nextToken(); 2124 if (FormatTok->Tok.is(tok::l_brace)) { 2125 nextToken(); 2126 parseBracedList(); 2127 } 2128 break; 2129 default: 2130 nextToken(); 2131 break; 2132 } 2133 } while (!eof()); 2134 } 2135 2136 void UnwrappedLineParser::parseIfThenElse() { 2137 auto HandleAttributes = [this]() { 2138 // Handle AttributeMacro, e.g. `if (x) UNLIKELY`. 2139 if (FormatTok->is(TT_AttributeMacro)) 2140 nextToken(); 2141 // Handle [[likely]] / [[unlikely]] attributes. 2142 if (FormatTok->is(tok::l_square) && tryToParseSimpleAttribute()) 2143 parseSquare(); 2144 }; 2145 2146 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 2147 nextToken(); 2148 if (FormatTok->Tok.isOneOf(tok::kw_constexpr, tok::identifier)) 2149 nextToken(); 2150 if (FormatTok->Tok.is(tok::l_paren)) 2151 parseParens(); 2152 HandleAttributes(); 2153 bool NeedsUnwrappedLine = false; 2154 if (FormatTok->Tok.is(tok::l_brace)) { 2155 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2156 parseBlock(); 2157 if (Style.BraceWrapping.BeforeElse) 2158 addUnwrappedLine(); 2159 else 2160 NeedsUnwrappedLine = true; 2161 } else { 2162 addUnwrappedLine(); 2163 ++Line->Level; 2164 parseStructuralElement(); 2165 --Line->Level; 2166 } 2167 if (FormatTok->Tok.is(tok::kw_else)) { 2168 nextToken(); 2169 HandleAttributes(); 2170 if (FormatTok->Tok.is(tok::l_brace)) { 2171 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2172 parseBlock(); 2173 addUnwrappedLine(); 2174 } else if (FormatTok->Tok.is(tok::kw_if)) { 2175 FormatToken *Previous = Tokens->getPreviousToken(); 2176 bool PrecededByComment = Previous && Previous->is(tok::comment); 2177 if (PrecededByComment) { 2178 addUnwrappedLine(); 2179 ++Line->Level; 2180 } 2181 parseIfThenElse(); 2182 if (PrecededByComment) 2183 --Line->Level; 2184 } else { 2185 addUnwrappedLine(); 2186 ++Line->Level; 2187 parseStructuralElement(); 2188 if (FormatTok->is(tok::eof)) 2189 addUnwrappedLine(); 2190 --Line->Level; 2191 } 2192 } else if (NeedsUnwrappedLine) { 2193 addUnwrappedLine(); 2194 } 2195 } 2196 2197 void UnwrappedLineParser::parseTryCatch() { 2198 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 2199 nextToken(); 2200 bool NeedsUnwrappedLine = false; 2201 if (FormatTok->is(tok::colon)) { 2202 // We are in a function try block, what comes is an initializer list. 2203 nextToken(); 2204 2205 // In case identifiers were removed by clang-tidy, what might follow is 2206 // multiple commas in sequence - before the first identifier. 2207 while (FormatTok->is(tok::comma)) 2208 nextToken(); 2209 2210 while (FormatTok->is(tok::identifier)) { 2211 nextToken(); 2212 if (FormatTok->is(tok::l_paren)) 2213 parseParens(); 2214 if (FormatTok->Previous && FormatTok->Previous->is(tok::identifier) && 2215 FormatTok->is(tok::l_brace)) { 2216 do { 2217 nextToken(); 2218 } while (!FormatTok->is(tok::r_brace)); 2219 nextToken(); 2220 } 2221 2222 // In case identifiers were removed by clang-tidy, what might follow is 2223 // multiple commas in sequence - after the first identifier. 2224 while (FormatTok->is(tok::comma)) 2225 nextToken(); 2226 } 2227 } 2228 // Parse try with resource. 2229 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) { 2230 parseParens(); 2231 } 2232 if (FormatTok->is(tok::l_brace)) { 2233 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2234 parseBlock(); 2235 if (Style.BraceWrapping.BeforeCatch) { 2236 addUnwrappedLine(); 2237 } else { 2238 NeedsUnwrappedLine = true; 2239 } 2240 } else if (!FormatTok->is(tok::kw_catch)) { 2241 // The C++ standard requires a compound-statement after a try. 2242 // If there's none, we try to assume there's a structuralElement 2243 // and try to continue. 2244 addUnwrappedLine(); 2245 ++Line->Level; 2246 parseStructuralElement(); 2247 --Line->Level; 2248 } 2249 while (1) { 2250 if (FormatTok->is(tok::at)) 2251 nextToken(); 2252 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 2253 tok::kw___finally) || 2254 ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2255 FormatTok->is(Keywords.kw_finally)) || 2256 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) || 2257 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally)))) 2258 break; 2259 nextToken(); 2260 while (FormatTok->isNot(tok::l_brace)) { 2261 if (FormatTok->is(tok::l_paren)) { 2262 parseParens(); 2263 continue; 2264 } 2265 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) 2266 return; 2267 nextToken(); 2268 } 2269 NeedsUnwrappedLine = false; 2270 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2271 parseBlock(); 2272 if (Style.BraceWrapping.BeforeCatch) 2273 addUnwrappedLine(); 2274 else 2275 NeedsUnwrappedLine = true; 2276 } 2277 if (NeedsUnwrappedLine) 2278 addUnwrappedLine(); 2279 } 2280 2281 void UnwrappedLineParser::parseNamespace() { 2282 assert(FormatTok->isOneOf(tok::kw_namespace, TT_NamespaceMacro) && 2283 "'namespace' expected"); 2284 2285 const FormatToken &InitialToken = *FormatTok; 2286 nextToken(); 2287 if (InitialToken.is(TT_NamespaceMacro)) { 2288 parseParens(); 2289 } else { 2290 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::kw_inline, 2291 tok::l_square, tok::period)) { 2292 if (FormatTok->is(tok::l_square)) 2293 parseSquare(); 2294 else 2295 nextToken(); 2296 } 2297 } 2298 if (FormatTok->Tok.is(tok::l_brace)) { 2299 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2300 addUnwrappedLine(); 2301 2302 unsigned AddLevels = 2303 Style.NamespaceIndentation == FormatStyle::NI_All || 2304 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 2305 DeclarationScopeStack.size() > 1) 2306 ? 1u 2307 : 0u; 2308 bool ManageWhitesmithsBraces = 2309 AddLevels == 0u && 2310 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths; 2311 2312 // If we're in Whitesmiths mode, indent the brace if we're not indenting 2313 // the whole block. 2314 if (ManageWhitesmithsBraces) 2315 ++Line->Level; 2316 2317 parseBlock(/*MustBeDeclaration=*/true, AddLevels, 2318 /*MunchSemi=*/true, 2319 /*UnindentWhitesmithsBraces=*/ManageWhitesmithsBraces); 2320 2321 // Munch the semicolon after a namespace. This is more common than one would 2322 // think. Putting the semicolon into its own line is very ugly. 2323 if (FormatTok->Tok.is(tok::semi)) 2324 nextToken(); 2325 2326 addUnwrappedLine(AddLevels > 0 ? LineLevel::Remove : LineLevel::Keep); 2327 2328 if (ManageWhitesmithsBraces) 2329 --Line->Level; 2330 } 2331 // FIXME: Add error handling. 2332 } 2333 2334 void UnwrappedLineParser::parseNew() { 2335 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 2336 nextToken(); 2337 2338 if (Style.isCSharp()) { 2339 do { 2340 if (FormatTok->is(tok::l_brace)) 2341 parseBracedList(); 2342 2343 if (FormatTok->isOneOf(tok::semi, tok::comma)) 2344 return; 2345 2346 nextToken(); 2347 } while (!eof()); 2348 } 2349 2350 if (Style.Language != FormatStyle::LK_Java) 2351 return; 2352 2353 // In Java, we can parse everything up to the parens, which aren't optional. 2354 do { 2355 // There should not be a ;, { or } before the new's open paren. 2356 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 2357 return; 2358 2359 // Consume the parens. 2360 if (FormatTok->is(tok::l_paren)) { 2361 parseParens(); 2362 2363 // If there is a class body of an anonymous class, consume that as child. 2364 if (FormatTok->is(tok::l_brace)) 2365 parseChildBlock(); 2366 return; 2367 } 2368 nextToken(); 2369 } while (!eof()); 2370 } 2371 2372 void UnwrappedLineParser::parseForOrWhileLoop() { 2373 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 2374 "'for', 'while' or foreach macro expected"); 2375 nextToken(); 2376 // JS' for await ( ... 2377 if (Style.isJavaScript() && FormatTok->is(Keywords.kw_await)) 2378 nextToken(); 2379 if (Style.isCpp() && FormatTok->is(tok::kw_co_await)) 2380 nextToken(); 2381 if (FormatTok->Tok.is(tok::l_paren)) 2382 parseParens(); 2383 if (FormatTok->Tok.is(tok::l_brace)) { 2384 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2385 parseBlock(); 2386 addUnwrappedLine(); 2387 } else { 2388 addUnwrappedLine(); 2389 ++Line->Level; 2390 parseStructuralElement(); 2391 --Line->Level; 2392 } 2393 } 2394 2395 void UnwrappedLineParser::parseDoWhile() { 2396 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 2397 nextToken(); 2398 if (FormatTok->Tok.is(tok::l_brace)) { 2399 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2400 parseBlock(); 2401 if (Style.BraceWrapping.BeforeWhile) 2402 addUnwrappedLine(); 2403 } else { 2404 addUnwrappedLine(); 2405 ++Line->Level; 2406 parseStructuralElement(); 2407 --Line->Level; 2408 } 2409 2410 // FIXME: Add error handling. 2411 if (!FormatTok->Tok.is(tok::kw_while)) { 2412 addUnwrappedLine(); 2413 return; 2414 } 2415 2416 // If in Whitesmiths mode, the line with the while() needs to be indented 2417 // to the same level as the block. 2418 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) 2419 ++Line->Level; 2420 2421 nextToken(); 2422 parseStructuralElement(); 2423 } 2424 2425 void UnwrappedLineParser::parseLabel(bool LeftAlignLabel) { 2426 nextToken(); 2427 unsigned OldLineLevel = Line->Level; 2428 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 2429 --Line->Level; 2430 if (LeftAlignLabel) 2431 Line->Level = 0; 2432 2433 if (!Style.IndentCaseBlocks && CommentsBeforeNextToken.empty() && 2434 FormatTok->Tok.is(tok::l_brace)) { 2435 2436 CompoundStatementIndenter Indenter(this, Line->Level, 2437 Style.BraceWrapping.AfterCaseLabel, 2438 Style.BraceWrapping.IndentBraces); 2439 parseBlock(); 2440 if (FormatTok->Tok.is(tok::kw_break)) { 2441 if (Style.BraceWrapping.AfterControlStatement == 2442 FormatStyle::BWACS_Always) { 2443 addUnwrappedLine(); 2444 if (!Style.IndentCaseBlocks && 2445 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) { 2446 Line->Level++; 2447 } 2448 } 2449 parseStructuralElement(); 2450 } 2451 addUnwrappedLine(); 2452 } else { 2453 if (FormatTok->is(tok::semi)) 2454 nextToken(); 2455 addUnwrappedLine(); 2456 } 2457 Line->Level = OldLineLevel; 2458 if (FormatTok->isNot(tok::l_brace)) { 2459 parseStructuralElement(); 2460 addUnwrappedLine(); 2461 } 2462 } 2463 2464 void UnwrappedLineParser::parseCaseLabel() { 2465 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 2466 2467 // FIXME: fix handling of complex expressions here. 2468 do { 2469 nextToken(); 2470 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 2471 parseLabel(); 2472 } 2473 2474 void UnwrappedLineParser::parseSwitch() { 2475 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 2476 nextToken(); 2477 if (FormatTok->Tok.is(tok::l_paren)) 2478 parseParens(); 2479 if (FormatTok->Tok.is(tok::l_brace)) { 2480 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2481 parseBlock(); 2482 addUnwrappedLine(); 2483 } else { 2484 addUnwrappedLine(); 2485 ++Line->Level; 2486 parseStructuralElement(); 2487 --Line->Level; 2488 } 2489 } 2490 2491 void UnwrappedLineParser::parseAccessSpecifier() { 2492 nextToken(); 2493 // Understand Qt's slots. 2494 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 2495 nextToken(); 2496 // Otherwise, we don't know what it is, and we'd better keep the next token. 2497 if (FormatTok->Tok.is(tok::colon)) 2498 nextToken(); 2499 addUnwrappedLine(); 2500 } 2501 2502 void UnwrappedLineParser::parseConcept() { 2503 assert(FormatTok->Tok.is(tok::kw_concept) && "'concept' expected"); 2504 nextToken(); 2505 if (!FormatTok->Tok.is(tok::identifier)) 2506 return; 2507 nextToken(); 2508 if (!FormatTok->Tok.is(tok::equal)) 2509 return; 2510 nextToken(); 2511 if (FormatTok->Tok.is(tok::kw_requires)) { 2512 nextToken(); 2513 parseRequiresExpression(Line->Level); 2514 } else { 2515 parseConstraintExpression(Line->Level); 2516 } 2517 } 2518 2519 void UnwrappedLineParser::parseRequiresExpression(unsigned int OriginalLevel) { 2520 // requires (R range) 2521 if (FormatTok->Tok.is(tok::l_paren)) { 2522 parseParens(); 2523 if (Style.IndentRequires && OriginalLevel != Line->Level) { 2524 addUnwrappedLine(); 2525 --Line->Level; 2526 } 2527 } 2528 2529 if (FormatTok->Tok.is(tok::l_brace)) { 2530 if (Style.BraceWrapping.AfterFunction) 2531 addUnwrappedLine(); 2532 FormatTok->setType(TT_FunctionLBrace); 2533 parseBlock(); 2534 addUnwrappedLine(); 2535 } else { 2536 parseConstraintExpression(OriginalLevel); 2537 } 2538 } 2539 2540 void UnwrappedLineParser::parseConstraintExpression( 2541 unsigned int OriginalLevel) { 2542 // requires Id<T> && Id<T> || Id<T> 2543 while ( 2544 FormatTok->isOneOf(tok::identifier, tok::kw_requires, tok::coloncolon)) { 2545 nextToken(); 2546 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::less, 2547 tok::greater, tok::comma, tok::ellipsis)) { 2548 if (FormatTok->Tok.is(tok::less)) { 2549 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 2550 /*ClosingBraceKind=*/tok::greater); 2551 continue; 2552 } 2553 nextToken(); 2554 } 2555 if (FormatTok->Tok.is(tok::kw_requires)) { 2556 parseRequiresExpression(OriginalLevel); 2557 } 2558 if (FormatTok->Tok.is(tok::less)) { 2559 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 2560 /*ClosingBraceKind=*/tok::greater); 2561 } 2562 2563 if (FormatTok->Tok.is(tok::l_paren)) { 2564 parseParens(); 2565 } 2566 if (FormatTok->Tok.is(tok::l_brace)) { 2567 if (Style.BraceWrapping.AfterFunction) 2568 addUnwrappedLine(); 2569 FormatTok->setType(TT_FunctionLBrace); 2570 parseBlock(); 2571 } 2572 if (FormatTok->Tok.is(tok::semi)) { 2573 // Eat any trailing semi. 2574 nextToken(); 2575 addUnwrappedLine(); 2576 } 2577 if (FormatTok->Tok.is(tok::colon)) { 2578 return; 2579 } 2580 if (!FormatTok->Tok.isOneOf(tok::ampamp, tok::pipepipe)) { 2581 if (FormatTok->Previous && 2582 !FormatTok->Previous->isOneOf(tok::identifier, tok::kw_requires, 2583 tok::coloncolon)) { 2584 addUnwrappedLine(); 2585 } 2586 if (Style.IndentRequires && OriginalLevel != Line->Level) { 2587 --Line->Level; 2588 } 2589 break; 2590 } else { 2591 FormatTok->setType(TT_ConstraintJunctions); 2592 } 2593 2594 nextToken(); 2595 } 2596 } 2597 2598 void UnwrappedLineParser::parseRequires() { 2599 assert(FormatTok->Tok.is(tok::kw_requires) && "'requires' expected"); 2600 2601 unsigned OriginalLevel = Line->Level; 2602 if (FormatTok->Previous && FormatTok->Previous->is(tok::greater)) { 2603 addUnwrappedLine(); 2604 if (Style.IndentRequires) { 2605 Line->Level++; 2606 } 2607 } 2608 nextToken(); 2609 2610 parseRequiresExpression(OriginalLevel); 2611 } 2612 2613 bool UnwrappedLineParser::parseEnum() { 2614 const FormatToken &InitialToken = *FormatTok; 2615 2616 // Won't be 'enum' for NS_ENUMs. 2617 if (FormatTok->Tok.is(tok::kw_enum)) 2618 nextToken(); 2619 2620 // In TypeScript, "enum" can also be used as property name, e.g. in interface 2621 // declarations. An "enum" keyword followed by a colon would be a syntax 2622 // error and thus assume it is just an identifier. 2623 if (Style.isJavaScript() && FormatTok->isOneOf(tok::colon, tok::question)) 2624 return false; 2625 2626 // In protobuf, "enum" can be used as a field name. 2627 if (Style.Language == FormatStyle::LK_Proto && FormatTok->is(tok::equal)) 2628 return false; 2629 2630 // Eat up enum class ... 2631 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct)) 2632 nextToken(); 2633 2634 while (FormatTok->Tok.getIdentifierInfo() || 2635 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 2636 tok::greater, tok::comma, tok::question)) { 2637 nextToken(); 2638 // We can have macros or attributes in between 'enum' and the enum name. 2639 if (FormatTok->is(tok::l_paren)) 2640 parseParens(); 2641 if (FormatTok->is(tok::identifier)) { 2642 nextToken(); 2643 // If there are two identifiers in a row, this is likely an elaborate 2644 // return type. In Java, this can be "implements", etc. 2645 if (Style.isCpp() && FormatTok->is(tok::identifier)) 2646 return false; 2647 } 2648 } 2649 2650 // Just a declaration or something is wrong. 2651 if (FormatTok->isNot(tok::l_brace)) 2652 return true; 2653 FormatTok->setBlockKind(BK_Block); 2654 2655 if (Style.Language == FormatStyle::LK_Java) { 2656 // Java enums are different. 2657 parseJavaEnumBody(); 2658 return true; 2659 } 2660 if (Style.Language == FormatStyle::LK_Proto) { 2661 parseBlock(/*MustBeDeclaration=*/true); 2662 return true; 2663 } 2664 2665 if (!Style.AllowShortEnumsOnASingleLine && 2666 ShouldBreakBeforeBrace(Style, InitialToken)) 2667 addUnwrappedLine(); 2668 // Parse enum body. 2669 nextToken(); 2670 if (!Style.AllowShortEnumsOnASingleLine) { 2671 addUnwrappedLine(); 2672 Line->Level += 1; 2673 } 2674 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true, 2675 /*IsEnum=*/true); 2676 if (!Style.AllowShortEnumsOnASingleLine) 2677 Line->Level -= 1; 2678 if (HasError) { 2679 if (FormatTok->is(tok::semi)) 2680 nextToken(); 2681 addUnwrappedLine(); 2682 } 2683 return true; 2684 2685 // There is no addUnwrappedLine() here so that we fall through to parsing a 2686 // structural element afterwards. Thus, in "enum A {} n, m;", 2687 // "} n, m;" will end up in one unwrapped line. 2688 } 2689 2690 bool UnwrappedLineParser::parseStructLike() { 2691 // parseRecord falls through and does not yet add an unwrapped line as a 2692 // record declaration or definition can start a structural element. 2693 parseRecord(); 2694 // This does not apply to Java, JavaScript and C#. 2695 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 2696 Style.isCSharp()) { 2697 if (FormatTok->is(tok::semi)) 2698 nextToken(); 2699 addUnwrappedLine(); 2700 return true; 2701 } 2702 return false; 2703 } 2704 2705 namespace { 2706 // A class used to set and restore the Token position when peeking 2707 // ahead in the token source. 2708 class ScopedTokenPosition { 2709 unsigned StoredPosition; 2710 FormatTokenSource *Tokens; 2711 2712 public: 2713 ScopedTokenPosition(FormatTokenSource *Tokens) : Tokens(Tokens) { 2714 assert(Tokens && "Tokens expected to not be null"); 2715 StoredPosition = Tokens->getPosition(); 2716 } 2717 2718 ~ScopedTokenPosition() { Tokens->setPosition(StoredPosition); } 2719 }; 2720 } // namespace 2721 2722 // Look to see if we have [[ by looking ahead, if 2723 // its not then rewind to the original position. 2724 bool UnwrappedLineParser::tryToParseSimpleAttribute() { 2725 ScopedTokenPosition AutoPosition(Tokens); 2726 FormatToken *Tok = Tokens->getNextToken(); 2727 // We already read the first [ check for the second. 2728 if (!Tok->is(tok::l_square)) { 2729 return false; 2730 } 2731 // Double check that the attribute is just something 2732 // fairly simple. 2733 while (Tok->isNot(tok::eof)) { 2734 if (Tok->is(tok::r_square)) { 2735 break; 2736 } 2737 Tok = Tokens->getNextToken(); 2738 } 2739 if (Tok->is(tok::eof)) 2740 return false; 2741 Tok = Tokens->getNextToken(); 2742 if (!Tok->is(tok::r_square)) { 2743 return false; 2744 } 2745 Tok = Tokens->getNextToken(); 2746 if (Tok->is(tok::semi)) { 2747 return false; 2748 } 2749 return true; 2750 } 2751 2752 void UnwrappedLineParser::parseJavaEnumBody() { 2753 // Determine whether the enum is simple, i.e. does not have a semicolon or 2754 // constants with class bodies. Simple enums can be formatted like braced 2755 // lists, contracted to a single line, etc. 2756 unsigned StoredPosition = Tokens->getPosition(); 2757 bool IsSimple = true; 2758 FormatToken *Tok = Tokens->getNextToken(); 2759 while (!Tok->is(tok::eof)) { 2760 if (Tok->is(tok::r_brace)) 2761 break; 2762 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 2763 IsSimple = false; 2764 break; 2765 } 2766 // FIXME: This will also mark enums with braces in the arguments to enum 2767 // constants as "not simple". This is probably fine in practice, though. 2768 Tok = Tokens->getNextToken(); 2769 } 2770 FormatTok = Tokens->setPosition(StoredPosition); 2771 2772 if (IsSimple) { 2773 nextToken(); 2774 parseBracedList(); 2775 addUnwrappedLine(); 2776 return; 2777 } 2778 2779 // Parse the body of a more complex enum. 2780 // First add a line for everything up to the "{". 2781 nextToken(); 2782 addUnwrappedLine(); 2783 ++Line->Level; 2784 2785 // Parse the enum constants. 2786 while (FormatTok) { 2787 if (FormatTok->is(tok::l_brace)) { 2788 // Parse the constant's class body. 2789 parseBlock(/*MustBeDeclaration=*/true, /*AddLevels=*/1u, 2790 /*MunchSemi=*/false); 2791 } else if (FormatTok->is(tok::l_paren)) { 2792 parseParens(); 2793 } else if (FormatTok->is(tok::comma)) { 2794 nextToken(); 2795 addUnwrappedLine(); 2796 } else if (FormatTok->is(tok::semi)) { 2797 nextToken(); 2798 addUnwrappedLine(); 2799 break; 2800 } else if (FormatTok->is(tok::r_brace)) { 2801 addUnwrappedLine(); 2802 break; 2803 } else { 2804 nextToken(); 2805 } 2806 } 2807 2808 // Parse the class body after the enum's ";" if any. 2809 parseLevel(/*HasOpeningBrace=*/true); 2810 nextToken(); 2811 --Line->Level; 2812 addUnwrappedLine(); 2813 } 2814 2815 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) { 2816 const FormatToken &InitialToken = *FormatTok; 2817 nextToken(); 2818 2819 // The actual identifier can be a nested name specifier, and in macros 2820 // it is often token-pasted. 2821 // An [[attribute]] can be before the identifier. 2822 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 2823 tok::kw___attribute, tok::kw___declspec, 2824 tok::kw_alignas, tok::l_square, tok::r_square) || 2825 ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2826 FormatTok->isOneOf(tok::period, tok::comma))) { 2827 if (Style.isJavaScript() && 2828 FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) { 2829 // JavaScript/TypeScript supports inline object types in 2830 // extends/implements positions: 2831 // class Foo implements {bar: number} { } 2832 nextToken(); 2833 if (FormatTok->is(tok::l_brace)) { 2834 tryToParseBracedList(); 2835 continue; 2836 } 2837 } 2838 bool IsNonMacroIdentifier = 2839 FormatTok->is(tok::identifier) && 2840 FormatTok->TokenText != FormatTok->TokenText.upper(); 2841 nextToken(); 2842 // We can have macros or attributes in between 'class' and the class name. 2843 if (!IsNonMacroIdentifier) { 2844 if (FormatTok->Tok.is(tok::l_paren)) { 2845 parseParens(); 2846 } else if (FormatTok->is(TT_AttributeSquare)) { 2847 parseSquare(); 2848 // Consume the closing TT_AttributeSquare. 2849 if (FormatTok->Next && FormatTok->is(TT_AttributeSquare)) 2850 nextToken(); 2851 } 2852 } 2853 } 2854 2855 // Note that parsing away template declarations here leads to incorrectly 2856 // accepting function declarations as record declarations. 2857 // In general, we cannot solve this problem. Consider: 2858 // class A<int> B() {} 2859 // which can be a function definition or a class definition when B() is a 2860 // macro. If we find enough real-world cases where this is a problem, we 2861 // can parse for the 'template' keyword in the beginning of the statement, 2862 // and thus rule out the record production in case there is no template 2863 // (this would still leave us with an ambiguity between template function 2864 // and class declarations). 2865 if (FormatTok->isOneOf(tok::colon, tok::less)) { 2866 while (!eof()) { 2867 if (FormatTok->is(tok::l_brace)) { 2868 calculateBraceTypes(/*ExpectClassBody=*/true); 2869 if (!tryToParseBracedList()) 2870 break; 2871 } 2872 if (FormatTok->Tok.is(tok::semi)) 2873 return; 2874 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where)) { 2875 addUnwrappedLine(); 2876 nextToken(); 2877 parseCSharpGenericTypeConstraint(); 2878 break; 2879 } 2880 nextToken(); 2881 } 2882 } 2883 if (FormatTok->Tok.is(tok::l_brace)) { 2884 if (ParseAsExpr) { 2885 parseChildBlock(); 2886 } else { 2887 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2888 addUnwrappedLine(); 2889 2890 unsigned AddLevels = Style.IndentAccessModifiers ? 2u : 1u; 2891 parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/false); 2892 } 2893 } 2894 // There is no addUnwrappedLine() here so that we fall through to parsing a 2895 // structural element afterwards. Thus, in "class A {} n, m;", 2896 // "} n, m;" will end up in one unwrapped line. 2897 } 2898 2899 void UnwrappedLineParser::parseObjCMethod() { 2900 assert(FormatTok->Tok.isOneOf(tok::l_paren, tok::identifier) && 2901 "'(' or identifier expected."); 2902 do { 2903 if (FormatTok->Tok.is(tok::semi)) { 2904 nextToken(); 2905 addUnwrappedLine(); 2906 return; 2907 } else if (FormatTok->Tok.is(tok::l_brace)) { 2908 if (Style.BraceWrapping.AfterFunction) 2909 addUnwrappedLine(); 2910 parseBlock(); 2911 addUnwrappedLine(); 2912 return; 2913 } else { 2914 nextToken(); 2915 } 2916 } while (!eof()); 2917 } 2918 2919 void UnwrappedLineParser::parseObjCProtocolList() { 2920 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 2921 do { 2922 nextToken(); 2923 // Early exit in case someone forgot a close angle. 2924 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 2925 FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) 2926 return; 2927 } while (!eof() && FormatTok->Tok.isNot(tok::greater)); 2928 nextToken(); // Skip '>'. 2929 } 2930 2931 void UnwrappedLineParser::parseObjCUntilAtEnd() { 2932 do { 2933 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 2934 nextToken(); 2935 addUnwrappedLine(); 2936 break; 2937 } 2938 if (FormatTok->is(tok::l_brace)) { 2939 parseBlock(); 2940 // In ObjC interfaces, nothing should be following the "}". 2941 addUnwrappedLine(); 2942 } else if (FormatTok->is(tok::r_brace)) { 2943 // Ignore stray "}". parseStructuralElement doesn't consume them. 2944 nextToken(); 2945 addUnwrappedLine(); 2946 } else if (FormatTok->isOneOf(tok::minus, tok::plus)) { 2947 nextToken(); 2948 parseObjCMethod(); 2949 } else { 2950 parseStructuralElement(); 2951 } 2952 } while (!eof()); 2953 } 2954 2955 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 2956 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_interface || 2957 FormatTok->Tok.getObjCKeywordID() == tok::objc_implementation); 2958 nextToken(); 2959 nextToken(); // interface name 2960 2961 // @interface can be followed by a lightweight generic 2962 // specialization list, then either a base class or a category. 2963 if (FormatTok->Tok.is(tok::less)) { 2964 parseObjCLightweightGenerics(); 2965 } 2966 if (FormatTok->Tok.is(tok::colon)) { 2967 nextToken(); 2968 nextToken(); // base class name 2969 // The base class can also have lightweight generics applied to it. 2970 if (FormatTok->Tok.is(tok::less)) { 2971 parseObjCLightweightGenerics(); 2972 } 2973 } else if (FormatTok->Tok.is(tok::l_paren)) 2974 // Skip category, if present. 2975 parseParens(); 2976 2977 if (FormatTok->Tok.is(tok::less)) 2978 parseObjCProtocolList(); 2979 2980 if (FormatTok->Tok.is(tok::l_brace)) { 2981 if (Style.BraceWrapping.AfterObjCDeclaration) 2982 addUnwrappedLine(); 2983 parseBlock(/*MustBeDeclaration=*/true); 2984 } 2985 2986 // With instance variables, this puts '}' on its own line. Without instance 2987 // variables, this ends the @interface line. 2988 addUnwrappedLine(); 2989 2990 parseObjCUntilAtEnd(); 2991 } 2992 2993 void UnwrappedLineParser::parseObjCLightweightGenerics() { 2994 assert(FormatTok->Tok.is(tok::less)); 2995 // Unlike protocol lists, generic parameterizations support 2996 // nested angles: 2997 // 2998 // @interface Foo<ValueType : id <NSCopying, NSSecureCoding>> : 2999 // NSObject <NSCopying, NSSecureCoding> 3000 // 3001 // so we need to count how many open angles we have left. 3002 unsigned NumOpenAngles = 1; 3003 do { 3004 nextToken(); 3005 // Early exit in case someone forgot a close angle. 3006 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 3007 FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) 3008 break; 3009 if (FormatTok->Tok.is(tok::less)) 3010 ++NumOpenAngles; 3011 else if (FormatTok->Tok.is(tok::greater)) { 3012 assert(NumOpenAngles > 0 && "'>' makes NumOpenAngles negative"); 3013 --NumOpenAngles; 3014 } 3015 } while (!eof() && NumOpenAngles != 0); 3016 nextToken(); // Skip '>'. 3017 } 3018 3019 // Returns true for the declaration/definition form of @protocol, 3020 // false for the expression form. 3021 bool UnwrappedLineParser::parseObjCProtocol() { 3022 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_protocol); 3023 nextToken(); 3024 3025 if (FormatTok->is(tok::l_paren)) 3026 // The expression form of @protocol, e.g. "Protocol* p = @protocol(foo);". 3027 return false; 3028 3029 // The definition/declaration form, 3030 // @protocol Foo 3031 // - (int)someMethod; 3032 // @end 3033 3034 nextToken(); // protocol name 3035 3036 if (FormatTok->Tok.is(tok::less)) 3037 parseObjCProtocolList(); 3038 3039 // Check for protocol declaration. 3040 if (FormatTok->Tok.is(tok::semi)) { 3041 nextToken(); 3042 addUnwrappedLine(); 3043 return true; 3044 } 3045 3046 addUnwrappedLine(); 3047 parseObjCUntilAtEnd(); 3048 return true; 3049 } 3050 3051 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 3052 bool IsImport = FormatTok->is(Keywords.kw_import); 3053 assert(IsImport || FormatTok->is(tok::kw_export)); 3054 nextToken(); 3055 3056 // Consume the "default" in "export default class/function". 3057 if (FormatTok->is(tok::kw_default)) 3058 nextToken(); 3059 3060 // Consume "async function", "function" and "default function", so that these 3061 // get parsed as free-standing JS functions, i.e. do not require a trailing 3062 // semicolon. 3063 if (FormatTok->is(Keywords.kw_async)) 3064 nextToken(); 3065 if (FormatTok->is(Keywords.kw_function)) { 3066 nextToken(); 3067 return; 3068 } 3069 3070 // For imports, `export *`, `export {...}`, consume the rest of the line up 3071 // to the terminating `;`. For everything else, just return and continue 3072 // parsing the structural element, i.e. the declaration or expression for 3073 // `export default`. 3074 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 3075 !FormatTok->isStringLiteral()) 3076 return; 3077 3078 while (!eof()) { 3079 if (FormatTok->is(tok::semi)) 3080 return; 3081 if (Line->Tokens.empty()) { 3082 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 3083 // import statement should terminate. 3084 return; 3085 } 3086 if (FormatTok->is(tok::l_brace)) { 3087 FormatTok->setBlockKind(BK_Block); 3088 nextToken(); 3089 parseBracedList(); 3090 } else { 3091 nextToken(); 3092 } 3093 } 3094 } 3095 3096 void UnwrappedLineParser::parseStatementMacro() { 3097 nextToken(); 3098 if (FormatTok->is(tok::l_paren)) 3099 parseParens(); 3100 if (FormatTok->is(tok::semi)) 3101 nextToken(); 3102 addUnwrappedLine(); 3103 } 3104 3105 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 3106 StringRef Prefix = "") { 3107 llvm::dbgs() << Prefix << "Line(" << Line.Level 3108 << ", FSC=" << Line.FirstStartColumn << ")" 3109 << (Line.InPPDirective ? " MACRO" : "") << ": "; 3110 for (const auto &Node : Line.Tokens) { 3111 llvm::dbgs() << Node.Tok->Tok.getName() << "[" 3112 << "T=" << static_cast<unsigned>(Node.Tok->getType()) 3113 << ", OC=" << Node.Tok->OriginalColumn << "] "; 3114 } 3115 for (const auto &Node : Line.Tokens) 3116 for (const auto &ChildNode : Node.Children) 3117 printDebugInfo(ChildNode, "\nChild: "); 3118 3119 llvm::dbgs() << "\n"; 3120 } 3121 3122 void UnwrappedLineParser::addUnwrappedLine(LineLevel AdjustLevel) { 3123 if (Line->Tokens.empty()) 3124 return; 3125 LLVM_DEBUG({ 3126 if (CurrentLines == &Lines) 3127 printDebugInfo(*Line); 3128 }); 3129 3130 // If this line closes a block when in Whitesmiths mode, remember that 3131 // information so that the level can be decreased after the line is added. 3132 // This has to happen after the addition of the line since the line itself 3133 // needs to be indented. 3134 bool ClosesWhitesmithsBlock = 3135 Line->MatchingOpeningBlockLineIndex != UnwrappedLine::kInvalidIndex && 3136 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths; 3137 3138 CurrentLines->push_back(std::move(*Line)); 3139 Line->Tokens.clear(); 3140 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex; 3141 Line->FirstStartColumn = 0; 3142 3143 if (ClosesWhitesmithsBlock && AdjustLevel == LineLevel::Remove) 3144 --Line->Level; 3145 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 3146 CurrentLines->append( 3147 std::make_move_iterator(PreprocessorDirectives.begin()), 3148 std::make_move_iterator(PreprocessorDirectives.end())); 3149 PreprocessorDirectives.clear(); 3150 } 3151 // Disconnect the current token from the last token on the previous line. 3152 FormatTok->Previous = nullptr; 3153 } 3154 3155 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 3156 3157 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 3158 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 3159 FormatTok.NewlinesBefore > 0; 3160 } 3161 3162 // Checks if \p FormatTok is a line comment that continues the line comment 3163 // section on \p Line. 3164 static bool 3165 continuesLineCommentSection(const FormatToken &FormatTok, 3166 const UnwrappedLine &Line, 3167 const llvm::Regex &CommentPragmasRegex) { 3168 if (Line.Tokens.empty()) 3169 return false; 3170 3171 StringRef IndentContent = FormatTok.TokenText; 3172 if (FormatTok.TokenText.startswith("//") || 3173 FormatTok.TokenText.startswith("/*")) 3174 IndentContent = FormatTok.TokenText.substr(2); 3175 if (CommentPragmasRegex.match(IndentContent)) 3176 return false; 3177 3178 // If Line starts with a line comment, then FormatTok continues the comment 3179 // section if its original column is greater or equal to the original start 3180 // column of the line. 3181 // 3182 // Define the min column token of a line as follows: if a line ends in '{' or 3183 // contains a '{' followed by a line comment, then the min column token is 3184 // that '{'. Otherwise, the min column token of the line is the first token of 3185 // the line. 3186 // 3187 // If Line starts with a token other than a line comment, then FormatTok 3188 // continues the comment section if its original column is greater than the 3189 // original start column of the min column token of the line. 3190 // 3191 // For example, the second line comment continues the first in these cases: 3192 // 3193 // // first line 3194 // // second line 3195 // 3196 // and: 3197 // 3198 // // first line 3199 // // second line 3200 // 3201 // and: 3202 // 3203 // int i; // first line 3204 // // second line 3205 // 3206 // and: 3207 // 3208 // do { // first line 3209 // // second line 3210 // int i; 3211 // } while (true); 3212 // 3213 // and: 3214 // 3215 // enum { 3216 // a, // first line 3217 // // second line 3218 // b 3219 // }; 3220 // 3221 // The second line comment doesn't continue the first in these cases: 3222 // 3223 // // first line 3224 // // second line 3225 // 3226 // and: 3227 // 3228 // int i; // first line 3229 // // second line 3230 // 3231 // and: 3232 // 3233 // do { // first line 3234 // // second line 3235 // int i; 3236 // } while (true); 3237 // 3238 // and: 3239 // 3240 // enum { 3241 // a, // first line 3242 // // second line 3243 // }; 3244 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 3245 3246 // Scan for '{//'. If found, use the column of '{' as a min column for line 3247 // comment section continuation. 3248 const FormatToken *PreviousToken = nullptr; 3249 for (const UnwrappedLineNode &Node : Line.Tokens) { 3250 if (PreviousToken && PreviousToken->is(tok::l_brace) && 3251 isLineComment(*Node.Tok)) { 3252 MinColumnToken = PreviousToken; 3253 break; 3254 } 3255 PreviousToken = Node.Tok; 3256 3257 // Grab the last newline preceding a token in this unwrapped line. 3258 if (Node.Tok->NewlinesBefore > 0) { 3259 MinColumnToken = Node.Tok; 3260 } 3261 } 3262 if (PreviousToken && PreviousToken->is(tok::l_brace)) { 3263 MinColumnToken = PreviousToken; 3264 } 3265 3266 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok, 3267 MinColumnToken); 3268 } 3269 3270 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 3271 bool JustComments = Line->Tokens.empty(); 3272 for (SmallVectorImpl<FormatToken *>::const_iterator 3273 I = CommentsBeforeNextToken.begin(), 3274 E = CommentsBeforeNextToken.end(); 3275 I != E; ++I) { 3276 // Line comments that belong to the same line comment section are put on the 3277 // same line since later we might want to reflow content between them. 3278 // Additional fine-grained breaking of line comment sections is controlled 3279 // by the class BreakableLineCommentSection in case it is desirable to keep 3280 // several line comment sections in the same unwrapped line. 3281 // 3282 // FIXME: Consider putting separate line comment sections as children to the 3283 // unwrapped line instead. 3284 (*I)->ContinuesLineCommentSection = 3285 continuesLineCommentSection(**I, *Line, CommentPragmasRegex); 3286 if (isOnNewLine(**I) && JustComments && !(*I)->ContinuesLineCommentSection) 3287 addUnwrappedLine(); 3288 pushToken(*I); 3289 } 3290 if (NewlineBeforeNext && JustComments) 3291 addUnwrappedLine(); 3292 CommentsBeforeNextToken.clear(); 3293 } 3294 3295 void UnwrappedLineParser::nextToken(int LevelDifference) { 3296 if (eof()) 3297 return; 3298 flushComments(isOnNewLine(*FormatTok)); 3299 pushToken(FormatTok); 3300 FormatToken *Previous = FormatTok; 3301 if (!Style.isJavaScript()) 3302 readToken(LevelDifference); 3303 else 3304 readTokenWithJavaScriptASI(); 3305 FormatTok->Previous = Previous; 3306 } 3307 3308 void UnwrappedLineParser::distributeComments( 3309 const SmallVectorImpl<FormatToken *> &Comments, 3310 const FormatToken *NextTok) { 3311 // Whether or not a line comment token continues a line is controlled by 3312 // the method continuesLineCommentSection, with the following caveat: 3313 // 3314 // Define a trail of Comments to be a nonempty proper postfix of Comments such 3315 // that each comment line from the trail is aligned with the next token, if 3316 // the next token exists. If a trail exists, the beginning of the maximal 3317 // trail is marked as a start of a new comment section. 3318 // 3319 // For example in this code: 3320 // 3321 // int a; // line about a 3322 // // line 1 about b 3323 // // line 2 about b 3324 // int b; 3325 // 3326 // the two lines about b form a maximal trail, so there are two sections, the 3327 // first one consisting of the single comment "// line about a" and the 3328 // second one consisting of the next two comments. 3329 if (Comments.empty()) 3330 return; 3331 bool ShouldPushCommentsInCurrentLine = true; 3332 bool HasTrailAlignedWithNextToken = false; 3333 unsigned StartOfTrailAlignedWithNextToken = 0; 3334 if (NextTok) { 3335 // We are skipping the first element intentionally. 3336 for (unsigned i = Comments.size() - 1; i > 0; --i) { 3337 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) { 3338 HasTrailAlignedWithNextToken = true; 3339 StartOfTrailAlignedWithNextToken = i; 3340 } 3341 } 3342 } 3343 for (unsigned i = 0, e = Comments.size(); i < e; ++i) { 3344 FormatToken *FormatTok = Comments[i]; 3345 if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) { 3346 FormatTok->ContinuesLineCommentSection = false; 3347 } else { 3348 FormatTok->ContinuesLineCommentSection = 3349 continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex); 3350 } 3351 if (!FormatTok->ContinuesLineCommentSection && 3352 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) { 3353 ShouldPushCommentsInCurrentLine = false; 3354 } 3355 if (ShouldPushCommentsInCurrentLine) { 3356 pushToken(FormatTok); 3357 } else { 3358 CommentsBeforeNextToken.push_back(FormatTok); 3359 } 3360 } 3361 } 3362 3363 void UnwrappedLineParser::readToken(int LevelDifference) { 3364 SmallVector<FormatToken *, 1> Comments; 3365 do { 3366 FormatTok = Tokens->getNextToken(); 3367 assert(FormatTok); 3368 while (FormatTok->getType() == TT_ConflictStart || 3369 FormatTok->getType() == TT_ConflictEnd || 3370 FormatTok->getType() == TT_ConflictAlternative) { 3371 if (FormatTok->getType() == TT_ConflictStart) { 3372 conditionalCompilationStart(/*Unreachable=*/false); 3373 } else if (FormatTok->getType() == TT_ConflictAlternative) { 3374 conditionalCompilationAlternative(); 3375 } else if (FormatTok->getType() == TT_ConflictEnd) { 3376 conditionalCompilationEnd(); 3377 } 3378 FormatTok = Tokens->getNextToken(); 3379 FormatTok->MustBreakBefore = true; 3380 } 3381 3382 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 3383 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 3384 distributeComments(Comments, FormatTok); 3385 Comments.clear(); 3386 // If there is an unfinished unwrapped line, we flush the preprocessor 3387 // directives only after that unwrapped line was finished later. 3388 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 3389 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 3390 assert((LevelDifference >= 0 || 3391 static_cast<unsigned>(-LevelDifference) <= Line->Level) && 3392 "LevelDifference makes Line->Level negative"); 3393 Line->Level += LevelDifference; 3394 // Comments stored before the preprocessor directive need to be output 3395 // before the preprocessor directive, at the same level as the 3396 // preprocessor directive, as we consider them to apply to the directive. 3397 if (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash && 3398 PPBranchLevel > 0) 3399 Line->Level += PPBranchLevel; 3400 flushComments(isOnNewLine(*FormatTok)); 3401 parsePPDirective(); 3402 } 3403 3404 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) && 3405 !Line->InPPDirective) { 3406 continue; 3407 } 3408 3409 if (!FormatTok->Tok.is(tok::comment)) { 3410 distributeComments(Comments, FormatTok); 3411 Comments.clear(); 3412 return; 3413 } 3414 3415 Comments.push_back(FormatTok); 3416 } while (!eof()); 3417 3418 distributeComments(Comments, nullptr); 3419 Comments.clear(); 3420 } 3421 3422 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 3423 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 3424 if (MustBreakBeforeNextToken) { 3425 Line->Tokens.back().Tok->MustBreakBefore = true; 3426 MustBreakBeforeNextToken = false; 3427 } 3428 } 3429 3430 } // end namespace format 3431 } // end namespace clang 3432