1 //===--- TokenAnnotator.cpp - Format C++ code -----------------------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 /// 10 /// \file 11 /// \brief This file implements a token annotator, i.e. creates 12 /// \c AnnotatedTokens out of \c FormatTokens with required extra information. 13 /// 14 //===----------------------------------------------------------------------===// 15 16 #include "TokenAnnotator.h" 17 #include "clang/Basic/SourceManager.h" 18 #include "llvm/Support/Debug.h" 19 20 #define DEBUG_TYPE "format-token-annotator" 21 22 namespace clang { 23 namespace format { 24 25 namespace { 26 27 /// \brief A parser that gathers additional information about tokens. 28 /// 29 /// The \c TokenAnnotator tries to match parenthesis and square brakets and 30 /// store a parenthesis levels. It also tries to resolve matching "<" and ">" 31 /// into template parameter lists. 32 class AnnotatingParser { 33 public: 34 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line, 35 const AdditionalKeywords &Keywords) 36 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false), 37 Keywords(Keywords) { 38 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false)); 39 resetTokenMetadata(CurrentToken); 40 } 41 42 private: 43 bool parseAngle() { 44 if (!CurrentToken) 45 return false; 46 ScopedContextCreator ContextCreator(*this, tok::less, 10); 47 FormatToken *Left = CurrentToken->Previous; 48 Contexts.back().IsExpression = false; 49 // If there's a template keyword before the opening angle bracket, this is a 50 // template parameter, not an argument. 51 Contexts.back().InTemplateArgument = 52 Left->Previous && Left->Previous->Tok.isNot(tok::kw_template); 53 54 if (Style.Language == FormatStyle::LK_Java && 55 CurrentToken->is(tok::question)) 56 next(); 57 58 while (CurrentToken) { 59 if (CurrentToken->is(tok::greater)) { 60 Left->MatchingParen = CurrentToken; 61 CurrentToken->MatchingParen = Left; 62 CurrentToken->Type = TT_TemplateCloser; 63 next(); 64 return true; 65 } 66 if (CurrentToken->is(tok::question) && 67 Style.Language == FormatStyle::LK_Java) { 68 next(); 69 continue; 70 } 71 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace, 72 tok::colon, tok::question)) 73 return false; 74 // If a && or || is found and interpreted as a binary operator, this set 75 // of angles is likely part of something like "a < b && c > d". If the 76 // angles are inside an expression, the ||/&& might also be a binary 77 // operator that was misinterpreted because we are parsing template 78 // parameters. 79 // FIXME: This is getting out of hand, write a decent parser. 80 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) && 81 CurrentToken->Previous->is(TT_BinaryOperator) && 82 Contexts[Contexts.size() - 2].IsExpression && 83 Line.First->isNot(tok::kw_template)) 84 return false; 85 updateParameterCount(Left, CurrentToken); 86 if (!consumeToken()) 87 return false; 88 } 89 return false; 90 } 91 92 bool parseParens(bool LookForDecls = false) { 93 if (!CurrentToken) 94 return false; 95 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1); 96 97 // FIXME: This is a bit of a hack. Do better. 98 Contexts.back().ColonIsForRangeExpr = 99 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr; 100 101 bool StartsObjCMethodExpr = false; 102 FormatToken *Left = CurrentToken->Previous; 103 if (CurrentToken->is(tok::caret)) { 104 // (^ can start a block type. 105 Left->Type = TT_ObjCBlockLParen; 106 } else if (FormatToken *MaybeSel = Left->Previous) { 107 // @selector( starts a selector. 108 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous && 109 MaybeSel->Previous->is(tok::at)) { 110 StartsObjCMethodExpr = true; 111 } 112 } 113 114 if (Left->Previous && 115 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_if, 116 tok::kw_while, tok::l_paren, tok::comma) || 117 Left->Previous->is(TT_BinaryOperator))) { 118 // static_assert, if and while usually contain expressions. 119 Contexts.back().IsExpression = true; 120 } else if (Line.InPPDirective && 121 (!Left->Previous || 122 !Left->Previous->isOneOf(tok::identifier, 123 TT_OverloadedOperator))) { 124 Contexts.back().IsExpression = true; 125 } else if (Left->Previous && Left->Previous->is(tok::r_square) && 126 Left->Previous->MatchingParen && 127 Left->Previous->MatchingParen->is(TT_LambdaLSquare)) { 128 // This is a parameter list of a lambda expression. 129 Contexts.back().IsExpression = false; 130 } else if (Contexts[Contexts.size() - 2].CaretFound) { 131 // This is the parameter list of an ObjC block. 132 Contexts.back().IsExpression = false; 133 } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) { 134 Left->Type = TT_AttributeParen; 135 } else if (Left->Previous && Left->Previous->IsForEachMacro) { 136 // The first argument to a foreach macro is a declaration. 137 Contexts.back().IsForEachMacro = true; 138 Contexts.back().IsExpression = false; 139 } else if (Left->Previous && Left->Previous->MatchingParen && 140 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) { 141 Contexts.back().IsExpression = false; 142 } 143 144 if (StartsObjCMethodExpr) { 145 Contexts.back().ColonIsObjCMethodExpr = true; 146 Left->Type = TT_ObjCMethodExpr; 147 } 148 149 bool MightBeFunctionType = CurrentToken->is(tok::star); 150 bool HasMultipleLines = false; 151 bool HasMultipleParametersOnALine = false; 152 while (CurrentToken) { 153 // LookForDecls is set when "if (" has been seen. Check for 154 // 'identifier' '*' 'identifier' followed by not '=' -- this 155 // '*' has to be a binary operator but determineStarAmpUsage() will 156 // categorize it as an unary operator, so set the right type here. 157 if (LookForDecls && CurrentToken->Next) { 158 FormatToken *Prev = CurrentToken->getPreviousNonComment(); 159 if (Prev) { 160 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 161 FormatToken *Next = CurrentToken->Next; 162 if (PrevPrev && PrevPrev->is(tok::identifier) && 163 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) && 164 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) { 165 Prev->Type = TT_BinaryOperator; 166 LookForDecls = false; 167 } 168 } 169 } 170 171 if (CurrentToken->Previous->is(TT_PointerOrReference) && 172 CurrentToken->Previous->Previous->isOneOf(tok::l_paren, 173 tok::coloncolon)) 174 MightBeFunctionType = true; 175 if (CurrentToken->Previous->is(TT_BinaryOperator)) 176 Contexts.back().IsExpression = true; 177 if (CurrentToken->is(tok::r_paren)) { 178 if (MightBeFunctionType && CurrentToken->Next && 179 (CurrentToken->Next->is(tok::l_paren) || 180 (CurrentToken->Next->is(tok::l_square) && 181 !Contexts.back().IsExpression))) 182 Left->Type = TT_FunctionTypeLParen; 183 Left->MatchingParen = CurrentToken; 184 CurrentToken->MatchingParen = Left; 185 186 if (StartsObjCMethodExpr) { 187 CurrentToken->Type = TT_ObjCMethodExpr; 188 if (Contexts.back().FirstObjCSelectorName) { 189 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 190 Contexts.back().LongestObjCSelectorName; 191 } 192 } 193 194 if (Left->is(TT_AttributeParen)) 195 CurrentToken->Type = TT_AttributeParen; 196 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation)) 197 CurrentToken->Type = TT_JavaAnnotation; 198 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation)) 199 CurrentToken->Type = TT_LeadingJavaAnnotation; 200 201 if (!HasMultipleLines) 202 Left->PackingKind = PPK_Inconclusive; 203 else if (HasMultipleParametersOnALine) 204 Left->PackingKind = PPK_BinPacked; 205 else 206 Left->PackingKind = PPK_OnePerLine; 207 208 next(); 209 return true; 210 } 211 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace)) 212 return false; 213 else if (CurrentToken->is(tok::l_brace)) 214 Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen 215 if (CurrentToken->is(tok::comma) && CurrentToken->Next && 216 !CurrentToken->Next->HasUnescapedNewline && 217 !CurrentToken->Next->isTrailingComment()) 218 HasMultipleParametersOnALine = true; 219 if (CurrentToken->isOneOf(tok::kw_const, tok::kw_auto) || 220 CurrentToken->isSimpleTypeSpecifier()) 221 Contexts.back().IsExpression = false; 222 FormatToken *Tok = CurrentToken; 223 if (!consumeToken()) 224 return false; 225 updateParameterCount(Left, Tok); 226 if (CurrentToken && CurrentToken->HasUnescapedNewline) 227 HasMultipleLines = true; 228 } 229 return false; 230 } 231 232 bool parseSquare() { 233 if (!CurrentToken) 234 return false; 235 236 // A '[' could be an index subscript (after an identifier or after 237 // ')' or ']'), it could be the start of an Objective-C method 238 // expression, or it could the the start of an Objective-C array literal. 239 FormatToken *Left = CurrentToken->Previous; 240 FormatToken *Parent = Left->getPreviousNonComment(); 241 bool StartsObjCMethodExpr = 242 Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) && 243 CurrentToken->isNot(tok::l_brace) && 244 (!Parent || 245 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren, 246 tok::kw_return, tok::kw_throw) || 247 Parent->isUnaryOperator() || 248 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) || 249 getBinOpPrecedence(Parent->Tok.getKind(), true, true) > prec::Unknown); 250 ScopedContextCreator ContextCreator(*this, tok::l_square, 10); 251 Contexts.back().IsExpression = true; 252 bool ColonFound = false; 253 254 if (StartsObjCMethodExpr) { 255 Contexts.back().ColonIsObjCMethodExpr = true; 256 Left->Type = TT_ObjCMethodExpr; 257 } else if (Parent && Parent->is(tok::at)) { 258 Left->Type = TT_ArrayInitializerLSquare; 259 } else if (Left->is(TT_Unknown)) { 260 Left->Type = TT_ArraySubscriptLSquare; 261 } 262 263 while (CurrentToken) { 264 if (CurrentToken->is(tok::r_square)) { 265 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_paren) && 266 Left->is(TT_ObjCMethodExpr)) { 267 // An ObjC method call is rarely followed by an open parenthesis. 268 // FIXME: Do we incorrectly label ":" with this? 269 StartsObjCMethodExpr = false; 270 Left->Type = TT_Unknown; 271 } 272 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) { 273 CurrentToken->Type = TT_ObjCMethodExpr; 274 // determineStarAmpUsage() thinks that '*' '[' is allocating an 275 // array of pointers, but if '[' starts a selector then '*' is a 276 // binary operator. 277 if (Parent && Parent->is(TT_PointerOrReference)) 278 Parent->Type = TT_BinaryOperator; 279 } 280 Left->MatchingParen = CurrentToken; 281 CurrentToken->MatchingParen = Left; 282 if (Contexts.back().FirstObjCSelectorName) { 283 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 284 Contexts.back().LongestObjCSelectorName; 285 if (Left->BlockParameterCount > 1) 286 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0; 287 } 288 next(); 289 return true; 290 } 291 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace)) 292 return false; 293 if (CurrentToken->is(tok::colon)) { 294 if (Left->is(TT_ArraySubscriptLSquare)) { 295 Left->Type = TT_ObjCMethodExpr; 296 StartsObjCMethodExpr = true; 297 Contexts.back().ColonIsObjCMethodExpr = true; 298 if (Parent && Parent->is(tok::r_paren)) 299 Parent->Type = TT_CastRParen; 300 } 301 ColonFound = true; 302 } 303 if (CurrentToken->is(tok::comma) && 304 Style.Language != FormatStyle::LK_Proto && 305 (Left->is(TT_ArraySubscriptLSquare) || 306 (Left->is(TT_ObjCMethodExpr) && !ColonFound))) 307 Left->Type = TT_ArrayInitializerLSquare; 308 FormatToken *Tok = CurrentToken; 309 if (!consumeToken()) 310 return false; 311 updateParameterCount(Left, Tok); 312 } 313 return false; 314 } 315 316 bool parseBrace() { 317 if (CurrentToken) { 318 FormatToken *Left = CurrentToken->Previous; 319 320 if (Contexts.back().CaretFound) 321 Left->Type = TT_ObjCBlockLBrace; 322 Contexts.back().CaretFound = false; 323 324 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1); 325 Contexts.back().ColonIsDictLiteral = true; 326 if (Left->BlockKind == BK_BracedInit) 327 Contexts.back().IsExpression = true; 328 329 while (CurrentToken) { 330 if (CurrentToken->is(tok::r_brace)) { 331 Left->MatchingParen = CurrentToken; 332 CurrentToken->MatchingParen = Left; 333 next(); 334 return true; 335 } 336 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 337 return false; 338 updateParameterCount(Left, CurrentToken); 339 if (CurrentToken->isOneOf(tok::colon, tok::l_brace)) { 340 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 341 if ((CurrentToken->is(tok::colon) || 342 Style.Language == FormatStyle::LK_Proto) && 343 Previous->is(tok::identifier)) 344 Previous->Type = TT_SelectorName; 345 if (CurrentToken->is(tok::colon)) 346 Left->Type = TT_DictLiteral; 347 } 348 if (!consumeToken()) 349 return false; 350 } 351 } 352 return true; 353 } 354 355 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 356 if (Current->is(TT_LambdaLSquare) || 357 (Current->is(tok::caret) && Current->is(TT_UnaryOperator)) || 358 (Style.Language == FormatStyle::LK_JavaScript && 359 Current->is(Keywords.kw_function))) { 360 ++Left->BlockParameterCount; 361 } 362 if (Current->is(tok::comma)) { 363 ++Left->ParameterCount; 364 if (!Left->Role) 365 Left->Role.reset(new CommaSeparatedList(Style)); 366 Left->Role->CommaFound(Current); 367 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 368 Left->ParameterCount = 1; 369 } 370 } 371 372 bool parseConditional() { 373 while (CurrentToken) { 374 if (CurrentToken->is(tok::colon)) { 375 CurrentToken->Type = TT_ConditionalExpr; 376 next(); 377 return true; 378 } 379 if (!consumeToken()) 380 return false; 381 } 382 return false; 383 } 384 385 bool parseTemplateDeclaration() { 386 if (CurrentToken && CurrentToken->is(tok::less)) { 387 CurrentToken->Type = TT_TemplateOpener; 388 next(); 389 if (!parseAngle()) 390 return false; 391 if (CurrentToken) 392 CurrentToken->Previous->ClosesTemplateDeclaration = true; 393 return true; 394 } 395 return false; 396 } 397 398 bool consumeToken() { 399 FormatToken *Tok = CurrentToken; 400 next(); 401 switch (Tok->Tok.getKind()) { 402 case tok::plus: 403 case tok::minus: 404 if (!Tok->Previous && Line.MustBeDeclaration) 405 Tok->Type = TT_ObjCMethodSpecifier; 406 break; 407 case tok::colon: 408 if (!Tok->Previous) 409 return false; 410 // Colons from ?: are handled in parseConditional(). 411 if (Tok->Previous->is(tok::r_paren) && Contexts.size() == 1 && 412 Line.First->isNot(tok::kw_case)) { 413 Tok->Type = TT_CtorInitializerColon; 414 } else if (Contexts.back().ColonIsDictLiteral) { 415 Tok->Type = TT_DictLiteral; 416 } else if (Contexts.back().ColonIsObjCMethodExpr || 417 Line.First->is(TT_ObjCMethodSpecifier)) { 418 Tok->Type = TT_ObjCMethodExpr; 419 Tok->Previous->Type = TT_SelectorName; 420 if (Tok->Previous->ColumnWidth > 421 Contexts.back().LongestObjCSelectorName) { 422 Contexts.back().LongestObjCSelectorName = Tok->Previous->ColumnWidth; 423 } 424 if (!Contexts.back().FirstObjCSelectorName) 425 Contexts.back().FirstObjCSelectorName = Tok->Previous; 426 } else if (Contexts.back().ColonIsForRangeExpr) { 427 Tok->Type = TT_RangeBasedForLoopColon; 428 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 429 Tok->Type = TT_BitFieldColon; 430 } else if (Contexts.size() == 1 && 431 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) { 432 Tok->Type = TT_InheritanceColon; 433 } else if (Tok->Previous->is(tok::identifier) && Tok->Next && 434 Tok->Next->isOneOf(tok::r_paren, tok::comma)) { 435 // This handles a special macro in ObjC code where selectors including 436 // the colon are passed as macro arguments. 437 Tok->Type = TT_ObjCMethodExpr; 438 } else if (Contexts.back().ContextKind == tok::l_paren) { 439 Tok->Type = TT_InlineASMColon; 440 } 441 break; 442 case tok::kw_if: 443 case tok::kw_while: 444 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 445 next(); 446 if (!parseParens(/*LookForDecls=*/true)) 447 return false; 448 } 449 break; 450 case tok::kw_for: 451 Contexts.back().ColonIsForRangeExpr = true; 452 next(); 453 if (!parseParens()) 454 return false; 455 break; 456 case tok::l_paren: 457 if (!parseParens()) 458 return false; 459 if (Line.MustBeDeclaration && Contexts.size() == 1 && 460 !Contexts.back().IsExpression && Line.First->isNot(TT_ObjCProperty) && 461 (!Tok->Previous || 462 !Tok->Previous->isOneOf(tok::kw_decltype, TT_LeadingJavaAnnotation))) 463 Line.MightBeFunctionDecl = true; 464 break; 465 case tok::l_square: 466 if (!parseSquare()) 467 return false; 468 break; 469 case tok::l_brace: 470 if (!parseBrace()) 471 return false; 472 break; 473 case tok::less: 474 if ((!Tok->Previous || 475 (!Tok->Previous->Tok.isLiteral() && 476 !(Tok->Previous->is(tok::r_paren) && Contexts.size() > 1))) && 477 parseAngle()) { 478 Tok->Type = TT_TemplateOpener; 479 } else { 480 Tok->Type = TT_BinaryOperator; 481 CurrentToken = Tok; 482 next(); 483 } 484 break; 485 case tok::r_paren: 486 case tok::r_square: 487 return false; 488 case tok::r_brace: 489 // Lines can start with '}'. 490 if (Tok->Previous) 491 return false; 492 break; 493 case tok::greater: 494 Tok->Type = TT_BinaryOperator; 495 break; 496 case tok::kw_operator: 497 while (CurrentToken && 498 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 499 if (CurrentToken->isOneOf(tok::star, tok::amp)) 500 CurrentToken->Type = TT_PointerOrReference; 501 consumeToken(); 502 if (CurrentToken && CurrentToken->Previous->is(TT_BinaryOperator)) 503 CurrentToken->Previous->Type = TT_OverloadedOperator; 504 } 505 if (CurrentToken) { 506 CurrentToken->Type = TT_OverloadedOperatorLParen; 507 if (CurrentToken->Previous->is(TT_BinaryOperator)) 508 CurrentToken->Previous->Type = TT_OverloadedOperator; 509 } 510 break; 511 case tok::question: 512 parseConditional(); 513 break; 514 case tok::kw_template: 515 parseTemplateDeclaration(); 516 break; 517 case tok::identifier: 518 if (Line.First->is(tok::kw_for) && Tok->is(Keywords.kw_in) && 519 Tok->Previous->isNot(tok::colon)) 520 Tok->Type = TT_ObjCForIn; 521 break; 522 case tok::comma: 523 if (Contexts.back().FirstStartOfName) 524 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 525 if (Contexts.back().InCtorInitializer) 526 Tok->Type = TT_CtorInitializerComma; 527 if (Contexts.back().IsForEachMacro) 528 Contexts.back().IsExpression = true; 529 break; 530 default: 531 break; 532 } 533 return true; 534 } 535 536 void parseIncludeDirective() { 537 if (CurrentToken && CurrentToken->is(tok::less)) { 538 next(); 539 while (CurrentToken) { 540 if (CurrentToken->isNot(tok::comment) || CurrentToken->Next) 541 CurrentToken->Type = TT_ImplicitStringLiteral; 542 next(); 543 } 544 } 545 } 546 547 void parseWarningOrError() { 548 next(); 549 // We still want to format the whitespace left of the first token of the 550 // warning or error. 551 next(); 552 while (CurrentToken) { 553 CurrentToken->Type = TT_ImplicitStringLiteral; 554 next(); 555 } 556 } 557 558 void parsePragma() { 559 next(); // Consume "pragma". 560 if (CurrentToken && CurrentToken->TokenText == "mark") { 561 next(); // Consume "mark". 562 next(); // Consume first token (so we fix leading whitespace). 563 while (CurrentToken) { 564 CurrentToken->Type = TT_ImplicitStringLiteral; 565 next(); 566 } 567 } 568 } 569 570 LineType parsePreprocessorDirective() { 571 LineType Type = LT_PreprocessorDirective; 572 next(); 573 if (!CurrentToken) 574 return Type; 575 if (CurrentToken->Tok.is(tok::numeric_constant)) { 576 CurrentToken->SpacesRequiredBefore = 1; 577 return Type; 578 } 579 // Hashes in the middle of a line can lead to any strange token 580 // sequence. 581 if (!CurrentToken->Tok.getIdentifierInfo()) 582 return Type; 583 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 584 case tok::pp_include: 585 case tok::pp_import: 586 next(); 587 parseIncludeDirective(); 588 Type = LT_ImportStatement; 589 break; 590 case tok::pp_error: 591 case tok::pp_warning: 592 parseWarningOrError(); 593 break; 594 case tok::pp_pragma: 595 parsePragma(); 596 break; 597 case tok::pp_if: 598 case tok::pp_elif: 599 Contexts.back().IsExpression = true; 600 parseLine(); 601 break; 602 default: 603 break; 604 } 605 while (CurrentToken) 606 next(); 607 return Type; 608 } 609 610 public: 611 LineType parseLine() { 612 if (CurrentToken->is(tok::hash)) { 613 return parsePreprocessorDirective(); 614 } 615 616 // Directly allow to 'import <string-literal>' to support protocol buffer 617 // definitions (code.google.com/p/protobuf) or missing "#" (either way we 618 // should not break the line). 619 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 620 if ((Style.Language == FormatStyle::LK_Java && 621 CurrentToken->is(Keywords.kw_package)) || 622 (Info && Info->getPPKeywordID() == tok::pp_import && 623 CurrentToken->Next && 624 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 625 tok::kw_static))) { 626 next(); 627 parseIncludeDirective(); 628 return LT_ImportStatement; 629 } 630 631 // If this line starts and ends in '<' and '>', respectively, it is likely 632 // part of "#define <a/b.h>". 633 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 634 parseIncludeDirective(); 635 return LT_ImportStatement; 636 } 637 638 bool KeywordVirtualFound = false; 639 bool ImportStatement = false; 640 while (CurrentToken) { 641 if (CurrentToken->is(tok::kw_virtual)) 642 KeywordVirtualFound = true; 643 if (IsImportStatement(*CurrentToken)) 644 ImportStatement = true; 645 if (!consumeToken()) 646 return LT_Invalid; 647 } 648 if (KeywordVirtualFound) 649 return LT_VirtualFunctionDecl; 650 if (ImportStatement) 651 return LT_ImportStatement; 652 653 if (Line.First->is(TT_ObjCMethodSpecifier)) { 654 if (Contexts.back().FirstObjCSelectorName) 655 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 656 Contexts.back().LongestObjCSelectorName; 657 return LT_ObjCMethodDecl; 658 } 659 660 return LT_Other; 661 } 662 663 private: 664 bool IsImportStatement(const FormatToken &Tok) { 665 // FIXME: Closure-library specific stuff should not be hard-coded but be 666 // configurable. 667 return Style.Language == FormatStyle::LK_JavaScript && 668 Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 669 Tok.Next->Next && (Tok.Next->Next->TokenText == "module" || 670 Tok.Next->Next->TokenText == "require" || 671 Tok.Next->Next->TokenText == "provide") && 672 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 673 } 674 675 void resetTokenMetadata(FormatToken *Token) { 676 if (!Token) 677 return; 678 679 // Reset token type in case we have already looked at it and then 680 // recovered from an error (e.g. failure to find the matching >). 681 if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_FunctionLBrace, 682 TT_ImplicitStringLiteral, TT_RegexLiteral, 683 TT_TrailingReturnArrow)) 684 CurrentToken->Type = TT_Unknown; 685 CurrentToken->Role.reset(); 686 CurrentToken->FakeLParens.clear(); 687 CurrentToken->FakeRParens = 0; 688 } 689 690 void next() { 691 if (CurrentToken) { 692 CurrentToken->NestingLevel = Contexts.size() - 1; 693 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 694 modifyContext(*CurrentToken); 695 determineTokenType(*CurrentToken); 696 CurrentToken = CurrentToken->Next; 697 } 698 699 resetTokenMetadata(CurrentToken); 700 } 701 702 /// \brief A struct to hold information valid in a specific context, e.g. 703 /// a pair of parenthesis. 704 struct Context { 705 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 706 bool IsExpression) 707 : ContextKind(ContextKind), BindingStrength(BindingStrength), 708 LongestObjCSelectorName(0), ColonIsForRangeExpr(false), 709 ColonIsDictLiteral(false), ColonIsObjCMethodExpr(false), 710 FirstObjCSelectorName(nullptr), FirstStartOfName(nullptr), 711 IsExpression(IsExpression), CanBeExpression(true), 712 InTemplateArgument(false), InCtorInitializer(false), 713 CaretFound(false), IsForEachMacro(false) {} 714 715 tok::TokenKind ContextKind; 716 unsigned BindingStrength; 717 unsigned LongestObjCSelectorName; 718 bool ColonIsForRangeExpr; 719 bool ColonIsDictLiteral; 720 bool ColonIsObjCMethodExpr; 721 FormatToken *FirstObjCSelectorName; 722 FormatToken *FirstStartOfName; 723 bool IsExpression; 724 bool CanBeExpression; 725 bool InTemplateArgument; 726 bool InCtorInitializer; 727 bool CaretFound; 728 bool IsForEachMacro; 729 }; 730 731 /// \brief Puts a new \c Context onto the stack \c Contexts for the lifetime 732 /// of each instance. 733 struct ScopedContextCreator { 734 AnnotatingParser &P; 735 736 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 737 unsigned Increase) 738 : P(P) { 739 P.Contexts.push_back(Context(ContextKind, 740 P.Contexts.back().BindingStrength + Increase, 741 P.Contexts.back().IsExpression)); 742 } 743 744 ~ScopedContextCreator() { P.Contexts.pop_back(); } 745 }; 746 747 void modifyContext(const FormatToken &Current) { 748 if (Current.getPrecedence() == prec::Assignment && 749 !Line.First->isOneOf(tok::kw_template, tok::kw_using, 750 TT_UnaryOperator) && 751 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 752 Contexts.back().IsExpression = true; 753 for (FormatToken *Previous = Current.Previous; 754 Previous && !Previous->isOneOf(tok::comma, tok::semi); 755 Previous = Previous->Previous) { 756 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 757 Previous = Previous->MatchingParen; 758 if (!Previous) 759 break; 760 } 761 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 762 Previous->isOneOf(tok::star, tok::amp) && Previous->Previous && 763 Previous->Previous->isNot(tok::equal)) 764 Previous->Type = TT_PointerOrReference; 765 } 766 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 767 Contexts.back().IsExpression = true; 768 } else if (Current.is(TT_TrailingReturnArrow)) { 769 Contexts.back().IsExpression = false; 770 } else if (Current.is(tok::l_paren) && !Line.MustBeDeclaration && 771 !Line.InPPDirective && 772 (!Current.Previous || 773 Current.Previous->isNot(tok::kw_decltype))) { 774 bool ParametersOfFunctionType = 775 Current.Previous && Current.Previous->is(tok::r_paren) && 776 Current.Previous->MatchingParen && 777 Current.Previous->MatchingParen->is(TT_FunctionTypeLParen); 778 bool IsForOrCatch = Current.Previous && 779 Current.Previous->isOneOf(tok::kw_for, tok::kw_catch); 780 Contexts.back().IsExpression = !ParametersOfFunctionType && !IsForOrCatch; 781 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 782 for (FormatToken *Previous = Current.Previous; 783 Previous && Previous->isOneOf(tok::star, tok::amp); 784 Previous = Previous->Previous) 785 Previous->Type = TT_PointerOrReference; 786 if (Line.MustBeDeclaration) 787 Contexts.back().IsExpression = Contexts.front().InCtorInitializer; 788 } else if (Current.Previous && 789 Current.Previous->is(TT_CtorInitializerColon)) { 790 Contexts.back().IsExpression = true; 791 Contexts.back().InCtorInitializer = true; 792 } else if (Current.is(tok::kw_new)) { 793 Contexts.back().CanBeExpression = false; 794 } else if (Current.is(tok::semi) || Current.is(tok::exclaim)) { 795 // This should be the condition or increment in a for-loop. 796 Contexts.back().IsExpression = true; 797 } 798 } 799 800 void determineTokenType(FormatToken &Current) { 801 if (!Current.is(TT_Unknown)) 802 // The token type is already known. 803 return; 804 805 // Line.MightBeFunctionDecl can only be true after the parentheses of a 806 // function declaration have been found. In this case, 'Current' is a 807 // trailing token of this declaration and thus cannot be a name. 808 if (Current.is(Keywords.kw_instanceof)) { 809 Current.Type = TT_BinaryOperator; 810 } else if (isStartOfName(Current) && 811 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 812 Contexts.back().FirstStartOfName = &Current; 813 Current.Type = TT_StartOfName; 814 } else if (Current.is(tok::kw_auto)) { 815 AutoFound = true; 816 } else if (Current.is(tok::arrow) && 817 Style.Language == FormatStyle::LK_Java) { 818 Current.Type = TT_LambdaArrow; 819 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 820 Current.NestingLevel == 0) { 821 Current.Type = TT_TrailingReturnArrow; 822 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 823 Current.Type = 824 determineStarAmpUsage(Current, Contexts.back().CanBeExpression && 825 Contexts.back().IsExpression, 826 Contexts.back().InTemplateArgument); 827 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 828 Current.Type = determinePlusMinusCaretUsage(Current); 829 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 830 Contexts.back().CaretFound = true; 831 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 832 Current.Type = determineIncrementUsage(Current); 833 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 834 Current.Type = TT_UnaryOperator; 835 } else if (Current.is(tok::question)) { 836 Current.Type = TT_ConditionalExpr; 837 } else if (Current.isBinaryOperator() && 838 (!Current.Previous || Current.Previous->isNot(tok::l_square))) { 839 Current.Type = TT_BinaryOperator; 840 } else if (Current.is(tok::comment)) { 841 if (Current.TokenText.startswith("//")) 842 Current.Type = TT_LineComment; 843 else 844 Current.Type = TT_BlockComment; 845 } else if (Current.is(tok::r_paren)) { 846 if (rParenEndsCast(Current)) 847 Current.Type = TT_CastRParen; 848 } else if (Current.is(tok::at) && Current.Next) { 849 switch (Current.Next->Tok.getObjCKeywordID()) { 850 case tok::objc_interface: 851 case tok::objc_implementation: 852 case tok::objc_protocol: 853 Current.Type = TT_ObjCDecl; 854 break; 855 case tok::objc_property: 856 Current.Type = TT_ObjCProperty; 857 break; 858 default: 859 break; 860 } 861 } else if (Current.is(tok::period)) { 862 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 863 if (PreviousNoComment && 864 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 865 Current.Type = TT_DesignatedInitializerPeriod; 866 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 867 Current.Previous->isOneOf(TT_JavaAnnotation, 868 TT_LeadingJavaAnnotation)) { 869 Current.Type = Current.Previous->Type; 870 } 871 } else if (Current.isOneOf(tok::identifier, tok::kw_const) && 872 Current.Previous && 873 !Current.Previous->isOneOf(tok::equal, tok::at) && 874 Line.MightBeFunctionDecl && Contexts.size() == 1) { 875 // Line.MightBeFunctionDecl can only be true after the parentheses of a 876 // function declaration have been found. 877 Current.Type = TT_TrailingAnnotation; 878 } else if (Style.Language == FormatStyle::LK_Java && Current.Previous) { 879 if (Current.Previous->is(tok::at) && 880 Current.isNot(Keywords.kw_interface)) { 881 const FormatToken &AtToken = *Current.Previous; 882 const FormatToken *Previous = AtToken.getPreviousNonComment(); 883 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 884 Current.Type = TT_LeadingJavaAnnotation; 885 else 886 Current.Type = TT_JavaAnnotation; 887 } else if (Current.Previous->is(tok::period) && 888 Current.Previous->isOneOf(TT_JavaAnnotation, 889 TT_LeadingJavaAnnotation)) { 890 Current.Type = Current.Previous->Type; 891 } 892 } 893 } 894 895 /// \brief Take a guess at whether \p Tok starts a name of a function or 896 /// variable declaration. 897 /// 898 /// This is a heuristic based on whether \p Tok is an identifier following 899 /// something that is likely a type. 900 bool isStartOfName(const FormatToken &Tok) { 901 if (Tok.isNot(tok::identifier) || !Tok.Previous) 902 return false; 903 904 if (Tok.Previous->is(TT_LeadingJavaAnnotation)) 905 return false; 906 907 // Skip "const" as it does not have an influence on whether this is a name. 908 FormatToken *PreviousNotConst = Tok.Previous; 909 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 910 PreviousNotConst = PreviousNotConst->Previous; 911 912 if (!PreviousNotConst) 913 return false; 914 915 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 916 PreviousNotConst->Previous && 917 PreviousNotConst->Previous->is(tok::hash); 918 919 if (PreviousNotConst->is(TT_TemplateCloser)) 920 return PreviousNotConst && PreviousNotConst->MatchingParen && 921 PreviousNotConst->MatchingParen->Previous && 922 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 923 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 924 925 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen && 926 PreviousNotConst->MatchingParen->Previous && 927 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype)) 928 return true; 929 930 return (!IsPPKeyword && PreviousNotConst->is(tok::identifier)) || 931 PreviousNotConst->is(TT_PointerOrReference) || 932 PreviousNotConst->isSimpleTypeSpecifier(); 933 } 934 935 /// \brief Determine whether ')' is ending a cast. 936 bool rParenEndsCast(const FormatToken &Tok) { 937 FormatToken *LeftOfParens = nullptr; 938 if (Tok.MatchingParen) 939 LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 940 if (LeftOfParens && LeftOfParens->is(tok::r_paren) && 941 LeftOfParens->MatchingParen) 942 LeftOfParens = LeftOfParens->MatchingParen->Previous; 943 if (LeftOfParens && LeftOfParens->is(tok::r_square) && 944 LeftOfParens->MatchingParen && 945 LeftOfParens->MatchingParen->is(TT_LambdaLSquare)) 946 return false; 947 if (Tok.Next) { 948 if (Tok.Next->is(tok::question)) 949 return false; 950 if (Style.Language == FormatStyle::LK_JavaScript && 951 Tok.Next->is(Keywords.kw_in)) 952 return false; 953 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 954 return true; 955 } 956 bool IsCast = false; 957 bool ParensAreEmpty = Tok.Previous == Tok.MatchingParen; 958 bool ParensAreType = 959 !Tok.Previous || 960 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) || 961 Tok.Previous->isSimpleTypeSpecifier(); 962 bool ParensCouldEndDecl = 963 Tok.Next && Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace); 964 bool IsSizeOfOrAlignOf = 965 LeftOfParens && LeftOfParens->isOneOf(tok::kw_sizeof, tok::kw_alignof); 966 if (ParensAreType && !ParensCouldEndDecl && !IsSizeOfOrAlignOf && 967 ((Contexts.size() > 1 && Contexts[Contexts.size() - 2].IsExpression) || 968 (Tok.Next && Tok.Next->isBinaryOperator()))) 969 IsCast = true; 970 else if (Tok.Next && Tok.Next->isNot(tok::string_literal) && 971 (Tok.Next->Tok.isLiteral() || 972 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 973 IsCast = true; 974 // If there is an identifier after the (), it is likely a cast, unless 975 // there is also an identifier before the (). 976 else if (LeftOfParens && Tok.Next && 977 (LeftOfParens->Tok.getIdentifierInfo() == nullptr || 978 LeftOfParens->is(tok::kw_return)) && 979 !LeftOfParens->isOneOf(TT_OverloadedOperator, tok::at, 980 TT_TemplateCloser)) { 981 if (Tok.Next->isOneOf(tok::identifier, tok::numeric_constant)) { 982 IsCast = true; 983 } else { 984 // Use heuristics to recognize c style casting. 985 FormatToken *Prev = Tok.Previous; 986 if (Prev && Prev->isOneOf(tok::amp, tok::star)) 987 Prev = Prev->Previous; 988 989 if (Prev && Tok.Next && Tok.Next->Next) { 990 bool NextIsUnary = Tok.Next->isUnaryOperator() || 991 Tok.Next->isOneOf(tok::amp, tok::star); 992 IsCast = 993 NextIsUnary && !Tok.Next->is(tok::plus) && 994 Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant); 995 } 996 997 for (; Prev != Tok.MatchingParen; Prev = Prev->Previous) { 998 if (!Prev || !Prev->isOneOf(tok::kw_const, tok::identifier)) { 999 IsCast = false; 1000 break; 1001 } 1002 } 1003 } 1004 } 1005 return IsCast && !ParensAreEmpty; 1006 } 1007 1008 /// \brief Return the type of the given token assuming it is * or &. 1009 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 1010 bool InTemplateArgument) { 1011 if (Style.Language == FormatStyle::LK_JavaScript) 1012 return TT_BinaryOperator; 1013 1014 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1015 if (!PrevToken) 1016 return TT_UnaryOperator; 1017 1018 const FormatToken *NextToken = Tok.getNextNonComment(); 1019 if (!NextToken || 1020 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 1021 return TT_Unknown; 1022 1023 if (PrevToken->is(tok::coloncolon)) 1024 return TT_PointerOrReference; 1025 1026 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 1027 tok::comma, tok::semi, tok::kw_return, tok::colon, 1028 tok::equal, tok::kw_delete, tok::kw_sizeof) || 1029 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 1030 TT_UnaryOperator, TT_CastRParen)) 1031 return TT_UnaryOperator; 1032 1033 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 1034 return TT_PointerOrReference; 1035 if (NextToken->isOneOf(tok::kw_operator, tok::comma)) 1036 return TT_PointerOrReference; 1037 1038 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen && 1039 PrevToken->MatchingParen->Previous && 1040 PrevToken->MatchingParen->Previous->isOneOf(tok::kw_typeof, 1041 tok::kw_decltype)) 1042 return TT_PointerOrReference; 1043 1044 if (PrevToken->Tok.isLiteral() || 1045 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 1046 tok::kw_false, tok::r_brace) || 1047 NextToken->Tok.isLiteral() || 1048 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 1049 NextToken->isUnaryOperator() || 1050 // If we know we're in a template argument, there are no named 1051 // declarations. Thus, having an identifier on the right-hand side 1052 // indicates a binary operator. 1053 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 1054 return TT_BinaryOperator; 1055 1056 // "&&(" is quite unlikely to be two successive unary "&". 1057 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren)) 1058 return TT_BinaryOperator; 1059 1060 // This catches some cases where evaluation order is used as control flow: 1061 // aaa && aaa->f(); 1062 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 1063 if (NextNextToken && NextNextToken->is(tok::arrow)) 1064 return TT_BinaryOperator; 1065 1066 // It is very unlikely that we are going to find a pointer or reference type 1067 // definition on the RHS of an assignment. 1068 if (IsExpression && !Contexts.back().CaretFound) 1069 return TT_BinaryOperator; 1070 1071 return TT_PointerOrReference; 1072 } 1073 1074 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 1075 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1076 if (!PrevToken || PrevToken->is(TT_CastRParen)) 1077 return TT_UnaryOperator; 1078 1079 // Use heuristics to recognize unary operators. 1080 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 1081 tok::question, tok::colon, tok::kw_return, 1082 tok::kw_case, tok::at, tok::l_brace)) 1083 return TT_UnaryOperator; 1084 1085 // There can't be two consecutive binary operators. 1086 if (PrevToken->is(TT_BinaryOperator)) 1087 return TT_UnaryOperator; 1088 1089 // Fall back to marking the token as binary operator. 1090 return TT_BinaryOperator; 1091 } 1092 1093 /// \brief Determine whether ++/-- are pre- or post-increments/-decrements. 1094 TokenType determineIncrementUsage(const FormatToken &Tok) { 1095 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1096 if (!PrevToken || PrevToken->is(TT_CastRParen)) 1097 return TT_UnaryOperator; 1098 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 1099 return TT_TrailingUnaryOperator; 1100 1101 return TT_UnaryOperator; 1102 } 1103 1104 SmallVector<Context, 8> Contexts; 1105 1106 const FormatStyle &Style; 1107 AnnotatedLine &Line; 1108 FormatToken *CurrentToken; 1109 bool AutoFound; 1110 const AdditionalKeywords &Keywords; 1111 }; 1112 1113 static int PrecedenceUnaryOperator = prec::PointerToMember + 1; 1114 static int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 1115 1116 /// \brief Parses binary expressions by inserting fake parenthesis based on 1117 /// operator precedence. 1118 class ExpressionParser { 1119 public: 1120 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 1121 AnnotatedLine &Line) 1122 : Style(Style), Keywords(Keywords), Current(Line.First) {} 1123 1124 /// \brief Parse expressions with the given operatore precedence. 1125 void parse(int Precedence = 0) { 1126 // Skip 'return' and ObjC selector colons as they are not part of a binary 1127 // expression. 1128 while (Current && (Current->is(tok::kw_return) || 1129 (Current->is(tok::colon) && 1130 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 1131 next(); 1132 1133 if (!Current || Precedence > PrecedenceArrowAndPeriod) 1134 return; 1135 1136 // Conditional expressions need to be parsed separately for proper nesting. 1137 if (Precedence == prec::Conditional) { 1138 parseConditionalExpr(); 1139 return; 1140 } 1141 1142 // Parse unary operators, which all have a higher precedence than binary 1143 // operators. 1144 if (Precedence == PrecedenceUnaryOperator) { 1145 parseUnaryOperator(); 1146 return; 1147 } 1148 1149 FormatToken *Start = Current; 1150 FormatToken *LatestOperator = nullptr; 1151 unsigned OperatorIndex = 0; 1152 1153 while (Current) { 1154 // Consume operators with higher precedence. 1155 parse(Precedence + 1); 1156 1157 int CurrentPrecedence = getCurrentPrecedence(); 1158 1159 if (Current && Current->is(TT_SelectorName) && 1160 Precedence == CurrentPrecedence) { 1161 if (LatestOperator) 1162 addFakeParenthesis(Start, prec::Level(Precedence)); 1163 Start = Current; 1164 } 1165 1166 // At the end of the line or when an operator with higher precedence is 1167 // found, insert fake parenthesis and return. 1168 if (!Current || (Current->closesScope() && Current->MatchingParen) || 1169 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 1170 (CurrentPrecedence == prec::Conditional && 1171 Precedence == prec::Assignment && Current->is(tok::colon))) { 1172 break; 1173 } 1174 1175 // Consume scopes: (), [], <> and {} 1176 if (Current->opensScope()) { 1177 while (Current && !Current->closesScope()) { 1178 next(); 1179 parse(); 1180 } 1181 next(); 1182 } else { 1183 // Operator found. 1184 if (CurrentPrecedence == Precedence) { 1185 LatestOperator = Current; 1186 Current->OperatorIndex = OperatorIndex; 1187 ++OperatorIndex; 1188 } 1189 next(/*SkipPastLeadingComments=*/Precedence > 0); 1190 } 1191 } 1192 1193 if (LatestOperator && (Current || Precedence > 0)) { 1194 LatestOperator->LastOperator = true; 1195 if (Precedence == PrecedenceArrowAndPeriod) { 1196 // Call expressions don't have a binary operator precedence. 1197 addFakeParenthesis(Start, prec::Unknown); 1198 } else { 1199 addFakeParenthesis(Start, prec::Level(Precedence)); 1200 } 1201 } 1202 } 1203 1204 private: 1205 /// \brief Gets the precedence (+1) of the given token for binary operators 1206 /// and other tokens that we treat like binary operators. 1207 int getCurrentPrecedence() { 1208 if (Current) { 1209 const FormatToken *NextNonComment = Current->getNextNonComment(); 1210 if (Current->is(TT_ConditionalExpr)) 1211 return prec::Conditional; 1212 else if (NextNonComment && NextNonComment->is(tok::colon) && 1213 NextNonComment->is(TT_DictLiteral)) 1214 return prec::Comma; 1215 else if (Current->is(TT_LambdaArrow)) 1216 return prec::Comma; 1217 else if (Current->isOneOf(tok::semi, TT_InlineASMColon, 1218 TT_SelectorName) || 1219 (Current->is(tok::comment) && NextNonComment && 1220 NextNonComment->is(TT_SelectorName))) 1221 return 0; 1222 else if (Current->is(TT_RangeBasedForLoopColon)) 1223 return prec::Comma; 1224 else if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 1225 return Current->getPrecedence(); 1226 else if (Current->isOneOf(tok::period, tok::arrow)) 1227 return PrecedenceArrowAndPeriod; 1228 else if (Style.Language == FormatStyle::LK_Java && 1229 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 1230 Keywords.kw_throws)) 1231 return 0; 1232 } 1233 return -1; 1234 } 1235 1236 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 1237 Start->FakeLParens.push_back(Precedence); 1238 if (Precedence > prec::Unknown) 1239 Start->StartsBinaryExpression = true; 1240 if (Current) { 1241 FormatToken *Previous = Current->Previous; 1242 while (Previous->is(tok::comment) && Previous->Previous) 1243 Previous = Previous->Previous; 1244 ++Previous->FakeRParens; 1245 if (Precedence > prec::Unknown) 1246 Previous->EndsBinaryExpression = true; 1247 } 1248 } 1249 1250 /// \brief Parse unary operator expressions and surround them with fake 1251 /// parentheses if appropriate. 1252 void parseUnaryOperator() { 1253 if (!Current || Current->isNot(TT_UnaryOperator)) { 1254 parse(PrecedenceArrowAndPeriod); 1255 return; 1256 } 1257 1258 FormatToken *Start = Current; 1259 next(); 1260 parseUnaryOperator(); 1261 1262 // The actual precedence doesn't matter. 1263 addFakeParenthesis(Start, prec::Unknown); 1264 } 1265 1266 void parseConditionalExpr() { 1267 while (Current && Current->isTrailingComment()) { 1268 next(); 1269 } 1270 FormatToken *Start = Current; 1271 parse(prec::LogicalOr); 1272 if (!Current || !Current->is(tok::question)) 1273 return; 1274 next(); 1275 parse(prec::Assignment); 1276 if (!Current || Current->isNot(TT_ConditionalExpr)) 1277 return; 1278 next(); 1279 parse(prec::Assignment); 1280 addFakeParenthesis(Start, prec::Conditional); 1281 } 1282 1283 void next(bool SkipPastLeadingComments = true) { 1284 if (Current) 1285 Current = Current->Next; 1286 while (Current && 1287 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 1288 Current->isTrailingComment()) 1289 Current = Current->Next; 1290 } 1291 1292 const FormatStyle &Style; 1293 const AdditionalKeywords &Keywords; 1294 FormatToken *Current; 1295 }; 1296 1297 } // end anonymous namespace 1298 1299 void TokenAnnotator::setCommentLineLevels( 1300 SmallVectorImpl<AnnotatedLine *> &Lines) { 1301 const AnnotatedLine *NextNonCommentLine = nullptr; 1302 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(), 1303 E = Lines.rend(); 1304 I != E; ++I) { 1305 if (NextNonCommentLine && (*I)->First->is(tok::comment) && 1306 (*I)->First->Next == nullptr) 1307 (*I)->Level = NextNonCommentLine->Level; 1308 else 1309 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr; 1310 1311 setCommentLineLevels((*I)->Children); 1312 } 1313 } 1314 1315 void TokenAnnotator::annotate(AnnotatedLine &Line) { 1316 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 1317 E = Line.Children.end(); 1318 I != E; ++I) { 1319 annotate(**I); 1320 } 1321 AnnotatingParser Parser(Style, Line, Keywords); 1322 Line.Type = Parser.parseLine(); 1323 if (Line.Type == LT_Invalid) 1324 return; 1325 1326 ExpressionParser ExprParser(Style, Keywords, Line); 1327 ExprParser.parse(); 1328 1329 if (Line.First->is(TT_ObjCMethodSpecifier)) 1330 Line.Type = LT_ObjCMethodDecl; 1331 else if (Line.First->is(TT_ObjCDecl)) 1332 Line.Type = LT_ObjCDecl; 1333 else if (Line.First->is(TT_ObjCProperty)) 1334 Line.Type = LT_ObjCProperty; 1335 1336 Line.First->SpacesRequiredBefore = 1; 1337 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 1338 } 1339 1340 // This function heuristically determines whether 'Current' starts the name of a 1341 // function declaration. 1342 static bool isFunctionDeclarationName(const FormatToken &Current) { 1343 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 1344 return false; 1345 const FormatToken *Next = Current.Next; 1346 for (; Next; Next = Next->Next) { 1347 if (Next->is(TT_TemplateOpener)) { 1348 Next = Next->MatchingParen; 1349 } else if (Next->is(tok::coloncolon)) { 1350 Next = Next->Next; 1351 if (!Next || !Next->is(tok::identifier)) 1352 return false; 1353 } else if (Next->is(tok::l_paren)) { 1354 break; 1355 } else { 1356 return false; 1357 } 1358 } 1359 if (!Next) 1360 return false; 1361 assert(Next->is(tok::l_paren)); 1362 if (Next->Next == Next->MatchingParen) 1363 return true; 1364 for (const FormatToken *Tok = Next->Next; Tok != Next->MatchingParen; 1365 Tok = Tok->Next) { 1366 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 1367 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName)) 1368 return true; 1369 if (Tok->isOneOf(tok::l_brace, tok::string_literal) || Tok->Tok.isLiteral()) 1370 return false; 1371 } 1372 return false; 1373 } 1374 1375 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 1376 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 1377 E = Line.Children.end(); 1378 I != E; ++I) { 1379 calculateFormattingInformation(**I); 1380 } 1381 1382 Line.First->TotalLength = 1383 Line.First->IsMultiline ? Style.ColumnLimit : Line.First->ColumnWidth; 1384 if (!Line.First->Next) 1385 return; 1386 FormatToken *Current = Line.First->Next; 1387 bool InFunctionDecl = Line.MightBeFunctionDecl; 1388 while (Current) { 1389 if (isFunctionDeclarationName(*Current)) 1390 Current->Type = TT_FunctionDeclarationName; 1391 if (Current->is(TT_LineComment)) { 1392 if (Current->Previous->BlockKind == BK_BracedInit && 1393 Current->Previous->opensScope()) 1394 Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1; 1395 else 1396 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 1397 1398 // If we find a trailing comment, iterate backwards to determine whether 1399 // it seems to relate to a specific parameter. If so, break before that 1400 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 1401 // to the previous line in: 1402 // SomeFunction(a, 1403 // b, // comment 1404 // c); 1405 if (!Current->HasUnescapedNewline) { 1406 for (FormatToken *Parameter = Current->Previous; Parameter; 1407 Parameter = Parameter->Previous) { 1408 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 1409 break; 1410 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 1411 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 1412 Parameter->HasUnescapedNewline) 1413 Parameter->MustBreakBefore = true; 1414 break; 1415 } 1416 } 1417 } 1418 } else if (Current->SpacesRequiredBefore == 0 && 1419 spaceRequiredBefore(Line, *Current)) { 1420 Current->SpacesRequiredBefore = 1; 1421 } 1422 1423 Current->MustBreakBefore = 1424 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 1425 1426 if (Style.AlwaysBreakAfterDefinitionReturnType && InFunctionDecl && 1427 Current->is(TT_FunctionDeclarationName) && 1428 !Line.Last->isOneOf(tok::semi, tok::comment)) // Only for definitions. 1429 // FIXME: Line.Last points to other characters than tok::semi 1430 // and tok::lbrace. 1431 Current->MustBreakBefore = true; 1432 1433 Current->CanBreakBefore = 1434 Current->MustBreakBefore || canBreakBefore(Line, *Current); 1435 unsigned ChildSize = 0; 1436 if (Current->Previous->Children.size() == 1) { 1437 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last; 1438 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 1439 : LastOfChild.TotalLength + 1; 1440 } 1441 const FormatToken *Prev = Current->Previous; 1442 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 1443 (Prev->Children.size() == 1 && 1444 Prev->Children[0]->First->MustBreakBefore) || 1445 Current->IsMultiline) 1446 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 1447 else 1448 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 1449 ChildSize + Current->SpacesRequiredBefore; 1450 1451 if (Current->is(TT_CtorInitializerColon)) 1452 InFunctionDecl = false; 1453 1454 // FIXME: Only calculate this if CanBreakBefore is true once static 1455 // initializers etc. are sorted out. 1456 // FIXME: Move magic numbers to a better place. 1457 Current->SplitPenalty = 20 * Current->BindingStrength + 1458 splitPenalty(Line, *Current, InFunctionDecl); 1459 1460 Current = Current->Next; 1461 } 1462 1463 calculateUnbreakableTailLengths(Line); 1464 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 1465 if (Current->Role) 1466 Current->Role->precomputeFormattingInfos(Current); 1467 } 1468 1469 DEBUG({ printDebugInfo(Line); }); 1470 } 1471 1472 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 1473 unsigned UnbreakableTailLength = 0; 1474 FormatToken *Current = Line.Last; 1475 while (Current) { 1476 Current->UnbreakableTailLength = UnbreakableTailLength; 1477 if (Current->CanBreakBefore || 1478 Current->isOneOf(tok::comment, tok::string_literal)) { 1479 UnbreakableTailLength = 0; 1480 } else { 1481 UnbreakableTailLength += 1482 Current->ColumnWidth + Current->SpacesRequiredBefore; 1483 } 1484 Current = Current->Previous; 1485 } 1486 } 1487 1488 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 1489 const FormatToken &Tok, 1490 bool InFunctionDecl) { 1491 const FormatToken &Left = *Tok.Previous; 1492 const FormatToken &Right = Tok; 1493 1494 if (Left.is(tok::semi)) 1495 return 0; 1496 1497 if (Style.Language == FormatStyle::LK_Java) { 1498 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 1499 return 1; 1500 if (Right.is(Keywords.kw_implements)) 1501 return 2; 1502 if (Left.is(tok::comma) && Left.NestingLevel == 0) 1503 return 3; 1504 } else if (Style.Language == FormatStyle::LK_JavaScript) { 1505 if (Right.is(Keywords.kw_function)) 1506 return 100; 1507 } 1508 1509 if (Left.is(tok::comma) || (Right.is(tok::identifier) && Right.Next && 1510 Right.Next->is(TT_DictLiteral))) 1511 return 1; 1512 if (Right.is(tok::l_square)) { 1513 if (Style.Language == FormatStyle::LK_Proto) 1514 return 1; 1515 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare)) 1516 return 500; 1517 } 1518 1519 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 1520 Right.is(tok::kw_operator)) { 1521 if (Line.First->is(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 1522 return 3; 1523 if (Left.is(TT_StartOfName)) 1524 return 20; 1525 if (InFunctionDecl && Right.NestingLevel == 0) 1526 return Style.PenaltyReturnTypeOnItsOwnLine; 1527 return 200; 1528 } 1529 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 1530 return 150; 1531 if (Left.is(TT_CastRParen)) 1532 return 100; 1533 if (Left.is(tok::coloncolon) || 1534 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 1535 return 500; 1536 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 1537 return 5000; 1538 1539 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon)) 1540 return 2; 1541 1542 if (Right.isMemberAccess()) { 1543 if (Left.is(tok::r_paren) && Left.MatchingParen && 1544 Left.MatchingParen->ParameterCount > 0) 1545 return 20; // Should be smaller than breaking at a nested comma. 1546 return 150; 1547 } 1548 1549 if (Right.is(TT_TrailingAnnotation) && 1550 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 1551 // Moving trailing annotations to the next line is fine for ObjC method 1552 // declarations. 1553 if (Line.First->is(TT_ObjCMethodSpecifier)) 1554 1555 return 10; 1556 // Generally, breaking before a trailing annotation is bad unless it is 1557 // function-like. It seems to be especially preferable to keep standard 1558 // annotations (i.e. "const", "final" and "override") on the same line. 1559 // Use a slightly higher penalty after ")" so that annotations like 1560 // "const override" are kept together. 1561 bool is_short_annotation = Right.TokenText.size() < 10; 1562 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 1563 } 1564 1565 // In for-loops, prefer breaking at ',' and ';'. 1566 if (Line.First->is(tok::kw_for) && Left.is(tok::equal)) 1567 return 4; 1568 1569 // In Objective-C method expressions, prefer breaking before "param:" over 1570 // breaking after it. 1571 if (Right.is(TT_SelectorName)) 1572 return 0; 1573 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 1574 return Line.MightBeFunctionDecl ? 50 : 500; 1575 1576 if (Left.is(tok::l_paren) && InFunctionDecl && Style.AlignAfterOpenBracket) 1577 return 100; 1578 if (Left.is(tok::equal) && InFunctionDecl) 1579 return 110; 1580 if (Right.is(tok::r_brace)) 1581 return 1; 1582 if (Left.is(TT_TemplateOpener)) 1583 return 100; 1584 if (Left.opensScope()) { 1585 if (!Style.AlignAfterOpenBracket) 1586 return 0; 1587 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 1588 : 19; 1589 } 1590 if (Left.is(TT_JavaAnnotation)) 1591 return 50; 1592 1593 if (Right.is(tok::lessless)) { 1594 if (Left.is(tok::string_literal)) { 1595 StringRef Content = Left.TokenText; 1596 if (Content.startswith("\"")) 1597 Content = Content.drop_front(1); 1598 if (Content.endswith("\"")) 1599 Content = Content.drop_back(1); 1600 Content = Content.trim(); 1601 if (Content.size() > 1 && 1602 (Content.back() == ':' || Content.back() == '=')) 1603 return 25; 1604 } 1605 return 1; // Breaking at a << is really cheap. 1606 } 1607 if (Left.is(TT_ConditionalExpr)) 1608 return prec::Conditional; 1609 prec::Level Level = Left.getPrecedence(); 1610 1611 if (Level != prec::Unknown) 1612 return Level; 1613 1614 return 3; 1615 } 1616 1617 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 1618 const FormatToken &Left, 1619 const FormatToken &Right) { 1620 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 1621 return true; 1622 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 1623 Left.Tok.getObjCKeywordID() == tok::objc_property) 1624 return true; 1625 if (Right.is(tok::hashhash)) 1626 return Left.is(tok::hash); 1627 if (Left.isOneOf(tok::hashhash, tok::hash)) 1628 return Right.is(tok::hash); 1629 if (Left.is(tok::l_paren) && Right.is(tok::r_paren)) 1630 return Style.SpaceInEmptyParentheses; 1631 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 1632 return (Right.is(TT_CastRParen) || 1633 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 1634 ? Style.SpacesInCStyleCastParentheses 1635 : Style.SpacesInParentheses; 1636 if (Right.isOneOf(tok::semi, tok::comma)) 1637 return false; 1638 if (Right.is(tok::less) && 1639 (Left.isOneOf(tok::kw_template, tok::r_paren) || 1640 (Line.Type == LT_ObjCDecl && Style.ObjCSpaceBeforeProtocolList))) 1641 return true; 1642 if (Left.isOneOf(tok::exclaim, tok::tilde)) 1643 return false; 1644 if (Left.is(tok::at) && 1645 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 1646 tok::numeric_constant, tok::l_paren, tok::l_brace, 1647 tok::kw_true, tok::kw_false)) 1648 return false; 1649 if (Left.is(tok::coloncolon)) 1650 return false; 1651 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) 1652 return false; 1653 if (Right.is(tok::ellipsis)) 1654 return Left.Tok.isLiteral(); 1655 if (Left.is(tok::l_square) && Right.is(tok::amp)) 1656 return false; 1657 if (Right.is(TT_PointerOrReference)) 1658 return Left.Tok.isLiteral() || 1659 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 1660 Style.PointerAlignment != FormatStyle::PAS_Left); 1661 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 1662 (!Left.is(TT_PointerOrReference) || 1663 Style.PointerAlignment != FormatStyle::PAS_Right)) 1664 return true; 1665 if (Left.is(TT_PointerOrReference)) 1666 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) || 1667 (!Right.isOneOf(TT_PointerOrReference, tok::l_paren) && 1668 Style.PointerAlignment != FormatStyle::PAS_Right && Left.Previous && 1669 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon)); 1670 if (Right.is(tok::star) && Left.is(tok::l_paren)) 1671 return false; 1672 if (Left.is(tok::l_square)) 1673 return (Left.is(TT_ArrayInitializerLSquare) && 1674 Style.SpacesInContainerLiterals && Right.isNot(tok::r_square)) || 1675 (Left.is(TT_ArraySubscriptLSquare) && Style.SpacesInSquareBrackets && 1676 Right.isNot(tok::r_square)); 1677 if (Right.is(tok::r_square)) 1678 return Right.MatchingParen && 1679 ((Style.SpacesInContainerLiterals && 1680 Right.MatchingParen->is(TT_ArrayInitializerLSquare)) || 1681 (Style.SpacesInSquareBrackets && 1682 Right.MatchingParen->is(TT_ArraySubscriptLSquare))); 1683 if (Right.is(tok::l_square) && 1684 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare) && 1685 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral)) 1686 return false; 1687 if (Left.is(tok::colon)) 1688 return !Left.is(TT_ObjCMethodExpr); 1689 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 1690 return !Left.Children.empty(); // No spaces in "{}". 1691 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) || 1692 (Right.is(tok::r_brace) && Right.MatchingParen && 1693 Right.MatchingParen->BlockKind != BK_Block)) 1694 return !Style.Cpp11BracedListStyle; 1695 if (Left.is(TT_BlockComment)) 1696 return !Left.TokenText.endswith("=*/"); 1697 if (Right.is(tok::l_paren)) { 1698 if (Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) 1699 return true; 1700 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) || 1701 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never && 1702 (Left.isOneOf(tok::kw_if, tok::kw_for, tok::kw_while, 1703 tok::kw_switch, tok::kw_case) || 1704 (Left.isOneOf(tok::kw_try, tok::kw_catch, tok::kw_new, 1705 tok::kw_delete) && 1706 (!Left.Previous || Left.Previous->isNot(tok::period))) || 1707 Left.IsForEachMacro)) || 1708 (Style.SpaceBeforeParens == FormatStyle::SBPO_Always && 1709 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword()) && 1710 Line.Type != LT_PreprocessorDirective); 1711 } 1712 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 1713 return false; 1714 if (Right.is(TT_UnaryOperator)) 1715 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 1716 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 1717 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 1718 tok::r_paren) || 1719 Left.isSimpleTypeSpecifier()) && 1720 Right.is(tok::l_brace) && Right.getNextNonComment() && 1721 Right.BlockKind != BK_Block) 1722 return false; 1723 if (Left.is(tok::period) || Right.is(tok::period)) 1724 return false; 1725 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L") 1726 return false; 1727 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 1728 Left.MatchingParen->Previous && 1729 Left.MatchingParen->Previous->is(tok::period)) 1730 // A.<B>DoSomething(); 1731 return false; 1732 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 1733 return false; 1734 return true; 1735 } 1736 1737 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 1738 const FormatToken &Right) { 1739 const FormatToken &Left = *Right.Previous; 1740 if (Style.Language == FormatStyle::LK_Proto) { 1741 if (Right.is(tok::period) && 1742 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 1743 Keywords.kw_repeated)) 1744 return true; 1745 if (Right.is(tok::l_paren) && 1746 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 1747 return true; 1748 } else if (Style.Language == FormatStyle::LK_JavaScript) { 1749 if (Left.is(Keywords.kw_var)) 1750 return true; 1751 } else if (Style.Language == FormatStyle::LK_Java) { 1752 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 1753 return true; 1754 if (Left.is(TT_LambdaArrow) || Right.is(TT_LambdaArrow)) 1755 return true; 1756 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 1757 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never; 1758 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 1759 tok::kw_protected) || 1760 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 1761 Keywords.kw_native)) && 1762 Right.is(TT_TemplateOpener)) 1763 return true; 1764 } 1765 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 1766 return true; // Never ever merge two identifiers. 1767 if (Left.is(TT_ImplicitStringLiteral)) 1768 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 1769 if (Line.Type == LT_ObjCMethodDecl) { 1770 if (Left.is(TT_ObjCMethodSpecifier)) 1771 return true; 1772 if (Left.is(tok::r_paren) && Right.is(tok::identifier)) 1773 // Don't space between ')' and <id> 1774 return false; 1775 } 1776 if (Line.Type == LT_ObjCProperty && 1777 (Right.is(tok::equal) || Left.is(tok::equal))) 1778 return false; 1779 1780 if (Right.is(TT_TrailingReturnArrow) || Left.is(TT_TrailingReturnArrow)) 1781 return true; 1782 if (Left.is(tok::comma)) 1783 return true; 1784 if (Right.is(tok::comma)) 1785 return false; 1786 if (Right.isOneOf(TT_CtorInitializerColon, TT_ObjCBlockLParen)) 1787 return true; 1788 if (Left.is(tok::kw_operator)) 1789 return Right.is(tok::coloncolon); 1790 if (Right.is(TT_OverloadedOperatorLParen)) 1791 return false; 1792 if (Right.is(tok::colon)) 1793 return !Line.First->isOneOf(tok::kw_case, tok::kw_default) && 1794 Right.getNextNonComment() && Right.isNot(TT_ObjCMethodExpr) && 1795 !Left.is(tok::question) && 1796 !(Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) && 1797 (Right.isNot(TT_DictLiteral) || Style.SpacesInContainerLiterals); 1798 if (Left.is(TT_UnaryOperator)) 1799 return Right.is(TT_BinaryOperator); 1800 if (Left.is(TT_CastRParen)) 1801 return Style.SpaceAfterCStyleCast || Right.is(TT_BinaryOperator); 1802 if (Left.is(tok::greater) && Right.is(tok::greater)) { 1803 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 1804 (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles); 1805 } 1806 if (Right.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 1807 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar)) 1808 return false; 1809 if (!Style.SpaceBeforeAssignmentOperators && 1810 Right.getPrecedence() == prec::Assignment) 1811 return false; 1812 if (Right.is(tok::coloncolon) && Left.isNot(tok::l_brace)) 1813 return (Left.is(TT_TemplateOpener) && 1814 Style.Standard == FormatStyle::LS_Cpp03) || 1815 !(Left.isOneOf(tok::identifier, tok::l_paren, tok::r_paren) || 1816 Left.isOneOf(TT_TemplateCloser, TT_TemplateOpener)); 1817 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 1818 return Style.SpacesInAngles; 1819 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 1820 Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr)) 1821 return true; 1822 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren)) 1823 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always; 1824 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 1825 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 1826 return false; 1827 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 1828 Line.First->is(tok::hash)) 1829 return true; 1830 if (Right.is(TT_TrailingUnaryOperator)) 1831 return false; 1832 if (Left.is(TT_RegexLiteral)) 1833 return false; 1834 return spaceRequiredBetween(Line, Left, Right); 1835 } 1836 1837 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 1838 static bool isAllmanBrace(const FormatToken &Tok) { 1839 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block && 1840 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral); 1841 } 1842 1843 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 1844 const FormatToken &Right) { 1845 const FormatToken &Left = *Right.Previous; 1846 if (Right.NewlinesBefore > 1) 1847 return true; 1848 1849 // If the last token before a '}' is a comma or a trailing comment, the 1850 // intention is to insert a line break after it in order to make shuffling 1851 // around entries easier. 1852 const FormatToken *BeforeClosingBrace = nullptr; 1853 if (Left.is(tok::l_brace) && Left.BlockKind != BK_Block && Left.MatchingParen) 1854 BeforeClosingBrace = Left.MatchingParen->Previous; 1855 else if (Right.is(tok::r_brace) && Right.BlockKind != BK_Block) 1856 BeforeClosingBrace = &Left; 1857 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 1858 BeforeClosingBrace->isTrailingComment())) 1859 return true; 1860 1861 if (Right.is(tok::comment)) 1862 return Left.BlockKind != BK_BracedInit && 1863 Left.isNot(TT_CtorInitializerColon) && 1864 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 1865 if (Right.Previous->isTrailingComment() || 1866 (Right.isStringLiteral() && Right.Previous->isStringLiteral())) 1867 return true; 1868 if (Right.Previous->IsUnterminatedLiteral) 1869 return true; 1870 if (Right.is(tok::lessless) && Right.Next && 1871 Right.Previous->is(tok::string_literal) && 1872 Right.Next->is(tok::string_literal)) 1873 return true; 1874 if (Right.Previous->ClosesTemplateDeclaration && 1875 Right.Previous->MatchingParen && 1876 Right.Previous->MatchingParen->NestingLevel == 0 && 1877 Style.AlwaysBreakTemplateDeclarations) 1878 return true; 1879 if ((Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) && 1880 Style.BreakConstructorInitializersBeforeComma && 1881 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 1882 return true; 1883 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 1884 // Raw string literals are special wrt. line breaks. The author has made a 1885 // deliberate choice and might have aligned the contents of the string 1886 // literal accordingly. Thus, we try keep existing line breaks. 1887 return Right.NewlinesBefore > 0; 1888 if (Right.Previous->is(tok::l_brace) && Right.NestingLevel == 1 && 1889 Style.Language == FormatStyle::LK_Proto) 1890 // Don't put enums onto single lines in protocol buffers. 1891 return true; 1892 if (Style.Language == FormatStyle::LK_JavaScript && Right.is(tok::r_brace) && 1893 Left.is(tok::l_brace) && !Left.Children.empty()) 1894 // Support AllowShortFunctionsOnASingleLine for JavaScript. 1895 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 1896 (Left.NestingLevel == 0 && Line.Level == 0 && 1897 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Inline); 1898 if (isAllmanBrace(Left) || isAllmanBrace(Right)) 1899 return Style.BreakBeforeBraces == FormatStyle::BS_Allman || 1900 Style.BreakBeforeBraces == FormatStyle::BS_GNU; 1901 if (Style.Language == FormatStyle::LK_Proto && Left.isNot(tok::l_brace) && 1902 Right.is(TT_SelectorName)) 1903 return true; 1904 if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine) 1905 return true; 1906 if (Right.is(tok::lessless) && Left.is(tok::identifier) && 1907 Left.TokenText == "endl") 1908 return true; 1909 1910 if (Style.Language == FormatStyle::LK_JavaScript) { 1911 // FIXME: This might apply to other languages and token kinds. 1912 if (Right.is(tok::char_constant) && Left.is(tok::plus) && Left.Previous && 1913 Left.Previous->is(tok::char_constant)) 1914 return true; 1915 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && 1916 Left.NestingLevel == 0) 1917 return true; 1918 } else if (Style.Language == FormatStyle::LK_Java) { 1919 if (Left.is(TT_LeadingJavaAnnotation) && 1920 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 1921 Line.Last->is(tok::l_brace)) 1922 return true; 1923 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 1924 Right.Next->is(tok::string_literal)) 1925 return true; 1926 } 1927 1928 return false; 1929 } 1930 1931 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 1932 const FormatToken &Right) { 1933 const FormatToken &Left = *Right.Previous; 1934 1935 if (Style.Language == FormatStyle::LK_Java) { 1936 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 1937 Keywords.kw_implements)) 1938 return false; 1939 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 1940 Keywords.kw_implements)) 1941 return true; 1942 } 1943 1944 if (Left.is(tok::at)) 1945 return false; 1946 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 1947 return false; 1948 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 1949 return !Right.is(tok::l_paren); 1950 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 1951 Right.is(tok::kw_operator)) 1952 return true; 1953 if (Right.isTrailingComment()) 1954 // We rely on MustBreakBefore being set correctly here as we should not 1955 // change the "binding" behavior of a comment. 1956 // The first comment in a braced lists is always interpreted as belonging to 1957 // the first list element. Otherwise, it should be placed outside of the 1958 // list. 1959 return Left.BlockKind == BK_BracedInit; 1960 if (Left.is(tok::question) && Right.is(tok::colon)) 1961 return false; 1962 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 1963 return Style.BreakBeforeTernaryOperators; 1964 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 1965 return !Style.BreakBeforeTernaryOperators; 1966 if (Right.is(TT_InheritanceColon)) 1967 return true; 1968 if (Right.is(tok::colon) && 1969 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 1970 return false; 1971 if (Left.is(tok::colon) && (Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr))) 1972 return true; 1973 if (Right.is(TT_SelectorName)) 1974 return true; 1975 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 1976 return true; 1977 if (Left.ClosesTemplateDeclaration) 1978 return true; 1979 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 1980 TT_OverloadedOperator)) 1981 return false; 1982 if (Left.is(TT_RangeBasedForLoopColon)) 1983 return true; 1984 if (Right.is(TT_RangeBasedForLoopColon)) 1985 return false; 1986 if (Left.isOneOf(TT_PointerOrReference, TT_TemplateCloser, 1987 TT_UnaryOperator) || 1988 Left.is(tok::kw_operator)) 1989 return false; 1990 if (Left.is(tok::equal) && Line.Type == LT_VirtualFunctionDecl) 1991 return false; 1992 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen)) 1993 return false; 1994 if (Left.is(tok::l_paren) && Left.Previous && 1995 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen) || 1996 Left.Previous->is(tok::kw_if))) 1997 return false; 1998 if (Right.is(TT_ImplicitStringLiteral)) 1999 return false; 2000 2001 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser)) 2002 return false; 2003 2004 // We only break before r_brace if there was a corresponding break before 2005 // the l_brace, which is tracked by BreakBeforeClosingBrace. 2006 if (Right.is(tok::r_brace)) 2007 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block; 2008 2009 // Allow breaking after a trailing annotation, e.g. after a method 2010 // declaration. 2011 if (Left.is(TT_TrailingAnnotation)) 2012 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 2013 tok::less, tok::coloncolon); 2014 2015 if (Right.is(tok::kw___attribute)) 2016 return true; 2017 2018 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 2019 return true; 2020 2021 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2022 return true; 2023 2024 if (Left.is(TT_CtorInitializerComma) && 2025 Style.BreakConstructorInitializersBeforeComma) 2026 return false; 2027 if (Right.is(TT_CtorInitializerComma) && 2028 Style.BreakConstructorInitializersBeforeComma) 2029 return true; 2030 if (Left.is(tok::greater) && Right.is(tok::greater) && 2031 Left.isNot(TT_TemplateCloser)) 2032 return false; 2033 if (Right.is(TT_BinaryOperator) && 2034 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 2035 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 2036 Right.getPrecedence() != prec::Assignment)) 2037 return true; 2038 if (Left.is(TT_ArrayInitializerLSquare)) 2039 return true; 2040 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 2041 return true; 2042 if (Left.isBinaryOperator() && !Left.isOneOf(tok::arrowstar, tok::lessless) && 2043 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 2044 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 2045 Left.getPrecedence() == prec::Assignment)) 2046 return true; 2047 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 2048 tok::kw_class, tok::kw_struct) || 2049 Right.isMemberAccess() || Right.is(TT_TrailingReturnArrow) || 2050 Right.isOneOf(tok::lessless, tok::colon, tok::l_square, tok::at) || 2051 (Left.is(tok::r_paren) && 2052 Right.isOneOf(tok::identifier, tok::kw_const)) || 2053 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)); 2054 } 2055 2056 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 2057 llvm::errs() << "AnnotatedTokens:\n"; 2058 const FormatToken *Tok = Line.First; 2059 while (Tok) { 2060 llvm::errs() << " M=" << Tok->MustBreakBefore 2061 << " C=" << Tok->CanBreakBefore << " T=" << Tok->Type 2062 << " S=" << Tok->SpacesRequiredBefore 2063 << " B=" << Tok->BlockParameterCount 2064 << " P=" << Tok->SplitPenalty << " Name=" << Tok->Tok.getName() 2065 << " L=" << Tok->TotalLength << " PPK=" << Tok->PackingKind 2066 << " FakeLParens="; 2067 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 2068 llvm::errs() << Tok->FakeLParens[i] << "/"; 2069 llvm::errs() << " FakeRParens=" << Tok->FakeRParens << "\n"; 2070 if (!Tok->Next) 2071 assert(Tok == Line.Last); 2072 Tok = Tok->Next; 2073 } 2074 llvm::errs() << "----\n"; 2075 } 2076 2077 } // namespace format 2078 } // namespace clang 2079