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