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.BreakBeforeBraces == FormatStyle::BS_Allman) { 158 Parser->addUnwrappedLine(); 159 } else if (Style.BreakBeforeBraces == FormatStyle::BS_GNU) { 160 Parser->addUnwrappedLine(); 161 ++LineLevel; 162 } 163 } 164 ~CompoundStatementIndenter() { LineLevel = OldLineLevel; } 165 166 private: 167 unsigned &LineLevel; 168 unsigned OldLineLevel; 169 }; 170 171 namespace { 172 173 class IndexedTokenSource : public FormatTokenSource { 174 public: 175 IndexedTokenSource(ArrayRef<FormatToken *> Tokens) 176 : Tokens(Tokens), Position(-1) {} 177 178 FormatToken *getNextToken() override { 179 ++Position; 180 return Tokens[Position]; 181 } 182 183 unsigned getPosition() override { 184 assert(Position >= 0); 185 return Position; 186 } 187 188 FormatToken *setPosition(unsigned P) override { 189 Position = P; 190 return Tokens[Position]; 191 } 192 193 void reset() { Position = -1; } 194 195 private: 196 ArrayRef<FormatToken *> Tokens; 197 int Position; 198 }; 199 200 } // end anonymous namespace 201 202 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style, 203 const AdditionalKeywords &Keywords, 204 ArrayRef<FormatToken *> Tokens, 205 UnwrappedLineConsumer &Callback) 206 : Line(new UnwrappedLine), MustBreakBeforeNextToken(false), 207 CurrentLines(&Lines), Style(Style), Keywords(Keywords), Tokens(nullptr), 208 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1) {} 209 210 void UnwrappedLineParser::reset() { 211 PPBranchLevel = -1; 212 Line.reset(new UnwrappedLine); 213 CommentsBeforeNextToken.clear(); 214 FormatTok = nullptr; 215 MustBreakBeforeNextToken = false; 216 PreprocessorDirectives.clear(); 217 CurrentLines = &Lines; 218 DeclarationScopeStack.clear(); 219 PPStack.clear(); 220 } 221 222 void UnwrappedLineParser::parse() { 223 IndexedTokenSource TokenSource(AllTokens); 224 do { 225 DEBUG(llvm::dbgs() << "----\n"); 226 reset(); 227 Tokens = &TokenSource; 228 TokenSource.reset(); 229 230 readToken(); 231 parseFile(); 232 // Create line with eof token. 233 pushToken(FormatTok); 234 addUnwrappedLine(); 235 236 for (SmallVectorImpl<UnwrappedLine>::iterator I = Lines.begin(), 237 E = Lines.end(); 238 I != E; ++I) { 239 Callback.consumeUnwrappedLine(*I); 240 } 241 Callback.finishRun(); 242 Lines.clear(); 243 while (!PPLevelBranchIndex.empty() && 244 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) { 245 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1); 246 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1); 247 } 248 if (!PPLevelBranchIndex.empty()) { 249 ++PPLevelBranchIndex.back(); 250 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size()); 251 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back()); 252 } 253 } while (!PPLevelBranchIndex.empty()); 254 } 255 256 void UnwrappedLineParser::parseFile() { 257 // The top-level context in a file always has declarations, except for pre- 258 // processor directives and JavaScript files. 259 bool MustBeDeclaration = 260 !Line->InPPDirective && Style.Language != FormatStyle::LK_JavaScript; 261 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 262 MustBeDeclaration); 263 parseLevel(/*HasOpeningBrace=*/false); 264 // Make sure to format the remaining tokens. 265 flushComments(true); 266 addUnwrappedLine(); 267 } 268 269 void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) { 270 bool SwitchLabelEncountered = false; 271 do { 272 tok::TokenKind kind = FormatTok->Tok.getKind(); 273 if (FormatTok->Type == TT_MacroBlockBegin) { 274 kind = tok::l_brace; 275 } else if (FormatTok->Type == TT_MacroBlockEnd) { 276 kind = tok::r_brace; 277 } 278 279 switch (kind) { 280 case tok::comment: 281 nextToken(); 282 addUnwrappedLine(); 283 break; 284 case tok::l_brace: 285 // FIXME: Add parameter whether this can happen - if this happens, we must 286 // be in a non-declaration context. 287 if (!FormatTok->is(TT_MacroBlockBegin) && tryToParseBracedList()) 288 continue; 289 parseBlock(/*MustBeDeclaration=*/false); 290 addUnwrappedLine(); 291 break; 292 case tok::r_brace: 293 if (HasOpeningBrace) 294 return; 295 nextToken(); 296 addUnwrappedLine(); 297 break; 298 case tok::kw_default: 299 case tok::kw_case: 300 if (!SwitchLabelEncountered && 301 (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1))) 302 ++Line->Level; 303 SwitchLabelEncountered = true; 304 parseStructuralElement(); 305 break; 306 default: 307 parseStructuralElement(); 308 break; 309 } 310 } while (!eof()); 311 } 312 313 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) { 314 // We'll parse forward through the tokens until we hit 315 // a closing brace or eof - note that getNextToken() will 316 // parse macros, so this will magically work inside macro 317 // definitions, too. 318 unsigned StoredPosition = Tokens->getPosition(); 319 FormatToken *Tok = FormatTok; 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 none-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 Tok->BlockKind = BK_Unknown; 337 LBraceStack.push_back(Tok); 338 break; 339 case tok::r_brace: 340 if (!LBraceStack.empty()) { 341 if (LBraceStack.back()->BlockKind == BK_Unknown) { 342 bool ProbablyBracedList = false; 343 if (Style.Language == FormatStyle::LK_Proto) { 344 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square); 345 } else { 346 // Using OriginalColumn to distinguish between ObjC methods and 347 // binary operators is a bit hacky. 348 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 349 NextTok->OriginalColumn == 0; 350 351 // If there is a comma, semicolon or right paren after the closing 352 // brace, we assume this is a braced initializer list. Note that 353 // regardless how we mark inner braces here, we will overwrite the 354 // BlockKind later if we parse a braced list (where all blocks 355 // inside are by default braced lists), or when we explicitly detect 356 // blocks (for example while parsing lambdas). 357 // 358 // We exclude + and - as they can be ObjC visibility modifiers. 359 ProbablyBracedList = 360 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 361 tok::r_paren, tok::r_square, tok::l_brace, 362 tok::l_paren, tok::ellipsis) || 363 (NextTok->is(tok::semi) && 364 (!ExpectClassBody || LBraceStack.size() != 1)) || 365 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 366 } 367 if (ProbablyBracedList) { 368 Tok->BlockKind = BK_BracedInit; 369 LBraceStack.back()->BlockKind = BK_BracedInit; 370 } else { 371 Tok->BlockKind = BK_Block; 372 LBraceStack.back()->BlockKind = BK_Block; 373 } 374 } 375 LBraceStack.pop_back(); 376 } 377 break; 378 case tok::at: 379 case tok::semi: 380 case tok::kw_if: 381 case tok::kw_while: 382 case tok::kw_for: 383 case tok::kw_switch: 384 case tok::kw_try: 385 case tok::kw___try: 386 if (!LBraceStack.empty()) 387 LBraceStack.back()->BlockKind = BK_Block; 388 break; 389 default: 390 break; 391 } 392 Tok = NextTok; 393 } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty()); 394 // Assume other blocks for all unclosed opening braces. 395 for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) { 396 if (LBraceStack[i]->BlockKind == BK_Unknown) 397 LBraceStack[i]->BlockKind = BK_Block; 398 } 399 400 FormatTok = Tokens->setPosition(StoredPosition); 401 } 402 403 void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, bool AddLevel, 404 bool MunchSemi) { 405 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) && 406 "'{' or macro block token expected"); 407 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 408 409 unsigned InitialLevel = Line->Level; 410 nextToken(); 411 412 if (MacroBlock && FormatTok->is(tok::l_paren)) 413 parseParens(); 414 415 addUnwrappedLine(); 416 417 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 418 MustBeDeclaration); 419 if (AddLevel) 420 ++Line->Level; 421 parseLevel(/*HasOpeningBrace=*/true); 422 423 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd) 424 : !FormatTok->is(tok::r_brace)) { 425 Line->Level = InitialLevel; 426 return; 427 } 428 429 nextToken(); // Munch the closing brace. 430 431 if (MacroBlock && FormatTok->is(tok::l_paren)) 432 parseParens(); 433 434 if (MunchSemi && FormatTok->Tok.is(tok::semi)) 435 nextToken(); 436 Line->Level = InitialLevel; 437 } 438 439 static bool isGoogScope(const UnwrappedLine &Line) { 440 // FIXME: Closure-library specific stuff should not be hard-coded but be 441 // configurable. 442 if (Line.Tokens.size() < 4) 443 return false; 444 auto I = Line.Tokens.begin(); 445 if (I->Tok->TokenText != "goog") 446 return false; 447 ++I; 448 if (I->Tok->isNot(tok::period)) 449 return false; 450 ++I; 451 if (I->Tok->TokenText != "scope") 452 return false; 453 ++I; 454 return I->Tok->is(tok::l_paren); 455 } 456 457 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 458 const FormatToken &InitialToken) { 459 switch (Style.BreakBeforeBraces) { 460 case FormatStyle::BS_Linux: 461 return InitialToken.isOneOf(tok::kw_namespace, tok::kw_class); 462 case FormatStyle::BS_Mozilla: 463 return InitialToken.isOneOf(tok::kw_class, tok::kw_struct, tok::kw_union); 464 case FormatStyle::BS_Allman: 465 case FormatStyle::BS_GNU: 466 return true; 467 default: 468 return false; 469 } 470 } 471 472 void UnwrappedLineParser::parseChildBlock() { 473 FormatTok->BlockKind = BK_Block; 474 nextToken(); 475 { 476 bool GoogScope = 477 Style.Language == FormatStyle::LK_JavaScript && isGoogScope(*Line); 478 ScopedLineState LineState(*this); 479 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 480 /*MustBeDeclaration=*/false); 481 Line->Level += GoogScope ? 0 : 1; 482 parseLevel(/*HasOpeningBrace=*/true); 483 flushComments(isOnNewLine(*FormatTok)); 484 Line->Level -= GoogScope ? 0 : 1; 485 } 486 nextToken(); 487 } 488 489 void UnwrappedLineParser::parsePPDirective() { 490 assert(FormatTok->Tok.is(tok::hash) && "'#' expected"); 491 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 492 nextToken(); 493 494 if (!FormatTok->Tok.getIdentifierInfo()) { 495 parsePPUnknown(); 496 return; 497 } 498 499 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 500 case tok::pp_define: 501 parsePPDefine(); 502 return; 503 case tok::pp_if: 504 parsePPIf(/*IfDef=*/false); 505 break; 506 case tok::pp_ifdef: 507 case tok::pp_ifndef: 508 parsePPIf(/*IfDef=*/true); 509 break; 510 case tok::pp_else: 511 parsePPElse(); 512 break; 513 case tok::pp_elif: 514 parsePPElIf(); 515 break; 516 case tok::pp_endif: 517 parsePPEndIf(); 518 break; 519 default: 520 parsePPUnknown(); 521 break; 522 } 523 } 524 525 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 526 if (Unreachable || (!PPStack.empty() && PPStack.back() == PP_Unreachable)) 527 PPStack.push_back(PP_Unreachable); 528 else 529 PPStack.push_back(PP_Conditional); 530 } 531 532 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 533 ++PPBranchLevel; 534 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 535 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 536 PPLevelBranchIndex.push_back(0); 537 PPLevelBranchCount.push_back(0); 538 } 539 PPChainBranchIndex.push(0); 540 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 541 conditionalCompilationCondition(Unreachable || Skip); 542 } 543 544 void UnwrappedLineParser::conditionalCompilationAlternative() { 545 if (!PPStack.empty()) 546 PPStack.pop_back(); 547 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 548 if (!PPChainBranchIndex.empty()) 549 ++PPChainBranchIndex.top(); 550 conditionalCompilationCondition( 551 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 552 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 553 } 554 555 void UnwrappedLineParser::conditionalCompilationEnd() { 556 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 557 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 558 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) { 559 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 560 } 561 } 562 // Guard against #endif's without #if. 563 if (PPBranchLevel > 0) 564 --PPBranchLevel; 565 if (!PPChainBranchIndex.empty()) 566 PPChainBranchIndex.pop(); 567 if (!PPStack.empty()) 568 PPStack.pop_back(); 569 } 570 571 void UnwrappedLineParser::parsePPIf(bool IfDef) { 572 nextToken(); 573 bool IsLiteralFalse = (FormatTok->Tok.isLiteral() && 574 FormatTok->Tok.getLiteralData() != nullptr && 575 StringRef(FormatTok->Tok.getLiteralData(), 576 FormatTok->Tok.getLength()) == "0") || 577 FormatTok->Tok.is(tok::kw_false); 578 conditionalCompilationStart(!IfDef && IsLiteralFalse); 579 parsePPUnknown(); 580 } 581 582 void UnwrappedLineParser::parsePPElse() { 583 conditionalCompilationAlternative(); 584 parsePPUnknown(); 585 } 586 587 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 588 589 void UnwrappedLineParser::parsePPEndIf() { 590 conditionalCompilationEnd(); 591 parsePPUnknown(); 592 } 593 594 void UnwrappedLineParser::parsePPDefine() { 595 nextToken(); 596 597 if (FormatTok->Tok.getKind() != tok::identifier) { 598 parsePPUnknown(); 599 return; 600 } 601 nextToken(); 602 if (FormatTok->Tok.getKind() == tok::l_paren && 603 FormatTok->WhitespaceRange.getBegin() == 604 FormatTok->WhitespaceRange.getEnd()) { 605 parseParens(); 606 } 607 addUnwrappedLine(); 608 Line->Level = 1; 609 610 // Errors during a preprocessor directive can only affect the layout of the 611 // preprocessor directive, and thus we ignore them. An alternative approach 612 // would be to use the same approach we use on the file level (no 613 // re-indentation if there was a structural error) within the macro 614 // definition. 615 parseFile(); 616 } 617 618 void UnwrappedLineParser::parsePPUnknown() { 619 do { 620 nextToken(); 621 } while (!eof()); 622 addUnwrappedLine(); 623 } 624 625 // Here we blacklist certain tokens that are not usually the first token in an 626 // unwrapped line. This is used in attempt to distinguish macro calls without 627 // trailing semicolons from other constructs split to several lines. 628 static bool tokenCanStartNewLine(const clang::Token &Tok) { 629 // Semicolon can be a null-statement, l_square can be a start of a macro or 630 // a C++11 attribute, but this doesn't seem to be common. 631 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 632 Tok.isNot(tok::l_square) && 633 // Tokens that can only be used as binary operators and a part of 634 // overloaded operator names. 635 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 636 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 637 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 638 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 639 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 640 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 641 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 642 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 643 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 644 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 645 Tok.isNot(tok::lesslessequal) && 646 // Colon is used in labels, base class lists, initializer lists, 647 // range-based for loops, ternary operator, but should never be the 648 // first token in an unwrapped line. 649 Tok.isNot(tok::colon) && 650 // 'noexcept' is a trailing annotation. 651 Tok.isNot(tok::kw_noexcept); 652 } 653 654 void UnwrappedLineParser::parseStructuralElement() { 655 assert(!FormatTok->Tok.is(tok::l_brace)); 656 switch (FormatTok->Tok.getKind()) { 657 case tok::at: 658 nextToken(); 659 if (FormatTok->Tok.is(tok::l_brace)) { 660 parseBracedList(); 661 break; 662 } 663 switch (FormatTok->Tok.getObjCKeywordID()) { 664 case tok::objc_public: 665 case tok::objc_protected: 666 case tok::objc_package: 667 case tok::objc_private: 668 return parseAccessSpecifier(); 669 case tok::objc_interface: 670 case tok::objc_implementation: 671 return parseObjCInterfaceOrImplementation(); 672 case tok::objc_protocol: 673 return parseObjCProtocol(); 674 case tok::objc_end: 675 return; // Handled by the caller. 676 case tok::objc_optional: 677 case tok::objc_required: 678 nextToken(); 679 addUnwrappedLine(); 680 return; 681 case tok::objc_autoreleasepool: 682 nextToken(); 683 if (FormatTok->Tok.is(tok::l_brace)) { 684 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman || 685 Style.BreakBeforeBraces == FormatStyle::BS_GNU) 686 addUnwrappedLine(); 687 parseBlock(/*MustBeDeclaration=*/false); 688 } 689 addUnwrappedLine(); 690 return; 691 case tok::objc_try: 692 // This branch isn't strictly necessary (the kw_try case below would 693 // do this too after the tok::at is parsed above). But be explicit. 694 parseTryCatch(); 695 return; 696 default: 697 break; 698 } 699 break; 700 case tok::kw_asm: 701 nextToken(); 702 if (FormatTok->is(tok::l_brace)) { 703 FormatTok->Type = TT_InlineASMBrace; 704 nextToken(); 705 while (FormatTok && FormatTok->isNot(tok::eof)) { 706 if (FormatTok->is(tok::r_brace)) { 707 FormatTok->Type = TT_InlineASMBrace; 708 nextToken(); 709 addUnwrappedLine(); 710 break; 711 } 712 FormatTok->Finalized = true; 713 nextToken(); 714 } 715 } 716 break; 717 case tok::kw_namespace: 718 parseNamespace(); 719 return; 720 case tok::kw_inline: 721 nextToken(); 722 if (FormatTok->Tok.is(tok::kw_namespace)) { 723 parseNamespace(); 724 return; 725 } 726 break; 727 case tok::kw_public: 728 case tok::kw_protected: 729 case tok::kw_private: 730 if (Style.Language == FormatStyle::LK_Java || 731 Style.Language == FormatStyle::LK_JavaScript) 732 nextToken(); 733 else 734 parseAccessSpecifier(); 735 return; 736 case tok::kw_if: 737 parseIfThenElse(); 738 return; 739 case tok::kw_for: 740 case tok::kw_while: 741 parseForOrWhileLoop(); 742 return; 743 case tok::kw_do: 744 parseDoWhile(); 745 return; 746 case tok::kw_switch: 747 parseSwitch(); 748 return; 749 case tok::kw_default: 750 nextToken(); 751 parseLabel(); 752 return; 753 case tok::kw_case: 754 parseCaseLabel(); 755 return; 756 case tok::kw_try: 757 case tok::kw___try: 758 parseTryCatch(); 759 return; 760 case tok::kw_extern: 761 nextToken(); 762 if (FormatTok->Tok.is(tok::string_literal)) { 763 nextToken(); 764 if (FormatTok->Tok.is(tok::l_brace)) { 765 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/false); 766 addUnwrappedLine(); 767 return; 768 } 769 } 770 break; 771 case tok::kw_export: 772 if (Style.Language == FormatStyle::LK_JavaScript) { 773 parseJavaScriptEs6ImportExport(); 774 return; 775 } 776 break; 777 case tok::identifier: 778 if (FormatTok->is(TT_ForEachMacro)) { 779 parseForOrWhileLoop(); 780 return; 781 } 782 if (FormatTok->is(TT_MacroBlockBegin)) { 783 parseBlock(/*MustBeDeclaration=*/false, /*AddLevel=*/true, 784 /*MunchSemi=*/false); 785 return; 786 } 787 if (Style.Language == FormatStyle::LK_JavaScript && 788 FormatTok->is(Keywords.kw_import)) { 789 parseJavaScriptEs6ImportExport(); 790 return; 791 } 792 if (FormatTok->is(Keywords.kw_signals)) { 793 nextToken(); 794 if (FormatTok->is(tok::colon)) { 795 nextToken(); 796 addUnwrappedLine(); 797 } 798 return; 799 } 800 // In all other cases, parse the declaration. 801 break; 802 default: 803 break; 804 } 805 do { 806 switch (FormatTok->Tok.getKind()) { 807 case tok::at: 808 nextToken(); 809 if (FormatTok->Tok.is(tok::l_brace)) 810 parseBracedList(); 811 break; 812 case tok::kw_enum: 813 // parseEnum falls through and does not yet add an unwrapped line as an 814 // enum definition can start a structural element. 815 parseEnum(); 816 // This only applies for C++. 817 if (Style.Language != FormatStyle::LK_Cpp) { 818 addUnwrappedLine(); 819 return; 820 } 821 break; 822 case tok::kw_typedef: 823 nextToken(); 824 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 825 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS)) 826 parseEnum(); 827 break; 828 case tok::kw_struct: 829 case tok::kw_union: 830 case tok::kw_class: 831 // parseRecord falls through and does not yet add an unwrapped line as a 832 // record declaration or definition can start a structural element. 833 parseRecord(); 834 // This does not apply for Java and JavaScript. 835 if (Style.Language == FormatStyle::LK_Java || 836 Style.Language == FormatStyle::LK_JavaScript) { 837 addUnwrappedLine(); 838 return; 839 } 840 break; 841 case tok::period: 842 nextToken(); 843 // In Java, classes have an implicit static member "class". 844 if (Style.Language == FormatStyle::LK_Java && FormatTok && 845 FormatTok->is(tok::kw_class)) 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::caret: 859 nextToken(); 860 if (FormatTok->Tok.isAnyIdentifier() || 861 FormatTok->isSimpleTypeSpecifier()) 862 nextToken(); 863 if (FormatTok->is(tok::l_paren)) 864 parseParens(); 865 if (FormatTok->is(tok::l_brace)) 866 parseChildBlock(); 867 break; 868 case tok::l_brace: 869 if (!tryToParseBracedList()) { 870 // A block outside of parentheses must be the last part of a 871 // structural element. 872 // FIXME: Figure out cases where this is not true, and add projections 873 // for them (the one we know is missing are lambdas). 874 if (Style.BreakBeforeBraces != FormatStyle::BS_Attach) 875 addUnwrappedLine(); 876 FormatTok->Type = TT_FunctionLBrace; 877 parseBlock(/*MustBeDeclaration=*/false); 878 addUnwrappedLine(); 879 return; 880 } 881 // Otherwise this was a braced init list, and the structural 882 // element continues. 883 break; 884 case tok::kw_try: 885 // We arrive here when parsing function-try blocks. 886 parseTryCatch(); 887 return; 888 case tok::identifier: { 889 if (FormatTok->is(TT_MacroBlockEnd)) { 890 addUnwrappedLine(); 891 return; 892 } 893 894 // Parse function literal unless 'function' is the first token in a line 895 // in which case this should be treated as a free-standing function. 896 if (Style.Language == FormatStyle::LK_JavaScript && 897 FormatTok->is(Keywords.kw_function) && Line->Tokens.size() > 0) { 898 tryToParseJSFunction(); 899 break; 900 } 901 if ((Style.Language == FormatStyle::LK_JavaScript || 902 Style.Language == FormatStyle::LK_Java) && 903 FormatTok->is(Keywords.kw_interface)) { 904 parseRecord(); 905 addUnwrappedLine(); 906 return; 907 } 908 909 StringRef Text = FormatTok->TokenText; 910 nextToken(); 911 if (Line->Tokens.size() == 1 && 912 // JS doesn't have macros, and within classes colons indicate fields, 913 // not labels. 914 Style.Language != FormatStyle::LK_JavaScript) { 915 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) { 916 parseLabel(); 917 return; 918 } 919 // Recognize function-like macro usages without trailing semicolon as 920 // well as free-standing macros like Q_OBJECT. 921 bool FunctionLike = FormatTok->is(tok::l_paren); 922 if (FunctionLike) 923 parseParens(); 924 925 bool FollowedByNewline = 926 CommentsBeforeNextToken.empty() 927 ? FormatTok->NewlinesBefore > 0 928 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 929 930 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 931 tokenCanStartNewLine(FormatTok->Tok) && Text == Text.upper()) { 932 addUnwrappedLine(); 933 return; 934 } 935 } 936 break; 937 } 938 case tok::equal: 939 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType 940 // TT_JsFatArrow. The always start an expression or a child block if 941 // followed by a curly. 942 if (FormatTok->is(TT_JsFatArrow)) { 943 nextToken(); 944 if (FormatTok->is(tok::l_brace)) 945 parseChildBlock(); 946 break; 947 } 948 949 nextToken(); 950 if (FormatTok->Tok.is(tok::l_brace)) { 951 parseBracedList(); 952 } 953 break; 954 case tok::l_square: 955 parseSquare(); 956 break; 957 case tok::kw_new: 958 parseNew(); 959 break; 960 default: 961 nextToken(); 962 break; 963 } 964 } while (!eof()); 965 } 966 967 bool UnwrappedLineParser::tryToParseLambda() { 968 if (Style.Language != FormatStyle::LK_Cpp) { 969 nextToken(); 970 return false; 971 } 972 // FIXME: This is a dirty way to access the previous token. Find a better 973 // solution. 974 if (!Line->Tokens.empty() && 975 (Line->Tokens.back().Tok->isOneOf(tok::identifier, tok::kw_operator, 976 tok::kw_new, tok::kw_delete) || 977 Line->Tokens.back().Tok->closesScope() || 978 Line->Tokens.back().Tok->isSimpleTypeSpecifier())) { 979 nextToken(); 980 return false; 981 } 982 assert(FormatTok->is(tok::l_square)); 983 FormatToken &LSquare = *FormatTok; 984 if (!tryToParseLambdaIntroducer()) 985 return false; 986 987 while (FormatTok->isNot(tok::l_brace)) { 988 if (FormatTok->isSimpleTypeSpecifier()) { 989 nextToken(); 990 continue; 991 } 992 switch (FormatTok->Tok.getKind()) { 993 case tok::l_brace: 994 break; 995 case tok::l_paren: 996 parseParens(); 997 break; 998 case tok::amp: 999 case tok::star: 1000 case tok::kw_const: 1001 case tok::comma: 1002 case tok::less: 1003 case tok::greater: 1004 case tok::identifier: 1005 case tok::numeric_constant: 1006 case tok::coloncolon: 1007 case tok::kw_mutable: 1008 nextToken(); 1009 break; 1010 case tok::arrow: 1011 FormatTok->Type = TT_LambdaArrow; 1012 nextToken(); 1013 break; 1014 default: 1015 return true; 1016 } 1017 } 1018 LSquare.Type = TT_LambdaLSquare; 1019 parseChildBlock(); 1020 return true; 1021 } 1022 1023 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 1024 nextToken(); 1025 if (FormatTok->is(tok::equal)) { 1026 nextToken(); 1027 if (FormatTok->is(tok::r_square)) { 1028 nextToken(); 1029 return true; 1030 } 1031 if (FormatTok->isNot(tok::comma)) 1032 return false; 1033 nextToken(); 1034 } else if (FormatTok->is(tok::amp)) { 1035 nextToken(); 1036 if (FormatTok->is(tok::r_square)) { 1037 nextToken(); 1038 return true; 1039 } 1040 if (!FormatTok->isOneOf(tok::comma, tok::identifier)) { 1041 return false; 1042 } 1043 if (FormatTok->is(tok::comma)) 1044 nextToken(); 1045 } else if (FormatTok->is(tok::r_square)) { 1046 nextToken(); 1047 return true; 1048 } 1049 do { 1050 if (FormatTok->is(tok::amp)) 1051 nextToken(); 1052 if (!FormatTok->isOneOf(tok::identifier, tok::kw_this)) 1053 return false; 1054 nextToken(); 1055 if (FormatTok->is(tok::ellipsis)) 1056 nextToken(); 1057 if (FormatTok->is(tok::comma)) { 1058 nextToken(); 1059 } else if (FormatTok->is(tok::r_square)) { 1060 nextToken(); 1061 return true; 1062 } else { 1063 return false; 1064 } 1065 } while (!eof()); 1066 return false; 1067 } 1068 1069 void UnwrappedLineParser::tryToParseJSFunction() { 1070 nextToken(); 1071 1072 // Consume function name. 1073 if (FormatTok->is(tok::identifier)) 1074 nextToken(); 1075 1076 if (FormatTok->isNot(tok::l_paren)) 1077 return; 1078 1079 // Parse formal parameter list. 1080 parseParens(); 1081 1082 if (FormatTok->is(tok::colon)) { 1083 // Parse a type definition. 1084 nextToken(); 1085 1086 // Eat the type declaration. For braced inline object types, balance braces, 1087 // otherwise just parse until finding an l_brace for the function body. 1088 if (FormatTok->is(tok::l_brace)) 1089 tryToParseBracedList(); 1090 else 1091 while (FormatTok->isNot(tok::l_brace) && !eof()) 1092 nextToken(); 1093 } 1094 1095 parseChildBlock(); 1096 } 1097 1098 bool UnwrappedLineParser::tryToParseBracedList() { 1099 if (FormatTok->BlockKind == BK_Unknown) 1100 calculateBraceTypes(); 1101 assert(FormatTok->BlockKind != BK_Unknown); 1102 if (FormatTok->BlockKind == BK_Block) 1103 return false; 1104 parseBracedList(); 1105 return true; 1106 } 1107 1108 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons) { 1109 bool HasError = false; 1110 nextToken(); 1111 1112 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 1113 // replace this by using parseAssigmentExpression() inside. 1114 do { 1115 if (Style.Language == FormatStyle::LK_JavaScript) { 1116 if (FormatTok->is(Keywords.kw_function)) { 1117 tryToParseJSFunction(); 1118 continue; 1119 } 1120 if (FormatTok->is(TT_JsFatArrow)) { 1121 nextToken(); 1122 // Fat arrows can be followed by simple expressions or by child blocks 1123 // in curly braces. 1124 if (FormatTok->is(tok::l_brace)) { 1125 parseChildBlock(); 1126 continue; 1127 } 1128 } 1129 } 1130 switch (FormatTok->Tok.getKind()) { 1131 case tok::caret: 1132 nextToken(); 1133 if (FormatTok->is(tok::l_brace)) { 1134 parseChildBlock(); 1135 } 1136 break; 1137 case tok::l_square: 1138 tryToParseLambda(); 1139 break; 1140 case tok::l_brace: 1141 // Assume there are no blocks inside a braced init list apart 1142 // from the ones we explicitly parse out (like lambdas). 1143 FormatTok->BlockKind = BK_BracedInit; 1144 parseBracedList(); 1145 break; 1146 case tok::l_paren: 1147 parseParens(); 1148 // JavaScript can just have free standing methods and getters/setters in 1149 // object literals. Detect them by a "{" following ")". 1150 if (Style.Language == FormatStyle::LK_JavaScript) { 1151 if (FormatTok->is(tok::l_brace)) 1152 parseChildBlock(); 1153 break; 1154 } 1155 break; 1156 case tok::r_brace: 1157 nextToken(); 1158 return !HasError; 1159 case tok::semi: 1160 HasError = true; 1161 if (!ContinueOnSemicolons) 1162 return !HasError; 1163 nextToken(); 1164 break; 1165 case tok::comma: 1166 nextToken(); 1167 break; 1168 default: 1169 nextToken(); 1170 break; 1171 } 1172 } while (!eof()); 1173 return false; 1174 } 1175 1176 void UnwrappedLineParser::parseParens() { 1177 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 1178 nextToken(); 1179 do { 1180 switch (FormatTok->Tok.getKind()) { 1181 case tok::l_paren: 1182 parseParens(); 1183 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 1184 parseChildBlock(); 1185 break; 1186 case tok::r_paren: 1187 nextToken(); 1188 return; 1189 case tok::r_brace: 1190 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1191 return; 1192 case tok::l_square: 1193 tryToParseLambda(); 1194 break; 1195 case tok::l_brace: 1196 if (!tryToParseBracedList()) 1197 parseChildBlock(); 1198 break; 1199 case tok::at: 1200 nextToken(); 1201 if (FormatTok->Tok.is(tok::l_brace)) 1202 parseBracedList(); 1203 break; 1204 case tok::identifier: 1205 if (Style.Language == FormatStyle::LK_JavaScript && 1206 FormatTok->is(Keywords.kw_function)) 1207 tryToParseJSFunction(); 1208 else 1209 nextToken(); 1210 break; 1211 default: 1212 nextToken(); 1213 break; 1214 } 1215 } while (!eof()); 1216 } 1217 1218 void UnwrappedLineParser::parseSquare() { 1219 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected."); 1220 if (tryToParseLambda()) 1221 return; 1222 do { 1223 switch (FormatTok->Tok.getKind()) { 1224 case tok::l_paren: 1225 parseParens(); 1226 break; 1227 case tok::r_square: 1228 nextToken(); 1229 return; 1230 case tok::r_brace: 1231 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1232 return; 1233 case tok::l_square: 1234 parseSquare(); 1235 break; 1236 case tok::l_brace: { 1237 if (!tryToParseBracedList()) 1238 parseChildBlock(); 1239 break; 1240 } 1241 case tok::at: 1242 nextToken(); 1243 if (FormatTok->Tok.is(tok::l_brace)) 1244 parseBracedList(); 1245 break; 1246 default: 1247 nextToken(); 1248 break; 1249 } 1250 } while (!eof()); 1251 } 1252 1253 void UnwrappedLineParser::parseIfThenElse() { 1254 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 1255 nextToken(); 1256 if (FormatTok->Tok.is(tok::l_paren)) 1257 parseParens(); 1258 bool NeedsUnwrappedLine = false; 1259 if (FormatTok->Tok.is(tok::l_brace)) { 1260 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1261 parseBlock(/*MustBeDeclaration=*/false); 1262 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman || 1263 Style.BreakBeforeBraces == FormatStyle::BS_GNU) { 1264 addUnwrappedLine(); 1265 } else { 1266 NeedsUnwrappedLine = true; 1267 } 1268 } else { 1269 addUnwrappedLine(); 1270 ++Line->Level; 1271 parseStructuralElement(); 1272 --Line->Level; 1273 } 1274 if (FormatTok->Tok.is(tok::kw_else)) { 1275 if (Style.BreakBeforeBraces == FormatStyle::BS_Stroustrup) 1276 addUnwrappedLine(); 1277 nextToken(); 1278 if (FormatTok->Tok.is(tok::l_brace)) { 1279 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1280 parseBlock(/*MustBeDeclaration=*/false); 1281 addUnwrappedLine(); 1282 } else if (FormatTok->Tok.is(tok::kw_if)) { 1283 parseIfThenElse(); 1284 } else { 1285 addUnwrappedLine(); 1286 ++Line->Level; 1287 parseStructuralElement(); 1288 --Line->Level; 1289 } 1290 } else if (NeedsUnwrappedLine) { 1291 addUnwrappedLine(); 1292 } 1293 } 1294 1295 void UnwrappedLineParser::parseTryCatch() { 1296 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 1297 nextToken(); 1298 bool NeedsUnwrappedLine = false; 1299 if (FormatTok->is(tok::colon)) { 1300 // We are in a function try block, what comes is an initializer list. 1301 nextToken(); 1302 while (FormatTok->is(tok::identifier)) { 1303 nextToken(); 1304 if (FormatTok->is(tok::l_paren)) 1305 parseParens(); 1306 if (FormatTok->is(tok::comma)) 1307 nextToken(); 1308 } 1309 } 1310 // Parse try with resource. 1311 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) { 1312 parseParens(); 1313 } 1314 if (FormatTok->is(tok::l_brace)) { 1315 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1316 parseBlock(/*MustBeDeclaration=*/false); 1317 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman || 1318 Style.BreakBeforeBraces == FormatStyle::BS_GNU || 1319 Style.BreakBeforeBraces == FormatStyle::BS_Stroustrup) { 1320 addUnwrappedLine(); 1321 } else { 1322 NeedsUnwrappedLine = true; 1323 } 1324 } else if (!FormatTok->is(tok::kw_catch)) { 1325 // The C++ standard requires a compound-statement after a try. 1326 // If there's none, we try to assume there's a structuralElement 1327 // and try to continue. 1328 addUnwrappedLine(); 1329 ++Line->Level; 1330 parseStructuralElement(); 1331 --Line->Level; 1332 } 1333 while (1) { 1334 if (FormatTok->is(tok::at)) 1335 nextToken(); 1336 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 1337 tok::kw___finally) || 1338 ((Style.Language == FormatStyle::LK_Java || 1339 Style.Language == FormatStyle::LK_JavaScript) && 1340 FormatTok->is(Keywords.kw_finally)) || 1341 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) || 1342 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally)))) 1343 break; 1344 nextToken(); 1345 while (FormatTok->isNot(tok::l_brace)) { 1346 if (FormatTok->is(tok::l_paren)) { 1347 parseParens(); 1348 continue; 1349 } 1350 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) 1351 return; 1352 nextToken(); 1353 } 1354 NeedsUnwrappedLine = false; 1355 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1356 parseBlock(/*MustBeDeclaration=*/false); 1357 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman || 1358 Style.BreakBeforeBraces == FormatStyle::BS_GNU || 1359 Style.BreakBeforeBraces == FormatStyle::BS_Stroustrup) { 1360 addUnwrappedLine(); 1361 } else { 1362 NeedsUnwrappedLine = true; 1363 } 1364 } 1365 if (NeedsUnwrappedLine) { 1366 addUnwrappedLine(); 1367 } 1368 } 1369 1370 void UnwrappedLineParser::parseNamespace() { 1371 assert(FormatTok->Tok.is(tok::kw_namespace) && "'namespace' expected"); 1372 1373 const FormatToken &InitialToken = *FormatTok; 1374 nextToken(); 1375 if (FormatTok->Tok.is(tok::identifier)) 1376 nextToken(); 1377 if (FormatTok->Tok.is(tok::l_brace)) { 1378 if (ShouldBreakBeforeBrace(Style, InitialToken)) 1379 addUnwrappedLine(); 1380 1381 bool AddLevel = Style.NamespaceIndentation == FormatStyle::NI_All || 1382 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 1383 DeclarationScopeStack.size() > 1); 1384 parseBlock(/*MustBeDeclaration=*/true, AddLevel); 1385 // Munch the semicolon after a namespace. This is more common than one would 1386 // think. Puttin the semicolon into its own line is very ugly. 1387 if (FormatTok->Tok.is(tok::semi)) 1388 nextToken(); 1389 addUnwrappedLine(); 1390 } 1391 // FIXME: Add error handling. 1392 } 1393 1394 void UnwrappedLineParser::parseNew() { 1395 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 1396 nextToken(); 1397 if (Style.Language != FormatStyle::LK_Java) 1398 return; 1399 1400 // In Java, we can parse everything up to the parens, which aren't optional. 1401 do { 1402 // There should not be a ;, { or } before the new's open paren. 1403 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 1404 return; 1405 1406 // Consume the parens. 1407 if (FormatTok->is(tok::l_paren)) { 1408 parseParens(); 1409 1410 // If there is a class body of an anonymous class, consume that as child. 1411 if (FormatTok->is(tok::l_brace)) 1412 parseChildBlock(); 1413 return; 1414 } 1415 nextToken(); 1416 } while (!eof()); 1417 } 1418 1419 void UnwrappedLineParser::parseForOrWhileLoop() { 1420 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 1421 "'for', 'while' or foreach macro expected"); 1422 nextToken(); 1423 if (FormatTok->Tok.is(tok::l_paren)) 1424 parseParens(); 1425 if (FormatTok->Tok.is(tok::l_brace)) { 1426 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1427 parseBlock(/*MustBeDeclaration=*/false); 1428 addUnwrappedLine(); 1429 } else { 1430 addUnwrappedLine(); 1431 ++Line->Level; 1432 parseStructuralElement(); 1433 --Line->Level; 1434 } 1435 } 1436 1437 void UnwrappedLineParser::parseDoWhile() { 1438 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 1439 nextToken(); 1440 if (FormatTok->Tok.is(tok::l_brace)) { 1441 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1442 parseBlock(/*MustBeDeclaration=*/false); 1443 if (Style.BreakBeforeBraces == FormatStyle::BS_GNU) 1444 addUnwrappedLine(); 1445 } else { 1446 addUnwrappedLine(); 1447 ++Line->Level; 1448 parseStructuralElement(); 1449 --Line->Level; 1450 } 1451 1452 // FIXME: Add error handling. 1453 if (!FormatTok->Tok.is(tok::kw_while)) { 1454 addUnwrappedLine(); 1455 return; 1456 } 1457 1458 nextToken(); 1459 parseStructuralElement(); 1460 } 1461 1462 void UnwrappedLineParser::parseLabel() { 1463 nextToken(); 1464 unsigned OldLineLevel = Line->Level; 1465 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 1466 --Line->Level; 1467 if (CommentsBeforeNextToken.empty() && FormatTok->Tok.is(tok::l_brace)) { 1468 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1469 parseBlock(/*MustBeDeclaration=*/false); 1470 if (FormatTok->Tok.is(tok::kw_break)) { 1471 // "break;" after "}" on its own line only for BS_Allman and BS_GNU 1472 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman || 1473 Style.BreakBeforeBraces == FormatStyle::BS_GNU) { 1474 addUnwrappedLine(); 1475 } 1476 parseStructuralElement(); 1477 } 1478 addUnwrappedLine(); 1479 } else { 1480 if (FormatTok->is(tok::semi)) 1481 nextToken(); 1482 addUnwrappedLine(); 1483 } 1484 Line->Level = OldLineLevel; 1485 } 1486 1487 void UnwrappedLineParser::parseCaseLabel() { 1488 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 1489 // FIXME: fix handling of complex expressions here. 1490 do { 1491 nextToken(); 1492 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 1493 parseLabel(); 1494 } 1495 1496 void UnwrappedLineParser::parseSwitch() { 1497 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 1498 nextToken(); 1499 if (FormatTok->Tok.is(tok::l_paren)) 1500 parseParens(); 1501 if (FormatTok->Tok.is(tok::l_brace)) { 1502 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1503 parseBlock(/*MustBeDeclaration=*/false); 1504 addUnwrappedLine(); 1505 } else { 1506 addUnwrappedLine(); 1507 ++Line->Level; 1508 parseStructuralElement(); 1509 --Line->Level; 1510 } 1511 } 1512 1513 void UnwrappedLineParser::parseAccessSpecifier() { 1514 nextToken(); 1515 // Understand Qt's slots. 1516 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 1517 nextToken(); 1518 // Otherwise, we don't know what it is, and we'd better keep the next token. 1519 if (FormatTok->Tok.is(tok::colon)) 1520 nextToken(); 1521 addUnwrappedLine(); 1522 } 1523 1524 void UnwrappedLineParser::parseEnum() { 1525 // Won't be 'enum' for NS_ENUMs. 1526 if (FormatTok->Tok.is(tok::kw_enum)) 1527 nextToken(); 1528 1529 // Eat up enum class ... 1530 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct)) 1531 nextToken(); 1532 1533 while (FormatTok->Tok.getIdentifierInfo() || 1534 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 1535 tok::greater, tok::comma, tok::question)) { 1536 nextToken(); 1537 // We can have macros or attributes in between 'enum' and the enum name. 1538 if (FormatTok->is(tok::l_paren)) 1539 parseParens(); 1540 if (FormatTok->is(tok::identifier)) { 1541 nextToken(); 1542 // If there are two identifiers in a row, this is likely an elaborate 1543 // return type. In Java, this can be "implements", etc. 1544 if (Style.Language == FormatStyle::LK_Cpp && 1545 FormatTok->is(tok::identifier)) 1546 return; 1547 } 1548 } 1549 1550 // Just a declaration or something is wrong. 1551 if (FormatTok->isNot(tok::l_brace)) 1552 return; 1553 FormatTok->BlockKind = BK_Block; 1554 1555 if (Style.Language == FormatStyle::LK_Java) { 1556 // Java enums are different. 1557 parseJavaEnumBody(); 1558 return; 1559 } else if (Style.Language == FormatStyle::LK_Proto) { 1560 parseBlock(/*MustBeDeclaration=*/true); 1561 return; 1562 } 1563 1564 // Parse enum body. 1565 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true); 1566 if (HasError) { 1567 if (FormatTok->is(tok::semi)) 1568 nextToken(); 1569 addUnwrappedLine(); 1570 } 1571 1572 // There is no addUnwrappedLine() here so that we fall through to parsing a 1573 // structural element afterwards. Thus, in "enum A {} n, m;", 1574 // "} n, m;" will end up in one unwrapped line. 1575 } 1576 1577 void UnwrappedLineParser::parseJavaEnumBody() { 1578 // Determine whether the enum is simple, i.e. does not have a semicolon or 1579 // constants with class bodies. Simple enums can be formatted like braced 1580 // lists, contracted to a single line, etc. 1581 unsigned StoredPosition = Tokens->getPosition(); 1582 bool IsSimple = true; 1583 FormatToken *Tok = Tokens->getNextToken(); 1584 while (Tok) { 1585 if (Tok->is(tok::r_brace)) 1586 break; 1587 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 1588 IsSimple = false; 1589 break; 1590 } 1591 // FIXME: This will also mark enums with braces in the arguments to enum 1592 // constants as "not simple". This is probably fine in practice, though. 1593 Tok = Tokens->getNextToken(); 1594 } 1595 FormatTok = Tokens->setPosition(StoredPosition); 1596 1597 if (IsSimple) { 1598 parseBracedList(); 1599 addUnwrappedLine(); 1600 return; 1601 } 1602 1603 // Parse the body of a more complex enum. 1604 // First add a line for everything up to the "{". 1605 nextToken(); 1606 addUnwrappedLine(); 1607 ++Line->Level; 1608 1609 // Parse the enum constants. 1610 while (FormatTok) { 1611 if (FormatTok->is(tok::l_brace)) { 1612 // Parse the constant's class body. 1613 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 1614 /*MunchSemi=*/false); 1615 } else if (FormatTok->is(tok::l_paren)) { 1616 parseParens(); 1617 } else if (FormatTok->is(tok::comma)) { 1618 nextToken(); 1619 addUnwrappedLine(); 1620 } else if (FormatTok->is(tok::semi)) { 1621 nextToken(); 1622 addUnwrappedLine(); 1623 break; 1624 } else if (FormatTok->is(tok::r_brace)) { 1625 addUnwrappedLine(); 1626 break; 1627 } else { 1628 nextToken(); 1629 } 1630 } 1631 1632 // Parse the class body after the enum's ";" if any. 1633 parseLevel(/*HasOpeningBrace=*/true); 1634 nextToken(); 1635 --Line->Level; 1636 addUnwrappedLine(); 1637 } 1638 1639 void UnwrappedLineParser::parseRecord() { 1640 const FormatToken &InitialToken = *FormatTok; 1641 nextToken(); 1642 1643 // The actual identifier can be a nested name specifier, and in macros 1644 // it is often token-pasted. 1645 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 1646 tok::kw___attribute, tok::kw___declspec, 1647 tok::kw_alignas) || 1648 ((Style.Language == FormatStyle::LK_Java || 1649 Style.Language == FormatStyle::LK_JavaScript) && 1650 FormatTok->isOneOf(tok::period, tok::comma))) { 1651 bool IsNonMacroIdentifier = 1652 FormatTok->is(tok::identifier) && 1653 FormatTok->TokenText != FormatTok->TokenText.upper(); 1654 nextToken(); 1655 // We can have macros or attributes in between 'class' and the class name. 1656 if (!IsNonMacroIdentifier && FormatTok->Tok.is(tok::l_paren)) 1657 parseParens(); 1658 } 1659 1660 // Note that parsing away template declarations here leads to incorrectly 1661 // accepting function declarations as record declarations. 1662 // In general, we cannot solve this problem. Consider: 1663 // class A<int> B() {} 1664 // which can be a function definition or a class definition when B() is a 1665 // macro. If we find enough real-world cases where this is a problem, we 1666 // can parse for the 'template' keyword in the beginning of the statement, 1667 // and thus rule out the record production in case there is no template 1668 // (this would still leave us with an ambiguity between template function 1669 // and class declarations). 1670 if (FormatTok->isOneOf(tok::colon, tok::less)) { 1671 while (!eof()) { 1672 if (FormatTok->is(tok::l_brace)) { 1673 calculateBraceTypes(/*ExpectClassBody=*/true); 1674 if (!tryToParseBracedList()) 1675 break; 1676 } 1677 if (FormatTok->Tok.is(tok::semi)) 1678 return; 1679 nextToken(); 1680 } 1681 } 1682 if (FormatTok->Tok.is(tok::l_brace)) { 1683 if (ShouldBreakBeforeBrace(Style, InitialToken)) 1684 addUnwrappedLine(); 1685 1686 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 1687 /*MunchSemi=*/false); 1688 } 1689 // There is no addUnwrappedLine() here so that we fall through to parsing a 1690 // structural element afterwards. Thus, in "class A {} n, m;", 1691 // "} n, m;" will end up in one unwrapped line. 1692 } 1693 1694 void UnwrappedLineParser::parseObjCProtocolList() { 1695 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 1696 do 1697 nextToken(); 1698 while (!eof() && FormatTok->Tok.isNot(tok::greater)); 1699 nextToken(); // Skip '>'. 1700 } 1701 1702 void UnwrappedLineParser::parseObjCUntilAtEnd() { 1703 do { 1704 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 1705 nextToken(); 1706 addUnwrappedLine(); 1707 break; 1708 } 1709 if (FormatTok->is(tok::l_brace)) { 1710 parseBlock(/*MustBeDeclaration=*/false); 1711 // In ObjC interfaces, nothing should be following the "}". 1712 addUnwrappedLine(); 1713 } else if (FormatTok->is(tok::r_brace)) { 1714 // Ignore stray "}". parseStructuralElement doesn't consume them. 1715 nextToken(); 1716 addUnwrappedLine(); 1717 } else { 1718 parseStructuralElement(); 1719 } 1720 } while (!eof()); 1721 } 1722 1723 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 1724 nextToken(); 1725 nextToken(); // interface name 1726 1727 // @interface can be followed by either a base class, or a category. 1728 if (FormatTok->Tok.is(tok::colon)) { 1729 nextToken(); 1730 nextToken(); // base class name 1731 } else if (FormatTok->Tok.is(tok::l_paren)) 1732 // Skip category, if present. 1733 parseParens(); 1734 1735 if (FormatTok->Tok.is(tok::less)) 1736 parseObjCProtocolList(); 1737 1738 if (FormatTok->Tok.is(tok::l_brace)) { 1739 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman || 1740 Style.BreakBeforeBraces == FormatStyle::BS_GNU) 1741 addUnwrappedLine(); 1742 parseBlock(/*MustBeDeclaration=*/true); 1743 } 1744 1745 // With instance variables, this puts '}' on its own line. Without instance 1746 // variables, this ends the @interface line. 1747 addUnwrappedLine(); 1748 1749 parseObjCUntilAtEnd(); 1750 } 1751 1752 void UnwrappedLineParser::parseObjCProtocol() { 1753 nextToken(); 1754 nextToken(); // protocol name 1755 1756 if (FormatTok->Tok.is(tok::less)) 1757 parseObjCProtocolList(); 1758 1759 // Check for protocol declaration. 1760 if (FormatTok->Tok.is(tok::semi)) { 1761 nextToken(); 1762 return addUnwrappedLine(); 1763 } 1764 1765 addUnwrappedLine(); 1766 parseObjCUntilAtEnd(); 1767 } 1768 1769 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 1770 assert(FormatTok->isOneOf(Keywords.kw_import, tok::kw_export)); 1771 nextToken(); 1772 1773 // Consume the "default" in "export default class/function". 1774 if (FormatTok->is(tok::kw_default)) 1775 nextToken(); 1776 1777 // Consume "function" and "default function", so that these get parsed as 1778 // free-standing JS functions, i.e. do not require a trailing semicolon. 1779 if (FormatTok->is(Keywords.kw_function)) { 1780 nextToken(); 1781 return; 1782 } 1783 1784 if (FormatTok->isOneOf(tok::kw_const, tok::kw_class, tok::kw_enum, 1785 Keywords.kw_var)) 1786 return; // Fall through to parsing the corresponding structure. 1787 1788 if (FormatTok->is(tok::l_brace)) { 1789 FormatTok->BlockKind = BK_Block; 1790 parseBracedList(); 1791 } 1792 1793 while (!eof() && FormatTok->isNot(tok::semi) && 1794 FormatTok->isNot(tok::l_brace)) { 1795 nextToken(); 1796 } 1797 } 1798 1799 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 1800 StringRef Prefix = "") { 1801 llvm::dbgs() << Prefix << "Line(" << Line.Level << ")" 1802 << (Line.InPPDirective ? " MACRO" : "") << ": "; 1803 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 1804 E = Line.Tokens.end(); 1805 I != E; ++I) { 1806 llvm::dbgs() << I->Tok->Tok.getName() << "[" << I->Tok->Type << "] "; 1807 } 1808 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 1809 E = Line.Tokens.end(); 1810 I != E; ++I) { 1811 const UnwrappedLineNode &Node = *I; 1812 for (SmallVectorImpl<UnwrappedLine>::const_iterator 1813 I = Node.Children.begin(), 1814 E = Node.Children.end(); 1815 I != E; ++I) { 1816 printDebugInfo(*I, "\nChild: "); 1817 } 1818 } 1819 llvm::dbgs() << "\n"; 1820 } 1821 1822 void UnwrappedLineParser::addUnwrappedLine() { 1823 if (Line->Tokens.empty()) 1824 return; 1825 DEBUG({ 1826 if (CurrentLines == &Lines) 1827 printDebugInfo(*Line); 1828 }); 1829 CurrentLines->push_back(std::move(*Line)); 1830 Line->Tokens.clear(); 1831 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 1832 CurrentLines->append( 1833 std::make_move_iterator(PreprocessorDirectives.begin()), 1834 std::make_move_iterator(PreprocessorDirectives.end())); 1835 PreprocessorDirectives.clear(); 1836 } 1837 } 1838 1839 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 1840 1841 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 1842 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 1843 FormatTok.NewlinesBefore > 0; 1844 } 1845 1846 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 1847 bool JustComments = Line->Tokens.empty(); 1848 for (SmallVectorImpl<FormatToken *>::const_iterator 1849 I = CommentsBeforeNextToken.begin(), 1850 E = CommentsBeforeNextToken.end(); 1851 I != E; ++I) { 1852 if (isOnNewLine(**I) && JustComments) 1853 addUnwrappedLine(); 1854 pushToken(*I); 1855 } 1856 if (NewlineBeforeNext && JustComments) 1857 addUnwrappedLine(); 1858 CommentsBeforeNextToken.clear(); 1859 } 1860 1861 void UnwrappedLineParser::nextToken() { 1862 if (eof()) 1863 return; 1864 flushComments(isOnNewLine(*FormatTok)); 1865 pushToken(FormatTok); 1866 readToken(); 1867 } 1868 1869 void UnwrappedLineParser::readToken() { 1870 bool CommentsInCurrentLine = true; 1871 do { 1872 FormatTok = Tokens->getNextToken(); 1873 assert(FormatTok); 1874 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 1875 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 1876 // If there is an unfinished unwrapped line, we flush the preprocessor 1877 // directives only after that unwrapped line was finished later. 1878 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 1879 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 1880 // Comments stored before the preprocessor directive need to be output 1881 // before the preprocessor directive, at the same level as the 1882 // preprocessor directive, as we consider them to apply to the directive. 1883 flushComments(isOnNewLine(*FormatTok)); 1884 parsePPDirective(); 1885 } 1886 while (FormatTok->Type == TT_ConflictStart || 1887 FormatTok->Type == TT_ConflictEnd || 1888 FormatTok->Type == TT_ConflictAlternative) { 1889 if (FormatTok->Type == TT_ConflictStart) { 1890 conditionalCompilationStart(/*Unreachable=*/false); 1891 } else if (FormatTok->Type == TT_ConflictAlternative) { 1892 conditionalCompilationAlternative(); 1893 } else if (FormatTok->Type == TT_ConflictEnd) { 1894 conditionalCompilationEnd(); 1895 } 1896 FormatTok = Tokens->getNextToken(); 1897 FormatTok->MustBreakBefore = true; 1898 } 1899 1900 if (!PPStack.empty() && (PPStack.back() == PP_Unreachable) && 1901 !Line->InPPDirective) { 1902 continue; 1903 } 1904 1905 if (!FormatTok->Tok.is(tok::comment)) 1906 return; 1907 if (isOnNewLine(*FormatTok) || FormatTok->IsFirst) { 1908 CommentsInCurrentLine = false; 1909 } 1910 if (CommentsInCurrentLine) { 1911 pushToken(FormatTok); 1912 } else { 1913 CommentsBeforeNextToken.push_back(FormatTok); 1914 } 1915 } while (!eof()); 1916 } 1917 1918 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 1919 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 1920 if (MustBreakBeforeNextToken) { 1921 Line->Tokens.back().Tok->MustBreakBefore = true; 1922 MustBreakBeforeNextToken = false; 1923 } 1924 } 1925 1926 } // end namespace format 1927 } // end namespace clang 1928