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