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