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 return false; 786 } 787 788 void UnwrappedLineParser::parseChildBlock() { 789 FormatTok->setBlockKind(BK_Block); 790 nextToken(); 791 { 792 bool SkipIndent = (Style.isJavaScript() && 793 (isGoogScope(*Line) || isIIFE(*Line, Keywords))); 794 ScopedLineState LineState(*this); 795 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 796 /*MustBeDeclaration=*/false); 797 Line->Level += SkipIndent ? 0 : 1; 798 parseLevel(/*HasOpeningBrace=*/true); 799 flushComments(isOnNewLine(*FormatTok)); 800 Line->Level -= SkipIndent ? 0 : 1; 801 } 802 nextToken(); 803 } 804 805 void UnwrappedLineParser::parsePPDirective() { 806 assert(FormatTok->Tok.is(tok::hash) && "'#' expected"); 807 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 808 809 nextToken(); 810 811 if (!FormatTok->Tok.getIdentifierInfo()) { 812 parsePPUnknown(); 813 return; 814 } 815 816 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 817 case tok::pp_define: 818 parsePPDefine(); 819 return; 820 case tok::pp_if: 821 parsePPIf(/*IfDef=*/false); 822 break; 823 case tok::pp_ifdef: 824 case tok::pp_ifndef: 825 parsePPIf(/*IfDef=*/true); 826 break; 827 case tok::pp_else: 828 parsePPElse(); 829 break; 830 case tok::pp_elifdef: 831 case tok::pp_elifndef: 832 case tok::pp_elif: 833 parsePPElIf(); 834 break; 835 case tok::pp_endif: 836 parsePPEndIf(); 837 break; 838 default: 839 parsePPUnknown(); 840 break; 841 } 842 } 843 844 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 845 size_t Line = CurrentLines->size(); 846 if (CurrentLines == &PreprocessorDirectives) 847 Line += Lines.size(); 848 849 if (Unreachable || 850 (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable)) 851 PPStack.push_back({PP_Unreachable, Line}); 852 else 853 PPStack.push_back({PP_Conditional, Line}); 854 } 855 856 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 857 ++PPBranchLevel; 858 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 859 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 860 PPLevelBranchIndex.push_back(0); 861 PPLevelBranchCount.push_back(0); 862 } 863 PPChainBranchIndex.push(0); 864 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 865 conditionalCompilationCondition(Unreachable || Skip); 866 } 867 868 void UnwrappedLineParser::conditionalCompilationAlternative() { 869 if (!PPStack.empty()) 870 PPStack.pop_back(); 871 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 872 if (!PPChainBranchIndex.empty()) 873 ++PPChainBranchIndex.top(); 874 conditionalCompilationCondition( 875 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 876 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 877 } 878 879 void UnwrappedLineParser::conditionalCompilationEnd() { 880 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 881 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 882 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) { 883 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 884 } 885 } 886 // Guard against #endif's without #if. 887 if (PPBranchLevel > -1) 888 --PPBranchLevel; 889 if (!PPChainBranchIndex.empty()) 890 PPChainBranchIndex.pop(); 891 if (!PPStack.empty()) 892 PPStack.pop_back(); 893 } 894 895 void UnwrappedLineParser::parsePPIf(bool IfDef) { 896 bool IfNDef = FormatTok->is(tok::pp_ifndef); 897 nextToken(); 898 bool Unreachable = false; 899 if (!IfDef && (FormatTok->is(tok::kw_false) || FormatTok->TokenText == "0")) 900 Unreachable = true; 901 if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG") 902 Unreachable = true; 903 conditionalCompilationStart(Unreachable); 904 FormatToken *IfCondition = FormatTok; 905 // If there's a #ifndef on the first line, and the only lines before it are 906 // comments, it could be an include guard. 907 bool MaybeIncludeGuard = IfNDef; 908 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) 909 for (auto &Line : Lines) { 910 if (!Line.Tokens.front().Tok->is(tok::comment)) { 911 MaybeIncludeGuard = false; 912 IncludeGuard = IG_Rejected; 913 break; 914 } 915 } 916 --PPBranchLevel; 917 parsePPUnknown(); 918 ++PPBranchLevel; 919 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) { 920 IncludeGuard = IG_IfNdefed; 921 IncludeGuardToken = IfCondition; 922 } 923 } 924 925 void UnwrappedLineParser::parsePPElse() { 926 // If a potential include guard has an #else, it's not an include guard. 927 if (IncludeGuard == IG_Defined && PPBranchLevel == 0) 928 IncludeGuard = IG_Rejected; 929 conditionalCompilationAlternative(); 930 if (PPBranchLevel > -1) 931 --PPBranchLevel; 932 parsePPUnknown(); 933 ++PPBranchLevel; 934 } 935 936 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 937 938 void UnwrappedLineParser::parsePPEndIf() { 939 conditionalCompilationEnd(); 940 parsePPUnknown(); 941 // If the #endif of a potential include guard is the last thing in the file, 942 // then we found an include guard. 943 if (IncludeGuard == IG_Defined && PPBranchLevel == -1 && Tokens->isEOF() && 944 Style.IndentPPDirectives != FormatStyle::PPDIS_None) 945 IncludeGuard = IG_Found; 946 } 947 948 void UnwrappedLineParser::parsePPDefine() { 949 nextToken(); 950 951 if (!FormatTok->Tok.getIdentifierInfo()) { 952 IncludeGuard = IG_Rejected; 953 IncludeGuardToken = nullptr; 954 parsePPUnknown(); 955 return; 956 } 957 958 if (IncludeGuard == IG_IfNdefed && 959 IncludeGuardToken->TokenText == FormatTok->TokenText) { 960 IncludeGuard = IG_Defined; 961 IncludeGuardToken = nullptr; 962 for (auto &Line : Lines) { 963 if (!Line.Tokens.front().Tok->isOneOf(tok::comment, tok::hash)) { 964 IncludeGuard = IG_Rejected; 965 break; 966 } 967 } 968 } 969 970 nextToken(); 971 if (FormatTok->Tok.getKind() == tok::l_paren && 972 FormatTok->WhitespaceRange.getBegin() == 973 FormatTok->WhitespaceRange.getEnd()) { 974 parseParens(); 975 } 976 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None) 977 Line->Level += PPBranchLevel + 1; 978 addUnwrappedLine(); 979 ++Line->Level; 980 981 // Errors during a preprocessor directive can only affect the layout of the 982 // preprocessor directive, and thus we ignore them. An alternative approach 983 // would be to use the same approach we use on the file level (no 984 // re-indentation if there was a structural error) within the macro 985 // definition. 986 parseFile(); 987 } 988 989 void UnwrappedLineParser::parsePPUnknown() { 990 do { 991 nextToken(); 992 } while (!eof()); 993 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None) 994 Line->Level += PPBranchLevel + 1; 995 addUnwrappedLine(); 996 } 997 998 // Here we exclude certain tokens that are not usually the first token in an 999 // unwrapped line. This is used in attempt to distinguish macro calls without 1000 // trailing semicolons from other constructs split to several lines. 1001 static bool tokenCanStartNewLine(const FormatToken &Tok) { 1002 // Semicolon can be a null-statement, l_square can be a start of a macro or 1003 // a C++11 attribute, but this doesn't seem to be common. 1004 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 1005 Tok.isNot(TT_AttributeSquare) && 1006 // Tokens that can only be used as binary operators and a part of 1007 // overloaded operator names. 1008 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 1009 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 1010 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 1011 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 1012 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 1013 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 1014 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 1015 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 1016 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 1017 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 1018 Tok.isNot(tok::lesslessequal) && 1019 // Colon is used in labels, base class lists, initializer lists, 1020 // range-based for loops, ternary operator, but should never be the 1021 // first token in an unwrapped line. 1022 Tok.isNot(tok::colon) && 1023 // 'noexcept' is a trailing annotation. 1024 Tok.isNot(tok::kw_noexcept); 1025 } 1026 1027 static bool mustBeJSIdent(const AdditionalKeywords &Keywords, 1028 const FormatToken *FormatTok) { 1029 // FIXME: This returns true for C/C++ keywords like 'struct'. 1030 return FormatTok->is(tok::identifier) && 1031 (FormatTok->Tok.getIdentifierInfo() == nullptr || 1032 !FormatTok->isOneOf( 1033 Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async, 1034 Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally, 1035 Keywords.kw_function, Keywords.kw_import, Keywords.kw_is, 1036 Keywords.kw_let, Keywords.kw_var, tok::kw_const, 1037 Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements, 1038 Keywords.kw_instanceof, Keywords.kw_interface, 1039 Keywords.kw_override, Keywords.kw_throws, Keywords.kw_from)); 1040 } 1041 1042 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords, 1043 const FormatToken *FormatTok) { 1044 return FormatTok->Tok.isLiteral() || 1045 FormatTok->isOneOf(tok::kw_true, tok::kw_false) || 1046 mustBeJSIdent(Keywords, FormatTok); 1047 } 1048 1049 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement 1050 // when encountered after a value (see mustBeJSIdentOrValue). 1051 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords, 1052 const FormatToken *FormatTok) { 1053 return FormatTok->isOneOf( 1054 tok::kw_return, Keywords.kw_yield, 1055 // conditionals 1056 tok::kw_if, tok::kw_else, 1057 // loops 1058 tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break, 1059 // switch/case 1060 tok::kw_switch, tok::kw_case, 1061 // exceptions 1062 tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally, 1063 // declaration 1064 tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let, 1065 Keywords.kw_async, Keywords.kw_function, 1066 // import/export 1067 Keywords.kw_import, tok::kw_export); 1068 } 1069 1070 // Checks whether a token is a type in K&R C (aka C78). 1071 static bool isC78Type(const FormatToken &Tok) { 1072 return Tok.isOneOf(tok::kw_char, tok::kw_short, tok::kw_int, tok::kw_long, 1073 tok::kw_unsigned, tok::kw_float, tok::kw_double, 1074 tok::identifier); 1075 } 1076 1077 // This function checks whether a token starts the first parameter declaration 1078 // in a K&R C (aka C78) function definition, e.g.: 1079 // int f(a, b) 1080 // short a, b; 1081 // { 1082 // return a + b; 1083 // } 1084 static bool isC78ParameterDecl(const FormatToken *Tok, const FormatToken *Next, 1085 const FormatToken *FuncName) { 1086 assert(Tok); 1087 assert(Next); 1088 assert(FuncName); 1089 1090 if (FuncName->isNot(tok::identifier)) 1091 return false; 1092 1093 const FormatToken *Prev = FuncName->Previous; 1094 if (!Prev || (Prev->isNot(tok::star) && !isC78Type(*Prev))) 1095 return false; 1096 1097 if (!isC78Type(*Tok) && 1098 !Tok->isOneOf(tok::kw_register, tok::kw_struct, tok::kw_union)) 1099 return false; 1100 1101 if (Next->isNot(tok::star) && !Next->Tok.getIdentifierInfo()) 1102 return false; 1103 1104 Tok = Tok->Previous; 1105 if (!Tok || Tok->isNot(tok::r_paren)) 1106 return false; 1107 1108 Tok = Tok->Previous; 1109 if (!Tok || Tok->isNot(tok::identifier)) 1110 return false; 1111 1112 return Tok->Previous && Tok->Previous->isOneOf(tok::l_paren, tok::comma); 1113 } 1114 1115 void UnwrappedLineParser::parseModuleImport() { 1116 nextToken(); 1117 while (!eof()) { 1118 if (FormatTok->is(tok::colon)) { 1119 FormatTok->setType(TT_ModulePartitionColon); 1120 } 1121 // Handle import <foo/bar.h> as we would an include statement. 1122 else if (FormatTok->is(tok::less)) { 1123 nextToken(); 1124 while (!FormatTok->isOneOf(tok::semi, tok::greater, tok::eof)) { 1125 // Mark tokens up to the trailing line comments as implicit string 1126 // literals. 1127 if (FormatTok->isNot(tok::comment) && 1128 !FormatTok->TokenText.startswith("//")) 1129 FormatTok->setType(TT_ImplicitStringLiteral); 1130 nextToken(); 1131 } 1132 } 1133 if (FormatTok->is(tok::semi)) { 1134 nextToken(); 1135 break; 1136 } 1137 nextToken(); 1138 } 1139 1140 addUnwrappedLine(); 1141 return; 1142 } 1143 1144 // readTokenWithJavaScriptASI reads the next token and terminates the current 1145 // line if JavaScript Automatic Semicolon Insertion must 1146 // happen between the current token and the next token. 1147 // 1148 // This method is conservative - it cannot cover all edge cases of JavaScript, 1149 // but only aims to correctly handle certain well known cases. It *must not* 1150 // return true in speculative cases. 1151 void UnwrappedLineParser::readTokenWithJavaScriptASI() { 1152 FormatToken *Previous = FormatTok; 1153 readToken(); 1154 FormatToken *Next = FormatTok; 1155 1156 bool IsOnSameLine = 1157 CommentsBeforeNextToken.empty() 1158 ? Next->NewlinesBefore == 0 1159 : CommentsBeforeNextToken.front()->NewlinesBefore == 0; 1160 if (IsOnSameLine) 1161 return; 1162 1163 bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous); 1164 bool PreviousStartsTemplateExpr = 1165 Previous->is(TT_TemplateString) && Previous->TokenText.endswith("${"); 1166 if (PreviousMustBeValue || Previous->is(tok::r_paren)) { 1167 // If the line contains an '@' sign, the previous token might be an 1168 // annotation, which can precede another identifier/value. 1169 bool HasAt = llvm::any_of(Line->Tokens, [](UnwrappedLineNode &LineNode) { 1170 return LineNode.Tok->is(tok::at); 1171 }); 1172 if (HasAt) 1173 return; 1174 } 1175 if (Next->is(tok::exclaim) && PreviousMustBeValue) 1176 return addUnwrappedLine(); 1177 bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next); 1178 bool NextEndsTemplateExpr = 1179 Next->is(TT_TemplateString) && Next->TokenText.startswith("}"); 1180 if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr && 1181 (PreviousMustBeValue || 1182 Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus, 1183 tok::minusminus))) 1184 return addUnwrappedLine(); 1185 if ((PreviousMustBeValue || Previous->is(tok::r_paren)) && 1186 isJSDeclOrStmt(Keywords, Next)) 1187 return addUnwrappedLine(); 1188 } 1189 1190 void UnwrappedLineParser::parseStructuralElement(bool IsTopLevel) { 1191 if (Style.Language == FormatStyle::LK_TableGen && 1192 FormatTok->is(tok::pp_include)) { 1193 nextToken(); 1194 if (FormatTok->is(tok::string_literal)) 1195 nextToken(); 1196 addUnwrappedLine(); 1197 return; 1198 } 1199 switch (FormatTok->Tok.getKind()) { 1200 case tok::kw_asm: 1201 nextToken(); 1202 if (FormatTok->is(tok::l_brace)) { 1203 FormatTok->setType(TT_InlineASMBrace); 1204 nextToken(); 1205 while (FormatTok && FormatTok->isNot(tok::eof)) { 1206 if (FormatTok->is(tok::r_brace)) { 1207 FormatTok->setType(TT_InlineASMBrace); 1208 nextToken(); 1209 addUnwrappedLine(); 1210 break; 1211 } 1212 FormatTok->Finalized = true; 1213 nextToken(); 1214 } 1215 } 1216 break; 1217 case tok::kw_namespace: 1218 parseNamespace(); 1219 return; 1220 case tok::kw_public: 1221 case tok::kw_protected: 1222 case tok::kw_private: 1223 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 1224 Style.isCSharp()) 1225 nextToken(); 1226 else 1227 parseAccessSpecifier(); 1228 return; 1229 case tok::kw_if: 1230 if (Style.isJavaScript() && Line->MustBeDeclaration) 1231 // field/method declaration. 1232 break; 1233 parseIfThenElse(); 1234 return; 1235 case tok::kw_for: 1236 case tok::kw_while: 1237 if (Style.isJavaScript() && Line->MustBeDeclaration) 1238 // field/method declaration. 1239 break; 1240 parseForOrWhileLoop(); 1241 return; 1242 case tok::kw_do: 1243 if (Style.isJavaScript() && Line->MustBeDeclaration) 1244 // field/method declaration. 1245 break; 1246 parseDoWhile(); 1247 return; 1248 case tok::kw_switch: 1249 if (Style.isJavaScript() && Line->MustBeDeclaration) 1250 // 'switch: string' field declaration. 1251 break; 1252 parseSwitch(); 1253 return; 1254 case tok::kw_default: 1255 if (Style.isJavaScript() && Line->MustBeDeclaration) 1256 // 'default: string' field declaration. 1257 break; 1258 nextToken(); 1259 if (FormatTok->is(tok::colon)) { 1260 parseLabel(); 1261 return; 1262 } 1263 // e.g. "default void f() {}" in a Java interface. 1264 break; 1265 case tok::kw_case: 1266 if (Style.isJavaScript() && Line->MustBeDeclaration) 1267 // 'case: string' field declaration. 1268 break; 1269 parseCaseLabel(); 1270 return; 1271 case tok::kw_try: 1272 case tok::kw___try: 1273 if (Style.isJavaScript() && Line->MustBeDeclaration) 1274 // field/method declaration. 1275 break; 1276 parseTryCatch(); 1277 return; 1278 case tok::kw_extern: 1279 nextToken(); 1280 if (FormatTok->Tok.is(tok::string_literal)) { 1281 nextToken(); 1282 if (FormatTok->Tok.is(tok::l_brace)) { 1283 if (Style.BraceWrapping.AfterExternBlock) 1284 addUnwrappedLine(); 1285 // Either we indent or for backwards compatibility we follow the 1286 // AfterExternBlock style. 1287 unsigned AddLevels = 1288 (Style.IndentExternBlock == FormatStyle::IEBS_Indent) || 1289 (Style.BraceWrapping.AfterExternBlock && 1290 Style.IndentExternBlock == 1291 FormatStyle::IEBS_AfterExternBlock) 1292 ? 1u 1293 : 0u; 1294 parseBlock(/*MustBeDeclaration=*/true, AddLevels); 1295 addUnwrappedLine(); 1296 return; 1297 } 1298 } 1299 break; 1300 case tok::kw_export: 1301 if (Style.isJavaScript()) { 1302 parseJavaScriptEs6ImportExport(); 1303 return; 1304 } 1305 if (!Style.isCpp()) 1306 break; 1307 // Handle C++ "(inline|export) namespace". 1308 LLVM_FALLTHROUGH; 1309 case tok::kw_inline: 1310 nextToken(); 1311 if (FormatTok->Tok.is(tok::kw_namespace)) { 1312 parseNamespace(); 1313 return; 1314 } 1315 break; 1316 case tok::identifier: 1317 if (FormatTok->is(TT_ForEachMacro)) { 1318 parseForOrWhileLoop(); 1319 return; 1320 } 1321 if (FormatTok->is(TT_MacroBlockBegin)) { 1322 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 1323 /*MunchSemi=*/false); 1324 return; 1325 } 1326 if (FormatTok->is(Keywords.kw_import)) { 1327 if (Style.isJavaScript()) { 1328 parseJavaScriptEs6ImportExport(); 1329 return; 1330 } 1331 if (Style.Language == FormatStyle::LK_Proto) { 1332 nextToken(); 1333 if (FormatTok->is(tok::kw_public)) 1334 nextToken(); 1335 if (!FormatTok->is(tok::string_literal)) 1336 return; 1337 nextToken(); 1338 if (FormatTok->is(tok::semi)) 1339 nextToken(); 1340 addUnwrappedLine(); 1341 return; 1342 } 1343 if (Style.isCpp()) { 1344 parseModuleImport(); 1345 return; 1346 } 1347 } 1348 if (Style.isCpp() && 1349 FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 1350 Keywords.kw_slots, Keywords.kw_qslots)) { 1351 nextToken(); 1352 if (FormatTok->is(tok::colon)) { 1353 nextToken(); 1354 addUnwrappedLine(); 1355 return; 1356 } 1357 } 1358 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1359 parseStatementMacro(); 1360 return; 1361 } 1362 if (Style.isCpp() && FormatTok->is(TT_NamespaceMacro)) { 1363 parseNamespace(); 1364 return; 1365 } 1366 // In all other cases, parse the declaration. 1367 break; 1368 default: 1369 break; 1370 } 1371 do { 1372 const FormatToken *Previous = FormatTok->Previous; 1373 switch (FormatTok->Tok.getKind()) { 1374 case tok::at: 1375 nextToken(); 1376 if (FormatTok->Tok.is(tok::l_brace)) { 1377 nextToken(); 1378 parseBracedList(); 1379 break; 1380 } else if (Style.Language == FormatStyle::LK_Java && 1381 FormatTok->is(Keywords.kw_interface)) { 1382 nextToken(); 1383 break; 1384 } 1385 switch (FormatTok->Tok.getObjCKeywordID()) { 1386 case tok::objc_public: 1387 case tok::objc_protected: 1388 case tok::objc_package: 1389 case tok::objc_private: 1390 return parseAccessSpecifier(); 1391 case tok::objc_interface: 1392 case tok::objc_implementation: 1393 return parseObjCInterfaceOrImplementation(); 1394 case tok::objc_protocol: 1395 if (parseObjCProtocol()) 1396 return; 1397 break; 1398 case tok::objc_end: 1399 return; // Handled by the caller. 1400 case tok::objc_optional: 1401 case tok::objc_required: 1402 nextToken(); 1403 addUnwrappedLine(); 1404 return; 1405 case tok::objc_autoreleasepool: 1406 nextToken(); 1407 if (FormatTok->Tok.is(tok::l_brace)) { 1408 if (Style.BraceWrapping.AfterControlStatement == 1409 FormatStyle::BWACS_Always) 1410 addUnwrappedLine(); 1411 parseBlock(); 1412 } 1413 addUnwrappedLine(); 1414 return; 1415 case tok::objc_synchronized: 1416 nextToken(); 1417 if (FormatTok->Tok.is(tok::l_paren)) 1418 // Skip synchronization object 1419 parseParens(); 1420 if (FormatTok->Tok.is(tok::l_brace)) { 1421 if (Style.BraceWrapping.AfterControlStatement == 1422 FormatStyle::BWACS_Always) 1423 addUnwrappedLine(); 1424 parseBlock(); 1425 } 1426 addUnwrappedLine(); 1427 return; 1428 case tok::objc_try: 1429 // This branch isn't strictly necessary (the kw_try case below would 1430 // do this too after the tok::at is parsed above). But be explicit. 1431 parseTryCatch(); 1432 return; 1433 default: 1434 break; 1435 } 1436 break; 1437 case tok::kw_concept: 1438 parseConcept(); 1439 return; 1440 case tok::kw_requires: 1441 parseRequires(); 1442 break; 1443 case tok::kw_enum: 1444 // Ignore if this is part of "template <enum ...". 1445 if (Previous && Previous->is(tok::less)) { 1446 nextToken(); 1447 break; 1448 } 1449 1450 // parseEnum falls through and does not yet add an unwrapped line as an 1451 // enum definition can start a structural element. 1452 if (!parseEnum()) 1453 break; 1454 // This only applies for C++. 1455 if (!Style.isCpp()) { 1456 addUnwrappedLine(); 1457 return; 1458 } 1459 break; 1460 case tok::kw_typedef: 1461 nextToken(); 1462 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 1463 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS, 1464 Keywords.kw_CF_CLOSED_ENUM, 1465 Keywords.kw_NS_CLOSED_ENUM)) 1466 parseEnum(); 1467 break; 1468 case tok::kw_struct: 1469 case tok::kw_union: 1470 case tok::kw_class: 1471 if (parseStructLike()) { 1472 return; 1473 } 1474 break; 1475 case tok::period: 1476 nextToken(); 1477 // In Java, classes have an implicit static member "class". 1478 if (Style.Language == FormatStyle::LK_Java && FormatTok && 1479 FormatTok->is(tok::kw_class)) 1480 nextToken(); 1481 if (Style.isJavaScript() && FormatTok && 1482 FormatTok->Tok.getIdentifierInfo()) 1483 // JavaScript only has pseudo keywords, all keywords are allowed to 1484 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 1485 nextToken(); 1486 break; 1487 case tok::semi: 1488 nextToken(); 1489 addUnwrappedLine(); 1490 return; 1491 case tok::r_brace: 1492 addUnwrappedLine(); 1493 return; 1494 case tok::l_paren: { 1495 parseParens(); 1496 // Break the unwrapped line if a K&R C function definition has a parameter 1497 // declaration. 1498 if (!IsTopLevel || !Style.isCpp() || !Previous || FormatTok->is(tok::eof)) 1499 break; 1500 if (isC78ParameterDecl(FormatTok, Tokens->peekNextToken(), Previous)) { 1501 addUnwrappedLine(); 1502 return; 1503 } 1504 break; 1505 } 1506 case tok::kw_operator: 1507 nextToken(); 1508 if (FormatTok->isBinaryOperator()) 1509 nextToken(); 1510 break; 1511 case tok::caret: 1512 nextToken(); 1513 if (FormatTok->Tok.isAnyIdentifier() || 1514 FormatTok->isSimpleTypeSpecifier()) 1515 nextToken(); 1516 if (FormatTok->is(tok::l_paren)) 1517 parseParens(); 1518 if (FormatTok->is(tok::l_brace)) 1519 parseChildBlock(); 1520 break; 1521 case tok::l_brace: 1522 if (!tryToParsePropertyAccessor() && !tryToParseBracedList()) { 1523 // A block outside of parentheses must be the last part of a 1524 // structural element. 1525 // FIXME: Figure out cases where this is not true, and add projections 1526 // for them (the one we know is missing are lambdas). 1527 if (Style.BraceWrapping.AfterFunction) 1528 addUnwrappedLine(); 1529 FormatTok->setType(TT_FunctionLBrace); 1530 parseBlock(); 1531 addUnwrappedLine(); 1532 return; 1533 } 1534 // Otherwise this was a braced init list, and the structural 1535 // element continues. 1536 break; 1537 case tok::kw_try: 1538 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1539 // field/method declaration. 1540 nextToken(); 1541 break; 1542 } 1543 // We arrive here when parsing function-try blocks. 1544 if (Style.BraceWrapping.AfterFunction) 1545 addUnwrappedLine(); 1546 parseTryCatch(); 1547 return; 1548 case tok::identifier: { 1549 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where) && 1550 Line->MustBeDeclaration) { 1551 addUnwrappedLine(); 1552 parseCSharpGenericTypeConstraint(); 1553 break; 1554 } 1555 if (FormatTok->is(TT_MacroBlockEnd)) { 1556 addUnwrappedLine(); 1557 return; 1558 } 1559 1560 // Function declarations (as opposed to function expressions) are parsed 1561 // on their own unwrapped line by continuing this loop. Function 1562 // expressions (functions that are not on their own line) must not create 1563 // a new unwrapped line, so they are special cased below. 1564 size_t TokenCount = Line->Tokens.size(); 1565 if (Style.isJavaScript() && FormatTok->is(Keywords.kw_function) && 1566 (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is( 1567 Keywords.kw_async)))) { 1568 tryToParseJSFunction(); 1569 break; 1570 } 1571 if ((Style.isJavaScript() || Style.Language == FormatStyle::LK_Java) && 1572 FormatTok->is(Keywords.kw_interface)) { 1573 if (Style.isJavaScript()) { 1574 // In JavaScript/TypeScript, "interface" can be used as a standalone 1575 // identifier, e.g. in `var interface = 1;`. If "interface" is 1576 // followed by another identifier, it is very like to be an actual 1577 // interface declaration. 1578 unsigned StoredPosition = Tokens->getPosition(); 1579 FormatToken *Next = Tokens->getNextToken(); 1580 FormatTok = Tokens->setPosition(StoredPosition); 1581 if (!mustBeJSIdent(Keywords, Next)) { 1582 nextToken(); 1583 break; 1584 } 1585 } 1586 parseRecord(); 1587 addUnwrappedLine(); 1588 return; 1589 } 1590 1591 if (FormatTok->is(Keywords.kw_interface)) { 1592 if (parseStructLike()) { 1593 return; 1594 } 1595 break; 1596 } 1597 1598 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1599 parseStatementMacro(); 1600 return; 1601 } 1602 1603 // See if the following token should start a new unwrapped line. 1604 StringRef Text = FormatTok->TokenText; 1605 nextToken(); 1606 1607 // JS doesn't have macros, and within classes colons indicate fields, not 1608 // labels. 1609 if (Style.isJavaScript()) 1610 break; 1611 1612 TokenCount = Line->Tokens.size(); 1613 if (TokenCount == 1 || 1614 (TokenCount == 2 && Line->Tokens.front().Tok->is(tok::comment))) { 1615 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) { 1616 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1617 parseLabel(!Style.IndentGotoLabels); 1618 return; 1619 } 1620 // Recognize function-like macro usages without trailing semicolon as 1621 // well as free-standing macros like Q_OBJECT. 1622 bool FunctionLike = FormatTok->is(tok::l_paren); 1623 if (FunctionLike) 1624 parseParens(); 1625 1626 bool FollowedByNewline = 1627 CommentsBeforeNextToken.empty() 1628 ? FormatTok->NewlinesBefore > 0 1629 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1630 1631 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1632 tokenCanStartNewLine(*FormatTok) && Text == Text.upper()) { 1633 addUnwrappedLine(); 1634 return; 1635 } 1636 } 1637 break; 1638 } 1639 case tok::equal: 1640 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType 1641 // TT_FatArrow. They always start an expression or a child block if 1642 // followed by a curly brace. 1643 if (FormatTok->is(TT_FatArrow)) { 1644 nextToken(); 1645 if (FormatTok->is(tok::l_brace)) { 1646 // C# may break after => if the next character is a newline. 1647 if (Style.isCSharp() && Style.BraceWrapping.AfterFunction == true) { 1648 // calling `addUnwrappedLine()` here causes odd parsing errors. 1649 FormatTok->MustBreakBefore = true; 1650 } 1651 parseChildBlock(); 1652 } 1653 break; 1654 } 1655 1656 nextToken(); 1657 if (FormatTok->Tok.is(tok::l_brace)) { 1658 // Block kind should probably be set to BK_BracedInit for any language. 1659 // C# needs this change to ensure that array initialisers and object 1660 // initialisers are indented the same way. 1661 if (Style.isCSharp()) 1662 FormatTok->setBlockKind(BK_BracedInit); 1663 nextToken(); 1664 parseBracedList(); 1665 } else if (Style.Language == FormatStyle::LK_Proto && 1666 FormatTok->Tok.is(tok::less)) { 1667 nextToken(); 1668 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 1669 /*ClosingBraceKind=*/tok::greater); 1670 } 1671 break; 1672 case tok::l_square: 1673 parseSquare(); 1674 break; 1675 case tok::kw_new: 1676 parseNew(); 1677 break; 1678 default: 1679 nextToken(); 1680 break; 1681 } 1682 } while (!eof()); 1683 } 1684 1685 bool UnwrappedLineParser::tryToParsePropertyAccessor() { 1686 assert(FormatTok->is(tok::l_brace)); 1687 if (!Style.isCSharp()) 1688 return false; 1689 // See if it's a property accessor. 1690 if (FormatTok->Previous->isNot(tok::identifier)) 1691 return false; 1692 1693 // See if we are inside a property accessor. 1694 // 1695 // Record the current tokenPosition so that we can advance and 1696 // reset the current token. `Next` is not set yet so we need 1697 // another way to advance along the token stream. 1698 unsigned int StoredPosition = Tokens->getPosition(); 1699 FormatToken *Tok = Tokens->getNextToken(); 1700 1701 // A trivial property accessor is of the form: 1702 // { [ACCESS_SPECIFIER] [get]; [ACCESS_SPECIFIER] [set] } 1703 // Track these as they do not require line breaks to be introduced. 1704 bool HasGetOrSet = false; 1705 bool IsTrivialPropertyAccessor = true; 1706 while (!eof()) { 1707 if (Tok->isOneOf(tok::semi, tok::kw_public, tok::kw_private, 1708 tok::kw_protected, Keywords.kw_internal, Keywords.kw_get, 1709 Keywords.kw_set)) { 1710 if (Tok->isOneOf(Keywords.kw_get, Keywords.kw_set)) 1711 HasGetOrSet = true; 1712 Tok = Tokens->getNextToken(); 1713 continue; 1714 } 1715 if (Tok->isNot(tok::r_brace)) 1716 IsTrivialPropertyAccessor = false; 1717 break; 1718 } 1719 1720 if (!HasGetOrSet) { 1721 Tokens->setPosition(StoredPosition); 1722 return false; 1723 } 1724 1725 // Try to parse the property accessor: 1726 // https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties 1727 Tokens->setPosition(StoredPosition); 1728 if (!IsTrivialPropertyAccessor && Style.BraceWrapping.AfterFunction == true) 1729 addUnwrappedLine(); 1730 nextToken(); 1731 do { 1732 switch (FormatTok->Tok.getKind()) { 1733 case tok::r_brace: 1734 nextToken(); 1735 if (FormatTok->is(tok::equal)) { 1736 while (!eof() && FormatTok->isNot(tok::semi)) 1737 nextToken(); 1738 nextToken(); 1739 } 1740 addUnwrappedLine(); 1741 return true; 1742 case tok::l_brace: 1743 ++Line->Level; 1744 parseBlock(/*MustBeDeclaration=*/true); 1745 addUnwrappedLine(); 1746 --Line->Level; 1747 break; 1748 case tok::equal: 1749 if (FormatTok->is(TT_FatArrow)) { 1750 ++Line->Level; 1751 do { 1752 nextToken(); 1753 } while (!eof() && FormatTok->isNot(tok::semi)); 1754 nextToken(); 1755 addUnwrappedLine(); 1756 --Line->Level; 1757 break; 1758 } 1759 nextToken(); 1760 break; 1761 default: 1762 if (FormatTok->isOneOf(Keywords.kw_get, Keywords.kw_set) && 1763 !IsTrivialPropertyAccessor) { 1764 // Non-trivial get/set needs to be on its own line. 1765 addUnwrappedLine(); 1766 } 1767 nextToken(); 1768 } 1769 } while (!eof()); 1770 1771 // Unreachable for well-formed code (paired '{' and '}'). 1772 return true; 1773 } 1774 1775 bool UnwrappedLineParser::tryToParseLambda() { 1776 if (!Style.isCpp()) { 1777 nextToken(); 1778 return false; 1779 } 1780 assert(FormatTok->is(tok::l_square)); 1781 FormatToken &LSquare = *FormatTok; 1782 if (!tryToParseLambdaIntroducer()) 1783 return false; 1784 1785 bool SeenArrow = false; 1786 1787 while (FormatTok->isNot(tok::l_brace)) { 1788 if (FormatTok->isSimpleTypeSpecifier()) { 1789 nextToken(); 1790 continue; 1791 } 1792 switch (FormatTok->Tok.getKind()) { 1793 case tok::l_brace: 1794 break; 1795 case tok::l_paren: 1796 parseParens(); 1797 break; 1798 case tok::amp: 1799 case tok::star: 1800 case tok::kw_const: 1801 case tok::comma: 1802 case tok::less: 1803 case tok::greater: 1804 case tok::identifier: 1805 case tok::numeric_constant: 1806 case tok::coloncolon: 1807 case tok::kw_class: 1808 case tok::kw_mutable: 1809 case tok::kw_noexcept: 1810 case tok::kw_template: 1811 case tok::kw_typename: 1812 nextToken(); 1813 break; 1814 // Specialization of a template with an integer parameter can contain 1815 // arithmetic, logical, comparison and ternary operators. 1816 // 1817 // FIXME: This also accepts sequences of operators that are not in the scope 1818 // of a template argument list. 1819 // 1820 // In a C++ lambda a template type can only occur after an arrow. We use 1821 // this as an heuristic to distinguish between Objective-C expressions 1822 // followed by an `a->b` expression, such as: 1823 // ([obj func:arg] + a->b) 1824 // Otherwise the code below would parse as a lambda. 1825 // 1826 // FIXME: This heuristic is incorrect for C++20 generic lambdas with 1827 // explicit template lists: []<bool b = true && false>(U &&u){} 1828 case tok::plus: 1829 case tok::minus: 1830 case tok::exclaim: 1831 case tok::tilde: 1832 case tok::slash: 1833 case tok::percent: 1834 case tok::lessless: 1835 case tok::pipe: 1836 case tok::pipepipe: 1837 case tok::ampamp: 1838 case tok::caret: 1839 case tok::equalequal: 1840 case tok::exclaimequal: 1841 case tok::greaterequal: 1842 case tok::lessequal: 1843 case tok::question: 1844 case tok::colon: 1845 case tok::ellipsis: 1846 case tok::kw_true: 1847 case tok::kw_false: 1848 if (SeenArrow) { 1849 nextToken(); 1850 break; 1851 } 1852 return true; 1853 case tok::arrow: 1854 // This might or might not actually be a lambda arrow (this could be an 1855 // ObjC method invocation followed by a dereferencing arrow). We might 1856 // reset this back to TT_Unknown in TokenAnnotator. 1857 FormatTok->setType(TT_LambdaArrow); 1858 SeenArrow = true; 1859 nextToken(); 1860 break; 1861 default: 1862 return true; 1863 } 1864 } 1865 FormatTok->setType(TT_LambdaLBrace); 1866 LSquare.setType(TT_LambdaLSquare); 1867 parseChildBlock(); 1868 return true; 1869 } 1870 1871 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 1872 const FormatToken *Previous = FormatTok->Previous; 1873 if (Previous && 1874 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 1875 tok::kw_delete, tok::l_square) || 1876 FormatTok->isCppStructuredBinding(Style) || Previous->closesScope() || 1877 Previous->isSimpleTypeSpecifier())) { 1878 nextToken(); 1879 return false; 1880 } 1881 nextToken(); 1882 if (FormatTok->is(tok::l_square)) { 1883 return false; 1884 } 1885 parseSquare(/*LambdaIntroducer=*/true); 1886 return true; 1887 } 1888 1889 void UnwrappedLineParser::tryToParseJSFunction() { 1890 assert(FormatTok->is(Keywords.kw_function) || 1891 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)); 1892 if (FormatTok->is(Keywords.kw_async)) 1893 nextToken(); 1894 // Consume "function". 1895 nextToken(); 1896 1897 // Consume * (generator function). Treat it like C++'s overloaded operators. 1898 if (FormatTok->is(tok::star)) { 1899 FormatTok->setType(TT_OverloadedOperator); 1900 nextToken(); 1901 } 1902 1903 // Consume function name. 1904 if (FormatTok->is(tok::identifier)) 1905 nextToken(); 1906 1907 if (FormatTok->isNot(tok::l_paren)) 1908 return; 1909 1910 // Parse formal parameter list. 1911 parseParens(); 1912 1913 if (FormatTok->is(tok::colon)) { 1914 // Parse a type definition. 1915 nextToken(); 1916 1917 // Eat the type declaration. For braced inline object types, balance braces, 1918 // otherwise just parse until finding an l_brace for the function body. 1919 if (FormatTok->is(tok::l_brace)) 1920 tryToParseBracedList(); 1921 else 1922 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 1923 nextToken(); 1924 } 1925 1926 if (FormatTok->is(tok::semi)) 1927 return; 1928 1929 parseChildBlock(); 1930 } 1931 1932 bool UnwrappedLineParser::tryToParseBracedList() { 1933 if (FormatTok->is(BK_Unknown)) 1934 calculateBraceTypes(); 1935 assert(FormatTok->isNot(BK_Unknown)); 1936 if (FormatTok->is(BK_Block)) 1937 return false; 1938 nextToken(); 1939 parseBracedList(); 1940 return true; 1941 } 1942 1943 bool UnwrappedLineParser::tryToParseCSharpLambda() { 1944 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType 1945 // TT_FatArrow. They always start an expression or a child block if 1946 // followed by a curly brace. 1947 nextToken(); 1948 if (FormatTok->isNot(tok::l_brace)) 1949 return false; 1950 // C# may break after => if the next character is a newline. 1951 if (Style.BraceWrapping.AfterFunction) { 1952 // calling `addUnwrappedLine()` here causes odd parsing errors. 1953 FormatTok->MustBreakBefore = true; 1954 } 1955 parseChildBlock(); 1956 return true; 1957 } 1958 1959 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons, 1960 bool IsEnum, 1961 tok::TokenKind ClosingBraceKind) { 1962 bool HasError = false; 1963 1964 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 1965 // replace this by using parseAssignmentExpression() inside. 1966 do { 1967 if (Style.isCSharp() && FormatTok->is(TT_FatArrow)) 1968 if (tryToParseCSharpLambda()) 1969 continue; 1970 if (Style.isJavaScript()) { 1971 if (FormatTok->is(Keywords.kw_function) || 1972 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) { 1973 tryToParseJSFunction(); 1974 continue; 1975 } 1976 if (FormatTok->is(TT_FatArrow)) { 1977 nextToken(); 1978 // Fat arrows can be followed by simple expressions or by child blocks 1979 // in curly braces. 1980 if (FormatTok->is(tok::l_brace)) { 1981 parseChildBlock(); 1982 continue; 1983 } 1984 } 1985 if (FormatTok->is(tok::l_brace)) { 1986 // Could be a method inside of a braced list `{a() { return 1; }}`. 1987 if (tryToParseBracedList()) 1988 continue; 1989 parseChildBlock(); 1990 } 1991 } 1992 if (FormatTok->Tok.getKind() == ClosingBraceKind) { 1993 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 1994 addUnwrappedLine(); 1995 nextToken(); 1996 return !HasError; 1997 } 1998 switch (FormatTok->Tok.getKind()) { 1999 case tok::caret: 2000 nextToken(); 2001 if (FormatTok->is(tok::l_brace)) { 2002 parseChildBlock(); 2003 } 2004 break; 2005 case tok::l_square: 2006 if (Style.isCSharp()) 2007 parseSquare(); 2008 else 2009 tryToParseLambda(); 2010 break; 2011 case tok::l_paren: 2012 parseParens(); 2013 // JavaScript can just have free standing methods and getters/setters in 2014 // object literals. Detect them by a "{" following ")". 2015 if (Style.isJavaScript()) { 2016 if (FormatTok->is(tok::l_brace)) 2017 parseChildBlock(); 2018 break; 2019 } 2020 break; 2021 case tok::l_brace: 2022 // Assume there are no blocks inside a braced init list apart 2023 // from the ones we explicitly parse out (like lambdas). 2024 FormatTok->setBlockKind(BK_BracedInit); 2025 nextToken(); 2026 parseBracedList(); 2027 break; 2028 case tok::less: 2029 if (Style.Language == FormatStyle::LK_Proto) { 2030 nextToken(); 2031 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 2032 /*ClosingBraceKind=*/tok::greater); 2033 } else { 2034 nextToken(); 2035 } 2036 break; 2037 case tok::semi: 2038 // JavaScript (or more precisely TypeScript) can have semicolons in braced 2039 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 2040 // used for error recovery if we have otherwise determined that this is 2041 // a braced list. 2042 if (Style.isJavaScript()) { 2043 nextToken(); 2044 break; 2045 } 2046 HasError = true; 2047 if (!ContinueOnSemicolons) 2048 return !HasError; 2049 nextToken(); 2050 break; 2051 case tok::comma: 2052 nextToken(); 2053 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 2054 addUnwrappedLine(); 2055 break; 2056 default: 2057 nextToken(); 2058 break; 2059 } 2060 } while (!eof()); 2061 return false; 2062 } 2063 2064 void UnwrappedLineParser::parseParens() { 2065 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 2066 nextToken(); 2067 do { 2068 switch (FormatTok->Tok.getKind()) { 2069 case tok::l_paren: 2070 parseParens(); 2071 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 2072 parseChildBlock(); 2073 break; 2074 case tok::r_paren: 2075 nextToken(); 2076 return; 2077 case tok::r_brace: 2078 // A "}" inside parenthesis is an error if there wasn't a matching "{". 2079 return; 2080 case tok::l_square: 2081 tryToParseLambda(); 2082 break; 2083 case tok::l_brace: 2084 if (!tryToParseBracedList()) 2085 parseChildBlock(); 2086 break; 2087 case tok::at: 2088 nextToken(); 2089 if (FormatTok->Tok.is(tok::l_brace)) { 2090 nextToken(); 2091 parseBracedList(); 2092 } 2093 break; 2094 case tok::equal: 2095 if (Style.isCSharp() && FormatTok->is(TT_FatArrow)) 2096 tryToParseCSharpLambda(); 2097 else 2098 nextToken(); 2099 break; 2100 case tok::kw_class: 2101 if (Style.isJavaScript()) 2102 parseRecord(/*ParseAsExpr=*/true); 2103 else 2104 nextToken(); 2105 break; 2106 case tok::identifier: 2107 if (Style.isJavaScript() && 2108 (FormatTok->is(Keywords.kw_function) || 2109 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function))) 2110 tryToParseJSFunction(); 2111 else 2112 nextToken(); 2113 break; 2114 default: 2115 nextToken(); 2116 break; 2117 } 2118 } while (!eof()); 2119 } 2120 2121 void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) { 2122 if (!LambdaIntroducer) { 2123 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected."); 2124 if (tryToParseLambda()) 2125 return; 2126 } 2127 do { 2128 switch (FormatTok->Tok.getKind()) { 2129 case tok::l_paren: 2130 parseParens(); 2131 break; 2132 case tok::r_square: 2133 nextToken(); 2134 return; 2135 case tok::r_brace: 2136 // A "}" inside parenthesis is an error if there wasn't a matching "{". 2137 return; 2138 case tok::l_square: 2139 parseSquare(); 2140 break; 2141 case tok::l_brace: { 2142 if (!tryToParseBracedList()) 2143 parseChildBlock(); 2144 break; 2145 } 2146 case tok::at: 2147 nextToken(); 2148 if (FormatTok->Tok.is(tok::l_brace)) { 2149 nextToken(); 2150 parseBracedList(); 2151 } 2152 break; 2153 default: 2154 nextToken(); 2155 break; 2156 } 2157 } while (!eof()); 2158 } 2159 2160 void UnwrappedLineParser::parseIfThenElse() { 2161 auto HandleAttributes = [this]() { 2162 // Handle AttributeMacro, e.g. `if (x) UNLIKELY`. 2163 if (FormatTok->is(TT_AttributeMacro)) 2164 nextToken(); 2165 // Handle [[likely]] / [[unlikely]] attributes. 2166 if (FormatTok->is(tok::l_square) && tryToParseSimpleAttribute()) 2167 parseSquare(); 2168 }; 2169 2170 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 2171 nextToken(); 2172 if (FormatTok->Tok.isOneOf(tok::kw_constexpr, tok::identifier)) 2173 nextToken(); 2174 if (FormatTok->Tok.is(tok::l_paren)) 2175 parseParens(); 2176 HandleAttributes(); 2177 bool NeedsUnwrappedLine = false; 2178 if (FormatTok->Tok.is(tok::l_brace)) { 2179 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2180 parseBlock(); 2181 if (Style.BraceWrapping.BeforeElse) 2182 addUnwrappedLine(); 2183 else 2184 NeedsUnwrappedLine = true; 2185 } else { 2186 addUnwrappedLine(); 2187 ++Line->Level; 2188 parseStructuralElement(); 2189 --Line->Level; 2190 } 2191 if (FormatTok->Tok.is(tok::kw_else)) { 2192 nextToken(); 2193 HandleAttributes(); 2194 if (FormatTok->Tok.is(tok::l_brace)) { 2195 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2196 parseBlock(); 2197 addUnwrappedLine(); 2198 } else if (FormatTok->Tok.is(tok::kw_if)) { 2199 FormatToken *Previous = Tokens->getPreviousToken(); 2200 bool PrecededByComment = Previous && Previous->is(tok::comment); 2201 if (PrecededByComment) { 2202 addUnwrappedLine(); 2203 ++Line->Level; 2204 } 2205 parseIfThenElse(); 2206 if (PrecededByComment) 2207 --Line->Level; 2208 } else { 2209 addUnwrappedLine(); 2210 ++Line->Level; 2211 parseStructuralElement(); 2212 if (FormatTok->is(tok::eof)) 2213 addUnwrappedLine(); 2214 --Line->Level; 2215 } 2216 } else if (NeedsUnwrappedLine) { 2217 addUnwrappedLine(); 2218 } 2219 } 2220 2221 void UnwrappedLineParser::parseTryCatch() { 2222 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 2223 nextToken(); 2224 bool NeedsUnwrappedLine = false; 2225 if (FormatTok->is(tok::colon)) { 2226 // We are in a function try block, what comes is an initializer list. 2227 nextToken(); 2228 2229 // In case identifiers were removed by clang-tidy, what might follow is 2230 // multiple commas in sequence - before the first identifier. 2231 while (FormatTok->is(tok::comma)) 2232 nextToken(); 2233 2234 while (FormatTok->is(tok::identifier)) { 2235 nextToken(); 2236 if (FormatTok->is(tok::l_paren)) 2237 parseParens(); 2238 if (FormatTok->Previous && FormatTok->Previous->is(tok::identifier) && 2239 FormatTok->is(tok::l_brace)) { 2240 do { 2241 nextToken(); 2242 } while (!FormatTok->is(tok::r_brace)); 2243 nextToken(); 2244 } 2245 2246 // In case identifiers were removed by clang-tidy, what might follow is 2247 // multiple commas in sequence - after the first identifier. 2248 while (FormatTok->is(tok::comma)) 2249 nextToken(); 2250 } 2251 } 2252 // Parse try with resource. 2253 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) { 2254 parseParens(); 2255 } 2256 if (FormatTok->is(tok::l_brace)) { 2257 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2258 parseBlock(); 2259 if (Style.BraceWrapping.BeforeCatch) { 2260 addUnwrappedLine(); 2261 } else { 2262 NeedsUnwrappedLine = true; 2263 } 2264 } else if (!FormatTok->is(tok::kw_catch)) { 2265 // The C++ standard requires a compound-statement after a try. 2266 // If there's none, we try to assume there's a structuralElement 2267 // and try to continue. 2268 addUnwrappedLine(); 2269 ++Line->Level; 2270 parseStructuralElement(); 2271 --Line->Level; 2272 } 2273 while (1) { 2274 if (FormatTok->is(tok::at)) 2275 nextToken(); 2276 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 2277 tok::kw___finally) || 2278 ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2279 FormatTok->is(Keywords.kw_finally)) || 2280 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) || 2281 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally)))) 2282 break; 2283 nextToken(); 2284 while (FormatTok->isNot(tok::l_brace)) { 2285 if (FormatTok->is(tok::l_paren)) { 2286 parseParens(); 2287 continue; 2288 } 2289 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) 2290 return; 2291 nextToken(); 2292 } 2293 NeedsUnwrappedLine = false; 2294 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2295 parseBlock(); 2296 if (Style.BraceWrapping.BeforeCatch) 2297 addUnwrappedLine(); 2298 else 2299 NeedsUnwrappedLine = true; 2300 } 2301 if (NeedsUnwrappedLine) 2302 addUnwrappedLine(); 2303 } 2304 2305 void UnwrappedLineParser::parseNamespace() { 2306 assert(FormatTok->isOneOf(tok::kw_namespace, TT_NamespaceMacro) && 2307 "'namespace' expected"); 2308 2309 const FormatToken &InitialToken = *FormatTok; 2310 nextToken(); 2311 if (InitialToken.is(TT_NamespaceMacro)) { 2312 parseParens(); 2313 } else { 2314 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::kw_inline, 2315 tok::l_square, tok::period)) { 2316 if (FormatTok->is(tok::l_square)) 2317 parseSquare(); 2318 else 2319 nextToken(); 2320 } 2321 } 2322 if (FormatTok->Tok.is(tok::l_brace)) { 2323 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2324 addUnwrappedLine(); 2325 2326 unsigned AddLevels = 2327 Style.NamespaceIndentation == FormatStyle::NI_All || 2328 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 2329 DeclarationScopeStack.size() > 1) 2330 ? 1u 2331 : 0u; 2332 bool ManageWhitesmithsBraces = 2333 AddLevels == 0u && 2334 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths; 2335 2336 // If we're in Whitesmiths mode, indent the brace if we're not indenting 2337 // the whole block. 2338 if (ManageWhitesmithsBraces) 2339 ++Line->Level; 2340 2341 parseBlock(/*MustBeDeclaration=*/true, AddLevels, 2342 /*MunchSemi=*/true, 2343 /*UnindentWhitesmithsBraces=*/ManageWhitesmithsBraces); 2344 2345 // Munch the semicolon after a namespace. This is more common than one would 2346 // think. Putting the semicolon into its own line is very ugly. 2347 if (FormatTok->Tok.is(tok::semi)) 2348 nextToken(); 2349 2350 addUnwrappedLine(AddLevels > 0 ? LineLevel::Remove : LineLevel::Keep); 2351 2352 if (ManageWhitesmithsBraces) 2353 --Line->Level; 2354 } 2355 // FIXME: Add error handling. 2356 } 2357 2358 void UnwrappedLineParser::parseNew() { 2359 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 2360 nextToken(); 2361 2362 if (Style.isCSharp()) { 2363 do { 2364 if (FormatTok->is(tok::l_brace)) 2365 parseBracedList(); 2366 2367 if (FormatTok->isOneOf(tok::semi, tok::comma)) 2368 return; 2369 2370 nextToken(); 2371 } while (!eof()); 2372 } 2373 2374 if (Style.Language != FormatStyle::LK_Java) 2375 return; 2376 2377 // In Java, we can parse everything up to the parens, which aren't optional. 2378 do { 2379 // There should not be a ;, { or } before the new's open paren. 2380 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 2381 return; 2382 2383 // Consume the parens. 2384 if (FormatTok->is(tok::l_paren)) { 2385 parseParens(); 2386 2387 // If there is a class body of an anonymous class, consume that as child. 2388 if (FormatTok->is(tok::l_brace)) 2389 parseChildBlock(); 2390 return; 2391 } 2392 nextToken(); 2393 } while (!eof()); 2394 } 2395 2396 void UnwrappedLineParser::parseForOrWhileLoop() { 2397 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 2398 "'for', 'while' or foreach macro expected"); 2399 nextToken(); 2400 // JS' for await ( ... 2401 if (Style.isJavaScript() && FormatTok->is(Keywords.kw_await)) 2402 nextToken(); 2403 if (Style.isCpp() && FormatTok->is(tok::kw_co_await)) 2404 nextToken(); 2405 if (FormatTok->Tok.is(tok::l_paren)) 2406 parseParens(); 2407 if (FormatTok->Tok.is(tok::l_brace)) { 2408 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2409 parseBlock(); 2410 addUnwrappedLine(); 2411 } else { 2412 addUnwrappedLine(); 2413 ++Line->Level; 2414 parseStructuralElement(); 2415 --Line->Level; 2416 } 2417 } 2418 2419 void UnwrappedLineParser::parseDoWhile() { 2420 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 2421 nextToken(); 2422 if (FormatTok->Tok.is(tok::l_brace)) { 2423 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2424 parseBlock(); 2425 if (Style.BraceWrapping.BeforeWhile) 2426 addUnwrappedLine(); 2427 } else { 2428 addUnwrappedLine(); 2429 ++Line->Level; 2430 parseStructuralElement(); 2431 --Line->Level; 2432 } 2433 2434 // FIXME: Add error handling. 2435 if (!FormatTok->Tok.is(tok::kw_while)) { 2436 addUnwrappedLine(); 2437 return; 2438 } 2439 2440 // If in Whitesmiths mode, the line with the while() needs to be indented 2441 // to the same level as the block. 2442 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) 2443 ++Line->Level; 2444 2445 nextToken(); 2446 parseStructuralElement(); 2447 } 2448 2449 void UnwrappedLineParser::parseLabel(bool LeftAlignLabel) { 2450 nextToken(); 2451 unsigned OldLineLevel = Line->Level; 2452 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 2453 --Line->Level; 2454 if (LeftAlignLabel) 2455 Line->Level = 0; 2456 2457 if (!Style.IndentCaseBlocks && CommentsBeforeNextToken.empty() && 2458 FormatTok->Tok.is(tok::l_brace)) { 2459 2460 CompoundStatementIndenter Indenter(this, Line->Level, 2461 Style.BraceWrapping.AfterCaseLabel, 2462 Style.BraceWrapping.IndentBraces); 2463 parseBlock(); 2464 if (FormatTok->Tok.is(tok::kw_break)) { 2465 if (Style.BraceWrapping.AfterControlStatement == 2466 FormatStyle::BWACS_Always) { 2467 addUnwrappedLine(); 2468 if (!Style.IndentCaseBlocks && 2469 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) { 2470 Line->Level++; 2471 } 2472 } 2473 parseStructuralElement(); 2474 } 2475 addUnwrappedLine(); 2476 } else { 2477 if (FormatTok->is(tok::semi)) 2478 nextToken(); 2479 addUnwrappedLine(); 2480 } 2481 Line->Level = OldLineLevel; 2482 if (FormatTok->isNot(tok::l_brace)) { 2483 parseStructuralElement(); 2484 addUnwrappedLine(); 2485 } 2486 } 2487 2488 void UnwrappedLineParser::parseCaseLabel() { 2489 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 2490 2491 // FIXME: fix handling of complex expressions here. 2492 do { 2493 nextToken(); 2494 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 2495 parseLabel(); 2496 } 2497 2498 void UnwrappedLineParser::parseSwitch() { 2499 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 2500 nextToken(); 2501 if (FormatTok->Tok.is(tok::l_paren)) 2502 parseParens(); 2503 if (FormatTok->Tok.is(tok::l_brace)) { 2504 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2505 parseBlock(); 2506 addUnwrappedLine(); 2507 } else { 2508 addUnwrappedLine(); 2509 ++Line->Level; 2510 parseStructuralElement(); 2511 --Line->Level; 2512 } 2513 } 2514 2515 void UnwrappedLineParser::parseAccessSpecifier() { 2516 nextToken(); 2517 // Understand Qt's slots. 2518 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 2519 nextToken(); 2520 // Otherwise, we don't know what it is, and we'd better keep the next token. 2521 if (FormatTok->Tok.is(tok::colon)) 2522 nextToken(); 2523 addUnwrappedLine(); 2524 } 2525 2526 void UnwrappedLineParser::parseConcept() { 2527 assert(FormatTok->Tok.is(tok::kw_concept) && "'concept' expected"); 2528 nextToken(); 2529 if (!FormatTok->Tok.is(tok::identifier)) 2530 return; 2531 nextToken(); 2532 if (!FormatTok->Tok.is(tok::equal)) 2533 return; 2534 nextToken(); 2535 if (FormatTok->Tok.is(tok::kw_requires)) { 2536 nextToken(); 2537 parseRequiresExpression(Line->Level); 2538 } else { 2539 parseConstraintExpression(Line->Level); 2540 } 2541 } 2542 2543 void UnwrappedLineParser::parseRequiresExpression(unsigned int OriginalLevel) { 2544 // requires (R range) 2545 if (FormatTok->Tok.is(tok::l_paren)) { 2546 parseParens(); 2547 if (Style.IndentRequires && OriginalLevel != Line->Level) { 2548 addUnwrappedLine(); 2549 --Line->Level; 2550 } 2551 } 2552 2553 if (FormatTok->Tok.is(tok::l_brace)) { 2554 if (Style.BraceWrapping.AfterFunction) 2555 addUnwrappedLine(); 2556 FormatTok->setType(TT_FunctionLBrace); 2557 parseBlock(); 2558 addUnwrappedLine(); 2559 } else { 2560 parseConstraintExpression(OriginalLevel); 2561 } 2562 } 2563 2564 void UnwrappedLineParser::parseConstraintExpression( 2565 unsigned int OriginalLevel) { 2566 // requires Id<T> && Id<T> || Id<T> 2567 while ( 2568 FormatTok->isOneOf(tok::identifier, tok::kw_requires, tok::coloncolon)) { 2569 nextToken(); 2570 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::less, 2571 tok::greater, tok::comma, tok::ellipsis)) { 2572 if (FormatTok->Tok.is(tok::less)) { 2573 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 2574 /*ClosingBraceKind=*/tok::greater); 2575 continue; 2576 } 2577 nextToken(); 2578 } 2579 if (FormatTok->Tok.is(tok::kw_requires)) { 2580 parseRequiresExpression(OriginalLevel); 2581 } 2582 if (FormatTok->Tok.is(tok::less)) { 2583 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 2584 /*ClosingBraceKind=*/tok::greater); 2585 } 2586 2587 if (FormatTok->Tok.is(tok::l_paren)) { 2588 parseParens(); 2589 } 2590 if (FormatTok->Tok.is(tok::l_brace)) { 2591 if (Style.BraceWrapping.AfterFunction) 2592 addUnwrappedLine(); 2593 FormatTok->setType(TT_FunctionLBrace); 2594 parseBlock(); 2595 } 2596 if (FormatTok->Tok.is(tok::semi)) { 2597 // Eat any trailing semi. 2598 nextToken(); 2599 addUnwrappedLine(); 2600 } 2601 if (FormatTok->Tok.is(tok::colon)) { 2602 return; 2603 } 2604 if (!FormatTok->Tok.isOneOf(tok::ampamp, tok::pipepipe)) { 2605 if (FormatTok->Previous && 2606 !FormatTok->Previous->isOneOf(tok::identifier, tok::kw_requires, 2607 tok::coloncolon)) { 2608 addUnwrappedLine(); 2609 } 2610 if (Style.IndentRequires && OriginalLevel != Line->Level) { 2611 --Line->Level; 2612 } 2613 break; 2614 } else { 2615 FormatTok->setType(TT_ConstraintJunctions); 2616 } 2617 2618 nextToken(); 2619 } 2620 } 2621 2622 void UnwrappedLineParser::parseRequires() { 2623 assert(FormatTok->Tok.is(tok::kw_requires) && "'requires' expected"); 2624 2625 unsigned OriginalLevel = Line->Level; 2626 if (FormatTok->Previous && FormatTok->Previous->is(tok::greater)) { 2627 addUnwrappedLine(); 2628 if (Style.IndentRequires) { 2629 Line->Level++; 2630 } 2631 } 2632 nextToken(); 2633 2634 parseRequiresExpression(OriginalLevel); 2635 } 2636 2637 bool UnwrappedLineParser::parseEnum() { 2638 // Won't be 'enum' for NS_ENUMs. 2639 if (FormatTok->Tok.is(tok::kw_enum)) 2640 nextToken(); 2641 2642 const FormatToken &InitialToken = *FormatTok; 2643 2644 // In TypeScript, "enum" can also be used as property name, e.g. in interface 2645 // declarations. An "enum" keyword followed by a colon would be a syntax 2646 // error and thus assume it is just an identifier. 2647 if (Style.isJavaScript() && FormatTok->isOneOf(tok::colon, tok::question)) 2648 return false; 2649 2650 // In protobuf, "enum" can be used as a field name. 2651 if (Style.Language == FormatStyle::LK_Proto && FormatTok->is(tok::equal)) 2652 return false; 2653 2654 // Eat up enum class ... 2655 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct)) 2656 nextToken(); 2657 2658 while (FormatTok->Tok.getIdentifierInfo() || 2659 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 2660 tok::greater, tok::comma, tok::question)) { 2661 nextToken(); 2662 // We can have macros or attributes in between 'enum' and the enum name. 2663 if (FormatTok->is(tok::l_paren)) 2664 parseParens(); 2665 if (FormatTok->is(tok::identifier)) { 2666 nextToken(); 2667 // If there are two identifiers in a row, this is likely an elaborate 2668 // return type. In Java, this can be "implements", etc. 2669 if (Style.isCpp() && FormatTok->is(tok::identifier)) 2670 return false; 2671 } 2672 } 2673 2674 // Just a declaration or something is wrong. 2675 if (FormatTok->isNot(tok::l_brace)) 2676 return true; 2677 FormatTok->setBlockKind(BK_Block); 2678 2679 if (Style.Language == FormatStyle::LK_Java) { 2680 // Java enums are different. 2681 parseJavaEnumBody(); 2682 return true; 2683 } 2684 if (Style.Language == FormatStyle::LK_Proto) { 2685 parseBlock(/*MustBeDeclaration=*/true); 2686 return true; 2687 } 2688 2689 if (!Style.AllowShortEnumsOnASingleLine && 2690 ShouldBreakBeforeBrace(Style, InitialToken)) 2691 addUnwrappedLine(); 2692 // Parse enum body. 2693 nextToken(); 2694 if (!Style.AllowShortEnumsOnASingleLine) { 2695 addUnwrappedLine(); 2696 Line->Level += 1; 2697 } 2698 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true, 2699 /*IsEnum=*/true); 2700 if (!Style.AllowShortEnumsOnASingleLine) 2701 Line->Level -= 1; 2702 if (HasError) { 2703 if (FormatTok->is(tok::semi)) 2704 nextToken(); 2705 addUnwrappedLine(); 2706 } 2707 return true; 2708 2709 // There is no addUnwrappedLine() here so that we fall through to parsing a 2710 // structural element afterwards. Thus, in "enum A {} n, m;", 2711 // "} n, m;" will end up in one unwrapped line. 2712 } 2713 2714 bool UnwrappedLineParser::parseStructLike() { 2715 // parseRecord falls through and does not yet add an unwrapped line as a 2716 // record declaration or definition can start a structural element. 2717 parseRecord(); 2718 // This does not apply to Java, JavaScript and C#. 2719 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 2720 Style.isCSharp()) { 2721 if (FormatTok->is(tok::semi)) 2722 nextToken(); 2723 addUnwrappedLine(); 2724 return true; 2725 } 2726 return false; 2727 } 2728 2729 namespace { 2730 // A class used to set and restore the Token position when peeking 2731 // ahead in the token source. 2732 class ScopedTokenPosition { 2733 unsigned StoredPosition; 2734 FormatTokenSource *Tokens; 2735 2736 public: 2737 ScopedTokenPosition(FormatTokenSource *Tokens) : Tokens(Tokens) { 2738 assert(Tokens && "Tokens expected to not be null"); 2739 StoredPosition = Tokens->getPosition(); 2740 } 2741 2742 ~ScopedTokenPosition() { Tokens->setPosition(StoredPosition); } 2743 }; 2744 } // namespace 2745 2746 // Look to see if we have [[ by looking ahead, if 2747 // its not then rewind to the original position. 2748 bool UnwrappedLineParser::tryToParseSimpleAttribute() { 2749 ScopedTokenPosition AutoPosition(Tokens); 2750 FormatToken *Tok = Tokens->getNextToken(); 2751 // We already read the first [ check for the second. 2752 if (!Tok->is(tok::l_square)) { 2753 return false; 2754 } 2755 // Double check that the attribute is just something 2756 // fairly simple. 2757 while (Tok->isNot(tok::eof)) { 2758 if (Tok->is(tok::r_square)) { 2759 break; 2760 } 2761 Tok = Tokens->getNextToken(); 2762 } 2763 if (Tok->is(tok::eof)) 2764 return false; 2765 Tok = Tokens->getNextToken(); 2766 if (!Tok->is(tok::r_square)) { 2767 return false; 2768 } 2769 Tok = Tokens->getNextToken(); 2770 if (Tok->is(tok::semi)) { 2771 return false; 2772 } 2773 return true; 2774 } 2775 2776 void UnwrappedLineParser::parseJavaEnumBody() { 2777 // Determine whether the enum is simple, i.e. does not have a semicolon or 2778 // constants with class bodies. Simple enums can be formatted like braced 2779 // lists, contracted to a single line, etc. 2780 unsigned StoredPosition = Tokens->getPosition(); 2781 bool IsSimple = true; 2782 FormatToken *Tok = Tokens->getNextToken(); 2783 while (!Tok->is(tok::eof)) { 2784 if (Tok->is(tok::r_brace)) 2785 break; 2786 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 2787 IsSimple = false; 2788 break; 2789 } 2790 // FIXME: This will also mark enums with braces in the arguments to enum 2791 // constants as "not simple". This is probably fine in practice, though. 2792 Tok = Tokens->getNextToken(); 2793 } 2794 FormatTok = Tokens->setPosition(StoredPosition); 2795 2796 if (IsSimple) { 2797 nextToken(); 2798 parseBracedList(); 2799 addUnwrappedLine(); 2800 return; 2801 } 2802 2803 // Parse the body of a more complex enum. 2804 // First add a line for everything up to the "{". 2805 nextToken(); 2806 addUnwrappedLine(); 2807 ++Line->Level; 2808 2809 // Parse the enum constants. 2810 while (FormatTok) { 2811 if (FormatTok->is(tok::l_brace)) { 2812 // Parse the constant's class body. 2813 parseBlock(/*MustBeDeclaration=*/true, /*AddLevels=*/1u, 2814 /*MunchSemi=*/false); 2815 } else if (FormatTok->is(tok::l_paren)) { 2816 parseParens(); 2817 } else if (FormatTok->is(tok::comma)) { 2818 nextToken(); 2819 addUnwrappedLine(); 2820 } else if (FormatTok->is(tok::semi)) { 2821 nextToken(); 2822 addUnwrappedLine(); 2823 break; 2824 } else if (FormatTok->is(tok::r_brace)) { 2825 addUnwrappedLine(); 2826 break; 2827 } else { 2828 nextToken(); 2829 } 2830 } 2831 2832 // Parse the class body after the enum's ";" if any. 2833 parseLevel(/*HasOpeningBrace=*/true); 2834 nextToken(); 2835 --Line->Level; 2836 addUnwrappedLine(); 2837 } 2838 2839 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) { 2840 const FormatToken &InitialToken = *FormatTok; 2841 nextToken(); 2842 2843 // The actual identifier can be a nested name specifier, and in macros 2844 // it is often token-pasted. 2845 // An [[attribute]] can be before the identifier. 2846 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 2847 tok::kw___attribute, tok::kw___declspec, 2848 tok::kw_alignas, tok::l_square, tok::r_square) || 2849 ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2850 FormatTok->isOneOf(tok::period, tok::comma))) { 2851 if (Style.isJavaScript() && 2852 FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) { 2853 // JavaScript/TypeScript supports inline object types in 2854 // extends/implements positions: 2855 // class Foo implements {bar: number} { } 2856 nextToken(); 2857 if (FormatTok->is(tok::l_brace)) { 2858 tryToParseBracedList(); 2859 continue; 2860 } 2861 } 2862 bool IsNonMacroIdentifier = 2863 FormatTok->is(tok::identifier) && 2864 FormatTok->TokenText != FormatTok->TokenText.upper(); 2865 nextToken(); 2866 // We can have macros or attributes in between 'class' and the class name. 2867 if (!IsNonMacroIdentifier) { 2868 if (FormatTok->Tok.is(tok::l_paren)) { 2869 parseParens(); 2870 } else if (FormatTok->is(TT_AttributeSquare)) { 2871 parseSquare(); 2872 // Consume the closing TT_AttributeSquare. 2873 if (FormatTok->Next && FormatTok->is(TT_AttributeSquare)) 2874 nextToken(); 2875 } 2876 } 2877 } 2878 2879 // Note that parsing away template declarations here leads to incorrectly 2880 // accepting function declarations as record declarations. 2881 // In general, we cannot solve this problem. Consider: 2882 // class A<int> B() {} 2883 // which can be a function definition or a class definition when B() is a 2884 // macro. If we find enough real-world cases where this is a problem, we 2885 // can parse for the 'template' keyword in the beginning of the statement, 2886 // and thus rule out the record production in case there is no template 2887 // (this would still leave us with an ambiguity between template function 2888 // and class declarations). 2889 if (FormatTok->isOneOf(tok::colon, tok::less)) { 2890 while (!eof()) { 2891 if (FormatTok->is(tok::l_brace)) { 2892 calculateBraceTypes(/*ExpectClassBody=*/true); 2893 if (!tryToParseBracedList()) 2894 break; 2895 } 2896 if (FormatTok->Tok.is(tok::semi)) 2897 return; 2898 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where)) { 2899 addUnwrappedLine(); 2900 nextToken(); 2901 parseCSharpGenericTypeConstraint(); 2902 break; 2903 } 2904 nextToken(); 2905 } 2906 } 2907 if (FormatTok->Tok.is(tok::l_brace)) { 2908 if (ParseAsExpr) { 2909 parseChildBlock(); 2910 } else { 2911 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2912 addUnwrappedLine(); 2913 2914 unsigned AddLevels = Style.IndentAccessModifiers ? 2u : 1u; 2915 parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/false); 2916 } 2917 } 2918 // There is no addUnwrappedLine() here so that we fall through to parsing a 2919 // structural element afterwards. Thus, in "class A {} n, m;", 2920 // "} n, m;" will end up in one unwrapped line. 2921 } 2922 2923 void UnwrappedLineParser::parseObjCMethod() { 2924 assert(FormatTok->Tok.isOneOf(tok::l_paren, tok::identifier) && 2925 "'(' or identifier expected."); 2926 do { 2927 if (FormatTok->Tok.is(tok::semi)) { 2928 nextToken(); 2929 addUnwrappedLine(); 2930 return; 2931 } else if (FormatTok->Tok.is(tok::l_brace)) { 2932 if (Style.BraceWrapping.AfterFunction) 2933 addUnwrappedLine(); 2934 parseBlock(); 2935 addUnwrappedLine(); 2936 return; 2937 } else { 2938 nextToken(); 2939 } 2940 } while (!eof()); 2941 } 2942 2943 void UnwrappedLineParser::parseObjCProtocolList() { 2944 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 2945 do { 2946 nextToken(); 2947 // Early exit in case someone forgot a close angle. 2948 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 2949 FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) 2950 return; 2951 } while (!eof() && FormatTok->Tok.isNot(tok::greater)); 2952 nextToken(); // Skip '>'. 2953 } 2954 2955 void UnwrappedLineParser::parseObjCUntilAtEnd() { 2956 do { 2957 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 2958 nextToken(); 2959 addUnwrappedLine(); 2960 break; 2961 } 2962 if (FormatTok->is(tok::l_brace)) { 2963 parseBlock(); 2964 // In ObjC interfaces, nothing should be following the "}". 2965 addUnwrappedLine(); 2966 } else if (FormatTok->is(tok::r_brace)) { 2967 // Ignore stray "}". parseStructuralElement doesn't consume them. 2968 nextToken(); 2969 addUnwrappedLine(); 2970 } else if (FormatTok->isOneOf(tok::minus, tok::plus)) { 2971 nextToken(); 2972 parseObjCMethod(); 2973 } else { 2974 parseStructuralElement(); 2975 } 2976 } while (!eof()); 2977 } 2978 2979 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 2980 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_interface || 2981 FormatTok->Tok.getObjCKeywordID() == tok::objc_implementation); 2982 nextToken(); 2983 nextToken(); // interface name 2984 2985 // @interface can be followed by a lightweight generic 2986 // specialization list, then either a base class or a category. 2987 if (FormatTok->Tok.is(tok::less)) { 2988 parseObjCLightweightGenerics(); 2989 } 2990 if (FormatTok->Tok.is(tok::colon)) { 2991 nextToken(); 2992 nextToken(); // base class name 2993 // The base class can also have lightweight generics applied to it. 2994 if (FormatTok->Tok.is(tok::less)) { 2995 parseObjCLightweightGenerics(); 2996 } 2997 } else if (FormatTok->Tok.is(tok::l_paren)) 2998 // Skip category, if present. 2999 parseParens(); 3000 3001 if (FormatTok->Tok.is(tok::less)) 3002 parseObjCProtocolList(); 3003 3004 if (FormatTok->Tok.is(tok::l_brace)) { 3005 if (Style.BraceWrapping.AfterObjCDeclaration) 3006 addUnwrappedLine(); 3007 parseBlock(/*MustBeDeclaration=*/true); 3008 } 3009 3010 // With instance variables, this puts '}' on its own line. Without instance 3011 // variables, this ends the @interface line. 3012 addUnwrappedLine(); 3013 3014 parseObjCUntilAtEnd(); 3015 } 3016 3017 void UnwrappedLineParser::parseObjCLightweightGenerics() { 3018 assert(FormatTok->Tok.is(tok::less)); 3019 // Unlike protocol lists, generic parameterizations support 3020 // nested angles: 3021 // 3022 // @interface Foo<ValueType : id <NSCopying, NSSecureCoding>> : 3023 // NSObject <NSCopying, NSSecureCoding> 3024 // 3025 // so we need to count how many open angles we have left. 3026 unsigned NumOpenAngles = 1; 3027 do { 3028 nextToken(); 3029 // Early exit in case someone forgot a close angle. 3030 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 3031 FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) 3032 break; 3033 if (FormatTok->Tok.is(tok::less)) 3034 ++NumOpenAngles; 3035 else if (FormatTok->Tok.is(tok::greater)) { 3036 assert(NumOpenAngles > 0 && "'>' makes NumOpenAngles negative"); 3037 --NumOpenAngles; 3038 } 3039 } while (!eof() && NumOpenAngles != 0); 3040 nextToken(); // Skip '>'. 3041 } 3042 3043 // Returns true for the declaration/definition form of @protocol, 3044 // false for the expression form. 3045 bool UnwrappedLineParser::parseObjCProtocol() { 3046 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_protocol); 3047 nextToken(); 3048 3049 if (FormatTok->is(tok::l_paren)) 3050 // The expression form of @protocol, e.g. "Protocol* p = @protocol(foo);". 3051 return false; 3052 3053 // The definition/declaration form, 3054 // @protocol Foo 3055 // - (int)someMethod; 3056 // @end 3057 3058 nextToken(); // protocol name 3059 3060 if (FormatTok->Tok.is(tok::less)) 3061 parseObjCProtocolList(); 3062 3063 // Check for protocol declaration. 3064 if (FormatTok->Tok.is(tok::semi)) { 3065 nextToken(); 3066 addUnwrappedLine(); 3067 return true; 3068 } 3069 3070 addUnwrappedLine(); 3071 parseObjCUntilAtEnd(); 3072 return true; 3073 } 3074 3075 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 3076 bool IsImport = FormatTok->is(Keywords.kw_import); 3077 assert(IsImport || FormatTok->is(tok::kw_export)); 3078 nextToken(); 3079 3080 // Consume the "default" in "export default class/function". 3081 if (FormatTok->is(tok::kw_default)) 3082 nextToken(); 3083 3084 // Consume "async function", "function" and "default function", so that these 3085 // get parsed as free-standing JS functions, i.e. do not require a trailing 3086 // semicolon. 3087 if (FormatTok->is(Keywords.kw_async)) 3088 nextToken(); 3089 if (FormatTok->is(Keywords.kw_function)) { 3090 nextToken(); 3091 return; 3092 } 3093 3094 // For imports, `export *`, `export {...}`, consume the rest of the line up 3095 // to the terminating `;`. For everything else, just return and continue 3096 // parsing the structural element, i.e. the declaration or expression for 3097 // `export default`. 3098 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 3099 !FormatTok->isStringLiteral()) 3100 return; 3101 3102 while (!eof()) { 3103 if (FormatTok->is(tok::semi)) 3104 return; 3105 if (Line->Tokens.empty()) { 3106 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 3107 // import statement should terminate. 3108 return; 3109 } 3110 if (FormatTok->is(tok::l_brace)) { 3111 FormatTok->setBlockKind(BK_Block); 3112 nextToken(); 3113 parseBracedList(); 3114 } else { 3115 nextToken(); 3116 } 3117 } 3118 } 3119 3120 void UnwrappedLineParser::parseStatementMacro() { 3121 nextToken(); 3122 if (FormatTok->is(tok::l_paren)) 3123 parseParens(); 3124 if (FormatTok->is(tok::semi)) 3125 nextToken(); 3126 addUnwrappedLine(); 3127 } 3128 3129 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 3130 StringRef Prefix = "") { 3131 llvm::dbgs() << Prefix << "Line(" << Line.Level 3132 << ", FSC=" << Line.FirstStartColumn << ")" 3133 << (Line.InPPDirective ? " MACRO" : "") << ": "; 3134 for (const auto &Node : Line.Tokens) { 3135 llvm::dbgs() << Node.Tok->Tok.getName() << "[" 3136 << "T=" << static_cast<unsigned>(Node.Tok->getType()) 3137 << ", OC=" << Node.Tok->OriginalColumn << "] "; 3138 } 3139 for (const auto &Node : Line.Tokens) 3140 for (const auto &ChildNode : Node.Children) 3141 printDebugInfo(ChildNode, "\nChild: "); 3142 3143 llvm::dbgs() << "\n"; 3144 } 3145 3146 void UnwrappedLineParser::addUnwrappedLine(LineLevel AdjustLevel) { 3147 if (Line->Tokens.empty()) 3148 return; 3149 LLVM_DEBUG({ 3150 if (CurrentLines == &Lines) 3151 printDebugInfo(*Line); 3152 }); 3153 3154 // If this line closes a block when in Whitesmiths mode, remember that 3155 // information so that the level can be decreased after the line is added. 3156 // This has to happen after the addition of the line since the line itself 3157 // needs to be indented. 3158 bool ClosesWhitesmithsBlock = 3159 Line->MatchingOpeningBlockLineIndex != UnwrappedLine::kInvalidIndex && 3160 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths; 3161 3162 CurrentLines->push_back(std::move(*Line)); 3163 Line->Tokens.clear(); 3164 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex; 3165 Line->FirstStartColumn = 0; 3166 3167 if (ClosesWhitesmithsBlock && AdjustLevel == LineLevel::Remove) 3168 --Line->Level; 3169 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 3170 CurrentLines->append( 3171 std::make_move_iterator(PreprocessorDirectives.begin()), 3172 std::make_move_iterator(PreprocessorDirectives.end())); 3173 PreprocessorDirectives.clear(); 3174 } 3175 // Disconnect the current token from the last token on the previous line. 3176 FormatTok->Previous = nullptr; 3177 } 3178 3179 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 3180 3181 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 3182 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 3183 FormatTok.NewlinesBefore > 0; 3184 } 3185 3186 // Checks if \p FormatTok is a line comment that continues the line comment 3187 // section on \p Line. 3188 static bool 3189 continuesLineCommentSection(const FormatToken &FormatTok, 3190 const UnwrappedLine &Line, 3191 const llvm::Regex &CommentPragmasRegex) { 3192 if (Line.Tokens.empty()) 3193 return false; 3194 3195 StringRef IndentContent = FormatTok.TokenText; 3196 if (FormatTok.TokenText.startswith("//") || 3197 FormatTok.TokenText.startswith("/*")) 3198 IndentContent = FormatTok.TokenText.substr(2); 3199 if (CommentPragmasRegex.match(IndentContent)) 3200 return false; 3201 3202 // If Line starts with a line comment, then FormatTok continues the comment 3203 // section if its original column is greater or equal to the original start 3204 // column of the line. 3205 // 3206 // Define the min column token of a line as follows: if a line ends in '{' or 3207 // contains a '{' followed by a line comment, then the min column token is 3208 // that '{'. Otherwise, the min column token of the line is the first token of 3209 // the line. 3210 // 3211 // If Line starts with a token other than a line comment, then FormatTok 3212 // continues the comment section if its original column is greater than the 3213 // original start column of the min column token of the line. 3214 // 3215 // For example, the second line comment continues the first in these cases: 3216 // 3217 // // first line 3218 // // second line 3219 // 3220 // and: 3221 // 3222 // // first line 3223 // // second line 3224 // 3225 // and: 3226 // 3227 // int i; // first line 3228 // // second line 3229 // 3230 // and: 3231 // 3232 // do { // first line 3233 // // second line 3234 // int i; 3235 // } while (true); 3236 // 3237 // and: 3238 // 3239 // enum { 3240 // a, // first line 3241 // // second line 3242 // b 3243 // }; 3244 // 3245 // The second line comment doesn't continue the first in these cases: 3246 // 3247 // // first line 3248 // // second line 3249 // 3250 // and: 3251 // 3252 // int i; // first line 3253 // // second line 3254 // 3255 // and: 3256 // 3257 // do { // first line 3258 // // second line 3259 // int i; 3260 // } while (true); 3261 // 3262 // and: 3263 // 3264 // enum { 3265 // a, // first line 3266 // // second line 3267 // }; 3268 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 3269 3270 // Scan for '{//'. If found, use the column of '{' as a min column for line 3271 // comment section continuation. 3272 const FormatToken *PreviousToken = nullptr; 3273 for (const UnwrappedLineNode &Node : Line.Tokens) { 3274 if (PreviousToken && PreviousToken->is(tok::l_brace) && 3275 isLineComment(*Node.Tok)) { 3276 MinColumnToken = PreviousToken; 3277 break; 3278 } 3279 PreviousToken = Node.Tok; 3280 3281 // Grab the last newline preceding a token in this unwrapped line. 3282 if (Node.Tok->NewlinesBefore > 0) { 3283 MinColumnToken = Node.Tok; 3284 } 3285 } 3286 if (PreviousToken && PreviousToken->is(tok::l_brace)) { 3287 MinColumnToken = PreviousToken; 3288 } 3289 3290 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok, 3291 MinColumnToken); 3292 } 3293 3294 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 3295 bool JustComments = Line->Tokens.empty(); 3296 for (SmallVectorImpl<FormatToken *>::const_iterator 3297 I = CommentsBeforeNextToken.begin(), 3298 E = CommentsBeforeNextToken.end(); 3299 I != E; ++I) { 3300 // Line comments that belong to the same line comment section are put on the 3301 // same line since later we might want to reflow content between them. 3302 // Additional fine-grained breaking of line comment sections is controlled 3303 // by the class BreakableLineCommentSection in case it is desirable to keep 3304 // several line comment sections in the same unwrapped line. 3305 // 3306 // FIXME: Consider putting separate line comment sections as children to the 3307 // unwrapped line instead. 3308 (*I)->ContinuesLineCommentSection = 3309 continuesLineCommentSection(**I, *Line, CommentPragmasRegex); 3310 if (isOnNewLine(**I) && JustComments && !(*I)->ContinuesLineCommentSection) 3311 addUnwrappedLine(); 3312 pushToken(*I); 3313 } 3314 if (NewlineBeforeNext && JustComments) 3315 addUnwrappedLine(); 3316 CommentsBeforeNextToken.clear(); 3317 } 3318 3319 void UnwrappedLineParser::nextToken(int LevelDifference) { 3320 if (eof()) 3321 return; 3322 flushComments(isOnNewLine(*FormatTok)); 3323 pushToken(FormatTok); 3324 FormatToken *Previous = FormatTok; 3325 if (!Style.isJavaScript()) 3326 readToken(LevelDifference); 3327 else 3328 readTokenWithJavaScriptASI(); 3329 FormatTok->Previous = Previous; 3330 } 3331 3332 void UnwrappedLineParser::distributeComments( 3333 const SmallVectorImpl<FormatToken *> &Comments, 3334 const FormatToken *NextTok) { 3335 // Whether or not a line comment token continues a line is controlled by 3336 // the method continuesLineCommentSection, with the following caveat: 3337 // 3338 // Define a trail of Comments to be a nonempty proper postfix of Comments such 3339 // that each comment line from the trail is aligned with the next token, if 3340 // the next token exists. If a trail exists, the beginning of the maximal 3341 // trail is marked as a start of a new comment section. 3342 // 3343 // For example in this code: 3344 // 3345 // int a; // line about a 3346 // // line 1 about b 3347 // // line 2 about b 3348 // int b; 3349 // 3350 // the two lines about b form a maximal trail, so there are two sections, the 3351 // first one consisting of the single comment "// line about a" and the 3352 // second one consisting of the next two comments. 3353 if (Comments.empty()) 3354 return; 3355 bool ShouldPushCommentsInCurrentLine = true; 3356 bool HasTrailAlignedWithNextToken = false; 3357 unsigned StartOfTrailAlignedWithNextToken = 0; 3358 if (NextTok) { 3359 // We are skipping the first element intentionally. 3360 for (unsigned i = Comments.size() - 1; i > 0; --i) { 3361 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) { 3362 HasTrailAlignedWithNextToken = true; 3363 StartOfTrailAlignedWithNextToken = i; 3364 } 3365 } 3366 } 3367 for (unsigned i = 0, e = Comments.size(); i < e; ++i) { 3368 FormatToken *FormatTok = Comments[i]; 3369 if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) { 3370 FormatTok->ContinuesLineCommentSection = false; 3371 } else { 3372 FormatTok->ContinuesLineCommentSection = 3373 continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex); 3374 } 3375 if (!FormatTok->ContinuesLineCommentSection && 3376 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) { 3377 ShouldPushCommentsInCurrentLine = false; 3378 } 3379 if (ShouldPushCommentsInCurrentLine) { 3380 pushToken(FormatTok); 3381 } else { 3382 CommentsBeforeNextToken.push_back(FormatTok); 3383 } 3384 } 3385 } 3386 3387 void UnwrappedLineParser::readToken(int LevelDifference) { 3388 SmallVector<FormatToken *, 1> Comments; 3389 do { 3390 FormatTok = Tokens->getNextToken(); 3391 assert(FormatTok); 3392 while (FormatTok->getType() == TT_ConflictStart || 3393 FormatTok->getType() == TT_ConflictEnd || 3394 FormatTok->getType() == TT_ConflictAlternative) { 3395 if (FormatTok->getType() == TT_ConflictStart) { 3396 conditionalCompilationStart(/*Unreachable=*/false); 3397 } else if (FormatTok->getType() == TT_ConflictAlternative) { 3398 conditionalCompilationAlternative(); 3399 } else if (FormatTok->getType() == TT_ConflictEnd) { 3400 conditionalCompilationEnd(); 3401 } 3402 FormatTok = Tokens->getNextToken(); 3403 FormatTok->MustBreakBefore = true; 3404 } 3405 3406 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 3407 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 3408 distributeComments(Comments, FormatTok); 3409 Comments.clear(); 3410 // If there is an unfinished unwrapped line, we flush the preprocessor 3411 // directives only after that unwrapped line was finished later. 3412 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 3413 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 3414 assert((LevelDifference >= 0 || 3415 static_cast<unsigned>(-LevelDifference) <= Line->Level) && 3416 "LevelDifference makes Line->Level negative"); 3417 Line->Level += LevelDifference; 3418 // Comments stored before the preprocessor directive need to be output 3419 // before the preprocessor directive, at the same level as the 3420 // preprocessor directive, as we consider them to apply to the directive. 3421 if (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash && 3422 PPBranchLevel > 0) 3423 Line->Level += PPBranchLevel; 3424 flushComments(isOnNewLine(*FormatTok)); 3425 parsePPDirective(); 3426 } 3427 3428 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) && 3429 !Line->InPPDirective) { 3430 continue; 3431 } 3432 3433 if (!FormatTok->Tok.is(tok::comment)) { 3434 distributeComments(Comments, FormatTok); 3435 Comments.clear(); 3436 return; 3437 } 3438 3439 Comments.push_back(FormatTok); 3440 } while (!eof()); 3441 3442 distributeComments(Comments, nullptr); 3443 Comments.clear(); 3444 } 3445 3446 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 3447 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 3448 if (MustBreakBeforeNextToken) { 3449 Line->Tokens.back().Tok->MustBreakBefore = true; 3450 MustBreakBeforeNextToken = false; 3451 } 3452 } 3453 3454 } // end namespace format 3455 } // end namespace clang 3456