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 const FormatToken *PrevTok = getPreviousToken(); 319 // Keep a stack of positions of lbrace tokens. We will 320 // update information about whether an lbrace starts a 321 // braced init list or a different block during the loop. 322 SmallVector<FormatToken *, 8> LBraceStack; 323 assert(Tok->Tok.is(tok::l_brace)); 324 do { 325 // Get next non-comment token. 326 FormatToken *NextTok; 327 unsigned ReadTokens = 0; 328 do { 329 NextTok = Tokens->getNextToken(); 330 ++ReadTokens; 331 } while (NextTok->is(tok::comment)); 332 333 switch (Tok->Tok.getKind()) { 334 case tok::l_brace: 335 if (Style.Language == FormatStyle::LK_JavaScript && PrevTok && 336 PrevTok->is(tok::colon)) 337 // In TypeScript's TypeMemberLists, there can be semicolons between the 338 // individual members. 339 Tok->BlockKind = BK_BracedInit; 340 else 341 Tok->BlockKind = BK_Unknown; 342 LBraceStack.push_back(Tok); 343 break; 344 case tok::r_brace: 345 if (LBraceStack.empty()) 346 break; 347 if (LBraceStack.back()->BlockKind == BK_Unknown) { 348 bool ProbablyBracedList = false; 349 if (Style.Language == FormatStyle::LK_Proto) { 350 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square); 351 } else { 352 // Using OriginalColumn to distinguish between ObjC methods and 353 // binary operators is a bit hacky. 354 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 355 NextTok->OriginalColumn == 0; 356 357 // If there is a comma, semicolon or right paren after the closing 358 // brace, we assume this is a braced initializer list. Note that 359 // regardless how we mark inner braces here, we will overwrite the 360 // BlockKind later if we parse a braced list (where all blocks 361 // inside are by default braced lists), or when we explicitly detect 362 // blocks (for example while parsing lambdas). 363 // 364 // We exclude + and - as they can be ObjC visibility modifiers. 365 ProbablyBracedList = 366 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 367 tok::r_paren, tok::r_square, tok::l_brace, 368 tok::l_square, tok::l_paren, tok::ellipsis) || 369 (NextTok->is(tok::semi) && 370 (!ExpectClassBody || LBraceStack.size() != 1)) || 371 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 372 } 373 if (ProbablyBracedList) { 374 Tok->BlockKind = BK_BracedInit; 375 LBraceStack.back()->BlockKind = BK_BracedInit; 376 } else { 377 Tok->BlockKind = BK_Block; 378 LBraceStack.back()->BlockKind = BK_Block; 379 } 380 } 381 LBraceStack.pop_back(); 382 break; 383 case tok::at: 384 case tok::semi: 385 case tok::kw_if: 386 case tok::kw_while: 387 case tok::kw_for: 388 case tok::kw_switch: 389 case tok::kw_try: 390 case tok::kw___try: 391 if (!LBraceStack.empty() && LBraceStack.back()->BlockKind == BK_Unknown) 392 LBraceStack.back()->BlockKind = BK_Block; 393 break; 394 default: 395 break; 396 } 397 PrevTok = Tok; 398 Tok = NextTok; 399 } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty()); 400 401 // Assume other blocks for all unclosed opening braces. 402 for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) { 403 if (LBraceStack[i]->BlockKind == BK_Unknown) 404 LBraceStack[i]->BlockKind = BK_Block; 405 } 406 407 FormatTok = Tokens->setPosition(StoredPosition); 408 } 409 410 void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, bool AddLevel, 411 bool MunchSemi) { 412 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) && 413 "'{' or macro block token expected"); 414 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 415 FormatTok->BlockKind = BK_Block; 416 417 unsigned InitialLevel = Line->Level; 418 nextToken(); 419 420 if (MacroBlock && FormatTok->is(tok::l_paren)) 421 parseParens(); 422 423 addUnwrappedLine(); 424 425 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 426 MustBeDeclaration); 427 if (AddLevel) 428 ++Line->Level; 429 parseLevel(/*HasOpeningBrace=*/true); 430 431 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd) 432 : !FormatTok->is(tok::r_brace)) { 433 Line->Level = InitialLevel; 434 FormatTok->BlockKind = BK_Block; 435 return; 436 } 437 438 nextToken(); // Munch the closing brace. 439 440 if (MacroBlock && FormatTok->is(tok::l_paren)) 441 parseParens(); 442 443 if (MunchSemi && FormatTok->Tok.is(tok::semi)) 444 nextToken(); 445 Line->Level = InitialLevel; 446 } 447 448 static bool isGoogScope(const UnwrappedLine &Line) { 449 // FIXME: Closure-library specific stuff should not be hard-coded but be 450 // configurable. 451 if (Line.Tokens.size() < 4) 452 return false; 453 auto I = Line.Tokens.begin(); 454 if (I->Tok->TokenText != "goog") 455 return false; 456 ++I; 457 if (I->Tok->isNot(tok::period)) 458 return false; 459 ++I; 460 if (I->Tok->TokenText != "scope") 461 return false; 462 ++I; 463 return I->Tok->is(tok::l_paren); 464 } 465 466 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 467 const FormatToken &InitialToken) { 468 if (InitialToken.is(tok::kw_namespace)) 469 return Style.BraceWrapping.AfterNamespace; 470 if (InitialToken.is(tok::kw_class)) 471 return Style.BraceWrapping.AfterClass; 472 if (InitialToken.is(tok::kw_union)) 473 return Style.BraceWrapping.AfterUnion; 474 if (InitialToken.is(tok::kw_struct)) 475 return Style.BraceWrapping.AfterStruct; 476 return false; 477 } 478 479 void UnwrappedLineParser::parseChildBlock() { 480 FormatTok->BlockKind = BK_Block; 481 nextToken(); 482 { 483 bool GoogScope = 484 Style.Language == FormatStyle::LK_JavaScript && isGoogScope(*Line); 485 ScopedLineState LineState(*this); 486 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 487 /*MustBeDeclaration=*/false); 488 Line->Level += GoogScope ? 0 : 1; 489 parseLevel(/*HasOpeningBrace=*/true); 490 flushComments(isOnNewLine(*FormatTok)); 491 Line->Level -= GoogScope ? 0 : 1; 492 } 493 nextToken(); 494 } 495 496 void UnwrappedLineParser::parsePPDirective() { 497 assert(FormatTok->Tok.is(tok::hash) && "'#' expected"); 498 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 499 nextToken(); 500 501 if (!FormatTok->Tok.getIdentifierInfo()) { 502 parsePPUnknown(); 503 return; 504 } 505 506 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 507 case tok::pp_define: 508 parsePPDefine(); 509 return; 510 case tok::pp_if: 511 parsePPIf(/*IfDef=*/false); 512 break; 513 case tok::pp_ifdef: 514 case tok::pp_ifndef: 515 parsePPIf(/*IfDef=*/true); 516 break; 517 case tok::pp_else: 518 parsePPElse(); 519 break; 520 case tok::pp_elif: 521 parsePPElIf(); 522 break; 523 case tok::pp_endif: 524 parsePPEndIf(); 525 break; 526 default: 527 parsePPUnknown(); 528 break; 529 } 530 } 531 532 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 533 if (Unreachable || (!PPStack.empty() && PPStack.back() == PP_Unreachable)) 534 PPStack.push_back(PP_Unreachable); 535 else 536 PPStack.push_back(PP_Conditional); 537 } 538 539 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 540 ++PPBranchLevel; 541 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 542 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 543 PPLevelBranchIndex.push_back(0); 544 PPLevelBranchCount.push_back(0); 545 } 546 PPChainBranchIndex.push(0); 547 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 548 conditionalCompilationCondition(Unreachable || Skip); 549 } 550 551 void UnwrappedLineParser::conditionalCompilationAlternative() { 552 if (!PPStack.empty()) 553 PPStack.pop_back(); 554 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 555 if (!PPChainBranchIndex.empty()) 556 ++PPChainBranchIndex.top(); 557 conditionalCompilationCondition( 558 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 559 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 560 } 561 562 void UnwrappedLineParser::conditionalCompilationEnd() { 563 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 564 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 565 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) { 566 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 567 } 568 } 569 // Guard against #endif's without #if. 570 if (PPBranchLevel > 0) 571 --PPBranchLevel; 572 if (!PPChainBranchIndex.empty()) 573 PPChainBranchIndex.pop(); 574 if (!PPStack.empty()) 575 PPStack.pop_back(); 576 } 577 578 void UnwrappedLineParser::parsePPIf(bool IfDef) { 579 nextToken(); 580 bool IsLiteralFalse = (FormatTok->Tok.isLiteral() && 581 FormatTok->Tok.getLiteralData() != nullptr && 582 StringRef(FormatTok->Tok.getLiteralData(), 583 FormatTok->Tok.getLength()) == "0") || 584 FormatTok->Tok.is(tok::kw_false); 585 conditionalCompilationStart(!IfDef && IsLiteralFalse); 586 parsePPUnknown(); 587 } 588 589 void UnwrappedLineParser::parsePPElse() { 590 conditionalCompilationAlternative(); 591 parsePPUnknown(); 592 } 593 594 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 595 596 void UnwrappedLineParser::parsePPEndIf() { 597 conditionalCompilationEnd(); 598 parsePPUnknown(); 599 } 600 601 void UnwrappedLineParser::parsePPDefine() { 602 nextToken(); 603 604 if (FormatTok->Tok.getKind() != tok::identifier) { 605 parsePPUnknown(); 606 return; 607 } 608 nextToken(); 609 if (FormatTok->Tok.getKind() == tok::l_paren && 610 FormatTok->WhitespaceRange.getBegin() == 611 FormatTok->WhitespaceRange.getEnd()) { 612 parseParens(); 613 } 614 addUnwrappedLine(); 615 Line->Level = 1; 616 617 // Errors during a preprocessor directive can only affect the layout of the 618 // preprocessor directive, and thus we ignore them. An alternative approach 619 // would be to use the same approach we use on the file level (no 620 // re-indentation if there was a structural error) within the macro 621 // definition. 622 parseFile(); 623 } 624 625 void UnwrappedLineParser::parsePPUnknown() { 626 do { 627 nextToken(); 628 } while (!eof()); 629 addUnwrappedLine(); 630 } 631 632 // Here we blacklist certain tokens that are not usually the first token in an 633 // unwrapped line. This is used in attempt to distinguish macro calls without 634 // trailing semicolons from other constructs split to several lines. 635 static bool tokenCanStartNewLine(const clang::Token &Tok) { 636 // Semicolon can be a null-statement, l_square can be a start of a macro or 637 // a C++11 attribute, but this doesn't seem to be common. 638 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 639 Tok.isNot(tok::l_square) && 640 // Tokens that can only be used as binary operators and a part of 641 // overloaded operator names. 642 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 643 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 644 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 645 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 646 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 647 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 648 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 649 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 650 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 651 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 652 Tok.isNot(tok::lesslessequal) && 653 // Colon is used in labels, base class lists, initializer lists, 654 // range-based for loops, ternary operator, but should never be the 655 // first token in an unwrapped line. 656 Tok.isNot(tok::colon) && 657 // 'noexcept' is a trailing annotation. 658 Tok.isNot(tok::kw_noexcept); 659 } 660 661 void UnwrappedLineParser::parseStructuralElement() { 662 assert(!FormatTok->is(tok::l_brace)); 663 if (Style.Language == FormatStyle::LK_TableGen && 664 FormatTok->is(tok::pp_include)) { 665 nextToken(); 666 if (FormatTok->is(tok::string_literal)) 667 nextToken(); 668 addUnwrappedLine(); 669 return; 670 } 671 switch (FormatTok->Tok.getKind()) { 672 case tok::at: 673 nextToken(); 674 if (FormatTok->Tok.is(tok::l_brace)) { 675 parseBracedList(); 676 break; 677 } 678 switch (FormatTok->Tok.getObjCKeywordID()) { 679 case tok::objc_public: 680 case tok::objc_protected: 681 case tok::objc_package: 682 case tok::objc_private: 683 return parseAccessSpecifier(); 684 case tok::objc_interface: 685 case tok::objc_implementation: 686 return parseObjCInterfaceOrImplementation(); 687 case tok::objc_protocol: 688 return parseObjCProtocol(); 689 case tok::objc_end: 690 return; // Handled by the caller. 691 case tok::objc_optional: 692 case tok::objc_required: 693 nextToken(); 694 addUnwrappedLine(); 695 return; 696 case tok::objc_autoreleasepool: 697 nextToken(); 698 if (FormatTok->Tok.is(tok::l_brace)) { 699 if (Style.BraceWrapping.AfterObjCDeclaration) 700 addUnwrappedLine(); 701 parseBlock(/*MustBeDeclaration=*/false); 702 } 703 addUnwrappedLine(); 704 return; 705 case tok::objc_try: 706 // This branch isn't strictly necessary (the kw_try case below would 707 // do this too after the tok::at is parsed above). But be explicit. 708 parseTryCatch(); 709 return; 710 default: 711 break; 712 } 713 break; 714 case tok::kw_asm: 715 nextToken(); 716 if (FormatTok->is(tok::l_brace)) { 717 FormatTok->Type = TT_InlineASMBrace; 718 nextToken(); 719 while (FormatTok && FormatTok->isNot(tok::eof)) { 720 if (FormatTok->is(tok::r_brace)) { 721 FormatTok->Type = TT_InlineASMBrace; 722 nextToken(); 723 addUnwrappedLine(); 724 break; 725 } 726 FormatTok->Finalized = true; 727 nextToken(); 728 } 729 } 730 break; 731 case tok::kw_namespace: 732 parseNamespace(); 733 return; 734 case tok::kw_inline: 735 nextToken(); 736 if (FormatTok->Tok.is(tok::kw_namespace)) { 737 parseNamespace(); 738 return; 739 } 740 break; 741 case tok::kw_public: 742 case tok::kw_protected: 743 case tok::kw_private: 744 if (Style.Language == FormatStyle::LK_Java || 745 Style.Language == FormatStyle::LK_JavaScript) 746 nextToken(); 747 else 748 parseAccessSpecifier(); 749 return; 750 case tok::kw_if: 751 parseIfThenElse(); 752 return; 753 case tok::kw_for: 754 case tok::kw_while: 755 parseForOrWhileLoop(); 756 return; 757 case tok::kw_do: 758 parseDoWhile(); 759 return; 760 case tok::kw_switch: 761 parseSwitch(); 762 return; 763 case tok::kw_default: 764 nextToken(); 765 parseLabel(); 766 return; 767 case tok::kw_case: 768 parseCaseLabel(); 769 return; 770 case tok::kw_try: 771 case tok::kw___try: 772 parseTryCatch(); 773 return; 774 case tok::kw_extern: 775 nextToken(); 776 if (FormatTok->Tok.is(tok::string_literal)) { 777 nextToken(); 778 if (FormatTok->Tok.is(tok::l_brace)) { 779 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/false); 780 addUnwrappedLine(); 781 return; 782 } 783 } 784 break; 785 case tok::kw_export: 786 if (Style.Language == FormatStyle::LK_JavaScript) { 787 parseJavaScriptEs6ImportExport(); 788 return; 789 } 790 break; 791 case tok::identifier: 792 if (FormatTok->is(TT_ForEachMacro)) { 793 parseForOrWhileLoop(); 794 return; 795 } 796 if (FormatTok->is(TT_MacroBlockBegin)) { 797 parseBlock(/*MustBeDeclaration=*/false, /*AddLevel=*/true, 798 /*MunchSemi=*/false); 799 return; 800 } 801 if (Style.Language == FormatStyle::LK_JavaScript && 802 FormatTok->is(Keywords.kw_import)) { 803 parseJavaScriptEs6ImportExport(); 804 return; 805 } 806 if (FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 807 Keywords.kw_slots, Keywords.kw_qslots)) { 808 nextToken(); 809 if (FormatTok->is(tok::colon)) { 810 nextToken(); 811 addUnwrappedLine(); 812 } 813 return; 814 } 815 // In all other cases, parse the declaration. 816 break; 817 default: 818 break; 819 } 820 do { 821 switch (FormatTok->Tok.getKind()) { 822 case tok::at: 823 nextToken(); 824 if (FormatTok->Tok.is(tok::l_brace)) 825 parseBracedList(); 826 break; 827 case tok::kw_enum: 828 // parseEnum falls through and does not yet add an unwrapped line as an 829 // enum definition can start a structural element. 830 if (!parseEnum()) 831 break; 832 // This only applies for C++. 833 if (Style.Language != FormatStyle::LK_Cpp) { 834 addUnwrappedLine(); 835 return; 836 } 837 break; 838 case tok::kw_typedef: 839 nextToken(); 840 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 841 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS)) 842 parseEnum(); 843 break; 844 case tok::kw_struct: 845 case tok::kw_union: 846 case tok::kw_class: 847 // parseRecord falls through and does not yet add an unwrapped line as a 848 // record declaration or definition can start a structural element. 849 parseRecord(); 850 // This does not apply for Java and JavaScript. 851 if (Style.Language == FormatStyle::LK_Java || 852 Style.Language == FormatStyle::LK_JavaScript) { 853 if (FormatTok->is(tok::semi)) 854 nextToken(); 855 addUnwrappedLine(); 856 return; 857 } 858 break; 859 case tok::period: 860 nextToken(); 861 // In Java, classes have an implicit static member "class". 862 if (Style.Language == FormatStyle::LK_Java && FormatTok && 863 FormatTok->is(tok::kw_class)) 864 nextToken(); 865 if (Style.Language == FormatStyle::LK_JavaScript && FormatTok && 866 FormatTok->Tok.getIdentifierInfo()) 867 // JavaScript only has pseudo keywords, all keywords are allowed to 868 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 869 nextToken(); 870 break; 871 case tok::semi: 872 nextToken(); 873 addUnwrappedLine(); 874 return; 875 case tok::r_brace: 876 addUnwrappedLine(); 877 return; 878 case tok::l_paren: 879 parseParens(); 880 break; 881 case tok::kw_operator: 882 nextToken(); 883 if (FormatTok->isBinaryOperator()) 884 nextToken(); 885 break; 886 case tok::caret: 887 nextToken(); 888 if (FormatTok->Tok.isAnyIdentifier() || 889 FormatTok->isSimpleTypeSpecifier()) 890 nextToken(); 891 if (FormatTok->is(tok::l_paren)) 892 parseParens(); 893 if (FormatTok->is(tok::l_brace)) 894 parseChildBlock(); 895 break; 896 case tok::l_brace: 897 if (!tryToParseBracedList()) { 898 // A block outside of parentheses must be the last part of a 899 // structural element. 900 // FIXME: Figure out cases where this is not true, and add projections 901 // for them (the one we know is missing are lambdas). 902 if (Style.BraceWrapping.AfterFunction) 903 addUnwrappedLine(); 904 FormatTok->Type = TT_FunctionLBrace; 905 parseBlock(/*MustBeDeclaration=*/false); 906 addUnwrappedLine(); 907 return; 908 } 909 // Otherwise this was a braced init list, and the structural 910 // element continues. 911 break; 912 case tok::kw_try: 913 // We arrive here when parsing function-try blocks. 914 parseTryCatch(); 915 return; 916 case tok::identifier: { 917 if (FormatTok->is(TT_MacroBlockEnd)) { 918 addUnwrappedLine(); 919 return; 920 } 921 922 // Parse function literal unless 'function' is the first token in a line 923 // in which case this should be treated as a free-standing function. 924 if (Style.Language == FormatStyle::LK_JavaScript && 925 FormatTok->is(Keywords.kw_function) && Line->Tokens.size() > 0) { 926 tryToParseJSFunction(); 927 break; 928 } 929 if ((Style.Language == FormatStyle::LK_JavaScript || 930 Style.Language == FormatStyle::LK_Java) && 931 FormatTok->is(Keywords.kw_interface)) { 932 parseRecord(); 933 addUnwrappedLine(); 934 return; 935 } 936 937 StringRef Text = FormatTok->TokenText; 938 nextToken(); 939 if (Line->Tokens.size() == 1 && 940 // JS doesn't have macros, and within classes colons indicate fields, 941 // not labels. 942 Style.Language != FormatStyle::LK_JavaScript) { 943 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) { 944 parseLabel(); 945 return; 946 } 947 // Recognize function-like macro usages without trailing semicolon as 948 // well as free-standing macros like Q_OBJECT. 949 bool FunctionLike = FormatTok->is(tok::l_paren); 950 if (FunctionLike) 951 parseParens(); 952 953 bool FollowedByNewline = 954 CommentsBeforeNextToken.empty() 955 ? FormatTok->NewlinesBefore > 0 956 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 957 958 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 959 tokenCanStartNewLine(FormatTok->Tok) && Text == Text.upper()) { 960 addUnwrappedLine(); 961 return; 962 } 963 } 964 break; 965 } 966 case tok::equal: 967 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType 968 // TT_JsFatArrow. The always start an expression or a child block if 969 // followed by a curly. 970 if (FormatTok->is(TT_JsFatArrow)) { 971 nextToken(); 972 if (FormatTok->is(tok::l_brace)) 973 parseChildBlock(); 974 break; 975 } 976 977 nextToken(); 978 if (FormatTok->Tok.is(tok::l_brace)) { 979 parseBracedList(); 980 } 981 break; 982 case tok::l_square: 983 parseSquare(); 984 break; 985 case tok::kw_new: 986 parseNew(); 987 break; 988 default: 989 nextToken(); 990 break; 991 } 992 } while (!eof()); 993 } 994 995 bool UnwrappedLineParser::tryToParseLambda() { 996 if (Style.Language != FormatStyle::LK_Cpp) { 997 nextToken(); 998 return false; 999 } 1000 const FormatToken* Previous = getPreviousToken(); 1001 if (Previous && 1002 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 1003 tok::kw_delete) || 1004 Previous->closesScope() || Previous->isSimpleTypeSpecifier())) { 1005 nextToken(); 1006 return false; 1007 } 1008 assert(FormatTok->is(tok::l_square)); 1009 FormatToken &LSquare = *FormatTok; 1010 if (!tryToParseLambdaIntroducer()) 1011 return false; 1012 1013 while (FormatTok->isNot(tok::l_brace)) { 1014 if (FormatTok->isSimpleTypeSpecifier()) { 1015 nextToken(); 1016 continue; 1017 } 1018 switch (FormatTok->Tok.getKind()) { 1019 case tok::l_brace: 1020 break; 1021 case tok::l_paren: 1022 parseParens(); 1023 break; 1024 case tok::amp: 1025 case tok::star: 1026 case tok::kw_const: 1027 case tok::comma: 1028 case tok::less: 1029 case tok::greater: 1030 case tok::identifier: 1031 case tok::numeric_constant: 1032 case tok::coloncolon: 1033 case tok::kw_mutable: 1034 nextToken(); 1035 break; 1036 case tok::arrow: 1037 FormatTok->Type = TT_LambdaArrow; 1038 nextToken(); 1039 break; 1040 default: 1041 return true; 1042 } 1043 } 1044 LSquare.Type = TT_LambdaLSquare; 1045 parseChildBlock(); 1046 return true; 1047 } 1048 1049 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 1050 nextToken(); 1051 if (FormatTok->is(tok::equal)) { 1052 nextToken(); 1053 if (FormatTok->is(tok::r_square)) { 1054 nextToken(); 1055 return true; 1056 } 1057 if (FormatTok->isNot(tok::comma)) 1058 return false; 1059 nextToken(); 1060 } else if (FormatTok->is(tok::amp)) { 1061 nextToken(); 1062 if (FormatTok->is(tok::r_square)) { 1063 nextToken(); 1064 return true; 1065 } 1066 if (!FormatTok->isOneOf(tok::comma, tok::identifier)) { 1067 return false; 1068 } 1069 if (FormatTok->is(tok::comma)) 1070 nextToken(); 1071 } else if (FormatTok->is(tok::r_square)) { 1072 nextToken(); 1073 return true; 1074 } 1075 do { 1076 if (FormatTok->is(tok::amp)) 1077 nextToken(); 1078 if (!FormatTok->isOneOf(tok::identifier, tok::kw_this)) 1079 return false; 1080 nextToken(); 1081 if (FormatTok->is(tok::ellipsis)) 1082 nextToken(); 1083 if (FormatTok->is(tok::comma)) { 1084 nextToken(); 1085 } else if (FormatTok->is(tok::r_square)) { 1086 nextToken(); 1087 return true; 1088 } else { 1089 return false; 1090 } 1091 } while (!eof()); 1092 return false; 1093 } 1094 1095 void UnwrappedLineParser::tryToParseJSFunction() { 1096 nextToken(); 1097 1098 // Consume function name. 1099 if (FormatTok->is(tok::identifier)) 1100 nextToken(); 1101 1102 if (FormatTok->isNot(tok::l_paren)) 1103 return; 1104 1105 // Parse formal parameter list. 1106 parseParens(); 1107 1108 if (FormatTok->is(tok::colon)) { 1109 // Parse a type definition. 1110 nextToken(); 1111 1112 // Eat the type declaration. For braced inline object types, balance braces, 1113 // otherwise just parse until finding an l_brace for the function body. 1114 if (FormatTok->is(tok::l_brace)) 1115 tryToParseBracedList(); 1116 else 1117 while (FormatTok->isNot(tok::l_brace) && !eof()) 1118 nextToken(); 1119 } 1120 1121 parseChildBlock(); 1122 } 1123 1124 bool UnwrappedLineParser::tryToParseBracedList() { 1125 if (FormatTok->BlockKind == BK_Unknown) 1126 calculateBraceTypes(); 1127 assert(FormatTok->BlockKind != BK_Unknown); 1128 if (FormatTok->BlockKind == BK_Block) 1129 return false; 1130 parseBracedList(); 1131 return true; 1132 } 1133 1134 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons) { 1135 bool HasError = false; 1136 nextToken(); 1137 1138 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 1139 // replace this by using parseAssigmentExpression() inside. 1140 do { 1141 if (Style.Language == FormatStyle::LK_JavaScript) { 1142 if (FormatTok->is(Keywords.kw_function)) { 1143 tryToParseJSFunction(); 1144 continue; 1145 } 1146 if (FormatTok->is(TT_JsFatArrow)) { 1147 nextToken(); 1148 // Fat arrows can be followed by simple expressions or by child blocks 1149 // in curly braces. 1150 if (FormatTok->is(tok::l_brace)) { 1151 parseChildBlock(); 1152 continue; 1153 } 1154 } 1155 } 1156 switch (FormatTok->Tok.getKind()) { 1157 case tok::caret: 1158 nextToken(); 1159 if (FormatTok->is(tok::l_brace)) { 1160 parseChildBlock(); 1161 } 1162 break; 1163 case tok::l_square: 1164 tryToParseLambda(); 1165 break; 1166 case tok::l_brace: 1167 // Assume there are no blocks inside a braced init list apart 1168 // from the ones we explicitly parse out (like lambdas). 1169 FormatTok->BlockKind = BK_BracedInit; 1170 parseBracedList(); 1171 break; 1172 case tok::l_paren: 1173 parseParens(); 1174 // JavaScript can just have free standing methods and getters/setters in 1175 // object literals. Detect them by a "{" following ")". 1176 if (Style.Language == FormatStyle::LK_JavaScript) { 1177 if (FormatTok->is(tok::l_brace)) 1178 parseChildBlock(); 1179 break; 1180 } 1181 break; 1182 case tok::r_brace: 1183 nextToken(); 1184 return !HasError; 1185 case tok::semi: 1186 // JavaScript (or more precisely TypeScript) can have semicolons in braced 1187 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 1188 // used for error recovery if we have otherwise determined that this is 1189 // a braced list. 1190 if (Style.Language == FormatStyle::LK_JavaScript) { 1191 nextToken(); 1192 break; 1193 } 1194 HasError = true; 1195 if (!ContinueOnSemicolons) 1196 return !HasError; 1197 nextToken(); 1198 break; 1199 case tok::comma: 1200 nextToken(); 1201 break; 1202 default: 1203 nextToken(); 1204 break; 1205 } 1206 } while (!eof()); 1207 return false; 1208 } 1209 1210 void UnwrappedLineParser::parseParens() { 1211 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 1212 nextToken(); 1213 do { 1214 switch (FormatTok->Tok.getKind()) { 1215 case tok::l_paren: 1216 parseParens(); 1217 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 1218 parseChildBlock(); 1219 break; 1220 case tok::r_paren: 1221 nextToken(); 1222 return; 1223 case tok::r_brace: 1224 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1225 return; 1226 case tok::l_square: 1227 tryToParseLambda(); 1228 break; 1229 case tok::l_brace: 1230 if (!tryToParseBracedList()) 1231 parseChildBlock(); 1232 break; 1233 case tok::at: 1234 nextToken(); 1235 if (FormatTok->Tok.is(tok::l_brace)) 1236 parseBracedList(); 1237 break; 1238 case tok::identifier: 1239 if (Style.Language == FormatStyle::LK_JavaScript && 1240 FormatTok->is(Keywords.kw_function)) 1241 tryToParseJSFunction(); 1242 else 1243 nextToken(); 1244 break; 1245 default: 1246 nextToken(); 1247 break; 1248 } 1249 } while (!eof()); 1250 } 1251 1252 void UnwrappedLineParser::parseSquare() { 1253 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected."); 1254 if (tryToParseLambda()) 1255 return; 1256 do { 1257 switch (FormatTok->Tok.getKind()) { 1258 case tok::l_paren: 1259 parseParens(); 1260 break; 1261 case tok::r_square: 1262 nextToken(); 1263 return; 1264 case tok::r_brace: 1265 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1266 return; 1267 case tok::l_square: 1268 parseSquare(); 1269 break; 1270 case tok::l_brace: { 1271 if (!tryToParseBracedList()) 1272 parseChildBlock(); 1273 break; 1274 } 1275 case tok::at: 1276 nextToken(); 1277 if (FormatTok->Tok.is(tok::l_brace)) 1278 parseBracedList(); 1279 break; 1280 default: 1281 nextToken(); 1282 break; 1283 } 1284 } while (!eof()); 1285 } 1286 1287 void UnwrappedLineParser::parseIfThenElse() { 1288 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 1289 nextToken(); 1290 if (FormatTok->Tok.is(tok::l_paren)) 1291 parseParens(); 1292 bool NeedsUnwrappedLine = false; 1293 if (FormatTok->Tok.is(tok::l_brace)) { 1294 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1295 parseBlock(/*MustBeDeclaration=*/false); 1296 if (Style.BraceWrapping.BeforeElse) 1297 addUnwrappedLine(); 1298 else 1299 NeedsUnwrappedLine = true; 1300 } else { 1301 addUnwrappedLine(); 1302 ++Line->Level; 1303 parseStructuralElement(); 1304 --Line->Level; 1305 } 1306 if (FormatTok->Tok.is(tok::kw_else)) { 1307 nextToken(); 1308 if (FormatTok->Tok.is(tok::l_brace)) { 1309 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1310 parseBlock(/*MustBeDeclaration=*/false); 1311 addUnwrappedLine(); 1312 } else if (FormatTok->Tok.is(tok::kw_if)) { 1313 parseIfThenElse(); 1314 } else { 1315 addUnwrappedLine(); 1316 ++Line->Level; 1317 parseStructuralElement(); 1318 --Line->Level; 1319 } 1320 } else if (NeedsUnwrappedLine) { 1321 addUnwrappedLine(); 1322 } 1323 } 1324 1325 void UnwrappedLineParser::parseTryCatch() { 1326 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 1327 nextToken(); 1328 bool NeedsUnwrappedLine = false; 1329 if (FormatTok->is(tok::colon)) { 1330 // We are in a function try block, what comes is an initializer list. 1331 nextToken(); 1332 while (FormatTok->is(tok::identifier)) { 1333 nextToken(); 1334 if (FormatTok->is(tok::l_paren)) 1335 parseParens(); 1336 if (FormatTok->is(tok::comma)) 1337 nextToken(); 1338 } 1339 } 1340 // Parse try with resource. 1341 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) { 1342 parseParens(); 1343 } 1344 if (FormatTok->is(tok::l_brace)) { 1345 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1346 parseBlock(/*MustBeDeclaration=*/false); 1347 if (Style.BraceWrapping.BeforeCatch) { 1348 addUnwrappedLine(); 1349 } else { 1350 NeedsUnwrappedLine = true; 1351 } 1352 } else if (!FormatTok->is(tok::kw_catch)) { 1353 // The C++ standard requires a compound-statement after a try. 1354 // If there's none, we try to assume there's a structuralElement 1355 // and try to continue. 1356 addUnwrappedLine(); 1357 ++Line->Level; 1358 parseStructuralElement(); 1359 --Line->Level; 1360 } 1361 while (1) { 1362 if (FormatTok->is(tok::at)) 1363 nextToken(); 1364 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 1365 tok::kw___finally) || 1366 ((Style.Language == FormatStyle::LK_Java || 1367 Style.Language == FormatStyle::LK_JavaScript) && 1368 FormatTok->is(Keywords.kw_finally)) || 1369 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) || 1370 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally)))) 1371 break; 1372 nextToken(); 1373 while (FormatTok->isNot(tok::l_brace)) { 1374 if (FormatTok->is(tok::l_paren)) { 1375 parseParens(); 1376 continue; 1377 } 1378 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) 1379 return; 1380 nextToken(); 1381 } 1382 NeedsUnwrappedLine = false; 1383 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1384 parseBlock(/*MustBeDeclaration=*/false); 1385 if (Style.BraceWrapping.BeforeCatch) 1386 addUnwrappedLine(); 1387 else 1388 NeedsUnwrappedLine = true; 1389 } 1390 if (NeedsUnwrappedLine) 1391 addUnwrappedLine(); 1392 } 1393 1394 void UnwrappedLineParser::parseNamespace() { 1395 assert(FormatTok->Tok.is(tok::kw_namespace) && "'namespace' expected"); 1396 1397 const FormatToken &InitialToken = *FormatTok; 1398 nextToken(); 1399 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon)) 1400 nextToken(); 1401 if (FormatTok->Tok.is(tok::l_brace)) { 1402 if (ShouldBreakBeforeBrace(Style, InitialToken)) 1403 addUnwrappedLine(); 1404 1405 bool AddLevel = Style.NamespaceIndentation == FormatStyle::NI_All || 1406 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 1407 DeclarationScopeStack.size() > 1); 1408 parseBlock(/*MustBeDeclaration=*/true, AddLevel); 1409 // Munch the semicolon after a namespace. This is more common than one would 1410 // think. Puttin the semicolon into its own line is very ugly. 1411 if (FormatTok->Tok.is(tok::semi)) 1412 nextToken(); 1413 addUnwrappedLine(); 1414 } 1415 // FIXME: Add error handling. 1416 } 1417 1418 void UnwrappedLineParser::parseNew() { 1419 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 1420 nextToken(); 1421 if (Style.Language != FormatStyle::LK_Java) 1422 return; 1423 1424 // In Java, we can parse everything up to the parens, which aren't optional. 1425 do { 1426 // There should not be a ;, { or } before the new's open paren. 1427 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 1428 return; 1429 1430 // Consume the parens. 1431 if (FormatTok->is(tok::l_paren)) { 1432 parseParens(); 1433 1434 // If there is a class body of an anonymous class, consume that as child. 1435 if (FormatTok->is(tok::l_brace)) 1436 parseChildBlock(); 1437 return; 1438 } 1439 nextToken(); 1440 } while (!eof()); 1441 } 1442 1443 void UnwrappedLineParser::parseForOrWhileLoop() { 1444 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 1445 "'for', 'while' or foreach macro expected"); 1446 nextToken(); 1447 if (FormatTok->Tok.is(tok::l_paren)) 1448 parseParens(); 1449 if (FormatTok->Tok.is(tok::l_brace)) { 1450 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1451 parseBlock(/*MustBeDeclaration=*/false); 1452 addUnwrappedLine(); 1453 } else { 1454 addUnwrappedLine(); 1455 ++Line->Level; 1456 parseStructuralElement(); 1457 --Line->Level; 1458 } 1459 } 1460 1461 void UnwrappedLineParser::parseDoWhile() { 1462 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 1463 nextToken(); 1464 if (FormatTok->Tok.is(tok::l_brace)) { 1465 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1466 parseBlock(/*MustBeDeclaration=*/false); 1467 if (Style.BraceWrapping.IndentBraces) 1468 addUnwrappedLine(); 1469 } else { 1470 addUnwrappedLine(); 1471 ++Line->Level; 1472 parseStructuralElement(); 1473 --Line->Level; 1474 } 1475 1476 // FIXME: Add error handling. 1477 if (!FormatTok->Tok.is(tok::kw_while)) { 1478 addUnwrappedLine(); 1479 return; 1480 } 1481 1482 nextToken(); 1483 parseStructuralElement(); 1484 } 1485 1486 void UnwrappedLineParser::parseLabel() { 1487 nextToken(); 1488 unsigned OldLineLevel = Line->Level; 1489 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 1490 --Line->Level; 1491 if (CommentsBeforeNextToken.empty() && FormatTok->Tok.is(tok::l_brace)) { 1492 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1493 parseBlock(/*MustBeDeclaration=*/false); 1494 if (FormatTok->Tok.is(tok::kw_break)) { 1495 if (Style.BraceWrapping.AfterControlStatement) 1496 addUnwrappedLine(); 1497 parseStructuralElement(); 1498 } 1499 addUnwrappedLine(); 1500 } else { 1501 if (FormatTok->is(tok::semi)) 1502 nextToken(); 1503 addUnwrappedLine(); 1504 } 1505 Line->Level = OldLineLevel; 1506 } 1507 1508 void UnwrappedLineParser::parseCaseLabel() { 1509 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 1510 // FIXME: fix handling of complex expressions here. 1511 do { 1512 nextToken(); 1513 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 1514 parseLabel(); 1515 } 1516 1517 void UnwrappedLineParser::parseSwitch() { 1518 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 1519 nextToken(); 1520 if (FormatTok->Tok.is(tok::l_paren)) 1521 parseParens(); 1522 if (FormatTok->Tok.is(tok::l_brace)) { 1523 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1524 parseBlock(/*MustBeDeclaration=*/false); 1525 addUnwrappedLine(); 1526 } else { 1527 addUnwrappedLine(); 1528 ++Line->Level; 1529 parseStructuralElement(); 1530 --Line->Level; 1531 } 1532 } 1533 1534 void UnwrappedLineParser::parseAccessSpecifier() { 1535 nextToken(); 1536 // Understand Qt's slots. 1537 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 1538 nextToken(); 1539 // Otherwise, we don't know what it is, and we'd better keep the next token. 1540 if (FormatTok->Tok.is(tok::colon)) 1541 nextToken(); 1542 addUnwrappedLine(); 1543 } 1544 1545 bool UnwrappedLineParser::parseEnum() { 1546 // Won't be 'enum' for NS_ENUMs. 1547 if (FormatTok->Tok.is(tok::kw_enum)) 1548 nextToken(); 1549 1550 // In TypeScript, "enum" can also be used as property name, e.g. in interface 1551 // declarations. An "enum" keyword followed by a colon would be a syntax 1552 // error and thus assume it is just an identifier. 1553 if (Style.Language == FormatStyle::LK_JavaScript && 1554 FormatTok->isOneOf(tok::colon, tok::question)) 1555 return false; 1556 1557 // Eat up enum class ... 1558 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct)) 1559 nextToken(); 1560 1561 while (FormatTok->Tok.getIdentifierInfo() || 1562 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 1563 tok::greater, tok::comma, tok::question)) { 1564 nextToken(); 1565 // We can have macros or attributes in between 'enum' and the enum name. 1566 if (FormatTok->is(tok::l_paren)) 1567 parseParens(); 1568 if (FormatTok->is(tok::identifier)) { 1569 nextToken(); 1570 // If there are two identifiers in a row, this is likely an elaborate 1571 // return type. In Java, this can be "implements", etc. 1572 if (Style.Language == FormatStyle::LK_Cpp && 1573 FormatTok->is(tok::identifier)) 1574 return false; 1575 } 1576 } 1577 1578 // Just a declaration or something is wrong. 1579 if (FormatTok->isNot(tok::l_brace)) 1580 return true; 1581 FormatTok->BlockKind = BK_Block; 1582 1583 if (Style.Language == FormatStyle::LK_Java) { 1584 // Java enums are different. 1585 parseJavaEnumBody(); 1586 return true; 1587 } 1588 if (Style.Language == FormatStyle::LK_Proto) { 1589 parseBlock(/*MustBeDeclaration=*/true); 1590 return true; 1591 } 1592 1593 // Parse enum body. 1594 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true); 1595 if (HasError) { 1596 if (FormatTok->is(tok::semi)) 1597 nextToken(); 1598 addUnwrappedLine(); 1599 } 1600 return true; 1601 1602 // There is no addUnwrappedLine() here so that we fall through to parsing a 1603 // structural element afterwards. Thus, in "enum A {} n, m;", 1604 // "} n, m;" will end up in one unwrapped line. 1605 } 1606 1607 void UnwrappedLineParser::parseJavaEnumBody() { 1608 // Determine whether the enum is simple, i.e. does not have a semicolon or 1609 // constants with class bodies. Simple enums can be formatted like braced 1610 // lists, contracted to a single line, etc. 1611 unsigned StoredPosition = Tokens->getPosition(); 1612 bool IsSimple = true; 1613 FormatToken *Tok = Tokens->getNextToken(); 1614 while (Tok) { 1615 if (Tok->is(tok::r_brace)) 1616 break; 1617 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 1618 IsSimple = false; 1619 break; 1620 } 1621 // FIXME: This will also mark enums with braces in the arguments to enum 1622 // constants as "not simple". This is probably fine in practice, though. 1623 Tok = Tokens->getNextToken(); 1624 } 1625 FormatTok = Tokens->setPosition(StoredPosition); 1626 1627 if (IsSimple) { 1628 parseBracedList(); 1629 addUnwrappedLine(); 1630 return; 1631 } 1632 1633 // Parse the body of a more complex enum. 1634 // First add a line for everything up to the "{". 1635 nextToken(); 1636 addUnwrappedLine(); 1637 ++Line->Level; 1638 1639 // Parse the enum constants. 1640 while (FormatTok) { 1641 if (FormatTok->is(tok::l_brace)) { 1642 // Parse the constant's class body. 1643 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 1644 /*MunchSemi=*/false); 1645 } else if (FormatTok->is(tok::l_paren)) { 1646 parseParens(); 1647 } else if (FormatTok->is(tok::comma)) { 1648 nextToken(); 1649 addUnwrappedLine(); 1650 } else if (FormatTok->is(tok::semi)) { 1651 nextToken(); 1652 addUnwrappedLine(); 1653 break; 1654 } else if (FormatTok->is(tok::r_brace)) { 1655 addUnwrappedLine(); 1656 break; 1657 } else { 1658 nextToken(); 1659 } 1660 } 1661 1662 // Parse the class body after the enum's ";" if any. 1663 parseLevel(/*HasOpeningBrace=*/true); 1664 nextToken(); 1665 --Line->Level; 1666 addUnwrappedLine(); 1667 } 1668 1669 void UnwrappedLineParser::parseRecord() { 1670 const FormatToken &InitialToken = *FormatTok; 1671 nextToken(); 1672 1673 // The actual identifier can be a nested name specifier, and in macros 1674 // it is often token-pasted. 1675 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 1676 tok::kw___attribute, tok::kw___declspec, 1677 tok::kw_alignas) || 1678 ((Style.Language == FormatStyle::LK_Java || 1679 Style.Language == FormatStyle::LK_JavaScript) && 1680 FormatTok->isOneOf(tok::period, tok::comma))) { 1681 bool IsNonMacroIdentifier = 1682 FormatTok->is(tok::identifier) && 1683 FormatTok->TokenText != FormatTok->TokenText.upper(); 1684 nextToken(); 1685 // We can have macros or attributes in between 'class' and the class name. 1686 if (!IsNonMacroIdentifier && FormatTok->Tok.is(tok::l_paren)) 1687 parseParens(); 1688 } 1689 1690 // Note that parsing away template declarations here leads to incorrectly 1691 // accepting function declarations as record declarations. 1692 // In general, we cannot solve this problem. Consider: 1693 // class A<int> B() {} 1694 // which can be a function definition or a class definition when B() is a 1695 // macro. If we find enough real-world cases where this is a problem, we 1696 // can parse for the 'template' keyword in the beginning of the statement, 1697 // and thus rule out the record production in case there is no template 1698 // (this would still leave us with an ambiguity between template function 1699 // and class declarations). 1700 if (FormatTok->isOneOf(tok::colon, tok::less)) { 1701 while (!eof()) { 1702 if (FormatTok->is(tok::l_brace)) { 1703 calculateBraceTypes(/*ExpectClassBody=*/true); 1704 if (!tryToParseBracedList()) 1705 break; 1706 } 1707 if (FormatTok->Tok.is(tok::semi)) 1708 return; 1709 nextToken(); 1710 } 1711 } 1712 if (FormatTok->Tok.is(tok::l_brace)) { 1713 if (ShouldBreakBeforeBrace(Style, InitialToken)) 1714 addUnwrappedLine(); 1715 1716 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 1717 /*MunchSemi=*/false); 1718 } 1719 // There is no addUnwrappedLine() here so that we fall through to parsing a 1720 // structural element afterwards. Thus, in "class A {} n, m;", 1721 // "} n, m;" will end up in one unwrapped line. 1722 } 1723 1724 void UnwrappedLineParser::parseObjCProtocolList() { 1725 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 1726 do 1727 nextToken(); 1728 while (!eof() && FormatTok->Tok.isNot(tok::greater)); 1729 nextToken(); // Skip '>'. 1730 } 1731 1732 void UnwrappedLineParser::parseObjCUntilAtEnd() { 1733 do { 1734 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 1735 nextToken(); 1736 addUnwrappedLine(); 1737 break; 1738 } 1739 if (FormatTok->is(tok::l_brace)) { 1740 parseBlock(/*MustBeDeclaration=*/false); 1741 // In ObjC interfaces, nothing should be following the "}". 1742 addUnwrappedLine(); 1743 } else if (FormatTok->is(tok::r_brace)) { 1744 // Ignore stray "}". parseStructuralElement doesn't consume them. 1745 nextToken(); 1746 addUnwrappedLine(); 1747 } else { 1748 parseStructuralElement(); 1749 } 1750 } while (!eof()); 1751 } 1752 1753 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 1754 nextToken(); 1755 nextToken(); // interface name 1756 1757 // @interface can be followed by either a base class, or a category. 1758 if (FormatTok->Tok.is(tok::colon)) { 1759 nextToken(); 1760 nextToken(); // base class name 1761 } else if (FormatTok->Tok.is(tok::l_paren)) 1762 // Skip category, if present. 1763 parseParens(); 1764 1765 if (FormatTok->Tok.is(tok::less)) 1766 parseObjCProtocolList(); 1767 1768 if (FormatTok->Tok.is(tok::l_brace)) { 1769 if (Style.BraceWrapping.AfterObjCDeclaration) 1770 addUnwrappedLine(); 1771 parseBlock(/*MustBeDeclaration=*/true); 1772 } 1773 1774 // With instance variables, this puts '}' on its own line. Without instance 1775 // variables, this ends the @interface line. 1776 addUnwrappedLine(); 1777 1778 parseObjCUntilAtEnd(); 1779 } 1780 1781 void UnwrappedLineParser::parseObjCProtocol() { 1782 nextToken(); 1783 nextToken(); // protocol name 1784 1785 if (FormatTok->Tok.is(tok::less)) 1786 parseObjCProtocolList(); 1787 1788 // Check for protocol declaration. 1789 if (FormatTok->Tok.is(tok::semi)) { 1790 nextToken(); 1791 return addUnwrappedLine(); 1792 } 1793 1794 addUnwrappedLine(); 1795 parseObjCUntilAtEnd(); 1796 } 1797 1798 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 1799 assert(FormatTok->isOneOf(Keywords.kw_import, tok::kw_export)); 1800 nextToken(); 1801 1802 // Consume the "default" in "export default class/function". 1803 if (FormatTok->is(tok::kw_default)) 1804 nextToken(); 1805 1806 // Consume "function" and "default function", so that these get parsed as 1807 // free-standing JS functions, i.e. do not require a trailing semicolon. 1808 if (FormatTok->is(Keywords.kw_function)) { 1809 nextToken(); 1810 return; 1811 } 1812 1813 // Consume the "abstract" in "export abstract class". 1814 if (FormatTok->is(Keywords.kw_abstract)) 1815 nextToken(); 1816 1817 if (FormatTok->isOneOf(tok::kw_const, tok::kw_class, tok::kw_enum, 1818 Keywords.kw_interface, Keywords.kw_let, 1819 Keywords.kw_var)) 1820 return; // Fall through to parsing the corresponding structure. 1821 1822 while (!eof() && FormatTok->isNot(tok::semi)) { 1823 if (FormatTok->is(tok::l_brace)) { 1824 FormatTok->BlockKind = BK_Block; 1825 parseBracedList(); 1826 } else { 1827 nextToken(); 1828 } 1829 } 1830 } 1831 1832 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 1833 StringRef Prefix = "") { 1834 llvm::dbgs() << Prefix << "Line(" << Line.Level << ")" 1835 << (Line.InPPDirective ? " MACRO" : "") << ": "; 1836 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 1837 E = Line.Tokens.end(); 1838 I != E; ++I) { 1839 llvm::dbgs() << I->Tok->Tok.getName() << "[" << I->Tok->Type << "] "; 1840 } 1841 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 1842 E = Line.Tokens.end(); 1843 I != E; ++I) { 1844 const UnwrappedLineNode &Node = *I; 1845 for (SmallVectorImpl<UnwrappedLine>::const_iterator 1846 I = Node.Children.begin(), 1847 E = Node.Children.end(); 1848 I != E; ++I) { 1849 printDebugInfo(*I, "\nChild: "); 1850 } 1851 } 1852 llvm::dbgs() << "\n"; 1853 } 1854 1855 void UnwrappedLineParser::addUnwrappedLine() { 1856 if (Line->Tokens.empty()) 1857 return; 1858 DEBUG({ 1859 if (CurrentLines == &Lines) 1860 printDebugInfo(*Line); 1861 }); 1862 CurrentLines->push_back(std::move(*Line)); 1863 Line->Tokens.clear(); 1864 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 1865 CurrentLines->append( 1866 std::make_move_iterator(PreprocessorDirectives.begin()), 1867 std::make_move_iterator(PreprocessorDirectives.end())); 1868 PreprocessorDirectives.clear(); 1869 } 1870 } 1871 1872 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 1873 1874 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 1875 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 1876 FormatTok.NewlinesBefore > 0; 1877 } 1878 1879 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 1880 bool JustComments = Line->Tokens.empty(); 1881 for (SmallVectorImpl<FormatToken *>::const_iterator 1882 I = CommentsBeforeNextToken.begin(), 1883 E = CommentsBeforeNextToken.end(); 1884 I != E; ++I) { 1885 if (isOnNewLine(**I) && JustComments) 1886 addUnwrappedLine(); 1887 pushToken(*I); 1888 } 1889 if (NewlineBeforeNext && JustComments) 1890 addUnwrappedLine(); 1891 CommentsBeforeNextToken.clear(); 1892 } 1893 1894 void UnwrappedLineParser::nextToken() { 1895 if (eof()) 1896 return; 1897 flushComments(isOnNewLine(*FormatTok)); 1898 pushToken(FormatTok); 1899 readToken(); 1900 } 1901 1902 const FormatToken *UnwrappedLineParser::getPreviousToken() { 1903 // FIXME: This is a dirty way to access the previous token. Find a better 1904 // solution. 1905 if (!Line || Line->Tokens.empty()) 1906 return nullptr; 1907 return Line->Tokens.back().Tok; 1908 } 1909 1910 void UnwrappedLineParser::readToken() { 1911 bool CommentsInCurrentLine = true; 1912 do { 1913 FormatTok = Tokens->getNextToken(); 1914 assert(FormatTok); 1915 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 1916 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 1917 // If there is an unfinished unwrapped line, we flush the preprocessor 1918 // directives only after that unwrapped line was finished later. 1919 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 1920 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 1921 // Comments stored before the preprocessor directive need to be output 1922 // before the preprocessor directive, at the same level as the 1923 // preprocessor directive, as we consider them to apply to the directive. 1924 flushComments(isOnNewLine(*FormatTok)); 1925 parsePPDirective(); 1926 } 1927 while (FormatTok->Type == TT_ConflictStart || 1928 FormatTok->Type == TT_ConflictEnd || 1929 FormatTok->Type == TT_ConflictAlternative) { 1930 if (FormatTok->Type == TT_ConflictStart) { 1931 conditionalCompilationStart(/*Unreachable=*/false); 1932 } else if (FormatTok->Type == TT_ConflictAlternative) { 1933 conditionalCompilationAlternative(); 1934 } else if (FormatTok->Type == TT_ConflictEnd) { 1935 conditionalCompilationEnd(); 1936 } 1937 FormatTok = Tokens->getNextToken(); 1938 FormatTok->MustBreakBefore = true; 1939 } 1940 1941 if (!PPStack.empty() && (PPStack.back() == PP_Unreachable) && 1942 !Line->InPPDirective) { 1943 continue; 1944 } 1945 1946 if (!FormatTok->Tok.is(tok::comment)) 1947 return; 1948 if (isOnNewLine(*FormatTok) || FormatTok->IsFirst) { 1949 CommentsInCurrentLine = false; 1950 } 1951 if (CommentsInCurrentLine) { 1952 pushToken(FormatTok); 1953 } else { 1954 CommentsBeforeNextToken.push_back(FormatTok); 1955 } 1956 } while (!eof()); 1957 } 1958 1959 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 1960 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 1961 if (MustBreakBeforeNextToken) { 1962 Line->Tokens.back().Tok->MustBreakBefore = true; 1963 MustBreakBeforeNextToken = false; 1964 } 1965 } 1966 1967 } // end namespace format 1968 } // end namespace clang 1969