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