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/ADT/SmallPtrSet.h" 19 #include "llvm/Support/Debug.h" 20 21 #define DEBUG_TYPE "format-token-annotator" 22 23 namespace clang { 24 namespace format { 25 26 namespace { 27 28 /// \brief A parser that gathers additional information about tokens. 29 /// 30 /// The \c TokenAnnotator tries to match parenthesis and square brakets and 31 /// store a parenthesis levels. It also tries to resolve matching "<" and ">" 32 /// into template parameter lists. 33 class AnnotatingParser { 34 public: 35 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line, 36 const AdditionalKeywords &Keywords) 37 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false), 38 Keywords(Keywords) { 39 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false)); 40 resetTokenMetadata(CurrentToken); 41 } 42 43 private: 44 bool parseAngle() { 45 if (!CurrentToken || !CurrentToken->Previous) 46 return false; 47 if (NonTemplateLess.count(CurrentToken->Previous)) 48 return false; 49 50 const FormatToken &Previous = *CurrentToken->Previous; // The '<'. 51 if (Previous.Previous) { 52 if (Previous.Previous->Tok.isLiteral()) 53 return false; 54 if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 && 55 (!Previous.Previous->MatchingParen || 56 !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen))) 57 return false; 58 } 59 60 FormatToken *Left = CurrentToken->Previous; 61 Left->ParentBracket = Contexts.back().ContextKind; 62 ScopedContextCreator ContextCreator(*this, tok::less, 12); 63 64 // If this angle is in the context of an expression, we need to be more 65 // hesitant to detect it as opening template parameters. 66 bool InExprContext = Contexts.back().IsExpression; 67 68 Contexts.back().IsExpression = false; 69 // If there's a template keyword before the opening angle bracket, this is a 70 // template parameter, not an argument. 71 Contexts.back().InTemplateArgument = 72 Left->Previous && Left->Previous->Tok.isNot(tok::kw_template); 73 74 if (Style.Language == FormatStyle::LK_Java && 75 CurrentToken->is(tok::question)) 76 next(); 77 78 while (CurrentToken) { 79 if (CurrentToken->is(tok::greater)) { 80 Left->MatchingParen = CurrentToken; 81 CurrentToken->MatchingParen = Left; 82 // In TT_Proto, we must distignuish between: 83 // map<key, value> 84 // msg < item: data > 85 // msg: < item: data > 86 // In TT_TextProto, map<key, value> does not occur. 87 if (Style.Language == FormatStyle::LK_TextProto || 88 (Style.Language == FormatStyle::LK_Proto && Left->Previous && 89 Left->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) 90 CurrentToken->Type = TT_DictLiteral; 91 else 92 CurrentToken->Type = TT_TemplateCloser; 93 next(); 94 return true; 95 } 96 if (CurrentToken->is(tok::question) && 97 Style.Language == FormatStyle::LK_Java) { 98 next(); 99 continue; 100 } 101 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) || 102 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext && 103 Style.Language != FormatStyle::LK_Proto && 104 Style.Language != FormatStyle::LK_TextProto)) 105 return false; 106 // If a && or || is found and interpreted as a binary operator, this set 107 // of angles is likely part of something like "a < b && c > d". If the 108 // angles are inside an expression, the ||/&& might also be a binary 109 // operator that was misinterpreted because we are parsing template 110 // parameters. 111 // FIXME: This is getting out of hand, write a decent parser. 112 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) && 113 CurrentToken->Previous->is(TT_BinaryOperator) && 114 Contexts[Contexts.size() - 2].IsExpression && 115 !Line.startsWith(tok::kw_template)) 116 return false; 117 updateParameterCount(Left, CurrentToken); 118 if (Style.Language == FormatStyle::LK_Proto) { 119 if (FormatToken *Previous = CurrentToken->getPreviousNonComment()) { 120 if (CurrentToken->is(tok::colon) || 121 (CurrentToken->isOneOf(tok::l_brace, tok::less) && 122 Previous->isNot(tok::colon))) 123 Previous->Type = TT_SelectorName; 124 } 125 } 126 if (!consumeToken()) 127 return false; 128 } 129 return false; 130 } 131 132 bool parseParens(bool LookForDecls = false) { 133 if (!CurrentToken) 134 return false; 135 FormatToken *Left = CurrentToken->Previous; 136 Left->ParentBracket = Contexts.back().ContextKind; 137 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1); 138 139 // FIXME: This is a bit of a hack. Do better. 140 Contexts.back().ColonIsForRangeExpr = 141 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr; 142 143 bool StartsObjCMethodExpr = false; 144 if (CurrentToken->is(tok::caret)) { 145 // (^ can start a block type. 146 Left->Type = TT_ObjCBlockLParen; 147 } else if (FormatToken *MaybeSel = Left->Previous) { 148 // @selector( starts a selector. 149 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous && 150 MaybeSel->Previous->is(tok::at)) { 151 StartsObjCMethodExpr = true; 152 } 153 } 154 155 if (Left->is(TT_OverloadedOperatorLParen)) { 156 Contexts.back().IsExpression = false; 157 } else if (Style.Language == FormatStyle::LK_JavaScript && 158 (Line.startsWith(Keywords.kw_type, tok::identifier) || 159 Line.startsWith(tok::kw_export, Keywords.kw_type, 160 tok::identifier))) { 161 // type X = (...); 162 // export type X = (...); 163 Contexts.back().IsExpression = false; 164 } else if (Left->Previous && 165 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_decltype, 166 tok::kw_if, tok::kw_while, tok::l_paren, 167 tok::comma) || 168 Left->Previous->endsSequence(tok::kw_constexpr, tok::kw_if) || 169 Left->Previous->is(TT_BinaryOperator))) { 170 // static_assert, if and while usually contain expressions. 171 Contexts.back().IsExpression = true; 172 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 173 (Left->Previous->is(Keywords.kw_function) || 174 (Left->Previous->endsSequence(tok::identifier, 175 Keywords.kw_function)))) { 176 // function(...) or function f(...) 177 Contexts.back().IsExpression = false; 178 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 179 Left->Previous->is(TT_JsTypeColon)) { 180 // let x: (SomeType); 181 Contexts.back().IsExpression = false; 182 } else if (Left->Previous && Left->Previous->is(tok::r_square) && 183 Left->Previous->MatchingParen && 184 Left->Previous->MatchingParen->is(TT_LambdaLSquare)) { 185 // This is a parameter list of a lambda expression. 186 Contexts.back().IsExpression = false; 187 } else if (Line.InPPDirective && 188 (!Left->Previous || !Left->Previous->is(tok::identifier))) { 189 Contexts.back().IsExpression = true; 190 } else if (Contexts[Contexts.size() - 2].CaretFound) { 191 // This is the parameter list of an ObjC block. 192 Contexts.back().IsExpression = false; 193 } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) { 194 Left->Type = TT_AttributeParen; 195 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) { 196 // The first argument to a foreach macro is a declaration. 197 Contexts.back().IsForEachMacro = true; 198 Contexts.back().IsExpression = false; 199 } else if (Left->Previous && Left->Previous->MatchingParen && 200 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) { 201 Contexts.back().IsExpression = false; 202 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) { 203 bool IsForOrCatch = 204 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch); 205 Contexts.back().IsExpression = !IsForOrCatch; 206 } 207 208 if (StartsObjCMethodExpr) { 209 Contexts.back().ColonIsObjCMethodExpr = true; 210 Left->Type = TT_ObjCMethodExpr; 211 } 212 213 bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression; 214 bool ProbablyFunctionType = CurrentToken->isOneOf(tok::star, tok::amp); 215 bool HasMultipleLines = false; 216 bool HasMultipleParametersOnALine = false; 217 bool MightBeObjCForRangeLoop = 218 Left->Previous && Left->Previous->is(tok::kw_for); 219 while (CurrentToken) { 220 // LookForDecls is set when "if (" has been seen. Check for 221 // 'identifier' '*' 'identifier' followed by not '=' -- this 222 // '*' has to be a binary operator but determineStarAmpUsage() will 223 // categorize it as an unary operator, so set the right type here. 224 if (LookForDecls && CurrentToken->Next) { 225 FormatToken *Prev = CurrentToken->getPreviousNonComment(); 226 if (Prev) { 227 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 228 FormatToken *Next = CurrentToken->Next; 229 if (PrevPrev && PrevPrev->is(tok::identifier) && 230 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) && 231 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) { 232 Prev->Type = TT_BinaryOperator; 233 LookForDecls = false; 234 } 235 } 236 } 237 238 if (CurrentToken->Previous->is(TT_PointerOrReference) && 239 CurrentToken->Previous->Previous->isOneOf(tok::l_paren, 240 tok::coloncolon)) 241 ProbablyFunctionType = true; 242 if (CurrentToken->is(tok::comma)) 243 MightBeFunctionType = false; 244 if (CurrentToken->Previous->is(TT_BinaryOperator)) 245 Contexts.back().IsExpression = true; 246 if (CurrentToken->is(tok::r_paren)) { 247 if (MightBeFunctionType && ProbablyFunctionType && CurrentToken->Next && 248 (CurrentToken->Next->is(tok::l_paren) || 249 (CurrentToken->Next->is(tok::l_square) && Line.MustBeDeclaration))) 250 Left->Type = TT_FunctionTypeLParen; 251 Left->MatchingParen = CurrentToken; 252 CurrentToken->MatchingParen = Left; 253 254 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) && 255 Left->Previous && Left->Previous->is(tok::l_paren)) { 256 // Detect the case where macros are used to generate lambdas or 257 // function bodies, e.g.: 258 // auto my_lambda = MARCO((Type *type, int i) { .. body .. }); 259 for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) { 260 if (Tok->is(TT_BinaryOperator) && 261 Tok->isOneOf(tok::star, tok::amp, tok::ampamp)) 262 Tok->Type = TT_PointerOrReference; 263 } 264 } 265 266 if (StartsObjCMethodExpr) { 267 CurrentToken->Type = TT_ObjCMethodExpr; 268 if (Contexts.back().FirstObjCSelectorName) { 269 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 270 Contexts.back().LongestObjCSelectorName; 271 } 272 } 273 274 if (Left->is(TT_AttributeParen)) 275 CurrentToken->Type = TT_AttributeParen; 276 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation)) 277 CurrentToken->Type = TT_JavaAnnotation; 278 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation)) 279 CurrentToken->Type = TT_LeadingJavaAnnotation; 280 281 if (!HasMultipleLines) 282 Left->PackingKind = PPK_Inconclusive; 283 else if (HasMultipleParametersOnALine) 284 Left->PackingKind = PPK_BinPacked; 285 else 286 Left->PackingKind = PPK_OnePerLine; 287 288 next(); 289 return true; 290 } 291 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace)) 292 return false; 293 294 if (CurrentToken->is(tok::l_brace)) 295 Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen 296 if (CurrentToken->is(tok::comma) && CurrentToken->Next && 297 !CurrentToken->Next->HasUnescapedNewline && 298 !CurrentToken->Next->isTrailingComment()) 299 HasMultipleParametersOnALine = true; 300 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) || 301 CurrentToken->Previous->isSimpleTypeSpecifier()) && 302 !CurrentToken->is(tok::l_brace)) 303 Contexts.back().IsExpression = false; 304 if (CurrentToken->isOneOf(tok::semi, tok::colon)) 305 MightBeObjCForRangeLoop = false; 306 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) 307 CurrentToken->Type = TT_ObjCForIn; 308 // When we discover a 'new', we set CanBeExpression to 'false' in order to 309 // parse the type correctly. Reset that after a comma. 310 if (CurrentToken->is(tok::comma)) 311 Contexts.back().CanBeExpression = true; 312 313 FormatToken *Tok = CurrentToken; 314 if (!consumeToken()) 315 return false; 316 updateParameterCount(Left, Tok); 317 if (CurrentToken && CurrentToken->HasUnescapedNewline) 318 HasMultipleLines = true; 319 } 320 return false; 321 } 322 323 bool parseSquare() { 324 if (!CurrentToken) 325 return false; 326 327 // A '[' could be an index subscript (after an identifier or after 328 // ')' or ']'), it could be the start of an Objective-C method 329 // expression, or it could the start of an Objective-C array literal. 330 FormatToken *Left = CurrentToken->Previous; 331 Left->ParentBracket = Contexts.back().ContextKind; 332 FormatToken *Parent = Left->getPreviousNonComment(); 333 334 // Cases where '>' is followed by '['. 335 // In C++, this can happen either in array of templates (foo<int>[10]) 336 // or when array is a nested template type (unique_ptr<type1<type2>[]>). 337 bool CppArrayTemplates = 338 Style.isCpp() && Parent && Parent->is(TT_TemplateCloser) && 339 (Contexts.back().CanBeExpression || Contexts.back().IsExpression || 340 Contexts.back().InTemplateArgument); 341 342 bool StartsObjCMethodExpr = 343 !CppArrayTemplates && Style.isCpp() && 344 Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) && 345 CurrentToken->isNot(tok::l_brace) && 346 (!Parent || 347 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren, 348 tok::kw_return, tok::kw_throw) || 349 Parent->isUnaryOperator() || 350 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) || 351 getBinOpPrecedence(Parent->Tok.getKind(), true, true) > prec::Unknown); 352 bool ColonFound = false; 353 354 unsigned BindingIncrease = 1; 355 if (Left->isCppStructuredBinding(Style)) { 356 Left->Type = TT_StructuredBindingLSquare; 357 } else if (Left->is(TT_Unknown)) { 358 if (StartsObjCMethodExpr) { 359 Left->Type = TT_ObjCMethodExpr; 360 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent && 361 Contexts.back().ContextKind == tok::l_brace && 362 Parent->isOneOf(tok::l_brace, tok::comma)) { 363 Left->Type = TT_JsComputedPropertyName; 364 } else if (Style.isCpp() && Contexts.back().ContextKind == tok::l_brace && 365 Parent && Parent->isOneOf(tok::l_brace, tok::comma)) { 366 Left->Type = TT_DesignatedInitializerLSquare; 367 } else if (CurrentToken->is(tok::r_square) && Parent && 368 Parent->is(TT_TemplateCloser)) { 369 Left->Type = TT_ArraySubscriptLSquare; 370 } else if (Style.Language == FormatStyle::LK_Proto || 371 (!CppArrayTemplates && Parent && 372 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at, 373 tok::comma, tok::l_paren, tok::l_square, 374 tok::question, tok::colon, tok::kw_return, 375 // Should only be relevant to JavaScript: 376 tok::kw_default))) { 377 Left->Type = TT_ArrayInitializerLSquare; 378 } else { 379 BindingIncrease = 10; 380 Left->Type = TT_ArraySubscriptLSquare; 381 } 382 } 383 384 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease); 385 Contexts.back().IsExpression = true; 386 if (Style.Language == FormatStyle::LK_JavaScript && Parent && 387 Parent->is(TT_JsTypeColon)) 388 Contexts.back().IsExpression = false; 389 390 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr; 391 392 while (CurrentToken) { 393 if (CurrentToken->is(tok::r_square)) { 394 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_paren) && 395 Left->is(TT_ObjCMethodExpr)) { 396 // An ObjC method call is rarely followed by an open parenthesis. 397 // FIXME: Do we incorrectly label ":" with this? 398 StartsObjCMethodExpr = false; 399 Left->Type = TT_Unknown; 400 } 401 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) { 402 CurrentToken->Type = TT_ObjCMethodExpr; 403 // determineStarAmpUsage() thinks that '*' '[' is allocating an 404 // array of pointers, but if '[' starts a selector then '*' is a 405 // binary operator. 406 if (Parent && Parent->is(TT_PointerOrReference)) 407 Parent->Type = TT_BinaryOperator; 408 } 409 Left->MatchingParen = CurrentToken; 410 CurrentToken->MatchingParen = Left; 411 if (Contexts.back().FirstObjCSelectorName) { 412 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 413 Contexts.back().LongestObjCSelectorName; 414 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts = 415 Left->ParameterCount; 416 if (Left->BlockParameterCount > 1) 417 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0; 418 } 419 next(); 420 return true; 421 } 422 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace)) 423 return false; 424 if (CurrentToken->is(tok::colon)) { 425 if (Left->isOneOf(TT_ArraySubscriptLSquare, 426 TT_DesignatedInitializerLSquare)) { 427 Left->Type = TT_ObjCMethodExpr; 428 StartsObjCMethodExpr = true; 429 // ParameterCount might have been set to 1 before expression was 430 // recognized as ObjCMethodExpr (as '1 + number of commas' formula is 431 // used for other expression types). Parameter counter has to be, 432 // therefore, reset to 0. 433 Left->ParameterCount = 0; 434 Contexts.back().ColonIsObjCMethodExpr = true; 435 if (Parent && Parent->is(tok::r_paren)) 436 Parent->Type = TT_CastRParen; 437 } 438 ColonFound = true; 439 } 440 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) && 441 !ColonFound) 442 Left->Type = TT_ArrayInitializerLSquare; 443 FormatToken *Tok = CurrentToken; 444 if (!consumeToken()) 445 return false; 446 updateParameterCount(Left, Tok); 447 } 448 return false; 449 } 450 451 bool parseBrace() { 452 if (CurrentToken) { 453 FormatToken *Left = CurrentToken->Previous; 454 Left->ParentBracket = Contexts.back().ContextKind; 455 456 if (Contexts.back().CaretFound) 457 Left->Type = TT_ObjCBlockLBrace; 458 Contexts.back().CaretFound = false; 459 460 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1); 461 Contexts.back().ColonIsDictLiteral = true; 462 if (Left->BlockKind == BK_BracedInit) 463 Contexts.back().IsExpression = true; 464 if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 465 Left->Previous->is(TT_JsTypeColon)) 466 Contexts.back().IsExpression = false; 467 468 while (CurrentToken) { 469 if (CurrentToken->is(tok::r_brace)) { 470 Left->MatchingParen = CurrentToken; 471 CurrentToken->MatchingParen = Left; 472 next(); 473 return true; 474 } 475 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 476 return false; 477 updateParameterCount(Left, CurrentToken); 478 if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) { 479 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 480 if (Previous->is(TT_JsTypeOptionalQuestion)) 481 Previous = Previous->getPreviousNonComment(); 482 if ((CurrentToken->is(tok::colon) && 483 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) || 484 Style.Language == FormatStyle::LK_Proto || 485 Style.Language == FormatStyle::LK_TextProto) { 486 Left->Type = TT_DictLiteral; 487 if (Previous->Tok.getIdentifierInfo() || 488 Previous->is(tok::string_literal)) 489 Previous->Type = TT_SelectorName; 490 } 491 if (CurrentToken->is(tok::colon) || 492 Style.Language == FormatStyle::LK_JavaScript) 493 Left->Type = TT_DictLiteral; 494 } 495 if (CurrentToken->is(tok::comma) && 496 Style.Language == FormatStyle::LK_JavaScript) 497 Left->Type = TT_DictLiteral; 498 if (!consumeToken()) 499 return false; 500 } 501 } 502 return true; 503 } 504 505 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 506 if (Current->is(tok::l_brace) && Current->BlockKind == BK_Block) 507 ++Left->BlockParameterCount; 508 if (Left->Type == TT_ObjCMethodExpr) { 509 if (Current->is(tok::colon)) 510 ++Left->ParameterCount; 511 } else if (Current->is(tok::comma)) { 512 ++Left->ParameterCount; 513 if (!Left->Role) 514 Left->Role.reset(new CommaSeparatedList(Style)); 515 Left->Role->CommaFound(Current); 516 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 517 Left->ParameterCount = 1; 518 } 519 } 520 521 bool parseConditional() { 522 while (CurrentToken) { 523 if (CurrentToken->is(tok::colon)) { 524 CurrentToken->Type = TT_ConditionalExpr; 525 next(); 526 return true; 527 } 528 if (!consumeToken()) 529 return false; 530 } 531 return false; 532 } 533 534 bool parseTemplateDeclaration() { 535 if (CurrentToken && CurrentToken->is(tok::less)) { 536 CurrentToken->Type = TT_TemplateOpener; 537 next(); 538 if (!parseAngle()) 539 return false; 540 if (CurrentToken) 541 CurrentToken->Previous->ClosesTemplateDeclaration = true; 542 return true; 543 } 544 return false; 545 } 546 547 bool consumeToken() { 548 FormatToken *Tok = CurrentToken; 549 next(); 550 switch (Tok->Tok.getKind()) { 551 case tok::plus: 552 case tok::minus: 553 if (!Tok->Previous && Line.MustBeDeclaration) 554 Tok->Type = TT_ObjCMethodSpecifier; 555 break; 556 case tok::colon: 557 if (!Tok->Previous) 558 return false; 559 // Colons from ?: are handled in parseConditional(). 560 if (Style.Language == FormatStyle::LK_JavaScript) { 561 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop 562 (Contexts.size() == 1 && // switch/case labels 563 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) || 564 Contexts.back().ContextKind == tok::l_paren || // function params 565 Contexts.back().ContextKind == tok::l_square || // array type 566 (!Contexts.back().IsExpression && 567 Contexts.back().ContextKind == tok::l_brace) || // object type 568 (Contexts.size() == 1 && 569 Line.MustBeDeclaration)) { // method/property declaration 570 Contexts.back().IsExpression = false; 571 Tok->Type = TT_JsTypeColon; 572 break; 573 } 574 } 575 if (Contexts.back().ColonIsDictLiteral || 576 Style.Language == FormatStyle::LK_Proto || 577 Style.Language == FormatStyle::LK_TextProto) { 578 Tok->Type = TT_DictLiteral; 579 if (Style.Language == FormatStyle::LK_TextProto) { 580 if (FormatToken *Previous = Tok->getPreviousNonComment()) 581 Previous->Type = TT_SelectorName; 582 } 583 } else if (Contexts.back().ColonIsObjCMethodExpr || 584 Line.startsWith(TT_ObjCMethodSpecifier)) { 585 Tok->Type = TT_ObjCMethodExpr; 586 const FormatToken *BeforePrevious = Tok->Previous->Previous; 587 if (!BeforePrevious || 588 !(BeforePrevious->is(TT_CastRParen) || 589 (BeforePrevious->is(TT_ObjCMethodExpr) && 590 BeforePrevious->is(tok::colon))) || 591 BeforePrevious->is(tok::r_square) || 592 Contexts.back().LongestObjCSelectorName == 0) { 593 Tok->Previous->Type = TT_SelectorName; 594 if (!Contexts.back().FirstObjCSelectorName) 595 Contexts.back().FirstObjCSelectorName = Tok->Previous; 596 else if (Tok->Previous->ColumnWidth > 597 Contexts.back().LongestObjCSelectorName) 598 Contexts.back().LongestObjCSelectorName = 599 Tok->Previous->ColumnWidth; 600 } 601 } else if (Contexts.back().ColonIsForRangeExpr) { 602 Tok->Type = TT_RangeBasedForLoopColon; 603 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 604 Tok->Type = TT_BitFieldColon; 605 } else if (Contexts.size() == 1 && 606 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) { 607 if (Tok->getPreviousNonComment()->isOneOf(tok::r_paren, 608 tok::kw_noexcept)) 609 Tok->Type = TT_CtorInitializerColon; 610 else 611 Tok->Type = TT_InheritanceColon; 612 } else if (Tok->Previous->is(tok::identifier) && Tok->Next && 613 Tok->Next->isOneOf(tok::r_paren, tok::comma)) { 614 // This handles a special macro in ObjC code where selectors including 615 // the colon are passed as macro arguments. 616 Tok->Type = TT_ObjCMethodExpr; 617 } else if (Contexts.back().ContextKind == tok::l_paren) { 618 Tok->Type = TT_InlineASMColon; 619 } 620 break; 621 case tok::pipe: 622 case tok::amp: 623 // | and & in declarations/type expressions represent union and 624 // intersection types, respectively. 625 if (Style.Language == FormatStyle::LK_JavaScript && 626 !Contexts.back().IsExpression) 627 Tok->Type = TT_JsTypeOperator; 628 break; 629 case tok::kw_if: 630 case tok::kw_while: 631 if (Tok->is(tok::kw_if) && CurrentToken && 632 CurrentToken->is(tok::kw_constexpr)) 633 next(); 634 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 635 next(); 636 if (!parseParens(/*LookForDecls=*/true)) 637 return false; 638 } 639 break; 640 case tok::kw_for: 641 if (Style.Language == FormatStyle::LK_JavaScript) { 642 // x.for and {for: ...} 643 if ((Tok->Previous && Tok->Previous->is(tok::period)) || 644 (Tok->Next && Tok->Next->is(tok::colon))) 645 break; 646 // JS' for await ( ... 647 if (CurrentToken && CurrentToken->is(Keywords.kw_await)) 648 next(); 649 } 650 Contexts.back().ColonIsForRangeExpr = true; 651 next(); 652 if (!parseParens()) 653 return false; 654 break; 655 case tok::l_paren: 656 // When faced with 'operator()()', the kw_operator handler incorrectly 657 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make 658 // the first two parens OverloadedOperators and the second l_paren an 659 // OverloadedOperatorLParen. 660 if (Tok->Previous && Tok->Previous->is(tok::r_paren) && 661 Tok->Previous->MatchingParen && 662 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) { 663 Tok->Previous->Type = TT_OverloadedOperator; 664 Tok->Previous->MatchingParen->Type = TT_OverloadedOperator; 665 Tok->Type = TT_OverloadedOperatorLParen; 666 } 667 668 if (!parseParens()) 669 return false; 670 if (Line.MustBeDeclaration && Contexts.size() == 1 && 671 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) && 672 (!Tok->Previous || 673 !Tok->Previous->isOneOf(tok::kw_decltype, tok::kw___attribute, 674 TT_LeadingJavaAnnotation))) 675 Line.MightBeFunctionDecl = true; 676 break; 677 case tok::l_square: 678 if (!parseSquare()) 679 return false; 680 break; 681 case tok::l_brace: 682 if (Style.Language == FormatStyle::LK_TextProto) { 683 FormatToken *Previous = Tok->getPreviousNonComment(); 684 if (Previous && Previous->Type != TT_DictLiteral) 685 Previous->Type = TT_SelectorName; 686 } 687 if (!parseBrace()) 688 return false; 689 break; 690 case tok::less: 691 if (parseAngle()) { 692 Tok->Type = TT_TemplateOpener; 693 // In TT_Proto, we must distignuish between: 694 // map<key, value> 695 // msg < item: data > 696 // msg: < item: data > 697 // In TT_TextProto, map<key, value> does not occur. 698 if (Style.Language == FormatStyle::LK_TextProto || 699 (Style.Language == FormatStyle::LK_Proto && Tok->Previous && 700 Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) { 701 Tok->Type = TT_DictLiteral; 702 FormatToken *Previous = Tok->getPreviousNonComment(); 703 if (Previous && Previous->Type != TT_DictLiteral) 704 Previous->Type = TT_SelectorName; 705 } 706 } else { 707 Tok->Type = TT_BinaryOperator; 708 NonTemplateLess.insert(Tok); 709 CurrentToken = Tok; 710 next(); 711 } 712 break; 713 case tok::r_paren: 714 case tok::r_square: 715 return false; 716 case tok::r_brace: 717 // Lines can start with '}'. 718 if (Tok->Previous) 719 return false; 720 break; 721 case tok::greater: 722 if (Style.Language != FormatStyle::LK_TextProto) 723 Tok->Type = TT_BinaryOperator; 724 break; 725 case tok::kw_operator: 726 while (CurrentToken && 727 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 728 if (CurrentToken->isOneOf(tok::star, tok::amp)) 729 CurrentToken->Type = TT_PointerOrReference; 730 consumeToken(); 731 if (CurrentToken && 732 CurrentToken->Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator, 733 tok::comma)) 734 CurrentToken->Previous->Type = TT_OverloadedOperator; 735 } 736 if (CurrentToken) { 737 CurrentToken->Type = TT_OverloadedOperatorLParen; 738 if (CurrentToken->Previous->is(TT_BinaryOperator)) 739 CurrentToken->Previous->Type = TT_OverloadedOperator; 740 } 741 break; 742 case tok::question: 743 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next && 744 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren, 745 tok::r_brace)) { 746 // Question marks before semicolons, colons, etc. indicate optional 747 // types (fields, parameters), e.g. 748 // function(x?: string, y?) {...} 749 // class X { y?; } 750 Tok->Type = TT_JsTypeOptionalQuestion; 751 break; 752 } 753 // Declarations cannot be conditional expressions, this can only be part 754 // of a type declaration. 755 if (Line.MustBeDeclaration && !Contexts.back().IsExpression && 756 Style.Language == FormatStyle::LK_JavaScript) 757 break; 758 parseConditional(); 759 break; 760 case tok::kw_template: 761 parseTemplateDeclaration(); 762 break; 763 case tok::comma: 764 if (Contexts.back().InCtorInitializer) 765 Tok->Type = TT_CtorInitializerComma; 766 else if (Contexts.back().InInheritanceList) 767 Tok->Type = TT_InheritanceComma; 768 else if (Contexts.back().FirstStartOfName && 769 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) { 770 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 771 Line.IsMultiVariableDeclStmt = true; 772 } 773 if (Contexts.back().IsForEachMacro) 774 Contexts.back().IsExpression = true; 775 break; 776 case tok::identifier: 777 if (Tok->isOneOf(Keywords.kw___has_include, 778 Keywords.kw___has_include_next)) { 779 parseHasInclude(); 780 } 781 break; 782 default: 783 break; 784 } 785 return true; 786 } 787 788 void parseIncludeDirective() { 789 if (CurrentToken && CurrentToken->is(tok::less)) { 790 next(); 791 while (CurrentToken) { 792 // Mark tokens up to the trailing line comments as implicit string 793 // literals. 794 if (CurrentToken->isNot(tok::comment) && 795 !CurrentToken->TokenText.startswith("//")) 796 CurrentToken->Type = TT_ImplicitStringLiteral; 797 next(); 798 } 799 } 800 } 801 802 void parseWarningOrError() { 803 next(); 804 // We still want to format the whitespace left of the first token of the 805 // warning or error. 806 next(); 807 while (CurrentToken) { 808 CurrentToken->Type = TT_ImplicitStringLiteral; 809 next(); 810 } 811 } 812 813 void parsePragma() { 814 next(); // Consume "pragma". 815 if (CurrentToken && 816 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 817 bool IsMark = CurrentToken->is(Keywords.kw_mark); 818 next(); // Consume "mark". 819 next(); // Consume first token (so we fix leading whitespace). 820 while (CurrentToken) { 821 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 822 CurrentToken->Type = TT_ImplicitStringLiteral; 823 next(); 824 } 825 } 826 } 827 828 void parseHasInclude() { 829 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 830 return; 831 next(); // '(' 832 parseIncludeDirective(); 833 next(); // ')' 834 } 835 836 LineType parsePreprocessorDirective() { 837 bool IsFirstToken = CurrentToken->IsFirst; 838 LineType Type = LT_PreprocessorDirective; 839 next(); 840 if (!CurrentToken) 841 return Type; 842 843 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) { 844 // JavaScript files can contain shebang lines of the form: 845 // #!/usr/bin/env node 846 // Treat these like C++ #include directives. 847 while (CurrentToken) { 848 // Tokens cannot be comments here. 849 CurrentToken->Type = TT_ImplicitStringLiteral; 850 next(); 851 } 852 return LT_ImportStatement; 853 } 854 855 if (CurrentToken->Tok.is(tok::numeric_constant)) { 856 CurrentToken->SpacesRequiredBefore = 1; 857 return Type; 858 } 859 // Hashes in the middle of a line can lead to any strange token 860 // sequence. 861 if (!CurrentToken->Tok.getIdentifierInfo()) 862 return Type; 863 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 864 case tok::pp_include: 865 case tok::pp_include_next: 866 case tok::pp_import: 867 next(); 868 parseIncludeDirective(); 869 Type = LT_ImportStatement; 870 break; 871 case tok::pp_error: 872 case tok::pp_warning: 873 parseWarningOrError(); 874 break; 875 case tok::pp_pragma: 876 parsePragma(); 877 break; 878 case tok::pp_if: 879 case tok::pp_elif: 880 Contexts.back().IsExpression = true; 881 parseLine(); 882 break; 883 default: 884 break; 885 } 886 while (CurrentToken) { 887 FormatToken *Tok = CurrentToken; 888 next(); 889 if (Tok->is(tok::l_paren)) 890 parseParens(); 891 else if (Tok->isOneOf(Keywords.kw___has_include, 892 Keywords.kw___has_include_next)) 893 parseHasInclude(); 894 } 895 return Type; 896 } 897 898 public: 899 LineType parseLine() { 900 NonTemplateLess.clear(); 901 if (CurrentToken->is(tok::hash)) 902 return parsePreprocessorDirective(); 903 904 // Directly allow to 'import <string-literal>' to support protocol buffer 905 // definitions (github.com/google/protobuf) or missing "#" (either way we 906 // should not break the line). 907 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 908 if ((Style.Language == FormatStyle::LK_Java && 909 CurrentToken->is(Keywords.kw_package)) || 910 (Info && Info->getPPKeywordID() == tok::pp_import && 911 CurrentToken->Next && 912 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 913 tok::kw_static))) { 914 next(); 915 parseIncludeDirective(); 916 return LT_ImportStatement; 917 } 918 919 // If this line starts and ends in '<' and '>', respectively, it is likely 920 // part of "#define <a/b.h>". 921 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 922 parseIncludeDirective(); 923 return LT_ImportStatement; 924 } 925 926 // In .proto files, top-level options are very similar to import statements 927 // and should not be line-wrapped. 928 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 929 CurrentToken->is(Keywords.kw_option)) { 930 next(); 931 if (CurrentToken && CurrentToken->is(tok::identifier)) 932 return LT_ImportStatement; 933 } 934 935 bool KeywordVirtualFound = false; 936 bool ImportStatement = false; 937 938 // import {...} from '...'; 939 if (Style.Language == FormatStyle::LK_JavaScript && 940 CurrentToken->is(Keywords.kw_import)) 941 ImportStatement = true; 942 943 while (CurrentToken) { 944 if (CurrentToken->is(tok::kw_virtual)) 945 KeywordVirtualFound = true; 946 if (Style.Language == FormatStyle::LK_JavaScript) { 947 // export {...} from '...'; 948 // An export followed by "from 'some string';" is a re-export from 949 // another module identified by a URI and is treated as a 950 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 951 // Just "export {...};" or "export class ..." should not be treated as 952 // an import in this sense. 953 if (Line.First->is(tok::kw_export) && 954 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 955 CurrentToken->Next->isStringLiteral()) 956 ImportStatement = true; 957 if (isClosureImportStatement(*CurrentToken)) 958 ImportStatement = true; 959 } 960 if (!consumeToken()) 961 return LT_Invalid; 962 } 963 if (KeywordVirtualFound) 964 return LT_VirtualFunctionDecl; 965 if (ImportStatement) 966 return LT_ImportStatement; 967 968 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 969 if (Contexts.back().FirstObjCSelectorName) 970 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 971 Contexts.back().LongestObjCSelectorName; 972 return LT_ObjCMethodDecl; 973 } 974 975 return LT_Other; 976 } 977 978 private: 979 bool isClosureImportStatement(const FormatToken &Tok) { 980 // FIXME: Closure-library specific stuff should not be hard-coded but be 981 // configurable. 982 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 983 Tok.Next->Next && 984 (Tok.Next->Next->TokenText == "module" || 985 Tok.Next->Next->TokenText == "provide" || 986 Tok.Next->Next->TokenText == "require" || 987 Tok.Next->Next->TokenText == "forwardDeclare") && 988 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 989 } 990 991 void resetTokenMetadata(FormatToken *Token) { 992 if (!Token) 993 return; 994 995 // Reset token type in case we have already looked at it and then 996 // recovered from an error (e.g. failure to find the matching >). 997 if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_ForEachMacro, 998 TT_FunctionLBrace, TT_ImplicitStringLiteral, 999 TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow, 1000 TT_OverloadedOperator, TT_RegexLiteral, 1001 TT_TemplateString, TT_ObjCStringLiteral)) 1002 CurrentToken->Type = TT_Unknown; 1003 CurrentToken->Role.reset(); 1004 CurrentToken->MatchingParen = nullptr; 1005 CurrentToken->FakeLParens.clear(); 1006 CurrentToken->FakeRParens = 0; 1007 } 1008 1009 void next() { 1010 if (CurrentToken) { 1011 CurrentToken->NestingLevel = Contexts.size() - 1; 1012 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1013 modifyContext(*CurrentToken); 1014 determineTokenType(*CurrentToken); 1015 CurrentToken = CurrentToken->Next; 1016 } 1017 1018 resetTokenMetadata(CurrentToken); 1019 } 1020 1021 /// \brief A struct to hold information valid in a specific context, e.g. 1022 /// a pair of parenthesis. 1023 struct Context { 1024 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1025 bool IsExpression) 1026 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1027 IsExpression(IsExpression) {} 1028 1029 tok::TokenKind ContextKind; 1030 unsigned BindingStrength; 1031 bool IsExpression; 1032 unsigned LongestObjCSelectorName = 0; 1033 bool ColonIsForRangeExpr = false; 1034 bool ColonIsDictLiteral = false; 1035 bool ColonIsObjCMethodExpr = false; 1036 FormatToken *FirstObjCSelectorName = nullptr; 1037 FormatToken *FirstStartOfName = nullptr; 1038 bool CanBeExpression = true; 1039 bool InTemplateArgument = false; 1040 bool InCtorInitializer = false; 1041 bool InInheritanceList = false; 1042 bool CaretFound = false; 1043 bool IsForEachMacro = false; 1044 }; 1045 1046 /// \brief Puts a new \c Context onto the stack \c Contexts for the lifetime 1047 /// of each instance. 1048 struct ScopedContextCreator { 1049 AnnotatingParser &P; 1050 1051 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1052 unsigned Increase) 1053 : P(P) { 1054 P.Contexts.push_back(Context(ContextKind, 1055 P.Contexts.back().BindingStrength + Increase, 1056 P.Contexts.back().IsExpression)); 1057 } 1058 1059 ~ScopedContextCreator() { P.Contexts.pop_back(); } 1060 }; 1061 1062 void modifyContext(const FormatToken &Current) { 1063 if (Current.getPrecedence() == prec::Assignment && 1064 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 1065 // Type aliases use `type X = ...;` in TypeScript and can be exported 1066 // using `export type ...`. 1067 !(Style.Language == FormatStyle::LK_JavaScript && 1068 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1069 Line.startsWith(tok::kw_export, Keywords.kw_type, 1070 tok::identifier))) && 1071 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 1072 Contexts.back().IsExpression = true; 1073 if (!Line.startsWith(TT_UnaryOperator)) { 1074 for (FormatToken *Previous = Current.Previous; 1075 Previous && Previous->Previous && 1076 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1077 Previous = Previous->Previous) { 1078 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1079 Previous = Previous->MatchingParen; 1080 if (!Previous) 1081 break; 1082 } 1083 if (Previous->opensScope()) 1084 break; 1085 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1086 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1087 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1088 Previous->Type = TT_PointerOrReference; 1089 } 1090 } 1091 } else if (Current.is(tok::lessless) && 1092 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1093 Contexts.back().IsExpression = true; 1094 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1095 Contexts.back().IsExpression = true; 1096 } else if (Current.is(TT_TrailingReturnArrow)) { 1097 Contexts.back().IsExpression = false; 1098 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1099 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1100 } else if (Current.Previous && 1101 Current.Previous->is(TT_CtorInitializerColon)) { 1102 Contexts.back().IsExpression = true; 1103 Contexts.back().InCtorInitializer = true; 1104 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1105 Contexts.back().InInheritanceList = true; 1106 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1107 for (FormatToken *Previous = Current.Previous; 1108 Previous && Previous->isOneOf(tok::star, tok::amp); 1109 Previous = Previous->Previous) 1110 Previous->Type = TT_PointerOrReference; 1111 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 1112 Contexts.back().IsExpression = false; 1113 } else if (Current.is(tok::kw_new)) { 1114 Contexts.back().CanBeExpression = false; 1115 } else if (Current.isOneOf(tok::semi, tok::exclaim)) { 1116 // This should be the condition or increment in a for-loop. 1117 Contexts.back().IsExpression = true; 1118 } 1119 } 1120 1121 void determineTokenType(FormatToken &Current) { 1122 if (!Current.is(TT_Unknown)) 1123 // The token type is already known. 1124 return; 1125 1126 if (Style.Language == FormatStyle::LK_JavaScript) { 1127 if (Current.is(tok::exclaim)) { 1128 if (Current.Previous && 1129 (Current.Previous->isOneOf(tok::identifier, tok::kw_namespace, 1130 tok::r_paren, tok::r_square, 1131 tok::r_brace) || 1132 Current.Previous->Tok.isLiteral())) { 1133 Current.Type = TT_JsNonNullAssertion; 1134 return; 1135 } 1136 if (Current.Next && 1137 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1138 Current.Type = TT_JsNonNullAssertion; 1139 return; 1140 } 1141 } 1142 } 1143 1144 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1145 // function declaration have been found. In this case, 'Current' is a 1146 // trailing token of this declaration and thus cannot be a name. 1147 if (Current.is(Keywords.kw_instanceof)) { 1148 Current.Type = TT_BinaryOperator; 1149 } else if (isStartOfName(Current) && 1150 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1151 Contexts.back().FirstStartOfName = &Current; 1152 Current.Type = TT_StartOfName; 1153 } else if (Current.is(tok::semi)) { 1154 // Reset FirstStartOfName after finding a semicolon so that a for loop 1155 // with multiple increment statements is not confused with a for loop 1156 // having multiple variable declarations. 1157 Contexts.back().FirstStartOfName = nullptr; 1158 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1159 AutoFound = true; 1160 } else if (Current.is(tok::arrow) && 1161 Style.Language == FormatStyle::LK_Java) { 1162 Current.Type = TT_LambdaArrow; 1163 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1164 Current.NestingLevel == 0) { 1165 Current.Type = TT_TrailingReturnArrow; 1166 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1167 Current.Type = determineStarAmpUsage(Current, 1168 Contexts.back().CanBeExpression && 1169 Contexts.back().IsExpression, 1170 Contexts.back().InTemplateArgument); 1171 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1172 Current.Type = determinePlusMinusCaretUsage(Current); 1173 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1174 Contexts.back().CaretFound = true; 1175 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1176 Current.Type = determineIncrementUsage(Current); 1177 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1178 Current.Type = TT_UnaryOperator; 1179 } else if (Current.is(tok::question)) { 1180 if (Style.Language == FormatStyle::LK_JavaScript && 1181 Line.MustBeDeclaration && !Contexts.back().IsExpression) { 1182 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1183 // on the interface, not a ternary expression. 1184 Current.Type = TT_JsTypeOptionalQuestion; 1185 } else { 1186 Current.Type = TT_ConditionalExpr; 1187 } 1188 } else if (Current.isBinaryOperator() && 1189 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1190 (!Current.is(tok::greater) && 1191 Style.Language != FormatStyle::LK_TextProto)) { 1192 Current.Type = TT_BinaryOperator; 1193 } else if (Current.is(tok::comment)) { 1194 if (Current.TokenText.startswith("/*")) { 1195 if (Current.TokenText.endswith("*/")) 1196 Current.Type = TT_BlockComment; 1197 else 1198 // The lexer has for some reason determined a comment here. But we 1199 // cannot really handle it, if it isn't properly terminated. 1200 Current.Tok.setKind(tok::unknown); 1201 } else { 1202 Current.Type = TT_LineComment; 1203 } 1204 } else if (Current.is(tok::r_paren)) { 1205 if (rParenEndsCast(Current)) 1206 Current.Type = TT_CastRParen; 1207 if (Current.MatchingParen && Current.Next && 1208 !Current.Next->isBinaryOperator() && 1209 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1210 tok::comma, tok::period, tok::arrow, 1211 tok::coloncolon)) 1212 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1213 // Make sure this isn't the return type of an Obj-C block declaration 1214 if (AfterParen->Tok.isNot(tok::caret)) { 1215 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1216 if (BeforeParen->is(tok::identifier) && 1217 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1218 (!BeforeParen->Previous || 1219 BeforeParen->Previous->ClosesTemplateDeclaration)) 1220 Current.Type = TT_FunctionAnnotationRParen; 1221 } 1222 } 1223 } else if (Current.is(tok::at) && Current.Next && 1224 Style.Language != FormatStyle::LK_JavaScript && 1225 Style.Language != FormatStyle::LK_Java) { 1226 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1227 // marks declarations and properties that need special formatting. 1228 switch (Current.Next->Tok.getObjCKeywordID()) { 1229 case tok::objc_interface: 1230 case tok::objc_implementation: 1231 case tok::objc_protocol: 1232 Current.Type = TT_ObjCDecl; 1233 break; 1234 case tok::objc_property: 1235 Current.Type = TT_ObjCProperty; 1236 break; 1237 default: 1238 break; 1239 } 1240 } else if (Current.is(tok::period)) { 1241 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1242 if (PreviousNoComment && 1243 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1244 Current.Type = TT_DesignatedInitializerPeriod; 1245 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1246 Current.Previous->isOneOf(TT_JavaAnnotation, 1247 TT_LeadingJavaAnnotation)) { 1248 Current.Type = Current.Previous->Type; 1249 } 1250 } else if (Current.isOneOf(tok::identifier, tok::kw_const) && 1251 Current.Previous && 1252 !Current.Previous->isOneOf(tok::equal, tok::at) && 1253 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1254 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1255 // function declaration have been found. 1256 Current.Type = TT_TrailingAnnotation; 1257 } else if ((Style.Language == FormatStyle::LK_Java || 1258 Style.Language == FormatStyle::LK_JavaScript) && 1259 Current.Previous) { 1260 if (Current.Previous->is(tok::at) && 1261 Current.isNot(Keywords.kw_interface)) { 1262 const FormatToken &AtToken = *Current.Previous; 1263 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1264 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1265 Current.Type = TT_LeadingJavaAnnotation; 1266 else 1267 Current.Type = TT_JavaAnnotation; 1268 } else if (Current.Previous->is(tok::period) && 1269 Current.Previous->isOneOf(TT_JavaAnnotation, 1270 TT_LeadingJavaAnnotation)) { 1271 Current.Type = Current.Previous->Type; 1272 } 1273 } 1274 } 1275 1276 /// \brief Take a guess at whether \p Tok starts a name of a function or 1277 /// variable declaration. 1278 /// 1279 /// This is a heuristic based on whether \p Tok is an identifier following 1280 /// something that is likely a type. 1281 bool isStartOfName(const FormatToken &Tok) { 1282 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1283 return false; 1284 1285 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1286 Keywords.kw_as)) 1287 return false; 1288 if (Style.Language == FormatStyle::LK_JavaScript && 1289 Tok.Previous->is(Keywords.kw_in)) 1290 return false; 1291 1292 // Skip "const" as it does not have an influence on whether this is a name. 1293 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1294 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1295 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1296 1297 if (!PreviousNotConst) 1298 return false; 1299 1300 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1301 PreviousNotConst->Previous && 1302 PreviousNotConst->Previous->is(tok::hash); 1303 1304 if (PreviousNotConst->is(TT_TemplateCloser)) 1305 return PreviousNotConst && PreviousNotConst->MatchingParen && 1306 PreviousNotConst->MatchingParen->Previous && 1307 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1308 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1309 1310 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen && 1311 PreviousNotConst->MatchingParen->Previous && 1312 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype)) 1313 return true; 1314 1315 return (!IsPPKeyword && 1316 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) || 1317 PreviousNotConst->is(TT_PointerOrReference) || 1318 PreviousNotConst->isSimpleTypeSpecifier(); 1319 } 1320 1321 /// \brief Determine whether ')' is ending a cast. 1322 bool rParenEndsCast(const FormatToken &Tok) { 1323 // C-style casts are only used in C++ and Java. 1324 if (!Style.isCpp() && Style.Language != FormatStyle::LK_Java) 1325 return false; 1326 1327 // Empty parens aren't casts and there are no casts at the end of the line. 1328 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1329 return false; 1330 1331 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1332 if (LeftOfParens) { 1333 // If there is a closing parenthesis left of the current parentheses, 1334 // look past it as these might be chained casts. 1335 if (LeftOfParens->is(tok::r_paren)) { 1336 if (!LeftOfParens->MatchingParen || 1337 !LeftOfParens->MatchingParen->Previous) 1338 return false; 1339 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1340 } 1341 1342 // If there is an identifier (or with a few exceptions a keyword) right 1343 // before the parentheses, this is unlikely to be a cast. 1344 if (LeftOfParens->Tok.getIdentifierInfo() && 1345 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1346 tok::kw_delete)) 1347 return false; 1348 1349 // Certain other tokens right before the parentheses are also signals that 1350 // this cannot be a cast. 1351 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1352 TT_TemplateCloser, tok::ellipsis)) 1353 return false; 1354 } 1355 1356 if (Tok.Next->is(tok::question)) 1357 return false; 1358 1359 // As Java has no function types, a "(" after the ")" likely means that this 1360 // is a cast. 1361 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1362 return true; 1363 1364 // If a (non-string) literal follows, this is likely a cast. 1365 if (Tok.Next->isNot(tok::string_literal) && 1366 (Tok.Next->Tok.isLiteral() || 1367 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1368 return true; 1369 1370 // Heuristically try to determine whether the parentheses contain a type. 1371 bool ParensAreType = 1372 !Tok.Previous || 1373 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) || 1374 Tok.Previous->isSimpleTypeSpecifier(); 1375 bool ParensCouldEndDecl = 1376 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1377 if (ParensAreType && !ParensCouldEndDecl) 1378 return true; 1379 1380 // At this point, we heuristically assume that there are no casts at the 1381 // start of the line. We assume that we have found most cases where there 1382 // are by the logic above, e.g. "(void)x;". 1383 if (!LeftOfParens) 1384 return false; 1385 1386 // Certain token types inside the parentheses mean that this can't be a 1387 // cast. 1388 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 1389 Token = Token->Next) 1390 if (Token->is(TT_BinaryOperator)) 1391 return false; 1392 1393 // If the following token is an identifier or 'this', this is a cast. All 1394 // cases where this can be something else are handled above. 1395 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1396 return true; 1397 1398 if (!Tok.Next->Next) 1399 return false; 1400 1401 // If the next token after the parenthesis is a unary operator, assume 1402 // that this is cast, unless there are unexpected tokens inside the 1403 // parenthesis. 1404 bool NextIsUnary = 1405 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 1406 if (!NextIsUnary || Tok.Next->is(tok::plus) || 1407 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 1408 return false; 1409 // Search for unexpected tokens. 1410 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 1411 Prev = Prev->Previous) { 1412 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 1413 return false; 1414 } 1415 return true; 1416 } 1417 1418 /// \brief Return the type of the given token assuming it is * or &. 1419 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 1420 bool InTemplateArgument) { 1421 if (Style.Language == FormatStyle::LK_JavaScript) 1422 return TT_BinaryOperator; 1423 1424 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1425 if (!PrevToken) 1426 return TT_UnaryOperator; 1427 1428 const FormatToken *NextToken = Tok.getNextNonComment(); 1429 if (!NextToken || 1430 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_const) || 1431 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 1432 return TT_PointerOrReference; 1433 1434 if (PrevToken->is(tok::coloncolon)) 1435 return TT_PointerOrReference; 1436 1437 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 1438 tok::comma, tok::semi, tok::kw_return, tok::colon, 1439 tok::equal, tok::kw_delete, tok::kw_sizeof, 1440 tok::kw_throw) || 1441 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 1442 TT_UnaryOperator, TT_CastRParen)) 1443 return TT_UnaryOperator; 1444 1445 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 1446 return TT_PointerOrReference; 1447 if (NextToken->is(tok::kw_operator) && !IsExpression) 1448 return TT_PointerOrReference; 1449 if (NextToken->isOneOf(tok::comma, tok::semi)) 1450 return TT_PointerOrReference; 1451 1452 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen) { 1453 FormatToken *TokenBeforeMatchingParen = 1454 PrevToken->MatchingParen->getPreviousNonComment(); 1455 if (TokenBeforeMatchingParen && 1456 TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype)) 1457 return TT_PointerOrReference; 1458 } 1459 1460 if (PrevToken->Tok.isLiteral() || 1461 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 1462 tok::kw_false, tok::r_brace) || 1463 NextToken->Tok.isLiteral() || 1464 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 1465 NextToken->isUnaryOperator() || 1466 // If we know we're in a template argument, there are no named 1467 // declarations. Thus, having an identifier on the right-hand side 1468 // indicates a binary operator. 1469 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 1470 return TT_BinaryOperator; 1471 1472 // "&&(" is quite unlikely to be two successive unary "&". 1473 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren)) 1474 return TT_BinaryOperator; 1475 1476 // This catches some cases where evaluation order is used as control flow: 1477 // aaa && aaa->f(); 1478 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 1479 if (NextNextToken && NextNextToken->is(tok::arrow)) 1480 return TT_BinaryOperator; 1481 1482 // It is very unlikely that we are going to find a pointer or reference type 1483 // definition on the RHS of an assignment. 1484 if (IsExpression && !Contexts.back().CaretFound) 1485 return TT_BinaryOperator; 1486 1487 return TT_PointerOrReference; 1488 } 1489 1490 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 1491 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1492 if (!PrevToken) 1493 return TT_UnaryOperator; 1494 1495 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator) && 1496 !PrevToken->is(tok::exclaim)) 1497 // There aren't any trailing unary operators except for TypeScript's 1498 // non-null operator (!). Thus, this must be squence of leading operators. 1499 return TT_UnaryOperator; 1500 1501 // Use heuristics to recognize unary operators. 1502 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 1503 tok::question, tok::colon, tok::kw_return, 1504 tok::kw_case, tok::at, tok::l_brace)) 1505 return TT_UnaryOperator; 1506 1507 // There can't be two consecutive binary operators. 1508 if (PrevToken->is(TT_BinaryOperator)) 1509 return TT_UnaryOperator; 1510 1511 // Fall back to marking the token as binary operator. 1512 return TT_BinaryOperator; 1513 } 1514 1515 /// \brief Determine whether ++/-- are pre- or post-increments/-decrements. 1516 TokenType determineIncrementUsage(const FormatToken &Tok) { 1517 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1518 if (!PrevToken || PrevToken->is(TT_CastRParen)) 1519 return TT_UnaryOperator; 1520 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 1521 return TT_TrailingUnaryOperator; 1522 1523 return TT_UnaryOperator; 1524 } 1525 1526 SmallVector<Context, 8> Contexts; 1527 1528 const FormatStyle &Style; 1529 AnnotatedLine &Line; 1530 FormatToken *CurrentToken; 1531 bool AutoFound; 1532 const AdditionalKeywords &Keywords; 1533 1534 // Set of "<" tokens that do not open a template parameter list. If parseAngle 1535 // determines that a specific token can't be a template opener, it will make 1536 // same decision irrespective of the decisions for tokens leading up to it. 1537 // Store this information to prevent this from causing exponential runtime. 1538 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 1539 }; 1540 1541 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 1542 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 1543 1544 /// \brief Parses binary expressions by inserting fake parenthesis based on 1545 /// operator precedence. 1546 class ExpressionParser { 1547 public: 1548 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 1549 AnnotatedLine &Line) 1550 : Style(Style), Keywords(Keywords), Current(Line.First) {} 1551 1552 /// \brief Parse expressions with the given operator precedence. 1553 void parse(int Precedence = 0) { 1554 // Skip 'return' and ObjC selector colons as they are not part of a binary 1555 // expression. 1556 while (Current && (Current->is(tok::kw_return) || 1557 (Current->is(tok::colon) && 1558 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 1559 next(); 1560 1561 if (!Current || Precedence > PrecedenceArrowAndPeriod) 1562 return; 1563 1564 // Conditional expressions need to be parsed separately for proper nesting. 1565 if (Precedence == prec::Conditional) { 1566 parseConditionalExpr(); 1567 return; 1568 } 1569 1570 // Parse unary operators, which all have a higher precedence than binary 1571 // operators. 1572 if (Precedence == PrecedenceUnaryOperator) { 1573 parseUnaryOperator(); 1574 return; 1575 } 1576 1577 FormatToken *Start = Current; 1578 FormatToken *LatestOperator = nullptr; 1579 unsigned OperatorIndex = 0; 1580 1581 while (Current) { 1582 // Consume operators with higher precedence. 1583 parse(Precedence + 1); 1584 1585 int CurrentPrecedence = getCurrentPrecedence(); 1586 1587 if (Current && Current->is(TT_SelectorName) && 1588 Precedence == CurrentPrecedence) { 1589 if (LatestOperator) 1590 addFakeParenthesis(Start, prec::Level(Precedence)); 1591 Start = Current; 1592 } 1593 1594 // At the end of the line or when an operator with higher precedence is 1595 // found, insert fake parenthesis and return. 1596 if (!Current || 1597 (Current->closesScope() && 1598 (Current->MatchingParen || Current->is(TT_TemplateString))) || 1599 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 1600 (CurrentPrecedence == prec::Conditional && 1601 Precedence == prec::Assignment && Current->is(tok::colon))) { 1602 break; 1603 } 1604 1605 // Consume scopes: (), [], <> and {} 1606 if (Current->opensScope()) { 1607 // In fragment of a JavaScript template string can look like '}..${' and 1608 // thus close a scope and open a new one at the same time. 1609 while (Current && (!Current->closesScope() || Current->opensScope())) { 1610 next(); 1611 parse(); 1612 } 1613 next(); 1614 } else { 1615 // Operator found. 1616 if (CurrentPrecedence == Precedence) { 1617 if (LatestOperator) 1618 LatestOperator->NextOperator = Current; 1619 LatestOperator = Current; 1620 Current->OperatorIndex = OperatorIndex; 1621 ++OperatorIndex; 1622 } 1623 next(/*SkipPastLeadingComments=*/Precedence > 0); 1624 } 1625 } 1626 1627 if (LatestOperator && (Current || Precedence > 0)) { 1628 // LatestOperator->LastOperator = true; 1629 if (Precedence == PrecedenceArrowAndPeriod) { 1630 // Call expressions don't have a binary operator precedence. 1631 addFakeParenthesis(Start, prec::Unknown); 1632 } else { 1633 addFakeParenthesis(Start, prec::Level(Precedence)); 1634 } 1635 } 1636 } 1637 1638 private: 1639 /// \brief Gets the precedence (+1) of the given token for binary operators 1640 /// and other tokens that we treat like binary operators. 1641 int getCurrentPrecedence() { 1642 if (Current) { 1643 const FormatToken *NextNonComment = Current->getNextNonComment(); 1644 if (Current->is(TT_ConditionalExpr)) 1645 return prec::Conditional; 1646 if (NextNonComment && Current->is(TT_SelectorName) && 1647 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 1648 ((Style.Language == FormatStyle::LK_Proto || 1649 Style.Language == FormatStyle::LK_TextProto) && 1650 NextNonComment->is(tok::less)))) 1651 return prec::Assignment; 1652 if (Current->is(TT_JsComputedPropertyName)) 1653 return prec::Assignment; 1654 if (Current->is(TT_LambdaArrow)) 1655 return prec::Comma; 1656 if (Current->is(TT_JsFatArrow)) 1657 return prec::Assignment; 1658 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 1659 (Current->is(tok::comment) && NextNonComment && 1660 NextNonComment->is(TT_SelectorName))) 1661 return 0; 1662 if (Current->is(TT_RangeBasedForLoopColon)) 1663 return prec::Comma; 1664 if ((Style.Language == FormatStyle::LK_Java || 1665 Style.Language == FormatStyle::LK_JavaScript) && 1666 Current->is(Keywords.kw_instanceof)) 1667 return prec::Relational; 1668 if (Style.Language == FormatStyle::LK_JavaScript && 1669 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 1670 return prec::Relational; 1671 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 1672 return Current->getPrecedence(); 1673 if (Current->isOneOf(tok::period, tok::arrow)) 1674 return PrecedenceArrowAndPeriod; 1675 if ((Style.Language == FormatStyle::LK_Java || 1676 Style.Language == FormatStyle::LK_JavaScript) && 1677 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 1678 Keywords.kw_throws)) 1679 return 0; 1680 } 1681 return -1; 1682 } 1683 1684 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 1685 Start->FakeLParens.push_back(Precedence); 1686 if (Precedence > prec::Unknown) 1687 Start->StartsBinaryExpression = true; 1688 if (Current) { 1689 FormatToken *Previous = Current->Previous; 1690 while (Previous->is(tok::comment) && Previous->Previous) 1691 Previous = Previous->Previous; 1692 ++Previous->FakeRParens; 1693 if (Precedence > prec::Unknown) 1694 Previous->EndsBinaryExpression = true; 1695 } 1696 } 1697 1698 /// \brief Parse unary operator expressions and surround them with fake 1699 /// parentheses if appropriate. 1700 void parseUnaryOperator() { 1701 llvm::SmallVector<FormatToken *, 2> Tokens; 1702 while (Current && Current->is(TT_UnaryOperator)) { 1703 Tokens.push_back(Current); 1704 next(); 1705 } 1706 parse(PrecedenceArrowAndPeriod); 1707 for (FormatToken *Token : llvm::reverse(Tokens)) 1708 // The actual precedence doesn't matter. 1709 addFakeParenthesis(Token, prec::Unknown); 1710 } 1711 1712 void parseConditionalExpr() { 1713 while (Current && Current->isTrailingComment()) { 1714 next(); 1715 } 1716 FormatToken *Start = Current; 1717 parse(prec::LogicalOr); 1718 if (!Current || !Current->is(tok::question)) 1719 return; 1720 next(); 1721 parse(prec::Assignment); 1722 if (!Current || Current->isNot(TT_ConditionalExpr)) 1723 return; 1724 next(); 1725 parse(prec::Assignment); 1726 addFakeParenthesis(Start, prec::Conditional); 1727 } 1728 1729 void next(bool SkipPastLeadingComments = true) { 1730 if (Current) 1731 Current = Current->Next; 1732 while (Current && 1733 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 1734 Current->isTrailingComment()) 1735 Current = Current->Next; 1736 } 1737 1738 const FormatStyle &Style; 1739 const AdditionalKeywords &Keywords; 1740 FormatToken *Current; 1741 }; 1742 1743 } // end anonymous namespace 1744 1745 void TokenAnnotator::setCommentLineLevels( 1746 SmallVectorImpl<AnnotatedLine *> &Lines) { 1747 const AnnotatedLine *NextNonCommentLine = nullptr; 1748 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(), 1749 E = Lines.rend(); 1750 I != E; ++I) { 1751 bool CommentLine = true; 1752 for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) { 1753 if (!Tok->is(tok::comment)) { 1754 CommentLine = false; 1755 break; 1756 } 1757 } 1758 1759 // If the comment is currently aligned with the line immediately following 1760 // it, that's probably intentional and we should keep it. 1761 if (NextNonCommentLine && CommentLine && 1762 NextNonCommentLine->First->NewlinesBefore <= 1 && 1763 NextNonCommentLine->First->OriginalColumn == 1764 (*I)->First->OriginalColumn) { 1765 // Align comments for preprocessor lines with the # in column 0. 1766 // Otherwise, align with the next line. 1767 (*I)->Level = (NextNonCommentLine->Type == LT_PreprocessorDirective || 1768 NextNonCommentLine->Type == LT_ImportStatement) 1769 ? 0 1770 : NextNonCommentLine->Level; 1771 } else { 1772 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr; 1773 } 1774 1775 setCommentLineLevels((*I)->Children); 1776 } 1777 } 1778 1779 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 1780 unsigned Result = 0; 1781 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 1782 Result = std::max(Result, Tok->NestingLevel); 1783 return Result; 1784 } 1785 1786 void TokenAnnotator::annotate(AnnotatedLine &Line) { 1787 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 1788 E = Line.Children.end(); 1789 I != E; ++I) { 1790 annotate(**I); 1791 } 1792 AnnotatingParser Parser(Style, Line, Keywords); 1793 Line.Type = Parser.parseLine(); 1794 1795 // With very deep nesting, ExpressionParser uses lots of stack and the 1796 // formatting algorithm is very slow. We're not going to do a good job here 1797 // anyway - it's probably generated code being formatted by mistake. 1798 // Just skip the whole line. 1799 if (maxNestingDepth(Line) > 50) 1800 Line.Type = LT_Invalid; 1801 1802 if (Line.Type == LT_Invalid) 1803 return; 1804 1805 ExpressionParser ExprParser(Style, Keywords, Line); 1806 ExprParser.parse(); 1807 1808 if (Line.startsWith(TT_ObjCMethodSpecifier)) 1809 Line.Type = LT_ObjCMethodDecl; 1810 else if (Line.startsWith(TT_ObjCDecl)) 1811 Line.Type = LT_ObjCDecl; 1812 else if (Line.startsWith(TT_ObjCProperty)) 1813 Line.Type = LT_ObjCProperty; 1814 1815 Line.First->SpacesRequiredBefore = 1; 1816 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 1817 } 1818 1819 // This function heuristically determines whether 'Current' starts the name of a 1820 // function declaration. 1821 static bool isFunctionDeclarationName(const FormatToken &Current, 1822 const AnnotatedLine &Line) { 1823 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 1824 for (; Next; Next = Next->Next) { 1825 if (Next->is(TT_OverloadedOperatorLParen)) 1826 return Next; 1827 if (Next->is(TT_OverloadedOperator)) 1828 continue; 1829 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 1830 // For 'new[]' and 'delete[]'. 1831 if (Next->Next && Next->Next->is(tok::l_square) && Next->Next->Next && 1832 Next->Next->Next->is(tok::r_square)) 1833 Next = Next->Next->Next; 1834 continue; 1835 } 1836 1837 break; 1838 } 1839 return nullptr; 1840 }; 1841 1842 // Find parentheses of parameter list. 1843 const FormatToken *Next = Current.Next; 1844 if (Current.is(tok::kw_operator)) { 1845 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 1846 return false; 1847 Next = skipOperatorName(Next); 1848 } else { 1849 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 1850 return false; 1851 for (; Next; Next = Next->Next) { 1852 if (Next->is(TT_TemplateOpener)) { 1853 Next = Next->MatchingParen; 1854 } else if (Next->is(tok::coloncolon)) { 1855 Next = Next->Next; 1856 if (!Next) 1857 return false; 1858 if (Next->is(tok::kw_operator)) { 1859 Next = skipOperatorName(Next->Next); 1860 break; 1861 } 1862 if (!Next->is(tok::identifier)) 1863 return false; 1864 } else if (Next->is(tok::l_paren)) { 1865 break; 1866 } else { 1867 return false; 1868 } 1869 } 1870 } 1871 1872 // Check whether parameter list can belong to a function declaration. 1873 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 1874 return false; 1875 // If the lines ends with "{", this is likely an function definition. 1876 if (Line.Last->is(tok::l_brace)) 1877 return true; 1878 if (Next->Next == Next->MatchingParen) 1879 return true; // Empty parentheses. 1880 // If there is an &/&& after the r_paren, this is likely a function. 1881 if (Next->MatchingParen->Next && 1882 Next->MatchingParen->Next->is(TT_PointerOrReference)) 1883 return true; 1884 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 1885 Tok = Tok->Next) { 1886 if (Tok->is(tok::l_paren) && Tok->MatchingParen) { 1887 Tok = Tok->MatchingParen; 1888 continue; 1889 } 1890 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 1891 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 1892 return true; 1893 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 1894 Tok->Tok.isLiteral()) 1895 return false; 1896 } 1897 return false; 1898 } 1899 1900 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 1901 assert(Line.MightBeFunctionDecl); 1902 1903 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 1904 Style.AlwaysBreakAfterReturnType == 1905 FormatStyle::RTBS_TopLevelDefinitions) && 1906 Line.Level > 0) 1907 return false; 1908 1909 switch (Style.AlwaysBreakAfterReturnType) { 1910 case FormatStyle::RTBS_None: 1911 return false; 1912 case FormatStyle::RTBS_All: 1913 case FormatStyle::RTBS_TopLevel: 1914 return true; 1915 case FormatStyle::RTBS_AllDefinitions: 1916 case FormatStyle::RTBS_TopLevelDefinitions: 1917 return Line.mightBeFunctionDefinition(); 1918 } 1919 1920 return false; 1921 } 1922 1923 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 1924 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 1925 E = Line.Children.end(); 1926 I != E; ++I) { 1927 calculateFormattingInformation(**I); 1928 } 1929 1930 Line.First->TotalLength = 1931 Line.First->IsMultiline ? Style.ColumnLimit 1932 : Line.FirstStartColumn + Line.First->ColumnWidth; 1933 FormatToken *Current = Line.First->Next; 1934 bool InFunctionDecl = Line.MightBeFunctionDecl; 1935 while (Current) { 1936 if (isFunctionDeclarationName(*Current, Line)) 1937 Current->Type = TT_FunctionDeclarationName; 1938 if (Current->is(TT_LineComment)) { 1939 if (Current->Previous->BlockKind == BK_BracedInit && 1940 Current->Previous->opensScope()) 1941 Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1; 1942 else 1943 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 1944 1945 // If we find a trailing comment, iterate backwards to determine whether 1946 // it seems to relate to a specific parameter. If so, break before that 1947 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 1948 // to the previous line in: 1949 // SomeFunction(a, 1950 // b, // comment 1951 // c); 1952 if (!Current->HasUnescapedNewline) { 1953 for (FormatToken *Parameter = Current->Previous; Parameter; 1954 Parameter = Parameter->Previous) { 1955 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 1956 break; 1957 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 1958 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 1959 Parameter->HasUnescapedNewline) 1960 Parameter->MustBreakBefore = true; 1961 break; 1962 } 1963 } 1964 } 1965 } else if (Current->SpacesRequiredBefore == 0 && 1966 spaceRequiredBefore(Line, *Current)) { 1967 Current->SpacesRequiredBefore = 1; 1968 } 1969 1970 Current->MustBreakBefore = 1971 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 1972 1973 if (!Current->MustBreakBefore && InFunctionDecl && 1974 Current->is(TT_FunctionDeclarationName)) 1975 Current->MustBreakBefore = mustBreakForReturnType(Line); 1976 1977 Current->CanBreakBefore = 1978 Current->MustBreakBefore || canBreakBefore(Line, *Current); 1979 unsigned ChildSize = 0; 1980 if (Current->Previous->Children.size() == 1) { 1981 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last; 1982 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 1983 : LastOfChild.TotalLength + 1; 1984 } 1985 const FormatToken *Prev = Current->Previous; 1986 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 1987 (Prev->Children.size() == 1 && 1988 Prev->Children[0]->First->MustBreakBefore) || 1989 Current->IsMultiline) 1990 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 1991 else 1992 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 1993 ChildSize + Current->SpacesRequiredBefore; 1994 1995 if (Current->is(TT_CtorInitializerColon)) 1996 InFunctionDecl = false; 1997 1998 // FIXME: Only calculate this if CanBreakBefore is true once static 1999 // initializers etc. are sorted out. 2000 // FIXME: Move magic numbers to a better place. 2001 Current->SplitPenalty = 20 * Current->BindingStrength + 2002 splitPenalty(Line, *Current, InFunctionDecl); 2003 2004 Current = Current->Next; 2005 } 2006 2007 calculateUnbreakableTailLengths(Line); 2008 unsigned IndentLevel = Line.Level; 2009 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2010 if (Current->Role) 2011 Current->Role->precomputeFormattingInfos(Current); 2012 if (Current->MatchingParen && 2013 Current->MatchingParen->opensBlockOrBlockTypeList(Style)) { 2014 assert(IndentLevel > 0); 2015 --IndentLevel; 2016 } 2017 Current->IndentLevel = IndentLevel; 2018 if (Current->opensBlockOrBlockTypeList(Style)) 2019 ++IndentLevel; 2020 } 2021 2022 DEBUG({ printDebugInfo(Line); }); 2023 } 2024 2025 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2026 unsigned UnbreakableTailLength = 0; 2027 FormatToken *Current = Line.Last; 2028 while (Current) { 2029 Current->UnbreakableTailLength = UnbreakableTailLength; 2030 if (Current->CanBreakBefore || 2031 Current->isOneOf(tok::comment, tok::string_literal)) { 2032 UnbreakableTailLength = 0; 2033 } else { 2034 UnbreakableTailLength += 2035 Current->ColumnWidth + Current->SpacesRequiredBefore; 2036 } 2037 Current = Current->Previous; 2038 } 2039 } 2040 2041 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2042 const FormatToken &Tok, 2043 bool InFunctionDecl) { 2044 const FormatToken &Left = *Tok.Previous; 2045 const FormatToken &Right = Tok; 2046 2047 if (Left.is(tok::semi)) 2048 return 0; 2049 2050 if (Style.Language == FormatStyle::LK_Java) { 2051 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2052 return 1; 2053 if (Right.is(Keywords.kw_implements)) 2054 return 2; 2055 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2056 return 3; 2057 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2058 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2059 return 100; 2060 if (Left.is(TT_JsTypeColon)) 2061 return 35; 2062 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2063 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2064 return 100; 2065 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2066 if (Left.opensScope() && Right.closesScope()) 2067 return 200; 2068 } 2069 2070 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2071 return 1; 2072 if (Right.is(tok::l_square)) { 2073 if (Style.Language == FormatStyle::LK_Proto) 2074 return 1; 2075 if (Left.is(tok::r_square)) 2076 return 200; 2077 // Slightly prefer formatting local lambda definitions like functions. 2078 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2079 return 35; 2080 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2081 TT_ArrayInitializerLSquare, 2082 TT_DesignatedInitializerLSquare)) 2083 return 500; 2084 } 2085 2086 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2087 Right.is(tok::kw_operator)) { 2088 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2089 return 3; 2090 if (Left.is(TT_StartOfName)) 2091 return 110; 2092 if (InFunctionDecl && Right.NestingLevel == 0) 2093 return Style.PenaltyReturnTypeOnItsOwnLine; 2094 return 200; 2095 } 2096 if (Right.is(TT_PointerOrReference)) 2097 return 190; 2098 if (Right.is(TT_LambdaArrow)) 2099 return 110; 2100 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2101 return 160; 2102 if (Left.is(TT_CastRParen)) 2103 return 100; 2104 if (Left.is(tok::coloncolon) || 2105 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2106 return 500; 2107 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2108 return 5000; 2109 if (Left.is(tok::comment)) 2110 return 1000; 2111 2112 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2113 TT_CtorInitializerColon)) 2114 return 2; 2115 2116 if (Right.isMemberAccess()) { 2117 // Breaking before the "./->" of a chained call/member access is reasonably 2118 // cheap, as formatting those with one call per line is generally 2119 // desirable. In particular, it should be cheaper to break before the call 2120 // than it is to break inside a call's parameters, which could lead to weird 2121 // "hanging" indents. The exception is the very last "./->" to support this 2122 // frequent pattern: 2123 // 2124 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2125 // dddddddd); 2126 // 2127 // which might otherwise be blown up onto many lines. Here, clang-format 2128 // won't produce "hanging" indents anyway as there is no other trailing 2129 // call. 2130 // 2131 // Also apply higher penalty is not a call as that might lead to a wrapping 2132 // like: 2133 // 2134 // aaaaaaa 2135 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2136 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2137 ? 150 2138 : 35; 2139 } 2140 2141 if (Right.is(TT_TrailingAnnotation) && 2142 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2143 // Moving trailing annotations to the next line is fine for ObjC method 2144 // declarations. 2145 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2146 return 10; 2147 // Generally, breaking before a trailing annotation is bad unless it is 2148 // function-like. It seems to be especially preferable to keep standard 2149 // annotations (i.e. "const", "final" and "override") on the same line. 2150 // Use a slightly higher penalty after ")" so that annotations like 2151 // "const override" are kept together. 2152 bool is_short_annotation = Right.TokenText.size() < 10; 2153 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 2154 } 2155 2156 // In for-loops, prefer breaking at ',' and ';'. 2157 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 2158 return 4; 2159 2160 // In Objective-C method expressions, prefer breaking before "param:" over 2161 // breaking after it. 2162 if (Right.is(TT_SelectorName)) 2163 return 0; 2164 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 2165 return Line.MightBeFunctionDecl ? 50 : 500; 2166 2167 if (Left.is(tok::l_paren) && InFunctionDecl && 2168 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 2169 return 100; 2170 if (Left.is(tok::l_paren) && Left.Previous && 2171 (Left.Previous->isOneOf(tok::kw_if, tok::kw_for) || 2172 Left.Previous->endsSequence(tok::kw_constexpr, tok::kw_if))) 2173 return 1000; 2174 if (Left.is(tok::equal) && InFunctionDecl) 2175 return 110; 2176 if (Right.is(tok::r_brace)) 2177 return 1; 2178 if (Left.is(TT_TemplateOpener)) 2179 return 100; 2180 if (Left.opensScope()) { 2181 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign) 2182 return 0; 2183 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 2184 : 19; 2185 } 2186 if (Left.is(TT_JavaAnnotation)) 2187 return 50; 2188 2189 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 2190 Left.Previous->isLabelString() && 2191 (Left.NextOperator || Left.OperatorIndex != 0)) 2192 return 50; 2193 if (Right.is(tok::plus) && Left.isLabelString() && 2194 (Right.NextOperator || Right.OperatorIndex != 0)) 2195 return 25; 2196 if (Left.is(tok::comma)) 2197 return 1; 2198 if (Right.is(tok::lessless) && Left.isLabelString() && 2199 (Right.NextOperator || Right.OperatorIndex != 1)) 2200 return 25; 2201 if (Right.is(tok::lessless)) { 2202 // Breaking at a << is really cheap. 2203 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 2204 // Slightly prefer to break before the first one in log-like statements. 2205 return 2; 2206 return 1; 2207 } 2208 if (Left.is(TT_ConditionalExpr)) 2209 return prec::Conditional; 2210 prec::Level Level = Left.getPrecedence(); 2211 if (Level == prec::Unknown) 2212 Level = Right.getPrecedence(); 2213 if (Level == prec::Assignment) 2214 return Style.PenaltyBreakAssignment; 2215 if (Level != prec::Unknown) 2216 return Level; 2217 2218 return 3; 2219 } 2220 2221 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 2222 const FormatToken &Left, 2223 const FormatToken &Right) { 2224 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 2225 return true; 2226 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 2227 return true; 2228 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 2229 Left.Tok.getObjCKeywordID() == tok::objc_property) 2230 return true; 2231 if (Right.is(tok::hashhash)) 2232 return Left.is(tok::hash); 2233 if (Left.isOneOf(tok::hashhash, tok::hash)) 2234 return Right.is(tok::hash); 2235 if (Left.is(tok::l_paren) && Right.is(tok::r_paren)) 2236 return Style.SpaceInEmptyParentheses; 2237 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 2238 return (Right.is(TT_CastRParen) || 2239 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 2240 ? Style.SpacesInCStyleCastParentheses 2241 : Style.SpacesInParentheses; 2242 if (Right.isOneOf(tok::semi, tok::comma)) 2243 return false; 2244 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl && 2245 Style.ObjCSpaceBeforeProtocolList) 2246 return true; 2247 if (Right.is(tok::less) && Left.is(tok::kw_template)) 2248 return Style.SpaceAfterTemplateKeyword; 2249 if (Left.isOneOf(tok::exclaim, tok::tilde)) 2250 return false; 2251 if (Left.is(tok::at) && 2252 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 2253 tok::numeric_constant, tok::l_paren, tok::l_brace, 2254 tok::kw_true, tok::kw_false)) 2255 return false; 2256 if (Left.is(tok::colon)) 2257 return !Left.is(TT_ObjCMethodExpr); 2258 if (Left.is(tok::coloncolon)) 2259 return false; 2260 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 2261 if (Style.Language == FormatStyle::LK_TextProto || 2262 (Style.Language == FormatStyle::LK_Proto && 2263 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 2264 // Format empty list as `<>`. 2265 if (Left.is(tok::less) && Right.is(tok::greater)) 2266 return false; 2267 return !Style.Cpp11BracedListStyle; 2268 } 2269 return false; 2270 } 2271 if (Right.is(tok::ellipsis)) 2272 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 2273 Left.Previous->is(tok::kw_case)); 2274 if (Left.is(tok::l_square) && Right.is(tok::amp)) 2275 return false; 2276 if (Right.is(TT_PointerOrReference)) { 2277 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 2278 if (!Left.MatchingParen) 2279 return true; 2280 FormatToken *TokenBeforeMatchingParen = 2281 Left.MatchingParen->getPreviousNonComment(); 2282 if (!TokenBeforeMatchingParen || 2283 !TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype)) 2284 return true; 2285 } 2286 return (Left.Tok.isLiteral() || 2287 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 2288 (Style.PointerAlignment != FormatStyle::PAS_Left || 2289 (Line.IsMultiVariableDeclStmt && 2290 (Left.NestingLevel == 0 || 2291 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for))))))); 2292 } 2293 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 2294 (!Left.is(TT_PointerOrReference) || 2295 (Style.PointerAlignment != FormatStyle::PAS_Right && 2296 !Line.IsMultiVariableDeclStmt))) 2297 return true; 2298 if (Left.is(TT_PointerOrReference)) 2299 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) || 2300 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 2301 !Right.is(TT_StartOfName)) || 2302 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) || 2303 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 2304 tok::l_paren) && 2305 (Style.PointerAlignment != FormatStyle::PAS_Right && 2306 !Line.IsMultiVariableDeclStmt) && 2307 Left.Previous && 2308 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon)); 2309 if (Right.is(tok::star) && Left.is(tok::l_paren)) 2310 return false; 2311 if (Left.is(tok::l_square)) 2312 return (Left.is(TT_ArrayInitializerLSquare) && 2313 Style.SpacesInContainerLiterals && Right.isNot(tok::r_square)) || 2314 (Left.isOneOf(TT_ArraySubscriptLSquare, 2315 TT_StructuredBindingLSquare) && 2316 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 2317 if (Right.is(tok::r_square)) 2318 return Right.MatchingParen && 2319 ((Style.SpacesInContainerLiterals && 2320 Right.MatchingParen->is(TT_ArrayInitializerLSquare)) || 2321 (Style.SpacesInSquareBrackets && 2322 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 2323 TT_StructuredBindingLSquare))); 2324 if (Right.is(tok::l_square) && 2325 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2326 TT_DesignatedInitializerLSquare, 2327 TT_StructuredBindingLSquare) && 2328 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral)) 2329 return false; 2330 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 2331 return !Left.Children.empty(); // No spaces in "{}". 2332 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) || 2333 (Right.is(tok::r_brace) && Right.MatchingParen && 2334 Right.MatchingParen->BlockKind != BK_Block)) 2335 return !Style.Cpp11BracedListStyle; 2336 if (Left.is(TT_BlockComment)) 2337 return !Left.TokenText.endswith("=*/"); 2338 if (Right.is(tok::l_paren)) { 2339 if (Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) 2340 return true; 2341 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) || 2342 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never && 2343 (Left.isOneOf(tok::kw_if, tok::pp_elif, tok::kw_for, tok::kw_while, 2344 tok::kw_switch, tok::kw_case, TT_ForEachMacro, 2345 TT_ObjCForIn) || 2346 Left.endsSequence(tok::kw_constexpr, tok::kw_if) || 2347 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch, 2348 tok::kw_new, tok::kw_delete) && 2349 (!Left.Previous || Left.Previous->isNot(tok::period))))) || 2350 (Style.SpaceBeforeParens == FormatStyle::SBPO_Always && 2351 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 2352 Left.is(tok::r_paren)) && 2353 Line.Type != LT_PreprocessorDirective); 2354 } 2355 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 2356 return false; 2357 if (Right.is(TT_UnaryOperator)) 2358 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 2359 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 2360 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 2361 tok::r_paren) || 2362 Left.isSimpleTypeSpecifier()) && 2363 Right.is(tok::l_brace) && Right.getNextNonComment() && 2364 Right.BlockKind != BK_Block) 2365 return false; 2366 if (Left.is(tok::period) || Right.is(tok::period)) 2367 return false; 2368 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L") 2369 return false; 2370 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 2371 Left.MatchingParen->Previous && 2372 Left.MatchingParen->Previous->is(tok::period)) 2373 // A.<B<C<...>>>DoSomething(); 2374 return false; 2375 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 2376 return false; 2377 return true; 2378 } 2379 2380 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 2381 const FormatToken &Right) { 2382 const FormatToken &Left = *Right.Previous; 2383 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 2384 return true; // Never ever merge two identifiers. 2385 if (Style.isCpp()) { 2386 if (Left.is(tok::kw_operator)) 2387 return Right.is(tok::coloncolon); 2388 } else if (Style.Language == FormatStyle::LK_Proto || 2389 Style.Language == FormatStyle::LK_TextProto) { 2390 if (Right.is(tok::period) && 2391 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 2392 Keywords.kw_repeated, Keywords.kw_extend)) 2393 return true; 2394 if (Right.is(tok::l_paren) && 2395 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 2396 return true; 2397 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 2398 return true; 2399 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2400 if (Left.is(TT_JsFatArrow)) 2401 return true; 2402 // for await ( ... 2403 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 2404 Left.Previous->is(tok::kw_for)) 2405 return true; 2406 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 2407 Right.MatchingParen) { 2408 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 2409 // An async arrow function, for example: `x = async () => foo();`, 2410 // as opposed to calling a function called async: `x = async();` 2411 if (Next && Next->is(TT_JsFatArrow)) 2412 return true; 2413 } 2414 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2415 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2416 return false; 2417 // In tagged template literals ("html`bar baz`"), there is no space between 2418 // the tag identifier and the template string. getIdentifierInfo makes sure 2419 // that the identifier is not a pseudo keyword like `yield`, either. 2420 if (Left.is(tok::identifier) && Keywords.IsJavaScriptIdentifier(Left) && 2421 Right.is(TT_TemplateString)) 2422 return false; 2423 if (Right.is(tok::star) && 2424 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 2425 return false; 2426 if (Right.isOneOf(tok::l_brace, tok::l_square) && 2427 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 2428 Keywords.kw_extends, Keywords.kw_implements)) 2429 return true; 2430 if (Right.is(tok::l_paren)) { 2431 // JS methods can use some keywords as names (e.g. `delete()`). 2432 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 2433 return false; 2434 // Valid JS method names can include keywords, e.g. `foo.delete()` or 2435 // `bar.instanceof()`. Recognize call positions by preceding period. 2436 if (Left.Previous && Left.Previous->is(tok::period) && 2437 Left.Tok.getIdentifierInfo()) 2438 return false; 2439 // Additional unary JavaScript operators that need a space after. 2440 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 2441 tok::kw_void)) 2442 return true; 2443 } 2444 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 2445 tok::kw_const) || 2446 // "of" is only a keyword if it appears after another identifier 2447 // (e.g. as "const x of y" in a for loop), or after a destructuring 2448 // operation (const [x, y] of z, const {a, b} of c). 2449 (Left.is(Keywords.kw_of) && Left.Previous && 2450 (Left.Previous->Tok.getIdentifierInfo() || 2451 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 2452 (!Left.Previous || !Left.Previous->is(tok::period))) 2453 return true; 2454 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 2455 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 2456 return false; 2457 if (Left.is(Keywords.kw_as) && 2458 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 2459 return true; 2460 if (Left.is(tok::kw_default) && Left.Previous && 2461 Left.Previous->is(tok::kw_export)) 2462 return true; 2463 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 2464 return true; 2465 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 2466 return false; 2467 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 2468 return false; 2469 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 2470 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 2471 return false; 2472 if (Left.is(tok::ellipsis)) 2473 return false; 2474 if (Left.is(TT_TemplateCloser) && 2475 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 2476 Keywords.kw_implements, Keywords.kw_extends)) 2477 // Type assertions ('<type>expr') are not followed by whitespace. Other 2478 // locations that should have whitespace following are identified by the 2479 // above set of follower tokens. 2480 return false; 2481 if (Right.is(TT_JsNonNullAssertion)) 2482 return false; 2483 if (Left.is(TT_JsNonNullAssertion) && 2484 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 2485 return true; // "x! as string", "x! in y" 2486 } else if (Style.Language == FormatStyle::LK_Java) { 2487 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 2488 return true; 2489 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 2490 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never; 2491 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 2492 tok::kw_protected) || 2493 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 2494 Keywords.kw_native)) && 2495 Right.is(TT_TemplateOpener)) 2496 return true; 2497 } 2498 if (Left.is(TT_ImplicitStringLiteral)) 2499 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2500 if (Line.Type == LT_ObjCMethodDecl) { 2501 if (Left.is(TT_ObjCMethodSpecifier)) 2502 return true; 2503 if (Left.is(tok::r_paren) && Right.is(tok::identifier)) 2504 // Don't space between ')' and <id> 2505 return false; 2506 } 2507 if (Line.Type == LT_ObjCProperty && 2508 (Right.is(tok::equal) || Left.is(tok::equal))) 2509 return false; 2510 2511 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 2512 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 2513 return true; 2514 if (Right.is(TT_OverloadedOperatorLParen)) 2515 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always; 2516 if (Left.is(tok::comma)) 2517 return true; 2518 if (Right.is(tok::comma)) 2519 return false; 2520 if (Right.isOneOf(TT_CtorInitializerColon, TT_ObjCBlockLParen)) 2521 return true; 2522 if (Right.is(tok::colon)) { 2523 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) || 2524 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 2525 return false; 2526 if (Right.is(TT_ObjCMethodExpr)) 2527 return false; 2528 if (Left.is(tok::question)) 2529 return false; 2530 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 2531 return false; 2532 if (Right.is(TT_DictLiteral)) 2533 return Style.SpacesInContainerLiterals; 2534 return true; 2535 } 2536 if (Left.is(TT_UnaryOperator)) 2537 return Right.is(TT_BinaryOperator); 2538 2539 // If the next token is a binary operator or a selector name, we have 2540 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 2541 if (Left.is(TT_CastRParen)) 2542 return Style.SpaceAfterCStyleCast || 2543 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 2544 2545 if (Left.is(tok::greater) && Right.is(tok::greater)) { 2546 if (Style.Language == FormatStyle::LK_TextProto || 2547 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 2548 return !Style.Cpp11BracedListStyle; 2549 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 2550 (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles); 2551 } 2552 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 2553 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 2554 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 2555 return false; 2556 if (!Style.SpaceBeforeAssignmentOperators && 2557 Right.getPrecedence() == prec::Assignment) 2558 return false; 2559 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 2560 // Generally don't remove existing spaces between an identifier and "::". 2561 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 2562 // this turns out to be too lenient, add analysis of the identifier itself. 2563 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2564 if (Right.is(tok::coloncolon) && !Left.isOneOf(tok::l_brace, tok::comment)) 2565 return (Left.is(TT_TemplateOpener) && 2566 Style.Standard == FormatStyle::LS_Cpp03) || 2567 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 2568 tok::kw___super, TT_TemplateCloser, 2569 TT_TemplateOpener)); 2570 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 2571 return Style.SpacesInAngles; 2572 // Space before TT_StructuredBindingLSquare. 2573 if (Right.is(TT_StructuredBindingLSquare)) 2574 return !Left.isOneOf(tok::amp, tok::ampamp) || 2575 Style.PointerAlignment != FormatStyle::PAS_Right; 2576 // Space before & or && following a TT_StructuredBindingLSquare. 2577 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 2578 Right.isOneOf(tok::amp, tok::ampamp)) 2579 return Style.PointerAlignment != FormatStyle::PAS_Left; 2580 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 2581 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 2582 !Right.is(tok::r_paren))) 2583 return true; 2584 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) && 2585 Right.isNot(TT_FunctionTypeLParen)) 2586 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always; 2587 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 2588 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 2589 return false; 2590 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 2591 Line.startsWith(tok::hash)) 2592 return true; 2593 if (Right.is(TT_TrailingUnaryOperator)) 2594 return false; 2595 if (Left.is(TT_RegexLiteral)) 2596 return false; 2597 return spaceRequiredBetween(Line, Left, Right); 2598 } 2599 2600 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 2601 static bool isAllmanBrace(const FormatToken &Tok) { 2602 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block && 2603 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral); 2604 } 2605 2606 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 2607 const FormatToken &Right) { 2608 const FormatToken &Left = *Right.Previous; 2609 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 2610 return true; 2611 2612 if (Style.Language == FormatStyle::LK_JavaScript) { 2613 // FIXME: This might apply to other languages and token kinds. 2614 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 2615 Left.Previous->is(tok::string_literal)) 2616 return true; 2617 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 2618 Left.Previous && Left.Previous->is(tok::equal) && 2619 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 2620 tok::kw_const) && 2621 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 2622 // above. 2623 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 2624 // Object literals on the top level of a file are treated as "enum-style". 2625 // Each key/value pair is put on a separate line, instead of bin-packing. 2626 return true; 2627 if (Left.is(tok::l_brace) && Line.Level == 0 && 2628 (Line.startsWith(tok::kw_enum) || 2629 Line.startsWith(tok::kw_const, tok::kw_enum) || 2630 Line.startsWith(tok::kw_export, tok::kw_enum) || 2631 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 2632 // JavaScript top-level enum key/value pairs are put on separate lines 2633 // instead of bin-packing. 2634 return true; 2635 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 2636 !Left.Children.empty()) 2637 // Support AllowShortFunctionsOnASingleLine for JavaScript. 2638 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 2639 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 2640 (Left.NestingLevel == 0 && Line.Level == 0 && 2641 Style.AllowShortFunctionsOnASingleLine & 2642 FormatStyle::SFS_InlineOnly); 2643 } else if (Style.Language == FormatStyle::LK_Java) { 2644 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 2645 Right.Next->is(tok::string_literal)) 2646 return true; 2647 } else if (Style.Language == FormatStyle::LK_Cpp || 2648 Style.Language == FormatStyle::LK_ObjC || 2649 Style.Language == FormatStyle::LK_Proto) { 2650 if (Left.isStringLiteral() && Right.isStringLiteral()) 2651 return true; 2652 } 2653 2654 // If the last token before a '}', ']', or ')' is a comma or a trailing 2655 // comment, the intention is to insert a line break after it in order to make 2656 // shuffling around entries easier. Import statements, especially in 2657 // JavaScript, can be an exception to this rule. 2658 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 2659 const FormatToken *BeforeClosingBrace = nullptr; 2660 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 2661 (Style.Language == FormatStyle::LK_JavaScript && 2662 Left.is(tok::l_paren))) && 2663 Left.BlockKind != BK_Block && Left.MatchingParen) 2664 BeforeClosingBrace = Left.MatchingParen->Previous; 2665 else if (Right.MatchingParen && 2666 (Right.MatchingParen->isOneOf(tok::l_brace, 2667 TT_ArrayInitializerLSquare) || 2668 (Style.Language == FormatStyle::LK_JavaScript && 2669 Right.MatchingParen->is(tok::l_paren)))) 2670 BeforeClosingBrace = &Left; 2671 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 2672 BeforeClosingBrace->isTrailingComment())) 2673 return true; 2674 } 2675 2676 if (Right.is(tok::comment)) 2677 return Left.BlockKind != BK_BracedInit && 2678 Left.isNot(TT_CtorInitializerColon) && 2679 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 2680 if (Left.isTrailingComment()) 2681 return true; 2682 if (Right.Previous->IsUnterminatedLiteral) 2683 return true; 2684 if (Right.is(tok::lessless) && Right.Next && 2685 Right.Previous->is(tok::string_literal) && 2686 Right.Next->is(tok::string_literal)) 2687 return true; 2688 if (Right.Previous->ClosesTemplateDeclaration && 2689 Right.Previous->MatchingParen && 2690 Right.Previous->MatchingParen->NestingLevel == 0 && 2691 Style.AlwaysBreakTemplateDeclarations) 2692 return true; 2693 if (Right.is(TT_CtorInitializerComma) && 2694 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 2695 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 2696 return true; 2697 if (Right.is(TT_CtorInitializerColon) && 2698 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 2699 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 2700 return true; 2701 // Break only if we have multiple inheritance. 2702 if (Style.BreakBeforeInheritanceComma && Right.is(TT_InheritanceComma)) 2703 return true; 2704 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 2705 // Raw string literals are special wrt. line breaks. The author has made a 2706 // deliberate choice and might have aligned the contents of the string 2707 // literal accordingly. Thus, we try keep existing line breaks. 2708 return Right.NewlinesBefore > 0; 2709 if ((Right.Previous->is(tok::l_brace) || 2710 (Right.Previous->is(tok::less) && Right.Previous->Previous && 2711 Right.Previous->Previous->is(tok::equal))) && 2712 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 2713 // Don't put enums or option definitions onto single lines in protocol 2714 // buffers. 2715 return true; 2716 } 2717 if (Right.is(TT_InlineASMBrace)) 2718 return Right.HasUnescapedNewline; 2719 if (isAllmanBrace(Left) || isAllmanBrace(Right)) 2720 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) || 2721 (Line.startsWith(tok::kw_typedef, tok::kw_enum) && 2722 Style.BraceWrapping.AfterEnum) || 2723 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 2724 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 2725 if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine) 2726 return true; 2727 2728 if ((Style.Language == FormatStyle::LK_Java || 2729 Style.Language == FormatStyle::LK_JavaScript) && 2730 Left.is(TT_LeadingJavaAnnotation) && 2731 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 2732 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 2733 return true; 2734 2735 return false; 2736 } 2737 2738 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 2739 const FormatToken &Right) { 2740 const FormatToken &Left = *Right.Previous; 2741 2742 // Language-specific stuff. 2743 if (Style.Language == FormatStyle::LK_Java) { 2744 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 2745 Keywords.kw_implements)) 2746 return false; 2747 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 2748 Keywords.kw_implements)) 2749 return true; 2750 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2751 const FormatToken *NonComment = Right.getPreviousNonComment(); 2752 if (NonComment && 2753 NonComment->isOneOf( 2754 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 2755 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 2756 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 2757 Keywords.kw_readonly, Keywords.kw_abstract, Keywords.kw_get, 2758 Keywords.kw_set, Keywords.kw_async, Keywords.kw_await)) 2759 return false; // Otherwise automatic semicolon insertion would trigger. 2760 if (Right.NestingLevel == 0 && 2761 (Left.Tok.getIdentifierInfo() || 2762 Left.isOneOf(tok::r_square, tok::r_paren)) && 2763 Right.isOneOf(tok::l_square, tok::l_paren)) 2764 return false; // Otherwise automatic semicolon insertion would trigger. 2765 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace)) 2766 return false; 2767 if (Left.is(TT_JsTypeColon)) 2768 return true; 2769 if (Right.NestingLevel == 0 && Right.is(Keywords.kw_is)) 2770 return false; 2771 if (Left.is(Keywords.kw_in)) 2772 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 2773 if (Right.is(Keywords.kw_in)) 2774 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 2775 if (Right.is(Keywords.kw_as)) 2776 return false; // must not break before as in 'x as type' casts 2777 if (Left.is(Keywords.kw_as)) 2778 return true; 2779 if (Left.is(TT_JsNonNullAssertion)) 2780 return true; 2781 if (Left.is(Keywords.kw_declare) && 2782 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 2783 Keywords.kw_function, tok::kw_class, tok::kw_enum, 2784 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 2785 Keywords.kw_let, tok::kw_const)) 2786 // See grammar for 'declare' statements at: 2787 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10 2788 return false; 2789 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 2790 Right.isOneOf(tok::identifier, tok::string_literal)) 2791 return false; // must not break in "module foo { ...}" 2792 if (Right.is(TT_TemplateString) && Right.closesScope()) 2793 return false; 2794 if (Left.is(TT_TemplateString) && Left.opensScope()) 2795 return true; 2796 } 2797 2798 if (Left.is(tok::at)) 2799 return false; 2800 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 2801 return false; 2802 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 2803 return !Right.is(tok::l_paren); 2804 if (Right.is(TT_PointerOrReference)) 2805 return Line.IsMultiVariableDeclStmt || 2806 (Style.PointerAlignment == FormatStyle::PAS_Right && 2807 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 2808 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2809 Right.is(tok::kw_operator)) 2810 return true; 2811 if (Left.is(TT_PointerOrReference)) 2812 return false; 2813 if (Right.isTrailingComment()) 2814 // We rely on MustBreakBefore being set correctly here as we should not 2815 // change the "binding" behavior of a comment. 2816 // The first comment in a braced lists is always interpreted as belonging to 2817 // the first list element. Otherwise, it should be placed outside of the 2818 // list. 2819 return Left.BlockKind == BK_BracedInit || 2820 (Left.is(TT_CtorInitializerColon) && 2821 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 2822 if (Left.is(tok::question) && Right.is(tok::colon)) 2823 return false; 2824 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 2825 return Style.BreakBeforeTernaryOperators; 2826 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 2827 return !Style.BreakBeforeTernaryOperators; 2828 if (Right.is(TT_InheritanceColon)) 2829 return true; 2830 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 2831 Left.isNot(TT_SelectorName)) 2832 return true; 2833 2834 if (Right.is(tok::colon) && 2835 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 2836 return false; 2837 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 2838 if ((Style.Language == FormatStyle::LK_Proto || 2839 Style.Language == FormatStyle::LK_TextProto) && 2840 Right.isStringLiteral()) 2841 return false; 2842 return true; 2843 } 2844 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 2845 Right.Next->is(TT_ObjCMethodExpr))) 2846 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 2847 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 2848 return true; 2849 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 2850 return true; 2851 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 2852 TT_OverloadedOperator)) 2853 return false; 2854 if (Left.is(TT_RangeBasedForLoopColon)) 2855 return true; 2856 if (Right.is(TT_RangeBasedForLoopColon)) 2857 return false; 2858 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 2859 return true; 2860 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 2861 Left.is(tok::kw_operator)) 2862 return false; 2863 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 2864 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 2865 return false; 2866 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen)) 2867 return false; 2868 if (Left.is(tok::l_paren) && Left.Previous && 2869 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 2870 return false; 2871 if (Right.is(TT_ImplicitStringLiteral)) 2872 return false; 2873 2874 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser)) 2875 return false; 2876 if (Right.is(tok::r_square) && Right.MatchingParen && 2877 Right.MatchingParen->is(TT_LambdaLSquare)) 2878 return false; 2879 2880 // We only break before r_brace if there was a corresponding break before 2881 // the l_brace, which is tracked by BreakBeforeClosingBrace. 2882 if (Right.is(tok::r_brace)) 2883 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block; 2884 2885 // Allow breaking after a trailing annotation, e.g. after a method 2886 // declaration. 2887 if (Left.is(TT_TrailingAnnotation)) 2888 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 2889 tok::less, tok::coloncolon); 2890 2891 if (Right.is(tok::kw___attribute)) 2892 return true; 2893 2894 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 2895 return true; 2896 2897 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2898 return true; 2899 2900 if (Left.is(TT_CtorInitializerColon)) 2901 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 2902 if (Right.is(TT_CtorInitializerColon)) 2903 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 2904 if (Left.is(TT_CtorInitializerComma) && 2905 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 2906 return false; 2907 if (Right.is(TT_CtorInitializerComma) && 2908 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 2909 return true; 2910 if (Left.is(TT_InheritanceComma) && Style.BreakBeforeInheritanceComma) 2911 return false; 2912 if (Right.is(TT_InheritanceComma) && Style.BreakBeforeInheritanceComma) 2913 return true; 2914 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 2915 (Left.is(tok::less) && Right.is(tok::less))) 2916 return false; 2917 if (Right.is(TT_BinaryOperator) && 2918 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 2919 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 2920 Right.getPrecedence() != prec::Assignment)) 2921 return true; 2922 if (Left.is(TT_ArrayInitializerLSquare)) 2923 return true; 2924 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 2925 return true; 2926 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 2927 !Left.isOneOf(tok::arrowstar, tok::lessless) && 2928 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 2929 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 2930 Left.getPrecedence() == prec::Assignment)) 2931 return true; 2932 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 2933 tok::kw_class, tok::kw_struct, tok::comment) || 2934 Right.isMemberAccess() || 2935 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 2936 tok::colon, tok::l_square, tok::at) || 2937 (Left.is(tok::r_paren) && 2938 Right.isOneOf(tok::identifier, tok::kw_const)) || 2939 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 2940 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 2941 } 2942 2943 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 2944 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 2945 const FormatToken *Tok = Line.First; 2946 while (Tok) { 2947 llvm::errs() << " M=" << Tok->MustBreakBefore 2948 << " C=" << Tok->CanBreakBefore 2949 << " T=" << getTokenTypeName(Tok->Type) 2950 << " S=" << Tok->SpacesRequiredBefore 2951 << " B=" << Tok->BlockParameterCount 2952 << " BK=" << Tok->BlockKind << " P=" << Tok->SplitPenalty 2953 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 2954 << " PPK=" << Tok->PackingKind << " FakeLParens="; 2955 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 2956 llvm::errs() << Tok->FakeLParens[i] << "/"; 2957 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 2958 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 2959 if (!Tok->Next) 2960 assert(Tok == Line.Last); 2961 Tok = Tok->Next; 2962 } 2963 llvm::errs() << "----\n"; 2964 } 2965 2966 } // namespace format 2967 } // namespace clang 2968