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 #define DEBUG_TYPE "format-parser" 17 18 #include "UnwrappedLineParser.h" 19 #include "llvm/Support/Debug.h" 20 21 namespace clang { 22 namespace format { 23 24 class FormatTokenSource { 25 public: 26 virtual ~FormatTokenSource() {} 27 virtual FormatToken *getNextToken() = 0; 28 29 virtual unsigned getPosition() = 0; 30 virtual FormatToken *setPosition(unsigned Position) = 0; 31 }; 32 33 namespace { 34 35 class ScopedDeclarationState { 36 public: 37 ScopedDeclarationState(UnwrappedLine &Line, std::vector<bool> &Stack, 38 bool MustBeDeclaration) 39 : Line(Line), Stack(Stack) { 40 Line.MustBeDeclaration = MustBeDeclaration; 41 Stack.push_back(MustBeDeclaration); 42 } 43 ~ScopedDeclarationState() { 44 Stack.pop_back(); 45 if (!Stack.empty()) 46 Line.MustBeDeclaration = Stack.back(); 47 else 48 Line.MustBeDeclaration = true; 49 } 50 51 private: 52 UnwrappedLine &Line; 53 std::vector<bool> &Stack; 54 }; 55 56 class ScopedMacroState : public FormatTokenSource { 57 public: 58 ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource, 59 FormatToken *&ResetToken, bool &StructuralError) 60 : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken), 61 PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource), 62 StructuralError(StructuralError), 63 PreviousStructuralError(StructuralError), Token(NULL) { 64 TokenSource = this; 65 Line.Level = 0; 66 Line.InPPDirective = true; 67 } 68 69 ~ScopedMacroState() { 70 TokenSource = PreviousTokenSource; 71 ResetToken = Token; 72 Line.InPPDirective = false; 73 Line.Level = PreviousLineLevel; 74 StructuralError = PreviousStructuralError; 75 } 76 77 virtual FormatToken *getNextToken() { 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 virtual unsigned getPosition() { return PreviousTokenSource->getPosition(); } 88 89 virtual FormatToken *setPosition(unsigned Position) { 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 bool &StructuralError; 114 bool PreviousStructuralError; 115 116 FormatToken *Token; 117 }; 118 119 } // end anonymous namespace 120 121 class ScopedLineState { 122 public: 123 ScopedLineState(UnwrappedLineParser &Parser, 124 bool SwitchToPreprocessorLines = false) 125 : Parser(Parser) { 126 OriginalLines = Parser.CurrentLines; 127 if (SwitchToPreprocessorLines) 128 Parser.CurrentLines = &Parser.PreprocessorDirectives; 129 else if (!Parser.Line->Tokens.empty()) 130 Parser.CurrentLines = &Parser.Line->Tokens.back().Children; 131 PreBlockLine = Parser.Line.take(); 132 Parser.Line.reset(new UnwrappedLine()); 133 Parser.Line->Level = PreBlockLine->Level; 134 Parser.Line->InPPDirective = PreBlockLine->InPPDirective; 135 } 136 137 ~ScopedLineState() { 138 if (!Parser.Line->Tokens.empty()) { 139 Parser.addUnwrappedLine(); 140 } 141 assert(Parser.Line->Tokens.empty()); 142 Parser.Line.reset(PreBlockLine); 143 if (Parser.CurrentLines == &Parser.PreprocessorDirectives) 144 Parser.MustBreakBeforeNextToken = true; 145 Parser.CurrentLines = OriginalLines; 146 } 147 148 private: 149 UnwrappedLineParser &Parser; 150 151 UnwrappedLine *PreBlockLine; 152 SmallVectorImpl<UnwrappedLine> *OriginalLines; 153 }; 154 155 namespace { 156 157 class IndexedTokenSource : public FormatTokenSource { 158 public: 159 IndexedTokenSource(ArrayRef<FormatToken *> Tokens) 160 : Tokens(Tokens), Position(-1) {} 161 162 virtual FormatToken *getNextToken() { 163 ++Position; 164 return Tokens[Position]; 165 } 166 167 virtual unsigned getPosition() { 168 assert(Position >= 0); 169 return Position; 170 } 171 172 virtual FormatToken *setPosition(unsigned P) { 173 Position = P; 174 return Tokens[Position]; 175 } 176 177 void reset() { Position = -1; } 178 179 private: 180 ArrayRef<FormatToken *> Tokens; 181 int Position; 182 }; 183 184 } // end anonymous namespace 185 186 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style, 187 ArrayRef<FormatToken *> Tokens, 188 UnwrappedLineConsumer &Callback) 189 : Line(new UnwrappedLine), MustBreakBeforeNextToken(false), 190 CurrentLines(&Lines), StructuralError(false), Style(Style), Tokens(NULL), 191 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1) {} 192 193 void UnwrappedLineParser::reset() { 194 PPBranchLevel = -1; 195 Line.reset(new UnwrappedLine); 196 CommentsBeforeNextToken.clear(); 197 FormatTok = NULL; 198 MustBreakBeforeNextToken = false; 199 PreprocessorDirectives.clear(); 200 CurrentLines = &Lines; 201 DeclarationScopeStack.clear(); 202 StructuralError = false; 203 PPStack.clear(); 204 } 205 206 bool UnwrappedLineParser::parse() { 207 IndexedTokenSource TokenSource(AllTokens); 208 do { 209 DEBUG(llvm::dbgs() << "----\n"); 210 reset(); 211 Tokens = &TokenSource; 212 TokenSource.reset(); 213 214 readToken(); 215 parseFile(); 216 // Create line with eof token. 217 pushToken(FormatTok); 218 addUnwrappedLine(); 219 220 for (SmallVectorImpl<UnwrappedLine>::iterator I = Lines.begin(), 221 E = Lines.end(); 222 I != E; ++I) { 223 Callback.consumeUnwrappedLine(*I); 224 } 225 Callback.finishRun(); 226 Lines.clear(); 227 while (!PPLevelBranchIndex.empty() && 228 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) { 229 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1); 230 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1); 231 } 232 if (!PPLevelBranchIndex.empty()) { 233 ++PPLevelBranchIndex.back(); 234 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size()); 235 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back()); 236 } 237 } while (!PPLevelBranchIndex.empty()); 238 239 return StructuralError; 240 } 241 242 void UnwrappedLineParser::parseFile() { 243 ScopedDeclarationState DeclarationState( 244 *Line, DeclarationScopeStack, 245 /*MustBeDeclaration=*/ !Line->InPPDirective); 246 parseLevel(/*HasOpeningBrace=*/false); 247 // Make sure to format the remaining tokens. 248 flushComments(true); 249 addUnwrappedLine(); 250 } 251 252 void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) { 253 bool SwitchLabelEncountered = false; 254 do { 255 switch (FormatTok->Tok.getKind()) { 256 case tok::comment: 257 nextToken(); 258 addUnwrappedLine(); 259 break; 260 case tok::l_brace: 261 // FIXME: Add parameter whether this can happen - if this happens, we must 262 // be in a non-declaration context. 263 parseBlock(/*MustBeDeclaration=*/false); 264 addUnwrappedLine(); 265 break; 266 case tok::r_brace: 267 if (HasOpeningBrace) 268 return; 269 StructuralError = true; 270 nextToken(); 271 addUnwrappedLine(); 272 break; 273 case tok::kw_default: 274 case tok::kw_case: 275 if (!SwitchLabelEncountered && 276 (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1))) 277 ++Line->Level; 278 SwitchLabelEncountered = true; 279 parseStructuralElement(); 280 break; 281 default: 282 parseStructuralElement(); 283 break; 284 } 285 } while (!eof()); 286 } 287 288 void UnwrappedLineParser::calculateBraceTypes() { 289 // We'll parse forward through the tokens until we hit 290 // a closing brace or eof - note that getNextToken() will 291 // parse macros, so this will magically work inside macro 292 // definitions, too. 293 unsigned StoredPosition = Tokens->getPosition(); 294 unsigned Position = StoredPosition; 295 FormatToken *Tok = FormatTok; 296 // Keep a stack of positions of lbrace tokens. We will 297 // update information about whether an lbrace starts a 298 // braced init list or a different block during the loop. 299 SmallVector<FormatToken *, 8> LBraceStack; 300 assert(Tok->Tok.is(tok::l_brace)); 301 do { 302 // Get next none-comment token. 303 FormatToken *NextTok; 304 unsigned ReadTokens = 0; 305 do { 306 NextTok = Tokens->getNextToken(); 307 ++ReadTokens; 308 } while (NextTok->is(tok::comment)); 309 310 switch (Tok->Tok.getKind()) { 311 case tok::l_brace: 312 LBraceStack.push_back(Tok); 313 break; 314 case tok::r_brace: 315 if (!LBraceStack.empty()) { 316 if (LBraceStack.back()->BlockKind == BK_Unknown) { 317 // If there is a comma, semicolon or right paren after the closing 318 // brace, we assume this is a braced initializer list. Note that 319 // regardless how we mark inner braces here, we will overwrite the 320 // BlockKind later if we parse a braced list (where all blocks inside 321 // are by default braced lists), or when we explicitly detect blocks 322 // (for example while parsing lambdas). 323 // 324 // We exclude + and - as they can be ObjC visibility modifiers. 325 if (NextTok->isOneOf(tok::comma, tok::semi, tok::r_paren, 326 tok::r_square, tok::l_brace, tok::colon) || 327 (NextTok->isBinaryOperator() && 328 !NextTok->isOneOf(tok::plus, tok::minus))) { 329 Tok->BlockKind = BK_BracedInit; 330 LBraceStack.back()->BlockKind = BK_BracedInit; 331 } else { 332 Tok->BlockKind = BK_Block; 333 LBraceStack.back()->BlockKind = BK_Block; 334 } 335 } 336 LBraceStack.pop_back(); 337 } 338 break; 339 case tok::semi: 340 case tok::kw_if: 341 case tok::kw_while: 342 case tok::kw_for: 343 case tok::kw_switch: 344 case tok::kw_try: 345 if (!LBraceStack.empty()) 346 LBraceStack.back()->BlockKind = BK_Block; 347 break; 348 default: 349 break; 350 } 351 Tok = NextTok; 352 Position += ReadTokens; 353 } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty()); 354 // Assume other blocks for all unclosed opening braces. 355 for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) { 356 if (LBraceStack[i]->BlockKind == BK_Unknown) 357 LBraceStack[i]->BlockKind = BK_Block; 358 } 359 360 FormatTok = Tokens->setPosition(StoredPosition); 361 } 362 363 void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, bool AddLevel, 364 bool MunchSemi) { 365 assert(FormatTok->Tok.is(tok::l_brace) && "'{' expected"); 366 unsigned InitialLevel = Line->Level; 367 nextToken(); 368 369 addUnwrappedLine(); 370 371 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 372 MustBeDeclaration); 373 if (AddLevel) 374 ++Line->Level; 375 parseLevel(/*HasOpeningBrace=*/true); 376 377 if (!FormatTok->Tok.is(tok::r_brace)) { 378 Line->Level = InitialLevel; 379 StructuralError = true; 380 return; 381 } 382 383 nextToken(); // Munch the closing brace. 384 if (MunchSemi && FormatTok->Tok.is(tok::semi)) 385 nextToken(); 386 Line->Level = InitialLevel; 387 } 388 389 void UnwrappedLineParser::parseChildBlock() { 390 FormatTok->BlockKind = BK_Block; 391 nextToken(); 392 { 393 ScopedLineState LineState(*this); 394 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 395 /*MustBeDeclaration=*/false); 396 Line->Level += 1; 397 parseLevel(/*HasOpeningBrace=*/true); 398 Line->Level -= 1; 399 } 400 nextToken(); 401 } 402 403 void UnwrappedLineParser::parsePPDirective() { 404 assert(FormatTok->Tok.is(tok::hash) && "'#' expected"); 405 ScopedMacroState MacroState(*Line, Tokens, FormatTok, StructuralError); 406 nextToken(); 407 408 if (FormatTok->Tok.getIdentifierInfo() == NULL) { 409 parsePPUnknown(); 410 return; 411 } 412 413 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 414 case tok::pp_define: 415 parsePPDefine(); 416 return; 417 case tok::pp_if: 418 parsePPIf(/*IfDef=*/false); 419 break; 420 case tok::pp_ifdef: 421 case tok::pp_ifndef: 422 parsePPIf(/*IfDef=*/true); 423 break; 424 case tok::pp_else: 425 parsePPElse(); 426 break; 427 case tok::pp_elif: 428 parsePPElIf(); 429 break; 430 case tok::pp_endif: 431 parsePPEndIf(); 432 break; 433 default: 434 parsePPUnknown(); 435 break; 436 } 437 } 438 439 void UnwrappedLineParser::pushPPConditional() { 440 if (!PPStack.empty() && PPStack.back() == PP_Unreachable) 441 PPStack.push_back(PP_Unreachable); 442 else 443 PPStack.push_back(PP_Conditional); 444 } 445 446 void UnwrappedLineParser::parsePPIf(bool IfDef) { 447 ++PPBranchLevel; 448 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 449 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 450 PPLevelBranchIndex.push_back(0); 451 PPLevelBranchCount.push_back(0); 452 } 453 PPChainBranchIndex.push(0); 454 nextToken(); 455 bool IsLiteralFalse = (FormatTok->Tok.isLiteral() && 456 StringRef(FormatTok->Tok.getLiteralData(), 457 FormatTok->Tok.getLength()) == "0") || 458 FormatTok->Tok.is(tok::kw_false); 459 if ((!IfDef && IsLiteralFalse) || PPLevelBranchIndex[PPBranchLevel] > 0) { 460 PPStack.push_back(PP_Unreachable); 461 } else { 462 pushPPConditional(); 463 } 464 parsePPUnknown(); 465 } 466 467 void UnwrappedLineParser::parsePPElse() { 468 if (!PPStack.empty()) 469 PPStack.pop_back(); 470 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 471 if (!PPChainBranchIndex.empty()) 472 ++PPChainBranchIndex.top(); 473 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 474 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()) { 475 PPStack.push_back(PP_Unreachable); 476 } else { 477 pushPPConditional(); 478 } 479 parsePPUnknown(); 480 } 481 482 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 483 484 void UnwrappedLineParser::parsePPEndIf() { 485 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 486 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 487 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) { 488 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 489 } 490 } 491 --PPBranchLevel; 492 if (!PPChainBranchIndex.empty()) 493 PPChainBranchIndex.pop(); 494 if (!PPStack.empty()) 495 PPStack.pop_back(); 496 parsePPUnknown(); 497 } 498 499 void UnwrappedLineParser::parsePPDefine() { 500 nextToken(); 501 502 if (FormatTok->Tok.getKind() != tok::identifier) { 503 parsePPUnknown(); 504 return; 505 } 506 nextToken(); 507 if (FormatTok->Tok.getKind() == tok::l_paren && 508 FormatTok->WhitespaceRange.getBegin() == 509 FormatTok->WhitespaceRange.getEnd()) { 510 parseParens(); 511 } 512 addUnwrappedLine(); 513 Line->Level = 1; 514 515 // Errors during a preprocessor directive can only affect the layout of the 516 // preprocessor directive, and thus we ignore them. An alternative approach 517 // would be to use the same approach we use on the file level (no 518 // re-indentation if there was a structural error) within the macro 519 // definition. 520 parseFile(); 521 } 522 523 void UnwrappedLineParser::parsePPUnknown() { 524 do { 525 nextToken(); 526 } while (!eof()); 527 addUnwrappedLine(); 528 } 529 530 // Here we blacklist certain tokens that are not usually the first token in an 531 // unwrapped line. This is used in attempt to distinguish macro calls without 532 // trailing semicolons from other constructs split to several lines. 533 bool tokenCanStartNewLine(clang::Token Tok) { 534 // Semicolon can be a null-statement, l_square can be a start of a macro or 535 // a C++11 attribute, but this doesn't seem to be common. 536 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 537 Tok.isNot(tok::l_square) && 538 // Tokens that can only be used as binary operators and a part of 539 // overloaded operator names. 540 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 541 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 542 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 543 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 544 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 545 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 546 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 547 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 548 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 549 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 550 Tok.isNot(tok::lesslessequal) && 551 // Colon is used in labels, base class lists, initializer lists, 552 // range-based for loops, ternary operator, but should never be the 553 // first token in an unwrapped line. 554 Tok.isNot(tok::colon); 555 } 556 557 void UnwrappedLineParser::parseStructuralElement() { 558 assert(!FormatTok->Tok.is(tok::l_brace)); 559 switch (FormatTok->Tok.getKind()) { 560 case tok::at: 561 nextToken(); 562 if (FormatTok->Tok.is(tok::l_brace)) { 563 parseBracedList(); 564 break; 565 } 566 switch (FormatTok->Tok.getObjCKeywordID()) { 567 case tok::objc_public: 568 case tok::objc_protected: 569 case tok::objc_package: 570 case tok::objc_private: 571 return parseAccessSpecifier(); 572 case tok::objc_interface: 573 case tok::objc_implementation: 574 return parseObjCInterfaceOrImplementation(); 575 case tok::objc_protocol: 576 return parseObjCProtocol(); 577 case tok::objc_end: 578 return; // Handled by the caller. 579 case tok::objc_optional: 580 case tok::objc_required: 581 nextToken(); 582 addUnwrappedLine(); 583 return; 584 default: 585 break; 586 } 587 break; 588 case tok::kw_namespace: 589 parseNamespace(); 590 return; 591 case tok::kw_inline: 592 nextToken(); 593 if (FormatTok->Tok.is(tok::kw_namespace)) { 594 parseNamespace(); 595 return; 596 } 597 break; 598 case tok::kw_public: 599 case tok::kw_protected: 600 case tok::kw_private: 601 parseAccessSpecifier(); 602 return; 603 case tok::kw_if: 604 parseIfThenElse(); 605 return; 606 case tok::kw_for: 607 case tok::kw_while: 608 parseForOrWhileLoop(); 609 return; 610 case tok::kw_do: 611 parseDoWhile(); 612 return; 613 case tok::kw_switch: 614 parseSwitch(); 615 return; 616 case tok::kw_default: 617 nextToken(); 618 parseLabel(); 619 return; 620 case tok::kw_case: 621 parseCaseLabel(); 622 return; 623 case tok::kw_return: 624 parseReturn(); 625 return; 626 case tok::kw_extern: 627 nextToken(); 628 if (FormatTok->Tok.is(tok::string_literal)) { 629 nextToken(); 630 if (FormatTok->Tok.is(tok::l_brace)) { 631 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/false); 632 addUnwrappedLine(); 633 return; 634 } 635 } 636 // In all other cases, parse the declaration. 637 break; 638 default: 639 break; 640 } 641 do { 642 switch (FormatTok->Tok.getKind()) { 643 case tok::at: 644 nextToken(); 645 if (FormatTok->Tok.is(tok::l_brace)) 646 parseBracedList(); 647 break; 648 case tok::kw_enum: 649 parseEnum(); 650 break; 651 case tok::kw_struct: 652 case tok::kw_union: 653 case tok::kw_class: 654 parseRecord(); 655 // A record declaration or definition is always the start of a structural 656 // element. 657 break; 658 case tok::semi: 659 nextToken(); 660 addUnwrappedLine(); 661 return; 662 case tok::r_brace: 663 addUnwrappedLine(); 664 return; 665 case tok::l_paren: 666 parseParens(); 667 break; 668 case tok::caret: 669 nextToken(); 670 if (FormatTok->is(tok::l_brace)) { 671 parseChildBlock(); 672 } 673 break; 674 case tok::l_brace: 675 if (!tryToParseBracedList()) { 676 // A block outside of parentheses must be the last part of a 677 // structural element. 678 // FIXME: Figure out cases where this is not true, and add projections 679 // for them (the one we know is missing are lambdas). 680 if (Style.BreakBeforeBraces == FormatStyle::BS_Linux || 681 Style.BreakBeforeBraces == FormatStyle::BS_Stroustrup || 682 Style.BreakBeforeBraces == FormatStyle::BS_Allman) 683 addUnwrappedLine(); 684 FormatTok->Type = TT_FunctionLBrace; 685 parseBlock(/*MustBeDeclaration=*/false); 686 addUnwrappedLine(); 687 return; 688 } 689 // Otherwise this was a braced init list, and the structural 690 // element continues. 691 break; 692 case tok::identifier: { 693 StringRef Text = FormatTok->TokenText; 694 nextToken(); 695 if (Line->Tokens.size() == 1) { 696 if (FormatTok->Tok.is(tok::colon)) { 697 parseLabel(); 698 return; 699 } 700 // Recognize function-like macro usages without trailing semicolon. 701 if (FormatTok->Tok.is(tok::l_paren)) { 702 parseParens(); 703 if (FormatTok->HasUnescapedNewline && 704 tokenCanStartNewLine(FormatTok->Tok)) { 705 addUnwrappedLine(); 706 return; 707 } 708 } else if (FormatTok->HasUnescapedNewline && Text.size() >= 5 && 709 Text == Text.upper()) { 710 // Recognize free-standing macros like Q_OBJECT. 711 addUnwrappedLine(); 712 return; 713 } 714 } 715 break; 716 } 717 case tok::equal: 718 nextToken(); 719 if (FormatTok->Tok.is(tok::l_brace)) { 720 parseBracedList(); 721 } 722 break; 723 case tok::l_square: 724 tryToParseLambda(); 725 break; 726 default: 727 nextToken(); 728 break; 729 } 730 } while (!eof()); 731 } 732 733 void UnwrappedLineParser::tryToParseLambda() { 734 // FIXME: This is a dirty way to access the previous token. Find a better 735 // solution. 736 if (!Line->Tokens.empty() && 737 Line->Tokens.back().Tok->isOneOf(tok::identifier, tok::kw_operator)) { 738 nextToken(); 739 return; 740 } 741 assert(FormatTok->is(tok::l_square)); 742 FormatToken &LSquare = *FormatTok; 743 if (!tryToParseLambdaIntroducer()) 744 return; 745 746 while (FormatTok->isNot(tok::l_brace)) { 747 switch (FormatTok->Tok.getKind()) { 748 case tok::l_brace: 749 break; 750 case tok::l_paren: 751 parseParens(); 752 break; 753 case tok::identifier: 754 case tok::kw_mutable: 755 nextToken(); 756 break; 757 default: 758 return; 759 } 760 } 761 LSquare.Type = TT_LambdaLSquare; 762 parseChildBlock(); 763 } 764 765 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 766 nextToken(); 767 if (FormatTok->is(tok::equal)) { 768 nextToken(); 769 if (FormatTok->is(tok::r_square)) { 770 nextToken(); 771 return true; 772 } 773 if (FormatTok->isNot(tok::comma)) 774 return false; 775 nextToken(); 776 } else if (FormatTok->is(tok::amp)) { 777 nextToken(); 778 if (FormatTok->is(tok::r_square)) { 779 nextToken(); 780 return true; 781 } 782 if (!FormatTok->isOneOf(tok::comma, tok::identifier)) { 783 return false; 784 } 785 if (FormatTok->is(tok::comma)) 786 nextToken(); 787 } else if (FormatTok->is(tok::r_square)) { 788 nextToken(); 789 return true; 790 } 791 do { 792 if (FormatTok->is(tok::amp)) 793 nextToken(); 794 if (!FormatTok->isOneOf(tok::identifier, tok::kw_this)) 795 return false; 796 nextToken(); 797 if (FormatTok->is(tok::comma)) { 798 nextToken(); 799 } else if (FormatTok->is(tok::r_square)) { 800 nextToken(); 801 return true; 802 } else { 803 return false; 804 } 805 } while (!eof()); 806 return false; 807 } 808 809 bool UnwrappedLineParser::tryToParseBracedList() { 810 if (FormatTok->BlockKind == BK_Unknown) 811 calculateBraceTypes(); 812 assert(FormatTok->BlockKind != BK_Unknown); 813 if (FormatTok->BlockKind == BK_Block) 814 return false; 815 parseBracedList(); 816 return true; 817 } 818 819 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons) { 820 bool HasError = false; 821 nextToken(); 822 823 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 824 // replace this by using parseAssigmentExpression() inside. 825 do { 826 // FIXME: When we start to support lambdas, we'll want to parse them away 827 // here, otherwise our bail-out scenarios below break. The better solution 828 // might be to just implement a more or less complete expression parser. 829 switch (FormatTok->Tok.getKind()) { 830 case tok::caret: 831 nextToken(); 832 if (FormatTok->is(tok::l_brace)) { 833 parseChildBlock(); 834 } 835 break; 836 case tok::l_square: 837 tryToParseLambda(); 838 break; 839 case tok::l_brace: 840 // Assume there are no blocks inside a braced init list apart 841 // from the ones we explicitly parse out (like lambdas). 842 FormatTok->BlockKind = BK_BracedInit; 843 parseBracedList(); 844 break; 845 case tok::r_brace: 846 nextToken(); 847 return !HasError; 848 case tok::semi: 849 HasError = true; 850 if (!ContinueOnSemicolons) 851 return !HasError; 852 nextToken(); 853 break; 854 case tok::comma: 855 nextToken(); 856 break; 857 default: 858 nextToken(); 859 break; 860 } 861 } while (!eof()); 862 return false; 863 } 864 865 void UnwrappedLineParser::parseReturn() { 866 nextToken(); 867 868 do { 869 switch (FormatTok->Tok.getKind()) { 870 case tok::l_brace: 871 parseBracedList(); 872 if (FormatTok->Tok.isNot(tok::semi)) { 873 // Assume missing ';'. 874 addUnwrappedLine(); 875 return; 876 } 877 break; 878 case tok::l_paren: 879 parseParens(); 880 break; 881 case tok::r_brace: 882 // Assume missing ';'. 883 addUnwrappedLine(); 884 return; 885 case tok::semi: 886 nextToken(); 887 addUnwrappedLine(); 888 return; 889 case tok::l_square: 890 tryToParseLambda(); 891 break; 892 default: 893 nextToken(); 894 break; 895 } 896 } while (!eof()); 897 } 898 899 void UnwrappedLineParser::parseParens() { 900 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 901 nextToken(); 902 do { 903 switch (FormatTok->Tok.getKind()) { 904 case tok::l_paren: 905 parseParens(); 906 break; 907 case tok::r_paren: 908 nextToken(); 909 return; 910 case tok::r_brace: 911 // A "}" inside parenthesis is an error if there wasn't a matching "{". 912 return; 913 case tok::l_square: 914 tryToParseLambda(); 915 break; 916 case tok::l_brace: { 917 if (!tryToParseBracedList()) { 918 parseChildBlock(); 919 } 920 break; 921 } 922 case tok::at: 923 nextToken(); 924 if (FormatTok->Tok.is(tok::l_brace)) 925 parseBracedList(); 926 break; 927 default: 928 nextToken(); 929 break; 930 } 931 } while (!eof()); 932 } 933 934 void UnwrappedLineParser::parseIfThenElse() { 935 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 936 nextToken(); 937 if (FormatTok->Tok.is(tok::l_paren)) 938 parseParens(); 939 bool NeedsUnwrappedLine = false; 940 if (FormatTok->Tok.is(tok::l_brace)) { 941 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman) 942 addUnwrappedLine(); 943 parseBlock(/*MustBeDeclaration=*/false); 944 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman) 945 addUnwrappedLine(); 946 else 947 NeedsUnwrappedLine = true; 948 } else { 949 addUnwrappedLine(); 950 ++Line->Level; 951 parseStructuralElement(); 952 --Line->Level; 953 } 954 if (FormatTok->Tok.is(tok::kw_else)) { 955 nextToken(); 956 if (FormatTok->Tok.is(tok::l_brace)) { 957 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman) 958 addUnwrappedLine(); 959 parseBlock(/*MustBeDeclaration=*/false); 960 addUnwrappedLine(); 961 } else if (FormatTok->Tok.is(tok::kw_if)) { 962 parseIfThenElse(); 963 } else { 964 addUnwrappedLine(); 965 ++Line->Level; 966 parseStructuralElement(); 967 --Line->Level; 968 } 969 } else if (NeedsUnwrappedLine) { 970 addUnwrappedLine(); 971 } 972 } 973 974 void UnwrappedLineParser::parseNamespace() { 975 assert(FormatTok->Tok.is(tok::kw_namespace) && "'namespace' expected"); 976 nextToken(); 977 if (FormatTok->Tok.is(tok::identifier)) 978 nextToken(); 979 if (FormatTok->Tok.is(tok::l_brace)) { 980 if (Style.BreakBeforeBraces == FormatStyle::BS_Linux || 981 Style.BreakBeforeBraces == FormatStyle::BS_Allman) 982 addUnwrappedLine(); 983 984 bool AddLevel = Style.NamespaceIndentation == FormatStyle::NI_All || 985 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 986 DeclarationScopeStack.size() > 1); 987 parseBlock(/*MustBeDeclaration=*/true, AddLevel); 988 // Munch the semicolon after a namespace. This is more common than one would 989 // think. Puttin the semicolon into its own line is very ugly. 990 if (FormatTok->Tok.is(tok::semi)) 991 nextToken(); 992 addUnwrappedLine(); 993 } 994 // FIXME: Add error handling. 995 } 996 997 void UnwrappedLineParser::parseForOrWhileLoop() { 998 assert((FormatTok->Tok.is(tok::kw_for) || FormatTok->Tok.is(tok::kw_while)) && 999 "'for' or 'while' expected"); 1000 nextToken(); 1001 if (FormatTok->Tok.is(tok::l_paren)) 1002 parseParens(); 1003 if (FormatTok->Tok.is(tok::l_brace)) { 1004 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman) 1005 addUnwrappedLine(); 1006 parseBlock(/*MustBeDeclaration=*/false); 1007 addUnwrappedLine(); 1008 } else { 1009 addUnwrappedLine(); 1010 ++Line->Level; 1011 parseStructuralElement(); 1012 --Line->Level; 1013 } 1014 } 1015 1016 void UnwrappedLineParser::parseDoWhile() { 1017 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 1018 nextToken(); 1019 if (FormatTok->Tok.is(tok::l_brace)) { 1020 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman) 1021 addUnwrappedLine(); 1022 parseBlock(/*MustBeDeclaration=*/false); 1023 } else { 1024 addUnwrappedLine(); 1025 ++Line->Level; 1026 parseStructuralElement(); 1027 --Line->Level; 1028 } 1029 1030 // FIXME: Add error handling. 1031 if (!FormatTok->Tok.is(tok::kw_while)) { 1032 addUnwrappedLine(); 1033 return; 1034 } 1035 1036 nextToken(); 1037 parseStructuralElement(); 1038 } 1039 1040 void UnwrappedLineParser::parseLabel() { 1041 nextToken(); 1042 unsigned OldLineLevel = Line->Level; 1043 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 1044 --Line->Level; 1045 if (CommentsBeforeNextToken.empty() && FormatTok->Tok.is(tok::l_brace)) { 1046 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman) 1047 addUnwrappedLine(); 1048 parseBlock(/*MustBeDeclaration=*/false); 1049 if (FormatTok->Tok.is(tok::kw_break)) { 1050 // "break;" after "}" on its own line only for BS_Allman 1051 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman) 1052 addUnwrappedLine(); 1053 parseStructuralElement(); 1054 } 1055 } 1056 addUnwrappedLine(); 1057 Line->Level = OldLineLevel; 1058 } 1059 1060 void UnwrappedLineParser::parseCaseLabel() { 1061 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 1062 // FIXME: fix handling of complex expressions here. 1063 do { 1064 nextToken(); 1065 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 1066 parseLabel(); 1067 } 1068 1069 void UnwrappedLineParser::parseSwitch() { 1070 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 1071 nextToken(); 1072 if (FormatTok->Tok.is(tok::l_paren)) 1073 parseParens(); 1074 if (FormatTok->Tok.is(tok::l_brace)) { 1075 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman) 1076 addUnwrappedLine(); 1077 parseBlock(/*MustBeDeclaration=*/false); 1078 addUnwrappedLine(); 1079 } else { 1080 addUnwrappedLine(); 1081 ++Line->Level; 1082 parseStructuralElement(); 1083 --Line->Level; 1084 } 1085 } 1086 1087 void UnwrappedLineParser::parseAccessSpecifier() { 1088 nextToken(); 1089 // Otherwise, we don't know what it is, and we'd better keep the next token. 1090 if (FormatTok->Tok.is(tok::colon)) 1091 nextToken(); 1092 addUnwrappedLine(); 1093 } 1094 1095 void UnwrappedLineParser::parseEnum() { 1096 nextToken(); 1097 // Eat up enum class ... 1098 if (FormatTok->Tok.is(tok::kw_class) || 1099 FormatTok->Tok.is(tok::kw_struct)) 1100 nextToken(); 1101 while (FormatTok->Tok.getIdentifierInfo() || 1102 FormatTok->isOneOf(tok::colon, tok::coloncolon)) { 1103 nextToken(); 1104 // We can have macros or attributes in between 'enum' and the enum name. 1105 if (FormatTok->Tok.is(tok::l_paren)) { 1106 parseParens(); 1107 } 1108 if (FormatTok->Tok.is(tok::identifier)) 1109 nextToken(); 1110 } 1111 if (FormatTok->Tok.is(tok::l_brace)) { 1112 FormatTok->BlockKind = BK_Block; 1113 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true); 1114 if (HasError) { 1115 if (FormatTok->is(tok::semi)) 1116 nextToken(); 1117 addUnwrappedLine(); 1118 } 1119 } 1120 // We fall through to parsing a structural element afterwards, so that in 1121 // enum A {} n, m; 1122 // "} n, m;" will end up in one unwrapped line. 1123 } 1124 1125 void UnwrappedLineParser::parseRecord() { 1126 nextToken(); 1127 if (FormatTok->Tok.is(tok::identifier) || 1128 FormatTok->Tok.is(tok::kw___attribute) || 1129 FormatTok->Tok.is(tok::kw___declspec) || 1130 FormatTok->Tok.is(tok::kw_alignas)) { 1131 nextToken(); 1132 // We can have macros or attributes in between 'class' and the class name. 1133 if (FormatTok->Tok.is(tok::l_paren)) { 1134 parseParens(); 1135 } 1136 // The actual identifier can be a nested name specifier, and in macros 1137 // it is often token-pasted. 1138 while (FormatTok->Tok.is(tok::identifier) || 1139 FormatTok->Tok.is(tok::coloncolon) || 1140 FormatTok->Tok.is(tok::hashhash)) 1141 nextToken(); 1142 1143 // Note that parsing away template declarations here leads to incorrectly 1144 // accepting function declarations as record declarations. 1145 // In general, we cannot solve this problem. Consider: 1146 // class A<int> B() {} 1147 // which can be a function definition or a class definition when B() is a 1148 // macro. If we find enough real-world cases where this is a problem, we 1149 // can parse for the 'template' keyword in the beginning of the statement, 1150 // and thus rule out the record production in case there is no template 1151 // (this would still leave us with an ambiguity between template function 1152 // and class declarations). 1153 if (FormatTok->Tok.is(tok::colon) || FormatTok->Tok.is(tok::less)) { 1154 while (!eof() && FormatTok->Tok.isNot(tok::l_brace)) { 1155 if (FormatTok->Tok.is(tok::semi)) 1156 return; 1157 nextToken(); 1158 } 1159 } 1160 } 1161 if (FormatTok->Tok.is(tok::l_brace)) { 1162 if (Style.BreakBeforeBraces == FormatStyle::BS_Linux || 1163 Style.BreakBeforeBraces == FormatStyle::BS_Allman) 1164 addUnwrappedLine(); 1165 1166 parseBlock(/*MustBeDeclaration=*/true, /*Addlevel=*/true, 1167 /*MunchSemi=*/false); 1168 } 1169 // We fall through to parsing a structural element afterwards, so 1170 // class A {} n, m; 1171 // will end up in one unwrapped line. 1172 } 1173 1174 void UnwrappedLineParser::parseObjCProtocolList() { 1175 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 1176 do 1177 nextToken(); 1178 while (!eof() && FormatTok->Tok.isNot(tok::greater)); 1179 nextToken(); // Skip '>'. 1180 } 1181 1182 void UnwrappedLineParser::parseObjCUntilAtEnd() { 1183 do { 1184 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 1185 nextToken(); 1186 addUnwrappedLine(); 1187 break; 1188 } 1189 if (FormatTok->is(tok::l_brace)) { 1190 parseBlock(/*MustBeDeclaration=*/false); 1191 // In ObjC interfaces, nothing should be following the "}". 1192 addUnwrappedLine(); 1193 } else { 1194 parseStructuralElement(); 1195 } 1196 } while (!eof()); 1197 } 1198 1199 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 1200 nextToken(); 1201 nextToken(); // interface name 1202 1203 // @interface can be followed by either a base class, or a category. 1204 if (FormatTok->Tok.is(tok::colon)) { 1205 nextToken(); 1206 nextToken(); // base class name 1207 } else if (FormatTok->Tok.is(tok::l_paren)) 1208 // Skip category, if present. 1209 parseParens(); 1210 1211 if (FormatTok->Tok.is(tok::less)) 1212 parseObjCProtocolList(); 1213 1214 // If instance variables are present, keep the '{' on the first line too. 1215 if (FormatTok->Tok.is(tok::l_brace)) 1216 parseBlock(/*MustBeDeclaration=*/true); 1217 1218 // With instance variables, this puts '}' on its own line. Without instance 1219 // variables, this ends the @interface line. 1220 addUnwrappedLine(); 1221 1222 parseObjCUntilAtEnd(); 1223 } 1224 1225 void UnwrappedLineParser::parseObjCProtocol() { 1226 nextToken(); 1227 nextToken(); // protocol name 1228 1229 if (FormatTok->Tok.is(tok::less)) 1230 parseObjCProtocolList(); 1231 1232 // Check for protocol declaration. 1233 if (FormatTok->Tok.is(tok::semi)) { 1234 nextToken(); 1235 return addUnwrappedLine(); 1236 } 1237 1238 addUnwrappedLine(); 1239 parseObjCUntilAtEnd(); 1240 } 1241 1242 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 1243 StringRef Prefix = "") { 1244 llvm::dbgs() << Prefix << "Line(" << Line.Level << ")" 1245 << (Line.InPPDirective ? " MACRO" : "") << ": "; 1246 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 1247 E = Line.Tokens.end(); 1248 I != E; ++I) { 1249 llvm::dbgs() << I->Tok->Tok.getName() << "[" << I->Tok->Type << "] "; 1250 } 1251 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 1252 E = Line.Tokens.end(); 1253 I != E; ++I) { 1254 const UnwrappedLineNode &Node = *I; 1255 for (SmallVectorImpl<UnwrappedLine>::const_iterator 1256 I = Node.Children.begin(), 1257 E = Node.Children.end(); 1258 I != E; ++I) { 1259 printDebugInfo(*I, "\nChild: "); 1260 } 1261 } 1262 llvm::dbgs() << "\n"; 1263 } 1264 1265 void UnwrappedLineParser::addUnwrappedLine() { 1266 if (Line->Tokens.empty()) 1267 return; 1268 DEBUG({ 1269 if (CurrentLines == &Lines) 1270 printDebugInfo(*Line); 1271 }); 1272 CurrentLines->push_back(*Line); 1273 Line->Tokens.clear(); 1274 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 1275 for (SmallVectorImpl<UnwrappedLine>::iterator 1276 I = PreprocessorDirectives.begin(), 1277 E = PreprocessorDirectives.end(); 1278 I != E; ++I) { 1279 CurrentLines->push_back(*I); 1280 } 1281 PreprocessorDirectives.clear(); 1282 } 1283 } 1284 1285 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 1286 1287 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 1288 bool JustComments = Line->Tokens.empty(); 1289 for (SmallVectorImpl<FormatToken *>::const_iterator 1290 I = CommentsBeforeNextToken.begin(), 1291 E = CommentsBeforeNextToken.end(); 1292 I != E; ++I) { 1293 if ((*I)->NewlinesBefore && JustComments) { 1294 addUnwrappedLine(); 1295 } 1296 pushToken(*I); 1297 } 1298 if (NewlineBeforeNext && JustComments) { 1299 addUnwrappedLine(); 1300 } 1301 CommentsBeforeNextToken.clear(); 1302 } 1303 1304 void UnwrappedLineParser::nextToken() { 1305 if (eof()) 1306 return; 1307 flushComments(FormatTok->NewlinesBefore > 0); 1308 pushToken(FormatTok); 1309 readToken(); 1310 } 1311 1312 void UnwrappedLineParser::readToken() { 1313 bool CommentsInCurrentLine = true; 1314 do { 1315 FormatTok = Tokens->getNextToken(); 1316 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 1317 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 1318 // If there is an unfinished unwrapped line, we flush the preprocessor 1319 // directives only after that unwrapped line was finished later. 1320 bool SwitchToPreprocessorLines = 1321 !Line->Tokens.empty() && CurrentLines == &Lines; 1322 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 1323 // Comments stored before the preprocessor directive need to be output 1324 // before the preprocessor directive, at the same level as the 1325 // preprocessor directive, as we consider them to apply to the directive. 1326 flushComments(FormatTok->NewlinesBefore > 0); 1327 parsePPDirective(); 1328 } 1329 1330 if (!PPStack.empty() && (PPStack.back() == PP_Unreachable) && 1331 !Line->InPPDirective) { 1332 continue; 1333 } 1334 1335 if (!FormatTok->Tok.is(tok::comment)) 1336 return; 1337 if (FormatTok->NewlinesBefore > 0 || FormatTok->IsFirst) { 1338 CommentsInCurrentLine = false; 1339 } 1340 if (CommentsInCurrentLine) { 1341 pushToken(FormatTok); 1342 } else { 1343 CommentsBeforeNextToken.push_back(FormatTok); 1344 } 1345 } while (!eof()); 1346 } 1347 1348 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 1349 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 1350 if (MustBreakBeforeNextToken) { 1351 Line->Tokens.back().Tok->MustBreakBefore = true; 1352 MustBreakBeforeNextToken = false; 1353 } 1354 } 1355 1356 } // end namespace format 1357 } // end namespace clang 1358