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