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