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 token can be used as an identifier in 30 /// an Objective-C \c \@selector, \c false otherwise. 31 /// 32 /// Because getFormattingLangOpts() always lexes source code as 33 /// Objective-C++, C++ keywords like \c new and \c delete are 34 /// lexed as tok::kw_*, not tok::identifier, even for Objective-C. 35 /// 36 /// For Objective-C and Objective-C++, both identifiers and keywords 37 /// are valid inside @selector(...) (or a macro which 38 /// invokes @selector(...)). So, we allow treat any identifier or 39 /// keyword as a potential Objective-C selector component. 40 static bool canBeObjCSelectorComponent(const FormatToken &Tok) { 41 return Tok.Tok.getIdentifierInfo() != nullptr; 42 } 43 44 /// With `Left` being '(', check if we're at either `[...](` or 45 /// `[...]<...>(`, where the [ opens a lambda capture list. 46 static bool isLambdaParameterList(const FormatToken *Left) { 47 // Skip <...> if present. 48 if (Left->Previous && Left->Previous->is(tok::greater) && 49 Left->Previous->MatchingParen && 50 Left->Previous->MatchingParen->is(TT_TemplateOpener)) 51 Left = Left->Previous->MatchingParen; 52 53 // Check for `[...]`. 54 return Left->Previous && Left->Previous->is(tok::r_square) && 55 Left->Previous->MatchingParen && 56 Left->Previous->MatchingParen->is(TT_LambdaLSquare); 57 } 58 59 /// Returns \c true if the token is followed by a boolean condition, \c false 60 /// otherwise. 61 static bool isKeywordWithCondition(const FormatToken &Tok) { 62 return Tok.isOneOf(tok::kw_if, tok::kw_for, tok::kw_while, tok::kw_switch, 63 tok::kw_constexpr, tok::kw_catch); 64 } 65 66 /// A parser that gathers additional information about tokens. 67 /// 68 /// The \c TokenAnnotator tries to match parenthesis and square brakets and 69 /// store a parenthesis levels. It also tries to resolve matching "<" and ">" 70 /// into template parameter lists. 71 class AnnotatingParser { 72 public: 73 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line, 74 const AdditionalKeywords &Keywords) 75 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false), 76 Keywords(Keywords) { 77 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false)); 78 resetTokenMetadata(); 79 } 80 81 private: 82 bool parseAngle() { 83 if (!CurrentToken || !CurrentToken->Previous) 84 return false; 85 if (NonTemplateLess.count(CurrentToken->Previous)) 86 return false; 87 88 const FormatToken &Previous = *CurrentToken->Previous; // The '<'. 89 if (Previous.Previous) { 90 if (Previous.Previous->Tok.isLiteral()) 91 return false; 92 if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 && 93 (!Previous.Previous->MatchingParen || 94 !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen))) 95 return false; 96 } 97 98 FormatToken *Left = CurrentToken->Previous; 99 Left->ParentBracket = Contexts.back().ContextKind; 100 ScopedContextCreator ContextCreator(*this, tok::less, 12); 101 102 // If this angle is in the context of an expression, we need to be more 103 // hesitant to detect it as opening template parameters. 104 bool InExprContext = Contexts.back().IsExpression; 105 106 Contexts.back().IsExpression = false; 107 // If there's a template keyword before the opening angle bracket, this is a 108 // template parameter, not an argument. 109 Contexts.back().InTemplateArgument = 110 Left->Previous && Left->Previous->Tok.isNot(tok::kw_template); 111 112 if (Style.Language == FormatStyle::LK_Java && 113 CurrentToken->is(tok::question)) 114 next(); 115 116 while (CurrentToken) { 117 if (CurrentToken->is(tok::greater)) { 118 // Try to do a better job at looking for ">>" within the condition of 119 // a statement. Conservatively insert spaces between consecutive ">" 120 // tokens to prevent splitting right bitshift operators and potentially 121 // altering program semantics. This check is overly conservative and 122 // will prevent spaces from being inserted in select nested template 123 // parameter cases, but should not alter program semantics. 124 if (CurrentToken->Next && CurrentToken->Next->is(tok::greater) && 125 Left->ParentBracket != tok::less && 126 (isKeywordWithCondition(*Line.First) || 127 CurrentToken->getStartOfNonWhitespace() == 128 CurrentToken->Next->getStartOfNonWhitespace().getLocWithOffset( 129 -1))) 130 return false; 131 Left->MatchingParen = CurrentToken; 132 CurrentToken->MatchingParen = Left; 133 // In TT_Proto, we must distignuish between: 134 // map<key, value> 135 // msg < item: data > 136 // msg: < item: data > 137 // In TT_TextProto, map<key, value> does not occur. 138 if (Style.Language == FormatStyle::LK_TextProto || 139 (Style.Language == FormatStyle::LK_Proto && Left->Previous && 140 Left->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) 141 CurrentToken->setType(TT_DictLiteral); 142 else 143 CurrentToken->setType(TT_TemplateCloser); 144 next(); 145 return true; 146 } 147 if (CurrentToken->is(tok::question) && 148 Style.Language == FormatStyle::LK_Java) { 149 next(); 150 continue; 151 } 152 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) || 153 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext && 154 !Style.isCSharp() && Style.Language != FormatStyle::LK_Proto && 155 Style.Language != FormatStyle::LK_TextProto)) 156 return false; 157 // If a && or || is found and interpreted as a binary operator, this set 158 // of angles is likely part of something like "a < b && c > d". If the 159 // angles are inside an expression, the ||/&& might also be a binary 160 // operator that was misinterpreted because we are parsing template 161 // parameters. 162 // FIXME: This is getting out of hand, write a decent parser. 163 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) && 164 CurrentToken->Previous->is(TT_BinaryOperator) && 165 Contexts[Contexts.size() - 2].IsExpression && 166 !Line.startsWith(tok::kw_template)) 167 return false; 168 updateParameterCount(Left, CurrentToken); 169 if (Style.Language == FormatStyle::LK_Proto) { 170 if (FormatToken *Previous = CurrentToken->getPreviousNonComment()) { 171 if (CurrentToken->is(tok::colon) || 172 (CurrentToken->isOneOf(tok::l_brace, tok::less) && 173 Previous->isNot(tok::colon))) 174 Previous->setType(TT_SelectorName); 175 } 176 } 177 if (!consumeToken()) 178 return false; 179 } 180 return false; 181 } 182 183 bool parseUntouchableParens() { 184 while (CurrentToken) { 185 CurrentToken->Finalized = true; 186 switch (CurrentToken->Tok.getKind()) { 187 case tok::l_paren: 188 next(); 189 if (!parseUntouchableParens()) 190 return false; 191 continue; 192 case tok::r_paren: 193 next(); 194 return true; 195 default: 196 // no-op 197 break; 198 } 199 next(); 200 } 201 return false; 202 } 203 204 bool parseParens(bool LookForDecls = false) { 205 if (!CurrentToken) 206 return false; 207 FormatToken *Left = CurrentToken->Previous; 208 assert(Left && "Unknown previous token"); 209 FormatToken *PrevNonComment = Left->getPreviousNonComment(); 210 Left->ParentBracket = Contexts.back().ContextKind; 211 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1); 212 213 // FIXME: This is a bit of a hack. Do better. 214 Contexts.back().ColonIsForRangeExpr = 215 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr; 216 217 if (Left->Previous && Left->Previous->is(TT_UntouchableMacroFunc)) { 218 Left->Finalized = true; 219 return parseUntouchableParens(); 220 } 221 222 bool StartsObjCMethodExpr = false; 223 if (FormatToken *MaybeSel = Left->Previous) { 224 // @selector( starts a selector. 225 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous && 226 MaybeSel->Previous->is(tok::at)) { 227 StartsObjCMethodExpr = true; 228 } 229 } 230 231 if (Left->is(TT_OverloadedOperatorLParen)) { 232 // Find the previous kw_operator token. 233 FormatToken *Prev = Left; 234 while (!Prev->is(tok::kw_operator)) { 235 Prev = Prev->Previous; 236 assert(Prev && "Expect a kw_operator prior to the OperatorLParen!"); 237 } 238 239 // If faced with "a.operator*(argument)" or "a->operator*(argument)", 240 // i.e. the operator is called as a member function, 241 // then the argument must be an expression. 242 bool OperatorCalledAsMemberFunction = 243 Prev->Previous && Prev->Previous->isOneOf(tok::period, tok::arrow); 244 Contexts.back().IsExpression = OperatorCalledAsMemberFunction; 245 } else if (Style.isJavaScript() && 246 (Line.startsWith(Keywords.kw_type, tok::identifier) || 247 Line.startsWith(tok::kw_export, Keywords.kw_type, 248 tok::identifier))) { 249 // type X = (...); 250 // export type X = (...); 251 Contexts.back().IsExpression = false; 252 } else if (Left->Previous && 253 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_while, 254 tok::l_paren, tok::comma) || 255 Left->Previous->isIf() || 256 Left->Previous->is(TT_BinaryOperator))) { 257 // static_assert, if and while usually contain expressions. 258 Contexts.back().IsExpression = true; 259 } else if (Style.isJavaScript() && Left->Previous && 260 (Left->Previous->is(Keywords.kw_function) || 261 (Left->Previous->endsSequence(tok::identifier, 262 Keywords.kw_function)))) { 263 // function(...) or function f(...) 264 Contexts.back().IsExpression = false; 265 } else if (Style.isJavaScript() && Left->Previous && 266 Left->Previous->is(TT_JsTypeColon)) { 267 // let x: (SomeType); 268 Contexts.back().IsExpression = false; 269 } else if (isLambdaParameterList(Left)) { 270 // This is a parameter list of a lambda expression. 271 Contexts.back().IsExpression = false; 272 } else if (Line.InPPDirective && 273 (!Left->Previous || !Left->Previous->is(tok::identifier))) { 274 Contexts.back().IsExpression = true; 275 } else if (Contexts[Contexts.size() - 2].CaretFound) { 276 // This is the parameter list of an ObjC block. 277 Contexts.back().IsExpression = false; 278 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) { 279 // The first argument to a foreach macro is a declaration. 280 Contexts.back().IsForEachMacro = true; 281 Contexts.back().IsExpression = false; 282 } else if (Left->Previous && Left->Previous->MatchingParen && 283 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) { 284 Contexts.back().IsExpression = false; 285 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) { 286 bool IsForOrCatch = 287 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch); 288 Contexts.back().IsExpression = !IsForOrCatch; 289 } 290 291 // Infer the role of the l_paren based on the previous token if we haven't 292 // detected one one yet. 293 if (PrevNonComment && Left->is(TT_Unknown)) { 294 if (PrevNonComment->is(tok::kw___attribute)) { 295 Left->setType(TT_AttributeParen); 296 } else if (PrevNonComment->isOneOf(TT_TypenameMacro, tok::kw_decltype, 297 tok::kw_typeof, tok::kw__Atomic, 298 tok::kw___underlying_type)) { 299 Left->setType(TT_TypeDeclarationParen); 300 // decltype() and typeof() usually contain expressions. 301 if (PrevNonComment->isOneOf(tok::kw_decltype, tok::kw_typeof)) 302 Contexts.back().IsExpression = true; 303 } 304 } 305 306 if (StartsObjCMethodExpr) { 307 Contexts.back().ColonIsObjCMethodExpr = true; 308 Left->setType(TT_ObjCMethodExpr); 309 } 310 311 // MightBeFunctionType and ProbablyFunctionType are used for 312 // function pointer and reference types as well as Objective-C 313 // block types: 314 // 315 // void (*FunctionPointer)(void); 316 // void (&FunctionReference)(void); 317 // void (&&FunctionReference)(void); 318 // void (^ObjCBlock)(void); 319 bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression; 320 bool ProbablyFunctionType = 321 CurrentToken->isOneOf(tok::star, tok::amp, tok::ampamp, tok::caret); 322 bool HasMultipleLines = false; 323 bool HasMultipleParametersOnALine = false; 324 bool MightBeObjCForRangeLoop = 325 Left->Previous && Left->Previous->is(tok::kw_for); 326 FormatToken *PossibleObjCForInToken = nullptr; 327 while (CurrentToken) { 328 // LookForDecls is set when "if (" has been seen. Check for 329 // 'identifier' '*' 'identifier' followed by not '=' -- this 330 // '*' has to be a binary operator but determineStarAmpUsage() will 331 // categorize it as an unary operator, so set the right type here. 332 if (LookForDecls && CurrentToken->Next) { 333 FormatToken *Prev = CurrentToken->getPreviousNonComment(); 334 if (Prev) { 335 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 336 FormatToken *Next = CurrentToken->Next; 337 if (PrevPrev && PrevPrev->is(tok::identifier) && 338 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) && 339 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) { 340 Prev->setType(TT_BinaryOperator); 341 LookForDecls = false; 342 } 343 } 344 } 345 346 if (CurrentToken->Previous->is(TT_PointerOrReference) && 347 CurrentToken->Previous->Previous->isOneOf(tok::l_paren, 348 tok::coloncolon)) 349 ProbablyFunctionType = true; 350 if (CurrentToken->is(tok::comma)) 351 MightBeFunctionType = false; 352 if (CurrentToken->Previous->is(TT_BinaryOperator)) 353 Contexts.back().IsExpression = true; 354 if (CurrentToken->is(tok::r_paren)) { 355 if (MightBeFunctionType && ProbablyFunctionType && CurrentToken->Next && 356 (CurrentToken->Next->is(tok::l_paren) || 357 (CurrentToken->Next->is(tok::l_square) && Line.MustBeDeclaration))) 358 Left->setType(Left->Next->is(tok::caret) ? TT_ObjCBlockLParen 359 : TT_FunctionTypeLParen); 360 Left->MatchingParen = CurrentToken; 361 CurrentToken->MatchingParen = Left; 362 363 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) && 364 Left->Previous && Left->Previous->is(tok::l_paren)) { 365 // Detect the case where macros are used to generate lambdas or 366 // function bodies, e.g.: 367 // auto my_lambda = MACRO((Type *type, int i) { .. body .. }); 368 for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) { 369 if (Tok->is(TT_BinaryOperator) && 370 Tok->isOneOf(tok::star, tok::amp, tok::ampamp)) 371 Tok->setType(TT_PointerOrReference); 372 } 373 } 374 375 if (StartsObjCMethodExpr) { 376 CurrentToken->setType(TT_ObjCMethodExpr); 377 if (Contexts.back().FirstObjCSelectorName) { 378 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 379 Contexts.back().LongestObjCSelectorName; 380 } 381 } 382 383 if (Left->is(TT_AttributeParen)) 384 CurrentToken->setType(TT_AttributeParen); 385 if (Left->is(TT_TypeDeclarationParen)) 386 CurrentToken->setType(TT_TypeDeclarationParen); 387 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation)) 388 CurrentToken->setType(TT_JavaAnnotation); 389 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation)) 390 CurrentToken->setType(TT_LeadingJavaAnnotation); 391 if (Left->Previous && Left->Previous->is(TT_AttributeSquare)) 392 CurrentToken->setType(TT_AttributeSquare); 393 394 if (!HasMultipleLines) 395 Left->setPackingKind(PPK_Inconclusive); 396 else if (HasMultipleParametersOnALine) 397 Left->setPackingKind(PPK_BinPacked); 398 else 399 Left->setPackingKind(PPK_OnePerLine); 400 401 next(); 402 return true; 403 } 404 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace)) 405 return false; 406 407 if (CurrentToken->is(tok::l_brace)) 408 Left->setType(TT_Unknown); // Not TT_ObjCBlockLParen 409 if (CurrentToken->is(tok::comma) && CurrentToken->Next && 410 !CurrentToken->Next->HasUnescapedNewline && 411 !CurrentToken->Next->isTrailingComment()) 412 HasMultipleParametersOnALine = true; 413 bool ProbablyFunctionTypeLParen = 414 (CurrentToken->is(tok::l_paren) && CurrentToken->Next && 415 CurrentToken->Next->isOneOf(tok::star, tok::amp, tok::caret)); 416 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) || 417 CurrentToken->Previous->isSimpleTypeSpecifier()) && 418 !(CurrentToken->is(tok::l_brace) || 419 (CurrentToken->is(tok::l_paren) && !ProbablyFunctionTypeLParen))) 420 Contexts.back().IsExpression = false; 421 if (CurrentToken->isOneOf(tok::semi, tok::colon)) { 422 MightBeObjCForRangeLoop = false; 423 if (PossibleObjCForInToken) { 424 PossibleObjCForInToken->setType(TT_Unknown); 425 PossibleObjCForInToken = nullptr; 426 } 427 } 428 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) { 429 PossibleObjCForInToken = CurrentToken; 430 PossibleObjCForInToken->setType(TT_ObjCForIn); 431 } 432 // When we discover a 'new', we set CanBeExpression to 'false' in order to 433 // parse the type correctly. Reset that after a comma. 434 if (CurrentToken->is(tok::comma)) 435 Contexts.back().CanBeExpression = true; 436 437 FormatToken *Tok = CurrentToken; 438 if (!consumeToken()) 439 return false; 440 updateParameterCount(Left, Tok); 441 if (CurrentToken && CurrentToken->HasUnescapedNewline) 442 HasMultipleLines = true; 443 } 444 return false; 445 } 446 447 bool isCSharpAttributeSpecifier(const FormatToken &Tok) { 448 if (!Style.isCSharp()) 449 return false; 450 451 // `identifier[i]` is not an attribute. 452 if (Tok.Previous && Tok.Previous->is(tok::identifier)) 453 return false; 454 455 // Chains of [] in `identifier[i][j][k]` are not attributes. 456 if (Tok.Previous && Tok.Previous->is(tok::r_square)) { 457 auto *MatchingParen = Tok.Previous->MatchingParen; 458 if (!MatchingParen || MatchingParen->is(TT_ArraySubscriptLSquare)) 459 return false; 460 } 461 462 const FormatToken *AttrTok = Tok.Next; 463 if (!AttrTok) 464 return false; 465 466 // Just an empty declaration e.g. string []. 467 if (AttrTok->is(tok::r_square)) 468 return false; 469 470 // Move along the tokens inbetween the '[' and ']' e.g. [STAThread]. 471 while (AttrTok && AttrTok->isNot(tok::r_square)) { 472 AttrTok = AttrTok->Next; 473 } 474 475 if (!AttrTok) 476 return false; 477 478 // Allow an attribute to be the only content of a file. 479 AttrTok = AttrTok->Next; 480 if (!AttrTok) 481 return true; 482 483 // Limit this to being an access modifier that follows. 484 if (AttrTok->isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 485 tok::comment, tok::kw_class, tok::kw_static, 486 tok::l_square, Keywords.kw_internal)) { 487 return true; 488 } 489 490 // incase its a [XXX] retval func(.... 491 if (AttrTok->Next && 492 AttrTok->Next->startsSequence(tok::identifier, tok::l_paren)) 493 return true; 494 495 return false; 496 } 497 498 bool isCpp11AttributeSpecifier(const FormatToken &Tok) { 499 if (!Style.isCpp() || !Tok.startsSequence(tok::l_square, tok::l_square)) 500 return false; 501 // The first square bracket is part of an ObjC array literal 502 if (Tok.Previous && Tok.Previous->is(tok::at)) { 503 return false; 504 } 505 const FormatToken *AttrTok = Tok.Next->Next; 506 if (!AttrTok) 507 return false; 508 // C++17 '[[using ns: foo, bar(baz, blech)]]' 509 // We assume nobody will name an ObjC variable 'using'. 510 if (AttrTok->startsSequence(tok::kw_using, tok::identifier, tok::colon)) 511 return true; 512 if (AttrTok->isNot(tok::identifier)) 513 return false; 514 while (AttrTok && !AttrTok->startsSequence(tok::r_square, tok::r_square)) { 515 // ObjC message send. We assume nobody will use : in a C++11 attribute 516 // specifier parameter, although this is technically valid: 517 // [[foo(:)]]. 518 if (AttrTok->is(tok::colon) || 519 AttrTok->startsSequence(tok::identifier, tok::identifier) || 520 AttrTok->startsSequence(tok::r_paren, tok::identifier)) 521 return false; 522 if (AttrTok->is(tok::ellipsis)) 523 return true; 524 AttrTok = AttrTok->Next; 525 } 526 return AttrTok && AttrTok->startsSequence(tok::r_square, tok::r_square); 527 } 528 529 bool parseSquare() { 530 if (!CurrentToken) 531 return false; 532 533 // A '[' could be an index subscript (after an identifier or after 534 // ')' or ']'), it could be the start of an Objective-C method 535 // expression, it could the start of an Objective-C array literal, 536 // or it could be a C++ attribute specifier [[foo::bar]]. 537 FormatToken *Left = CurrentToken->Previous; 538 Left->ParentBracket = Contexts.back().ContextKind; 539 FormatToken *Parent = Left->getPreviousNonComment(); 540 541 // Cases where '>' is followed by '['. 542 // In C++, this can happen either in array of templates (foo<int>[10]) 543 // or when array is a nested template type (unique_ptr<type1<type2>[]>). 544 bool CppArrayTemplates = 545 Style.isCpp() && Parent && Parent->is(TT_TemplateCloser) && 546 (Contexts.back().CanBeExpression || Contexts.back().IsExpression || 547 Contexts.back().InTemplateArgument); 548 549 bool IsCpp11AttributeSpecifier = isCpp11AttributeSpecifier(*Left) || 550 Contexts.back().InCpp11AttributeSpecifier; 551 552 // Treat C# Attributes [STAThread] much like C++ attributes [[...]]. 553 bool IsCSharpAttributeSpecifier = 554 isCSharpAttributeSpecifier(*Left) || 555 Contexts.back().InCSharpAttributeSpecifier; 556 557 bool InsideInlineASM = Line.startsWith(tok::kw_asm); 558 bool IsCppStructuredBinding = Left->isCppStructuredBinding(Style); 559 bool StartsObjCMethodExpr = 560 !IsCppStructuredBinding && !InsideInlineASM && !CppArrayTemplates && 561 Style.isCpp() && !IsCpp11AttributeSpecifier && 562 !IsCSharpAttributeSpecifier && Contexts.back().CanBeExpression && 563 Left->isNot(TT_LambdaLSquare) && 564 !CurrentToken->isOneOf(tok::l_brace, tok::r_square) && 565 (!Parent || 566 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren, 567 tok::kw_return, tok::kw_throw) || 568 Parent->isUnaryOperator() || 569 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 570 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) || 571 (getBinOpPrecedence(Parent->Tok.getKind(), true, true) > 572 prec::Unknown)); 573 bool ColonFound = false; 574 575 unsigned BindingIncrease = 1; 576 if (IsCppStructuredBinding) { 577 Left->setType(TT_StructuredBindingLSquare); 578 } else if (Left->is(TT_Unknown)) { 579 if (StartsObjCMethodExpr) { 580 Left->setType(TT_ObjCMethodExpr); 581 } else if (InsideInlineASM) { 582 Left->setType(TT_InlineASMSymbolicNameLSquare); 583 } else if (IsCpp11AttributeSpecifier) { 584 Left->setType(TT_AttributeSquare); 585 } else if (Style.isJavaScript() && Parent && 586 Contexts.back().ContextKind == tok::l_brace && 587 Parent->isOneOf(tok::l_brace, tok::comma)) { 588 Left->setType(TT_JsComputedPropertyName); 589 } else if (Style.isCpp() && Contexts.back().ContextKind == tok::l_brace && 590 Parent && Parent->isOneOf(tok::l_brace, tok::comma)) { 591 Left->setType(TT_DesignatedInitializerLSquare); 592 } else if (IsCSharpAttributeSpecifier) { 593 Left->setType(TT_AttributeSquare); 594 } else if (CurrentToken->is(tok::r_square) && Parent && 595 Parent->is(TT_TemplateCloser)) { 596 Left->setType(TT_ArraySubscriptLSquare); 597 } else if (Style.Language == FormatStyle::LK_Proto || 598 Style.Language == FormatStyle::LK_TextProto) { 599 // Square braces in LK_Proto can either be message field attributes: 600 // 601 // optional Aaa aaa = 1 [ 602 // (aaa) = aaa 603 // ]; 604 // 605 // extensions 123 [ 606 // (aaa) = aaa 607 // ]; 608 // 609 // or text proto extensions (in options): 610 // 611 // option (Aaa.options) = { 612 // [type.type/type] { 613 // key: value 614 // } 615 // } 616 // 617 // or repeated fields (in options): 618 // 619 // option (Aaa.options) = { 620 // keys: [ 1, 2, 3 ] 621 // } 622 // 623 // In the first and the third case we want to spread the contents inside 624 // the square braces; in the second we want to keep them inline. 625 Left->setType(TT_ArrayInitializerLSquare); 626 if (!Left->endsSequence(tok::l_square, tok::numeric_constant, 627 tok::equal) && 628 !Left->endsSequence(tok::l_square, tok::numeric_constant, 629 tok::identifier) && 630 !Left->endsSequence(tok::l_square, tok::colon, TT_SelectorName)) { 631 Left->setType(TT_ProtoExtensionLSquare); 632 BindingIncrease = 10; 633 } 634 } else if (!CppArrayTemplates && Parent && 635 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at, 636 tok::comma, tok::l_paren, tok::l_square, 637 tok::question, tok::colon, tok::kw_return, 638 // Should only be relevant to JavaScript: 639 tok::kw_default)) { 640 Left->setType(TT_ArrayInitializerLSquare); 641 } else { 642 BindingIncrease = 10; 643 Left->setType(TT_ArraySubscriptLSquare); 644 } 645 } 646 647 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease); 648 Contexts.back().IsExpression = true; 649 if (Style.isJavaScript() && Parent && Parent->is(TT_JsTypeColon)) 650 Contexts.back().IsExpression = false; 651 652 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr; 653 Contexts.back().InCpp11AttributeSpecifier = IsCpp11AttributeSpecifier; 654 Contexts.back().InCSharpAttributeSpecifier = IsCSharpAttributeSpecifier; 655 656 while (CurrentToken) { 657 if (CurrentToken->is(tok::r_square)) { 658 if (IsCpp11AttributeSpecifier) 659 CurrentToken->setType(TT_AttributeSquare); 660 if (IsCSharpAttributeSpecifier) 661 CurrentToken->setType(TT_AttributeSquare); 662 else if (((CurrentToken->Next && 663 CurrentToken->Next->is(tok::l_paren)) || 664 (CurrentToken->Previous && 665 CurrentToken->Previous->Previous == Left)) && 666 Left->is(TT_ObjCMethodExpr)) { 667 // An ObjC method call is rarely followed by an open parenthesis. It 668 // also can't be composed of just one token, unless it's a macro that 669 // will be expanded to more tokens. 670 // FIXME: Do we incorrectly label ":" with this? 671 StartsObjCMethodExpr = false; 672 Left->setType(TT_Unknown); 673 } 674 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) { 675 CurrentToken->setType(TT_ObjCMethodExpr); 676 // If we haven't seen a colon yet, make sure the last identifier 677 // before the r_square is tagged as a selector name component. 678 if (!ColonFound && CurrentToken->Previous && 679 CurrentToken->Previous->is(TT_Unknown) && 680 canBeObjCSelectorComponent(*CurrentToken->Previous)) 681 CurrentToken->Previous->setType(TT_SelectorName); 682 // determineStarAmpUsage() thinks that '*' '[' is allocating an 683 // array of pointers, but if '[' starts a selector then '*' is a 684 // binary operator. 685 if (Parent && Parent->is(TT_PointerOrReference)) 686 Parent->setType(TT_BinaryOperator); 687 } 688 // An arrow after an ObjC method expression is not a lambda arrow. 689 if (CurrentToken->getType() == TT_ObjCMethodExpr && 690 CurrentToken->Next && CurrentToken->Next->is(TT_LambdaArrow)) 691 CurrentToken->Next->setType(TT_Unknown); 692 Left->MatchingParen = CurrentToken; 693 CurrentToken->MatchingParen = Left; 694 // FirstObjCSelectorName is set when a colon is found. This does 695 // not work, however, when the method has no parameters. 696 // Here, we set FirstObjCSelectorName when the end of the method call is 697 // reached, in case it was not set already. 698 if (!Contexts.back().FirstObjCSelectorName) { 699 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 700 if (Previous && Previous->is(TT_SelectorName)) { 701 Previous->ObjCSelectorNameParts = 1; 702 Contexts.back().FirstObjCSelectorName = Previous; 703 } 704 } else { 705 Left->ParameterCount = 706 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 707 } 708 if (Contexts.back().FirstObjCSelectorName) { 709 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 710 Contexts.back().LongestObjCSelectorName; 711 if (Left->BlockParameterCount > 1) 712 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0; 713 } 714 next(); 715 return true; 716 } 717 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace)) 718 return false; 719 if (CurrentToken->is(tok::colon)) { 720 if (IsCpp11AttributeSpecifier && 721 CurrentToken->endsSequence(tok::colon, tok::identifier, 722 tok::kw_using)) { 723 // Remember that this is a [[using ns: foo]] C++ attribute, so we 724 // don't add a space before the colon (unlike other colons). 725 CurrentToken->setType(TT_AttributeColon); 726 } else if (Left->isOneOf(TT_ArraySubscriptLSquare, 727 TT_DesignatedInitializerLSquare)) { 728 Left->setType(TT_ObjCMethodExpr); 729 StartsObjCMethodExpr = true; 730 Contexts.back().ColonIsObjCMethodExpr = true; 731 if (Parent && Parent->is(tok::r_paren)) 732 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 733 Parent->setType(TT_CastRParen); 734 } 735 ColonFound = true; 736 } 737 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) && 738 !ColonFound) 739 Left->setType(TT_ArrayInitializerLSquare); 740 FormatToken *Tok = CurrentToken; 741 if (!consumeToken()) 742 return false; 743 updateParameterCount(Left, Tok); 744 } 745 return false; 746 } 747 748 bool couldBeInStructArrayInitializer() const { 749 if (Contexts.size() < 2) 750 return false; 751 // We want to back up no more then 2 context levels i.e. 752 // . { { <- 753 const auto End = std::next(Contexts.rbegin(), 2); 754 auto Last = Contexts.rbegin(); 755 unsigned Depth = 0; 756 for (; Last != End; ++Last) { 757 if (Last->ContextKind == tok::l_brace) 758 ++Depth; 759 } 760 return Depth == 2 && Last->ContextKind != tok::l_brace; 761 } 762 763 bool parseBrace() { 764 if (CurrentToken) { 765 FormatToken *Left = CurrentToken->Previous; 766 Left->ParentBracket = Contexts.back().ContextKind; 767 768 if (Contexts.back().CaretFound) 769 Left->setType(TT_ObjCBlockLBrace); 770 Contexts.back().CaretFound = false; 771 772 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1); 773 Contexts.back().ColonIsDictLiteral = true; 774 if (Left->is(BK_BracedInit)) 775 Contexts.back().IsExpression = true; 776 if (Style.isJavaScript() && Left->Previous && 777 Left->Previous->is(TT_JsTypeColon)) 778 Contexts.back().IsExpression = false; 779 780 unsigned CommaCount = 0; 781 while (CurrentToken) { 782 if (CurrentToken->is(tok::r_brace)) { 783 Left->MatchingParen = CurrentToken; 784 CurrentToken->MatchingParen = Left; 785 if (Style.AlignArrayOfStructures != FormatStyle::AIAS_None) { 786 if (Left->ParentBracket == tok::l_brace && 787 couldBeInStructArrayInitializer() && CommaCount > 0) { 788 Contexts.back().InStructArrayInitializer = true; 789 } 790 } 791 next(); 792 return true; 793 } 794 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 795 return false; 796 updateParameterCount(Left, CurrentToken); 797 if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) { 798 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 799 if (Previous->is(TT_JsTypeOptionalQuestion)) 800 Previous = Previous->getPreviousNonComment(); 801 if ((CurrentToken->is(tok::colon) && 802 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) || 803 Style.Language == FormatStyle::LK_Proto || 804 Style.Language == FormatStyle::LK_TextProto) { 805 Left->setType(TT_DictLiteral); 806 if (Previous->Tok.getIdentifierInfo() || 807 Previous->is(tok::string_literal)) 808 Previous->setType(TT_SelectorName); 809 } 810 if (CurrentToken->is(tok::colon) || Style.isJavaScript()) 811 Left->setType(TT_DictLiteral); 812 } 813 if (CurrentToken->is(tok::comma)) { 814 if (Style.isJavaScript()) 815 Left->setType(TT_DictLiteral); 816 ++CommaCount; 817 } 818 if (!consumeToken()) 819 return false; 820 } 821 } 822 return true; 823 } 824 825 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 826 // For ObjC methods, the number of parameters is calculated differently as 827 // method declarations have a different structure (the parameters are not 828 // inside a bracket scope). 829 if (Current->is(tok::l_brace) && Current->is(BK_Block)) 830 ++Left->BlockParameterCount; 831 if (Current->is(tok::comma)) { 832 ++Left->ParameterCount; 833 if (!Left->Role) 834 Left->Role.reset(new CommaSeparatedList(Style)); 835 Left->Role->CommaFound(Current); 836 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 837 Left->ParameterCount = 1; 838 } 839 } 840 841 bool parseConditional() { 842 while (CurrentToken) { 843 if (CurrentToken->is(tok::colon)) { 844 CurrentToken->setType(TT_ConditionalExpr); 845 next(); 846 return true; 847 } 848 if (!consumeToken()) 849 return false; 850 } 851 return false; 852 } 853 854 bool parseTemplateDeclaration() { 855 if (CurrentToken && CurrentToken->is(tok::less)) { 856 CurrentToken->setType(TT_TemplateOpener); 857 next(); 858 if (!parseAngle()) 859 return false; 860 if (CurrentToken) 861 CurrentToken->Previous->ClosesTemplateDeclaration = true; 862 return true; 863 } 864 return false; 865 } 866 867 bool consumeToken() { 868 FormatToken *Tok = CurrentToken; 869 next(); 870 switch (Tok->Tok.getKind()) { 871 case tok::plus: 872 case tok::minus: 873 if (!Tok->Previous && Line.MustBeDeclaration) 874 Tok->setType(TT_ObjCMethodSpecifier); 875 break; 876 case tok::colon: 877 if (!Tok->Previous) 878 return false; 879 // Colons from ?: are handled in parseConditional(). 880 if (Style.isJavaScript()) { 881 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop 882 (Contexts.size() == 1 && // switch/case labels 883 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) || 884 Contexts.back().ContextKind == tok::l_paren || // function params 885 Contexts.back().ContextKind == tok::l_square || // array type 886 (!Contexts.back().IsExpression && 887 Contexts.back().ContextKind == tok::l_brace) || // object type 888 (Contexts.size() == 1 && 889 Line.MustBeDeclaration)) { // method/property declaration 890 Contexts.back().IsExpression = false; 891 Tok->setType(TT_JsTypeColon); 892 break; 893 } 894 } else if (Style.isCSharp()) { 895 if (Contexts.back().InCSharpAttributeSpecifier) { 896 Tok->setType(TT_AttributeColon); 897 break; 898 } 899 if (Contexts.back().ContextKind == tok::l_paren) { 900 Tok->setType(TT_CSharpNamedArgumentColon); 901 break; 902 } 903 } 904 if (Line.First->isOneOf(Keywords.kw_module, Keywords.kw_import) || 905 Line.First->startsSequence(tok::kw_export, Keywords.kw_module) || 906 Line.First->startsSequence(tok::kw_export, Keywords.kw_import)) { 907 Tok->setType(TT_ModulePartitionColon); 908 } else if (Contexts.back().ColonIsDictLiteral || 909 Style.Language == FormatStyle::LK_Proto || 910 Style.Language == FormatStyle::LK_TextProto) { 911 Tok->setType(TT_DictLiteral); 912 if (Style.Language == FormatStyle::LK_TextProto) { 913 if (FormatToken *Previous = Tok->getPreviousNonComment()) 914 Previous->setType(TT_SelectorName); 915 } 916 } else if (Contexts.back().ColonIsObjCMethodExpr || 917 Line.startsWith(TT_ObjCMethodSpecifier)) { 918 Tok->setType(TT_ObjCMethodExpr); 919 const FormatToken *BeforePrevious = Tok->Previous->Previous; 920 // Ensure we tag all identifiers in method declarations as 921 // TT_SelectorName. 922 bool UnknownIdentifierInMethodDeclaration = 923 Line.startsWith(TT_ObjCMethodSpecifier) && 924 Tok->Previous->is(tok::identifier) && Tok->Previous->is(TT_Unknown); 925 if (!BeforePrevious || 926 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 927 !(BeforePrevious->is(TT_CastRParen) || 928 (BeforePrevious->is(TT_ObjCMethodExpr) && 929 BeforePrevious->is(tok::colon))) || 930 BeforePrevious->is(tok::r_square) || 931 Contexts.back().LongestObjCSelectorName == 0 || 932 UnknownIdentifierInMethodDeclaration) { 933 Tok->Previous->setType(TT_SelectorName); 934 if (!Contexts.back().FirstObjCSelectorName) 935 Contexts.back().FirstObjCSelectorName = Tok->Previous; 936 else if (Tok->Previous->ColumnWidth > 937 Contexts.back().LongestObjCSelectorName) 938 Contexts.back().LongestObjCSelectorName = 939 Tok->Previous->ColumnWidth; 940 Tok->Previous->ParameterIndex = 941 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 942 ++Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 943 } 944 } else if (Contexts.back().ColonIsForRangeExpr) { 945 Tok->setType(TT_RangeBasedForLoopColon); 946 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 947 Tok->setType(TT_BitFieldColon); 948 } else if (Contexts.size() == 1 && 949 !Line.First->isOneOf(tok::kw_enum, tok::kw_case, 950 tok::kw_default)) { 951 FormatToken *Prev = Tok->getPreviousNonComment(); 952 if (!Prev) 953 break; 954 if (Prev->isOneOf(tok::r_paren, tok::kw_noexcept)) 955 Tok->setType(TT_CtorInitializerColon); 956 else if (Prev->is(tok::kw_try)) { 957 // Member initializer list within function try block. 958 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 959 if (!PrevPrev) 960 break; 961 if (PrevPrev && PrevPrev->isOneOf(tok::r_paren, tok::kw_noexcept)) 962 Tok->setType(TT_CtorInitializerColon); 963 } else 964 Tok->setType(TT_InheritanceColon); 965 } else if (canBeObjCSelectorComponent(*Tok->Previous) && Tok->Next && 966 (Tok->Next->isOneOf(tok::r_paren, tok::comma) || 967 (canBeObjCSelectorComponent(*Tok->Next) && Tok->Next->Next && 968 Tok->Next->Next->is(tok::colon)))) { 969 // This handles a special macro in ObjC code where selectors including 970 // the colon are passed as macro arguments. 971 Tok->setType(TT_ObjCMethodExpr); 972 } else if (Contexts.back().ContextKind == tok::l_paren) { 973 Tok->setType(TT_InlineASMColon); 974 } 975 break; 976 case tok::pipe: 977 case tok::amp: 978 // | and & in declarations/type expressions represent union and 979 // intersection types, respectively. 980 if (Style.isJavaScript() && !Contexts.back().IsExpression) 981 Tok->setType(TT_JsTypeOperator); 982 break; 983 case tok::kw_if: 984 case tok::kw_while: 985 if (Tok->is(tok::kw_if) && CurrentToken && 986 CurrentToken->isOneOf(tok::kw_constexpr, tok::identifier)) 987 next(); 988 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 989 next(); 990 if (!parseParens(/*LookForDecls=*/true)) 991 return false; 992 } 993 break; 994 case tok::kw_for: 995 if (Style.isJavaScript()) { 996 // x.for and {for: ...} 997 if ((Tok->Previous && Tok->Previous->is(tok::period)) || 998 (Tok->Next && Tok->Next->is(tok::colon))) 999 break; 1000 // JS' for await ( ... 1001 if (CurrentToken && CurrentToken->is(Keywords.kw_await)) 1002 next(); 1003 } 1004 if (Style.isCpp() && CurrentToken && CurrentToken->is(tok::kw_co_await)) 1005 next(); 1006 Contexts.back().ColonIsForRangeExpr = true; 1007 next(); 1008 if (!parseParens()) 1009 return false; 1010 break; 1011 case tok::l_paren: 1012 // When faced with 'operator()()', the kw_operator handler incorrectly 1013 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make 1014 // the first two parens OverloadedOperators and the second l_paren an 1015 // OverloadedOperatorLParen. 1016 if (Tok->Previous && Tok->Previous->is(tok::r_paren) && 1017 Tok->Previous->MatchingParen && 1018 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) { 1019 Tok->Previous->setType(TT_OverloadedOperator); 1020 Tok->Previous->MatchingParen->setType(TT_OverloadedOperator); 1021 Tok->setType(TT_OverloadedOperatorLParen); 1022 } 1023 1024 if (!parseParens()) 1025 return false; 1026 if (Line.MustBeDeclaration && Contexts.size() == 1 && 1027 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) && 1028 !Tok->is(TT_TypeDeclarationParen) && 1029 (!Tok->Previous || !Tok->Previous->isOneOf(tok::kw___attribute, 1030 TT_LeadingJavaAnnotation))) 1031 Line.MightBeFunctionDecl = true; 1032 break; 1033 case tok::l_square: 1034 if (!parseSquare()) 1035 return false; 1036 break; 1037 case tok::l_brace: 1038 if (Style.Language == FormatStyle::LK_TextProto) { 1039 FormatToken *Previous = Tok->getPreviousNonComment(); 1040 if (Previous && Previous->getType() != TT_DictLiteral) 1041 Previous->setType(TT_SelectorName); 1042 } 1043 if (!parseBrace()) 1044 return false; 1045 break; 1046 case tok::less: 1047 if (parseAngle()) { 1048 Tok->setType(TT_TemplateOpener); 1049 // In TT_Proto, we must distignuish between: 1050 // map<key, value> 1051 // msg < item: data > 1052 // msg: < item: data > 1053 // In TT_TextProto, map<key, value> does not occur. 1054 if (Style.Language == FormatStyle::LK_TextProto || 1055 (Style.Language == FormatStyle::LK_Proto && Tok->Previous && 1056 Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) { 1057 Tok->setType(TT_DictLiteral); 1058 FormatToken *Previous = Tok->getPreviousNonComment(); 1059 if (Previous && Previous->getType() != TT_DictLiteral) 1060 Previous->setType(TT_SelectorName); 1061 } 1062 } else { 1063 Tok->setType(TT_BinaryOperator); 1064 NonTemplateLess.insert(Tok); 1065 CurrentToken = Tok; 1066 next(); 1067 } 1068 break; 1069 case tok::r_paren: 1070 case tok::r_square: 1071 return false; 1072 case tok::r_brace: 1073 // Lines can start with '}'. 1074 if (Tok->Previous) 1075 return false; 1076 break; 1077 case tok::greater: 1078 if (Style.Language != FormatStyle::LK_TextProto) 1079 Tok->setType(TT_BinaryOperator); 1080 if (Tok->Previous && Tok->Previous->is(TT_TemplateCloser)) 1081 Tok->SpacesRequiredBefore = 1; 1082 break; 1083 case tok::kw_operator: 1084 if (Style.Language == FormatStyle::LK_TextProto || 1085 Style.Language == FormatStyle::LK_Proto) 1086 break; 1087 while (CurrentToken && 1088 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 1089 if (CurrentToken->isOneOf(tok::star, tok::amp)) 1090 CurrentToken->setType(TT_PointerOrReference); 1091 consumeToken(); 1092 if (CurrentToken && CurrentToken->is(tok::comma) && 1093 CurrentToken->Previous->isNot(tok::kw_operator)) 1094 break; 1095 if (CurrentToken && CurrentToken->Previous->isOneOf( 1096 TT_BinaryOperator, TT_UnaryOperator, tok::comma, 1097 tok::star, tok::arrow, tok::amp, tok::ampamp)) 1098 CurrentToken->Previous->setType(TT_OverloadedOperator); 1099 } 1100 if (CurrentToken && CurrentToken->is(tok::l_paren)) 1101 CurrentToken->setType(TT_OverloadedOperatorLParen); 1102 if (CurrentToken && CurrentToken->Previous->is(TT_BinaryOperator)) 1103 CurrentToken->Previous->setType(TT_OverloadedOperator); 1104 break; 1105 case tok::question: 1106 if (Style.isJavaScript() && Tok->Next && 1107 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren, 1108 tok::r_brace)) { 1109 // Question marks before semicolons, colons, etc. indicate optional 1110 // types (fields, parameters), e.g. 1111 // function(x?: string, y?) {...} 1112 // class X { y?; } 1113 Tok->setType(TT_JsTypeOptionalQuestion); 1114 break; 1115 } 1116 // Declarations cannot be conditional expressions, this can only be part 1117 // of a type declaration. 1118 if (Line.MustBeDeclaration && !Contexts.back().IsExpression && 1119 Style.isJavaScript()) 1120 break; 1121 if (Style.isCSharp()) { 1122 // `Type?)`, `Type?>`, `Type? name;` and `Type? name =` can only be 1123 // nullable types. 1124 // Line.MustBeDeclaration will be true for `Type? name;`. 1125 if ((!Contexts.back().IsExpression && Line.MustBeDeclaration) || 1126 (Tok->Next && Tok->Next->isOneOf(tok::r_paren, tok::greater)) || 1127 (Tok->Next && Tok->Next->is(tok::identifier) && Tok->Next->Next && 1128 Tok->Next->Next->is(tok::equal))) { 1129 Tok->setType(TT_CSharpNullable); 1130 break; 1131 } 1132 } 1133 parseConditional(); 1134 break; 1135 case tok::kw_template: 1136 parseTemplateDeclaration(); 1137 break; 1138 case tok::comma: 1139 if (Contexts.back().InCtorInitializer) 1140 Tok->setType(TT_CtorInitializerComma); 1141 else if (Contexts.back().InInheritanceList) 1142 Tok->setType(TT_InheritanceComma); 1143 else if (Contexts.back().FirstStartOfName && 1144 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) { 1145 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 1146 Line.IsMultiVariableDeclStmt = true; 1147 } 1148 if (Contexts.back().IsForEachMacro) 1149 Contexts.back().IsExpression = true; 1150 break; 1151 case tok::identifier: 1152 if (Tok->isOneOf(Keywords.kw___has_include, 1153 Keywords.kw___has_include_next)) { 1154 parseHasInclude(); 1155 } 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 default: 1163 break; 1164 } 1165 return true; 1166 } 1167 1168 void parseCSharpGenericTypeConstraint() { 1169 int OpenAngleBracketsCount = 0; 1170 while (CurrentToken) { 1171 if (CurrentToken->is(tok::less)) { 1172 // parseAngle is too greedy and will consume the whole line. 1173 CurrentToken->setType(TT_TemplateOpener); 1174 ++OpenAngleBracketsCount; 1175 next(); 1176 } else if (CurrentToken->is(tok::greater)) { 1177 CurrentToken->setType(TT_TemplateCloser); 1178 --OpenAngleBracketsCount; 1179 next(); 1180 } else if (CurrentToken->is(tok::comma) && OpenAngleBracketsCount == 0) { 1181 // We allow line breaks after GenericTypeConstraintComma's 1182 // so do not flag commas in Generics as GenericTypeConstraintComma's. 1183 CurrentToken->setType(TT_CSharpGenericTypeConstraintComma); 1184 next(); 1185 } else if (CurrentToken->is(Keywords.kw_where)) { 1186 CurrentToken->setType(TT_CSharpGenericTypeConstraint); 1187 next(); 1188 } else if (CurrentToken->is(tok::colon)) { 1189 CurrentToken->setType(TT_CSharpGenericTypeConstraintColon); 1190 next(); 1191 } else { 1192 next(); 1193 } 1194 } 1195 } 1196 1197 void parseIncludeDirective() { 1198 if (CurrentToken && CurrentToken->is(tok::less)) { 1199 next(); 1200 while (CurrentToken) { 1201 // Mark tokens up to the trailing line comments as implicit string 1202 // literals. 1203 if (CurrentToken->isNot(tok::comment) && 1204 !CurrentToken->TokenText.startswith("//")) 1205 CurrentToken->setType(TT_ImplicitStringLiteral); 1206 next(); 1207 } 1208 } 1209 } 1210 1211 void parseWarningOrError() { 1212 next(); 1213 // We still want to format the whitespace left of the first token of the 1214 // warning or error. 1215 next(); 1216 while (CurrentToken) { 1217 CurrentToken->setType(TT_ImplicitStringLiteral); 1218 next(); 1219 } 1220 } 1221 1222 void parsePragma() { 1223 next(); // Consume "pragma". 1224 if (CurrentToken && 1225 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 1226 bool IsMark = CurrentToken->is(Keywords.kw_mark); 1227 next(); // Consume "mark". 1228 next(); // Consume first token (so we fix leading whitespace). 1229 while (CurrentToken) { 1230 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 1231 CurrentToken->setType(TT_ImplicitStringLiteral); 1232 next(); 1233 } 1234 } 1235 } 1236 1237 void parseHasInclude() { 1238 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 1239 return; 1240 next(); // '(' 1241 parseIncludeDirective(); 1242 next(); // ')' 1243 } 1244 1245 LineType parsePreprocessorDirective() { 1246 bool IsFirstToken = CurrentToken->IsFirst; 1247 LineType Type = LT_PreprocessorDirective; 1248 next(); 1249 if (!CurrentToken) 1250 return Type; 1251 1252 if (Style.isJavaScript() && IsFirstToken) { 1253 // JavaScript files can contain shebang lines of the form: 1254 // #!/usr/bin/env node 1255 // Treat these like C++ #include directives. 1256 while (CurrentToken) { 1257 // Tokens cannot be comments here. 1258 CurrentToken->setType(TT_ImplicitStringLiteral); 1259 next(); 1260 } 1261 return LT_ImportStatement; 1262 } 1263 1264 if (CurrentToken->Tok.is(tok::numeric_constant)) { 1265 CurrentToken->SpacesRequiredBefore = 1; 1266 return Type; 1267 } 1268 // Hashes in the middle of a line can lead to any strange token 1269 // sequence. 1270 if (!CurrentToken->Tok.getIdentifierInfo()) 1271 return Type; 1272 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 1273 case tok::pp_include: 1274 case tok::pp_include_next: 1275 case tok::pp_import: 1276 next(); 1277 parseIncludeDirective(); 1278 Type = LT_ImportStatement; 1279 break; 1280 case tok::pp_error: 1281 case tok::pp_warning: 1282 parseWarningOrError(); 1283 break; 1284 case tok::pp_pragma: 1285 parsePragma(); 1286 break; 1287 case tok::pp_if: 1288 case tok::pp_elif: 1289 Contexts.back().IsExpression = true; 1290 next(); 1291 parseLine(); 1292 break; 1293 default: 1294 break; 1295 } 1296 while (CurrentToken) { 1297 FormatToken *Tok = CurrentToken; 1298 next(); 1299 if (Tok->is(tok::l_paren)) 1300 parseParens(); 1301 else if (Tok->isOneOf(Keywords.kw___has_include, 1302 Keywords.kw___has_include_next)) 1303 parseHasInclude(); 1304 } 1305 return Type; 1306 } 1307 1308 public: 1309 LineType parseLine() { 1310 if (!CurrentToken) 1311 return LT_Invalid; 1312 NonTemplateLess.clear(); 1313 if (CurrentToken->is(tok::hash)) 1314 return parsePreprocessorDirective(); 1315 1316 // Directly allow to 'import <string-literal>' to support protocol buffer 1317 // definitions (github.com/google/protobuf) or missing "#" (either way we 1318 // should not break the line). 1319 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 1320 if ((Style.Language == FormatStyle::LK_Java && 1321 CurrentToken->is(Keywords.kw_package)) || 1322 (Info && Info->getPPKeywordID() == tok::pp_import && 1323 CurrentToken->Next && 1324 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 1325 tok::kw_static))) { 1326 next(); 1327 parseIncludeDirective(); 1328 return LT_ImportStatement; 1329 } 1330 1331 // If this line starts and ends in '<' and '>', respectively, it is likely 1332 // part of "#define <a/b.h>". 1333 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 1334 parseIncludeDirective(); 1335 return LT_ImportStatement; 1336 } 1337 1338 // In .proto files, top-level options and package statements are very 1339 // similar to import statements and should not be line-wrapped. 1340 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 1341 CurrentToken->isOneOf(Keywords.kw_option, Keywords.kw_package)) { 1342 next(); 1343 if (CurrentToken && CurrentToken->is(tok::identifier)) { 1344 while (CurrentToken) 1345 next(); 1346 return LT_ImportStatement; 1347 } 1348 } 1349 1350 bool KeywordVirtualFound = false; 1351 bool ImportStatement = false; 1352 1353 // import {...} from '...'; 1354 if (Style.isJavaScript() && CurrentToken->is(Keywords.kw_import)) 1355 ImportStatement = true; 1356 1357 while (CurrentToken) { 1358 if (CurrentToken->is(tok::kw_virtual)) 1359 KeywordVirtualFound = true; 1360 if (Style.isJavaScript()) { 1361 // export {...} from '...'; 1362 // An export followed by "from 'some string';" is a re-export from 1363 // another module identified by a URI and is treated as a 1364 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 1365 // Just "export {...};" or "export class ..." should not be treated as 1366 // an import in this sense. 1367 if (Line.First->is(tok::kw_export) && 1368 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 1369 CurrentToken->Next->isStringLiteral()) 1370 ImportStatement = true; 1371 if (isClosureImportStatement(*CurrentToken)) 1372 ImportStatement = true; 1373 } 1374 if (!consumeToken()) 1375 return LT_Invalid; 1376 } 1377 if (KeywordVirtualFound) 1378 return LT_VirtualFunctionDecl; 1379 if (ImportStatement) 1380 return LT_ImportStatement; 1381 1382 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 1383 if (Contexts.back().FirstObjCSelectorName) 1384 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 1385 Contexts.back().LongestObjCSelectorName; 1386 return LT_ObjCMethodDecl; 1387 } 1388 1389 for (const auto &ctx : Contexts) { 1390 if (ctx.InStructArrayInitializer) { 1391 return LT_ArrayOfStructInitializer; 1392 } 1393 } 1394 1395 return LT_Other; 1396 } 1397 1398 private: 1399 bool isClosureImportStatement(const FormatToken &Tok) { 1400 // FIXME: Closure-library specific stuff should not be hard-coded but be 1401 // configurable. 1402 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 1403 Tok.Next->Next && 1404 (Tok.Next->Next->TokenText == "module" || 1405 Tok.Next->Next->TokenText == "provide" || 1406 Tok.Next->Next->TokenText == "require" || 1407 Tok.Next->Next->TokenText == "requireType" || 1408 Tok.Next->Next->TokenText == "forwardDeclare") && 1409 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 1410 } 1411 1412 void resetTokenMetadata() { 1413 if (!CurrentToken) 1414 return; 1415 1416 // Reset token type in case we have already looked at it and then 1417 // recovered from an error (e.g. failure to find the matching >). 1418 if (!CurrentToken->isOneOf( 1419 TT_LambdaLSquare, TT_LambdaLBrace, TT_AttributeMacro, TT_IfMacro, 1420 TT_ForEachMacro, TT_TypenameMacro, TT_FunctionLBrace, 1421 TT_ImplicitStringLiteral, TT_InlineASMBrace, TT_FatArrow, 1422 TT_LambdaArrow, TT_NamespaceMacro, TT_OverloadedOperator, 1423 TT_RegexLiteral, TT_TemplateString, TT_ObjCStringLiteral, 1424 TT_UntouchableMacroFunc, TT_ConstraintJunctions, 1425 TT_StatementAttributeLikeMacro)) 1426 CurrentToken->setType(TT_Unknown); 1427 CurrentToken->Role.reset(); 1428 CurrentToken->MatchingParen = nullptr; 1429 CurrentToken->FakeLParens.clear(); 1430 CurrentToken->FakeRParens = 0; 1431 } 1432 1433 void next() { 1434 if (!CurrentToken) 1435 return; 1436 1437 CurrentToken->NestingLevel = Contexts.size() - 1; 1438 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1439 modifyContext(*CurrentToken); 1440 determineTokenType(*CurrentToken); 1441 CurrentToken = CurrentToken->Next; 1442 1443 resetTokenMetadata(); 1444 } 1445 1446 /// A struct to hold information valid in a specific context, e.g. 1447 /// a pair of parenthesis. 1448 struct Context { 1449 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1450 bool IsExpression) 1451 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1452 IsExpression(IsExpression) {} 1453 1454 tok::TokenKind ContextKind; 1455 unsigned BindingStrength; 1456 bool IsExpression; 1457 unsigned LongestObjCSelectorName = 0; 1458 bool ColonIsForRangeExpr = false; 1459 bool ColonIsDictLiteral = false; 1460 bool ColonIsObjCMethodExpr = false; 1461 FormatToken *FirstObjCSelectorName = nullptr; 1462 FormatToken *FirstStartOfName = nullptr; 1463 bool CanBeExpression = true; 1464 bool InTemplateArgument = false; 1465 bool InCtorInitializer = false; 1466 bool InInheritanceList = false; 1467 bool CaretFound = false; 1468 bool IsForEachMacro = false; 1469 bool InCpp11AttributeSpecifier = false; 1470 bool InCSharpAttributeSpecifier = false; 1471 bool InStructArrayInitializer = false; 1472 }; 1473 1474 /// Puts a new \c Context onto the stack \c Contexts for the lifetime 1475 /// of each instance. 1476 struct ScopedContextCreator { 1477 AnnotatingParser &P; 1478 1479 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1480 unsigned Increase) 1481 : P(P) { 1482 P.Contexts.push_back(Context(ContextKind, 1483 P.Contexts.back().BindingStrength + Increase, 1484 P.Contexts.back().IsExpression)); 1485 } 1486 1487 ~ScopedContextCreator() { 1488 if (P.Style.AlignArrayOfStructures != FormatStyle::AIAS_None) { 1489 if (P.Contexts.back().InStructArrayInitializer) { 1490 P.Contexts.pop_back(); 1491 P.Contexts.back().InStructArrayInitializer = true; 1492 return; 1493 } 1494 } 1495 P.Contexts.pop_back(); 1496 } 1497 }; 1498 1499 void modifyContext(const FormatToken &Current) { 1500 if (Current.getPrecedence() == prec::Assignment && 1501 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 1502 // Type aliases use `type X = ...;` in TypeScript and can be exported 1503 // using `export type ...`. 1504 !(Style.isJavaScript() && 1505 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1506 Line.startsWith(tok::kw_export, Keywords.kw_type, 1507 tok::identifier))) && 1508 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 1509 Contexts.back().IsExpression = true; 1510 if (!Line.startsWith(TT_UnaryOperator)) { 1511 for (FormatToken *Previous = Current.Previous; 1512 Previous && Previous->Previous && 1513 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1514 Previous = Previous->Previous) { 1515 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1516 Previous = Previous->MatchingParen; 1517 if (!Previous) 1518 break; 1519 } 1520 if (Previous->opensScope()) 1521 break; 1522 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1523 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1524 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1525 Previous->setType(TT_PointerOrReference); 1526 } 1527 } 1528 } else if (Current.is(tok::lessless) && 1529 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1530 Contexts.back().IsExpression = true; 1531 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1532 Contexts.back().IsExpression = true; 1533 } else if (Current.is(TT_TrailingReturnArrow)) { 1534 Contexts.back().IsExpression = false; 1535 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1536 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1537 } else if (Current.Previous && 1538 Current.Previous->is(TT_CtorInitializerColon)) { 1539 Contexts.back().IsExpression = true; 1540 Contexts.back().InCtorInitializer = true; 1541 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1542 Contexts.back().InInheritanceList = true; 1543 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1544 for (FormatToken *Previous = Current.Previous; 1545 Previous && Previous->isOneOf(tok::star, tok::amp); 1546 Previous = Previous->Previous) 1547 Previous->setType(TT_PointerOrReference); 1548 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 1549 Contexts.back().IsExpression = false; 1550 } else if (Current.is(tok::kw_new)) { 1551 Contexts.back().CanBeExpression = false; 1552 } else if (Current.is(tok::semi) || 1553 (Current.is(tok::exclaim) && Current.Previous && 1554 !Current.Previous->is(tok::kw_operator))) { 1555 // This should be the condition or increment in a for-loop. 1556 // But not operator !() (can't use TT_OverloadedOperator here as its not 1557 // been annotated yet). 1558 Contexts.back().IsExpression = true; 1559 } 1560 } 1561 1562 static FormatToken *untilMatchingParen(FormatToken *Current) { 1563 // Used when `MatchingParen` is not yet established. 1564 int ParenLevel = 0; 1565 while (Current) { 1566 if (Current->is(tok::l_paren)) 1567 ParenLevel++; 1568 if (Current->is(tok::r_paren)) 1569 ParenLevel--; 1570 if (ParenLevel < 1) 1571 break; 1572 Current = Current->Next; 1573 } 1574 return Current; 1575 } 1576 1577 static bool isDeductionGuide(FormatToken &Current) { 1578 // Look for a deduction guide template<T> A(...) -> A<...>; 1579 if (Current.Previous && Current.Previous->is(tok::r_paren) && 1580 Current.startsSequence(tok::arrow, tok::identifier, tok::less)) { 1581 // Find the TemplateCloser. 1582 FormatToken *TemplateCloser = Current.Next->Next; 1583 int NestingLevel = 0; 1584 while (TemplateCloser) { 1585 // Skip over an expressions in parens A<(3 < 2)>; 1586 if (TemplateCloser->is(tok::l_paren)) { 1587 // No Matching Paren yet so skip to matching paren 1588 TemplateCloser = untilMatchingParen(TemplateCloser); 1589 if (!TemplateCloser) 1590 break; 1591 } 1592 if (TemplateCloser->is(tok::less)) 1593 NestingLevel++; 1594 if (TemplateCloser->is(tok::greater)) 1595 NestingLevel--; 1596 if (NestingLevel < 1) 1597 break; 1598 TemplateCloser = TemplateCloser->Next; 1599 } 1600 // Assuming we have found the end of the template ensure its followed 1601 // with a semi-colon. 1602 if (TemplateCloser && TemplateCloser->Next && 1603 TemplateCloser->Next->is(tok::semi) && 1604 Current.Previous->MatchingParen) { 1605 // Determine if the identifier `A` prior to the A<..>; is the same as 1606 // prior to the A(..) 1607 FormatToken *LeadingIdentifier = 1608 Current.Previous->MatchingParen->Previous; 1609 1610 // Differentiate a deduction guide by seeing the 1611 // > of the template prior to the leading identifier. 1612 if (LeadingIdentifier) { 1613 FormatToken *PriorLeadingIdentifier = LeadingIdentifier->Previous; 1614 // Skip back past explicit decoration 1615 if (PriorLeadingIdentifier && 1616 PriorLeadingIdentifier->is(tok::kw_explicit)) 1617 PriorLeadingIdentifier = PriorLeadingIdentifier->Previous; 1618 1619 return (PriorLeadingIdentifier && 1620 PriorLeadingIdentifier->is(TT_TemplateCloser) && 1621 LeadingIdentifier->TokenText == Current.Next->TokenText); 1622 } 1623 } 1624 } 1625 return false; 1626 } 1627 1628 void determineTokenType(FormatToken &Current) { 1629 if (!Current.is(TT_Unknown)) 1630 // The token type is already known. 1631 return; 1632 1633 if ((Style.isJavaScript() || Style.isCSharp()) && 1634 Current.is(tok::exclaim)) { 1635 if (Current.Previous) { 1636 bool IsIdentifier = 1637 Style.isJavaScript() 1638 ? Keywords.IsJavaScriptIdentifier( 1639 *Current.Previous, /* AcceptIdentifierName= */ true) 1640 : Current.Previous->is(tok::identifier); 1641 if (IsIdentifier || 1642 Current.Previous->isOneOf( 1643 tok::kw_namespace, tok::r_paren, tok::r_square, tok::r_brace, 1644 tok::kw_false, tok::kw_true, Keywords.kw_type, Keywords.kw_get, 1645 Keywords.kw_set) || 1646 Current.Previous->Tok.isLiteral()) { 1647 Current.setType(TT_NonNullAssertion); 1648 return; 1649 } 1650 } 1651 if (Current.Next && 1652 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1653 Current.setType(TT_NonNullAssertion); 1654 return; 1655 } 1656 } 1657 1658 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1659 // function declaration have been found. In this case, 'Current' is a 1660 // trailing token of this declaration and thus cannot be a name. 1661 if (Current.is(Keywords.kw_instanceof)) { 1662 Current.setType(TT_BinaryOperator); 1663 } else if (isStartOfName(Current) && 1664 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1665 Contexts.back().FirstStartOfName = &Current; 1666 Current.setType(TT_StartOfName); 1667 } else if (Current.is(tok::semi)) { 1668 // Reset FirstStartOfName after finding a semicolon so that a for loop 1669 // with multiple increment statements is not confused with a for loop 1670 // having multiple variable declarations. 1671 Contexts.back().FirstStartOfName = nullptr; 1672 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1673 AutoFound = true; 1674 } else if (Current.is(tok::arrow) && 1675 Style.Language == FormatStyle::LK_Java) { 1676 Current.setType(TT_LambdaArrow); 1677 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1678 Current.NestingLevel == 0 && 1679 !Current.Previous->isOneOf(tok::kw_operator, tok::identifier)) { 1680 // not auto operator->() -> xxx; 1681 Current.setType(TT_TrailingReturnArrow); 1682 } else if (Current.is(tok::arrow) && Current.Previous && 1683 Current.Previous->is(tok::r_brace)) { 1684 // Concept implicit conversion constraint needs to be treated like 1685 // a trailing return type ... } -> <type>. 1686 Current.setType(TT_TrailingReturnArrow); 1687 } else if (isDeductionGuide(Current)) { 1688 // Deduction guides trailing arrow " A(...) -> A<T>;". 1689 Current.setType(TT_TrailingReturnArrow); 1690 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1691 Current.setType(determineStarAmpUsage( 1692 Current, 1693 Contexts.back().CanBeExpression && Contexts.back().IsExpression, 1694 Contexts.back().InTemplateArgument)); 1695 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1696 Current.setType(determinePlusMinusCaretUsage(Current)); 1697 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1698 Contexts.back().CaretFound = true; 1699 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1700 Current.setType(determineIncrementUsage(Current)); 1701 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1702 Current.setType(TT_UnaryOperator); 1703 } else if (Current.is(tok::question)) { 1704 if (Style.isJavaScript() && Line.MustBeDeclaration && 1705 !Contexts.back().IsExpression) { 1706 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1707 // on the interface, not a ternary expression. 1708 Current.setType(TT_JsTypeOptionalQuestion); 1709 } else { 1710 Current.setType(TT_ConditionalExpr); 1711 } 1712 } else if (Current.isBinaryOperator() && 1713 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1714 (!Current.is(tok::greater) && 1715 Style.Language != FormatStyle::LK_TextProto)) { 1716 Current.setType(TT_BinaryOperator); 1717 } else if (Current.is(tok::comment)) { 1718 if (Current.TokenText.startswith("/*")) { 1719 if (Current.TokenText.endswith("*/")) 1720 Current.setType(TT_BlockComment); 1721 else 1722 // The lexer has for some reason determined a comment here. But we 1723 // cannot really handle it, if it isn't properly terminated. 1724 Current.Tok.setKind(tok::unknown); 1725 } else { 1726 Current.setType(TT_LineComment); 1727 } 1728 } else if (Current.is(tok::r_paren)) { 1729 if (rParenEndsCast(Current)) 1730 Current.setType(TT_CastRParen); 1731 if (Current.MatchingParen && Current.Next && 1732 !Current.Next->isBinaryOperator() && 1733 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1734 tok::comma, tok::period, tok::arrow, 1735 tok::coloncolon)) 1736 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1737 // Make sure this isn't the return type of an Obj-C block declaration 1738 if (AfterParen->Tok.isNot(tok::caret)) { 1739 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1740 if (BeforeParen->is(tok::identifier) && 1741 !BeforeParen->is(TT_TypenameMacro) && 1742 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1743 (!BeforeParen->Previous || 1744 BeforeParen->Previous->ClosesTemplateDeclaration)) 1745 Current.setType(TT_FunctionAnnotationRParen); 1746 } 1747 } 1748 } else if (Current.is(tok::at) && Current.Next && !Style.isJavaScript() && 1749 Style.Language != FormatStyle::LK_Java) { 1750 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1751 // marks declarations and properties that need special formatting. 1752 switch (Current.Next->Tok.getObjCKeywordID()) { 1753 case tok::objc_interface: 1754 case tok::objc_implementation: 1755 case tok::objc_protocol: 1756 Current.setType(TT_ObjCDecl); 1757 break; 1758 case tok::objc_property: 1759 Current.setType(TT_ObjCProperty); 1760 break; 1761 default: 1762 break; 1763 } 1764 } else if (Current.is(tok::period)) { 1765 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1766 if (PreviousNoComment && 1767 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1768 Current.setType(TT_DesignatedInitializerPeriod); 1769 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1770 Current.Previous->isOneOf(TT_JavaAnnotation, 1771 TT_LeadingJavaAnnotation)) { 1772 Current.setType(Current.Previous->getType()); 1773 } 1774 } else if (canBeObjCSelectorComponent(Current) && 1775 // FIXME(bug 36976): ObjC return types shouldn't use 1776 // TT_CastRParen. 1777 Current.Previous && Current.Previous->is(TT_CastRParen) && 1778 Current.Previous->MatchingParen && 1779 Current.Previous->MatchingParen->Previous && 1780 Current.Previous->MatchingParen->Previous->is( 1781 TT_ObjCMethodSpecifier)) { 1782 // This is the first part of an Objective-C selector name. (If there's no 1783 // colon after this, this is the only place which annotates the identifier 1784 // as a selector.) 1785 Current.setType(TT_SelectorName); 1786 } else if (Current.isOneOf(tok::identifier, tok::kw_const, tok::kw_noexcept, 1787 tok::kw_requires) && 1788 Current.Previous && 1789 !Current.Previous->isOneOf(tok::equal, tok::at) && 1790 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1791 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1792 // function declaration have been found. 1793 Current.setType(TT_TrailingAnnotation); 1794 } else if ((Style.Language == FormatStyle::LK_Java || 1795 Style.isJavaScript()) && 1796 Current.Previous) { 1797 if (Current.Previous->is(tok::at) && 1798 Current.isNot(Keywords.kw_interface)) { 1799 const FormatToken &AtToken = *Current.Previous; 1800 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1801 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1802 Current.setType(TT_LeadingJavaAnnotation); 1803 else 1804 Current.setType(TT_JavaAnnotation); 1805 } else if (Current.Previous->is(tok::period) && 1806 Current.Previous->isOneOf(TT_JavaAnnotation, 1807 TT_LeadingJavaAnnotation)) { 1808 Current.setType(Current.Previous->getType()); 1809 } 1810 } 1811 } 1812 1813 /// Take a guess at whether \p Tok starts a name of a function or 1814 /// variable declaration. 1815 /// 1816 /// This is a heuristic based on whether \p Tok is an identifier following 1817 /// something that is likely a type. 1818 bool isStartOfName(const FormatToken &Tok) { 1819 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1820 return false; 1821 1822 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1823 Keywords.kw_as)) 1824 return false; 1825 if (Style.isJavaScript() && Tok.Previous->is(Keywords.kw_in)) 1826 return false; 1827 1828 // Skip "const" as it does not have an influence on whether this is a name. 1829 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1830 1831 // For javascript const can be like "let" or "var" 1832 if (!Style.isJavaScript()) 1833 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1834 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1835 1836 if (!PreviousNotConst) 1837 return false; 1838 1839 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1840 PreviousNotConst->Previous && 1841 PreviousNotConst->Previous->is(tok::hash); 1842 1843 if (PreviousNotConst->is(TT_TemplateCloser)) 1844 return PreviousNotConst && PreviousNotConst->MatchingParen && 1845 PreviousNotConst->MatchingParen->Previous && 1846 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1847 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1848 1849 if (PreviousNotConst->is(tok::r_paren) && 1850 PreviousNotConst->is(TT_TypeDeclarationParen)) 1851 return true; 1852 1853 // If is a preprocess keyword like #define. 1854 if (IsPPKeyword) 1855 return false; 1856 1857 // int a or auto a. 1858 if (PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) 1859 return true; 1860 1861 // *a or &a or &&a. 1862 if (PreviousNotConst->is(TT_PointerOrReference)) 1863 return true; 1864 1865 // MyClass a; 1866 if (PreviousNotConst->isSimpleTypeSpecifier()) 1867 return true; 1868 1869 // const a = in JavaScript. 1870 return (Style.isJavaScript() && PreviousNotConst->is(tok::kw_const)); 1871 } 1872 1873 /// Determine whether ')' is ending a cast. 1874 bool rParenEndsCast(const FormatToken &Tok) { 1875 // C-style casts are only used in C++, C# and Java. 1876 if (!Style.isCSharp() && !Style.isCpp() && 1877 Style.Language != FormatStyle::LK_Java) 1878 return false; 1879 1880 // Empty parens aren't casts and there are no casts at the end of the line. 1881 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1882 return false; 1883 1884 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1885 if (LeftOfParens) { 1886 // If there is a closing parenthesis left of the current 1887 // parentheses, look past it as these might be chained casts. 1888 if (LeftOfParens->is(tok::r_paren) && 1889 LeftOfParens->isNot(TT_CastRParen)) { 1890 if (!LeftOfParens->MatchingParen || 1891 !LeftOfParens->MatchingParen->Previous) 1892 return false; 1893 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1894 } 1895 1896 // If there is an identifier (or with a few exceptions a keyword) right 1897 // before the parentheses, this is unlikely to be a cast. 1898 if (LeftOfParens->Tok.getIdentifierInfo() && 1899 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1900 tok::kw_delete)) 1901 return false; 1902 1903 // Certain other tokens right before the parentheses are also signals that 1904 // this cannot be a cast. 1905 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1906 TT_TemplateCloser, tok::ellipsis)) 1907 return false; 1908 } 1909 1910 if (Tok.Next->is(tok::question)) 1911 return false; 1912 1913 // `foreach((A a, B b) in someList)` should not be seen as a cast. 1914 if (Tok.Next->is(Keywords.kw_in) && Style.isCSharp()) 1915 return false; 1916 1917 // Functions which end with decorations like volatile, noexcept are unlikely 1918 // to be casts. 1919 if (Tok.Next->isOneOf(tok::kw_noexcept, tok::kw_volatile, tok::kw_const, 1920 tok::kw_requires, tok::kw_throw, tok::arrow, 1921 Keywords.kw_override, Keywords.kw_final) || 1922 isCpp11AttributeSpecifier(*Tok.Next)) 1923 return false; 1924 1925 // As Java has no function types, a "(" after the ")" likely means that this 1926 // is a cast. 1927 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1928 return true; 1929 1930 // If a (non-string) literal follows, this is likely a cast. 1931 if (Tok.Next->isNot(tok::string_literal) && 1932 (Tok.Next->Tok.isLiteral() || 1933 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1934 return true; 1935 1936 // Heuristically try to determine whether the parentheses contain a type. 1937 auto IsQualifiedPointerOrReference = [](FormatToken *T) { 1938 // This is used to handle cases such as x = (foo *const)&y; 1939 assert(!T->isSimpleTypeSpecifier() && "Should have already been checked"); 1940 // Strip trailing qualifiers such as const or volatile when checking 1941 // whether the parens could be a cast to a pointer/reference type. 1942 while (T) { 1943 if (T->is(TT_AttributeParen)) { 1944 // Handle `x = (foo *__attribute__((foo)))&v;`: 1945 if (T->MatchingParen && T->MatchingParen->Previous && 1946 T->MatchingParen->Previous->is(tok::kw___attribute)) { 1947 T = T->MatchingParen->Previous->Previous; 1948 continue; 1949 } 1950 } else if (T->is(TT_AttributeSquare)) { 1951 // Handle `x = (foo *[[clang::foo]])&v;`: 1952 if (T->MatchingParen && T->MatchingParen->Previous) { 1953 T = T->MatchingParen->Previous; 1954 continue; 1955 } 1956 } else if (T->canBePointerOrReferenceQualifier()) { 1957 T = T->Previous; 1958 continue; 1959 } 1960 break; 1961 } 1962 return T && T->is(TT_PointerOrReference); 1963 }; 1964 bool ParensAreType = 1965 !Tok.Previous || 1966 Tok.Previous->isOneOf(TT_TemplateCloser, TT_TypeDeclarationParen) || 1967 Tok.Previous->isSimpleTypeSpecifier() || 1968 IsQualifiedPointerOrReference(Tok.Previous); 1969 bool ParensCouldEndDecl = 1970 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1971 if (ParensAreType && !ParensCouldEndDecl) 1972 return true; 1973 1974 // At this point, we heuristically assume that there are no casts at the 1975 // start of the line. We assume that we have found most cases where there 1976 // are by the logic above, e.g. "(void)x;". 1977 if (!LeftOfParens) 1978 return false; 1979 1980 // Certain token types inside the parentheses mean that this can't be a 1981 // cast. 1982 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 1983 Token = Token->Next) 1984 if (Token->is(TT_BinaryOperator)) 1985 return false; 1986 1987 // If the following token is an identifier or 'this', this is a cast. All 1988 // cases where this can be something else are handled above. 1989 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1990 return true; 1991 1992 // Look for a cast `( x ) (`. 1993 if (Tok.Next->is(tok::l_paren) && Tok.Previous && Tok.Previous->Previous) { 1994 if (Tok.Previous->is(tok::identifier) && 1995 Tok.Previous->Previous->is(tok::l_paren)) 1996 return true; 1997 } 1998 1999 if (!Tok.Next->Next) 2000 return false; 2001 2002 // If the next token after the parenthesis is a unary operator, assume 2003 // that this is cast, unless there are unexpected tokens inside the 2004 // parenthesis. 2005 bool NextIsUnary = 2006 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 2007 if (!NextIsUnary || Tok.Next->is(tok::plus) || 2008 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 2009 return false; 2010 // Search for unexpected tokens. 2011 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 2012 Prev = Prev->Previous) { 2013 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 2014 return false; 2015 } 2016 return true; 2017 } 2018 2019 /// Return the type of the given token assuming it is * or &. 2020 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 2021 bool InTemplateArgument) { 2022 if (Style.isJavaScript()) 2023 return TT_BinaryOperator; 2024 2025 // && in C# must be a binary operator. 2026 if (Style.isCSharp() && Tok.is(tok::ampamp)) 2027 return TT_BinaryOperator; 2028 2029 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2030 if (!PrevToken) 2031 return TT_UnaryOperator; 2032 2033 const FormatToken *NextToken = Tok.getNextNonComment(); 2034 if (!NextToken || 2035 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_noexcept) || 2036 NextToken->canBePointerOrReferenceQualifier() || 2037 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 2038 return TT_PointerOrReference; 2039 2040 if (PrevToken->is(tok::coloncolon)) 2041 return TT_PointerOrReference; 2042 2043 if (PrevToken->is(tok::r_paren) && PrevToken->is(TT_TypeDeclarationParen)) 2044 return TT_PointerOrReference; 2045 2046 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 2047 tok::comma, tok::semi, tok::kw_return, tok::colon, 2048 tok::kw_co_return, tok::kw_co_await, 2049 tok::kw_co_yield, tok::equal, tok::kw_delete, 2050 tok::kw_sizeof, tok::kw_throw, TT_BinaryOperator, 2051 TT_ConditionalExpr, TT_UnaryOperator, TT_CastRParen)) 2052 return TT_UnaryOperator; 2053 2054 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 2055 return TT_PointerOrReference; 2056 if (NextToken->is(tok::kw_operator) && !IsExpression) 2057 return TT_PointerOrReference; 2058 if (NextToken->isOneOf(tok::comma, tok::semi)) 2059 return TT_PointerOrReference; 2060 2061 if (PrevToken->Tok.isLiteral() || 2062 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 2063 tok::kw_false, tok::r_brace) || 2064 NextToken->Tok.isLiteral() || 2065 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 2066 NextToken->isUnaryOperator() || 2067 // If we know we're in a template argument, there are no named 2068 // declarations. Thus, having an identifier on the right-hand side 2069 // indicates a binary operator. 2070 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 2071 return TT_BinaryOperator; 2072 2073 // "&&(" is quite unlikely to be two successive unary "&". 2074 if (Tok.is(tok::ampamp) && NextToken->is(tok::l_paren)) 2075 return TT_BinaryOperator; 2076 2077 // This catches some cases where evaluation order is used as control flow: 2078 // aaa && aaa->f(); 2079 if (NextToken->Tok.isAnyIdentifier()) { 2080 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 2081 if (NextNextToken && NextNextToken->is(tok::arrow)) 2082 return TT_BinaryOperator; 2083 } 2084 2085 // It is very unlikely that we are going to find a pointer or reference type 2086 // definition on the RHS of an assignment. 2087 if (IsExpression && !Contexts.back().CaretFound) 2088 return TT_BinaryOperator; 2089 2090 return TT_PointerOrReference; 2091 } 2092 2093 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 2094 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2095 if (!PrevToken) 2096 return TT_UnaryOperator; 2097 2098 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 2099 // This must be a sequence of leading unary operators. 2100 return TT_UnaryOperator; 2101 2102 // Use heuristics to recognize unary operators. 2103 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 2104 tok::question, tok::colon, tok::kw_return, 2105 tok::kw_case, tok::at, tok::l_brace, tok::kw_throw, 2106 tok::kw_co_return, tok::kw_co_yield)) 2107 return TT_UnaryOperator; 2108 2109 // There can't be two consecutive binary operators. 2110 if (PrevToken->is(TT_BinaryOperator)) 2111 return TT_UnaryOperator; 2112 2113 // Fall back to marking the token as binary operator. 2114 return TT_BinaryOperator; 2115 } 2116 2117 /// Determine whether ++/-- are pre- or post-increments/-decrements. 2118 TokenType determineIncrementUsage(const FormatToken &Tok) { 2119 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2120 if (!PrevToken || PrevToken->is(TT_CastRParen)) 2121 return TT_UnaryOperator; 2122 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 2123 return TT_TrailingUnaryOperator; 2124 2125 return TT_UnaryOperator; 2126 } 2127 2128 SmallVector<Context, 8> Contexts; 2129 2130 const FormatStyle &Style; 2131 AnnotatedLine &Line; 2132 FormatToken *CurrentToken; 2133 bool AutoFound; 2134 const AdditionalKeywords &Keywords; 2135 2136 // Set of "<" tokens that do not open a template parameter list. If parseAngle 2137 // determines that a specific token can't be a template opener, it will make 2138 // same decision irrespective of the decisions for tokens leading up to it. 2139 // Store this information to prevent this from causing exponential runtime. 2140 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 2141 }; 2142 2143 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 2144 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 2145 2146 /// Parses binary expressions by inserting fake parenthesis based on 2147 /// operator precedence. 2148 class ExpressionParser { 2149 public: 2150 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 2151 AnnotatedLine &Line) 2152 : Style(Style), Keywords(Keywords), Current(Line.First) {} 2153 2154 /// Parse expressions with the given operator precedence. 2155 void parse(int Precedence = 0) { 2156 // Skip 'return' and ObjC selector colons as they are not part of a binary 2157 // expression. 2158 while (Current && (Current->is(tok::kw_return) || 2159 (Current->is(tok::colon) && 2160 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 2161 next(); 2162 2163 if (!Current || Precedence > PrecedenceArrowAndPeriod) 2164 return; 2165 2166 // Conditional expressions need to be parsed separately for proper nesting. 2167 if (Precedence == prec::Conditional) { 2168 parseConditionalExpr(); 2169 return; 2170 } 2171 2172 // Parse unary operators, which all have a higher precedence than binary 2173 // operators. 2174 if (Precedence == PrecedenceUnaryOperator) { 2175 parseUnaryOperator(); 2176 return; 2177 } 2178 2179 FormatToken *Start = Current; 2180 FormatToken *LatestOperator = nullptr; 2181 unsigned OperatorIndex = 0; 2182 2183 while (Current) { 2184 // Consume operators with higher precedence. 2185 parse(Precedence + 1); 2186 2187 int CurrentPrecedence = getCurrentPrecedence(); 2188 2189 if (Precedence == CurrentPrecedence && Current && 2190 Current->is(TT_SelectorName)) { 2191 if (LatestOperator) 2192 addFakeParenthesis(Start, prec::Level(Precedence)); 2193 Start = Current; 2194 } 2195 2196 // At the end of the line or when an operator with higher precedence is 2197 // found, insert fake parenthesis and return. 2198 if (!Current || 2199 (Current->closesScope() && 2200 (Current->MatchingParen || Current->is(TT_TemplateString))) || 2201 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 2202 (CurrentPrecedence == prec::Conditional && 2203 Precedence == prec::Assignment && Current->is(tok::colon))) { 2204 break; 2205 } 2206 2207 // Consume scopes: (), [], <> and {} 2208 if (Current->opensScope()) { 2209 // In fragment of a JavaScript template string can look like '}..${' and 2210 // thus close a scope and open a new one at the same time. 2211 while (Current && (!Current->closesScope() || Current->opensScope())) { 2212 next(); 2213 parse(); 2214 } 2215 next(); 2216 } else { 2217 // Operator found. 2218 if (CurrentPrecedence == Precedence) { 2219 if (LatestOperator) 2220 LatestOperator->NextOperator = Current; 2221 LatestOperator = Current; 2222 Current->OperatorIndex = OperatorIndex; 2223 ++OperatorIndex; 2224 } 2225 next(/*SkipPastLeadingComments=*/Precedence > 0); 2226 } 2227 } 2228 2229 if (LatestOperator && (Current || Precedence > 0)) { 2230 // LatestOperator->LastOperator = true; 2231 if (Precedence == PrecedenceArrowAndPeriod) { 2232 // Call expressions don't have a binary operator precedence. 2233 addFakeParenthesis(Start, prec::Unknown); 2234 } else { 2235 addFakeParenthesis(Start, prec::Level(Precedence)); 2236 } 2237 } 2238 } 2239 2240 private: 2241 /// Gets the precedence (+1) of the given token for binary operators 2242 /// and other tokens that we treat like binary operators. 2243 int getCurrentPrecedence() { 2244 if (Current) { 2245 const FormatToken *NextNonComment = Current->getNextNonComment(); 2246 if (Current->is(TT_ConditionalExpr)) 2247 return prec::Conditional; 2248 if (NextNonComment && Current->is(TT_SelectorName) && 2249 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 2250 ((Style.Language == FormatStyle::LK_Proto || 2251 Style.Language == FormatStyle::LK_TextProto) && 2252 NextNonComment->is(tok::less)))) 2253 return prec::Assignment; 2254 if (Current->is(TT_JsComputedPropertyName)) 2255 return prec::Assignment; 2256 if (Current->is(TT_LambdaArrow)) 2257 return prec::Comma; 2258 if (Current->is(TT_FatArrow)) 2259 return prec::Assignment; 2260 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 2261 (Current->is(tok::comment) && NextNonComment && 2262 NextNonComment->is(TT_SelectorName))) 2263 return 0; 2264 if (Current->is(TT_RangeBasedForLoopColon)) 2265 return prec::Comma; 2266 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2267 Current->is(Keywords.kw_instanceof)) 2268 return prec::Relational; 2269 if (Style.isJavaScript() && 2270 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 2271 return prec::Relational; 2272 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 2273 return Current->getPrecedence(); 2274 if (Current->isOneOf(tok::period, tok::arrow)) 2275 return PrecedenceArrowAndPeriod; 2276 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2277 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 2278 Keywords.kw_throws)) 2279 return 0; 2280 } 2281 return -1; 2282 } 2283 2284 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 2285 Start->FakeLParens.push_back(Precedence); 2286 if (Precedence > prec::Unknown) 2287 Start->StartsBinaryExpression = true; 2288 if (Current) { 2289 FormatToken *Previous = Current->Previous; 2290 while (Previous->is(tok::comment) && Previous->Previous) 2291 Previous = Previous->Previous; 2292 ++Previous->FakeRParens; 2293 if (Precedence > prec::Unknown) 2294 Previous->EndsBinaryExpression = true; 2295 } 2296 } 2297 2298 /// Parse unary operator expressions and surround them with fake 2299 /// parentheses if appropriate. 2300 void parseUnaryOperator() { 2301 llvm::SmallVector<FormatToken *, 2> Tokens; 2302 while (Current && Current->is(TT_UnaryOperator)) { 2303 Tokens.push_back(Current); 2304 next(); 2305 } 2306 parse(PrecedenceArrowAndPeriod); 2307 for (FormatToken *Token : llvm::reverse(Tokens)) 2308 // The actual precedence doesn't matter. 2309 addFakeParenthesis(Token, prec::Unknown); 2310 } 2311 2312 void parseConditionalExpr() { 2313 while (Current && Current->isTrailingComment()) { 2314 next(); 2315 } 2316 FormatToken *Start = Current; 2317 parse(prec::LogicalOr); 2318 if (!Current || !Current->is(tok::question)) 2319 return; 2320 next(); 2321 parse(prec::Assignment); 2322 if (!Current || Current->isNot(TT_ConditionalExpr)) 2323 return; 2324 next(); 2325 parse(prec::Assignment); 2326 addFakeParenthesis(Start, prec::Conditional); 2327 } 2328 2329 void next(bool SkipPastLeadingComments = true) { 2330 if (Current) 2331 Current = Current->Next; 2332 while (Current && 2333 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 2334 Current->isTrailingComment()) 2335 Current = Current->Next; 2336 } 2337 2338 const FormatStyle &Style; 2339 const AdditionalKeywords &Keywords; 2340 FormatToken *Current; 2341 }; 2342 2343 } // end anonymous namespace 2344 2345 void TokenAnnotator::setCommentLineLevels( 2346 SmallVectorImpl<AnnotatedLine *> &Lines) { 2347 const AnnotatedLine *NextNonCommentLine = nullptr; 2348 for (AnnotatedLine *AL : llvm::reverse(Lines)) { 2349 bool CommentLine = true; 2350 for (const FormatToken *Tok = AL->First; Tok; Tok = Tok->Next) { 2351 if (!Tok->is(tok::comment)) { 2352 CommentLine = false; 2353 break; 2354 } 2355 } 2356 2357 // If the comment is currently aligned with the line immediately following 2358 // it, that's probably intentional and we should keep it. 2359 if (NextNonCommentLine && CommentLine && 2360 NextNonCommentLine->First->NewlinesBefore <= 1 && 2361 NextNonCommentLine->First->OriginalColumn == 2362 AL->First->OriginalColumn) { 2363 // Align comments for preprocessor lines with the # in column 0 if 2364 // preprocessor lines are not indented. Otherwise, align with the next 2365 // line. 2366 AL->Level = (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 2367 (NextNonCommentLine->Type == LT_PreprocessorDirective || 2368 NextNonCommentLine->Type == LT_ImportStatement)) 2369 ? 0 2370 : NextNonCommentLine->Level; 2371 } else { 2372 NextNonCommentLine = AL->First->isNot(tok::r_brace) ? AL : nullptr; 2373 } 2374 2375 setCommentLineLevels(AL->Children); 2376 } 2377 } 2378 2379 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 2380 unsigned Result = 0; 2381 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 2382 Result = std::max(Result, Tok->NestingLevel); 2383 return Result; 2384 } 2385 2386 void TokenAnnotator::annotate(AnnotatedLine &Line) { 2387 for (auto &Child : Line.Children) 2388 annotate(*Child); 2389 2390 AnnotatingParser Parser(Style, Line, Keywords); 2391 Line.Type = Parser.parseLine(); 2392 2393 // With very deep nesting, ExpressionParser uses lots of stack and the 2394 // formatting algorithm is very slow. We're not going to do a good job here 2395 // anyway - it's probably generated code being formatted by mistake. 2396 // Just skip the whole line. 2397 if (maxNestingDepth(Line) > 50) 2398 Line.Type = LT_Invalid; 2399 2400 if (Line.Type == LT_Invalid) 2401 return; 2402 2403 ExpressionParser ExprParser(Style, Keywords, Line); 2404 ExprParser.parse(); 2405 2406 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2407 Line.Type = LT_ObjCMethodDecl; 2408 else if (Line.startsWith(TT_ObjCDecl)) 2409 Line.Type = LT_ObjCDecl; 2410 else if (Line.startsWith(TT_ObjCProperty)) 2411 Line.Type = LT_ObjCProperty; 2412 2413 Line.First->SpacesRequiredBefore = 1; 2414 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 2415 } 2416 2417 // This function heuristically determines whether 'Current' starts the name of a 2418 // function declaration. 2419 static bool isFunctionDeclarationName(bool IsCpp, const FormatToken &Current, 2420 const AnnotatedLine &Line) { 2421 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 2422 for (; Next; Next = Next->Next) { 2423 if (Next->is(TT_OverloadedOperatorLParen)) 2424 return Next; 2425 if (Next->is(TT_OverloadedOperator)) 2426 continue; 2427 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 2428 // For 'new[]' and 'delete[]'. 2429 if (Next->Next && 2430 Next->Next->startsSequence(tok::l_square, tok::r_square)) 2431 Next = Next->Next->Next; 2432 continue; 2433 } 2434 if (Next->startsSequence(tok::l_square, tok::r_square)) { 2435 // For operator[](). 2436 Next = Next->Next; 2437 continue; 2438 } 2439 if ((Next->isSimpleTypeSpecifier() || Next->is(tok::identifier)) && 2440 Next->Next && Next->Next->isOneOf(tok::star, tok::amp, tok::ampamp)) { 2441 // For operator void*(), operator char*(), operator Foo*(). 2442 Next = Next->Next; 2443 continue; 2444 } 2445 if (Next->is(TT_TemplateOpener) && Next->MatchingParen) { 2446 Next = Next->MatchingParen; 2447 continue; 2448 } 2449 2450 break; 2451 } 2452 return nullptr; 2453 }; 2454 2455 // Find parentheses of parameter list. 2456 const FormatToken *Next = Current.Next; 2457 if (Current.is(tok::kw_operator)) { 2458 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 2459 return false; 2460 Next = skipOperatorName(Next); 2461 } else { 2462 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 2463 return false; 2464 for (; Next; Next = Next->Next) { 2465 if (Next->is(TT_TemplateOpener)) { 2466 Next = Next->MatchingParen; 2467 } else if (Next->is(tok::coloncolon)) { 2468 Next = Next->Next; 2469 if (!Next) 2470 return false; 2471 if (Next->is(tok::kw_operator)) { 2472 Next = skipOperatorName(Next->Next); 2473 break; 2474 } 2475 if (!Next->is(tok::identifier)) 2476 return false; 2477 } else if (Next->is(tok::l_paren)) { 2478 break; 2479 } else { 2480 return false; 2481 } 2482 } 2483 } 2484 2485 // Check whether parameter list can belong to a function declaration. 2486 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 2487 return false; 2488 // If the lines ends with "{", this is likely a function definition. 2489 if (Line.Last->is(tok::l_brace)) 2490 return true; 2491 if (Next->Next == Next->MatchingParen) 2492 return true; // Empty parentheses. 2493 // If there is an &/&& after the r_paren, this is likely a function. 2494 if (Next->MatchingParen->Next && 2495 Next->MatchingParen->Next->is(TT_PointerOrReference)) 2496 return true; 2497 2498 // Check for K&R C function definitions (and C++ function definitions with 2499 // unnamed parameters), e.g.: 2500 // int f(i) 2501 // { 2502 // return i + 1; 2503 // } 2504 // bool g(size_t = 0, bool b = false) 2505 // { 2506 // return !b; 2507 // } 2508 if (IsCpp && Next->Next && Next->Next->is(tok::identifier) && 2509 !Line.endsWith(tok::semi)) 2510 return true; 2511 2512 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 2513 Tok = Tok->Next) { 2514 if (Tok->is(TT_TypeDeclarationParen)) 2515 return true; 2516 if (Tok->isOneOf(tok::l_paren, TT_TemplateOpener) && Tok->MatchingParen) { 2517 Tok = Tok->MatchingParen; 2518 continue; 2519 } 2520 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 2521 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 2522 return true; 2523 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 2524 Tok->Tok.isLiteral()) 2525 return false; 2526 } 2527 return false; 2528 } 2529 2530 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 2531 assert(Line.MightBeFunctionDecl); 2532 2533 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 2534 Style.AlwaysBreakAfterReturnType == 2535 FormatStyle::RTBS_TopLevelDefinitions) && 2536 Line.Level > 0) 2537 return false; 2538 2539 switch (Style.AlwaysBreakAfterReturnType) { 2540 case FormatStyle::RTBS_None: 2541 return false; 2542 case FormatStyle::RTBS_All: 2543 case FormatStyle::RTBS_TopLevel: 2544 return true; 2545 case FormatStyle::RTBS_AllDefinitions: 2546 case FormatStyle::RTBS_TopLevelDefinitions: 2547 return Line.mightBeFunctionDefinition(); 2548 } 2549 2550 return false; 2551 } 2552 2553 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 2554 for (AnnotatedLine *ChildLine : Line.Children) 2555 calculateFormattingInformation(*ChildLine); 2556 2557 Line.First->TotalLength = 2558 Line.First->IsMultiline ? Style.ColumnLimit 2559 : Line.FirstStartColumn + Line.First->ColumnWidth; 2560 FormatToken *Current = Line.First->Next; 2561 bool InFunctionDecl = Line.MightBeFunctionDecl; 2562 bool AlignArrayOfStructures = 2563 (Style.AlignArrayOfStructures != FormatStyle::AIAS_None && 2564 Line.Type == LT_ArrayOfStructInitializer); 2565 if (AlignArrayOfStructures) 2566 calculateArrayInitializerColumnList(Line); 2567 2568 while (Current) { 2569 if (isFunctionDeclarationName(Style.isCpp(), *Current, Line)) 2570 Current->setType(TT_FunctionDeclarationName); 2571 const FormatToken *Prev = Current->Previous; 2572 if (Current->is(TT_LineComment)) { 2573 if (Prev->is(BK_BracedInit) && Prev->opensScope()) 2574 Current->SpacesRequiredBefore = 2575 (Style.Cpp11BracedListStyle && !Style.SpacesInParentheses) ? 0 : 1; 2576 else 2577 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2578 2579 // If we find a trailing comment, iterate backwards to determine whether 2580 // it seems to relate to a specific parameter. If so, break before that 2581 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2582 // to the previous line in: 2583 // SomeFunction(a, 2584 // b, // comment 2585 // c); 2586 if (!Current->HasUnescapedNewline) { 2587 for (FormatToken *Parameter = Current->Previous; Parameter; 2588 Parameter = Parameter->Previous) { 2589 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 2590 break; 2591 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 2592 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 2593 Parameter->HasUnescapedNewline) 2594 Parameter->MustBreakBefore = true; 2595 break; 2596 } 2597 } 2598 } 2599 } else if (Current->SpacesRequiredBefore == 0 && 2600 spaceRequiredBefore(Line, *Current)) { 2601 Current->SpacesRequiredBefore = 1; 2602 } 2603 2604 Current->MustBreakBefore = 2605 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 2606 2607 if (!Current->MustBreakBefore && InFunctionDecl && 2608 Current->is(TT_FunctionDeclarationName)) 2609 Current->MustBreakBefore = mustBreakForReturnType(Line); 2610 2611 Current->CanBreakBefore = 2612 Current->MustBreakBefore || canBreakBefore(Line, *Current); 2613 unsigned ChildSize = 0; 2614 if (Prev->Children.size() == 1) { 2615 FormatToken &LastOfChild = *Prev->Children[0]->Last; 2616 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 2617 : LastOfChild.TotalLength + 1; 2618 } 2619 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 2620 (Prev->Children.size() == 1 && 2621 Prev->Children[0]->First->MustBreakBefore) || 2622 Current->IsMultiline) 2623 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 2624 else 2625 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 2626 ChildSize + Current->SpacesRequiredBefore; 2627 2628 if (Current->is(TT_CtorInitializerColon)) 2629 InFunctionDecl = false; 2630 2631 // FIXME: Only calculate this if CanBreakBefore is true once static 2632 // initializers etc. are sorted out. 2633 // FIXME: Move magic numbers to a better place. 2634 2635 // Reduce penalty for aligning ObjC method arguments using the colon 2636 // alignment as this is the canonical way (still prefer fitting everything 2637 // into one line if possible). Trying to fit a whole expression into one 2638 // line should not force other line breaks (e.g. when ObjC method 2639 // expression is a part of other expression). 2640 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl); 2641 if (Style.Language == FormatStyle::LK_ObjC && 2642 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) { 2643 if (Current->ParameterIndex == 1) 2644 Current->SplitPenalty += 5 * Current->BindingStrength; 2645 } else { 2646 Current->SplitPenalty += 20 * Current->BindingStrength; 2647 } 2648 2649 Current = Current->Next; 2650 } 2651 2652 calculateUnbreakableTailLengths(Line); 2653 unsigned IndentLevel = Line.Level; 2654 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2655 if (Current->Role) 2656 Current->Role->precomputeFormattingInfos(Current); 2657 if (Current->MatchingParen && 2658 Current->MatchingParen->opensBlockOrBlockTypeList(Style) && 2659 IndentLevel > 0) { 2660 --IndentLevel; 2661 } 2662 Current->IndentLevel = IndentLevel; 2663 if (Current->opensBlockOrBlockTypeList(Style)) 2664 ++IndentLevel; 2665 } 2666 2667 LLVM_DEBUG({ printDebugInfo(Line); }); 2668 } 2669 2670 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2671 unsigned UnbreakableTailLength = 0; 2672 FormatToken *Current = Line.Last; 2673 while (Current) { 2674 Current->UnbreakableTailLength = UnbreakableTailLength; 2675 if (Current->CanBreakBefore || 2676 Current->isOneOf(tok::comment, tok::string_literal)) { 2677 UnbreakableTailLength = 0; 2678 } else { 2679 UnbreakableTailLength += 2680 Current->ColumnWidth + Current->SpacesRequiredBefore; 2681 } 2682 Current = Current->Previous; 2683 } 2684 } 2685 2686 void TokenAnnotator::calculateArrayInitializerColumnList(AnnotatedLine &Line) { 2687 if (Line.First == Line.Last) { 2688 return; 2689 } 2690 auto *CurrentToken = Line.First; 2691 CurrentToken->ArrayInitializerLineStart = true; 2692 unsigned Depth = 0; 2693 while (CurrentToken != nullptr && CurrentToken != Line.Last) { 2694 if (CurrentToken->is(tok::l_brace)) { 2695 CurrentToken->IsArrayInitializer = true; 2696 if (CurrentToken->Next != nullptr) 2697 CurrentToken->Next->MustBreakBefore = true; 2698 CurrentToken = 2699 calculateInitializerColumnList(Line, CurrentToken->Next, Depth + 1); 2700 } else { 2701 CurrentToken = CurrentToken->Next; 2702 } 2703 } 2704 } 2705 2706 FormatToken *TokenAnnotator::calculateInitializerColumnList( 2707 AnnotatedLine &Line, FormatToken *CurrentToken, unsigned Depth) { 2708 while (CurrentToken != nullptr && CurrentToken != Line.Last) { 2709 if (CurrentToken->is(tok::l_brace)) 2710 ++Depth; 2711 else if (CurrentToken->is(tok::r_brace)) 2712 --Depth; 2713 if (Depth == 2 && CurrentToken->isOneOf(tok::l_brace, tok::comma)) { 2714 CurrentToken = CurrentToken->Next; 2715 if (CurrentToken == nullptr) 2716 break; 2717 CurrentToken->StartsColumn = true; 2718 CurrentToken = CurrentToken->Previous; 2719 } 2720 CurrentToken = CurrentToken->Next; 2721 } 2722 return CurrentToken; 2723 } 2724 2725 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2726 const FormatToken &Tok, 2727 bool InFunctionDecl) { 2728 const FormatToken &Left = *Tok.Previous; 2729 const FormatToken &Right = Tok; 2730 2731 if (Left.is(tok::semi)) 2732 return 0; 2733 2734 if (Style.Language == FormatStyle::LK_Java) { 2735 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2736 return 1; 2737 if (Right.is(Keywords.kw_implements)) 2738 return 2; 2739 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2740 return 3; 2741 } else if (Style.isJavaScript()) { 2742 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2743 return 100; 2744 if (Left.is(TT_JsTypeColon)) 2745 return 35; 2746 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2747 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2748 return 100; 2749 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2750 if (Left.opensScope() && Right.closesScope()) 2751 return 200; 2752 } 2753 2754 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2755 return 1; 2756 if (Right.is(tok::l_square)) { 2757 if (Style.Language == FormatStyle::LK_Proto) 2758 return 1; 2759 if (Left.is(tok::r_square)) 2760 return 200; 2761 // Slightly prefer formatting local lambda definitions like functions. 2762 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2763 return 35; 2764 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2765 TT_ArrayInitializerLSquare, 2766 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) 2767 return 500; 2768 } 2769 2770 if (Left.is(tok::coloncolon) || 2771 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2772 return 500; 2773 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2774 Right.is(tok::kw_operator)) { 2775 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2776 return 3; 2777 if (Left.is(TT_StartOfName)) 2778 return 110; 2779 if (InFunctionDecl && Right.NestingLevel == 0) 2780 return Style.PenaltyReturnTypeOnItsOwnLine; 2781 return 200; 2782 } 2783 if (Right.is(TT_PointerOrReference)) 2784 return 190; 2785 if (Right.is(TT_LambdaArrow)) 2786 return 110; 2787 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2788 return 160; 2789 if (Left.is(TT_CastRParen)) 2790 return 100; 2791 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2792 return 5000; 2793 if (Left.is(tok::comment)) 2794 return 1000; 2795 2796 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2797 TT_CtorInitializerColon)) 2798 return 2; 2799 2800 if (Right.isMemberAccess()) { 2801 // Breaking before the "./->" of a chained call/member access is reasonably 2802 // cheap, as formatting those with one call per line is generally 2803 // desirable. In particular, it should be cheaper to break before the call 2804 // than it is to break inside a call's parameters, which could lead to weird 2805 // "hanging" indents. The exception is the very last "./->" to support this 2806 // frequent pattern: 2807 // 2808 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2809 // dddddddd); 2810 // 2811 // which might otherwise be blown up onto many lines. Here, clang-format 2812 // won't produce "hanging" indents anyway as there is no other trailing 2813 // call. 2814 // 2815 // Also apply higher penalty is not a call as that might lead to a wrapping 2816 // like: 2817 // 2818 // aaaaaaa 2819 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2820 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2821 ? 150 2822 : 35; 2823 } 2824 2825 if (Right.is(TT_TrailingAnnotation) && 2826 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2827 // Moving trailing annotations to the next line is fine for ObjC method 2828 // declarations. 2829 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2830 return 10; 2831 // Generally, breaking before a trailing annotation is bad unless it is 2832 // function-like. It seems to be especially preferable to keep standard 2833 // annotations (i.e. "const", "final" and "override") on the same line. 2834 // Use a slightly higher penalty after ")" so that annotations like 2835 // "const override" are kept together. 2836 bool is_short_annotation = Right.TokenText.size() < 10; 2837 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 2838 } 2839 2840 // In for-loops, prefer breaking at ',' and ';'. 2841 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 2842 return 4; 2843 2844 // In Objective-C method expressions, prefer breaking before "param:" over 2845 // breaking after it. 2846 if (Right.is(TT_SelectorName)) 2847 return 0; 2848 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 2849 return Line.MightBeFunctionDecl ? 50 : 500; 2850 2851 // In Objective-C type declarations, avoid breaking after the category's 2852 // open paren (we'll prefer breaking after the protocol list's opening 2853 // angle bracket, if present). 2854 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous && 2855 Left.Previous->isOneOf(tok::identifier, tok::greater)) 2856 return 500; 2857 2858 if (Left.is(tok::l_paren) && Style.PenaltyBreakOpenParenthesis != 0) 2859 return Style.PenaltyBreakOpenParenthesis; 2860 if (Left.is(tok::l_paren) && InFunctionDecl && 2861 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 2862 return 100; 2863 if (Left.is(tok::l_paren) && Left.Previous && 2864 (Left.Previous->is(tok::kw_for) || Left.Previous->isIf())) 2865 return 1000; 2866 if (Left.is(tok::equal) && InFunctionDecl) 2867 return 110; 2868 if (Right.is(tok::r_brace)) 2869 return 1; 2870 if (Left.is(TT_TemplateOpener)) 2871 return 100; 2872 if (Left.opensScope()) { 2873 // If we aren't aligning after opening parens/braces we can always break 2874 // here unless the style does not want us to place all arguments on the 2875 // next line. 2876 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign && 2877 (Left.ParameterCount <= 1 || Style.AllowAllArgumentsOnNextLine)) 2878 return 0; 2879 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle) 2880 return 19; 2881 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 2882 : 19; 2883 } 2884 if (Left.is(TT_JavaAnnotation)) 2885 return 50; 2886 2887 if (Left.is(TT_UnaryOperator)) 2888 return 60; 2889 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 2890 Left.Previous->isLabelString() && 2891 (Left.NextOperator || Left.OperatorIndex != 0)) 2892 return 50; 2893 if (Right.is(tok::plus) && Left.isLabelString() && 2894 (Right.NextOperator || Right.OperatorIndex != 0)) 2895 return 25; 2896 if (Left.is(tok::comma)) 2897 return 1; 2898 if (Right.is(tok::lessless) && Left.isLabelString() && 2899 (Right.NextOperator || Right.OperatorIndex != 1)) 2900 return 25; 2901 if (Right.is(tok::lessless)) { 2902 // Breaking at a << is really cheap. 2903 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 2904 // Slightly prefer to break before the first one in log-like statements. 2905 return 2; 2906 return 1; 2907 } 2908 if (Left.ClosesTemplateDeclaration) 2909 return Style.PenaltyBreakTemplateDeclaration; 2910 if (Left.is(TT_ConditionalExpr)) 2911 return prec::Conditional; 2912 prec::Level Level = Left.getPrecedence(); 2913 if (Level == prec::Unknown) 2914 Level = Right.getPrecedence(); 2915 if (Level == prec::Assignment) 2916 return Style.PenaltyBreakAssignment; 2917 if (Level != prec::Unknown) 2918 return Level; 2919 2920 return 3; 2921 } 2922 2923 bool TokenAnnotator::spaceRequiredBeforeParens(const FormatToken &Right) const { 2924 if (Style.SpaceBeforeParens == FormatStyle::SBPO_Always) 2925 return true; 2926 if (Right.is(TT_OverloadedOperatorLParen) && 2927 Style.SpaceBeforeParensOptions.AfterOverloadedOperator) 2928 return true; 2929 if (Style.SpaceBeforeParensOptions.BeforeNonEmptyParentheses && 2930 Right.ParameterCount > 0) 2931 return true; 2932 return false; 2933 } 2934 2935 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 2936 const FormatToken &Left, 2937 const FormatToken &Right) { 2938 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 2939 return true; 2940 if (Style.isJson() && Left.is(tok::string_literal) && Right.is(tok::colon)) 2941 return false; 2942 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 2943 return true; 2944 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 2945 Left.Tok.getObjCKeywordID() == tok::objc_property) 2946 return true; 2947 if (Right.is(tok::hashhash)) 2948 return Left.is(tok::hash); 2949 if (Left.isOneOf(tok::hashhash, tok::hash)) 2950 return Right.is(tok::hash); 2951 if ((Left.is(tok::l_paren) && Right.is(tok::r_paren)) || 2952 (Left.is(tok::l_brace) && Left.isNot(BK_Block) && 2953 Right.is(tok::r_brace) && Right.isNot(BK_Block))) 2954 return Style.SpaceInEmptyParentheses; 2955 if (Style.SpacesInConditionalStatement) { 2956 if (Left.is(tok::l_paren) && Left.Previous && 2957 isKeywordWithCondition(*Left.Previous)) 2958 return true; 2959 if (Right.is(tok::r_paren) && Right.MatchingParen && 2960 Right.MatchingParen->Previous && 2961 isKeywordWithCondition(*Right.MatchingParen->Previous)) 2962 return true; 2963 } 2964 2965 // auto{x} auto(x) 2966 if (Left.is(tok::kw_auto) && Right.isOneOf(tok::l_paren, tok::l_brace)) 2967 return false; 2968 2969 // operator co_await(x) 2970 if (Right.is(tok::l_paren) && Left.is(tok::kw_co_await) && Left.Previous && 2971 Left.Previous->is(tok::kw_operator)) 2972 return false; 2973 // co_await (x), co_yield (x), co_return (x) 2974 if (Left.isOneOf(tok::kw_co_await, tok::kw_co_yield, tok::kw_co_return) && 2975 Right.isNot(tok::semi)) 2976 return true; 2977 // requires clause Concept1<T> && Concept2<T> 2978 if (Left.is(TT_ConstraintJunctions) && Right.is(tok::identifier)) 2979 return true; 2980 2981 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 2982 return (Right.is(TT_CastRParen) || 2983 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 2984 ? Style.SpacesInCStyleCastParentheses 2985 : Style.SpacesInParentheses; 2986 if (Right.isOneOf(tok::semi, tok::comma)) 2987 return false; 2988 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) { 2989 bool IsLightweightGeneric = Right.MatchingParen && 2990 Right.MatchingParen->Next && 2991 Right.MatchingParen->Next->is(tok::colon); 2992 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList; 2993 } 2994 if (Right.is(tok::less) && Left.is(tok::kw_template)) 2995 return Style.SpaceAfterTemplateKeyword; 2996 if (Left.isOneOf(tok::exclaim, tok::tilde)) 2997 return false; 2998 if (Left.is(tok::at) && 2999 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 3000 tok::numeric_constant, tok::l_paren, tok::l_brace, 3001 tok::kw_true, tok::kw_false)) 3002 return false; 3003 if (Left.is(tok::colon)) 3004 return !Left.is(TT_ObjCMethodExpr); 3005 if (Left.is(tok::coloncolon)) 3006 return false; 3007 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 3008 if (Style.Language == FormatStyle::LK_TextProto || 3009 (Style.Language == FormatStyle::LK_Proto && 3010 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 3011 // Format empty list as `<>`. 3012 if (Left.is(tok::less) && Right.is(tok::greater)) 3013 return false; 3014 return !Style.Cpp11BracedListStyle; 3015 } 3016 return false; 3017 } 3018 if (Right.is(tok::ellipsis)) 3019 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 3020 Left.Previous->is(tok::kw_case)); 3021 if (Left.is(tok::l_square) && Right.is(tok::amp)) 3022 return Style.SpacesInSquareBrackets; 3023 if (Right.is(TT_PointerOrReference)) { 3024 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 3025 if (!Left.MatchingParen) 3026 return true; 3027 FormatToken *TokenBeforeMatchingParen = 3028 Left.MatchingParen->getPreviousNonComment(); 3029 if (!TokenBeforeMatchingParen || !Left.is(TT_TypeDeclarationParen)) 3030 return true; 3031 } 3032 // Add a space if the previous token is a pointer qualifier or the closing 3033 // parenthesis of __attribute__(()) expression and the style requires spaces 3034 // after pointer qualifiers. 3035 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_After || 3036 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 3037 (Left.is(TT_AttributeParen) || Left.canBePointerOrReferenceQualifier())) 3038 return true; 3039 if (Left.Tok.isLiteral()) 3040 return true; 3041 // for (auto a = 0, b = 0; const auto & c : {1, 2, 3}) 3042 if (Left.isTypeOrIdentifier() && Right.Next && Right.Next->Next && 3043 Right.Next->Next->is(TT_RangeBasedForLoopColon)) 3044 return getTokenPointerOrReferenceAlignment(Right) != 3045 FormatStyle::PAS_Left; 3046 return ( 3047 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 3048 (getTokenPointerOrReferenceAlignment(Right) != FormatStyle::PAS_Left || 3049 (Line.IsMultiVariableDeclStmt && 3050 (Left.NestingLevel == 0 || 3051 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for))))))); 3052 } 3053 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 3054 (!Left.is(TT_PointerOrReference) || 3055 (getTokenPointerOrReferenceAlignment(Left) != FormatStyle::PAS_Right && 3056 !Line.IsMultiVariableDeclStmt))) 3057 return true; 3058 if (Left.is(TT_PointerOrReference)) { 3059 // Add a space if the next token is a pointer qualifier and the style 3060 // requires spaces before pointer qualifiers. 3061 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Before || 3062 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 3063 Right.canBePointerOrReferenceQualifier()) 3064 return true; 3065 // & 1 3066 if (Right.Tok.isLiteral()) 3067 return true; 3068 // & /* comment 3069 if (Right.is(TT_BlockComment)) 3070 return true; 3071 // foo() -> const Bar * override/final 3072 if (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 3073 !Right.is(TT_StartOfName)) 3074 return true; 3075 // & { 3076 if (Right.is(tok::l_brace) && Right.is(BK_Block)) 3077 return true; 3078 // for (auto a = 0, b = 0; const auto& c : {1, 2, 3}) 3079 if (Left.Previous && Left.Previous->isTypeOrIdentifier() && Right.Next && 3080 Right.Next->is(TT_RangeBasedForLoopColon)) 3081 return getTokenPointerOrReferenceAlignment(Left) != 3082 FormatStyle::PAS_Right; 3083 return !Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 3084 tok::l_paren) && 3085 (getTokenPointerOrReferenceAlignment(Left) != 3086 FormatStyle::PAS_Right && 3087 !Line.IsMultiVariableDeclStmt) && 3088 Left.Previous && 3089 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon, 3090 tok::l_square); 3091 } 3092 // Ensure right pointer alignment with ellipsis e.g. int *...P 3093 if (Left.is(tok::ellipsis) && Left.Previous && 3094 Left.Previous->isOneOf(tok::star, tok::amp, tok::ampamp)) 3095 return Style.PointerAlignment != FormatStyle::PAS_Right; 3096 3097 if (Right.is(tok::star) && Left.is(tok::l_paren)) 3098 return false; 3099 if (Left.is(tok::star) && Right.isOneOf(tok::star, tok::amp, tok::ampamp)) 3100 return false; 3101 if (Right.isOneOf(tok::star, tok::amp, tok::ampamp)) { 3102 const FormatToken *Previous = &Left; 3103 while (Previous && !Previous->is(tok::kw_operator)) { 3104 if (Previous->is(tok::identifier) || Previous->isSimpleTypeSpecifier()) { 3105 Previous = Previous->getPreviousNonComment(); 3106 continue; 3107 } 3108 if (Previous->is(TT_TemplateCloser) && Previous->MatchingParen) { 3109 Previous = Previous->MatchingParen->getPreviousNonComment(); 3110 continue; 3111 } 3112 if (Previous->is(tok::coloncolon)) { 3113 Previous = Previous->getPreviousNonComment(); 3114 continue; 3115 } 3116 break; 3117 } 3118 // Space between the type and the * in: 3119 // operator void*() 3120 // operator char*() 3121 // operator void const*() 3122 // operator void volatile*() 3123 // operator /*comment*/ const char*() 3124 // operator volatile /*comment*/ char*() 3125 // operator Foo*() 3126 // operator C<T>*() 3127 // operator std::Foo*() 3128 // operator C<T>::D<U>*() 3129 // dependent on PointerAlignment style. 3130 if (Previous) { 3131 if (Previous->endsSequence(tok::kw_operator)) 3132 return (Style.PointerAlignment != FormatStyle::PAS_Left); 3133 if (Previous->is(tok::kw_const) || Previous->is(tok::kw_volatile)) 3134 return (Style.PointerAlignment != FormatStyle::PAS_Left) || 3135 (Style.SpaceAroundPointerQualifiers == 3136 FormatStyle::SAPQ_After) || 3137 (Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both); 3138 } 3139 } 3140 const auto SpaceRequiredForArrayInitializerLSquare = 3141 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 3142 return Style.SpacesInContainerLiterals || 3143 ((Style.Language == FormatStyle::LK_Proto || 3144 Style.Language == FormatStyle::LK_TextProto) && 3145 !Style.Cpp11BracedListStyle && 3146 LSquareTok.endsSequence(tok::l_square, tok::colon, 3147 TT_SelectorName)); 3148 }; 3149 if (Left.is(tok::l_square)) 3150 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 3151 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 3152 (Left.isOneOf(TT_ArraySubscriptLSquare, TT_StructuredBindingLSquare, 3153 TT_LambdaLSquare) && 3154 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 3155 if (Right.is(tok::r_square)) 3156 return Right.MatchingParen && 3157 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 3158 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 3159 Style)) || 3160 (Style.SpacesInSquareBrackets && 3161 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 3162 TT_StructuredBindingLSquare, 3163 TT_LambdaLSquare)) || 3164 Right.MatchingParen->is(TT_AttributeParen)); 3165 if (Right.is(tok::l_square) && 3166 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 3167 TT_DesignatedInitializerLSquare, 3168 TT_StructuredBindingLSquare, TT_AttributeSquare) && 3169 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral) && 3170 !(!Left.is(tok::r_square) && Style.SpaceBeforeSquareBrackets && 3171 Right.is(TT_ArraySubscriptLSquare))) 3172 return false; 3173 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 3174 return !Left.Children.empty(); // No spaces in "{}". 3175 if ((Left.is(tok::l_brace) && Left.isNot(BK_Block)) || 3176 (Right.is(tok::r_brace) && Right.MatchingParen && 3177 Right.MatchingParen->isNot(BK_Block))) 3178 return Style.Cpp11BracedListStyle ? Style.SpacesInParentheses : true; 3179 if (Left.is(TT_BlockComment)) 3180 // No whitespace in x(/*foo=*/1), except for JavaScript. 3181 return Style.isJavaScript() || !Left.TokenText.endswith("=*/"); 3182 3183 // Space between template and attribute. 3184 // e.g. template <typename T> [[nodiscard]] ... 3185 if (Left.is(TT_TemplateCloser) && Right.is(TT_AttributeSquare)) 3186 return true; 3187 // Space before parentheses common for all languages 3188 if (Right.is(tok::l_paren)) { 3189 if (Left.is(TT_TemplateCloser) && Right.isNot(TT_FunctionTypeLParen)) 3190 return spaceRequiredBeforeParens(Right); 3191 if (Left.is(tok::kw_requires)) 3192 return spaceRequiredBeforeParens(Right); 3193 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 3194 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) 3195 return true; 3196 if (Left.is(TT_ForEachMacro)) 3197 return (Style.SpaceBeforeParensOptions.AfterForeachMacros || 3198 spaceRequiredBeforeParens(Right)); 3199 if (Left.is(TT_IfMacro)) 3200 return (Style.SpaceBeforeParensOptions.AfterIfMacros || 3201 spaceRequiredBeforeParens(Right)); 3202 if (Line.Type == LT_ObjCDecl) 3203 return true; 3204 if (Left.is(tok::semi)) 3205 return true; 3206 if (Left.isOneOf(tok::pp_elif, tok::kw_for, tok::kw_while, tok::kw_switch, 3207 tok::kw_case, TT_ForEachMacro, TT_ObjCForIn)) 3208 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3209 spaceRequiredBeforeParens(Right); 3210 if (Left.isIf(Line.Type != LT_PreprocessorDirective)) 3211 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3212 spaceRequiredBeforeParens(Right); 3213 3214 // TODO add Operator overloading specific Options to 3215 // SpaceBeforeParensOptions 3216 if (Right.is(TT_OverloadedOperatorLParen)) 3217 return spaceRequiredBeforeParens(Right); 3218 // Function declaration or definition 3219 if (Line.MightBeFunctionDecl && (Left.is(TT_FunctionDeclarationName))) { 3220 if (Line.mightBeFunctionDefinition()) 3221 return Style.SpaceBeforeParensOptions.AfterFunctionDefinitionName || 3222 spaceRequiredBeforeParens(Right); 3223 else 3224 return Style.SpaceBeforeParensOptions.AfterFunctionDeclarationName || 3225 spaceRequiredBeforeParens(Right); 3226 } 3227 // Lambda 3228 if (Line.Type != LT_PreprocessorDirective && Left.is(tok::r_square) && 3229 Left.MatchingParen && Left.MatchingParen->is(TT_LambdaLSquare)) 3230 return Style.SpaceBeforeParensOptions.AfterFunctionDefinitionName || 3231 spaceRequiredBeforeParens(Right); 3232 if (!Left.Previous || Left.Previous->isNot(tok::period)) { 3233 if (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch)) 3234 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3235 spaceRequiredBeforeParens(Right); 3236 if (Left.isOneOf(tok::kw_new, tok::kw_delete)) 3237 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never || 3238 spaceRequiredBeforeParens(Right); 3239 } 3240 if (Line.Type != LT_PreprocessorDirective && 3241 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 3242 Left.is(tok::r_paren) || Left.isSimpleTypeSpecifier())) 3243 return spaceRequiredBeforeParens(Right); 3244 return false; 3245 } 3246 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 3247 return false; 3248 if (Right.is(TT_UnaryOperator)) 3249 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 3250 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 3251 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 3252 tok::r_paren) || 3253 Left.isSimpleTypeSpecifier()) && 3254 Right.is(tok::l_brace) && Right.getNextNonComment() && 3255 Right.isNot(BK_Block)) 3256 return false; 3257 if (Left.is(tok::period) || Right.is(tok::period)) 3258 return false; 3259 // u#str, U#str, L#str, u8#str 3260 // uR#str, UR#str, LR#str, u8R#str 3261 if (Right.is(tok::hash) && Left.is(tok::identifier) && 3262 (Left.TokenText == "L" || Left.TokenText == "u" || 3263 Left.TokenText == "U" || Left.TokenText == "u8" || 3264 Left.TokenText == "LR" || Left.TokenText == "uR" || 3265 Left.TokenText == "UR" || Left.TokenText == "u8R")) 3266 return false; 3267 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 3268 Left.MatchingParen->Previous && 3269 (Left.MatchingParen->Previous->is(tok::period) || 3270 Left.MatchingParen->Previous->is(tok::coloncolon))) 3271 // Java call to generic function with explicit type: 3272 // A.<B<C<...>>>DoSomething(); 3273 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference. 3274 return false; 3275 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 3276 return false; 3277 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at)) 3278 // Objective-C dictionary literal -> no space after opening brace. 3279 return false; 3280 if (Right.is(tok::r_brace) && Right.MatchingParen && 3281 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at)) 3282 // Objective-C dictionary literal -> no space before closing brace. 3283 return false; 3284 if (Right.getType() == TT_TrailingAnnotation && 3285 Right.isOneOf(tok::amp, tok::ampamp) && 3286 Left.isOneOf(tok::kw_const, tok::kw_volatile) && 3287 (!Right.Next || Right.Next->is(tok::semi))) 3288 // Match const and volatile ref-qualifiers without any additional 3289 // qualifiers such as 3290 // void Fn() const &; 3291 return getTokenReferenceAlignment(Right) != FormatStyle::PAS_Left; 3292 3293 return true; 3294 } 3295 3296 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 3297 const FormatToken &Right) { 3298 const FormatToken &Left = *Right.Previous; 3299 auto HasExistingWhitespace = [&Right]() { 3300 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 3301 }; 3302 3303 // If the token is finalized don't touch it (as it could be in a 3304 // clang-format-off section). 3305 if (Left.Finalized) 3306 return HasExistingWhitespace(); 3307 3308 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 3309 return true; // Never ever merge two identifiers. 3310 3311 // Leave a space between * and /* to avoid C4138 `comment end` found outside 3312 // of comment. 3313 if (Left.is(tok::star) && Right.is(tok::comment)) 3314 return true; 3315 3316 if (Style.isCpp()) { 3317 // Space between import <iostream>. 3318 // or import .....; 3319 if (Left.is(Keywords.kw_import) && Right.isOneOf(tok::less, tok::ellipsis)) 3320 return true; 3321 // No space between module :. 3322 if (Left.isOneOf(Keywords.kw_module, Keywords.kw_import) && 3323 Right.is(TT_ModulePartitionColon)) 3324 return true; 3325 // No space between import foo:bar but keep a space between import :bar; 3326 if (Left.is(tok::identifier) && Right.is(TT_ModulePartitionColon)) 3327 return false; 3328 // No space between :bar; 3329 if (Left.is(TT_ModulePartitionColon) && 3330 Right.isOneOf(tok::identifier, tok::kw_private)) 3331 return false; 3332 if (Left.is(tok::ellipsis) && Right.is(tok::identifier) && 3333 Line.First->is(Keywords.kw_import)) 3334 return false; 3335 3336 if (Left.is(tok::kw_operator)) 3337 return Right.is(tok::coloncolon); 3338 if (Right.is(tok::l_brace) && Right.is(BK_BracedInit) && 3339 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList) 3340 return true; 3341 } else if (Style.Language == FormatStyle::LK_Proto || 3342 Style.Language == FormatStyle::LK_TextProto) { 3343 if (Right.is(tok::period) && 3344 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 3345 Keywords.kw_repeated, Keywords.kw_extend)) 3346 return true; 3347 if (Right.is(tok::l_paren) && 3348 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 3349 return true; 3350 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 3351 return true; 3352 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 3353 if (Left.is(tok::slash) || Right.is(tok::slash)) 3354 return false; 3355 if (Left.MatchingParen && 3356 Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 3357 Right.isOneOf(tok::l_brace, tok::less)) 3358 return !Style.Cpp11BracedListStyle; 3359 // A percent is probably part of a formatting specification, such as %lld. 3360 if (Left.is(tok::percent)) 3361 return false; 3362 // Preserve the existence of a space before a percent for cases like 0x%04x 3363 // and "%d %d" 3364 if (Left.is(tok::numeric_constant) && Right.is(tok::percent)) 3365 return HasExistingWhitespace(); 3366 } else if (Style.isJson()) { 3367 if (Right.is(tok::colon)) 3368 return false; 3369 } else if (Style.isCSharp()) { 3370 // Require spaces around '{' and before '}' unless they appear in 3371 // interpolated strings. Interpolated strings are merged into a single token 3372 // so cannot have spaces inserted by this function. 3373 3374 // No space between 'this' and '[' 3375 if (Left.is(tok::kw_this) && Right.is(tok::l_square)) 3376 return false; 3377 3378 // No space between 'new' and '(' 3379 if (Left.is(tok::kw_new) && Right.is(tok::l_paren)) 3380 return false; 3381 3382 // Space before { (including space within '{ {'). 3383 if (Right.is(tok::l_brace)) 3384 return true; 3385 3386 // Spaces inside braces. 3387 if (Left.is(tok::l_brace) && Right.isNot(tok::r_brace)) 3388 return true; 3389 3390 if (Left.isNot(tok::l_brace) && Right.is(tok::r_brace)) 3391 return true; 3392 3393 // Spaces around '=>'. 3394 if (Left.is(TT_FatArrow) || Right.is(TT_FatArrow)) 3395 return true; 3396 3397 // No spaces around attribute target colons 3398 if (Left.is(TT_AttributeColon) || Right.is(TT_AttributeColon)) 3399 return false; 3400 3401 // space between type and variable e.g. Dictionary<string,string> foo; 3402 if (Left.is(TT_TemplateCloser) && Right.is(TT_StartOfName)) 3403 return true; 3404 3405 // spaces inside square brackets. 3406 if (Left.is(tok::l_square) || Right.is(tok::r_square)) 3407 return Style.SpacesInSquareBrackets; 3408 3409 // No space before ? in nullable types. 3410 if (Right.is(TT_CSharpNullable)) 3411 return false; 3412 3413 // No space before null forgiving '!'. 3414 if (Right.is(TT_NonNullAssertion)) 3415 return false; 3416 3417 // No space between consecutive commas '[,,]'. 3418 if (Left.is(tok::comma) && Right.is(tok::comma)) 3419 return false; 3420 3421 // space after var in `var (key, value)` 3422 if (Left.is(Keywords.kw_var) && Right.is(tok::l_paren)) 3423 return true; 3424 3425 // space between keywords and paren e.g. "using (" 3426 if (Right.is(tok::l_paren)) 3427 if (Left.isOneOf(tok::kw_using, Keywords.kw_async, Keywords.kw_when, 3428 Keywords.kw_lock)) 3429 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3430 spaceRequiredBeforeParens(Right); 3431 3432 // space between method modifier and opening parenthesis of a tuple return 3433 // type 3434 if (Left.isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 3435 tok::kw_virtual, tok::kw_extern, tok::kw_static, 3436 Keywords.kw_internal, Keywords.kw_abstract, 3437 Keywords.kw_sealed, Keywords.kw_override, 3438 Keywords.kw_async, Keywords.kw_unsafe) && 3439 Right.is(tok::l_paren)) 3440 return true; 3441 } else if (Style.isJavaScript()) { 3442 if (Left.is(TT_FatArrow)) 3443 return true; 3444 // for await ( ... 3445 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 3446 Left.Previous->is(tok::kw_for)) 3447 return true; 3448 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 3449 Right.MatchingParen) { 3450 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 3451 // An async arrow function, for example: `x = async () => foo();`, 3452 // as opposed to calling a function called async: `x = async();` 3453 if (Next && Next->is(TT_FatArrow)) 3454 return true; 3455 } 3456 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 3457 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 3458 return false; 3459 // In tagged template literals ("html`bar baz`"), there is no space between 3460 // the tag identifier and the template string. 3461 if (Keywords.IsJavaScriptIdentifier(Left, 3462 /* AcceptIdentifierName= */ false) && 3463 Right.is(TT_TemplateString)) 3464 return false; 3465 if (Right.is(tok::star) && 3466 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 3467 return false; 3468 if (Right.isOneOf(tok::l_brace, tok::l_square) && 3469 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 3470 Keywords.kw_extends, Keywords.kw_implements)) 3471 return true; 3472 if (Right.is(tok::l_paren)) { 3473 // JS methods can use some keywords as names (e.g. `delete()`). 3474 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 3475 return false; 3476 // Valid JS method names can include keywords, e.g. `foo.delete()` or 3477 // `bar.instanceof()`. Recognize call positions by preceding period. 3478 if (Left.Previous && Left.Previous->is(tok::period) && 3479 Left.Tok.getIdentifierInfo()) 3480 return false; 3481 // Additional unary JavaScript operators that need a space after. 3482 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 3483 tok::kw_void)) 3484 return true; 3485 } 3486 // `foo as const;` casts into a const type. 3487 if (Left.endsSequence(tok::kw_const, Keywords.kw_as)) { 3488 return false; 3489 } 3490 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 3491 tok::kw_const) || 3492 // "of" is only a keyword if it appears after another identifier 3493 // (e.g. as "const x of y" in a for loop), or after a destructuring 3494 // operation (const [x, y] of z, const {a, b} of c). 3495 (Left.is(Keywords.kw_of) && Left.Previous && 3496 (Left.Previous->Tok.is(tok::identifier) || 3497 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 3498 (!Left.Previous || !Left.Previous->is(tok::period))) 3499 return true; 3500 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 3501 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 3502 return false; 3503 if (Left.is(Keywords.kw_as) && 3504 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 3505 return true; 3506 if (Left.is(tok::kw_default) && Left.Previous && 3507 Left.Previous->is(tok::kw_export)) 3508 return true; 3509 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 3510 return true; 3511 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 3512 return false; 3513 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 3514 return false; 3515 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 3516 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 3517 return false; 3518 if (Left.is(tok::ellipsis)) 3519 return false; 3520 if (Left.is(TT_TemplateCloser) && 3521 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 3522 Keywords.kw_implements, Keywords.kw_extends)) 3523 // Type assertions ('<type>expr') are not followed by whitespace. Other 3524 // locations that should have whitespace following are identified by the 3525 // above set of follower tokens. 3526 return false; 3527 if (Right.is(TT_NonNullAssertion)) 3528 return false; 3529 if (Left.is(TT_NonNullAssertion) && 3530 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 3531 return true; // "x! as string", "x! in y" 3532 } else if (Style.Language == FormatStyle::LK_Java) { 3533 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 3534 return true; 3535 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 3536 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3537 spaceRequiredBeforeParens(Right); 3538 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 3539 tok::kw_protected) || 3540 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 3541 Keywords.kw_native)) && 3542 Right.is(TT_TemplateOpener)) 3543 return true; 3544 } 3545 if (Left.is(TT_ImplicitStringLiteral)) 3546 return HasExistingWhitespace(); 3547 if (Line.Type == LT_ObjCMethodDecl) { 3548 if (Left.is(TT_ObjCMethodSpecifier)) 3549 return true; 3550 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right)) 3551 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a 3552 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class 3553 // method declaration. 3554 return false; 3555 } 3556 if (Line.Type == LT_ObjCProperty && 3557 (Right.is(tok::equal) || Left.is(tok::equal))) 3558 return false; 3559 3560 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 3561 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 3562 return true; 3563 if (Left.is(tok::comma) && !Right.is(TT_OverloadedOperatorLParen)) 3564 return true; 3565 if (Right.is(tok::comma)) 3566 return false; 3567 if (Right.is(TT_ObjCBlockLParen)) 3568 return true; 3569 if (Right.is(TT_CtorInitializerColon)) 3570 return Style.SpaceBeforeCtorInitializerColon; 3571 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 3572 return false; 3573 if (Right.is(TT_RangeBasedForLoopColon) && 3574 !Style.SpaceBeforeRangeBasedForLoopColon) 3575 return false; 3576 if (Left.is(TT_BitFieldColon)) 3577 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 3578 Style.BitFieldColonSpacing == FormatStyle::BFCS_After; 3579 if (Right.is(tok::colon)) { 3580 if (Line.First->isOneOf(tok::kw_default, tok::kw_case)) 3581 return Style.SpaceBeforeCaseColon; 3582 if (!Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 3583 return false; 3584 if (Right.is(TT_ObjCMethodExpr)) 3585 return false; 3586 if (Left.is(tok::question)) 3587 return false; 3588 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 3589 return false; 3590 if (Right.is(TT_DictLiteral)) 3591 return Style.SpacesInContainerLiterals; 3592 if (Right.is(TT_AttributeColon)) 3593 return false; 3594 if (Right.is(TT_CSharpNamedArgumentColon)) 3595 return false; 3596 if (Right.is(TT_BitFieldColon)) 3597 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 3598 Style.BitFieldColonSpacing == FormatStyle::BFCS_Before; 3599 return true; 3600 } 3601 // Do not merge "- -" into "--". 3602 if ((Left.isOneOf(tok::minus, tok::minusminus) && 3603 Right.isOneOf(tok::minus, tok::minusminus)) || 3604 (Left.isOneOf(tok::plus, tok::plusplus) && 3605 Right.isOneOf(tok::plus, tok::plusplus))) 3606 return true; 3607 if (Left.is(TT_UnaryOperator)) { 3608 if (!Right.is(tok::l_paren)) { 3609 // The alternative operators for ~ and ! are "compl" and "not". 3610 // If they are used instead, we do not want to combine them with 3611 // the token to the right, unless that is a left paren. 3612 if (Left.is(tok::exclaim) && Left.TokenText == "not") 3613 return true; 3614 if (Left.is(tok::tilde) && Left.TokenText == "compl") 3615 return true; 3616 // Lambda captures allow for a lone &, so "&]" needs to be properly 3617 // handled. 3618 if (Left.is(tok::amp) && Right.is(tok::r_square)) 3619 return Style.SpacesInSquareBrackets; 3620 } 3621 return (Style.SpaceAfterLogicalNot && Left.is(tok::exclaim)) || 3622 Right.is(TT_BinaryOperator); 3623 } 3624 3625 // If the next token is a binary operator or a selector name, we have 3626 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 3627 if (Left.is(TT_CastRParen)) 3628 return Style.SpaceAfterCStyleCast || 3629 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 3630 3631 auto ShouldAddSpacesInAngles = [this, &HasExistingWhitespace]() { 3632 if (this->Style.SpacesInAngles == FormatStyle::SIAS_Always) 3633 return true; 3634 if (this->Style.SpacesInAngles == FormatStyle::SIAS_Leave) 3635 return HasExistingWhitespace(); 3636 return false; 3637 }; 3638 3639 if (Left.is(tok::greater) && Right.is(tok::greater)) { 3640 if (Style.Language == FormatStyle::LK_TextProto || 3641 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 3642 return !Style.Cpp11BracedListStyle; 3643 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 3644 ((Style.Standard < FormatStyle::LS_Cpp11) || 3645 ShouldAddSpacesInAngles()); 3646 } 3647 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 3648 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 3649 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 3650 return false; 3651 if (!Style.SpaceBeforeAssignmentOperators && Left.isNot(TT_TemplateCloser) && 3652 Right.getPrecedence() == prec::Assignment) 3653 return false; 3654 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) && 3655 (Left.is(tok::identifier) || Left.is(tok::kw_this))) 3656 return false; 3657 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 3658 // Generally don't remove existing spaces between an identifier and "::". 3659 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 3660 // this turns out to be too lenient, add analysis of the identifier itself. 3661 return HasExistingWhitespace(); 3662 if (Right.is(tok::coloncolon) && 3663 !Left.isOneOf(tok::l_brace, tok::comment, tok::l_paren)) 3664 // Put a space between < and :: in vector< ::std::string > 3665 return (Left.is(TT_TemplateOpener) && 3666 ((Style.Standard < FormatStyle::LS_Cpp11) || 3667 ShouldAddSpacesInAngles())) || 3668 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 3669 tok::kw___super, TT_TemplateOpener, 3670 TT_TemplateCloser)) || 3671 (Left.is(tok::l_paren) && Style.SpacesInParentheses); 3672 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 3673 return ShouldAddSpacesInAngles(); 3674 // Space before TT_StructuredBindingLSquare. 3675 if (Right.is(TT_StructuredBindingLSquare)) 3676 return !Left.isOneOf(tok::amp, tok::ampamp) || 3677 getTokenReferenceAlignment(Left) != FormatStyle::PAS_Right; 3678 // Space before & or && following a TT_StructuredBindingLSquare. 3679 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 3680 Right.isOneOf(tok::amp, tok::ampamp)) 3681 return getTokenReferenceAlignment(Right) != FormatStyle::PAS_Left; 3682 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 3683 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 3684 !Right.is(tok::r_paren))) 3685 return true; 3686 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 3687 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 3688 return false; 3689 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 3690 Line.startsWith(tok::hash)) 3691 return true; 3692 if (Right.is(TT_TrailingUnaryOperator)) 3693 return false; 3694 if (Left.is(TT_RegexLiteral)) 3695 return false; 3696 return spaceRequiredBetween(Line, Left, Right); 3697 } 3698 3699 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 3700 static bool isAllmanBrace(const FormatToken &Tok) { 3701 return Tok.is(tok::l_brace) && Tok.is(BK_Block) && 3702 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_LambdaLBrace, TT_DictLiteral); 3703 } 3704 3705 // Returns 'true' if 'Tok' is a function argument. 3706 static bool IsFunctionArgument(const FormatToken &Tok) { 3707 return Tok.MatchingParen && Tok.MatchingParen->Next && 3708 Tok.MatchingParen->Next->isOneOf(tok::comma, tok::r_paren); 3709 } 3710 3711 static bool 3712 isItAnEmptyLambdaAllowed(const FormatToken &Tok, 3713 FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3714 return Tok.Children.empty() && ShortLambdaOption != FormatStyle::SLS_None; 3715 } 3716 3717 static bool isAllmanLambdaBrace(const FormatToken &Tok) { 3718 return (Tok.is(tok::l_brace) && Tok.is(BK_Block) && 3719 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral)); 3720 } 3721 3722 // Returns the first token on the line that is not a comment. 3723 static const FormatToken *getFirstNonComment(const AnnotatedLine &Line) { 3724 const FormatToken *Next = Line.First; 3725 if (!Next) 3726 return Next; 3727 if (Next->is(tok::comment)) 3728 Next = Next->getNextNonComment(); 3729 return Next; 3730 } 3731 3732 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 3733 const FormatToken &Right) { 3734 const FormatToken &Left = *Right.Previous; 3735 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 3736 return true; 3737 3738 if (Style.isCSharp()) { 3739 if (Left.is(TT_FatArrow) && Right.is(tok::l_brace) && 3740 Style.BraceWrapping.AfterFunction) 3741 return true; 3742 if (Right.is(TT_CSharpNamedArgumentColon) || 3743 Left.is(TT_CSharpNamedArgumentColon)) 3744 return false; 3745 if (Right.is(TT_CSharpGenericTypeConstraint)) 3746 return true; 3747 if (Right.Next && Right.Next->is(TT_FatArrow) && 3748 (Right.is(tok::numeric_constant) || 3749 (Right.is(tok::identifier) && Right.TokenText == "_"))) 3750 return true; 3751 3752 // Break after C# [...] and before public/protected/private/internal. 3753 if (Left.is(TT_AttributeSquare) && Left.is(tok::r_square) && 3754 (Right.isAccessSpecifier(/*ColonRequired=*/false) || 3755 Right.is(Keywords.kw_internal))) 3756 return true; 3757 // Break between ] and [ but only when there are really 2 attributes. 3758 if (Left.is(TT_AttributeSquare) && Right.is(TT_AttributeSquare) && 3759 Left.is(tok::r_square) && Right.is(tok::l_square)) 3760 return true; 3761 3762 } else if (Style.isJavaScript()) { 3763 // FIXME: This might apply to other languages and token kinds. 3764 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 3765 Left.Previous->is(tok::string_literal)) 3766 return true; 3767 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 3768 Left.Previous && Left.Previous->is(tok::equal) && 3769 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 3770 tok::kw_const) && 3771 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 3772 // above. 3773 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 3774 // Object literals on the top level of a file are treated as "enum-style". 3775 // Each key/value pair is put on a separate line, instead of bin-packing. 3776 return true; 3777 if (Left.is(tok::l_brace) && Line.Level == 0 && 3778 (Line.startsWith(tok::kw_enum) || 3779 Line.startsWith(tok::kw_const, tok::kw_enum) || 3780 Line.startsWith(tok::kw_export, tok::kw_enum) || 3781 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 3782 // JavaScript top-level enum key/value pairs are put on separate lines 3783 // instead of bin-packing. 3784 return true; 3785 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && Left.Previous && 3786 Left.Previous->is(TT_FatArrow)) { 3787 // JS arrow function (=> {...}). 3788 switch (Style.AllowShortLambdasOnASingleLine) { 3789 case FormatStyle::SLS_All: 3790 return false; 3791 case FormatStyle::SLS_None: 3792 return true; 3793 case FormatStyle::SLS_Empty: 3794 return !Left.Children.empty(); 3795 case FormatStyle::SLS_Inline: 3796 // allow one-lining inline (e.g. in function call args) and empty arrow 3797 // functions. 3798 return (Left.NestingLevel == 0 && Line.Level == 0) && 3799 !Left.Children.empty(); 3800 } 3801 llvm_unreachable("Unknown FormatStyle::ShortLambdaStyle enum"); 3802 } 3803 3804 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 3805 !Left.Children.empty()) 3806 // Support AllowShortFunctionsOnASingleLine for JavaScript. 3807 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 3808 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 3809 (Left.NestingLevel == 0 && Line.Level == 0 && 3810 Style.AllowShortFunctionsOnASingleLine & 3811 FormatStyle::SFS_InlineOnly); 3812 } else if (Style.Language == FormatStyle::LK_Java) { 3813 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 3814 Right.Next->is(tok::string_literal)) 3815 return true; 3816 } else if (Style.Language == FormatStyle::LK_Cpp || 3817 Style.Language == FormatStyle::LK_ObjC || 3818 Style.Language == FormatStyle::LK_Proto || 3819 Style.Language == FormatStyle::LK_TableGen || 3820 Style.Language == FormatStyle::LK_TextProto) { 3821 if (Left.isStringLiteral() && Right.isStringLiteral()) 3822 return true; 3823 } 3824 3825 // Basic JSON newline processing. 3826 if (Style.isJson()) { 3827 // Always break after a JSON record opener. 3828 // { 3829 // } 3830 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace)) 3831 return true; 3832 // Always break after a JSON array opener. 3833 // [ 3834 // ] 3835 if (Left.is(TT_ArrayInitializerLSquare) && Left.is(tok::l_square) && 3836 !Right.is(tok::r_square)) 3837 return true; 3838 // Always break after successive entries. 3839 // 1, 3840 // 2 3841 if (Left.is(tok::comma)) 3842 return true; 3843 } 3844 3845 // If the last token before a '}', ']', or ')' is a comma or a trailing 3846 // comment, the intention is to insert a line break after it in order to make 3847 // shuffling around entries easier. Import statements, especially in 3848 // JavaScript, can be an exception to this rule. 3849 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 3850 const FormatToken *BeforeClosingBrace = nullptr; 3851 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 3852 (Style.isJavaScript() && Left.is(tok::l_paren))) && 3853 Left.isNot(BK_Block) && Left.MatchingParen) 3854 BeforeClosingBrace = Left.MatchingParen->Previous; 3855 else if (Right.MatchingParen && 3856 (Right.MatchingParen->isOneOf(tok::l_brace, 3857 TT_ArrayInitializerLSquare) || 3858 (Style.isJavaScript() && Right.MatchingParen->is(tok::l_paren)))) 3859 BeforeClosingBrace = &Left; 3860 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 3861 BeforeClosingBrace->isTrailingComment())) 3862 return true; 3863 } 3864 3865 if (Right.is(tok::comment)) 3866 return Left.isNot(BK_BracedInit) && Left.isNot(TT_CtorInitializerColon) && 3867 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 3868 if (Left.isTrailingComment()) 3869 return true; 3870 if (Left.IsUnterminatedLiteral) 3871 return true; 3872 if (Right.is(tok::lessless) && Right.Next && Left.is(tok::string_literal) && 3873 Right.Next->is(tok::string_literal)) 3874 return true; 3875 // Can break after template<> declaration 3876 if (Left.ClosesTemplateDeclaration && Left.MatchingParen && 3877 Left.MatchingParen->NestingLevel == 0) { 3878 // Put concepts on the next line e.g. 3879 // template<typename T> 3880 // concept ... 3881 if (Right.is(tok::kw_concept)) 3882 return Style.BreakBeforeConceptDeclarations; 3883 return (Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes); 3884 } 3885 if (Style.PackConstructorInitializers == FormatStyle::PCIS_Never) { 3886 if (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeColon && 3887 (Left.is(TT_CtorInitializerComma) || Right.is(TT_CtorInitializerColon))) 3888 return true; 3889 3890 if (Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon && 3891 Left.isOneOf(TT_CtorInitializerColon, TT_CtorInitializerComma)) 3892 return true; 3893 } 3894 if (Style.PackConstructorInitializers < FormatStyle::PCIS_CurrentLine && 3895 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3896 Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) 3897 return true; 3898 // Break only if we have multiple inheritance. 3899 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 3900 Right.is(TT_InheritanceComma)) 3901 return true; 3902 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterComma && 3903 Left.is(TT_InheritanceComma)) 3904 return true; 3905 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 3906 // Multiline raw string literals are special wrt. line breaks. The author 3907 // has made a deliberate choice and might have aligned the contents of the 3908 // string literal accordingly. Thus, we try keep existing line breaks. 3909 return Right.IsMultiline && Right.NewlinesBefore > 0; 3910 if ((Left.is(tok::l_brace) || (Left.is(tok::less) && Left.Previous && 3911 Left.Previous->is(tok::equal))) && 3912 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 3913 // Don't put enums or option definitions onto single lines in protocol 3914 // buffers. 3915 return true; 3916 } 3917 if (Right.is(TT_InlineASMBrace)) 3918 return Right.HasUnescapedNewline; 3919 3920 if (isAllmanBrace(Left) || isAllmanBrace(Right)) { 3921 auto FirstNonComment = getFirstNonComment(Line); 3922 bool AccessSpecifier = 3923 FirstNonComment && 3924 FirstNonComment->isOneOf(Keywords.kw_internal, tok::kw_public, 3925 tok::kw_private, tok::kw_protected); 3926 3927 if (Style.BraceWrapping.AfterEnum) { 3928 if (Line.startsWith(tok::kw_enum) || 3929 Line.startsWith(tok::kw_typedef, tok::kw_enum)) 3930 return true; 3931 // Ensure BraceWrapping for `public enum A {`. 3932 if (AccessSpecifier && FirstNonComment->Next && 3933 FirstNonComment->Next->is(tok::kw_enum)) 3934 return true; 3935 } 3936 3937 // Ensure BraceWrapping for `public interface A {`. 3938 if (Style.BraceWrapping.AfterClass && 3939 ((AccessSpecifier && FirstNonComment->Next && 3940 FirstNonComment->Next->is(Keywords.kw_interface)) || 3941 Line.startsWith(Keywords.kw_interface))) 3942 return true; 3943 3944 return (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 3945 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 3946 } 3947 3948 if (Left.is(TT_ObjCBlockLBrace) && 3949 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 3950 return true; 3951 3952 // Ensure wrapping after __attribute__((XX)) and @interface etc. 3953 if (Left.is(TT_AttributeParen) && Right.is(TT_ObjCDecl)) 3954 return true; 3955 3956 if (Left.is(TT_LambdaLBrace)) { 3957 if (IsFunctionArgument(Left) && 3958 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline) 3959 return false; 3960 3961 if (Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_None || 3962 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline || 3963 (!Left.Children.empty() && 3964 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Empty)) 3965 return true; 3966 } 3967 3968 if (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace) && 3969 Left.isOneOf(tok::star, tok::amp, tok::ampamp, TT_TemplateCloser)) { 3970 return true; 3971 } 3972 3973 // Put multiple Java annotation on a new line. 3974 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 3975 Left.is(TT_LeadingJavaAnnotation) && 3976 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 3977 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 3978 return true; 3979 3980 if (Right.is(TT_ProtoExtensionLSquare)) 3981 return true; 3982 3983 // In text proto instances if a submessage contains at least 2 entries and at 3984 // least one of them is a submessage, like A { ... B { ... } ... }, 3985 // put all of the entries of A on separate lines by forcing the selector of 3986 // the submessage B to be put on a newline. 3987 // 3988 // Example: these can stay on one line: 3989 // a { scalar_1: 1 scalar_2: 2 } 3990 // a { b { key: value } } 3991 // 3992 // and these entries need to be on a new line even if putting them all in one 3993 // line is under the column limit: 3994 // a { 3995 // scalar: 1 3996 // b { key: value } 3997 // } 3998 // 3999 // We enforce this by breaking before a submessage field that has previous 4000 // siblings, *and* breaking before a field that follows a submessage field. 4001 // 4002 // Be careful to exclude the case [proto.ext] { ... } since the `]` is 4003 // the TT_SelectorName there, but we don't want to break inside the brackets. 4004 // 4005 // Another edge case is @submessage { key: value }, which is a common 4006 // substitution placeholder. In this case we want to keep `@` and `submessage` 4007 // together. 4008 // 4009 // We ensure elsewhere that extensions are always on their own line. 4010 if ((Style.Language == FormatStyle::LK_Proto || 4011 Style.Language == FormatStyle::LK_TextProto) && 4012 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) { 4013 // Keep `@submessage` together in: 4014 // @submessage { key: value } 4015 if (Left.is(tok::at)) 4016 return false; 4017 // Look for the scope opener after selector in cases like: 4018 // selector { ... 4019 // selector: { ... 4020 // selector: @base { ... 4021 FormatToken *LBrace = Right.Next; 4022 if (LBrace && LBrace->is(tok::colon)) { 4023 LBrace = LBrace->Next; 4024 if (LBrace && LBrace->is(tok::at)) { 4025 LBrace = LBrace->Next; 4026 if (LBrace) 4027 LBrace = LBrace->Next; 4028 } 4029 } 4030 if (LBrace && 4031 // The scope opener is one of {, [, <: 4032 // selector { ... } 4033 // selector [ ... ] 4034 // selector < ... > 4035 // 4036 // In case of selector { ... }, the l_brace is TT_DictLiteral. 4037 // In case of an empty selector {}, the l_brace is not TT_DictLiteral, 4038 // so we check for immediately following r_brace. 4039 ((LBrace->is(tok::l_brace) && 4040 (LBrace->is(TT_DictLiteral) || 4041 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) || 4042 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) { 4043 // If Left.ParameterCount is 0, then this submessage entry is not the 4044 // first in its parent submessage, and we want to break before this entry. 4045 // If Left.ParameterCount is greater than 0, then its parent submessage 4046 // might contain 1 or more entries and we want to break before this entry 4047 // if it contains at least 2 entries. We deal with this case later by 4048 // detecting and breaking before the next entry in the parent submessage. 4049 if (Left.ParameterCount == 0) 4050 return true; 4051 // However, if this submessage is the first entry in its parent 4052 // submessage, Left.ParameterCount might be 1 in some cases. 4053 // We deal with this case later by detecting an entry 4054 // following a closing paren of this submessage. 4055 } 4056 4057 // If this is an entry immediately following a submessage, it will be 4058 // preceded by a closing paren of that submessage, like in: 4059 // left---. .---right 4060 // v v 4061 // sub: { ... } key: value 4062 // If there was a comment between `}` an `key` above, then `key` would be 4063 // put on a new line anyways. 4064 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square)) 4065 return true; 4066 } 4067 4068 // Deal with lambda arguments in C++ - we want consistent line breaks whether 4069 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced 4070 // as aggressive line breaks are placed when the lambda is not the last arg. 4071 if ((Style.Language == FormatStyle::LK_Cpp || 4072 Style.Language == FormatStyle::LK_ObjC) && 4073 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 && 4074 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) { 4075 // Multiple lambdas in the same function call force line breaks. 4076 if (Left.BlockParameterCount > 1) 4077 return true; 4078 4079 // A lambda followed by another arg forces a line break. 4080 if (!Left.Role) 4081 return false; 4082 auto Comma = Left.Role->lastComma(); 4083 if (!Comma) 4084 return false; 4085 auto Next = Comma->getNextNonComment(); 4086 if (!Next) 4087 return false; 4088 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret)) 4089 return true; 4090 } 4091 4092 return false; 4093 } 4094 4095 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 4096 const FormatToken &Right) { 4097 const FormatToken &Left = *Right.Previous; 4098 // Language-specific stuff. 4099 if (Style.isCSharp()) { 4100 if (Left.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon) || 4101 Right.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon)) 4102 return false; 4103 // Only break after commas for generic type constraints. 4104 if (Line.First->is(TT_CSharpGenericTypeConstraint)) 4105 return Left.is(TT_CSharpGenericTypeConstraintComma); 4106 // Keep nullable operators attached to their identifiers. 4107 if (Right.is(TT_CSharpNullable)) { 4108 return false; 4109 } 4110 } else if (Style.Language == FormatStyle::LK_Java) { 4111 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 4112 Keywords.kw_implements)) 4113 return false; 4114 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 4115 Keywords.kw_implements)) 4116 return true; 4117 } else if (Style.isJavaScript()) { 4118 const FormatToken *NonComment = Right.getPreviousNonComment(); 4119 if (NonComment && 4120 NonComment->isOneOf( 4121 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 4122 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 4123 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 4124 Keywords.kw_readonly, Keywords.kw_override, Keywords.kw_abstract, 4125 Keywords.kw_get, Keywords.kw_set, Keywords.kw_async, 4126 Keywords.kw_await)) 4127 return false; // Otherwise automatic semicolon insertion would trigger. 4128 if (Right.NestingLevel == 0 && 4129 (Left.Tok.getIdentifierInfo() || 4130 Left.isOneOf(tok::r_square, tok::r_paren)) && 4131 Right.isOneOf(tok::l_square, tok::l_paren)) 4132 return false; // Otherwise automatic semicolon insertion would trigger. 4133 if (NonComment && NonComment->is(tok::identifier) && 4134 NonComment->TokenText == "asserts") 4135 return false; 4136 if (Left.is(TT_FatArrow) && Right.is(tok::l_brace)) 4137 return false; 4138 if (Left.is(TT_JsTypeColon)) 4139 return true; 4140 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;"). 4141 if (Left.is(tok::exclaim) && Right.is(tok::colon)) 4142 return false; 4143 // Look for is type annotations like: 4144 // function f(): a is B { ... } 4145 // Do not break before is in these cases. 4146 if (Right.is(Keywords.kw_is)) { 4147 const FormatToken *Next = Right.getNextNonComment(); 4148 // If `is` is followed by a colon, it's likely that it's a dict key, so 4149 // ignore it for this check. 4150 // For example this is common in Polymer: 4151 // Polymer({ 4152 // is: 'name', 4153 // ... 4154 // }); 4155 if (!Next || !Next->is(tok::colon)) 4156 return false; 4157 } 4158 if (Left.is(Keywords.kw_in)) 4159 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 4160 if (Right.is(Keywords.kw_in)) 4161 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 4162 if (Right.is(Keywords.kw_as)) 4163 return false; // must not break before as in 'x as type' casts 4164 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) { 4165 // extends and infer can appear as keywords in conditional types: 4166 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types 4167 // do not break before them, as the expressions are subject to ASI. 4168 return false; 4169 } 4170 if (Left.is(Keywords.kw_as)) 4171 return true; 4172 if (Left.is(TT_NonNullAssertion)) 4173 return true; 4174 if (Left.is(Keywords.kw_declare) && 4175 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 4176 Keywords.kw_function, tok::kw_class, tok::kw_enum, 4177 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 4178 Keywords.kw_let, tok::kw_const)) 4179 // See grammar for 'declare' statements at: 4180 // https://github.com/Microsoft/TypeScript/blob/main/doc/spec-ARCHIVED.md#A.10 4181 return false; 4182 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 4183 Right.isOneOf(tok::identifier, tok::string_literal)) 4184 return false; // must not break in "module foo { ...}" 4185 if (Right.is(TT_TemplateString) && Right.closesScope()) 4186 return false; 4187 // Don't split tagged template literal so there is a break between the tag 4188 // identifier and template string. 4189 if (Left.is(tok::identifier) && Right.is(TT_TemplateString)) { 4190 return false; 4191 } 4192 if (Left.is(TT_TemplateString) && Left.opensScope()) 4193 return true; 4194 } 4195 4196 if (Left.is(tok::at)) 4197 return false; 4198 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 4199 return false; 4200 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 4201 return !Right.is(tok::l_paren); 4202 if (Right.is(TT_PointerOrReference)) 4203 return Line.IsMultiVariableDeclStmt || 4204 (getTokenPointerOrReferenceAlignment(Right) == 4205 FormatStyle::PAS_Right && 4206 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 4207 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 4208 Right.is(tok::kw_operator)) 4209 return true; 4210 if (Left.is(TT_PointerOrReference)) 4211 return false; 4212 if (Right.isTrailingComment()) 4213 // We rely on MustBreakBefore being set correctly here as we should not 4214 // change the "binding" behavior of a comment. 4215 // The first comment in a braced lists is always interpreted as belonging to 4216 // the first list element. Otherwise, it should be placed outside of the 4217 // list. 4218 return Left.is(BK_BracedInit) || 4219 (Left.is(TT_CtorInitializerColon) && 4220 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 4221 if (Left.is(tok::question) && Right.is(tok::colon)) 4222 return false; 4223 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 4224 return Style.BreakBeforeTernaryOperators; 4225 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 4226 return !Style.BreakBeforeTernaryOperators; 4227 if (Left.is(TT_InheritanceColon)) 4228 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon; 4229 if (Right.is(TT_InheritanceColon)) 4230 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon; 4231 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 4232 Left.isNot(TT_SelectorName)) 4233 return true; 4234 4235 if (Right.is(tok::colon) && 4236 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 4237 return false; 4238 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 4239 if (Style.Language == FormatStyle::LK_Proto || 4240 Style.Language == FormatStyle::LK_TextProto) { 4241 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 4242 return false; 4243 // Prevent cases like: 4244 // 4245 // submessage: 4246 // { key: valueeeeeeeeeeee } 4247 // 4248 // when the snippet does not fit into one line. 4249 // Prefer: 4250 // 4251 // submessage: { 4252 // key: valueeeeeeeeeeee 4253 // } 4254 // 4255 // instead, even if it is longer by one line. 4256 // 4257 // Note that this allows allows the "{" to go over the column limit 4258 // when the column limit is just between ":" and "{", but that does 4259 // not happen too often and alternative formattings in this case are 4260 // not much better. 4261 // 4262 // The code covers the cases: 4263 // 4264 // submessage: { ... } 4265 // submessage: < ... > 4266 // repeated: [ ... ] 4267 if (((Right.is(tok::l_brace) || Right.is(tok::less)) && 4268 Right.is(TT_DictLiteral)) || 4269 Right.is(TT_ArrayInitializerLSquare)) 4270 return false; 4271 } 4272 return true; 4273 } 4274 if (Right.is(tok::r_square) && Right.MatchingParen && 4275 Right.MatchingParen->is(TT_ProtoExtensionLSquare)) 4276 return false; 4277 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 4278 Right.Next->is(TT_ObjCMethodExpr))) 4279 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 4280 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 4281 return true; 4282 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 4283 return true; 4284 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 4285 TT_OverloadedOperator)) 4286 return false; 4287 if (Left.is(TT_RangeBasedForLoopColon)) 4288 return true; 4289 if (Right.is(TT_RangeBasedForLoopColon)) 4290 return false; 4291 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 4292 return true; 4293 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 4294 Left.is(tok::kw_operator)) 4295 return false; 4296 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 4297 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 4298 return false; 4299 if (Left.is(tok::equal) && Right.is(tok::l_brace) && 4300 !Style.Cpp11BracedListStyle) 4301 return false; 4302 if (Left.is(tok::l_paren) && 4303 Left.isOneOf(TT_AttributeParen, TT_TypeDeclarationParen)) 4304 return false; 4305 if (Left.is(tok::l_paren) && Left.Previous && 4306 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 4307 return false; 4308 if (Right.is(TT_ImplicitStringLiteral)) 4309 return false; 4310 4311 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser)) 4312 return false; 4313 if (Right.is(tok::r_square) && Right.MatchingParen && 4314 Right.MatchingParen->is(TT_LambdaLSquare)) 4315 return false; 4316 4317 // We only break before r_brace if there was a corresponding break before 4318 // the l_brace, which is tracked by BreakBeforeClosingBrace. 4319 if (Right.is(tok::r_brace)) 4320 return Right.MatchingParen && Right.MatchingParen->is(BK_Block); 4321 4322 // Allow breaking after a trailing annotation, e.g. after a method 4323 // declaration. 4324 if (Left.is(TT_TrailingAnnotation)) 4325 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 4326 tok::less, tok::coloncolon); 4327 4328 if (Right.is(tok::kw___attribute) || 4329 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) 4330 return !Left.is(TT_AttributeSquare); 4331 4332 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 4333 return true; 4334 4335 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 4336 return true; 4337 4338 if (Left.is(TT_CtorInitializerColon)) 4339 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 4340 if (Right.is(TT_CtorInitializerColon)) 4341 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 4342 if (Left.is(TT_CtorInitializerComma) && 4343 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 4344 return false; 4345 if (Right.is(TT_CtorInitializerComma) && 4346 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 4347 return true; 4348 if (Left.is(TT_InheritanceComma) && 4349 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 4350 return false; 4351 if (Right.is(TT_InheritanceComma) && 4352 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 4353 return true; 4354 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 4355 (Left.is(tok::less) && Right.is(tok::less))) 4356 return false; 4357 if (Right.is(TT_BinaryOperator) && 4358 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 4359 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 4360 Right.getPrecedence() != prec::Assignment)) 4361 return true; 4362 if (Left.is(TT_ArrayInitializerLSquare)) 4363 return true; 4364 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 4365 return true; 4366 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 4367 !Left.isOneOf(tok::arrowstar, tok::lessless) && 4368 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 4369 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 4370 Left.getPrecedence() == prec::Assignment)) 4371 return true; 4372 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 4373 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) 4374 return false; 4375 4376 auto ShortLambdaOption = Style.AllowShortLambdasOnASingleLine; 4377 if (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace)) { 4378 if (isAllmanLambdaBrace(Left)) 4379 return !isItAnEmptyLambdaAllowed(Left, ShortLambdaOption); 4380 if (isAllmanLambdaBrace(Right)) 4381 return !isItAnEmptyLambdaAllowed(Right, ShortLambdaOption); 4382 } 4383 4384 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 4385 tok::kw_class, tok::kw_struct, tok::comment) || 4386 Right.isMemberAccess() || 4387 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 4388 tok::colon, tok::l_square, tok::at) || 4389 (Left.is(tok::r_paren) && 4390 Right.isOneOf(tok::identifier, tok::kw_const)) || 4391 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 4392 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 4393 } 4394 4395 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 4396 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 4397 const FormatToken *Tok = Line.First; 4398 while (Tok) { 4399 llvm::errs() << " M=" << Tok->MustBreakBefore 4400 << " C=" << Tok->CanBreakBefore 4401 << " T=" << getTokenTypeName(Tok->getType()) 4402 << " S=" << Tok->SpacesRequiredBefore 4403 << " F=" << Tok->Finalized << " B=" << Tok->BlockParameterCount 4404 << " BK=" << Tok->getBlockKind() << " P=" << Tok->SplitPenalty 4405 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 4406 << " PPK=" << Tok->getPackingKind() << " FakeLParens="; 4407 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 4408 llvm::errs() << Tok->FakeLParens[i] << "/"; 4409 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 4410 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo(); 4411 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 4412 if (!Tok->Next) 4413 assert(Tok == Line.Last); 4414 Tok = Tok->Next; 4415 } 4416 llvm::errs() << "----\n"; 4417 } 4418 4419 FormatStyle::PointerAlignmentStyle 4420 TokenAnnotator::getTokenReferenceAlignment(const FormatToken &Reference) { 4421 assert(Reference.isOneOf(tok::amp, tok::ampamp)); 4422 switch (Style.ReferenceAlignment) { 4423 case FormatStyle::RAS_Pointer: 4424 return Style.PointerAlignment; 4425 case FormatStyle::RAS_Left: 4426 return FormatStyle::PAS_Left; 4427 case FormatStyle::RAS_Right: 4428 return FormatStyle::PAS_Right; 4429 case FormatStyle::RAS_Middle: 4430 return FormatStyle::PAS_Middle; 4431 } 4432 assert(0); //"Unhandled value of ReferenceAlignment" 4433 return Style.PointerAlignment; 4434 } 4435 4436 FormatStyle::PointerAlignmentStyle 4437 TokenAnnotator::getTokenPointerOrReferenceAlignment( 4438 const FormatToken &PointerOrReference) { 4439 if (PointerOrReference.isOneOf(tok::amp, tok::ampamp)) { 4440 switch (Style.ReferenceAlignment) { 4441 case FormatStyle::RAS_Pointer: 4442 return Style.PointerAlignment; 4443 case FormatStyle::RAS_Left: 4444 return FormatStyle::PAS_Left; 4445 case FormatStyle::RAS_Right: 4446 return FormatStyle::PAS_Right; 4447 case FormatStyle::RAS_Middle: 4448 return FormatStyle::PAS_Middle; 4449 } 4450 } 4451 assert(PointerOrReference.is(tok::star)); 4452 return Style.PointerAlignment; 4453 } 4454 4455 } // namespace format 4456 } // namespace clang 4457