1 //===--- TokenAnnotator.cpp - Format C++ code -----------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// 9 /// \file 10 /// This file implements a token annotator, i.e. creates 11 /// \c AnnotatedTokens out of \c FormatTokens with required extra information. 12 /// 13 //===----------------------------------------------------------------------===// 14 15 #include "TokenAnnotator.h" 16 #include "FormatToken.h" 17 #include "clang/Basic/SourceManager.h" 18 #include "clang/Basic/TokenKinds.h" 19 #include "llvm/ADT/SmallPtrSet.h" 20 #include "llvm/Support/Debug.h" 21 22 #define DEBUG_TYPE "format-token-annotator" 23 24 namespace clang { 25 namespace format { 26 27 namespace { 28 29 /// Returns \c true if the line starts with a token that can start a statement 30 /// with an initializer. 31 static bool startsWithInitStatement(const AnnotatedLine &Line) { 32 return Line.startsWith(tok::kw_for) || Line.startsWith(tok::kw_if) || 33 Line.startsWith(tok::kw_switch); 34 } 35 36 /// Returns \c true if the token can be used as an identifier in 37 /// an Objective-C \c \@selector, \c false otherwise. 38 /// 39 /// Because getFormattingLangOpts() always lexes source code as 40 /// Objective-C++, C++ keywords like \c new and \c delete are 41 /// lexed as tok::kw_*, not tok::identifier, even for Objective-C. 42 /// 43 /// For Objective-C and Objective-C++, both identifiers and keywords 44 /// are valid inside @selector(...) (or a macro which 45 /// invokes @selector(...)). So, we allow treat any identifier or 46 /// keyword as a potential Objective-C selector component. 47 static bool canBeObjCSelectorComponent(const FormatToken &Tok) { 48 return Tok.Tok.getIdentifierInfo() != nullptr; 49 } 50 51 /// With `Left` being '(', check if we're at either `[...](` or 52 /// `[...]<...>(`, where the [ opens a lambda capture list. 53 static bool isLambdaParameterList(const FormatToken *Left) { 54 // Skip <...> if present. 55 if (Left->Previous && Left->Previous->is(tok::greater) && 56 Left->Previous->MatchingParen && 57 Left->Previous->MatchingParen->is(TT_TemplateOpener)) { 58 Left = Left->Previous->MatchingParen; 59 } 60 61 // Check for `[...]`. 62 return Left->Previous && Left->Previous->is(tok::r_square) && 63 Left->Previous->MatchingParen && 64 Left->Previous->MatchingParen->is(TT_LambdaLSquare); 65 } 66 67 /// Returns \c true if the token is followed by a boolean condition, \c false 68 /// otherwise. 69 static bool isKeywordWithCondition(const FormatToken &Tok) { 70 return Tok.isOneOf(tok::kw_if, tok::kw_for, tok::kw_while, tok::kw_switch, 71 tok::kw_constexpr, tok::kw_catch); 72 } 73 74 /// Returns \c true if the token starts a C++ attribute, \c false otherwise. 75 static bool isCppAttribute(bool IsCpp, const FormatToken &Tok) { 76 if (!IsCpp || !Tok.startsSequence(tok::l_square, tok::l_square)) 77 return false; 78 // The first square bracket is part of an ObjC array literal 79 if (Tok.Previous && Tok.Previous->is(tok::at)) 80 return false; 81 const FormatToken *AttrTok = Tok.Next->Next; 82 if (!AttrTok) 83 return false; 84 // C++17 '[[using ns: foo, bar(baz, blech)]]' 85 // We assume nobody will name an ObjC variable 'using'. 86 if (AttrTok->startsSequence(tok::kw_using, tok::identifier, tok::colon)) 87 return true; 88 if (AttrTok->isNot(tok::identifier)) 89 return false; 90 while (AttrTok && !AttrTok->startsSequence(tok::r_square, tok::r_square)) { 91 // ObjC message send. We assume nobody will use : in a C++11 attribute 92 // specifier parameter, although this is technically valid: 93 // [[foo(:)]]. 94 if (AttrTok->is(tok::colon) || 95 AttrTok->startsSequence(tok::identifier, tok::identifier) || 96 AttrTok->startsSequence(tok::r_paren, tok::identifier)) { 97 return false; 98 } 99 if (AttrTok->is(tok::ellipsis)) 100 return true; 101 AttrTok = AttrTok->Next; 102 } 103 return AttrTok && AttrTok->startsSequence(tok::r_square, tok::r_square); 104 } 105 106 /// A parser that gathers additional information about tokens. 107 /// 108 /// The \c TokenAnnotator tries to match parenthesis and square brakets and 109 /// store a parenthesis levels. It also tries to resolve matching "<" and ">" 110 /// into template parameter lists. 111 class AnnotatingParser { 112 public: 113 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line, 114 const AdditionalKeywords &Keywords) 115 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false), 116 Keywords(Keywords) { 117 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false)); 118 resetTokenMetadata(); 119 } 120 121 private: 122 bool parseAngle() { 123 if (!CurrentToken || !CurrentToken->Previous) 124 return false; 125 if (NonTemplateLess.count(CurrentToken->Previous)) 126 return false; 127 128 const FormatToken &Previous = *CurrentToken->Previous; // The '<'. 129 if (Previous.Previous) { 130 if (Previous.Previous->Tok.isLiteral()) 131 return false; 132 if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 && 133 (!Previous.Previous->MatchingParen || 134 !Previous.Previous->MatchingParen->is( 135 TT_OverloadedOperatorLParen))) { 136 return false; 137 } 138 } 139 140 FormatToken *Left = CurrentToken->Previous; 141 Left->ParentBracket = Contexts.back().ContextKind; 142 ScopedContextCreator ContextCreator(*this, tok::less, 12); 143 144 // If this angle is in the context of an expression, we need to be more 145 // hesitant to detect it as opening template parameters. 146 bool InExprContext = Contexts.back().IsExpression; 147 148 Contexts.back().IsExpression = false; 149 // If there's a template keyword before the opening angle bracket, this is a 150 // template parameter, not an argument. 151 if (Left->Previous && Left->Previous->isNot(tok::kw_template)) 152 Contexts.back().ContextType = Context::TemplateArgument; 153 154 if (Style.Language == FormatStyle::LK_Java && 155 CurrentToken->is(tok::question)) { 156 next(); 157 } 158 159 while (CurrentToken) { 160 if (CurrentToken->is(tok::greater)) { 161 // Try to do a better job at looking for ">>" within the condition of 162 // a statement. Conservatively insert spaces between consecutive ">" 163 // tokens to prevent splitting right bitshift operators and potentially 164 // altering program semantics. This check is overly conservative and 165 // will prevent spaces from being inserted in select nested template 166 // parameter cases, but should not alter program semantics. 167 if (CurrentToken->Next && CurrentToken->Next->is(tok::greater) && 168 Left->ParentBracket != tok::less && 169 CurrentToken->getStartOfNonWhitespace() == 170 CurrentToken->Next->getStartOfNonWhitespace().getLocWithOffset( 171 -1)) { 172 return false; 173 } 174 Left->MatchingParen = CurrentToken; 175 CurrentToken->MatchingParen = Left; 176 // In TT_Proto, we must distignuish between: 177 // map<key, value> 178 // msg < item: data > 179 // msg: < item: data > 180 // In TT_TextProto, map<key, value> does not occur. 181 if (Style.Language == FormatStyle::LK_TextProto || 182 (Style.Language == FormatStyle::LK_Proto && Left->Previous && 183 Left->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) { 184 CurrentToken->setType(TT_DictLiteral); 185 } else { 186 CurrentToken->setType(TT_TemplateCloser); 187 } 188 next(); 189 return true; 190 } 191 if (CurrentToken->is(tok::question) && 192 Style.Language == FormatStyle::LK_Java) { 193 next(); 194 continue; 195 } 196 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) || 197 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext && 198 !Style.isCSharp() && Style.Language != FormatStyle::LK_Proto && 199 Style.Language != FormatStyle::LK_TextProto)) { 200 return false; 201 } 202 // If a && or || is found and interpreted as a binary operator, this set 203 // of angles is likely part of something like "a < b && c > d". If the 204 // angles are inside an expression, the ||/&& might also be a binary 205 // operator that was misinterpreted because we are parsing template 206 // parameters. 207 // FIXME: This is getting out of hand, write a decent parser. 208 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) && 209 CurrentToken->Previous->is(TT_BinaryOperator) && 210 Contexts[Contexts.size() - 2].IsExpression && 211 !Line.startsWith(tok::kw_template)) { 212 return false; 213 } 214 updateParameterCount(Left, CurrentToken); 215 if (Style.Language == FormatStyle::LK_Proto) { 216 if (FormatToken *Previous = CurrentToken->getPreviousNonComment()) { 217 if (CurrentToken->is(tok::colon) || 218 (CurrentToken->isOneOf(tok::l_brace, tok::less) && 219 Previous->isNot(tok::colon))) { 220 Previous->setType(TT_SelectorName); 221 } 222 } 223 } 224 if (!consumeToken()) 225 return false; 226 } 227 return false; 228 } 229 230 bool parseUntouchableParens() { 231 while (CurrentToken) { 232 CurrentToken->Finalized = true; 233 switch (CurrentToken->Tok.getKind()) { 234 case tok::l_paren: 235 next(); 236 if (!parseUntouchableParens()) 237 return false; 238 continue; 239 case tok::r_paren: 240 next(); 241 return true; 242 default: 243 // no-op 244 break; 245 } 246 next(); 247 } 248 return false; 249 } 250 251 bool parseParens(bool LookForDecls = false) { 252 if (!CurrentToken) 253 return false; 254 assert(CurrentToken->Previous && "Unknown previous token"); 255 FormatToken &OpeningParen = *CurrentToken->Previous; 256 assert(OpeningParen.is(tok::l_paren)); 257 FormatToken *PrevNonComment = OpeningParen.getPreviousNonComment(); 258 OpeningParen.ParentBracket = Contexts.back().ContextKind; 259 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1); 260 261 // FIXME: This is a bit of a hack. Do better. 262 Contexts.back().ColonIsForRangeExpr = 263 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr; 264 265 if (OpeningParen.Previous && 266 OpeningParen.Previous->is(TT_UntouchableMacroFunc)) { 267 OpeningParen.Finalized = true; 268 return parseUntouchableParens(); 269 } 270 271 bool StartsObjCMethodExpr = false; 272 if (!Style.isVerilog()) { 273 if (FormatToken *MaybeSel = OpeningParen.Previous) { 274 // @selector( starts a selector. 275 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && 276 MaybeSel->Previous && MaybeSel->Previous->is(tok::at)) { 277 StartsObjCMethodExpr = true; 278 } 279 } 280 } 281 282 if (OpeningParen.is(TT_OverloadedOperatorLParen)) { 283 // Find the previous kw_operator token. 284 FormatToken *Prev = &OpeningParen; 285 while (!Prev->is(tok::kw_operator)) { 286 Prev = Prev->Previous; 287 assert(Prev && "Expect a kw_operator prior to the OperatorLParen!"); 288 } 289 290 // If faced with "a.operator*(argument)" or "a->operator*(argument)", 291 // i.e. the operator is called as a member function, 292 // then the argument must be an expression. 293 bool OperatorCalledAsMemberFunction = 294 Prev->Previous && Prev->Previous->isOneOf(tok::period, tok::arrow); 295 Contexts.back().IsExpression = OperatorCalledAsMemberFunction; 296 } else if (Style.isJavaScript() && 297 (Line.startsWith(Keywords.kw_type, tok::identifier) || 298 Line.startsWith(tok::kw_export, Keywords.kw_type, 299 tok::identifier))) { 300 // type X = (...); 301 // export type X = (...); 302 Contexts.back().IsExpression = false; 303 } else if (OpeningParen.Previous && 304 (OpeningParen.Previous->isOneOf(tok::kw_static_assert, 305 tok::kw_while, tok::l_paren, 306 tok::comma, TT_BinaryOperator) || 307 OpeningParen.Previous->isIf())) { 308 // static_assert, if and while usually contain expressions. 309 Contexts.back().IsExpression = true; 310 } else if (Style.isJavaScript() && OpeningParen.Previous && 311 (OpeningParen.Previous->is(Keywords.kw_function) || 312 (OpeningParen.Previous->endsSequence(tok::identifier, 313 Keywords.kw_function)))) { 314 // function(...) or function f(...) 315 Contexts.back().IsExpression = false; 316 } else if (Style.isJavaScript() && OpeningParen.Previous && 317 OpeningParen.Previous->is(TT_JsTypeColon)) { 318 // let x: (SomeType); 319 Contexts.back().IsExpression = false; 320 } else if (isLambdaParameterList(&OpeningParen)) { 321 // This is a parameter list of a lambda expression. 322 Contexts.back().IsExpression = false; 323 } else if (OpeningParen.is(TT_RequiresExpressionLParen)) { 324 Contexts.back().IsExpression = false; 325 } else if (OpeningParen.Previous && 326 OpeningParen.Previous->is(tok::kw__Generic)) { 327 Contexts.back().ContextType = Context::C11GenericSelection; 328 Contexts.back().IsExpression = true; 329 } else if (Line.InPPDirective && 330 (!OpeningParen.Previous || 331 !OpeningParen.Previous->is(tok::identifier))) { 332 Contexts.back().IsExpression = true; 333 } else if (Contexts[Contexts.size() - 2].CaretFound) { 334 // This is the parameter list of an ObjC block. 335 Contexts.back().IsExpression = false; 336 } else if (OpeningParen.Previous && 337 OpeningParen.Previous->is(TT_ForEachMacro)) { 338 // The first argument to a foreach macro is a declaration. 339 Contexts.back().ContextType = Context::ForEachMacro; 340 Contexts.back().IsExpression = false; 341 } else if (OpeningParen.Previous && OpeningParen.Previous->MatchingParen && 342 OpeningParen.Previous->MatchingParen->isOneOf( 343 TT_ObjCBlockLParen, TT_FunctionTypeLParen)) { 344 Contexts.back().IsExpression = false; 345 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) { 346 bool IsForOrCatch = 347 OpeningParen.Previous && 348 OpeningParen.Previous->isOneOf(tok::kw_for, tok::kw_catch); 349 Contexts.back().IsExpression = !IsForOrCatch; 350 } 351 352 // Infer the role of the l_paren based on the previous token if we haven't 353 // detected one yet. 354 if (PrevNonComment && OpeningParen.is(TT_Unknown)) { 355 if (PrevNonComment->is(tok::kw___attribute)) { 356 OpeningParen.setType(TT_AttributeParen); 357 } else if (PrevNonComment->isOneOf(TT_TypenameMacro, tok::kw_decltype, 358 tok::kw_typeof, 359 #define TRANSFORM_TYPE_TRAIT_DEF(_, Trait) tok::kw___##Trait, 360 #include "clang/Basic/TransformTypeTraits.def" 361 tok::kw__Atomic)) { 362 OpeningParen.setType(TT_TypeDeclarationParen); 363 // decltype() and typeof() usually contain expressions. 364 if (PrevNonComment->isOneOf(tok::kw_decltype, tok::kw_typeof)) 365 Contexts.back().IsExpression = true; 366 } 367 } 368 369 if (StartsObjCMethodExpr) { 370 Contexts.back().ColonIsObjCMethodExpr = true; 371 OpeningParen.setType(TT_ObjCMethodExpr); 372 } 373 374 // MightBeFunctionType and ProbablyFunctionType are used for 375 // function pointer and reference types as well as Objective-C 376 // block types: 377 // 378 // void (*FunctionPointer)(void); 379 // void (&FunctionReference)(void); 380 // void (&&FunctionReference)(void); 381 // void (^ObjCBlock)(void); 382 bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression; 383 bool ProbablyFunctionType = 384 CurrentToken->isOneOf(tok::star, tok::amp, tok::ampamp, tok::caret); 385 bool HasMultipleLines = false; 386 bool HasMultipleParametersOnALine = false; 387 bool MightBeObjCForRangeLoop = 388 OpeningParen.Previous && OpeningParen.Previous->is(tok::kw_for); 389 FormatToken *PossibleObjCForInToken = nullptr; 390 while (CurrentToken) { 391 // LookForDecls is set when "if (" has been seen. Check for 392 // 'identifier' '*' 'identifier' followed by not '=' -- this 393 // '*' has to be a binary operator but determineStarAmpUsage() will 394 // categorize it as an unary operator, so set the right type here. 395 if (LookForDecls && CurrentToken->Next) { 396 FormatToken *Prev = CurrentToken->getPreviousNonComment(); 397 if (Prev) { 398 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 399 FormatToken *Next = CurrentToken->Next; 400 if (PrevPrev && PrevPrev->is(tok::identifier) && 401 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) && 402 CurrentToken->is(tok::identifier) && 403 !Next->isOneOf(tok::equal, tok::l_brace)) { 404 Prev->setType(TT_BinaryOperator); 405 LookForDecls = false; 406 } 407 } 408 } 409 410 if (CurrentToken->Previous->is(TT_PointerOrReference) && 411 CurrentToken->Previous->Previous->isOneOf(tok::l_paren, 412 tok::coloncolon)) { 413 ProbablyFunctionType = true; 414 } 415 if (CurrentToken->is(tok::comma)) 416 MightBeFunctionType = false; 417 if (CurrentToken->Previous->is(TT_BinaryOperator)) 418 Contexts.back().IsExpression = true; 419 if (CurrentToken->is(tok::r_paren)) { 420 if (OpeningParen.isNot(TT_CppCastLParen) && MightBeFunctionType && 421 ProbablyFunctionType && CurrentToken->Next && 422 (CurrentToken->Next->is(tok::l_paren) || 423 (CurrentToken->Next->is(tok::l_square) && 424 Line.MustBeDeclaration))) { 425 OpeningParen.setType(OpeningParen.Next->is(tok::caret) 426 ? TT_ObjCBlockLParen 427 : TT_FunctionTypeLParen); 428 } 429 OpeningParen.MatchingParen = CurrentToken; 430 CurrentToken->MatchingParen = &OpeningParen; 431 432 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) && 433 OpeningParen.Previous && OpeningParen.Previous->is(tok::l_paren)) { 434 // Detect the case where macros are used to generate lambdas or 435 // function bodies, e.g.: 436 // auto my_lambda = MACRO((Type *type, int i) { .. body .. }); 437 for (FormatToken *Tok = &OpeningParen; Tok != CurrentToken; 438 Tok = Tok->Next) { 439 if (Tok->is(TT_BinaryOperator) && 440 Tok->isOneOf(tok::star, tok::amp, tok::ampamp)) { 441 Tok->setType(TT_PointerOrReference); 442 } 443 } 444 } 445 446 if (StartsObjCMethodExpr) { 447 CurrentToken->setType(TT_ObjCMethodExpr); 448 if (Contexts.back().FirstObjCSelectorName) { 449 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 450 Contexts.back().LongestObjCSelectorName; 451 } 452 } 453 454 if (OpeningParen.is(TT_AttributeParen)) 455 CurrentToken->setType(TT_AttributeParen); 456 if (OpeningParen.is(TT_TypeDeclarationParen)) 457 CurrentToken->setType(TT_TypeDeclarationParen); 458 if (OpeningParen.Previous && 459 OpeningParen.Previous->is(TT_JavaAnnotation)) { 460 CurrentToken->setType(TT_JavaAnnotation); 461 } 462 if (OpeningParen.Previous && 463 OpeningParen.Previous->is(TT_LeadingJavaAnnotation)) { 464 CurrentToken->setType(TT_LeadingJavaAnnotation); 465 } 466 if (OpeningParen.Previous && 467 OpeningParen.Previous->is(TT_AttributeSquare)) { 468 CurrentToken->setType(TT_AttributeSquare); 469 } 470 471 if (!HasMultipleLines) 472 OpeningParen.setPackingKind(PPK_Inconclusive); 473 else if (HasMultipleParametersOnALine) 474 OpeningParen.setPackingKind(PPK_BinPacked); 475 else 476 OpeningParen.setPackingKind(PPK_OnePerLine); 477 478 next(); 479 return true; 480 } 481 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace)) 482 return false; 483 484 if (CurrentToken->is(tok::l_brace) && OpeningParen.is(TT_ObjCBlockLParen)) 485 OpeningParen.setType(TT_Unknown); 486 if (CurrentToken->is(tok::comma) && CurrentToken->Next && 487 !CurrentToken->Next->HasUnescapedNewline && 488 !CurrentToken->Next->isTrailingComment()) { 489 HasMultipleParametersOnALine = true; 490 } 491 bool ProbablyFunctionTypeLParen = 492 (CurrentToken->is(tok::l_paren) && CurrentToken->Next && 493 CurrentToken->Next->isOneOf(tok::star, tok::amp, tok::caret)); 494 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) || 495 CurrentToken->Previous->isSimpleTypeSpecifier()) && 496 !(CurrentToken->is(tok::l_brace) || 497 (CurrentToken->is(tok::l_paren) && !ProbablyFunctionTypeLParen))) { 498 Contexts.back().IsExpression = false; 499 } 500 if (CurrentToken->isOneOf(tok::semi, tok::colon)) { 501 MightBeObjCForRangeLoop = false; 502 if (PossibleObjCForInToken) { 503 PossibleObjCForInToken->setType(TT_Unknown); 504 PossibleObjCForInToken = nullptr; 505 } 506 } 507 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) { 508 PossibleObjCForInToken = CurrentToken; 509 PossibleObjCForInToken->setType(TT_ObjCForIn); 510 } 511 // When we discover a 'new', we set CanBeExpression to 'false' in order to 512 // parse the type correctly. Reset that after a comma. 513 if (CurrentToken->is(tok::comma)) 514 Contexts.back().CanBeExpression = true; 515 516 FormatToken *Tok = CurrentToken; 517 if (!consumeToken()) 518 return false; 519 updateParameterCount(&OpeningParen, Tok); 520 if (CurrentToken && CurrentToken->HasUnescapedNewline) 521 HasMultipleLines = true; 522 } 523 return false; 524 } 525 526 bool isCSharpAttributeSpecifier(const FormatToken &Tok) { 527 if (!Style.isCSharp()) 528 return false; 529 530 // `identifier[i]` is not an attribute. 531 if (Tok.Previous && Tok.Previous->is(tok::identifier)) 532 return false; 533 534 // Chains of [] in `identifier[i][j][k]` are not attributes. 535 if (Tok.Previous && Tok.Previous->is(tok::r_square)) { 536 auto *MatchingParen = Tok.Previous->MatchingParen; 537 if (!MatchingParen || MatchingParen->is(TT_ArraySubscriptLSquare)) 538 return false; 539 } 540 541 const FormatToken *AttrTok = Tok.Next; 542 if (!AttrTok) 543 return false; 544 545 // Just an empty declaration e.g. string []. 546 if (AttrTok->is(tok::r_square)) 547 return false; 548 549 // Move along the tokens inbetween the '[' and ']' e.g. [STAThread]. 550 while (AttrTok && AttrTok->isNot(tok::r_square)) 551 AttrTok = AttrTok->Next; 552 553 if (!AttrTok) 554 return false; 555 556 // Allow an attribute to be the only content of a file. 557 AttrTok = AttrTok->Next; 558 if (!AttrTok) 559 return true; 560 561 // Limit this to being an access modifier that follows. 562 if (AttrTok->isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 563 tok::comment, tok::kw_class, tok::kw_static, 564 tok::l_square, Keywords.kw_internal)) { 565 return true; 566 } 567 568 // incase its a [XXX] retval func(.... 569 if (AttrTok->Next && 570 AttrTok->Next->startsSequence(tok::identifier, tok::l_paren)) { 571 return true; 572 } 573 574 return false; 575 } 576 577 bool isCpp11AttributeSpecifier(const FormatToken &Tok) { 578 return isCppAttribute(Style.isCpp(), Tok); 579 } 580 581 bool parseSquare() { 582 if (!CurrentToken) 583 return false; 584 585 // A '[' could be an index subscript (after an identifier or after 586 // ')' or ']'), it could be the start of an Objective-C method 587 // expression, it could the start of an Objective-C array literal, 588 // or it could be a C++ attribute specifier [[foo::bar]]. 589 FormatToken *Left = CurrentToken->Previous; 590 Left->ParentBracket = Contexts.back().ContextKind; 591 FormatToken *Parent = Left->getPreviousNonComment(); 592 593 // Cases where '>' is followed by '['. 594 // In C++, this can happen either in array of templates (foo<int>[10]) 595 // or when array is a nested template type (unique_ptr<type1<type2>[]>). 596 bool CppArrayTemplates = 597 Style.isCpp() && Parent && Parent->is(TT_TemplateCloser) && 598 (Contexts.back().CanBeExpression || Contexts.back().IsExpression || 599 Contexts.back().ContextType == Context::TemplateArgument); 600 601 const bool IsInnerSquare = Contexts.back().InCpp11AttributeSpecifier; 602 const bool IsCpp11AttributeSpecifier = 603 isCpp11AttributeSpecifier(*Left) || IsInnerSquare; 604 605 // Treat C# Attributes [STAThread] much like C++ attributes [[...]]. 606 bool IsCSharpAttributeSpecifier = 607 isCSharpAttributeSpecifier(*Left) || 608 Contexts.back().InCSharpAttributeSpecifier; 609 610 bool InsideInlineASM = Line.startsWith(tok::kw_asm); 611 bool IsCppStructuredBinding = Left->isCppStructuredBinding(Style); 612 bool StartsObjCMethodExpr = 613 !IsCppStructuredBinding && !InsideInlineASM && !CppArrayTemplates && 614 Style.isCpp() && !IsCpp11AttributeSpecifier && 615 !IsCSharpAttributeSpecifier && Contexts.back().CanBeExpression && 616 Left->isNot(TT_LambdaLSquare) && 617 !CurrentToken->isOneOf(tok::l_brace, tok::r_square) && 618 (!Parent || 619 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren, 620 tok::kw_return, tok::kw_throw) || 621 Parent->isUnaryOperator() || 622 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 623 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) || 624 (getBinOpPrecedence(Parent->Tok.getKind(), true, true) > 625 prec::Unknown)); 626 bool ColonFound = false; 627 628 unsigned BindingIncrease = 1; 629 if (IsCppStructuredBinding) { 630 Left->setType(TT_StructuredBindingLSquare); 631 } else if (Left->is(TT_Unknown)) { 632 if (StartsObjCMethodExpr) { 633 Left->setType(TT_ObjCMethodExpr); 634 } else if (InsideInlineASM) { 635 Left->setType(TT_InlineASMSymbolicNameLSquare); 636 } else if (IsCpp11AttributeSpecifier) { 637 Left->setType(TT_AttributeSquare); 638 if (!IsInnerSquare && Left->Previous) 639 Left->Previous->EndsCppAttributeGroup = false; 640 } else if (Style.isJavaScript() && Parent && 641 Contexts.back().ContextKind == tok::l_brace && 642 Parent->isOneOf(tok::l_brace, tok::comma)) { 643 Left->setType(TT_JsComputedPropertyName); 644 } else if (Style.isCpp() && Contexts.back().ContextKind == tok::l_brace && 645 Parent && Parent->isOneOf(tok::l_brace, tok::comma)) { 646 Left->setType(TT_DesignatedInitializerLSquare); 647 } else if (IsCSharpAttributeSpecifier) { 648 Left->setType(TT_AttributeSquare); 649 } else if (CurrentToken->is(tok::r_square) && Parent && 650 Parent->is(TT_TemplateCloser)) { 651 Left->setType(TT_ArraySubscriptLSquare); 652 } else if (Style.Language == FormatStyle::LK_Proto || 653 Style.Language == FormatStyle::LK_TextProto) { 654 // Square braces in LK_Proto can either be message field attributes: 655 // 656 // optional Aaa aaa = 1 [ 657 // (aaa) = aaa 658 // ]; 659 // 660 // extensions 123 [ 661 // (aaa) = aaa 662 // ]; 663 // 664 // or text proto extensions (in options): 665 // 666 // option (Aaa.options) = { 667 // [type.type/type] { 668 // key: value 669 // } 670 // } 671 // 672 // or repeated fields (in options): 673 // 674 // option (Aaa.options) = { 675 // keys: [ 1, 2, 3 ] 676 // } 677 // 678 // In the first and the third case we want to spread the contents inside 679 // the square braces; in the second we want to keep them inline. 680 Left->setType(TT_ArrayInitializerLSquare); 681 if (!Left->endsSequence(tok::l_square, tok::numeric_constant, 682 tok::equal) && 683 !Left->endsSequence(tok::l_square, tok::numeric_constant, 684 tok::identifier) && 685 !Left->endsSequence(tok::l_square, tok::colon, TT_SelectorName)) { 686 Left->setType(TT_ProtoExtensionLSquare); 687 BindingIncrease = 10; 688 } 689 } else if (!CppArrayTemplates && Parent && 690 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at, 691 tok::comma, tok::l_paren, tok::l_square, 692 tok::question, tok::colon, tok::kw_return, 693 // Should only be relevant to JavaScript: 694 tok::kw_default)) { 695 Left->setType(TT_ArrayInitializerLSquare); 696 } else { 697 BindingIncrease = 10; 698 Left->setType(TT_ArraySubscriptLSquare); 699 } 700 } 701 702 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease); 703 Contexts.back().IsExpression = true; 704 if (Style.isJavaScript() && Parent && Parent->is(TT_JsTypeColon)) 705 Contexts.back().IsExpression = false; 706 707 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr; 708 Contexts.back().InCpp11AttributeSpecifier = IsCpp11AttributeSpecifier; 709 Contexts.back().InCSharpAttributeSpecifier = IsCSharpAttributeSpecifier; 710 711 while (CurrentToken) { 712 if (CurrentToken->is(tok::r_square)) { 713 if (IsCpp11AttributeSpecifier) { 714 CurrentToken->setType(TT_AttributeSquare); 715 if (!IsInnerSquare) 716 CurrentToken->EndsCppAttributeGroup = true; 717 } 718 if (IsCSharpAttributeSpecifier) { 719 CurrentToken->setType(TT_AttributeSquare); 720 } else if (((CurrentToken->Next && 721 CurrentToken->Next->is(tok::l_paren)) || 722 (CurrentToken->Previous && 723 CurrentToken->Previous->Previous == Left)) && 724 Left->is(TT_ObjCMethodExpr)) { 725 // An ObjC method call is rarely followed by an open parenthesis. It 726 // also can't be composed of just one token, unless it's a macro that 727 // will be expanded to more tokens. 728 // FIXME: Do we incorrectly label ":" with this? 729 StartsObjCMethodExpr = false; 730 Left->setType(TT_Unknown); 731 } 732 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) { 733 CurrentToken->setType(TT_ObjCMethodExpr); 734 // If we haven't seen a colon yet, make sure the last identifier 735 // before the r_square is tagged as a selector name component. 736 if (!ColonFound && CurrentToken->Previous && 737 CurrentToken->Previous->is(TT_Unknown) && 738 canBeObjCSelectorComponent(*CurrentToken->Previous)) { 739 CurrentToken->Previous->setType(TT_SelectorName); 740 } 741 // determineStarAmpUsage() thinks that '*' '[' is allocating an 742 // array of pointers, but if '[' starts a selector then '*' is a 743 // binary operator. 744 if (Parent && Parent->is(TT_PointerOrReference)) 745 Parent->overwriteFixedType(TT_BinaryOperator); 746 } 747 // An arrow after an ObjC method expression is not a lambda arrow. 748 if (CurrentToken->getType() == TT_ObjCMethodExpr && 749 CurrentToken->Next && CurrentToken->Next->is(TT_LambdaArrow)) { 750 CurrentToken->Next->overwriteFixedType(TT_Unknown); 751 } 752 Left->MatchingParen = CurrentToken; 753 CurrentToken->MatchingParen = Left; 754 // FirstObjCSelectorName is set when a colon is found. This does 755 // not work, however, when the method has no parameters. 756 // Here, we set FirstObjCSelectorName when the end of the method call is 757 // reached, in case it was not set already. 758 if (!Contexts.back().FirstObjCSelectorName) { 759 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 760 if (Previous && Previous->is(TT_SelectorName)) { 761 Previous->ObjCSelectorNameParts = 1; 762 Contexts.back().FirstObjCSelectorName = Previous; 763 } 764 } else { 765 Left->ParameterCount = 766 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 767 } 768 if (Contexts.back().FirstObjCSelectorName) { 769 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 770 Contexts.back().LongestObjCSelectorName; 771 if (Left->BlockParameterCount > 1) 772 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0; 773 } 774 next(); 775 return true; 776 } 777 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace)) 778 return false; 779 if (CurrentToken->is(tok::colon)) { 780 if (IsCpp11AttributeSpecifier && 781 CurrentToken->endsSequence(tok::colon, tok::identifier, 782 tok::kw_using)) { 783 // Remember that this is a [[using ns: foo]] C++ attribute, so we 784 // don't add a space before the colon (unlike other colons). 785 CurrentToken->setType(TT_AttributeColon); 786 } else if (!Style.isVerilog() && !Line.InPragmaDirective && 787 Left->isOneOf(TT_ArraySubscriptLSquare, 788 TT_DesignatedInitializerLSquare)) { 789 Left->setType(TT_ObjCMethodExpr); 790 StartsObjCMethodExpr = true; 791 Contexts.back().ColonIsObjCMethodExpr = true; 792 if (Parent && Parent->is(tok::r_paren)) { 793 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 794 Parent->setType(TT_CastRParen); 795 } 796 } 797 ColonFound = true; 798 } 799 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) && 800 !ColonFound) { 801 Left->setType(TT_ArrayInitializerLSquare); 802 } 803 FormatToken *Tok = CurrentToken; 804 if (!consumeToken()) 805 return false; 806 updateParameterCount(Left, Tok); 807 } 808 return false; 809 } 810 811 bool couldBeInStructArrayInitializer() const { 812 if (Contexts.size() < 2) 813 return false; 814 // We want to back up no more then 2 context levels i.e. 815 // . { { <- 816 const auto End = std::next(Contexts.rbegin(), 2); 817 auto Last = Contexts.rbegin(); 818 unsigned Depth = 0; 819 for (; Last != End; ++Last) 820 if (Last->ContextKind == tok::l_brace) 821 ++Depth; 822 return Depth == 2 && Last->ContextKind != tok::l_brace; 823 } 824 825 bool parseBrace() { 826 if (!CurrentToken) 827 return true; 828 829 assert(CurrentToken->Previous); 830 FormatToken &OpeningBrace = *CurrentToken->Previous; 831 assert(OpeningBrace.is(tok::l_brace)); 832 OpeningBrace.ParentBracket = Contexts.back().ContextKind; 833 834 if (Contexts.back().CaretFound) 835 OpeningBrace.overwriteFixedType(TT_ObjCBlockLBrace); 836 Contexts.back().CaretFound = false; 837 838 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1); 839 Contexts.back().ColonIsDictLiteral = true; 840 if (OpeningBrace.is(BK_BracedInit)) 841 Contexts.back().IsExpression = true; 842 if (Style.isJavaScript() && OpeningBrace.Previous && 843 OpeningBrace.Previous->is(TT_JsTypeColon)) { 844 Contexts.back().IsExpression = false; 845 } 846 847 unsigned CommaCount = 0; 848 while (CurrentToken) { 849 if (CurrentToken->is(tok::r_brace)) { 850 assert(OpeningBrace.Optional == CurrentToken->Optional); 851 OpeningBrace.MatchingParen = CurrentToken; 852 CurrentToken->MatchingParen = &OpeningBrace; 853 if (Style.AlignArrayOfStructures != FormatStyle::AIAS_None) { 854 if (OpeningBrace.ParentBracket == tok::l_brace && 855 couldBeInStructArrayInitializer() && CommaCount > 0) { 856 Contexts.back().ContextType = Context::StructArrayInitializer; 857 } 858 } 859 next(); 860 return true; 861 } 862 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 863 return false; 864 updateParameterCount(&OpeningBrace, CurrentToken); 865 if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) { 866 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 867 if (Previous->is(TT_JsTypeOptionalQuestion)) 868 Previous = Previous->getPreviousNonComment(); 869 if ((CurrentToken->is(tok::colon) && 870 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) || 871 Style.Language == FormatStyle::LK_Proto || 872 Style.Language == FormatStyle::LK_TextProto) { 873 OpeningBrace.setType(TT_DictLiteral); 874 if (Previous->Tok.getIdentifierInfo() || 875 Previous->is(tok::string_literal)) { 876 Previous->setType(TT_SelectorName); 877 } 878 } 879 if (CurrentToken->is(tok::colon) && OpeningBrace.is(TT_Unknown)) 880 OpeningBrace.setType(TT_DictLiteral); 881 else if (Style.isJavaScript()) 882 OpeningBrace.overwriteFixedType(TT_DictLiteral); 883 } 884 if (CurrentToken->is(tok::comma)) { 885 if (Style.isJavaScript()) 886 OpeningBrace.overwriteFixedType(TT_DictLiteral); 887 ++CommaCount; 888 } 889 if (!consumeToken()) 890 return false; 891 } 892 return true; 893 } 894 895 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 896 // For ObjC methods, the number of parameters is calculated differently as 897 // method declarations have a different structure (the parameters are not 898 // inside a bracket scope). 899 if (Current->is(tok::l_brace) && Current->is(BK_Block)) 900 ++Left->BlockParameterCount; 901 if (Current->is(tok::comma)) { 902 ++Left->ParameterCount; 903 if (!Left->Role) 904 Left->Role.reset(new CommaSeparatedList(Style)); 905 Left->Role->CommaFound(Current); 906 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 907 Left->ParameterCount = 1; 908 } 909 } 910 911 bool parseConditional() { 912 while (CurrentToken) { 913 if (CurrentToken->is(tok::colon)) { 914 CurrentToken->setType(TT_ConditionalExpr); 915 next(); 916 return true; 917 } 918 if (!consumeToken()) 919 return false; 920 } 921 return false; 922 } 923 924 bool parseTemplateDeclaration() { 925 if (CurrentToken && CurrentToken->is(tok::less)) { 926 CurrentToken->setType(TT_TemplateOpener); 927 next(); 928 if (!parseAngle()) 929 return false; 930 if (CurrentToken) 931 CurrentToken->Previous->ClosesTemplateDeclaration = true; 932 return true; 933 } 934 return false; 935 } 936 937 bool consumeToken() { 938 FormatToken *Tok = CurrentToken; 939 next(); 940 // In Verilog primitives' state tables, `:`, `?`, and `-` aren't normal 941 // operators. 942 if (Tok->is(TT_VerilogTableItem)) 943 return true; 944 switch (Tok->Tok.getKind()) { 945 case tok::plus: 946 case tok::minus: 947 if (!Tok->Previous && Line.MustBeDeclaration) 948 Tok->setType(TT_ObjCMethodSpecifier); 949 break; 950 case tok::colon: 951 if (!Tok->Previous) 952 return false; 953 // Colons from ?: are handled in parseConditional(). 954 if (Style.isJavaScript()) { 955 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop 956 (Contexts.size() == 1 && // switch/case labels 957 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) || 958 Contexts.back().ContextKind == tok::l_paren || // function params 959 Contexts.back().ContextKind == tok::l_square || // array type 960 (!Contexts.back().IsExpression && 961 Contexts.back().ContextKind == tok::l_brace) || // object type 962 (Contexts.size() == 1 && 963 Line.MustBeDeclaration)) { // method/property declaration 964 Contexts.back().IsExpression = false; 965 Tok->setType(TT_JsTypeColon); 966 break; 967 } 968 } else if (Style.isCSharp()) { 969 if (Contexts.back().InCSharpAttributeSpecifier) { 970 Tok->setType(TT_AttributeColon); 971 break; 972 } 973 if (Contexts.back().ContextKind == tok::l_paren) { 974 Tok->setType(TT_CSharpNamedArgumentColon); 975 break; 976 } 977 } else if (Style.isVerilog() && Tok->isNot(TT_BinaryOperator)) { 978 // The distribution weight operators are labeled 979 // TT_BinaryOperator by the lexer. 980 if (Keywords.isVerilogEnd(*Tok->Previous) || 981 Keywords.isVerilogBegin(*Tok->Previous)) { 982 Tok->setType(TT_VerilogBlockLabelColon); 983 } else if (Contexts.back().ContextKind == tok::l_square) { 984 Tok->setType(TT_BitFieldColon); 985 } else if (Contexts.back().ColonIsDictLiteral) { 986 Tok->setType(TT_DictLiteral); 987 } else if (Contexts.size() == 1) { 988 // In Verilog a case label doesn't have the case keyword. We 989 // assume a colon following an expression is a case label. 990 // Colons from ?: are annotated in parseConditional(). 991 Tok->setType(TT_GotoLabelColon); 992 if (Line.Level > 1 || (!Line.InPPDirective && Line.Level > 0)) 993 --Line.Level; 994 } 995 break; 996 } 997 if (Line.First->isOneOf(Keywords.kw_module, Keywords.kw_import) || 998 Line.First->startsSequence(tok::kw_export, Keywords.kw_module) || 999 Line.First->startsSequence(tok::kw_export, Keywords.kw_import)) { 1000 Tok->setType(TT_ModulePartitionColon); 1001 } else if (Contexts.back().ColonIsDictLiteral || 1002 Style.Language == FormatStyle::LK_Proto || 1003 Style.Language == FormatStyle::LK_TextProto) { 1004 Tok->setType(TT_DictLiteral); 1005 if (Style.Language == FormatStyle::LK_TextProto) { 1006 if (FormatToken *Previous = Tok->getPreviousNonComment()) 1007 Previous->setType(TT_SelectorName); 1008 } 1009 } else if (Contexts.back().ColonIsObjCMethodExpr || 1010 Line.startsWith(TT_ObjCMethodSpecifier)) { 1011 Tok->setType(TT_ObjCMethodExpr); 1012 const FormatToken *BeforePrevious = Tok->Previous->Previous; 1013 // Ensure we tag all identifiers in method declarations as 1014 // TT_SelectorName. 1015 bool UnknownIdentifierInMethodDeclaration = 1016 Line.startsWith(TT_ObjCMethodSpecifier) && 1017 Tok->Previous->is(tok::identifier) && Tok->Previous->is(TT_Unknown); 1018 if (!BeforePrevious || 1019 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 1020 !(BeforePrevious->is(TT_CastRParen) || 1021 (BeforePrevious->is(TT_ObjCMethodExpr) && 1022 BeforePrevious->is(tok::colon))) || 1023 BeforePrevious->is(tok::r_square) || 1024 Contexts.back().LongestObjCSelectorName == 0 || 1025 UnknownIdentifierInMethodDeclaration) { 1026 Tok->Previous->setType(TT_SelectorName); 1027 if (!Contexts.back().FirstObjCSelectorName) { 1028 Contexts.back().FirstObjCSelectorName = Tok->Previous; 1029 } else if (Tok->Previous->ColumnWidth > 1030 Contexts.back().LongestObjCSelectorName) { 1031 Contexts.back().LongestObjCSelectorName = 1032 Tok->Previous->ColumnWidth; 1033 } 1034 Tok->Previous->ParameterIndex = 1035 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 1036 ++Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 1037 } 1038 } else if (Contexts.back().ColonIsForRangeExpr) { 1039 Tok->setType(TT_RangeBasedForLoopColon); 1040 } else if (Contexts.back().ContextType == Context::C11GenericSelection) { 1041 Tok->setType(TT_GenericSelectionColon); 1042 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 1043 Tok->setType(TT_BitFieldColon); 1044 } else if (Contexts.size() == 1 && 1045 !Line.First->isOneOf(tok::kw_enum, tok::kw_case, 1046 tok::kw_default)) { 1047 FormatToken *Prev = Tok->getPreviousNonComment(); 1048 if (!Prev) 1049 break; 1050 if (Prev->isOneOf(tok::r_paren, tok::kw_noexcept) || 1051 Prev->ClosesRequiresClause) { 1052 Tok->setType(TT_CtorInitializerColon); 1053 } else if (Prev->is(tok::kw_try)) { 1054 // Member initializer list within function try block. 1055 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 1056 if (!PrevPrev) 1057 break; 1058 if (PrevPrev && PrevPrev->isOneOf(tok::r_paren, tok::kw_noexcept)) 1059 Tok->setType(TT_CtorInitializerColon); 1060 } else { 1061 Tok->setType(TT_InheritanceColon); 1062 } 1063 } else if (canBeObjCSelectorComponent(*Tok->Previous) && Tok->Next && 1064 (Tok->Next->isOneOf(tok::r_paren, tok::comma) || 1065 (canBeObjCSelectorComponent(*Tok->Next) && Tok->Next->Next && 1066 Tok->Next->Next->is(tok::colon)))) { 1067 // This handles a special macro in ObjC code where selectors including 1068 // the colon are passed as macro arguments. 1069 Tok->setType(TT_ObjCMethodExpr); 1070 } else if (Contexts.back().ContextKind == tok::l_paren && 1071 !Line.InPragmaDirective) { 1072 Tok->setType(TT_InlineASMColon); 1073 } 1074 break; 1075 case tok::pipe: 1076 case tok::amp: 1077 // | and & in declarations/type expressions represent union and 1078 // intersection types, respectively. 1079 if (Style.isJavaScript() && !Contexts.back().IsExpression) 1080 Tok->setType(TT_JsTypeOperator); 1081 break; 1082 case tok::kw_if: 1083 if (CurrentToken && 1084 CurrentToken->isOneOf(tok::kw_constexpr, tok::identifier)) { 1085 next(); 1086 } 1087 [[fallthrough]]; 1088 case tok::kw_while: 1089 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 1090 next(); 1091 if (!parseParens(/*LookForDecls=*/true)) 1092 return false; 1093 } 1094 break; 1095 case tok::kw_for: 1096 if (Style.isJavaScript()) { 1097 // x.for and {for: ...} 1098 if ((Tok->Previous && Tok->Previous->is(tok::period)) || 1099 (Tok->Next && Tok->Next->is(tok::colon))) { 1100 break; 1101 } 1102 // JS' for await ( ... 1103 if (CurrentToken && CurrentToken->is(Keywords.kw_await)) 1104 next(); 1105 } 1106 if (Style.isCpp() && CurrentToken && CurrentToken->is(tok::kw_co_await)) 1107 next(); 1108 Contexts.back().ColonIsForRangeExpr = true; 1109 if (!CurrentToken || CurrentToken->isNot(tok::l_paren)) 1110 return false; 1111 next(); 1112 if (!parseParens()) 1113 return false; 1114 break; 1115 case tok::l_paren: 1116 // When faced with 'operator()()', the kw_operator handler incorrectly 1117 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make 1118 // the first two parens OverloadedOperators and the second l_paren an 1119 // OverloadedOperatorLParen. 1120 if (Tok->Previous && Tok->Previous->is(tok::r_paren) && 1121 Tok->Previous->MatchingParen && 1122 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) { 1123 Tok->Previous->setType(TT_OverloadedOperator); 1124 Tok->Previous->MatchingParen->setType(TT_OverloadedOperator); 1125 Tok->setType(TT_OverloadedOperatorLParen); 1126 } 1127 1128 if (!parseParens()) 1129 return false; 1130 if (Line.MustBeDeclaration && Contexts.size() == 1 && 1131 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) && 1132 !Tok->isOneOf(TT_TypeDeclarationParen, TT_RequiresExpressionLParen) && 1133 (!Tok->Previous || 1134 !Tok->Previous->isOneOf(tok::kw___attribute, TT_RequiresClause, 1135 TT_LeadingJavaAnnotation))) { 1136 Line.MightBeFunctionDecl = true; 1137 } 1138 break; 1139 case tok::l_square: 1140 if (!parseSquare()) 1141 return false; 1142 break; 1143 case tok::l_brace: 1144 if (Style.Language == FormatStyle::LK_TextProto) { 1145 FormatToken *Previous = Tok->getPreviousNonComment(); 1146 if (Previous && Previous->getType() != TT_DictLiteral) 1147 Previous->setType(TT_SelectorName); 1148 } 1149 if (!parseBrace()) 1150 return false; 1151 break; 1152 case tok::less: 1153 if (parseAngle()) { 1154 Tok->setType(TT_TemplateOpener); 1155 // In TT_Proto, we must distignuish between: 1156 // map<key, value> 1157 // msg < item: data > 1158 // msg: < item: data > 1159 // In TT_TextProto, map<key, value> does not occur. 1160 if (Style.Language == FormatStyle::LK_TextProto || 1161 (Style.Language == FormatStyle::LK_Proto && Tok->Previous && 1162 Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) { 1163 Tok->setType(TT_DictLiteral); 1164 FormatToken *Previous = Tok->getPreviousNonComment(); 1165 if (Previous && Previous->getType() != TT_DictLiteral) 1166 Previous->setType(TT_SelectorName); 1167 } 1168 } else { 1169 Tok->setType(TT_BinaryOperator); 1170 NonTemplateLess.insert(Tok); 1171 CurrentToken = Tok; 1172 next(); 1173 } 1174 break; 1175 case tok::r_paren: 1176 case tok::r_square: 1177 return false; 1178 case tok::r_brace: 1179 // Lines can start with '}'. 1180 if (Tok->Previous) 1181 return false; 1182 break; 1183 case tok::greater: 1184 if (Style.Language != FormatStyle::LK_TextProto) 1185 Tok->setType(TT_BinaryOperator); 1186 if (Tok->Previous && Tok->Previous->is(TT_TemplateCloser)) 1187 Tok->SpacesRequiredBefore = 1; 1188 break; 1189 case tok::kw_operator: 1190 if (Style.Language == FormatStyle::LK_TextProto || 1191 Style.Language == FormatStyle::LK_Proto) { 1192 break; 1193 } 1194 while (CurrentToken && 1195 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 1196 if (CurrentToken->isOneOf(tok::star, tok::amp)) 1197 CurrentToken->setType(TT_PointerOrReference); 1198 consumeToken(); 1199 if (!CurrentToken) 1200 continue; 1201 if (CurrentToken->is(tok::comma) && 1202 CurrentToken->Previous->isNot(tok::kw_operator)) { 1203 break; 1204 } 1205 if (CurrentToken->Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator, 1206 tok::comma, tok::star, tok::arrow, 1207 tok::amp, tok::ampamp) || 1208 // User defined literal. 1209 CurrentToken->Previous->TokenText.startswith("\"\"")) { 1210 CurrentToken->Previous->setType(TT_OverloadedOperator); 1211 } 1212 } 1213 if (CurrentToken && CurrentToken->is(tok::l_paren)) 1214 CurrentToken->setType(TT_OverloadedOperatorLParen); 1215 if (CurrentToken && CurrentToken->Previous->is(TT_BinaryOperator)) 1216 CurrentToken->Previous->setType(TT_OverloadedOperator); 1217 break; 1218 case tok::question: 1219 if (Style.isJavaScript() && Tok->Next && 1220 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren, 1221 tok::r_brace)) { 1222 // Question marks before semicolons, colons, etc. indicate optional 1223 // types (fields, parameters), e.g. 1224 // function(x?: string, y?) {...} 1225 // class X { y?; } 1226 Tok->setType(TT_JsTypeOptionalQuestion); 1227 break; 1228 } 1229 // Declarations cannot be conditional expressions, this can only be part 1230 // of a type declaration. 1231 if (Line.MustBeDeclaration && !Contexts.back().IsExpression && 1232 Style.isJavaScript()) { 1233 break; 1234 } 1235 if (Style.isCSharp()) { 1236 // `Type?)`, `Type?>`, `Type? name;` and `Type? name =` can only be 1237 // nullable types. 1238 // Line.MustBeDeclaration will be true for `Type? name;`. 1239 if ((!Contexts.back().IsExpression && Line.MustBeDeclaration) || 1240 (Tok->Next && Tok->Next->isOneOf(tok::r_paren, tok::greater)) || 1241 (Tok->Next && Tok->Next->is(tok::identifier) && Tok->Next->Next && 1242 Tok->Next->Next->is(tok::equal))) { 1243 Tok->setType(TT_CSharpNullable); 1244 break; 1245 } 1246 } 1247 parseConditional(); 1248 break; 1249 case tok::kw_template: 1250 parseTemplateDeclaration(); 1251 break; 1252 case tok::comma: 1253 switch (Contexts.back().ContextType) { 1254 case Context::CtorInitializer: 1255 Tok->setType(TT_CtorInitializerComma); 1256 break; 1257 case Context::InheritanceList: 1258 Tok->setType(TT_InheritanceComma); 1259 break; 1260 default: 1261 if (Contexts.back().FirstStartOfName && 1262 (Contexts.size() == 1 || startsWithInitStatement(Line))) { 1263 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 1264 Line.IsMultiVariableDeclStmt = true; 1265 } 1266 break; 1267 } 1268 if (Contexts.back().ContextType == Context::ForEachMacro) 1269 Contexts.back().IsExpression = true; 1270 break; 1271 case tok::kw_default: 1272 // Unindent case labels. 1273 if (Style.isVerilog() && Keywords.isVerilogEndOfLabel(*Tok) && 1274 (Line.Level > 1 || (!Line.InPPDirective && Line.Level > 0))) { 1275 --Line.Level; 1276 } 1277 break; 1278 case tok::identifier: 1279 if (Tok->isOneOf(Keywords.kw___has_include, 1280 Keywords.kw___has_include_next)) { 1281 parseHasInclude(); 1282 } 1283 if (Style.isCSharp() && Tok->is(Keywords.kw_where) && Tok->Next && 1284 Tok->Next->isNot(tok::l_paren)) { 1285 Tok->setType(TT_CSharpGenericTypeConstraint); 1286 parseCSharpGenericTypeConstraint(); 1287 if (Tok->getPreviousNonComment() == nullptr) 1288 Line.IsContinuation = true; 1289 } 1290 break; 1291 case tok::arrow: 1292 if (Tok->isNot(TT_LambdaArrow) && Tok->Previous && 1293 Tok->Previous->is(tok::kw_noexcept)) { 1294 Tok->setType(TT_TrailingReturnArrow); 1295 } 1296 break; 1297 case tok::eof: 1298 if (Style.InsertNewlineAtEOF && Tok->NewlinesBefore == 0) 1299 Tok->NewlinesBefore = 1; 1300 break; 1301 default: 1302 break; 1303 } 1304 return true; 1305 } 1306 1307 void parseCSharpGenericTypeConstraint() { 1308 int OpenAngleBracketsCount = 0; 1309 while (CurrentToken) { 1310 if (CurrentToken->is(tok::less)) { 1311 // parseAngle is too greedy and will consume the whole line. 1312 CurrentToken->setType(TT_TemplateOpener); 1313 ++OpenAngleBracketsCount; 1314 next(); 1315 } else if (CurrentToken->is(tok::greater)) { 1316 CurrentToken->setType(TT_TemplateCloser); 1317 --OpenAngleBracketsCount; 1318 next(); 1319 } else if (CurrentToken->is(tok::comma) && OpenAngleBracketsCount == 0) { 1320 // We allow line breaks after GenericTypeConstraintComma's 1321 // so do not flag commas in Generics as GenericTypeConstraintComma's. 1322 CurrentToken->setType(TT_CSharpGenericTypeConstraintComma); 1323 next(); 1324 } else if (CurrentToken->is(Keywords.kw_where)) { 1325 CurrentToken->setType(TT_CSharpGenericTypeConstraint); 1326 next(); 1327 } else if (CurrentToken->is(tok::colon)) { 1328 CurrentToken->setType(TT_CSharpGenericTypeConstraintColon); 1329 next(); 1330 } else { 1331 next(); 1332 } 1333 } 1334 } 1335 1336 void parseIncludeDirective() { 1337 if (CurrentToken && CurrentToken->is(tok::less)) { 1338 next(); 1339 while (CurrentToken) { 1340 // Mark tokens up to the trailing line comments as implicit string 1341 // literals. 1342 if (CurrentToken->isNot(tok::comment) && 1343 !CurrentToken->TokenText.startswith("//")) { 1344 CurrentToken->setType(TT_ImplicitStringLiteral); 1345 } 1346 next(); 1347 } 1348 } 1349 } 1350 1351 void parseWarningOrError() { 1352 next(); 1353 // We still want to format the whitespace left of the first token of the 1354 // warning or error. 1355 next(); 1356 while (CurrentToken) { 1357 CurrentToken->setType(TT_ImplicitStringLiteral); 1358 next(); 1359 } 1360 } 1361 1362 void parsePragma() { 1363 next(); // Consume "pragma". 1364 if (CurrentToken && 1365 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option, 1366 Keywords.kw_region)) { 1367 bool IsMarkOrRegion = 1368 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_region); 1369 next(); 1370 next(); // Consume first token (so we fix leading whitespace). 1371 while (CurrentToken) { 1372 if (IsMarkOrRegion || CurrentToken->Previous->is(TT_BinaryOperator)) 1373 CurrentToken->setType(TT_ImplicitStringLiteral); 1374 next(); 1375 } 1376 } 1377 } 1378 1379 void parseHasInclude() { 1380 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 1381 return; 1382 next(); // '(' 1383 parseIncludeDirective(); 1384 next(); // ')' 1385 } 1386 1387 LineType parsePreprocessorDirective() { 1388 bool IsFirstToken = CurrentToken->IsFirst; 1389 LineType Type = LT_PreprocessorDirective; 1390 next(); 1391 if (!CurrentToken) 1392 return Type; 1393 1394 if (Style.isJavaScript() && IsFirstToken) { 1395 // JavaScript files can contain shebang lines of the form: 1396 // #!/usr/bin/env node 1397 // Treat these like C++ #include directives. 1398 while (CurrentToken) { 1399 // Tokens cannot be comments here. 1400 CurrentToken->setType(TT_ImplicitStringLiteral); 1401 next(); 1402 } 1403 return LT_ImportStatement; 1404 } 1405 1406 if (CurrentToken->is(tok::numeric_constant)) { 1407 CurrentToken->SpacesRequiredBefore = 1; 1408 return Type; 1409 } 1410 // Hashes in the middle of a line can lead to any strange token 1411 // sequence. 1412 if (!CurrentToken->Tok.getIdentifierInfo()) 1413 return Type; 1414 // In Verilog macro expansions start with a backtick just like preprocessor 1415 // directives. Thus we stop if the word is not a preprocessor directive. 1416 if (Style.isVerilog() && !Keywords.isVerilogPPDirective(*CurrentToken)) 1417 return LT_Invalid; 1418 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 1419 case tok::pp_include: 1420 case tok::pp_include_next: 1421 case tok::pp_import: 1422 next(); 1423 parseIncludeDirective(); 1424 Type = LT_ImportStatement; 1425 break; 1426 case tok::pp_error: 1427 case tok::pp_warning: 1428 parseWarningOrError(); 1429 break; 1430 case tok::pp_pragma: 1431 parsePragma(); 1432 break; 1433 case tok::pp_if: 1434 case tok::pp_elif: 1435 Contexts.back().IsExpression = true; 1436 next(); 1437 parseLine(); 1438 break; 1439 default: 1440 break; 1441 } 1442 while (CurrentToken) { 1443 FormatToken *Tok = CurrentToken; 1444 next(); 1445 if (Tok->is(tok::l_paren)) { 1446 parseParens(); 1447 } else if (Tok->isOneOf(Keywords.kw___has_include, 1448 Keywords.kw___has_include_next)) { 1449 parseHasInclude(); 1450 } 1451 } 1452 return Type; 1453 } 1454 1455 public: 1456 LineType parseLine() { 1457 if (!CurrentToken) 1458 return LT_Invalid; 1459 NonTemplateLess.clear(); 1460 if (!Line.InMacroBody && CurrentToken->is(tok::hash)) { 1461 // We were not yet allowed to use C++17 optional when this was being 1462 // written. So we used LT_Invalid to mark that the line is not a 1463 // preprocessor directive. 1464 auto Type = parsePreprocessorDirective(); 1465 if (Type != LT_Invalid) 1466 return Type; 1467 } 1468 1469 // Directly allow to 'import <string-literal>' to support protocol buffer 1470 // definitions (github.com/google/protobuf) or missing "#" (either way we 1471 // should not break the line). 1472 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 1473 if ((Style.Language == FormatStyle::LK_Java && 1474 CurrentToken->is(Keywords.kw_package)) || 1475 (!Style.isVerilog() && Info && 1476 Info->getPPKeywordID() == tok::pp_import && CurrentToken->Next && 1477 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 1478 tok::kw_static))) { 1479 next(); 1480 parseIncludeDirective(); 1481 return LT_ImportStatement; 1482 } 1483 1484 // If this line starts and ends in '<' and '>', respectively, it is likely 1485 // part of "#define <a/b.h>". 1486 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 1487 parseIncludeDirective(); 1488 return LT_ImportStatement; 1489 } 1490 1491 // In .proto files, top-level options and package statements are very 1492 // similar to import statements and should not be line-wrapped. 1493 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 1494 CurrentToken->isOneOf(Keywords.kw_option, Keywords.kw_package)) { 1495 next(); 1496 if (CurrentToken && CurrentToken->is(tok::identifier)) { 1497 while (CurrentToken) 1498 next(); 1499 return LT_ImportStatement; 1500 } 1501 } 1502 1503 bool KeywordVirtualFound = false; 1504 bool ImportStatement = false; 1505 1506 // import {...} from '...'; 1507 if (Style.isJavaScript() && CurrentToken->is(Keywords.kw_import)) 1508 ImportStatement = true; 1509 1510 while (CurrentToken) { 1511 if (CurrentToken->is(tok::kw_virtual)) 1512 KeywordVirtualFound = true; 1513 if (Style.isJavaScript()) { 1514 // export {...} from '...'; 1515 // An export followed by "from 'some string';" is a re-export from 1516 // another module identified by a URI and is treated as a 1517 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 1518 // Just "export {...};" or "export class ..." should not be treated as 1519 // an import in this sense. 1520 if (Line.First->is(tok::kw_export) && 1521 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 1522 CurrentToken->Next->isStringLiteral()) { 1523 ImportStatement = true; 1524 } 1525 if (isClosureImportStatement(*CurrentToken)) 1526 ImportStatement = true; 1527 } 1528 if (!consumeToken()) 1529 return LT_Invalid; 1530 } 1531 if (KeywordVirtualFound) 1532 return LT_VirtualFunctionDecl; 1533 if (ImportStatement) 1534 return LT_ImportStatement; 1535 1536 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 1537 if (Contexts.back().FirstObjCSelectorName) { 1538 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 1539 Contexts.back().LongestObjCSelectorName; 1540 } 1541 return LT_ObjCMethodDecl; 1542 } 1543 1544 for (const auto &ctx : Contexts) 1545 if (ctx.ContextType == Context::StructArrayInitializer) 1546 return LT_ArrayOfStructInitializer; 1547 1548 return LT_Other; 1549 } 1550 1551 private: 1552 bool isClosureImportStatement(const FormatToken &Tok) { 1553 // FIXME: Closure-library specific stuff should not be hard-coded but be 1554 // configurable. 1555 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 1556 Tok.Next->Next && 1557 (Tok.Next->Next->TokenText == "module" || 1558 Tok.Next->Next->TokenText == "provide" || 1559 Tok.Next->Next->TokenText == "require" || 1560 Tok.Next->Next->TokenText == "requireType" || 1561 Tok.Next->Next->TokenText == "forwardDeclare") && 1562 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 1563 } 1564 1565 void resetTokenMetadata() { 1566 if (!CurrentToken) 1567 return; 1568 1569 // Reset token type in case we have already looked at it and then 1570 // recovered from an error (e.g. failure to find the matching >). 1571 if (!CurrentToken->isTypeFinalized() && 1572 !CurrentToken->isOneOf( 1573 TT_LambdaLSquare, TT_LambdaLBrace, TT_AttributeMacro, TT_IfMacro, 1574 TT_ForEachMacro, TT_TypenameMacro, TT_FunctionLBrace, 1575 TT_ImplicitStringLiteral, TT_InlineASMBrace, TT_FatArrow, 1576 TT_LambdaArrow, TT_NamespaceMacro, TT_OverloadedOperator, 1577 TT_RegexLiteral, TT_TemplateString, TT_ObjCStringLiteral, 1578 TT_UntouchableMacroFunc, TT_StatementAttributeLikeMacro, 1579 TT_FunctionLikeOrFreestandingMacro, TT_ClassLBrace, TT_EnumLBrace, 1580 TT_RecordLBrace, TT_StructLBrace, TT_UnionLBrace, TT_RequiresClause, 1581 TT_RequiresClauseInARequiresExpression, TT_RequiresExpression, 1582 TT_RequiresExpressionLParen, TT_RequiresExpressionLBrace, 1583 TT_CompoundRequirementLBrace, TT_BracedListLBrace)) { 1584 CurrentToken->setType(TT_Unknown); 1585 } 1586 CurrentToken->Role.reset(); 1587 CurrentToken->MatchingParen = nullptr; 1588 CurrentToken->FakeLParens.clear(); 1589 CurrentToken->FakeRParens = 0; 1590 } 1591 1592 void next() { 1593 if (!CurrentToken) 1594 return; 1595 1596 CurrentToken->NestingLevel = Contexts.size() - 1; 1597 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1598 modifyContext(*CurrentToken); 1599 determineTokenType(*CurrentToken); 1600 CurrentToken = CurrentToken->Next; 1601 1602 resetTokenMetadata(); 1603 } 1604 1605 /// A struct to hold information valid in a specific context, e.g. 1606 /// a pair of parenthesis. 1607 struct Context { 1608 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1609 bool IsExpression) 1610 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1611 IsExpression(IsExpression) {} 1612 1613 tok::TokenKind ContextKind; 1614 unsigned BindingStrength; 1615 bool IsExpression; 1616 unsigned LongestObjCSelectorName = 0; 1617 bool ColonIsForRangeExpr = false; 1618 bool ColonIsDictLiteral = false; 1619 bool ColonIsObjCMethodExpr = false; 1620 FormatToken *FirstObjCSelectorName = nullptr; 1621 FormatToken *FirstStartOfName = nullptr; 1622 bool CanBeExpression = true; 1623 bool CaretFound = false; 1624 bool InCpp11AttributeSpecifier = false; 1625 bool InCSharpAttributeSpecifier = false; 1626 enum { 1627 Unknown, 1628 // Like the part after `:` in a constructor. 1629 // Context(...) : IsExpression(IsExpression) 1630 CtorInitializer, 1631 // Like in the parentheses in a foreach. 1632 ForEachMacro, 1633 // Like the inheritance list in a class declaration. 1634 // class Input : public IO 1635 InheritanceList, 1636 // Like in the braced list. 1637 // int x[] = {}; 1638 StructArrayInitializer, 1639 // Like in `static_cast<int>`. 1640 TemplateArgument, 1641 // C11 _Generic selection. 1642 C11GenericSelection, 1643 } ContextType = Unknown; 1644 }; 1645 1646 /// Puts a new \c Context onto the stack \c Contexts for the lifetime 1647 /// of each instance. 1648 struct ScopedContextCreator { 1649 AnnotatingParser &P; 1650 1651 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1652 unsigned Increase) 1653 : P(P) { 1654 P.Contexts.push_back(Context(ContextKind, 1655 P.Contexts.back().BindingStrength + Increase, 1656 P.Contexts.back().IsExpression)); 1657 } 1658 1659 ~ScopedContextCreator() { 1660 if (P.Style.AlignArrayOfStructures != FormatStyle::AIAS_None) { 1661 if (P.Contexts.back().ContextType == Context::StructArrayInitializer) { 1662 P.Contexts.pop_back(); 1663 P.Contexts.back().ContextType = Context::StructArrayInitializer; 1664 return; 1665 } 1666 } 1667 P.Contexts.pop_back(); 1668 } 1669 }; 1670 1671 void modifyContext(const FormatToken &Current) { 1672 auto AssignmentStartsExpression = [&]() { 1673 if (Current.getPrecedence() != prec::Assignment) 1674 return false; 1675 1676 if (Line.First->isOneOf(tok::kw_using, tok::kw_return)) 1677 return false; 1678 if (Line.First->is(tok::kw_template)) { 1679 assert(Current.Previous); 1680 if (Current.Previous->is(tok::kw_operator)) { 1681 // `template ... operator=` cannot be an expression. 1682 return false; 1683 } 1684 1685 // `template` keyword can start a variable template. 1686 const FormatToken *Tok = Line.First->getNextNonComment(); 1687 assert(Tok); // Current token is on the same line. 1688 if (Tok->isNot(TT_TemplateOpener)) { 1689 // Explicit template instantiations do not have `<>`. 1690 return false; 1691 } 1692 1693 Tok = Tok->MatchingParen; 1694 if (!Tok) 1695 return false; 1696 Tok = Tok->getNextNonComment(); 1697 if (!Tok) 1698 return false; 1699 1700 if (Tok->isOneOf(tok::kw_class, tok::kw_enum, tok::kw_struct, 1701 tok::kw_using)) { 1702 return false; 1703 } 1704 1705 return true; 1706 } 1707 1708 // Type aliases use `type X = ...;` in TypeScript and can be exported 1709 // using `export type ...`. 1710 if (Style.isJavaScript() && 1711 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1712 Line.startsWith(tok::kw_export, Keywords.kw_type, 1713 tok::identifier))) { 1714 return false; 1715 } 1716 1717 return !Current.Previous || Current.Previous->isNot(tok::kw_operator); 1718 }; 1719 1720 if (AssignmentStartsExpression()) { 1721 Contexts.back().IsExpression = true; 1722 if (!Line.startsWith(TT_UnaryOperator)) { 1723 for (FormatToken *Previous = Current.Previous; 1724 Previous && Previous->Previous && 1725 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1726 Previous = Previous->Previous) { 1727 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1728 Previous = Previous->MatchingParen; 1729 if (!Previous) 1730 break; 1731 } 1732 if (Previous->opensScope()) 1733 break; 1734 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1735 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1736 Previous->Previous && Previous->Previous->isNot(tok::equal)) { 1737 Previous->setType(TT_PointerOrReference); 1738 } 1739 } 1740 } 1741 } else if (Current.is(tok::lessless) && 1742 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1743 Contexts.back().IsExpression = true; 1744 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1745 Contexts.back().IsExpression = true; 1746 } else if (Current.is(TT_TrailingReturnArrow)) { 1747 Contexts.back().IsExpression = false; 1748 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1749 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1750 } else if (Current.Previous && 1751 Current.Previous->is(TT_CtorInitializerColon)) { 1752 Contexts.back().IsExpression = true; 1753 Contexts.back().ContextType = Context::CtorInitializer; 1754 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1755 Contexts.back().ContextType = Context::InheritanceList; 1756 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1757 for (FormatToken *Previous = Current.Previous; 1758 Previous && Previous->isOneOf(tok::star, tok::amp); 1759 Previous = Previous->Previous) { 1760 Previous->setType(TT_PointerOrReference); 1761 } 1762 if (Line.MustBeDeclaration && 1763 Contexts.front().ContextType != Context::CtorInitializer) { 1764 Contexts.back().IsExpression = false; 1765 } 1766 } else if (Current.is(tok::kw_new)) { 1767 Contexts.back().CanBeExpression = false; 1768 } else if (Current.is(tok::semi) || 1769 (Current.is(tok::exclaim) && Current.Previous && 1770 !Current.Previous->is(tok::kw_operator))) { 1771 // This should be the condition or increment in a for-loop. 1772 // But not operator !() (can't use TT_OverloadedOperator here as its not 1773 // been annotated yet). 1774 Contexts.back().IsExpression = true; 1775 } 1776 } 1777 1778 static FormatToken *untilMatchingParen(FormatToken *Current) { 1779 // Used when `MatchingParen` is not yet established. 1780 int ParenLevel = 0; 1781 while (Current) { 1782 if (Current->is(tok::l_paren)) 1783 ++ParenLevel; 1784 if (Current->is(tok::r_paren)) 1785 --ParenLevel; 1786 if (ParenLevel < 1) 1787 break; 1788 Current = Current->Next; 1789 } 1790 return Current; 1791 } 1792 1793 static bool isDeductionGuide(FormatToken &Current) { 1794 // Look for a deduction guide template<T> A(...) -> A<...>; 1795 if (Current.Previous && Current.Previous->is(tok::r_paren) && 1796 Current.startsSequence(tok::arrow, tok::identifier, tok::less)) { 1797 // Find the TemplateCloser. 1798 FormatToken *TemplateCloser = Current.Next->Next; 1799 int NestingLevel = 0; 1800 while (TemplateCloser) { 1801 // Skip over an expressions in parens A<(3 < 2)>; 1802 if (TemplateCloser->is(tok::l_paren)) { 1803 // No Matching Paren yet so skip to matching paren 1804 TemplateCloser = untilMatchingParen(TemplateCloser); 1805 if (!TemplateCloser) 1806 break; 1807 } 1808 if (TemplateCloser->is(tok::less)) 1809 ++NestingLevel; 1810 if (TemplateCloser->is(tok::greater)) 1811 --NestingLevel; 1812 if (NestingLevel < 1) 1813 break; 1814 TemplateCloser = TemplateCloser->Next; 1815 } 1816 // Assuming we have found the end of the template ensure its followed 1817 // with a semi-colon. 1818 if (TemplateCloser && TemplateCloser->Next && 1819 TemplateCloser->Next->is(tok::semi) && 1820 Current.Previous->MatchingParen) { 1821 // Determine if the identifier `A` prior to the A<..>; is the same as 1822 // prior to the A(..) 1823 FormatToken *LeadingIdentifier = 1824 Current.Previous->MatchingParen->Previous; 1825 1826 return LeadingIdentifier && 1827 LeadingIdentifier->TokenText == Current.Next->TokenText; 1828 } 1829 } 1830 return false; 1831 } 1832 1833 void determineTokenType(FormatToken &Current) { 1834 if (!Current.is(TT_Unknown)) { 1835 // The token type is already known. 1836 return; 1837 } 1838 1839 if ((Style.isJavaScript() || Style.isCSharp()) && 1840 Current.is(tok::exclaim)) { 1841 if (Current.Previous) { 1842 bool IsIdentifier = 1843 Style.isJavaScript() 1844 ? Keywords.IsJavaScriptIdentifier( 1845 *Current.Previous, /* AcceptIdentifierName= */ true) 1846 : Current.Previous->is(tok::identifier); 1847 if (IsIdentifier || 1848 Current.Previous->isOneOf( 1849 tok::kw_default, tok::kw_namespace, tok::r_paren, tok::r_square, 1850 tok::r_brace, tok::kw_false, tok::kw_true, Keywords.kw_type, 1851 Keywords.kw_get, Keywords.kw_init, Keywords.kw_set) || 1852 Current.Previous->Tok.isLiteral()) { 1853 Current.setType(TT_NonNullAssertion); 1854 return; 1855 } 1856 } 1857 if (Current.Next && 1858 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1859 Current.setType(TT_NonNullAssertion); 1860 return; 1861 } 1862 } 1863 1864 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1865 // function declaration have been found. In this case, 'Current' is a 1866 // trailing token of this declaration and thus cannot be a name. 1867 if (Current.is(Keywords.kw_instanceof)) { 1868 Current.setType(TT_BinaryOperator); 1869 } else if (isStartOfName(Current) && 1870 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1871 Contexts.back().FirstStartOfName = &Current; 1872 Current.setType(TT_StartOfName); 1873 } else if (Current.is(tok::semi)) { 1874 // Reset FirstStartOfName after finding a semicolon so that a for loop 1875 // with multiple increment statements is not confused with a for loop 1876 // having multiple variable declarations. 1877 Contexts.back().FirstStartOfName = nullptr; 1878 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1879 AutoFound = true; 1880 } else if (Current.is(tok::arrow) && 1881 Style.Language == FormatStyle::LK_Java) { 1882 Current.setType(TT_LambdaArrow); 1883 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1884 Current.NestingLevel == 0 && 1885 !Current.Previous->isOneOf(tok::kw_operator, tok::identifier)) { 1886 // not auto operator->() -> xxx; 1887 Current.setType(TT_TrailingReturnArrow); 1888 } else if (Current.is(tok::arrow) && Current.Previous && 1889 Current.Previous->is(tok::r_brace)) { 1890 // Concept implicit conversion constraint needs to be treated like 1891 // a trailing return type ... } -> <type>. 1892 Current.setType(TT_TrailingReturnArrow); 1893 } else if (isDeductionGuide(Current)) { 1894 // Deduction guides trailing arrow " A(...) -> A<T>;". 1895 Current.setType(TT_TrailingReturnArrow); 1896 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1897 Current.setType(determineStarAmpUsage( 1898 Current, 1899 Contexts.back().CanBeExpression && Contexts.back().IsExpression, 1900 Contexts.back().ContextType == Context::TemplateArgument)); 1901 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret) || 1902 (Style.isVerilog() && Current.is(tok::pipe))) { 1903 Current.setType(determinePlusMinusCaretUsage(Current)); 1904 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1905 Contexts.back().CaretFound = true; 1906 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1907 Current.setType(determineIncrementUsage(Current)); 1908 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1909 Current.setType(TT_UnaryOperator); 1910 } else if (Current.is(tok::question)) { 1911 if (Style.isJavaScript() && Line.MustBeDeclaration && 1912 !Contexts.back().IsExpression) { 1913 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1914 // on the interface, not a ternary expression. 1915 Current.setType(TT_JsTypeOptionalQuestion); 1916 } else { 1917 Current.setType(TT_ConditionalExpr); 1918 } 1919 } else if (Current.isBinaryOperator() && 1920 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1921 (!Current.is(tok::greater) && 1922 Style.Language != FormatStyle::LK_TextProto)) { 1923 Current.setType(TT_BinaryOperator); 1924 } else if (Current.is(tok::comment)) { 1925 if (Current.TokenText.startswith("/*")) { 1926 if (Current.TokenText.endswith("*/")) { 1927 Current.setType(TT_BlockComment); 1928 } else { 1929 // The lexer has for some reason determined a comment here. But we 1930 // cannot really handle it, if it isn't properly terminated. 1931 Current.Tok.setKind(tok::unknown); 1932 } 1933 } else { 1934 Current.setType(TT_LineComment); 1935 } 1936 } else if (Current.is(tok::l_paren)) { 1937 if (lParenStartsCppCast(Current)) 1938 Current.setType(TT_CppCastLParen); 1939 } else if (Current.is(tok::r_paren)) { 1940 if (rParenEndsCast(Current)) 1941 Current.setType(TT_CastRParen); 1942 if (Current.MatchingParen && Current.Next && 1943 !Current.Next->isBinaryOperator() && 1944 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1945 tok::comma, tok::period, tok::arrow, 1946 tok::coloncolon, tok::kw_noexcept)) { 1947 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1948 // Make sure this isn't the return type of an Obj-C block declaration 1949 if (AfterParen->isNot(tok::caret)) { 1950 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) { 1951 if (BeforeParen->is(tok::identifier) && 1952 !BeforeParen->is(TT_TypenameMacro) && 1953 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1954 (!BeforeParen->Previous || 1955 BeforeParen->Previous->ClosesTemplateDeclaration)) { 1956 Current.setType(TT_FunctionAnnotationRParen); 1957 } 1958 } 1959 } 1960 } 1961 } 1962 } else if (Current.is(tok::at) && Current.Next && !Style.isJavaScript() && 1963 Style.Language != FormatStyle::LK_Java) { 1964 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1965 // marks declarations and properties that need special formatting. 1966 switch (Current.Next->Tok.getObjCKeywordID()) { 1967 case tok::objc_interface: 1968 case tok::objc_implementation: 1969 case tok::objc_protocol: 1970 Current.setType(TT_ObjCDecl); 1971 break; 1972 case tok::objc_property: 1973 Current.setType(TT_ObjCProperty); 1974 break; 1975 default: 1976 break; 1977 } 1978 } else if (Current.is(tok::period)) { 1979 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1980 if (PreviousNoComment && 1981 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) { 1982 Current.setType(TT_DesignatedInitializerPeriod); 1983 } else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1984 Current.Previous->isOneOf(TT_JavaAnnotation, 1985 TT_LeadingJavaAnnotation)) { 1986 Current.setType(Current.Previous->getType()); 1987 } 1988 } else if (canBeObjCSelectorComponent(Current) && 1989 // FIXME(bug 36976): ObjC return types shouldn't use 1990 // TT_CastRParen. 1991 Current.Previous && Current.Previous->is(TT_CastRParen) && 1992 Current.Previous->MatchingParen && 1993 Current.Previous->MatchingParen->Previous && 1994 Current.Previous->MatchingParen->Previous->is( 1995 TT_ObjCMethodSpecifier)) { 1996 // This is the first part of an Objective-C selector name. (If there's no 1997 // colon after this, this is the only place which annotates the identifier 1998 // as a selector.) 1999 Current.setType(TT_SelectorName); 2000 } else if (Current.isOneOf(tok::identifier, tok::kw_const, tok::kw_noexcept, 2001 tok::kw_requires) && 2002 Current.Previous && 2003 !Current.Previous->isOneOf(tok::equal, tok::at, 2004 TT_CtorInitializerComma, 2005 TT_CtorInitializerColon) && 2006 Line.MightBeFunctionDecl && Contexts.size() == 1) { 2007 // Line.MightBeFunctionDecl can only be true after the parentheses of a 2008 // function declaration have been found. 2009 Current.setType(TT_TrailingAnnotation); 2010 } else if ((Style.Language == FormatStyle::LK_Java || 2011 Style.isJavaScript()) && 2012 Current.Previous) { 2013 if (Current.Previous->is(tok::at) && 2014 Current.isNot(Keywords.kw_interface)) { 2015 const FormatToken &AtToken = *Current.Previous; 2016 const FormatToken *Previous = AtToken.getPreviousNonComment(); 2017 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 2018 Current.setType(TT_LeadingJavaAnnotation); 2019 else 2020 Current.setType(TT_JavaAnnotation); 2021 } else if (Current.Previous->is(tok::period) && 2022 Current.Previous->isOneOf(TT_JavaAnnotation, 2023 TT_LeadingJavaAnnotation)) { 2024 Current.setType(Current.Previous->getType()); 2025 } 2026 } 2027 } 2028 2029 /// Take a guess at whether \p Tok starts a name of a function or 2030 /// variable declaration. 2031 /// 2032 /// This is a heuristic based on whether \p Tok is an identifier following 2033 /// something that is likely a type. 2034 bool isStartOfName(const FormatToken &Tok) { 2035 if (Tok.isNot(tok::identifier) || !Tok.Previous) 2036 return false; 2037 2038 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 2039 Keywords.kw_as)) { 2040 return false; 2041 } 2042 if (Style.isJavaScript() && Tok.Previous->is(Keywords.kw_in)) 2043 return false; 2044 2045 // Skip "const" as it does not have an influence on whether this is a name. 2046 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 2047 2048 // For javascript const can be like "let" or "var" 2049 if (!Style.isJavaScript()) 2050 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 2051 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 2052 2053 if (!PreviousNotConst) 2054 return false; 2055 2056 if (PreviousNotConst->ClosesRequiresClause) 2057 return false; 2058 2059 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 2060 PreviousNotConst->Previous && 2061 PreviousNotConst->Previous->is(tok::hash); 2062 2063 if (PreviousNotConst->is(TT_TemplateCloser)) { 2064 return PreviousNotConst && PreviousNotConst->MatchingParen && 2065 PreviousNotConst->MatchingParen->Previous && 2066 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 2067 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 2068 } 2069 2070 if (PreviousNotConst->is(tok::r_paren) && 2071 PreviousNotConst->is(TT_TypeDeclarationParen)) { 2072 return true; 2073 } 2074 2075 // If is a preprocess keyword like #define. 2076 if (IsPPKeyword) 2077 return false; 2078 2079 // int a or auto a. 2080 if (PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) 2081 return true; 2082 2083 // *a or &a or &&a. 2084 if (PreviousNotConst->is(TT_PointerOrReference)) 2085 return true; 2086 2087 // MyClass a; 2088 if (PreviousNotConst->isSimpleTypeSpecifier()) 2089 return true; 2090 2091 // type[] a in Java 2092 if (Style.Language == FormatStyle::LK_Java && 2093 PreviousNotConst->is(tok::r_square)) { 2094 return true; 2095 } 2096 2097 // const a = in JavaScript. 2098 return Style.isJavaScript() && PreviousNotConst->is(tok::kw_const); 2099 } 2100 2101 /// Determine whether '(' is starting a C++ cast. 2102 bool lParenStartsCppCast(const FormatToken &Tok) { 2103 // C-style casts are only used in C++. 2104 if (!Style.isCpp()) 2105 return false; 2106 2107 FormatToken *LeftOfParens = Tok.getPreviousNonComment(); 2108 if (LeftOfParens && LeftOfParens->is(TT_TemplateCloser) && 2109 LeftOfParens->MatchingParen) { 2110 auto *Prev = LeftOfParens->MatchingParen->getPreviousNonComment(); 2111 if (Prev && 2112 Prev->isOneOf(tok::kw_const_cast, tok::kw_dynamic_cast, 2113 tok::kw_reinterpret_cast, tok::kw_static_cast)) { 2114 // FIXME: Maybe we should handle identifiers ending with "_cast", 2115 // e.g. any_cast? 2116 return true; 2117 } 2118 } 2119 return false; 2120 } 2121 2122 /// Determine whether ')' is ending a cast. 2123 bool rParenEndsCast(const FormatToken &Tok) { 2124 // C-style casts are only used in C++, C# and Java. 2125 if (!Style.isCSharp() && !Style.isCpp() && 2126 Style.Language != FormatStyle::LK_Java) { 2127 return false; 2128 } 2129 2130 // Empty parens aren't casts and there are no casts at the end of the line. 2131 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 2132 return false; 2133 2134 if (Tok.MatchingParen->is(TT_OverloadedOperatorLParen)) 2135 return false; 2136 2137 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 2138 if (LeftOfParens) { 2139 // If there is a closing parenthesis left of the current 2140 // parentheses, look past it as these might be chained casts. 2141 if (LeftOfParens->is(tok::r_paren) && 2142 LeftOfParens->isNot(TT_CastRParen)) { 2143 if (!LeftOfParens->MatchingParen || 2144 !LeftOfParens->MatchingParen->Previous) { 2145 return false; 2146 } 2147 LeftOfParens = LeftOfParens->MatchingParen->Previous; 2148 } 2149 2150 if (LeftOfParens->is(tok::r_square)) { 2151 // delete[] (void *)ptr; 2152 auto MayBeArrayDelete = [](FormatToken *Tok) -> FormatToken * { 2153 if (Tok->isNot(tok::r_square)) 2154 return nullptr; 2155 2156 Tok = Tok->getPreviousNonComment(); 2157 if (!Tok || Tok->isNot(tok::l_square)) 2158 return nullptr; 2159 2160 Tok = Tok->getPreviousNonComment(); 2161 if (!Tok || Tok->isNot(tok::kw_delete)) 2162 return nullptr; 2163 return Tok; 2164 }; 2165 if (FormatToken *MaybeDelete = MayBeArrayDelete(LeftOfParens)) 2166 LeftOfParens = MaybeDelete; 2167 } 2168 2169 // The Condition directly below this one will see the operator arguments 2170 // as a (void *foo) cast. 2171 // void operator delete(void *foo) ATTRIB; 2172 if (LeftOfParens->Tok.getIdentifierInfo() && LeftOfParens->Previous && 2173 LeftOfParens->Previous->is(tok::kw_operator)) { 2174 return false; 2175 } 2176 2177 // If there is an identifier (or with a few exceptions a keyword) right 2178 // before the parentheses, this is unlikely to be a cast. 2179 if (LeftOfParens->Tok.getIdentifierInfo() && 2180 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 2181 tok::kw_delete, tok::kw_throw)) { 2182 return false; 2183 } 2184 2185 // Certain other tokens right before the parentheses are also signals that 2186 // this cannot be a cast. 2187 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 2188 TT_TemplateCloser, tok::ellipsis)) { 2189 return false; 2190 } 2191 } 2192 2193 if (Tok.Next->is(tok::question)) 2194 return false; 2195 2196 // `foreach((A a, B b) in someList)` should not be seen as a cast. 2197 if (Tok.Next->is(Keywords.kw_in) && Style.isCSharp()) 2198 return false; 2199 2200 // Functions which end with decorations like volatile, noexcept are unlikely 2201 // to be casts. 2202 if (Tok.Next->isOneOf(tok::kw_noexcept, tok::kw_volatile, tok::kw_const, 2203 tok::kw_requires, tok::kw_throw, tok::arrow, 2204 Keywords.kw_override, Keywords.kw_final) || 2205 isCpp11AttributeSpecifier(*Tok.Next)) { 2206 return false; 2207 } 2208 2209 // As Java has no function types, a "(" after the ")" likely means that this 2210 // is a cast. 2211 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 2212 return true; 2213 2214 // If a (non-string) literal follows, this is likely a cast. 2215 if (Tok.Next->isNot(tok::string_literal) && 2216 (Tok.Next->Tok.isLiteral() || 2217 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) { 2218 return true; 2219 } 2220 2221 // Heuristically try to determine whether the parentheses contain a type. 2222 auto IsQualifiedPointerOrReference = [](FormatToken *T) { 2223 // This is used to handle cases such as x = (foo *const)&y; 2224 assert(!T->isSimpleTypeSpecifier() && "Should have already been checked"); 2225 // Strip trailing qualifiers such as const or volatile when checking 2226 // whether the parens could be a cast to a pointer/reference type. 2227 while (T) { 2228 if (T->is(TT_AttributeParen)) { 2229 // Handle `x = (foo *__attribute__((foo)))&v;`: 2230 if (T->MatchingParen && T->MatchingParen->Previous && 2231 T->MatchingParen->Previous->is(tok::kw___attribute)) { 2232 T = T->MatchingParen->Previous->Previous; 2233 continue; 2234 } 2235 } else if (T->is(TT_AttributeSquare)) { 2236 // Handle `x = (foo *[[clang::foo]])&v;`: 2237 if (T->MatchingParen && T->MatchingParen->Previous) { 2238 T = T->MatchingParen->Previous; 2239 continue; 2240 } 2241 } else if (T->canBePointerOrReferenceQualifier()) { 2242 T = T->Previous; 2243 continue; 2244 } 2245 break; 2246 } 2247 return T && T->is(TT_PointerOrReference); 2248 }; 2249 bool ParensAreType = 2250 !Tok.Previous || 2251 Tok.Previous->isOneOf(TT_TemplateCloser, TT_TypeDeclarationParen) || 2252 Tok.Previous->isSimpleTypeSpecifier() || 2253 IsQualifiedPointerOrReference(Tok.Previous); 2254 bool ParensCouldEndDecl = 2255 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 2256 if (ParensAreType && !ParensCouldEndDecl) 2257 return true; 2258 2259 // At this point, we heuristically assume that there are no casts at the 2260 // start of the line. We assume that we have found most cases where there 2261 // are by the logic above, e.g. "(void)x;". 2262 if (!LeftOfParens) 2263 return false; 2264 2265 // Certain token types inside the parentheses mean that this can't be a 2266 // cast. 2267 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 2268 Token = Token->Next) { 2269 if (Token->is(TT_BinaryOperator)) 2270 return false; 2271 } 2272 2273 // If the following token is an identifier or 'this', this is a cast. All 2274 // cases where this can be something else are handled above. 2275 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 2276 return true; 2277 2278 // Look for a cast `( x ) (`. 2279 if (Tok.Next->is(tok::l_paren) && Tok.Previous && Tok.Previous->Previous) { 2280 if (Tok.Previous->is(tok::identifier) && 2281 Tok.Previous->Previous->is(tok::l_paren)) { 2282 return true; 2283 } 2284 } 2285 2286 if (!Tok.Next->Next) 2287 return false; 2288 2289 // If the next token after the parenthesis is a unary operator, assume 2290 // that this is cast, unless there are unexpected tokens inside the 2291 // parenthesis. 2292 bool NextIsUnary = 2293 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 2294 if (!NextIsUnary || Tok.Next->is(tok::plus) || 2295 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) { 2296 return false; 2297 } 2298 // Search for unexpected tokens. 2299 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 2300 Prev = Prev->Previous) { 2301 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 2302 return false; 2303 } 2304 return true; 2305 } 2306 2307 /// Returns true if the token is used as a unary operator. 2308 bool determineUnaryOperatorByUsage(const FormatToken &Tok) { 2309 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2310 if (!PrevToken) 2311 return true; 2312 2313 // These keywords are deliberately not included here because they may 2314 // precede only one of unary star/amp and plus/minus but not both. They are 2315 // either included in determineStarAmpUsage or determinePlusMinusCaretUsage. 2316 // 2317 // @ - It may be followed by a unary `-` in Objective-C literals. We don't 2318 // know how they can be followed by a star or amp. 2319 if (PrevToken->isOneOf( 2320 TT_ConditionalExpr, tok::l_paren, tok::comma, tok::colon, tok::semi, 2321 tok::equal, tok::question, tok::l_square, tok::l_brace, 2322 tok::kw_case, tok::kw_co_await, tok::kw_co_return, tok::kw_co_yield, 2323 tok::kw_delete, tok::kw_return, tok::kw_throw)) { 2324 return true; 2325 } 2326 2327 // We put sizeof here instead of only in determineStarAmpUsage. In the cases 2328 // where the unary `+` operator is overloaded, it is reasonable to write 2329 // things like `sizeof +x`. Like commit 446d6ec996c6c3. 2330 if (PrevToken->is(tok::kw_sizeof)) 2331 return true; 2332 2333 // A sequence of leading unary operators. 2334 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 2335 return true; 2336 2337 // There can't be two consecutive binary operators. 2338 if (PrevToken->is(TT_BinaryOperator)) 2339 return true; 2340 2341 return false; 2342 } 2343 2344 /// Return the type of the given token assuming it is * or &. 2345 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 2346 bool InTemplateArgument) { 2347 if (Style.isJavaScript()) 2348 return TT_BinaryOperator; 2349 2350 // && in C# must be a binary operator. 2351 if (Style.isCSharp() && Tok.is(tok::ampamp)) 2352 return TT_BinaryOperator; 2353 2354 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2355 if (!PrevToken) 2356 return TT_UnaryOperator; 2357 2358 const FormatToken *NextToken = Tok.getNextNonComment(); 2359 2360 if (InTemplateArgument && NextToken && NextToken->is(tok::kw_noexcept)) 2361 return TT_BinaryOperator; 2362 2363 if (!NextToken || 2364 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_noexcept, tok::comma, 2365 tok::r_paren) || 2366 NextToken->canBePointerOrReferenceQualifier() || 2367 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) { 2368 return TT_PointerOrReference; 2369 } 2370 2371 if (PrevToken->is(tok::coloncolon)) 2372 return TT_PointerOrReference; 2373 2374 if (PrevToken->is(tok::r_paren) && PrevToken->is(TT_TypeDeclarationParen)) 2375 return TT_PointerOrReference; 2376 2377 if (determineUnaryOperatorByUsage(Tok)) 2378 return TT_UnaryOperator; 2379 2380 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 2381 return TT_PointerOrReference; 2382 if (NextToken->is(tok::kw_operator) && !IsExpression) 2383 return TT_PointerOrReference; 2384 if (NextToken->isOneOf(tok::comma, tok::semi)) 2385 return TT_PointerOrReference; 2386 2387 // After right braces, star tokens are likely to be pointers to struct, 2388 // union, or class. 2389 // struct {} *ptr; 2390 // This by itself is not sufficient to distinguish from multiplication 2391 // following a brace-initialized expression, as in: 2392 // int i = int{42} * 2; 2393 // In the struct case, the part of the struct declaration until the `{` and 2394 // the `}` are put on separate unwrapped lines; in the brace-initialized 2395 // case, the matching `{` is on the same unwrapped line, so check for the 2396 // presence of the matching brace to distinguish between those. 2397 if (PrevToken->is(tok::r_brace) && Tok.is(tok::star) && 2398 !PrevToken->MatchingParen) { 2399 return TT_PointerOrReference; 2400 } 2401 2402 // if (Class* obj { function() }) 2403 if (PrevToken->Tok.isAnyIdentifier() && NextToken->Tok.isAnyIdentifier() && 2404 NextToken->Next && NextToken->Next->is(tok::l_brace)) { 2405 return TT_PointerOrReference; 2406 } 2407 2408 if (PrevToken->endsSequence(tok::r_square, tok::l_square, tok::kw_delete)) 2409 return TT_UnaryOperator; 2410 2411 if (PrevToken->Tok.isLiteral() || 2412 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 2413 tok::kw_false, tok::r_brace)) { 2414 return TT_BinaryOperator; 2415 } 2416 2417 const FormatToken *NextNonParen = NextToken; 2418 while (NextNonParen && NextNonParen->is(tok::l_paren)) 2419 NextNonParen = NextNonParen->getNextNonComment(); 2420 if (NextNonParen && (NextNonParen->Tok.isLiteral() || 2421 NextNonParen->isOneOf(tok::kw_true, tok::kw_false) || 2422 NextNonParen->isUnaryOperator())) { 2423 return TT_BinaryOperator; 2424 } 2425 2426 // If we know we're in a template argument, there are no named declarations. 2427 // Thus, having an identifier on the right-hand side indicates a binary 2428 // operator. 2429 if (InTemplateArgument && NextToken->Tok.isAnyIdentifier()) 2430 return TT_BinaryOperator; 2431 2432 // "&&(" is quite unlikely to be two successive unary "&". 2433 if (Tok.is(tok::ampamp) && NextToken->is(tok::l_paren)) 2434 return TT_BinaryOperator; 2435 2436 // This catches some cases where evaluation order is used as control flow: 2437 // aaa && aaa->f(); 2438 if (NextToken->Tok.isAnyIdentifier()) { 2439 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 2440 if (NextNextToken && NextNextToken->is(tok::arrow)) 2441 return TT_BinaryOperator; 2442 } 2443 2444 // It is very unlikely that we are going to find a pointer or reference type 2445 // definition on the RHS of an assignment. 2446 if (IsExpression && !Contexts.back().CaretFound) 2447 return TT_BinaryOperator; 2448 2449 return TT_PointerOrReference; 2450 } 2451 2452 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 2453 if (determineUnaryOperatorByUsage(Tok)) 2454 return TT_UnaryOperator; 2455 2456 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2457 if (!PrevToken) 2458 return TT_UnaryOperator; 2459 2460 if (PrevToken->is(tok::at)) 2461 return TT_UnaryOperator; 2462 2463 // Fall back to marking the token as binary operator. 2464 return TT_BinaryOperator; 2465 } 2466 2467 /// Determine whether ++/-- are pre- or post-increments/-decrements. 2468 TokenType determineIncrementUsage(const FormatToken &Tok) { 2469 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2470 if (!PrevToken || PrevToken->is(TT_CastRParen)) 2471 return TT_UnaryOperator; 2472 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 2473 return TT_TrailingUnaryOperator; 2474 2475 return TT_UnaryOperator; 2476 } 2477 2478 SmallVector<Context, 8> Contexts; 2479 2480 const FormatStyle &Style; 2481 AnnotatedLine &Line; 2482 FormatToken *CurrentToken; 2483 bool AutoFound; 2484 const AdditionalKeywords &Keywords; 2485 2486 // Set of "<" tokens that do not open a template parameter list. If parseAngle 2487 // determines that a specific token can't be a template opener, it will make 2488 // same decision irrespective of the decisions for tokens leading up to it. 2489 // Store this information to prevent this from causing exponential runtime. 2490 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 2491 }; 2492 2493 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 2494 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 2495 2496 /// Parses binary expressions by inserting fake parenthesis based on 2497 /// operator precedence. 2498 class ExpressionParser { 2499 public: 2500 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 2501 AnnotatedLine &Line) 2502 : Style(Style), Keywords(Keywords), Line(Line), Current(Line.First) {} 2503 2504 /// Parse expressions with the given operator precedence. 2505 void parse(int Precedence = 0) { 2506 // Skip 'return' and ObjC selector colons as they are not part of a binary 2507 // expression. 2508 while (Current && (Current->is(tok::kw_return) || 2509 (Current->is(tok::colon) && 2510 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) { 2511 next(); 2512 } 2513 2514 if (!Current || Precedence > PrecedenceArrowAndPeriod) 2515 return; 2516 2517 // Conditional expressions need to be parsed separately for proper nesting. 2518 if (Precedence == prec::Conditional) { 2519 parseConditionalExpr(); 2520 return; 2521 } 2522 2523 // Parse unary operators, which all have a higher precedence than binary 2524 // operators. 2525 if (Precedence == PrecedenceUnaryOperator) { 2526 parseUnaryOperator(); 2527 return; 2528 } 2529 2530 FormatToken *Start = Current; 2531 FormatToken *LatestOperator = nullptr; 2532 unsigned OperatorIndex = 0; 2533 2534 while (Current) { 2535 // Consume operators with higher precedence. 2536 parse(Precedence + 1); 2537 2538 int CurrentPrecedence = getCurrentPrecedence(); 2539 2540 if (Precedence == CurrentPrecedence && Current && 2541 Current->is(TT_SelectorName)) { 2542 if (LatestOperator) 2543 addFakeParenthesis(Start, prec::Level(Precedence)); 2544 Start = Current; 2545 } 2546 2547 // At the end of the line or when an operator with higher precedence is 2548 // found, insert fake parenthesis and return. 2549 if (!Current || 2550 (Current->closesScope() && 2551 (Current->MatchingParen || Current->is(TT_TemplateString))) || 2552 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 2553 (CurrentPrecedence == prec::Conditional && 2554 Precedence == prec::Assignment && Current->is(tok::colon))) { 2555 break; 2556 } 2557 2558 // Consume scopes: (), [], <> and {} 2559 // In addition to that we handle require clauses as scope, so that the 2560 // constraints in that are correctly indented. 2561 if (Current->opensScope() || 2562 Current->isOneOf(TT_RequiresClause, 2563 TT_RequiresClauseInARequiresExpression)) { 2564 // In fragment of a JavaScript template string can look like '}..${' and 2565 // thus close a scope and open a new one at the same time. 2566 while (Current && (!Current->closesScope() || Current->opensScope())) { 2567 next(); 2568 parse(); 2569 } 2570 next(); 2571 } else { 2572 // Operator found. 2573 if (CurrentPrecedence == Precedence) { 2574 if (LatestOperator) 2575 LatestOperator->NextOperator = Current; 2576 LatestOperator = Current; 2577 Current->OperatorIndex = OperatorIndex; 2578 ++OperatorIndex; 2579 } 2580 next(/*SkipPastLeadingComments=*/Precedence > 0); 2581 } 2582 } 2583 2584 if (LatestOperator && (Current || Precedence > 0)) { 2585 // The requires clauses do not neccessarily end in a semicolon or a brace, 2586 // but just go over to struct/class or a function declaration, we need to 2587 // intervene so that the fake right paren is inserted correctly. 2588 auto End = 2589 (Start->Previous && 2590 Start->Previous->isOneOf(TT_RequiresClause, 2591 TT_RequiresClauseInARequiresExpression)) 2592 ? [this]() { 2593 auto Ret = Current ? Current : Line.Last; 2594 while (!Ret->ClosesRequiresClause && Ret->Previous) 2595 Ret = Ret->Previous; 2596 return Ret; 2597 }() 2598 : nullptr; 2599 2600 if (Precedence == PrecedenceArrowAndPeriod) { 2601 // Call expressions don't have a binary operator precedence. 2602 addFakeParenthesis(Start, prec::Unknown, End); 2603 } else { 2604 addFakeParenthesis(Start, prec::Level(Precedence), End); 2605 } 2606 } 2607 } 2608 2609 private: 2610 /// Gets the precedence (+1) of the given token for binary operators 2611 /// and other tokens that we treat like binary operators. 2612 int getCurrentPrecedence() { 2613 if (Current) { 2614 const FormatToken *NextNonComment = Current->getNextNonComment(); 2615 if (Current->is(TT_ConditionalExpr)) 2616 return prec::Conditional; 2617 if (NextNonComment && Current->is(TT_SelectorName) && 2618 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 2619 ((Style.Language == FormatStyle::LK_Proto || 2620 Style.Language == FormatStyle::LK_TextProto) && 2621 NextNonComment->is(tok::less)))) { 2622 return prec::Assignment; 2623 } 2624 if (Current->is(TT_JsComputedPropertyName)) 2625 return prec::Assignment; 2626 if (Current->is(TT_LambdaArrow)) 2627 return prec::Comma; 2628 if (Current->is(TT_FatArrow)) 2629 return prec::Assignment; 2630 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 2631 (Current->is(tok::comment) && NextNonComment && 2632 NextNonComment->is(TT_SelectorName))) { 2633 return 0; 2634 } 2635 if (Current->is(TT_RangeBasedForLoopColon)) 2636 return prec::Comma; 2637 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2638 Current->is(Keywords.kw_instanceof)) { 2639 return prec::Relational; 2640 } 2641 if (Style.isJavaScript() && 2642 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) { 2643 return prec::Relational; 2644 } 2645 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 2646 return Current->getPrecedence(); 2647 if (Current->isOneOf(tok::period, tok::arrow) && 2648 Current->isNot(TT_TrailingReturnArrow)) { 2649 return PrecedenceArrowAndPeriod; 2650 } 2651 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2652 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 2653 Keywords.kw_throws)) { 2654 return 0; 2655 } 2656 // In Verilog case labels are not on separate lines straight out of 2657 // UnwrappedLineParser. The colon is not part of an expression. 2658 if (Style.isVerilog() && Current->is(tok::colon)) 2659 return 0; 2660 } 2661 return -1; 2662 } 2663 2664 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence, 2665 FormatToken *End = nullptr) { 2666 Start->FakeLParens.push_back(Precedence); 2667 if (Precedence > prec::Unknown) 2668 Start->StartsBinaryExpression = true; 2669 if (!End && Current) 2670 End = Current->getPreviousNonComment(); 2671 if (End) { 2672 ++End->FakeRParens; 2673 if (Precedence > prec::Unknown) 2674 End->EndsBinaryExpression = true; 2675 } 2676 } 2677 2678 /// Parse unary operator expressions and surround them with fake 2679 /// parentheses if appropriate. 2680 void parseUnaryOperator() { 2681 llvm::SmallVector<FormatToken *, 2> Tokens; 2682 while (Current && Current->is(TT_UnaryOperator)) { 2683 Tokens.push_back(Current); 2684 next(); 2685 } 2686 parse(PrecedenceArrowAndPeriod); 2687 for (FormatToken *Token : llvm::reverse(Tokens)) { 2688 // The actual precedence doesn't matter. 2689 addFakeParenthesis(Token, prec::Unknown); 2690 } 2691 } 2692 2693 void parseConditionalExpr() { 2694 while (Current && Current->isTrailingComment()) 2695 next(); 2696 FormatToken *Start = Current; 2697 parse(prec::LogicalOr); 2698 if (!Current || !Current->is(tok::question)) 2699 return; 2700 next(); 2701 parse(prec::Assignment); 2702 if (!Current || Current->isNot(TT_ConditionalExpr)) 2703 return; 2704 next(); 2705 parse(prec::Assignment); 2706 addFakeParenthesis(Start, prec::Conditional); 2707 } 2708 2709 void next(bool SkipPastLeadingComments = true) { 2710 if (Current) 2711 Current = Current->Next; 2712 while (Current && 2713 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 2714 Current->isTrailingComment()) { 2715 Current = Current->Next; 2716 } 2717 } 2718 2719 const FormatStyle &Style; 2720 const AdditionalKeywords &Keywords; 2721 const AnnotatedLine &Line; 2722 FormatToken *Current; 2723 }; 2724 2725 } // end anonymous namespace 2726 2727 void TokenAnnotator::setCommentLineLevels( 2728 SmallVectorImpl<AnnotatedLine *> &Lines) const { 2729 const AnnotatedLine *NextNonCommentLine = nullptr; 2730 for (AnnotatedLine *Line : llvm::reverse(Lines)) { 2731 assert(Line->First); 2732 2733 // If the comment is currently aligned with the line immediately following 2734 // it, that's probably intentional and we should keep it. 2735 if (NextNonCommentLine && Line->isComment() && 2736 NextNonCommentLine->First->NewlinesBefore <= 1 && 2737 NextNonCommentLine->First->OriginalColumn == 2738 Line->First->OriginalColumn) { 2739 const bool PPDirectiveOrImportStmt = 2740 NextNonCommentLine->Type == LT_PreprocessorDirective || 2741 NextNonCommentLine->Type == LT_ImportStatement; 2742 if (PPDirectiveOrImportStmt) 2743 Line->Type = LT_CommentAbovePPDirective; 2744 // Align comments for preprocessor lines with the # in column 0 if 2745 // preprocessor lines are not indented. Otherwise, align with the next 2746 // line. 2747 Line->Level = Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 2748 PPDirectiveOrImportStmt 2749 ? 0 2750 : NextNonCommentLine->Level; 2751 } else { 2752 NextNonCommentLine = Line->First->isNot(tok::r_brace) ? Line : nullptr; 2753 } 2754 2755 setCommentLineLevels(Line->Children); 2756 } 2757 } 2758 2759 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 2760 unsigned Result = 0; 2761 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 2762 Result = std::max(Result, Tok->NestingLevel); 2763 return Result; 2764 } 2765 2766 void TokenAnnotator::annotate(AnnotatedLine &Line) const { 2767 for (auto &Child : Line.Children) 2768 annotate(*Child); 2769 2770 AnnotatingParser Parser(Style, Line, Keywords); 2771 Line.Type = Parser.parseLine(); 2772 2773 // With very deep nesting, ExpressionParser uses lots of stack and the 2774 // formatting algorithm is very slow. We're not going to do a good job here 2775 // anyway - it's probably generated code being formatted by mistake. 2776 // Just skip the whole line. 2777 if (maxNestingDepth(Line) > 50) 2778 Line.Type = LT_Invalid; 2779 2780 if (Line.Type == LT_Invalid) 2781 return; 2782 2783 ExpressionParser ExprParser(Style, Keywords, Line); 2784 ExprParser.parse(); 2785 2786 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2787 Line.Type = LT_ObjCMethodDecl; 2788 else if (Line.startsWith(TT_ObjCDecl)) 2789 Line.Type = LT_ObjCDecl; 2790 else if (Line.startsWith(TT_ObjCProperty)) 2791 Line.Type = LT_ObjCProperty; 2792 2793 Line.First->SpacesRequiredBefore = 1; 2794 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 2795 } 2796 2797 // This function heuristically determines whether 'Current' starts the name of a 2798 // function declaration. 2799 static bool isFunctionDeclarationName(bool IsCpp, const FormatToken &Current, 2800 const AnnotatedLine &Line) { 2801 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 2802 for (; Next; Next = Next->Next) { 2803 if (Next->is(TT_OverloadedOperatorLParen)) 2804 return Next; 2805 if (Next->is(TT_OverloadedOperator)) 2806 continue; 2807 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 2808 // For 'new[]' and 'delete[]'. 2809 if (Next->Next && 2810 Next->Next->startsSequence(tok::l_square, tok::r_square)) { 2811 Next = Next->Next->Next; 2812 } 2813 continue; 2814 } 2815 if (Next->startsSequence(tok::l_square, tok::r_square)) { 2816 // For operator[](). 2817 Next = Next->Next; 2818 continue; 2819 } 2820 if ((Next->isSimpleTypeSpecifier() || Next->is(tok::identifier)) && 2821 Next->Next && Next->Next->isOneOf(tok::star, tok::amp, tok::ampamp)) { 2822 // For operator void*(), operator char*(), operator Foo*(). 2823 Next = Next->Next; 2824 continue; 2825 } 2826 if (Next->is(TT_TemplateOpener) && Next->MatchingParen) { 2827 Next = Next->MatchingParen; 2828 continue; 2829 } 2830 2831 break; 2832 } 2833 return nullptr; 2834 }; 2835 2836 // Find parentheses of parameter list. 2837 const FormatToken *Next = Current.Next; 2838 if (Current.is(tok::kw_operator)) { 2839 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 2840 return false; 2841 Next = skipOperatorName(Next); 2842 } else { 2843 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 2844 return false; 2845 for (; Next; Next = Next->Next) { 2846 if (Next->is(TT_TemplateOpener) && Next->MatchingParen) { 2847 Next = Next->MatchingParen; 2848 } else if (Next->is(tok::coloncolon)) { 2849 Next = Next->Next; 2850 if (!Next) 2851 return false; 2852 if (Next->is(tok::kw_operator)) { 2853 Next = skipOperatorName(Next->Next); 2854 break; 2855 } 2856 if (!Next->is(tok::identifier)) 2857 return false; 2858 } else if (isCppAttribute(IsCpp, *Next)) { 2859 Next = Next->MatchingParen; 2860 if (!Next) 2861 return false; 2862 } else if (Next->is(tok::l_paren)) { 2863 break; 2864 } else { 2865 return false; 2866 } 2867 } 2868 } 2869 2870 // Check whether parameter list can belong to a function declaration. 2871 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 2872 return false; 2873 // If the lines ends with "{", this is likely a function definition. 2874 if (Line.Last->is(tok::l_brace)) 2875 return true; 2876 if (Next->Next == Next->MatchingParen) 2877 return true; // Empty parentheses. 2878 // If there is an &/&& after the r_paren, this is likely a function. 2879 if (Next->MatchingParen->Next && 2880 Next->MatchingParen->Next->is(TT_PointerOrReference)) { 2881 return true; 2882 } 2883 2884 // Check for K&R C function definitions (and C++ function definitions with 2885 // unnamed parameters), e.g.: 2886 // int f(i) 2887 // { 2888 // return i + 1; 2889 // } 2890 // bool g(size_t = 0, bool b = false) 2891 // { 2892 // return !b; 2893 // } 2894 if (IsCpp && Next->Next && Next->Next->is(tok::identifier) && 2895 !Line.endsWith(tok::semi)) { 2896 return true; 2897 } 2898 2899 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 2900 Tok = Tok->Next) { 2901 if (Tok->is(TT_TypeDeclarationParen)) 2902 return true; 2903 if (Tok->isOneOf(tok::l_paren, TT_TemplateOpener) && Tok->MatchingParen) { 2904 Tok = Tok->MatchingParen; 2905 continue; 2906 } 2907 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 2908 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) { 2909 return true; 2910 } 2911 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 2912 Tok->Tok.isLiteral()) { 2913 return false; 2914 } 2915 } 2916 return false; 2917 } 2918 2919 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 2920 assert(Line.MightBeFunctionDecl); 2921 2922 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 2923 Style.AlwaysBreakAfterReturnType == 2924 FormatStyle::RTBS_TopLevelDefinitions) && 2925 Line.Level > 0) { 2926 return false; 2927 } 2928 2929 switch (Style.AlwaysBreakAfterReturnType) { 2930 case FormatStyle::RTBS_None: 2931 return false; 2932 case FormatStyle::RTBS_All: 2933 case FormatStyle::RTBS_TopLevel: 2934 return true; 2935 case FormatStyle::RTBS_AllDefinitions: 2936 case FormatStyle::RTBS_TopLevelDefinitions: 2937 return Line.mightBeFunctionDefinition(); 2938 } 2939 2940 return false; 2941 } 2942 2943 static bool mustBreakAfterAttributes(const FormatToken &Tok, 2944 const FormatStyle &Style) { 2945 switch (Style.BreakAfterAttributes) { 2946 case FormatStyle::ABS_Always: 2947 return true; 2948 case FormatStyle::ABS_Leave: 2949 return Tok.NewlinesBefore > 0; 2950 default: 2951 return false; 2952 } 2953 } 2954 2955 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) const { 2956 for (AnnotatedLine *ChildLine : Line.Children) 2957 calculateFormattingInformation(*ChildLine); 2958 2959 Line.First->TotalLength = 2960 Line.First->IsMultiline ? Style.ColumnLimit 2961 : Line.FirstStartColumn + Line.First->ColumnWidth; 2962 FormatToken *Current = Line.First->Next; 2963 bool InFunctionDecl = Line.MightBeFunctionDecl; 2964 bool AlignArrayOfStructures = 2965 (Style.AlignArrayOfStructures != FormatStyle::AIAS_None && 2966 Line.Type == LT_ArrayOfStructInitializer); 2967 if (AlignArrayOfStructures) 2968 calculateArrayInitializerColumnList(Line); 2969 2970 for (FormatToken *Tok = Current, *AfterLastAttribute = nullptr; Tok; 2971 Tok = Tok->Next) { 2972 if (isFunctionDeclarationName(Style.isCpp(), *Tok, Line)) { 2973 Tok->setType(TT_FunctionDeclarationName); 2974 if (AfterLastAttribute && 2975 mustBreakAfterAttributes(*AfterLastAttribute, Style)) { 2976 AfterLastAttribute->MustBreakBefore = true; 2977 Line.ReturnTypeWrapped = true; 2978 } 2979 break; 2980 } 2981 if (Tok->Previous->EndsCppAttributeGroup) 2982 AfterLastAttribute = Tok; 2983 } 2984 2985 while (Current) { 2986 const FormatToken *Prev = Current->Previous; 2987 if (Current->is(TT_LineComment)) { 2988 if (Prev->is(BK_BracedInit) && Prev->opensScope()) { 2989 Current->SpacesRequiredBefore = 2990 (Style.Cpp11BracedListStyle && !Style.SpacesInParentheses) ? 0 : 1; 2991 } else { 2992 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2993 } 2994 2995 // If we find a trailing comment, iterate backwards to determine whether 2996 // it seems to relate to a specific parameter. If so, break before that 2997 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2998 // to the previous line in: 2999 // SomeFunction(a, 3000 // b, // comment 3001 // c); 3002 if (!Current->HasUnescapedNewline) { 3003 for (FormatToken *Parameter = Current->Previous; Parameter; 3004 Parameter = Parameter->Previous) { 3005 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 3006 break; 3007 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 3008 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 3009 Parameter->HasUnescapedNewline) { 3010 Parameter->MustBreakBefore = true; 3011 } 3012 break; 3013 } 3014 } 3015 } 3016 } else if (Current->SpacesRequiredBefore == 0 && 3017 spaceRequiredBefore(Line, *Current)) { 3018 Current->SpacesRequiredBefore = 1; 3019 } 3020 3021 const auto &Children = Prev->Children; 3022 if (!Children.empty() && Children.back()->Last->is(TT_LineComment)) { 3023 Current->MustBreakBefore = true; 3024 } else { 3025 Current->MustBreakBefore = 3026 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 3027 if (!Current->MustBreakBefore && InFunctionDecl && 3028 Current->is(TT_FunctionDeclarationName)) { 3029 Current->MustBreakBefore = mustBreakForReturnType(Line); 3030 } 3031 } 3032 3033 Current->CanBreakBefore = 3034 Current->MustBreakBefore || canBreakBefore(Line, *Current); 3035 unsigned ChildSize = 0; 3036 if (Prev->Children.size() == 1) { 3037 FormatToken &LastOfChild = *Prev->Children[0]->Last; 3038 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 3039 : LastOfChild.TotalLength + 1; 3040 } 3041 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 3042 (Prev->Children.size() == 1 && 3043 Prev->Children[0]->First->MustBreakBefore) || 3044 Current->IsMultiline) { 3045 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 3046 } else { 3047 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 3048 ChildSize + Current->SpacesRequiredBefore; 3049 } 3050 3051 if (Current->is(TT_CtorInitializerColon)) 3052 InFunctionDecl = false; 3053 3054 // FIXME: Only calculate this if CanBreakBefore is true once static 3055 // initializers etc. are sorted out. 3056 // FIXME: Move magic numbers to a better place. 3057 3058 // Reduce penalty for aligning ObjC method arguments using the colon 3059 // alignment as this is the canonical way (still prefer fitting everything 3060 // into one line if possible). Trying to fit a whole expression into one 3061 // line should not force other line breaks (e.g. when ObjC method 3062 // expression is a part of other expression). 3063 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl); 3064 if (Style.Language == FormatStyle::LK_ObjC && 3065 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) { 3066 if (Current->ParameterIndex == 1) 3067 Current->SplitPenalty += 5 * Current->BindingStrength; 3068 } else { 3069 Current->SplitPenalty += 20 * Current->BindingStrength; 3070 } 3071 3072 Current = Current->Next; 3073 } 3074 3075 calculateUnbreakableTailLengths(Line); 3076 unsigned IndentLevel = Line.Level; 3077 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 3078 if (Current->Role) 3079 Current->Role->precomputeFormattingInfos(Current); 3080 if (Current->MatchingParen && 3081 Current->MatchingParen->opensBlockOrBlockTypeList(Style) && 3082 IndentLevel > 0) { 3083 --IndentLevel; 3084 } 3085 Current->IndentLevel = IndentLevel; 3086 if (Current->opensBlockOrBlockTypeList(Style)) 3087 ++IndentLevel; 3088 } 3089 3090 LLVM_DEBUG({ printDebugInfo(Line); }); 3091 } 3092 3093 void TokenAnnotator::calculateUnbreakableTailLengths( 3094 AnnotatedLine &Line) const { 3095 unsigned UnbreakableTailLength = 0; 3096 FormatToken *Current = Line.Last; 3097 while (Current) { 3098 Current->UnbreakableTailLength = UnbreakableTailLength; 3099 if (Current->CanBreakBefore || 3100 Current->isOneOf(tok::comment, tok::string_literal)) { 3101 UnbreakableTailLength = 0; 3102 } else { 3103 UnbreakableTailLength += 3104 Current->ColumnWidth + Current->SpacesRequiredBefore; 3105 } 3106 Current = Current->Previous; 3107 } 3108 } 3109 3110 void TokenAnnotator::calculateArrayInitializerColumnList( 3111 AnnotatedLine &Line) const { 3112 if (Line.First == Line.Last) 3113 return; 3114 auto *CurrentToken = Line.First; 3115 CurrentToken->ArrayInitializerLineStart = true; 3116 unsigned Depth = 0; 3117 while (CurrentToken != nullptr && CurrentToken != Line.Last) { 3118 if (CurrentToken->is(tok::l_brace)) { 3119 CurrentToken->IsArrayInitializer = true; 3120 if (CurrentToken->Next != nullptr) 3121 CurrentToken->Next->MustBreakBefore = true; 3122 CurrentToken = 3123 calculateInitializerColumnList(Line, CurrentToken->Next, Depth + 1); 3124 } else { 3125 CurrentToken = CurrentToken->Next; 3126 } 3127 } 3128 } 3129 3130 FormatToken *TokenAnnotator::calculateInitializerColumnList( 3131 AnnotatedLine &Line, FormatToken *CurrentToken, unsigned Depth) const { 3132 while (CurrentToken != nullptr && CurrentToken != Line.Last) { 3133 if (CurrentToken->is(tok::l_brace)) 3134 ++Depth; 3135 else if (CurrentToken->is(tok::r_brace)) 3136 --Depth; 3137 if (Depth == 2 && CurrentToken->isOneOf(tok::l_brace, tok::comma)) { 3138 CurrentToken = CurrentToken->Next; 3139 if (CurrentToken == nullptr) 3140 break; 3141 CurrentToken->StartsColumn = true; 3142 CurrentToken = CurrentToken->Previous; 3143 } 3144 CurrentToken = CurrentToken->Next; 3145 } 3146 return CurrentToken; 3147 } 3148 3149 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 3150 const FormatToken &Tok, 3151 bool InFunctionDecl) const { 3152 const FormatToken &Left = *Tok.Previous; 3153 const FormatToken &Right = Tok; 3154 3155 if (Left.is(tok::semi)) 3156 return 0; 3157 3158 // Language specific handling. 3159 if (Style.Language == FormatStyle::LK_Java) { 3160 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 3161 return 1; 3162 if (Right.is(Keywords.kw_implements)) 3163 return 2; 3164 if (Left.is(tok::comma) && Left.NestingLevel == 0) 3165 return 3; 3166 } else if (Style.isJavaScript()) { 3167 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 3168 return 100; 3169 if (Left.is(TT_JsTypeColon)) 3170 return 35; 3171 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 3172 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) { 3173 return 100; 3174 } 3175 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 3176 if (Left.opensScope() && Right.closesScope()) 3177 return 200; 3178 } else if (Style.isProto()) { 3179 if (Right.is(tok::l_square)) 3180 return 1; 3181 if (Right.is(tok::period)) 3182 return 500; 3183 } 3184 3185 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 3186 return 1; 3187 if (Right.is(tok::l_square)) { 3188 if (Left.is(tok::r_square)) 3189 return 200; 3190 // Slightly prefer formatting local lambda definitions like functions. 3191 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 3192 return 35; 3193 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 3194 TT_ArrayInitializerLSquare, 3195 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) { 3196 return 500; 3197 } 3198 } 3199 3200 if (Left.is(tok::coloncolon)) 3201 return 500; 3202 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 3203 Right.is(tok::kw_operator)) { 3204 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 3205 return 3; 3206 if (Left.is(TT_StartOfName)) 3207 return 110; 3208 if (InFunctionDecl && Right.NestingLevel == 0) 3209 return Style.PenaltyReturnTypeOnItsOwnLine; 3210 return 200; 3211 } 3212 if (Right.is(TT_PointerOrReference)) 3213 return 190; 3214 if (Right.is(TT_LambdaArrow)) 3215 return 110; 3216 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 3217 return 160; 3218 if (Left.is(TT_CastRParen)) 3219 return 100; 3220 if (Left.isOneOf(tok::kw_class, tok::kw_struct, tok::kw_union)) 3221 return 5000; 3222 if (Left.is(tok::comment)) 3223 return 1000; 3224 3225 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 3226 TT_CtorInitializerColon)) { 3227 return 2; 3228 } 3229 3230 if (Right.isMemberAccess()) { 3231 // Breaking before the "./->" of a chained call/member access is reasonably 3232 // cheap, as formatting those with one call per line is generally 3233 // desirable. In particular, it should be cheaper to break before the call 3234 // than it is to break inside a call's parameters, which could lead to weird 3235 // "hanging" indents. The exception is the very last "./->" to support this 3236 // frequent pattern: 3237 // 3238 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 3239 // dddddddd); 3240 // 3241 // which might otherwise be blown up onto many lines. Here, clang-format 3242 // won't produce "hanging" indents anyway as there is no other trailing 3243 // call. 3244 // 3245 // Also apply higher penalty is not a call as that might lead to a wrapping 3246 // like: 3247 // 3248 // aaaaaaa 3249 // .aaaaaaaaa.bbbbbbbb(cccccccc); 3250 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 3251 ? 150 3252 : 35; 3253 } 3254 3255 if (Right.is(TT_TrailingAnnotation) && 3256 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 3257 // Moving trailing annotations to the next line is fine for ObjC method 3258 // declarations. 3259 if (Line.startsWith(TT_ObjCMethodSpecifier)) 3260 return 10; 3261 // Generally, breaking before a trailing annotation is bad unless it is 3262 // function-like. It seems to be especially preferable to keep standard 3263 // annotations (i.e. "const", "final" and "override") on the same line. 3264 // Use a slightly higher penalty after ")" so that annotations like 3265 // "const override" are kept together. 3266 bool is_short_annotation = Right.TokenText.size() < 10; 3267 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 3268 } 3269 3270 // In for-loops, prefer breaking at ',' and ';'. 3271 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 3272 return 4; 3273 3274 // In Objective-C method expressions, prefer breaking before "param:" over 3275 // breaking after it. 3276 if (Right.is(TT_SelectorName)) 3277 return 0; 3278 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 3279 return Line.MightBeFunctionDecl ? 50 : 500; 3280 3281 // In Objective-C type declarations, avoid breaking after the category's 3282 // open paren (we'll prefer breaking after the protocol list's opening 3283 // angle bracket, if present). 3284 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous && 3285 Left.Previous->isOneOf(tok::identifier, tok::greater)) { 3286 return 500; 3287 } 3288 3289 if (Left.is(tok::l_paren) && Style.PenaltyBreakOpenParenthesis != 0) 3290 return Style.PenaltyBreakOpenParenthesis; 3291 if (Left.is(tok::l_paren) && InFunctionDecl && 3292 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) { 3293 return 100; 3294 } 3295 if (Left.is(tok::l_paren) && Left.Previous && 3296 (Left.Previous->isOneOf(tok::kw_for, tok::kw__Generic) || 3297 Left.Previous->isIf())) { 3298 return 1000; 3299 } 3300 if (Left.is(tok::equal) && InFunctionDecl) 3301 return 110; 3302 if (Right.is(tok::r_brace)) 3303 return 1; 3304 if (Left.is(TT_TemplateOpener)) 3305 return 100; 3306 if (Left.opensScope()) { 3307 // If we aren't aligning after opening parens/braces we can always break 3308 // here unless the style does not want us to place all arguments on the 3309 // next line. 3310 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign && 3311 (Left.ParameterCount <= 1 || Style.AllowAllArgumentsOnNextLine)) { 3312 return 0; 3313 } 3314 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle) 3315 return 19; 3316 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 3317 : 19; 3318 } 3319 if (Left.is(TT_JavaAnnotation)) 3320 return 50; 3321 3322 if (Left.is(TT_UnaryOperator)) 3323 return 60; 3324 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 3325 Left.Previous->isLabelString() && 3326 (Left.NextOperator || Left.OperatorIndex != 0)) { 3327 return 50; 3328 } 3329 if (Right.is(tok::plus) && Left.isLabelString() && 3330 (Right.NextOperator || Right.OperatorIndex != 0)) { 3331 return 25; 3332 } 3333 if (Left.is(tok::comma)) 3334 return 1; 3335 if (Right.is(tok::lessless) && Left.isLabelString() && 3336 (Right.NextOperator || Right.OperatorIndex != 1)) { 3337 return 25; 3338 } 3339 if (Right.is(tok::lessless)) { 3340 // Breaking at a << is really cheap. 3341 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) { 3342 // Slightly prefer to break before the first one in log-like statements. 3343 return 2; 3344 } 3345 return 1; 3346 } 3347 if (Left.ClosesTemplateDeclaration) 3348 return Style.PenaltyBreakTemplateDeclaration; 3349 if (Left.ClosesRequiresClause) 3350 return 0; 3351 if (Left.is(TT_ConditionalExpr)) 3352 return prec::Conditional; 3353 prec::Level Level = Left.getPrecedence(); 3354 if (Level == prec::Unknown) 3355 Level = Right.getPrecedence(); 3356 if (Level == prec::Assignment) 3357 return Style.PenaltyBreakAssignment; 3358 if (Level != prec::Unknown) 3359 return Level; 3360 3361 return 3; 3362 } 3363 3364 bool TokenAnnotator::spaceRequiredBeforeParens(const FormatToken &Right) const { 3365 if (Style.SpaceBeforeParens == FormatStyle::SBPO_Always) 3366 return true; 3367 if (Right.is(TT_OverloadedOperatorLParen) && 3368 Style.SpaceBeforeParensOptions.AfterOverloadedOperator) { 3369 return true; 3370 } 3371 if (Style.SpaceBeforeParensOptions.BeforeNonEmptyParentheses && 3372 Right.ParameterCount > 0) { 3373 return true; 3374 } 3375 return false; 3376 } 3377 3378 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 3379 const FormatToken &Left, 3380 const FormatToken &Right) const { 3381 if (Left.is(tok::kw_return) && 3382 !Right.isOneOf(tok::semi, tok::r_paren, tok::hashhash)) { 3383 return true; 3384 } 3385 if (Left.is(tok::kw_throw) && Right.is(tok::l_paren) && Right.MatchingParen && 3386 Right.MatchingParen->is(TT_CastRParen)) { 3387 return true; 3388 } 3389 if (Style.isJson() && Left.is(tok::string_literal) && Right.is(tok::colon)) 3390 return false; 3391 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 3392 return true; 3393 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 3394 Left.Tok.getObjCKeywordID() == tok::objc_property) { 3395 return true; 3396 } 3397 if (Right.is(tok::hashhash)) 3398 return Left.is(tok::hash); 3399 if (Left.isOneOf(tok::hashhash, tok::hash)) 3400 return Right.is(tok::hash); 3401 if ((Left.is(tok::l_paren) && Right.is(tok::r_paren)) || 3402 (Left.is(tok::l_brace) && Left.isNot(BK_Block) && 3403 Right.is(tok::r_brace) && Right.isNot(BK_Block))) { 3404 return Style.SpaceInEmptyParentheses; 3405 } 3406 if (Style.SpacesInConditionalStatement) { 3407 const FormatToken *LeftParen = nullptr; 3408 if (Left.is(tok::l_paren)) 3409 LeftParen = &Left; 3410 else if (Right.is(tok::r_paren) && Right.MatchingParen) 3411 LeftParen = Right.MatchingParen; 3412 if (LeftParen && LeftParen->Previous && 3413 isKeywordWithCondition(*LeftParen->Previous)) { 3414 return true; 3415 } 3416 } 3417 3418 // trailing return type 'auto': []() -> auto {}, auto foo() -> auto {} 3419 if (Left.is(tok::kw_auto) && Right.isOneOf(TT_LambdaLBrace, TT_FunctionLBrace, 3420 // function return type 'auto' 3421 TT_FunctionTypeLParen)) { 3422 return true; 3423 } 3424 3425 // auto{x} auto(x) 3426 if (Left.is(tok::kw_auto) && Right.isOneOf(tok::l_paren, tok::l_brace)) 3427 return false; 3428 3429 // operator co_await(x) 3430 if (Right.is(tok::l_paren) && Left.is(tok::kw_co_await) && Left.Previous && 3431 Left.Previous->is(tok::kw_operator)) { 3432 return false; 3433 } 3434 // co_await (x), co_yield (x), co_return (x) 3435 if (Left.isOneOf(tok::kw_co_await, tok::kw_co_yield, tok::kw_co_return) && 3436 !Right.isOneOf(tok::semi, tok::r_paren)) { 3437 return true; 3438 } 3439 3440 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) { 3441 return (Right.is(TT_CastRParen) || 3442 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 3443 ? Style.SpacesInCStyleCastParentheses 3444 : Style.SpacesInParentheses; 3445 } 3446 if (Right.isOneOf(tok::semi, tok::comma)) 3447 return false; 3448 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) { 3449 bool IsLightweightGeneric = Right.MatchingParen && 3450 Right.MatchingParen->Next && 3451 Right.MatchingParen->Next->is(tok::colon); 3452 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList; 3453 } 3454 if (Right.is(tok::less) && Left.is(tok::kw_template)) 3455 return Style.SpaceAfterTemplateKeyword; 3456 if (Left.isOneOf(tok::exclaim, tok::tilde)) 3457 return false; 3458 if (Left.is(tok::at) && 3459 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 3460 tok::numeric_constant, tok::l_paren, tok::l_brace, 3461 tok::kw_true, tok::kw_false)) { 3462 return false; 3463 } 3464 if (Left.is(tok::colon)) 3465 return !Left.is(TT_ObjCMethodExpr); 3466 if (Left.is(tok::coloncolon)) 3467 return false; 3468 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 3469 if (Style.Language == FormatStyle::LK_TextProto || 3470 (Style.Language == FormatStyle::LK_Proto && 3471 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 3472 // Format empty list as `<>`. 3473 if (Left.is(tok::less) && Right.is(tok::greater)) 3474 return false; 3475 return !Style.Cpp11BracedListStyle; 3476 } 3477 // Don't attempt to format operator<(), as it is handled later. 3478 if (Right.isNot(TT_OverloadedOperatorLParen)) 3479 return false; 3480 } 3481 if (Right.is(tok::ellipsis)) { 3482 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 3483 Left.Previous->is(tok::kw_case)); 3484 } 3485 if (Left.is(tok::l_square) && Right.is(tok::amp)) 3486 return Style.SpacesInSquareBrackets; 3487 if (Right.is(TT_PointerOrReference)) { 3488 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 3489 if (!Left.MatchingParen) 3490 return true; 3491 FormatToken *TokenBeforeMatchingParen = 3492 Left.MatchingParen->getPreviousNonComment(); 3493 if (!TokenBeforeMatchingParen || !Left.is(TT_TypeDeclarationParen)) 3494 return true; 3495 } 3496 // Add a space if the previous token is a pointer qualifier or the closing 3497 // parenthesis of __attribute__(()) expression and the style requires spaces 3498 // after pointer qualifiers. 3499 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_After || 3500 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 3501 (Left.is(TT_AttributeParen) || 3502 Left.canBePointerOrReferenceQualifier())) { 3503 return true; 3504 } 3505 if (Left.Tok.isLiteral()) 3506 return true; 3507 // for (auto a = 0, b = 0; const auto & c : {1, 2, 3}) 3508 if (Left.isTypeOrIdentifier() && Right.Next && Right.Next->Next && 3509 Right.Next->Next->is(TT_RangeBasedForLoopColon)) { 3510 return getTokenPointerOrReferenceAlignment(Right) != 3511 FormatStyle::PAS_Left; 3512 } 3513 return !Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 3514 (getTokenPointerOrReferenceAlignment(Right) != 3515 FormatStyle::PAS_Left || 3516 (Line.IsMultiVariableDeclStmt && 3517 (Left.NestingLevel == 0 || 3518 (Left.NestingLevel == 1 && startsWithInitStatement(Line))))); 3519 } 3520 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 3521 (!Left.is(TT_PointerOrReference) || 3522 (getTokenPointerOrReferenceAlignment(Left) != FormatStyle::PAS_Right && 3523 !Line.IsMultiVariableDeclStmt))) { 3524 return true; 3525 } 3526 if (Left.is(TT_PointerOrReference)) { 3527 // Add a space if the next token is a pointer qualifier and the style 3528 // requires spaces before pointer qualifiers. 3529 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Before || 3530 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 3531 Right.canBePointerOrReferenceQualifier()) { 3532 return true; 3533 } 3534 // & 1 3535 if (Right.Tok.isLiteral()) 3536 return true; 3537 // & /* comment 3538 if (Right.is(TT_BlockComment)) 3539 return true; 3540 // foo() -> const Bar * override/final 3541 if (Right.isOneOf(Keywords.kw_override, Keywords.kw_final, 3542 tok::kw_noexcept) && 3543 !Right.is(TT_StartOfName)) { 3544 return true; 3545 } 3546 // & { 3547 if (Right.is(tok::l_brace) && Right.is(BK_Block)) 3548 return true; 3549 // for (auto a = 0, b = 0; const auto& c : {1, 2, 3}) 3550 if (Left.Previous && Left.Previous->isTypeOrIdentifier() && Right.Next && 3551 Right.Next->is(TT_RangeBasedForLoopColon)) { 3552 return getTokenPointerOrReferenceAlignment(Left) != 3553 FormatStyle::PAS_Right; 3554 } 3555 if (Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 3556 tok::l_paren)) { 3557 return false; 3558 } 3559 if (getTokenPointerOrReferenceAlignment(Left) == FormatStyle::PAS_Right) 3560 return false; 3561 // FIXME: Setting IsMultiVariableDeclStmt for the whole line is error-prone, 3562 // because it does not take into account nested scopes like lambdas. 3563 // In multi-variable declaration statements, attach */& to the variable 3564 // independently of the style. However, avoid doing it if we are in a nested 3565 // scope, e.g. lambda. We still need to special-case statements with 3566 // initializers. 3567 if (Line.IsMultiVariableDeclStmt && 3568 (Left.NestingLevel == Line.First->NestingLevel || 3569 ((Left.NestingLevel == Line.First->NestingLevel + 1) && 3570 startsWithInitStatement(Line)))) { 3571 return false; 3572 } 3573 return Left.Previous && !Left.Previous->isOneOf( 3574 tok::l_paren, tok::coloncolon, tok::l_square); 3575 } 3576 // Ensure right pointer alignment with ellipsis e.g. int *...P 3577 if (Left.is(tok::ellipsis) && Left.Previous && 3578 Left.Previous->isOneOf(tok::star, tok::amp, tok::ampamp)) { 3579 return Style.PointerAlignment != FormatStyle::PAS_Right; 3580 } 3581 3582 if (Right.is(tok::star) && Left.is(tok::l_paren)) 3583 return false; 3584 if (Left.is(tok::star) && Right.isOneOf(tok::star, tok::amp, tok::ampamp)) 3585 return false; 3586 if (Right.isOneOf(tok::star, tok::amp, tok::ampamp)) { 3587 const FormatToken *Previous = &Left; 3588 while (Previous && !Previous->is(tok::kw_operator)) { 3589 if (Previous->is(tok::identifier) || Previous->isSimpleTypeSpecifier()) { 3590 Previous = Previous->getPreviousNonComment(); 3591 continue; 3592 } 3593 if (Previous->is(TT_TemplateCloser) && Previous->MatchingParen) { 3594 Previous = Previous->MatchingParen->getPreviousNonComment(); 3595 continue; 3596 } 3597 if (Previous->is(tok::coloncolon)) { 3598 Previous = Previous->getPreviousNonComment(); 3599 continue; 3600 } 3601 break; 3602 } 3603 // Space between the type and the * in: 3604 // operator void*() 3605 // operator char*() 3606 // operator void const*() 3607 // operator void volatile*() 3608 // operator /*comment*/ const char*() 3609 // operator volatile /*comment*/ char*() 3610 // operator Foo*() 3611 // operator C<T>*() 3612 // operator std::Foo*() 3613 // operator C<T>::D<U>*() 3614 // dependent on PointerAlignment style. 3615 if (Previous) { 3616 if (Previous->endsSequence(tok::kw_operator)) 3617 return Style.PointerAlignment != FormatStyle::PAS_Left; 3618 if (Previous->is(tok::kw_const) || Previous->is(tok::kw_volatile)) { 3619 return (Style.PointerAlignment != FormatStyle::PAS_Left) || 3620 (Style.SpaceAroundPointerQualifiers == 3621 FormatStyle::SAPQ_After) || 3622 (Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both); 3623 } 3624 } 3625 } 3626 const auto SpaceRequiredForArrayInitializerLSquare = 3627 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 3628 return Style.SpacesInContainerLiterals || 3629 ((Style.Language == FormatStyle::LK_Proto || 3630 Style.Language == FormatStyle::LK_TextProto) && 3631 !Style.Cpp11BracedListStyle && 3632 LSquareTok.endsSequence(tok::l_square, tok::colon, 3633 TT_SelectorName)); 3634 }; 3635 if (Left.is(tok::l_square)) { 3636 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 3637 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 3638 (Left.isOneOf(TT_ArraySubscriptLSquare, TT_StructuredBindingLSquare, 3639 TT_LambdaLSquare) && 3640 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 3641 } 3642 if (Right.is(tok::r_square)) { 3643 return Right.MatchingParen && 3644 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 3645 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 3646 Style)) || 3647 (Style.SpacesInSquareBrackets && 3648 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 3649 TT_StructuredBindingLSquare, 3650 TT_LambdaLSquare)) || 3651 Right.MatchingParen->is(TT_AttributeParen)); 3652 } 3653 if (Right.is(tok::l_square) && 3654 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 3655 TT_DesignatedInitializerLSquare, 3656 TT_StructuredBindingLSquare, TT_AttributeSquare) && 3657 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral) && 3658 !(!Left.is(tok::r_square) && Style.SpaceBeforeSquareBrackets && 3659 Right.is(TT_ArraySubscriptLSquare))) { 3660 return false; 3661 } 3662 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 3663 return !Left.Children.empty(); // No spaces in "{}". 3664 if ((Left.is(tok::l_brace) && Left.isNot(BK_Block)) || 3665 (Right.is(tok::r_brace) && Right.MatchingParen && 3666 Right.MatchingParen->isNot(BK_Block))) { 3667 return Style.Cpp11BracedListStyle ? Style.SpacesInParentheses : true; 3668 } 3669 if (Left.is(TT_BlockComment)) { 3670 // No whitespace in x(/*foo=*/1), except for JavaScript. 3671 return Style.isJavaScript() || !Left.TokenText.endswith("=*/"); 3672 } 3673 3674 // Space between template and attribute. 3675 // e.g. template <typename T> [[nodiscard]] ... 3676 if (Left.is(TT_TemplateCloser) && Right.is(TT_AttributeSquare)) 3677 return true; 3678 // Space before parentheses common for all languages 3679 if (Right.is(tok::l_paren)) { 3680 if (Left.is(TT_TemplateCloser) && Right.isNot(TT_FunctionTypeLParen)) 3681 return spaceRequiredBeforeParens(Right); 3682 if (Left.isOneOf(TT_RequiresClause, 3683 TT_RequiresClauseInARequiresExpression)) { 3684 return Style.SpaceBeforeParensOptions.AfterRequiresInClause || 3685 spaceRequiredBeforeParens(Right); 3686 } 3687 if (Left.is(TT_RequiresExpression)) { 3688 return Style.SpaceBeforeParensOptions.AfterRequiresInExpression || 3689 spaceRequiredBeforeParens(Right); 3690 } 3691 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 3692 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) { 3693 return true; 3694 } 3695 if (Left.is(TT_ForEachMacro)) { 3696 return Style.SpaceBeforeParensOptions.AfterForeachMacros || 3697 spaceRequiredBeforeParens(Right); 3698 } 3699 if (Left.is(TT_IfMacro)) { 3700 return Style.SpaceBeforeParensOptions.AfterIfMacros || 3701 spaceRequiredBeforeParens(Right); 3702 } 3703 if (Line.Type == LT_ObjCDecl) 3704 return true; 3705 if (Left.is(tok::semi)) 3706 return true; 3707 if (Left.isOneOf(tok::pp_elif, tok::kw_for, tok::kw_while, tok::kw_switch, 3708 tok::kw_case, TT_ForEachMacro, TT_ObjCForIn) || 3709 Left.isIf(Line.Type != LT_PreprocessorDirective) || 3710 Right.is(TT_ConditionLParen)) { 3711 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3712 spaceRequiredBeforeParens(Right); 3713 } 3714 3715 // TODO add Operator overloading specific Options to 3716 // SpaceBeforeParensOptions 3717 if (Right.is(TT_OverloadedOperatorLParen)) 3718 return spaceRequiredBeforeParens(Right); 3719 // Function declaration or definition 3720 if (Line.MightBeFunctionDecl && (Left.is(TT_FunctionDeclarationName))) { 3721 if (Line.mightBeFunctionDefinition()) { 3722 return Style.SpaceBeforeParensOptions.AfterFunctionDefinitionName || 3723 spaceRequiredBeforeParens(Right); 3724 } else { 3725 return Style.SpaceBeforeParensOptions.AfterFunctionDeclarationName || 3726 spaceRequiredBeforeParens(Right); 3727 } 3728 } 3729 // Lambda 3730 if (Line.Type != LT_PreprocessorDirective && Left.is(tok::r_square) && 3731 Left.MatchingParen && Left.MatchingParen->is(TT_LambdaLSquare)) { 3732 return Style.SpaceBeforeParensOptions.AfterFunctionDefinitionName || 3733 spaceRequiredBeforeParens(Right); 3734 } 3735 if (!Left.Previous || Left.Previous->isNot(tok::period)) { 3736 if (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch)) { 3737 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3738 spaceRequiredBeforeParens(Right); 3739 } 3740 if (Left.isOneOf(tok::kw_new, tok::kw_delete)) { 3741 return ((!Line.MightBeFunctionDecl || !Left.Previous) && 3742 Style.SpaceBeforeParens != FormatStyle::SBPO_Never) || 3743 spaceRequiredBeforeParens(Right); 3744 } 3745 3746 if (Left.is(tok::r_square) && Left.MatchingParen && 3747 Left.MatchingParen->Previous && 3748 Left.MatchingParen->Previous->is(tok::kw_delete)) { 3749 return (Style.SpaceBeforeParens != FormatStyle::SBPO_Never) || 3750 spaceRequiredBeforeParens(Right); 3751 } 3752 } 3753 // Handle builtins like identifiers. 3754 if (Line.Type != LT_PreprocessorDirective && 3755 (Left.Tok.getIdentifierInfo() || Left.is(tok::r_paren))) { 3756 return spaceRequiredBeforeParens(Right); 3757 } 3758 return false; 3759 } 3760 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 3761 return false; 3762 if (Right.is(TT_UnaryOperator)) { 3763 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 3764 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 3765 } 3766 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 3767 tok::r_paren) || 3768 Left.isSimpleTypeSpecifier()) && 3769 Right.is(tok::l_brace) && Right.getNextNonComment() && 3770 Right.isNot(BK_Block)) { 3771 return false; 3772 } 3773 if (Left.is(tok::period) || Right.is(tok::period)) 3774 return false; 3775 // u#str, U#str, L#str, u8#str 3776 // uR#str, UR#str, LR#str, u8R#str 3777 if (Right.is(tok::hash) && Left.is(tok::identifier) && 3778 (Left.TokenText == "L" || Left.TokenText == "u" || 3779 Left.TokenText == "U" || Left.TokenText == "u8" || 3780 Left.TokenText == "LR" || Left.TokenText == "uR" || 3781 Left.TokenText == "UR" || Left.TokenText == "u8R")) { 3782 return false; 3783 } 3784 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 3785 Left.MatchingParen->Previous && 3786 (Left.MatchingParen->Previous->is(tok::period) || 3787 Left.MatchingParen->Previous->is(tok::coloncolon))) { 3788 // Java call to generic function with explicit type: 3789 // A.<B<C<...>>>DoSomething(); 3790 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference. 3791 return false; 3792 } 3793 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 3794 return false; 3795 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at)) { 3796 // Objective-C dictionary literal -> no space after opening brace. 3797 return false; 3798 } 3799 if (Right.is(tok::r_brace) && Right.MatchingParen && 3800 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at)) { 3801 // Objective-C dictionary literal -> no space before closing brace. 3802 return false; 3803 } 3804 if (Right.getType() == TT_TrailingAnnotation && 3805 Right.isOneOf(tok::amp, tok::ampamp) && 3806 Left.isOneOf(tok::kw_const, tok::kw_volatile) && 3807 (!Right.Next || Right.Next->is(tok::semi))) { 3808 // Match const and volatile ref-qualifiers without any additional 3809 // qualifiers such as 3810 // void Fn() const &; 3811 return getTokenReferenceAlignment(Right) != FormatStyle::PAS_Left; 3812 } 3813 3814 return true; 3815 } 3816 3817 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 3818 const FormatToken &Right) const { 3819 const FormatToken &Left = *Right.Previous; 3820 3821 // If the token is finalized don't touch it (as it could be in a 3822 // clang-format-off section). 3823 if (Left.Finalized) 3824 return Right.hasWhitespaceBefore(); 3825 3826 // Never ever merge two words. 3827 if (Keywords.isWordLike(Right) && Keywords.isWordLike(Left)) 3828 return true; 3829 3830 // Leave a space between * and /* to avoid C4138 `comment end` found outside 3831 // of comment. 3832 if (Left.is(tok::star) && Right.is(tok::comment)) 3833 return true; 3834 3835 if (Style.isCpp()) { 3836 // Space between UDL and dot: auto b = 4s .count(); 3837 if (Right.is(tok::period) && Left.is(tok::numeric_constant)) 3838 return true; 3839 // Space between import <iostream>. 3840 // or import .....; 3841 if (Left.is(Keywords.kw_import) && Right.isOneOf(tok::less, tok::ellipsis)) 3842 return true; 3843 // Space between `module :` and `import :`. 3844 if (Left.isOneOf(Keywords.kw_module, Keywords.kw_import) && 3845 Right.is(TT_ModulePartitionColon)) { 3846 return true; 3847 } 3848 // No space between import foo:bar but keep a space between import :bar; 3849 if (Left.is(tok::identifier) && Right.is(TT_ModulePartitionColon)) 3850 return false; 3851 // No space between :bar; 3852 if (Left.is(TT_ModulePartitionColon) && 3853 Right.isOneOf(tok::identifier, tok::kw_private)) { 3854 return false; 3855 } 3856 if (Left.is(tok::ellipsis) && Right.is(tok::identifier) && 3857 Line.First->is(Keywords.kw_import)) { 3858 return false; 3859 } 3860 // Space in __attribute__((attr)) ::type. 3861 if (Left.is(TT_AttributeParen) && Right.is(tok::coloncolon)) 3862 return true; 3863 3864 if (Left.is(tok::kw_operator)) 3865 return Right.is(tok::coloncolon); 3866 if (Right.is(tok::l_brace) && Right.is(BK_BracedInit) && 3867 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList) { 3868 return true; 3869 } 3870 if (Left.is(tok::less) && Left.is(TT_OverloadedOperator) && 3871 Right.is(TT_TemplateOpener)) { 3872 return true; 3873 } 3874 } else if (Style.Language == FormatStyle::LK_Proto || 3875 Style.Language == FormatStyle::LK_TextProto) { 3876 if (Right.is(tok::period) && 3877 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 3878 Keywords.kw_repeated, Keywords.kw_extend)) { 3879 return true; 3880 } 3881 if (Right.is(tok::l_paren) && 3882 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) { 3883 return true; 3884 } 3885 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 3886 return true; 3887 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 3888 if (Left.is(tok::slash) || Right.is(tok::slash)) 3889 return false; 3890 if (Left.MatchingParen && 3891 Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 3892 Right.isOneOf(tok::l_brace, tok::less)) { 3893 return !Style.Cpp11BracedListStyle; 3894 } 3895 // A percent is probably part of a formatting specification, such as %lld. 3896 if (Left.is(tok::percent)) 3897 return false; 3898 // Preserve the existence of a space before a percent for cases like 0x%04x 3899 // and "%d %d" 3900 if (Left.is(tok::numeric_constant) && Right.is(tok::percent)) 3901 return Right.hasWhitespaceBefore(); 3902 } else if (Style.isJson()) { 3903 if (Right.is(tok::colon)) 3904 return false; 3905 } else if (Style.isCSharp()) { 3906 // Require spaces around '{' and before '}' unless they appear in 3907 // interpolated strings. Interpolated strings are merged into a single token 3908 // so cannot have spaces inserted by this function. 3909 3910 // No space between 'this' and '[' 3911 if (Left.is(tok::kw_this) && Right.is(tok::l_square)) 3912 return false; 3913 3914 // No space between 'new' and '(' 3915 if (Left.is(tok::kw_new) && Right.is(tok::l_paren)) 3916 return false; 3917 3918 // Space before { (including space within '{ {'). 3919 if (Right.is(tok::l_brace)) 3920 return true; 3921 3922 // Spaces inside braces. 3923 if (Left.is(tok::l_brace) && Right.isNot(tok::r_brace)) 3924 return true; 3925 3926 if (Left.isNot(tok::l_brace) && Right.is(tok::r_brace)) 3927 return true; 3928 3929 // Spaces around '=>'. 3930 if (Left.is(TT_FatArrow) || Right.is(TT_FatArrow)) 3931 return true; 3932 3933 // No spaces around attribute target colons 3934 if (Left.is(TT_AttributeColon) || Right.is(TT_AttributeColon)) 3935 return false; 3936 3937 // space between type and variable e.g. Dictionary<string,string> foo; 3938 if (Left.is(TT_TemplateCloser) && Right.is(TT_StartOfName)) 3939 return true; 3940 3941 // spaces inside square brackets. 3942 if (Left.is(tok::l_square) || Right.is(tok::r_square)) 3943 return Style.SpacesInSquareBrackets; 3944 3945 // No space before ? in nullable types. 3946 if (Right.is(TT_CSharpNullable)) 3947 return false; 3948 3949 // No space before null forgiving '!'. 3950 if (Right.is(TT_NonNullAssertion)) 3951 return false; 3952 3953 // No space between consecutive commas '[,,]'. 3954 if (Left.is(tok::comma) && Right.is(tok::comma)) 3955 return false; 3956 3957 // space after var in `var (key, value)` 3958 if (Left.is(Keywords.kw_var) && Right.is(tok::l_paren)) 3959 return true; 3960 3961 // space between keywords and paren e.g. "using (" 3962 if (Right.is(tok::l_paren)) { 3963 if (Left.isOneOf(tok::kw_using, Keywords.kw_async, Keywords.kw_when, 3964 Keywords.kw_lock)) { 3965 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3966 spaceRequiredBeforeParens(Right); 3967 } 3968 } 3969 3970 // space between method modifier and opening parenthesis of a tuple return 3971 // type 3972 if (Left.isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 3973 tok::kw_virtual, tok::kw_extern, tok::kw_static, 3974 Keywords.kw_internal, Keywords.kw_abstract, 3975 Keywords.kw_sealed, Keywords.kw_override, 3976 Keywords.kw_async, Keywords.kw_unsafe) && 3977 Right.is(tok::l_paren)) { 3978 return true; 3979 } 3980 } else if (Style.isJavaScript()) { 3981 if (Left.is(TT_FatArrow)) 3982 return true; 3983 // for await ( ... 3984 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 3985 Left.Previous->is(tok::kw_for)) { 3986 return true; 3987 } 3988 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 3989 Right.MatchingParen) { 3990 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 3991 // An async arrow function, for example: `x = async () => foo();`, 3992 // as opposed to calling a function called async: `x = async();` 3993 if (Next && Next->is(TT_FatArrow)) 3994 return true; 3995 } 3996 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 3997 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) { 3998 return false; 3999 } 4000 // In tagged template literals ("html`bar baz`"), there is no space between 4001 // the tag identifier and the template string. 4002 if (Keywords.IsJavaScriptIdentifier(Left, 4003 /* AcceptIdentifierName= */ false) && 4004 Right.is(TT_TemplateString)) { 4005 return false; 4006 } 4007 if (Right.is(tok::star) && 4008 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) { 4009 return false; 4010 } 4011 if (Right.isOneOf(tok::l_brace, tok::l_square) && 4012 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 4013 Keywords.kw_extends, Keywords.kw_implements)) { 4014 return true; 4015 } 4016 if (Right.is(tok::l_paren)) { 4017 // JS methods can use some keywords as names (e.g. `delete()`). 4018 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 4019 return false; 4020 // Valid JS method names can include keywords, e.g. `foo.delete()` or 4021 // `bar.instanceof()`. Recognize call positions by preceding period. 4022 if (Left.Previous && Left.Previous->is(tok::period) && 4023 Left.Tok.getIdentifierInfo()) { 4024 return false; 4025 } 4026 // Additional unary JavaScript operators that need a space after. 4027 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 4028 tok::kw_void)) { 4029 return true; 4030 } 4031 } 4032 // `foo as const;` casts into a const type. 4033 if (Left.endsSequence(tok::kw_const, Keywords.kw_as)) 4034 return false; 4035 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 4036 tok::kw_const) || 4037 // "of" is only a keyword if it appears after another identifier 4038 // (e.g. as "const x of y" in a for loop), or after a destructuring 4039 // operation (const [x, y] of z, const {a, b} of c). 4040 (Left.is(Keywords.kw_of) && Left.Previous && 4041 (Left.Previous->is(tok::identifier) || 4042 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 4043 (!Left.Previous || !Left.Previous->is(tok::period))) { 4044 return true; 4045 } 4046 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 4047 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) { 4048 return false; 4049 } 4050 if (Left.is(Keywords.kw_as) && 4051 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) { 4052 return true; 4053 } 4054 if (Left.is(tok::kw_default) && Left.Previous && 4055 Left.Previous->is(tok::kw_export)) { 4056 return true; 4057 } 4058 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 4059 return true; 4060 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 4061 return false; 4062 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 4063 return false; 4064 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 4065 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) { 4066 return false; 4067 } 4068 if (Left.is(tok::ellipsis)) 4069 return false; 4070 if (Left.is(TT_TemplateCloser) && 4071 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 4072 Keywords.kw_implements, Keywords.kw_extends)) { 4073 // Type assertions ('<type>expr') are not followed by whitespace. Other 4074 // locations that should have whitespace following are identified by the 4075 // above set of follower tokens. 4076 return false; 4077 } 4078 if (Right.is(TT_NonNullAssertion)) 4079 return false; 4080 if (Left.is(TT_NonNullAssertion) && 4081 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) { 4082 return true; // "x! as string", "x! in y" 4083 } 4084 } else if (Style.Language == FormatStyle::LK_Java) { 4085 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 4086 return true; 4087 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) { 4088 return Style.SpaceBeforeParensOptions.AfterControlStatements || 4089 spaceRequiredBeforeParens(Right); 4090 } 4091 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 4092 tok::kw_protected) || 4093 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 4094 Keywords.kw_native)) && 4095 Right.is(TT_TemplateOpener)) { 4096 return true; 4097 } 4098 } else if (Style.isVerilog()) { 4099 // Add space between things in a primitive's state table unless in a 4100 // transition like `(0?)`. 4101 if ((Left.is(TT_VerilogTableItem) && 4102 !Right.isOneOf(tok::r_paren, tok::semi)) || 4103 (Right.is(TT_VerilogTableItem) && Left.isNot(tok::l_paren))) { 4104 const FormatToken *Next = Right.getNextNonComment(); 4105 return !(Next && Next->is(tok::r_paren)); 4106 } 4107 // Don't add space within a delay like `#0`. 4108 if (Left.isNot(TT_BinaryOperator) && 4109 Left.isOneOf(Keywords.kw_verilogHash, Keywords.kw_verilogHashHash)) { 4110 return false; 4111 } 4112 // Add space after a delay. 4113 if (!Right.is(tok::semi) && 4114 (Left.endsSequence(tok::numeric_constant, Keywords.kw_verilogHash) || 4115 Left.endsSequence(tok::numeric_constant, 4116 Keywords.kw_verilogHashHash) || 4117 (Left.is(tok::r_paren) && Left.MatchingParen && 4118 Left.MatchingParen->endsSequence(tok::l_paren, tok::at)))) { 4119 return true; 4120 } 4121 // Don't add embedded spaces in a number literal like `16'h1?ax` or an array 4122 // literal like `'{}`. 4123 if (Left.is(Keywords.kw_apostrophe) || 4124 (Left.is(TT_VerilogNumberBase) && Right.is(tok::numeric_constant))) { 4125 return false; 4126 } 4127 // Add space between the type name and dimension like `logic [1:0]`. 4128 if (Right.is(tok::l_square) && 4129 Left.isOneOf(TT_VerilogDimensionedTypeName, Keywords.kw_function)) { 4130 return true; 4131 } 4132 // Don't add spaces between a casting type and the quote or repetition count 4133 // and the brace. 4134 if ((Right.is(Keywords.kw_apostrophe) || 4135 (Right.is(BK_BracedInit) && Right.is(tok::l_brace))) && 4136 !(Left.isOneOf(Keywords.kw_assign, Keywords.kw_unique) || 4137 Keywords.isVerilogWordOperator(Left)) && 4138 (Left.isOneOf(tok::r_square, tok::r_paren, tok::r_brace, 4139 tok::numeric_constant) || 4140 Keywords.isWordLike(Left))) { 4141 return false; 4142 } 4143 // Add space in attribute like `(* ASYNC_REG = "TRUE" *)`. 4144 if (Left.endsSequence(tok::star, tok::l_paren) && Right.is(tok::identifier)) 4145 return true; 4146 } 4147 if (Left.is(TT_ImplicitStringLiteral)) 4148 return Right.hasWhitespaceBefore(); 4149 if (Line.Type == LT_ObjCMethodDecl) { 4150 if (Left.is(TT_ObjCMethodSpecifier)) 4151 return true; 4152 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right)) { 4153 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a 4154 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class 4155 // method declaration. 4156 return false; 4157 } 4158 } 4159 if (Line.Type == LT_ObjCProperty && 4160 (Right.is(tok::equal) || Left.is(tok::equal))) { 4161 return false; 4162 } 4163 4164 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 4165 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) { 4166 return true; 4167 } 4168 if (Left.is(tok::comma) && !Right.is(TT_OverloadedOperatorLParen)) 4169 return true; 4170 if (Right.is(tok::comma)) 4171 return false; 4172 if (Right.is(TT_ObjCBlockLParen)) 4173 return true; 4174 if (Right.is(TT_CtorInitializerColon)) 4175 return Style.SpaceBeforeCtorInitializerColon; 4176 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 4177 return false; 4178 if (Right.is(TT_RangeBasedForLoopColon) && 4179 !Style.SpaceBeforeRangeBasedForLoopColon) { 4180 return false; 4181 } 4182 if (Left.is(TT_BitFieldColon)) { 4183 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 4184 Style.BitFieldColonSpacing == FormatStyle::BFCS_After; 4185 } 4186 if (Right.is(tok::colon)) { 4187 if (Right.is(TT_GotoLabelColon) || 4188 (!Style.isVerilog() && 4189 Line.First->isOneOf(tok::kw_default, tok::kw_case))) { 4190 return Style.SpaceBeforeCaseColon; 4191 } 4192 if (Line.First->isOneOf(tok::kw_default, tok::kw_case)) 4193 return Style.SpaceBeforeCaseColon; 4194 const FormatToken *Next = Right.getNextNonComment(); 4195 if (!Next || Next->is(tok::semi)) 4196 return false; 4197 if (Right.is(TT_ObjCMethodExpr)) 4198 return false; 4199 if (Left.is(tok::question)) 4200 return false; 4201 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 4202 return false; 4203 if (Right.is(TT_DictLiteral)) 4204 return Style.SpacesInContainerLiterals; 4205 if (Right.is(TT_AttributeColon)) 4206 return false; 4207 if (Right.is(TT_CSharpNamedArgumentColon)) 4208 return false; 4209 if (Right.is(TT_GenericSelectionColon)) 4210 return false; 4211 if (Right.is(TT_BitFieldColon)) { 4212 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 4213 Style.BitFieldColonSpacing == FormatStyle::BFCS_Before; 4214 } 4215 return true; 4216 } 4217 // Do not merge "- -" into "--". 4218 if ((Left.isOneOf(tok::minus, tok::minusminus) && 4219 Right.isOneOf(tok::minus, tok::minusminus)) || 4220 (Left.isOneOf(tok::plus, tok::plusplus) && 4221 Right.isOneOf(tok::plus, tok::plusplus))) { 4222 return true; 4223 } 4224 if (Left.is(TT_UnaryOperator)) { 4225 if (!Right.is(tok::l_paren)) { 4226 // The alternative operators for ~ and ! are "compl" and "not". 4227 // If they are used instead, we do not want to combine them with 4228 // the token to the right, unless that is a left paren. 4229 if (Left.is(tok::exclaim) && Left.TokenText == "not") 4230 return true; 4231 if (Left.is(tok::tilde) && Left.TokenText == "compl") 4232 return true; 4233 // Lambda captures allow for a lone &, so "&]" needs to be properly 4234 // handled. 4235 if (Left.is(tok::amp) && Right.is(tok::r_square)) 4236 return Style.SpacesInSquareBrackets; 4237 } 4238 return (Style.SpaceAfterLogicalNot && Left.is(tok::exclaim)) || 4239 Right.is(TT_BinaryOperator); 4240 } 4241 4242 // If the next token is a binary operator or a selector name, we have 4243 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 4244 if (Left.is(TT_CastRParen)) { 4245 return Style.SpaceAfterCStyleCast || 4246 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 4247 } 4248 4249 auto ShouldAddSpacesInAngles = [this, &Right]() { 4250 if (this->Style.SpacesInAngles == FormatStyle::SIAS_Always) 4251 return true; 4252 if (this->Style.SpacesInAngles == FormatStyle::SIAS_Leave) 4253 return Right.hasWhitespaceBefore(); 4254 return false; 4255 }; 4256 4257 if (Left.is(tok::greater) && Right.is(tok::greater)) { 4258 if (Style.Language == FormatStyle::LK_TextProto || 4259 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) { 4260 return !Style.Cpp11BracedListStyle; 4261 } 4262 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 4263 ((Style.Standard < FormatStyle::LS_Cpp11) || 4264 ShouldAddSpacesInAngles()); 4265 } 4266 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 4267 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 4268 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) { 4269 return false; 4270 } 4271 if (!Style.SpaceBeforeAssignmentOperators && Left.isNot(TT_TemplateCloser) && 4272 Right.getPrecedence() == prec::Assignment) { 4273 return false; 4274 } 4275 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) && 4276 (Left.is(tok::identifier) || Left.is(tok::kw_this))) { 4277 return false; 4278 } 4279 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) { 4280 // Generally don't remove existing spaces between an identifier and "::". 4281 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 4282 // this turns out to be too lenient, add analysis of the identifier itself. 4283 return Right.hasWhitespaceBefore(); 4284 } 4285 if (Right.is(tok::coloncolon) && 4286 !Left.isOneOf(tok::l_brace, tok::comment, tok::l_paren)) { 4287 // Put a space between < and :: in vector< ::std::string > 4288 return (Left.is(TT_TemplateOpener) && 4289 ((Style.Standard < FormatStyle::LS_Cpp11) || 4290 ShouldAddSpacesInAngles())) || 4291 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 4292 tok::kw___super, TT_TemplateOpener, 4293 TT_TemplateCloser)) || 4294 (Left.is(tok::l_paren) && Style.SpacesInParentheses); 4295 } 4296 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 4297 return ShouldAddSpacesInAngles(); 4298 // Space before TT_StructuredBindingLSquare. 4299 if (Right.is(TT_StructuredBindingLSquare)) { 4300 return !Left.isOneOf(tok::amp, tok::ampamp) || 4301 getTokenReferenceAlignment(Left) != FormatStyle::PAS_Right; 4302 } 4303 // Space before & or && following a TT_StructuredBindingLSquare. 4304 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 4305 Right.isOneOf(tok::amp, tok::ampamp)) { 4306 return getTokenReferenceAlignment(Right) != FormatStyle::PAS_Left; 4307 } 4308 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 4309 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 4310 !Right.is(tok::r_paren))) { 4311 return true; 4312 } 4313 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 4314 Left.MatchingParen && 4315 Left.MatchingParen->is(TT_OverloadedOperatorLParen)) { 4316 return false; 4317 } 4318 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 4319 Line.Type == LT_ImportStatement) { 4320 return true; 4321 } 4322 if (Right.is(TT_TrailingUnaryOperator)) 4323 return false; 4324 if (Left.is(TT_RegexLiteral)) 4325 return false; 4326 return spaceRequiredBetween(Line, Left, Right); 4327 } 4328 4329 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 4330 static bool isAllmanBrace(const FormatToken &Tok) { 4331 return Tok.is(tok::l_brace) && Tok.is(BK_Block) && 4332 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_LambdaLBrace, TT_DictLiteral); 4333 } 4334 4335 // Returns 'true' if 'Tok' is a function argument. 4336 static bool IsFunctionArgument(const FormatToken &Tok) { 4337 return Tok.MatchingParen && Tok.MatchingParen->Next && 4338 Tok.MatchingParen->Next->isOneOf(tok::comma, tok::r_paren); 4339 } 4340 4341 static bool 4342 isItAnEmptyLambdaAllowed(const FormatToken &Tok, 4343 FormatStyle::ShortLambdaStyle ShortLambdaOption) { 4344 return Tok.Children.empty() && ShortLambdaOption != FormatStyle::SLS_None; 4345 } 4346 4347 static bool isAllmanLambdaBrace(const FormatToken &Tok) { 4348 return Tok.is(tok::l_brace) && Tok.is(BK_Block) && 4349 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral); 4350 } 4351 4352 // Returns the first token on the line that is not a comment. 4353 static const FormatToken *getFirstNonComment(const AnnotatedLine &Line) { 4354 const FormatToken *Next = Line.First; 4355 if (!Next) 4356 return Next; 4357 if (Next->is(tok::comment)) 4358 Next = Next->getNextNonComment(); 4359 return Next; 4360 } 4361 4362 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 4363 const FormatToken &Right) const { 4364 const FormatToken &Left = *Right.Previous; 4365 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 4366 return true; 4367 4368 if (Style.isCSharp()) { 4369 if (Left.is(TT_FatArrow) && Right.is(tok::l_brace) && 4370 Style.BraceWrapping.AfterFunction) { 4371 return true; 4372 } 4373 if (Right.is(TT_CSharpNamedArgumentColon) || 4374 Left.is(TT_CSharpNamedArgumentColon)) { 4375 return false; 4376 } 4377 if (Right.is(TT_CSharpGenericTypeConstraint)) 4378 return true; 4379 if (Right.Next && Right.Next->is(TT_FatArrow) && 4380 (Right.is(tok::numeric_constant) || 4381 (Right.is(tok::identifier) && Right.TokenText == "_"))) { 4382 return true; 4383 } 4384 4385 // Break after C# [...] and before public/protected/private/internal. 4386 if (Left.is(TT_AttributeSquare) && Left.is(tok::r_square) && 4387 (Right.isAccessSpecifier(/*ColonRequired=*/false) || 4388 Right.is(Keywords.kw_internal))) { 4389 return true; 4390 } 4391 // Break between ] and [ but only when there are really 2 attributes. 4392 if (Left.is(TT_AttributeSquare) && Right.is(TT_AttributeSquare) && 4393 Left.is(tok::r_square) && Right.is(tok::l_square)) { 4394 return true; 4395 } 4396 4397 } else if (Style.isJavaScript()) { 4398 // FIXME: This might apply to other languages and token kinds. 4399 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 4400 Left.Previous->is(tok::string_literal)) { 4401 return true; 4402 } 4403 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 4404 Left.Previous && Left.Previous->is(tok::equal) && 4405 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 4406 tok::kw_const) && 4407 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 4408 // above. 4409 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) { 4410 // Object literals on the top level of a file are treated as "enum-style". 4411 // Each key/value pair is put on a separate line, instead of bin-packing. 4412 return true; 4413 } 4414 if (Left.is(tok::l_brace) && Line.Level == 0 && 4415 (Line.startsWith(tok::kw_enum) || 4416 Line.startsWith(tok::kw_const, tok::kw_enum) || 4417 Line.startsWith(tok::kw_export, tok::kw_enum) || 4418 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) { 4419 // JavaScript top-level enum key/value pairs are put on separate lines 4420 // instead of bin-packing. 4421 return true; 4422 } 4423 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && Left.Previous && 4424 Left.Previous->is(TT_FatArrow)) { 4425 // JS arrow function (=> {...}). 4426 switch (Style.AllowShortLambdasOnASingleLine) { 4427 case FormatStyle::SLS_All: 4428 return false; 4429 case FormatStyle::SLS_None: 4430 return true; 4431 case FormatStyle::SLS_Empty: 4432 return !Left.Children.empty(); 4433 case FormatStyle::SLS_Inline: 4434 // allow one-lining inline (e.g. in function call args) and empty arrow 4435 // functions. 4436 return (Left.NestingLevel == 0 && Line.Level == 0) && 4437 !Left.Children.empty(); 4438 } 4439 llvm_unreachable("Unknown FormatStyle::ShortLambdaStyle enum"); 4440 } 4441 4442 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 4443 !Left.Children.empty()) { 4444 // Support AllowShortFunctionsOnASingleLine for JavaScript. 4445 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 4446 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 4447 (Left.NestingLevel == 0 && Line.Level == 0 && 4448 Style.AllowShortFunctionsOnASingleLine & 4449 FormatStyle::SFS_InlineOnly); 4450 } 4451 } else if (Style.Language == FormatStyle::LK_Java) { 4452 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 4453 Right.Next->is(tok::string_literal)) { 4454 return true; 4455 } 4456 } else if (Style.isVerilog()) { 4457 // Break after labels. In Verilog labels don't have the 'case' keyword, so 4458 // it is hard to identify them in UnwrappedLineParser. 4459 if (!Keywords.isVerilogBegin(Right) && Keywords.isVerilogEndOfLabel(Left)) 4460 return true; 4461 } else if (Style.Language == FormatStyle::LK_Cpp || 4462 Style.Language == FormatStyle::LK_ObjC || 4463 Style.Language == FormatStyle::LK_Proto || 4464 Style.Language == FormatStyle::LK_TableGen || 4465 Style.Language == FormatStyle::LK_TextProto) { 4466 if (Left.isStringLiteral() && Right.isStringLiteral()) 4467 return true; 4468 } 4469 4470 // Basic JSON newline processing. 4471 if (Style.isJson()) { 4472 // Always break after a JSON record opener. 4473 // { 4474 // } 4475 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace)) 4476 return true; 4477 // Always break after a JSON array opener based on BreakArrays. 4478 if ((Left.is(TT_ArrayInitializerLSquare) && Left.is(tok::l_square) && 4479 Right.isNot(tok::r_square)) || 4480 Left.is(tok::comma)) { 4481 if (Right.is(tok::l_brace)) 4482 return true; 4483 // scan to the right if an we see an object or an array inside 4484 // then break. 4485 for (const auto *Tok = &Right; Tok; Tok = Tok->Next) { 4486 if (Tok->isOneOf(tok::l_brace, tok::l_square)) 4487 return true; 4488 if (Tok->isOneOf(tok::r_brace, tok::r_square)) 4489 break; 4490 } 4491 return Style.BreakArrays; 4492 } 4493 } 4494 4495 if (Line.startsWith(tok::kw_asm) && Right.is(TT_InlineASMColon) && 4496 Style.BreakBeforeInlineASMColon == FormatStyle::BBIAS_Always) { 4497 return true; 4498 } 4499 4500 // If the last token before a '}', ']', or ')' is a comma or a trailing 4501 // comment, the intention is to insert a line break after it in order to make 4502 // shuffling around entries easier. Import statements, especially in 4503 // JavaScript, can be an exception to this rule. 4504 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 4505 const FormatToken *BeforeClosingBrace = nullptr; 4506 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 4507 (Style.isJavaScript() && Left.is(tok::l_paren))) && 4508 Left.isNot(BK_Block) && Left.MatchingParen) { 4509 BeforeClosingBrace = Left.MatchingParen->Previous; 4510 } else if (Right.MatchingParen && 4511 (Right.MatchingParen->isOneOf(tok::l_brace, 4512 TT_ArrayInitializerLSquare) || 4513 (Style.isJavaScript() && 4514 Right.MatchingParen->is(tok::l_paren)))) { 4515 BeforeClosingBrace = &Left; 4516 } 4517 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 4518 BeforeClosingBrace->isTrailingComment())) { 4519 return true; 4520 } 4521 } 4522 4523 if (Right.is(tok::comment)) { 4524 return Left.isNot(BK_BracedInit) && Left.isNot(TT_CtorInitializerColon) && 4525 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 4526 } 4527 if (Left.isTrailingComment()) 4528 return true; 4529 if (Left.IsUnterminatedLiteral) 4530 return true; 4531 if (Right.is(tok::lessless) && Right.Next && Left.is(tok::string_literal) && 4532 Right.Next->is(tok::string_literal)) { 4533 return true; 4534 } 4535 if (Right.is(TT_RequiresClause)) { 4536 switch (Style.RequiresClausePosition) { 4537 case FormatStyle::RCPS_OwnLine: 4538 case FormatStyle::RCPS_WithFollowing: 4539 return true; 4540 default: 4541 break; 4542 } 4543 } 4544 // Can break after template<> declaration 4545 if (Left.ClosesTemplateDeclaration && Left.MatchingParen && 4546 Left.MatchingParen->NestingLevel == 0) { 4547 // Put concepts on the next line e.g. 4548 // template<typename T> 4549 // concept ... 4550 if (Right.is(tok::kw_concept)) 4551 return Style.BreakBeforeConceptDeclarations == FormatStyle::BBCDS_Always; 4552 return Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes; 4553 } 4554 if (Left.ClosesRequiresClause && Right.isNot(tok::semi)) { 4555 switch (Style.RequiresClausePosition) { 4556 case FormatStyle::RCPS_OwnLine: 4557 case FormatStyle::RCPS_WithPreceding: 4558 return true; 4559 default: 4560 break; 4561 } 4562 } 4563 if (Style.PackConstructorInitializers == FormatStyle::PCIS_Never) { 4564 if (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeColon && 4565 (Left.is(TT_CtorInitializerComma) || 4566 Right.is(TT_CtorInitializerColon))) { 4567 return true; 4568 } 4569 4570 if (Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon && 4571 Left.isOneOf(TT_CtorInitializerColon, TT_CtorInitializerComma)) { 4572 return true; 4573 } 4574 } 4575 if (Style.PackConstructorInitializers < FormatStyle::PCIS_CurrentLine && 4576 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 4577 Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) { 4578 return true; 4579 } 4580 // Break only if we have multiple inheritance. 4581 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 4582 Right.is(TT_InheritanceComma)) { 4583 return true; 4584 } 4585 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterComma && 4586 Left.is(TT_InheritanceComma)) { 4587 return true; 4588 } 4589 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) { 4590 // Multiline raw string literals are special wrt. line breaks. The author 4591 // has made a deliberate choice and might have aligned the contents of the 4592 // string literal accordingly. Thus, we try keep existing line breaks. 4593 return Right.IsMultiline && Right.NewlinesBefore > 0; 4594 } 4595 if ((Left.is(tok::l_brace) || (Left.is(tok::less) && Left.Previous && 4596 Left.Previous->is(tok::equal))) && 4597 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 4598 // Don't put enums or option definitions onto single lines in protocol 4599 // buffers. 4600 return true; 4601 } 4602 if (Right.is(TT_InlineASMBrace)) 4603 return Right.HasUnescapedNewline; 4604 4605 if (isAllmanBrace(Left) || isAllmanBrace(Right)) { 4606 auto FirstNonComment = getFirstNonComment(Line); 4607 bool AccessSpecifier = 4608 FirstNonComment && 4609 FirstNonComment->isOneOf(Keywords.kw_internal, tok::kw_public, 4610 tok::kw_private, tok::kw_protected); 4611 4612 if (Style.BraceWrapping.AfterEnum) { 4613 if (Line.startsWith(tok::kw_enum) || 4614 Line.startsWith(tok::kw_typedef, tok::kw_enum)) { 4615 return true; 4616 } 4617 // Ensure BraceWrapping for `public enum A {`. 4618 if (AccessSpecifier && FirstNonComment->Next && 4619 FirstNonComment->Next->is(tok::kw_enum)) { 4620 return true; 4621 } 4622 } 4623 4624 // Ensure BraceWrapping for `public interface A {`. 4625 if (Style.BraceWrapping.AfterClass && 4626 ((AccessSpecifier && FirstNonComment->Next && 4627 FirstNonComment->Next->is(Keywords.kw_interface)) || 4628 Line.startsWith(Keywords.kw_interface))) { 4629 return true; 4630 } 4631 4632 return (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 4633 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 4634 } 4635 4636 if (Left.is(TT_ObjCBlockLBrace) && 4637 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) { 4638 return true; 4639 } 4640 4641 // Ensure wrapping after __attribute__((XX)) and @interface etc. 4642 if (Left.is(TT_AttributeParen) && Right.is(TT_ObjCDecl)) 4643 return true; 4644 4645 if (Left.is(TT_LambdaLBrace)) { 4646 if (IsFunctionArgument(Left) && 4647 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline) { 4648 return false; 4649 } 4650 4651 if (Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_None || 4652 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline || 4653 (!Left.Children.empty() && 4654 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Empty)) { 4655 return true; 4656 } 4657 } 4658 4659 if (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace) && 4660 Left.isOneOf(tok::star, tok::amp, tok::ampamp, TT_TemplateCloser)) { 4661 return true; 4662 } 4663 4664 // Put multiple Java annotation on a new line. 4665 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 4666 Left.is(TT_LeadingJavaAnnotation) && 4667 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 4668 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) { 4669 return true; 4670 } 4671 4672 if (Right.is(TT_ProtoExtensionLSquare)) 4673 return true; 4674 4675 // In text proto instances if a submessage contains at least 2 entries and at 4676 // least one of them is a submessage, like A { ... B { ... } ... }, 4677 // put all of the entries of A on separate lines by forcing the selector of 4678 // the submessage B to be put on a newline. 4679 // 4680 // Example: these can stay on one line: 4681 // a { scalar_1: 1 scalar_2: 2 } 4682 // a { b { key: value } } 4683 // 4684 // and these entries need to be on a new line even if putting them all in one 4685 // line is under the column limit: 4686 // a { 4687 // scalar: 1 4688 // b { key: value } 4689 // } 4690 // 4691 // We enforce this by breaking before a submessage field that has previous 4692 // siblings, *and* breaking before a field that follows a submessage field. 4693 // 4694 // Be careful to exclude the case [proto.ext] { ... } since the `]` is 4695 // the TT_SelectorName there, but we don't want to break inside the brackets. 4696 // 4697 // Another edge case is @submessage { key: value }, which is a common 4698 // substitution placeholder. In this case we want to keep `@` and `submessage` 4699 // together. 4700 // 4701 // We ensure elsewhere that extensions are always on their own line. 4702 if ((Style.Language == FormatStyle::LK_Proto || 4703 Style.Language == FormatStyle::LK_TextProto) && 4704 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) { 4705 // Keep `@submessage` together in: 4706 // @submessage { key: value } 4707 if (Left.is(tok::at)) 4708 return false; 4709 // Look for the scope opener after selector in cases like: 4710 // selector { ... 4711 // selector: { ... 4712 // selector: @base { ... 4713 FormatToken *LBrace = Right.Next; 4714 if (LBrace && LBrace->is(tok::colon)) { 4715 LBrace = LBrace->Next; 4716 if (LBrace && LBrace->is(tok::at)) { 4717 LBrace = LBrace->Next; 4718 if (LBrace) 4719 LBrace = LBrace->Next; 4720 } 4721 } 4722 if (LBrace && 4723 // The scope opener is one of {, [, <: 4724 // selector { ... } 4725 // selector [ ... ] 4726 // selector < ... > 4727 // 4728 // In case of selector { ... }, the l_brace is TT_DictLiteral. 4729 // In case of an empty selector {}, the l_brace is not TT_DictLiteral, 4730 // so we check for immediately following r_brace. 4731 ((LBrace->is(tok::l_brace) && 4732 (LBrace->is(TT_DictLiteral) || 4733 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) || 4734 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) { 4735 // If Left.ParameterCount is 0, then this submessage entry is not the 4736 // first in its parent submessage, and we want to break before this entry. 4737 // If Left.ParameterCount is greater than 0, then its parent submessage 4738 // might contain 1 or more entries and we want to break before this entry 4739 // if it contains at least 2 entries. We deal with this case later by 4740 // detecting and breaking before the next entry in the parent submessage. 4741 if (Left.ParameterCount == 0) 4742 return true; 4743 // However, if this submessage is the first entry in its parent 4744 // submessage, Left.ParameterCount might be 1 in some cases. 4745 // We deal with this case later by detecting an entry 4746 // following a closing paren of this submessage. 4747 } 4748 4749 // If this is an entry immediately following a submessage, it will be 4750 // preceded by a closing paren of that submessage, like in: 4751 // left---. .---right 4752 // v v 4753 // sub: { ... } key: value 4754 // If there was a comment between `}` an `key` above, then `key` would be 4755 // put on a new line anyways. 4756 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square)) 4757 return true; 4758 } 4759 4760 // Deal with lambda arguments in C++ - we want consistent line breaks whether 4761 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced 4762 // as aggressive line breaks are placed when the lambda is not the last arg. 4763 if ((Style.Language == FormatStyle::LK_Cpp || 4764 Style.Language == FormatStyle::LK_ObjC) && 4765 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 && 4766 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) { 4767 // Multiple lambdas in the same function call force line breaks. 4768 if (Left.BlockParameterCount > 1) 4769 return true; 4770 4771 // A lambda followed by another arg forces a line break. 4772 if (!Left.Role) 4773 return false; 4774 auto Comma = Left.Role->lastComma(); 4775 if (!Comma) 4776 return false; 4777 auto Next = Comma->getNextNonComment(); 4778 if (!Next) 4779 return false; 4780 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret)) 4781 return true; 4782 } 4783 4784 return false; 4785 } 4786 4787 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 4788 const FormatToken &Right) const { 4789 const FormatToken &Left = *Right.Previous; 4790 // Language-specific stuff. 4791 if (Style.isCSharp()) { 4792 if (Left.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon) || 4793 Right.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon)) { 4794 return false; 4795 } 4796 // Only break after commas for generic type constraints. 4797 if (Line.First->is(TT_CSharpGenericTypeConstraint)) 4798 return Left.is(TT_CSharpGenericTypeConstraintComma); 4799 // Keep nullable operators attached to their identifiers. 4800 if (Right.is(TT_CSharpNullable)) 4801 return false; 4802 } else if (Style.Language == FormatStyle::LK_Java) { 4803 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 4804 Keywords.kw_implements)) { 4805 return false; 4806 } 4807 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 4808 Keywords.kw_implements)) { 4809 return true; 4810 } 4811 } else if (Style.isJavaScript()) { 4812 const FormatToken *NonComment = Right.getPreviousNonComment(); 4813 if (NonComment && 4814 NonComment->isOneOf( 4815 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 4816 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 4817 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 4818 Keywords.kw_readonly, Keywords.kw_override, Keywords.kw_abstract, 4819 Keywords.kw_get, Keywords.kw_set, Keywords.kw_async, 4820 Keywords.kw_await)) { 4821 return false; // Otherwise automatic semicolon insertion would trigger. 4822 } 4823 if (Right.NestingLevel == 0 && 4824 (Left.Tok.getIdentifierInfo() || 4825 Left.isOneOf(tok::r_square, tok::r_paren)) && 4826 Right.isOneOf(tok::l_square, tok::l_paren)) { 4827 return false; // Otherwise automatic semicolon insertion would trigger. 4828 } 4829 if (NonComment && NonComment->is(tok::identifier) && 4830 NonComment->TokenText == "asserts") { 4831 return false; 4832 } 4833 if (Left.is(TT_FatArrow) && Right.is(tok::l_brace)) 4834 return false; 4835 if (Left.is(TT_JsTypeColon)) 4836 return true; 4837 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;"). 4838 if (Left.is(tok::exclaim) && Right.is(tok::colon)) 4839 return false; 4840 // Look for is type annotations like: 4841 // function f(): a is B { ... } 4842 // Do not break before is in these cases. 4843 if (Right.is(Keywords.kw_is)) { 4844 const FormatToken *Next = Right.getNextNonComment(); 4845 // If `is` is followed by a colon, it's likely that it's a dict key, so 4846 // ignore it for this check. 4847 // For example this is common in Polymer: 4848 // Polymer({ 4849 // is: 'name', 4850 // ... 4851 // }); 4852 if (!Next || !Next->is(tok::colon)) 4853 return false; 4854 } 4855 if (Left.is(Keywords.kw_in)) 4856 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 4857 if (Right.is(Keywords.kw_in)) 4858 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 4859 if (Right.is(Keywords.kw_as)) 4860 return false; // must not break before as in 'x as type' casts 4861 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) { 4862 // extends and infer can appear as keywords in conditional types: 4863 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types 4864 // do not break before them, as the expressions are subject to ASI. 4865 return false; 4866 } 4867 if (Left.is(Keywords.kw_as)) 4868 return true; 4869 if (Left.is(TT_NonNullAssertion)) 4870 return true; 4871 if (Left.is(Keywords.kw_declare) && 4872 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 4873 Keywords.kw_function, tok::kw_class, tok::kw_enum, 4874 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 4875 Keywords.kw_let, tok::kw_const)) { 4876 // See grammar for 'declare' statements at: 4877 // https://github.com/Microsoft/TypeScript/blob/main/doc/spec-ARCHIVED.md#A.10 4878 return false; 4879 } 4880 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 4881 Right.isOneOf(tok::identifier, tok::string_literal)) { 4882 return false; // must not break in "module foo { ...}" 4883 } 4884 if (Right.is(TT_TemplateString) && Right.closesScope()) 4885 return false; 4886 // Don't split tagged template literal so there is a break between the tag 4887 // identifier and template string. 4888 if (Left.is(tok::identifier) && Right.is(TT_TemplateString)) 4889 return false; 4890 if (Left.is(TT_TemplateString) && Left.opensScope()) 4891 return true; 4892 } 4893 4894 if (Left.is(tok::at)) 4895 return false; 4896 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 4897 return false; 4898 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 4899 return !Right.is(tok::l_paren); 4900 if (Right.is(TT_PointerOrReference)) { 4901 return Line.IsMultiVariableDeclStmt || 4902 (getTokenPointerOrReferenceAlignment(Right) == 4903 FormatStyle::PAS_Right && 4904 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 4905 } 4906 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 4907 Right.is(tok::kw_operator)) { 4908 return true; 4909 } 4910 if (Left.is(TT_PointerOrReference)) 4911 return false; 4912 if (Right.isTrailingComment()) { 4913 // We rely on MustBreakBefore being set correctly here as we should not 4914 // change the "binding" behavior of a comment. 4915 // The first comment in a braced lists is always interpreted as belonging to 4916 // the first list element. Otherwise, it should be placed outside of the 4917 // list. 4918 return Left.is(BK_BracedInit) || 4919 (Left.is(TT_CtorInitializerColon) && Right.NewlinesBefore > 0 && 4920 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 4921 } 4922 if (Left.is(tok::question) && Right.is(tok::colon)) 4923 return false; 4924 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 4925 return Style.BreakBeforeTernaryOperators; 4926 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 4927 return !Style.BreakBeforeTernaryOperators; 4928 if (Left.is(TT_InheritanceColon)) 4929 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon; 4930 if (Right.is(TT_InheritanceColon)) 4931 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon; 4932 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 4933 Left.isNot(TT_SelectorName)) { 4934 return true; 4935 } 4936 4937 if (Right.is(tok::colon) && 4938 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) { 4939 return false; 4940 } 4941 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 4942 if (Style.Language == FormatStyle::LK_Proto || 4943 Style.Language == FormatStyle::LK_TextProto) { 4944 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 4945 return false; 4946 // Prevent cases like: 4947 // 4948 // submessage: 4949 // { key: valueeeeeeeeeeee } 4950 // 4951 // when the snippet does not fit into one line. 4952 // Prefer: 4953 // 4954 // submessage: { 4955 // key: valueeeeeeeeeeee 4956 // } 4957 // 4958 // instead, even if it is longer by one line. 4959 // 4960 // Note that this allows the "{" to go over the column limit 4961 // when the column limit is just between ":" and "{", but that does 4962 // not happen too often and alternative formattings in this case are 4963 // not much better. 4964 // 4965 // The code covers the cases: 4966 // 4967 // submessage: { ... } 4968 // submessage: < ... > 4969 // repeated: [ ... ] 4970 if (((Right.is(tok::l_brace) || Right.is(tok::less)) && 4971 Right.is(TT_DictLiteral)) || 4972 Right.is(TT_ArrayInitializerLSquare)) { 4973 return false; 4974 } 4975 } 4976 return true; 4977 } 4978 if (Right.is(tok::r_square) && Right.MatchingParen && 4979 Right.MatchingParen->is(TT_ProtoExtensionLSquare)) { 4980 return false; 4981 } 4982 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 4983 Right.Next->is(TT_ObjCMethodExpr))) { 4984 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 4985 } 4986 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 4987 return true; 4988 if (Right.is(tok::kw_concept)) 4989 return Style.BreakBeforeConceptDeclarations != FormatStyle::BBCDS_Never; 4990 if (Right.is(TT_RequiresClause)) 4991 return true; 4992 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 4993 return true; 4994 if (Left.ClosesRequiresClause) 4995 return true; 4996 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 4997 TT_OverloadedOperator)) { 4998 return false; 4999 } 5000 if (Left.is(TT_RangeBasedForLoopColon)) 5001 return true; 5002 if (Right.is(TT_RangeBasedForLoopColon)) 5003 return false; 5004 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 5005 return true; 5006 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 5007 (Left.is(tok::less) && Right.is(tok::less))) { 5008 return false; 5009 } 5010 if (Right.is(TT_BinaryOperator) && 5011 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 5012 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 5013 Right.getPrecedence() != prec::Assignment)) { 5014 return true; 5015 } 5016 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 5017 Left.is(tok::kw_operator)) { 5018 return false; 5019 } 5020 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 5021 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) { 5022 return false; 5023 } 5024 if (Left.is(tok::equal) && Right.is(tok::l_brace) && 5025 !Style.Cpp11BracedListStyle) { 5026 return false; 5027 } 5028 if (Left.is(tok::l_paren) && 5029 Left.isOneOf(TT_AttributeParen, TT_TypeDeclarationParen)) { 5030 return false; 5031 } 5032 if (Left.is(tok::l_paren) && Left.Previous && 5033 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) { 5034 return false; 5035 } 5036 if (Right.is(TT_ImplicitStringLiteral)) 5037 return false; 5038 5039 if (Right.is(TT_TemplateCloser)) 5040 return false; 5041 if (Right.is(tok::r_square) && Right.MatchingParen && 5042 Right.MatchingParen->is(TT_LambdaLSquare)) { 5043 return false; 5044 } 5045 5046 // We only break before r_brace if there was a corresponding break before 5047 // the l_brace, which is tracked by BreakBeforeClosingBrace. 5048 if (Right.is(tok::r_brace)) 5049 return Right.MatchingParen && Right.MatchingParen->is(BK_Block); 5050 5051 // We only break before r_paren if we're in a block indented context. 5052 if (Right.is(tok::r_paren)) { 5053 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_BlockIndent || 5054 !Right.MatchingParen) { 5055 return false; 5056 } 5057 auto Next = Right.Next; 5058 if (Next && Next->is(tok::r_paren)) 5059 Next = Next->Next; 5060 if (Next && Next->is(tok::l_paren)) 5061 return false; 5062 const FormatToken *Previous = Right.MatchingParen->Previous; 5063 return !(Previous && (Previous->is(tok::kw_for) || Previous->isIf())); 5064 } 5065 5066 // Allow breaking after a trailing annotation, e.g. after a method 5067 // declaration. 5068 if (Left.is(TT_TrailingAnnotation)) { 5069 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 5070 tok::less, tok::coloncolon); 5071 } 5072 5073 if (Right.is(tok::kw___attribute) || 5074 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) { 5075 return !Left.is(TT_AttributeSquare); 5076 } 5077 5078 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 5079 return true; 5080 5081 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 5082 return true; 5083 5084 if (Left.is(TT_CtorInitializerColon)) { 5085 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon && 5086 (!Right.isTrailingComment() || Right.NewlinesBefore > 0); 5087 } 5088 if (Right.is(TT_CtorInitializerColon)) 5089 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 5090 if (Left.is(TT_CtorInitializerComma) && 5091 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) { 5092 return false; 5093 } 5094 if (Right.is(TT_CtorInitializerComma) && 5095 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) { 5096 return true; 5097 } 5098 if (Left.is(TT_InheritanceComma) && 5099 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) { 5100 return false; 5101 } 5102 if (Right.is(TT_InheritanceComma) && 5103 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) { 5104 return true; 5105 } 5106 if (Left.is(TT_ArrayInitializerLSquare)) 5107 return true; 5108 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 5109 return true; 5110 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 5111 !Left.isOneOf(tok::arrowstar, tok::lessless) && 5112 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 5113 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 5114 Left.getPrecedence() == prec::Assignment)) { 5115 return true; 5116 } 5117 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 5118 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) { 5119 return false; 5120 } 5121 5122 auto ShortLambdaOption = Style.AllowShortLambdasOnASingleLine; 5123 if (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace)) { 5124 if (isAllmanLambdaBrace(Left)) 5125 return !isItAnEmptyLambdaAllowed(Left, ShortLambdaOption); 5126 if (isAllmanLambdaBrace(Right)) 5127 return !isItAnEmptyLambdaAllowed(Right, ShortLambdaOption); 5128 } 5129 5130 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 5131 tok::kw_class, tok::kw_struct, tok::comment) || 5132 Right.isMemberAccess() || 5133 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 5134 tok::colon, tok::l_square, tok::at) || 5135 (Left.is(tok::r_paren) && 5136 Right.isOneOf(tok::identifier, tok::kw_const)) || 5137 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 5138 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 5139 } 5140 5141 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) const { 5142 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << ", P=" << Line.PPLevel 5143 << ", T=" << Line.Type << ", C=" << Line.IsContinuation 5144 << "):\n"; 5145 const FormatToken *Tok = Line.First; 5146 while (Tok) { 5147 llvm::errs() << " M=" << Tok->MustBreakBefore 5148 << " C=" << Tok->CanBreakBefore 5149 << " T=" << getTokenTypeName(Tok->getType()) 5150 << " S=" << Tok->SpacesRequiredBefore 5151 << " F=" << Tok->Finalized << " B=" << Tok->BlockParameterCount 5152 << " BK=" << Tok->getBlockKind() << " P=" << Tok->SplitPenalty 5153 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 5154 << " PPK=" << Tok->getPackingKind() << " FakeLParens="; 5155 for (prec::Level LParen : Tok->FakeLParens) 5156 llvm::errs() << LParen << "/"; 5157 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 5158 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo(); 5159 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 5160 if (!Tok->Next) 5161 assert(Tok == Line.Last); 5162 Tok = Tok->Next; 5163 } 5164 llvm::errs() << "----\n"; 5165 } 5166 5167 FormatStyle::PointerAlignmentStyle 5168 TokenAnnotator::getTokenReferenceAlignment(const FormatToken &Reference) const { 5169 assert(Reference.isOneOf(tok::amp, tok::ampamp)); 5170 switch (Style.ReferenceAlignment) { 5171 case FormatStyle::RAS_Pointer: 5172 return Style.PointerAlignment; 5173 case FormatStyle::RAS_Left: 5174 return FormatStyle::PAS_Left; 5175 case FormatStyle::RAS_Right: 5176 return FormatStyle::PAS_Right; 5177 case FormatStyle::RAS_Middle: 5178 return FormatStyle::PAS_Middle; 5179 } 5180 assert(0); //"Unhandled value of ReferenceAlignment" 5181 return Style.PointerAlignment; 5182 } 5183 5184 FormatStyle::PointerAlignmentStyle 5185 TokenAnnotator::getTokenPointerOrReferenceAlignment( 5186 const FormatToken &PointerOrReference) const { 5187 if (PointerOrReference.isOneOf(tok::amp, tok::ampamp)) { 5188 switch (Style.ReferenceAlignment) { 5189 case FormatStyle::RAS_Pointer: 5190 return Style.PointerAlignment; 5191 case FormatStyle::RAS_Left: 5192 return FormatStyle::PAS_Left; 5193 case FormatStyle::RAS_Right: 5194 return FormatStyle::PAS_Right; 5195 case FormatStyle::RAS_Middle: 5196 return FormatStyle::PAS_Middle; 5197 } 5198 } 5199 assert(PointerOrReference.is(tok::star)); 5200 return Style.PointerAlignment; 5201 } 5202 5203 } // namespace format 5204 } // namespace clang 5205