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->Type == TT_MacroBlockBegin) { 373 kind = tok::l_brace; 374 } else if (FormatTok->Type == 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 blacklist 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 clang::Token &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(tok::l_square) && 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->Type = TT_InlineASMBrace; 1037 nextToken(); 1038 while (FormatTok && FormatTok->isNot(tok::eof)) { 1039 if (FormatTok->is(tok::r_brace)) { 1040 FormatTok->Type = 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.BraceWrapping.AfterExternBlock) { 1117 addUnwrappedLine(); 1118 parseBlock(/*MustBeDeclaration=*/true); 1119 } else { 1120 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/false); 1121 } 1122 addUnwrappedLine(); 1123 return; 1124 } 1125 } 1126 break; 1127 case tok::kw_export: 1128 if (Style.Language == FormatStyle::LK_JavaScript) { 1129 parseJavaScriptEs6ImportExport(); 1130 return; 1131 } 1132 if (!Style.isCpp()) 1133 break; 1134 // Handle C++ "(inline|export) namespace". 1135 LLVM_FALLTHROUGH; 1136 case tok::kw_inline: 1137 nextToken(); 1138 if (FormatTok->Tok.is(tok::kw_namespace)) { 1139 parseNamespace(); 1140 return; 1141 } 1142 break; 1143 case tok::identifier: 1144 if (FormatTok->is(TT_ForEachMacro)) { 1145 parseForOrWhileLoop(); 1146 return; 1147 } 1148 if (FormatTok->is(TT_MacroBlockBegin)) { 1149 parseBlock(/*MustBeDeclaration=*/false, /*AddLevel=*/true, 1150 /*MunchSemi=*/false); 1151 return; 1152 } 1153 if (FormatTok->is(Keywords.kw_import)) { 1154 if (Style.Language == FormatStyle::LK_JavaScript) { 1155 parseJavaScriptEs6ImportExport(); 1156 return; 1157 } 1158 if (Style.Language == FormatStyle::LK_Proto) { 1159 nextToken(); 1160 if (FormatTok->is(tok::kw_public)) 1161 nextToken(); 1162 if (!FormatTok->is(tok::string_literal)) 1163 return; 1164 nextToken(); 1165 if (FormatTok->is(tok::semi)) 1166 nextToken(); 1167 addUnwrappedLine(); 1168 return; 1169 } 1170 } 1171 if (Style.isCpp() && 1172 FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 1173 Keywords.kw_slots, Keywords.kw_qslots)) { 1174 nextToken(); 1175 if (FormatTok->is(tok::colon)) { 1176 nextToken(); 1177 addUnwrappedLine(); 1178 return; 1179 } 1180 } 1181 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1182 parseStatementMacro(); 1183 return; 1184 } 1185 if (Style.isCpp() && FormatTok->is(TT_NamespaceMacro)) { 1186 parseNamespace(); 1187 return; 1188 } 1189 // In all other cases, parse the declaration. 1190 break; 1191 default: 1192 break; 1193 } 1194 do { 1195 const FormatToken *Previous = FormatTok->Previous; 1196 switch (FormatTok->Tok.getKind()) { 1197 case tok::at: 1198 nextToken(); 1199 if (FormatTok->Tok.is(tok::l_brace)) { 1200 nextToken(); 1201 parseBracedList(); 1202 break; 1203 } else if (Style.Language == FormatStyle::LK_Java && 1204 FormatTok->is(Keywords.kw_interface)) { 1205 nextToken(); 1206 break; 1207 } 1208 switch (FormatTok->Tok.getObjCKeywordID()) { 1209 case tok::objc_public: 1210 case tok::objc_protected: 1211 case tok::objc_package: 1212 case tok::objc_private: 1213 return parseAccessSpecifier(); 1214 case tok::objc_interface: 1215 case tok::objc_implementation: 1216 return parseObjCInterfaceOrImplementation(); 1217 case tok::objc_protocol: 1218 if (parseObjCProtocol()) 1219 return; 1220 break; 1221 case tok::objc_end: 1222 return; // Handled by the caller. 1223 case tok::objc_optional: 1224 case tok::objc_required: 1225 nextToken(); 1226 addUnwrappedLine(); 1227 return; 1228 case tok::objc_autoreleasepool: 1229 nextToken(); 1230 if (FormatTok->Tok.is(tok::l_brace)) { 1231 if (Style.BraceWrapping.AfterControlStatement == 1232 FormatStyle::BWACS_Always) 1233 addUnwrappedLine(); 1234 parseBlock(/*MustBeDeclaration=*/false); 1235 } 1236 addUnwrappedLine(); 1237 return; 1238 case tok::objc_synchronized: 1239 nextToken(); 1240 if (FormatTok->Tok.is(tok::l_paren)) 1241 // Skip synchronization object 1242 parseParens(); 1243 if (FormatTok->Tok.is(tok::l_brace)) { 1244 if (Style.BraceWrapping.AfterControlStatement == 1245 FormatStyle::BWACS_Always) 1246 addUnwrappedLine(); 1247 parseBlock(/*MustBeDeclaration=*/false); 1248 } 1249 addUnwrappedLine(); 1250 return; 1251 case tok::objc_try: 1252 // This branch isn't strictly necessary (the kw_try case below would 1253 // do this too after the tok::at is parsed above). But be explicit. 1254 parseTryCatch(); 1255 return; 1256 default: 1257 break; 1258 } 1259 break; 1260 case tok::kw_enum: 1261 // Ignore if this is part of "template <enum ...". 1262 if (Previous && Previous->is(tok::less)) { 1263 nextToken(); 1264 break; 1265 } 1266 1267 // parseEnum falls through and does not yet add an unwrapped line as an 1268 // enum definition can start a structural element. 1269 if (!parseEnum()) 1270 break; 1271 // This only applies for C++. 1272 if (!Style.isCpp()) { 1273 addUnwrappedLine(); 1274 return; 1275 } 1276 break; 1277 case tok::kw_typedef: 1278 nextToken(); 1279 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 1280 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS, 1281 Keywords.kw_CF_CLOSED_ENUM, 1282 Keywords.kw_NS_CLOSED_ENUM)) 1283 parseEnum(); 1284 break; 1285 case tok::kw_struct: 1286 case tok::kw_union: 1287 case tok::kw_class: 1288 // parseRecord falls through and does not yet add an unwrapped line as a 1289 // record declaration or definition can start a structural element. 1290 parseRecord(); 1291 // This does not apply for Java, JavaScript and C#. 1292 if (Style.Language == FormatStyle::LK_Java || 1293 Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp()) { 1294 if (FormatTok->is(tok::semi)) 1295 nextToken(); 1296 addUnwrappedLine(); 1297 return; 1298 } 1299 break; 1300 case tok::period: 1301 nextToken(); 1302 // In Java, classes have an implicit static member "class". 1303 if (Style.Language == FormatStyle::LK_Java && FormatTok && 1304 FormatTok->is(tok::kw_class)) 1305 nextToken(); 1306 if (Style.Language == FormatStyle::LK_JavaScript && FormatTok && 1307 FormatTok->Tok.getIdentifierInfo()) 1308 // JavaScript only has pseudo keywords, all keywords are allowed to 1309 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 1310 nextToken(); 1311 break; 1312 case tok::semi: 1313 nextToken(); 1314 addUnwrappedLine(); 1315 return; 1316 case tok::r_brace: 1317 addUnwrappedLine(); 1318 return; 1319 case tok::l_paren: 1320 parseParens(); 1321 break; 1322 case tok::kw_operator: 1323 nextToken(); 1324 if (FormatTok->isBinaryOperator()) 1325 nextToken(); 1326 break; 1327 case tok::caret: 1328 nextToken(); 1329 if (FormatTok->Tok.isAnyIdentifier() || 1330 FormatTok->isSimpleTypeSpecifier()) 1331 nextToken(); 1332 if (FormatTok->is(tok::l_paren)) 1333 parseParens(); 1334 if (FormatTok->is(tok::l_brace)) 1335 parseChildBlock(); 1336 break; 1337 case tok::l_brace: 1338 if (!tryToParsePropertyAccessor() && !tryToParseBracedList()) { 1339 // A block outside of parentheses must be the last part of a 1340 // structural element. 1341 // FIXME: Figure out cases where this is not true, and add projections 1342 // for them (the one we know is missing are lambdas). 1343 if (Style.BraceWrapping.AfterFunction) 1344 addUnwrappedLine(); 1345 FormatTok->Type = TT_FunctionLBrace; 1346 parseBlock(/*MustBeDeclaration=*/false); 1347 addUnwrappedLine(); 1348 return; 1349 } 1350 // Otherwise this was a braced init list, and the structural 1351 // element continues. 1352 break; 1353 case tok::kw_try: 1354 if (Style.Language == FormatStyle::LK_JavaScript && 1355 Line->MustBeDeclaration) { 1356 // field/method declaration. 1357 nextToken(); 1358 break; 1359 } 1360 // We arrive here when parsing function-try blocks. 1361 if (Style.BraceWrapping.AfterFunction) 1362 addUnwrappedLine(); 1363 parseTryCatch(); 1364 return; 1365 case tok::identifier: { 1366 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where) && 1367 Line->MustBeDeclaration) { 1368 addUnwrappedLine(); 1369 parseCSharpGenericTypeConstraint(); 1370 break; 1371 } 1372 if (FormatTok->is(TT_MacroBlockEnd)) { 1373 addUnwrappedLine(); 1374 return; 1375 } 1376 1377 // Function declarations (as opposed to function expressions) are parsed 1378 // on their own unwrapped line by continuing this loop. Function 1379 // expressions (functions that are not on their own line) must not create 1380 // a new unwrapped line, so they are special cased below. 1381 size_t TokenCount = Line->Tokens.size(); 1382 if (Style.Language == FormatStyle::LK_JavaScript && 1383 FormatTok->is(Keywords.kw_function) && 1384 (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is( 1385 Keywords.kw_async)))) { 1386 tryToParseJSFunction(); 1387 break; 1388 } 1389 if ((Style.Language == FormatStyle::LK_JavaScript || 1390 Style.Language == FormatStyle::LK_Java) && 1391 FormatTok->is(Keywords.kw_interface)) { 1392 if (Style.Language == FormatStyle::LK_JavaScript) { 1393 // In JavaScript/TypeScript, "interface" can be used as a standalone 1394 // identifier, e.g. in `var interface = 1;`. If "interface" is 1395 // followed by another identifier, it is very like to be an actual 1396 // interface declaration. 1397 unsigned StoredPosition = Tokens->getPosition(); 1398 FormatToken *Next = Tokens->getNextToken(); 1399 FormatTok = Tokens->setPosition(StoredPosition); 1400 if (Next && !mustBeJSIdent(Keywords, Next)) { 1401 nextToken(); 1402 break; 1403 } 1404 } 1405 parseRecord(); 1406 addUnwrappedLine(); 1407 return; 1408 } 1409 1410 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1411 parseStatementMacro(); 1412 return; 1413 } 1414 1415 // See if the following token should start a new unwrapped line. 1416 StringRef Text = FormatTok->TokenText; 1417 nextToken(); 1418 1419 // JS doesn't have macros, and within classes colons indicate fields, not 1420 // labels. 1421 if (Style.Language == FormatStyle::LK_JavaScript) 1422 break; 1423 1424 TokenCount = Line->Tokens.size(); 1425 if (TokenCount == 1 || 1426 (TokenCount == 2 && Line->Tokens.front().Tok->is(tok::comment))) { 1427 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) { 1428 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1429 parseLabel(!Style.IndentGotoLabels); 1430 return; 1431 } 1432 // Recognize function-like macro usages without trailing semicolon as 1433 // well as free-standing macros like Q_OBJECT. 1434 bool FunctionLike = FormatTok->is(tok::l_paren); 1435 if (FunctionLike) 1436 parseParens(); 1437 1438 bool FollowedByNewline = 1439 CommentsBeforeNextToken.empty() 1440 ? FormatTok->NewlinesBefore > 0 1441 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1442 1443 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1444 tokenCanStartNewLine(FormatTok->Tok) && Text == Text.upper()) { 1445 addUnwrappedLine(); 1446 return; 1447 } 1448 } 1449 break; 1450 } 1451 case tok::equal: 1452 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType 1453 // TT_JsFatArrow. The always start an expression or a child block if 1454 // followed by a curly. 1455 if (FormatTok->is(TT_JsFatArrow)) { 1456 nextToken(); 1457 if (FormatTok->is(tok::l_brace)) 1458 parseChildBlock(); 1459 break; 1460 } 1461 1462 nextToken(); 1463 if (FormatTok->Tok.is(tok::l_brace)) { 1464 // Block kind should probably be set to BK_BracedInit for any language. 1465 // C# needs this change to ensure that array initialisers and object 1466 // initialisers are indented the same way. 1467 if (Style.isCSharp()) 1468 FormatTok->BlockKind = BK_BracedInit; 1469 nextToken(); 1470 parseBracedList(); 1471 } else if (Style.Language == FormatStyle::LK_Proto && 1472 FormatTok->Tok.is(tok::less)) { 1473 nextToken(); 1474 parseBracedList(/*ContinueOnSemicolons=*/false, 1475 /*ClosingBraceKind=*/tok::greater); 1476 } 1477 break; 1478 case tok::l_square: 1479 parseSquare(); 1480 break; 1481 case tok::kw_new: 1482 parseNew(); 1483 break; 1484 default: 1485 nextToken(); 1486 break; 1487 } 1488 } while (!eof()); 1489 } 1490 1491 bool UnwrappedLineParser::tryToParsePropertyAccessor() { 1492 assert(FormatTok->is(tok::l_brace)); 1493 if (!Style.isCSharp()) 1494 return false; 1495 // See if it's a property accessor. 1496 if (FormatTok->Previous->isNot(tok::identifier)) 1497 return false; 1498 1499 // See if we are inside a property accessor. 1500 // 1501 // Record the current tokenPosition so that we can advance and 1502 // reset the current token. `Next` is not set yet so we need 1503 // another way to advance along the token stream. 1504 unsigned int StoredPosition = Tokens->getPosition(); 1505 FormatToken *Tok = Tokens->getNextToken(); 1506 1507 // A trivial property accessor is of the form: 1508 // { [ACCESS_SPECIFIER] [get]; [ACCESS_SPECIFIER] [set] } 1509 // Track these as they do not require line breaks to be introduced. 1510 bool HasGetOrSet = false; 1511 bool IsTrivialPropertyAccessor = true; 1512 while (!eof()) { 1513 if (Tok->isOneOf(tok::semi, tok::kw_public, tok::kw_private, 1514 tok::kw_protected, Keywords.kw_internal, Keywords.kw_get, 1515 Keywords.kw_set)) { 1516 if (Tok->isOneOf(Keywords.kw_get, Keywords.kw_set)) 1517 HasGetOrSet = true; 1518 Tok = Tokens->getNextToken(); 1519 continue; 1520 } 1521 if (Tok->isNot(tok::r_brace)) 1522 IsTrivialPropertyAccessor = false; 1523 break; 1524 } 1525 1526 if (!HasGetOrSet) { 1527 Tokens->setPosition(StoredPosition); 1528 return false; 1529 } 1530 1531 // Try to parse the property accessor: 1532 // https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties 1533 Tokens->setPosition(StoredPosition); 1534 nextToken(); 1535 do { 1536 switch (FormatTok->Tok.getKind()) { 1537 case tok::r_brace: 1538 nextToken(); 1539 if (FormatTok->is(tok::equal)) { 1540 while (!eof() && FormatTok->isNot(tok::semi)) 1541 nextToken(); 1542 nextToken(); 1543 } 1544 addUnwrappedLine(); 1545 return true; 1546 case tok::l_brace: 1547 ++Line->Level; 1548 parseBlock(/*MustBeDeclaration=*/true); 1549 addUnwrappedLine(); 1550 --Line->Level; 1551 break; 1552 case tok::equal: 1553 if (FormatTok->is(TT_JsFatArrow)) { 1554 ++Line->Level; 1555 do { 1556 nextToken(); 1557 } while (!eof() && FormatTok->isNot(tok::semi)); 1558 nextToken(); 1559 addUnwrappedLine(); 1560 --Line->Level; 1561 break; 1562 } 1563 nextToken(); 1564 break; 1565 default: 1566 if (FormatTok->isOneOf(Keywords.kw_get, Keywords.kw_set) && 1567 !IsTrivialPropertyAccessor) { 1568 // Non-trivial get/set needs to be on its own line. 1569 addUnwrappedLine(); 1570 } 1571 nextToken(); 1572 } 1573 } while (!eof()); 1574 1575 // Unreachable for well-formed code (paired '{' and '}'). 1576 return true; 1577 } 1578 1579 bool UnwrappedLineParser::tryToParseLambda() { 1580 if (!Style.isCpp()) { 1581 nextToken(); 1582 return false; 1583 } 1584 assert(FormatTok->is(tok::l_square)); 1585 FormatToken &LSquare = *FormatTok; 1586 if (!tryToParseLambdaIntroducer()) 1587 return false; 1588 1589 bool SeenArrow = false; 1590 1591 while (FormatTok->isNot(tok::l_brace)) { 1592 if (FormatTok->isSimpleTypeSpecifier()) { 1593 nextToken(); 1594 continue; 1595 } 1596 switch (FormatTok->Tok.getKind()) { 1597 case tok::l_brace: 1598 break; 1599 case tok::l_paren: 1600 parseParens(); 1601 break; 1602 case tok::amp: 1603 case tok::star: 1604 case tok::kw_const: 1605 case tok::comma: 1606 case tok::less: 1607 case tok::greater: 1608 case tok::identifier: 1609 case tok::numeric_constant: 1610 case tok::coloncolon: 1611 case tok::kw_class: 1612 case tok::kw_mutable: 1613 case tok::kw_noexcept: 1614 case tok::kw_template: 1615 case tok::kw_typename: 1616 nextToken(); 1617 break; 1618 // Specialization of a template with an integer parameter can contain 1619 // arithmetic, logical, comparison and ternary operators. 1620 // 1621 // FIXME: This also accepts sequences of operators that are not in the scope 1622 // of a template argument list. 1623 // 1624 // In a C++ lambda a template type can only occur after an arrow. We use 1625 // this as an heuristic to distinguish between Objective-C expressions 1626 // followed by an `a->b` expression, such as: 1627 // ([obj func:arg] + a->b) 1628 // Otherwise the code below would parse as a lambda. 1629 // 1630 // FIXME: This heuristic is incorrect for C++20 generic lambdas with 1631 // explicit template lists: []<bool b = true && false>(U &&u){} 1632 case tok::plus: 1633 case tok::minus: 1634 case tok::exclaim: 1635 case tok::tilde: 1636 case tok::slash: 1637 case tok::percent: 1638 case tok::lessless: 1639 case tok::pipe: 1640 case tok::pipepipe: 1641 case tok::ampamp: 1642 case tok::caret: 1643 case tok::equalequal: 1644 case tok::exclaimequal: 1645 case tok::greaterequal: 1646 case tok::lessequal: 1647 case tok::question: 1648 case tok::colon: 1649 case tok::kw_true: 1650 case tok::kw_false: 1651 if (SeenArrow) { 1652 nextToken(); 1653 break; 1654 } 1655 return true; 1656 case tok::arrow: 1657 // This might or might not actually be a lambda arrow (this could be an 1658 // ObjC method invocation followed by a dereferencing arrow). We might 1659 // reset this back to TT_Unknown in TokenAnnotator. 1660 FormatTok->Type = TT_LambdaArrow; 1661 SeenArrow = true; 1662 nextToken(); 1663 break; 1664 default: 1665 return true; 1666 } 1667 } 1668 FormatTok->Type = TT_LambdaLBrace; 1669 LSquare.Type = TT_LambdaLSquare; 1670 parseChildBlock(); 1671 return true; 1672 } 1673 1674 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 1675 const FormatToken *Previous = FormatTok->Previous; 1676 if (Previous && 1677 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 1678 tok::kw_delete, tok::l_square) || 1679 FormatTok->isCppStructuredBinding(Style) || Previous->closesScope() || 1680 Previous->isSimpleTypeSpecifier())) { 1681 nextToken(); 1682 return false; 1683 } 1684 nextToken(); 1685 if (FormatTok->is(tok::l_square)) { 1686 return false; 1687 } 1688 parseSquare(/*LambdaIntroducer=*/true); 1689 return true; 1690 } 1691 1692 void UnwrappedLineParser::tryToParseJSFunction() { 1693 assert(FormatTok->is(Keywords.kw_function) || 1694 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)); 1695 if (FormatTok->is(Keywords.kw_async)) 1696 nextToken(); 1697 // Consume "function". 1698 nextToken(); 1699 1700 // Consume * (generator function). Treat it like C++'s overloaded operators. 1701 if (FormatTok->is(tok::star)) { 1702 FormatTok->Type = TT_OverloadedOperator; 1703 nextToken(); 1704 } 1705 1706 // Consume function name. 1707 if (FormatTok->is(tok::identifier)) 1708 nextToken(); 1709 1710 if (FormatTok->isNot(tok::l_paren)) 1711 return; 1712 1713 // Parse formal parameter list. 1714 parseParens(); 1715 1716 if (FormatTok->is(tok::colon)) { 1717 // Parse a type definition. 1718 nextToken(); 1719 1720 // Eat the type declaration. For braced inline object types, balance braces, 1721 // otherwise just parse until finding an l_brace for the function body. 1722 if (FormatTok->is(tok::l_brace)) 1723 tryToParseBracedList(); 1724 else 1725 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 1726 nextToken(); 1727 } 1728 1729 if (FormatTok->is(tok::semi)) 1730 return; 1731 1732 parseChildBlock(); 1733 } 1734 1735 bool UnwrappedLineParser::tryToParseBracedList() { 1736 if (FormatTok->BlockKind == BK_Unknown) 1737 calculateBraceTypes(); 1738 assert(FormatTok->BlockKind != BK_Unknown); 1739 if (FormatTok->BlockKind == BK_Block) 1740 return false; 1741 nextToken(); 1742 parseBracedList(); 1743 return true; 1744 } 1745 1746 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons, 1747 tok::TokenKind ClosingBraceKind) { 1748 bool HasError = false; 1749 1750 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 1751 // replace this by using parseAssigmentExpression() inside. 1752 do { 1753 if (Style.isCSharp()) { 1754 if (FormatTok->is(TT_JsFatArrow)) { 1755 nextToken(); 1756 // Fat arrows can be followed by simple expressions or by child blocks 1757 // in curly braces. 1758 if (FormatTok->is(tok::l_brace)) { 1759 parseChildBlock(); 1760 continue; 1761 } 1762 } 1763 } 1764 if (Style.Language == FormatStyle::LK_JavaScript) { 1765 if (FormatTok->is(Keywords.kw_function) || 1766 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) { 1767 tryToParseJSFunction(); 1768 continue; 1769 } 1770 if (FormatTok->is(TT_JsFatArrow)) { 1771 nextToken(); 1772 // Fat arrows can be followed by simple expressions or by child blocks 1773 // in curly braces. 1774 if (FormatTok->is(tok::l_brace)) { 1775 parseChildBlock(); 1776 continue; 1777 } 1778 } 1779 if (FormatTok->is(tok::l_brace)) { 1780 // Could be a method inside of a braced list `{a() { return 1; }}`. 1781 if (tryToParseBracedList()) 1782 continue; 1783 parseChildBlock(); 1784 } 1785 } 1786 if (FormatTok->Tok.getKind() == ClosingBraceKind) { 1787 nextToken(); 1788 return !HasError; 1789 } 1790 switch (FormatTok->Tok.getKind()) { 1791 case tok::caret: 1792 nextToken(); 1793 if (FormatTok->is(tok::l_brace)) { 1794 parseChildBlock(); 1795 } 1796 break; 1797 case tok::l_square: 1798 if (Style.isCSharp()) 1799 parseSquare(); 1800 else 1801 tryToParseLambda(); 1802 break; 1803 case tok::l_paren: 1804 parseParens(); 1805 // JavaScript can just have free standing methods and getters/setters in 1806 // object literals. Detect them by a "{" following ")". 1807 if (Style.Language == FormatStyle::LK_JavaScript) { 1808 if (FormatTok->is(tok::l_brace)) 1809 parseChildBlock(); 1810 break; 1811 } 1812 break; 1813 case tok::l_brace: 1814 // Assume there are no blocks inside a braced init list apart 1815 // from the ones we explicitly parse out (like lambdas). 1816 FormatTok->BlockKind = BK_BracedInit; 1817 nextToken(); 1818 parseBracedList(); 1819 break; 1820 case tok::less: 1821 if (Style.Language == FormatStyle::LK_Proto) { 1822 nextToken(); 1823 parseBracedList(/*ContinueOnSemicolons=*/false, 1824 /*ClosingBraceKind=*/tok::greater); 1825 } else { 1826 nextToken(); 1827 } 1828 break; 1829 case tok::semi: 1830 // JavaScript (or more precisely TypeScript) can have semicolons in braced 1831 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 1832 // used for error recovery if we have otherwise determined that this is 1833 // a braced list. 1834 if (Style.Language == FormatStyle::LK_JavaScript) { 1835 nextToken(); 1836 break; 1837 } 1838 HasError = true; 1839 if (!ContinueOnSemicolons) 1840 return !HasError; 1841 nextToken(); 1842 break; 1843 case tok::comma: 1844 nextToken(); 1845 break; 1846 default: 1847 nextToken(); 1848 break; 1849 } 1850 } while (!eof()); 1851 return false; 1852 } 1853 1854 void UnwrappedLineParser::parseParens() { 1855 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 1856 nextToken(); 1857 do { 1858 switch (FormatTok->Tok.getKind()) { 1859 case tok::l_paren: 1860 parseParens(); 1861 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 1862 parseChildBlock(); 1863 break; 1864 case tok::r_paren: 1865 nextToken(); 1866 return; 1867 case tok::r_brace: 1868 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1869 return; 1870 case tok::l_square: 1871 tryToParseLambda(); 1872 break; 1873 case tok::l_brace: 1874 if (!tryToParseBracedList()) 1875 parseChildBlock(); 1876 break; 1877 case tok::at: 1878 nextToken(); 1879 if (FormatTok->Tok.is(tok::l_brace)) { 1880 nextToken(); 1881 parseBracedList(); 1882 } 1883 break; 1884 case tok::kw_class: 1885 if (Style.Language == FormatStyle::LK_JavaScript) 1886 parseRecord(/*ParseAsExpr=*/true); 1887 else 1888 nextToken(); 1889 break; 1890 case tok::identifier: 1891 if (Style.Language == FormatStyle::LK_JavaScript && 1892 (FormatTok->is(Keywords.kw_function) || 1893 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function))) 1894 tryToParseJSFunction(); 1895 else 1896 nextToken(); 1897 break; 1898 default: 1899 nextToken(); 1900 break; 1901 } 1902 } while (!eof()); 1903 } 1904 1905 void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) { 1906 if (!LambdaIntroducer) { 1907 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected."); 1908 if (tryToParseLambda()) 1909 return; 1910 } 1911 do { 1912 switch (FormatTok->Tok.getKind()) { 1913 case tok::l_paren: 1914 parseParens(); 1915 break; 1916 case tok::r_square: 1917 nextToken(); 1918 return; 1919 case tok::r_brace: 1920 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1921 return; 1922 case tok::l_square: 1923 parseSquare(); 1924 break; 1925 case tok::l_brace: { 1926 if (!tryToParseBracedList()) 1927 parseChildBlock(); 1928 break; 1929 } 1930 case tok::at: 1931 nextToken(); 1932 if (FormatTok->Tok.is(tok::l_brace)) { 1933 nextToken(); 1934 parseBracedList(); 1935 } 1936 break; 1937 default: 1938 nextToken(); 1939 break; 1940 } 1941 } while (!eof()); 1942 } 1943 1944 void UnwrappedLineParser::parseIfThenElse() { 1945 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 1946 nextToken(); 1947 if (FormatTok->Tok.isOneOf(tok::kw_constexpr, tok::identifier)) 1948 nextToken(); 1949 if (FormatTok->Tok.is(tok::l_paren)) 1950 parseParens(); 1951 bool NeedsUnwrappedLine = false; 1952 if (FormatTok->Tok.is(tok::l_brace)) { 1953 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1954 parseBlock(/*MustBeDeclaration=*/false); 1955 if (Style.BraceWrapping.BeforeElse) 1956 addUnwrappedLine(); 1957 else 1958 NeedsUnwrappedLine = true; 1959 } else { 1960 addUnwrappedLine(); 1961 ++Line->Level; 1962 parseStructuralElement(); 1963 --Line->Level; 1964 } 1965 if (FormatTok->Tok.is(tok::kw_else)) { 1966 nextToken(); 1967 if (FormatTok->Tok.is(tok::l_brace)) { 1968 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1969 parseBlock(/*MustBeDeclaration=*/false); 1970 addUnwrappedLine(); 1971 } else if (FormatTok->Tok.is(tok::kw_if)) { 1972 parseIfThenElse(); 1973 } else { 1974 addUnwrappedLine(); 1975 ++Line->Level; 1976 parseStructuralElement(); 1977 if (FormatTok->is(tok::eof)) 1978 addUnwrappedLine(); 1979 --Line->Level; 1980 } 1981 } else if (NeedsUnwrappedLine) { 1982 addUnwrappedLine(); 1983 } 1984 } 1985 1986 void UnwrappedLineParser::parseTryCatch() { 1987 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 1988 nextToken(); 1989 bool NeedsUnwrappedLine = false; 1990 if (FormatTok->is(tok::colon)) { 1991 // We are in a function try block, what comes is an initializer list. 1992 nextToken(); 1993 1994 // In case identifiers were removed by clang-tidy, what might follow is 1995 // multiple commas in sequence - before the first identifier. 1996 while (FormatTok->is(tok::comma)) 1997 nextToken(); 1998 1999 while (FormatTok->is(tok::identifier)) { 2000 nextToken(); 2001 if (FormatTok->is(tok::l_paren)) 2002 parseParens(); 2003 2004 // In case identifiers were removed by clang-tidy, what might follow is 2005 // multiple commas in sequence - after the first identifier. 2006 while (FormatTok->is(tok::comma)) 2007 nextToken(); 2008 } 2009 } 2010 // Parse try with resource. 2011 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) { 2012 parseParens(); 2013 } 2014 if (FormatTok->is(tok::l_brace)) { 2015 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2016 parseBlock(/*MustBeDeclaration=*/false); 2017 if (Style.BraceWrapping.BeforeCatch) { 2018 addUnwrappedLine(); 2019 } else { 2020 NeedsUnwrappedLine = true; 2021 } 2022 } else if (!FormatTok->is(tok::kw_catch)) { 2023 // The C++ standard requires a compound-statement after a try. 2024 // If there's none, we try to assume there's a structuralElement 2025 // and try to continue. 2026 addUnwrappedLine(); 2027 ++Line->Level; 2028 parseStructuralElement(); 2029 --Line->Level; 2030 } 2031 while (1) { 2032 if (FormatTok->is(tok::at)) 2033 nextToken(); 2034 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 2035 tok::kw___finally) || 2036 ((Style.Language == FormatStyle::LK_Java || 2037 Style.Language == FormatStyle::LK_JavaScript) && 2038 FormatTok->is(Keywords.kw_finally)) || 2039 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) || 2040 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally)))) 2041 break; 2042 nextToken(); 2043 while (FormatTok->isNot(tok::l_brace)) { 2044 if (FormatTok->is(tok::l_paren)) { 2045 parseParens(); 2046 continue; 2047 } 2048 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) 2049 return; 2050 nextToken(); 2051 } 2052 NeedsUnwrappedLine = false; 2053 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2054 parseBlock(/*MustBeDeclaration=*/false); 2055 if (Style.BraceWrapping.BeforeCatch) 2056 addUnwrappedLine(); 2057 else 2058 NeedsUnwrappedLine = true; 2059 } 2060 if (NeedsUnwrappedLine) 2061 addUnwrappedLine(); 2062 } 2063 2064 void UnwrappedLineParser::parseNamespace() { 2065 assert(FormatTok->isOneOf(tok::kw_namespace, TT_NamespaceMacro) && 2066 "'namespace' expected"); 2067 2068 const FormatToken &InitialToken = *FormatTok; 2069 nextToken(); 2070 if (InitialToken.is(TT_NamespaceMacro)) { 2071 parseParens(); 2072 } else { 2073 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::kw_inline, 2074 tok::l_square)) { 2075 if (FormatTok->is(tok::l_square)) 2076 parseSquare(); 2077 else 2078 nextToken(); 2079 } 2080 } 2081 if (FormatTok->Tok.is(tok::l_brace)) { 2082 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2083 addUnwrappedLine(); 2084 2085 bool AddLevel = Style.NamespaceIndentation == FormatStyle::NI_All || 2086 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 2087 DeclarationScopeStack.size() > 1); 2088 parseBlock(/*MustBeDeclaration=*/true, AddLevel); 2089 // Munch the semicolon after a namespace. This is more common than one would 2090 // think. Putting the semicolon into its own line is very ugly. 2091 if (FormatTok->Tok.is(tok::semi)) 2092 nextToken(); 2093 addUnwrappedLine(); 2094 } 2095 // FIXME: Add error handling. 2096 } 2097 2098 void UnwrappedLineParser::parseNew() { 2099 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 2100 nextToken(); 2101 2102 if (Style.isCSharp()) { 2103 do { 2104 if (FormatTok->is(tok::l_brace)) 2105 parseBracedList(); 2106 2107 if (FormatTok->isOneOf(tok::semi, tok::comma)) 2108 return; 2109 2110 nextToken(); 2111 } while (!eof()); 2112 } 2113 2114 if (Style.Language != FormatStyle::LK_Java) 2115 return; 2116 2117 // In Java, we can parse everything up to the parens, which aren't optional. 2118 do { 2119 // There should not be a ;, { or } before the new's open paren. 2120 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 2121 return; 2122 2123 // Consume the parens. 2124 if (FormatTok->is(tok::l_paren)) { 2125 parseParens(); 2126 2127 // If there is a class body of an anonymous class, consume that as child. 2128 if (FormatTok->is(tok::l_brace)) 2129 parseChildBlock(); 2130 return; 2131 } 2132 nextToken(); 2133 } while (!eof()); 2134 } 2135 2136 void UnwrappedLineParser::parseForOrWhileLoop() { 2137 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 2138 "'for', 'while' or foreach macro expected"); 2139 nextToken(); 2140 // JS' for await ( ... 2141 if (Style.Language == FormatStyle::LK_JavaScript && 2142 FormatTok->is(Keywords.kw_await)) 2143 nextToken(); 2144 if (FormatTok->Tok.is(tok::l_paren)) 2145 parseParens(); 2146 if (FormatTok->Tok.is(tok::l_brace)) { 2147 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2148 parseBlock(/*MustBeDeclaration=*/false); 2149 addUnwrappedLine(); 2150 } else { 2151 addUnwrappedLine(); 2152 ++Line->Level; 2153 parseStructuralElement(); 2154 --Line->Level; 2155 } 2156 } 2157 2158 void UnwrappedLineParser::parseDoWhile() { 2159 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 2160 nextToken(); 2161 if (FormatTok->Tok.is(tok::l_brace)) { 2162 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2163 parseBlock(/*MustBeDeclaration=*/false); 2164 if (Style.BraceWrapping.IndentBraces) 2165 addUnwrappedLine(); 2166 } else { 2167 addUnwrappedLine(); 2168 ++Line->Level; 2169 parseStructuralElement(); 2170 --Line->Level; 2171 } 2172 2173 // FIXME: Add error handling. 2174 if (!FormatTok->Tok.is(tok::kw_while)) { 2175 addUnwrappedLine(); 2176 return; 2177 } 2178 2179 nextToken(); 2180 parseStructuralElement(); 2181 } 2182 2183 void UnwrappedLineParser::parseLabel(bool LeftAlignLabel) { 2184 nextToken(); 2185 unsigned OldLineLevel = Line->Level; 2186 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 2187 --Line->Level; 2188 if (LeftAlignLabel) 2189 Line->Level = 0; 2190 if (!Style.IndentCaseBlocks && CommentsBeforeNextToken.empty() && 2191 FormatTok->Tok.is(tok::l_brace)) { 2192 CompoundStatementIndenter Indenter(this, Line->Level, 2193 Style.BraceWrapping.AfterCaseLabel, 2194 Style.BraceWrapping.IndentBraces); 2195 parseBlock(/*MustBeDeclaration=*/false); 2196 if (FormatTok->Tok.is(tok::kw_break)) { 2197 if (Style.BraceWrapping.AfterControlStatement == 2198 FormatStyle::BWACS_Always) 2199 addUnwrappedLine(); 2200 parseStructuralElement(); 2201 } 2202 addUnwrappedLine(); 2203 } else { 2204 if (FormatTok->is(tok::semi)) 2205 nextToken(); 2206 addUnwrappedLine(); 2207 } 2208 Line->Level = OldLineLevel; 2209 if (FormatTok->isNot(tok::l_brace)) { 2210 parseStructuralElement(); 2211 addUnwrappedLine(); 2212 } 2213 } 2214 2215 void UnwrappedLineParser::parseCaseLabel() { 2216 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 2217 // FIXME: fix handling of complex expressions here. 2218 do { 2219 nextToken(); 2220 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 2221 parseLabel(); 2222 } 2223 2224 void UnwrappedLineParser::parseSwitch() { 2225 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 2226 nextToken(); 2227 if (FormatTok->Tok.is(tok::l_paren)) 2228 parseParens(); 2229 if (FormatTok->Tok.is(tok::l_brace)) { 2230 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2231 parseBlock(/*MustBeDeclaration=*/false); 2232 addUnwrappedLine(); 2233 } else { 2234 addUnwrappedLine(); 2235 ++Line->Level; 2236 parseStructuralElement(); 2237 --Line->Level; 2238 } 2239 } 2240 2241 void UnwrappedLineParser::parseAccessSpecifier() { 2242 nextToken(); 2243 // Understand Qt's slots. 2244 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 2245 nextToken(); 2246 // Otherwise, we don't know what it is, and we'd better keep the next token. 2247 if (FormatTok->Tok.is(tok::colon)) 2248 nextToken(); 2249 addUnwrappedLine(); 2250 } 2251 2252 bool UnwrappedLineParser::parseEnum() { 2253 // Won't be 'enum' for NS_ENUMs. 2254 if (FormatTok->Tok.is(tok::kw_enum)) 2255 nextToken(); 2256 2257 // In TypeScript, "enum" can also be used as property name, e.g. in interface 2258 // declarations. An "enum" keyword followed by a colon would be a syntax 2259 // error and thus assume it is just an identifier. 2260 if (Style.Language == FormatStyle::LK_JavaScript && 2261 FormatTok->isOneOf(tok::colon, tok::question)) 2262 return false; 2263 2264 // In protobuf, "enum" can be used as a field name. 2265 if (Style.Language == FormatStyle::LK_Proto && FormatTok->is(tok::equal)) 2266 return false; 2267 2268 // Eat up enum class ... 2269 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct)) 2270 nextToken(); 2271 2272 while (FormatTok->Tok.getIdentifierInfo() || 2273 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 2274 tok::greater, tok::comma, tok::question)) { 2275 nextToken(); 2276 // We can have macros or attributes in between 'enum' and the enum name. 2277 if (FormatTok->is(tok::l_paren)) 2278 parseParens(); 2279 if (FormatTok->is(tok::identifier)) { 2280 nextToken(); 2281 // If there are two identifiers in a row, this is likely an elaborate 2282 // return type. In Java, this can be "implements", etc. 2283 if (Style.isCpp() && FormatTok->is(tok::identifier)) 2284 return false; 2285 } 2286 } 2287 2288 // Just a declaration or something is wrong. 2289 if (FormatTok->isNot(tok::l_brace)) 2290 return true; 2291 FormatTok->BlockKind = BK_Block; 2292 2293 if (Style.Language == FormatStyle::LK_Java) { 2294 // Java enums are different. 2295 parseJavaEnumBody(); 2296 return true; 2297 } 2298 if (Style.Language == FormatStyle::LK_Proto) { 2299 parseBlock(/*MustBeDeclaration=*/true); 2300 return true; 2301 } 2302 2303 // Parse enum body. 2304 nextToken(); 2305 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true); 2306 if (HasError) { 2307 if (FormatTok->is(tok::semi)) 2308 nextToken(); 2309 addUnwrappedLine(); 2310 } 2311 return true; 2312 2313 // There is no addUnwrappedLine() here so that we fall through to parsing a 2314 // structural element afterwards. Thus, in "enum A {} n, m;", 2315 // "} n, m;" will end up in one unwrapped line. 2316 } 2317 2318 void UnwrappedLineParser::parseJavaEnumBody() { 2319 // Determine whether the enum is simple, i.e. does not have a semicolon or 2320 // constants with class bodies. Simple enums can be formatted like braced 2321 // lists, contracted to a single line, etc. 2322 unsigned StoredPosition = Tokens->getPosition(); 2323 bool IsSimple = true; 2324 FormatToken *Tok = Tokens->getNextToken(); 2325 while (Tok) { 2326 if (Tok->is(tok::r_brace)) 2327 break; 2328 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 2329 IsSimple = false; 2330 break; 2331 } 2332 // FIXME: This will also mark enums with braces in the arguments to enum 2333 // constants as "not simple". This is probably fine in practice, though. 2334 Tok = Tokens->getNextToken(); 2335 } 2336 FormatTok = Tokens->setPosition(StoredPosition); 2337 2338 if (IsSimple) { 2339 nextToken(); 2340 parseBracedList(); 2341 addUnwrappedLine(); 2342 return; 2343 } 2344 2345 // Parse the body of a more complex enum. 2346 // First add a line for everything up to the "{". 2347 nextToken(); 2348 addUnwrappedLine(); 2349 ++Line->Level; 2350 2351 // Parse the enum constants. 2352 while (FormatTok) { 2353 if (FormatTok->is(tok::l_brace)) { 2354 // Parse the constant's class body. 2355 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 2356 /*MunchSemi=*/false); 2357 } else if (FormatTok->is(tok::l_paren)) { 2358 parseParens(); 2359 } else if (FormatTok->is(tok::comma)) { 2360 nextToken(); 2361 addUnwrappedLine(); 2362 } else if (FormatTok->is(tok::semi)) { 2363 nextToken(); 2364 addUnwrappedLine(); 2365 break; 2366 } else if (FormatTok->is(tok::r_brace)) { 2367 addUnwrappedLine(); 2368 break; 2369 } else { 2370 nextToken(); 2371 } 2372 } 2373 2374 // Parse the class body after the enum's ";" if any. 2375 parseLevel(/*HasOpeningBrace=*/true); 2376 nextToken(); 2377 --Line->Level; 2378 addUnwrappedLine(); 2379 } 2380 2381 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) { 2382 const FormatToken &InitialToken = *FormatTok; 2383 nextToken(); 2384 2385 // The actual identifier can be a nested name specifier, and in macros 2386 // it is often token-pasted. 2387 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 2388 tok::kw___attribute, tok::kw___declspec, 2389 tok::kw_alignas) || 2390 ((Style.Language == FormatStyle::LK_Java || 2391 Style.Language == FormatStyle::LK_JavaScript) && 2392 FormatTok->isOneOf(tok::period, tok::comma))) { 2393 if (Style.Language == FormatStyle::LK_JavaScript && 2394 FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) { 2395 // JavaScript/TypeScript supports inline object types in 2396 // extends/implements positions: 2397 // class Foo implements {bar: number} { } 2398 nextToken(); 2399 if (FormatTok->is(tok::l_brace)) { 2400 tryToParseBracedList(); 2401 continue; 2402 } 2403 } 2404 bool IsNonMacroIdentifier = 2405 FormatTok->is(tok::identifier) && 2406 FormatTok->TokenText != FormatTok->TokenText.upper(); 2407 nextToken(); 2408 // We can have macros or attributes in between 'class' and the class name. 2409 if (!IsNonMacroIdentifier && FormatTok->Tok.is(tok::l_paren)) 2410 parseParens(); 2411 } 2412 2413 // Note that parsing away template declarations here leads to incorrectly 2414 // accepting function declarations as record declarations. 2415 // In general, we cannot solve this problem. Consider: 2416 // class A<int> B() {} 2417 // which can be a function definition or a class definition when B() is a 2418 // macro. If we find enough real-world cases where this is a problem, we 2419 // can parse for the 'template' keyword in the beginning of the statement, 2420 // and thus rule out the record production in case there is no template 2421 // (this would still leave us with an ambiguity between template function 2422 // and class declarations). 2423 if (FormatTok->isOneOf(tok::colon, tok::less)) { 2424 while (!eof()) { 2425 if (FormatTok->is(tok::l_brace)) { 2426 calculateBraceTypes(/*ExpectClassBody=*/true); 2427 if (!tryToParseBracedList()) 2428 break; 2429 } 2430 if (FormatTok->Tok.is(tok::semi)) 2431 return; 2432 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where)) { 2433 addUnwrappedLine(); 2434 nextToken(); 2435 parseCSharpGenericTypeConstraint(); 2436 break; 2437 } 2438 nextToken(); 2439 } 2440 } 2441 if (FormatTok->Tok.is(tok::l_brace)) { 2442 if (ParseAsExpr) { 2443 parseChildBlock(); 2444 } else { 2445 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2446 addUnwrappedLine(); 2447 2448 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 2449 /*MunchSemi=*/false); 2450 } 2451 } 2452 // There is no addUnwrappedLine() here so that we fall through to parsing a 2453 // structural element afterwards. Thus, in "class A {} n, m;", 2454 // "} n, m;" will end up in one unwrapped line. 2455 } 2456 2457 void UnwrappedLineParser::parseObjCMethod() { 2458 assert(FormatTok->Tok.isOneOf(tok::l_paren, tok::identifier) && 2459 "'(' or identifier expected."); 2460 do { 2461 if (FormatTok->Tok.is(tok::semi)) { 2462 nextToken(); 2463 addUnwrappedLine(); 2464 return; 2465 } else if (FormatTok->Tok.is(tok::l_brace)) { 2466 if (Style.BraceWrapping.AfterFunction) 2467 addUnwrappedLine(); 2468 parseBlock(/*MustBeDeclaration=*/false); 2469 addUnwrappedLine(); 2470 return; 2471 } else { 2472 nextToken(); 2473 } 2474 } while (!eof()); 2475 } 2476 2477 void UnwrappedLineParser::parseObjCProtocolList() { 2478 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 2479 do { 2480 nextToken(); 2481 // Early exit in case someone forgot a close angle. 2482 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 2483 FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) 2484 return; 2485 } while (!eof() && FormatTok->Tok.isNot(tok::greater)); 2486 nextToken(); // Skip '>'. 2487 } 2488 2489 void UnwrappedLineParser::parseObjCUntilAtEnd() { 2490 do { 2491 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 2492 nextToken(); 2493 addUnwrappedLine(); 2494 break; 2495 } 2496 if (FormatTok->is(tok::l_brace)) { 2497 parseBlock(/*MustBeDeclaration=*/false); 2498 // In ObjC interfaces, nothing should be following the "}". 2499 addUnwrappedLine(); 2500 } else if (FormatTok->is(tok::r_brace)) { 2501 // Ignore stray "}". parseStructuralElement doesn't consume them. 2502 nextToken(); 2503 addUnwrappedLine(); 2504 } else if (FormatTok->isOneOf(tok::minus, tok::plus)) { 2505 nextToken(); 2506 parseObjCMethod(); 2507 } else { 2508 parseStructuralElement(); 2509 } 2510 } while (!eof()); 2511 } 2512 2513 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 2514 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_interface || 2515 FormatTok->Tok.getObjCKeywordID() == tok::objc_implementation); 2516 nextToken(); 2517 nextToken(); // interface name 2518 2519 // @interface can be followed by a lightweight generic 2520 // specialization list, then either a base class or a category. 2521 if (FormatTok->Tok.is(tok::less)) { 2522 // Unlike protocol lists, generic parameterizations support 2523 // nested angles: 2524 // 2525 // @interface Foo<ValueType : id <NSCopying, NSSecureCoding>> : 2526 // NSObject <NSCopying, NSSecureCoding> 2527 // 2528 // so we need to count how many open angles we have left. 2529 unsigned NumOpenAngles = 1; 2530 do { 2531 nextToken(); 2532 // Early exit in case someone forgot a close angle. 2533 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 2534 FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) 2535 break; 2536 if (FormatTok->Tok.is(tok::less)) 2537 ++NumOpenAngles; 2538 else if (FormatTok->Tok.is(tok::greater)) { 2539 assert(NumOpenAngles > 0 && "'>' makes NumOpenAngles negative"); 2540 --NumOpenAngles; 2541 } 2542 } while (!eof() && NumOpenAngles != 0); 2543 nextToken(); // Skip '>'. 2544 } 2545 if (FormatTok->Tok.is(tok::colon)) { 2546 nextToken(); 2547 nextToken(); // base class name 2548 } else if (FormatTok->Tok.is(tok::l_paren)) 2549 // Skip category, if present. 2550 parseParens(); 2551 2552 if (FormatTok->Tok.is(tok::less)) 2553 parseObjCProtocolList(); 2554 2555 if (FormatTok->Tok.is(tok::l_brace)) { 2556 if (Style.BraceWrapping.AfterObjCDeclaration) 2557 addUnwrappedLine(); 2558 parseBlock(/*MustBeDeclaration=*/true); 2559 } 2560 2561 // With instance variables, this puts '}' on its own line. Without instance 2562 // variables, this ends the @interface line. 2563 addUnwrappedLine(); 2564 2565 parseObjCUntilAtEnd(); 2566 } 2567 2568 // Returns true for the declaration/definition form of @protocol, 2569 // false for the expression form. 2570 bool UnwrappedLineParser::parseObjCProtocol() { 2571 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_protocol); 2572 nextToken(); 2573 2574 if (FormatTok->is(tok::l_paren)) 2575 // The expression form of @protocol, e.g. "Protocol* p = @protocol(foo);". 2576 return false; 2577 2578 // The definition/declaration form, 2579 // @protocol Foo 2580 // - (int)someMethod; 2581 // @end 2582 2583 nextToken(); // protocol name 2584 2585 if (FormatTok->Tok.is(tok::less)) 2586 parseObjCProtocolList(); 2587 2588 // Check for protocol declaration. 2589 if (FormatTok->Tok.is(tok::semi)) { 2590 nextToken(); 2591 addUnwrappedLine(); 2592 return true; 2593 } 2594 2595 addUnwrappedLine(); 2596 parseObjCUntilAtEnd(); 2597 return true; 2598 } 2599 2600 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 2601 bool IsImport = FormatTok->is(Keywords.kw_import); 2602 assert(IsImport || FormatTok->is(tok::kw_export)); 2603 nextToken(); 2604 2605 // Consume the "default" in "export default class/function". 2606 if (FormatTok->is(tok::kw_default)) 2607 nextToken(); 2608 2609 // Consume "async function", "function" and "default function", so that these 2610 // get parsed as free-standing JS functions, i.e. do not require a trailing 2611 // semicolon. 2612 if (FormatTok->is(Keywords.kw_async)) 2613 nextToken(); 2614 if (FormatTok->is(Keywords.kw_function)) { 2615 nextToken(); 2616 return; 2617 } 2618 2619 // For imports, `export *`, `export {...}`, consume the rest of the line up 2620 // to the terminating `;`. For everything else, just return and continue 2621 // parsing the structural element, i.e. the declaration or expression for 2622 // `export default`. 2623 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 2624 !FormatTok->isStringLiteral()) 2625 return; 2626 2627 while (!eof()) { 2628 if (FormatTok->is(tok::semi)) 2629 return; 2630 if (Line->Tokens.empty()) { 2631 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 2632 // import statement should terminate. 2633 return; 2634 } 2635 if (FormatTok->is(tok::l_brace)) { 2636 FormatTok->BlockKind = BK_Block; 2637 nextToken(); 2638 parseBracedList(); 2639 } else { 2640 nextToken(); 2641 } 2642 } 2643 } 2644 2645 void UnwrappedLineParser::parseStatementMacro() { 2646 nextToken(); 2647 if (FormatTok->is(tok::l_paren)) 2648 parseParens(); 2649 if (FormatTok->is(tok::semi)) 2650 nextToken(); 2651 addUnwrappedLine(); 2652 } 2653 2654 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 2655 StringRef Prefix = "") { 2656 llvm::dbgs() << Prefix << "Line(" << Line.Level 2657 << ", FSC=" << Line.FirstStartColumn << ")" 2658 << (Line.InPPDirective ? " MACRO" : "") << ": "; 2659 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2660 E = Line.Tokens.end(); 2661 I != E; ++I) { 2662 llvm::dbgs() << I->Tok->Tok.getName() << "[" 2663 << "T=" << I->Tok->Type << ", OC=" << I->Tok->OriginalColumn 2664 << "] "; 2665 } 2666 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2667 E = Line.Tokens.end(); 2668 I != E; ++I) { 2669 const UnwrappedLineNode &Node = *I; 2670 for (SmallVectorImpl<UnwrappedLine>::const_iterator 2671 I = Node.Children.begin(), 2672 E = Node.Children.end(); 2673 I != E; ++I) { 2674 printDebugInfo(*I, "\nChild: "); 2675 } 2676 } 2677 llvm::dbgs() << "\n"; 2678 } 2679 2680 void UnwrappedLineParser::addUnwrappedLine() { 2681 if (Line->Tokens.empty()) 2682 return; 2683 LLVM_DEBUG({ 2684 if (CurrentLines == &Lines) 2685 printDebugInfo(*Line); 2686 }); 2687 CurrentLines->push_back(std::move(*Line)); 2688 Line->Tokens.clear(); 2689 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex; 2690 Line->FirstStartColumn = 0; 2691 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 2692 CurrentLines->append( 2693 std::make_move_iterator(PreprocessorDirectives.begin()), 2694 std::make_move_iterator(PreprocessorDirectives.end())); 2695 PreprocessorDirectives.clear(); 2696 } 2697 // Disconnect the current token from the last token on the previous line. 2698 FormatTok->Previous = nullptr; 2699 } 2700 2701 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 2702 2703 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 2704 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 2705 FormatTok.NewlinesBefore > 0; 2706 } 2707 2708 // Checks if \p FormatTok is a line comment that continues the line comment 2709 // section on \p Line. 2710 static bool 2711 continuesLineCommentSection(const FormatToken &FormatTok, 2712 const UnwrappedLine &Line, 2713 const llvm::Regex &CommentPragmasRegex) { 2714 if (Line.Tokens.empty()) 2715 return false; 2716 2717 StringRef IndentContent = FormatTok.TokenText; 2718 if (FormatTok.TokenText.startswith("//") || 2719 FormatTok.TokenText.startswith("/*")) 2720 IndentContent = FormatTok.TokenText.substr(2); 2721 if (CommentPragmasRegex.match(IndentContent)) 2722 return false; 2723 2724 // If Line starts with a line comment, then FormatTok continues the comment 2725 // section if its original column is greater or equal to the original start 2726 // column of the line. 2727 // 2728 // Define the min column token of a line as follows: if a line ends in '{' or 2729 // contains a '{' followed by a line comment, then the min column token is 2730 // that '{'. Otherwise, the min column token of the line is the first token of 2731 // the line. 2732 // 2733 // If Line starts with a token other than a line comment, then FormatTok 2734 // continues the comment section if its original column is greater than the 2735 // original start column of the min column token of the line. 2736 // 2737 // For example, the second line comment continues the first in these cases: 2738 // 2739 // // first line 2740 // // second line 2741 // 2742 // and: 2743 // 2744 // // first line 2745 // // second line 2746 // 2747 // and: 2748 // 2749 // int i; // first line 2750 // // second line 2751 // 2752 // and: 2753 // 2754 // do { // first line 2755 // // second line 2756 // int i; 2757 // } while (true); 2758 // 2759 // and: 2760 // 2761 // enum { 2762 // a, // first line 2763 // // second line 2764 // b 2765 // }; 2766 // 2767 // The second line comment doesn't continue the first in these cases: 2768 // 2769 // // first line 2770 // // second line 2771 // 2772 // and: 2773 // 2774 // int i; // first line 2775 // // second line 2776 // 2777 // and: 2778 // 2779 // do { // first line 2780 // // second line 2781 // int i; 2782 // } while (true); 2783 // 2784 // and: 2785 // 2786 // enum { 2787 // a, // first line 2788 // // second line 2789 // }; 2790 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 2791 2792 // Scan for '{//'. If found, use the column of '{' as a min column for line 2793 // comment section continuation. 2794 const FormatToken *PreviousToken = nullptr; 2795 for (const UnwrappedLineNode &Node : Line.Tokens) { 2796 if (PreviousToken && PreviousToken->is(tok::l_brace) && 2797 isLineComment(*Node.Tok)) { 2798 MinColumnToken = PreviousToken; 2799 break; 2800 } 2801 PreviousToken = Node.Tok; 2802 2803 // Grab the last newline preceding a token in this unwrapped line. 2804 if (Node.Tok->NewlinesBefore > 0) { 2805 MinColumnToken = Node.Tok; 2806 } 2807 } 2808 if (PreviousToken && PreviousToken->is(tok::l_brace)) { 2809 MinColumnToken = PreviousToken; 2810 } 2811 2812 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok, 2813 MinColumnToken); 2814 } 2815 2816 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 2817 bool JustComments = Line->Tokens.empty(); 2818 for (SmallVectorImpl<FormatToken *>::const_iterator 2819 I = CommentsBeforeNextToken.begin(), 2820 E = CommentsBeforeNextToken.end(); 2821 I != E; ++I) { 2822 // Line comments that belong to the same line comment section are put on the 2823 // same line since later we might want to reflow content between them. 2824 // Additional fine-grained breaking of line comment sections is controlled 2825 // by the class BreakableLineCommentSection in case it is desirable to keep 2826 // several line comment sections in the same unwrapped line. 2827 // 2828 // FIXME: Consider putting separate line comment sections as children to the 2829 // unwrapped line instead. 2830 (*I)->ContinuesLineCommentSection = 2831 continuesLineCommentSection(**I, *Line, CommentPragmasRegex); 2832 if (isOnNewLine(**I) && JustComments && !(*I)->ContinuesLineCommentSection) 2833 addUnwrappedLine(); 2834 pushToken(*I); 2835 } 2836 if (NewlineBeforeNext && JustComments) 2837 addUnwrappedLine(); 2838 CommentsBeforeNextToken.clear(); 2839 } 2840 2841 void UnwrappedLineParser::nextToken(int LevelDifference) { 2842 if (eof()) 2843 return; 2844 flushComments(isOnNewLine(*FormatTok)); 2845 pushToken(FormatTok); 2846 FormatToken *Previous = FormatTok; 2847 if (Style.Language != FormatStyle::LK_JavaScript) 2848 readToken(LevelDifference); 2849 else 2850 readTokenWithJavaScriptASI(); 2851 FormatTok->Previous = Previous; 2852 } 2853 2854 void UnwrappedLineParser::distributeComments( 2855 const SmallVectorImpl<FormatToken *> &Comments, 2856 const FormatToken *NextTok) { 2857 // Whether or not a line comment token continues a line is controlled by 2858 // the method continuesLineCommentSection, with the following caveat: 2859 // 2860 // Define a trail of Comments to be a nonempty proper postfix of Comments such 2861 // that each comment line from the trail is aligned with the next token, if 2862 // the next token exists. If a trail exists, the beginning of the maximal 2863 // trail is marked as a start of a new comment section. 2864 // 2865 // For example in this code: 2866 // 2867 // int a; // line about a 2868 // // line 1 about b 2869 // // line 2 about b 2870 // int b; 2871 // 2872 // the two lines about b form a maximal trail, so there are two sections, the 2873 // first one consisting of the single comment "// line about a" and the 2874 // second one consisting of the next two comments. 2875 if (Comments.empty()) 2876 return; 2877 bool ShouldPushCommentsInCurrentLine = true; 2878 bool HasTrailAlignedWithNextToken = false; 2879 unsigned StartOfTrailAlignedWithNextToken = 0; 2880 if (NextTok) { 2881 // We are skipping the first element intentionally. 2882 for (unsigned i = Comments.size() - 1; i > 0; --i) { 2883 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) { 2884 HasTrailAlignedWithNextToken = true; 2885 StartOfTrailAlignedWithNextToken = i; 2886 } 2887 } 2888 } 2889 for (unsigned i = 0, e = Comments.size(); i < e; ++i) { 2890 FormatToken *FormatTok = Comments[i]; 2891 if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) { 2892 FormatTok->ContinuesLineCommentSection = false; 2893 } else { 2894 FormatTok->ContinuesLineCommentSection = 2895 continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex); 2896 } 2897 if (!FormatTok->ContinuesLineCommentSection && 2898 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) { 2899 ShouldPushCommentsInCurrentLine = false; 2900 } 2901 if (ShouldPushCommentsInCurrentLine) { 2902 pushToken(FormatTok); 2903 } else { 2904 CommentsBeforeNextToken.push_back(FormatTok); 2905 } 2906 } 2907 } 2908 2909 void UnwrappedLineParser::readToken(int LevelDifference) { 2910 SmallVector<FormatToken *, 1> Comments; 2911 do { 2912 FormatTok = Tokens->getNextToken(); 2913 assert(FormatTok); 2914 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 2915 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 2916 distributeComments(Comments, FormatTok); 2917 Comments.clear(); 2918 // If there is an unfinished unwrapped line, we flush the preprocessor 2919 // directives only after that unwrapped line was finished later. 2920 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 2921 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 2922 assert((LevelDifference >= 0 || 2923 static_cast<unsigned>(-LevelDifference) <= Line->Level) && 2924 "LevelDifference makes Line->Level negative"); 2925 Line->Level += LevelDifference; 2926 // Comments stored before the preprocessor directive need to be output 2927 // before the preprocessor directive, at the same level as the 2928 // preprocessor directive, as we consider them to apply to the directive. 2929 if (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash && 2930 PPBranchLevel > 0) 2931 Line->Level += PPBranchLevel; 2932 flushComments(isOnNewLine(*FormatTok)); 2933 parsePPDirective(); 2934 } 2935 while (FormatTok->Type == TT_ConflictStart || 2936 FormatTok->Type == TT_ConflictEnd || 2937 FormatTok->Type == TT_ConflictAlternative) { 2938 if (FormatTok->Type == TT_ConflictStart) { 2939 conditionalCompilationStart(/*Unreachable=*/false); 2940 } else if (FormatTok->Type == TT_ConflictAlternative) { 2941 conditionalCompilationAlternative(); 2942 } else if (FormatTok->Type == TT_ConflictEnd) { 2943 conditionalCompilationEnd(); 2944 } 2945 FormatTok = Tokens->getNextToken(); 2946 FormatTok->MustBreakBefore = true; 2947 } 2948 2949 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) && 2950 !Line->InPPDirective) { 2951 continue; 2952 } 2953 2954 if (!FormatTok->Tok.is(tok::comment)) { 2955 distributeComments(Comments, FormatTok); 2956 Comments.clear(); 2957 return; 2958 } 2959 2960 Comments.push_back(FormatTok); 2961 } while (!eof()); 2962 2963 distributeComments(Comments, nullptr); 2964 Comments.clear(); 2965 } 2966 2967 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 2968 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 2969 if (MustBreakBeforeNextToken) { 2970 Line->Tokens.back().Tok->MustBreakBefore = true; 2971 MustBreakBeforeNextToken = false; 2972 } 2973 } 2974 2975 } // end namespace format 2976 } // end namespace clang 2977