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), Tokens(nullptr), 206 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1) {} 207 208 void UnwrappedLineParser::reset() { 209 PPBranchLevel = -1; 210 Line.reset(new UnwrappedLine); 211 CommentsBeforeNextToken.clear(); 212 FormatTok = nullptr; 213 MustBreakBeforeNextToken = false; 214 PreprocessorDirectives.clear(); 215 CurrentLines = &Lines; 216 DeclarationScopeStack.clear(); 217 PPStack.clear(); 218 } 219 220 void UnwrappedLineParser::parse() { 221 IndexedTokenSource TokenSource(AllTokens); 222 do { 223 DEBUG(llvm::dbgs() << "----\n"); 224 reset(); 225 Tokens = &TokenSource; 226 TokenSource.reset(); 227 228 readToken(); 229 parseFile(); 230 // Create line with eof token. 231 pushToken(FormatTok); 232 addUnwrappedLine(); 233 234 for (SmallVectorImpl<UnwrappedLine>::iterator I = Lines.begin(), 235 E = Lines.end(); 236 I != E; ++I) { 237 Callback.consumeUnwrappedLine(*I); 238 } 239 Callback.finishRun(); 240 Lines.clear(); 241 while (!PPLevelBranchIndex.empty() && 242 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) { 243 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1); 244 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1); 245 } 246 if (!PPLevelBranchIndex.empty()) { 247 ++PPLevelBranchIndex.back(); 248 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size()); 249 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back()); 250 } 251 } while (!PPLevelBranchIndex.empty()); 252 } 253 254 void UnwrappedLineParser::parseFile() { 255 // The top-level context in a file always has declarations, except for pre- 256 // processor directives and JavaScript files. 257 bool MustBeDeclaration = 258 !Line->InPPDirective && Style.Language != FormatStyle::LK_JavaScript; 259 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 260 MustBeDeclaration); 261 parseLevel(/*HasOpeningBrace=*/false); 262 // Make sure to format the remaining tokens. 263 flushComments(true); 264 addUnwrappedLine(); 265 } 266 267 void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) { 268 bool SwitchLabelEncountered = false; 269 do { 270 tok::TokenKind kind = FormatTok->Tok.getKind(); 271 if (FormatTok->Type == TT_MacroBlockBegin) { 272 kind = tok::l_brace; 273 } else if (FormatTok->Type == TT_MacroBlockEnd) { 274 kind = tok::r_brace; 275 } 276 277 switch (kind) { 278 case tok::comment: 279 nextToken(); 280 addUnwrappedLine(); 281 break; 282 case tok::l_brace: 283 // FIXME: Add parameter whether this can happen - if this happens, we must 284 // be in a non-declaration context. 285 if (!FormatTok->is(TT_MacroBlockBegin) && tryToParseBracedList()) 286 continue; 287 parseBlock(/*MustBeDeclaration=*/false); 288 addUnwrappedLine(); 289 break; 290 case tok::r_brace: 291 if (HasOpeningBrace) 292 return; 293 nextToken(); 294 addUnwrappedLine(); 295 break; 296 case tok::kw_default: 297 case tok::kw_case: 298 if (!SwitchLabelEncountered && 299 (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1))) 300 ++Line->Level; 301 SwitchLabelEncountered = true; 302 parseStructuralElement(); 303 break; 304 default: 305 parseStructuralElement(); 306 break; 307 } 308 } while (!eof()); 309 } 310 311 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) { 312 // We'll parse forward through the tokens until we hit 313 // a closing brace or eof - note that getNextToken() will 314 // parse macros, so this will magically work inside macro 315 // definitions, too. 316 unsigned StoredPosition = Tokens->getPosition(); 317 FormatToken *Tok = FormatTok; 318 // Keep a stack of positions of lbrace tokens. We will 319 // update information about whether an lbrace starts a 320 // braced init list or a different block during the loop. 321 SmallVector<FormatToken *, 8> LBraceStack; 322 assert(Tok->Tok.is(tok::l_brace)); 323 do { 324 // Get next none-comment token. 325 FormatToken *NextTok; 326 unsigned ReadTokens = 0; 327 do { 328 NextTok = Tokens->getNextToken(); 329 ++ReadTokens; 330 } while (NextTok->is(tok::comment)); 331 332 switch (Tok->Tok.getKind()) { 333 case tok::l_brace: 334 Tok->BlockKind = BK_Unknown; 335 LBraceStack.push_back(Tok); 336 break; 337 case tok::r_brace: 338 if (!LBraceStack.empty()) { 339 if (LBraceStack.back()->BlockKind == BK_Unknown) { 340 bool ProbablyBracedList = false; 341 if (Style.Language == FormatStyle::LK_Proto) { 342 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square); 343 } else { 344 // Using OriginalColumn to distinguish between ObjC methods and 345 // binary operators is a bit hacky. 346 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 347 NextTok->OriginalColumn == 0; 348 349 // If there is a comma, semicolon or right paren after the closing 350 // brace, we assume this is a braced initializer list. Note that 351 // regardless how we mark inner braces here, we will overwrite the 352 // BlockKind later if we parse a braced list (where all blocks 353 // inside are by default braced lists), or when we explicitly detect 354 // blocks (for example while parsing lambdas). 355 // 356 // We exclude + and - as they can be ObjC visibility modifiers. 357 ProbablyBracedList = 358 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 359 tok::r_paren, tok::r_square, tok::l_brace, 360 tok::l_paren, tok::ellipsis) || 361 (NextTok->is(tok::semi) && 362 (!ExpectClassBody || LBraceStack.size() != 1)) || 363 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 364 } 365 if (ProbablyBracedList) { 366 Tok->BlockKind = BK_BracedInit; 367 LBraceStack.back()->BlockKind = BK_BracedInit; 368 } else { 369 Tok->BlockKind = BK_Block; 370 LBraceStack.back()->BlockKind = BK_Block; 371 } 372 } 373 LBraceStack.pop_back(); 374 } 375 break; 376 case tok::at: 377 case tok::semi: 378 case tok::kw_if: 379 case tok::kw_while: 380 case tok::kw_for: 381 case tok::kw_switch: 382 case tok::kw_try: 383 case tok::kw___try: 384 if (!LBraceStack.empty()) 385 LBraceStack.back()->BlockKind = BK_Block; 386 break; 387 default: 388 break; 389 } 390 Tok = NextTok; 391 } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty()); 392 // Assume other blocks for all unclosed opening braces. 393 for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) { 394 if (LBraceStack[i]->BlockKind == BK_Unknown) 395 LBraceStack[i]->BlockKind = BK_Block; 396 } 397 398 FormatTok = Tokens->setPosition(StoredPosition); 399 } 400 401 void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, bool AddLevel, 402 bool MunchSemi) { 403 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) && 404 "'{' or macro block token expected"); 405 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 406 407 unsigned InitialLevel = Line->Level; 408 nextToken(); 409 410 if (MacroBlock && FormatTok->is(tok::l_paren)) 411 parseParens(); 412 413 addUnwrappedLine(); 414 415 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 416 MustBeDeclaration); 417 if (AddLevel) 418 ++Line->Level; 419 parseLevel(/*HasOpeningBrace=*/true); 420 421 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd) 422 : !FormatTok->is(tok::r_brace)) { 423 Line->Level = InitialLevel; 424 return; 425 } 426 427 nextToken(); // Munch the closing brace. 428 429 if (MacroBlock && FormatTok->is(tok::l_paren)) 430 parseParens(); 431 432 if (MunchSemi && FormatTok->Tok.is(tok::semi)) 433 nextToken(); 434 Line->Level = InitialLevel; 435 } 436 437 static bool isGoogScope(const UnwrappedLine &Line) { 438 // FIXME: Closure-library specific stuff should not be hard-coded but be 439 // configurable. 440 if (Line.Tokens.size() < 4) 441 return false; 442 auto I = Line.Tokens.begin(); 443 if (I->Tok->TokenText != "goog") 444 return false; 445 ++I; 446 if (I->Tok->isNot(tok::period)) 447 return false; 448 ++I; 449 if (I->Tok->TokenText != "scope") 450 return false; 451 ++I; 452 return I->Tok->is(tok::l_paren); 453 } 454 455 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 456 const FormatToken &InitialToken) { 457 if (InitialToken.is(tok::kw_namespace)) 458 return Style.BraceWrapping.AfterNamespace; 459 if (InitialToken.is(tok::kw_class)) 460 return Style.BraceWrapping.AfterClass; 461 if (InitialToken.is(tok::kw_union)) 462 return Style.BraceWrapping.AfterUnion; 463 if (InitialToken.is(tok::kw_struct)) 464 return Style.BraceWrapping.AfterStruct; 465 return false; 466 } 467 468 void UnwrappedLineParser::parseChildBlock() { 469 FormatTok->BlockKind = BK_Block; 470 nextToken(); 471 { 472 bool GoogScope = 473 Style.Language == FormatStyle::LK_JavaScript && isGoogScope(*Line); 474 ScopedLineState LineState(*this); 475 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 476 /*MustBeDeclaration=*/false); 477 Line->Level += GoogScope ? 0 : 1; 478 parseLevel(/*HasOpeningBrace=*/true); 479 flushComments(isOnNewLine(*FormatTok)); 480 Line->Level -= GoogScope ? 0 : 1; 481 } 482 nextToken(); 483 } 484 485 void UnwrappedLineParser::parsePPDirective() { 486 assert(FormatTok->Tok.is(tok::hash) && "'#' expected"); 487 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 488 nextToken(); 489 490 if (!FormatTok->Tok.getIdentifierInfo()) { 491 parsePPUnknown(); 492 return; 493 } 494 495 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 496 case tok::pp_define: 497 parsePPDefine(); 498 return; 499 case tok::pp_if: 500 parsePPIf(/*IfDef=*/false); 501 break; 502 case tok::pp_ifdef: 503 case tok::pp_ifndef: 504 parsePPIf(/*IfDef=*/true); 505 break; 506 case tok::pp_else: 507 parsePPElse(); 508 break; 509 case tok::pp_elif: 510 parsePPElIf(); 511 break; 512 case tok::pp_endif: 513 parsePPEndIf(); 514 break; 515 default: 516 parsePPUnknown(); 517 break; 518 } 519 } 520 521 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 522 if (Unreachable || (!PPStack.empty() && PPStack.back() == PP_Unreachable)) 523 PPStack.push_back(PP_Unreachable); 524 else 525 PPStack.push_back(PP_Conditional); 526 } 527 528 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 529 ++PPBranchLevel; 530 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 531 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 532 PPLevelBranchIndex.push_back(0); 533 PPLevelBranchCount.push_back(0); 534 } 535 PPChainBranchIndex.push(0); 536 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 537 conditionalCompilationCondition(Unreachable || Skip); 538 } 539 540 void UnwrappedLineParser::conditionalCompilationAlternative() { 541 if (!PPStack.empty()) 542 PPStack.pop_back(); 543 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 544 if (!PPChainBranchIndex.empty()) 545 ++PPChainBranchIndex.top(); 546 conditionalCompilationCondition( 547 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 548 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 549 } 550 551 void UnwrappedLineParser::conditionalCompilationEnd() { 552 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 553 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 554 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) { 555 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 556 } 557 } 558 // Guard against #endif's without #if. 559 if (PPBranchLevel > 0) 560 --PPBranchLevel; 561 if (!PPChainBranchIndex.empty()) 562 PPChainBranchIndex.pop(); 563 if (!PPStack.empty()) 564 PPStack.pop_back(); 565 } 566 567 void UnwrappedLineParser::parsePPIf(bool IfDef) { 568 nextToken(); 569 bool IsLiteralFalse = (FormatTok->Tok.isLiteral() && 570 FormatTok->Tok.getLiteralData() != nullptr && 571 StringRef(FormatTok->Tok.getLiteralData(), 572 FormatTok->Tok.getLength()) == "0") || 573 FormatTok->Tok.is(tok::kw_false); 574 conditionalCompilationStart(!IfDef && IsLiteralFalse); 575 parsePPUnknown(); 576 } 577 578 void UnwrappedLineParser::parsePPElse() { 579 conditionalCompilationAlternative(); 580 parsePPUnknown(); 581 } 582 583 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 584 585 void UnwrappedLineParser::parsePPEndIf() { 586 conditionalCompilationEnd(); 587 parsePPUnknown(); 588 } 589 590 void UnwrappedLineParser::parsePPDefine() { 591 nextToken(); 592 593 if (FormatTok->Tok.getKind() != tok::identifier) { 594 parsePPUnknown(); 595 return; 596 } 597 nextToken(); 598 if (FormatTok->Tok.getKind() == tok::l_paren && 599 FormatTok->WhitespaceRange.getBegin() == 600 FormatTok->WhitespaceRange.getEnd()) { 601 parseParens(); 602 } 603 addUnwrappedLine(); 604 Line->Level = 1; 605 606 // Errors during a preprocessor directive can only affect the layout of the 607 // preprocessor directive, and thus we ignore them. An alternative approach 608 // would be to use the same approach we use on the file level (no 609 // re-indentation if there was a structural error) within the macro 610 // definition. 611 parseFile(); 612 } 613 614 void UnwrappedLineParser::parsePPUnknown() { 615 do { 616 nextToken(); 617 } while (!eof()); 618 addUnwrappedLine(); 619 } 620 621 // Here we blacklist certain tokens that are not usually the first token in an 622 // unwrapped line. This is used in attempt to distinguish macro calls without 623 // trailing semicolons from other constructs split to several lines. 624 static bool tokenCanStartNewLine(const clang::Token &Tok) { 625 // Semicolon can be a null-statement, l_square can be a start of a macro or 626 // a C++11 attribute, but this doesn't seem to be common. 627 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 628 Tok.isNot(tok::l_square) && 629 // Tokens that can only be used as binary operators and a part of 630 // overloaded operator names. 631 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 632 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 633 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 634 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 635 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 636 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 637 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 638 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 639 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 640 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 641 Tok.isNot(tok::lesslessequal) && 642 // Colon is used in labels, base class lists, initializer lists, 643 // range-based for loops, ternary operator, but should never be the 644 // first token in an unwrapped line. 645 Tok.isNot(tok::colon) && 646 // 'noexcept' is a trailing annotation. 647 Tok.isNot(tok::kw_noexcept); 648 } 649 650 void UnwrappedLineParser::parseStructuralElement() { 651 assert(!FormatTok->Tok.is(tok::l_brace)); 652 switch (FormatTok->Tok.getKind()) { 653 case tok::at: 654 nextToken(); 655 if (FormatTok->Tok.is(tok::l_brace)) { 656 parseBracedList(); 657 break; 658 } 659 switch (FormatTok->Tok.getObjCKeywordID()) { 660 case tok::objc_public: 661 case tok::objc_protected: 662 case tok::objc_package: 663 case tok::objc_private: 664 return parseAccessSpecifier(); 665 case tok::objc_interface: 666 case tok::objc_implementation: 667 return parseObjCInterfaceOrImplementation(); 668 case tok::objc_protocol: 669 return parseObjCProtocol(); 670 case tok::objc_end: 671 return; // Handled by the caller. 672 case tok::objc_optional: 673 case tok::objc_required: 674 nextToken(); 675 addUnwrappedLine(); 676 return; 677 case tok::objc_autoreleasepool: 678 nextToken(); 679 if (FormatTok->Tok.is(tok::l_brace)) { 680 if (Style.BraceWrapping.AfterObjCDeclaration) 681 addUnwrappedLine(); 682 parseBlock(/*MustBeDeclaration=*/false); 683 } 684 addUnwrappedLine(); 685 return; 686 case tok::objc_try: 687 // This branch isn't strictly necessary (the kw_try case below would 688 // do this too after the tok::at is parsed above). But be explicit. 689 parseTryCatch(); 690 return; 691 default: 692 break; 693 } 694 break; 695 case tok::kw_asm: 696 nextToken(); 697 if (FormatTok->is(tok::l_brace)) { 698 FormatTok->Type = TT_InlineASMBrace; 699 nextToken(); 700 while (FormatTok && FormatTok->isNot(tok::eof)) { 701 if (FormatTok->is(tok::r_brace)) { 702 FormatTok->Type = TT_InlineASMBrace; 703 nextToken(); 704 addUnwrappedLine(); 705 break; 706 } 707 FormatTok->Finalized = true; 708 nextToken(); 709 } 710 } 711 break; 712 case tok::kw_namespace: 713 parseNamespace(); 714 return; 715 case tok::kw_inline: 716 nextToken(); 717 if (FormatTok->Tok.is(tok::kw_namespace)) { 718 parseNamespace(); 719 return; 720 } 721 break; 722 case tok::kw_public: 723 case tok::kw_protected: 724 case tok::kw_private: 725 if (Style.Language == FormatStyle::LK_Java || 726 Style.Language == FormatStyle::LK_JavaScript) 727 nextToken(); 728 else 729 parseAccessSpecifier(); 730 return; 731 case tok::kw_if: 732 parseIfThenElse(); 733 return; 734 case tok::kw_for: 735 case tok::kw_while: 736 parseForOrWhileLoop(); 737 return; 738 case tok::kw_do: 739 parseDoWhile(); 740 return; 741 case tok::kw_switch: 742 parseSwitch(); 743 return; 744 case tok::kw_default: 745 nextToken(); 746 parseLabel(); 747 return; 748 case tok::kw_case: 749 parseCaseLabel(); 750 return; 751 case tok::kw_try: 752 case tok::kw___try: 753 parseTryCatch(); 754 return; 755 case tok::kw_extern: 756 nextToken(); 757 if (FormatTok->Tok.is(tok::string_literal)) { 758 nextToken(); 759 if (FormatTok->Tok.is(tok::l_brace)) { 760 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/false); 761 addUnwrappedLine(); 762 return; 763 } 764 } 765 break; 766 case tok::kw_export: 767 if (Style.Language == FormatStyle::LK_JavaScript) { 768 parseJavaScriptEs6ImportExport(); 769 return; 770 } 771 break; 772 case tok::identifier: 773 if (FormatTok->is(TT_ForEachMacro)) { 774 parseForOrWhileLoop(); 775 return; 776 } 777 if (FormatTok->is(TT_MacroBlockBegin)) { 778 parseBlock(/*MustBeDeclaration=*/false, /*AddLevel=*/true, 779 /*MunchSemi=*/false); 780 return; 781 } 782 if (Style.Language == FormatStyle::LK_JavaScript && 783 FormatTok->is(Keywords.kw_import)) { 784 parseJavaScriptEs6ImportExport(); 785 return; 786 } 787 if (FormatTok->is(Keywords.kw_signals)) { 788 nextToken(); 789 if (FormatTok->is(tok::colon)) { 790 nextToken(); 791 addUnwrappedLine(); 792 } 793 return; 794 } 795 // In all other cases, parse the declaration. 796 break; 797 default: 798 break; 799 } 800 do { 801 switch (FormatTok->Tok.getKind()) { 802 case tok::at: 803 nextToken(); 804 if (FormatTok->Tok.is(tok::l_brace)) 805 parseBracedList(); 806 break; 807 case tok::kw_enum: 808 // parseEnum falls through and does not yet add an unwrapped line as an 809 // enum definition can start a structural element. 810 parseEnum(); 811 // This only applies for C++. 812 if (Style.Language != FormatStyle::LK_Cpp) { 813 addUnwrappedLine(); 814 return; 815 } 816 break; 817 case tok::kw_typedef: 818 nextToken(); 819 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 820 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS)) 821 parseEnum(); 822 break; 823 case tok::kw_struct: 824 case tok::kw_union: 825 case tok::kw_class: 826 // parseRecord falls through and does not yet add an unwrapped line as a 827 // record declaration or definition can start a structural element. 828 parseRecord(); 829 // This does not apply for Java and JavaScript. 830 if (Style.Language == FormatStyle::LK_Java || 831 Style.Language == FormatStyle::LK_JavaScript) { 832 addUnwrappedLine(); 833 return; 834 } 835 break; 836 case tok::period: 837 nextToken(); 838 // In Java, classes have an implicit static member "class". 839 if (Style.Language == FormatStyle::LK_Java && FormatTok && 840 FormatTok->is(tok::kw_class)) 841 nextToken(); 842 if (Style.Language == FormatStyle::LK_JavaScript && FormatTok && 843 FormatTok->Tok.getIdentifierInfo()) 844 // JavaScript only has pseudo keywords, all keywords are allowed to 845 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 846 nextToken(); 847 break; 848 case tok::semi: 849 nextToken(); 850 addUnwrappedLine(); 851 return; 852 case tok::r_brace: 853 addUnwrappedLine(); 854 return; 855 case tok::l_paren: 856 parseParens(); 857 break; 858 case tok::kw_operator: 859 nextToken(); 860 if (FormatTok->isBinaryOperator()) 861 nextToken(); 862 break; 863 case tok::caret: 864 nextToken(); 865 if (FormatTok->Tok.isAnyIdentifier() || 866 FormatTok->isSimpleTypeSpecifier()) 867 nextToken(); 868 if (FormatTok->is(tok::l_paren)) 869 parseParens(); 870 if (FormatTok->is(tok::l_brace)) 871 parseChildBlock(); 872 break; 873 case tok::l_brace: 874 if (!tryToParseBracedList()) { 875 // A block outside of parentheses must be the last part of a 876 // structural element. 877 // FIXME: Figure out cases where this is not true, and add projections 878 // for them (the one we know is missing are lambdas). 879 if (Style.BraceWrapping.AfterFunction) 880 addUnwrappedLine(); 881 FormatTok->Type = TT_FunctionLBrace; 882 parseBlock(/*MustBeDeclaration=*/false); 883 addUnwrappedLine(); 884 return; 885 } 886 // Otherwise this was a braced init list, and the structural 887 // element continues. 888 break; 889 case tok::kw_try: 890 // We arrive here when parsing function-try blocks. 891 parseTryCatch(); 892 return; 893 case tok::identifier: { 894 if (FormatTok->is(TT_MacroBlockEnd)) { 895 addUnwrappedLine(); 896 return; 897 } 898 899 // Parse function literal unless 'function' is the first token in a line 900 // in which case this should be treated as a free-standing function. 901 if (Style.Language == FormatStyle::LK_JavaScript && 902 FormatTok->is(Keywords.kw_function) && Line->Tokens.size() > 0) { 903 tryToParseJSFunction(); 904 break; 905 } 906 if ((Style.Language == FormatStyle::LK_JavaScript || 907 Style.Language == FormatStyle::LK_Java) && 908 FormatTok->is(Keywords.kw_interface)) { 909 parseRecord(); 910 addUnwrappedLine(); 911 return; 912 } 913 914 StringRef Text = FormatTok->TokenText; 915 nextToken(); 916 if (Line->Tokens.size() == 1 && 917 // JS doesn't have macros, and within classes colons indicate fields, 918 // not labels. 919 Style.Language != FormatStyle::LK_JavaScript) { 920 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) { 921 parseLabel(); 922 return; 923 } 924 // Recognize function-like macro usages without trailing semicolon as 925 // well as free-standing macros like Q_OBJECT. 926 bool FunctionLike = FormatTok->is(tok::l_paren); 927 if (FunctionLike) 928 parseParens(); 929 930 bool FollowedByNewline = 931 CommentsBeforeNextToken.empty() 932 ? FormatTok->NewlinesBefore > 0 933 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 934 935 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 936 tokenCanStartNewLine(FormatTok->Tok) && Text == Text.upper()) { 937 addUnwrappedLine(); 938 return; 939 } 940 } 941 break; 942 } 943 case tok::equal: 944 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType 945 // TT_JsFatArrow. The always start an expression or a child block if 946 // followed by a curly. 947 if (FormatTok->is(TT_JsFatArrow)) { 948 nextToken(); 949 if (FormatTok->is(tok::l_brace)) 950 parseChildBlock(); 951 break; 952 } 953 954 nextToken(); 955 if (FormatTok->Tok.is(tok::l_brace)) { 956 parseBracedList(); 957 } 958 break; 959 case tok::l_square: 960 parseSquare(); 961 break; 962 case tok::kw_new: 963 parseNew(); 964 break; 965 default: 966 nextToken(); 967 break; 968 } 969 } while (!eof()); 970 } 971 972 bool UnwrappedLineParser::tryToParseLambda() { 973 if (Style.Language != FormatStyle::LK_Cpp) { 974 nextToken(); 975 return false; 976 } 977 // FIXME: This is a dirty way to access the previous token. Find a better 978 // solution. 979 if (!Line->Tokens.empty() && 980 (Line->Tokens.back().Tok->isOneOf(tok::identifier, tok::kw_operator, 981 tok::kw_new, tok::kw_delete) || 982 Line->Tokens.back().Tok->closesScope() || 983 Line->Tokens.back().Tok->isSimpleTypeSpecifier())) { 984 nextToken(); 985 return false; 986 } 987 assert(FormatTok->is(tok::l_square)); 988 FormatToken &LSquare = *FormatTok; 989 if (!tryToParseLambdaIntroducer()) 990 return false; 991 992 while (FormatTok->isNot(tok::l_brace)) { 993 if (FormatTok->isSimpleTypeSpecifier()) { 994 nextToken(); 995 continue; 996 } 997 switch (FormatTok->Tok.getKind()) { 998 case tok::l_brace: 999 break; 1000 case tok::l_paren: 1001 parseParens(); 1002 break; 1003 case tok::amp: 1004 case tok::star: 1005 case tok::kw_const: 1006 case tok::comma: 1007 case tok::less: 1008 case tok::greater: 1009 case tok::identifier: 1010 case tok::numeric_constant: 1011 case tok::coloncolon: 1012 case tok::kw_mutable: 1013 nextToken(); 1014 break; 1015 case tok::arrow: 1016 FormatTok->Type = TT_LambdaArrow; 1017 nextToken(); 1018 break; 1019 default: 1020 return true; 1021 } 1022 } 1023 LSquare.Type = TT_LambdaLSquare; 1024 parseChildBlock(); 1025 return true; 1026 } 1027 1028 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 1029 nextToken(); 1030 if (FormatTok->is(tok::equal)) { 1031 nextToken(); 1032 if (FormatTok->is(tok::r_square)) { 1033 nextToken(); 1034 return true; 1035 } 1036 if (FormatTok->isNot(tok::comma)) 1037 return false; 1038 nextToken(); 1039 } else if (FormatTok->is(tok::amp)) { 1040 nextToken(); 1041 if (FormatTok->is(tok::r_square)) { 1042 nextToken(); 1043 return true; 1044 } 1045 if (!FormatTok->isOneOf(tok::comma, tok::identifier)) { 1046 return false; 1047 } 1048 if (FormatTok->is(tok::comma)) 1049 nextToken(); 1050 } else if (FormatTok->is(tok::r_square)) { 1051 nextToken(); 1052 return true; 1053 } 1054 do { 1055 if (FormatTok->is(tok::amp)) 1056 nextToken(); 1057 if (!FormatTok->isOneOf(tok::identifier, tok::kw_this)) 1058 return false; 1059 nextToken(); 1060 if (FormatTok->is(tok::ellipsis)) 1061 nextToken(); 1062 if (FormatTok->is(tok::comma)) { 1063 nextToken(); 1064 } else if (FormatTok->is(tok::r_square)) { 1065 nextToken(); 1066 return true; 1067 } else { 1068 return false; 1069 } 1070 } while (!eof()); 1071 return false; 1072 } 1073 1074 void UnwrappedLineParser::tryToParseJSFunction() { 1075 nextToken(); 1076 1077 // Consume function name. 1078 if (FormatTok->is(tok::identifier)) 1079 nextToken(); 1080 1081 if (FormatTok->isNot(tok::l_paren)) 1082 return; 1083 1084 // Parse formal parameter list. 1085 parseParens(); 1086 1087 if (FormatTok->is(tok::colon)) { 1088 // Parse a type definition. 1089 nextToken(); 1090 1091 // Eat the type declaration. For braced inline object types, balance braces, 1092 // otherwise just parse until finding an l_brace for the function body. 1093 if (FormatTok->is(tok::l_brace)) 1094 tryToParseBracedList(); 1095 else 1096 while (FormatTok->isNot(tok::l_brace) && !eof()) 1097 nextToken(); 1098 } 1099 1100 parseChildBlock(); 1101 } 1102 1103 bool UnwrappedLineParser::tryToParseBracedList() { 1104 if (FormatTok->BlockKind == BK_Unknown) 1105 calculateBraceTypes(); 1106 assert(FormatTok->BlockKind != BK_Unknown); 1107 if (FormatTok->BlockKind == BK_Block) 1108 return false; 1109 parseBracedList(); 1110 return true; 1111 } 1112 1113 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons) { 1114 bool HasError = false; 1115 nextToken(); 1116 1117 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 1118 // replace this by using parseAssigmentExpression() inside. 1119 do { 1120 if (Style.Language == FormatStyle::LK_JavaScript) { 1121 if (FormatTok->is(Keywords.kw_function)) { 1122 tryToParseJSFunction(); 1123 continue; 1124 } 1125 if (FormatTok->is(TT_JsFatArrow)) { 1126 nextToken(); 1127 // Fat arrows can be followed by simple expressions or by child blocks 1128 // in curly braces. 1129 if (FormatTok->is(tok::l_brace)) { 1130 parseChildBlock(); 1131 continue; 1132 } 1133 } 1134 } 1135 switch (FormatTok->Tok.getKind()) { 1136 case tok::caret: 1137 nextToken(); 1138 if (FormatTok->is(tok::l_brace)) { 1139 parseChildBlock(); 1140 } 1141 break; 1142 case tok::l_square: 1143 tryToParseLambda(); 1144 break; 1145 case tok::l_brace: 1146 // Assume there are no blocks inside a braced init list apart 1147 // from the ones we explicitly parse out (like lambdas). 1148 FormatTok->BlockKind = BK_BracedInit; 1149 parseBracedList(); 1150 break; 1151 case tok::l_paren: 1152 parseParens(); 1153 // JavaScript can just have free standing methods and getters/setters in 1154 // object literals. Detect them by a "{" following ")". 1155 if (Style.Language == FormatStyle::LK_JavaScript) { 1156 if (FormatTok->is(tok::l_brace)) 1157 parseChildBlock(); 1158 break; 1159 } 1160 break; 1161 case tok::r_brace: 1162 nextToken(); 1163 return !HasError; 1164 case tok::semi: 1165 HasError = true; 1166 if (!ContinueOnSemicolons) 1167 return !HasError; 1168 nextToken(); 1169 break; 1170 case tok::comma: 1171 nextToken(); 1172 break; 1173 default: 1174 nextToken(); 1175 break; 1176 } 1177 } while (!eof()); 1178 return false; 1179 } 1180 1181 void UnwrappedLineParser::parseParens() { 1182 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 1183 nextToken(); 1184 do { 1185 switch (FormatTok->Tok.getKind()) { 1186 case tok::l_paren: 1187 parseParens(); 1188 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 1189 parseChildBlock(); 1190 break; 1191 case tok::r_paren: 1192 nextToken(); 1193 return; 1194 case tok::r_brace: 1195 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1196 return; 1197 case tok::l_square: 1198 tryToParseLambda(); 1199 break; 1200 case tok::l_brace: 1201 if (!tryToParseBracedList()) 1202 parseChildBlock(); 1203 break; 1204 case tok::at: 1205 nextToken(); 1206 if (FormatTok->Tok.is(tok::l_brace)) 1207 parseBracedList(); 1208 break; 1209 case tok::identifier: 1210 if (Style.Language == FormatStyle::LK_JavaScript && 1211 FormatTok->is(Keywords.kw_function)) 1212 tryToParseJSFunction(); 1213 else 1214 nextToken(); 1215 break; 1216 default: 1217 nextToken(); 1218 break; 1219 } 1220 } while (!eof()); 1221 } 1222 1223 void UnwrappedLineParser::parseSquare() { 1224 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected."); 1225 if (tryToParseLambda()) 1226 return; 1227 do { 1228 switch (FormatTok->Tok.getKind()) { 1229 case tok::l_paren: 1230 parseParens(); 1231 break; 1232 case tok::r_square: 1233 nextToken(); 1234 return; 1235 case tok::r_brace: 1236 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1237 return; 1238 case tok::l_square: 1239 parseSquare(); 1240 break; 1241 case tok::l_brace: { 1242 if (!tryToParseBracedList()) 1243 parseChildBlock(); 1244 break; 1245 } 1246 case tok::at: 1247 nextToken(); 1248 if (FormatTok->Tok.is(tok::l_brace)) 1249 parseBracedList(); 1250 break; 1251 default: 1252 nextToken(); 1253 break; 1254 } 1255 } while (!eof()); 1256 } 1257 1258 void UnwrappedLineParser::parseIfThenElse() { 1259 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 1260 nextToken(); 1261 if (FormatTok->Tok.is(tok::l_paren)) 1262 parseParens(); 1263 bool NeedsUnwrappedLine = false; 1264 if (FormatTok->Tok.is(tok::l_brace)) { 1265 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1266 parseBlock(/*MustBeDeclaration=*/false); 1267 if (Style.BraceWrapping.BeforeElse) 1268 addUnwrappedLine(); 1269 else 1270 NeedsUnwrappedLine = true; 1271 } else { 1272 addUnwrappedLine(); 1273 ++Line->Level; 1274 parseStructuralElement(); 1275 --Line->Level; 1276 } 1277 if (FormatTok->Tok.is(tok::kw_else)) { 1278 nextToken(); 1279 if (FormatTok->Tok.is(tok::l_brace)) { 1280 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1281 parseBlock(/*MustBeDeclaration=*/false); 1282 addUnwrappedLine(); 1283 } else if (FormatTok->Tok.is(tok::kw_if)) { 1284 parseIfThenElse(); 1285 } else { 1286 addUnwrappedLine(); 1287 ++Line->Level; 1288 parseStructuralElement(); 1289 --Line->Level; 1290 } 1291 } else if (NeedsUnwrappedLine) { 1292 addUnwrappedLine(); 1293 } 1294 } 1295 1296 void UnwrappedLineParser::parseTryCatch() { 1297 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 1298 nextToken(); 1299 bool NeedsUnwrappedLine = false; 1300 if (FormatTok->is(tok::colon)) { 1301 // We are in a function try block, what comes is an initializer list. 1302 nextToken(); 1303 while (FormatTok->is(tok::identifier)) { 1304 nextToken(); 1305 if (FormatTok->is(tok::l_paren)) 1306 parseParens(); 1307 if (FormatTok->is(tok::comma)) 1308 nextToken(); 1309 } 1310 } 1311 // Parse try with resource. 1312 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) { 1313 parseParens(); 1314 } 1315 if (FormatTok->is(tok::l_brace)) { 1316 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1317 parseBlock(/*MustBeDeclaration=*/false); 1318 if (Style.BraceWrapping.BeforeCatch) { 1319 addUnwrappedLine(); 1320 } else { 1321 NeedsUnwrappedLine = true; 1322 } 1323 } else if (!FormatTok->is(tok::kw_catch)) { 1324 // The C++ standard requires a compound-statement after a try. 1325 // If there's none, we try to assume there's a structuralElement 1326 // and try to continue. 1327 addUnwrappedLine(); 1328 ++Line->Level; 1329 parseStructuralElement(); 1330 --Line->Level; 1331 } 1332 while (1) { 1333 if (FormatTok->is(tok::at)) 1334 nextToken(); 1335 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 1336 tok::kw___finally) || 1337 ((Style.Language == FormatStyle::LK_Java || 1338 Style.Language == FormatStyle::LK_JavaScript) && 1339 FormatTok->is(Keywords.kw_finally)) || 1340 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) || 1341 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally)))) 1342 break; 1343 nextToken(); 1344 while (FormatTok->isNot(tok::l_brace)) { 1345 if (FormatTok->is(tok::l_paren)) { 1346 parseParens(); 1347 continue; 1348 } 1349 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) 1350 return; 1351 nextToken(); 1352 } 1353 NeedsUnwrappedLine = false; 1354 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1355 parseBlock(/*MustBeDeclaration=*/false); 1356 if (Style.BraceWrapping.BeforeCatch) 1357 addUnwrappedLine(); 1358 else 1359 NeedsUnwrappedLine = true; 1360 } 1361 if (NeedsUnwrappedLine) 1362 addUnwrappedLine(); 1363 } 1364 1365 void UnwrappedLineParser::parseNamespace() { 1366 assert(FormatTok->Tok.is(tok::kw_namespace) && "'namespace' expected"); 1367 1368 const FormatToken &InitialToken = *FormatTok; 1369 nextToken(); 1370 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon)) 1371 nextToken(); 1372 if (FormatTok->Tok.is(tok::l_brace)) { 1373 if (ShouldBreakBeforeBrace(Style, InitialToken)) 1374 addUnwrappedLine(); 1375 1376 bool AddLevel = Style.NamespaceIndentation == FormatStyle::NI_All || 1377 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 1378 DeclarationScopeStack.size() > 1); 1379 parseBlock(/*MustBeDeclaration=*/true, AddLevel); 1380 // Munch the semicolon after a namespace. This is more common than one would 1381 // think. Puttin the semicolon into its own line is very ugly. 1382 if (FormatTok->Tok.is(tok::semi)) 1383 nextToken(); 1384 addUnwrappedLine(); 1385 } 1386 // FIXME: Add error handling. 1387 } 1388 1389 void UnwrappedLineParser::parseNew() { 1390 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 1391 nextToken(); 1392 if (Style.Language != FormatStyle::LK_Java) 1393 return; 1394 1395 // In Java, we can parse everything up to the parens, which aren't optional. 1396 do { 1397 // There should not be a ;, { or } before the new's open paren. 1398 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 1399 return; 1400 1401 // Consume the parens. 1402 if (FormatTok->is(tok::l_paren)) { 1403 parseParens(); 1404 1405 // If there is a class body of an anonymous class, consume that as child. 1406 if (FormatTok->is(tok::l_brace)) 1407 parseChildBlock(); 1408 return; 1409 } 1410 nextToken(); 1411 } while (!eof()); 1412 } 1413 1414 void UnwrappedLineParser::parseForOrWhileLoop() { 1415 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 1416 "'for', 'while' or foreach macro expected"); 1417 nextToken(); 1418 if (FormatTok->Tok.is(tok::l_paren)) 1419 parseParens(); 1420 if (FormatTok->Tok.is(tok::l_brace)) { 1421 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1422 parseBlock(/*MustBeDeclaration=*/false); 1423 addUnwrappedLine(); 1424 } else { 1425 addUnwrappedLine(); 1426 ++Line->Level; 1427 parseStructuralElement(); 1428 --Line->Level; 1429 } 1430 } 1431 1432 void UnwrappedLineParser::parseDoWhile() { 1433 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 1434 nextToken(); 1435 if (FormatTok->Tok.is(tok::l_brace)) { 1436 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1437 parseBlock(/*MustBeDeclaration=*/false); 1438 if (Style.BraceWrapping.IndentBraces) 1439 addUnwrappedLine(); 1440 } else { 1441 addUnwrappedLine(); 1442 ++Line->Level; 1443 parseStructuralElement(); 1444 --Line->Level; 1445 } 1446 1447 // FIXME: Add error handling. 1448 if (!FormatTok->Tok.is(tok::kw_while)) { 1449 addUnwrappedLine(); 1450 return; 1451 } 1452 1453 nextToken(); 1454 parseStructuralElement(); 1455 } 1456 1457 void UnwrappedLineParser::parseLabel() { 1458 nextToken(); 1459 unsigned OldLineLevel = Line->Level; 1460 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 1461 --Line->Level; 1462 if (CommentsBeforeNextToken.empty() && FormatTok->Tok.is(tok::l_brace)) { 1463 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1464 parseBlock(/*MustBeDeclaration=*/false); 1465 if (FormatTok->Tok.is(tok::kw_break)) { 1466 if (Style.BraceWrapping.AfterControlStatement) 1467 addUnwrappedLine(); 1468 parseStructuralElement(); 1469 } 1470 addUnwrappedLine(); 1471 } else { 1472 if (FormatTok->is(tok::semi)) 1473 nextToken(); 1474 addUnwrappedLine(); 1475 } 1476 Line->Level = OldLineLevel; 1477 } 1478 1479 void UnwrappedLineParser::parseCaseLabel() { 1480 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 1481 // FIXME: fix handling of complex expressions here. 1482 do { 1483 nextToken(); 1484 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 1485 parseLabel(); 1486 } 1487 1488 void UnwrappedLineParser::parseSwitch() { 1489 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 1490 nextToken(); 1491 if (FormatTok->Tok.is(tok::l_paren)) 1492 parseParens(); 1493 if (FormatTok->Tok.is(tok::l_brace)) { 1494 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1495 parseBlock(/*MustBeDeclaration=*/false); 1496 addUnwrappedLine(); 1497 } else { 1498 addUnwrappedLine(); 1499 ++Line->Level; 1500 parseStructuralElement(); 1501 --Line->Level; 1502 } 1503 } 1504 1505 void UnwrappedLineParser::parseAccessSpecifier() { 1506 nextToken(); 1507 // Understand Qt's slots. 1508 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 1509 nextToken(); 1510 // Otherwise, we don't know what it is, and we'd better keep the next token. 1511 if (FormatTok->Tok.is(tok::colon)) 1512 nextToken(); 1513 addUnwrappedLine(); 1514 } 1515 1516 void UnwrappedLineParser::parseEnum() { 1517 // Won't be 'enum' for NS_ENUMs. 1518 if (FormatTok->Tok.is(tok::kw_enum)) 1519 nextToken(); 1520 1521 // Eat up enum class ... 1522 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct)) 1523 nextToken(); 1524 1525 while (FormatTok->Tok.getIdentifierInfo() || 1526 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 1527 tok::greater, tok::comma, tok::question)) { 1528 nextToken(); 1529 // We can have macros or attributes in between 'enum' and the enum name. 1530 if (FormatTok->is(tok::l_paren)) 1531 parseParens(); 1532 if (FormatTok->is(tok::identifier)) { 1533 nextToken(); 1534 // If there are two identifiers in a row, this is likely an elaborate 1535 // return type. In Java, this can be "implements", etc. 1536 if (Style.Language == FormatStyle::LK_Cpp && 1537 FormatTok->is(tok::identifier)) 1538 return; 1539 } 1540 } 1541 1542 // Just a declaration or something is wrong. 1543 if (FormatTok->isNot(tok::l_brace)) 1544 return; 1545 FormatTok->BlockKind = BK_Block; 1546 1547 if (Style.Language == FormatStyle::LK_Java) { 1548 // Java enums are different. 1549 parseJavaEnumBody(); 1550 return; 1551 } else if (Style.Language == FormatStyle::LK_Proto) { 1552 parseBlock(/*MustBeDeclaration=*/true); 1553 return; 1554 } 1555 1556 // Parse enum body. 1557 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true); 1558 if (HasError) { 1559 if (FormatTok->is(tok::semi)) 1560 nextToken(); 1561 addUnwrappedLine(); 1562 } 1563 1564 // There is no addUnwrappedLine() here so that we fall through to parsing a 1565 // structural element afterwards. Thus, in "enum A {} n, m;", 1566 // "} n, m;" will end up in one unwrapped line. 1567 } 1568 1569 void UnwrappedLineParser::parseJavaEnumBody() { 1570 // Determine whether the enum is simple, i.e. does not have a semicolon or 1571 // constants with class bodies. Simple enums can be formatted like braced 1572 // lists, contracted to a single line, etc. 1573 unsigned StoredPosition = Tokens->getPosition(); 1574 bool IsSimple = true; 1575 FormatToken *Tok = Tokens->getNextToken(); 1576 while (Tok) { 1577 if (Tok->is(tok::r_brace)) 1578 break; 1579 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 1580 IsSimple = false; 1581 break; 1582 } 1583 // FIXME: This will also mark enums with braces in the arguments to enum 1584 // constants as "not simple". This is probably fine in practice, though. 1585 Tok = Tokens->getNextToken(); 1586 } 1587 FormatTok = Tokens->setPosition(StoredPosition); 1588 1589 if (IsSimple) { 1590 parseBracedList(); 1591 addUnwrappedLine(); 1592 return; 1593 } 1594 1595 // Parse the body of a more complex enum. 1596 // First add a line for everything up to the "{". 1597 nextToken(); 1598 addUnwrappedLine(); 1599 ++Line->Level; 1600 1601 // Parse the enum constants. 1602 while (FormatTok) { 1603 if (FormatTok->is(tok::l_brace)) { 1604 // Parse the constant's class body. 1605 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 1606 /*MunchSemi=*/false); 1607 } else if (FormatTok->is(tok::l_paren)) { 1608 parseParens(); 1609 } else if (FormatTok->is(tok::comma)) { 1610 nextToken(); 1611 addUnwrappedLine(); 1612 } else if (FormatTok->is(tok::semi)) { 1613 nextToken(); 1614 addUnwrappedLine(); 1615 break; 1616 } else if (FormatTok->is(tok::r_brace)) { 1617 addUnwrappedLine(); 1618 break; 1619 } else { 1620 nextToken(); 1621 } 1622 } 1623 1624 // Parse the class body after the enum's ";" if any. 1625 parseLevel(/*HasOpeningBrace=*/true); 1626 nextToken(); 1627 --Line->Level; 1628 addUnwrappedLine(); 1629 } 1630 1631 void UnwrappedLineParser::parseRecord() { 1632 const FormatToken &InitialToken = *FormatTok; 1633 nextToken(); 1634 1635 // The actual identifier can be a nested name specifier, and in macros 1636 // it is often token-pasted. 1637 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 1638 tok::kw___attribute, tok::kw___declspec, 1639 tok::kw_alignas) || 1640 ((Style.Language == FormatStyle::LK_Java || 1641 Style.Language == FormatStyle::LK_JavaScript) && 1642 FormatTok->isOneOf(tok::period, tok::comma))) { 1643 bool IsNonMacroIdentifier = 1644 FormatTok->is(tok::identifier) && 1645 FormatTok->TokenText != FormatTok->TokenText.upper(); 1646 nextToken(); 1647 // We can have macros or attributes in between 'class' and the class name. 1648 if (!IsNonMacroIdentifier && FormatTok->Tok.is(tok::l_paren)) 1649 parseParens(); 1650 } 1651 1652 // Note that parsing away template declarations here leads to incorrectly 1653 // accepting function declarations as record declarations. 1654 // In general, we cannot solve this problem. Consider: 1655 // class A<int> B() {} 1656 // which can be a function definition or a class definition when B() is a 1657 // macro. If we find enough real-world cases where this is a problem, we 1658 // can parse for the 'template' keyword in the beginning of the statement, 1659 // and thus rule out the record production in case there is no template 1660 // (this would still leave us with an ambiguity between template function 1661 // and class declarations). 1662 if (FormatTok->isOneOf(tok::colon, tok::less)) { 1663 while (!eof()) { 1664 if (FormatTok->is(tok::l_brace)) { 1665 calculateBraceTypes(/*ExpectClassBody=*/true); 1666 if (!tryToParseBracedList()) 1667 break; 1668 } 1669 if (FormatTok->Tok.is(tok::semi)) 1670 return; 1671 nextToken(); 1672 } 1673 } 1674 if (FormatTok->Tok.is(tok::l_brace)) { 1675 if (ShouldBreakBeforeBrace(Style, InitialToken)) 1676 addUnwrappedLine(); 1677 1678 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 1679 /*MunchSemi=*/false); 1680 } 1681 // There is no addUnwrappedLine() here so that we fall through to parsing a 1682 // structural element afterwards. Thus, in "class A {} n, m;", 1683 // "} n, m;" will end up in one unwrapped line. 1684 } 1685 1686 void UnwrappedLineParser::parseObjCProtocolList() { 1687 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 1688 do 1689 nextToken(); 1690 while (!eof() && FormatTok->Tok.isNot(tok::greater)); 1691 nextToken(); // Skip '>'. 1692 } 1693 1694 void UnwrappedLineParser::parseObjCUntilAtEnd() { 1695 do { 1696 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 1697 nextToken(); 1698 addUnwrappedLine(); 1699 break; 1700 } 1701 if (FormatTok->is(tok::l_brace)) { 1702 parseBlock(/*MustBeDeclaration=*/false); 1703 // In ObjC interfaces, nothing should be following the "}". 1704 addUnwrappedLine(); 1705 } else if (FormatTok->is(tok::r_brace)) { 1706 // Ignore stray "}". parseStructuralElement doesn't consume them. 1707 nextToken(); 1708 addUnwrappedLine(); 1709 } else { 1710 parseStructuralElement(); 1711 } 1712 } while (!eof()); 1713 } 1714 1715 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 1716 nextToken(); 1717 nextToken(); // interface name 1718 1719 // @interface can be followed by either a base class, or a category. 1720 if (FormatTok->Tok.is(tok::colon)) { 1721 nextToken(); 1722 nextToken(); // base class name 1723 } else if (FormatTok->Tok.is(tok::l_paren)) 1724 // Skip category, if present. 1725 parseParens(); 1726 1727 if (FormatTok->Tok.is(tok::less)) 1728 parseObjCProtocolList(); 1729 1730 if (FormatTok->Tok.is(tok::l_brace)) { 1731 if (Style.BraceWrapping.AfterObjCDeclaration) 1732 addUnwrappedLine(); 1733 parseBlock(/*MustBeDeclaration=*/true); 1734 } 1735 1736 // With instance variables, this puts '}' on its own line. Without instance 1737 // variables, this ends the @interface line. 1738 addUnwrappedLine(); 1739 1740 parseObjCUntilAtEnd(); 1741 } 1742 1743 void UnwrappedLineParser::parseObjCProtocol() { 1744 nextToken(); 1745 nextToken(); // protocol name 1746 1747 if (FormatTok->Tok.is(tok::less)) 1748 parseObjCProtocolList(); 1749 1750 // Check for protocol declaration. 1751 if (FormatTok->Tok.is(tok::semi)) { 1752 nextToken(); 1753 return addUnwrappedLine(); 1754 } 1755 1756 addUnwrappedLine(); 1757 parseObjCUntilAtEnd(); 1758 } 1759 1760 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 1761 assert(FormatTok->isOneOf(Keywords.kw_import, tok::kw_export)); 1762 nextToken(); 1763 1764 // Consume the "default" in "export default class/function". 1765 if (FormatTok->is(tok::kw_default)) 1766 nextToken(); 1767 1768 // Consume "function" and "default function", so that these get parsed as 1769 // free-standing JS functions, i.e. do not require a trailing semicolon. 1770 if (FormatTok->is(Keywords.kw_function)) { 1771 nextToken(); 1772 return; 1773 } 1774 1775 if (FormatTok->isOneOf(tok::kw_const, tok::kw_class, tok::kw_enum, 1776 Keywords.kw_let, Keywords.kw_var)) 1777 return; // Fall through to parsing the corresponding structure. 1778 1779 if (FormatTok->is(tok::l_brace)) { 1780 FormatTok->BlockKind = BK_Block; 1781 parseBracedList(); 1782 } 1783 1784 while (!eof() && FormatTok->isNot(tok::semi) && 1785 FormatTok->isNot(tok::l_brace)) { 1786 nextToken(); 1787 } 1788 } 1789 1790 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 1791 StringRef Prefix = "") { 1792 llvm::dbgs() << Prefix << "Line(" << Line.Level << ")" 1793 << (Line.InPPDirective ? " MACRO" : "") << ": "; 1794 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 1795 E = Line.Tokens.end(); 1796 I != E; ++I) { 1797 llvm::dbgs() << I->Tok->Tok.getName() << "[" << I->Tok->Type << "] "; 1798 } 1799 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 1800 E = Line.Tokens.end(); 1801 I != E; ++I) { 1802 const UnwrappedLineNode &Node = *I; 1803 for (SmallVectorImpl<UnwrappedLine>::const_iterator 1804 I = Node.Children.begin(), 1805 E = Node.Children.end(); 1806 I != E; ++I) { 1807 printDebugInfo(*I, "\nChild: "); 1808 } 1809 } 1810 llvm::dbgs() << "\n"; 1811 } 1812 1813 void UnwrappedLineParser::addUnwrappedLine() { 1814 if (Line->Tokens.empty()) 1815 return; 1816 DEBUG({ 1817 if (CurrentLines == &Lines) 1818 printDebugInfo(*Line); 1819 }); 1820 CurrentLines->push_back(std::move(*Line)); 1821 Line->Tokens.clear(); 1822 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 1823 CurrentLines->append( 1824 std::make_move_iterator(PreprocessorDirectives.begin()), 1825 std::make_move_iterator(PreprocessorDirectives.end())); 1826 PreprocessorDirectives.clear(); 1827 } 1828 } 1829 1830 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 1831 1832 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 1833 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 1834 FormatTok.NewlinesBefore > 0; 1835 } 1836 1837 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 1838 bool JustComments = Line->Tokens.empty(); 1839 for (SmallVectorImpl<FormatToken *>::const_iterator 1840 I = CommentsBeforeNextToken.begin(), 1841 E = CommentsBeforeNextToken.end(); 1842 I != E; ++I) { 1843 if (isOnNewLine(**I) && JustComments) 1844 addUnwrappedLine(); 1845 pushToken(*I); 1846 } 1847 if (NewlineBeforeNext && JustComments) 1848 addUnwrappedLine(); 1849 CommentsBeforeNextToken.clear(); 1850 } 1851 1852 void UnwrappedLineParser::nextToken() { 1853 if (eof()) 1854 return; 1855 flushComments(isOnNewLine(*FormatTok)); 1856 pushToken(FormatTok); 1857 readToken(); 1858 } 1859 1860 void UnwrappedLineParser::readToken() { 1861 bool CommentsInCurrentLine = true; 1862 do { 1863 FormatTok = Tokens->getNextToken(); 1864 assert(FormatTok); 1865 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 1866 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 1867 // If there is an unfinished unwrapped line, we flush the preprocessor 1868 // directives only after that unwrapped line was finished later. 1869 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 1870 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 1871 // Comments stored before the preprocessor directive need to be output 1872 // before the preprocessor directive, at the same level as the 1873 // preprocessor directive, as we consider them to apply to the directive. 1874 flushComments(isOnNewLine(*FormatTok)); 1875 parsePPDirective(); 1876 } 1877 while (FormatTok->Type == TT_ConflictStart || 1878 FormatTok->Type == TT_ConflictEnd || 1879 FormatTok->Type == TT_ConflictAlternative) { 1880 if (FormatTok->Type == TT_ConflictStart) { 1881 conditionalCompilationStart(/*Unreachable=*/false); 1882 } else if (FormatTok->Type == TT_ConflictAlternative) { 1883 conditionalCompilationAlternative(); 1884 } else if (FormatTok->Type == TT_ConflictEnd) { 1885 conditionalCompilationEnd(); 1886 } 1887 FormatTok = Tokens->getNextToken(); 1888 FormatTok->MustBreakBefore = true; 1889 } 1890 1891 if (!PPStack.empty() && (PPStack.back() == PP_Unreachable) && 1892 !Line->InPPDirective) { 1893 continue; 1894 } 1895 1896 if (!FormatTok->Tok.is(tok::comment)) 1897 return; 1898 if (isOnNewLine(*FormatTok) || FormatTok->IsFirst) { 1899 CommentsInCurrentLine = false; 1900 } 1901 if (CommentsInCurrentLine) { 1902 pushToken(FormatTok); 1903 } else { 1904 CommentsBeforeNextToken.push_back(FormatTok); 1905 } 1906 } while (!eof()); 1907 } 1908 1909 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 1910 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 1911 if (MustBreakBeforeNextToken) { 1912 Line->Tokens.back().Tok->MustBreakBefore = true; 1913 MustBreakBeforeNextToken = false; 1914 } 1915 } 1916 1917 } // end namespace format 1918 } // end namespace clang 1919