1 //===--- TokenAnnotator.cpp - Format C++ code -----------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// 9 /// \file 10 /// This file implements a token annotator, i.e. creates 11 /// \c AnnotatedTokens out of \c FormatTokens with required extra information. 12 /// 13 //===----------------------------------------------------------------------===// 14 15 #include "TokenAnnotator.h" 16 #include "FormatToken.h" 17 #include "clang/Basic/SourceManager.h" 18 #include "clang/Basic/TokenKinds.h" 19 #include "llvm/ADT/SmallPtrSet.h" 20 #include "llvm/Support/Debug.h" 21 22 #define DEBUG_TYPE "format-token-annotator" 23 24 namespace clang { 25 namespace format { 26 27 namespace { 28 29 /// Returns \c true if the line starts with a token that can start a statement 30 /// with an initializer. 31 static bool startsWithInitStatement(const AnnotatedLine &Line) { 32 return Line.startsWith(tok::kw_for) || Line.startsWith(tok::kw_if) || 33 Line.startsWith(tok::kw_switch); 34 } 35 36 /// Returns \c true if the token can be used as an identifier in 37 /// an Objective-C \c \@selector, \c false otherwise. 38 /// 39 /// Because getFormattingLangOpts() always lexes source code as 40 /// Objective-C++, C++ keywords like \c new and \c delete are 41 /// lexed as tok::kw_*, not tok::identifier, even for Objective-C. 42 /// 43 /// For Objective-C and Objective-C++, both identifiers and keywords 44 /// are valid inside @selector(...) (or a macro which 45 /// invokes @selector(...)). So, we allow treat any identifier or 46 /// keyword as a potential Objective-C selector component. 47 static bool canBeObjCSelectorComponent(const FormatToken &Tok) { 48 return Tok.Tok.getIdentifierInfo() != nullptr; 49 } 50 51 /// With `Left` being '(', check if we're at either `[...](` or 52 /// `[...]<...>(`, where the [ opens a lambda capture list. 53 static bool isLambdaParameterList(const FormatToken *Left) { 54 // Skip <...> if present. 55 if (Left->Previous && Left->Previous->is(tok::greater) && 56 Left->Previous->MatchingParen && 57 Left->Previous->MatchingParen->is(TT_TemplateOpener)) 58 Left = Left->Previous->MatchingParen; 59 60 // Check for `[...]`. 61 return Left->Previous && Left->Previous->is(tok::r_square) && 62 Left->Previous->MatchingParen && 63 Left->Previous->MatchingParen->is(TT_LambdaLSquare); 64 } 65 66 /// Returns \c true if the token is followed by a boolean condition, \c false 67 /// otherwise. 68 static bool isKeywordWithCondition(const FormatToken &Tok) { 69 return Tok.isOneOf(tok::kw_if, tok::kw_for, tok::kw_while, tok::kw_switch, 70 tok::kw_constexpr, tok::kw_catch); 71 } 72 73 /// A parser that gathers additional information about tokens. 74 /// 75 /// The \c TokenAnnotator tries to match parenthesis and square brakets and 76 /// store a parenthesis levels. It also tries to resolve matching "<" and ">" 77 /// into template parameter lists. 78 class AnnotatingParser { 79 public: 80 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line, 81 const AdditionalKeywords &Keywords) 82 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false), 83 Keywords(Keywords) { 84 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false)); 85 resetTokenMetadata(); 86 } 87 88 private: 89 bool parseAngle() { 90 if (!CurrentToken || !CurrentToken->Previous) 91 return false; 92 if (NonTemplateLess.count(CurrentToken->Previous)) 93 return false; 94 95 const FormatToken &Previous = *CurrentToken->Previous; // The '<'. 96 if (Previous.Previous) { 97 if (Previous.Previous->Tok.isLiteral()) 98 return false; 99 if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 && 100 (!Previous.Previous->MatchingParen || 101 !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen))) 102 return false; 103 } 104 105 FormatToken *Left = CurrentToken->Previous; 106 Left->ParentBracket = Contexts.back().ContextKind; 107 ScopedContextCreator ContextCreator(*this, tok::less, 12); 108 109 // If this angle is in the context of an expression, we need to be more 110 // hesitant to detect it as opening template parameters. 111 bool InExprContext = Contexts.back().IsExpression; 112 113 Contexts.back().IsExpression = false; 114 // If there's a template keyword before the opening angle bracket, this is a 115 // template parameter, not an argument. 116 if (Left->Previous && Left->Previous->isNot(tok::kw_template)) 117 Contexts.back().ContextType = Context::TemplateArgument; 118 119 if (Style.Language == FormatStyle::LK_Java && 120 CurrentToken->is(tok::question)) 121 next(); 122 123 while (CurrentToken) { 124 if (CurrentToken->is(tok::greater)) { 125 // Try to do a better job at looking for ">>" within the condition of 126 // a statement. Conservatively insert spaces between consecutive ">" 127 // tokens to prevent splitting right bitshift operators and potentially 128 // altering program semantics. This check is overly conservative and 129 // will prevent spaces from being inserted in select nested template 130 // parameter cases, but should not alter program semantics. 131 if (CurrentToken->Next && CurrentToken->Next->is(tok::greater) && 132 Left->ParentBracket != tok::less && 133 (isKeywordWithCondition(*Line.First) || 134 CurrentToken->getStartOfNonWhitespace() == 135 CurrentToken->Next->getStartOfNonWhitespace().getLocWithOffset( 136 -1))) 137 return false; 138 Left->MatchingParen = CurrentToken; 139 CurrentToken->MatchingParen = Left; 140 // In TT_Proto, we must distignuish between: 141 // map<key, value> 142 // msg < item: data > 143 // msg: < item: data > 144 // In TT_TextProto, map<key, value> does not occur. 145 if (Style.Language == FormatStyle::LK_TextProto || 146 (Style.Language == FormatStyle::LK_Proto && Left->Previous && 147 Left->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) 148 CurrentToken->setType(TT_DictLiteral); 149 else 150 CurrentToken->setType(TT_TemplateCloser); 151 next(); 152 return true; 153 } 154 if (CurrentToken->is(tok::question) && 155 Style.Language == FormatStyle::LK_Java) { 156 next(); 157 continue; 158 } 159 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) || 160 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext && 161 !Style.isCSharp() && Style.Language != FormatStyle::LK_Proto && 162 Style.Language != FormatStyle::LK_TextProto)) 163 return false; 164 // If a && or || is found and interpreted as a binary operator, this set 165 // of angles is likely part of something like "a < b && c > d". If the 166 // angles are inside an expression, the ||/&& might also be a binary 167 // operator that was misinterpreted because we are parsing template 168 // parameters. 169 // FIXME: This is getting out of hand, write a decent parser. 170 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) && 171 CurrentToken->Previous->is(TT_BinaryOperator) && 172 Contexts[Contexts.size() - 2].IsExpression && 173 !Line.startsWith(tok::kw_template)) 174 return false; 175 updateParameterCount(Left, CurrentToken); 176 if (Style.Language == FormatStyle::LK_Proto) { 177 if (FormatToken *Previous = CurrentToken->getPreviousNonComment()) { 178 if (CurrentToken->is(tok::colon) || 179 (CurrentToken->isOneOf(tok::l_brace, tok::less) && 180 Previous->isNot(tok::colon))) 181 Previous->setType(TT_SelectorName); 182 } 183 } 184 if (!consumeToken()) 185 return false; 186 } 187 return false; 188 } 189 190 bool parseUntouchableParens() { 191 while (CurrentToken) { 192 CurrentToken->Finalized = true; 193 switch (CurrentToken->Tok.getKind()) { 194 case tok::l_paren: 195 next(); 196 if (!parseUntouchableParens()) 197 return false; 198 continue; 199 case tok::r_paren: 200 next(); 201 return true; 202 default: 203 // no-op 204 break; 205 } 206 next(); 207 } 208 return false; 209 } 210 211 bool parseParens(bool LookForDecls = false) { 212 if (!CurrentToken) 213 return false; 214 assert(CurrentToken->Previous && "Unknown previous token"); 215 FormatToken &OpeningParen = *CurrentToken->Previous; 216 assert(OpeningParen.is(tok::l_paren)); 217 FormatToken *PrevNonComment = OpeningParen.getPreviousNonComment(); 218 OpeningParen.ParentBracket = Contexts.back().ContextKind; 219 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1); 220 221 // FIXME: This is a bit of a hack. Do better. 222 Contexts.back().ColonIsForRangeExpr = 223 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr; 224 225 if (OpeningParen.Previous && 226 OpeningParen.Previous->is(TT_UntouchableMacroFunc)) { 227 OpeningParen.Finalized = true; 228 return parseUntouchableParens(); 229 } 230 231 bool StartsObjCMethodExpr = false; 232 if (FormatToken *MaybeSel = OpeningParen.Previous) { 233 // @selector( starts a selector. 234 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous && 235 MaybeSel->Previous->is(tok::at)) 236 StartsObjCMethodExpr = true; 237 } 238 239 if (OpeningParen.is(TT_OverloadedOperatorLParen)) { 240 // Find the previous kw_operator token. 241 FormatToken *Prev = &OpeningParen; 242 while (!Prev->is(tok::kw_operator)) { 243 Prev = Prev->Previous; 244 assert(Prev && "Expect a kw_operator prior to the OperatorLParen!"); 245 } 246 247 // If faced with "a.operator*(argument)" or "a->operator*(argument)", 248 // i.e. the operator is called as a member function, 249 // then the argument must be an expression. 250 bool OperatorCalledAsMemberFunction = 251 Prev->Previous && Prev->Previous->isOneOf(tok::period, tok::arrow); 252 Contexts.back().IsExpression = OperatorCalledAsMemberFunction; 253 } else if (Style.isJavaScript() && 254 (Line.startsWith(Keywords.kw_type, tok::identifier) || 255 Line.startsWith(tok::kw_export, Keywords.kw_type, 256 tok::identifier))) { 257 // type X = (...); 258 // export type X = (...); 259 Contexts.back().IsExpression = false; 260 } else if (OpeningParen.Previous && 261 (OpeningParen.Previous->isOneOf(tok::kw_static_assert, 262 tok::kw_while, tok::l_paren, 263 tok::comma) || 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().ContextType = Context::ForEachMacro; 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().ContextType == Context::TemplateArgument); 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().ContextType = Context::StructArrayInitializer; 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 switch (Contexts.back().ContextType) { 1161 case Context::CtorInitializer: 1162 Tok->setType(TT_CtorInitializerComma); 1163 break; 1164 case Context::InheritanceList: 1165 Tok->setType(TT_InheritanceComma); 1166 break; 1167 default: 1168 if (Contexts.back().FirstStartOfName && 1169 (Contexts.size() == 1 || startsWithInitStatement(Line))) { 1170 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 1171 Line.IsMultiVariableDeclStmt = true; 1172 } 1173 break; 1174 } 1175 if (Contexts.back().ContextType == Context::ForEachMacro) 1176 Contexts.back().IsExpression = true; 1177 break; 1178 case tok::identifier: 1179 if (Tok->isOneOf(Keywords.kw___has_include, 1180 Keywords.kw___has_include_next)) 1181 parseHasInclude(); 1182 if (Style.isCSharp() && Tok->is(Keywords.kw_where) && Tok->Next && 1183 Tok->Next->isNot(tok::l_paren)) { 1184 Tok->setType(TT_CSharpGenericTypeConstraint); 1185 parseCSharpGenericTypeConstraint(); 1186 } 1187 break; 1188 case tok::arrow: 1189 if (Tok->Previous && Tok->Previous->is(tok::kw_noexcept)) 1190 Tok->setType(TT_TrailingReturnArrow); 1191 break; 1192 default: 1193 break; 1194 } 1195 return true; 1196 } 1197 1198 void parseCSharpGenericTypeConstraint() { 1199 int OpenAngleBracketsCount = 0; 1200 while (CurrentToken) { 1201 if (CurrentToken->is(tok::less)) { 1202 // parseAngle is too greedy and will consume the whole line. 1203 CurrentToken->setType(TT_TemplateOpener); 1204 ++OpenAngleBracketsCount; 1205 next(); 1206 } else if (CurrentToken->is(tok::greater)) { 1207 CurrentToken->setType(TT_TemplateCloser); 1208 --OpenAngleBracketsCount; 1209 next(); 1210 } else if (CurrentToken->is(tok::comma) && OpenAngleBracketsCount == 0) { 1211 // We allow line breaks after GenericTypeConstraintComma's 1212 // so do not flag commas in Generics as GenericTypeConstraintComma's. 1213 CurrentToken->setType(TT_CSharpGenericTypeConstraintComma); 1214 next(); 1215 } else if (CurrentToken->is(Keywords.kw_where)) { 1216 CurrentToken->setType(TT_CSharpGenericTypeConstraint); 1217 next(); 1218 } else if (CurrentToken->is(tok::colon)) { 1219 CurrentToken->setType(TT_CSharpGenericTypeConstraintColon); 1220 next(); 1221 } else { 1222 next(); 1223 } 1224 } 1225 } 1226 1227 void parseIncludeDirective() { 1228 if (CurrentToken && CurrentToken->is(tok::less)) { 1229 next(); 1230 while (CurrentToken) { 1231 // Mark tokens up to the trailing line comments as implicit string 1232 // literals. 1233 if (CurrentToken->isNot(tok::comment) && 1234 !CurrentToken->TokenText.startswith("//")) 1235 CurrentToken->setType(TT_ImplicitStringLiteral); 1236 next(); 1237 } 1238 } 1239 } 1240 1241 void parseWarningOrError() { 1242 next(); 1243 // We still want to format the whitespace left of the first token of the 1244 // warning or error. 1245 next(); 1246 while (CurrentToken) { 1247 CurrentToken->setType(TT_ImplicitStringLiteral); 1248 next(); 1249 } 1250 } 1251 1252 void parsePragma() { 1253 next(); // Consume "pragma". 1254 if (CurrentToken && 1255 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 1256 bool IsMark = CurrentToken->is(Keywords.kw_mark); 1257 next(); // Consume "mark". 1258 next(); // Consume first token (so we fix leading whitespace). 1259 while (CurrentToken) { 1260 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 1261 CurrentToken->setType(TT_ImplicitStringLiteral); 1262 next(); 1263 } 1264 } 1265 } 1266 1267 void parseHasInclude() { 1268 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 1269 return; 1270 next(); // '(' 1271 parseIncludeDirective(); 1272 next(); // ')' 1273 } 1274 1275 LineType parsePreprocessorDirective() { 1276 bool IsFirstToken = CurrentToken->IsFirst; 1277 LineType Type = LT_PreprocessorDirective; 1278 next(); 1279 if (!CurrentToken) 1280 return Type; 1281 1282 if (Style.isJavaScript() && IsFirstToken) { 1283 // JavaScript files can contain shebang lines of the form: 1284 // #!/usr/bin/env node 1285 // Treat these like C++ #include directives. 1286 while (CurrentToken) { 1287 // Tokens cannot be comments here. 1288 CurrentToken->setType(TT_ImplicitStringLiteral); 1289 next(); 1290 } 1291 return LT_ImportStatement; 1292 } 1293 1294 if (CurrentToken->is(tok::numeric_constant)) { 1295 CurrentToken->SpacesRequiredBefore = 1; 1296 return Type; 1297 } 1298 // Hashes in the middle of a line can lead to any strange token 1299 // sequence. 1300 if (!CurrentToken->Tok.getIdentifierInfo()) 1301 return Type; 1302 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 1303 case tok::pp_include: 1304 case tok::pp_include_next: 1305 case tok::pp_import: 1306 next(); 1307 parseIncludeDirective(); 1308 Type = LT_ImportStatement; 1309 break; 1310 case tok::pp_error: 1311 case tok::pp_warning: 1312 parseWarningOrError(); 1313 break; 1314 case tok::pp_pragma: 1315 parsePragma(); 1316 break; 1317 case tok::pp_if: 1318 case tok::pp_elif: 1319 Contexts.back().IsExpression = true; 1320 next(); 1321 parseLine(); 1322 break; 1323 default: 1324 break; 1325 } 1326 while (CurrentToken) { 1327 FormatToken *Tok = CurrentToken; 1328 next(); 1329 if (Tok->is(tok::l_paren)) 1330 parseParens(); 1331 else if (Tok->isOneOf(Keywords.kw___has_include, 1332 Keywords.kw___has_include_next)) 1333 parseHasInclude(); 1334 } 1335 return Type; 1336 } 1337 1338 public: 1339 LineType parseLine() { 1340 if (!CurrentToken) 1341 return LT_Invalid; 1342 NonTemplateLess.clear(); 1343 if (CurrentToken->is(tok::hash)) 1344 return parsePreprocessorDirective(); 1345 1346 // Directly allow to 'import <string-literal>' to support protocol buffer 1347 // definitions (github.com/google/protobuf) or missing "#" (either way we 1348 // should not break the line). 1349 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 1350 if ((Style.Language == FormatStyle::LK_Java && 1351 CurrentToken->is(Keywords.kw_package)) || 1352 (Info && Info->getPPKeywordID() == tok::pp_import && 1353 CurrentToken->Next && 1354 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 1355 tok::kw_static))) { 1356 next(); 1357 parseIncludeDirective(); 1358 return LT_ImportStatement; 1359 } 1360 1361 // If this line starts and ends in '<' and '>', respectively, it is likely 1362 // part of "#define <a/b.h>". 1363 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 1364 parseIncludeDirective(); 1365 return LT_ImportStatement; 1366 } 1367 1368 // In .proto files, top-level options and package statements are very 1369 // similar to import statements and should not be line-wrapped. 1370 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 1371 CurrentToken->isOneOf(Keywords.kw_option, Keywords.kw_package)) { 1372 next(); 1373 if (CurrentToken && CurrentToken->is(tok::identifier)) { 1374 while (CurrentToken) 1375 next(); 1376 return LT_ImportStatement; 1377 } 1378 } 1379 1380 bool KeywordVirtualFound = false; 1381 bool ImportStatement = false; 1382 1383 // import {...} from '...'; 1384 if (Style.isJavaScript() && CurrentToken->is(Keywords.kw_import)) 1385 ImportStatement = true; 1386 1387 while (CurrentToken) { 1388 if (CurrentToken->is(tok::kw_virtual)) 1389 KeywordVirtualFound = true; 1390 if (Style.isJavaScript()) { 1391 // export {...} from '...'; 1392 // An export followed by "from 'some string';" is a re-export from 1393 // another module identified by a URI and is treated as a 1394 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 1395 // Just "export {...};" or "export class ..." should not be treated as 1396 // an import in this sense. 1397 if (Line.First->is(tok::kw_export) && 1398 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 1399 CurrentToken->Next->isStringLiteral()) 1400 ImportStatement = true; 1401 if (isClosureImportStatement(*CurrentToken)) 1402 ImportStatement = true; 1403 } 1404 if (!consumeToken()) 1405 return LT_Invalid; 1406 } 1407 if (KeywordVirtualFound) 1408 return LT_VirtualFunctionDecl; 1409 if (ImportStatement) 1410 return LT_ImportStatement; 1411 1412 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 1413 if (Contexts.back().FirstObjCSelectorName) 1414 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 1415 Contexts.back().LongestObjCSelectorName; 1416 return LT_ObjCMethodDecl; 1417 } 1418 1419 for (const auto &ctx : Contexts) 1420 if (ctx.ContextType == Context::StructArrayInitializer) 1421 return LT_ArrayOfStructInitializer; 1422 1423 return LT_Other; 1424 } 1425 1426 private: 1427 bool isClosureImportStatement(const FormatToken &Tok) { 1428 // FIXME: Closure-library specific stuff should not be hard-coded but be 1429 // configurable. 1430 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 1431 Tok.Next->Next && 1432 (Tok.Next->Next->TokenText == "module" || 1433 Tok.Next->Next->TokenText == "provide" || 1434 Tok.Next->Next->TokenText == "require" || 1435 Tok.Next->Next->TokenText == "requireType" || 1436 Tok.Next->Next->TokenText == "forwardDeclare") && 1437 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 1438 } 1439 1440 void resetTokenMetadata() { 1441 if (!CurrentToken) 1442 return; 1443 1444 // Reset token type in case we have already looked at it and then 1445 // recovered from an error (e.g. failure to find the matching >). 1446 if (!CurrentToken->isTypeFinalized() && 1447 !CurrentToken->isOneOf( 1448 TT_LambdaLSquare, TT_LambdaLBrace, TT_AttributeMacro, TT_IfMacro, 1449 TT_ForEachMacro, TT_TypenameMacro, TT_FunctionLBrace, 1450 TT_ImplicitStringLiteral, TT_InlineASMBrace, TT_FatArrow, 1451 TT_LambdaArrow, TT_NamespaceMacro, TT_OverloadedOperator, 1452 TT_RegexLiteral, TT_TemplateString, TT_ObjCStringLiteral, 1453 TT_UntouchableMacroFunc, TT_StatementAttributeLikeMacro, 1454 TT_FunctionLikeOrFreestandingMacro, TT_ClassLBrace, TT_EnumLBrace, 1455 TT_RecordLBrace, TT_StructLBrace, TT_UnionLBrace, TT_RequiresClause, 1456 TT_RequiresClauseInARequiresExpression, TT_RequiresExpression, 1457 TT_RequiresExpressionLParen, TT_RequiresExpressionLBrace, 1458 TT_CompoundRequirementLBrace, TT_BracedListLBrace)) 1459 CurrentToken->setType(TT_Unknown); 1460 CurrentToken->Role.reset(); 1461 CurrentToken->MatchingParen = nullptr; 1462 CurrentToken->FakeLParens.clear(); 1463 CurrentToken->FakeRParens = 0; 1464 } 1465 1466 void next() { 1467 if (!CurrentToken) 1468 return; 1469 1470 CurrentToken->NestingLevel = Contexts.size() - 1; 1471 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1472 modifyContext(*CurrentToken); 1473 determineTokenType(*CurrentToken); 1474 CurrentToken = CurrentToken->Next; 1475 1476 resetTokenMetadata(); 1477 } 1478 1479 /// A struct to hold information valid in a specific context, e.g. 1480 /// a pair of parenthesis. 1481 struct Context { 1482 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1483 bool IsExpression) 1484 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1485 IsExpression(IsExpression) {} 1486 1487 tok::TokenKind ContextKind; 1488 unsigned BindingStrength; 1489 bool IsExpression; 1490 unsigned LongestObjCSelectorName = 0; 1491 bool ColonIsForRangeExpr = false; 1492 bool ColonIsDictLiteral = false; 1493 bool ColonIsObjCMethodExpr = false; 1494 FormatToken *FirstObjCSelectorName = nullptr; 1495 FormatToken *FirstStartOfName = nullptr; 1496 bool CanBeExpression = true; 1497 bool CaretFound = false; 1498 bool InCpp11AttributeSpecifier = false; 1499 bool InCSharpAttributeSpecifier = false; 1500 enum { 1501 Unknown, 1502 // Like the part after `:` in a constructor. 1503 // Context(...) : IsExpression(IsExpression) 1504 CtorInitializer, 1505 // Like in the parentheses in a foreach. 1506 ForEachMacro, 1507 // Like the inheritance list in a class declaration. 1508 // class Input : public IO 1509 InheritanceList, 1510 // Like in the braced list. 1511 // int x[] = {}; 1512 StructArrayInitializer, 1513 // Like in `static_cast<int>`. 1514 TemplateArgument, 1515 } ContextType = Unknown; 1516 }; 1517 1518 /// Puts a new \c Context onto the stack \c Contexts for the lifetime 1519 /// of each instance. 1520 struct ScopedContextCreator { 1521 AnnotatingParser &P; 1522 1523 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1524 unsigned Increase) 1525 : P(P) { 1526 P.Contexts.push_back(Context(ContextKind, 1527 P.Contexts.back().BindingStrength + Increase, 1528 P.Contexts.back().IsExpression)); 1529 } 1530 1531 ~ScopedContextCreator() { 1532 if (P.Style.AlignArrayOfStructures != FormatStyle::AIAS_None) { 1533 if (P.Contexts.back().ContextType == Context::StructArrayInitializer) { 1534 P.Contexts.pop_back(); 1535 P.Contexts.back().ContextType = Context::StructArrayInitializer; 1536 return; 1537 } 1538 } 1539 P.Contexts.pop_back(); 1540 } 1541 }; 1542 1543 void modifyContext(const FormatToken &Current) { 1544 auto AssignmentStartsExpression = [&]() { 1545 if (Current.getPrecedence() != prec::Assignment) 1546 return false; 1547 1548 if (Line.First->isOneOf(tok::kw_using, tok::kw_return)) 1549 return false; 1550 if (Line.First->is(tok::kw_template)) { 1551 assert(Current.Previous); 1552 if (Current.Previous->is(tok::kw_operator)) { 1553 // `template ... operator=` cannot be an expression. 1554 return false; 1555 } 1556 1557 // `template` keyword can start a variable template. 1558 const FormatToken *Tok = Line.First->getNextNonComment(); 1559 assert(Tok); // Current token is on the same line. 1560 if (Tok->isNot(TT_TemplateOpener)) { 1561 // Explicit template instantiations do not have `<>`. 1562 return false; 1563 } 1564 1565 Tok = Tok->MatchingParen; 1566 if (!Tok) 1567 return false; 1568 Tok = Tok->getNextNonComment(); 1569 if (!Tok) 1570 return false; 1571 1572 if (Tok->isOneOf(tok::kw_class, tok::kw_enum, tok::kw_concept, 1573 tok::kw_struct, tok::kw_using)) 1574 return false; 1575 1576 return true; 1577 } 1578 1579 // Type aliases use `type X = ...;` in TypeScript and can be exported 1580 // using `export type ...`. 1581 if (Style.isJavaScript() && 1582 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1583 Line.startsWith(tok::kw_export, Keywords.kw_type, tok::identifier))) 1584 return false; 1585 1586 return !Current.Previous || Current.Previous->isNot(tok::kw_operator); 1587 }; 1588 1589 if (AssignmentStartsExpression()) { 1590 Contexts.back().IsExpression = true; 1591 if (!Line.startsWith(TT_UnaryOperator)) { 1592 for (FormatToken *Previous = Current.Previous; 1593 Previous && Previous->Previous && 1594 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1595 Previous = Previous->Previous) { 1596 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1597 Previous = Previous->MatchingParen; 1598 if (!Previous) 1599 break; 1600 } 1601 if (Previous->opensScope()) 1602 break; 1603 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1604 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1605 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1606 Previous->setType(TT_PointerOrReference); 1607 } 1608 } 1609 } else if (Current.is(tok::lessless) && 1610 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1611 Contexts.back().IsExpression = true; 1612 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1613 Contexts.back().IsExpression = true; 1614 } else if (Current.is(TT_TrailingReturnArrow)) { 1615 Contexts.back().IsExpression = false; 1616 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1617 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1618 } else if (Current.Previous && 1619 Current.Previous->is(TT_CtorInitializerColon)) { 1620 Contexts.back().IsExpression = true; 1621 Contexts.back().ContextType = Context::CtorInitializer; 1622 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1623 Contexts.back().ContextType = Context::InheritanceList; 1624 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1625 for (FormatToken *Previous = Current.Previous; 1626 Previous && Previous->isOneOf(tok::star, tok::amp); 1627 Previous = Previous->Previous) 1628 Previous->setType(TT_PointerOrReference); 1629 if (Line.MustBeDeclaration && 1630 Contexts.front().ContextType != Context::CtorInitializer) 1631 Contexts.back().IsExpression = false; 1632 } else if (Current.is(tok::kw_new)) { 1633 Contexts.back().CanBeExpression = false; 1634 } else if (Current.is(tok::semi) || 1635 (Current.is(tok::exclaim) && Current.Previous && 1636 !Current.Previous->is(tok::kw_operator))) { 1637 // This should be the condition or increment in a for-loop. 1638 // But not operator !() (can't use TT_OverloadedOperator here as its not 1639 // been annotated yet). 1640 Contexts.back().IsExpression = true; 1641 } 1642 } 1643 1644 static FormatToken *untilMatchingParen(FormatToken *Current) { 1645 // Used when `MatchingParen` is not yet established. 1646 int ParenLevel = 0; 1647 while (Current) { 1648 if (Current->is(tok::l_paren)) 1649 ++ParenLevel; 1650 if (Current->is(tok::r_paren)) 1651 --ParenLevel; 1652 if (ParenLevel < 1) 1653 break; 1654 Current = Current->Next; 1655 } 1656 return Current; 1657 } 1658 1659 static bool isDeductionGuide(FormatToken &Current) { 1660 // Look for a deduction guide template<T> A(...) -> A<...>; 1661 if (Current.Previous && Current.Previous->is(tok::r_paren) && 1662 Current.startsSequence(tok::arrow, tok::identifier, tok::less)) { 1663 // Find the TemplateCloser. 1664 FormatToken *TemplateCloser = Current.Next->Next; 1665 int NestingLevel = 0; 1666 while (TemplateCloser) { 1667 // Skip over an expressions in parens A<(3 < 2)>; 1668 if (TemplateCloser->is(tok::l_paren)) { 1669 // No Matching Paren yet so skip to matching paren 1670 TemplateCloser = untilMatchingParen(TemplateCloser); 1671 if (!TemplateCloser) 1672 break; 1673 } 1674 if (TemplateCloser->is(tok::less)) 1675 ++NestingLevel; 1676 if (TemplateCloser->is(tok::greater)) 1677 --NestingLevel; 1678 if (NestingLevel < 1) 1679 break; 1680 TemplateCloser = TemplateCloser->Next; 1681 } 1682 // Assuming we have found the end of the template ensure its followed 1683 // with a semi-colon. 1684 if (TemplateCloser && TemplateCloser->Next && 1685 TemplateCloser->Next->is(tok::semi) && 1686 Current.Previous->MatchingParen) { 1687 // Determine if the identifier `A` prior to the A<..>; is the same as 1688 // prior to the A(..) 1689 FormatToken *LeadingIdentifier = 1690 Current.Previous->MatchingParen->Previous; 1691 1692 // Differentiate a deduction guide by seeing the 1693 // > of the template prior to the leading identifier. 1694 if (LeadingIdentifier) { 1695 FormatToken *PriorLeadingIdentifier = LeadingIdentifier->Previous; 1696 // Skip back past explicit decoration 1697 if (PriorLeadingIdentifier && 1698 PriorLeadingIdentifier->is(tok::kw_explicit)) 1699 PriorLeadingIdentifier = PriorLeadingIdentifier->Previous; 1700 1701 return PriorLeadingIdentifier && 1702 (PriorLeadingIdentifier->is(TT_TemplateCloser) || 1703 PriorLeadingIdentifier->ClosesRequiresClause) && 1704 LeadingIdentifier->TokenText == Current.Next->TokenText; 1705 } 1706 } 1707 } 1708 return false; 1709 } 1710 1711 void determineTokenType(FormatToken &Current) { 1712 if (!Current.is(TT_Unknown)) 1713 // The token type is already known. 1714 return; 1715 1716 if ((Style.isJavaScript() || Style.isCSharp()) && 1717 Current.is(tok::exclaim)) { 1718 if (Current.Previous) { 1719 bool IsIdentifier = 1720 Style.isJavaScript() 1721 ? Keywords.IsJavaScriptIdentifier( 1722 *Current.Previous, /* AcceptIdentifierName= */ true) 1723 : Current.Previous->is(tok::identifier); 1724 if (IsIdentifier || 1725 Current.Previous->isOneOf( 1726 tok::kw_default, tok::kw_namespace, tok::r_paren, tok::r_square, 1727 tok::r_brace, tok::kw_false, tok::kw_true, Keywords.kw_type, 1728 Keywords.kw_get, Keywords.kw_init, Keywords.kw_set) || 1729 Current.Previous->Tok.isLiteral()) { 1730 Current.setType(TT_NonNullAssertion); 1731 return; 1732 } 1733 } 1734 if (Current.Next && 1735 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1736 Current.setType(TT_NonNullAssertion); 1737 return; 1738 } 1739 } 1740 1741 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1742 // function declaration have been found. In this case, 'Current' is a 1743 // trailing token of this declaration and thus cannot be a name. 1744 if (Current.is(Keywords.kw_instanceof)) { 1745 Current.setType(TT_BinaryOperator); 1746 } else if (isStartOfName(Current) && 1747 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1748 Contexts.back().FirstStartOfName = &Current; 1749 Current.setType(TT_StartOfName); 1750 } else if (Current.is(tok::semi)) { 1751 // Reset FirstStartOfName after finding a semicolon so that a for loop 1752 // with multiple increment statements is not confused with a for loop 1753 // having multiple variable declarations. 1754 Contexts.back().FirstStartOfName = nullptr; 1755 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1756 AutoFound = true; 1757 } else if (Current.is(tok::arrow) && 1758 Style.Language == FormatStyle::LK_Java) { 1759 Current.setType(TT_LambdaArrow); 1760 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1761 Current.NestingLevel == 0 && 1762 !Current.Previous->isOneOf(tok::kw_operator, tok::identifier)) { 1763 // not auto operator->() -> xxx; 1764 Current.setType(TT_TrailingReturnArrow); 1765 } else if (Current.is(tok::arrow) && Current.Previous && 1766 Current.Previous->is(tok::r_brace)) { 1767 // Concept implicit conversion constraint needs to be treated like 1768 // a trailing return type ... } -> <type>. 1769 Current.setType(TT_TrailingReturnArrow); 1770 } else if (isDeductionGuide(Current)) { 1771 // Deduction guides trailing arrow " A(...) -> A<T>;". 1772 Current.setType(TT_TrailingReturnArrow); 1773 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1774 Current.setType(determineStarAmpUsage( 1775 Current, 1776 Contexts.back().CanBeExpression && Contexts.back().IsExpression, 1777 Contexts.back().ContextType == Context::TemplateArgument)); 1778 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1779 Current.setType(determinePlusMinusCaretUsage(Current)); 1780 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1781 Contexts.back().CaretFound = true; 1782 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1783 Current.setType(determineIncrementUsage(Current)); 1784 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1785 Current.setType(TT_UnaryOperator); 1786 } else if (Current.is(tok::question)) { 1787 if (Style.isJavaScript() && Line.MustBeDeclaration && 1788 !Contexts.back().IsExpression) { 1789 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1790 // on the interface, not a ternary expression. 1791 Current.setType(TT_JsTypeOptionalQuestion); 1792 } else { 1793 Current.setType(TT_ConditionalExpr); 1794 } 1795 } else if (Current.isBinaryOperator() && 1796 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1797 (!Current.is(tok::greater) && 1798 Style.Language != FormatStyle::LK_TextProto)) { 1799 Current.setType(TT_BinaryOperator); 1800 } else if (Current.is(tok::comment)) { 1801 if (Current.TokenText.startswith("/*")) 1802 if (Current.TokenText.endswith("*/")) 1803 Current.setType(TT_BlockComment); 1804 else 1805 // The lexer has for some reason determined a comment here. But we 1806 // cannot really handle it, if it isn't properly terminated. 1807 Current.Tok.setKind(tok::unknown); 1808 else 1809 Current.setType(TT_LineComment); 1810 } else if (Current.is(tok::l_paren)) { 1811 if (lParenStartsCppCast(Current)) 1812 Current.setType(TT_CppCastLParen); 1813 } else if (Current.is(tok::r_paren)) { 1814 if (rParenEndsCast(Current)) 1815 Current.setType(TT_CastRParen); 1816 if (Current.MatchingParen && Current.Next && 1817 !Current.Next->isBinaryOperator() && 1818 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1819 tok::comma, tok::period, tok::arrow, 1820 tok::coloncolon)) 1821 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1822 // Make sure this isn't the return type of an Obj-C block declaration 1823 if (AfterParen->isNot(tok::caret)) { 1824 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1825 if (BeforeParen->is(tok::identifier) && 1826 !BeforeParen->is(TT_TypenameMacro) && 1827 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1828 (!BeforeParen->Previous || 1829 BeforeParen->Previous->ClosesTemplateDeclaration)) 1830 Current.setType(TT_FunctionAnnotationRParen); 1831 } 1832 } 1833 } else if (Current.is(tok::at) && Current.Next && !Style.isJavaScript() && 1834 Style.Language != FormatStyle::LK_Java) { 1835 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1836 // marks declarations and properties that need special formatting. 1837 switch (Current.Next->Tok.getObjCKeywordID()) { 1838 case tok::objc_interface: 1839 case tok::objc_implementation: 1840 case tok::objc_protocol: 1841 Current.setType(TT_ObjCDecl); 1842 break; 1843 case tok::objc_property: 1844 Current.setType(TT_ObjCProperty); 1845 break; 1846 default: 1847 break; 1848 } 1849 } else if (Current.is(tok::period)) { 1850 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1851 if (PreviousNoComment && 1852 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1853 Current.setType(TT_DesignatedInitializerPeriod); 1854 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1855 Current.Previous->isOneOf(TT_JavaAnnotation, 1856 TT_LeadingJavaAnnotation)) 1857 Current.setType(Current.Previous->getType()); 1858 } else if (canBeObjCSelectorComponent(Current) && 1859 // FIXME(bug 36976): ObjC return types shouldn't use 1860 // TT_CastRParen. 1861 Current.Previous && Current.Previous->is(TT_CastRParen) && 1862 Current.Previous->MatchingParen && 1863 Current.Previous->MatchingParen->Previous && 1864 Current.Previous->MatchingParen->Previous->is( 1865 TT_ObjCMethodSpecifier)) { 1866 // This is the first part of an Objective-C selector name. (If there's no 1867 // colon after this, this is the only place which annotates the identifier 1868 // as a selector.) 1869 Current.setType(TT_SelectorName); 1870 } else if (Current.isOneOf(tok::identifier, tok::kw_const, tok::kw_noexcept, 1871 tok::kw_requires) && 1872 Current.Previous && 1873 !Current.Previous->isOneOf(tok::equal, tok::at) && 1874 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1875 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1876 // function declaration have been found. 1877 Current.setType(TT_TrailingAnnotation); 1878 } else if ((Style.Language == FormatStyle::LK_Java || 1879 Style.isJavaScript()) && 1880 Current.Previous) { 1881 if (Current.Previous->is(tok::at) && 1882 Current.isNot(Keywords.kw_interface)) { 1883 const FormatToken &AtToken = *Current.Previous; 1884 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1885 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1886 Current.setType(TT_LeadingJavaAnnotation); 1887 else 1888 Current.setType(TT_JavaAnnotation); 1889 } else if (Current.Previous->is(tok::period) && 1890 Current.Previous->isOneOf(TT_JavaAnnotation, 1891 TT_LeadingJavaAnnotation)) { 1892 Current.setType(Current.Previous->getType()); 1893 } 1894 } 1895 } 1896 1897 /// Take a guess at whether \p Tok starts a name of a function or 1898 /// variable declaration. 1899 /// 1900 /// This is a heuristic based on whether \p Tok is an identifier following 1901 /// something that is likely a type. 1902 bool isStartOfName(const FormatToken &Tok) { 1903 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1904 return false; 1905 1906 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1907 Keywords.kw_as)) 1908 return false; 1909 if (Style.isJavaScript() && Tok.Previous->is(Keywords.kw_in)) 1910 return false; 1911 1912 // Skip "const" as it does not have an influence on whether this is a name. 1913 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1914 1915 // For javascript const can be like "let" or "var" 1916 if (!Style.isJavaScript()) 1917 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1918 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1919 1920 if (!PreviousNotConst) 1921 return false; 1922 1923 if (PreviousNotConst->ClosesRequiresClause) 1924 return false; 1925 1926 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1927 PreviousNotConst->Previous && 1928 PreviousNotConst->Previous->is(tok::hash); 1929 1930 if (PreviousNotConst->is(TT_TemplateCloser)) 1931 return PreviousNotConst && PreviousNotConst->MatchingParen && 1932 PreviousNotConst->MatchingParen->Previous && 1933 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1934 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1935 1936 if (PreviousNotConst->is(tok::r_paren) && 1937 PreviousNotConst->is(TT_TypeDeclarationParen)) 1938 return true; 1939 1940 // If is a preprocess keyword like #define. 1941 if (IsPPKeyword) 1942 return false; 1943 1944 // int a or auto a. 1945 if (PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) 1946 return true; 1947 1948 // *a or &a or &&a. 1949 if (PreviousNotConst->is(TT_PointerOrReference)) 1950 return true; 1951 1952 // MyClass a; 1953 if (PreviousNotConst->isSimpleTypeSpecifier()) 1954 return true; 1955 1956 // const a = in JavaScript. 1957 return Style.isJavaScript() && PreviousNotConst->is(tok::kw_const); 1958 } 1959 1960 /// Determine whether '(' is starting a C++ cast. 1961 bool lParenStartsCppCast(const FormatToken &Tok) { 1962 // C-style casts are only used in C++. 1963 if (!Style.isCpp()) 1964 return false; 1965 1966 FormatToken *LeftOfParens = Tok.getPreviousNonComment(); 1967 if (LeftOfParens && LeftOfParens->is(TT_TemplateCloser) && 1968 LeftOfParens->MatchingParen) { 1969 auto *Prev = LeftOfParens->MatchingParen->getPreviousNonComment(); 1970 if (Prev && Prev->isOneOf(tok::kw_const_cast, tok::kw_dynamic_cast, 1971 tok::kw_reinterpret_cast, tok::kw_static_cast)) 1972 // FIXME: Maybe we should handle identifiers ending with "_cast", 1973 // e.g. any_cast? 1974 return true; 1975 } 1976 return false; 1977 } 1978 1979 /// Determine whether ')' is ending a cast. 1980 bool rParenEndsCast(const FormatToken &Tok) { 1981 // C-style casts are only used in C++, C# and Java. 1982 if (!Style.isCSharp() && !Style.isCpp() && 1983 Style.Language != FormatStyle::LK_Java) 1984 return false; 1985 1986 // Empty parens aren't casts and there are no casts at the end of the line. 1987 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1988 return false; 1989 1990 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1991 if (LeftOfParens) { 1992 // If there is a closing parenthesis left of the current 1993 // parentheses, look past it as these might be chained casts. 1994 if (LeftOfParens->is(tok::r_paren) && 1995 LeftOfParens->isNot(TT_CastRParen)) { 1996 if (!LeftOfParens->MatchingParen || 1997 !LeftOfParens->MatchingParen->Previous) 1998 return false; 1999 LeftOfParens = LeftOfParens->MatchingParen->Previous; 2000 } 2001 2002 if (LeftOfParens->is(tok::r_square)) { 2003 // delete[] (void *)ptr; 2004 auto MayBeArrayDelete = [](FormatToken *Tok) -> FormatToken * { 2005 if (Tok->isNot(tok::r_square)) 2006 return nullptr; 2007 2008 Tok = Tok->getPreviousNonComment(); 2009 if (!Tok || Tok->isNot(tok::l_square)) 2010 return nullptr; 2011 2012 Tok = Tok->getPreviousNonComment(); 2013 if (!Tok || Tok->isNot(tok::kw_delete)) 2014 return nullptr; 2015 return Tok; 2016 }; 2017 if (FormatToken *MaybeDelete = MayBeArrayDelete(LeftOfParens)) 2018 LeftOfParens = MaybeDelete; 2019 } 2020 2021 // The Condition directly below this one will see the operator arguments 2022 // as a (void *foo) cast. 2023 // void operator delete(void *foo) ATTRIB; 2024 if (LeftOfParens->Tok.getIdentifierInfo() && LeftOfParens->Previous && 2025 LeftOfParens->Previous->is(tok::kw_operator)) 2026 return false; 2027 2028 // If there is an identifier (or with a few exceptions a keyword) right 2029 // before the parentheses, this is unlikely to be a cast. 2030 if (LeftOfParens->Tok.getIdentifierInfo() && 2031 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 2032 tok::kw_delete)) 2033 return false; 2034 2035 // Certain other tokens right before the parentheses are also signals that 2036 // this cannot be a cast. 2037 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 2038 TT_TemplateCloser, tok::ellipsis)) 2039 return false; 2040 } 2041 2042 if (Tok.Next->is(tok::question)) 2043 return false; 2044 2045 // `foreach((A a, B b) in someList)` should not be seen as a cast. 2046 if (Tok.Next->is(Keywords.kw_in) && Style.isCSharp()) 2047 return false; 2048 2049 // Functions which end with decorations like volatile, noexcept are unlikely 2050 // to be casts. 2051 if (Tok.Next->isOneOf(tok::kw_noexcept, tok::kw_volatile, tok::kw_const, 2052 tok::kw_requires, tok::kw_throw, tok::arrow, 2053 Keywords.kw_override, Keywords.kw_final) || 2054 isCpp11AttributeSpecifier(*Tok.Next)) 2055 return false; 2056 2057 // As Java has no function types, a "(" after the ")" likely means that this 2058 // is a cast. 2059 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 2060 return true; 2061 2062 // If a (non-string) literal follows, this is likely a cast. 2063 if (Tok.Next->isNot(tok::string_literal) && 2064 (Tok.Next->Tok.isLiteral() || 2065 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 2066 return true; 2067 2068 // Heuristically try to determine whether the parentheses contain a type. 2069 auto IsQualifiedPointerOrReference = [](FormatToken *T) { 2070 // This is used to handle cases such as x = (foo *const)&y; 2071 assert(!T->isSimpleTypeSpecifier() && "Should have already been checked"); 2072 // Strip trailing qualifiers such as const or volatile when checking 2073 // whether the parens could be a cast to a pointer/reference type. 2074 while (T) { 2075 if (T->is(TT_AttributeParen)) { 2076 // Handle `x = (foo *__attribute__((foo)))&v;`: 2077 if (T->MatchingParen && T->MatchingParen->Previous && 2078 T->MatchingParen->Previous->is(tok::kw___attribute)) { 2079 T = T->MatchingParen->Previous->Previous; 2080 continue; 2081 } 2082 } else if (T->is(TT_AttributeSquare)) { 2083 // Handle `x = (foo *[[clang::foo]])&v;`: 2084 if (T->MatchingParen && T->MatchingParen->Previous) { 2085 T = T->MatchingParen->Previous; 2086 continue; 2087 } 2088 } else if (T->canBePointerOrReferenceQualifier()) { 2089 T = T->Previous; 2090 continue; 2091 } 2092 break; 2093 } 2094 return T && T->is(TT_PointerOrReference); 2095 }; 2096 bool ParensAreType = 2097 !Tok.Previous || 2098 Tok.Previous->isOneOf(TT_TemplateCloser, TT_TypeDeclarationParen) || 2099 Tok.Previous->isSimpleTypeSpecifier() || 2100 IsQualifiedPointerOrReference(Tok.Previous); 2101 bool ParensCouldEndDecl = 2102 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 2103 if (ParensAreType && !ParensCouldEndDecl) 2104 return true; 2105 2106 // At this point, we heuristically assume that there are no casts at the 2107 // start of the line. We assume that we have found most cases where there 2108 // are by the logic above, e.g. "(void)x;". 2109 if (!LeftOfParens) 2110 return false; 2111 2112 // Certain token types inside the parentheses mean that this can't be a 2113 // cast. 2114 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 2115 Token = Token->Next) 2116 if (Token->is(TT_BinaryOperator)) 2117 return false; 2118 2119 // If the following token is an identifier or 'this', this is a cast. All 2120 // cases where this can be something else are handled above. 2121 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 2122 return true; 2123 2124 // Look for a cast `( x ) (`. 2125 if (Tok.Next->is(tok::l_paren) && Tok.Previous && Tok.Previous->Previous) { 2126 if (Tok.Previous->is(tok::identifier) && 2127 Tok.Previous->Previous->is(tok::l_paren)) 2128 return true; 2129 } 2130 2131 if (!Tok.Next->Next) 2132 return false; 2133 2134 // If the next token after the parenthesis is a unary operator, assume 2135 // that this is cast, unless there are unexpected tokens inside the 2136 // parenthesis. 2137 bool NextIsUnary = 2138 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 2139 if (!NextIsUnary || Tok.Next->is(tok::plus) || 2140 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 2141 return false; 2142 // Search for unexpected tokens. 2143 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 2144 Prev = Prev->Previous) 2145 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 2146 return false; 2147 return true; 2148 } 2149 2150 /// Return the type of the given token assuming it is * or &. 2151 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 2152 bool InTemplateArgument) { 2153 if (Style.isJavaScript()) 2154 return TT_BinaryOperator; 2155 2156 // && in C# must be a binary operator. 2157 if (Style.isCSharp() && Tok.is(tok::ampamp)) 2158 return TT_BinaryOperator; 2159 2160 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2161 if (!PrevToken) 2162 return TT_UnaryOperator; 2163 2164 const FormatToken *NextToken = Tok.getNextNonComment(); 2165 2166 if (InTemplateArgument && NextToken && NextToken->is(tok::kw_noexcept)) 2167 return TT_BinaryOperator; 2168 2169 if (!NextToken || 2170 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_noexcept) || 2171 NextToken->canBePointerOrReferenceQualifier() || 2172 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 2173 return TT_PointerOrReference; 2174 2175 if (PrevToken->is(tok::coloncolon)) 2176 return TT_PointerOrReference; 2177 2178 if (PrevToken->is(tok::r_paren) && PrevToken->is(TT_TypeDeclarationParen)) 2179 return TT_PointerOrReference; 2180 2181 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 2182 tok::comma, tok::semi, tok::kw_return, tok::colon, 2183 tok::kw_co_return, tok::kw_co_await, 2184 tok::kw_co_yield, tok::equal, tok::kw_delete, 2185 tok::kw_sizeof, tok::kw_throw, TT_BinaryOperator, 2186 TT_ConditionalExpr, TT_UnaryOperator, TT_CastRParen)) 2187 return TT_UnaryOperator; 2188 2189 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 2190 return TT_PointerOrReference; 2191 if (NextToken->is(tok::kw_operator) && !IsExpression) 2192 return TT_PointerOrReference; 2193 if (NextToken->isOneOf(tok::comma, tok::semi)) 2194 return TT_PointerOrReference; 2195 2196 if (PrevToken->Tok.isLiteral() || 2197 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 2198 tok::kw_false, tok::r_brace)) 2199 return TT_BinaryOperator; 2200 2201 const FormatToken *NextNonParen = NextToken; 2202 while (NextNonParen && NextNonParen->is(tok::l_paren)) 2203 NextNonParen = NextNonParen->getNextNonComment(); 2204 if (NextNonParen && (NextNonParen->Tok.isLiteral() || 2205 NextNonParen->isOneOf(tok::kw_true, tok::kw_false) || 2206 NextNonParen->isUnaryOperator())) 2207 return TT_BinaryOperator; 2208 2209 // If we know we're in a template argument, there are no named declarations. 2210 // Thus, having an identifier on the right-hand side indicates a binary 2211 // operator. 2212 if (InTemplateArgument && NextToken->Tok.isAnyIdentifier()) 2213 return TT_BinaryOperator; 2214 2215 // "&&(" is quite unlikely to be two successive unary "&". 2216 if (Tok.is(tok::ampamp) && NextToken->is(tok::l_paren)) 2217 return TT_BinaryOperator; 2218 2219 // This catches some cases where evaluation order is used as control flow: 2220 // aaa && aaa->f(); 2221 if (NextToken->Tok.isAnyIdentifier()) { 2222 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 2223 if (NextNextToken && NextNextToken->is(tok::arrow)) 2224 return TT_BinaryOperator; 2225 } 2226 2227 // It is very unlikely that we are going to find a pointer or reference type 2228 // definition on the RHS of an assignment. 2229 if (IsExpression && !Contexts.back().CaretFound) 2230 return TT_BinaryOperator; 2231 2232 return TT_PointerOrReference; 2233 } 2234 2235 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 2236 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2237 if (!PrevToken) 2238 return TT_UnaryOperator; 2239 2240 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 2241 // This must be a sequence of leading unary operators. 2242 return TT_UnaryOperator; 2243 2244 // Use heuristics to recognize unary operators. 2245 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 2246 tok::question, tok::colon, tok::kw_return, 2247 tok::kw_case, tok::at, tok::l_brace, tok::kw_throw, 2248 tok::kw_co_return, tok::kw_co_yield)) 2249 return TT_UnaryOperator; 2250 2251 // There can't be two consecutive binary operators. 2252 if (PrevToken->is(TT_BinaryOperator)) 2253 return TT_UnaryOperator; 2254 2255 // Fall back to marking the token as binary operator. 2256 return TT_BinaryOperator; 2257 } 2258 2259 /// Determine whether ++/-- are pre- or post-increments/-decrements. 2260 TokenType determineIncrementUsage(const FormatToken &Tok) { 2261 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2262 if (!PrevToken || PrevToken->is(TT_CastRParen)) 2263 return TT_UnaryOperator; 2264 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 2265 return TT_TrailingUnaryOperator; 2266 2267 return TT_UnaryOperator; 2268 } 2269 2270 SmallVector<Context, 8> Contexts; 2271 2272 const FormatStyle &Style; 2273 AnnotatedLine &Line; 2274 FormatToken *CurrentToken; 2275 bool AutoFound; 2276 const AdditionalKeywords &Keywords; 2277 2278 // Set of "<" tokens that do not open a template parameter list. If parseAngle 2279 // determines that a specific token can't be a template opener, it will make 2280 // same decision irrespective of the decisions for tokens leading up to it. 2281 // Store this information to prevent this from causing exponential runtime. 2282 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 2283 }; 2284 2285 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 2286 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 2287 2288 /// Parses binary expressions by inserting fake parenthesis based on 2289 /// operator precedence. 2290 class ExpressionParser { 2291 public: 2292 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 2293 AnnotatedLine &Line) 2294 : Style(Style), Keywords(Keywords), Line(Line), Current(Line.First) {} 2295 2296 /// Parse expressions with the given operator precedence. 2297 void parse(int Precedence = 0) { 2298 // Skip 'return' and ObjC selector colons as they are not part of a binary 2299 // expression. 2300 while (Current && (Current->is(tok::kw_return) || 2301 (Current->is(tok::colon) && 2302 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 2303 next(); 2304 2305 if (!Current || Precedence > PrecedenceArrowAndPeriod) 2306 return; 2307 2308 // Conditional expressions need to be parsed separately for proper nesting. 2309 if (Precedence == prec::Conditional) { 2310 parseConditionalExpr(); 2311 return; 2312 } 2313 2314 // Parse unary operators, which all have a higher precedence than binary 2315 // operators. 2316 if (Precedence == PrecedenceUnaryOperator) { 2317 parseUnaryOperator(); 2318 return; 2319 } 2320 2321 FormatToken *Start = Current; 2322 FormatToken *LatestOperator = nullptr; 2323 unsigned OperatorIndex = 0; 2324 2325 while (Current) { 2326 // Consume operators with higher precedence. 2327 parse(Precedence + 1); 2328 2329 int CurrentPrecedence = getCurrentPrecedence(); 2330 2331 if (Precedence == CurrentPrecedence && Current && 2332 Current->is(TT_SelectorName)) { 2333 if (LatestOperator) 2334 addFakeParenthesis(Start, prec::Level(Precedence)); 2335 Start = Current; 2336 } 2337 2338 // At the end of the line or when an operator with higher precedence is 2339 // found, insert fake parenthesis and return. 2340 if (!Current || 2341 (Current->closesScope() && 2342 (Current->MatchingParen || Current->is(TT_TemplateString))) || 2343 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 2344 (CurrentPrecedence == prec::Conditional && 2345 Precedence == prec::Assignment && Current->is(tok::colon))) 2346 break; 2347 2348 // Consume scopes: (), [], <> and {} 2349 // In addition to that we handle require clauses as scope, so that the 2350 // constraints in that are correctly indented. 2351 if (Current->opensScope() || 2352 Current->isOneOf(TT_RequiresClause, 2353 TT_RequiresClauseInARequiresExpression)) { 2354 // In fragment of a JavaScript template string can look like '}..${' and 2355 // thus close a scope and open a new one at the same time. 2356 while (Current && (!Current->closesScope() || Current->opensScope())) { 2357 next(); 2358 parse(); 2359 } 2360 next(); 2361 } else { 2362 // Operator found. 2363 if (CurrentPrecedence == Precedence) { 2364 if (LatestOperator) 2365 LatestOperator->NextOperator = Current; 2366 LatestOperator = Current; 2367 Current->OperatorIndex = OperatorIndex; 2368 ++OperatorIndex; 2369 } 2370 next(/*SkipPastLeadingComments=*/Precedence > 0); 2371 } 2372 } 2373 2374 if (LatestOperator && (Current || Precedence > 0)) { 2375 // The requires clauses do not neccessarily end in a semicolon or a brace, 2376 // but just go over to struct/class or a function declaration, we need to 2377 // intervene so that the fake right paren is inserted correctly. 2378 auto End = 2379 (Start->Previous && 2380 Start->Previous->isOneOf(TT_RequiresClause, 2381 TT_RequiresClauseInARequiresExpression)) 2382 ? [this](){ 2383 auto Ret = Current ? Current : Line.Last; 2384 while (!Ret->ClosesRequiresClause && Ret->Previous) 2385 Ret = Ret->Previous; 2386 return Ret; 2387 }() 2388 : nullptr; 2389 2390 if (Precedence == PrecedenceArrowAndPeriod) { 2391 // Call expressions don't have a binary operator precedence. 2392 addFakeParenthesis(Start, prec::Unknown, End); 2393 } else { 2394 addFakeParenthesis(Start, prec::Level(Precedence), End); 2395 } 2396 } 2397 } 2398 2399 private: 2400 /// Gets the precedence (+1) of the given token for binary operators 2401 /// and other tokens that we treat like binary operators. 2402 int getCurrentPrecedence() { 2403 if (Current) { 2404 const FormatToken *NextNonComment = Current->getNextNonComment(); 2405 if (Current->is(TT_ConditionalExpr)) 2406 return prec::Conditional; 2407 if (NextNonComment && Current->is(TT_SelectorName) && 2408 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 2409 ((Style.Language == FormatStyle::LK_Proto || 2410 Style.Language == FormatStyle::LK_TextProto) && 2411 NextNonComment->is(tok::less)))) 2412 return prec::Assignment; 2413 if (Current->is(TT_JsComputedPropertyName)) 2414 return prec::Assignment; 2415 if (Current->is(TT_LambdaArrow)) 2416 return prec::Comma; 2417 if (Current->is(TT_FatArrow)) 2418 return prec::Assignment; 2419 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 2420 (Current->is(tok::comment) && NextNonComment && 2421 NextNonComment->is(TT_SelectorName))) 2422 return 0; 2423 if (Current->is(TT_RangeBasedForLoopColon)) 2424 return prec::Comma; 2425 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2426 Current->is(Keywords.kw_instanceof)) 2427 return prec::Relational; 2428 if (Style.isJavaScript() && 2429 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 2430 return prec::Relational; 2431 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 2432 return Current->getPrecedence(); 2433 if (Current->isOneOf(tok::period, tok::arrow)) 2434 return PrecedenceArrowAndPeriod; 2435 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2436 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 2437 Keywords.kw_throws)) 2438 return 0; 2439 } 2440 return -1; 2441 } 2442 2443 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence, 2444 FormatToken *End = nullptr) { 2445 Start->FakeLParens.push_back(Precedence); 2446 if (Precedence > prec::Unknown) 2447 Start->StartsBinaryExpression = true; 2448 if (!End && Current) 2449 End = Current->getPreviousNonComment(); 2450 if (End) { 2451 ++End->FakeRParens; 2452 if (Precedence > prec::Unknown) 2453 End->EndsBinaryExpression = true; 2454 } 2455 } 2456 2457 /// Parse unary operator expressions and surround them with fake 2458 /// parentheses if appropriate. 2459 void parseUnaryOperator() { 2460 llvm::SmallVector<FormatToken *, 2> Tokens; 2461 while (Current && Current->is(TT_UnaryOperator)) { 2462 Tokens.push_back(Current); 2463 next(); 2464 } 2465 parse(PrecedenceArrowAndPeriod); 2466 for (FormatToken *Token : llvm::reverse(Tokens)) 2467 // The actual precedence doesn't matter. 2468 addFakeParenthesis(Token, prec::Unknown); 2469 } 2470 2471 void parseConditionalExpr() { 2472 while (Current && Current->isTrailingComment()) 2473 next(); 2474 FormatToken *Start = Current; 2475 parse(prec::LogicalOr); 2476 if (!Current || !Current->is(tok::question)) 2477 return; 2478 next(); 2479 parse(prec::Assignment); 2480 if (!Current || Current->isNot(TT_ConditionalExpr)) 2481 return; 2482 next(); 2483 parse(prec::Assignment); 2484 addFakeParenthesis(Start, prec::Conditional); 2485 } 2486 2487 void next(bool SkipPastLeadingComments = true) { 2488 if (Current) 2489 Current = Current->Next; 2490 while (Current && 2491 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 2492 Current->isTrailingComment()) 2493 Current = Current->Next; 2494 } 2495 2496 const FormatStyle &Style; 2497 const AdditionalKeywords &Keywords; 2498 const AnnotatedLine &Line; 2499 FormatToken *Current; 2500 }; 2501 2502 } // end anonymous namespace 2503 2504 void TokenAnnotator::setCommentLineLevels( 2505 SmallVectorImpl<AnnotatedLine *> &Lines) { 2506 const AnnotatedLine *NextNonCommentLine = nullptr; 2507 for (AnnotatedLine *Line : llvm::reverse(Lines)) { 2508 assert(Line->First); 2509 bool CommentLine = true; 2510 for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) { 2511 if (!Tok->is(tok::comment)) { 2512 CommentLine = false; 2513 break; 2514 } 2515 } 2516 2517 // If the comment is currently aligned with the line immediately following 2518 // it, that's probably intentional and we should keep it. 2519 if (NextNonCommentLine && CommentLine && 2520 NextNonCommentLine->First->NewlinesBefore <= 1 && 2521 NextNonCommentLine->First->OriginalColumn == 2522 Line->First->OriginalColumn) { 2523 // Align comments for preprocessor lines with the # in column 0 if 2524 // preprocessor lines are not indented. Otherwise, align with the next 2525 // line. 2526 Line->Level = 2527 (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 2528 (NextNonCommentLine->Type == LT_PreprocessorDirective || 2529 NextNonCommentLine->Type == LT_ImportStatement)) 2530 ? 0 2531 : NextNonCommentLine->Level; 2532 } else { 2533 NextNonCommentLine = Line->First->isNot(tok::r_brace) ? Line : nullptr; 2534 } 2535 2536 setCommentLineLevels(Line->Children); 2537 } 2538 } 2539 2540 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 2541 unsigned Result = 0; 2542 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 2543 Result = std::max(Result, Tok->NestingLevel); 2544 return Result; 2545 } 2546 2547 void TokenAnnotator::annotate(AnnotatedLine &Line) { 2548 for (auto &Child : Line.Children) 2549 annotate(*Child); 2550 2551 AnnotatingParser Parser(Style, Line, Keywords); 2552 Line.Type = Parser.parseLine(); 2553 2554 // With very deep nesting, ExpressionParser uses lots of stack and the 2555 // formatting algorithm is very slow. We're not going to do a good job here 2556 // anyway - it's probably generated code being formatted by mistake. 2557 // Just skip the whole line. 2558 if (maxNestingDepth(Line) > 50) 2559 Line.Type = LT_Invalid; 2560 2561 if (Line.Type == LT_Invalid) 2562 return; 2563 2564 ExpressionParser ExprParser(Style, Keywords, Line); 2565 ExprParser.parse(); 2566 2567 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2568 Line.Type = LT_ObjCMethodDecl; 2569 else if (Line.startsWith(TT_ObjCDecl)) 2570 Line.Type = LT_ObjCDecl; 2571 else if (Line.startsWith(TT_ObjCProperty)) 2572 Line.Type = LT_ObjCProperty; 2573 2574 Line.First->SpacesRequiredBefore = 1; 2575 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 2576 } 2577 2578 // This function heuristically determines whether 'Current' starts the name of a 2579 // function declaration. 2580 static bool isFunctionDeclarationName(bool IsCpp, const FormatToken &Current, 2581 const AnnotatedLine &Line) { 2582 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 2583 for (; Next; Next = Next->Next) { 2584 if (Next->is(TT_OverloadedOperatorLParen)) 2585 return Next; 2586 if (Next->is(TT_OverloadedOperator)) 2587 continue; 2588 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 2589 // For 'new[]' and 'delete[]'. 2590 if (Next->Next && 2591 Next->Next->startsSequence(tok::l_square, tok::r_square)) 2592 Next = Next->Next->Next; 2593 continue; 2594 } 2595 if (Next->startsSequence(tok::l_square, tok::r_square)) { 2596 // For operator[](). 2597 Next = Next->Next; 2598 continue; 2599 } 2600 if ((Next->isSimpleTypeSpecifier() || Next->is(tok::identifier)) && 2601 Next->Next && Next->Next->isOneOf(tok::star, tok::amp, tok::ampamp)) { 2602 // For operator void*(), operator char*(), operator Foo*(). 2603 Next = Next->Next; 2604 continue; 2605 } 2606 if (Next->is(TT_TemplateOpener) && Next->MatchingParen) { 2607 Next = Next->MatchingParen; 2608 continue; 2609 } 2610 2611 break; 2612 } 2613 return nullptr; 2614 }; 2615 2616 // Find parentheses of parameter list. 2617 const FormatToken *Next = Current.Next; 2618 if (Current.is(tok::kw_operator)) { 2619 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 2620 return false; 2621 Next = skipOperatorName(Next); 2622 } else { 2623 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 2624 return false; 2625 for (; Next; Next = Next->Next) { 2626 if (Next->is(TT_TemplateOpener)) { 2627 Next = Next->MatchingParen; 2628 } else if (Next->is(tok::coloncolon)) { 2629 Next = Next->Next; 2630 if (!Next) 2631 return false; 2632 if (Next->is(tok::kw_operator)) { 2633 Next = skipOperatorName(Next->Next); 2634 break; 2635 } 2636 if (!Next->is(tok::identifier)) 2637 return false; 2638 } else if (Next->is(tok::l_paren)) { 2639 break; 2640 } else { 2641 return false; 2642 } 2643 } 2644 } 2645 2646 // Check whether parameter list can belong to a function declaration. 2647 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 2648 return false; 2649 // If the lines ends with "{", this is likely a function definition. 2650 if (Line.Last->is(tok::l_brace)) 2651 return true; 2652 if (Next->Next == Next->MatchingParen) 2653 return true; // Empty parentheses. 2654 // If there is an &/&& after the r_paren, this is likely a function. 2655 if (Next->MatchingParen->Next && 2656 Next->MatchingParen->Next->is(TT_PointerOrReference)) 2657 return true; 2658 2659 // Check for K&R C function definitions (and C++ function definitions with 2660 // unnamed parameters), e.g.: 2661 // int f(i) 2662 // { 2663 // return i + 1; 2664 // } 2665 // bool g(size_t = 0, bool b = false) 2666 // { 2667 // return !b; 2668 // } 2669 if (IsCpp && Next->Next && Next->Next->is(tok::identifier) && 2670 !Line.endsWith(tok::semi)) 2671 return true; 2672 2673 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 2674 Tok = Tok->Next) { 2675 if (Tok->is(TT_TypeDeclarationParen)) 2676 return true; 2677 if (Tok->isOneOf(tok::l_paren, TT_TemplateOpener) && Tok->MatchingParen) { 2678 Tok = Tok->MatchingParen; 2679 continue; 2680 } 2681 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 2682 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 2683 return true; 2684 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 2685 Tok->Tok.isLiteral()) 2686 return false; 2687 } 2688 return false; 2689 } 2690 2691 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 2692 assert(Line.MightBeFunctionDecl); 2693 2694 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 2695 Style.AlwaysBreakAfterReturnType == 2696 FormatStyle::RTBS_TopLevelDefinitions) && 2697 Line.Level > 0) 2698 return false; 2699 2700 switch (Style.AlwaysBreakAfterReturnType) { 2701 case FormatStyle::RTBS_None: 2702 return false; 2703 case FormatStyle::RTBS_All: 2704 case FormatStyle::RTBS_TopLevel: 2705 return true; 2706 case FormatStyle::RTBS_AllDefinitions: 2707 case FormatStyle::RTBS_TopLevelDefinitions: 2708 return Line.mightBeFunctionDefinition(); 2709 } 2710 2711 return false; 2712 } 2713 2714 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 2715 for (AnnotatedLine *ChildLine : Line.Children) 2716 calculateFormattingInformation(*ChildLine); 2717 2718 Line.First->TotalLength = 2719 Line.First->IsMultiline ? Style.ColumnLimit 2720 : Line.FirstStartColumn + Line.First->ColumnWidth; 2721 FormatToken *Current = Line.First->Next; 2722 bool InFunctionDecl = Line.MightBeFunctionDecl; 2723 bool AlignArrayOfStructures = 2724 (Style.AlignArrayOfStructures != FormatStyle::AIAS_None && 2725 Line.Type == LT_ArrayOfStructInitializer); 2726 if (AlignArrayOfStructures) 2727 calculateArrayInitializerColumnList(Line); 2728 2729 while (Current) { 2730 if (isFunctionDeclarationName(Style.isCpp(), *Current, Line)) 2731 Current->setType(TT_FunctionDeclarationName); 2732 const FormatToken *Prev = Current->Previous; 2733 if (Current->is(TT_LineComment)) { 2734 if (Prev->is(BK_BracedInit) && Prev->opensScope()) 2735 Current->SpacesRequiredBefore = 2736 (Style.Cpp11BracedListStyle && !Style.SpacesInParentheses) ? 0 : 1; 2737 else 2738 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2739 2740 // If we find a trailing comment, iterate backwards to determine whether 2741 // it seems to relate to a specific parameter. If so, break before that 2742 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2743 // to the previous line in: 2744 // SomeFunction(a, 2745 // b, // comment 2746 // c); 2747 if (!Current->HasUnescapedNewline) { 2748 for (FormatToken *Parameter = Current->Previous; Parameter; 2749 Parameter = Parameter->Previous) { 2750 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 2751 break; 2752 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 2753 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 2754 Parameter->HasUnescapedNewline) 2755 Parameter->MustBreakBefore = true; 2756 break; 2757 } 2758 } 2759 } 2760 } else if (Current->SpacesRequiredBefore == 0 && 2761 spaceRequiredBefore(Line, *Current)) { 2762 Current->SpacesRequiredBefore = 1; 2763 } 2764 2765 const auto &Children = Prev->Children; 2766 if (!Children.empty() && Children.back()->Last->is(TT_LineComment)) { 2767 Current->MustBreakBefore = true; 2768 } else { 2769 Current->MustBreakBefore = 2770 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 2771 if (!Current->MustBreakBefore && InFunctionDecl && 2772 Current->is(TT_FunctionDeclarationName)) 2773 Current->MustBreakBefore = mustBreakForReturnType(Line); 2774 } 2775 2776 Current->CanBreakBefore = 2777 Current->MustBreakBefore || canBreakBefore(Line, *Current); 2778 unsigned ChildSize = 0; 2779 if (Prev->Children.size() == 1) { 2780 FormatToken &LastOfChild = *Prev->Children[0]->Last; 2781 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 2782 : LastOfChild.TotalLength + 1; 2783 } 2784 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 2785 (Prev->Children.size() == 1 && 2786 Prev->Children[0]->First->MustBreakBefore) || 2787 Current->IsMultiline) 2788 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 2789 else 2790 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 2791 ChildSize + Current->SpacesRequiredBefore; 2792 2793 if (Current->is(TT_CtorInitializerColon)) 2794 InFunctionDecl = false; 2795 2796 // FIXME: Only calculate this if CanBreakBefore is true once static 2797 // initializers etc. are sorted out. 2798 // FIXME: Move magic numbers to a better place. 2799 2800 // Reduce penalty for aligning ObjC method arguments using the colon 2801 // alignment as this is the canonical way (still prefer fitting everything 2802 // into one line if possible). Trying to fit a whole expression into one 2803 // line should not force other line breaks (e.g. when ObjC method 2804 // expression is a part of other expression). 2805 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl); 2806 if (Style.Language == FormatStyle::LK_ObjC && 2807 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) { 2808 if (Current->ParameterIndex == 1) 2809 Current->SplitPenalty += 5 * Current->BindingStrength; 2810 } else { 2811 Current->SplitPenalty += 20 * Current->BindingStrength; 2812 } 2813 2814 Current = Current->Next; 2815 } 2816 2817 calculateUnbreakableTailLengths(Line); 2818 unsigned IndentLevel = Line.Level; 2819 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2820 if (Current->Role) 2821 Current->Role->precomputeFormattingInfos(Current); 2822 if (Current->MatchingParen && 2823 Current->MatchingParen->opensBlockOrBlockTypeList(Style) && 2824 IndentLevel > 0) 2825 --IndentLevel; 2826 Current->IndentLevel = IndentLevel; 2827 if (Current->opensBlockOrBlockTypeList(Style)) 2828 ++IndentLevel; 2829 } 2830 2831 LLVM_DEBUG({ printDebugInfo(Line); }); 2832 } 2833 2834 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2835 unsigned UnbreakableTailLength = 0; 2836 FormatToken *Current = Line.Last; 2837 while (Current) { 2838 Current->UnbreakableTailLength = UnbreakableTailLength; 2839 if (Current->CanBreakBefore || 2840 Current->isOneOf(tok::comment, tok::string_literal)) { 2841 UnbreakableTailLength = 0; 2842 } else { 2843 UnbreakableTailLength += 2844 Current->ColumnWidth + Current->SpacesRequiredBefore; 2845 } 2846 Current = Current->Previous; 2847 } 2848 } 2849 2850 void TokenAnnotator::calculateArrayInitializerColumnList(AnnotatedLine &Line) { 2851 if (Line.First == Line.Last) 2852 return; 2853 auto *CurrentToken = Line.First; 2854 CurrentToken->ArrayInitializerLineStart = true; 2855 unsigned Depth = 0; 2856 while (CurrentToken != nullptr && CurrentToken != Line.Last) { 2857 if (CurrentToken->is(tok::l_brace)) { 2858 CurrentToken->IsArrayInitializer = true; 2859 if (CurrentToken->Next != nullptr) 2860 CurrentToken->Next->MustBreakBefore = true; 2861 CurrentToken = 2862 calculateInitializerColumnList(Line, CurrentToken->Next, Depth + 1); 2863 } else { 2864 CurrentToken = CurrentToken->Next; 2865 } 2866 } 2867 } 2868 2869 FormatToken *TokenAnnotator::calculateInitializerColumnList( 2870 AnnotatedLine &Line, FormatToken *CurrentToken, unsigned Depth) { 2871 while (CurrentToken != nullptr && CurrentToken != Line.Last) { 2872 if (CurrentToken->is(tok::l_brace)) 2873 ++Depth; 2874 else if (CurrentToken->is(tok::r_brace)) 2875 --Depth; 2876 if (Depth == 2 && CurrentToken->isOneOf(tok::l_brace, tok::comma)) { 2877 CurrentToken = CurrentToken->Next; 2878 if (CurrentToken == nullptr) 2879 break; 2880 CurrentToken->StartsColumn = true; 2881 CurrentToken = CurrentToken->Previous; 2882 } 2883 CurrentToken = CurrentToken->Next; 2884 } 2885 return CurrentToken; 2886 } 2887 2888 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2889 const FormatToken &Tok, 2890 bool InFunctionDecl) { 2891 const FormatToken &Left = *Tok.Previous; 2892 const FormatToken &Right = Tok; 2893 2894 if (Left.is(tok::semi)) 2895 return 0; 2896 2897 if (Style.Language == FormatStyle::LK_Java) { 2898 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2899 return 1; 2900 if (Right.is(Keywords.kw_implements)) 2901 return 2; 2902 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2903 return 3; 2904 } else if (Style.isJavaScript()) { 2905 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2906 return 100; 2907 if (Left.is(TT_JsTypeColon)) 2908 return 35; 2909 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2910 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2911 return 100; 2912 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2913 if (Left.opensScope() && Right.closesScope()) 2914 return 200; 2915 } 2916 2917 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2918 return 1; 2919 if (Right.is(tok::l_square)) { 2920 if (Style.Language == FormatStyle::LK_Proto) 2921 return 1; 2922 if (Left.is(tok::r_square)) 2923 return 200; 2924 // Slightly prefer formatting local lambda definitions like functions. 2925 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2926 return 35; 2927 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2928 TT_ArrayInitializerLSquare, 2929 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) 2930 return 500; 2931 } 2932 2933 if (Left.is(tok::coloncolon) || 2934 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2935 return 500; 2936 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2937 Right.is(tok::kw_operator)) { 2938 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2939 return 3; 2940 if (Left.is(TT_StartOfName)) 2941 return 110; 2942 if (InFunctionDecl && Right.NestingLevel == 0) 2943 return Style.PenaltyReturnTypeOnItsOwnLine; 2944 return 200; 2945 } 2946 if (Right.is(TT_PointerOrReference)) 2947 return 190; 2948 if (Right.is(TT_LambdaArrow)) 2949 return 110; 2950 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2951 return 160; 2952 if (Left.is(TT_CastRParen)) 2953 return 100; 2954 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2955 return 5000; 2956 if (Left.is(tok::comment)) 2957 return 1000; 2958 2959 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2960 TT_CtorInitializerColon)) 2961 return 2; 2962 2963 if (Right.isMemberAccess()) { 2964 // Breaking before the "./->" of a chained call/member access is reasonably 2965 // cheap, as formatting those with one call per line is generally 2966 // desirable. In particular, it should be cheaper to break before the call 2967 // than it is to break inside a call's parameters, which could lead to weird 2968 // "hanging" indents. The exception is the very last "./->" to support this 2969 // frequent pattern: 2970 // 2971 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2972 // dddddddd); 2973 // 2974 // which might otherwise be blown up onto many lines. Here, clang-format 2975 // won't produce "hanging" indents anyway as there is no other trailing 2976 // call. 2977 // 2978 // Also apply higher penalty is not a call as that might lead to a wrapping 2979 // like: 2980 // 2981 // aaaaaaa 2982 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2983 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2984 ? 150 2985 : 35; 2986 } 2987 2988 if (Right.is(TT_TrailingAnnotation) && 2989 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2990 // Moving trailing annotations to the next line is fine for ObjC method 2991 // declarations. 2992 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2993 return 10; 2994 // Generally, breaking before a trailing annotation is bad unless it is 2995 // function-like. It seems to be especially preferable to keep standard 2996 // annotations (i.e. "const", "final" and "override") on the same line. 2997 // Use a slightly higher penalty after ")" so that annotations like 2998 // "const override" are kept together. 2999 bool is_short_annotation = Right.TokenText.size() < 10; 3000 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 3001 } 3002 3003 // In for-loops, prefer breaking at ',' and ';'. 3004 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 3005 return 4; 3006 3007 // In Objective-C method expressions, prefer breaking before "param:" over 3008 // breaking after it. 3009 if (Right.is(TT_SelectorName)) 3010 return 0; 3011 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 3012 return Line.MightBeFunctionDecl ? 50 : 500; 3013 3014 // In Objective-C type declarations, avoid breaking after the category's 3015 // open paren (we'll prefer breaking after the protocol list's opening 3016 // angle bracket, if present). 3017 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous && 3018 Left.Previous->isOneOf(tok::identifier, tok::greater)) 3019 return 500; 3020 3021 if (Left.is(tok::l_paren) && Style.PenaltyBreakOpenParenthesis != 0) 3022 return Style.PenaltyBreakOpenParenthesis; 3023 if (Left.is(tok::l_paren) && InFunctionDecl && 3024 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 3025 return 100; 3026 if (Left.is(tok::l_paren) && Left.Previous && 3027 (Left.Previous->is(tok::kw_for) || Left.Previous->isIf())) 3028 return 1000; 3029 if (Left.is(tok::equal) && InFunctionDecl) 3030 return 110; 3031 if (Right.is(tok::r_brace)) 3032 return 1; 3033 if (Left.is(TT_TemplateOpener)) 3034 return 100; 3035 if (Left.opensScope()) { 3036 // If we aren't aligning after opening parens/braces we can always break 3037 // here unless the style does not want us to place all arguments on the 3038 // next line. 3039 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign && 3040 (Left.ParameterCount <= 1 || Style.AllowAllArgumentsOnNextLine)) 3041 return 0; 3042 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle) 3043 return 19; 3044 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 3045 : 19; 3046 } 3047 if (Left.is(TT_JavaAnnotation)) 3048 return 50; 3049 3050 if (Left.is(TT_UnaryOperator)) 3051 return 60; 3052 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 3053 Left.Previous->isLabelString() && 3054 (Left.NextOperator || Left.OperatorIndex != 0)) 3055 return 50; 3056 if (Right.is(tok::plus) && Left.isLabelString() && 3057 (Right.NextOperator || Right.OperatorIndex != 0)) 3058 return 25; 3059 if (Left.is(tok::comma)) 3060 return 1; 3061 if (Right.is(tok::lessless) && Left.isLabelString() && 3062 (Right.NextOperator || Right.OperatorIndex != 1)) 3063 return 25; 3064 if (Right.is(tok::lessless)) { 3065 // Breaking at a << is really cheap. 3066 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 3067 // Slightly prefer to break before the first one in log-like statements. 3068 return 2; 3069 return 1; 3070 } 3071 if (Left.ClosesTemplateDeclaration) 3072 return Style.PenaltyBreakTemplateDeclaration; 3073 if (Left.ClosesRequiresClause) 3074 return 0; 3075 if (Left.is(TT_ConditionalExpr)) 3076 return prec::Conditional; 3077 prec::Level Level = Left.getPrecedence(); 3078 if (Level == prec::Unknown) 3079 Level = Right.getPrecedence(); 3080 if (Level == prec::Assignment) 3081 return Style.PenaltyBreakAssignment; 3082 if (Level != prec::Unknown) 3083 return Level; 3084 3085 return 3; 3086 } 3087 3088 bool TokenAnnotator::spaceRequiredBeforeParens(const FormatToken &Right) const { 3089 if (Style.SpaceBeforeParens == FormatStyle::SBPO_Always) 3090 return true; 3091 if (Right.is(TT_OverloadedOperatorLParen) && 3092 Style.SpaceBeforeParensOptions.AfterOverloadedOperator) 3093 return true; 3094 if (Style.SpaceBeforeParensOptions.BeforeNonEmptyParentheses && 3095 Right.ParameterCount > 0) 3096 return true; 3097 return false; 3098 } 3099 3100 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 3101 const FormatToken &Left, 3102 const FormatToken &Right) { 3103 if (Left.is(tok::kw_return) && 3104 !Right.isOneOf(tok::semi, tok::r_paren, tok::hashhash)) 3105 return true; 3106 if (Style.isJson() && Left.is(tok::string_literal) && Right.is(tok::colon)) 3107 return false; 3108 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 3109 return true; 3110 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 3111 Left.Tok.getObjCKeywordID() == tok::objc_property) 3112 return true; 3113 if (Right.is(tok::hashhash)) 3114 return Left.is(tok::hash); 3115 if (Left.isOneOf(tok::hashhash, tok::hash)) 3116 return Right.is(tok::hash); 3117 if ((Left.is(tok::l_paren) && Right.is(tok::r_paren)) || 3118 (Left.is(tok::l_brace) && Left.isNot(BK_Block) && 3119 Right.is(tok::r_brace) && Right.isNot(BK_Block))) 3120 return Style.SpaceInEmptyParentheses; 3121 if (Style.SpacesInConditionalStatement) { 3122 if (Left.is(tok::l_paren) && Left.Previous && 3123 isKeywordWithCondition(*Left.Previous)) 3124 return true; 3125 if (Right.is(tok::r_paren) && Right.MatchingParen && 3126 Right.MatchingParen->Previous && 3127 isKeywordWithCondition(*Right.MatchingParen->Previous)) 3128 return true; 3129 } 3130 3131 // auto{x} auto(x) 3132 if (Left.is(tok::kw_auto) && Right.isOneOf(tok::l_paren, tok::l_brace)) 3133 return false; 3134 3135 // operator co_await(x) 3136 if (Right.is(tok::l_paren) && Left.is(tok::kw_co_await) && Left.Previous && 3137 Left.Previous->is(tok::kw_operator)) 3138 return false; 3139 // co_await (x), co_yield (x), co_return (x) 3140 if (Left.isOneOf(tok::kw_co_await, tok::kw_co_yield, tok::kw_co_return) && 3141 !Right.isOneOf(tok::semi, tok::r_paren)) 3142 return true; 3143 3144 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 3145 return (Right.is(TT_CastRParen) || 3146 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 3147 ? Style.SpacesInCStyleCastParentheses 3148 : Style.SpacesInParentheses; 3149 if (Right.isOneOf(tok::semi, tok::comma)) 3150 return false; 3151 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) { 3152 bool IsLightweightGeneric = Right.MatchingParen && 3153 Right.MatchingParen->Next && 3154 Right.MatchingParen->Next->is(tok::colon); 3155 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList; 3156 } 3157 if (Right.is(tok::less) && Left.is(tok::kw_template)) 3158 return Style.SpaceAfterTemplateKeyword; 3159 if (Left.isOneOf(tok::exclaim, tok::tilde)) 3160 return false; 3161 if (Left.is(tok::at) && 3162 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 3163 tok::numeric_constant, tok::l_paren, tok::l_brace, 3164 tok::kw_true, tok::kw_false)) 3165 return false; 3166 if (Left.is(tok::colon)) 3167 return !Left.is(TT_ObjCMethodExpr); 3168 if (Left.is(tok::coloncolon)) 3169 return false; 3170 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 3171 if (Style.Language == FormatStyle::LK_TextProto || 3172 (Style.Language == FormatStyle::LK_Proto && 3173 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 3174 // Format empty list as `<>`. 3175 if (Left.is(tok::less) && Right.is(tok::greater)) 3176 return false; 3177 return !Style.Cpp11BracedListStyle; 3178 } 3179 return false; 3180 } 3181 if (Right.is(tok::ellipsis)) 3182 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 3183 Left.Previous->is(tok::kw_case)); 3184 if (Left.is(tok::l_square) && Right.is(tok::amp)) 3185 return Style.SpacesInSquareBrackets; 3186 if (Right.is(TT_PointerOrReference)) { 3187 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 3188 if (!Left.MatchingParen) 3189 return true; 3190 FormatToken *TokenBeforeMatchingParen = 3191 Left.MatchingParen->getPreviousNonComment(); 3192 if (!TokenBeforeMatchingParen || !Left.is(TT_TypeDeclarationParen)) 3193 return true; 3194 } 3195 // Add a space if the previous token is a pointer qualifier or the closing 3196 // parenthesis of __attribute__(()) expression and the style requires spaces 3197 // after pointer qualifiers. 3198 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_After || 3199 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 3200 (Left.is(TT_AttributeParen) || Left.canBePointerOrReferenceQualifier())) 3201 return true; 3202 if (Left.Tok.isLiteral()) 3203 return true; 3204 // for (auto a = 0, b = 0; const auto & c : {1, 2, 3}) 3205 if (Left.isTypeOrIdentifier() && Right.Next && Right.Next->Next && 3206 Right.Next->Next->is(TT_RangeBasedForLoopColon)) 3207 return getTokenPointerOrReferenceAlignment(Right) != 3208 FormatStyle::PAS_Left; 3209 return !Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 3210 (getTokenPointerOrReferenceAlignment(Right) != 3211 FormatStyle::PAS_Left || 3212 (Line.IsMultiVariableDeclStmt && 3213 (Left.NestingLevel == 0 || 3214 (Left.NestingLevel == 1 && startsWithInitStatement(Line))))); 3215 } 3216 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 3217 (!Left.is(TT_PointerOrReference) || 3218 (getTokenPointerOrReferenceAlignment(Left) != FormatStyle::PAS_Right && 3219 !Line.IsMultiVariableDeclStmt))) 3220 return true; 3221 if (Left.is(TT_PointerOrReference)) { 3222 // Add a space if the next token is a pointer qualifier and the style 3223 // requires spaces before pointer qualifiers. 3224 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Before || 3225 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 3226 Right.canBePointerOrReferenceQualifier()) 3227 return true; 3228 // & 1 3229 if (Right.Tok.isLiteral()) 3230 return true; 3231 // & /* comment 3232 if (Right.is(TT_BlockComment)) 3233 return true; 3234 // foo() -> const Bar * override/final 3235 if (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 3236 !Right.is(TT_StartOfName)) 3237 return true; 3238 // & { 3239 if (Right.is(tok::l_brace) && Right.is(BK_Block)) 3240 return true; 3241 // for (auto a = 0, b = 0; const auto& c : {1, 2, 3}) 3242 if (Left.Previous && Left.Previous->isTypeOrIdentifier() && Right.Next && 3243 Right.Next->is(TT_RangeBasedForLoopColon)) 3244 return getTokenPointerOrReferenceAlignment(Left) != 3245 FormatStyle::PAS_Right; 3246 if (Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 3247 tok::l_paren)) 3248 return false; 3249 if (getTokenPointerOrReferenceAlignment(Left) == FormatStyle::PAS_Right) 3250 return false; 3251 // FIXME: Setting IsMultiVariableDeclStmt for the whole line is error-prone, 3252 // because it does not take into account nested scopes like lambdas. 3253 // In multi-variable declaration statements, attach */& to the variable 3254 // independently of the style. However, avoid doing it if we are in a nested 3255 // scope, e.g. lambda. We still need to special-case statements with 3256 // initializers. 3257 if (Line.IsMultiVariableDeclStmt && 3258 (Left.NestingLevel == Line.First->NestingLevel || 3259 ((Left.NestingLevel == Line.First->NestingLevel + 1) && 3260 startsWithInitStatement(Line)))) 3261 return false; 3262 return Left.Previous && !Left.Previous->isOneOf( 3263 tok::l_paren, tok::coloncolon, tok::l_square); 3264 } 3265 // Ensure right pointer alignment with ellipsis e.g. int *...P 3266 if (Left.is(tok::ellipsis) && Left.Previous && 3267 Left.Previous->isOneOf(tok::star, tok::amp, tok::ampamp)) 3268 return Style.PointerAlignment != FormatStyle::PAS_Right; 3269 3270 if (Right.is(tok::star) && Left.is(tok::l_paren)) 3271 return false; 3272 if (Left.is(tok::star) && Right.isOneOf(tok::star, tok::amp, tok::ampamp)) 3273 return false; 3274 if (Right.isOneOf(tok::star, tok::amp, tok::ampamp)) { 3275 const FormatToken *Previous = &Left; 3276 while (Previous && !Previous->is(tok::kw_operator)) { 3277 if (Previous->is(tok::identifier) || Previous->isSimpleTypeSpecifier()) { 3278 Previous = Previous->getPreviousNonComment(); 3279 continue; 3280 } 3281 if (Previous->is(TT_TemplateCloser) && Previous->MatchingParen) { 3282 Previous = Previous->MatchingParen->getPreviousNonComment(); 3283 continue; 3284 } 3285 if (Previous->is(tok::coloncolon)) { 3286 Previous = Previous->getPreviousNonComment(); 3287 continue; 3288 } 3289 break; 3290 } 3291 // Space between the type and the * in: 3292 // operator void*() 3293 // operator char*() 3294 // operator void const*() 3295 // operator void volatile*() 3296 // operator /*comment*/ const char*() 3297 // operator volatile /*comment*/ char*() 3298 // operator Foo*() 3299 // operator C<T>*() 3300 // operator std::Foo*() 3301 // operator C<T>::D<U>*() 3302 // dependent on PointerAlignment style. 3303 if (Previous) { 3304 if (Previous->endsSequence(tok::kw_operator)) 3305 return Style.PointerAlignment != FormatStyle::PAS_Left; 3306 if (Previous->is(tok::kw_const) || Previous->is(tok::kw_volatile)) 3307 return (Style.PointerAlignment != FormatStyle::PAS_Left) || 3308 (Style.SpaceAroundPointerQualifiers == 3309 FormatStyle::SAPQ_After) || 3310 (Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both); 3311 } 3312 } 3313 const auto SpaceRequiredForArrayInitializerLSquare = 3314 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 3315 return Style.SpacesInContainerLiterals || 3316 ((Style.Language == FormatStyle::LK_Proto || 3317 Style.Language == FormatStyle::LK_TextProto) && 3318 !Style.Cpp11BracedListStyle && 3319 LSquareTok.endsSequence(tok::l_square, tok::colon, 3320 TT_SelectorName)); 3321 }; 3322 if (Left.is(tok::l_square)) 3323 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 3324 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 3325 (Left.isOneOf(TT_ArraySubscriptLSquare, TT_StructuredBindingLSquare, 3326 TT_LambdaLSquare) && 3327 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 3328 if (Right.is(tok::r_square)) 3329 return Right.MatchingParen && 3330 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 3331 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 3332 Style)) || 3333 (Style.SpacesInSquareBrackets && 3334 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 3335 TT_StructuredBindingLSquare, 3336 TT_LambdaLSquare)) || 3337 Right.MatchingParen->is(TT_AttributeParen)); 3338 if (Right.is(tok::l_square) && 3339 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 3340 TT_DesignatedInitializerLSquare, 3341 TT_StructuredBindingLSquare, TT_AttributeSquare) && 3342 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral) && 3343 !(!Left.is(tok::r_square) && Style.SpaceBeforeSquareBrackets && 3344 Right.is(TT_ArraySubscriptLSquare))) 3345 return false; 3346 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 3347 return !Left.Children.empty(); // No spaces in "{}". 3348 if ((Left.is(tok::l_brace) && Left.isNot(BK_Block)) || 3349 (Right.is(tok::r_brace) && Right.MatchingParen && 3350 Right.MatchingParen->isNot(BK_Block))) 3351 return Style.Cpp11BracedListStyle ? Style.SpacesInParentheses : true; 3352 if (Left.is(TT_BlockComment)) 3353 // No whitespace in x(/*foo=*/1), except for JavaScript. 3354 return Style.isJavaScript() || !Left.TokenText.endswith("=*/"); 3355 3356 // Space between template and attribute. 3357 // e.g. template <typename T> [[nodiscard]] ... 3358 if (Left.is(TT_TemplateCloser) && Right.is(TT_AttributeSquare)) 3359 return true; 3360 // Space before parentheses common for all languages 3361 if (Right.is(tok::l_paren)) { 3362 if (Left.is(TT_TemplateCloser) && Right.isNot(TT_FunctionTypeLParen)) 3363 return spaceRequiredBeforeParens(Right); 3364 if (Left.isOneOf(TT_RequiresClause, TT_RequiresClauseInARequiresExpression)) 3365 return Style.SpaceBeforeParensOptions.AfterRequiresInClause || 3366 spaceRequiredBeforeParens(Right); 3367 if (Left.is(TT_RequiresExpression)) 3368 return Style.SpaceBeforeParensOptions.AfterRequiresInExpression || 3369 spaceRequiredBeforeParens(Right); 3370 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 3371 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) 3372 return true; 3373 if (Left.is(TT_ForEachMacro)) 3374 return Style.SpaceBeforeParensOptions.AfterForeachMacros || 3375 spaceRequiredBeforeParens(Right); 3376 if (Left.is(TT_IfMacro)) 3377 return Style.SpaceBeforeParensOptions.AfterIfMacros || 3378 spaceRequiredBeforeParens(Right); 3379 if (Line.Type == LT_ObjCDecl) 3380 return true; 3381 if (Left.is(tok::semi)) 3382 return true; 3383 if (Left.isOneOf(tok::pp_elif, tok::kw_for, tok::kw_while, tok::kw_switch, 3384 tok::kw_case, TT_ForEachMacro, TT_ObjCForIn)) 3385 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3386 spaceRequiredBeforeParens(Right); 3387 if (Left.isIf(Line.Type != LT_PreprocessorDirective)) 3388 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3389 spaceRequiredBeforeParens(Right); 3390 3391 // TODO add Operator overloading specific Options to 3392 // SpaceBeforeParensOptions 3393 if (Right.is(TT_OverloadedOperatorLParen)) 3394 return spaceRequiredBeforeParens(Right); 3395 // Function declaration or definition 3396 if (Line.MightBeFunctionDecl && (Left.is(TT_FunctionDeclarationName))) { 3397 if (Line.mightBeFunctionDefinition()) 3398 return Style.SpaceBeforeParensOptions.AfterFunctionDefinitionName || 3399 spaceRequiredBeforeParens(Right); 3400 else 3401 return Style.SpaceBeforeParensOptions.AfterFunctionDeclarationName || 3402 spaceRequiredBeforeParens(Right); 3403 } 3404 // Lambda 3405 if (Line.Type != LT_PreprocessorDirective && Left.is(tok::r_square) && 3406 Left.MatchingParen && Left.MatchingParen->is(TT_LambdaLSquare)) 3407 return Style.SpaceBeforeParensOptions.AfterFunctionDefinitionName || 3408 spaceRequiredBeforeParens(Right); 3409 if (!Left.Previous || Left.Previous->isNot(tok::period)) { 3410 if (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch)) 3411 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3412 spaceRequiredBeforeParens(Right); 3413 if (Left.isOneOf(tok::kw_new, tok::kw_delete)) 3414 return ((!Line.MightBeFunctionDecl || !Left.Previous) && 3415 Style.SpaceBeforeParens != FormatStyle::SBPO_Never) || 3416 spaceRequiredBeforeParens(Right); 3417 3418 if (Left.is(tok::r_square) && Left.MatchingParen && 3419 Left.MatchingParen->Previous && 3420 Left.MatchingParen->Previous->is(tok::kw_delete)) 3421 return (Style.SpaceBeforeParens != FormatStyle::SBPO_Never) || 3422 spaceRequiredBeforeParens(Right); 3423 } 3424 if (Line.Type != LT_PreprocessorDirective && 3425 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 3426 Left.is(tok::r_paren) || Left.isSimpleTypeSpecifier())) 3427 return spaceRequiredBeforeParens(Right); 3428 return false; 3429 } 3430 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 3431 return false; 3432 if (Right.is(TT_UnaryOperator)) 3433 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 3434 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 3435 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 3436 tok::r_paren) || 3437 Left.isSimpleTypeSpecifier()) && 3438 Right.is(tok::l_brace) && Right.getNextNonComment() && 3439 Right.isNot(BK_Block)) 3440 return false; 3441 if (Left.is(tok::period) || Right.is(tok::period)) 3442 return false; 3443 // u#str, U#str, L#str, u8#str 3444 // uR#str, UR#str, LR#str, u8R#str 3445 if (Right.is(tok::hash) && Left.is(tok::identifier) && 3446 (Left.TokenText == "L" || Left.TokenText == "u" || 3447 Left.TokenText == "U" || Left.TokenText == "u8" || 3448 Left.TokenText == "LR" || Left.TokenText == "uR" || 3449 Left.TokenText == "UR" || Left.TokenText == "u8R")) 3450 return false; 3451 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 3452 Left.MatchingParen->Previous && 3453 (Left.MatchingParen->Previous->is(tok::period) || 3454 Left.MatchingParen->Previous->is(tok::coloncolon))) 3455 // Java call to generic function with explicit type: 3456 // A.<B<C<...>>>DoSomething(); 3457 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference. 3458 return false; 3459 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 3460 return false; 3461 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at)) 3462 // Objective-C dictionary literal -> no space after opening brace. 3463 return false; 3464 if (Right.is(tok::r_brace) && Right.MatchingParen && 3465 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at)) 3466 // Objective-C dictionary literal -> no space before closing brace. 3467 return false; 3468 if (Right.getType() == TT_TrailingAnnotation && 3469 Right.isOneOf(tok::amp, tok::ampamp) && 3470 Left.isOneOf(tok::kw_const, tok::kw_volatile) && 3471 (!Right.Next || Right.Next->is(tok::semi))) 3472 // Match const and volatile ref-qualifiers without any additional 3473 // qualifiers such as 3474 // void Fn() const &; 3475 return getTokenReferenceAlignment(Right) != FormatStyle::PAS_Left; 3476 3477 return true; 3478 } 3479 3480 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 3481 const FormatToken &Right) { 3482 const FormatToken &Left = *Right.Previous; 3483 3484 // If the token is finalized don't touch it (as it could be in a 3485 // clang-format-off section). 3486 if (Left.Finalized) 3487 return Right.hasWhitespaceBefore(); 3488 3489 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 3490 return true; // Never ever merge two identifiers. 3491 3492 // Leave a space between * and /* to avoid C4138 `comment end` found outside 3493 // of comment. 3494 if (Left.is(tok::star) && Right.is(tok::comment)) 3495 return true; 3496 3497 if (Style.isCpp()) { 3498 // Space between import <iostream>. 3499 // or import .....; 3500 if (Left.is(Keywords.kw_import) && Right.isOneOf(tok::less, tok::ellipsis)) 3501 return true; 3502 // Space between `module :` and `import :`. 3503 if (Left.isOneOf(Keywords.kw_module, Keywords.kw_import) && 3504 Right.is(TT_ModulePartitionColon)) 3505 return true; 3506 // No space between import foo:bar but keep a space between import :bar; 3507 if (Left.is(tok::identifier) && Right.is(TT_ModulePartitionColon)) 3508 return false; 3509 // No space between :bar; 3510 if (Left.is(TT_ModulePartitionColon) && 3511 Right.isOneOf(tok::identifier, tok::kw_private)) 3512 return false; 3513 if (Left.is(tok::ellipsis) && Right.is(tok::identifier) && 3514 Line.First->is(Keywords.kw_import)) 3515 return false; 3516 // Space in __attribute__((attr)) ::type. 3517 if (Left.is(TT_AttributeParen) && Right.is(tok::coloncolon)) 3518 return true; 3519 3520 if (Left.is(tok::kw_operator)) 3521 return Right.is(tok::coloncolon); 3522 if (Right.is(tok::l_brace) && Right.is(BK_BracedInit) && 3523 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList) 3524 return true; 3525 if (Left.is(tok::less) && Left.is(TT_OverloadedOperator) && 3526 Right.is(TT_TemplateOpener)) 3527 return true; 3528 } else if (Style.Language == FormatStyle::LK_Proto || 3529 Style.Language == FormatStyle::LK_TextProto) { 3530 if (Right.is(tok::period) && 3531 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 3532 Keywords.kw_repeated, Keywords.kw_extend)) 3533 return true; 3534 if (Right.is(tok::l_paren) && 3535 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 3536 return true; 3537 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 3538 return true; 3539 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 3540 if (Left.is(tok::slash) || Right.is(tok::slash)) 3541 return false; 3542 if (Left.MatchingParen && 3543 Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 3544 Right.isOneOf(tok::l_brace, tok::less)) 3545 return !Style.Cpp11BracedListStyle; 3546 // A percent is probably part of a formatting specification, such as %lld. 3547 if (Left.is(tok::percent)) 3548 return false; 3549 // Preserve the existence of a space before a percent for cases like 0x%04x 3550 // and "%d %d" 3551 if (Left.is(tok::numeric_constant) && Right.is(tok::percent)) 3552 return Right.hasWhitespaceBefore(); 3553 } else if (Style.isJson()) { 3554 if (Right.is(tok::colon)) 3555 return false; 3556 } else if (Style.isCSharp()) { 3557 // Require spaces around '{' and before '}' unless they appear in 3558 // interpolated strings. Interpolated strings are merged into a single token 3559 // so cannot have spaces inserted by this function. 3560 3561 // No space between 'this' and '[' 3562 if (Left.is(tok::kw_this) && Right.is(tok::l_square)) 3563 return false; 3564 3565 // No space between 'new' and '(' 3566 if (Left.is(tok::kw_new) && Right.is(tok::l_paren)) 3567 return false; 3568 3569 // Space before { (including space within '{ {'). 3570 if (Right.is(tok::l_brace)) 3571 return true; 3572 3573 // Spaces inside braces. 3574 if (Left.is(tok::l_brace) && Right.isNot(tok::r_brace)) 3575 return true; 3576 3577 if (Left.isNot(tok::l_brace) && Right.is(tok::r_brace)) 3578 return true; 3579 3580 // Spaces around '=>'. 3581 if (Left.is(TT_FatArrow) || Right.is(TT_FatArrow)) 3582 return true; 3583 3584 // No spaces around attribute target colons 3585 if (Left.is(TT_AttributeColon) || Right.is(TT_AttributeColon)) 3586 return false; 3587 3588 // space between type and variable e.g. Dictionary<string,string> foo; 3589 if (Left.is(TT_TemplateCloser) && Right.is(TT_StartOfName)) 3590 return true; 3591 3592 // spaces inside square brackets. 3593 if (Left.is(tok::l_square) || Right.is(tok::r_square)) 3594 return Style.SpacesInSquareBrackets; 3595 3596 // No space before ? in nullable types. 3597 if (Right.is(TT_CSharpNullable)) 3598 return false; 3599 3600 // No space before null forgiving '!'. 3601 if (Right.is(TT_NonNullAssertion)) 3602 return false; 3603 3604 // No space between consecutive commas '[,,]'. 3605 if (Left.is(tok::comma) && Right.is(tok::comma)) 3606 return false; 3607 3608 // space after var in `var (key, value)` 3609 if (Left.is(Keywords.kw_var) && Right.is(tok::l_paren)) 3610 return true; 3611 3612 // space between keywords and paren e.g. "using (" 3613 if (Right.is(tok::l_paren)) 3614 if (Left.isOneOf(tok::kw_using, Keywords.kw_async, Keywords.kw_when, 3615 Keywords.kw_lock)) 3616 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3617 spaceRequiredBeforeParens(Right); 3618 3619 // space between method modifier and opening parenthesis of a tuple return 3620 // type 3621 if (Left.isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 3622 tok::kw_virtual, tok::kw_extern, tok::kw_static, 3623 Keywords.kw_internal, Keywords.kw_abstract, 3624 Keywords.kw_sealed, Keywords.kw_override, 3625 Keywords.kw_async, Keywords.kw_unsafe) && 3626 Right.is(tok::l_paren)) 3627 return true; 3628 } else if (Style.isJavaScript()) { 3629 if (Left.is(TT_FatArrow)) 3630 return true; 3631 // for await ( ... 3632 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 3633 Left.Previous->is(tok::kw_for)) 3634 return true; 3635 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 3636 Right.MatchingParen) { 3637 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 3638 // An async arrow function, for example: `x = async () => foo();`, 3639 // as opposed to calling a function called async: `x = async();` 3640 if (Next && Next->is(TT_FatArrow)) 3641 return true; 3642 } 3643 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 3644 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 3645 return false; 3646 // In tagged template literals ("html`bar baz`"), there is no space between 3647 // the tag identifier and the template string. 3648 if (Keywords.IsJavaScriptIdentifier(Left, 3649 /* AcceptIdentifierName= */ false) && 3650 Right.is(TT_TemplateString)) 3651 return false; 3652 if (Right.is(tok::star) && 3653 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 3654 return false; 3655 if (Right.isOneOf(tok::l_brace, tok::l_square) && 3656 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 3657 Keywords.kw_extends, Keywords.kw_implements)) 3658 return true; 3659 if (Right.is(tok::l_paren)) { 3660 // JS methods can use some keywords as names (e.g. `delete()`). 3661 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 3662 return false; 3663 // Valid JS method names can include keywords, e.g. `foo.delete()` or 3664 // `bar.instanceof()`. Recognize call positions by preceding period. 3665 if (Left.Previous && Left.Previous->is(tok::period) && 3666 Left.Tok.getIdentifierInfo()) 3667 return false; 3668 // Additional unary JavaScript operators that need a space after. 3669 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 3670 tok::kw_void)) 3671 return true; 3672 } 3673 // `foo as const;` casts into a const type. 3674 if (Left.endsSequence(tok::kw_const, Keywords.kw_as)) 3675 return false; 3676 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 3677 tok::kw_const) || 3678 // "of" is only a keyword if it appears after another identifier 3679 // (e.g. as "const x of y" in a for loop), or after a destructuring 3680 // operation (const [x, y] of z, const {a, b} of c). 3681 (Left.is(Keywords.kw_of) && Left.Previous && 3682 (Left.Previous->is(tok::identifier) || 3683 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 3684 (!Left.Previous || !Left.Previous->is(tok::period))) 3685 return true; 3686 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 3687 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 3688 return false; 3689 if (Left.is(Keywords.kw_as) && 3690 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 3691 return true; 3692 if (Left.is(tok::kw_default) && Left.Previous && 3693 Left.Previous->is(tok::kw_export)) 3694 return true; 3695 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 3696 return true; 3697 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 3698 return false; 3699 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 3700 return false; 3701 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 3702 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 3703 return false; 3704 if (Left.is(tok::ellipsis)) 3705 return false; 3706 if (Left.is(TT_TemplateCloser) && 3707 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 3708 Keywords.kw_implements, Keywords.kw_extends)) 3709 // Type assertions ('<type>expr') are not followed by whitespace. Other 3710 // locations that should have whitespace following are identified by the 3711 // above set of follower tokens. 3712 return false; 3713 if (Right.is(TT_NonNullAssertion)) 3714 return false; 3715 if (Left.is(TT_NonNullAssertion) && 3716 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 3717 return true; // "x! as string", "x! in y" 3718 } else if (Style.Language == FormatStyle::LK_Java) { 3719 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 3720 return true; 3721 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 3722 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3723 spaceRequiredBeforeParens(Right); 3724 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 3725 tok::kw_protected) || 3726 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 3727 Keywords.kw_native)) && 3728 Right.is(TT_TemplateOpener)) 3729 return true; 3730 } 3731 if (Left.is(TT_ImplicitStringLiteral)) 3732 return Right.hasWhitespaceBefore(); 3733 if (Line.Type == LT_ObjCMethodDecl) { 3734 if (Left.is(TT_ObjCMethodSpecifier)) 3735 return true; 3736 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right)) 3737 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a 3738 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class 3739 // method declaration. 3740 return false; 3741 } 3742 if (Line.Type == LT_ObjCProperty && 3743 (Right.is(tok::equal) || Left.is(tok::equal))) 3744 return false; 3745 3746 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 3747 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 3748 return true; 3749 if (Left.is(tok::comma) && !Right.is(TT_OverloadedOperatorLParen)) 3750 return true; 3751 if (Right.is(tok::comma)) 3752 return false; 3753 if (Right.is(TT_ObjCBlockLParen)) 3754 return true; 3755 if (Right.is(TT_CtorInitializerColon)) 3756 return Style.SpaceBeforeCtorInitializerColon; 3757 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 3758 return false; 3759 if (Right.is(TT_RangeBasedForLoopColon) && 3760 !Style.SpaceBeforeRangeBasedForLoopColon) 3761 return false; 3762 if (Left.is(TT_BitFieldColon)) 3763 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 3764 Style.BitFieldColonSpacing == FormatStyle::BFCS_After; 3765 if (Right.is(tok::colon)) { 3766 if (Line.First->isOneOf(tok::kw_default, tok::kw_case)) 3767 return Style.SpaceBeforeCaseColon; 3768 const FormatToken *Next = Right.getNextNonComment(); 3769 if (!Next || Next->is(tok::semi)) 3770 return false; 3771 if (Right.is(TT_ObjCMethodExpr)) 3772 return false; 3773 if (Left.is(tok::question)) 3774 return false; 3775 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 3776 return false; 3777 if (Right.is(TT_DictLiteral)) 3778 return Style.SpacesInContainerLiterals; 3779 if (Right.is(TT_AttributeColon)) 3780 return false; 3781 if (Right.is(TT_CSharpNamedArgumentColon)) 3782 return false; 3783 if (Right.is(TT_BitFieldColon)) 3784 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 3785 Style.BitFieldColonSpacing == FormatStyle::BFCS_Before; 3786 return true; 3787 } 3788 // Do not merge "- -" into "--". 3789 if ((Left.isOneOf(tok::minus, tok::minusminus) && 3790 Right.isOneOf(tok::minus, tok::minusminus)) || 3791 (Left.isOneOf(tok::plus, tok::plusplus) && 3792 Right.isOneOf(tok::plus, tok::plusplus))) 3793 return true; 3794 if (Left.is(TT_UnaryOperator)) { 3795 if (!Right.is(tok::l_paren)) { 3796 // The alternative operators for ~ and ! are "compl" and "not". 3797 // If they are used instead, we do not want to combine them with 3798 // the token to the right, unless that is a left paren. 3799 if (Left.is(tok::exclaim) && Left.TokenText == "not") 3800 return true; 3801 if (Left.is(tok::tilde) && Left.TokenText == "compl") 3802 return true; 3803 // Lambda captures allow for a lone &, so "&]" needs to be properly 3804 // handled. 3805 if (Left.is(tok::amp) && Right.is(tok::r_square)) 3806 return Style.SpacesInSquareBrackets; 3807 } 3808 return (Style.SpaceAfterLogicalNot && Left.is(tok::exclaim)) || 3809 Right.is(TT_BinaryOperator); 3810 } 3811 3812 // If the next token is a binary operator or a selector name, we have 3813 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 3814 if (Left.is(TT_CastRParen)) 3815 return Style.SpaceAfterCStyleCast || 3816 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 3817 3818 auto ShouldAddSpacesInAngles = [this, &Right]() { 3819 if (this->Style.SpacesInAngles == FormatStyle::SIAS_Always) 3820 return true; 3821 if (this->Style.SpacesInAngles == FormatStyle::SIAS_Leave) 3822 return Right.hasWhitespaceBefore(); 3823 return false; 3824 }; 3825 3826 if (Left.is(tok::greater) && Right.is(tok::greater)) { 3827 if (Style.Language == FormatStyle::LK_TextProto || 3828 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 3829 return !Style.Cpp11BracedListStyle; 3830 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 3831 ((Style.Standard < FormatStyle::LS_Cpp11) || 3832 ShouldAddSpacesInAngles()); 3833 } 3834 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 3835 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 3836 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 3837 return false; 3838 if (!Style.SpaceBeforeAssignmentOperators && Left.isNot(TT_TemplateCloser) && 3839 Right.getPrecedence() == prec::Assignment) 3840 return false; 3841 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) && 3842 (Left.is(tok::identifier) || Left.is(tok::kw_this))) 3843 return false; 3844 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 3845 // Generally don't remove existing spaces between an identifier and "::". 3846 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 3847 // this turns out to be too lenient, add analysis of the identifier itself. 3848 return Right.hasWhitespaceBefore(); 3849 if (Right.is(tok::coloncolon) && 3850 !Left.isOneOf(tok::l_brace, tok::comment, tok::l_paren)) 3851 // Put a space between < and :: in vector< ::std::string > 3852 return (Left.is(TT_TemplateOpener) && 3853 ((Style.Standard < FormatStyle::LS_Cpp11) || 3854 ShouldAddSpacesInAngles())) || 3855 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 3856 tok::kw___super, TT_TemplateOpener, 3857 TT_TemplateCloser)) || 3858 (Left.is(tok::l_paren) && Style.SpacesInParentheses); 3859 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 3860 return ShouldAddSpacesInAngles(); 3861 // Space before TT_StructuredBindingLSquare. 3862 if (Right.is(TT_StructuredBindingLSquare)) 3863 return !Left.isOneOf(tok::amp, tok::ampamp) || 3864 getTokenReferenceAlignment(Left) != FormatStyle::PAS_Right; 3865 // Space before & or && following a TT_StructuredBindingLSquare. 3866 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 3867 Right.isOneOf(tok::amp, tok::ampamp)) 3868 return getTokenReferenceAlignment(Right) != FormatStyle::PAS_Left; 3869 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 3870 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 3871 !Right.is(tok::r_paren))) 3872 return true; 3873 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 3874 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 3875 return false; 3876 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 3877 Line.startsWith(tok::hash)) 3878 return true; 3879 if (Right.is(TT_TrailingUnaryOperator)) 3880 return false; 3881 if (Left.is(TT_RegexLiteral)) 3882 return false; 3883 return spaceRequiredBetween(Line, Left, Right); 3884 } 3885 3886 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 3887 static bool isAllmanBrace(const FormatToken &Tok) { 3888 return Tok.is(tok::l_brace) && Tok.is(BK_Block) && 3889 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_LambdaLBrace, TT_DictLiteral); 3890 } 3891 3892 // Returns 'true' if 'Tok' is a function argument. 3893 static bool IsFunctionArgument(const FormatToken &Tok) { 3894 return Tok.MatchingParen && Tok.MatchingParen->Next && 3895 Tok.MatchingParen->Next->isOneOf(tok::comma, tok::r_paren); 3896 } 3897 3898 static bool 3899 isItAnEmptyLambdaAllowed(const FormatToken &Tok, 3900 FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3901 return Tok.Children.empty() && ShortLambdaOption != FormatStyle::SLS_None; 3902 } 3903 3904 static bool isAllmanLambdaBrace(const FormatToken &Tok) { 3905 return Tok.is(tok::l_brace) && Tok.is(BK_Block) && 3906 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral); 3907 } 3908 3909 // Returns the first token on the line that is not a comment. 3910 static const FormatToken *getFirstNonComment(const AnnotatedLine &Line) { 3911 const FormatToken *Next = Line.First; 3912 if (!Next) 3913 return Next; 3914 if (Next->is(tok::comment)) 3915 Next = Next->getNextNonComment(); 3916 return Next; 3917 } 3918 3919 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 3920 const FormatToken &Right) { 3921 const FormatToken &Left = *Right.Previous; 3922 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 3923 return true; 3924 3925 if (Style.isCSharp()) { 3926 if (Left.is(TT_FatArrow) && Right.is(tok::l_brace) && 3927 Style.BraceWrapping.AfterFunction) 3928 return true; 3929 if (Right.is(TT_CSharpNamedArgumentColon) || 3930 Left.is(TT_CSharpNamedArgumentColon)) 3931 return false; 3932 if (Right.is(TT_CSharpGenericTypeConstraint)) 3933 return true; 3934 if (Right.Next && Right.Next->is(TT_FatArrow) && 3935 (Right.is(tok::numeric_constant) || 3936 (Right.is(tok::identifier) && Right.TokenText == "_"))) 3937 return true; 3938 3939 // Break after C# [...] and before public/protected/private/internal. 3940 if (Left.is(TT_AttributeSquare) && Left.is(tok::r_square) && 3941 (Right.isAccessSpecifier(/*ColonRequired=*/false) || 3942 Right.is(Keywords.kw_internal))) 3943 return true; 3944 // Break between ] and [ but only when there are really 2 attributes. 3945 if (Left.is(TT_AttributeSquare) && Right.is(TT_AttributeSquare) && 3946 Left.is(tok::r_square) && Right.is(tok::l_square)) 3947 return true; 3948 3949 } else if (Style.isJavaScript()) { 3950 // FIXME: This might apply to other languages and token kinds. 3951 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 3952 Left.Previous->is(tok::string_literal)) 3953 return true; 3954 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 3955 Left.Previous && Left.Previous->is(tok::equal) && 3956 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 3957 tok::kw_const) && 3958 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 3959 // above. 3960 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 3961 // Object literals on the top level of a file are treated as "enum-style". 3962 // Each key/value pair is put on a separate line, instead of bin-packing. 3963 return true; 3964 if (Left.is(tok::l_brace) && Line.Level == 0 && 3965 (Line.startsWith(tok::kw_enum) || 3966 Line.startsWith(tok::kw_const, tok::kw_enum) || 3967 Line.startsWith(tok::kw_export, tok::kw_enum) || 3968 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 3969 // JavaScript top-level enum key/value pairs are put on separate lines 3970 // instead of bin-packing. 3971 return true; 3972 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && Left.Previous && 3973 Left.Previous->is(TT_FatArrow)) { 3974 // JS arrow function (=> {...}). 3975 switch (Style.AllowShortLambdasOnASingleLine) { 3976 case FormatStyle::SLS_All: 3977 return false; 3978 case FormatStyle::SLS_None: 3979 return true; 3980 case FormatStyle::SLS_Empty: 3981 return !Left.Children.empty(); 3982 case FormatStyle::SLS_Inline: 3983 // allow one-lining inline (e.g. in function call args) and empty arrow 3984 // functions. 3985 return (Left.NestingLevel == 0 && Line.Level == 0) && 3986 !Left.Children.empty(); 3987 } 3988 llvm_unreachable("Unknown FormatStyle::ShortLambdaStyle enum"); 3989 } 3990 3991 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 3992 !Left.Children.empty()) 3993 // Support AllowShortFunctionsOnASingleLine for JavaScript. 3994 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 3995 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 3996 (Left.NestingLevel == 0 && Line.Level == 0 && 3997 Style.AllowShortFunctionsOnASingleLine & 3998 FormatStyle::SFS_InlineOnly); 3999 } else if (Style.Language == FormatStyle::LK_Java) { 4000 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 4001 Right.Next->is(tok::string_literal)) 4002 return true; 4003 } else if (Style.Language == FormatStyle::LK_Cpp || 4004 Style.Language == FormatStyle::LK_ObjC || 4005 Style.Language == FormatStyle::LK_Proto || 4006 Style.Language == FormatStyle::LK_TableGen || 4007 Style.Language == FormatStyle::LK_TextProto) { 4008 if (Left.isStringLiteral() && Right.isStringLiteral()) 4009 return true; 4010 } 4011 4012 // Basic JSON newline processing. 4013 if (Style.isJson()) { 4014 // Always break after a JSON record opener. 4015 // { 4016 // } 4017 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace)) 4018 return true; 4019 // Always break after a JSON array opener. 4020 // [ 4021 // ] 4022 if (Left.is(TT_ArrayInitializerLSquare) && Left.is(tok::l_square) && 4023 !Right.is(tok::r_square)) 4024 return true; 4025 // Always break after successive entries. 4026 // 1, 4027 // 2 4028 if (Left.is(tok::comma)) 4029 return true; 4030 } 4031 4032 // If the last token before a '}', ']', or ')' is a comma or a trailing 4033 // comment, the intention is to insert a line break after it in order to make 4034 // shuffling around entries easier. Import statements, especially in 4035 // JavaScript, can be an exception to this rule. 4036 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 4037 const FormatToken *BeforeClosingBrace = nullptr; 4038 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 4039 (Style.isJavaScript() && Left.is(tok::l_paren))) && 4040 Left.isNot(BK_Block) && Left.MatchingParen) 4041 BeforeClosingBrace = Left.MatchingParen->Previous; 4042 else if (Right.MatchingParen && 4043 (Right.MatchingParen->isOneOf(tok::l_brace, 4044 TT_ArrayInitializerLSquare) || 4045 (Style.isJavaScript() && Right.MatchingParen->is(tok::l_paren)))) 4046 BeforeClosingBrace = &Left; 4047 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 4048 BeforeClosingBrace->isTrailingComment())) 4049 return true; 4050 } 4051 4052 if (Right.is(tok::comment)) 4053 return Left.isNot(BK_BracedInit) && Left.isNot(TT_CtorInitializerColon) && 4054 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 4055 if (Left.isTrailingComment()) 4056 return true; 4057 if (Left.IsUnterminatedLiteral) 4058 return true; 4059 if (Right.is(tok::lessless) && Right.Next && Left.is(tok::string_literal) && 4060 Right.Next->is(tok::string_literal)) 4061 return true; 4062 if (Right.is(TT_RequiresClause)) { 4063 switch (Style.RequiresClausePosition) { 4064 case FormatStyle::RCPS_OwnLine: 4065 case FormatStyle::RCPS_WithFollowing: 4066 return true; 4067 default: 4068 break; 4069 } 4070 } 4071 // Can break after template<> declaration 4072 if (Left.ClosesTemplateDeclaration && Left.MatchingParen && 4073 Left.MatchingParen->NestingLevel == 0) { 4074 // Put concepts on the next line e.g. 4075 // template<typename T> 4076 // concept ... 4077 if (Right.is(tok::kw_concept)) 4078 return Style.BreakBeforeConceptDeclarations == FormatStyle::BBCDS_Always; 4079 return Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes; 4080 } 4081 if (Left.ClosesRequiresClause && Right.isNot(tok::semi)) { 4082 switch (Style.RequiresClausePosition) { 4083 case FormatStyle::RCPS_OwnLine: 4084 case FormatStyle::RCPS_WithPreceding: 4085 return true; 4086 default: 4087 break; 4088 } 4089 } 4090 if (Style.PackConstructorInitializers == FormatStyle::PCIS_Never) { 4091 if (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeColon && 4092 (Left.is(TT_CtorInitializerComma) || Right.is(TT_CtorInitializerColon))) 4093 return true; 4094 4095 if (Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon && 4096 Left.isOneOf(TT_CtorInitializerColon, TT_CtorInitializerComma)) 4097 return true; 4098 } 4099 if (Style.PackConstructorInitializers < FormatStyle::PCIS_CurrentLine && 4100 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 4101 Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) 4102 return true; 4103 // Break only if we have multiple inheritance. 4104 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 4105 Right.is(TT_InheritanceComma)) 4106 return true; 4107 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterComma && 4108 Left.is(TT_InheritanceComma)) 4109 return true; 4110 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 4111 // Multiline raw string literals are special wrt. line breaks. The author 4112 // has made a deliberate choice and might have aligned the contents of the 4113 // string literal accordingly. Thus, we try keep existing line breaks. 4114 return Right.IsMultiline && Right.NewlinesBefore > 0; 4115 if ((Left.is(tok::l_brace) || (Left.is(tok::less) && Left.Previous && 4116 Left.Previous->is(tok::equal))) && 4117 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 4118 // Don't put enums or option definitions onto single lines in protocol 4119 // buffers. 4120 return true; 4121 } 4122 if (Right.is(TT_InlineASMBrace)) 4123 return Right.HasUnescapedNewline; 4124 4125 if (isAllmanBrace(Left) || isAllmanBrace(Right)) { 4126 auto FirstNonComment = getFirstNonComment(Line); 4127 bool AccessSpecifier = 4128 FirstNonComment && 4129 FirstNonComment->isOneOf(Keywords.kw_internal, tok::kw_public, 4130 tok::kw_private, tok::kw_protected); 4131 4132 if (Style.BraceWrapping.AfterEnum) { 4133 if (Line.startsWith(tok::kw_enum) || 4134 Line.startsWith(tok::kw_typedef, tok::kw_enum)) 4135 return true; 4136 // Ensure BraceWrapping for `public enum A {`. 4137 if (AccessSpecifier && FirstNonComment->Next && 4138 FirstNonComment->Next->is(tok::kw_enum)) 4139 return true; 4140 } 4141 4142 // Ensure BraceWrapping for `public interface A {`. 4143 if (Style.BraceWrapping.AfterClass && 4144 ((AccessSpecifier && FirstNonComment->Next && 4145 FirstNonComment->Next->is(Keywords.kw_interface)) || 4146 Line.startsWith(Keywords.kw_interface))) 4147 return true; 4148 4149 return (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 4150 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 4151 } 4152 4153 if (Left.is(TT_ObjCBlockLBrace) && 4154 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 4155 return true; 4156 4157 // Ensure wrapping after __attribute__((XX)) and @interface etc. 4158 if (Left.is(TT_AttributeParen) && Right.is(TT_ObjCDecl)) 4159 return true; 4160 4161 if (Left.is(TT_LambdaLBrace)) { 4162 if (IsFunctionArgument(Left) && 4163 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline) 4164 return false; 4165 4166 if (Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_None || 4167 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline || 4168 (!Left.Children.empty() && 4169 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Empty)) 4170 return true; 4171 } 4172 4173 if (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace) && 4174 Left.isOneOf(tok::star, tok::amp, tok::ampamp, TT_TemplateCloser)) 4175 return true; 4176 4177 // Put multiple Java annotation on a new line. 4178 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 4179 Left.is(TT_LeadingJavaAnnotation) && 4180 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 4181 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 4182 return true; 4183 4184 if (Right.is(TT_ProtoExtensionLSquare)) 4185 return true; 4186 4187 // In text proto instances if a submessage contains at least 2 entries and at 4188 // least one of them is a submessage, like A { ... B { ... } ... }, 4189 // put all of the entries of A on separate lines by forcing the selector of 4190 // the submessage B to be put on a newline. 4191 // 4192 // Example: these can stay on one line: 4193 // a { scalar_1: 1 scalar_2: 2 } 4194 // a { b { key: value } } 4195 // 4196 // and these entries need to be on a new line even if putting them all in one 4197 // line is under the column limit: 4198 // a { 4199 // scalar: 1 4200 // b { key: value } 4201 // } 4202 // 4203 // We enforce this by breaking before a submessage field that has previous 4204 // siblings, *and* breaking before a field that follows a submessage field. 4205 // 4206 // Be careful to exclude the case [proto.ext] { ... } since the `]` is 4207 // the TT_SelectorName there, but we don't want to break inside the brackets. 4208 // 4209 // Another edge case is @submessage { key: value }, which is a common 4210 // substitution placeholder. In this case we want to keep `@` and `submessage` 4211 // together. 4212 // 4213 // We ensure elsewhere that extensions are always on their own line. 4214 if ((Style.Language == FormatStyle::LK_Proto || 4215 Style.Language == FormatStyle::LK_TextProto) && 4216 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) { 4217 // Keep `@submessage` together in: 4218 // @submessage { key: value } 4219 if (Left.is(tok::at)) 4220 return false; 4221 // Look for the scope opener after selector in cases like: 4222 // selector { ... 4223 // selector: { ... 4224 // selector: @base { ... 4225 FormatToken *LBrace = Right.Next; 4226 if (LBrace && LBrace->is(tok::colon)) { 4227 LBrace = LBrace->Next; 4228 if (LBrace && LBrace->is(tok::at)) { 4229 LBrace = LBrace->Next; 4230 if (LBrace) 4231 LBrace = LBrace->Next; 4232 } 4233 } 4234 if (LBrace && 4235 // The scope opener is one of {, [, <: 4236 // selector { ... } 4237 // selector [ ... ] 4238 // selector < ... > 4239 // 4240 // In case of selector { ... }, the l_brace is TT_DictLiteral. 4241 // In case of an empty selector {}, the l_brace is not TT_DictLiteral, 4242 // so we check for immediately following r_brace. 4243 ((LBrace->is(tok::l_brace) && 4244 (LBrace->is(TT_DictLiteral) || 4245 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) || 4246 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) { 4247 // If Left.ParameterCount is 0, then this submessage entry is not the 4248 // first in its parent submessage, and we want to break before this entry. 4249 // If Left.ParameterCount is greater than 0, then its parent submessage 4250 // might contain 1 or more entries and we want to break before this entry 4251 // if it contains at least 2 entries. We deal with this case later by 4252 // detecting and breaking before the next entry in the parent submessage. 4253 if (Left.ParameterCount == 0) 4254 return true; 4255 // However, if this submessage is the first entry in its parent 4256 // submessage, Left.ParameterCount might be 1 in some cases. 4257 // We deal with this case later by detecting an entry 4258 // following a closing paren of this submessage. 4259 } 4260 4261 // If this is an entry immediately following a submessage, it will be 4262 // preceded by a closing paren of that submessage, like in: 4263 // left---. .---right 4264 // v v 4265 // sub: { ... } key: value 4266 // If there was a comment between `}` an `key` above, then `key` would be 4267 // put on a new line anyways. 4268 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square)) 4269 return true; 4270 } 4271 4272 // Deal with lambda arguments in C++ - we want consistent line breaks whether 4273 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced 4274 // as aggressive line breaks are placed when the lambda is not the last arg. 4275 if ((Style.Language == FormatStyle::LK_Cpp || 4276 Style.Language == FormatStyle::LK_ObjC) && 4277 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 && 4278 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) { 4279 // Multiple lambdas in the same function call force line breaks. 4280 if (Left.BlockParameterCount > 1) 4281 return true; 4282 4283 // A lambda followed by another arg forces a line break. 4284 if (!Left.Role) 4285 return false; 4286 auto Comma = Left.Role->lastComma(); 4287 if (!Comma) 4288 return false; 4289 auto Next = Comma->getNextNonComment(); 4290 if (!Next) 4291 return false; 4292 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret)) 4293 return true; 4294 } 4295 4296 return false; 4297 } 4298 4299 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 4300 const FormatToken &Right) { 4301 const FormatToken &Left = *Right.Previous; 4302 // Language-specific stuff. 4303 if (Style.isCSharp()) { 4304 if (Left.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon) || 4305 Right.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon)) 4306 return false; 4307 // Only break after commas for generic type constraints. 4308 if (Line.First->is(TT_CSharpGenericTypeConstraint)) 4309 return Left.is(TT_CSharpGenericTypeConstraintComma); 4310 // Keep nullable operators attached to their identifiers. 4311 if (Right.is(TT_CSharpNullable)) 4312 return false; 4313 } else if (Style.Language == FormatStyle::LK_Java) { 4314 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 4315 Keywords.kw_implements)) 4316 return false; 4317 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 4318 Keywords.kw_implements)) 4319 return true; 4320 } else if (Style.isJavaScript()) { 4321 const FormatToken *NonComment = Right.getPreviousNonComment(); 4322 if (NonComment && 4323 NonComment->isOneOf( 4324 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 4325 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 4326 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 4327 Keywords.kw_readonly, Keywords.kw_override, Keywords.kw_abstract, 4328 Keywords.kw_get, Keywords.kw_set, Keywords.kw_async, 4329 Keywords.kw_await)) 4330 return false; // Otherwise automatic semicolon insertion would trigger. 4331 if (Right.NestingLevel == 0 && 4332 (Left.Tok.getIdentifierInfo() || 4333 Left.isOneOf(tok::r_square, tok::r_paren)) && 4334 Right.isOneOf(tok::l_square, tok::l_paren)) 4335 return false; // Otherwise automatic semicolon insertion would trigger. 4336 if (NonComment && NonComment->is(tok::identifier) && 4337 NonComment->TokenText == "asserts") 4338 return false; 4339 if (Left.is(TT_FatArrow) && Right.is(tok::l_brace)) 4340 return false; 4341 if (Left.is(TT_JsTypeColon)) 4342 return true; 4343 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;"). 4344 if (Left.is(tok::exclaim) && Right.is(tok::colon)) 4345 return false; 4346 // Look for is type annotations like: 4347 // function f(): a is B { ... } 4348 // Do not break before is in these cases. 4349 if (Right.is(Keywords.kw_is)) { 4350 const FormatToken *Next = Right.getNextNonComment(); 4351 // If `is` is followed by a colon, it's likely that it's a dict key, so 4352 // ignore it for this check. 4353 // For example this is common in Polymer: 4354 // Polymer({ 4355 // is: 'name', 4356 // ... 4357 // }); 4358 if (!Next || !Next->is(tok::colon)) 4359 return false; 4360 } 4361 if (Left.is(Keywords.kw_in)) 4362 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 4363 if (Right.is(Keywords.kw_in)) 4364 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 4365 if (Right.is(Keywords.kw_as)) 4366 return false; // must not break before as in 'x as type' casts 4367 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) { 4368 // extends and infer can appear as keywords in conditional types: 4369 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types 4370 // do not break before them, as the expressions are subject to ASI. 4371 return false; 4372 } 4373 if (Left.is(Keywords.kw_as)) 4374 return true; 4375 if (Left.is(TT_NonNullAssertion)) 4376 return true; 4377 if (Left.is(Keywords.kw_declare) && 4378 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 4379 Keywords.kw_function, tok::kw_class, tok::kw_enum, 4380 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 4381 Keywords.kw_let, tok::kw_const)) 4382 // See grammar for 'declare' statements at: 4383 // https://github.com/Microsoft/TypeScript/blob/main/doc/spec-ARCHIVED.md#A.10 4384 return false; 4385 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 4386 Right.isOneOf(tok::identifier, tok::string_literal)) 4387 return false; // must not break in "module foo { ...}" 4388 if (Right.is(TT_TemplateString) && Right.closesScope()) 4389 return false; 4390 // Don't split tagged template literal so there is a break between the tag 4391 // identifier and template string. 4392 if (Left.is(tok::identifier) && Right.is(TT_TemplateString)) 4393 return false; 4394 if (Left.is(TT_TemplateString) && Left.opensScope()) 4395 return true; 4396 } 4397 4398 if (Left.is(tok::at)) 4399 return false; 4400 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 4401 return false; 4402 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 4403 return !Right.is(tok::l_paren); 4404 if (Right.is(TT_PointerOrReference)) 4405 return Line.IsMultiVariableDeclStmt || 4406 (getTokenPointerOrReferenceAlignment(Right) == 4407 FormatStyle::PAS_Right && 4408 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 4409 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 4410 Right.is(tok::kw_operator)) 4411 return true; 4412 if (Left.is(TT_PointerOrReference)) 4413 return false; 4414 if (Right.isTrailingComment()) 4415 // We rely on MustBreakBefore being set correctly here as we should not 4416 // change the "binding" behavior of a comment. 4417 // The first comment in a braced lists is always interpreted as belonging to 4418 // the first list element. Otherwise, it should be placed outside of the 4419 // list. 4420 return Left.is(BK_BracedInit) || 4421 (Left.is(TT_CtorInitializerColon) && 4422 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 4423 if (Left.is(tok::question) && Right.is(tok::colon)) 4424 return false; 4425 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 4426 return Style.BreakBeforeTernaryOperators; 4427 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 4428 return !Style.BreakBeforeTernaryOperators; 4429 if (Left.is(TT_InheritanceColon)) 4430 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon; 4431 if (Right.is(TT_InheritanceColon)) 4432 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon; 4433 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 4434 Left.isNot(TT_SelectorName)) 4435 return true; 4436 4437 if (Right.is(tok::colon) && 4438 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 4439 return false; 4440 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 4441 if (Style.Language == FormatStyle::LK_Proto || 4442 Style.Language == FormatStyle::LK_TextProto) { 4443 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 4444 return false; 4445 // Prevent cases like: 4446 // 4447 // submessage: 4448 // { key: valueeeeeeeeeeee } 4449 // 4450 // when the snippet does not fit into one line. 4451 // Prefer: 4452 // 4453 // submessage: { 4454 // key: valueeeeeeeeeeee 4455 // } 4456 // 4457 // instead, even if it is longer by one line. 4458 // 4459 // Note that this allows allows the "{" to go over the column limit 4460 // when the column limit is just between ":" and "{", but that does 4461 // not happen too often and alternative formattings in this case are 4462 // not much better. 4463 // 4464 // The code covers the cases: 4465 // 4466 // submessage: { ... } 4467 // submessage: < ... > 4468 // repeated: [ ... ] 4469 if (((Right.is(tok::l_brace) || Right.is(tok::less)) && 4470 Right.is(TT_DictLiteral)) || 4471 Right.is(TT_ArrayInitializerLSquare)) 4472 return false; 4473 } 4474 return true; 4475 } 4476 if (Right.is(tok::r_square) && Right.MatchingParen && 4477 Right.MatchingParen->is(TT_ProtoExtensionLSquare)) 4478 return false; 4479 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 4480 Right.Next->is(TT_ObjCMethodExpr))) 4481 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 4482 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 4483 return true; 4484 if (Right.is(tok::kw_concept)) 4485 return Style.BreakBeforeConceptDeclarations != FormatStyle::BBCDS_Never; 4486 if (Right.is(TT_RequiresClause)) 4487 return true; 4488 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 4489 return true; 4490 if (Left.ClosesRequiresClause) 4491 return true; 4492 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 4493 TT_OverloadedOperator)) 4494 return false; 4495 if (Left.is(TT_RangeBasedForLoopColon)) 4496 return true; 4497 if (Right.is(TT_RangeBasedForLoopColon)) 4498 return false; 4499 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 4500 return true; 4501 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 4502 (Left.is(tok::less) && Right.is(tok::less))) 4503 return false; 4504 if (Right.is(TT_BinaryOperator) && 4505 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 4506 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 4507 Right.getPrecedence() != prec::Assignment)) 4508 return true; 4509 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 4510 Left.is(tok::kw_operator)) 4511 return false; 4512 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 4513 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 4514 return false; 4515 if (Left.is(tok::equal) && Right.is(tok::l_brace) && 4516 !Style.Cpp11BracedListStyle) 4517 return false; 4518 if (Left.is(tok::l_paren) && 4519 Left.isOneOf(TT_AttributeParen, TT_TypeDeclarationParen)) 4520 return false; 4521 if (Left.is(tok::l_paren) && Left.Previous && 4522 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 4523 return false; 4524 if (Right.is(TT_ImplicitStringLiteral)) 4525 return false; 4526 4527 if (Right.is(TT_TemplateCloser)) 4528 return false; 4529 if (Right.is(tok::r_square) && Right.MatchingParen && 4530 Right.MatchingParen->is(TT_LambdaLSquare)) 4531 return false; 4532 4533 // We only break before r_brace if there was a corresponding break before 4534 // the l_brace, which is tracked by BreakBeforeClosingBrace. 4535 if (Right.is(tok::r_brace)) 4536 return Right.MatchingParen && Right.MatchingParen->is(BK_Block); 4537 4538 // We only break before r_paren if we're in a block indented context. 4539 if (Right.is(tok::r_paren)) { 4540 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent) 4541 return Right.MatchingParen && 4542 !(Right.MatchingParen->Previous && 4543 (Right.MatchingParen->Previous->is(tok::kw_for) || 4544 Right.MatchingParen->Previous->isIf())); 4545 4546 return false; 4547 } 4548 4549 // Allow breaking after a trailing annotation, e.g. after a method 4550 // declaration. 4551 if (Left.is(TT_TrailingAnnotation)) 4552 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 4553 tok::less, tok::coloncolon); 4554 4555 if (Right.is(tok::kw___attribute) || 4556 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) 4557 return !Left.is(TT_AttributeSquare); 4558 4559 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 4560 return true; 4561 4562 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 4563 return true; 4564 4565 if (Left.is(TT_CtorInitializerColon)) 4566 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 4567 if (Right.is(TT_CtorInitializerColon)) 4568 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 4569 if (Left.is(TT_CtorInitializerComma) && 4570 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 4571 return false; 4572 if (Right.is(TT_CtorInitializerComma) && 4573 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 4574 return true; 4575 if (Left.is(TT_InheritanceComma) && 4576 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 4577 return false; 4578 if (Right.is(TT_InheritanceComma) && 4579 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 4580 return true; 4581 if (Left.is(TT_ArrayInitializerLSquare)) 4582 return true; 4583 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 4584 return true; 4585 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 4586 !Left.isOneOf(tok::arrowstar, tok::lessless) && 4587 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 4588 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 4589 Left.getPrecedence() == prec::Assignment)) 4590 return true; 4591 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 4592 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) 4593 return false; 4594 4595 auto ShortLambdaOption = Style.AllowShortLambdasOnASingleLine; 4596 if (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace)) { 4597 if (isAllmanLambdaBrace(Left)) 4598 return !isItAnEmptyLambdaAllowed(Left, ShortLambdaOption); 4599 if (isAllmanLambdaBrace(Right)) 4600 return !isItAnEmptyLambdaAllowed(Right, ShortLambdaOption); 4601 } 4602 4603 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 4604 tok::kw_class, tok::kw_struct, tok::comment) || 4605 Right.isMemberAccess() || 4606 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 4607 tok::colon, tok::l_square, tok::at) || 4608 (Left.is(tok::r_paren) && 4609 Right.isOneOf(tok::identifier, tok::kw_const)) || 4610 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 4611 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 4612 } 4613 4614 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 4615 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 4616 const FormatToken *Tok = Line.First; 4617 while (Tok) { 4618 llvm::errs() << " M=" << Tok->MustBreakBefore 4619 << " C=" << Tok->CanBreakBefore 4620 << " T=" << getTokenTypeName(Tok->getType()) 4621 << " S=" << Tok->SpacesRequiredBefore 4622 << " F=" << Tok->Finalized << " B=" << Tok->BlockParameterCount 4623 << " BK=" << Tok->getBlockKind() << " P=" << Tok->SplitPenalty 4624 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 4625 << " PPK=" << Tok->getPackingKind() << " FakeLParens="; 4626 for (prec::Level LParen : Tok->FakeLParens) 4627 llvm::errs() << LParen << "/"; 4628 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 4629 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo(); 4630 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 4631 if (!Tok->Next) 4632 assert(Tok == Line.Last); 4633 Tok = Tok->Next; 4634 } 4635 llvm::errs() << "----\n"; 4636 } 4637 4638 FormatStyle::PointerAlignmentStyle 4639 TokenAnnotator::getTokenReferenceAlignment(const FormatToken &Reference) { 4640 assert(Reference.isOneOf(tok::amp, tok::ampamp)); 4641 switch (Style.ReferenceAlignment) { 4642 case FormatStyle::RAS_Pointer: 4643 return Style.PointerAlignment; 4644 case FormatStyle::RAS_Left: 4645 return FormatStyle::PAS_Left; 4646 case FormatStyle::RAS_Right: 4647 return FormatStyle::PAS_Right; 4648 case FormatStyle::RAS_Middle: 4649 return FormatStyle::PAS_Middle; 4650 } 4651 assert(0); //"Unhandled value of ReferenceAlignment" 4652 return Style.PointerAlignment; 4653 } 4654 4655 FormatStyle::PointerAlignmentStyle 4656 TokenAnnotator::getTokenPointerOrReferenceAlignment( 4657 const FormatToken &PointerOrReference) { 4658 if (PointerOrReference.isOneOf(tok::amp, tok::ampamp)) { 4659 switch (Style.ReferenceAlignment) { 4660 case FormatStyle::RAS_Pointer: 4661 return Style.PointerAlignment; 4662 case FormatStyle::RAS_Left: 4663 return FormatStyle::PAS_Left; 4664 case FormatStyle::RAS_Right: 4665 return FormatStyle::PAS_Right; 4666 case FormatStyle::RAS_Middle: 4667 return FormatStyle::PAS_Middle; 4668 } 4669 } 4670 assert(PointerOrReference.is(tok::star)); 4671 return Style.PointerAlignment; 4672 } 4673 4674 } // namespace format 4675 } // namespace clang 4676