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