1 //===--- UnwrappedLineParser.cpp - Format C++ code ------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// 9 /// \file 10 /// This file contains the implementation of the UnwrappedLineParser, 11 /// which turns a stream of tokens into UnwrappedLines. 12 /// 13 //===----------------------------------------------------------------------===// 14 15 #include "UnwrappedLineParser.h" 16 #include "FormatToken.h" 17 #include "llvm/ADT/STLExtras.h" 18 #include "llvm/Support/Debug.h" 19 #include "llvm/Support/raw_ostream.h" 20 21 #include <algorithm> 22 23 #define DEBUG_TYPE "format-parser" 24 25 namespace clang { 26 namespace format { 27 28 class FormatTokenSource { 29 public: 30 virtual ~FormatTokenSource() {} 31 virtual FormatToken *getNextToken() = 0; 32 33 virtual unsigned getPosition() = 0; 34 virtual FormatToken *setPosition(unsigned Position) = 0; 35 }; 36 37 namespace { 38 39 class ScopedDeclarationState { 40 public: 41 ScopedDeclarationState(UnwrappedLine &Line, std::vector<bool> &Stack, 42 bool MustBeDeclaration) 43 : Line(Line), Stack(Stack) { 44 Line.MustBeDeclaration = MustBeDeclaration; 45 Stack.push_back(MustBeDeclaration); 46 } 47 ~ScopedDeclarationState() { 48 Stack.pop_back(); 49 if (!Stack.empty()) 50 Line.MustBeDeclaration = Stack.back(); 51 else 52 Line.MustBeDeclaration = true; 53 } 54 55 private: 56 UnwrappedLine &Line; 57 std::vector<bool> &Stack; 58 }; 59 60 static bool isLineComment(const FormatToken &FormatTok) { 61 return FormatTok.is(tok::comment) && !FormatTok.TokenText.startswith("/*"); 62 } 63 64 // Checks if \p FormatTok is a line comment that continues the line comment 65 // \p Previous. The original column of \p MinColumnToken is used to determine 66 // whether \p FormatTok is indented enough to the right to continue \p Previous. 67 static bool continuesLineComment(const FormatToken &FormatTok, 68 const FormatToken *Previous, 69 const FormatToken *MinColumnToken) { 70 if (!Previous || !MinColumnToken) 71 return false; 72 unsigned MinContinueColumn = 73 MinColumnToken->OriginalColumn + (isLineComment(*MinColumnToken) ? 0 : 1); 74 return isLineComment(FormatTok) && FormatTok.NewlinesBefore == 1 && 75 isLineComment(*Previous) && 76 FormatTok.OriginalColumn >= MinContinueColumn; 77 } 78 79 class ScopedMacroState : public FormatTokenSource { 80 public: 81 ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource, 82 FormatToken *&ResetToken) 83 : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken), 84 PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource), 85 Token(nullptr), PreviousToken(nullptr) { 86 FakeEOF.Tok.startToken(); 87 FakeEOF.Tok.setKind(tok::eof); 88 TokenSource = this; 89 Line.Level = 0; 90 Line.InPPDirective = true; 91 } 92 93 ~ScopedMacroState() override { 94 TokenSource = PreviousTokenSource; 95 ResetToken = Token; 96 Line.InPPDirective = false; 97 Line.Level = PreviousLineLevel; 98 } 99 100 FormatToken *getNextToken() override { 101 // The \c UnwrappedLineParser guards against this by never calling 102 // \c getNextToken() after it has encountered the first eof token. 103 assert(!eof()); 104 PreviousToken = Token; 105 Token = PreviousTokenSource->getNextToken(); 106 if (eof()) 107 return &FakeEOF; 108 return Token; 109 } 110 111 unsigned getPosition() override { return PreviousTokenSource->getPosition(); } 112 113 FormatToken *setPosition(unsigned Position) override { 114 PreviousToken = nullptr; 115 Token = PreviousTokenSource->setPosition(Position); 116 return Token; 117 } 118 119 private: 120 bool eof() { 121 return Token && Token->HasUnescapedNewline && 122 !continuesLineComment(*Token, PreviousToken, 123 /*MinColumnToken=*/PreviousToken); 124 } 125 126 FormatToken FakeEOF; 127 UnwrappedLine &Line; 128 FormatTokenSource *&TokenSource; 129 FormatToken *&ResetToken; 130 unsigned PreviousLineLevel; 131 FormatTokenSource *PreviousTokenSource; 132 133 FormatToken *Token; 134 FormatToken *PreviousToken; 135 }; 136 137 } // end anonymous namespace 138 139 class ScopedLineState { 140 public: 141 ScopedLineState(UnwrappedLineParser &Parser, 142 bool SwitchToPreprocessorLines = false) 143 : Parser(Parser), OriginalLines(Parser.CurrentLines) { 144 if (SwitchToPreprocessorLines) 145 Parser.CurrentLines = &Parser.PreprocessorDirectives; 146 else if (!Parser.Line->Tokens.empty()) 147 Parser.CurrentLines = &Parser.Line->Tokens.back().Children; 148 PreBlockLine = std::move(Parser.Line); 149 Parser.Line = std::make_unique<UnwrappedLine>(); 150 Parser.Line->Level = PreBlockLine->Level; 151 Parser.Line->InPPDirective = PreBlockLine->InPPDirective; 152 } 153 154 ~ScopedLineState() { 155 if (!Parser.Line->Tokens.empty()) { 156 Parser.addUnwrappedLine(); 157 } 158 assert(Parser.Line->Tokens.empty()); 159 Parser.Line = std::move(PreBlockLine); 160 if (Parser.CurrentLines == &Parser.PreprocessorDirectives) 161 Parser.MustBreakBeforeNextToken = true; 162 Parser.CurrentLines = OriginalLines; 163 } 164 165 private: 166 UnwrappedLineParser &Parser; 167 168 std::unique_ptr<UnwrappedLine> PreBlockLine; 169 SmallVectorImpl<UnwrappedLine> *OriginalLines; 170 }; 171 172 class CompoundStatementIndenter { 173 public: 174 CompoundStatementIndenter(UnwrappedLineParser *Parser, 175 const FormatStyle &Style, unsigned &LineLevel) 176 : CompoundStatementIndenter(Parser, LineLevel, 177 Style.BraceWrapping.AfterControlStatement, 178 Style.BraceWrapping.IndentBraces) {} 179 CompoundStatementIndenter(UnwrappedLineParser *Parser, unsigned &LineLevel, 180 bool WrapBrace, bool IndentBrace) 181 : LineLevel(LineLevel), OldLineLevel(LineLevel) { 182 if (WrapBrace) 183 Parser->addUnwrappedLine(); 184 if (IndentBrace) 185 ++LineLevel; 186 } 187 ~CompoundStatementIndenter() { LineLevel = OldLineLevel; } 188 189 private: 190 unsigned &LineLevel; 191 unsigned OldLineLevel; 192 }; 193 194 namespace { 195 196 class IndexedTokenSource : public FormatTokenSource { 197 public: 198 IndexedTokenSource(ArrayRef<FormatToken *> Tokens) 199 : Tokens(Tokens), Position(-1) {} 200 201 FormatToken *getNextToken() override { 202 ++Position; 203 return Tokens[Position]; 204 } 205 206 unsigned getPosition() override { 207 assert(Position >= 0); 208 return Position; 209 } 210 211 FormatToken *setPosition(unsigned P) override { 212 Position = P; 213 return Tokens[Position]; 214 } 215 216 void reset() { Position = -1; } 217 218 private: 219 ArrayRef<FormatToken *> Tokens; 220 int Position; 221 }; 222 223 } // end anonymous namespace 224 225 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style, 226 const AdditionalKeywords &Keywords, 227 unsigned FirstStartColumn, 228 ArrayRef<FormatToken *> Tokens, 229 UnwrappedLineConsumer &Callback) 230 : Line(new UnwrappedLine), MustBreakBeforeNextToken(false), 231 CurrentLines(&Lines), Style(Style), Keywords(Keywords), 232 CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr), 233 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1), 234 IncludeGuard(Style.IndentPPDirectives == FormatStyle::PPDIS_None 235 ? IG_Rejected 236 : IG_Inited), 237 IncludeGuardToken(nullptr), FirstStartColumn(FirstStartColumn) {} 238 239 void UnwrappedLineParser::reset() { 240 PPBranchLevel = -1; 241 IncludeGuard = Style.IndentPPDirectives == FormatStyle::PPDIS_None 242 ? IG_Rejected 243 : IG_Inited; 244 IncludeGuardToken = nullptr; 245 Line.reset(new UnwrappedLine); 246 CommentsBeforeNextToken.clear(); 247 FormatTok = nullptr; 248 MustBreakBeforeNextToken = false; 249 PreprocessorDirectives.clear(); 250 CurrentLines = &Lines; 251 DeclarationScopeStack.clear(); 252 PPStack.clear(); 253 Line->FirstStartColumn = FirstStartColumn; 254 } 255 256 void UnwrappedLineParser::parse() { 257 IndexedTokenSource TokenSource(AllTokens); 258 Line->FirstStartColumn = FirstStartColumn; 259 do { 260 LLVM_DEBUG(llvm::dbgs() << "----\n"); 261 reset(); 262 Tokens = &TokenSource; 263 TokenSource.reset(); 264 265 readToken(); 266 parseFile(); 267 268 // If we found an include guard then all preprocessor directives (other than 269 // the guard) are over-indented by one. 270 if (IncludeGuard == IG_Found) 271 for (auto &Line : Lines) 272 if (Line.InPPDirective && Line.Level > 0) 273 --Line.Level; 274 275 // Create line with eof token. 276 pushToken(FormatTok); 277 addUnwrappedLine(); 278 279 for (SmallVectorImpl<UnwrappedLine>::iterator I = Lines.begin(), 280 E = Lines.end(); 281 I != E; ++I) { 282 Callback.consumeUnwrappedLine(*I); 283 } 284 Callback.finishRun(); 285 Lines.clear(); 286 while (!PPLevelBranchIndex.empty() && 287 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) { 288 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1); 289 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1); 290 } 291 if (!PPLevelBranchIndex.empty()) { 292 ++PPLevelBranchIndex.back(); 293 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size()); 294 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back()); 295 } 296 } while (!PPLevelBranchIndex.empty()); 297 } 298 299 void UnwrappedLineParser::parseFile() { 300 // The top-level context in a file always has declarations, except for pre- 301 // processor directives and JavaScript files. 302 bool MustBeDeclaration = 303 !Line->InPPDirective && Style.Language != FormatStyle::LK_JavaScript; 304 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 305 MustBeDeclaration); 306 if (Style.Language == FormatStyle::LK_TextProto) 307 parseBracedList(); 308 else 309 parseLevel(/*HasOpeningBrace=*/false); 310 // Make sure to format the remaining tokens. 311 // 312 // LK_TextProto is special since its top-level is parsed as the body of a 313 // braced list, which does not necessarily have natural line separators such 314 // as a semicolon. Comments after the last entry that have been determined to 315 // not belong to that line, as in: 316 // key: value 317 // // endfile comment 318 // do not have a chance to be put on a line of their own until this point. 319 // Here we add this newline before end-of-file comments. 320 if (Style.Language == FormatStyle::LK_TextProto && 321 !CommentsBeforeNextToken.empty()) 322 addUnwrappedLine(); 323 flushComments(true); 324 addUnwrappedLine(); 325 } 326 327 void UnwrappedLineParser::parseCSharpGenericTypeConstraint() { 328 do { 329 switch (FormatTok->Tok.getKind()) { 330 case tok::l_brace: 331 return; 332 default: 333 if (FormatTok->is(Keywords.kw_where)) { 334 addUnwrappedLine(); 335 nextToken(); 336 parseCSharpGenericTypeConstraint(); 337 break; 338 } 339 nextToken(); 340 break; 341 } 342 } while (!eof()); 343 } 344 345 void UnwrappedLineParser::parseCSharpAttribute() { 346 int UnpairedSquareBrackets = 1; 347 do { 348 switch (FormatTok->Tok.getKind()) { 349 case tok::r_square: 350 nextToken(); 351 --UnpairedSquareBrackets; 352 if (UnpairedSquareBrackets == 0) { 353 addUnwrappedLine(); 354 return; 355 } 356 break; 357 case tok::l_square: 358 ++UnpairedSquareBrackets; 359 nextToken(); 360 break; 361 default: 362 nextToken(); 363 break; 364 } 365 } while (!eof()); 366 } 367 368 void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) { 369 bool SwitchLabelEncountered = false; 370 do { 371 tok::TokenKind kind = FormatTok->Tok.getKind(); 372 if (FormatTok->getType() == TT_MacroBlockBegin) { 373 kind = tok::l_brace; 374 } else if (FormatTok->getType() == TT_MacroBlockEnd) { 375 kind = tok::r_brace; 376 } 377 378 switch (kind) { 379 case tok::comment: 380 nextToken(); 381 addUnwrappedLine(); 382 break; 383 case tok::l_brace: 384 // FIXME: Add parameter whether this can happen - if this happens, we must 385 // be in a non-declaration context. 386 if (!FormatTok->is(TT_MacroBlockBegin) && tryToParseBracedList()) 387 continue; 388 parseBlock(/*MustBeDeclaration=*/false); 389 addUnwrappedLine(); 390 break; 391 case tok::r_brace: 392 if (HasOpeningBrace) 393 return; 394 nextToken(); 395 addUnwrappedLine(); 396 break; 397 case tok::kw_default: { 398 unsigned StoredPosition = Tokens->getPosition(); 399 FormatToken *Next; 400 do { 401 Next = Tokens->getNextToken(); 402 } while (Next && Next->is(tok::comment)); 403 FormatTok = Tokens->setPosition(StoredPosition); 404 if (Next && Next->isNot(tok::colon)) { 405 // default not followed by ':' is not a case label; treat it like 406 // an identifier. 407 parseStructuralElement(); 408 break; 409 } 410 // Else, if it is 'default:', fall through to the case handling. 411 LLVM_FALLTHROUGH; 412 } 413 case tok::kw_case: 414 if (Style.Language == FormatStyle::LK_JavaScript && 415 Line->MustBeDeclaration) { 416 // A 'case: string' style field declaration. 417 parseStructuralElement(); 418 break; 419 } 420 if (!SwitchLabelEncountered && 421 (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1))) 422 ++Line->Level; 423 SwitchLabelEncountered = true; 424 parseStructuralElement(); 425 break; 426 case tok::l_square: 427 if (Style.isCSharp()) { 428 nextToken(); 429 parseCSharpAttribute(); 430 break; 431 } 432 LLVM_FALLTHROUGH; 433 default: 434 parseStructuralElement(); 435 break; 436 } 437 } while (!eof()); 438 } 439 440 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) { 441 // We'll parse forward through the tokens until we hit 442 // a closing brace or eof - note that getNextToken() will 443 // parse macros, so this will magically work inside macro 444 // definitions, too. 445 unsigned StoredPosition = Tokens->getPosition(); 446 FormatToken *Tok = FormatTok; 447 const FormatToken *PrevTok = Tok->Previous; 448 // Keep a stack of positions of lbrace tokens. We will 449 // update information about whether an lbrace starts a 450 // braced init list or a different block during the loop. 451 SmallVector<FormatToken *, 8> LBraceStack; 452 assert(Tok->Tok.is(tok::l_brace)); 453 do { 454 // Get next non-comment token. 455 FormatToken *NextTok; 456 unsigned ReadTokens = 0; 457 do { 458 NextTok = Tokens->getNextToken(); 459 ++ReadTokens; 460 } while (NextTok->is(tok::comment)); 461 462 switch (Tok->Tok.getKind()) { 463 case tok::l_brace: 464 if (Style.Language == FormatStyle::LK_JavaScript && PrevTok) { 465 if (PrevTok->isOneOf(tok::colon, tok::less)) 466 // A ':' indicates this code is in a type, or a braced list 467 // following a label in an object literal ({a: {b: 1}}). 468 // A '<' could be an object used in a comparison, but that is nonsense 469 // code (can never return true), so more likely it is a generic type 470 // argument (`X<{a: string; b: number}>`). 471 // The code below could be confused by semicolons between the 472 // individual members in a type member list, which would normally 473 // trigger BK_Block. In both cases, this must be parsed as an inline 474 // braced init. 475 Tok->BlockKind = BK_BracedInit; 476 else if (PrevTok->is(tok::r_paren)) 477 // `) { }` can only occur in function or method declarations in JS. 478 Tok->BlockKind = BK_Block; 479 } else { 480 Tok->BlockKind = BK_Unknown; 481 } 482 LBraceStack.push_back(Tok); 483 break; 484 case tok::r_brace: 485 if (LBraceStack.empty()) 486 break; 487 if (LBraceStack.back()->BlockKind == BK_Unknown) { 488 bool ProbablyBracedList = false; 489 if (Style.Language == FormatStyle::LK_Proto) { 490 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square); 491 } else { 492 // Using OriginalColumn to distinguish between ObjC methods and 493 // binary operators is a bit hacky. 494 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 495 NextTok->OriginalColumn == 0; 496 497 // If there is a comma, semicolon or right paren after the closing 498 // brace, we assume this is a braced initializer list. Note that 499 // regardless how we mark inner braces here, we will overwrite the 500 // BlockKind later if we parse a braced list (where all blocks 501 // inside are by default braced lists), or when we explicitly detect 502 // blocks (for example while parsing lambdas). 503 // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a 504 // braced list in JS. 505 ProbablyBracedList = 506 (Style.Language == FormatStyle::LK_JavaScript && 507 NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in, 508 Keywords.kw_as)) || 509 (Style.isCpp() && NextTok->is(tok::l_paren)) || 510 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 511 tok::r_paren, tok::r_square, tok::l_brace, 512 tok::ellipsis) || 513 (NextTok->is(tok::identifier) && 514 !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace)) || 515 (NextTok->is(tok::semi) && 516 (!ExpectClassBody || LBraceStack.size() != 1)) || 517 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 518 if (!Style.isCSharp() && NextTok->is(tok::l_square)) { 519 // We can have an array subscript after a braced init 520 // list, but C++11 attributes are expected after blocks. 521 NextTok = Tokens->getNextToken(); 522 ++ReadTokens; 523 ProbablyBracedList = NextTok->isNot(tok::l_square); 524 } 525 } 526 if (ProbablyBracedList) { 527 Tok->BlockKind = BK_BracedInit; 528 LBraceStack.back()->BlockKind = BK_BracedInit; 529 } else { 530 Tok->BlockKind = BK_Block; 531 LBraceStack.back()->BlockKind = BK_Block; 532 } 533 } 534 LBraceStack.pop_back(); 535 break; 536 case tok::identifier: 537 if (!Tok->is(TT_StatementMacro)) 538 break; 539 LLVM_FALLTHROUGH; 540 case tok::at: 541 case tok::semi: 542 case tok::kw_if: 543 case tok::kw_while: 544 case tok::kw_for: 545 case tok::kw_switch: 546 case tok::kw_try: 547 case tok::kw___try: 548 if (!LBraceStack.empty() && LBraceStack.back()->BlockKind == BK_Unknown) 549 LBraceStack.back()->BlockKind = BK_Block; 550 break; 551 default: 552 break; 553 } 554 PrevTok = Tok; 555 Tok = NextTok; 556 } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty()); 557 558 // Assume other blocks for all unclosed opening braces. 559 for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) { 560 if (LBraceStack[i]->BlockKind == BK_Unknown) 561 LBraceStack[i]->BlockKind = BK_Block; 562 } 563 564 FormatTok = Tokens->setPosition(StoredPosition); 565 } 566 567 template <class T> 568 static inline void hash_combine(std::size_t &seed, const T &v) { 569 std::hash<T> hasher; 570 seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2); 571 } 572 573 size_t UnwrappedLineParser::computePPHash() const { 574 size_t h = 0; 575 for (const auto &i : PPStack) { 576 hash_combine(h, size_t(i.Kind)); 577 hash_combine(h, i.Line); 578 } 579 return h; 580 } 581 582 void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, bool AddLevel, 583 bool MunchSemi) { 584 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) && 585 "'{' or macro block token expected"); 586 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 587 FormatTok->BlockKind = BK_Block; 588 589 size_t PPStartHash = computePPHash(); 590 591 unsigned InitialLevel = Line->Level; 592 nextToken(/*LevelDifference=*/AddLevel ? 1 : 0); 593 594 if (MacroBlock && FormatTok->is(tok::l_paren)) 595 parseParens(); 596 597 size_t NbPreprocessorDirectives = 598 CurrentLines == &Lines ? PreprocessorDirectives.size() : 0; 599 addUnwrappedLine(); 600 size_t OpeningLineIndex = 601 CurrentLines->empty() 602 ? (UnwrappedLine::kInvalidIndex) 603 : (CurrentLines->size() - 1 - NbPreprocessorDirectives); 604 605 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 606 MustBeDeclaration); 607 if (AddLevel) 608 ++Line->Level; 609 parseLevel(/*HasOpeningBrace=*/true); 610 611 if (eof()) 612 return; 613 614 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd) 615 : !FormatTok->is(tok::r_brace)) { 616 Line->Level = InitialLevel; 617 FormatTok->BlockKind = BK_Block; 618 return; 619 } 620 621 size_t PPEndHash = computePPHash(); 622 623 // Munch the closing brace. 624 nextToken(/*LevelDifference=*/AddLevel ? -1 : 0); 625 626 if (MacroBlock && FormatTok->is(tok::l_paren)) 627 parseParens(); 628 629 if (MunchSemi && FormatTok->Tok.is(tok::semi)) 630 nextToken(); 631 Line->Level = InitialLevel; 632 633 if (PPStartHash == PPEndHash) { 634 Line->MatchingOpeningBlockLineIndex = OpeningLineIndex; 635 if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) { 636 // Update the opening line to add the forward reference as well 637 (*CurrentLines)[OpeningLineIndex].MatchingClosingBlockLineIndex = 638 CurrentLines->size() - 1; 639 } 640 } 641 } 642 643 static bool isGoogScope(const UnwrappedLine &Line) { 644 // FIXME: Closure-library specific stuff should not be hard-coded but be 645 // configurable. 646 if (Line.Tokens.size() < 4) 647 return false; 648 auto I = Line.Tokens.begin(); 649 if (I->Tok->TokenText != "goog") 650 return false; 651 ++I; 652 if (I->Tok->isNot(tok::period)) 653 return false; 654 ++I; 655 if (I->Tok->TokenText != "scope") 656 return false; 657 ++I; 658 return I->Tok->is(tok::l_paren); 659 } 660 661 static bool isIIFE(const UnwrappedLine &Line, 662 const AdditionalKeywords &Keywords) { 663 // Look for the start of an immediately invoked anonymous function. 664 // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression 665 // This is commonly done in JavaScript to create a new, anonymous scope. 666 // Example: (function() { ... })() 667 if (Line.Tokens.size() < 3) 668 return false; 669 auto I = Line.Tokens.begin(); 670 if (I->Tok->isNot(tok::l_paren)) 671 return false; 672 ++I; 673 if (I->Tok->isNot(Keywords.kw_function)) 674 return false; 675 ++I; 676 return I->Tok->is(tok::l_paren); 677 } 678 679 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 680 const FormatToken &InitialToken) { 681 if (InitialToken.isOneOf(tok::kw_namespace, TT_NamespaceMacro)) 682 return Style.BraceWrapping.AfterNamespace; 683 if (InitialToken.is(tok::kw_class)) 684 return Style.BraceWrapping.AfterClass; 685 if (InitialToken.is(tok::kw_union)) 686 return Style.BraceWrapping.AfterUnion; 687 if (InitialToken.is(tok::kw_struct)) 688 return Style.BraceWrapping.AfterStruct; 689 return false; 690 } 691 692 void UnwrappedLineParser::parseChildBlock() { 693 FormatTok->BlockKind = BK_Block; 694 nextToken(); 695 { 696 bool SkipIndent = (Style.Language == FormatStyle::LK_JavaScript && 697 (isGoogScope(*Line) || isIIFE(*Line, Keywords))); 698 ScopedLineState LineState(*this); 699 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 700 /*MustBeDeclaration=*/false); 701 Line->Level += SkipIndent ? 0 : 1; 702 parseLevel(/*HasOpeningBrace=*/true); 703 flushComments(isOnNewLine(*FormatTok)); 704 Line->Level -= SkipIndent ? 0 : 1; 705 } 706 nextToken(); 707 } 708 709 void UnwrappedLineParser::parsePPDirective() { 710 assert(FormatTok->Tok.is(tok::hash) && "'#' expected"); 711 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 712 713 nextToken(); 714 715 if (!FormatTok->Tok.getIdentifierInfo()) { 716 parsePPUnknown(); 717 return; 718 } 719 720 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 721 case tok::pp_define: 722 parsePPDefine(); 723 return; 724 case tok::pp_if: 725 parsePPIf(/*IfDef=*/false); 726 break; 727 case tok::pp_ifdef: 728 case tok::pp_ifndef: 729 parsePPIf(/*IfDef=*/true); 730 break; 731 case tok::pp_else: 732 parsePPElse(); 733 break; 734 case tok::pp_elif: 735 parsePPElIf(); 736 break; 737 case tok::pp_endif: 738 parsePPEndIf(); 739 break; 740 default: 741 parsePPUnknown(); 742 break; 743 } 744 } 745 746 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 747 size_t Line = CurrentLines->size(); 748 if (CurrentLines == &PreprocessorDirectives) 749 Line += Lines.size(); 750 751 if (Unreachable || 752 (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable)) 753 PPStack.push_back({PP_Unreachable, Line}); 754 else 755 PPStack.push_back({PP_Conditional, Line}); 756 } 757 758 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 759 ++PPBranchLevel; 760 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 761 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 762 PPLevelBranchIndex.push_back(0); 763 PPLevelBranchCount.push_back(0); 764 } 765 PPChainBranchIndex.push(0); 766 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 767 conditionalCompilationCondition(Unreachable || Skip); 768 } 769 770 void UnwrappedLineParser::conditionalCompilationAlternative() { 771 if (!PPStack.empty()) 772 PPStack.pop_back(); 773 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 774 if (!PPChainBranchIndex.empty()) 775 ++PPChainBranchIndex.top(); 776 conditionalCompilationCondition( 777 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 778 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 779 } 780 781 void UnwrappedLineParser::conditionalCompilationEnd() { 782 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 783 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 784 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) { 785 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 786 } 787 } 788 // Guard against #endif's without #if. 789 if (PPBranchLevel > -1) 790 --PPBranchLevel; 791 if (!PPChainBranchIndex.empty()) 792 PPChainBranchIndex.pop(); 793 if (!PPStack.empty()) 794 PPStack.pop_back(); 795 } 796 797 void UnwrappedLineParser::parsePPIf(bool IfDef) { 798 bool IfNDef = FormatTok->is(tok::pp_ifndef); 799 nextToken(); 800 bool Unreachable = false; 801 if (!IfDef && (FormatTok->is(tok::kw_false) || FormatTok->TokenText == "0")) 802 Unreachable = true; 803 if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG") 804 Unreachable = true; 805 conditionalCompilationStart(Unreachable); 806 FormatToken *IfCondition = FormatTok; 807 // If there's a #ifndef on the first line, and the only lines before it are 808 // comments, it could be an include guard. 809 bool MaybeIncludeGuard = IfNDef; 810 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) 811 for (auto &Line : Lines) { 812 if (!Line.Tokens.front().Tok->is(tok::comment)) { 813 MaybeIncludeGuard = false; 814 IncludeGuard = IG_Rejected; 815 break; 816 } 817 } 818 --PPBranchLevel; 819 parsePPUnknown(); 820 ++PPBranchLevel; 821 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) { 822 IncludeGuard = IG_IfNdefed; 823 IncludeGuardToken = IfCondition; 824 } 825 } 826 827 void UnwrappedLineParser::parsePPElse() { 828 // If a potential include guard has an #else, it's not an include guard. 829 if (IncludeGuard == IG_Defined && PPBranchLevel == 0) 830 IncludeGuard = IG_Rejected; 831 conditionalCompilationAlternative(); 832 if (PPBranchLevel > -1) 833 --PPBranchLevel; 834 parsePPUnknown(); 835 ++PPBranchLevel; 836 } 837 838 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 839 840 void UnwrappedLineParser::parsePPEndIf() { 841 conditionalCompilationEnd(); 842 parsePPUnknown(); 843 // If the #endif of a potential include guard is the last thing in the file, 844 // then we found an include guard. 845 unsigned TokenPosition = Tokens->getPosition(); 846 FormatToken *PeekNext = AllTokens[TokenPosition]; 847 if (IncludeGuard == IG_Defined && PPBranchLevel == -1 && 848 PeekNext->is(tok::eof) && 849 Style.IndentPPDirectives != FormatStyle::PPDIS_None) 850 IncludeGuard = IG_Found; 851 } 852 853 void UnwrappedLineParser::parsePPDefine() { 854 nextToken(); 855 856 if (!FormatTok->Tok.getIdentifierInfo()) { 857 IncludeGuard = IG_Rejected; 858 IncludeGuardToken = nullptr; 859 parsePPUnknown(); 860 return; 861 } 862 863 if (IncludeGuard == IG_IfNdefed && 864 IncludeGuardToken->TokenText == FormatTok->TokenText) { 865 IncludeGuard = IG_Defined; 866 IncludeGuardToken = nullptr; 867 for (auto &Line : Lines) { 868 if (!Line.Tokens.front().Tok->isOneOf(tok::comment, tok::hash)) { 869 IncludeGuard = IG_Rejected; 870 break; 871 } 872 } 873 } 874 875 nextToken(); 876 if (FormatTok->Tok.getKind() == tok::l_paren && 877 FormatTok->WhitespaceRange.getBegin() == 878 FormatTok->WhitespaceRange.getEnd()) { 879 parseParens(); 880 } 881 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None) 882 Line->Level += PPBranchLevel + 1; 883 addUnwrappedLine(); 884 ++Line->Level; 885 886 // Errors during a preprocessor directive can only affect the layout of the 887 // preprocessor directive, and thus we ignore them. An alternative approach 888 // would be to use the same approach we use on the file level (no 889 // re-indentation if there was a structural error) within the macro 890 // definition. 891 parseFile(); 892 } 893 894 void UnwrappedLineParser::parsePPUnknown() { 895 do { 896 nextToken(); 897 } while (!eof()); 898 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None) 899 Line->Level += PPBranchLevel + 1; 900 addUnwrappedLine(); 901 } 902 903 // Here we exclude certain tokens that are not usually the first token in an 904 // unwrapped line. This is used in attempt to distinguish macro calls without 905 // trailing semicolons from other constructs split to several lines. 906 static bool tokenCanStartNewLine(const FormatToken &Tok) { 907 // Semicolon can be a null-statement, l_square can be a start of a macro or 908 // a C++11 attribute, but this doesn't seem to be common. 909 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 910 Tok.isNot(TT_AttributeSquare) && 911 // Tokens that can only be used as binary operators and a part of 912 // overloaded operator names. 913 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 914 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 915 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 916 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 917 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 918 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 919 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 920 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 921 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 922 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 923 Tok.isNot(tok::lesslessequal) && 924 // Colon is used in labels, base class lists, initializer lists, 925 // range-based for loops, ternary operator, but should never be the 926 // first token in an unwrapped line. 927 Tok.isNot(tok::colon) && 928 // 'noexcept' is a trailing annotation. 929 Tok.isNot(tok::kw_noexcept); 930 } 931 932 static bool mustBeJSIdent(const AdditionalKeywords &Keywords, 933 const FormatToken *FormatTok) { 934 // FIXME: This returns true for C/C++ keywords like 'struct'. 935 return FormatTok->is(tok::identifier) && 936 (FormatTok->Tok.getIdentifierInfo() == nullptr || 937 !FormatTok->isOneOf( 938 Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async, 939 Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally, 940 Keywords.kw_function, Keywords.kw_import, Keywords.kw_is, 941 Keywords.kw_let, Keywords.kw_var, tok::kw_const, 942 Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements, 943 Keywords.kw_instanceof, Keywords.kw_interface, Keywords.kw_throws, 944 Keywords.kw_from)); 945 } 946 947 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords, 948 const FormatToken *FormatTok) { 949 return FormatTok->Tok.isLiteral() || 950 FormatTok->isOneOf(tok::kw_true, tok::kw_false) || 951 mustBeJSIdent(Keywords, FormatTok); 952 } 953 954 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement 955 // when encountered after a value (see mustBeJSIdentOrValue). 956 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords, 957 const FormatToken *FormatTok) { 958 return FormatTok->isOneOf( 959 tok::kw_return, Keywords.kw_yield, 960 // conditionals 961 tok::kw_if, tok::kw_else, 962 // loops 963 tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break, 964 // switch/case 965 tok::kw_switch, tok::kw_case, 966 // exceptions 967 tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally, 968 // declaration 969 tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let, 970 Keywords.kw_async, Keywords.kw_function, 971 // import/export 972 Keywords.kw_import, tok::kw_export); 973 } 974 975 // readTokenWithJavaScriptASI reads the next token and terminates the current 976 // line if JavaScript Automatic Semicolon Insertion must 977 // happen between the current token and the next token. 978 // 979 // This method is conservative - it cannot cover all edge cases of JavaScript, 980 // but only aims to correctly handle certain well known cases. It *must not* 981 // return true in speculative cases. 982 void UnwrappedLineParser::readTokenWithJavaScriptASI() { 983 FormatToken *Previous = FormatTok; 984 readToken(); 985 FormatToken *Next = FormatTok; 986 987 bool IsOnSameLine = 988 CommentsBeforeNextToken.empty() 989 ? Next->NewlinesBefore == 0 990 : CommentsBeforeNextToken.front()->NewlinesBefore == 0; 991 if (IsOnSameLine) 992 return; 993 994 bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous); 995 bool PreviousStartsTemplateExpr = 996 Previous->is(TT_TemplateString) && Previous->TokenText.endswith("${"); 997 if (PreviousMustBeValue || Previous->is(tok::r_paren)) { 998 // If the line contains an '@' sign, the previous token might be an 999 // annotation, which can precede another identifier/value. 1000 bool HasAt = std::find_if(Line->Tokens.begin(), Line->Tokens.end(), 1001 [](UnwrappedLineNode &LineNode) { 1002 return LineNode.Tok->is(tok::at); 1003 }) != Line->Tokens.end(); 1004 if (HasAt) 1005 return; 1006 } 1007 if (Next->is(tok::exclaim) && PreviousMustBeValue) 1008 return addUnwrappedLine(); 1009 bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next); 1010 bool NextEndsTemplateExpr = 1011 Next->is(TT_TemplateString) && Next->TokenText.startswith("}"); 1012 if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr && 1013 (PreviousMustBeValue || 1014 Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus, 1015 tok::minusminus))) 1016 return addUnwrappedLine(); 1017 if ((PreviousMustBeValue || Previous->is(tok::r_paren)) && 1018 isJSDeclOrStmt(Keywords, Next)) 1019 return addUnwrappedLine(); 1020 } 1021 1022 void UnwrappedLineParser::parseStructuralElement() { 1023 assert(!FormatTok->is(tok::l_brace)); 1024 if (Style.Language == FormatStyle::LK_TableGen && 1025 FormatTok->is(tok::pp_include)) { 1026 nextToken(); 1027 if (FormatTok->is(tok::string_literal)) 1028 nextToken(); 1029 addUnwrappedLine(); 1030 return; 1031 } 1032 switch (FormatTok->Tok.getKind()) { 1033 case tok::kw_asm: 1034 nextToken(); 1035 if (FormatTok->is(tok::l_brace)) { 1036 FormatTok->setType(TT_InlineASMBrace); 1037 nextToken(); 1038 while (FormatTok && FormatTok->isNot(tok::eof)) { 1039 if (FormatTok->is(tok::r_brace)) { 1040 FormatTok->setType(TT_InlineASMBrace); 1041 nextToken(); 1042 addUnwrappedLine(); 1043 break; 1044 } 1045 FormatTok->Finalized = true; 1046 nextToken(); 1047 } 1048 } 1049 break; 1050 case tok::kw_namespace: 1051 parseNamespace(); 1052 return; 1053 case tok::kw_public: 1054 case tok::kw_protected: 1055 case tok::kw_private: 1056 if (Style.Language == FormatStyle::LK_Java || 1057 Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp()) 1058 nextToken(); 1059 else 1060 parseAccessSpecifier(); 1061 return; 1062 case tok::kw_if: 1063 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1064 // field/method declaration. 1065 break; 1066 parseIfThenElse(); 1067 return; 1068 case tok::kw_for: 1069 case tok::kw_while: 1070 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1071 // field/method declaration. 1072 break; 1073 parseForOrWhileLoop(); 1074 return; 1075 case tok::kw_do: 1076 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1077 // field/method declaration. 1078 break; 1079 parseDoWhile(); 1080 return; 1081 case tok::kw_switch: 1082 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1083 // 'switch: string' field declaration. 1084 break; 1085 parseSwitch(); 1086 return; 1087 case tok::kw_default: 1088 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1089 // 'default: string' field declaration. 1090 break; 1091 nextToken(); 1092 if (FormatTok->is(tok::colon)) { 1093 parseLabel(); 1094 return; 1095 } 1096 // e.g. "default void f() {}" in a Java interface. 1097 break; 1098 case tok::kw_case: 1099 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1100 // 'case: string' field declaration. 1101 break; 1102 parseCaseLabel(); 1103 return; 1104 case tok::kw_try: 1105 case tok::kw___try: 1106 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1107 // field/method declaration. 1108 break; 1109 parseTryCatch(); 1110 return; 1111 case tok::kw_extern: 1112 nextToken(); 1113 if (FormatTok->Tok.is(tok::string_literal)) { 1114 nextToken(); 1115 if (FormatTok->Tok.is(tok::l_brace)) { 1116 if (!Style.IndentExternBlock) { 1117 if (Style.BraceWrapping.AfterExternBlock) { 1118 addUnwrappedLine(); 1119 } 1120 parseBlock(/*MustBeDeclaration=*/true, 1121 /*AddLevel=*/Style.BraceWrapping.AfterExternBlock); 1122 } else { 1123 parseBlock(/*MustBeDeclaration=*/true, 1124 /*AddLevel=*/Style.IndentExternBlock == 1125 FormatStyle::IEBS_Indent); 1126 } 1127 addUnwrappedLine(); 1128 return; 1129 } 1130 } 1131 break; 1132 case tok::kw_export: 1133 if (Style.Language == FormatStyle::LK_JavaScript) { 1134 parseJavaScriptEs6ImportExport(); 1135 return; 1136 } 1137 if (!Style.isCpp()) 1138 break; 1139 // Handle C++ "(inline|export) namespace". 1140 LLVM_FALLTHROUGH; 1141 case tok::kw_inline: 1142 nextToken(); 1143 if (FormatTok->Tok.is(tok::kw_namespace)) { 1144 parseNamespace(); 1145 return; 1146 } 1147 break; 1148 case tok::identifier: 1149 if (FormatTok->is(TT_ForEachMacro)) { 1150 parseForOrWhileLoop(); 1151 return; 1152 } 1153 if (FormatTok->is(TT_MacroBlockBegin)) { 1154 parseBlock(/*MustBeDeclaration=*/false, /*AddLevel=*/true, 1155 /*MunchSemi=*/false); 1156 return; 1157 } 1158 if (FormatTok->is(Keywords.kw_import)) { 1159 if (Style.Language == FormatStyle::LK_JavaScript) { 1160 parseJavaScriptEs6ImportExport(); 1161 return; 1162 } 1163 if (Style.Language == FormatStyle::LK_Proto) { 1164 nextToken(); 1165 if (FormatTok->is(tok::kw_public)) 1166 nextToken(); 1167 if (!FormatTok->is(tok::string_literal)) 1168 return; 1169 nextToken(); 1170 if (FormatTok->is(tok::semi)) 1171 nextToken(); 1172 addUnwrappedLine(); 1173 return; 1174 } 1175 } 1176 if (Style.isCpp() && 1177 FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 1178 Keywords.kw_slots, Keywords.kw_qslots)) { 1179 nextToken(); 1180 if (FormatTok->is(tok::colon)) { 1181 nextToken(); 1182 addUnwrappedLine(); 1183 return; 1184 } 1185 } 1186 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1187 parseStatementMacro(); 1188 return; 1189 } 1190 if (Style.isCpp() && FormatTok->is(TT_NamespaceMacro)) { 1191 parseNamespace(); 1192 return; 1193 } 1194 // In all other cases, parse the declaration. 1195 break; 1196 default: 1197 break; 1198 } 1199 do { 1200 const FormatToken *Previous = FormatTok->Previous; 1201 switch (FormatTok->Tok.getKind()) { 1202 case tok::at: 1203 nextToken(); 1204 if (FormatTok->Tok.is(tok::l_brace)) { 1205 nextToken(); 1206 parseBracedList(); 1207 break; 1208 } else if (Style.Language == FormatStyle::LK_Java && 1209 FormatTok->is(Keywords.kw_interface)) { 1210 nextToken(); 1211 break; 1212 } 1213 switch (FormatTok->Tok.getObjCKeywordID()) { 1214 case tok::objc_public: 1215 case tok::objc_protected: 1216 case tok::objc_package: 1217 case tok::objc_private: 1218 return parseAccessSpecifier(); 1219 case tok::objc_interface: 1220 case tok::objc_implementation: 1221 return parseObjCInterfaceOrImplementation(); 1222 case tok::objc_protocol: 1223 if (parseObjCProtocol()) 1224 return; 1225 break; 1226 case tok::objc_end: 1227 return; // Handled by the caller. 1228 case tok::objc_optional: 1229 case tok::objc_required: 1230 nextToken(); 1231 addUnwrappedLine(); 1232 return; 1233 case tok::objc_autoreleasepool: 1234 nextToken(); 1235 if (FormatTok->Tok.is(tok::l_brace)) { 1236 if (Style.BraceWrapping.AfterControlStatement == 1237 FormatStyle::BWACS_Always) 1238 addUnwrappedLine(); 1239 parseBlock(/*MustBeDeclaration=*/false); 1240 } 1241 addUnwrappedLine(); 1242 return; 1243 case tok::objc_synchronized: 1244 nextToken(); 1245 if (FormatTok->Tok.is(tok::l_paren)) 1246 // Skip synchronization object 1247 parseParens(); 1248 if (FormatTok->Tok.is(tok::l_brace)) { 1249 if (Style.BraceWrapping.AfterControlStatement == 1250 FormatStyle::BWACS_Always) 1251 addUnwrappedLine(); 1252 parseBlock(/*MustBeDeclaration=*/false); 1253 } 1254 addUnwrappedLine(); 1255 return; 1256 case tok::objc_try: 1257 // This branch isn't strictly necessary (the kw_try case below would 1258 // do this too after the tok::at is parsed above). But be explicit. 1259 parseTryCatch(); 1260 return; 1261 default: 1262 break; 1263 } 1264 break; 1265 case tok::kw_enum: 1266 // Ignore if this is part of "template <enum ...". 1267 if (Previous && Previous->is(tok::less)) { 1268 nextToken(); 1269 break; 1270 } 1271 1272 // parseEnum falls through and does not yet add an unwrapped line as an 1273 // enum definition can start a structural element. 1274 if (!parseEnum()) 1275 break; 1276 // This only applies for C++. 1277 if (!Style.isCpp()) { 1278 addUnwrappedLine(); 1279 return; 1280 } 1281 break; 1282 case tok::kw_typedef: 1283 nextToken(); 1284 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 1285 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS, 1286 Keywords.kw_CF_CLOSED_ENUM, 1287 Keywords.kw_NS_CLOSED_ENUM)) 1288 parseEnum(); 1289 break; 1290 case tok::kw_struct: 1291 case tok::kw_union: 1292 case tok::kw_class: 1293 // parseRecord falls through and does not yet add an unwrapped line as a 1294 // record declaration or definition can start a structural element. 1295 parseRecord(); 1296 // This does not apply for Java, JavaScript and C#. 1297 if (Style.Language == FormatStyle::LK_Java || 1298 Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp()) { 1299 if (FormatTok->is(tok::semi)) 1300 nextToken(); 1301 addUnwrappedLine(); 1302 return; 1303 } 1304 break; 1305 case tok::period: 1306 nextToken(); 1307 // In Java, classes have an implicit static member "class". 1308 if (Style.Language == FormatStyle::LK_Java && FormatTok && 1309 FormatTok->is(tok::kw_class)) 1310 nextToken(); 1311 if (Style.Language == FormatStyle::LK_JavaScript && FormatTok && 1312 FormatTok->Tok.getIdentifierInfo()) 1313 // JavaScript only has pseudo keywords, all keywords are allowed to 1314 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 1315 nextToken(); 1316 break; 1317 case tok::semi: 1318 nextToken(); 1319 addUnwrappedLine(); 1320 return; 1321 case tok::r_brace: 1322 addUnwrappedLine(); 1323 return; 1324 case tok::l_paren: 1325 parseParens(); 1326 break; 1327 case tok::kw_operator: 1328 nextToken(); 1329 if (FormatTok->isBinaryOperator()) 1330 nextToken(); 1331 break; 1332 case tok::caret: 1333 nextToken(); 1334 if (FormatTok->Tok.isAnyIdentifier() || 1335 FormatTok->isSimpleTypeSpecifier()) 1336 nextToken(); 1337 if (FormatTok->is(tok::l_paren)) 1338 parseParens(); 1339 if (FormatTok->is(tok::l_brace)) 1340 parseChildBlock(); 1341 break; 1342 case tok::l_brace: 1343 if (!tryToParsePropertyAccessor() && !tryToParseBracedList()) { 1344 // A block outside of parentheses must be the last part of a 1345 // structural element. 1346 // FIXME: Figure out cases where this is not true, and add projections 1347 // for them (the one we know is missing are lambdas). 1348 if (Style.BraceWrapping.AfterFunction) 1349 addUnwrappedLine(); 1350 FormatTok->setType(TT_FunctionLBrace); 1351 parseBlock(/*MustBeDeclaration=*/false); 1352 addUnwrappedLine(); 1353 return; 1354 } 1355 // Otherwise this was a braced init list, and the structural 1356 // element continues. 1357 break; 1358 case tok::kw_try: 1359 if (Style.Language == FormatStyle::LK_JavaScript && 1360 Line->MustBeDeclaration) { 1361 // field/method declaration. 1362 nextToken(); 1363 break; 1364 } 1365 // We arrive here when parsing function-try blocks. 1366 if (Style.BraceWrapping.AfterFunction) 1367 addUnwrappedLine(); 1368 parseTryCatch(); 1369 return; 1370 case tok::identifier: { 1371 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where) && 1372 Line->MustBeDeclaration) { 1373 addUnwrappedLine(); 1374 parseCSharpGenericTypeConstraint(); 1375 break; 1376 } 1377 if (FormatTok->is(TT_MacroBlockEnd)) { 1378 addUnwrappedLine(); 1379 return; 1380 } 1381 1382 // Function declarations (as opposed to function expressions) are parsed 1383 // on their own unwrapped line by continuing this loop. Function 1384 // expressions (functions that are not on their own line) must not create 1385 // a new unwrapped line, so they are special cased below. 1386 size_t TokenCount = Line->Tokens.size(); 1387 if (Style.Language == FormatStyle::LK_JavaScript && 1388 FormatTok->is(Keywords.kw_function) && 1389 (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is( 1390 Keywords.kw_async)))) { 1391 tryToParseJSFunction(); 1392 break; 1393 } 1394 if ((Style.Language == FormatStyle::LK_JavaScript || 1395 Style.Language == FormatStyle::LK_Java) && 1396 FormatTok->is(Keywords.kw_interface)) { 1397 if (Style.Language == FormatStyle::LK_JavaScript) { 1398 // In JavaScript/TypeScript, "interface" can be used as a standalone 1399 // identifier, e.g. in `var interface = 1;`. If "interface" is 1400 // followed by another identifier, it is very like to be an actual 1401 // interface declaration. 1402 unsigned StoredPosition = Tokens->getPosition(); 1403 FormatToken *Next = Tokens->getNextToken(); 1404 FormatTok = Tokens->setPosition(StoredPosition); 1405 if (Next && !mustBeJSIdent(Keywords, Next)) { 1406 nextToken(); 1407 break; 1408 } 1409 } 1410 parseRecord(); 1411 addUnwrappedLine(); 1412 return; 1413 } 1414 1415 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1416 parseStatementMacro(); 1417 return; 1418 } 1419 1420 // See if the following token should start a new unwrapped line. 1421 StringRef Text = FormatTok->TokenText; 1422 nextToken(); 1423 1424 // JS doesn't have macros, and within classes colons indicate fields, not 1425 // labels. 1426 if (Style.Language == FormatStyle::LK_JavaScript) 1427 break; 1428 1429 TokenCount = Line->Tokens.size(); 1430 if (TokenCount == 1 || 1431 (TokenCount == 2 && Line->Tokens.front().Tok->is(tok::comment))) { 1432 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) { 1433 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1434 parseLabel(!Style.IndentGotoLabels); 1435 return; 1436 } 1437 // Recognize function-like macro usages without trailing semicolon as 1438 // well as free-standing macros like Q_OBJECT. 1439 bool FunctionLike = FormatTok->is(tok::l_paren); 1440 if (FunctionLike) 1441 parseParens(); 1442 1443 bool FollowedByNewline = 1444 CommentsBeforeNextToken.empty() 1445 ? FormatTok->NewlinesBefore > 0 1446 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1447 1448 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1449 tokenCanStartNewLine(*FormatTok) && Text == Text.upper()) { 1450 addUnwrappedLine(); 1451 return; 1452 } 1453 } 1454 break; 1455 } 1456 case tok::equal: 1457 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType 1458 // TT_JsFatArrow. The always start an expression or a child block if 1459 // followed by a curly. 1460 if (FormatTok->is(TT_JsFatArrow)) { 1461 nextToken(); 1462 if (FormatTok->is(tok::l_brace)) { 1463 // C# may break after => if the next character is a newline. 1464 if (Style.isCSharp() && Style.BraceWrapping.AfterFunction == true) { 1465 // calling `addUnwrappedLine()` here causes odd parsing errors. 1466 FormatTok->MustBreakBefore = true; 1467 } 1468 parseChildBlock(); 1469 } 1470 break; 1471 } 1472 1473 nextToken(); 1474 if (FormatTok->Tok.is(tok::l_brace)) { 1475 // Block kind should probably be set to BK_BracedInit for any language. 1476 // C# needs this change to ensure that array initialisers and object 1477 // initialisers are indented the same way. 1478 if (Style.isCSharp()) 1479 FormatTok->BlockKind = BK_BracedInit; 1480 nextToken(); 1481 parseBracedList(); 1482 } else if (Style.Language == FormatStyle::LK_Proto && 1483 FormatTok->Tok.is(tok::less)) { 1484 nextToken(); 1485 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 1486 /*ClosingBraceKind=*/tok::greater); 1487 } 1488 break; 1489 case tok::l_square: 1490 parseSquare(); 1491 break; 1492 case tok::kw_new: 1493 parseNew(); 1494 break; 1495 default: 1496 nextToken(); 1497 break; 1498 } 1499 } while (!eof()); 1500 } 1501 1502 bool UnwrappedLineParser::tryToParsePropertyAccessor() { 1503 assert(FormatTok->is(tok::l_brace)); 1504 if (!Style.isCSharp()) 1505 return false; 1506 // See if it's a property accessor. 1507 if (FormatTok->Previous->isNot(tok::identifier)) 1508 return false; 1509 1510 // See if we are inside a property accessor. 1511 // 1512 // Record the current tokenPosition so that we can advance and 1513 // reset the current token. `Next` is not set yet so we need 1514 // another way to advance along the token stream. 1515 unsigned int StoredPosition = Tokens->getPosition(); 1516 FormatToken *Tok = Tokens->getNextToken(); 1517 1518 // A trivial property accessor is of the form: 1519 // { [ACCESS_SPECIFIER] [get]; [ACCESS_SPECIFIER] [set] } 1520 // Track these as they do not require line breaks to be introduced. 1521 bool HasGetOrSet = false; 1522 bool IsTrivialPropertyAccessor = true; 1523 while (!eof()) { 1524 if (Tok->isOneOf(tok::semi, tok::kw_public, tok::kw_private, 1525 tok::kw_protected, Keywords.kw_internal, Keywords.kw_get, 1526 Keywords.kw_set)) { 1527 if (Tok->isOneOf(Keywords.kw_get, Keywords.kw_set)) 1528 HasGetOrSet = true; 1529 Tok = Tokens->getNextToken(); 1530 continue; 1531 } 1532 if (Tok->isNot(tok::r_brace)) 1533 IsTrivialPropertyAccessor = false; 1534 break; 1535 } 1536 1537 if (!HasGetOrSet) { 1538 Tokens->setPosition(StoredPosition); 1539 return false; 1540 } 1541 1542 // Try to parse the property accessor: 1543 // https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties 1544 Tokens->setPosition(StoredPosition); 1545 if (!IsTrivialPropertyAccessor && Style.BraceWrapping.AfterFunction == true) 1546 addUnwrappedLine(); 1547 nextToken(); 1548 do { 1549 switch (FormatTok->Tok.getKind()) { 1550 case tok::r_brace: 1551 nextToken(); 1552 if (FormatTok->is(tok::equal)) { 1553 while (!eof() && FormatTok->isNot(tok::semi)) 1554 nextToken(); 1555 nextToken(); 1556 } 1557 addUnwrappedLine(); 1558 return true; 1559 case tok::l_brace: 1560 ++Line->Level; 1561 parseBlock(/*MustBeDeclaration=*/true); 1562 addUnwrappedLine(); 1563 --Line->Level; 1564 break; 1565 case tok::equal: 1566 if (FormatTok->is(TT_JsFatArrow)) { 1567 ++Line->Level; 1568 do { 1569 nextToken(); 1570 } while (!eof() && FormatTok->isNot(tok::semi)); 1571 nextToken(); 1572 addUnwrappedLine(); 1573 --Line->Level; 1574 break; 1575 } 1576 nextToken(); 1577 break; 1578 default: 1579 if (FormatTok->isOneOf(Keywords.kw_get, Keywords.kw_set) && 1580 !IsTrivialPropertyAccessor) { 1581 // Non-trivial get/set needs to be on its own line. 1582 addUnwrappedLine(); 1583 } 1584 nextToken(); 1585 } 1586 } while (!eof()); 1587 1588 // Unreachable for well-formed code (paired '{' and '}'). 1589 return true; 1590 } 1591 1592 bool UnwrappedLineParser::tryToParseLambda() { 1593 if (!Style.isCpp()) { 1594 nextToken(); 1595 return false; 1596 } 1597 assert(FormatTok->is(tok::l_square)); 1598 FormatToken &LSquare = *FormatTok; 1599 if (!tryToParseLambdaIntroducer()) 1600 return false; 1601 1602 bool SeenArrow = false; 1603 1604 while (FormatTok->isNot(tok::l_brace)) { 1605 if (FormatTok->isSimpleTypeSpecifier()) { 1606 nextToken(); 1607 continue; 1608 } 1609 switch (FormatTok->Tok.getKind()) { 1610 case tok::l_brace: 1611 break; 1612 case tok::l_paren: 1613 parseParens(); 1614 break; 1615 case tok::amp: 1616 case tok::star: 1617 case tok::kw_const: 1618 case tok::comma: 1619 case tok::less: 1620 case tok::greater: 1621 case tok::identifier: 1622 case tok::numeric_constant: 1623 case tok::coloncolon: 1624 case tok::kw_class: 1625 case tok::kw_mutable: 1626 case tok::kw_noexcept: 1627 case tok::kw_template: 1628 case tok::kw_typename: 1629 nextToken(); 1630 break; 1631 // Specialization of a template with an integer parameter can contain 1632 // arithmetic, logical, comparison and ternary operators. 1633 // 1634 // FIXME: This also accepts sequences of operators that are not in the scope 1635 // of a template argument list. 1636 // 1637 // In a C++ lambda a template type can only occur after an arrow. We use 1638 // this as an heuristic to distinguish between Objective-C expressions 1639 // followed by an `a->b` expression, such as: 1640 // ([obj func:arg] + a->b) 1641 // Otherwise the code below would parse as a lambda. 1642 // 1643 // FIXME: This heuristic is incorrect for C++20 generic lambdas with 1644 // explicit template lists: []<bool b = true && false>(U &&u){} 1645 case tok::plus: 1646 case tok::minus: 1647 case tok::exclaim: 1648 case tok::tilde: 1649 case tok::slash: 1650 case tok::percent: 1651 case tok::lessless: 1652 case tok::pipe: 1653 case tok::pipepipe: 1654 case tok::ampamp: 1655 case tok::caret: 1656 case tok::equalequal: 1657 case tok::exclaimequal: 1658 case tok::greaterequal: 1659 case tok::lessequal: 1660 case tok::question: 1661 case tok::colon: 1662 case tok::ellipsis: 1663 case tok::kw_true: 1664 case tok::kw_false: 1665 if (SeenArrow) { 1666 nextToken(); 1667 break; 1668 } 1669 return true; 1670 case tok::arrow: 1671 // This might or might not actually be a lambda arrow (this could be an 1672 // ObjC method invocation followed by a dereferencing arrow). We might 1673 // reset this back to TT_Unknown in TokenAnnotator. 1674 FormatTok->setType(TT_LambdaArrow); 1675 SeenArrow = true; 1676 nextToken(); 1677 break; 1678 default: 1679 return true; 1680 } 1681 } 1682 FormatTok->setType(TT_LambdaLBrace); 1683 LSquare.setType(TT_LambdaLSquare); 1684 parseChildBlock(); 1685 return true; 1686 } 1687 1688 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 1689 const FormatToken *Previous = FormatTok->Previous; 1690 if (Previous && 1691 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 1692 tok::kw_delete, tok::l_square) || 1693 FormatTok->isCppStructuredBinding(Style) || Previous->closesScope() || 1694 Previous->isSimpleTypeSpecifier())) { 1695 nextToken(); 1696 return false; 1697 } 1698 nextToken(); 1699 if (FormatTok->is(tok::l_square)) { 1700 return false; 1701 } 1702 parseSquare(/*LambdaIntroducer=*/true); 1703 return true; 1704 } 1705 1706 void UnwrappedLineParser::tryToParseJSFunction() { 1707 assert(FormatTok->is(Keywords.kw_function) || 1708 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)); 1709 if (FormatTok->is(Keywords.kw_async)) 1710 nextToken(); 1711 // Consume "function". 1712 nextToken(); 1713 1714 // Consume * (generator function). Treat it like C++'s overloaded operators. 1715 if (FormatTok->is(tok::star)) { 1716 FormatTok->setType(TT_OverloadedOperator); 1717 nextToken(); 1718 } 1719 1720 // Consume function name. 1721 if (FormatTok->is(tok::identifier)) 1722 nextToken(); 1723 1724 if (FormatTok->isNot(tok::l_paren)) 1725 return; 1726 1727 // Parse formal parameter list. 1728 parseParens(); 1729 1730 if (FormatTok->is(tok::colon)) { 1731 // Parse a type definition. 1732 nextToken(); 1733 1734 // Eat the type declaration. For braced inline object types, balance braces, 1735 // otherwise just parse until finding an l_brace for the function body. 1736 if (FormatTok->is(tok::l_brace)) 1737 tryToParseBracedList(); 1738 else 1739 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 1740 nextToken(); 1741 } 1742 1743 if (FormatTok->is(tok::semi)) 1744 return; 1745 1746 parseChildBlock(); 1747 } 1748 1749 bool UnwrappedLineParser::tryToParseBracedList() { 1750 if (FormatTok->BlockKind == BK_Unknown) 1751 calculateBraceTypes(); 1752 assert(FormatTok->BlockKind != BK_Unknown); 1753 if (FormatTok->BlockKind == BK_Block) 1754 return false; 1755 nextToken(); 1756 parseBracedList(); 1757 return true; 1758 } 1759 1760 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons, 1761 bool IsEnum, 1762 tok::TokenKind ClosingBraceKind) { 1763 bool HasError = false; 1764 1765 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 1766 // replace this by using parseAssigmentExpression() inside. 1767 do { 1768 if (Style.isCSharp()) { 1769 if (FormatTok->is(TT_JsFatArrow)) { 1770 nextToken(); 1771 // Fat arrows can be followed by simple expressions or by child blocks 1772 // in curly braces. 1773 if (FormatTok->is(tok::l_brace)) { 1774 parseChildBlock(); 1775 continue; 1776 } 1777 } 1778 } 1779 if (Style.Language == FormatStyle::LK_JavaScript) { 1780 if (FormatTok->is(Keywords.kw_function) || 1781 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) { 1782 tryToParseJSFunction(); 1783 continue; 1784 } 1785 if (FormatTok->is(TT_JsFatArrow)) { 1786 nextToken(); 1787 // Fat arrows can be followed by simple expressions or by child blocks 1788 // in curly braces. 1789 if (FormatTok->is(tok::l_brace)) { 1790 parseChildBlock(); 1791 continue; 1792 } 1793 } 1794 if (FormatTok->is(tok::l_brace)) { 1795 // Could be a method inside of a braced list `{a() { return 1; }}`. 1796 if (tryToParseBracedList()) 1797 continue; 1798 parseChildBlock(); 1799 } 1800 } 1801 if (FormatTok->Tok.getKind() == ClosingBraceKind) { 1802 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 1803 addUnwrappedLine(); 1804 nextToken(); 1805 return !HasError; 1806 } 1807 switch (FormatTok->Tok.getKind()) { 1808 case tok::caret: 1809 nextToken(); 1810 if (FormatTok->is(tok::l_brace)) { 1811 parseChildBlock(); 1812 } 1813 break; 1814 case tok::l_square: 1815 if (Style.isCSharp()) 1816 parseSquare(); 1817 else 1818 tryToParseLambda(); 1819 break; 1820 case tok::l_paren: 1821 parseParens(); 1822 // JavaScript can just have free standing methods and getters/setters in 1823 // object literals. Detect them by a "{" following ")". 1824 if (Style.Language == FormatStyle::LK_JavaScript) { 1825 if (FormatTok->is(tok::l_brace)) 1826 parseChildBlock(); 1827 break; 1828 } 1829 break; 1830 case tok::l_brace: 1831 // Assume there are no blocks inside a braced init list apart 1832 // from the ones we explicitly parse out (like lambdas). 1833 FormatTok->BlockKind = BK_BracedInit; 1834 nextToken(); 1835 parseBracedList(); 1836 break; 1837 case tok::less: 1838 if (Style.Language == FormatStyle::LK_Proto) { 1839 nextToken(); 1840 parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false, 1841 /*ClosingBraceKind=*/tok::greater); 1842 } else { 1843 nextToken(); 1844 } 1845 break; 1846 case tok::semi: 1847 // JavaScript (or more precisely TypeScript) can have semicolons in braced 1848 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 1849 // used for error recovery if we have otherwise determined that this is 1850 // a braced list. 1851 if (Style.Language == FormatStyle::LK_JavaScript) { 1852 nextToken(); 1853 break; 1854 } 1855 HasError = true; 1856 if (!ContinueOnSemicolons) 1857 return !HasError; 1858 nextToken(); 1859 break; 1860 case tok::comma: 1861 nextToken(); 1862 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 1863 addUnwrappedLine(); 1864 break; 1865 default: 1866 nextToken(); 1867 break; 1868 } 1869 } while (!eof()); 1870 return false; 1871 } 1872 1873 void UnwrappedLineParser::parseParens() { 1874 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 1875 nextToken(); 1876 do { 1877 switch (FormatTok->Tok.getKind()) { 1878 case tok::l_paren: 1879 parseParens(); 1880 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 1881 parseChildBlock(); 1882 break; 1883 case tok::r_paren: 1884 nextToken(); 1885 return; 1886 case tok::r_brace: 1887 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1888 return; 1889 case tok::l_square: 1890 tryToParseLambda(); 1891 break; 1892 case tok::l_brace: 1893 if (!tryToParseBracedList()) 1894 parseChildBlock(); 1895 break; 1896 case tok::at: 1897 nextToken(); 1898 if (FormatTok->Tok.is(tok::l_brace)) { 1899 nextToken(); 1900 parseBracedList(); 1901 } 1902 break; 1903 case tok::kw_class: 1904 if (Style.Language == FormatStyle::LK_JavaScript) 1905 parseRecord(/*ParseAsExpr=*/true); 1906 else 1907 nextToken(); 1908 break; 1909 case tok::identifier: 1910 if (Style.Language == FormatStyle::LK_JavaScript && 1911 (FormatTok->is(Keywords.kw_function) || 1912 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function))) 1913 tryToParseJSFunction(); 1914 else 1915 nextToken(); 1916 break; 1917 default: 1918 nextToken(); 1919 break; 1920 } 1921 } while (!eof()); 1922 } 1923 1924 void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) { 1925 if (!LambdaIntroducer) { 1926 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected."); 1927 if (tryToParseLambda()) 1928 return; 1929 } 1930 do { 1931 switch (FormatTok->Tok.getKind()) { 1932 case tok::l_paren: 1933 parseParens(); 1934 break; 1935 case tok::r_square: 1936 nextToken(); 1937 return; 1938 case tok::r_brace: 1939 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1940 return; 1941 case tok::l_square: 1942 parseSquare(); 1943 break; 1944 case tok::l_brace: { 1945 if (!tryToParseBracedList()) 1946 parseChildBlock(); 1947 break; 1948 } 1949 case tok::at: 1950 nextToken(); 1951 if (FormatTok->Tok.is(tok::l_brace)) { 1952 nextToken(); 1953 parseBracedList(); 1954 } 1955 break; 1956 default: 1957 nextToken(); 1958 break; 1959 } 1960 } while (!eof()); 1961 } 1962 1963 void UnwrappedLineParser::parseIfThenElse() { 1964 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 1965 nextToken(); 1966 if (FormatTok->Tok.isOneOf(tok::kw_constexpr, tok::identifier)) 1967 nextToken(); 1968 if (FormatTok->Tok.is(tok::l_paren)) 1969 parseParens(); 1970 // handle [[likely]] / [[unlikely]] 1971 if (FormatTok->is(tok::l_square) && tryToParseSimpleAttribute()) 1972 parseSquare(); 1973 bool NeedsUnwrappedLine = false; 1974 if (FormatTok->Tok.is(tok::l_brace)) { 1975 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1976 parseBlock(/*MustBeDeclaration=*/false); 1977 if (Style.BraceWrapping.BeforeElse) 1978 addUnwrappedLine(); 1979 else 1980 NeedsUnwrappedLine = true; 1981 } else { 1982 addUnwrappedLine(); 1983 ++Line->Level; 1984 parseStructuralElement(); 1985 --Line->Level; 1986 } 1987 if (FormatTok->Tok.is(tok::kw_else)) { 1988 nextToken(); 1989 // handle [[likely]] / [[unlikely]] 1990 if (FormatTok->Tok.is(tok::l_square) && tryToParseSimpleAttribute()) 1991 parseSquare(); 1992 if (FormatTok->Tok.is(tok::l_brace)) { 1993 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1994 parseBlock(/*MustBeDeclaration=*/false); 1995 addUnwrappedLine(); 1996 } else if (FormatTok->Tok.is(tok::kw_if)) { 1997 parseIfThenElse(); 1998 } else { 1999 addUnwrappedLine(); 2000 ++Line->Level; 2001 parseStructuralElement(); 2002 if (FormatTok->is(tok::eof)) 2003 addUnwrappedLine(); 2004 --Line->Level; 2005 } 2006 } else if (NeedsUnwrappedLine) { 2007 addUnwrappedLine(); 2008 } 2009 } 2010 2011 void UnwrappedLineParser::parseTryCatch() { 2012 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 2013 nextToken(); 2014 bool NeedsUnwrappedLine = false; 2015 if (FormatTok->is(tok::colon)) { 2016 // We are in a function try block, what comes is an initializer list. 2017 nextToken(); 2018 2019 // In case identifiers were removed by clang-tidy, what might follow is 2020 // multiple commas in sequence - before the first identifier. 2021 while (FormatTok->is(tok::comma)) 2022 nextToken(); 2023 2024 while (FormatTok->is(tok::identifier)) { 2025 nextToken(); 2026 if (FormatTok->is(tok::l_paren)) 2027 parseParens(); 2028 2029 // In case identifiers were removed by clang-tidy, what might follow is 2030 // multiple commas in sequence - after the first identifier. 2031 while (FormatTok->is(tok::comma)) 2032 nextToken(); 2033 } 2034 } 2035 // Parse try with resource. 2036 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) { 2037 parseParens(); 2038 } 2039 if (FormatTok->is(tok::l_brace)) { 2040 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2041 parseBlock(/*MustBeDeclaration=*/false); 2042 if (Style.BraceWrapping.BeforeCatch) { 2043 addUnwrappedLine(); 2044 } else { 2045 NeedsUnwrappedLine = true; 2046 } 2047 } else if (!FormatTok->is(tok::kw_catch)) { 2048 // The C++ standard requires a compound-statement after a try. 2049 // If there's none, we try to assume there's a structuralElement 2050 // and try to continue. 2051 addUnwrappedLine(); 2052 ++Line->Level; 2053 parseStructuralElement(); 2054 --Line->Level; 2055 } 2056 while (1) { 2057 if (FormatTok->is(tok::at)) 2058 nextToken(); 2059 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 2060 tok::kw___finally) || 2061 ((Style.Language == FormatStyle::LK_Java || 2062 Style.Language == FormatStyle::LK_JavaScript) && 2063 FormatTok->is(Keywords.kw_finally)) || 2064 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) || 2065 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally)))) 2066 break; 2067 nextToken(); 2068 while (FormatTok->isNot(tok::l_brace)) { 2069 if (FormatTok->is(tok::l_paren)) { 2070 parseParens(); 2071 continue; 2072 } 2073 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) 2074 return; 2075 nextToken(); 2076 } 2077 NeedsUnwrappedLine = false; 2078 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2079 parseBlock(/*MustBeDeclaration=*/false); 2080 if (Style.BraceWrapping.BeforeCatch) 2081 addUnwrappedLine(); 2082 else 2083 NeedsUnwrappedLine = true; 2084 } 2085 if (NeedsUnwrappedLine) 2086 addUnwrappedLine(); 2087 } 2088 2089 void UnwrappedLineParser::parseNamespace() { 2090 assert(FormatTok->isOneOf(tok::kw_namespace, TT_NamespaceMacro) && 2091 "'namespace' expected"); 2092 2093 const FormatToken &InitialToken = *FormatTok; 2094 nextToken(); 2095 if (InitialToken.is(TT_NamespaceMacro)) { 2096 parseParens(); 2097 } else { 2098 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::kw_inline, 2099 tok::l_square)) { 2100 if (FormatTok->is(tok::l_square)) 2101 parseSquare(); 2102 else 2103 nextToken(); 2104 } 2105 } 2106 if (FormatTok->Tok.is(tok::l_brace)) { 2107 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2108 addUnwrappedLine(); 2109 2110 bool AddLevel = Style.NamespaceIndentation == FormatStyle::NI_All || 2111 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 2112 DeclarationScopeStack.size() > 1); 2113 parseBlock(/*MustBeDeclaration=*/true, AddLevel); 2114 // Munch the semicolon after a namespace. This is more common than one would 2115 // think. Putting the semicolon into its own line is very ugly. 2116 if (FormatTok->Tok.is(tok::semi)) 2117 nextToken(); 2118 addUnwrappedLine(); 2119 } 2120 // FIXME: Add error handling. 2121 } 2122 2123 void UnwrappedLineParser::parseNew() { 2124 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 2125 nextToken(); 2126 2127 if (Style.isCSharp()) { 2128 do { 2129 if (FormatTok->is(tok::l_brace)) 2130 parseBracedList(); 2131 2132 if (FormatTok->isOneOf(tok::semi, tok::comma)) 2133 return; 2134 2135 nextToken(); 2136 } while (!eof()); 2137 } 2138 2139 if (Style.Language != FormatStyle::LK_Java) 2140 return; 2141 2142 // In Java, we can parse everything up to the parens, which aren't optional. 2143 do { 2144 // There should not be a ;, { or } before the new's open paren. 2145 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 2146 return; 2147 2148 // Consume the parens. 2149 if (FormatTok->is(tok::l_paren)) { 2150 parseParens(); 2151 2152 // If there is a class body of an anonymous class, consume that as child. 2153 if (FormatTok->is(tok::l_brace)) 2154 parseChildBlock(); 2155 return; 2156 } 2157 nextToken(); 2158 } while (!eof()); 2159 } 2160 2161 void UnwrappedLineParser::parseForOrWhileLoop() { 2162 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 2163 "'for', 'while' or foreach macro expected"); 2164 nextToken(); 2165 // JS' for await ( ... 2166 if (Style.Language == FormatStyle::LK_JavaScript && 2167 FormatTok->is(Keywords.kw_await)) 2168 nextToken(); 2169 if (FormatTok->Tok.is(tok::l_paren)) 2170 parseParens(); 2171 if (FormatTok->Tok.is(tok::l_brace)) { 2172 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2173 parseBlock(/*MustBeDeclaration=*/false); 2174 addUnwrappedLine(); 2175 } else { 2176 addUnwrappedLine(); 2177 ++Line->Level; 2178 parseStructuralElement(); 2179 --Line->Level; 2180 } 2181 } 2182 2183 void UnwrappedLineParser::parseDoWhile() { 2184 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 2185 nextToken(); 2186 if (FormatTok->Tok.is(tok::l_brace)) { 2187 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2188 parseBlock(/*MustBeDeclaration=*/false); 2189 if (Style.BraceWrapping.BeforeWhile) 2190 addUnwrappedLine(); 2191 } else { 2192 addUnwrappedLine(); 2193 ++Line->Level; 2194 parseStructuralElement(); 2195 --Line->Level; 2196 } 2197 2198 // FIXME: Add error handling. 2199 if (!FormatTok->Tok.is(tok::kw_while)) { 2200 addUnwrappedLine(); 2201 return; 2202 } 2203 2204 nextToken(); 2205 parseStructuralElement(); 2206 } 2207 2208 void UnwrappedLineParser::parseLabel(bool LeftAlignLabel) { 2209 nextToken(); 2210 unsigned OldLineLevel = Line->Level; 2211 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 2212 --Line->Level; 2213 if (LeftAlignLabel) 2214 Line->Level = 0; 2215 if (!Style.IndentCaseBlocks && CommentsBeforeNextToken.empty() && 2216 FormatTok->Tok.is(tok::l_brace)) { 2217 CompoundStatementIndenter Indenter(this, Line->Level, 2218 Style.BraceWrapping.AfterCaseLabel, 2219 Style.BraceWrapping.IndentBraces); 2220 parseBlock(/*MustBeDeclaration=*/false); 2221 if (FormatTok->Tok.is(tok::kw_break)) { 2222 if (Style.BraceWrapping.AfterControlStatement == 2223 FormatStyle::BWACS_Always) { 2224 addUnwrappedLine(); 2225 if (!Style.IndentCaseBlocks && 2226 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) { 2227 Line->Level++; 2228 } 2229 } 2230 parseStructuralElement(); 2231 } 2232 addUnwrappedLine(); 2233 } else { 2234 if (FormatTok->is(tok::semi)) 2235 nextToken(); 2236 addUnwrappedLine(); 2237 } 2238 Line->Level = OldLineLevel; 2239 if (FormatTok->isNot(tok::l_brace)) { 2240 parseStructuralElement(); 2241 addUnwrappedLine(); 2242 } 2243 } 2244 2245 void UnwrappedLineParser::parseCaseLabel() { 2246 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 2247 // FIXME: fix handling of complex expressions here. 2248 do { 2249 nextToken(); 2250 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 2251 parseLabel(); 2252 } 2253 2254 void UnwrappedLineParser::parseSwitch() { 2255 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 2256 nextToken(); 2257 if (FormatTok->Tok.is(tok::l_paren)) 2258 parseParens(); 2259 if (FormatTok->Tok.is(tok::l_brace)) { 2260 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2261 parseBlock(/*MustBeDeclaration=*/false); 2262 addUnwrappedLine(); 2263 } else { 2264 addUnwrappedLine(); 2265 ++Line->Level; 2266 parseStructuralElement(); 2267 --Line->Level; 2268 } 2269 } 2270 2271 void UnwrappedLineParser::parseAccessSpecifier() { 2272 nextToken(); 2273 // Understand Qt's slots. 2274 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 2275 nextToken(); 2276 // Otherwise, we don't know what it is, and we'd better keep the next token. 2277 if (FormatTok->Tok.is(tok::colon)) 2278 nextToken(); 2279 addUnwrappedLine(); 2280 } 2281 2282 bool UnwrappedLineParser::parseEnum() { 2283 // Won't be 'enum' for NS_ENUMs. 2284 if (FormatTok->Tok.is(tok::kw_enum)) 2285 nextToken(); 2286 2287 // In TypeScript, "enum" can also be used as property name, e.g. in interface 2288 // declarations. An "enum" keyword followed by a colon would be a syntax 2289 // error and thus assume it is just an identifier. 2290 if (Style.Language == FormatStyle::LK_JavaScript && 2291 FormatTok->isOneOf(tok::colon, tok::question)) 2292 return false; 2293 2294 // In protobuf, "enum" can be used as a field name. 2295 if (Style.Language == FormatStyle::LK_Proto && FormatTok->is(tok::equal)) 2296 return false; 2297 2298 // Eat up enum class ... 2299 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct)) 2300 nextToken(); 2301 2302 while (FormatTok->Tok.getIdentifierInfo() || 2303 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 2304 tok::greater, tok::comma, tok::question)) { 2305 nextToken(); 2306 // We can have macros or attributes in between 'enum' and the enum name. 2307 if (FormatTok->is(tok::l_paren)) 2308 parseParens(); 2309 if (FormatTok->is(tok::identifier)) { 2310 nextToken(); 2311 // If there are two identifiers in a row, this is likely an elaborate 2312 // return type. In Java, this can be "implements", etc. 2313 if (Style.isCpp() && FormatTok->is(tok::identifier)) 2314 return false; 2315 } 2316 } 2317 2318 // Just a declaration or something is wrong. 2319 if (FormatTok->isNot(tok::l_brace)) 2320 return true; 2321 FormatTok->BlockKind = BK_Block; 2322 2323 if (Style.Language == FormatStyle::LK_Java) { 2324 // Java enums are different. 2325 parseJavaEnumBody(); 2326 return true; 2327 } 2328 if (Style.Language == FormatStyle::LK_Proto) { 2329 parseBlock(/*MustBeDeclaration=*/true); 2330 return true; 2331 } 2332 2333 if (!Style.AllowShortEnumsOnASingleLine) 2334 addUnwrappedLine(); 2335 // Parse enum body. 2336 nextToken(); 2337 if (!Style.AllowShortEnumsOnASingleLine) { 2338 addUnwrappedLine(); 2339 Line->Level += 1; 2340 } 2341 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true, 2342 /*IsEnum=*/true); 2343 if (!Style.AllowShortEnumsOnASingleLine) 2344 Line->Level -= 1; 2345 if (HasError) { 2346 if (FormatTok->is(tok::semi)) 2347 nextToken(); 2348 addUnwrappedLine(); 2349 } 2350 return true; 2351 2352 // There is no addUnwrappedLine() here so that we fall through to parsing a 2353 // structural element afterwards. Thus, in "enum A {} n, m;", 2354 // "} n, m;" will end up in one unwrapped line. 2355 } 2356 2357 namespace { 2358 // A class used to set and restore the Token position when peeking 2359 // ahead in the token source. 2360 class ScopedTokenPosition { 2361 unsigned StoredPosition; 2362 FormatTokenSource *Tokens; 2363 2364 public: 2365 ScopedTokenPosition(FormatTokenSource *Tokens) : Tokens(Tokens) { 2366 assert(Tokens && "Tokens expected to not be null"); 2367 StoredPosition = Tokens->getPosition(); 2368 } 2369 2370 ~ScopedTokenPosition() { Tokens->setPosition(StoredPosition); } 2371 }; 2372 } // namespace 2373 2374 // Look to see if we have [[ by looking ahead, if 2375 // its not then rewind to the original position. 2376 bool UnwrappedLineParser::tryToParseSimpleAttribute() { 2377 ScopedTokenPosition AutoPosition(Tokens); 2378 FormatToken *Tok = Tokens->getNextToken(); 2379 // We already read the first [ check for the second. 2380 if (Tok && !Tok->is(tok::l_square)) { 2381 return false; 2382 } 2383 // Double check that the attribute is just something 2384 // fairly simple. 2385 while (Tok) { 2386 if (Tok->is(tok::r_square)) { 2387 break; 2388 } 2389 Tok = Tokens->getNextToken(); 2390 } 2391 Tok = Tokens->getNextToken(); 2392 if (Tok && !Tok->is(tok::r_square)) { 2393 return false; 2394 } 2395 Tok = Tokens->getNextToken(); 2396 if (Tok && Tok->is(tok::semi)) { 2397 return false; 2398 } 2399 return true; 2400 } 2401 2402 void UnwrappedLineParser::parseJavaEnumBody() { 2403 // Determine whether the enum is simple, i.e. does not have a semicolon or 2404 // constants with class bodies. Simple enums can be formatted like braced 2405 // lists, contracted to a single line, etc. 2406 unsigned StoredPosition = Tokens->getPosition(); 2407 bool IsSimple = true; 2408 FormatToken *Tok = Tokens->getNextToken(); 2409 while (Tok) { 2410 if (Tok->is(tok::r_brace)) 2411 break; 2412 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 2413 IsSimple = false; 2414 break; 2415 } 2416 // FIXME: This will also mark enums with braces in the arguments to enum 2417 // constants as "not simple". This is probably fine in practice, though. 2418 Tok = Tokens->getNextToken(); 2419 } 2420 FormatTok = Tokens->setPosition(StoredPosition); 2421 2422 if (IsSimple) { 2423 nextToken(); 2424 parseBracedList(); 2425 addUnwrappedLine(); 2426 return; 2427 } 2428 2429 // Parse the body of a more complex enum. 2430 // First add a line for everything up to the "{". 2431 nextToken(); 2432 addUnwrappedLine(); 2433 ++Line->Level; 2434 2435 // Parse the enum constants. 2436 while (FormatTok) { 2437 if (FormatTok->is(tok::l_brace)) { 2438 // Parse the constant's class body. 2439 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 2440 /*MunchSemi=*/false); 2441 } else if (FormatTok->is(tok::l_paren)) { 2442 parseParens(); 2443 } else if (FormatTok->is(tok::comma)) { 2444 nextToken(); 2445 addUnwrappedLine(); 2446 } else if (FormatTok->is(tok::semi)) { 2447 nextToken(); 2448 addUnwrappedLine(); 2449 break; 2450 } else if (FormatTok->is(tok::r_brace)) { 2451 addUnwrappedLine(); 2452 break; 2453 } else { 2454 nextToken(); 2455 } 2456 } 2457 2458 // Parse the class body after the enum's ";" if any. 2459 parseLevel(/*HasOpeningBrace=*/true); 2460 nextToken(); 2461 --Line->Level; 2462 addUnwrappedLine(); 2463 } 2464 2465 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) { 2466 const FormatToken &InitialToken = *FormatTok; 2467 nextToken(); 2468 2469 // The actual identifier can be a nested name specifier, and in macros 2470 // it is often token-pasted. 2471 // An [[attribute]] can be before the identifier. 2472 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 2473 tok::kw___attribute, tok::kw___declspec, 2474 tok::kw_alignas, tok::l_square, tok::r_square) || 2475 ((Style.Language == FormatStyle::LK_Java || 2476 Style.Language == FormatStyle::LK_JavaScript) && 2477 FormatTok->isOneOf(tok::period, tok::comma))) { 2478 if (Style.Language == FormatStyle::LK_JavaScript && 2479 FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) { 2480 // JavaScript/TypeScript supports inline object types in 2481 // extends/implements positions: 2482 // class Foo implements {bar: number} { } 2483 nextToken(); 2484 if (FormatTok->is(tok::l_brace)) { 2485 tryToParseBracedList(); 2486 continue; 2487 } 2488 } 2489 bool IsNonMacroIdentifier = 2490 FormatTok->is(tok::identifier) && 2491 FormatTok->TokenText != FormatTok->TokenText.upper(); 2492 nextToken(); 2493 // We can have macros or attributes in between 'class' and the class name. 2494 if (!IsNonMacroIdentifier) { 2495 if (FormatTok->Tok.is(tok::l_paren)) { 2496 parseParens(); 2497 } else if (FormatTok->is(TT_AttributeSquare)) { 2498 parseSquare(); 2499 // Consume the closing TT_AttributeSquare. 2500 if (FormatTok->Next && FormatTok->is(TT_AttributeSquare)) 2501 nextToken(); 2502 } 2503 } 2504 } 2505 2506 // Note that parsing away template declarations here leads to incorrectly 2507 // accepting function declarations as record declarations. 2508 // In general, we cannot solve this problem. Consider: 2509 // class A<int> B() {} 2510 // which can be a function definition or a class definition when B() is a 2511 // macro. If we find enough real-world cases where this is a problem, we 2512 // can parse for the 'template' keyword in the beginning of the statement, 2513 // and thus rule out the record production in case there is no template 2514 // (this would still leave us with an ambiguity between template function 2515 // and class declarations). 2516 if (FormatTok->isOneOf(tok::colon, tok::less)) { 2517 while (!eof()) { 2518 if (FormatTok->is(tok::l_brace)) { 2519 calculateBraceTypes(/*ExpectClassBody=*/true); 2520 if (!tryToParseBracedList()) 2521 break; 2522 } 2523 if (FormatTok->Tok.is(tok::semi)) 2524 return; 2525 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where)) { 2526 addUnwrappedLine(); 2527 nextToken(); 2528 parseCSharpGenericTypeConstraint(); 2529 break; 2530 } 2531 nextToken(); 2532 } 2533 } 2534 if (FormatTok->Tok.is(tok::l_brace)) { 2535 if (ParseAsExpr) { 2536 parseChildBlock(); 2537 } else { 2538 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2539 addUnwrappedLine(); 2540 2541 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 2542 /*MunchSemi=*/false); 2543 } 2544 } 2545 // There is no addUnwrappedLine() here so that we fall through to parsing a 2546 // structural element afterwards. Thus, in "class A {} n, m;", 2547 // "} n, m;" will end up in one unwrapped line. 2548 } 2549 2550 void UnwrappedLineParser::parseObjCMethod() { 2551 assert(FormatTok->Tok.isOneOf(tok::l_paren, tok::identifier) && 2552 "'(' or identifier expected."); 2553 do { 2554 if (FormatTok->Tok.is(tok::semi)) { 2555 nextToken(); 2556 addUnwrappedLine(); 2557 return; 2558 } else if (FormatTok->Tok.is(tok::l_brace)) { 2559 if (Style.BraceWrapping.AfterFunction) 2560 addUnwrappedLine(); 2561 parseBlock(/*MustBeDeclaration=*/false); 2562 addUnwrappedLine(); 2563 return; 2564 } else { 2565 nextToken(); 2566 } 2567 } while (!eof()); 2568 } 2569 2570 void UnwrappedLineParser::parseObjCProtocolList() { 2571 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 2572 do { 2573 nextToken(); 2574 // Early exit in case someone forgot a close angle. 2575 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 2576 FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) 2577 return; 2578 } while (!eof() && FormatTok->Tok.isNot(tok::greater)); 2579 nextToken(); // Skip '>'. 2580 } 2581 2582 void UnwrappedLineParser::parseObjCUntilAtEnd() { 2583 do { 2584 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 2585 nextToken(); 2586 addUnwrappedLine(); 2587 break; 2588 } 2589 if (FormatTok->is(tok::l_brace)) { 2590 parseBlock(/*MustBeDeclaration=*/false); 2591 // In ObjC interfaces, nothing should be following the "}". 2592 addUnwrappedLine(); 2593 } else if (FormatTok->is(tok::r_brace)) { 2594 // Ignore stray "}". parseStructuralElement doesn't consume them. 2595 nextToken(); 2596 addUnwrappedLine(); 2597 } else if (FormatTok->isOneOf(tok::minus, tok::plus)) { 2598 nextToken(); 2599 parseObjCMethod(); 2600 } else { 2601 parseStructuralElement(); 2602 } 2603 } while (!eof()); 2604 } 2605 2606 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 2607 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_interface || 2608 FormatTok->Tok.getObjCKeywordID() == tok::objc_implementation); 2609 nextToken(); 2610 nextToken(); // interface name 2611 2612 // @interface can be followed by a lightweight generic 2613 // specialization list, then either a base class or a category. 2614 if (FormatTok->Tok.is(tok::less)) { 2615 // Unlike protocol lists, generic parameterizations support 2616 // nested angles: 2617 // 2618 // @interface Foo<ValueType : id <NSCopying, NSSecureCoding>> : 2619 // NSObject <NSCopying, NSSecureCoding> 2620 // 2621 // so we need to count how many open angles we have left. 2622 unsigned NumOpenAngles = 1; 2623 do { 2624 nextToken(); 2625 // Early exit in case someone forgot a close angle. 2626 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 2627 FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) 2628 break; 2629 if (FormatTok->Tok.is(tok::less)) 2630 ++NumOpenAngles; 2631 else if (FormatTok->Tok.is(tok::greater)) { 2632 assert(NumOpenAngles > 0 && "'>' makes NumOpenAngles negative"); 2633 --NumOpenAngles; 2634 } 2635 } while (!eof() && NumOpenAngles != 0); 2636 nextToken(); // Skip '>'. 2637 } 2638 if (FormatTok->Tok.is(tok::colon)) { 2639 nextToken(); 2640 nextToken(); // base class name 2641 } else if (FormatTok->Tok.is(tok::l_paren)) 2642 // Skip category, if present. 2643 parseParens(); 2644 2645 if (FormatTok->Tok.is(tok::less)) 2646 parseObjCProtocolList(); 2647 2648 if (FormatTok->Tok.is(tok::l_brace)) { 2649 if (Style.BraceWrapping.AfterObjCDeclaration) 2650 addUnwrappedLine(); 2651 parseBlock(/*MustBeDeclaration=*/true); 2652 } 2653 2654 // With instance variables, this puts '}' on its own line. Without instance 2655 // variables, this ends the @interface line. 2656 addUnwrappedLine(); 2657 2658 parseObjCUntilAtEnd(); 2659 } 2660 2661 // Returns true for the declaration/definition form of @protocol, 2662 // false for the expression form. 2663 bool UnwrappedLineParser::parseObjCProtocol() { 2664 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_protocol); 2665 nextToken(); 2666 2667 if (FormatTok->is(tok::l_paren)) 2668 // The expression form of @protocol, e.g. "Protocol* p = @protocol(foo);". 2669 return false; 2670 2671 // The definition/declaration form, 2672 // @protocol Foo 2673 // - (int)someMethod; 2674 // @end 2675 2676 nextToken(); // protocol name 2677 2678 if (FormatTok->Tok.is(tok::less)) 2679 parseObjCProtocolList(); 2680 2681 // Check for protocol declaration. 2682 if (FormatTok->Tok.is(tok::semi)) { 2683 nextToken(); 2684 addUnwrappedLine(); 2685 return true; 2686 } 2687 2688 addUnwrappedLine(); 2689 parseObjCUntilAtEnd(); 2690 return true; 2691 } 2692 2693 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 2694 bool IsImport = FormatTok->is(Keywords.kw_import); 2695 assert(IsImport || FormatTok->is(tok::kw_export)); 2696 nextToken(); 2697 2698 // Consume the "default" in "export default class/function". 2699 if (FormatTok->is(tok::kw_default)) 2700 nextToken(); 2701 2702 // Consume "async function", "function" and "default function", so that these 2703 // get parsed as free-standing JS functions, i.e. do not require a trailing 2704 // semicolon. 2705 if (FormatTok->is(Keywords.kw_async)) 2706 nextToken(); 2707 if (FormatTok->is(Keywords.kw_function)) { 2708 nextToken(); 2709 return; 2710 } 2711 2712 // For imports, `export *`, `export {...}`, consume the rest of the line up 2713 // to the terminating `;`. For everything else, just return and continue 2714 // parsing the structural element, i.e. the declaration or expression for 2715 // `export default`. 2716 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 2717 !FormatTok->isStringLiteral()) 2718 return; 2719 2720 while (!eof()) { 2721 if (FormatTok->is(tok::semi)) 2722 return; 2723 if (Line->Tokens.empty()) { 2724 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 2725 // import statement should terminate. 2726 return; 2727 } 2728 if (FormatTok->is(tok::l_brace)) { 2729 FormatTok->BlockKind = BK_Block; 2730 nextToken(); 2731 parseBracedList(); 2732 } else { 2733 nextToken(); 2734 } 2735 } 2736 } 2737 2738 void UnwrappedLineParser::parseStatementMacro() { 2739 nextToken(); 2740 if (FormatTok->is(tok::l_paren)) 2741 parseParens(); 2742 if (FormatTok->is(tok::semi)) 2743 nextToken(); 2744 addUnwrappedLine(); 2745 } 2746 2747 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 2748 StringRef Prefix = "") { 2749 llvm::dbgs() << Prefix << "Line(" << Line.Level 2750 << ", FSC=" << Line.FirstStartColumn << ")" 2751 << (Line.InPPDirective ? " MACRO" : "") << ": "; 2752 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2753 E = Line.Tokens.end(); 2754 I != E; ++I) { 2755 llvm::dbgs() << I->Tok->Tok.getName() << "[" 2756 << "T=" << I->Tok->getType() 2757 << ", OC=" << I->Tok->OriginalColumn << "] "; 2758 } 2759 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2760 E = Line.Tokens.end(); 2761 I != E; ++I) { 2762 const UnwrappedLineNode &Node = *I; 2763 for (SmallVectorImpl<UnwrappedLine>::const_iterator 2764 I = Node.Children.begin(), 2765 E = Node.Children.end(); 2766 I != E; ++I) { 2767 printDebugInfo(*I, "\nChild: "); 2768 } 2769 } 2770 llvm::dbgs() << "\n"; 2771 } 2772 2773 void UnwrappedLineParser::addUnwrappedLine() { 2774 if (Line->Tokens.empty()) 2775 return; 2776 LLVM_DEBUG({ 2777 if (CurrentLines == &Lines) 2778 printDebugInfo(*Line); 2779 }); 2780 CurrentLines->push_back(std::move(*Line)); 2781 Line->Tokens.clear(); 2782 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex; 2783 Line->FirstStartColumn = 0; 2784 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 2785 CurrentLines->append( 2786 std::make_move_iterator(PreprocessorDirectives.begin()), 2787 std::make_move_iterator(PreprocessorDirectives.end())); 2788 PreprocessorDirectives.clear(); 2789 } 2790 // Disconnect the current token from the last token on the previous line. 2791 FormatTok->Previous = nullptr; 2792 } 2793 2794 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 2795 2796 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 2797 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 2798 FormatTok.NewlinesBefore > 0; 2799 } 2800 2801 // Checks if \p FormatTok is a line comment that continues the line comment 2802 // section on \p Line. 2803 static bool 2804 continuesLineCommentSection(const FormatToken &FormatTok, 2805 const UnwrappedLine &Line, 2806 const llvm::Regex &CommentPragmasRegex) { 2807 if (Line.Tokens.empty()) 2808 return false; 2809 2810 StringRef IndentContent = FormatTok.TokenText; 2811 if (FormatTok.TokenText.startswith("//") || 2812 FormatTok.TokenText.startswith("/*")) 2813 IndentContent = FormatTok.TokenText.substr(2); 2814 if (CommentPragmasRegex.match(IndentContent)) 2815 return false; 2816 2817 // If Line starts with a line comment, then FormatTok continues the comment 2818 // section if its original column is greater or equal to the original start 2819 // column of the line. 2820 // 2821 // Define the min column token of a line as follows: if a line ends in '{' or 2822 // contains a '{' followed by a line comment, then the min column token is 2823 // that '{'. Otherwise, the min column token of the line is the first token of 2824 // the line. 2825 // 2826 // If Line starts with a token other than a line comment, then FormatTok 2827 // continues the comment section if its original column is greater than the 2828 // original start column of the min column token of the line. 2829 // 2830 // For example, the second line comment continues the first in these cases: 2831 // 2832 // // first line 2833 // // second line 2834 // 2835 // and: 2836 // 2837 // // first line 2838 // // second line 2839 // 2840 // and: 2841 // 2842 // int i; // first line 2843 // // second line 2844 // 2845 // and: 2846 // 2847 // do { // first line 2848 // // second line 2849 // int i; 2850 // } while (true); 2851 // 2852 // and: 2853 // 2854 // enum { 2855 // a, // first line 2856 // // second line 2857 // b 2858 // }; 2859 // 2860 // The second line comment doesn't continue the first in these cases: 2861 // 2862 // // first line 2863 // // second line 2864 // 2865 // and: 2866 // 2867 // int i; // first line 2868 // // second line 2869 // 2870 // and: 2871 // 2872 // do { // first line 2873 // // second line 2874 // int i; 2875 // } while (true); 2876 // 2877 // and: 2878 // 2879 // enum { 2880 // a, // first line 2881 // // second line 2882 // }; 2883 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 2884 2885 // Scan for '{//'. If found, use the column of '{' as a min column for line 2886 // comment section continuation. 2887 const FormatToken *PreviousToken = nullptr; 2888 for (const UnwrappedLineNode &Node : Line.Tokens) { 2889 if (PreviousToken && PreviousToken->is(tok::l_brace) && 2890 isLineComment(*Node.Tok)) { 2891 MinColumnToken = PreviousToken; 2892 break; 2893 } 2894 PreviousToken = Node.Tok; 2895 2896 // Grab the last newline preceding a token in this unwrapped line. 2897 if (Node.Tok->NewlinesBefore > 0) { 2898 MinColumnToken = Node.Tok; 2899 } 2900 } 2901 if (PreviousToken && PreviousToken->is(tok::l_brace)) { 2902 MinColumnToken = PreviousToken; 2903 } 2904 2905 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok, 2906 MinColumnToken); 2907 } 2908 2909 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 2910 bool JustComments = Line->Tokens.empty(); 2911 for (SmallVectorImpl<FormatToken *>::const_iterator 2912 I = CommentsBeforeNextToken.begin(), 2913 E = CommentsBeforeNextToken.end(); 2914 I != E; ++I) { 2915 // Line comments that belong to the same line comment section are put on the 2916 // same line since later we might want to reflow content between them. 2917 // Additional fine-grained breaking of line comment sections is controlled 2918 // by the class BreakableLineCommentSection in case it is desirable to keep 2919 // several line comment sections in the same unwrapped line. 2920 // 2921 // FIXME: Consider putting separate line comment sections as children to the 2922 // unwrapped line instead. 2923 (*I)->ContinuesLineCommentSection = 2924 continuesLineCommentSection(**I, *Line, CommentPragmasRegex); 2925 if (isOnNewLine(**I) && JustComments && !(*I)->ContinuesLineCommentSection) 2926 addUnwrappedLine(); 2927 pushToken(*I); 2928 } 2929 if (NewlineBeforeNext && JustComments) 2930 addUnwrappedLine(); 2931 CommentsBeforeNextToken.clear(); 2932 } 2933 2934 void UnwrappedLineParser::nextToken(int LevelDifference) { 2935 if (eof()) 2936 return; 2937 flushComments(isOnNewLine(*FormatTok)); 2938 pushToken(FormatTok); 2939 FormatToken *Previous = FormatTok; 2940 if (Style.Language != FormatStyle::LK_JavaScript) 2941 readToken(LevelDifference); 2942 else 2943 readTokenWithJavaScriptASI(); 2944 FormatTok->Previous = Previous; 2945 } 2946 2947 void UnwrappedLineParser::distributeComments( 2948 const SmallVectorImpl<FormatToken *> &Comments, 2949 const FormatToken *NextTok) { 2950 // Whether or not a line comment token continues a line is controlled by 2951 // the method continuesLineCommentSection, with the following caveat: 2952 // 2953 // Define a trail of Comments to be a nonempty proper postfix of Comments such 2954 // that each comment line from the trail is aligned with the next token, if 2955 // the next token exists. If a trail exists, the beginning of the maximal 2956 // trail is marked as a start of a new comment section. 2957 // 2958 // For example in this code: 2959 // 2960 // int a; // line about a 2961 // // line 1 about b 2962 // // line 2 about b 2963 // int b; 2964 // 2965 // the two lines about b form a maximal trail, so there are two sections, the 2966 // first one consisting of the single comment "// line about a" and the 2967 // second one consisting of the next two comments. 2968 if (Comments.empty()) 2969 return; 2970 bool ShouldPushCommentsInCurrentLine = true; 2971 bool HasTrailAlignedWithNextToken = false; 2972 unsigned StartOfTrailAlignedWithNextToken = 0; 2973 if (NextTok) { 2974 // We are skipping the first element intentionally. 2975 for (unsigned i = Comments.size() - 1; i > 0; --i) { 2976 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) { 2977 HasTrailAlignedWithNextToken = true; 2978 StartOfTrailAlignedWithNextToken = i; 2979 } 2980 } 2981 } 2982 for (unsigned i = 0, e = Comments.size(); i < e; ++i) { 2983 FormatToken *FormatTok = Comments[i]; 2984 if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) { 2985 FormatTok->ContinuesLineCommentSection = false; 2986 } else { 2987 FormatTok->ContinuesLineCommentSection = 2988 continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex); 2989 } 2990 if (!FormatTok->ContinuesLineCommentSection && 2991 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) { 2992 ShouldPushCommentsInCurrentLine = false; 2993 } 2994 if (ShouldPushCommentsInCurrentLine) { 2995 pushToken(FormatTok); 2996 } else { 2997 CommentsBeforeNextToken.push_back(FormatTok); 2998 } 2999 } 3000 } 3001 3002 void UnwrappedLineParser::readToken(int LevelDifference) { 3003 SmallVector<FormatToken *, 1> Comments; 3004 do { 3005 FormatTok = Tokens->getNextToken(); 3006 assert(FormatTok); 3007 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 3008 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 3009 distributeComments(Comments, FormatTok); 3010 Comments.clear(); 3011 // If there is an unfinished unwrapped line, we flush the preprocessor 3012 // directives only after that unwrapped line was finished later. 3013 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 3014 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 3015 assert((LevelDifference >= 0 || 3016 static_cast<unsigned>(-LevelDifference) <= Line->Level) && 3017 "LevelDifference makes Line->Level negative"); 3018 Line->Level += LevelDifference; 3019 // Comments stored before the preprocessor directive need to be output 3020 // before the preprocessor directive, at the same level as the 3021 // preprocessor directive, as we consider them to apply to the directive. 3022 if (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash && 3023 PPBranchLevel > 0) 3024 Line->Level += PPBranchLevel; 3025 flushComments(isOnNewLine(*FormatTok)); 3026 parsePPDirective(); 3027 } 3028 while (FormatTok->getType() == TT_ConflictStart || 3029 FormatTok->getType() == TT_ConflictEnd || 3030 FormatTok->getType() == TT_ConflictAlternative) { 3031 if (FormatTok->getType() == TT_ConflictStart) { 3032 conditionalCompilationStart(/*Unreachable=*/false); 3033 } else if (FormatTok->getType() == TT_ConflictAlternative) { 3034 conditionalCompilationAlternative(); 3035 } else if (FormatTok->getType() == TT_ConflictEnd) { 3036 conditionalCompilationEnd(); 3037 } 3038 FormatTok = Tokens->getNextToken(); 3039 FormatTok->MustBreakBefore = true; 3040 } 3041 3042 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) && 3043 !Line->InPPDirective) { 3044 continue; 3045 } 3046 3047 if (!FormatTok->Tok.is(tok::comment)) { 3048 distributeComments(Comments, FormatTok); 3049 Comments.clear(); 3050 return; 3051 } 3052 3053 Comments.push_back(FormatTok); 3054 } while (!eof()); 3055 3056 distributeComments(Comments, nullptr); 3057 Comments.clear(); 3058 } 3059 3060 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 3061 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 3062 if (MustBreakBeforeNextToken) { 3063 Line->Tokens.back().Tok->MustBreakBefore = true; 3064 MustBreakBeforeNextToken = false; 3065 } 3066 } 3067 3068 } // end namespace format 3069 } // end namespace clang 3070