1 //===--- UnwrappedLineParser.cpp - Format C++ code ------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// 9 /// \file 10 /// This file contains the implementation of the UnwrappedLineParser, 11 /// which turns a stream of tokens into UnwrappedLines. 12 /// 13 //===----------------------------------------------------------------------===// 14 15 #include "UnwrappedLineParser.h" 16 #include "FormatToken.h" 17 #include "FormatTokenLexer.h" 18 #include "FormatTokenSource.h" 19 #include "Macros.h" 20 #include "TokenAnnotator.h" 21 #include "clang/Basic/TokenKinds.h" 22 #include "llvm/ADT/STLExtras.h" 23 #include "llvm/ADT/StringRef.h" 24 #include "llvm/Support/Debug.h" 25 #include "llvm/Support/raw_os_ostream.h" 26 #include "llvm/Support/raw_ostream.h" 27 28 #include <algorithm> 29 #include <utility> 30 31 #define DEBUG_TYPE "format-parser" 32 33 namespace clang { 34 namespace format { 35 36 namespace { 37 38 void printLine(llvm::raw_ostream &OS, const UnwrappedLine &Line, 39 StringRef Prefix = "", bool PrintText = false) { 40 OS << Prefix << "Line(" << Line.Level << ", FSC=" << Line.FirstStartColumn 41 << ")" << (Line.InPPDirective ? " MACRO" : "") << ": "; 42 bool NewLine = false; 43 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 44 E = Line.Tokens.end(); 45 I != E; ++I) { 46 if (NewLine) { 47 OS << Prefix; 48 NewLine = false; 49 } 50 OS << I->Tok->Tok.getName() << "[" << "T=" << (unsigned)I->Tok->getType() 51 << ", OC=" << I->Tok->OriginalColumn << ", \"" << I->Tok->TokenText 52 << "\"] "; 53 for (SmallVectorImpl<UnwrappedLine>::const_iterator 54 CI = I->Children.begin(), 55 CE = I->Children.end(); 56 CI != CE; ++CI) { 57 OS << "\n"; 58 printLine(OS, *CI, (Prefix + " ").str()); 59 NewLine = true; 60 } 61 } 62 if (!NewLine) 63 OS << "\n"; 64 } 65 66 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line) { 67 printLine(llvm::dbgs(), Line); 68 } 69 70 class ScopedDeclarationState { 71 public: 72 ScopedDeclarationState(UnwrappedLine &Line, llvm::BitVector &Stack, 73 bool MustBeDeclaration) 74 : Line(Line), Stack(Stack) { 75 Line.MustBeDeclaration = MustBeDeclaration; 76 Stack.push_back(MustBeDeclaration); 77 } 78 ~ScopedDeclarationState() { 79 Stack.pop_back(); 80 if (!Stack.empty()) 81 Line.MustBeDeclaration = Stack.back(); 82 else 83 Line.MustBeDeclaration = true; 84 } 85 86 private: 87 UnwrappedLine &Line; 88 llvm::BitVector &Stack; 89 }; 90 91 } // end anonymous namespace 92 93 class ScopedLineState { 94 public: 95 ScopedLineState(UnwrappedLineParser &Parser, 96 bool SwitchToPreprocessorLines = false) 97 : Parser(Parser), OriginalLines(Parser.CurrentLines) { 98 if (SwitchToPreprocessorLines) 99 Parser.CurrentLines = &Parser.PreprocessorDirectives; 100 else if (!Parser.Line->Tokens.empty()) 101 Parser.CurrentLines = &Parser.Line->Tokens.back().Children; 102 PreBlockLine = std::move(Parser.Line); 103 Parser.Line = std::make_unique<UnwrappedLine>(); 104 Parser.Line->Level = PreBlockLine->Level; 105 Parser.Line->PPLevel = PreBlockLine->PPLevel; 106 Parser.Line->InPPDirective = PreBlockLine->InPPDirective; 107 Parser.Line->InMacroBody = PreBlockLine->InMacroBody; 108 } 109 110 ~ScopedLineState() { 111 if (!Parser.Line->Tokens.empty()) 112 Parser.addUnwrappedLine(); 113 assert(Parser.Line->Tokens.empty()); 114 Parser.Line = std::move(PreBlockLine); 115 if (Parser.CurrentLines == &Parser.PreprocessorDirectives) 116 Parser.MustBreakBeforeNextToken = true; 117 Parser.CurrentLines = OriginalLines; 118 } 119 120 private: 121 UnwrappedLineParser &Parser; 122 123 std::unique_ptr<UnwrappedLine> PreBlockLine; 124 SmallVectorImpl<UnwrappedLine> *OriginalLines; 125 }; 126 127 class CompoundStatementIndenter { 128 public: 129 CompoundStatementIndenter(UnwrappedLineParser *Parser, 130 const FormatStyle &Style, unsigned &LineLevel) 131 : CompoundStatementIndenter(Parser, LineLevel, 132 Style.BraceWrapping.AfterControlStatement, 133 Style.BraceWrapping.IndentBraces) {} 134 CompoundStatementIndenter(UnwrappedLineParser *Parser, unsigned &LineLevel, 135 bool WrapBrace, bool IndentBrace) 136 : LineLevel(LineLevel), OldLineLevel(LineLevel) { 137 if (WrapBrace) 138 Parser->addUnwrappedLine(); 139 if (IndentBrace) 140 ++LineLevel; 141 } 142 ~CompoundStatementIndenter() { LineLevel = OldLineLevel; } 143 144 private: 145 unsigned &LineLevel; 146 unsigned OldLineLevel; 147 }; 148 149 UnwrappedLineParser::UnwrappedLineParser( 150 SourceManager &SourceMgr, const FormatStyle &Style, 151 const AdditionalKeywords &Keywords, unsigned FirstStartColumn, 152 ArrayRef<FormatToken *> Tokens, UnwrappedLineConsumer &Callback, 153 llvm::SpecificBumpPtrAllocator<FormatToken> &Allocator, 154 IdentifierTable &IdentTable) 155 : Line(new UnwrappedLine), MustBreakBeforeNextToken(false), 156 CurrentLines(&Lines), Style(Style), Keywords(Keywords), 157 CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr), 158 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1), 159 IncludeGuard(Style.IndentPPDirectives == FormatStyle::PPDIS_None 160 ? IG_Rejected 161 : IG_Inited), 162 IncludeGuardToken(nullptr), FirstStartColumn(FirstStartColumn), 163 Macros(Style.Macros, SourceMgr, Style, Allocator, IdentTable) {} 164 165 void UnwrappedLineParser::reset() { 166 PPBranchLevel = -1; 167 IncludeGuard = Style.IndentPPDirectives == FormatStyle::PPDIS_None 168 ? IG_Rejected 169 : IG_Inited; 170 IncludeGuardToken = nullptr; 171 Line.reset(new UnwrappedLine); 172 CommentsBeforeNextToken.clear(); 173 FormatTok = nullptr; 174 MustBreakBeforeNextToken = false; 175 IsDecltypeAutoFunction = false; 176 PreprocessorDirectives.clear(); 177 CurrentLines = &Lines; 178 DeclarationScopeStack.clear(); 179 NestedTooDeep.clear(); 180 NestedLambdas.clear(); 181 PPStack.clear(); 182 Line->FirstStartColumn = FirstStartColumn; 183 184 if (!Unexpanded.empty()) 185 for (FormatToken *Token : AllTokens) 186 Token->MacroCtx.reset(); 187 CurrentExpandedLines.clear(); 188 ExpandedLines.clear(); 189 Unexpanded.clear(); 190 InExpansion = false; 191 Reconstruct.reset(); 192 } 193 194 void UnwrappedLineParser::parse() { 195 IndexedTokenSource TokenSource(AllTokens); 196 Line->FirstStartColumn = FirstStartColumn; 197 do { 198 LLVM_DEBUG(llvm::dbgs() << "----\n"); 199 reset(); 200 Tokens = &TokenSource; 201 TokenSource.reset(); 202 203 readToken(); 204 parseFile(); 205 206 // If we found an include guard then all preprocessor directives (other than 207 // the guard) are over-indented by one. 208 if (IncludeGuard == IG_Found) { 209 for (auto &Line : Lines) 210 if (Line.InPPDirective && Line.Level > 0) 211 --Line.Level; 212 } 213 214 // Create line with eof token. 215 assert(eof()); 216 pushToken(FormatTok); 217 addUnwrappedLine(); 218 219 // In a first run, format everything with the lines containing macro calls 220 // replaced by the expansion. 221 if (!ExpandedLines.empty()) { 222 LLVM_DEBUG(llvm::dbgs() << "Expanded lines:\n"); 223 for (const auto &Line : Lines) { 224 if (!Line.Tokens.empty()) { 225 auto it = ExpandedLines.find(Line.Tokens.begin()->Tok); 226 if (it != ExpandedLines.end()) { 227 for (const auto &Expanded : it->second) { 228 LLVM_DEBUG(printDebugInfo(Expanded)); 229 Callback.consumeUnwrappedLine(Expanded); 230 } 231 continue; 232 } 233 } 234 LLVM_DEBUG(printDebugInfo(Line)); 235 Callback.consumeUnwrappedLine(Line); 236 } 237 Callback.finishRun(); 238 } 239 240 LLVM_DEBUG(llvm::dbgs() << "Unwrapped lines:\n"); 241 for (const UnwrappedLine &Line : Lines) { 242 LLVM_DEBUG(printDebugInfo(Line)); 243 Callback.consumeUnwrappedLine(Line); 244 } 245 Callback.finishRun(); 246 Lines.clear(); 247 while (!PPLevelBranchIndex.empty() && 248 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) { 249 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1); 250 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1); 251 } 252 if (!PPLevelBranchIndex.empty()) { 253 ++PPLevelBranchIndex.back(); 254 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size()); 255 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back()); 256 } 257 } while (!PPLevelBranchIndex.empty()); 258 } 259 260 void UnwrappedLineParser::parseFile() { 261 // The top-level context in a file always has declarations, except for pre- 262 // processor directives and JavaScript files. 263 bool MustBeDeclaration = !Line->InPPDirective && !Style.isJavaScript(); 264 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 265 MustBeDeclaration); 266 if (Style.Language == FormatStyle::LK_TextProto) 267 parseBracedList(); 268 else 269 parseLevel(); 270 // Make sure to format the remaining tokens. 271 // 272 // LK_TextProto is special since its top-level is parsed as the body of a 273 // braced list, which does not necessarily have natural line separators such 274 // as a semicolon. Comments after the last entry that have been determined to 275 // not belong to that line, as in: 276 // key: value 277 // // endfile comment 278 // do not have a chance to be put on a line of their own until this point. 279 // Here we add this newline before end-of-file comments. 280 if (Style.Language == FormatStyle::LK_TextProto && 281 !CommentsBeforeNextToken.empty()) { 282 addUnwrappedLine(); 283 } 284 flushComments(true); 285 addUnwrappedLine(); 286 } 287 288 void UnwrappedLineParser::parseCSharpGenericTypeConstraint() { 289 do { 290 switch (FormatTok->Tok.getKind()) { 291 case tok::l_brace: 292 return; 293 default: 294 if (FormatTok->is(Keywords.kw_where)) { 295 addUnwrappedLine(); 296 nextToken(); 297 parseCSharpGenericTypeConstraint(); 298 break; 299 } 300 nextToken(); 301 break; 302 } 303 } while (!eof()); 304 } 305 306 void UnwrappedLineParser::parseCSharpAttribute() { 307 int UnpairedSquareBrackets = 1; 308 do { 309 switch (FormatTok->Tok.getKind()) { 310 case tok::r_square: 311 nextToken(); 312 --UnpairedSquareBrackets; 313 if (UnpairedSquareBrackets == 0) { 314 addUnwrappedLine(); 315 return; 316 } 317 break; 318 case tok::l_square: 319 ++UnpairedSquareBrackets; 320 nextToken(); 321 break; 322 default: 323 nextToken(); 324 break; 325 } 326 } while (!eof()); 327 } 328 329 bool UnwrappedLineParser::precededByCommentOrPPDirective() const { 330 if (!Lines.empty() && Lines.back().InPPDirective) 331 return true; 332 333 const FormatToken *Previous = Tokens->getPreviousToken(); 334 return Previous && Previous->is(tok::comment) && 335 (Previous->IsMultiline || Previous->NewlinesBefore > 0); 336 } 337 338 /// \brief Parses a level, that is ???. 339 /// \param OpeningBrace Opening brace (\p nullptr if absent) of that level. 340 /// \param IfKind The \p if statement kind in the level. 341 /// \param IfLeftBrace The left brace of the \p if block in the level. 342 /// \returns true if a simple block of if/else/for/while, or false otherwise. 343 /// (A simple block has a single statement.) 344 bool UnwrappedLineParser::parseLevel(const FormatToken *OpeningBrace, 345 IfStmtKind *IfKind, 346 FormatToken **IfLeftBrace) { 347 const bool InRequiresExpression = 348 OpeningBrace && OpeningBrace->is(TT_RequiresExpressionLBrace); 349 const bool IsPrecededByCommentOrPPDirective = 350 !Style.RemoveBracesLLVM || precededByCommentOrPPDirective(); 351 FormatToken *IfLBrace = nullptr; 352 bool HasDoWhile = false; 353 bool HasLabel = false; 354 unsigned StatementCount = 0; 355 bool SwitchLabelEncountered = false; 356 357 do { 358 if (FormatTok->isAttribute()) { 359 nextToken(); 360 continue; 361 } 362 tok::TokenKind kind = FormatTok->Tok.getKind(); 363 if (FormatTok->getType() == TT_MacroBlockBegin) 364 kind = tok::l_brace; 365 else if (FormatTok->getType() == TT_MacroBlockEnd) 366 kind = tok::r_brace; 367 368 auto ParseDefault = [this, OpeningBrace, IfKind, &IfLBrace, &HasDoWhile, 369 &HasLabel, &StatementCount] { 370 parseStructuralElement(OpeningBrace, IfKind, &IfLBrace, 371 HasDoWhile ? nullptr : &HasDoWhile, 372 HasLabel ? nullptr : &HasLabel); 373 ++StatementCount; 374 assert(StatementCount > 0 && "StatementCount overflow!"); 375 }; 376 377 switch (kind) { 378 case tok::comment: 379 nextToken(); 380 addUnwrappedLine(); 381 break; 382 case tok::l_brace: 383 if (InRequiresExpression) { 384 FormatTok->setFinalizedType(TT_RequiresExpressionLBrace); 385 } else if (FormatTok->Previous && 386 FormatTok->Previous->ClosesRequiresClause) { 387 // We need the 'default' case here to correctly parse a function 388 // l_brace. 389 ParseDefault(); 390 continue; 391 } 392 if (!InRequiresExpression && FormatTok->isNot(TT_MacroBlockBegin) && 393 tryToParseBracedList()) { 394 continue; 395 } 396 parseBlock(); 397 ++StatementCount; 398 assert(StatementCount > 0 && "StatementCount overflow!"); 399 addUnwrappedLine(); 400 break; 401 case tok::r_brace: 402 if (OpeningBrace) { 403 if (!Style.RemoveBracesLLVM || Line->InPPDirective || 404 !OpeningBrace->isOneOf(TT_ControlStatementLBrace, TT_ElseLBrace)) { 405 return false; 406 } 407 if (FormatTok->isNot(tok::r_brace) || StatementCount != 1 || HasLabel || 408 HasDoWhile || IsPrecededByCommentOrPPDirective || 409 precededByCommentOrPPDirective()) { 410 return false; 411 } 412 const FormatToken *Next = Tokens->peekNextToken(); 413 if (Next->is(tok::comment) && Next->NewlinesBefore == 0) 414 return false; 415 if (IfLeftBrace) 416 *IfLeftBrace = IfLBrace; 417 return true; 418 } 419 nextToken(); 420 addUnwrappedLine(); 421 break; 422 case tok::kw_default: { 423 unsigned StoredPosition = Tokens->getPosition(); 424 FormatToken *Next; 425 do { 426 Next = Tokens->getNextToken(); 427 assert(Next); 428 } while (Next->is(tok::comment)); 429 FormatTok = Tokens->setPosition(StoredPosition); 430 if (Next->isNot(tok::colon)) { 431 // default not followed by ':' is not a case label; treat it like 432 // an identifier. 433 parseStructuralElement(); 434 break; 435 } 436 // Else, if it is 'default:', fall through to the case handling. 437 [[fallthrough]]; 438 } 439 case tok::kw_case: 440 if (Style.Language == FormatStyle::LK_Proto || Style.isVerilog() || 441 (Style.isJavaScript() && Line->MustBeDeclaration)) { 442 // Proto: there are no switch/case statements 443 // Verilog: Case labels don't have this word. We handle case 444 // labels including default in TokenAnnotator. 445 // JavaScript: A 'case: string' style field declaration. 446 ParseDefault(); 447 break; 448 } 449 if (!SwitchLabelEncountered && 450 (Style.IndentCaseLabels || 451 (Line->InPPDirective && Line->Level == 1))) { 452 ++Line->Level; 453 } 454 SwitchLabelEncountered = true; 455 parseStructuralElement(); 456 break; 457 case tok::l_square: 458 if (Style.isCSharp()) { 459 nextToken(); 460 parseCSharpAttribute(); 461 break; 462 } 463 if (handleCppAttributes()) 464 break; 465 [[fallthrough]]; 466 default: 467 ParseDefault(); 468 break; 469 } 470 } while (!eof()); 471 472 return false; 473 } 474 475 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) { 476 // We'll parse forward through the tokens until we hit 477 // a closing brace or eof - note that getNextToken() will 478 // parse macros, so this will magically work inside macro 479 // definitions, too. 480 unsigned StoredPosition = Tokens->getPosition(); 481 FormatToken *Tok = FormatTok; 482 const FormatToken *PrevTok = Tok->Previous; 483 // Keep a stack of positions of lbrace tokens. We will 484 // update information about whether an lbrace starts a 485 // braced init list or a different block during the loop. 486 struct StackEntry { 487 FormatToken *Tok; 488 const FormatToken *PrevTok; 489 }; 490 SmallVector<StackEntry, 8> LBraceStack; 491 assert(Tok->is(tok::l_brace)); 492 do { 493 // Get next non-comment, non-preprocessor token. 494 FormatToken *NextTok; 495 do { 496 NextTok = Tokens->getNextToken(); 497 } while (NextTok->is(tok::comment)); 498 while (NextTok->is(tok::hash) && !Line->InMacroBody) { 499 NextTok = Tokens->getNextToken(); 500 do { 501 NextTok = Tokens->getNextToken(); 502 } while (NextTok->is(tok::comment) || 503 (NextTok->NewlinesBefore == 0 && NextTok->isNot(tok::eof))); 504 } 505 506 switch (Tok->Tok.getKind()) { 507 case tok::l_brace: 508 if (Style.isJavaScript() && PrevTok) { 509 if (PrevTok->isOneOf(tok::colon, tok::less)) { 510 // A ':' indicates this code is in a type, or a braced list 511 // following a label in an object literal ({a: {b: 1}}). 512 // A '<' could be an object used in a comparison, but that is nonsense 513 // code (can never return true), so more likely it is a generic type 514 // argument (`X<{a: string; b: number}>`). 515 // The code below could be confused by semicolons between the 516 // individual members in a type member list, which would normally 517 // trigger BK_Block. In both cases, this must be parsed as an inline 518 // braced init. 519 Tok->setBlockKind(BK_BracedInit); 520 } else if (PrevTok->is(tok::r_paren)) { 521 // `) { }` can only occur in function or method declarations in JS. 522 Tok->setBlockKind(BK_Block); 523 } 524 } else { 525 Tok->setBlockKind(BK_Unknown); 526 } 527 LBraceStack.push_back({Tok, PrevTok}); 528 break; 529 case tok::r_brace: 530 if (LBraceStack.empty()) 531 break; 532 if (LBraceStack.back().Tok->is(BK_Unknown)) { 533 bool ProbablyBracedList = false; 534 if (Style.Language == FormatStyle::LK_Proto) { 535 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square); 536 } else { 537 // Skip NextTok over preprocessor lines, otherwise we may not 538 // properly diagnose the block as a braced intializer 539 // if the comma separator appears after the pp directive. 540 while (NextTok->is(tok::hash)) { 541 ScopedMacroState MacroState(*Line, Tokens, NextTok); 542 do { 543 NextTok = Tokens->getNextToken(); 544 } while (NextTok->isNot(tok::eof)); 545 } 546 547 // Using OriginalColumn to distinguish between ObjC methods and 548 // binary operators is a bit hacky. 549 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 550 NextTok->OriginalColumn == 0; 551 552 // Try to detect a braced list. Note that regardless how we mark inner 553 // braces here, we will overwrite the BlockKind later if we parse a 554 // braced list (where all blocks inside are by default braced lists), 555 // or when we explicitly detect blocks (for example while parsing 556 // lambdas). 557 558 // If we already marked the opening brace as braced list, the closing 559 // must also be part of it. 560 ProbablyBracedList = LBraceStack.back().Tok->is(TT_BracedListLBrace); 561 562 ProbablyBracedList = ProbablyBracedList || 563 (Style.isJavaScript() && 564 NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in, 565 Keywords.kw_as)); 566 ProbablyBracedList = ProbablyBracedList || 567 (Style.isCpp() && NextTok->is(tok::l_paren)); 568 569 // If there is a comma, semicolon or right paren after the closing 570 // brace, we assume this is a braced initializer list. 571 // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a 572 // braced list in JS. 573 ProbablyBracedList = 574 ProbablyBracedList || 575 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 576 tok::r_paren, tok::r_square, tok::ellipsis); 577 578 // Distinguish between braced list in a constructor initializer list 579 // followed by constructor body, or just adjacent blocks. 580 ProbablyBracedList = 581 ProbablyBracedList || 582 (NextTok->is(tok::l_brace) && LBraceStack.back().PrevTok && 583 LBraceStack.back().PrevTok->isOneOf(tok::identifier, 584 tok::greater)); 585 586 ProbablyBracedList = 587 ProbablyBracedList || 588 (NextTok->is(tok::identifier) && 589 !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace)); 590 591 ProbablyBracedList = ProbablyBracedList || 592 (NextTok->is(tok::semi) && 593 (!ExpectClassBody || LBraceStack.size() != 1)); 594 595 ProbablyBracedList = 596 ProbablyBracedList || 597 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 598 599 if (!Style.isCSharp() && NextTok->is(tok::l_square)) { 600 // We can have an array subscript after a braced init 601 // list, but C++11 attributes are expected after blocks. 602 NextTok = Tokens->getNextToken(); 603 ProbablyBracedList = NextTok->isNot(tok::l_square); 604 } 605 } 606 if (ProbablyBracedList) { 607 Tok->setBlockKind(BK_BracedInit); 608 LBraceStack.back().Tok->setBlockKind(BK_BracedInit); 609 } else { 610 Tok->setBlockKind(BK_Block); 611 LBraceStack.back().Tok->setBlockKind(BK_Block); 612 } 613 } 614 LBraceStack.pop_back(); 615 break; 616 case tok::identifier: 617 if (Tok->isNot(TT_StatementMacro)) 618 break; 619 [[fallthrough]]; 620 case tok::at: 621 case tok::semi: 622 case tok::kw_if: 623 case tok::kw_while: 624 case tok::kw_for: 625 case tok::kw_switch: 626 case tok::kw_try: 627 case tok::kw___try: 628 if (!LBraceStack.empty() && LBraceStack.back().Tok->is(BK_Unknown)) 629 LBraceStack.back().Tok->setBlockKind(BK_Block); 630 break; 631 default: 632 break; 633 } 634 PrevTok = Tok; 635 Tok = NextTok; 636 } while (Tok->isNot(tok::eof) && !LBraceStack.empty()); 637 638 // Assume other blocks for all unclosed opening braces. 639 for (const auto &Entry : LBraceStack) 640 if (Entry.Tok->is(BK_Unknown)) 641 Entry.Tok->setBlockKind(BK_Block); 642 643 FormatTok = Tokens->setPosition(StoredPosition); 644 } 645 646 // Sets the token type of the directly previous right brace. 647 void UnwrappedLineParser::setPreviousRBraceType(TokenType Type) { 648 if (auto Prev = FormatTok->getPreviousNonComment(); 649 Prev && Prev->is(tok::r_brace)) { 650 Prev->setFinalizedType(Type); 651 } 652 } 653 654 template <class T> 655 static inline void hash_combine(std::size_t &seed, const T &v) { 656 std::hash<T> hasher; 657 seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2); 658 } 659 660 size_t UnwrappedLineParser::computePPHash() const { 661 size_t h = 0; 662 for (const auto &i : PPStack) { 663 hash_combine(h, size_t(i.Kind)); 664 hash_combine(h, i.Line); 665 } 666 return h; 667 } 668 669 // Checks whether \p ParsedLine might fit on a single line. If \p OpeningBrace 670 // is not null, subtracts its length (plus the preceding space) when computing 671 // the length of \p ParsedLine. We must clone the tokens of \p ParsedLine before 672 // running the token annotator on it so that we can restore them afterward. 673 bool UnwrappedLineParser::mightFitOnOneLine( 674 UnwrappedLine &ParsedLine, const FormatToken *OpeningBrace) const { 675 const auto ColumnLimit = Style.ColumnLimit; 676 if (ColumnLimit == 0) 677 return true; 678 679 auto &Tokens = ParsedLine.Tokens; 680 assert(!Tokens.empty()); 681 682 const auto *LastToken = Tokens.back().Tok; 683 assert(LastToken); 684 685 SmallVector<UnwrappedLineNode> SavedTokens(Tokens.size()); 686 687 int Index = 0; 688 for (const auto &Token : Tokens) { 689 assert(Token.Tok); 690 auto &SavedToken = SavedTokens[Index++]; 691 SavedToken.Tok = new FormatToken; 692 SavedToken.Tok->copyFrom(*Token.Tok); 693 SavedToken.Children = std::move(Token.Children); 694 } 695 696 AnnotatedLine Line(ParsedLine); 697 assert(Line.Last == LastToken); 698 699 TokenAnnotator Annotator(Style, Keywords); 700 Annotator.annotate(Line); 701 Annotator.calculateFormattingInformation(Line); 702 703 auto Length = LastToken->TotalLength; 704 if (OpeningBrace) { 705 assert(OpeningBrace != Tokens.front().Tok); 706 if (auto Prev = OpeningBrace->Previous; 707 Prev && Prev->TotalLength + ColumnLimit == OpeningBrace->TotalLength) { 708 Length -= ColumnLimit; 709 } 710 Length -= OpeningBrace->TokenText.size() + 1; 711 } 712 713 if (const auto *FirstToken = Line.First; FirstToken->is(tok::r_brace)) { 714 assert(!OpeningBrace || OpeningBrace->is(TT_ControlStatementLBrace)); 715 Length -= FirstToken->TokenText.size() + 1; 716 } 717 718 Index = 0; 719 for (auto &Token : Tokens) { 720 const auto &SavedToken = SavedTokens[Index++]; 721 Token.Tok->copyFrom(*SavedToken.Tok); 722 Token.Children = std::move(SavedToken.Children); 723 delete SavedToken.Tok; 724 } 725 726 // If these change PPLevel needs to be used for get correct indentation. 727 assert(!Line.InMacroBody); 728 assert(!Line.InPPDirective); 729 return Line.Level * Style.IndentWidth + Length <= ColumnLimit; 730 } 731 732 FormatToken *UnwrappedLineParser::parseBlock(bool MustBeDeclaration, 733 unsigned AddLevels, bool MunchSemi, 734 bool KeepBraces, 735 IfStmtKind *IfKind, 736 bool UnindentWhitesmithsBraces) { 737 auto HandleVerilogBlockLabel = [this]() { 738 // ":" name 739 if (Style.isVerilog() && FormatTok->is(tok::colon)) { 740 nextToken(); 741 if (Keywords.isVerilogIdentifier(*FormatTok)) 742 nextToken(); 743 } 744 }; 745 746 // Whether this is a Verilog-specific block that has a special header like a 747 // module. 748 const bool VerilogHierarchy = 749 Style.isVerilog() && Keywords.isVerilogHierarchy(*FormatTok); 750 assert((FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) || 751 (Style.isVerilog() && 752 (Keywords.isVerilogBegin(*FormatTok) || VerilogHierarchy))) && 753 "'{' or macro block token expected"); 754 FormatToken *Tok = FormatTok; 755 const bool FollowedByComment = Tokens->peekNextToken()->is(tok::comment); 756 auto Index = CurrentLines->size(); 757 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 758 FormatTok->setBlockKind(BK_Block); 759 760 // For Whitesmiths mode, jump to the next level prior to skipping over the 761 // braces. 762 if (!VerilogHierarchy && AddLevels > 0 && 763 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) { 764 ++Line->Level; 765 } 766 767 size_t PPStartHash = computePPHash(); 768 769 const unsigned InitialLevel = Line->Level; 770 if (VerilogHierarchy) { 771 AddLevels += parseVerilogHierarchyHeader(); 772 } else { 773 nextToken(/*LevelDifference=*/AddLevels); 774 HandleVerilogBlockLabel(); 775 } 776 777 // Bail out if there are too many levels. Otherwise, the stack might overflow. 778 if (Line->Level > 300) 779 return nullptr; 780 781 if (MacroBlock && FormatTok->is(tok::l_paren)) 782 parseParens(); 783 784 size_t NbPreprocessorDirectives = 785 !parsingPPDirective() ? PreprocessorDirectives.size() : 0; 786 addUnwrappedLine(); 787 size_t OpeningLineIndex = 788 CurrentLines->empty() 789 ? (UnwrappedLine::kInvalidIndex) 790 : (CurrentLines->size() - 1 - NbPreprocessorDirectives); 791 792 // Whitesmiths is weird here. The brace needs to be indented for the namespace 793 // block, but the block itself may not be indented depending on the style 794 // settings. This allows the format to back up one level in those cases. 795 if (UnindentWhitesmithsBraces) 796 --Line->Level; 797 798 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 799 MustBeDeclaration); 800 if (AddLevels > 0u && Style.BreakBeforeBraces != FormatStyle::BS_Whitesmiths) 801 Line->Level += AddLevels; 802 803 FormatToken *IfLBrace = nullptr; 804 const bool SimpleBlock = parseLevel(Tok, IfKind, &IfLBrace); 805 806 if (eof()) 807 return IfLBrace; 808 809 if (MacroBlock ? FormatTok->isNot(TT_MacroBlockEnd) 810 : FormatTok->isNot(tok::r_brace)) { 811 Line->Level = InitialLevel; 812 FormatTok->setBlockKind(BK_Block); 813 return IfLBrace; 814 } 815 816 if (FormatTok->is(tok::r_brace) && Tok->is(TT_NamespaceLBrace)) 817 FormatTok->setFinalizedType(TT_NamespaceRBrace); 818 819 const bool IsFunctionRBrace = 820 FormatTok->is(tok::r_brace) && Tok->is(TT_FunctionLBrace); 821 822 auto RemoveBraces = [=]() mutable { 823 if (!SimpleBlock) 824 return false; 825 assert(Tok->isOneOf(TT_ControlStatementLBrace, TT_ElseLBrace)); 826 assert(FormatTok->is(tok::r_brace)); 827 const bool WrappedOpeningBrace = !Tok->Previous; 828 if (WrappedOpeningBrace && FollowedByComment) 829 return false; 830 const bool HasRequiredIfBraces = IfLBrace && !IfLBrace->Optional; 831 if (KeepBraces && !HasRequiredIfBraces) 832 return false; 833 if (Tok->isNot(TT_ElseLBrace) || !HasRequiredIfBraces) { 834 const FormatToken *Previous = Tokens->getPreviousToken(); 835 assert(Previous); 836 if (Previous->is(tok::r_brace) && !Previous->Optional) 837 return false; 838 } 839 assert(!CurrentLines->empty()); 840 auto &LastLine = CurrentLines->back(); 841 if (LastLine.Level == InitialLevel + 1 && !mightFitOnOneLine(LastLine)) 842 return false; 843 if (Tok->is(TT_ElseLBrace)) 844 return true; 845 if (WrappedOpeningBrace) { 846 assert(Index > 0); 847 --Index; // The line above the wrapped l_brace. 848 Tok = nullptr; 849 } 850 return mightFitOnOneLine((*CurrentLines)[Index], Tok); 851 }; 852 if (RemoveBraces()) { 853 Tok->MatchingParen = FormatTok; 854 FormatTok->MatchingParen = Tok; 855 } 856 857 size_t PPEndHash = computePPHash(); 858 859 // Munch the closing brace. 860 nextToken(/*LevelDifference=*/-AddLevels); 861 862 // When this is a function block and there is an unnecessary semicolon 863 // afterwards then mark it as optional (so the RemoveSemi pass can get rid of 864 // it later). 865 if (Style.RemoveSemicolon && IsFunctionRBrace) { 866 while (FormatTok->is(tok::semi)) { 867 FormatTok->Optional = true; 868 nextToken(); 869 } 870 } 871 872 HandleVerilogBlockLabel(); 873 874 if (MacroBlock && FormatTok->is(tok::l_paren)) 875 parseParens(); 876 877 Line->Level = InitialLevel; 878 879 if (FormatTok->is(tok::kw_noexcept)) { 880 // A noexcept in a requires expression. 881 nextToken(); 882 } 883 884 if (FormatTok->is(tok::arrow)) { 885 // Following the } or noexcept we can find a trailing return type arrow 886 // as part of an implicit conversion constraint. 887 nextToken(); 888 parseStructuralElement(); 889 } 890 891 if (MunchSemi && FormatTok->is(tok::semi)) 892 nextToken(); 893 894 if (PPStartHash == PPEndHash) { 895 Line->MatchingOpeningBlockLineIndex = OpeningLineIndex; 896 if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) { 897 // Update the opening line to add the forward reference as well 898 (*CurrentLines)[OpeningLineIndex].MatchingClosingBlockLineIndex = 899 CurrentLines->size() - 1; 900 } 901 } 902 903 return IfLBrace; 904 } 905 906 static bool isGoogScope(const UnwrappedLine &Line) { 907 // FIXME: Closure-library specific stuff should not be hard-coded but be 908 // configurable. 909 if (Line.Tokens.size() < 4) 910 return false; 911 auto I = Line.Tokens.begin(); 912 if (I->Tok->TokenText != "goog") 913 return false; 914 ++I; 915 if (I->Tok->isNot(tok::period)) 916 return false; 917 ++I; 918 if (I->Tok->TokenText != "scope") 919 return false; 920 ++I; 921 return I->Tok->is(tok::l_paren); 922 } 923 924 static bool isIIFE(const UnwrappedLine &Line, 925 const AdditionalKeywords &Keywords) { 926 // Look for the start of an immediately invoked anonymous function. 927 // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression 928 // This is commonly done in JavaScript to create a new, anonymous scope. 929 // Example: (function() { ... })() 930 if (Line.Tokens.size() < 3) 931 return false; 932 auto I = Line.Tokens.begin(); 933 if (I->Tok->isNot(tok::l_paren)) 934 return false; 935 ++I; 936 if (I->Tok->isNot(Keywords.kw_function)) 937 return false; 938 ++I; 939 return I->Tok->is(tok::l_paren); 940 } 941 942 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 943 const FormatToken &InitialToken) { 944 tok::TokenKind Kind = InitialToken.Tok.getKind(); 945 if (InitialToken.is(TT_NamespaceMacro)) 946 Kind = tok::kw_namespace; 947 948 switch (Kind) { 949 case tok::kw_namespace: 950 return Style.BraceWrapping.AfterNamespace; 951 case tok::kw_class: 952 return Style.BraceWrapping.AfterClass; 953 case tok::kw_union: 954 return Style.BraceWrapping.AfterUnion; 955 case tok::kw_struct: 956 return Style.BraceWrapping.AfterStruct; 957 case tok::kw_enum: 958 return Style.BraceWrapping.AfterEnum; 959 default: 960 return false; 961 } 962 } 963 964 void UnwrappedLineParser::parseChildBlock() { 965 assert(FormatTok->is(tok::l_brace)); 966 FormatTok->setBlockKind(BK_Block); 967 const FormatToken *OpeningBrace = FormatTok; 968 nextToken(); 969 { 970 bool SkipIndent = (Style.isJavaScript() && 971 (isGoogScope(*Line) || isIIFE(*Line, Keywords))); 972 ScopedLineState LineState(*this); 973 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 974 /*MustBeDeclaration=*/false); 975 Line->Level += SkipIndent ? 0 : 1; 976 parseLevel(OpeningBrace); 977 flushComments(isOnNewLine(*FormatTok)); 978 Line->Level -= SkipIndent ? 0 : 1; 979 } 980 nextToken(); 981 } 982 983 void UnwrappedLineParser::parsePPDirective() { 984 assert(FormatTok->is(tok::hash) && "'#' expected"); 985 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 986 987 nextToken(); 988 989 if (!FormatTok->Tok.getIdentifierInfo()) { 990 parsePPUnknown(); 991 return; 992 } 993 994 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 995 case tok::pp_define: 996 parsePPDefine(); 997 return; 998 case tok::pp_if: 999 parsePPIf(/*IfDef=*/false); 1000 break; 1001 case tok::pp_ifdef: 1002 case tok::pp_ifndef: 1003 parsePPIf(/*IfDef=*/true); 1004 break; 1005 case tok::pp_else: 1006 case tok::pp_elifdef: 1007 case tok::pp_elifndef: 1008 case tok::pp_elif: 1009 parsePPElse(); 1010 break; 1011 case tok::pp_endif: 1012 parsePPEndIf(); 1013 break; 1014 case tok::pp_pragma: 1015 parsePPPragma(); 1016 break; 1017 default: 1018 parsePPUnknown(); 1019 break; 1020 } 1021 } 1022 1023 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 1024 size_t Line = CurrentLines->size(); 1025 if (CurrentLines == &PreprocessorDirectives) 1026 Line += Lines.size(); 1027 1028 if (Unreachable || 1029 (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable)) { 1030 PPStack.push_back({PP_Unreachable, Line}); 1031 } else { 1032 PPStack.push_back({PP_Conditional, Line}); 1033 } 1034 } 1035 1036 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 1037 ++PPBranchLevel; 1038 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 1039 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 1040 PPLevelBranchIndex.push_back(0); 1041 PPLevelBranchCount.push_back(0); 1042 } 1043 PPChainBranchIndex.push(Unreachable ? -1 : 0); 1044 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 1045 conditionalCompilationCondition(Unreachable || Skip); 1046 } 1047 1048 void UnwrappedLineParser::conditionalCompilationAlternative() { 1049 if (!PPStack.empty()) 1050 PPStack.pop_back(); 1051 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 1052 if (!PPChainBranchIndex.empty()) 1053 ++PPChainBranchIndex.top(); 1054 conditionalCompilationCondition( 1055 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 1056 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 1057 } 1058 1059 void UnwrappedLineParser::conditionalCompilationEnd() { 1060 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 1061 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 1062 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) 1063 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 1064 } 1065 // Guard against #endif's without #if. 1066 if (PPBranchLevel > -1) 1067 --PPBranchLevel; 1068 if (!PPChainBranchIndex.empty()) 1069 PPChainBranchIndex.pop(); 1070 if (!PPStack.empty()) 1071 PPStack.pop_back(); 1072 } 1073 1074 void UnwrappedLineParser::parsePPIf(bool IfDef) { 1075 bool IfNDef = FormatTok->is(tok::pp_ifndef); 1076 nextToken(); 1077 bool Unreachable = false; 1078 if (!IfDef && (FormatTok->is(tok::kw_false) || FormatTok->TokenText == "0")) 1079 Unreachable = true; 1080 if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG") 1081 Unreachable = true; 1082 conditionalCompilationStart(Unreachable); 1083 FormatToken *IfCondition = FormatTok; 1084 // If there's a #ifndef on the first line, and the only lines before it are 1085 // comments, it could be an include guard. 1086 bool MaybeIncludeGuard = IfNDef; 1087 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) { 1088 for (auto &Line : Lines) { 1089 if (Line.Tokens.front().Tok->isNot(tok::comment)) { 1090 MaybeIncludeGuard = false; 1091 IncludeGuard = IG_Rejected; 1092 break; 1093 } 1094 } 1095 } 1096 --PPBranchLevel; 1097 parsePPUnknown(); 1098 ++PPBranchLevel; 1099 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) { 1100 IncludeGuard = IG_IfNdefed; 1101 IncludeGuardToken = IfCondition; 1102 } 1103 } 1104 1105 void UnwrappedLineParser::parsePPElse() { 1106 // If a potential include guard has an #else, it's not an include guard. 1107 if (IncludeGuard == IG_Defined && PPBranchLevel == 0) 1108 IncludeGuard = IG_Rejected; 1109 // Don't crash when there is an #else without an #if. 1110 assert(PPBranchLevel >= -1); 1111 if (PPBranchLevel == -1) 1112 conditionalCompilationStart(/*Unreachable=*/true); 1113 conditionalCompilationAlternative(); 1114 --PPBranchLevel; 1115 parsePPUnknown(); 1116 ++PPBranchLevel; 1117 } 1118 1119 void UnwrappedLineParser::parsePPEndIf() { 1120 conditionalCompilationEnd(); 1121 parsePPUnknown(); 1122 // If the #endif of a potential include guard is the last thing in the file, 1123 // then we found an include guard. 1124 if (IncludeGuard == IG_Defined && PPBranchLevel == -1 && Tokens->isEOF() && 1125 Style.IndentPPDirectives != FormatStyle::PPDIS_None) { 1126 IncludeGuard = IG_Found; 1127 } 1128 } 1129 1130 void UnwrappedLineParser::parsePPDefine() { 1131 nextToken(); 1132 1133 if (!FormatTok->Tok.getIdentifierInfo()) { 1134 IncludeGuard = IG_Rejected; 1135 IncludeGuardToken = nullptr; 1136 parsePPUnknown(); 1137 return; 1138 } 1139 1140 if (IncludeGuard == IG_IfNdefed && 1141 IncludeGuardToken->TokenText == FormatTok->TokenText) { 1142 IncludeGuard = IG_Defined; 1143 IncludeGuardToken = nullptr; 1144 for (auto &Line : Lines) { 1145 if (!Line.Tokens.front().Tok->isOneOf(tok::comment, tok::hash)) { 1146 IncludeGuard = IG_Rejected; 1147 break; 1148 } 1149 } 1150 } 1151 1152 // In the context of a define, even keywords should be treated as normal 1153 // identifiers. Setting the kind to identifier is not enough, because we need 1154 // to treat additional keywords like __except as well, which are already 1155 // identifiers. Setting the identifier info to null interferes with include 1156 // guard processing above, and changes preprocessing nesting. 1157 FormatTok->Tok.setKind(tok::identifier); 1158 FormatTok->Tok.setIdentifierInfo(Keywords.kw_internal_ident_after_define); 1159 nextToken(); 1160 if (FormatTok->Tok.getKind() == tok::l_paren && 1161 !FormatTok->hasWhitespaceBefore()) { 1162 parseParens(); 1163 } 1164 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None) 1165 Line->Level += PPBranchLevel + 1; 1166 addUnwrappedLine(); 1167 ++Line->Level; 1168 1169 Line->PPLevel = PPBranchLevel + (IncludeGuard == IG_Defined ? 0 : 1); 1170 assert((int)Line->PPLevel >= 0); 1171 Line->InMacroBody = true; 1172 1173 if (FormatTok->is(tok::identifier) && 1174 Tokens->peekNextToken()->is(tok::colon)) { 1175 nextToken(); 1176 nextToken(); 1177 } 1178 1179 // Errors during a preprocessor directive can only affect the layout of the 1180 // preprocessor directive, and thus we ignore them. An alternative approach 1181 // would be to use the same approach we use on the file level (no 1182 // re-indentation if there was a structural error) within the macro 1183 // definition. 1184 parseFile(); 1185 } 1186 1187 void UnwrappedLineParser::parsePPPragma() { 1188 Line->InPragmaDirective = true; 1189 parsePPUnknown(); 1190 } 1191 1192 void UnwrappedLineParser::parsePPUnknown() { 1193 do { 1194 nextToken(); 1195 } while (!eof()); 1196 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None) 1197 Line->Level += PPBranchLevel + 1; 1198 addUnwrappedLine(); 1199 } 1200 1201 // Here we exclude certain tokens that are not usually the first token in an 1202 // unwrapped line. This is used in attempt to distinguish macro calls without 1203 // trailing semicolons from other constructs split to several lines. 1204 static bool tokenCanStartNewLine(const FormatToken &Tok) { 1205 // Semicolon can be a null-statement, l_square can be a start of a macro or 1206 // a C++11 attribute, but this doesn't seem to be common. 1207 assert(Tok.isNot(TT_AttributeSquare)); 1208 return !Tok.isOneOf(tok::semi, tok::l_brace, 1209 // Tokens that can only be used as binary operators and a 1210 // part of overloaded operator names. 1211 tok::period, tok::periodstar, tok::arrow, tok::arrowstar, 1212 tok::less, tok::greater, tok::slash, tok::percent, 1213 tok::lessless, tok::greatergreater, tok::equal, 1214 tok::plusequal, tok::minusequal, tok::starequal, 1215 tok::slashequal, tok::percentequal, tok::ampequal, 1216 tok::pipeequal, tok::caretequal, tok::greatergreaterequal, 1217 tok::lesslessequal, 1218 // Colon is used in labels, base class lists, initializer 1219 // lists, range-based for loops, ternary operator, but 1220 // should never be the first token in an unwrapped line. 1221 tok::colon, 1222 // 'noexcept' is a trailing annotation. 1223 tok::kw_noexcept); 1224 } 1225 1226 static bool mustBeJSIdent(const AdditionalKeywords &Keywords, 1227 const FormatToken *FormatTok) { 1228 // FIXME: This returns true for C/C++ keywords like 'struct'. 1229 return FormatTok->is(tok::identifier) && 1230 (!FormatTok->Tok.getIdentifierInfo() || 1231 !FormatTok->isOneOf( 1232 Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async, 1233 Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally, 1234 Keywords.kw_function, Keywords.kw_import, Keywords.kw_is, 1235 Keywords.kw_let, Keywords.kw_var, tok::kw_const, 1236 Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements, 1237 Keywords.kw_instanceof, Keywords.kw_interface, 1238 Keywords.kw_override, Keywords.kw_throws, Keywords.kw_from)); 1239 } 1240 1241 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords, 1242 const FormatToken *FormatTok) { 1243 return FormatTok->Tok.isLiteral() || 1244 FormatTok->isOneOf(tok::kw_true, tok::kw_false) || 1245 mustBeJSIdent(Keywords, FormatTok); 1246 } 1247 1248 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement 1249 // when encountered after a value (see mustBeJSIdentOrValue). 1250 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords, 1251 const FormatToken *FormatTok) { 1252 return FormatTok->isOneOf( 1253 tok::kw_return, Keywords.kw_yield, 1254 // conditionals 1255 tok::kw_if, tok::kw_else, 1256 // loops 1257 tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break, 1258 // switch/case 1259 tok::kw_switch, tok::kw_case, 1260 // exceptions 1261 tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally, 1262 // declaration 1263 tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let, 1264 Keywords.kw_async, Keywords.kw_function, 1265 // import/export 1266 Keywords.kw_import, tok::kw_export); 1267 } 1268 1269 // Checks whether a token is a type in K&R C (aka C78). 1270 static bool isC78Type(const FormatToken &Tok) { 1271 return Tok.isOneOf(tok::kw_char, tok::kw_short, tok::kw_int, tok::kw_long, 1272 tok::kw_unsigned, tok::kw_float, tok::kw_double, 1273 tok::identifier); 1274 } 1275 1276 // This function checks whether a token starts the first parameter declaration 1277 // in a K&R C (aka C78) function definition, e.g.: 1278 // int f(a, b) 1279 // short a, b; 1280 // { 1281 // return a + b; 1282 // } 1283 static bool isC78ParameterDecl(const FormatToken *Tok, const FormatToken *Next, 1284 const FormatToken *FuncName) { 1285 assert(Tok); 1286 assert(Next); 1287 assert(FuncName); 1288 1289 if (FuncName->isNot(tok::identifier)) 1290 return false; 1291 1292 const FormatToken *Prev = FuncName->Previous; 1293 if (!Prev || (Prev->isNot(tok::star) && !isC78Type(*Prev))) 1294 return false; 1295 1296 if (!isC78Type(*Tok) && 1297 !Tok->isOneOf(tok::kw_register, tok::kw_struct, tok::kw_union)) { 1298 return false; 1299 } 1300 1301 if (Next->isNot(tok::star) && !Next->Tok.getIdentifierInfo()) 1302 return false; 1303 1304 Tok = Tok->Previous; 1305 if (!Tok || Tok->isNot(tok::r_paren)) 1306 return false; 1307 1308 Tok = Tok->Previous; 1309 if (!Tok || Tok->isNot(tok::identifier)) 1310 return false; 1311 1312 return Tok->Previous && Tok->Previous->isOneOf(tok::l_paren, tok::comma); 1313 } 1314 1315 bool UnwrappedLineParser::parseModuleImport() { 1316 assert(FormatTok->is(Keywords.kw_import) && "'import' expected"); 1317 1318 if (auto Token = Tokens->peekNextToken(/*SkipComment=*/true); 1319 !Token->Tok.getIdentifierInfo() && 1320 !Token->isOneOf(tok::colon, tok::less, tok::string_literal)) { 1321 return false; 1322 } 1323 1324 nextToken(); 1325 while (!eof()) { 1326 if (FormatTok->is(tok::colon)) { 1327 FormatTok->setFinalizedType(TT_ModulePartitionColon); 1328 } 1329 // Handle import <foo/bar.h> as we would an include statement. 1330 else if (FormatTok->is(tok::less)) { 1331 nextToken(); 1332 while (!FormatTok->isOneOf(tok::semi, tok::greater, tok::eof)) { 1333 // Mark tokens up to the trailing line comments as implicit string 1334 // literals. 1335 if (FormatTok->isNot(tok::comment) && 1336 !FormatTok->TokenText.starts_with("//")) { 1337 FormatTok->setFinalizedType(TT_ImplicitStringLiteral); 1338 } 1339 nextToken(); 1340 } 1341 } 1342 if (FormatTok->is(tok::semi)) { 1343 nextToken(); 1344 break; 1345 } 1346 nextToken(); 1347 } 1348 1349 addUnwrappedLine(); 1350 return true; 1351 } 1352 1353 // readTokenWithJavaScriptASI reads the next token and terminates the current 1354 // line if JavaScript Automatic Semicolon Insertion must 1355 // happen between the current token and the next token. 1356 // 1357 // This method is conservative - it cannot cover all edge cases of JavaScript, 1358 // but only aims to correctly handle certain well known cases. It *must not* 1359 // return true in speculative cases. 1360 void UnwrappedLineParser::readTokenWithJavaScriptASI() { 1361 FormatToken *Previous = FormatTok; 1362 readToken(); 1363 FormatToken *Next = FormatTok; 1364 1365 bool IsOnSameLine = 1366 CommentsBeforeNextToken.empty() 1367 ? Next->NewlinesBefore == 0 1368 : CommentsBeforeNextToken.front()->NewlinesBefore == 0; 1369 if (IsOnSameLine) 1370 return; 1371 1372 bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous); 1373 bool PreviousStartsTemplateExpr = 1374 Previous->is(TT_TemplateString) && Previous->TokenText.ends_with("${"); 1375 if (PreviousMustBeValue || Previous->is(tok::r_paren)) { 1376 // If the line contains an '@' sign, the previous token might be an 1377 // annotation, which can precede another identifier/value. 1378 bool HasAt = llvm::any_of(Line->Tokens, [](UnwrappedLineNode &LineNode) { 1379 return LineNode.Tok->is(tok::at); 1380 }); 1381 if (HasAt) 1382 return; 1383 } 1384 if (Next->is(tok::exclaim) && PreviousMustBeValue) 1385 return addUnwrappedLine(); 1386 bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next); 1387 bool NextEndsTemplateExpr = 1388 Next->is(TT_TemplateString) && Next->TokenText.starts_with("}"); 1389 if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr && 1390 (PreviousMustBeValue || 1391 Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus, 1392 tok::minusminus))) { 1393 return addUnwrappedLine(); 1394 } 1395 if ((PreviousMustBeValue || Previous->is(tok::r_paren)) && 1396 isJSDeclOrStmt(Keywords, Next)) { 1397 return addUnwrappedLine(); 1398 } 1399 } 1400 1401 void UnwrappedLineParser::parseStructuralElement( 1402 const FormatToken *OpeningBrace, IfStmtKind *IfKind, 1403 FormatToken **IfLeftBrace, bool *HasDoWhile, bool *HasLabel) { 1404 if (Style.Language == FormatStyle::LK_TableGen && 1405 FormatTok->is(tok::pp_include)) { 1406 nextToken(); 1407 if (FormatTok->is(tok::string_literal)) 1408 nextToken(); 1409 addUnwrappedLine(); 1410 return; 1411 } 1412 1413 if (Style.isCpp()) { 1414 while (FormatTok->is(tok::l_square) && handleCppAttributes()) { 1415 } 1416 } else if (Style.isVerilog()) { 1417 if (Keywords.isVerilogStructuredProcedure(*FormatTok)) { 1418 parseForOrWhileLoop(/*HasParens=*/false); 1419 return; 1420 } 1421 if (FormatTok->isOneOf(Keywords.kw_foreach, Keywords.kw_repeat)) { 1422 parseForOrWhileLoop(); 1423 return; 1424 } 1425 if (FormatTok->isOneOf(tok::kw_restrict, Keywords.kw_assert, 1426 Keywords.kw_assume, Keywords.kw_cover)) { 1427 parseIfThenElse(IfKind, /*KeepBraces=*/false, /*IsVerilogAssert=*/true); 1428 return; 1429 } 1430 1431 // Skip things that can exist before keywords like 'if' and 'case'. 1432 while (true) { 1433 if (FormatTok->isOneOf(Keywords.kw_priority, Keywords.kw_unique, 1434 Keywords.kw_unique0)) { 1435 nextToken(); 1436 } else if (FormatTok->is(tok::l_paren) && 1437 Tokens->peekNextToken()->is(tok::star)) { 1438 parseParens(); 1439 } else { 1440 break; 1441 } 1442 } 1443 } 1444 1445 // Tokens that only make sense at the beginning of a line. 1446 switch (FormatTok->Tok.getKind()) { 1447 case tok::kw_asm: 1448 nextToken(); 1449 if (FormatTok->is(tok::l_brace)) { 1450 FormatTok->setFinalizedType(TT_InlineASMBrace); 1451 nextToken(); 1452 while (FormatTok && !eof()) { 1453 if (FormatTok->is(tok::r_brace)) { 1454 FormatTok->setFinalizedType(TT_InlineASMBrace); 1455 nextToken(); 1456 addUnwrappedLine(); 1457 break; 1458 } 1459 FormatTok->Finalized = true; 1460 nextToken(); 1461 } 1462 } 1463 break; 1464 case tok::kw_namespace: 1465 parseNamespace(); 1466 return; 1467 case tok::kw_public: 1468 case tok::kw_protected: 1469 case tok::kw_private: 1470 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 1471 Style.isCSharp()) { 1472 nextToken(); 1473 } else { 1474 parseAccessSpecifier(); 1475 } 1476 return; 1477 case tok::kw_if: { 1478 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1479 // field/method declaration. 1480 break; 1481 } 1482 FormatToken *Tok = parseIfThenElse(IfKind); 1483 if (IfLeftBrace) 1484 *IfLeftBrace = Tok; 1485 return; 1486 } 1487 case tok::kw_for: 1488 case tok::kw_while: 1489 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1490 // field/method declaration. 1491 break; 1492 } 1493 parseForOrWhileLoop(); 1494 return; 1495 case tok::kw_do: 1496 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1497 // field/method declaration. 1498 break; 1499 } 1500 parseDoWhile(); 1501 if (HasDoWhile) 1502 *HasDoWhile = true; 1503 return; 1504 case tok::kw_switch: 1505 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1506 // 'switch: string' field declaration. 1507 break; 1508 } 1509 parseSwitch(); 1510 return; 1511 case tok::kw_default: 1512 // In Verilog default along with other labels are handled in the next loop. 1513 if (Style.isVerilog()) 1514 break; 1515 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1516 // 'default: string' field declaration. 1517 break; 1518 } 1519 nextToken(); 1520 if (FormatTok->is(tok::colon)) { 1521 FormatTok->setFinalizedType(TT_CaseLabelColon); 1522 parseLabel(); 1523 return; 1524 } 1525 // e.g. "default void f() {}" in a Java interface. 1526 break; 1527 case tok::kw_case: 1528 // Proto: there are no switch/case statements. 1529 if (Style.Language == FormatStyle::LK_Proto) { 1530 nextToken(); 1531 return; 1532 } 1533 if (Style.isVerilog()) { 1534 parseBlock(); 1535 addUnwrappedLine(); 1536 return; 1537 } 1538 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1539 // 'case: string' field declaration. 1540 nextToken(); 1541 break; 1542 } 1543 parseCaseLabel(); 1544 return; 1545 case tok::kw_try: 1546 case tok::kw___try: 1547 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1548 // field/method declaration. 1549 break; 1550 } 1551 parseTryCatch(); 1552 return; 1553 case tok::kw_extern: 1554 nextToken(); 1555 if (Style.isVerilog()) { 1556 // In Verilog and extern module declaration looks like a start of module. 1557 // But there is no body and endmodule. So we handle it separately. 1558 if (Keywords.isVerilogHierarchy(*FormatTok)) { 1559 parseVerilogHierarchyHeader(); 1560 return; 1561 } 1562 } else if (FormatTok->is(tok::string_literal)) { 1563 nextToken(); 1564 if (FormatTok->is(tok::l_brace)) { 1565 if (Style.BraceWrapping.AfterExternBlock) 1566 addUnwrappedLine(); 1567 // Either we indent or for backwards compatibility we follow the 1568 // AfterExternBlock style. 1569 unsigned AddLevels = 1570 (Style.IndentExternBlock == FormatStyle::IEBS_Indent) || 1571 (Style.BraceWrapping.AfterExternBlock && 1572 Style.IndentExternBlock == 1573 FormatStyle::IEBS_AfterExternBlock) 1574 ? 1u 1575 : 0u; 1576 parseBlock(/*MustBeDeclaration=*/true, AddLevels); 1577 addUnwrappedLine(); 1578 return; 1579 } 1580 } 1581 break; 1582 case tok::kw_export: 1583 if (Style.isJavaScript()) { 1584 parseJavaScriptEs6ImportExport(); 1585 return; 1586 } 1587 if (Style.isCpp()) { 1588 nextToken(); 1589 if (FormatTok->is(tok::kw_namespace)) { 1590 parseNamespace(); 1591 return; 1592 } 1593 if (FormatTok->is(Keywords.kw_import) && parseModuleImport()) 1594 return; 1595 } 1596 break; 1597 case tok::kw_inline: 1598 nextToken(); 1599 if (FormatTok->is(tok::kw_namespace)) { 1600 parseNamespace(); 1601 return; 1602 } 1603 break; 1604 case tok::identifier: 1605 if (FormatTok->is(TT_ForEachMacro)) { 1606 parseForOrWhileLoop(); 1607 return; 1608 } 1609 if (FormatTok->is(TT_MacroBlockBegin)) { 1610 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 1611 /*MunchSemi=*/false); 1612 return; 1613 } 1614 if (FormatTok->is(Keywords.kw_import)) { 1615 if (Style.isJavaScript()) { 1616 parseJavaScriptEs6ImportExport(); 1617 return; 1618 } 1619 if (Style.Language == FormatStyle::LK_Proto) { 1620 nextToken(); 1621 if (FormatTok->is(tok::kw_public)) 1622 nextToken(); 1623 if (FormatTok->isNot(tok::string_literal)) 1624 return; 1625 nextToken(); 1626 if (FormatTok->is(tok::semi)) 1627 nextToken(); 1628 addUnwrappedLine(); 1629 return; 1630 } 1631 if (Style.isCpp() && parseModuleImport()) 1632 return; 1633 } 1634 if (Style.isCpp() && 1635 FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 1636 Keywords.kw_slots, Keywords.kw_qslots)) { 1637 nextToken(); 1638 if (FormatTok->is(tok::colon)) { 1639 nextToken(); 1640 addUnwrappedLine(); 1641 return; 1642 } 1643 } 1644 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1645 parseStatementMacro(); 1646 return; 1647 } 1648 if (Style.isCpp() && FormatTok->is(TT_NamespaceMacro)) { 1649 parseNamespace(); 1650 return; 1651 } 1652 // In Verilog labels can be any expression, so we don't do them here. 1653 // JS doesn't have macros, and within classes colons indicate fields, not 1654 // labels. 1655 if (!Style.isJavaScript() && !Style.isVerilog() && 1656 Tokens->peekNextToken()->is(tok::colon) && !Line->MustBeDeclaration) { 1657 nextToken(); 1658 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1659 FormatTok->setFinalizedType(TT_GotoLabelColon); 1660 parseLabel(!Style.IndentGotoLabels); 1661 if (HasLabel) 1662 *HasLabel = true; 1663 return; 1664 } 1665 // In all other cases, parse the declaration. 1666 break; 1667 default: 1668 break; 1669 } 1670 1671 const bool InRequiresExpression = 1672 OpeningBrace && OpeningBrace->is(TT_RequiresExpressionLBrace); 1673 do { 1674 const FormatToken *Previous = FormatTok->Previous; 1675 switch (FormatTok->Tok.getKind()) { 1676 case tok::at: 1677 nextToken(); 1678 if (FormatTok->is(tok::l_brace)) { 1679 nextToken(); 1680 parseBracedList(); 1681 break; 1682 } else if (Style.Language == FormatStyle::LK_Java && 1683 FormatTok->is(Keywords.kw_interface)) { 1684 nextToken(); 1685 break; 1686 } 1687 switch (FormatTok->Tok.getObjCKeywordID()) { 1688 case tok::objc_public: 1689 case tok::objc_protected: 1690 case tok::objc_package: 1691 case tok::objc_private: 1692 return parseAccessSpecifier(); 1693 case tok::objc_interface: 1694 case tok::objc_implementation: 1695 return parseObjCInterfaceOrImplementation(); 1696 case tok::objc_protocol: 1697 if (parseObjCProtocol()) 1698 return; 1699 break; 1700 case tok::objc_end: 1701 return; // Handled by the caller. 1702 case tok::objc_optional: 1703 case tok::objc_required: 1704 nextToken(); 1705 addUnwrappedLine(); 1706 return; 1707 case tok::objc_autoreleasepool: 1708 nextToken(); 1709 if (FormatTok->is(tok::l_brace)) { 1710 if (Style.BraceWrapping.AfterControlStatement == 1711 FormatStyle::BWACS_Always) { 1712 addUnwrappedLine(); 1713 } 1714 parseBlock(); 1715 } 1716 addUnwrappedLine(); 1717 return; 1718 case tok::objc_synchronized: 1719 nextToken(); 1720 if (FormatTok->is(tok::l_paren)) { 1721 // Skip synchronization object 1722 parseParens(); 1723 } 1724 if (FormatTok->is(tok::l_brace)) { 1725 if (Style.BraceWrapping.AfterControlStatement == 1726 FormatStyle::BWACS_Always) { 1727 addUnwrappedLine(); 1728 } 1729 parseBlock(); 1730 } 1731 addUnwrappedLine(); 1732 return; 1733 case tok::objc_try: 1734 // This branch isn't strictly necessary (the kw_try case below would 1735 // do this too after the tok::at is parsed above). But be explicit. 1736 parseTryCatch(); 1737 return; 1738 default: 1739 break; 1740 } 1741 break; 1742 case tok::kw_requires: { 1743 if (Style.isCpp()) { 1744 bool ParsedClause = parseRequires(); 1745 if (ParsedClause) 1746 return; 1747 } else { 1748 nextToken(); 1749 } 1750 break; 1751 } 1752 case tok::kw_enum: 1753 // Ignore if this is part of "template <enum ...". 1754 if (Previous && Previous->is(tok::less)) { 1755 nextToken(); 1756 break; 1757 } 1758 1759 // parseEnum falls through and does not yet add an unwrapped line as an 1760 // enum definition can start a structural element. 1761 if (!parseEnum()) 1762 break; 1763 // This only applies to C++ and Verilog. 1764 if (!Style.isCpp() && !Style.isVerilog()) { 1765 addUnwrappedLine(); 1766 return; 1767 } 1768 break; 1769 case tok::kw_typedef: 1770 nextToken(); 1771 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 1772 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS, 1773 Keywords.kw_CF_CLOSED_ENUM, 1774 Keywords.kw_NS_CLOSED_ENUM)) { 1775 parseEnum(); 1776 } 1777 break; 1778 case tok::kw_class: 1779 if (Style.isVerilog()) { 1780 parseBlock(); 1781 addUnwrappedLine(); 1782 return; 1783 } 1784 [[fallthrough]]; 1785 case tok::kw_struct: 1786 case tok::kw_union: 1787 if (parseStructLike()) 1788 return; 1789 break; 1790 case tok::kw_decltype: 1791 nextToken(); 1792 if (FormatTok->is(tok::l_paren)) { 1793 parseParens(); 1794 assert(FormatTok->Previous); 1795 if (FormatTok->Previous->endsSequence(tok::r_paren, tok::kw_auto, 1796 tok::l_paren)) { 1797 Line->SeenDecltypeAuto = true; 1798 } 1799 } 1800 break; 1801 case tok::period: 1802 nextToken(); 1803 // In Java, classes have an implicit static member "class". 1804 if (Style.Language == FormatStyle::LK_Java && FormatTok && 1805 FormatTok->is(tok::kw_class)) { 1806 nextToken(); 1807 } 1808 if (Style.isJavaScript() && FormatTok && 1809 FormatTok->Tok.getIdentifierInfo()) { 1810 // JavaScript only has pseudo keywords, all keywords are allowed to 1811 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 1812 nextToken(); 1813 } 1814 break; 1815 case tok::semi: 1816 nextToken(); 1817 addUnwrappedLine(); 1818 return; 1819 case tok::r_brace: 1820 addUnwrappedLine(); 1821 return; 1822 case tok::l_paren: { 1823 parseParens(); 1824 // Break the unwrapped line if a K&R C function definition has a parameter 1825 // declaration. 1826 if (OpeningBrace || !Style.isCpp() || !Previous || eof()) 1827 break; 1828 if (isC78ParameterDecl(FormatTok, 1829 Tokens->peekNextToken(/*SkipComment=*/true), 1830 Previous)) { 1831 addUnwrappedLine(); 1832 return; 1833 } 1834 break; 1835 } 1836 case tok::kw_operator: 1837 nextToken(); 1838 if (FormatTok->isBinaryOperator()) 1839 nextToken(); 1840 break; 1841 case tok::caret: 1842 nextToken(); 1843 // Block return type. 1844 if (FormatTok->Tok.isAnyIdentifier() || 1845 FormatTok->isSimpleTypeSpecifier()) { 1846 nextToken(); 1847 // Return types: pointers are ok too. 1848 while (FormatTok->is(tok::star)) 1849 nextToken(); 1850 } 1851 // Block argument list. 1852 if (FormatTok->is(tok::l_paren)) 1853 parseParens(); 1854 // Block body. 1855 if (FormatTok->is(tok::l_brace)) 1856 parseChildBlock(); 1857 break; 1858 case tok::l_brace: 1859 if (InRequiresExpression) 1860 FormatTok->setFinalizedType(TT_BracedListLBrace); 1861 if (!tryToParsePropertyAccessor() && !tryToParseBracedList()) { 1862 IsDecltypeAutoFunction = Line->SeenDecltypeAuto; 1863 // A block outside of parentheses must be the last part of a 1864 // structural element. 1865 // FIXME: Figure out cases where this is not true, and add projections 1866 // for them (the one we know is missing are lambdas). 1867 if (Style.Language == FormatStyle::LK_Java && 1868 Line->Tokens.front().Tok->is(Keywords.kw_synchronized)) { 1869 // If necessary, we could set the type to something different than 1870 // TT_FunctionLBrace. 1871 if (Style.BraceWrapping.AfterControlStatement == 1872 FormatStyle::BWACS_Always) { 1873 addUnwrappedLine(); 1874 } 1875 } else if (Style.BraceWrapping.AfterFunction) { 1876 addUnwrappedLine(); 1877 } 1878 FormatTok->setFinalizedType(TT_FunctionLBrace); 1879 parseBlock(); 1880 IsDecltypeAutoFunction = false; 1881 addUnwrappedLine(); 1882 return; 1883 } 1884 // Otherwise this was a braced init list, and the structural 1885 // element continues. 1886 break; 1887 case tok::kw_try: 1888 if (Style.isJavaScript() && Line->MustBeDeclaration) { 1889 // field/method declaration. 1890 nextToken(); 1891 break; 1892 } 1893 // We arrive here when parsing function-try blocks. 1894 if (Style.BraceWrapping.AfterFunction) 1895 addUnwrappedLine(); 1896 parseTryCatch(); 1897 return; 1898 case tok::identifier: { 1899 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where) && 1900 Line->MustBeDeclaration) { 1901 addUnwrappedLine(); 1902 parseCSharpGenericTypeConstraint(); 1903 break; 1904 } 1905 if (FormatTok->is(TT_MacroBlockEnd)) { 1906 addUnwrappedLine(); 1907 return; 1908 } 1909 1910 // Function declarations (as opposed to function expressions) are parsed 1911 // on their own unwrapped line by continuing this loop. Function 1912 // expressions (functions that are not on their own line) must not create 1913 // a new unwrapped line, so they are special cased below. 1914 size_t TokenCount = Line->Tokens.size(); 1915 if (Style.isJavaScript() && FormatTok->is(Keywords.kw_function) && 1916 (TokenCount > 1 || 1917 (TokenCount == 1 && 1918 Line->Tokens.front().Tok->isNot(Keywords.kw_async)))) { 1919 tryToParseJSFunction(); 1920 break; 1921 } 1922 if ((Style.isJavaScript() || Style.Language == FormatStyle::LK_Java) && 1923 FormatTok->is(Keywords.kw_interface)) { 1924 if (Style.isJavaScript()) { 1925 // In JavaScript/TypeScript, "interface" can be used as a standalone 1926 // identifier, e.g. in `var interface = 1;`. If "interface" is 1927 // followed by another identifier, it is very like to be an actual 1928 // interface declaration. 1929 unsigned StoredPosition = Tokens->getPosition(); 1930 FormatToken *Next = Tokens->getNextToken(); 1931 FormatTok = Tokens->setPosition(StoredPosition); 1932 if (!mustBeJSIdent(Keywords, Next)) { 1933 nextToken(); 1934 break; 1935 } 1936 } 1937 parseRecord(); 1938 addUnwrappedLine(); 1939 return; 1940 } 1941 1942 if (Style.isVerilog()) { 1943 if (FormatTok->is(Keywords.kw_table)) { 1944 parseVerilogTable(); 1945 return; 1946 } 1947 if (Keywords.isVerilogBegin(*FormatTok) || 1948 Keywords.isVerilogHierarchy(*FormatTok)) { 1949 parseBlock(); 1950 addUnwrappedLine(); 1951 return; 1952 } 1953 } 1954 1955 if (!Style.isCpp() && FormatTok->is(Keywords.kw_interface)) { 1956 if (parseStructLike()) 1957 return; 1958 break; 1959 } 1960 1961 if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) { 1962 parseStatementMacro(); 1963 return; 1964 } 1965 1966 // See if the following token should start a new unwrapped line. 1967 StringRef Text = FormatTok->TokenText; 1968 1969 FormatToken *PreviousToken = FormatTok; 1970 nextToken(); 1971 1972 // JS doesn't have macros, and within classes colons indicate fields, not 1973 // labels. 1974 if (Style.isJavaScript()) 1975 break; 1976 1977 auto OneTokenSoFar = [&]() { 1978 auto I = Line->Tokens.begin(), E = Line->Tokens.end(); 1979 while (I != E && I->Tok->is(tok::comment)) 1980 ++I; 1981 if (Style.isVerilog()) 1982 while (I != E && I->Tok->is(tok::hash)) 1983 ++I; 1984 return I != E && (++I == E); 1985 }; 1986 if (OneTokenSoFar()) { 1987 // Recognize function-like macro usages without trailing semicolon as 1988 // well as free-standing macros like Q_OBJECT. 1989 bool FunctionLike = FormatTok->is(tok::l_paren); 1990 if (FunctionLike) 1991 parseParens(); 1992 1993 bool FollowedByNewline = 1994 CommentsBeforeNextToken.empty() 1995 ? FormatTok->NewlinesBefore > 0 1996 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1997 1998 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1999 tokenCanStartNewLine(*FormatTok) && Text == Text.upper()) { 2000 if (PreviousToken->isNot(TT_UntouchableMacroFunc)) 2001 PreviousToken->setFinalizedType(TT_FunctionLikeOrFreestandingMacro); 2002 addUnwrappedLine(); 2003 return; 2004 } 2005 } 2006 break; 2007 } 2008 case tok::equal: 2009 if ((Style.isJavaScript() || Style.isCSharp()) && 2010 FormatTok->is(TT_FatArrow)) { 2011 tryToParseChildBlock(); 2012 break; 2013 } 2014 2015 nextToken(); 2016 if (FormatTok->is(tok::l_brace)) { 2017 // Block kind should probably be set to BK_BracedInit for any language. 2018 // C# needs this change to ensure that array initialisers and object 2019 // initialisers are indented the same way. 2020 if (Style.isCSharp()) 2021 FormatTok->setBlockKind(BK_BracedInit); 2022 nextToken(); 2023 parseBracedList(); 2024 } else if (Style.Language == FormatStyle::LK_Proto && 2025 FormatTok->is(tok::less)) { 2026 nextToken(); 2027 parseBracedList(/*IsAngleBracket=*/true); 2028 } 2029 break; 2030 case tok::l_square: 2031 parseSquare(); 2032 break; 2033 case tok::kw_new: 2034 parseNew(); 2035 break; 2036 case tok::kw_case: 2037 // Proto: there are no switch/case statements. 2038 if (Style.Language == FormatStyle::LK_Proto) { 2039 nextToken(); 2040 return; 2041 } 2042 // In Verilog switch is called case. 2043 if (Style.isVerilog()) { 2044 parseBlock(); 2045 addUnwrappedLine(); 2046 return; 2047 } 2048 if (Style.isJavaScript() && Line->MustBeDeclaration) { 2049 // 'case: string' field declaration. 2050 nextToken(); 2051 break; 2052 } 2053 parseCaseLabel(); 2054 break; 2055 case tok::kw_default: 2056 nextToken(); 2057 if (Style.isVerilog()) { 2058 if (FormatTok->is(tok::colon)) { 2059 // The label will be handled in the next iteration. 2060 break; 2061 } 2062 if (FormatTok->is(Keywords.kw_clocking)) { 2063 // A default clocking block. 2064 parseBlock(); 2065 addUnwrappedLine(); 2066 return; 2067 } 2068 parseVerilogCaseLabel(); 2069 return; 2070 } 2071 break; 2072 case tok::colon: 2073 nextToken(); 2074 if (Style.isVerilog()) { 2075 parseVerilogCaseLabel(); 2076 return; 2077 } 2078 break; 2079 default: 2080 nextToken(); 2081 break; 2082 } 2083 } while (!eof()); 2084 } 2085 2086 bool UnwrappedLineParser::tryToParsePropertyAccessor() { 2087 assert(FormatTok->is(tok::l_brace)); 2088 if (!Style.isCSharp()) 2089 return false; 2090 // See if it's a property accessor. 2091 if (FormatTok->Previous->isNot(tok::identifier)) 2092 return false; 2093 2094 // See if we are inside a property accessor. 2095 // 2096 // Record the current tokenPosition so that we can advance and 2097 // reset the current token. `Next` is not set yet so we need 2098 // another way to advance along the token stream. 2099 unsigned int StoredPosition = Tokens->getPosition(); 2100 FormatToken *Tok = Tokens->getNextToken(); 2101 2102 // A trivial property accessor is of the form: 2103 // { [ACCESS_SPECIFIER] [get]; [ACCESS_SPECIFIER] [set|init] } 2104 // Track these as they do not require line breaks to be introduced. 2105 bool HasSpecialAccessor = false; 2106 bool IsTrivialPropertyAccessor = true; 2107 while (!eof()) { 2108 if (Tok->isOneOf(tok::semi, tok::kw_public, tok::kw_private, 2109 tok::kw_protected, Keywords.kw_internal, Keywords.kw_get, 2110 Keywords.kw_init, Keywords.kw_set)) { 2111 if (Tok->isOneOf(Keywords.kw_get, Keywords.kw_init, Keywords.kw_set)) 2112 HasSpecialAccessor = true; 2113 Tok = Tokens->getNextToken(); 2114 continue; 2115 } 2116 if (Tok->isNot(tok::r_brace)) 2117 IsTrivialPropertyAccessor = false; 2118 break; 2119 } 2120 2121 if (!HasSpecialAccessor) { 2122 Tokens->setPosition(StoredPosition); 2123 return false; 2124 } 2125 2126 // Try to parse the property accessor: 2127 // https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties 2128 Tokens->setPosition(StoredPosition); 2129 if (!IsTrivialPropertyAccessor && Style.BraceWrapping.AfterFunction) 2130 addUnwrappedLine(); 2131 nextToken(); 2132 do { 2133 switch (FormatTok->Tok.getKind()) { 2134 case tok::r_brace: 2135 nextToken(); 2136 if (FormatTok->is(tok::equal)) { 2137 while (!eof() && FormatTok->isNot(tok::semi)) 2138 nextToken(); 2139 nextToken(); 2140 } 2141 addUnwrappedLine(); 2142 return true; 2143 case tok::l_brace: 2144 ++Line->Level; 2145 parseBlock(/*MustBeDeclaration=*/true); 2146 addUnwrappedLine(); 2147 --Line->Level; 2148 break; 2149 case tok::equal: 2150 if (FormatTok->is(TT_FatArrow)) { 2151 ++Line->Level; 2152 do { 2153 nextToken(); 2154 } while (!eof() && FormatTok->isNot(tok::semi)); 2155 nextToken(); 2156 addUnwrappedLine(); 2157 --Line->Level; 2158 break; 2159 } 2160 nextToken(); 2161 break; 2162 default: 2163 if (FormatTok->isOneOf(Keywords.kw_get, Keywords.kw_init, 2164 Keywords.kw_set) && 2165 !IsTrivialPropertyAccessor) { 2166 // Non-trivial get/set needs to be on its own line. 2167 addUnwrappedLine(); 2168 } 2169 nextToken(); 2170 } 2171 } while (!eof()); 2172 2173 // Unreachable for well-formed code (paired '{' and '}'). 2174 return true; 2175 } 2176 2177 bool UnwrappedLineParser::tryToParseLambda() { 2178 assert(FormatTok->is(tok::l_square)); 2179 if (!Style.isCpp()) { 2180 nextToken(); 2181 return false; 2182 } 2183 FormatToken &LSquare = *FormatTok; 2184 if (!tryToParseLambdaIntroducer()) 2185 return false; 2186 2187 bool SeenArrow = false; 2188 bool InTemplateParameterList = false; 2189 2190 while (FormatTok->isNot(tok::l_brace)) { 2191 if (FormatTok->isSimpleTypeSpecifier()) { 2192 nextToken(); 2193 continue; 2194 } 2195 switch (FormatTok->Tok.getKind()) { 2196 case tok::l_brace: 2197 break; 2198 case tok::l_paren: 2199 parseParens(/*AmpAmpTokenType=*/TT_PointerOrReference); 2200 break; 2201 case tok::l_square: 2202 parseSquare(); 2203 break; 2204 case tok::less: 2205 assert(FormatTok->Previous); 2206 if (FormatTok->Previous->is(tok::r_square)) 2207 InTemplateParameterList = true; 2208 nextToken(); 2209 break; 2210 case tok::kw_auto: 2211 case tok::kw_class: 2212 case tok::kw_template: 2213 case tok::kw_typename: 2214 case tok::amp: 2215 case tok::star: 2216 case tok::kw_const: 2217 case tok::kw_constexpr: 2218 case tok::kw_consteval: 2219 case tok::comma: 2220 case tok::greater: 2221 case tok::identifier: 2222 case tok::numeric_constant: 2223 case tok::coloncolon: 2224 case tok::kw_mutable: 2225 case tok::kw_noexcept: 2226 case tok::kw_static: 2227 nextToken(); 2228 break; 2229 // Specialization of a template with an integer parameter can contain 2230 // arithmetic, logical, comparison and ternary operators. 2231 // 2232 // FIXME: This also accepts sequences of operators that are not in the scope 2233 // of a template argument list. 2234 // 2235 // In a C++ lambda a template type can only occur after an arrow. We use 2236 // this as an heuristic to distinguish between Objective-C expressions 2237 // followed by an `a->b` expression, such as: 2238 // ([obj func:arg] + a->b) 2239 // Otherwise the code below would parse as a lambda. 2240 case tok::plus: 2241 case tok::minus: 2242 case tok::exclaim: 2243 case tok::tilde: 2244 case tok::slash: 2245 case tok::percent: 2246 case tok::lessless: 2247 case tok::pipe: 2248 case tok::pipepipe: 2249 case tok::ampamp: 2250 case tok::caret: 2251 case tok::equalequal: 2252 case tok::exclaimequal: 2253 case tok::greaterequal: 2254 case tok::lessequal: 2255 case tok::question: 2256 case tok::colon: 2257 case tok::ellipsis: 2258 case tok::kw_true: 2259 case tok::kw_false: 2260 if (SeenArrow || InTemplateParameterList) { 2261 nextToken(); 2262 break; 2263 } 2264 return true; 2265 case tok::arrow: 2266 // This might or might not actually be a lambda arrow (this could be an 2267 // ObjC method invocation followed by a dereferencing arrow). We might 2268 // reset this back to TT_Unknown in TokenAnnotator. 2269 FormatTok->setFinalizedType(TT_TrailingReturnArrow); 2270 SeenArrow = true; 2271 nextToken(); 2272 break; 2273 case tok::kw_requires: { 2274 auto *RequiresToken = FormatTok; 2275 nextToken(); 2276 parseRequiresClause(RequiresToken); 2277 break; 2278 } 2279 case tok::equal: 2280 if (!InTemplateParameterList) 2281 return true; 2282 nextToken(); 2283 break; 2284 default: 2285 return true; 2286 } 2287 } 2288 2289 FormatTok->setFinalizedType(TT_LambdaLBrace); 2290 LSquare.setFinalizedType(TT_LambdaLSquare); 2291 2292 NestedLambdas.push_back(Line->SeenDecltypeAuto); 2293 parseChildBlock(); 2294 assert(!NestedLambdas.empty()); 2295 NestedLambdas.pop_back(); 2296 2297 return true; 2298 } 2299 2300 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 2301 const FormatToken *Previous = FormatTok->Previous; 2302 const FormatToken *LeftSquare = FormatTok; 2303 nextToken(); 2304 if ((Previous && ((Previous->Tok.getIdentifierInfo() && 2305 !Previous->isOneOf(tok::kw_return, tok::kw_co_await, 2306 tok::kw_co_yield, tok::kw_co_return)) || 2307 Previous->closesScope())) || 2308 LeftSquare->isCppStructuredBinding(Style)) { 2309 return false; 2310 } 2311 if (FormatTok->is(tok::l_square)) 2312 return false; 2313 if (FormatTok->is(tok::r_square)) { 2314 const FormatToken *Next = Tokens->peekNextToken(/*SkipComment=*/true); 2315 if (Next->is(tok::greater)) 2316 return false; 2317 } 2318 parseSquare(/*LambdaIntroducer=*/true); 2319 return true; 2320 } 2321 2322 void UnwrappedLineParser::tryToParseJSFunction() { 2323 assert(FormatTok->is(Keywords.kw_function)); 2324 if (FormatTok->is(Keywords.kw_async)) 2325 nextToken(); 2326 // Consume "function". 2327 nextToken(); 2328 2329 // Consume * (generator function). Treat it like C++'s overloaded operators. 2330 if (FormatTok->is(tok::star)) { 2331 FormatTok->setFinalizedType(TT_OverloadedOperator); 2332 nextToken(); 2333 } 2334 2335 // Consume function name. 2336 if (FormatTok->is(tok::identifier)) 2337 nextToken(); 2338 2339 if (FormatTok->isNot(tok::l_paren)) 2340 return; 2341 2342 // Parse formal parameter list. 2343 parseParens(); 2344 2345 if (FormatTok->is(tok::colon)) { 2346 // Parse a type definition. 2347 nextToken(); 2348 2349 // Eat the type declaration. For braced inline object types, balance braces, 2350 // otherwise just parse until finding an l_brace for the function body. 2351 if (FormatTok->is(tok::l_brace)) 2352 tryToParseBracedList(); 2353 else 2354 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 2355 nextToken(); 2356 } 2357 2358 if (FormatTok->is(tok::semi)) 2359 return; 2360 2361 parseChildBlock(); 2362 } 2363 2364 bool UnwrappedLineParser::tryToParseBracedList() { 2365 if (FormatTok->is(BK_Unknown)) 2366 calculateBraceTypes(); 2367 assert(FormatTok->isNot(BK_Unknown)); 2368 if (FormatTok->is(BK_Block)) 2369 return false; 2370 nextToken(); 2371 parseBracedList(); 2372 return true; 2373 } 2374 2375 bool UnwrappedLineParser::tryToParseChildBlock() { 2376 assert(Style.isJavaScript() || Style.isCSharp()); 2377 assert(FormatTok->is(TT_FatArrow)); 2378 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType TT_FatArrow. 2379 // They always start an expression or a child block if followed by a curly 2380 // brace. 2381 nextToken(); 2382 if (FormatTok->isNot(tok::l_brace)) 2383 return false; 2384 parseChildBlock(); 2385 return true; 2386 } 2387 2388 bool UnwrappedLineParser::parseBracedList(bool IsAngleBracket, bool IsEnum) { 2389 bool HasError = false; 2390 2391 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 2392 // replace this by using parseAssignmentExpression() inside. 2393 do { 2394 if (Style.isCSharp() && FormatTok->is(TT_FatArrow) && 2395 tryToParseChildBlock()) { 2396 continue; 2397 } 2398 if (Style.isJavaScript()) { 2399 if (FormatTok->is(Keywords.kw_function)) { 2400 tryToParseJSFunction(); 2401 continue; 2402 } 2403 if (FormatTok->is(tok::l_brace)) { 2404 // Could be a method inside of a braced list `{a() { return 1; }}`. 2405 if (tryToParseBracedList()) 2406 continue; 2407 parseChildBlock(); 2408 } 2409 } 2410 if (FormatTok->is(IsAngleBracket ? tok::greater : tok::r_brace)) { 2411 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 2412 addUnwrappedLine(); 2413 nextToken(); 2414 return !HasError; 2415 } 2416 switch (FormatTok->Tok.getKind()) { 2417 case tok::l_square: 2418 if (Style.isCSharp()) 2419 parseSquare(); 2420 else 2421 tryToParseLambda(); 2422 break; 2423 case tok::l_paren: 2424 parseParens(); 2425 // JavaScript can just have free standing methods and getters/setters in 2426 // object literals. Detect them by a "{" following ")". 2427 if (Style.isJavaScript()) { 2428 if (FormatTok->is(tok::l_brace)) 2429 parseChildBlock(); 2430 break; 2431 } 2432 break; 2433 case tok::l_brace: 2434 // Assume there are no blocks inside a braced init list apart 2435 // from the ones we explicitly parse out (like lambdas). 2436 FormatTok->setBlockKind(BK_BracedInit); 2437 nextToken(); 2438 parseBracedList(); 2439 break; 2440 case tok::less: 2441 nextToken(); 2442 if (IsAngleBracket) 2443 parseBracedList(/*IsAngleBracket=*/true); 2444 break; 2445 case tok::semi: 2446 // JavaScript (or more precisely TypeScript) can have semicolons in braced 2447 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 2448 // used for error recovery if we have otherwise determined that this is 2449 // a braced list. 2450 if (Style.isJavaScript()) { 2451 nextToken(); 2452 break; 2453 } 2454 HasError = true; 2455 if (!IsEnum) 2456 return false; 2457 nextToken(); 2458 break; 2459 case tok::comma: 2460 nextToken(); 2461 if (IsEnum && !Style.AllowShortEnumsOnASingleLine) 2462 addUnwrappedLine(); 2463 break; 2464 default: 2465 nextToken(); 2466 break; 2467 } 2468 } while (!eof()); 2469 return false; 2470 } 2471 2472 /// \brief Parses a pair of parentheses (and everything between them). 2473 /// \param AmpAmpTokenType If different than TT_Unknown sets this type for all 2474 /// double ampersands. This applies for all nested scopes as well. 2475 /// 2476 /// Returns whether there is a `=` token between the parentheses. 2477 bool UnwrappedLineParser::parseParens(TokenType AmpAmpTokenType) { 2478 assert(FormatTok->is(tok::l_paren) && "'(' expected."); 2479 auto *LeftParen = FormatTok; 2480 bool SeenEqual = false; 2481 const bool MightBeStmtExpr = Tokens->peekNextToken()->is(tok::l_brace); 2482 nextToken(); 2483 do { 2484 switch (FormatTok->Tok.getKind()) { 2485 case tok::l_paren: 2486 if (parseParens(AmpAmpTokenType)) 2487 SeenEqual = true; 2488 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 2489 parseChildBlock(); 2490 break; 2491 case tok::r_paren: 2492 if (!MightBeStmtExpr && 2493 Style.RemoveParentheses > FormatStyle::RPS_Leave) { 2494 const auto *Prev = LeftParen->Previous; 2495 const auto *Next = Tokens->peekNextToken(); 2496 const bool DoubleParens = 2497 Prev && Prev->is(tok::l_paren) && Next && Next->is(tok::r_paren); 2498 const auto *PrevPrev = Prev ? Prev->getPreviousNonComment() : nullptr; 2499 const bool Blacklisted = 2500 PrevPrev && 2501 (PrevPrev->isOneOf(tok::kw___attribute, tok::kw_decltype) || 2502 (SeenEqual && 2503 (PrevPrev->isOneOf(tok::kw_if, tok::kw_while) || 2504 PrevPrev->endsSequence(tok::kw_constexpr, tok::kw_if)))); 2505 const bool ReturnParens = 2506 Style.RemoveParentheses == FormatStyle::RPS_ReturnStatement && 2507 ((NestedLambdas.empty() && !IsDecltypeAutoFunction) || 2508 (!NestedLambdas.empty() && !NestedLambdas.back())) && 2509 Prev && Prev->isOneOf(tok::kw_return, tok::kw_co_return) && Next && 2510 Next->is(tok::semi); 2511 if ((DoubleParens && !Blacklisted) || ReturnParens) { 2512 LeftParen->Optional = true; 2513 FormatTok->Optional = true; 2514 } 2515 } 2516 nextToken(); 2517 return SeenEqual; 2518 case tok::r_brace: 2519 // A "}" inside parenthesis is an error if there wasn't a matching "{". 2520 return SeenEqual; 2521 case tok::l_square: 2522 tryToParseLambda(); 2523 break; 2524 case tok::l_brace: 2525 if (!tryToParseBracedList()) 2526 parseChildBlock(); 2527 break; 2528 case tok::at: 2529 nextToken(); 2530 if (FormatTok->is(tok::l_brace)) { 2531 nextToken(); 2532 parseBracedList(); 2533 } 2534 break; 2535 case tok::equal: 2536 SeenEqual = true; 2537 if (Style.isCSharp() && FormatTok->is(TT_FatArrow)) 2538 tryToParseChildBlock(); 2539 else 2540 nextToken(); 2541 break; 2542 case tok::kw_class: 2543 if (Style.isJavaScript()) 2544 parseRecord(/*ParseAsExpr=*/true); 2545 else 2546 nextToken(); 2547 break; 2548 case tok::identifier: 2549 if (Style.isJavaScript() && (FormatTok->is(Keywords.kw_function))) 2550 tryToParseJSFunction(); 2551 else 2552 nextToken(); 2553 break; 2554 case tok::kw_requires: { 2555 auto RequiresToken = FormatTok; 2556 nextToken(); 2557 parseRequiresExpression(RequiresToken); 2558 break; 2559 } 2560 case tok::ampamp: 2561 if (AmpAmpTokenType != TT_Unknown) 2562 FormatTok->setFinalizedType(AmpAmpTokenType); 2563 [[fallthrough]]; 2564 default: 2565 nextToken(); 2566 break; 2567 } 2568 } while (!eof()); 2569 return SeenEqual; 2570 } 2571 2572 void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) { 2573 if (!LambdaIntroducer) { 2574 assert(FormatTok->is(tok::l_square) && "'[' expected."); 2575 if (tryToParseLambda()) 2576 return; 2577 } 2578 do { 2579 switch (FormatTok->Tok.getKind()) { 2580 case tok::l_paren: 2581 parseParens(); 2582 break; 2583 case tok::r_square: 2584 nextToken(); 2585 return; 2586 case tok::r_brace: 2587 // A "}" inside parenthesis is an error if there wasn't a matching "{". 2588 return; 2589 case tok::l_square: 2590 parseSquare(); 2591 break; 2592 case tok::l_brace: { 2593 if (!tryToParseBracedList()) 2594 parseChildBlock(); 2595 break; 2596 } 2597 case tok::at: 2598 nextToken(); 2599 if (FormatTok->is(tok::l_brace)) { 2600 nextToken(); 2601 parseBracedList(); 2602 } 2603 break; 2604 default: 2605 nextToken(); 2606 break; 2607 } 2608 } while (!eof()); 2609 } 2610 2611 void UnwrappedLineParser::keepAncestorBraces() { 2612 if (!Style.RemoveBracesLLVM) 2613 return; 2614 2615 const int MaxNestingLevels = 2; 2616 const int Size = NestedTooDeep.size(); 2617 if (Size >= MaxNestingLevels) 2618 NestedTooDeep[Size - MaxNestingLevels] = true; 2619 NestedTooDeep.push_back(false); 2620 } 2621 2622 static FormatToken *getLastNonComment(const UnwrappedLine &Line) { 2623 for (const auto &Token : llvm::reverse(Line.Tokens)) 2624 if (Token.Tok->isNot(tok::comment)) 2625 return Token.Tok; 2626 2627 return nullptr; 2628 } 2629 2630 void UnwrappedLineParser::parseUnbracedBody(bool CheckEOF) { 2631 FormatToken *Tok = nullptr; 2632 2633 if (Style.InsertBraces && !Line->InPPDirective && !Line->Tokens.empty() && 2634 PreprocessorDirectives.empty() && FormatTok->isNot(tok::semi)) { 2635 Tok = Style.BraceWrapping.AfterControlStatement == FormatStyle::BWACS_Never 2636 ? getLastNonComment(*Line) 2637 : Line->Tokens.back().Tok; 2638 assert(Tok); 2639 if (Tok->BraceCount < 0) { 2640 assert(Tok->BraceCount == -1); 2641 Tok = nullptr; 2642 } else { 2643 Tok->BraceCount = -1; 2644 } 2645 } 2646 2647 addUnwrappedLine(); 2648 ++Line->Level; 2649 parseStructuralElement(); 2650 2651 if (Tok) { 2652 assert(!Line->InPPDirective); 2653 Tok = nullptr; 2654 for (const auto &L : llvm::reverse(*CurrentLines)) { 2655 if (!L.InPPDirective && getLastNonComment(L)) { 2656 Tok = L.Tokens.back().Tok; 2657 break; 2658 } 2659 } 2660 assert(Tok); 2661 ++Tok->BraceCount; 2662 } 2663 2664 if (CheckEOF && eof()) 2665 addUnwrappedLine(); 2666 2667 --Line->Level; 2668 } 2669 2670 static void markOptionalBraces(FormatToken *LeftBrace) { 2671 if (!LeftBrace) 2672 return; 2673 2674 assert(LeftBrace->is(tok::l_brace)); 2675 2676 FormatToken *RightBrace = LeftBrace->MatchingParen; 2677 if (!RightBrace) { 2678 assert(!LeftBrace->Optional); 2679 return; 2680 } 2681 2682 assert(RightBrace->is(tok::r_brace)); 2683 assert(RightBrace->MatchingParen == LeftBrace); 2684 assert(LeftBrace->Optional == RightBrace->Optional); 2685 2686 LeftBrace->Optional = true; 2687 RightBrace->Optional = true; 2688 } 2689 2690 void UnwrappedLineParser::handleAttributes() { 2691 // Handle AttributeMacro, e.g. `if (x) UNLIKELY`. 2692 if (FormatTok->isAttribute()) 2693 nextToken(); 2694 else if (FormatTok->is(tok::l_square)) 2695 handleCppAttributes(); 2696 } 2697 2698 bool UnwrappedLineParser::handleCppAttributes() { 2699 // Handle [[likely]] / [[unlikely]] attributes. 2700 assert(FormatTok->is(tok::l_square)); 2701 if (!tryToParseSimpleAttribute()) 2702 return false; 2703 parseSquare(); 2704 return true; 2705 } 2706 2707 /// Returns whether \c Tok begins a block. 2708 bool UnwrappedLineParser::isBlockBegin(const FormatToken &Tok) const { 2709 // FIXME: rename the function or make 2710 // Tok.isOneOf(tok::l_brace, TT_MacroBlockBegin) work. 2711 return Style.isVerilog() ? Keywords.isVerilogBegin(Tok) 2712 : Tok.is(tok::l_brace); 2713 } 2714 2715 FormatToken *UnwrappedLineParser::parseIfThenElse(IfStmtKind *IfKind, 2716 bool KeepBraces, 2717 bool IsVerilogAssert) { 2718 assert((FormatTok->is(tok::kw_if) || 2719 (Style.isVerilog() && 2720 FormatTok->isOneOf(tok::kw_restrict, Keywords.kw_assert, 2721 Keywords.kw_assume, Keywords.kw_cover))) && 2722 "'if' expected"); 2723 nextToken(); 2724 2725 if (IsVerilogAssert) { 2726 // Handle `assert #0` and `assert final`. 2727 if (FormatTok->is(Keywords.kw_verilogHash)) { 2728 nextToken(); 2729 if (FormatTok->is(tok::numeric_constant)) 2730 nextToken(); 2731 } else if (FormatTok->isOneOf(Keywords.kw_final, Keywords.kw_property, 2732 Keywords.kw_sequence)) { 2733 nextToken(); 2734 } 2735 } 2736 2737 // Handle `if !consteval`. 2738 if (FormatTok->is(tok::exclaim)) 2739 nextToken(); 2740 2741 bool KeepIfBraces = true; 2742 if (FormatTok->is(tok::kw_consteval)) { 2743 nextToken(); 2744 } else { 2745 KeepIfBraces = !Style.RemoveBracesLLVM || KeepBraces; 2746 if (FormatTok->isOneOf(tok::kw_constexpr, tok::identifier)) 2747 nextToken(); 2748 if (FormatTok->is(tok::l_paren)) { 2749 FormatTok->setFinalizedType(TT_ConditionLParen); 2750 parseParens(); 2751 } 2752 } 2753 handleAttributes(); 2754 // The then action is optional in Verilog assert statements. 2755 if (IsVerilogAssert && FormatTok->is(tok::semi)) { 2756 nextToken(); 2757 addUnwrappedLine(); 2758 return nullptr; 2759 } 2760 2761 bool NeedsUnwrappedLine = false; 2762 keepAncestorBraces(); 2763 2764 FormatToken *IfLeftBrace = nullptr; 2765 IfStmtKind IfBlockKind = IfStmtKind::NotIf; 2766 2767 if (isBlockBegin(*FormatTok)) { 2768 FormatTok->setFinalizedType(TT_ControlStatementLBrace); 2769 IfLeftBrace = FormatTok; 2770 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2771 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 2772 /*MunchSemi=*/true, KeepIfBraces, &IfBlockKind); 2773 setPreviousRBraceType(TT_ControlStatementRBrace); 2774 if (Style.BraceWrapping.BeforeElse) 2775 addUnwrappedLine(); 2776 else 2777 NeedsUnwrappedLine = true; 2778 } else if (IsVerilogAssert && FormatTok->is(tok::kw_else)) { 2779 addUnwrappedLine(); 2780 } else { 2781 parseUnbracedBody(); 2782 } 2783 2784 if (Style.RemoveBracesLLVM) { 2785 assert(!NestedTooDeep.empty()); 2786 KeepIfBraces = KeepIfBraces || 2787 (IfLeftBrace && !IfLeftBrace->MatchingParen) || 2788 NestedTooDeep.back() || IfBlockKind == IfStmtKind::IfOnly || 2789 IfBlockKind == IfStmtKind::IfElseIf; 2790 } 2791 2792 bool KeepElseBraces = KeepIfBraces; 2793 FormatToken *ElseLeftBrace = nullptr; 2794 IfStmtKind Kind = IfStmtKind::IfOnly; 2795 2796 if (FormatTok->is(tok::kw_else)) { 2797 if (Style.RemoveBracesLLVM) { 2798 NestedTooDeep.back() = false; 2799 Kind = IfStmtKind::IfElse; 2800 } 2801 nextToken(); 2802 handleAttributes(); 2803 if (isBlockBegin(*FormatTok)) { 2804 const bool FollowedByIf = Tokens->peekNextToken()->is(tok::kw_if); 2805 FormatTok->setFinalizedType(TT_ElseLBrace); 2806 ElseLeftBrace = FormatTok; 2807 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2808 IfStmtKind ElseBlockKind = IfStmtKind::NotIf; 2809 FormatToken *IfLBrace = 2810 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 2811 /*MunchSemi=*/true, KeepElseBraces, &ElseBlockKind); 2812 setPreviousRBraceType(TT_ElseRBrace); 2813 if (FormatTok->is(tok::kw_else)) { 2814 KeepElseBraces = KeepElseBraces || 2815 ElseBlockKind == IfStmtKind::IfOnly || 2816 ElseBlockKind == IfStmtKind::IfElseIf; 2817 } else if (FollowedByIf && IfLBrace && !IfLBrace->Optional) { 2818 KeepElseBraces = true; 2819 assert(ElseLeftBrace->MatchingParen); 2820 markOptionalBraces(ElseLeftBrace); 2821 } 2822 addUnwrappedLine(); 2823 } else if (!IsVerilogAssert && FormatTok->is(tok::kw_if)) { 2824 const FormatToken *Previous = Tokens->getPreviousToken(); 2825 assert(Previous); 2826 const bool IsPrecededByComment = Previous->is(tok::comment); 2827 if (IsPrecededByComment) { 2828 addUnwrappedLine(); 2829 ++Line->Level; 2830 } 2831 bool TooDeep = true; 2832 if (Style.RemoveBracesLLVM) { 2833 Kind = IfStmtKind::IfElseIf; 2834 TooDeep = NestedTooDeep.pop_back_val(); 2835 } 2836 ElseLeftBrace = parseIfThenElse(/*IfKind=*/nullptr, KeepIfBraces); 2837 if (Style.RemoveBracesLLVM) 2838 NestedTooDeep.push_back(TooDeep); 2839 if (IsPrecededByComment) 2840 --Line->Level; 2841 } else { 2842 parseUnbracedBody(/*CheckEOF=*/true); 2843 } 2844 } else { 2845 KeepIfBraces = KeepIfBraces || IfBlockKind == IfStmtKind::IfElse; 2846 if (NeedsUnwrappedLine) 2847 addUnwrappedLine(); 2848 } 2849 2850 if (!Style.RemoveBracesLLVM) 2851 return nullptr; 2852 2853 assert(!NestedTooDeep.empty()); 2854 KeepElseBraces = KeepElseBraces || 2855 (ElseLeftBrace && !ElseLeftBrace->MatchingParen) || 2856 NestedTooDeep.back(); 2857 2858 NestedTooDeep.pop_back(); 2859 2860 if (!KeepIfBraces && !KeepElseBraces) { 2861 markOptionalBraces(IfLeftBrace); 2862 markOptionalBraces(ElseLeftBrace); 2863 } else if (IfLeftBrace) { 2864 FormatToken *IfRightBrace = IfLeftBrace->MatchingParen; 2865 if (IfRightBrace) { 2866 assert(IfRightBrace->MatchingParen == IfLeftBrace); 2867 assert(!IfLeftBrace->Optional); 2868 assert(!IfRightBrace->Optional); 2869 IfLeftBrace->MatchingParen = nullptr; 2870 IfRightBrace->MatchingParen = nullptr; 2871 } 2872 } 2873 2874 if (IfKind) 2875 *IfKind = Kind; 2876 2877 return IfLeftBrace; 2878 } 2879 2880 void UnwrappedLineParser::parseTryCatch() { 2881 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 2882 nextToken(); 2883 bool NeedsUnwrappedLine = false; 2884 if (FormatTok->is(tok::colon)) { 2885 // We are in a function try block, what comes is an initializer list. 2886 nextToken(); 2887 2888 // In case identifiers were removed by clang-tidy, what might follow is 2889 // multiple commas in sequence - before the first identifier. 2890 while (FormatTok->is(tok::comma)) 2891 nextToken(); 2892 2893 while (FormatTok->is(tok::identifier)) { 2894 nextToken(); 2895 if (FormatTok->is(tok::l_paren)) 2896 parseParens(); 2897 if (FormatTok->Previous && FormatTok->Previous->is(tok::identifier) && 2898 FormatTok->is(tok::l_brace)) { 2899 do { 2900 nextToken(); 2901 } while (FormatTok->isNot(tok::r_brace)); 2902 nextToken(); 2903 } 2904 2905 // In case identifiers were removed by clang-tidy, what might follow is 2906 // multiple commas in sequence - after the first identifier. 2907 while (FormatTok->is(tok::comma)) 2908 nextToken(); 2909 } 2910 } 2911 // Parse try with resource. 2912 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) 2913 parseParens(); 2914 2915 keepAncestorBraces(); 2916 2917 if (FormatTok->is(tok::l_brace)) { 2918 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2919 parseBlock(); 2920 if (Style.BraceWrapping.BeforeCatch) 2921 addUnwrappedLine(); 2922 else 2923 NeedsUnwrappedLine = true; 2924 } else if (FormatTok->isNot(tok::kw_catch)) { 2925 // The C++ standard requires a compound-statement after a try. 2926 // If there's none, we try to assume there's a structuralElement 2927 // and try to continue. 2928 addUnwrappedLine(); 2929 ++Line->Level; 2930 parseStructuralElement(); 2931 --Line->Level; 2932 } 2933 while (true) { 2934 if (FormatTok->is(tok::at)) 2935 nextToken(); 2936 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 2937 tok::kw___finally) || 2938 ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2939 FormatTok->is(Keywords.kw_finally)) || 2940 (FormatTok->isObjCAtKeyword(tok::objc_catch) || 2941 FormatTok->isObjCAtKeyword(tok::objc_finally)))) { 2942 break; 2943 } 2944 nextToken(); 2945 while (FormatTok->isNot(tok::l_brace)) { 2946 if (FormatTok->is(tok::l_paren)) { 2947 parseParens(); 2948 continue; 2949 } 2950 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) { 2951 if (Style.RemoveBracesLLVM) 2952 NestedTooDeep.pop_back(); 2953 return; 2954 } 2955 nextToken(); 2956 } 2957 NeedsUnwrappedLine = false; 2958 Line->MustBeDeclaration = false; 2959 CompoundStatementIndenter Indenter(this, Style, Line->Level); 2960 parseBlock(); 2961 if (Style.BraceWrapping.BeforeCatch) 2962 addUnwrappedLine(); 2963 else 2964 NeedsUnwrappedLine = true; 2965 } 2966 2967 if (Style.RemoveBracesLLVM) 2968 NestedTooDeep.pop_back(); 2969 2970 if (NeedsUnwrappedLine) 2971 addUnwrappedLine(); 2972 } 2973 2974 void UnwrappedLineParser::parseNamespace() { 2975 assert(FormatTok->isOneOf(tok::kw_namespace, TT_NamespaceMacro) && 2976 "'namespace' expected"); 2977 2978 const FormatToken &InitialToken = *FormatTok; 2979 nextToken(); 2980 if (InitialToken.is(TT_NamespaceMacro)) { 2981 parseParens(); 2982 } else { 2983 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::kw_inline, 2984 tok::l_square, tok::period, tok::l_paren) || 2985 (Style.isCSharp() && FormatTok->is(tok::kw_union))) { 2986 if (FormatTok->is(tok::l_square)) 2987 parseSquare(); 2988 else if (FormatTok->is(tok::l_paren)) 2989 parseParens(); 2990 else 2991 nextToken(); 2992 } 2993 } 2994 if (FormatTok->is(tok::l_brace)) { 2995 FormatTok->setFinalizedType(TT_NamespaceLBrace); 2996 2997 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2998 addUnwrappedLine(); 2999 3000 unsigned AddLevels = 3001 Style.NamespaceIndentation == FormatStyle::NI_All || 3002 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 3003 DeclarationScopeStack.size() > 1) 3004 ? 1u 3005 : 0u; 3006 bool ManageWhitesmithsBraces = 3007 AddLevels == 0u && 3008 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths; 3009 3010 // If we're in Whitesmiths mode, indent the brace if we're not indenting 3011 // the whole block. 3012 if (ManageWhitesmithsBraces) 3013 ++Line->Level; 3014 3015 // Munch the semicolon after a namespace. This is more common than one would 3016 // think. Putting the semicolon into its own line is very ugly. 3017 parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/true, 3018 /*KeepBraces=*/true, /*IfKind=*/nullptr, 3019 ManageWhitesmithsBraces); 3020 3021 addUnwrappedLine(AddLevels > 0 ? LineLevel::Remove : LineLevel::Keep); 3022 3023 if (ManageWhitesmithsBraces) 3024 --Line->Level; 3025 } 3026 // FIXME: Add error handling. 3027 } 3028 3029 void UnwrappedLineParser::parseNew() { 3030 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 3031 nextToken(); 3032 3033 if (Style.isCSharp()) { 3034 do { 3035 // Handle constructor invocation, e.g. `new(field: value)`. 3036 if (FormatTok->is(tok::l_paren)) 3037 parseParens(); 3038 3039 // Handle array initialization syntax, e.g. `new[] {10, 20, 30}`. 3040 if (FormatTok->is(tok::l_brace)) 3041 parseBracedList(); 3042 3043 if (FormatTok->isOneOf(tok::semi, tok::comma)) 3044 return; 3045 3046 nextToken(); 3047 } while (!eof()); 3048 } 3049 3050 if (Style.Language != FormatStyle::LK_Java) 3051 return; 3052 3053 // In Java, we can parse everything up to the parens, which aren't optional. 3054 do { 3055 // There should not be a ;, { or } before the new's open paren. 3056 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 3057 return; 3058 3059 // Consume the parens. 3060 if (FormatTok->is(tok::l_paren)) { 3061 parseParens(); 3062 3063 // If there is a class body of an anonymous class, consume that as child. 3064 if (FormatTok->is(tok::l_brace)) 3065 parseChildBlock(); 3066 return; 3067 } 3068 nextToken(); 3069 } while (!eof()); 3070 } 3071 3072 void UnwrappedLineParser::parseLoopBody(bool KeepBraces, bool WrapRightBrace) { 3073 keepAncestorBraces(); 3074 3075 if (isBlockBegin(*FormatTok)) { 3076 FormatTok->setFinalizedType(TT_ControlStatementLBrace); 3077 FormatToken *LeftBrace = FormatTok; 3078 CompoundStatementIndenter Indenter(this, Style, Line->Level); 3079 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u, 3080 /*MunchSemi=*/true, KeepBraces); 3081 setPreviousRBraceType(TT_ControlStatementRBrace); 3082 if (!KeepBraces) { 3083 assert(!NestedTooDeep.empty()); 3084 if (!NestedTooDeep.back()) 3085 markOptionalBraces(LeftBrace); 3086 } 3087 if (WrapRightBrace) 3088 addUnwrappedLine(); 3089 } else { 3090 parseUnbracedBody(); 3091 } 3092 3093 if (!KeepBraces) 3094 NestedTooDeep.pop_back(); 3095 } 3096 3097 void UnwrappedLineParser::parseForOrWhileLoop(bool HasParens) { 3098 assert((FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) || 3099 (Style.isVerilog() && 3100 FormatTok->isOneOf(Keywords.kw_always, Keywords.kw_always_comb, 3101 Keywords.kw_always_ff, Keywords.kw_always_latch, 3102 Keywords.kw_final, Keywords.kw_initial, 3103 Keywords.kw_foreach, Keywords.kw_forever, 3104 Keywords.kw_repeat))) && 3105 "'for', 'while' or foreach macro expected"); 3106 const bool KeepBraces = !Style.RemoveBracesLLVM || 3107 !FormatTok->isOneOf(tok::kw_for, tok::kw_while); 3108 3109 nextToken(); 3110 // JS' for await ( ... 3111 if (Style.isJavaScript() && FormatTok->is(Keywords.kw_await)) 3112 nextToken(); 3113 if (Style.isCpp() && FormatTok->is(tok::kw_co_await)) 3114 nextToken(); 3115 if (HasParens && FormatTok->is(tok::l_paren)) { 3116 // The type is only set for Verilog basically because we were afraid to 3117 // change the existing behavior for loops. See the discussion on D121756 for 3118 // details. 3119 if (Style.isVerilog()) 3120 FormatTok->setFinalizedType(TT_ConditionLParen); 3121 parseParens(); 3122 } 3123 3124 if (Style.isVerilog()) { 3125 // Event control. 3126 parseVerilogSensitivityList(); 3127 } else if (Style.AllowShortLoopsOnASingleLine && FormatTok->is(tok::semi) && 3128 Tokens->getPreviousToken()->is(tok::r_paren)) { 3129 nextToken(); 3130 addUnwrappedLine(); 3131 return; 3132 } 3133 3134 handleAttributes(); 3135 parseLoopBody(KeepBraces, /*WrapRightBrace=*/true); 3136 } 3137 3138 void UnwrappedLineParser::parseDoWhile() { 3139 assert(FormatTok->is(tok::kw_do) && "'do' expected"); 3140 nextToken(); 3141 3142 parseLoopBody(/*KeepBraces=*/true, Style.BraceWrapping.BeforeWhile); 3143 3144 // FIXME: Add error handling. 3145 if (FormatTok->isNot(tok::kw_while)) { 3146 addUnwrappedLine(); 3147 return; 3148 } 3149 3150 FormatTok->setFinalizedType(TT_DoWhile); 3151 3152 // If in Whitesmiths mode, the line with the while() needs to be indented 3153 // to the same level as the block. 3154 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) 3155 ++Line->Level; 3156 3157 nextToken(); 3158 parseStructuralElement(); 3159 } 3160 3161 void UnwrappedLineParser::parseLabel(bool LeftAlignLabel) { 3162 nextToken(); 3163 unsigned OldLineLevel = Line->Level; 3164 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 3165 --Line->Level; 3166 if (LeftAlignLabel) 3167 Line->Level = 0; 3168 3169 if (!Style.IndentCaseBlocks && CommentsBeforeNextToken.empty() && 3170 FormatTok->is(tok::l_brace)) { 3171 3172 CompoundStatementIndenter Indenter(this, Line->Level, 3173 Style.BraceWrapping.AfterCaseLabel, 3174 Style.BraceWrapping.IndentBraces); 3175 parseBlock(); 3176 if (FormatTok->is(tok::kw_break)) { 3177 if (Style.BraceWrapping.AfterControlStatement == 3178 FormatStyle::BWACS_Always) { 3179 addUnwrappedLine(); 3180 if (!Style.IndentCaseBlocks && 3181 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) { 3182 ++Line->Level; 3183 } 3184 } 3185 parseStructuralElement(); 3186 } 3187 addUnwrappedLine(); 3188 } else { 3189 if (FormatTok->is(tok::semi)) 3190 nextToken(); 3191 addUnwrappedLine(); 3192 } 3193 Line->Level = OldLineLevel; 3194 if (FormatTok->isNot(tok::l_brace)) { 3195 parseStructuralElement(); 3196 addUnwrappedLine(); 3197 } 3198 } 3199 3200 void UnwrappedLineParser::parseCaseLabel() { 3201 assert(FormatTok->is(tok::kw_case) && "'case' expected"); 3202 3203 // FIXME: fix handling of complex expressions here. 3204 do { 3205 nextToken(); 3206 if (FormatTok->is(tok::colon)) { 3207 FormatTok->setFinalizedType(TT_CaseLabelColon); 3208 break; 3209 } 3210 } while (!eof()); 3211 parseLabel(); 3212 } 3213 3214 void UnwrappedLineParser::parseSwitch() { 3215 assert(FormatTok->is(tok::kw_switch) && "'switch' expected"); 3216 nextToken(); 3217 if (FormatTok->is(tok::l_paren)) 3218 parseParens(); 3219 3220 keepAncestorBraces(); 3221 3222 if (FormatTok->is(tok::l_brace)) { 3223 CompoundStatementIndenter Indenter(this, Style, Line->Level); 3224 FormatTok->setFinalizedType(TT_ControlStatementLBrace); 3225 parseBlock(); 3226 setPreviousRBraceType(TT_ControlStatementRBrace); 3227 addUnwrappedLine(); 3228 } else { 3229 addUnwrappedLine(); 3230 ++Line->Level; 3231 parseStructuralElement(); 3232 --Line->Level; 3233 } 3234 3235 if (Style.RemoveBracesLLVM) 3236 NestedTooDeep.pop_back(); 3237 } 3238 3239 // Operators that can follow a C variable. 3240 static bool isCOperatorFollowingVar(tok::TokenKind kind) { 3241 switch (kind) { 3242 case tok::ampamp: 3243 case tok::ampequal: 3244 case tok::arrow: 3245 case tok::caret: 3246 case tok::caretequal: 3247 case tok::comma: 3248 case tok::ellipsis: 3249 case tok::equal: 3250 case tok::equalequal: 3251 case tok::exclaim: 3252 case tok::exclaimequal: 3253 case tok::greater: 3254 case tok::greaterequal: 3255 case tok::greatergreater: 3256 case tok::greatergreaterequal: 3257 case tok::l_paren: 3258 case tok::l_square: 3259 case tok::less: 3260 case tok::lessequal: 3261 case tok::lessless: 3262 case tok::lesslessequal: 3263 case tok::minus: 3264 case tok::minusequal: 3265 case tok::minusminus: 3266 case tok::percent: 3267 case tok::percentequal: 3268 case tok::period: 3269 case tok::pipe: 3270 case tok::pipeequal: 3271 case tok::pipepipe: 3272 case tok::plus: 3273 case tok::plusequal: 3274 case tok::plusplus: 3275 case tok::question: 3276 case tok::r_brace: 3277 case tok::r_paren: 3278 case tok::r_square: 3279 case tok::semi: 3280 case tok::slash: 3281 case tok::slashequal: 3282 case tok::star: 3283 case tok::starequal: 3284 return true; 3285 default: 3286 return false; 3287 } 3288 } 3289 3290 void UnwrappedLineParser::parseAccessSpecifier() { 3291 FormatToken *AccessSpecifierCandidate = FormatTok; 3292 nextToken(); 3293 // Understand Qt's slots. 3294 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 3295 nextToken(); 3296 // Otherwise, we don't know what it is, and we'd better keep the next token. 3297 if (FormatTok->is(tok::colon)) { 3298 nextToken(); 3299 addUnwrappedLine(); 3300 } else if (FormatTok->isNot(tok::coloncolon) && 3301 !isCOperatorFollowingVar(FormatTok->Tok.getKind())) { 3302 // Not a variable name nor namespace name. 3303 addUnwrappedLine(); 3304 } else if (AccessSpecifierCandidate) { 3305 // Consider the access specifier to be a C identifier. 3306 AccessSpecifierCandidate->Tok.setKind(tok::identifier); 3307 } 3308 } 3309 3310 /// \brief Parses a requires, decides if it is a clause or an expression. 3311 /// \pre The current token has to be the requires keyword. 3312 /// \returns true if it parsed a clause. 3313 bool clang::format::UnwrappedLineParser::parseRequires() { 3314 assert(FormatTok->is(tok::kw_requires) && "'requires' expected"); 3315 auto RequiresToken = FormatTok; 3316 3317 // We try to guess if it is a requires clause, or a requires expression. For 3318 // that we first consume the keyword and check the next token. 3319 nextToken(); 3320 3321 switch (FormatTok->Tok.getKind()) { 3322 case tok::l_brace: 3323 // This can only be an expression, never a clause. 3324 parseRequiresExpression(RequiresToken); 3325 return false; 3326 case tok::l_paren: 3327 // Clauses and expression can start with a paren, it's unclear what we have. 3328 break; 3329 default: 3330 // All other tokens can only be a clause. 3331 parseRequiresClause(RequiresToken); 3332 return true; 3333 } 3334 3335 // Looking forward we would have to decide if there are function declaration 3336 // like arguments to the requires expression: 3337 // requires (T t) { 3338 // Or there is a constraint expression for the requires clause: 3339 // requires (C<T> && ... 3340 3341 // But first let's look behind. 3342 auto *PreviousNonComment = RequiresToken->getPreviousNonComment(); 3343 3344 if (!PreviousNonComment || 3345 PreviousNonComment->is(TT_RequiresExpressionLBrace)) { 3346 // If there is no token, or an expression left brace, we are a requires 3347 // clause within a requires expression. 3348 parseRequiresClause(RequiresToken); 3349 return true; 3350 } 3351 3352 switch (PreviousNonComment->Tok.getKind()) { 3353 case tok::greater: 3354 case tok::r_paren: 3355 case tok::kw_noexcept: 3356 case tok::kw_const: 3357 // This is a requires clause. 3358 parseRequiresClause(RequiresToken); 3359 return true; 3360 case tok::amp: 3361 case tok::ampamp: { 3362 // This can be either: 3363 // if (... && requires (T t) ...) 3364 // Or 3365 // void member(...) && requires (C<T> ... 3366 // We check the one token before that for a const: 3367 // void member(...) const && requires (C<T> ... 3368 auto PrevPrev = PreviousNonComment->getPreviousNonComment(); 3369 if (PrevPrev && PrevPrev->is(tok::kw_const)) { 3370 parseRequiresClause(RequiresToken); 3371 return true; 3372 } 3373 break; 3374 } 3375 default: 3376 if (PreviousNonComment->isTypeOrIdentifier()) { 3377 // This is a requires clause. 3378 parseRequiresClause(RequiresToken); 3379 return true; 3380 } 3381 // It's an expression. 3382 parseRequiresExpression(RequiresToken); 3383 return false; 3384 } 3385 3386 // Now we look forward and try to check if the paren content is a parameter 3387 // list. The parameters can be cv-qualified and contain references or 3388 // pointers. 3389 // So we want basically to check for TYPE NAME, but TYPE can contain all kinds 3390 // of stuff: typename, const, *, &, &&, ::, identifiers. 3391 3392 unsigned StoredPosition = Tokens->getPosition(); 3393 FormatToken *NextToken = Tokens->getNextToken(); 3394 int Lookahead = 0; 3395 auto PeekNext = [&Lookahead, &NextToken, this] { 3396 ++Lookahead; 3397 NextToken = Tokens->getNextToken(); 3398 }; 3399 3400 bool FoundType = false; 3401 bool LastWasColonColon = false; 3402 int OpenAngles = 0; 3403 3404 for (; Lookahead < 50; PeekNext()) { 3405 switch (NextToken->Tok.getKind()) { 3406 case tok::kw_volatile: 3407 case tok::kw_const: 3408 case tok::comma: 3409 if (OpenAngles == 0) { 3410 FormatTok = Tokens->setPosition(StoredPosition); 3411 parseRequiresExpression(RequiresToken); 3412 return false; 3413 } 3414 break; 3415 case tok::r_paren: 3416 case tok::pipepipe: 3417 FormatTok = Tokens->setPosition(StoredPosition); 3418 parseRequiresClause(RequiresToken); 3419 return true; 3420 case tok::eof: 3421 // Break out of the loop. 3422 Lookahead = 50; 3423 break; 3424 case tok::coloncolon: 3425 LastWasColonColon = true; 3426 break; 3427 case tok::identifier: 3428 if (FoundType && !LastWasColonColon && OpenAngles == 0) { 3429 FormatTok = Tokens->setPosition(StoredPosition); 3430 parseRequiresExpression(RequiresToken); 3431 return false; 3432 } 3433 FoundType = true; 3434 LastWasColonColon = false; 3435 break; 3436 case tok::less: 3437 ++OpenAngles; 3438 break; 3439 case tok::greater: 3440 --OpenAngles; 3441 break; 3442 default: 3443 if (NextToken->isSimpleTypeSpecifier()) { 3444 FormatTok = Tokens->setPosition(StoredPosition); 3445 parseRequiresExpression(RequiresToken); 3446 return false; 3447 } 3448 break; 3449 } 3450 } 3451 // This seems to be a complicated expression, just assume it's a clause. 3452 FormatTok = Tokens->setPosition(StoredPosition); 3453 parseRequiresClause(RequiresToken); 3454 return true; 3455 } 3456 3457 /// \brief Parses a requires clause. 3458 /// \param RequiresToken The requires keyword token, which starts this clause. 3459 /// \pre We need to be on the next token after the requires keyword. 3460 /// \sa parseRequiresExpression 3461 /// 3462 /// Returns if it either has finished parsing the clause, or it detects, that 3463 /// the clause is incorrect. 3464 void UnwrappedLineParser::parseRequiresClause(FormatToken *RequiresToken) { 3465 assert(FormatTok->getPreviousNonComment() == RequiresToken); 3466 assert(RequiresToken->is(tok::kw_requires) && "'requires' expected"); 3467 3468 // If there is no previous token, we are within a requires expression, 3469 // otherwise we will always have the template or function declaration in front 3470 // of it. 3471 bool InRequiresExpression = 3472 !RequiresToken->Previous || 3473 RequiresToken->Previous->is(TT_RequiresExpressionLBrace); 3474 3475 RequiresToken->setFinalizedType(InRequiresExpression 3476 ? TT_RequiresClauseInARequiresExpression 3477 : TT_RequiresClause); 3478 3479 // NOTE: parseConstraintExpression is only ever called from this function. 3480 // It could be inlined into here. 3481 parseConstraintExpression(); 3482 3483 if (!InRequiresExpression) 3484 FormatTok->Previous->ClosesRequiresClause = true; 3485 } 3486 3487 /// \brief Parses a requires expression. 3488 /// \param RequiresToken The requires keyword token, which starts this clause. 3489 /// \pre We need to be on the next token after the requires keyword. 3490 /// \sa parseRequiresClause 3491 /// 3492 /// Returns if it either has finished parsing the expression, or it detects, 3493 /// that the expression is incorrect. 3494 void UnwrappedLineParser::parseRequiresExpression(FormatToken *RequiresToken) { 3495 assert(FormatTok->getPreviousNonComment() == RequiresToken); 3496 assert(RequiresToken->is(tok::kw_requires) && "'requires' expected"); 3497 3498 RequiresToken->setFinalizedType(TT_RequiresExpression); 3499 3500 if (FormatTok->is(tok::l_paren)) { 3501 FormatTok->setFinalizedType(TT_RequiresExpressionLParen); 3502 parseParens(); 3503 } 3504 3505 if (FormatTok->is(tok::l_brace)) { 3506 FormatTok->setFinalizedType(TT_RequiresExpressionLBrace); 3507 parseChildBlock(); 3508 } 3509 } 3510 3511 /// \brief Parses a constraint expression. 3512 /// 3513 /// This is the body of a requires clause. It returns, when the parsing is 3514 /// complete, or the expression is incorrect. 3515 void UnwrappedLineParser::parseConstraintExpression() { 3516 // The special handling for lambdas is needed since tryToParseLambda() eats a 3517 // token and if a requires expression is the last part of a requires clause 3518 // and followed by an attribute like [[nodiscard]] the ClosesRequiresClause is 3519 // not set on the correct token. Thus we need to be aware if we even expect a 3520 // lambda to be possible. 3521 // template <typename T> requires requires { ... } [[nodiscard]] ...; 3522 bool LambdaNextTimeAllowed = true; 3523 3524 // Within lambda declarations, it is permitted to put a requires clause after 3525 // its template parameter list, which would place the requires clause right 3526 // before the parentheses of the parameters of the lambda declaration. Thus, 3527 // we track if we expect to see grouping parentheses at all. 3528 // Without this check, `requires foo<T> (T t)` in the below example would be 3529 // seen as the whole requires clause, accidentally eating the parameters of 3530 // the lambda. 3531 // [&]<typename T> requires foo<T> (T t) { ... }; 3532 bool TopLevelParensAllowed = true; 3533 3534 do { 3535 bool LambdaThisTimeAllowed = std::exchange(LambdaNextTimeAllowed, false); 3536 3537 switch (FormatTok->Tok.getKind()) { 3538 case tok::kw_requires: { 3539 auto RequiresToken = FormatTok; 3540 nextToken(); 3541 parseRequiresExpression(RequiresToken); 3542 break; 3543 } 3544 3545 case tok::l_paren: 3546 if (!TopLevelParensAllowed) 3547 return; 3548 parseParens(/*AmpAmpTokenType=*/TT_BinaryOperator); 3549 TopLevelParensAllowed = false; 3550 break; 3551 3552 case tok::l_square: 3553 if (!LambdaThisTimeAllowed || !tryToParseLambda()) 3554 return; 3555 break; 3556 3557 case tok::kw_const: 3558 case tok::semi: 3559 case tok::kw_class: 3560 case tok::kw_struct: 3561 case tok::kw_union: 3562 return; 3563 3564 case tok::l_brace: 3565 // Potential function body. 3566 return; 3567 3568 case tok::ampamp: 3569 case tok::pipepipe: 3570 FormatTok->setFinalizedType(TT_BinaryOperator); 3571 nextToken(); 3572 LambdaNextTimeAllowed = true; 3573 TopLevelParensAllowed = true; 3574 break; 3575 3576 case tok::comma: 3577 case tok::comment: 3578 LambdaNextTimeAllowed = LambdaThisTimeAllowed; 3579 nextToken(); 3580 break; 3581 3582 case tok::kw_sizeof: 3583 case tok::greater: 3584 case tok::greaterequal: 3585 case tok::greatergreater: 3586 case tok::less: 3587 case tok::lessequal: 3588 case tok::lessless: 3589 case tok::equalequal: 3590 case tok::exclaim: 3591 case tok::exclaimequal: 3592 case tok::plus: 3593 case tok::minus: 3594 case tok::star: 3595 case tok::slash: 3596 LambdaNextTimeAllowed = true; 3597 TopLevelParensAllowed = true; 3598 // Just eat them. 3599 nextToken(); 3600 break; 3601 3602 case tok::numeric_constant: 3603 case tok::coloncolon: 3604 case tok::kw_true: 3605 case tok::kw_false: 3606 TopLevelParensAllowed = false; 3607 // Just eat them. 3608 nextToken(); 3609 break; 3610 3611 case tok::kw_static_cast: 3612 case tok::kw_const_cast: 3613 case tok::kw_reinterpret_cast: 3614 case tok::kw_dynamic_cast: 3615 nextToken(); 3616 if (FormatTok->isNot(tok::less)) 3617 return; 3618 3619 nextToken(); 3620 parseBracedList(/*IsAngleBracket=*/true); 3621 break; 3622 3623 default: 3624 if (!FormatTok->Tok.getIdentifierInfo()) { 3625 // Identifiers are part of the default case, we check for more then 3626 // tok::identifier to handle builtin type traits. 3627 return; 3628 } 3629 3630 // We need to differentiate identifiers for a template deduction guide, 3631 // variables, or function return types (the constraint expression has 3632 // ended before that), and basically all other cases. But it's easier to 3633 // check the other way around. 3634 assert(FormatTok->Previous); 3635 switch (FormatTok->Previous->Tok.getKind()) { 3636 case tok::coloncolon: // Nested identifier. 3637 case tok::ampamp: // Start of a function or variable for the 3638 case tok::pipepipe: // constraint expression. (binary) 3639 case tok::exclaim: // The same as above, but unary. 3640 case tok::kw_requires: // Initial identifier of a requires clause. 3641 case tok::equal: // Initial identifier of a concept declaration. 3642 break; 3643 default: 3644 return; 3645 } 3646 3647 // Read identifier with optional template declaration. 3648 nextToken(); 3649 if (FormatTok->is(tok::less)) { 3650 nextToken(); 3651 parseBracedList(/*IsAngleBracket=*/true); 3652 } 3653 TopLevelParensAllowed = false; 3654 break; 3655 } 3656 } while (!eof()); 3657 } 3658 3659 bool UnwrappedLineParser::parseEnum() { 3660 const FormatToken &InitialToken = *FormatTok; 3661 3662 // Won't be 'enum' for NS_ENUMs. 3663 if (FormatTok->is(tok::kw_enum)) 3664 nextToken(); 3665 3666 // In TypeScript, "enum" can also be used as property name, e.g. in interface 3667 // declarations. An "enum" keyword followed by a colon would be a syntax 3668 // error and thus assume it is just an identifier. 3669 if (Style.isJavaScript() && FormatTok->isOneOf(tok::colon, tok::question)) 3670 return false; 3671 3672 // In protobuf, "enum" can be used as a field name. 3673 if (Style.Language == FormatStyle::LK_Proto && FormatTok->is(tok::equal)) 3674 return false; 3675 3676 // Eat up enum class ... 3677 if (FormatTok->isOneOf(tok::kw_class, tok::kw_struct)) 3678 nextToken(); 3679 3680 while (FormatTok->Tok.getIdentifierInfo() || 3681 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 3682 tok::greater, tok::comma, tok::question, 3683 tok::l_square, tok::r_square)) { 3684 if (Style.isVerilog()) { 3685 FormatTok->setFinalizedType(TT_VerilogDimensionedTypeName); 3686 nextToken(); 3687 // In Verilog the base type can have dimensions. 3688 while (FormatTok->is(tok::l_square)) 3689 parseSquare(); 3690 } else { 3691 nextToken(); 3692 } 3693 // We can have macros or attributes in between 'enum' and the enum name. 3694 if (FormatTok->is(tok::l_paren)) 3695 parseParens(); 3696 assert(FormatTok->isNot(TT_AttributeSquare)); 3697 if (FormatTok->is(tok::identifier)) { 3698 nextToken(); 3699 // If there are two identifiers in a row, this is likely an elaborate 3700 // return type. In Java, this can be "implements", etc. 3701 if (Style.isCpp() && FormatTok->is(tok::identifier)) 3702 return false; 3703 } 3704 } 3705 3706 // Just a declaration or something is wrong. 3707 if (FormatTok->isNot(tok::l_brace)) 3708 return true; 3709 FormatTok->setFinalizedType(TT_EnumLBrace); 3710 FormatTok->setBlockKind(BK_Block); 3711 3712 if (Style.Language == FormatStyle::LK_Java) { 3713 // Java enums are different. 3714 parseJavaEnumBody(); 3715 return true; 3716 } 3717 if (Style.Language == FormatStyle::LK_Proto) { 3718 parseBlock(/*MustBeDeclaration=*/true); 3719 return true; 3720 } 3721 3722 if (!Style.AllowShortEnumsOnASingleLine && 3723 ShouldBreakBeforeBrace(Style, InitialToken)) { 3724 addUnwrappedLine(); 3725 } 3726 // Parse enum body. 3727 nextToken(); 3728 if (!Style.AllowShortEnumsOnASingleLine) { 3729 addUnwrappedLine(); 3730 Line->Level += 1; 3731 } 3732 bool HasError = !parseBracedList(/*IsAngleBracket=*/false, /*IsEnum=*/true); 3733 if (!Style.AllowShortEnumsOnASingleLine) 3734 Line->Level -= 1; 3735 if (HasError) { 3736 if (FormatTok->is(tok::semi)) 3737 nextToken(); 3738 addUnwrappedLine(); 3739 } 3740 setPreviousRBraceType(TT_EnumRBrace); 3741 return true; 3742 3743 // There is no addUnwrappedLine() here so that we fall through to parsing a 3744 // structural element afterwards. Thus, in "enum A {} n, m;", 3745 // "} n, m;" will end up in one unwrapped line. 3746 } 3747 3748 bool UnwrappedLineParser::parseStructLike() { 3749 // parseRecord falls through and does not yet add an unwrapped line as a 3750 // record declaration or definition can start a structural element. 3751 parseRecord(); 3752 // This does not apply to Java, JavaScript and C#. 3753 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 3754 Style.isCSharp()) { 3755 if (FormatTok->is(tok::semi)) 3756 nextToken(); 3757 addUnwrappedLine(); 3758 return true; 3759 } 3760 return false; 3761 } 3762 3763 namespace { 3764 // A class used to set and restore the Token position when peeking 3765 // ahead in the token source. 3766 class ScopedTokenPosition { 3767 unsigned StoredPosition; 3768 FormatTokenSource *Tokens; 3769 3770 public: 3771 ScopedTokenPosition(FormatTokenSource *Tokens) : Tokens(Tokens) { 3772 assert(Tokens && "Tokens expected to not be null"); 3773 StoredPosition = Tokens->getPosition(); 3774 } 3775 3776 ~ScopedTokenPosition() { Tokens->setPosition(StoredPosition); } 3777 }; 3778 } // namespace 3779 3780 // Look to see if we have [[ by looking ahead, if 3781 // its not then rewind to the original position. 3782 bool UnwrappedLineParser::tryToParseSimpleAttribute() { 3783 ScopedTokenPosition AutoPosition(Tokens); 3784 FormatToken *Tok = Tokens->getNextToken(); 3785 // We already read the first [ check for the second. 3786 if (Tok->isNot(tok::l_square)) 3787 return false; 3788 // Double check that the attribute is just something 3789 // fairly simple. 3790 while (Tok->isNot(tok::eof)) { 3791 if (Tok->is(tok::r_square)) 3792 break; 3793 Tok = Tokens->getNextToken(); 3794 } 3795 if (Tok->is(tok::eof)) 3796 return false; 3797 Tok = Tokens->getNextToken(); 3798 if (Tok->isNot(tok::r_square)) 3799 return false; 3800 Tok = Tokens->getNextToken(); 3801 if (Tok->is(tok::semi)) 3802 return false; 3803 return true; 3804 } 3805 3806 void UnwrappedLineParser::parseJavaEnumBody() { 3807 assert(FormatTok->is(tok::l_brace)); 3808 const FormatToken *OpeningBrace = FormatTok; 3809 3810 // Determine whether the enum is simple, i.e. does not have a semicolon or 3811 // constants with class bodies. Simple enums can be formatted like braced 3812 // lists, contracted to a single line, etc. 3813 unsigned StoredPosition = Tokens->getPosition(); 3814 bool IsSimple = true; 3815 FormatToken *Tok = Tokens->getNextToken(); 3816 while (Tok->isNot(tok::eof)) { 3817 if (Tok->is(tok::r_brace)) 3818 break; 3819 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 3820 IsSimple = false; 3821 break; 3822 } 3823 // FIXME: This will also mark enums with braces in the arguments to enum 3824 // constants as "not simple". This is probably fine in practice, though. 3825 Tok = Tokens->getNextToken(); 3826 } 3827 FormatTok = Tokens->setPosition(StoredPosition); 3828 3829 if (IsSimple) { 3830 nextToken(); 3831 parseBracedList(); 3832 addUnwrappedLine(); 3833 return; 3834 } 3835 3836 // Parse the body of a more complex enum. 3837 // First add a line for everything up to the "{". 3838 nextToken(); 3839 addUnwrappedLine(); 3840 ++Line->Level; 3841 3842 // Parse the enum constants. 3843 while (!eof()) { 3844 if (FormatTok->is(tok::l_brace)) { 3845 // Parse the constant's class body. 3846 parseBlock(/*MustBeDeclaration=*/true, /*AddLevels=*/1u, 3847 /*MunchSemi=*/false); 3848 } else if (FormatTok->is(tok::l_paren)) { 3849 parseParens(); 3850 } else if (FormatTok->is(tok::comma)) { 3851 nextToken(); 3852 addUnwrappedLine(); 3853 } else if (FormatTok->is(tok::semi)) { 3854 nextToken(); 3855 addUnwrappedLine(); 3856 break; 3857 } else if (FormatTok->is(tok::r_brace)) { 3858 addUnwrappedLine(); 3859 break; 3860 } else { 3861 nextToken(); 3862 } 3863 } 3864 3865 // Parse the class body after the enum's ";" if any. 3866 parseLevel(OpeningBrace); 3867 nextToken(); 3868 --Line->Level; 3869 addUnwrappedLine(); 3870 } 3871 3872 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) { 3873 const FormatToken &InitialToken = *FormatTok; 3874 nextToken(); 3875 3876 // The actual identifier can be a nested name specifier, and in macros 3877 // it is often token-pasted. 3878 // An [[attribute]] can be before the identifier. 3879 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 3880 tok::kw_alignas, tok::l_square) || 3881 FormatTok->isAttribute() || 3882 ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 3883 FormatTok->isOneOf(tok::period, tok::comma))) { 3884 if (Style.isJavaScript() && 3885 FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) { 3886 // JavaScript/TypeScript supports inline object types in 3887 // extends/implements positions: 3888 // class Foo implements {bar: number} { } 3889 nextToken(); 3890 if (FormatTok->is(tok::l_brace)) { 3891 tryToParseBracedList(); 3892 continue; 3893 } 3894 } 3895 if (FormatTok->is(tok::l_square) && handleCppAttributes()) 3896 continue; 3897 bool IsNonMacroIdentifier = 3898 FormatTok->is(tok::identifier) && 3899 FormatTok->TokenText != FormatTok->TokenText.upper(); 3900 nextToken(); 3901 // We can have macros in between 'class' and the class name. 3902 if (!IsNonMacroIdentifier && FormatTok->is(tok::l_paren)) 3903 parseParens(); 3904 } 3905 3906 // Note that parsing away template declarations here leads to incorrectly 3907 // accepting function declarations as record declarations. 3908 // In general, we cannot solve this problem. Consider: 3909 // class A<int> B() {} 3910 // which can be a function definition or a class definition when B() is a 3911 // macro. If we find enough real-world cases where this is a problem, we 3912 // can parse for the 'template' keyword in the beginning of the statement, 3913 // and thus rule out the record production in case there is no template 3914 // (this would still leave us with an ambiguity between template function 3915 // and class declarations). 3916 if (FormatTok->isOneOf(tok::colon, tok::less)) { 3917 do { 3918 if (FormatTok->is(tok::l_brace)) { 3919 calculateBraceTypes(/*ExpectClassBody=*/true); 3920 if (!tryToParseBracedList()) 3921 break; 3922 } 3923 if (FormatTok->is(tok::l_square)) { 3924 FormatToken *Previous = FormatTok->Previous; 3925 if (!Previous || 3926 !(Previous->is(tok::r_paren) || Previous->isTypeOrIdentifier())) { 3927 // Don't try parsing a lambda if we had a closing parenthesis before, 3928 // it was probably a pointer to an array: int (*)[]. 3929 if (!tryToParseLambda()) 3930 continue; 3931 } else { 3932 parseSquare(); 3933 continue; 3934 } 3935 } 3936 if (FormatTok->is(tok::semi)) 3937 return; 3938 if (Style.isCSharp() && FormatTok->is(Keywords.kw_where)) { 3939 addUnwrappedLine(); 3940 nextToken(); 3941 parseCSharpGenericTypeConstraint(); 3942 break; 3943 } 3944 nextToken(); 3945 } while (!eof()); 3946 } 3947 3948 auto GetBraceTypes = 3949 [](const FormatToken &RecordTok) -> std::pair<TokenType, TokenType> { 3950 switch (RecordTok.Tok.getKind()) { 3951 case tok::kw_class: 3952 return {TT_ClassLBrace, TT_ClassRBrace}; 3953 case tok::kw_struct: 3954 return {TT_StructLBrace, TT_StructRBrace}; 3955 case tok::kw_union: 3956 return {TT_UnionLBrace, TT_UnionRBrace}; 3957 default: 3958 // Useful for e.g. interface. 3959 return {TT_RecordLBrace, TT_RecordRBrace}; 3960 } 3961 }; 3962 if (FormatTok->is(tok::l_brace)) { 3963 auto [OpenBraceType, ClosingBraceType] = GetBraceTypes(InitialToken); 3964 FormatTok->setFinalizedType(OpenBraceType); 3965 if (ParseAsExpr) { 3966 parseChildBlock(); 3967 } else { 3968 if (ShouldBreakBeforeBrace(Style, InitialToken)) 3969 addUnwrappedLine(); 3970 3971 unsigned AddLevels = Style.IndentAccessModifiers ? 2u : 1u; 3972 parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/false); 3973 } 3974 setPreviousRBraceType(ClosingBraceType); 3975 } 3976 // There is no addUnwrappedLine() here so that we fall through to parsing a 3977 // structural element afterwards. Thus, in "class A {} n, m;", 3978 // "} n, m;" will end up in one unwrapped line. 3979 } 3980 3981 void UnwrappedLineParser::parseObjCMethod() { 3982 assert(FormatTok->isOneOf(tok::l_paren, tok::identifier) && 3983 "'(' or identifier expected."); 3984 do { 3985 if (FormatTok->is(tok::semi)) { 3986 nextToken(); 3987 addUnwrappedLine(); 3988 return; 3989 } else if (FormatTok->is(tok::l_brace)) { 3990 if (Style.BraceWrapping.AfterFunction) 3991 addUnwrappedLine(); 3992 parseBlock(); 3993 addUnwrappedLine(); 3994 return; 3995 } else { 3996 nextToken(); 3997 } 3998 } while (!eof()); 3999 } 4000 4001 void UnwrappedLineParser::parseObjCProtocolList() { 4002 assert(FormatTok->is(tok::less) && "'<' expected."); 4003 do { 4004 nextToken(); 4005 // Early exit in case someone forgot a close angle. 4006 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 4007 FormatTok->isObjCAtKeyword(tok::objc_end)) { 4008 return; 4009 } 4010 } while (!eof() && FormatTok->isNot(tok::greater)); 4011 nextToken(); // Skip '>'. 4012 } 4013 4014 void UnwrappedLineParser::parseObjCUntilAtEnd() { 4015 do { 4016 if (FormatTok->isObjCAtKeyword(tok::objc_end)) { 4017 nextToken(); 4018 addUnwrappedLine(); 4019 break; 4020 } 4021 if (FormatTok->is(tok::l_brace)) { 4022 parseBlock(); 4023 // In ObjC interfaces, nothing should be following the "}". 4024 addUnwrappedLine(); 4025 } else if (FormatTok->is(tok::r_brace)) { 4026 // Ignore stray "}". parseStructuralElement doesn't consume them. 4027 nextToken(); 4028 addUnwrappedLine(); 4029 } else if (FormatTok->isOneOf(tok::minus, tok::plus)) { 4030 nextToken(); 4031 parseObjCMethod(); 4032 } else { 4033 parseStructuralElement(); 4034 } 4035 } while (!eof()); 4036 } 4037 4038 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 4039 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_interface || 4040 FormatTok->Tok.getObjCKeywordID() == tok::objc_implementation); 4041 nextToken(); 4042 nextToken(); // interface name 4043 4044 // @interface can be followed by a lightweight generic 4045 // specialization list, then either a base class or a category. 4046 if (FormatTok->is(tok::less)) 4047 parseObjCLightweightGenerics(); 4048 if (FormatTok->is(tok::colon)) { 4049 nextToken(); 4050 nextToken(); // base class name 4051 // The base class can also have lightweight generics applied to it. 4052 if (FormatTok->is(tok::less)) 4053 parseObjCLightweightGenerics(); 4054 } else if (FormatTok->is(tok::l_paren)) { 4055 // Skip category, if present. 4056 parseParens(); 4057 } 4058 4059 if (FormatTok->is(tok::less)) 4060 parseObjCProtocolList(); 4061 4062 if (FormatTok->is(tok::l_brace)) { 4063 if (Style.BraceWrapping.AfterObjCDeclaration) 4064 addUnwrappedLine(); 4065 parseBlock(/*MustBeDeclaration=*/true); 4066 } 4067 4068 // With instance variables, this puts '}' on its own line. Without instance 4069 // variables, this ends the @interface line. 4070 addUnwrappedLine(); 4071 4072 parseObjCUntilAtEnd(); 4073 } 4074 4075 void UnwrappedLineParser::parseObjCLightweightGenerics() { 4076 assert(FormatTok->is(tok::less)); 4077 // Unlike protocol lists, generic parameterizations support 4078 // nested angles: 4079 // 4080 // @interface Foo<ValueType : id <NSCopying, NSSecureCoding>> : 4081 // NSObject <NSCopying, NSSecureCoding> 4082 // 4083 // so we need to count how many open angles we have left. 4084 unsigned NumOpenAngles = 1; 4085 do { 4086 nextToken(); 4087 // Early exit in case someone forgot a close angle. 4088 if (FormatTok->isOneOf(tok::semi, tok::l_brace) || 4089 FormatTok->isObjCAtKeyword(tok::objc_end)) { 4090 break; 4091 } 4092 if (FormatTok->is(tok::less)) { 4093 ++NumOpenAngles; 4094 } else if (FormatTok->is(tok::greater)) { 4095 assert(NumOpenAngles > 0 && "'>' makes NumOpenAngles negative"); 4096 --NumOpenAngles; 4097 } 4098 } while (!eof() && NumOpenAngles != 0); 4099 nextToken(); // Skip '>'. 4100 } 4101 4102 // Returns true for the declaration/definition form of @protocol, 4103 // false for the expression form. 4104 bool UnwrappedLineParser::parseObjCProtocol() { 4105 assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_protocol); 4106 nextToken(); 4107 4108 if (FormatTok->is(tok::l_paren)) { 4109 // The expression form of @protocol, e.g. "Protocol* p = @protocol(foo);". 4110 return false; 4111 } 4112 4113 // The definition/declaration form, 4114 // @protocol Foo 4115 // - (int)someMethod; 4116 // @end 4117 4118 nextToken(); // protocol name 4119 4120 if (FormatTok->is(tok::less)) 4121 parseObjCProtocolList(); 4122 4123 // Check for protocol declaration. 4124 if (FormatTok->is(tok::semi)) { 4125 nextToken(); 4126 addUnwrappedLine(); 4127 return true; 4128 } 4129 4130 addUnwrappedLine(); 4131 parseObjCUntilAtEnd(); 4132 return true; 4133 } 4134 4135 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 4136 bool IsImport = FormatTok->is(Keywords.kw_import); 4137 assert(IsImport || FormatTok->is(tok::kw_export)); 4138 nextToken(); 4139 4140 // Consume the "default" in "export default class/function". 4141 if (FormatTok->is(tok::kw_default)) 4142 nextToken(); 4143 4144 // Consume "async function", "function" and "default function", so that these 4145 // get parsed as free-standing JS functions, i.e. do not require a trailing 4146 // semicolon. 4147 if (FormatTok->is(Keywords.kw_async)) 4148 nextToken(); 4149 if (FormatTok->is(Keywords.kw_function)) { 4150 nextToken(); 4151 return; 4152 } 4153 4154 // For imports, `export *`, `export {...}`, consume the rest of the line up 4155 // to the terminating `;`. For everything else, just return and continue 4156 // parsing the structural element, i.e. the declaration or expression for 4157 // `export default`. 4158 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 4159 !FormatTok->isStringLiteral() && 4160 !(FormatTok->is(Keywords.kw_type) && 4161 Tokens->peekNextToken()->isOneOf(tok::l_brace, tok::star))) { 4162 return; 4163 } 4164 4165 while (!eof()) { 4166 if (FormatTok->is(tok::semi)) 4167 return; 4168 if (Line->Tokens.empty()) { 4169 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 4170 // import statement should terminate. 4171 return; 4172 } 4173 if (FormatTok->is(tok::l_brace)) { 4174 FormatTok->setBlockKind(BK_Block); 4175 nextToken(); 4176 parseBracedList(); 4177 } else { 4178 nextToken(); 4179 } 4180 } 4181 } 4182 4183 void UnwrappedLineParser::parseStatementMacro() { 4184 nextToken(); 4185 if (FormatTok->is(tok::l_paren)) 4186 parseParens(); 4187 if (FormatTok->is(tok::semi)) 4188 nextToken(); 4189 addUnwrappedLine(); 4190 } 4191 4192 void UnwrappedLineParser::parseVerilogHierarchyIdentifier() { 4193 // consume things like a::`b.c[d:e] or a::* 4194 while (true) { 4195 if (FormatTok->isOneOf(tok::star, tok::period, tok::periodstar, 4196 tok::coloncolon, tok::hash) || 4197 Keywords.isVerilogIdentifier(*FormatTok)) { 4198 nextToken(); 4199 } else if (FormatTok->is(tok::l_square)) { 4200 parseSquare(); 4201 } else { 4202 break; 4203 } 4204 } 4205 } 4206 4207 void UnwrappedLineParser::parseVerilogSensitivityList() { 4208 if (FormatTok->isNot(tok::at)) 4209 return; 4210 nextToken(); 4211 // A block event expression has 2 at signs. 4212 if (FormatTok->is(tok::at)) 4213 nextToken(); 4214 switch (FormatTok->Tok.getKind()) { 4215 case tok::star: 4216 nextToken(); 4217 break; 4218 case tok::l_paren: 4219 parseParens(); 4220 break; 4221 default: 4222 parseVerilogHierarchyIdentifier(); 4223 break; 4224 } 4225 } 4226 4227 unsigned UnwrappedLineParser::parseVerilogHierarchyHeader() { 4228 unsigned AddLevels = 0; 4229 4230 if (FormatTok->is(Keywords.kw_clocking)) { 4231 nextToken(); 4232 if (Keywords.isVerilogIdentifier(*FormatTok)) 4233 nextToken(); 4234 parseVerilogSensitivityList(); 4235 if (FormatTok->is(tok::semi)) 4236 nextToken(); 4237 } else if (FormatTok->isOneOf(tok::kw_case, Keywords.kw_casex, 4238 Keywords.kw_casez, Keywords.kw_randcase, 4239 Keywords.kw_randsequence)) { 4240 if (Style.IndentCaseLabels) 4241 AddLevels++; 4242 nextToken(); 4243 if (FormatTok->is(tok::l_paren)) { 4244 FormatTok->setFinalizedType(TT_ConditionLParen); 4245 parseParens(); 4246 } 4247 if (FormatTok->isOneOf(Keywords.kw_inside, Keywords.kw_matches)) 4248 nextToken(); 4249 // The case header has no semicolon. 4250 } else { 4251 // "module" etc. 4252 nextToken(); 4253 // all the words like the name of the module and specifiers like 4254 // "automatic" and the width of function return type 4255 while (true) { 4256 if (FormatTok->is(tok::l_square)) { 4257 auto Prev = FormatTok->getPreviousNonComment(); 4258 if (Prev && Keywords.isVerilogIdentifier(*Prev)) 4259 Prev->setFinalizedType(TT_VerilogDimensionedTypeName); 4260 parseSquare(); 4261 } else if (Keywords.isVerilogIdentifier(*FormatTok) || 4262 FormatTok->isOneOf(Keywords.kw_automatic, tok::kw_static)) { 4263 nextToken(); 4264 } else { 4265 break; 4266 } 4267 } 4268 4269 auto NewLine = [this]() { 4270 addUnwrappedLine(); 4271 Line->IsContinuation = true; 4272 }; 4273 4274 // package imports 4275 while (FormatTok->is(Keywords.kw_import)) { 4276 NewLine(); 4277 nextToken(); 4278 parseVerilogHierarchyIdentifier(); 4279 if (FormatTok->is(tok::semi)) 4280 nextToken(); 4281 } 4282 4283 // parameters and ports 4284 if (FormatTok->is(Keywords.kw_verilogHash)) { 4285 NewLine(); 4286 nextToken(); 4287 if (FormatTok->is(tok::l_paren)) { 4288 FormatTok->setFinalizedType(TT_VerilogMultiLineListLParen); 4289 parseParens(); 4290 } 4291 } 4292 if (FormatTok->is(tok::l_paren)) { 4293 NewLine(); 4294 FormatTok->setFinalizedType(TT_VerilogMultiLineListLParen); 4295 parseParens(); 4296 } 4297 4298 // extends and implements 4299 if (FormatTok->is(Keywords.kw_extends)) { 4300 NewLine(); 4301 nextToken(); 4302 parseVerilogHierarchyIdentifier(); 4303 if (FormatTok->is(tok::l_paren)) 4304 parseParens(); 4305 } 4306 if (FormatTok->is(Keywords.kw_implements)) { 4307 NewLine(); 4308 do { 4309 nextToken(); 4310 parseVerilogHierarchyIdentifier(); 4311 } while (FormatTok->is(tok::comma)); 4312 } 4313 4314 // Coverage event for cover groups. 4315 if (FormatTok->is(tok::at)) { 4316 NewLine(); 4317 parseVerilogSensitivityList(); 4318 } 4319 4320 if (FormatTok->is(tok::semi)) 4321 nextToken(/*LevelDifference=*/1); 4322 addUnwrappedLine(); 4323 } 4324 4325 return AddLevels; 4326 } 4327 4328 void UnwrappedLineParser::parseVerilogTable() { 4329 assert(FormatTok->is(Keywords.kw_table)); 4330 nextToken(/*LevelDifference=*/1); 4331 addUnwrappedLine(); 4332 4333 auto InitialLevel = Line->Level++; 4334 while (!eof() && !Keywords.isVerilogEnd(*FormatTok)) { 4335 FormatToken *Tok = FormatTok; 4336 nextToken(); 4337 if (Tok->is(tok::semi)) 4338 addUnwrappedLine(); 4339 else if (Tok->isOneOf(tok::star, tok::colon, tok::question, tok::minus)) 4340 Tok->setFinalizedType(TT_VerilogTableItem); 4341 } 4342 Line->Level = InitialLevel; 4343 nextToken(/*LevelDifference=*/-1); 4344 addUnwrappedLine(); 4345 } 4346 4347 void UnwrappedLineParser::parseVerilogCaseLabel() { 4348 // The label will get unindented in AnnotatingParser. If there are no leading 4349 // spaces, indent the rest here so that things inside the block will be 4350 // indented relative to things outside. We don't use parseLabel because we 4351 // don't know whether this colon is a label or a ternary expression at this 4352 // point. 4353 auto OrigLevel = Line->Level; 4354 auto FirstLine = CurrentLines->size(); 4355 if (Line->Level == 0 || (Line->InPPDirective && Line->Level <= 1)) 4356 ++Line->Level; 4357 else if (!Style.IndentCaseBlocks && Keywords.isVerilogBegin(*FormatTok)) 4358 --Line->Level; 4359 parseStructuralElement(); 4360 // Restore the indentation in both the new line and the line that has the 4361 // label. 4362 if (CurrentLines->size() > FirstLine) 4363 (*CurrentLines)[FirstLine].Level = OrigLevel; 4364 Line->Level = OrigLevel; 4365 } 4366 4367 bool UnwrappedLineParser::containsExpansion(const UnwrappedLine &Line) const { 4368 for (const auto &N : Line.Tokens) { 4369 if (N.Tok->MacroCtx) 4370 return true; 4371 for (const UnwrappedLine &Child : N.Children) 4372 if (containsExpansion(Child)) 4373 return true; 4374 } 4375 return false; 4376 } 4377 4378 void UnwrappedLineParser::addUnwrappedLine(LineLevel AdjustLevel) { 4379 if (Line->Tokens.empty()) 4380 return; 4381 LLVM_DEBUG({ 4382 if (!parsingPPDirective()) { 4383 llvm::dbgs() << "Adding unwrapped line:\n"; 4384 printDebugInfo(*Line); 4385 } 4386 }); 4387 4388 // If this line closes a block when in Whitesmiths mode, remember that 4389 // information so that the level can be decreased after the line is added. 4390 // This has to happen after the addition of the line since the line itself 4391 // needs to be indented. 4392 bool ClosesWhitesmithsBlock = 4393 Line->MatchingOpeningBlockLineIndex != UnwrappedLine::kInvalidIndex && 4394 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths; 4395 4396 // If the current line was expanded from a macro call, we use it to 4397 // reconstruct an unwrapped line from the structure of the expanded unwrapped 4398 // line and the unexpanded token stream. 4399 if (!parsingPPDirective() && !InExpansion && containsExpansion(*Line)) { 4400 if (!Reconstruct) 4401 Reconstruct.emplace(Line->Level, Unexpanded); 4402 Reconstruct->addLine(*Line); 4403 4404 // While the reconstructed unexpanded lines are stored in the normal 4405 // flow of lines, the expanded lines are stored on the side to be analyzed 4406 // in an extra step. 4407 CurrentExpandedLines.push_back(std::move(*Line)); 4408 4409 if (Reconstruct->finished()) { 4410 UnwrappedLine Reconstructed = std::move(*Reconstruct).takeResult(); 4411 assert(!Reconstructed.Tokens.empty() && 4412 "Reconstructed must at least contain the macro identifier."); 4413 assert(!parsingPPDirective()); 4414 LLVM_DEBUG({ 4415 llvm::dbgs() << "Adding unexpanded line:\n"; 4416 printDebugInfo(Reconstructed); 4417 }); 4418 ExpandedLines[Reconstructed.Tokens.begin()->Tok] = CurrentExpandedLines; 4419 Lines.push_back(std::move(Reconstructed)); 4420 CurrentExpandedLines.clear(); 4421 Reconstruct.reset(); 4422 } 4423 } else { 4424 // At the top level we only get here when no unexpansion is going on, or 4425 // when conditional formatting led to unfinished macro reconstructions. 4426 assert(!Reconstruct || (CurrentLines != &Lines) || PPStack.size() > 0); 4427 CurrentLines->push_back(std::move(*Line)); 4428 } 4429 Line->Tokens.clear(); 4430 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex; 4431 Line->FirstStartColumn = 0; 4432 Line->IsContinuation = false; 4433 Line->SeenDecltypeAuto = false; 4434 4435 if (ClosesWhitesmithsBlock && AdjustLevel == LineLevel::Remove) 4436 --Line->Level; 4437 if (!parsingPPDirective() && !PreprocessorDirectives.empty()) { 4438 CurrentLines->append( 4439 std::make_move_iterator(PreprocessorDirectives.begin()), 4440 std::make_move_iterator(PreprocessorDirectives.end())); 4441 PreprocessorDirectives.clear(); 4442 } 4443 // Disconnect the current token from the last token on the previous line. 4444 FormatTok->Previous = nullptr; 4445 } 4446 4447 bool UnwrappedLineParser::eof() const { return FormatTok->is(tok::eof); } 4448 4449 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 4450 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 4451 FormatTok.NewlinesBefore > 0; 4452 } 4453 4454 // Checks if \p FormatTok is a line comment that continues the line comment 4455 // section on \p Line. 4456 static bool 4457 continuesLineCommentSection(const FormatToken &FormatTok, 4458 const UnwrappedLine &Line, 4459 const llvm::Regex &CommentPragmasRegex) { 4460 if (Line.Tokens.empty()) 4461 return false; 4462 4463 StringRef IndentContent = FormatTok.TokenText; 4464 if (FormatTok.TokenText.starts_with("//") || 4465 FormatTok.TokenText.starts_with("/*")) { 4466 IndentContent = FormatTok.TokenText.substr(2); 4467 } 4468 if (CommentPragmasRegex.match(IndentContent)) 4469 return false; 4470 4471 // If Line starts with a line comment, then FormatTok continues the comment 4472 // section if its original column is greater or equal to the original start 4473 // column of the line. 4474 // 4475 // Define the min column token of a line as follows: if a line ends in '{' or 4476 // contains a '{' followed by a line comment, then the min column token is 4477 // that '{'. Otherwise, the min column token of the line is the first token of 4478 // the line. 4479 // 4480 // If Line starts with a token other than a line comment, then FormatTok 4481 // continues the comment section if its original column is greater than the 4482 // original start column of the min column token of the line. 4483 // 4484 // For example, the second line comment continues the first in these cases: 4485 // 4486 // // first line 4487 // // second line 4488 // 4489 // and: 4490 // 4491 // // first line 4492 // // second line 4493 // 4494 // and: 4495 // 4496 // int i; // first line 4497 // // second line 4498 // 4499 // and: 4500 // 4501 // do { // first line 4502 // // second line 4503 // int i; 4504 // } while (true); 4505 // 4506 // and: 4507 // 4508 // enum { 4509 // a, // first line 4510 // // second line 4511 // b 4512 // }; 4513 // 4514 // The second line comment doesn't continue the first in these cases: 4515 // 4516 // // first line 4517 // // second line 4518 // 4519 // and: 4520 // 4521 // int i; // first line 4522 // // second line 4523 // 4524 // and: 4525 // 4526 // do { // first line 4527 // // second line 4528 // int i; 4529 // } while (true); 4530 // 4531 // and: 4532 // 4533 // enum { 4534 // a, // first line 4535 // // second line 4536 // }; 4537 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 4538 4539 // Scan for '{//'. If found, use the column of '{' as a min column for line 4540 // comment section continuation. 4541 const FormatToken *PreviousToken = nullptr; 4542 for (const UnwrappedLineNode &Node : Line.Tokens) { 4543 if (PreviousToken && PreviousToken->is(tok::l_brace) && 4544 isLineComment(*Node.Tok)) { 4545 MinColumnToken = PreviousToken; 4546 break; 4547 } 4548 PreviousToken = Node.Tok; 4549 4550 // Grab the last newline preceding a token in this unwrapped line. 4551 if (Node.Tok->NewlinesBefore > 0) 4552 MinColumnToken = Node.Tok; 4553 } 4554 if (PreviousToken && PreviousToken->is(tok::l_brace)) 4555 MinColumnToken = PreviousToken; 4556 4557 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok, 4558 MinColumnToken); 4559 } 4560 4561 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 4562 bool JustComments = Line->Tokens.empty(); 4563 for (FormatToken *Tok : CommentsBeforeNextToken) { 4564 // Line comments that belong to the same line comment section are put on the 4565 // same line since later we might want to reflow content between them. 4566 // Additional fine-grained breaking of line comment sections is controlled 4567 // by the class BreakableLineCommentSection in case it is desirable to keep 4568 // several line comment sections in the same unwrapped line. 4569 // 4570 // FIXME: Consider putting separate line comment sections as children to the 4571 // unwrapped line instead. 4572 Tok->ContinuesLineCommentSection = 4573 continuesLineCommentSection(*Tok, *Line, CommentPragmasRegex); 4574 if (isOnNewLine(*Tok) && JustComments && !Tok->ContinuesLineCommentSection) 4575 addUnwrappedLine(); 4576 pushToken(Tok); 4577 } 4578 if (NewlineBeforeNext && JustComments) 4579 addUnwrappedLine(); 4580 CommentsBeforeNextToken.clear(); 4581 } 4582 4583 void UnwrappedLineParser::nextToken(int LevelDifference) { 4584 if (eof()) 4585 return; 4586 flushComments(isOnNewLine(*FormatTok)); 4587 pushToken(FormatTok); 4588 FormatToken *Previous = FormatTok; 4589 if (!Style.isJavaScript()) 4590 readToken(LevelDifference); 4591 else 4592 readTokenWithJavaScriptASI(); 4593 FormatTok->Previous = Previous; 4594 if (Style.isVerilog()) { 4595 // Blocks in Verilog can have `begin` and `end` instead of braces. For 4596 // keywords like `begin`, we can't treat them the same as left braces 4597 // because some contexts require one of them. For example structs use 4598 // braces and if blocks use keywords, and a left brace can occur in an if 4599 // statement, but it is not a block. For keywords like `end`, we simply 4600 // treat them the same as right braces. 4601 if (Keywords.isVerilogEnd(*FormatTok)) 4602 FormatTok->Tok.setKind(tok::r_brace); 4603 } 4604 } 4605 4606 void UnwrappedLineParser::distributeComments( 4607 const SmallVectorImpl<FormatToken *> &Comments, 4608 const FormatToken *NextTok) { 4609 // Whether or not a line comment token continues a line is controlled by 4610 // the method continuesLineCommentSection, with the following caveat: 4611 // 4612 // Define a trail of Comments to be a nonempty proper postfix of Comments such 4613 // that each comment line from the trail is aligned with the next token, if 4614 // the next token exists. If a trail exists, the beginning of the maximal 4615 // trail is marked as a start of a new comment section. 4616 // 4617 // For example in this code: 4618 // 4619 // int a; // line about a 4620 // // line 1 about b 4621 // // line 2 about b 4622 // int b; 4623 // 4624 // the two lines about b form a maximal trail, so there are two sections, the 4625 // first one consisting of the single comment "// line about a" and the 4626 // second one consisting of the next two comments. 4627 if (Comments.empty()) 4628 return; 4629 bool ShouldPushCommentsInCurrentLine = true; 4630 bool HasTrailAlignedWithNextToken = false; 4631 unsigned StartOfTrailAlignedWithNextToken = 0; 4632 if (NextTok) { 4633 // We are skipping the first element intentionally. 4634 for (unsigned i = Comments.size() - 1; i > 0; --i) { 4635 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) { 4636 HasTrailAlignedWithNextToken = true; 4637 StartOfTrailAlignedWithNextToken = i; 4638 } 4639 } 4640 } 4641 for (unsigned i = 0, e = Comments.size(); i < e; ++i) { 4642 FormatToken *FormatTok = Comments[i]; 4643 if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) { 4644 FormatTok->ContinuesLineCommentSection = false; 4645 } else { 4646 FormatTok->ContinuesLineCommentSection = 4647 continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex); 4648 } 4649 if (!FormatTok->ContinuesLineCommentSection && 4650 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) { 4651 ShouldPushCommentsInCurrentLine = false; 4652 } 4653 if (ShouldPushCommentsInCurrentLine) 4654 pushToken(FormatTok); 4655 else 4656 CommentsBeforeNextToken.push_back(FormatTok); 4657 } 4658 } 4659 4660 void UnwrappedLineParser::readToken(int LevelDifference) { 4661 SmallVector<FormatToken *, 1> Comments; 4662 bool PreviousWasComment = false; 4663 bool FirstNonCommentOnLine = false; 4664 do { 4665 FormatTok = Tokens->getNextToken(); 4666 assert(FormatTok); 4667 while (FormatTok->getType() == TT_ConflictStart || 4668 FormatTok->getType() == TT_ConflictEnd || 4669 FormatTok->getType() == TT_ConflictAlternative) { 4670 if (FormatTok->getType() == TT_ConflictStart) 4671 conditionalCompilationStart(/*Unreachable=*/false); 4672 else if (FormatTok->getType() == TT_ConflictAlternative) 4673 conditionalCompilationAlternative(); 4674 else if (FormatTok->getType() == TT_ConflictEnd) 4675 conditionalCompilationEnd(); 4676 FormatTok = Tokens->getNextToken(); 4677 FormatTok->MustBreakBefore = true; 4678 } 4679 4680 auto IsFirstNonCommentOnLine = [](bool FirstNonCommentOnLine, 4681 const FormatToken &Tok, 4682 bool PreviousWasComment) { 4683 auto IsFirstOnLine = [](const FormatToken &Tok) { 4684 return Tok.HasUnescapedNewline || Tok.IsFirst; 4685 }; 4686 4687 // Consider preprocessor directives preceded by block comments as first 4688 // on line. 4689 if (PreviousWasComment) 4690 return FirstNonCommentOnLine || IsFirstOnLine(Tok); 4691 return IsFirstOnLine(Tok); 4692 }; 4693 4694 FirstNonCommentOnLine = IsFirstNonCommentOnLine( 4695 FirstNonCommentOnLine, *FormatTok, PreviousWasComment); 4696 PreviousWasComment = FormatTok->is(tok::comment); 4697 4698 while (!Line->InPPDirective && FormatTok->is(tok::hash) && 4699 (!Style.isVerilog() || 4700 Keywords.isVerilogPPDirective(*Tokens->peekNextToken())) && 4701 FirstNonCommentOnLine) { 4702 distributeComments(Comments, FormatTok); 4703 Comments.clear(); 4704 // If there is an unfinished unwrapped line, we flush the preprocessor 4705 // directives only after that unwrapped line was finished later. 4706 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 4707 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 4708 assert((LevelDifference >= 0 || 4709 static_cast<unsigned>(-LevelDifference) <= Line->Level) && 4710 "LevelDifference makes Line->Level negative"); 4711 Line->Level += LevelDifference; 4712 // Comments stored before the preprocessor directive need to be output 4713 // before the preprocessor directive, at the same level as the 4714 // preprocessor directive, as we consider them to apply to the directive. 4715 if (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash && 4716 PPBranchLevel > 0) { 4717 Line->Level += PPBranchLevel; 4718 } 4719 flushComments(isOnNewLine(*FormatTok)); 4720 parsePPDirective(); 4721 PreviousWasComment = FormatTok->is(tok::comment); 4722 FirstNonCommentOnLine = IsFirstNonCommentOnLine( 4723 FirstNonCommentOnLine, *FormatTok, PreviousWasComment); 4724 } 4725 4726 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) && 4727 !Line->InPPDirective) { 4728 continue; 4729 } 4730 4731 if (FormatTok->is(tok::identifier) && 4732 Macros.defined(FormatTok->TokenText) && 4733 // FIXME: Allow expanding macros in preprocessor directives. 4734 !Line->InPPDirective) { 4735 FormatToken *ID = FormatTok; 4736 unsigned Position = Tokens->getPosition(); 4737 4738 // To correctly parse the code, we need to replace the tokens of the macro 4739 // call with its expansion. 4740 auto PreCall = std::move(Line); 4741 Line.reset(new UnwrappedLine); 4742 bool OldInExpansion = InExpansion; 4743 InExpansion = true; 4744 // We parse the macro call into a new line. 4745 auto Args = parseMacroCall(); 4746 InExpansion = OldInExpansion; 4747 assert(Line->Tokens.front().Tok == ID); 4748 // And remember the unexpanded macro call tokens. 4749 auto UnexpandedLine = std::move(Line); 4750 // Reset to the old line. 4751 Line = std::move(PreCall); 4752 4753 LLVM_DEBUG({ 4754 llvm::dbgs() << "Macro call: " << ID->TokenText << "("; 4755 if (Args) { 4756 llvm::dbgs() << "("; 4757 for (const auto &Arg : Args.value()) 4758 for (const auto &T : Arg) 4759 llvm::dbgs() << T->TokenText << " "; 4760 llvm::dbgs() << ")"; 4761 } 4762 llvm::dbgs() << "\n"; 4763 }); 4764 if (Macros.objectLike(ID->TokenText) && Args && 4765 !Macros.hasArity(ID->TokenText, Args->size())) { 4766 // The macro is either 4767 // - object-like, but we got argumnets, or 4768 // - overloaded to be both object-like and function-like, but none of 4769 // the function-like arities match the number of arguments. 4770 // Thus, expand as object-like macro. 4771 LLVM_DEBUG(llvm::dbgs() 4772 << "Macro \"" << ID->TokenText 4773 << "\" not overloaded for arity " << Args->size() 4774 << "or not function-like, using object-like overload."); 4775 Args.reset(); 4776 UnexpandedLine->Tokens.resize(1); 4777 Tokens->setPosition(Position); 4778 nextToken(); 4779 assert(!Args && Macros.objectLike(ID->TokenText)); 4780 } 4781 if ((!Args && Macros.objectLike(ID->TokenText)) || 4782 (Args && Macros.hasArity(ID->TokenText, Args->size()))) { 4783 // Next, we insert the expanded tokens in the token stream at the 4784 // current position, and continue parsing. 4785 Unexpanded[ID] = std::move(UnexpandedLine); 4786 SmallVector<FormatToken *, 8> Expansion = 4787 Macros.expand(ID, std::move(Args)); 4788 if (!Expansion.empty()) 4789 FormatTok = Tokens->insertTokens(Expansion); 4790 4791 LLVM_DEBUG({ 4792 llvm::dbgs() << "Expanded: "; 4793 for (const auto &T : Expansion) 4794 llvm::dbgs() << T->TokenText << " "; 4795 llvm::dbgs() << "\n"; 4796 }); 4797 } else { 4798 LLVM_DEBUG({ 4799 llvm::dbgs() << "Did not expand macro \"" << ID->TokenText 4800 << "\", because it was used "; 4801 if (Args) 4802 llvm::dbgs() << "with " << Args->size(); 4803 else 4804 llvm::dbgs() << "without"; 4805 llvm::dbgs() << " arguments, which doesn't match any definition.\n"; 4806 }); 4807 Tokens->setPosition(Position); 4808 FormatTok = ID; 4809 } 4810 } 4811 4812 if (FormatTok->isNot(tok::comment)) { 4813 distributeComments(Comments, FormatTok); 4814 Comments.clear(); 4815 return; 4816 } 4817 4818 Comments.push_back(FormatTok); 4819 } while (!eof()); 4820 4821 distributeComments(Comments, nullptr); 4822 Comments.clear(); 4823 } 4824 4825 namespace { 4826 template <typename Iterator> 4827 void pushTokens(Iterator Begin, Iterator End, 4828 llvm::SmallVectorImpl<FormatToken *> &Into) { 4829 for (auto I = Begin; I != End; ++I) { 4830 Into.push_back(I->Tok); 4831 for (const auto &Child : I->Children) 4832 pushTokens(Child.Tokens.begin(), Child.Tokens.end(), Into); 4833 } 4834 } 4835 } // namespace 4836 4837 std::optional<llvm::SmallVector<llvm::SmallVector<FormatToken *, 8>, 1>> 4838 UnwrappedLineParser::parseMacroCall() { 4839 std::optional<llvm::SmallVector<llvm::SmallVector<FormatToken *, 8>, 1>> Args; 4840 assert(Line->Tokens.empty()); 4841 nextToken(); 4842 if (FormatTok->isNot(tok::l_paren)) 4843 return Args; 4844 unsigned Position = Tokens->getPosition(); 4845 FormatToken *Tok = FormatTok; 4846 nextToken(); 4847 Args.emplace(); 4848 auto ArgStart = std::prev(Line->Tokens.end()); 4849 4850 int Parens = 0; 4851 do { 4852 switch (FormatTok->Tok.getKind()) { 4853 case tok::l_paren: 4854 ++Parens; 4855 nextToken(); 4856 break; 4857 case tok::r_paren: { 4858 if (Parens > 0) { 4859 --Parens; 4860 nextToken(); 4861 break; 4862 } 4863 Args->push_back({}); 4864 pushTokens(std::next(ArgStart), Line->Tokens.end(), Args->back()); 4865 nextToken(); 4866 return Args; 4867 } 4868 case tok::comma: { 4869 if (Parens > 0) { 4870 nextToken(); 4871 break; 4872 } 4873 Args->push_back({}); 4874 pushTokens(std::next(ArgStart), Line->Tokens.end(), Args->back()); 4875 nextToken(); 4876 ArgStart = std::prev(Line->Tokens.end()); 4877 break; 4878 } 4879 default: 4880 nextToken(); 4881 break; 4882 } 4883 } while (!eof()); 4884 Line->Tokens.resize(1); 4885 Tokens->setPosition(Position); 4886 FormatTok = Tok; 4887 return {}; 4888 } 4889 4890 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 4891 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 4892 if (MustBreakBeforeNextToken) { 4893 Line->Tokens.back().Tok->MustBreakBefore = true; 4894 MustBreakBeforeNextToken = false; 4895 } 4896 } 4897 4898 } // end namespace format 4899 } // end namespace clang 4900