1 //===--- UnwrappedLineParser.cpp - Format C++ code ------------------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 /// 10 /// \file 11 /// \brief This file contains the implementation of the UnwrappedLineParser, 12 /// which turns a stream of tokens into UnwrappedLines. 13 /// 14 //===----------------------------------------------------------------------===// 15 16 #include "UnwrappedLineParser.h" 17 #include "llvm/ADT/STLExtras.h" 18 #include "llvm/Support/Debug.h" 19 #include "llvm/Support/raw_ostream.h" 20 21 #define DEBUG_TYPE "format-parser" 22 23 namespace clang { 24 namespace format { 25 26 class FormatTokenSource { 27 public: 28 virtual ~FormatTokenSource() {} 29 virtual FormatToken *getNextToken() = 0; 30 31 virtual unsigned getPosition() = 0; 32 virtual FormatToken *setPosition(unsigned Position) = 0; 33 }; 34 35 namespace { 36 37 class ScopedDeclarationState { 38 public: 39 ScopedDeclarationState(UnwrappedLine &Line, std::vector<bool> &Stack, 40 bool MustBeDeclaration) 41 : Line(Line), Stack(Stack) { 42 Line.MustBeDeclaration = MustBeDeclaration; 43 Stack.push_back(MustBeDeclaration); 44 } 45 ~ScopedDeclarationState() { 46 Stack.pop_back(); 47 if (!Stack.empty()) 48 Line.MustBeDeclaration = Stack.back(); 49 else 50 Line.MustBeDeclaration = true; 51 } 52 53 private: 54 UnwrappedLine &Line; 55 std::vector<bool> &Stack; 56 }; 57 58 class ScopedMacroState : public FormatTokenSource { 59 public: 60 ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource, 61 FormatToken *&ResetToken) 62 : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken), 63 PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource), 64 Token(nullptr) { 65 TokenSource = this; 66 Line.Level = 0; 67 Line.InPPDirective = true; 68 } 69 70 ~ScopedMacroState() override { 71 TokenSource = PreviousTokenSource; 72 ResetToken = Token; 73 Line.InPPDirective = false; 74 Line.Level = PreviousLineLevel; 75 } 76 77 FormatToken *getNextToken() override { 78 // The \c UnwrappedLineParser guards against this by never calling 79 // \c getNextToken() after it has encountered the first eof token. 80 assert(!eof()); 81 Token = PreviousTokenSource->getNextToken(); 82 if (eof()) 83 return getFakeEOF(); 84 return Token; 85 } 86 87 unsigned getPosition() override { return PreviousTokenSource->getPosition(); } 88 89 FormatToken *setPosition(unsigned Position) override { 90 Token = PreviousTokenSource->setPosition(Position); 91 return Token; 92 } 93 94 private: 95 bool eof() { return Token && Token->HasUnescapedNewline; } 96 97 FormatToken *getFakeEOF() { 98 static bool EOFInitialized = false; 99 static FormatToken FormatTok; 100 if (!EOFInitialized) { 101 FormatTok.Tok.startToken(); 102 FormatTok.Tok.setKind(tok::eof); 103 EOFInitialized = true; 104 } 105 return &FormatTok; 106 } 107 108 UnwrappedLine &Line; 109 FormatTokenSource *&TokenSource; 110 FormatToken *&ResetToken; 111 unsigned PreviousLineLevel; 112 FormatTokenSource *PreviousTokenSource; 113 114 FormatToken *Token; 115 }; 116 117 } // end anonymous namespace 118 119 class ScopedLineState { 120 public: 121 ScopedLineState(UnwrappedLineParser &Parser, 122 bool SwitchToPreprocessorLines = false) 123 : Parser(Parser), OriginalLines(Parser.CurrentLines) { 124 if (SwitchToPreprocessorLines) 125 Parser.CurrentLines = &Parser.PreprocessorDirectives; 126 else if (!Parser.Line->Tokens.empty()) 127 Parser.CurrentLines = &Parser.Line->Tokens.back().Children; 128 PreBlockLine = std::move(Parser.Line); 129 Parser.Line = llvm::make_unique<UnwrappedLine>(); 130 Parser.Line->Level = PreBlockLine->Level; 131 Parser.Line->InPPDirective = PreBlockLine->InPPDirective; 132 } 133 134 ~ScopedLineState() { 135 if (!Parser.Line->Tokens.empty()) { 136 Parser.addUnwrappedLine(); 137 } 138 assert(Parser.Line->Tokens.empty()); 139 Parser.Line = std::move(PreBlockLine); 140 if (Parser.CurrentLines == &Parser.PreprocessorDirectives) 141 Parser.MustBreakBeforeNextToken = true; 142 Parser.CurrentLines = OriginalLines; 143 } 144 145 private: 146 UnwrappedLineParser &Parser; 147 148 std::unique_ptr<UnwrappedLine> PreBlockLine; 149 SmallVectorImpl<UnwrappedLine> *OriginalLines; 150 }; 151 152 class CompoundStatementIndenter { 153 public: 154 CompoundStatementIndenter(UnwrappedLineParser *Parser, 155 const FormatStyle &Style, unsigned &LineLevel) 156 : LineLevel(LineLevel), OldLineLevel(LineLevel) { 157 if (Style.BraceWrapping.AfterControlStatement) 158 Parser->addUnwrappedLine(); 159 if (Style.BraceWrapping.IndentBraces) 160 ++LineLevel; 161 } 162 ~CompoundStatementIndenter() { LineLevel = OldLineLevel; } 163 164 private: 165 unsigned &LineLevel; 166 unsigned OldLineLevel; 167 }; 168 169 namespace { 170 171 class IndexedTokenSource : public FormatTokenSource { 172 public: 173 IndexedTokenSource(ArrayRef<FormatToken *> Tokens) 174 : Tokens(Tokens), Position(-1) {} 175 176 FormatToken *getNextToken() override { 177 ++Position; 178 return Tokens[Position]; 179 } 180 181 unsigned getPosition() override { 182 assert(Position >= 0); 183 return Position; 184 } 185 186 FormatToken *setPosition(unsigned P) override { 187 Position = P; 188 return Tokens[Position]; 189 } 190 191 void reset() { Position = -1; } 192 193 private: 194 ArrayRef<FormatToken *> Tokens; 195 int Position; 196 }; 197 198 } // end anonymous namespace 199 200 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style, 201 const AdditionalKeywords &Keywords, 202 ArrayRef<FormatToken *> Tokens, 203 UnwrappedLineConsumer &Callback) 204 : Line(new UnwrappedLine), MustBreakBeforeNextToken(false), 205 CurrentLines(&Lines), Style(Style), Keywords(Keywords), 206 CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr), 207 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1) {} 208 209 void UnwrappedLineParser::reset() { 210 PPBranchLevel = -1; 211 Line.reset(new UnwrappedLine); 212 CommentsBeforeNextToken.clear(); 213 FormatTok = nullptr; 214 MustBreakBeforeNextToken = false; 215 PreprocessorDirectives.clear(); 216 CurrentLines = &Lines; 217 DeclarationScopeStack.clear(); 218 PPStack.clear(); 219 } 220 221 void UnwrappedLineParser::parse() { 222 IndexedTokenSource TokenSource(AllTokens); 223 do { 224 DEBUG(llvm::dbgs() << "----\n"); 225 reset(); 226 Tokens = &TokenSource; 227 TokenSource.reset(); 228 229 readToken(); 230 parseFile(); 231 // Create line with eof token. 232 pushToken(FormatTok); 233 addUnwrappedLine(); 234 235 for (SmallVectorImpl<UnwrappedLine>::iterator I = Lines.begin(), 236 E = Lines.end(); 237 I != E; ++I) { 238 Callback.consumeUnwrappedLine(*I); 239 } 240 Callback.finishRun(); 241 Lines.clear(); 242 while (!PPLevelBranchIndex.empty() && 243 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) { 244 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1); 245 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1); 246 } 247 if (!PPLevelBranchIndex.empty()) { 248 ++PPLevelBranchIndex.back(); 249 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size()); 250 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back()); 251 } 252 } while (!PPLevelBranchIndex.empty()); 253 } 254 255 void UnwrappedLineParser::parseFile() { 256 // The top-level context in a file always has declarations, except for pre- 257 // processor directives and JavaScript files. 258 bool MustBeDeclaration = 259 !Line->InPPDirective && Style.Language != FormatStyle::LK_JavaScript; 260 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 261 MustBeDeclaration); 262 parseLevel(/*HasOpeningBrace=*/false); 263 // Make sure to format the remaining tokens. 264 flushComments(true); 265 addUnwrappedLine(); 266 } 267 268 void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) { 269 bool SwitchLabelEncountered = false; 270 do { 271 tok::TokenKind kind = FormatTok->Tok.getKind(); 272 if (FormatTok->Type == TT_MacroBlockBegin) { 273 kind = tok::l_brace; 274 } else if (FormatTok->Type == TT_MacroBlockEnd) { 275 kind = tok::r_brace; 276 } 277 278 switch (kind) { 279 case tok::comment: 280 nextToken(); 281 addUnwrappedLine(); 282 break; 283 case tok::l_brace: 284 // FIXME: Add parameter whether this can happen - if this happens, we must 285 // be in a non-declaration context. 286 if (!FormatTok->is(TT_MacroBlockBegin) && tryToParseBracedList()) 287 continue; 288 parseBlock(/*MustBeDeclaration=*/false); 289 addUnwrappedLine(); 290 break; 291 case tok::r_brace: 292 if (HasOpeningBrace) 293 return; 294 nextToken(); 295 addUnwrappedLine(); 296 break; 297 case tok::kw_default: 298 case tok::kw_case: 299 if (!SwitchLabelEncountered && 300 (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1))) 301 ++Line->Level; 302 SwitchLabelEncountered = true; 303 parseStructuralElement(); 304 break; 305 default: 306 parseStructuralElement(); 307 break; 308 } 309 } while (!eof()); 310 } 311 312 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) { 313 // We'll parse forward through the tokens until we hit 314 // a closing brace or eof - note that getNextToken() will 315 // parse macros, so this will magically work inside macro 316 // definitions, too. 317 unsigned StoredPosition = Tokens->getPosition(); 318 FormatToken *Tok = FormatTok; 319 const FormatToken *PrevTok = getPreviousToken(); 320 // Keep a stack of positions of lbrace tokens. We will 321 // update information about whether an lbrace starts a 322 // braced init list or a different block during the loop. 323 SmallVector<FormatToken *, 8> LBraceStack; 324 assert(Tok->Tok.is(tok::l_brace)); 325 do { 326 // Get next non-comment token. 327 FormatToken *NextTok; 328 unsigned ReadTokens = 0; 329 do { 330 NextTok = Tokens->getNextToken(); 331 ++ReadTokens; 332 } while (NextTok->is(tok::comment)); 333 334 switch (Tok->Tok.getKind()) { 335 case tok::l_brace: 336 if (Style.Language == FormatStyle::LK_JavaScript && PrevTok && 337 PrevTok->is(tok::colon)) 338 // In TypeScript's TypeMemberLists, there can be semicolons between the 339 // individual members. 340 Tok->BlockKind = BK_BracedInit; 341 else 342 Tok->BlockKind = BK_Unknown; 343 LBraceStack.push_back(Tok); 344 break; 345 case tok::r_brace: 346 if (LBraceStack.empty()) 347 break; 348 if (LBraceStack.back()->BlockKind == BK_Unknown) { 349 bool ProbablyBracedList = false; 350 if (Style.Language == FormatStyle::LK_Proto) { 351 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square); 352 } else { 353 // Using OriginalColumn to distinguish between ObjC methods and 354 // binary operators is a bit hacky. 355 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 356 NextTok->OriginalColumn == 0; 357 358 // If there is a comma, semicolon or right paren after the closing 359 // brace, we assume this is a braced initializer list. Note that 360 // regardless how we mark inner braces here, we will overwrite the 361 // BlockKind later if we parse a braced list (where all blocks 362 // inside are by default braced lists), or when we explicitly detect 363 // blocks (for example while parsing lambdas). 364 ProbablyBracedList = 365 (Style.Language == FormatStyle::LK_JavaScript && 366 NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in, 367 Keywords.kw_as)) || 368 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 369 tok::r_paren, tok::r_square, tok::l_brace, 370 tok::l_square, tok::l_paren, tok::ellipsis) || 371 (NextTok->is(tok::identifier) && 372 !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace)) || 373 (NextTok->is(tok::semi) && 374 (!ExpectClassBody || LBraceStack.size() != 1)) || 375 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 376 } 377 if (ProbablyBracedList) { 378 Tok->BlockKind = BK_BracedInit; 379 LBraceStack.back()->BlockKind = BK_BracedInit; 380 } else { 381 Tok->BlockKind = BK_Block; 382 LBraceStack.back()->BlockKind = BK_Block; 383 } 384 } 385 LBraceStack.pop_back(); 386 break; 387 case tok::at: 388 case tok::semi: 389 case tok::kw_if: 390 case tok::kw_while: 391 case tok::kw_for: 392 case tok::kw_switch: 393 case tok::kw_try: 394 case tok::kw___try: 395 if (!LBraceStack.empty() && LBraceStack.back()->BlockKind == BK_Unknown) 396 LBraceStack.back()->BlockKind = BK_Block; 397 break; 398 default: 399 break; 400 } 401 PrevTok = Tok; 402 Tok = NextTok; 403 } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty()); 404 405 // Assume other blocks for all unclosed opening braces. 406 for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) { 407 if (LBraceStack[i]->BlockKind == BK_Unknown) 408 LBraceStack[i]->BlockKind = BK_Block; 409 } 410 411 FormatTok = Tokens->setPosition(StoredPosition); 412 } 413 414 void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, bool AddLevel, 415 bool MunchSemi) { 416 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) && 417 "'{' or macro block token expected"); 418 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 419 FormatTok->BlockKind = BK_Block; 420 421 unsigned InitialLevel = Line->Level; 422 nextToken(); 423 424 if (MacroBlock && FormatTok->is(tok::l_paren)) 425 parseParens(); 426 427 addUnwrappedLine(); 428 429 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 430 MustBeDeclaration); 431 if (AddLevel) 432 ++Line->Level; 433 parseLevel(/*HasOpeningBrace=*/true); 434 435 if (eof()) 436 return; 437 438 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd) 439 : !FormatTok->is(tok::r_brace)) { 440 Line->Level = InitialLevel; 441 FormatTok->BlockKind = BK_Block; 442 return; 443 } 444 445 nextToken(); // Munch the closing brace. 446 447 if (MacroBlock && FormatTok->is(tok::l_paren)) 448 parseParens(); 449 450 if (MunchSemi && FormatTok->Tok.is(tok::semi)) 451 nextToken(); 452 Line->Level = InitialLevel; 453 } 454 455 static bool isGoogScope(const UnwrappedLine &Line) { 456 // FIXME: Closure-library specific stuff should not be hard-coded but be 457 // configurable. 458 if (Line.Tokens.size() < 4) 459 return false; 460 auto I = Line.Tokens.begin(); 461 if (I->Tok->TokenText != "goog") 462 return false; 463 ++I; 464 if (I->Tok->isNot(tok::period)) 465 return false; 466 ++I; 467 if (I->Tok->TokenText != "scope") 468 return false; 469 ++I; 470 return I->Tok->is(tok::l_paren); 471 } 472 473 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 474 const FormatToken &InitialToken) { 475 if (InitialToken.is(tok::kw_namespace)) 476 return Style.BraceWrapping.AfterNamespace; 477 if (InitialToken.is(tok::kw_class)) 478 return Style.BraceWrapping.AfterClass; 479 if (InitialToken.is(tok::kw_union)) 480 return Style.BraceWrapping.AfterUnion; 481 if (InitialToken.is(tok::kw_struct)) 482 return Style.BraceWrapping.AfterStruct; 483 return false; 484 } 485 486 void UnwrappedLineParser::parseChildBlock() { 487 FormatTok->BlockKind = BK_Block; 488 nextToken(); 489 { 490 bool GoogScope = 491 Style.Language == FormatStyle::LK_JavaScript && isGoogScope(*Line); 492 ScopedLineState LineState(*this); 493 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 494 /*MustBeDeclaration=*/false); 495 Line->Level += GoogScope ? 0 : 1; 496 parseLevel(/*HasOpeningBrace=*/true); 497 flushComments(isOnNewLine(*FormatTok)); 498 Line->Level -= GoogScope ? 0 : 1; 499 } 500 nextToken(); 501 } 502 503 void UnwrappedLineParser::parsePPDirective() { 504 assert(FormatTok->Tok.is(tok::hash) && "'#' expected"); 505 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 506 nextToken(); 507 508 if (!FormatTok->Tok.getIdentifierInfo()) { 509 parsePPUnknown(); 510 return; 511 } 512 513 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 514 case tok::pp_define: 515 parsePPDefine(); 516 return; 517 case tok::pp_if: 518 parsePPIf(/*IfDef=*/false); 519 break; 520 case tok::pp_ifdef: 521 case tok::pp_ifndef: 522 parsePPIf(/*IfDef=*/true); 523 break; 524 case tok::pp_else: 525 parsePPElse(); 526 break; 527 case tok::pp_elif: 528 parsePPElIf(); 529 break; 530 case tok::pp_endif: 531 parsePPEndIf(); 532 break; 533 default: 534 parsePPUnknown(); 535 break; 536 } 537 } 538 539 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 540 if (Unreachable || (!PPStack.empty() && PPStack.back() == PP_Unreachable)) 541 PPStack.push_back(PP_Unreachable); 542 else 543 PPStack.push_back(PP_Conditional); 544 } 545 546 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 547 ++PPBranchLevel; 548 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 549 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 550 PPLevelBranchIndex.push_back(0); 551 PPLevelBranchCount.push_back(0); 552 } 553 PPChainBranchIndex.push(0); 554 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 555 conditionalCompilationCondition(Unreachable || Skip); 556 } 557 558 void UnwrappedLineParser::conditionalCompilationAlternative() { 559 if (!PPStack.empty()) 560 PPStack.pop_back(); 561 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 562 if (!PPChainBranchIndex.empty()) 563 ++PPChainBranchIndex.top(); 564 conditionalCompilationCondition( 565 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 566 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 567 } 568 569 void UnwrappedLineParser::conditionalCompilationEnd() { 570 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 571 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 572 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) { 573 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 574 } 575 } 576 // Guard against #endif's without #if. 577 if (PPBranchLevel > 0) 578 --PPBranchLevel; 579 if (!PPChainBranchIndex.empty()) 580 PPChainBranchIndex.pop(); 581 if (!PPStack.empty()) 582 PPStack.pop_back(); 583 } 584 585 void UnwrappedLineParser::parsePPIf(bool IfDef) { 586 nextToken(); 587 bool IsLiteralFalse = (FormatTok->Tok.isLiteral() && 588 FormatTok->Tok.getLiteralData() != nullptr && 589 StringRef(FormatTok->Tok.getLiteralData(), 590 FormatTok->Tok.getLength()) == "0") || 591 FormatTok->Tok.is(tok::kw_false); 592 conditionalCompilationStart(!IfDef && IsLiteralFalse); 593 parsePPUnknown(); 594 } 595 596 void UnwrappedLineParser::parsePPElse() { 597 conditionalCompilationAlternative(); 598 parsePPUnknown(); 599 } 600 601 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 602 603 void UnwrappedLineParser::parsePPEndIf() { 604 conditionalCompilationEnd(); 605 parsePPUnknown(); 606 } 607 608 void UnwrappedLineParser::parsePPDefine() { 609 nextToken(); 610 611 if (FormatTok->Tok.getKind() != tok::identifier) { 612 parsePPUnknown(); 613 return; 614 } 615 nextToken(); 616 if (FormatTok->Tok.getKind() == tok::l_paren && 617 FormatTok->WhitespaceRange.getBegin() == 618 FormatTok->WhitespaceRange.getEnd()) { 619 parseParens(); 620 } 621 addUnwrappedLine(); 622 Line->Level = 1; 623 624 // Errors during a preprocessor directive can only affect the layout of the 625 // preprocessor directive, and thus we ignore them. An alternative approach 626 // would be to use the same approach we use on the file level (no 627 // re-indentation if there was a structural error) within the macro 628 // definition. 629 parseFile(); 630 } 631 632 void UnwrappedLineParser::parsePPUnknown() { 633 do { 634 nextToken(); 635 } while (!eof()); 636 addUnwrappedLine(); 637 } 638 639 // Here we blacklist certain tokens that are not usually the first token in an 640 // unwrapped line. This is used in attempt to distinguish macro calls without 641 // trailing semicolons from other constructs split to several lines. 642 static bool tokenCanStartNewLine(const clang::Token &Tok) { 643 // Semicolon can be a null-statement, l_square can be a start of a macro or 644 // a C++11 attribute, but this doesn't seem to be common. 645 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 646 Tok.isNot(tok::l_square) && 647 // Tokens that can only be used as binary operators and a part of 648 // overloaded operator names. 649 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 650 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 651 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 652 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 653 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 654 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 655 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 656 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 657 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 658 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 659 Tok.isNot(tok::lesslessequal) && 660 // Colon is used in labels, base class lists, initializer lists, 661 // range-based for loops, ternary operator, but should never be the 662 // first token in an unwrapped line. 663 Tok.isNot(tok::colon) && 664 // 'noexcept' is a trailing annotation. 665 Tok.isNot(tok::kw_noexcept); 666 } 667 668 static bool mustBeJSIdent(const AdditionalKeywords &Keywords, 669 const FormatToken *FormatTok) { 670 // FIXME: This returns true for C/C++ keywords like 'struct'. 671 return FormatTok->is(tok::identifier) && 672 (FormatTok->Tok.getIdentifierInfo() == nullptr || 673 !FormatTok->isOneOf( 674 Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async, 675 Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally, 676 Keywords.kw_function, Keywords.kw_import, Keywords.kw_is, 677 Keywords.kw_let, Keywords.kw_var, tok::kw_const, 678 Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements, 679 Keywords.kw_instanceof, Keywords.kw_interface, 680 Keywords.kw_throws)); 681 } 682 683 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords, 684 const FormatToken *FormatTok) { 685 return FormatTok->Tok.isLiteral() || 686 FormatTok->isOneOf(tok::kw_true, tok::kw_false) || 687 mustBeJSIdent(Keywords, FormatTok); 688 } 689 690 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement 691 // when encountered after a value (see mustBeJSIdentOrValue). 692 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords, 693 const FormatToken *FormatTok) { 694 return FormatTok->isOneOf( 695 tok::kw_return, Keywords.kw_yield, 696 // conditionals 697 tok::kw_if, tok::kw_else, 698 // loops 699 tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break, 700 // switch/case 701 tok::kw_switch, tok::kw_case, 702 // exceptions 703 tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally, 704 // declaration 705 tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let, 706 Keywords.kw_async, Keywords.kw_function, 707 // import/export 708 Keywords.kw_import, tok::kw_export); 709 } 710 711 // readTokenWithJavaScriptASI reads the next token and terminates the current 712 // line if JavaScript Automatic Semicolon Insertion must 713 // happen between the current token and the next token. 714 // 715 // This method is conservative - it cannot cover all edge cases of JavaScript, 716 // but only aims to correctly handle certain well known cases. It *must not* 717 // return true in speculative cases. 718 void UnwrappedLineParser::readTokenWithJavaScriptASI() { 719 FormatToken *Previous = FormatTok; 720 readToken(); 721 FormatToken *Next = FormatTok; 722 723 bool IsOnSameLine = 724 CommentsBeforeNextToken.empty() 725 ? Next->NewlinesBefore == 0 726 : CommentsBeforeNextToken.front()->NewlinesBefore == 0; 727 if (IsOnSameLine) 728 return; 729 730 bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous); 731 bool PreviousStartsTemplateExpr = 732 Previous->is(TT_TemplateString) && Previous->TokenText.endswith("${"); 733 if (PreviousMustBeValue && Line && Line->Tokens.size() > 1) { 734 // If the token before the previous one is an '@', the previous token is an 735 // annotation and can precede another identifier/value. 736 const FormatToken *PrePrevious = std::prev(Line->Tokens.end(), 2)->Tok; 737 if (PrePrevious->is(tok::at)) 738 return; 739 } 740 if (Next->is(tok::exclaim) && PreviousMustBeValue) 741 return addUnwrappedLine(); 742 bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next); 743 bool NextEndsTemplateExpr = 744 Next->is(TT_TemplateString) && Next->TokenText.startswith("}"); 745 if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr && 746 (PreviousMustBeValue || 747 Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus, 748 tok::minusminus))) 749 return addUnwrappedLine(); 750 if (PreviousMustBeValue && isJSDeclOrStmt(Keywords, Next)) 751 return addUnwrappedLine(); 752 } 753 754 void UnwrappedLineParser::parseStructuralElement() { 755 assert(!FormatTok->is(tok::l_brace)); 756 if (Style.Language == FormatStyle::LK_TableGen && 757 FormatTok->is(tok::pp_include)) { 758 nextToken(); 759 if (FormatTok->is(tok::string_literal)) 760 nextToken(); 761 addUnwrappedLine(); 762 return; 763 } 764 switch (FormatTok->Tok.getKind()) { 765 case tok::at: 766 nextToken(); 767 if (FormatTok->Tok.is(tok::l_brace)) { 768 parseBracedList(); 769 break; 770 } 771 switch (FormatTok->Tok.getObjCKeywordID()) { 772 case tok::objc_public: 773 case tok::objc_protected: 774 case tok::objc_package: 775 case tok::objc_private: 776 return parseAccessSpecifier(); 777 case tok::objc_interface: 778 case tok::objc_implementation: 779 return parseObjCInterfaceOrImplementation(); 780 case tok::objc_protocol: 781 return parseObjCProtocol(); 782 case tok::objc_end: 783 return; // Handled by the caller. 784 case tok::objc_optional: 785 case tok::objc_required: 786 nextToken(); 787 addUnwrappedLine(); 788 return; 789 case tok::objc_autoreleasepool: 790 nextToken(); 791 if (FormatTok->Tok.is(tok::l_brace)) { 792 if (Style.BraceWrapping.AfterObjCDeclaration) 793 addUnwrappedLine(); 794 parseBlock(/*MustBeDeclaration=*/false); 795 } 796 addUnwrappedLine(); 797 return; 798 case tok::objc_try: 799 // This branch isn't strictly necessary (the kw_try case below would 800 // do this too after the tok::at is parsed above). But be explicit. 801 parseTryCatch(); 802 return; 803 default: 804 break; 805 } 806 break; 807 case tok::kw_asm: 808 nextToken(); 809 if (FormatTok->is(tok::l_brace)) { 810 FormatTok->Type = TT_InlineASMBrace; 811 nextToken(); 812 while (FormatTok && FormatTok->isNot(tok::eof)) { 813 if (FormatTok->is(tok::r_brace)) { 814 FormatTok->Type = TT_InlineASMBrace; 815 nextToken(); 816 addUnwrappedLine(); 817 break; 818 } 819 FormatTok->Finalized = true; 820 nextToken(); 821 } 822 } 823 break; 824 case tok::kw_namespace: 825 parseNamespace(); 826 return; 827 case tok::kw_inline: 828 nextToken(); 829 if (FormatTok->Tok.is(tok::kw_namespace)) { 830 parseNamespace(); 831 return; 832 } 833 break; 834 case tok::kw_public: 835 case tok::kw_protected: 836 case tok::kw_private: 837 if (Style.Language == FormatStyle::LK_Java || 838 Style.Language == FormatStyle::LK_JavaScript) 839 nextToken(); 840 else 841 parseAccessSpecifier(); 842 return; 843 case tok::kw_if: 844 parseIfThenElse(); 845 return; 846 case tok::kw_for: 847 case tok::kw_while: 848 parseForOrWhileLoop(); 849 return; 850 case tok::kw_do: 851 parseDoWhile(); 852 return; 853 case tok::kw_switch: 854 parseSwitch(); 855 return; 856 case tok::kw_default: 857 nextToken(); 858 parseLabel(); 859 return; 860 case tok::kw_case: 861 parseCaseLabel(); 862 return; 863 case tok::kw_try: 864 case tok::kw___try: 865 parseTryCatch(); 866 return; 867 case tok::kw_extern: 868 nextToken(); 869 if (FormatTok->Tok.is(tok::string_literal)) { 870 nextToken(); 871 if (FormatTok->Tok.is(tok::l_brace)) { 872 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/false); 873 addUnwrappedLine(); 874 return; 875 } 876 } 877 break; 878 case tok::kw_export: 879 if (Style.Language == FormatStyle::LK_JavaScript) { 880 parseJavaScriptEs6ImportExport(); 881 return; 882 } 883 break; 884 case tok::identifier: 885 if (FormatTok->is(TT_ForEachMacro)) { 886 parseForOrWhileLoop(); 887 return; 888 } 889 if (FormatTok->is(TT_MacroBlockBegin)) { 890 parseBlock(/*MustBeDeclaration=*/false, /*AddLevel=*/true, 891 /*MunchSemi=*/false); 892 return; 893 } 894 if (FormatTok->is(Keywords.kw_import)) { 895 if (Style.Language == FormatStyle::LK_JavaScript) { 896 parseJavaScriptEs6ImportExport(); 897 return; 898 } 899 if (Style.Language == FormatStyle::LK_Proto) { 900 nextToken(); 901 if (FormatTok->is(tok::kw_public)) 902 nextToken(); 903 if (!FormatTok->is(tok::string_literal)) 904 return; 905 nextToken(); 906 if (FormatTok->is(tok::semi)) 907 nextToken(); 908 addUnwrappedLine(); 909 return; 910 } 911 } 912 if (FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 913 Keywords.kw_slots, Keywords.kw_qslots)) { 914 nextToken(); 915 if (FormatTok->is(tok::colon)) { 916 nextToken(); 917 addUnwrappedLine(); 918 return; 919 } 920 } 921 // In all other cases, parse the declaration. 922 break; 923 default: 924 break; 925 } 926 do { 927 const FormatToken *Previous = getPreviousToken(); 928 switch (FormatTok->Tok.getKind()) { 929 case tok::at: 930 nextToken(); 931 if (FormatTok->Tok.is(tok::l_brace)) 932 parseBracedList(); 933 break; 934 case tok::kw_enum: 935 // Ignore if this is part of "template <enum ...". 936 if (Previous && Previous->is(tok::less)) { 937 nextToken(); 938 break; 939 } 940 941 // parseEnum falls through and does not yet add an unwrapped line as an 942 // enum definition can start a structural element. 943 if (!parseEnum()) 944 break; 945 // This only applies for C++. 946 if (Style.Language != FormatStyle::LK_Cpp) { 947 addUnwrappedLine(); 948 return; 949 } 950 break; 951 case tok::kw_typedef: 952 nextToken(); 953 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 954 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS)) 955 parseEnum(); 956 break; 957 case tok::kw_struct: 958 case tok::kw_union: 959 case tok::kw_class: 960 // parseRecord falls through and does not yet add an unwrapped line as a 961 // record declaration or definition can start a structural element. 962 parseRecord(); 963 // This does not apply for Java and JavaScript. 964 if (Style.Language == FormatStyle::LK_Java || 965 Style.Language == FormatStyle::LK_JavaScript) { 966 if (FormatTok->is(tok::semi)) 967 nextToken(); 968 addUnwrappedLine(); 969 return; 970 } 971 break; 972 case tok::period: 973 nextToken(); 974 // In Java, classes have an implicit static member "class". 975 if (Style.Language == FormatStyle::LK_Java && FormatTok && 976 FormatTok->is(tok::kw_class)) 977 nextToken(); 978 if (Style.Language == FormatStyle::LK_JavaScript && FormatTok && 979 FormatTok->Tok.getIdentifierInfo()) 980 // JavaScript only has pseudo keywords, all keywords are allowed to 981 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 982 nextToken(); 983 break; 984 case tok::semi: 985 nextToken(); 986 addUnwrappedLine(); 987 return; 988 case tok::r_brace: 989 addUnwrappedLine(); 990 return; 991 case tok::l_paren: 992 parseParens(); 993 break; 994 case tok::kw_operator: 995 nextToken(); 996 if (FormatTok->isBinaryOperator()) 997 nextToken(); 998 break; 999 case tok::caret: 1000 nextToken(); 1001 if (FormatTok->Tok.isAnyIdentifier() || 1002 FormatTok->isSimpleTypeSpecifier()) 1003 nextToken(); 1004 if (FormatTok->is(tok::l_paren)) 1005 parseParens(); 1006 if (FormatTok->is(tok::l_brace)) 1007 parseChildBlock(); 1008 break; 1009 case tok::l_brace: 1010 if (!tryToParseBracedList()) { 1011 // A block outside of parentheses must be the last part of a 1012 // structural element. 1013 // FIXME: Figure out cases where this is not true, and add projections 1014 // for them (the one we know is missing are lambdas). 1015 if (Style.BraceWrapping.AfterFunction) 1016 addUnwrappedLine(); 1017 FormatTok->Type = TT_FunctionLBrace; 1018 parseBlock(/*MustBeDeclaration=*/false); 1019 addUnwrappedLine(); 1020 return; 1021 } 1022 // Otherwise this was a braced init list, and the structural 1023 // element continues. 1024 break; 1025 case tok::kw_try: 1026 // We arrive here when parsing function-try blocks. 1027 parseTryCatch(); 1028 return; 1029 case tok::identifier: { 1030 if (FormatTok->is(TT_MacroBlockEnd)) { 1031 addUnwrappedLine(); 1032 return; 1033 } 1034 1035 // Parse function literal unless 'function' is the first token in a line 1036 // in which case this should be treated as a free-standing function. 1037 if (Style.Language == FormatStyle::LK_JavaScript && 1038 (FormatTok->is(Keywords.kw_function) || 1039 FormatTok->startsSequence(Keywords.kw_async, 1040 Keywords.kw_function)) && 1041 Line->Tokens.size() > 0) { 1042 tryToParseJSFunction(); 1043 break; 1044 } 1045 if ((Style.Language == FormatStyle::LK_JavaScript || 1046 Style.Language == FormatStyle::LK_Java) && 1047 FormatTok->is(Keywords.kw_interface)) { 1048 if (Style.Language == FormatStyle::LK_JavaScript) { 1049 // In JavaScript/TypeScript, "interface" can be used as a standalone 1050 // identifier, e.g. in `var interface = 1;`. If "interface" is 1051 // followed by another identifier, it is very like to be an actual 1052 // interface declaration. 1053 unsigned StoredPosition = Tokens->getPosition(); 1054 FormatToken *Next = Tokens->getNextToken(); 1055 FormatTok = Tokens->setPosition(StoredPosition); 1056 if (Next && !mustBeJSIdent(Keywords, Next)) { 1057 nextToken(); 1058 break; 1059 } 1060 } 1061 parseRecord(); 1062 addUnwrappedLine(); 1063 return; 1064 } 1065 1066 // See if the following token should start a new unwrapped line. 1067 StringRef Text = FormatTok->TokenText; 1068 nextToken(); 1069 if (Line->Tokens.size() == 1 && 1070 // JS doesn't have macros, and within classes colons indicate fields, 1071 // not labels. 1072 Style.Language != FormatStyle::LK_JavaScript) { 1073 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) { 1074 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1075 parseLabel(); 1076 return; 1077 } 1078 // Recognize function-like macro usages without trailing semicolon as 1079 // well as free-standing macros like Q_OBJECT. 1080 bool FunctionLike = FormatTok->is(tok::l_paren); 1081 if (FunctionLike) 1082 parseParens(); 1083 1084 bool FollowedByNewline = 1085 CommentsBeforeNextToken.empty() 1086 ? FormatTok->NewlinesBefore > 0 1087 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1088 1089 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1090 tokenCanStartNewLine(FormatTok->Tok) && Text == Text.upper()) { 1091 addUnwrappedLine(); 1092 return; 1093 } 1094 } 1095 break; 1096 } 1097 case tok::equal: 1098 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType 1099 // TT_JsFatArrow. The always start an expression or a child block if 1100 // followed by a curly. 1101 if (FormatTok->is(TT_JsFatArrow)) { 1102 nextToken(); 1103 if (FormatTok->is(tok::l_brace)) 1104 parseChildBlock(); 1105 break; 1106 } 1107 1108 nextToken(); 1109 if (FormatTok->Tok.is(tok::l_brace)) { 1110 parseBracedList(); 1111 } 1112 break; 1113 case tok::l_square: 1114 parseSquare(); 1115 break; 1116 case tok::kw_new: 1117 parseNew(); 1118 break; 1119 default: 1120 nextToken(); 1121 break; 1122 } 1123 } while (!eof()); 1124 } 1125 1126 bool UnwrappedLineParser::tryToParseLambda() { 1127 if (Style.Language != FormatStyle::LK_Cpp) { 1128 nextToken(); 1129 return false; 1130 } 1131 const FormatToken* Previous = getPreviousToken(); 1132 if (Previous && 1133 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 1134 tok::kw_delete) || 1135 Previous->closesScope() || Previous->isSimpleTypeSpecifier())) { 1136 nextToken(); 1137 return false; 1138 } 1139 assert(FormatTok->is(tok::l_square)); 1140 FormatToken &LSquare = *FormatTok; 1141 if (!tryToParseLambdaIntroducer()) 1142 return false; 1143 1144 while (FormatTok->isNot(tok::l_brace)) { 1145 if (FormatTok->isSimpleTypeSpecifier()) { 1146 nextToken(); 1147 continue; 1148 } 1149 switch (FormatTok->Tok.getKind()) { 1150 case tok::l_brace: 1151 break; 1152 case tok::l_paren: 1153 parseParens(); 1154 break; 1155 case tok::amp: 1156 case tok::star: 1157 case tok::kw_const: 1158 case tok::comma: 1159 case tok::less: 1160 case tok::greater: 1161 case tok::identifier: 1162 case tok::numeric_constant: 1163 case tok::coloncolon: 1164 case tok::kw_mutable: 1165 nextToken(); 1166 break; 1167 case tok::arrow: 1168 FormatTok->Type = TT_LambdaArrow; 1169 nextToken(); 1170 break; 1171 default: 1172 return true; 1173 } 1174 } 1175 LSquare.Type = TT_LambdaLSquare; 1176 parseChildBlock(); 1177 return true; 1178 } 1179 1180 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 1181 nextToken(); 1182 if (FormatTok->is(tok::equal)) { 1183 nextToken(); 1184 if (FormatTok->is(tok::r_square)) { 1185 nextToken(); 1186 return true; 1187 } 1188 if (FormatTok->isNot(tok::comma)) 1189 return false; 1190 nextToken(); 1191 } else if (FormatTok->is(tok::amp)) { 1192 nextToken(); 1193 if (FormatTok->is(tok::r_square)) { 1194 nextToken(); 1195 return true; 1196 } 1197 if (!FormatTok->isOneOf(tok::comma, tok::identifier)) { 1198 return false; 1199 } 1200 if (FormatTok->is(tok::comma)) 1201 nextToken(); 1202 } else if (FormatTok->is(tok::r_square)) { 1203 nextToken(); 1204 return true; 1205 } 1206 do { 1207 if (FormatTok->is(tok::amp)) 1208 nextToken(); 1209 if (!FormatTok->isOneOf(tok::identifier, tok::kw_this)) 1210 return false; 1211 nextToken(); 1212 if (FormatTok->is(tok::ellipsis)) 1213 nextToken(); 1214 if (FormatTok->is(tok::comma)) { 1215 nextToken(); 1216 } else if (FormatTok->is(tok::r_square)) { 1217 nextToken(); 1218 return true; 1219 } else { 1220 return false; 1221 } 1222 } while (!eof()); 1223 return false; 1224 } 1225 1226 void UnwrappedLineParser::tryToParseJSFunction() { 1227 assert(FormatTok->is(Keywords.kw_function) || 1228 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)); 1229 if (FormatTok->is(Keywords.kw_async)) 1230 nextToken(); 1231 // Consume "function". 1232 nextToken(); 1233 1234 // Consume * (generator function). Treat it like C++'s overloaded operators. 1235 if (FormatTok->is(tok::star)) { 1236 FormatTok->Type = TT_OverloadedOperator; 1237 nextToken(); 1238 } 1239 1240 // Consume function name. 1241 if (FormatTok->is(tok::identifier)) 1242 nextToken(); 1243 1244 if (FormatTok->isNot(tok::l_paren)) 1245 return; 1246 1247 // Parse formal parameter list. 1248 parseParens(); 1249 1250 if (FormatTok->is(tok::colon)) { 1251 // Parse a type definition. 1252 nextToken(); 1253 1254 // Eat the type declaration. For braced inline object types, balance braces, 1255 // otherwise just parse until finding an l_brace for the function body. 1256 if (FormatTok->is(tok::l_brace)) 1257 tryToParseBracedList(); 1258 else 1259 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 1260 nextToken(); 1261 } 1262 1263 if (FormatTok->is(tok::semi)) 1264 return; 1265 1266 parseChildBlock(); 1267 } 1268 1269 bool UnwrappedLineParser::tryToParseBracedList() { 1270 if (FormatTok->BlockKind == BK_Unknown) 1271 calculateBraceTypes(); 1272 assert(FormatTok->BlockKind != BK_Unknown); 1273 if (FormatTok->BlockKind == BK_Block) 1274 return false; 1275 parseBracedList(); 1276 return true; 1277 } 1278 1279 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons) { 1280 bool HasError = false; 1281 nextToken(); 1282 1283 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 1284 // replace this by using parseAssigmentExpression() inside. 1285 do { 1286 if (Style.Language == FormatStyle::LK_JavaScript) { 1287 if (FormatTok->is(Keywords.kw_function) || 1288 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) { 1289 tryToParseJSFunction(); 1290 continue; 1291 } 1292 if (FormatTok->is(TT_JsFatArrow)) { 1293 nextToken(); 1294 // Fat arrows can be followed by simple expressions or by child blocks 1295 // in curly braces. 1296 if (FormatTok->is(tok::l_brace)) { 1297 parseChildBlock(); 1298 continue; 1299 } 1300 } 1301 } 1302 switch (FormatTok->Tok.getKind()) { 1303 case tok::caret: 1304 nextToken(); 1305 if (FormatTok->is(tok::l_brace)) { 1306 parseChildBlock(); 1307 } 1308 break; 1309 case tok::l_square: 1310 tryToParseLambda(); 1311 break; 1312 case tok::l_brace: 1313 // Assume there are no blocks inside a braced init list apart 1314 // from the ones we explicitly parse out (like lambdas). 1315 FormatTok->BlockKind = BK_BracedInit; 1316 parseBracedList(); 1317 break; 1318 case tok::l_paren: 1319 parseParens(); 1320 // JavaScript can just have free standing methods and getters/setters in 1321 // object literals. Detect them by a "{" following ")". 1322 if (Style.Language == FormatStyle::LK_JavaScript) { 1323 if (FormatTok->is(tok::l_brace)) 1324 parseChildBlock(); 1325 break; 1326 } 1327 break; 1328 case tok::r_brace: 1329 nextToken(); 1330 return !HasError; 1331 case tok::semi: 1332 // JavaScript (or more precisely TypeScript) can have semicolons in braced 1333 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 1334 // used for error recovery if we have otherwise determined that this is 1335 // a braced list. 1336 if (Style.Language == FormatStyle::LK_JavaScript) { 1337 nextToken(); 1338 break; 1339 } 1340 HasError = true; 1341 if (!ContinueOnSemicolons) 1342 return !HasError; 1343 nextToken(); 1344 break; 1345 case tok::comma: 1346 nextToken(); 1347 break; 1348 default: 1349 nextToken(); 1350 break; 1351 } 1352 } while (!eof()); 1353 return false; 1354 } 1355 1356 void UnwrappedLineParser::parseParens() { 1357 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 1358 nextToken(); 1359 do { 1360 switch (FormatTok->Tok.getKind()) { 1361 case tok::l_paren: 1362 parseParens(); 1363 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 1364 parseChildBlock(); 1365 break; 1366 case tok::r_paren: 1367 nextToken(); 1368 return; 1369 case tok::r_brace: 1370 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1371 return; 1372 case tok::l_square: 1373 tryToParseLambda(); 1374 break; 1375 case tok::l_brace: 1376 if (!tryToParseBracedList()) 1377 parseChildBlock(); 1378 break; 1379 case tok::at: 1380 nextToken(); 1381 if (FormatTok->Tok.is(tok::l_brace)) 1382 parseBracedList(); 1383 break; 1384 case tok::identifier: 1385 if (Style.Language == FormatStyle::LK_JavaScript && 1386 (FormatTok->is(Keywords.kw_function) || 1387 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function))) 1388 tryToParseJSFunction(); 1389 else 1390 nextToken(); 1391 break; 1392 default: 1393 nextToken(); 1394 break; 1395 } 1396 } while (!eof()); 1397 } 1398 1399 void UnwrappedLineParser::parseSquare() { 1400 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected."); 1401 if (tryToParseLambda()) 1402 return; 1403 do { 1404 switch (FormatTok->Tok.getKind()) { 1405 case tok::l_paren: 1406 parseParens(); 1407 break; 1408 case tok::r_square: 1409 nextToken(); 1410 return; 1411 case tok::r_brace: 1412 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1413 return; 1414 case tok::l_square: 1415 parseSquare(); 1416 break; 1417 case tok::l_brace: { 1418 if (!tryToParseBracedList()) 1419 parseChildBlock(); 1420 break; 1421 } 1422 case tok::at: 1423 nextToken(); 1424 if (FormatTok->Tok.is(tok::l_brace)) 1425 parseBracedList(); 1426 break; 1427 default: 1428 nextToken(); 1429 break; 1430 } 1431 } while (!eof()); 1432 } 1433 1434 void UnwrappedLineParser::parseIfThenElse() { 1435 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 1436 nextToken(); 1437 if (FormatTok->Tok.is(tok::l_paren)) 1438 parseParens(); 1439 bool NeedsUnwrappedLine = false; 1440 if (FormatTok->Tok.is(tok::l_brace)) { 1441 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1442 parseBlock(/*MustBeDeclaration=*/false); 1443 if (Style.BraceWrapping.BeforeElse) 1444 addUnwrappedLine(); 1445 else 1446 NeedsUnwrappedLine = true; 1447 } else { 1448 addUnwrappedLine(); 1449 ++Line->Level; 1450 parseStructuralElement(); 1451 --Line->Level; 1452 } 1453 if (FormatTok->Tok.is(tok::kw_else)) { 1454 nextToken(); 1455 if (FormatTok->Tok.is(tok::l_brace)) { 1456 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1457 parseBlock(/*MustBeDeclaration=*/false); 1458 addUnwrappedLine(); 1459 } else if (FormatTok->Tok.is(tok::kw_if)) { 1460 parseIfThenElse(); 1461 } else { 1462 addUnwrappedLine(); 1463 ++Line->Level; 1464 parseStructuralElement(); 1465 if (FormatTok->is(tok::eof)) 1466 addUnwrappedLine(); 1467 --Line->Level; 1468 } 1469 } else if (NeedsUnwrappedLine) { 1470 addUnwrappedLine(); 1471 } 1472 } 1473 1474 void UnwrappedLineParser::parseTryCatch() { 1475 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 1476 nextToken(); 1477 bool NeedsUnwrappedLine = false; 1478 if (FormatTok->is(tok::colon)) { 1479 // We are in a function try block, what comes is an initializer list. 1480 nextToken(); 1481 while (FormatTok->is(tok::identifier)) { 1482 nextToken(); 1483 if (FormatTok->is(tok::l_paren)) 1484 parseParens(); 1485 if (FormatTok->is(tok::comma)) 1486 nextToken(); 1487 } 1488 } 1489 // Parse try with resource. 1490 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) { 1491 parseParens(); 1492 } 1493 if (FormatTok->is(tok::l_brace)) { 1494 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1495 parseBlock(/*MustBeDeclaration=*/false); 1496 if (Style.BraceWrapping.BeforeCatch) { 1497 addUnwrappedLine(); 1498 } else { 1499 NeedsUnwrappedLine = true; 1500 } 1501 } else if (!FormatTok->is(tok::kw_catch)) { 1502 // The C++ standard requires a compound-statement after a try. 1503 // If there's none, we try to assume there's a structuralElement 1504 // and try to continue. 1505 addUnwrappedLine(); 1506 ++Line->Level; 1507 parseStructuralElement(); 1508 --Line->Level; 1509 } 1510 while (1) { 1511 if (FormatTok->is(tok::at)) 1512 nextToken(); 1513 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 1514 tok::kw___finally) || 1515 ((Style.Language == FormatStyle::LK_Java || 1516 Style.Language == FormatStyle::LK_JavaScript) && 1517 FormatTok->is(Keywords.kw_finally)) || 1518 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) || 1519 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally)))) 1520 break; 1521 nextToken(); 1522 while (FormatTok->isNot(tok::l_brace)) { 1523 if (FormatTok->is(tok::l_paren)) { 1524 parseParens(); 1525 continue; 1526 } 1527 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) 1528 return; 1529 nextToken(); 1530 } 1531 NeedsUnwrappedLine = false; 1532 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1533 parseBlock(/*MustBeDeclaration=*/false); 1534 if (Style.BraceWrapping.BeforeCatch) 1535 addUnwrappedLine(); 1536 else 1537 NeedsUnwrappedLine = true; 1538 } 1539 if (NeedsUnwrappedLine) 1540 addUnwrappedLine(); 1541 } 1542 1543 void UnwrappedLineParser::parseNamespace() { 1544 assert(FormatTok->Tok.is(tok::kw_namespace) && "'namespace' expected"); 1545 1546 const FormatToken &InitialToken = *FormatTok; 1547 nextToken(); 1548 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon)) 1549 nextToken(); 1550 if (FormatTok->Tok.is(tok::l_brace)) { 1551 if (ShouldBreakBeforeBrace(Style, InitialToken)) 1552 addUnwrappedLine(); 1553 1554 bool AddLevel = Style.NamespaceIndentation == FormatStyle::NI_All || 1555 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 1556 DeclarationScopeStack.size() > 1); 1557 parseBlock(/*MustBeDeclaration=*/true, AddLevel); 1558 // Munch the semicolon after a namespace. This is more common than one would 1559 // think. Puttin the semicolon into its own line is very ugly. 1560 if (FormatTok->Tok.is(tok::semi)) 1561 nextToken(); 1562 addUnwrappedLine(); 1563 } 1564 // FIXME: Add error handling. 1565 } 1566 1567 void UnwrappedLineParser::parseNew() { 1568 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 1569 nextToken(); 1570 if (Style.Language != FormatStyle::LK_Java) 1571 return; 1572 1573 // In Java, we can parse everything up to the parens, which aren't optional. 1574 do { 1575 // There should not be a ;, { or } before the new's open paren. 1576 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 1577 return; 1578 1579 // Consume the parens. 1580 if (FormatTok->is(tok::l_paren)) { 1581 parseParens(); 1582 1583 // If there is a class body of an anonymous class, consume that as child. 1584 if (FormatTok->is(tok::l_brace)) 1585 parseChildBlock(); 1586 return; 1587 } 1588 nextToken(); 1589 } while (!eof()); 1590 } 1591 1592 void UnwrappedLineParser::parseForOrWhileLoop() { 1593 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 1594 "'for', 'while' or foreach macro expected"); 1595 nextToken(); 1596 if (FormatTok->Tok.is(tok::l_paren)) 1597 parseParens(); 1598 if (FormatTok->Tok.is(tok::l_brace)) { 1599 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1600 parseBlock(/*MustBeDeclaration=*/false); 1601 addUnwrappedLine(); 1602 } else { 1603 addUnwrappedLine(); 1604 ++Line->Level; 1605 parseStructuralElement(); 1606 --Line->Level; 1607 } 1608 } 1609 1610 void UnwrappedLineParser::parseDoWhile() { 1611 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 1612 nextToken(); 1613 if (FormatTok->Tok.is(tok::l_brace)) { 1614 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1615 parseBlock(/*MustBeDeclaration=*/false); 1616 if (Style.BraceWrapping.IndentBraces) 1617 addUnwrappedLine(); 1618 } else { 1619 addUnwrappedLine(); 1620 ++Line->Level; 1621 parseStructuralElement(); 1622 --Line->Level; 1623 } 1624 1625 // FIXME: Add error handling. 1626 if (!FormatTok->Tok.is(tok::kw_while)) { 1627 addUnwrappedLine(); 1628 return; 1629 } 1630 1631 nextToken(); 1632 parseStructuralElement(); 1633 } 1634 1635 void UnwrappedLineParser::parseLabel() { 1636 nextToken(); 1637 unsigned OldLineLevel = Line->Level; 1638 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 1639 --Line->Level; 1640 if (CommentsBeforeNextToken.empty() && FormatTok->Tok.is(tok::l_brace)) { 1641 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1642 parseBlock(/*MustBeDeclaration=*/false); 1643 if (FormatTok->Tok.is(tok::kw_break)) { 1644 if (Style.BraceWrapping.AfterControlStatement) 1645 addUnwrappedLine(); 1646 parseStructuralElement(); 1647 } 1648 addUnwrappedLine(); 1649 } else { 1650 if (FormatTok->is(tok::semi)) 1651 nextToken(); 1652 addUnwrappedLine(); 1653 } 1654 Line->Level = OldLineLevel; 1655 if (FormatTok->isNot(tok::l_brace)) { 1656 parseStructuralElement(); 1657 addUnwrappedLine(); 1658 } 1659 } 1660 1661 void UnwrappedLineParser::parseCaseLabel() { 1662 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 1663 // FIXME: fix handling of complex expressions here. 1664 do { 1665 nextToken(); 1666 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 1667 parseLabel(); 1668 } 1669 1670 void UnwrappedLineParser::parseSwitch() { 1671 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 1672 nextToken(); 1673 if (FormatTok->Tok.is(tok::l_paren)) 1674 parseParens(); 1675 if (FormatTok->Tok.is(tok::l_brace)) { 1676 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1677 parseBlock(/*MustBeDeclaration=*/false); 1678 addUnwrappedLine(); 1679 } else { 1680 addUnwrappedLine(); 1681 ++Line->Level; 1682 parseStructuralElement(); 1683 --Line->Level; 1684 } 1685 } 1686 1687 void UnwrappedLineParser::parseAccessSpecifier() { 1688 nextToken(); 1689 // Understand Qt's slots. 1690 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 1691 nextToken(); 1692 // Otherwise, we don't know what it is, and we'd better keep the next token. 1693 if (FormatTok->Tok.is(tok::colon)) 1694 nextToken(); 1695 addUnwrappedLine(); 1696 } 1697 1698 bool UnwrappedLineParser::parseEnum() { 1699 // Won't be 'enum' for NS_ENUMs. 1700 if (FormatTok->Tok.is(tok::kw_enum)) 1701 nextToken(); 1702 1703 // In TypeScript, "enum" can also be used as property name, e.g. in interface 1704 // declarations. An "enum" keyword followed by a colon would be a syntax 1705 // error and thus assume it is just an identifier. 1706 if (Style.Language == FormatStyle::LK_JavaScript && 1707 FormatTok->isOneOf(tok::colon, tok::question)) 1708 return false; 1709 1710 // Eat up enum class ... 1711 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct)) 1712 nextToken(); 1713 1714 while (FormatTok->Tok.getIdentifierInfo() || 1715 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 1716 tok::greater, tok::comma, tok::question)) { 1717 nextToken(); 1718 // We can have macros or attributes in between 'enum' and the enum name. 1719 if (FormatTok->is(tok::l_paren)) 1720 parseParens(); 1721 if (FormatTok->is(tok::identifier)) { 1722 nextToken(); 1723 // If there are two identifiers in a row, this is likely an elaborate 1724 // return type. In Java, this can be "implements", etc. 1725 if (Style.Language == FormatStyle::LK_Cpp && 1726 FormatTok->is(tok::identifier)) 1727 return false; 1728 } 1729 } 1730 1731 // Just a declaration or something is wrong. 1732 if (FormatTok->isNot(tok::l_brace)) 1733 return true; 1734 FormatTok->BlockKind = BK_Block; 1735 1736 if (Style.Language == FormatStyle::LK_Java) { 1737 // Java enums are different. 1738 parseJavaEnumBody(); 1739 return true; 1740 } 1741 if (Style.Language == FormatStyle::LK_Proto) { 1742 parseBlock(/*MustBeDeclaration=*/true); 1743 return true; 1744 } 1745 1746 // Parse enum body. 1747 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true); 1748 if (HasError) { 1749 if (FormatTok->is(tok::semi)) 1750 nextToken(); 1751 addUnwrappedLine(); 1752 } 1753 return true; 1754 1755 // There is no addUnwrappedLine() here so that we fall through to parsing a 1756 // structural element afterwards. Thus, in "enum A {} n, m;", 1757 // "} n, m;" will end up in one unwrapped line. 1758 } 1759 1760 void UnwrappedLineParser::parseJavaEnumBody() { 1761 // Determine whether the enum is simple, i.e. does not have a semicolon or 1762 // constants with class bodies. Simple enums can be formatted like braced 1763 // lists, contracted to a single line, etc. 1764 unsigned StoredPosition = Tokens->getPosition(); 1765 bool IsSimple = true; 1766 FormatToken *Tok = Tokens->getNextToken(); 1767 while (Tok) { 1768 if (Tok->is(tok::r_brace)) 1769 break; 1770 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 1771 IsSimple = false; 1772 break; 1773 } 1774 // FIXME: This will also mark enums with braces in the arguments to enum 1775 // constants as "not simple". This is probably fine in practice, though. 1776 Tok = Tokens->getNextToken(); 1777 } 1778 FormatTok = Tokens->setPosition(StoredPosition); 1779 1780 if (IsSimple) { 1781 parseBracedList(); 1782 addUnwrappedLine(); 1783 return; 1784 } 1785 1786 // Parse the body of a more complex enum. 1787 // First add a line for everything up to the "{". 1788 nextToken(); 1789 addUnwrappedLine(); 1790 ++Line->Level; 1791 1792 // Parse the enum constants. 1793 while (FormatTok) { 1794 if (FormatTok->is(tok::l_brace)) { 1795 // Parse the constant's class body. 1796 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 1797 /*MunchSemi=*/false); 1798 } else if (FormatTok->is(tok::l_paren)) { 1799 parseParens(); 1800 } else if (FormatTok->is(tok::comma)) { 1801 nextToken(); 1802 addUnwrappedLine(); 1803 } else if (FormatTok->is(tok::semi)) { 1804 nextToken(); 1805 addUnwrappedLine(); 1806 break; 1807 } else if (FormatTok->is(tok::r_brace)) { 1808 addUnwrappedLine(); 1809 break; 1810 } else { 1811 nextToken(); 1812 } 1813 } 1814 1815 // Parse the class body after the enum's ";" if any. 1816 parseLevel(/*HasOpeningBrace=*/true); 1817 nextToken(); 1818 --Line->Level; 1819 addUnwrappedLine(); 1820 } 1821 1822 void UnwrappedLineParser::parseRecord() { 1823 const FormatToken &InitialToken = *FormatTok; 1824 nextToken(); 1825 1826 // The actual identifier can be a nested name specifier, and in macros 1827 // it is often token-pasted. 1828 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 1829 tok::kw___attribute, tok::kw___declspec, 1830 tok::kw_alignas) || 1831 ((Style.Language == FormatStyle::LK_Java || 1832 Style.Language == FormatStyle::LK_JavaScript) && 1833 FormatTok->isOneOf(tok::period, tok::comma))) { 1834 bool IsNonMacroIdentifier = 1835 FormatTok->is(tok::identifier) && 1836 FormatTok->TokenText != FormatTok->TokenText.upper(); 1837 nextToken(); 1838 // We can have macros or attributes in between 'class' and the class name. 1839 if (!IsNonMacroIdentifier && FormatTok->Tok.is(tok::l_paren)) 1840 parseParens(); 1841 } 1842 1843 // Note that parsing away template declarations here leads to incorrectly 1844 // accepting function declarations as record declarations. 1845 // In general, we cannot solve this problem. Consider: 1846 // class A<int> B() {} 1847 // which can be a function definition or a class definition when B() is a 1848 // macro. If we find enough real-world cases where this is a problem, we 1849 // can parse for the 'template' keyword in the beginning of the statement, 1850 // and thus rule out the record production in case there is no template 1851 // (this would still leave us with an ambiguity between template function 1852 // and class declarations). 1853 if (FormatTok->isOneOf(tok::colon, tok::less)) { 1854 while (!eof()) { 1855 if (FormatTok->is(tok::l_brace)) { 1856 calculateBraceTypes(/*ExpectClassBody=*/true); 1857 if (!tryToParseBracedList()) 1858 break; 1859 } 1860 if (FormatTok->Tok.is(tok::semi)) 1861 return; 1862 nextToken(); 1863 } 1864 } 1865 if (FormatTok->Tok.is(tok::l_brace)) { 1866 if (ShouldBreakBeforeBrace(Style, InitialToken)) 1867 addUnwrappedLine(); 1868 1869 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 1870 /*MunchSemi=*/false); 1871 } 1872 // There is no addUnwrappedLine() here so that we fall through to parsing a 1873 // structural element afterwards. Thus, in "class A {} n, m;", 1874 // "} n, m;" will end up in one unwrapped line. 1875 } 1876 1877 void UnwrappedLineParser::parseObjCProtocolList() { 1878 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 1879 do 1880 nextToken(); 1881 while (!eof() && FormatTok->Tok.isNot(tok::greater)); 1882 nextToken(); // Skip '>'. 1883 } 1884 1885 void UnwrappedLineParser::parseObjCUntilAtEnd() { 1886 do { 1887 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 1888 nextToken(); 1889 addUnwrappedLine(); 1890 break; 1891 } 1892 if (FormatTok->is(tok::l_brace)) { 1893 parseBlock(/*MustBeDeclaration=*/false); 1894 // In ObjC interfaces, nothing should be following the "}". 1895 addUnwrappedLine(); 1896 } else if (FormatTok->is(tok::r_brace)) { 1897 // Ignore stray "}". parseStructuralElement doesn't consume them. 1898 nextToken(); 1899 addUnwrappedLine(); 1900 } else { 1901 parseStructuralElement(); 1902 } 1903 } while (!eof()); 1904 } 1905 1906 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 1907 nextToken(); 1908 nextToken(); // interface name 1909 1910 // @interface can be followed by either a base class, or a category. 1911 if (FormatTok->Tok.is(tok::colon)) { 1912 nextToken(); 1913 nextToken(); // base class name 1914 } else if (FormatTok->Tok.is(tok::l_paren)) 1915 // Skip category, if present. 1916 parseParens(); 1917 1918 if (FormatTok->Tok.is(tok::less)) 1919 parseObjCProtocolList(); 1920 1921 if (FormatTok->Tok.is(tok::l_brace)) { 1922 if (Style.BraceWrapping.AfterObjCDeclaration) 1923 addUnwrappedLine(); 1924 parseBlock(/*MustBeDeclaration=*/true); 1925 } 1926 1927 // With instance variables, this puts '}' on its own line. Without instance 1928 // variables, this ends the @interface line. 1929 addUnwrappedLine(); 1930 1931 parseObjCUntilAtEnd(); 1932 } 1933 1934 void UnwrappedLineParser::parseObjCProtocol() { 1935 nextToken(); 1936 nextToken(); // protocol name 1937 1938 if (FormatTok->Tok.is(tok::less)) 1939 parseObjCProtocolList(); 1940 1941 // Check for protocol declaration. 1942 if (FormatTok->Tok.is(tok::semi)) { 1943 nextToken(); 1944 return addUnwrappedLine(); 1945 } 1946 1947 addUnwrappedLine(); 1948 parseObjCUntilAtEnd(); 1949 } 1950 1951 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 1952 bool IsImport = FormatTok->is(Keywords.kw_import); 1953 assert(IsImport || FormatTok->is(tok::kw_export)); 1954 nextToken(); 1955 1956 // Consume the "default" in "export default class/function". 1957 if (FormatTok->is(tok::kw_default)) 1958 nextToken(); 1959 1960 // Consume "async function", "function" and "default function", so that these 1961 // get parsed as free-standing JS functions, i.e. do not require a trailing 1962 // semicolon. 1963 if (FormatTok->is(Keywords.kw_async)) 1964 nextToken(); 1965 if (FormatTok->is(Keywords.kw_function)) { 1966 nextToken(); 1967 return; 1968 } 1969 1970 // For imports, `export *`, `export {...}`, consume the rest of the line up 1971 // to the terminating `;`. For everything else, just return and continue 1972 // parsing the structural element, i.e. the declaration or expression for 1973 // `export default`. 1974 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 1975 !FormatTok->isStringLiteral()) 1976 return; 1977 1978 while (!eof()) { 1979 if (FormatTok->is(tok::semi)) 1980 return; 1981 if (Line->Tokens.size() == 0) { 1982 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 1983 // import statement should terminate. 1984 return; 1985 } 1986 if (FormatTok->is(tok::l_brace)) { 1987 FormatTok->BlockKind = BK_Block; 1988 parseBracedList(); 1989 } else { 1990 nextToken(); 1991 } 1992 } 1993 } 1994 1995 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 1996 StringRef Prefix = "") { 1997 llvm::dbgs() << Prefix << "Line(" << Line.Level << ")" 1998 << (Line.InPPDirective ? " MACRO" : "") << ": "; 1999 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2000 E = Line.Tokens.end(); 2001 I != E; ++I) { 2002 llvm::dbgs() << I->Tok->Tok.getName() << "[" 2003 << "T=" << I->Tok->Type 2004 << ", OC=" << I->Tok->OriginalColumn << "] "; 2005 } 2006 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2007 E = Line.Tokens.end(); 2008 I != E; ++I) { 2009 const UnwrappedLineNode &Node = *I; 2010 for (SmallVectorImpl<UnwrappedLine>::const_iterator 2011 I = Node.Children.begin(), 2012 E = Node.Children.end(); 2013 I != E; ++I) { 2014 printDebugInfo(*I, "\nChild: "); 2015 } 2016 } 2017 llvm::dbgs() << "\n"; 2018 } 2019 2020 void UnwrappedLineParser::addUnwrappedLine() { 2021 if (Line->Tokens.empty()) 2022 return; 2023 DEBUG({ 2024 if (CurrentLines == &Lines) 2025 printDebugInfo(*Line); 2026 }); 2027 CurrentLines->push_back(std::move(*Line)); 2028 Line->Tokens.clear(); 2029 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 2030 CurrentLines->append( 2031 std::make_move_iterator(PreprocessorDirectives.begin()), 2032 std::make_move_iterator(PreprocessorDirectives.end())); 2033 PreprocessorDirectives.clear(); 2034 } 2035 } 2036 2037 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 2038 2039 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 2040 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 2041 FormatTok.NewlinesBefore > 0; 2042 } 2043 2044 static bool isLineComment(const FormatToken &FormatTok) { 2045 return FormatTok.is(tok::comment) && 2046 FormatTok.TokenText.startswith("//"); 2047 } 2048 2049 // Checks if \p FormatTok is a line comment that continues the line comment 2050 // section on \p Line. 2051 static bool continuesLineComment(const FormatToken &FormatTok, 2052 const UnwrappedLine &Line, 2053 llvm::Regex &CommentPragmasRegex) { 2054 if (Line.Tokens.empty()) 2055 return false; 2056 2057 StringRef IndentContent = FormatTok.TokenText; 2058 if (FormatTok.TokenText.startswith("//") || 2059 FormatTok.TokenText.startswith("/*")) 2060 IndentContent = FormatTok.TokenText.substr(2); 2061 if (CommentPragmasRegex.match(IndentContent)) 2062 return false; 2063 2064 // If Line starts with a line comment, then FormatTok continues the comment 2065 // section if its original column is greater or equal to the original start 2066 // column of the line. 2067 // 2068 // Define the min column token of a line as follows: if a line ends in '{' or 2069 // contains a '{' followed by a line comment, then the min column token is 2070 // that '{'. Otherwise, the min column token of the line is the first token of 2071 // the line. 2072 // 2073 // If Line starts with a token other than a line comment, then FormatTok 2074 // continues the comment section if its original column is greater than the 2075 // original start column of the min column token of the line. 2076 // 2077 // For example, the second line comment continues the first in these cases: 2078 // 2079 // // first line 2080 // // second line 2081 // 2082 // and: 2083 // 2084 // // first line 2085 // // second line 2086 // 2087 // and: 2088 // 2089 // int i; // first line 2090 // // second line 2091 // 2092 // and: 2093 // 2094 // do { // first line 2095 // // second line 2096 // int i; 2097 // } while (true); 2098 // 2099 // and: 2100 // 2101 // enum { 2102 // a, // first line 2103 // // second line 2104 // b 2105 // }; 2106 // 2107 // The second line comment doesn't continue the first in these cases: 2108 // 2109 // // first line 2110 // // second line 2111 // 2112 // and: 2113 // 2114 // int i; // first line 2115 // // second line 2116 // 2117 // and: 2118 // 2119 // do { // first line 2120 // // second line 2121 // int i; 2122 // } while (true); 2123 // 2124 // and: 2125 // 2126 // enum { 2127 // a, // first line 2128 // // second line 2129 // }; 2130 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 2131 2132 // Scan for '{//'. If found, use the column of '{' as a min column for line 2133 // comment section continuation. 2134 const FormatToken *PreviousToken = nullptr; 2135 for (const UnwrappedLineNode Node : Line.Tokens) { 2136 if (PreviousToken && PreviousToken->is(tok::l_brace) && 2137 isLineComment(*Node.Tok)) { 2138 MinColumnToken = PreviousToken; 2139 break; 2140 } 2141 PreviousToken = Node.Tok; 2142 2143 // Grab the last newline preceding a token in this unwrapped line. 2144 if (Node.Tok->NewlinesBefore > 0) { 2145 MinColumnToken = Node.Tok; 2146 } 2147 } 2148 if (PreviousToken && PreviousToken->is(tok::l_brace)) { 2149 MinColumnToken = PreviousToken; 2150 } 2151 2152 unsigned MinContinueColumn = 2153 MinColumnToken->OriginalColumn + 2154 (isLineComment(*MinColumnToken) ? 0 : 1); 2155 return isLineComment(FormatTok) && FormatTok.NewlinesBefore == 1 && 2156 isLineComment(*(Line.Tokens.back().Tok)) && 2157 FormatTok.OriginalColumn >= MinContinueColumn; 2158 } 2159 2160 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 2161 bool JustComments = Line->Tokens.empty(); 2162 for (SmallVectorImpl<FormatToken *>::const_iterator 2163 I = CommentsBeforeNextToken.begin(), 2164 E = CommentsBeforeNextToken.end(); 2165 I != E; ++I) { 2166 // Line comments that belong to the same line comment section are put on the 2167 // same line since later we might want to reflow content between them. 2168 // Additional fine-grained breaking of line comment sections is controlled 2169 // by the class BreakableLineCommentSection in case it is desirable to keep 2170 // several line comment sections in the same unwrapped line. 2171 // 2172 // FIXME: Consider putting separate line comment sections as children to the 2173 // unwrapped line instead. 2174 (*I)->ContinuesLineCommentSection = 2175 continuesLineComment(**I, *Line, CommentPragmasRegex); 2176 if (isOnNewLine(**I) && JustComments && !(*I)->ContinuesLineCommentSection) 2177 addUnwrappedLine(); 2178 pushToken(*I); 2179 } 2180 if (NewlineBeforeNext && JustComments) 2181 addUnwrappedLine(); 2182 CommentsBeforeNextToken.clear(); 2183 } 2184 2185 void UnwrappedLineParser::nextToken() { 2186 if (eof()) 2187 return; 2188 flushComments(isOnNewLine(*FormatTok)); 2189 pushToken(FormatTok); 2190 if (Style.Language != FormatStyle::LK_JavaScript) 2191 readToken(); 2192 else 2193 readTokenWithJavaScriptASI(); 2194 } 2195 2196 const FormatToken *UnwrappedLineParser::getPreviousToken() { 2197 // FIXME: This is a dirty way to access the previous token. Find a better 2198 // solution. 2199 if (!Line || Line->Tokens.empty()) 2200 return nullptr; 2201 return Line->Tokens.back().Tok; 2202 } 2203 2204 void UnwrappedLineParser::readToken() { 2205 bool CommentsInCurrentLine = true; 2206 do { 2207 FormatTok = Tokens->getNextToken(); 2208 assert(FormatTok); 2209 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 2210 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 2211 // If there is an unfinished unwrapped line, we flush the preprocessor 2212 // directives only after that unwrapped line was finished later. 2213 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 2214 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 2215 // Comments stored before the preprocessor directive need to be output 2216 // before the preprocessor directive, at the same level as the 2217 // preprocessor directive, as we consider them to apply to the directive. 2218 flushComments(isOnNewLine(*FormatTok)); 2219 parsePPDirective(); 2220 } 2221 while (FormatTok->Type == TT_ConflictStart || 2222 FormatTok->Type == TT_ConflictEnd || 2223 FormatTok->Type == TT_ConflictAlternative) { 2224 if (FormatTok->Type == TT_ConflictStart) { 2225 conditionalCompilationStart(/*Unreachable=*/false); 2226 } else if (FormatTok->Type == TT_ConflictAlternative) { 2227 conditionalCompilationAlternative(); 2228 } else if (FormatTok->Type == TT_ConflictEnd) { 2229 conditionalCompilationEnd(); 2230 } 2231 FormatTok = Tokens->getNextToken(); 2232 FormatTok->MustBreakBefore = true; 2233 } 2234 2235 if (!PPStack.empty() && (PPStack.back() == PP_Unreachable) && 2236 !Line->InPPDirective) { 2237 continue; 2238 } 2239 2240 if (!FormatTok->Tok.is(tok::comment)) 2241 return; 2242 FormatTok->ContinuesLineCommentSection = 2243 continuesLineComment(*FormatTok, *Line, CommentPragmasRegex); 2244 if (!FormatTok->ContinuesLineCommentSection && 2245 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) { 2246 CommentsInCurrentLine = false; 2247 } 2248 if (CommentsInCurrentLine) { 2249 pushToken(FormatTok); 2250 } else { 2251 CommentsBeforeNextToken.push_back(FormatTok); 2252 } 2253 } while (!eof()); 2254 } 2255 2256 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 2257 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 2258 if (MustBreakBeforeNextToken) { 2259 Line->Tokens.back().Tok->MustBreakBefore = true; 2260 MustBreakBeforeNextToken = false; 2261 } 2262 } 2263 2264 } // end namespace format 2265 } // end namespace clang 2266