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 assert(Left->Optional == CurrentToken->Optional); 784 Left->MatchingParen = CurrentToken; 785 CurrentToken->MatchingParen = Left; 786 if (Style.AlignArrayOfStructures != FormatStyle::AIAS_None) { 787 if (Left->ParentBracket == tok::l_brace && 788 couldBeInStructArrayInitializer() && CommaCount > 0) { 789 Contexts.back().InStructArrayInitializer = true; 790 } 791 } 792 next(); 793 return true; 794 } 795 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 796 return false; 797 updateParameterCount(Left, CurrentToken); 798 if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) { 799 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 800 if (Previous->is(TT_JsTypeOptionalQuestion)) 801 Previous = Previous->getPreviousNonComment(); 802 if ((CurrentToken->is(tok::colon) && 803 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) || 804 Style.Language == FormatStyle::LK_Proto || 805 Style.Language == FormatStyle::LK_TextProto) { 806 Left->setType(TT_DictLiteral); 807 if (Previous->Tok.getIdentifierInfo() || 808 Previous->is(tok::string_literal)) 809 Previous->setType(TT_SelectorName); 810 } 811 if (CurrentToken->is(tok::colon) || Style.isJavaScript()) 812 Left->setType(TT_DictLiteral); 813 } 814 if (CurrentToken->is(tok::comma)) { 815 if (Style.isJavaScript()) 816 Left->setType(TT_DictLiteral); 817 ++CommaCount; 818 } 819 if (!consumeToken()) 820 return false; 821 } 822 } 823 return true; 824 } 825 826 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 827 // For ObjC methods, the number of parameters is calculated differently as 828 // method declarations have a different structure (the parameters are not 829 // inside a bracket scope). 830 if (Current->is(tok::l_brace) && Current->is(BK_Block)) 831 ++Left->BlockParameterCount; 832 if (Current->is(tok::comma)) { 833 ++Left->ParameterCount; 834 if (!Left->Role) 835 Left->Role.reset(new CommaSeparatedList(Style)); 836 Left->Role->CommaFound(Current); 837 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 838 Left->ParameterCount = 1; 839 } 840 } 841 842 bool parseConditional() { 843 while (CurrentToken) { 844 if (CurrentToken->is(tok::colon)) { 845 CurrentToken->setType(TT_ConditionalExpr); 846 next(); 847 return true; 848 } 849 if (!consumeToken()) 850 return false; 851 } 852 return false; 853 } 854 855 bool parseTemplateDeclaration() { 856 if (CurrentToken && CurrentToken->is(tok::less)) { 857 CurrentToken->setType(TT_TemplateOpener); 858 next(); 859 if (!parseAngle()) 860 return false; 861 if (CurrentToken) 862 CurrentToken->Previous->ClosesTemplateDeclaration = true; 863 return true; 864 } 865 return false; 866 } 867 868 bool consumeToken() { 869 FormatToken *Tok = CurrentToken; 870 next(); 871 switch (Tok->Tok.getKind()) { 872 case tok::plus: 873 case tok::minus: 874 if (!Tok->Previous && Line.MustBeDeclaration) 875 Tok->setType(TT_ObjCMethodSpecifier); 876 break; 877 case tok::colon: 878 if (!Tok->Previous) 879 return false; 880 // Colons from ?: are handled in parseConditional(). 881 if (Style.isJavaScript()) { 882 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop 883 (Contexts.size() == 1 && // switch/case labels 884 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) || 885 Contexts.back().ContextKind == tok::l_paren || // function params 886 Contexts.back().ContextKind == tok::l_square || // array type 887 (!Contexts.back().IsExpression && 888 Contexts.back().ContextKind == tok::l_brace) || // object type 889 (Contexts.size() == 1 && 890 Line.MustBeDeclaration)) { // method/property declaration 891 Contexts.back().IsExpression = false; 892 Tok->setType(TT_JsTypeColon); 893 break; 894 } 895 } else if (Style.isCSharp()) { 896 if (Contexts.back().InCSharpAttributeSpecifier) { 897 Tok->setType(TT_AttributeColon); 898 break; 899 } 900 if (Contexts.back().ContextKind == tok::l_paren) { 901 Tok->setType(TT_CSharpNamedArgumentColon); 902 break; 903 } 904 } 905 if (Line.First->isOneOf(Keywords.kw_module, Keywords.kw_import) || 906 Line.First->startsSequence(tok::kw_export, Keywords.kw_module) || 907 Line.First->startsSequence(tok::kw_export, Keywords.kw_import)) { 908 Tok->setType(TT_ModulePartitionColon); 909 } else if (Contexts.back().ColonIsDictLiteral || 910 Style.Language == FormatStyle::LK_Proto || 911 Style.Language == FormatStyle::LK_TextProto) { 912 Tok->setType(TT_DictLiteral); 913 if (Style.Language == FormatStyle::LK_TextProto) { 914 if (FormatToken *Previous = Tok->getPreviousNonComment()) 915 Previous->setType(TT_SelectorName); 916 } 917 } else if (Contexts.back().ColonIsObjCMethodExpr || 918 Line.startsWith(TT_ObjCMethodSpecifier)) { 919 Tok->setType(TT_ObjCMethodExpr); 920 const FormatToken *BeforePrevious = Tok->Previous->Previous; 921 // Ensure we tag all identifiers in method declarations as 922 // TT_SelectorName. 923 bool UnknownIdentifierInMethodDeclaration = 924 Line.startsWith(TT_ObjCMethodSpecifier) && 925 Tok->Previous->is(tok::identifier) && Tok->Previous->is(TT_Unknown); 926 if (!BeforePrevious || 927 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 928 !(BeforePrevious->is(TT_CastRParen) || 929 (BeforePrevious->is(TT_ObjCMethodExpr) && 930 BeforePrevious->is(tok::colon))) || 931 BeforePrevious->is(tok::r_square) || 932 Contexts.back().LongestObjCSelectorName == 0 || 933 UnknownIdentifierInMethodDeclaration) { 934 Tok->Previous->setType(TT_SelectorName); 935 if (!Contexts.back().FirstObjCSelectorName) 936 Contexts.back().FirstObjCSelectorName = Tok->Previous; 937 else if (Tok->Previous->ColumnWidth > 938 Contexts.back().LongestObjCSelectorName) 939 Contexts.back().LongestObjCSelectorName = 940 Tok->Previous->ColumnWidth; 941 Tok->Previous->ParameterIndex = 942 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 943 ++Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 944 } 945 } else if (Contexts.back().ColonIsForRangeExpr) { 946 Tok->setType(TT_RangeBasedForLoopColon); 947 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 948 Tok->setType(TT_BitFieldColon); 949 } else if (Contexts.size() == 1 && 950 !Line.First->isOneOf(tok::kw_enum, tok::kw_case, 951 tok::kw_default)) { 952 FormatToken *Prev = Tok->getPreviousNonComment(); 953 if (!Prev) 954 break; 955 if (Prev->isOneOf(tok::r_paren, tok::kw_noexcept)) 956 Tok->setType(TT_CtorInitializerColon); 957 else if (Prev->is(tok::kw_try)) { 958 // Member initializer list within function try block. 959 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 960 if (!PrevPrev) 961 break; 962 if (PrevPrev && PrevPrev->isOneOf(tok::r_paren, tok::kw_noexcept)) 963 Tok->setType(TT_CtorInitializerColon); 964 } else 965 Tok->setType(TT_InheritanceColon); 966 } else if (canBeObjCSelectorComponent(*Tok->Previous) && Tok->Next && 967 (Tok->Next->isOneOf(tok::r_paren, tok::comma) || 968 (canBeObjCSelectorComponent(*Tok->Next) && Tok->Next->Next && 969 Tok->Next->Next->is(tok::colon)))) { 970 // This handles a special macro in ObjC code where selectors including 971 // the colon are passed as macro arguments. 972 Tok->setType(TT_ObjCMethodExpr); 973 } else if (Contexts.back().ContextKind == tok::l_paren) { 974 Tok->setType(TT_InlineASMColon); 975 } 976 break; 977 case tok::pipe: 978 case tok::amp: 979 // | and & in declarations/type expressions represent union and 980 // intersection types, respectively. 981 if (Style.isJavaScript() && !Contexts.back().IsExpression) 982 Tok->setType(TT_JsTypeOperator); 983 break; 984 case tok::kw_if: 985 case tok::kw_while: 986 if (Tok->is(tok::kw_if) && CurrentToken && 987 CurrentToken->isOneOf(tok::kw_constexpr, tok::identifier)) 988 next(); 989 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 990 next(); 991 if (!parseParens(/*LookForDecls=*/true)) 992 return false; 993 } 994 break; 995 case tok::kw_for: 996 if (Style.isJavaScript()) { 997 // x.for and {for: ...} 998 if ((Tok->Previous && Tok->Previous->is(tok::period)) || 999 (Tok->Next && Tok->Next->is(tok::colon))) 1000 break; 1001 // JS' for await ( ... 1002 if (CurrentToken && CurrentToken->is(Keywords.kw_await)) 1003 next(); 1004 } 1005 if (Style.isCpp() && CurrentToken && CurrentToken->is(tok::kw_co_await)) 1006 next(); 1007 Contexts.back().ColonIsForRangeExpr = true; 1008 next(); 1009 if (!parseParens()) 1010 return false; 1011 break; 1012 case tok::l_paren: 1013 // When faced with 'operator()()', the kw_operator handler incorrectly 1014 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make 1015 // the first two parens OverloadedOperators and the second l_paren an 1016 // OverloadedOperatorLParen. 1017 if (Tok->Previous && Tok->Previous->is(tok::r_paren) && 1018 Tok->Previous->MatchingParen && 1019 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) { 1020 Tok->Previous->setType(TT_OverloadedOperator); 1021 Tok->Previous->MatchingParen->setType(TT_OverloadedOperator); 1022 Tok->setType(TT_OverloadedOperatorLParen); 1023 } 1024 1025 if (!parseParens()) 1026 return false; 1027 if (Line.MustBeDeclaration && Contexts.size() == 1 && 1028 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) && 1029 !Tok->is(TT_TypeDeclarationParen) && 1030 (!Tok->Previous || !Tok->Previous->isOneOf(tok::kw___attribute, 1031 TT_LeadingJavaAnnotation))) 1032 Line.MightBeFunctionDecl = true; 1033 break; 1034 case tok::l_square: 1035 if (!parseSquare()) 1036 return false; 1037 break; 1038 case tok::l_brace: 1039 if (Style.Language == FormatStyle::LK_TextProto) { 1040 FormatToken *Previous = Tok->getPreviousNonComment(); 1041 if (Previous && Previous->getType() != TT_DictLiteral) 1042 Previous->setType(TT_SelectorName); 1043 } 1044 if (!parseBrace()) 1045 return false; 1046 break; 1047 case tok::less: 1048 if (parseAngle()) { 1049 Tok->setType(TT_TemplateOpener); 1050 // In TT_Proto, we must distignuish between: 1051 // map<key, value> 1052 // msg < item: data > 1053 // msg: < item: data > 1054 // In TT_TextProto, map<key, value> does not occur. 1055 if (Style.Language == FormatStyle::LK_TextProto || 1056 (Style.Language == FormatStyle::LK_Proto && Tok->Previous && 1057 Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) { 1058 Tok->setType(TT_DictLiteral); 1059 FormatToken *Previous = Tok->getPreviousNonComment(); 1060 if (Previous && Previous->getType() != TT_DictLiteral) 1061 Previous->setType(TT_SelectorName); 1062 } 1063 } else { 1064 Tok->setType(TT_BinaryOperator); 1065 NonTemplateLess.insert(Tok); 1066 CurrentToken = Tok; 1067 next(); 1068 } 1069 break; 1070 case tok::r_paren: 1071 case tok::r_square: 1072 return false; 1073 case tok::r_brace: 1074 // Lines can start with '}'. 1075 if (Tok->Previous) 1076 return false; 1077 break; 1078 case tok::greater: 1079 if (Style.Language != FormatStyle::LK_TextProto) 1080 Tok->setType(TT_BinaryOperator); 1081 if (Tok->Previous && Tok->Previous->is(TT_TemplateCloser)) 1082 Tok->SpacesRequiredBefore = 1; 1083 break; 1084 case tok::kw_operator: 1085 if (Style.Language == FormatStyle::LK_TextProto || 1086 Style.Language == FormatStyle::LK_Proto) 1087 break; 1088 while (CurrentToken && 1089 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 1090 if (CurrentToken->isOneOf(tok::star, tok::amp)) 1091 CurrentToken->setType(TT_PointerOrReference); 1092 consumeToken(); 1093 if (CurrentToken && CurrentToken->is(tok::comma) && 1094 CurrentToken->Previous->isNot(tok::kw_operator)) 1095 break; 1096 if (CurrentToken && CurrentToken->Previous->isOneOf( 1097 TT_BinaryOperator, TT_UnaryOperator, tok::comma, 1098 tok::star, tok::arrow, tok::amp, tok::ampamp)) 1099 CurrentToken->Previous->setType(TT_OverloadedOperator); 1100 } 1101 if (CurrentToken && CurrentToken->is(tok::l_paren)) 1102 CurrentToken->setType(TT_OverloadedOperatorLParen); 1103 if (CurrentToken && CurrentToken->Previous->is(TT_BinaryOperator)) 1104 CurrentToken->Previous->setType(TT_OverloadedOperator); 1105 break; 1106 case tok::question: 1107 if (Style.isJavaScript() && Tok->Next && 1108 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren, 1109 tok::r_brace)) { 1110 // Question marks before semicolons, colons, etc. indicate optional 1111 // types (fields, parameters), e.g. 1112 // function(x?: string, y?) {...} 1113 // class X { y?; } 1114 Tok->setType(TT_JsTypeOptionalQuestion); 1115 break; 1116 } 1117 // Declarations cannot be conditional expressions, this can only be part 1118 // of a type declaration. 1119 if (Line.MustBeDeclaration && !Contexts.back().IsExpression && 1120 Style.isJavaScript()) 1121 break; 1122 if (Style.isCSharp()) { 1123 // `Type?)`, `Type?>`, `Type? name;` and `Type? name =` can only be 1124 // nullable types. 1125 // Line.MustBeDeclaration will be true for `Type? name;`. 1126 if ((!Contexts.back().IsExpression && Line.MustBeDeclaration) || 1127 (Tok->Next && Tok->Next->isOneOf(tok::r_paren, tok::greater)) || 1128 (Tok->Next && Tok->Next->is(tok::identifier) && Tok->Next->Next && 1129 Tok->Next->Next->is(tok::equal))) { 1130 Tok->setType(TT_CSharpNullable); 1131 break; 1132 } 1133 } 1134 parseConditional(); 1135 break; 1136 case tok::kw_template: 1137 parseTemplateDeclaration(); 1138 break; 1139 case tok::comma: 1140 if (Contexts.back().InCtorInitializer) 1141 Tok->setType(TT_CtorInitializerComma); 1142 else if (Contexts.back().InInheritanceList) 1143 Tok->setType(TT_InheritanceComma); 1144 else if (Contexts.back().FirstStartOfName && 1145 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) { 1146 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 1147 Line.IsMultiVariableDeclStmt = true; 1148 } 1149 if (Contexts.back().IsForEachMacro) 1150 Contexts.back().IsExpression = true; 1151 break; 1152 case tok::identifier: 1153 if (Tok->isOneOf(Keywords.kw___has_include, 1154 Keywords.kw___has_include_next)) { 1155 parseHasInclude(); 1156 } 1157 if (Style.isCSharp() && Tok->is(Keywords.kw_where) && Tok->Next && 1158 Tok->Next->isNot(tok::l_paren)) { 1159 Tok->setType(TT_CSharpGenericTypeConstraint); 1160 parseCSharpGenericTypeConstraint(); 1161 } 1162 break; 1163 default: 1164 break; 1165 } 1166 return true; 1167 } 1168 1169 void parseCSharpGenericTypeConstraint() { 1170 int OpenAngleBracketsCount = 0; 1171 while (CurrentToken) { 1172 if (CurrentToken->is(tok::less)) { 1173 // parseAngle is too greedy and will consume the whole line. 1174 CurrentToken->setType(TT_TemplateOpener); 1175 ++OpenAngleBracketsCount; 1176 next(); 1177 } else if (CurrentToken->is(tok::greater)) { 1178 CurrentToken->setType(TT_TemplateCloser); 1179 --OpenAngleBracketsCount; 1180 next(); 1181 } else if (CurrentToken->is(tok::comma) && OpenAngleBracketsCount == 0) { 1182 // We allow line breaks after GenericTypeConstraintComma's 1183 // so do not flag commas in Generics as GenericTypeConstraintComma's. 1184 CurrentToken->setType(TT_CSharpGenericTypeConstraintComma); 1185 next(); 1186 } else if (CurrentToken->is(Keywords.kw_where)) { 1187 CurrentToken->setType(TT_CSharpGenericTypeConstraint); 1188 next(); 1189 } else if (CurrentToken->is(tok::colon)) { 1190 CurrentToken->setType(TT_CSharpGenericTypeConstraintColon); 1191 next(); 1192 } else { 1193 next(); 1194 } 1195 } 1196 } 1197 1198 void parseIncludeDirective() { 1199 if (CurrentToken && CurrentToken->is(tok::less)) { 1200 next(); 1201 while (CurrentToken) { 1202 // Mark tokens up to the trailing line comments as implicit string 1203 // literals. 1204 if (CurrentToken->isNot(tok::comment) && 1205 !CurrentToken->TokenText.startswith("//")) 1206 CurrentToken->setType(TT_ImplicitStringLiteral); 1207 next(); 1208 } 1209 } 1210 } 1211 1212 void parseWarningOrError() { 1213 next(); 1214 // We still want to format the whitespace left of the first token of the 1215 // warning or error. 1216 next(); 1217 while (CurrentToken) { 1218 CurrentToken->setType(TT_ImplicitStringLiteral); 1219 next(); 1220 } 1221 } 1222 1223 void parsePragma() { 1224 next(); // Consume "pragma". 1225 if (CurrentToken && 1226 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 1227 bool IsMark = CurrentToken->is(Keywords.kw_mark); 1228 next(); // Consume "mark". 1229 next(); // Consume first token (so we fix leading whitespace). 1230 while (CurrentToken) { 1231 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 1232 CurrentToken->setType(TT_ImplicitStringLiteral); 1233 next(); 1234 } 1235 } 1236 } 1237 1238 void parseHasInclude() { 1239 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 1240 return; 1241 next(); // '(' 1242 parseIncludeDirective(); 1243 next(); // ')' 1244 } 1245 1246 LineType parsePreprocessorDirective() { 1247 bool IsFirstToken = CurrentToken->IsFirst; 1248 LineType Type = LT_PreprocessorDirective; 1249 next(); 1250 if (!CurrentToken) 1251 return Type; 1252 1253 if (Style.isJavaScript() && IsFirstToken) { 1254 // JavaScript files can contain shebang lines of the form: 1255 // #!/usr/bin/env node 1256 // Treat these like C++ #include directives. 1257 while (CurrentToken) { 1258 // Tokens cannot be comments here. 1259 CurrentToken->setType(TT_ImplicitStringLiteral); 1260 next(); 1261 } 1262 return LT_ImportStatement; 1263 } 1264 1265 if (CurrentToken->Tok.is(tok::numeric_constant)) { 1266 CurrentToken->SpacesRequiredBefore = 1; 1267 return Type; 1268 } 1269 // Hashes in the middle of a line can lead to any strange token 1270 // sequence. 1271 if (!CurrentToken->Tok.getIdentifierInfo()) 1272 return Type; 1273 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 1274 case tok::pp_include: 1275 case tok::pp_include_next: 1276 case tok::pp_import: 1277 next(); 1278 parseIncludeDirective(); 1279 Type = LT_ImportStatement; 1280 break; 1281 case tok::pp_error: 1282 case tok::pp_warning: 1283 parseWarningOrError(); 1284 break; 1285 case tok::pp_pragma: 1286 parsePragma(); 1287 break; 1288 case tok::pp_if: 1289 case tok::pp_elif: 1290 Contexts.back().IsExpression = true; 1291 next(); 1292 parseLine(); 1293 break; 1294 default: 1295 break; 1296 } 1297 while (CurrentToken) { 1298 FormatToken *Tok = CurrentToken; 1299 next(); 1300 if (Tok->is(tok::l_paren)) 1301 parseParens(); 1302 else if (Tok->isOneOf(Keywords.kw___has_include, 1303 Keywords.kw___has_include_next)) 1304 parseHasInclude(); 1305 } 1306 return Type; 1307 } 1308 1309 public: 1310 LineType parseLine() { 1311 if (!CurrentToken) 1312 return LT_Invalid; 1313 NonTemplateLess.clear(); 1314 if (CurrentToken->is(tok::hash)) 1315 return parsePreprocessorDirective(); 1316 1317 // Directly allow to 'import <string-literal>' to support protocol buffer 1318 // definitions (github.com/google/protobuf) or missing "#" (either way we 1319 // should not break the line). 1320 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 1321 if ((Style.Language == FormatStyle::LK_Java && 1322 CurrentToken->is(Keywords.kw_package)) || 1323 (Info && Info->getPPKeywordID() == tok::pp_import && 1324 CurrentToken->Next && 1325 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 1326 tok::kw_static))) { 1327 next(); 1328 parseIncludeDirective(); 1329 return LT_ImportStatement; 1330 } 1331 1332 // If this line starts and ends in '<' and '>', respectively, it is likely 1333 // part of "#define <a/b.h>". 1334 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 1335 parseIncludeDirective(); 1336 return LT_ImportStatement; 1337 } 1338 1339 // In .proto files, top-level options and package statements are very 1340 // similar to import statements and should not be line-wrapped. 1341 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 1342 CurrentToken->isOneOf(Keywords.kw_option, Keywords.kw_package)) { 1343 next(); 1344 if (CurrentToken && CurrentToken->is(tok::identifier)) { 1345 while (CurrentToken) 1346 next(); 1347 return LT_ImportStatement; 1348 } 1349 } 1350 1351 bool KeywordVirtualFound = false; 1352 bool ImportStatement = false; 1353 1354 // import {...} from '...'; 1355 if (Style.isJavaScript() && CurrentToken->is(Keywords.kw_import)) 1356 ImportStatement = true; 1357 1358 while (CurrentToken) { 1359 if (CurrentToken->is(tok::kw_virtual)) 1360 KeywordVirtualFound = true; 1361 if (Style.isJavaScript()) { 1362 // export {...} from '...'; 1363 // An export followed by "from 'some string';" is a re-export from 1364 // another module identified by a URI and is treated as a 1365 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 1366 // Just "export {...};" or "export class ..." should not be treated as 1367 // an import in this sense. 1368 if (Line.First->is(tok::kw_export) && 1369 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 1370 CurrentToken->Next->isStringLiteral()) 1371 ImportStatement = true; 1372 if (isClosureImportStatement(*CurrentToken)) 1373 ImportStatement = true; 1374 } 1375 if (!consumeToken()) 1376 return LT_Invalid; 1377 } 1378 if (KeywordVirtualFound) 1379 return LT_VirtualFunctionDecl; 1380 if (ImportStatement) 1381 return LT_ImportStatement; 1382 1383 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 1384 if (Contexts.back().FirstObjCSelectorName) 1385 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 1386 Contexts.back().LongestObjCSelectorName; 1387 return LT_ObjCMethodDecl; 1388 } 1389 1390 for (const auto &ctx : Contexts) { 1391 if (ctx.InStructArrayInitializer) { 1392 return LT_ArrayOfStructInitializer; 1393 } 1394 } 1395 1396 return LT_Other; 1397 } 1398 1399 private: 1400 bool isClosureImportStatement(const FormatToken &Tok) { 1401 // FIXME: Closure-library specific stuff should not be hard-coded but be 1402 // configurable. 1403 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 1404 Tok.Next->Next && 1405 (Tok.Next->Next->TokenText == "module" || 1406 Tok.Next->Next->TokenText == "provide" || 1407 Tok.Next->Next->TokenText == "require" || 1408 Tok.Next->Next->TokenText == "requireType" || 1409 Tok.Next->Next->TokenText == "forwardDeclare") && 1410 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 1411 } 1412 1413 void resetTokenMetadata() { 1414 if (!CurrentToken) 1415 return; 1416 1417 // Reset token type in case we have already looked at it and then 1418 // recovered from an error (e.g. failure to find the matching >). 1419 if (!CurrentToken->isOneOf( 1420 TT_LambdaLSquare, TT_LambdaLBrace, TT_AttributeMacro, TT_IfMacro, 1421 TT_ForEachMacro, TT_TypenameMacro, TT_FunctionLBrace, 1422 TT_ImplicitStringLiteral, TT_InlineASMBrace, TT_FatArrow, 1423 TT_LambdaArrow, TT_NamespaceMacro, TT_OverloadedOperator, 1424 TT_RegexLiteral, TT_TemplateString, TT_ObjCStringLiteral, 1425 TT_UntouchableMacroFunc, TT_ConstraintJunctions, 1426 TT_StatementAttributeLikeMacro, TT_FunctionLikeOrFreestandingMacro)) 1427 CurrentToken->setType(TT_Unknown); 1428 CurrentToken->Role.reset(); 1429 CurrentToken->MatchingParen = nullptr; 1430 CurrentToken->FakeLParens.clear(); 1431 CurrentToken->FakeRParens = 0; 1432 } 1433 1434 void next() { 1435 if (!CurrentToken) 1436 return; 1437 1438 CurrentToken->NestingLevel = Contexts.size() - 1; 1439 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1440 modifyContext(*CurrentToken); 1441 determineTokenType(*CurrentToken); 1442 CurrentToken = CurrentToken->Next; 1443 1444 resetTokenMetadata(); 1445 } 1446 1447 /// A struct to hold information valid in a specific context, e.g. 1448 /// a pair of parenthesis. 1449 struct Context { 1450 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1451 bool IsExpression) 1452 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1453 IsExpression(IsExpression) {} 1454 1455 tok::TokenKind ContextKind; 1456 unsigned BindingStrength; 1457 bool IsExpression; 1458 unsigned LongestObjCSelectorName = 0; 1459 bool ColonIsForRangeExpr = false; 1460 bool ColonIsDictLiteral = false; 1461 bool ColonIsObjCMethodExpr = false; 1462 FormatToken *FirstObjCSelectorName = nullptr; 1463 FormatToken *FirstStartOfName = nullptr; 1464 bool CanBeExpression = true; 1465 bool InTemplateArgument = false; 1466 bool InCtorInitializer = false; 1467 bool InInheritanceList = false; 1468 bool CaretFound = false; 1469 bool IsForEachMacro = false; 1470 bool InCpp11AttributeSpecifier = false; 1471 bool InCSharpAttributeSpecifier = false; 1472 bool InStructArrayInitializer = false; 1473 }; 1474 1475 /// Puts a new \c Context onto the stack \c Contexts for the lifetime 1476 /// of each instance. 1477 struct ScopedContextCreator { 1478 AnnotatingParser &P; 1479 1480 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1481 unsigned Increase) 1482 : P(P) { 1483 P.Contexts.push_back(Context(ContextKind, 1484 P.Contexts.back().BindingStrength + Increase, 1485 P.Contexts.back().IsExpression)); 1486 } 1487 1488 ~ScopedContextCreator() { 1489 if (P.Style.AlignArrayOfStructures != FormatStyle::AIAS_None) { 1490 if (P.Contexts.back().InStructArrayInitializer) { 1491 P.Contexts.pop_back(); 1492 P.Contexts.back().InStructArrayInitializer = true; 1493 return; 1494 } 1495 } 1496 P.Contexts.pop_back(); 1497 } 1498 }; 1499 1500 void modifyContext(const FormatToken &Current) { 1501 if (Current.getPrecedence() == prec::Assignment && 1502 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 1503 // Type aliases use `type X = ...;` in TypeScript and can be exported 1504 // using `export type ...`. 1505 !(Style.isJavaScript() && 1506 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1507 Line.startsWith(tok::kw_export, Keywords.kw_type, 1508 tok::identifier))) && 1509 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 1510 Contexts.back().IsExpression = true; 1511 if (!Line.startsWith(TT_UnaryOperator)) { 1512 for (FormatToken *Previous = Current.Previous; 1513 Previous && Previous->Previous && 1514 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1515 Previous = Previous->Previous) { 1516 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1517 Previous = Previous->MatchingParen; 1518 if (!Previous) 1519 break; 1520 } 1521 if (Previous->opensScope()) 1522 break; 1523 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1524 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1525 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1526 Previous->setType(TT_PointerOrReference); 1527 } 1528 } 1529 } else if (Current.is(tok::lessless) && 1530 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1531 Contexts.back().IsExpression = true; 1532 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1533 Contexts.back().IsExpression = true; 1534 } else if (Current.is(TT_TrailingReturnArrow)) { 1535 Contexts.back().IsExpression = false; 1536 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1537 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1538 } else if (Current.Previous && 1539 Current.Previous->is(TT_CtorInitializerColon)) { 1540 Contexts.back().IsExpression = true; 1541 Contexts.back().InCtorInitializer = true; 1542 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1543 Contexts.back().InInheritanceList = true; 1544 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1545 for (FormatToken *Previous = Current.Previous; 1546 Previous && Previous->isOneOf(tok::star, tok::amp); 1547 Previous = Previous->Previous) 1548 Previous->setType(TT_PointerOrReference); 1549 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 1550 Contexts.back().IsExpression = false; 1551 } else if (Current.is(tok::kw_new)) { 1552 Contexts.back().CanBeExpression = false; 1553 } else if (Current.is(tok::semi) || 1554 (Current.is(tok::exclaim) && Current.Previous && 1555 !Current.Previous->is(tok::kw_operator))) { 1556 // This should be the condition or increment in a for-loop. 1557 // But not operator !() (can't use TT_OverloadedOperator here as its not 1558 // been annotated yet). 1559 Contexts.back().IsExpression = true; 1560 } 1561 } 1562 1563 static FormatToken *untilMatchingParen(FormatToken *Current) { 1564 // Used when `MatchingParen` is not yet established. 1565 int ParenLevel = 0; 1566 while (Current) { 1567 if (Current->is(tok::l_paren)) 1568 ++ParenLevel; 1569 if (Current->is(tok::r_paren)) 1570 --ParenLevel; 1571 if (ParenLevel < 1) 1572 break; 1573 Current = Current->Next; 1574 } 1575 return Current; 1576 } 1577 1578 static bool isDeductionGuide(FormatToken &Current) { 1579 // Look for a deduction guide template<T> A(...) -> A<...>; 1580 if (Current.Previous && Current.Previous->is(tok::r_paren) && 1581 Current.startsSequence(tok::arrow, tok::identifier, tok::less)) { 1582 // Find the TemplateCloser. 1583 FormatToken *TemplateCloser = Current.Next->Next; 1584 int NestingLevel = 0; 1585 while (TemplateCloser) { 1586 // Skip over an expressions in parens A<(3 < 2)>; 1587 if (TemplateCloser->is(tok::l_paren)) { 1588 // No Matching Paren yet so skip to matching paren 1589 TemplateCloser = untilMatchingParen(TemplateCloser); 1590 if (!TemplateCloser) 1591 break; 1592 } 1593 if (TemplateCloser->is(tok::less)) 1594 ++NestingLevel; 1595 if (TemplateCloser->is(tok::greater)) 1596 --NestingLevel; 1597 if (NestingLevel < 1) 1598 break; 1599 TemplateCloser = TemplateCloser->Next; 1600 } 1601 // Assuming we have found the end of the template ensure its followed 1602 // with a semi-colon. 1603 if (TemplateCloser && TemplateCloser->Next && 1604 TemplateCloser->Next->is(tok::semi) && 1605 Current.Previous->MatchingParen) { 1606 // Determine if the identifier `A` prior to the A<..>; is the same as 1607 // prior to the A(..) 1608 FormatToken *LeadingIdentifier = 1609 Current.Previous->MatchingParen->Previous; 1610 1611 // Differentiate a deduction guide by seeing the 1612 // > of the template prior to the leading identifier. 1613 if (LeadingIdentifier) { 1614 FormatToken *PriorLeadingIdentifier = LeadingIdentifier->Previous; 1615 // Skip back past explicit decoration 1616 if (PriorLeadingIdentifier && 1617 PriorLeadingIdentifier->is(tok::kw_explicit)) 1618 PriorLeadingIdentifier = PriorLeadingIdentifier->Previous; 1619 1620 return (PriorLeadingIdentifier && 1621 PriorLeadingIdentifier->is(TT_TemplateCloser) && 1622 LeadingIdentifier->TokenText == Current.Next->TokenText); 1623 } 1624 } 1625 } 1626 return false; 1627 } 1628 1629 void determineTokenType(FormatToken &Current) { 1630 if (!Current.is(TT_Unknown)) 1631 // The token type is already known. 1632 return; 1633 1634 if ((Style.isJavaScript() || Style.isCSharp()) && 1635 Current.is(tok::exclaim)) { 1636 if (Current.Previous) { 1637 bool IsIdentifier = 1638 Style.isJavaScript() 1639 ? Keywords.IsJavaScriptIdentifier( 1640 *Current.Previous, /* AcceptIdentifierName= */ true) 1641 : Current.Previous->is(tok::identifier); 1642 if (IsIdentifier || 1643 Current.Previous->isOneOf( 1644 tok::kw_namespace, tok::r_paren, tok::r_square, tok::r_brace, 1645 tok::kw_false, tok::kw_true, Keywords.kw_type, Keywords.kw_get, 1646 Keywords.kw_set) || 1647 Current.Previous->Tok.isLiteral()) { 1648 Current.setType(TT_NonNullAssertion); 1649 return; 1650 } 1651 } 1652 if (Current.Next && 1653 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1654 Current.setType(TT_NonNullAssertion); 1655 return; 1656 } 1657 } 1658 1659 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1660 // function declaration have been found. In this case, 'Current' is a 1661 // trailing token of this declaration and thus cannot be a name. 1662 if (Current.is(Keywords.kw_instanceof)) { 1663 Current.setType(TT_BinaryOperator); 1664 } else if (isStartOfName(Current) && 1665 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1666 Contexts.back().FirstStartOfName = &Current; 1667 Current.setType(TT_StartOfName); 1668 } else if (Current.is(tok::semi)) { 1669 // Reset FirstStartOfName after finding a semicolon so that a for loop 1670 // with multiple increment statements is not confused with a for loop 1671 // having multiple variable declarations. 1672 Contexts.back().FirstStartOfName = nullptr; 1673 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1674 AutoFound = true; 1675 } else if (Current.is(tok::arrow) && 1676 Style.Language == FormatStyle::LK_Java) { 1677 Current.setType(TT_LambdaArrow); 1678 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1679 Current.NestingLevel == 0 && 1680 !Current.Previous->isOneOf(tok::kw_operator, tok::identifier)) { 1681 // not auto operator->() -> xxx; 1682 Current.setType(TT_TrailingReturnArrow); 1683 } else if (Current.is(tok::arrow) && Current.Previous && 1684 Current.Previous->is(tok::r_brace)) { 1685 // Concept implicit conversion constraint needs to be treated like 1686 // a trailing return type ... } -> <type>. 1687 Current.setType(TT_TrailingReturnArrow); 1688 } else if (isDeductionGuide(Current)) { 1689 // Deduction guides trailing arrow " A(...) -> A<T>;". 1690 Current.setType(TT_TrailingReturnArrow); 1691 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1692 Current.setType(determineStarAmpUsage( 1693 Current, 1694 Contexts.back().CanBeExpression && Contexts.back().IsExpression, 1695 Contexts.back().InTemplateArgument)); 1696 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1697 Current.setType(determinePlusMinusCaretUsage(Current)); 1698 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1699 Contexts.back().CaretFound = true; 1700 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1701 Current.setType(determineIncrementUsage(Current)); 1702 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1703 Current.setType(TT_UnaryOperator); 1704 } else if (Current.is(tok::question)) { 1705 if (Style.isJavaScript() && Line.MustBeDeclaration && 1706 !Contexts.back().IsExpression) { 1707 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1708 // on the interface, not a ternary expression. 1709 Current.setType(TT_JsTypeOptionalQuestion); 1710 } else { 1711 Current.setType(TT_ConditionalExpr); 1712 } 1713 } else if (Current.isBinaryOperator() && 1714 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1715 (!Current.is(tok::greater) && 1716 Style.Language != FormatStyle::LK_TextProto)) { 1717 Current.setType(TT_BinaryOperator); 1718 } else if (Current.is(tok::comment)) { 1719 if (Current.TokenText.startswith("/*")) { 1720 if (Current.TokenText.endswith("*/")) 1721 Current.setType(TT_BlockComment); 1722 else 1723 // The lexer has for some reason determined a comment here. But we 1724 // cannot really handle it, if it isn't properly terminated. 1725 Current.Tok.setKind(tok::unknown); 1726 } else { 1727 Current.setType(TT_LineComment); 1728 } 1729 } else if (Current.is(tok::r_paren)) { 1730 if (rParenEndsCast(Current)) 1731 Current.setType(TT_CastRParen); 1732 if (Current.MatchingParen && Current.Next && 1733 !Current.Next->isBinaryOperator() && 1734 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1735 tok::comma, tok::period, tok::arrow, 1736 tok::coloncolon)) 1737 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1738 // Make sure this isn't the return type of an Obj-C block declaration 1739 if (AfterParen->Tok.isNot(tok::caret)) { 1740 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1741 if (BeforeParen->is(tok::identifier) && 1742 !BeforeParen->is(TT_TypenameMacro) && 1743 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1744 (!BeforeParen->Previous || 1745 BeforeParen->Previous->ClosesTemplateDeclaration)) 1746 Current.setType(TT_FunctionAnnotationRParen); 1747 } 1748 } 1749 } else if (Current.is(tok::at) && Current.Next && !Style.isJavaScript() && 1750 Style.Language != FormatStyle::LK_Java) { 1751 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1752 // marks declarations and properties that need special formatting. 1753 switch (Current.Next->Tok.getObjCKeywordID()) { 1754 case tok::objc_interface: 1755 case tok::objc_implementation: 1756 case tok::objc_protocol: 1757 Current.setType(TT_ObjCDecl); 1758 break; 1759 case tok::objc_property: 1760 Current.setType(TT_ObjCProperty); 1761 break; 1762 default: 1763 break; 1764 } 1765 } else if (Current.is(tok::period)) { 1766 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1767 if (PreviousNoComment && 1768 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1769 Current.setType(TT_DesignatedInitializerPeriod); 1770 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1771 Current.Previous->isOneOf(TT_JavaAnnotation, 1772 TT_LeadingJavaAnnotation)) { 1773 Current.setType(Current.Previous->getType()); 1774 } 1775 } else if (canBeObjCSelectorComponent(Current) && 1776 // FIXME(bug 36976): ObjC return types shouldn't use 1777 // TT_CastRParen. 1778 Current.Previous && Current.Previous->is(TT_CastRParen) && 1779 Current.Previous->MatchingParen && 1780 Current.Previous->MatchingParen->Previous && 1781 Current.Previous->MatchingParen->Previous->is( 1782 TT_ObjCMethodSpecifier)) { 1783 // This is the first part of an Objective-C selector name. (If there's no 1784 // colon after this, this is the only place which annotates the identifier 1785 // as a selector.) 1786 Current.setType(TT_SelectorName); 1787 } else if (Current.isOneOf(tok::identifier, tok::kw_const, tok::kw_noexcept, 1788 tok::kw_requires) && 1789 Current.Previous && 1790 !Current.Previous->isOneOf(tok::equal, tok::at) && 1791 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1792 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1793 // function declaration have been found. 1794 Current.setType(TT_TrailingAnnotation); 1795 } else if ((Style.Language == FormatStyle::LK_Java || 1796 Style.isJavaScript()) && 1797 Current.Previous) { 1798 if (Current.Previous->is(tok::at) && 1799 Current.isNot(Keywords.kw_interface)) { 1800 const FormatToken &AtToken = *Current.Previous; 1801 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1802 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1803 Current.setType(TT_LeadingJavaAnnotation); 1804 else 1805 Current.setType(TT_JavaAnnotation); 1806 } else if (Current.Previous->is(tok::period) && 1807 Current.Previous->isOneOf(TT_JavaAnnotation, 1808 TT_LeadingJavaAnnotation)) { 1809 Current.setType(Current.Previous->getType()); 1810 } 1811 } 1812 } 1813 1814 /// Take a guess at whether \p Tok starts a name of a function or 1815 /// variable declaration. 1816 /// 1817 /// This is a heuristic based on whether \p Tok is an identifier following 1818 /// something that is likely a type. 1819 bool isStartOfName(const FormatToken &Tok) { 1820 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1821 return false; 1822 1823 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1824 Keywords.kw_as)) 1825 return false; 1826 if (Style.isJavaScript() && Tok.Previous->is(Keywords.kw_in)) 1827 return false; 1828 1829 // Skip "const" as it does not have an influence on whether this is a name. 1830 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1831 1832 // For javascript const can be like "let" or "var" 1833 if (!Style.isJavaScript()) 1834 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1835 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1836 1837 if (!PreviousNotConst) 1838 return false; 1839 1840 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1841 PreviousNotConst->Previous && 1842 PreviousNotConst->Previous->is(tok::hash); 1843 1844 if (PreviousNotConst->is(TT_TemplateCloser)) 1845 return PreviousNotConst && PreviousNotConst->MatchingParen && 1846 PreviousNotConst->MatchingParen->Previous && 1847 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1848 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1849 1850 if (PreviousNotConst->is(tok::r_paren) && 1851 PreviousNotConst->is(TT_TypeDeclarationParen)) 1852 return true; 1853 1854 // If is a preprocess keyword like #define. 1855 if (IsPPKeyword) 1856 return false; 1857 1858 // int a or auto a. 1859 if (PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) 1860 return true; 1861 1862 // *a or &a or &&a. 1863 if (PreviousNotConst->is(TT_PointerOrReference)) 1864 return true; 1865 1866 // MyClass a; 1867 if (PreviousNotConst->isSimpleTypeSpecifier()) 1868 return true; 1869 1870 // const a = in JavaScript. 1871 return (Style.isJavaScript() && PreviousNotConst->is(tok::kw_const)); 1872 } 1873 1874 /// Determine whether ')' is ending a cast. 1875 bool rParenEndsCast(const FormatToken &Tok) { 1876 // C-style casts are only used in C++, C# and Java. 1877 if (!Style.isCSharp() && !Style.isCpp() && 1878 Style.Language != FormatStyle::LK_Java) 1879 return false; 1880 1881 // Empty parens aren't casts and there are no casts at the end of the line. 1882 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1883 return false; 1884 1885 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1886 if (LeftOfParens) { 1887 // If there is a closing parenthesis left of the current 1888 // parentheses, look past it as these might be chained casts. 1889 if (LeftOfParens->is(tok::r_paren) && 1890 LeftOfParens->isNot(TT_CastRParen)) { 1891 if (!LeftOfParens->MatchingParen || 1892 !LeftOfParens->MatchingParen->Previous) 1893 return false; 1894 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1895 } 1896 1897 // The Condition directly below this one will see the operator arguments 1898 // as a (void *foo) cast. 1899 // void operator delete(void *foo) ATTRIB; 1900 if (LeftOfParens->Tok.getIdentifierInfo() && LeftOfParens->Previous && 1901 LeftOfParens->Previous->is(tok::kw_operator)) 1902 return false; 1903 1904 // If there is an identifier (or with a few exceptions a keyword) right 1905 // before the parentheses, this is unlikely to be a cast. 1906 if (LeftOfParens->Tok.getIdentifierInfo() && 1907 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1908 tok::kw_delete)) 1909 return false; 1910 1911 // Certain other tokens right before the parentheses are also signals that 1912 // this cannot be a cast. 1913 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1914 TT_TemplateCloser, tok::ellipsis)) 1915 return false; 1916 } 1917 1918 if (Tok.Next->is(tok::question)) 1919 return false; 1920 1921 // `foreach((A a, B b) in someList)` should not be seen as a cast. 1922 if (Tok.Next->is(Keywords.kw_in) && Style.isCSharp()) 1923 return false; 1924 1925 // Functions which end with decorations like volatile, noexcept are unlikely 1926 // to be casts. 1927 if (Tok.Next->isOneOf(tok::kw_noexcept, tok::kw_volatile, tok::kw_const, 1928 tok::kw_requires, tok::kw_throw, tok::arrow, 1929 Keywords.kw_override, Keywords.kw_final) || 1930 isCpp11AttributeSpecifier(*Tok.Next)) 1931 return false; 1932 1933 // As Java has no function types, a "(" after the ")" likely means that this 1934 // is a cast. 1935 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1936 return true; 1937 1938 // If a (non-string) literal follows, this is likely a cast. 1939 if (Tok.Next->isNot(tok::string_literal) && 1940 (Tok.Next->Tok.isLiteral() || 1941 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1942 return true; 1943 1944 // Heuristically try to determine whether the parentheses contain a type. 1945 auto IsQualifiedPointerOrReference = [](FormatToken *T) { 1946 // This is used to handle cases such as x = (foo *const)&y; 1947 assert(!T->isSimpleTypeSpecifier() && "Should have already been checked"); 1948 // Strip trailing qualifiers such as const or volatile when checking 1949 // whether the parens could be a cast to a pointer/reference type. 1950 while (T) { 1951 if (T->is(TT_AttributeParen)) { 1952 // Handle `x = (foo *__attribute__((foo)))&v;`: 1953 if (T->MatchingParen && T->MatchingParen->Previous && 1954 T->MatchingParen->Previous->is(tok::kw___attribute)) { 1955 T = T->MatchingParen->Previous->Previous; 1956 continue; 1957 } 1958 } else if (T->is(TT_AttributeSquare)) { 1959 // Handle `x = (foo *[[clang::foo]])&v;`: 1960 if (T->MatchingParen && T->MatchingParen->Previous) { 1961 T = T->MatchingParen->Previous; 1962 continue; 1963 } 1964 } else if (T->canBePointerOrReferenceQualifier()) { 1965 T = T->Previous; 1966 continue; 1967 } 1968 break; 1969 } 1970 return T && T->is(TT_PointerOrReference); 1971 }; 1972 bool ParensAreType = 1973 !Tok.Previous || 1974 Tok.Previous->isOneOf(TT_TemplateCloser, TT_TypeDeclarationParen) || 1975 Tok.Previous->isSimpleTypeSpecifier() || 1976 IsQualifiedPointerOrReference(Tok.Previous); 1977 bool ParensCouldEndDecl = 1978 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1979 if (ParensAreType && !ParensCouldEndDecl) 1980 return true; 1981 1982 // At this point, we heuristically assume that there are no casts at the 1983 // start of the line. We assume that we have found most cases where there 1984 // are by the logic above, e.g. "(void)x;". 1985 if (!LeftOfParens) 1986 return false; 1987 1988 // Certain token types inside the parentheses mean that this can't be a 1989 // cast. 1990 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 1991 Token = Token->Next) 1992 if (Token->is(TT_BinaryOperator)) 1993 return false; 1994 1995 // If the following token is an identifier or 'this', this is a cast. All 1996 // cases where this can be something else are handled above. 1997 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1998 return true; 1999 2000 // Look for a cast `( x ) (`. 2001 if (Tok.Next->is(tok::l_paren) && Tok.Previous && Tok.Previous->Previous) { 2002 if (Tok.Previous->is(tok::identifier) && 2003 Tok.Previous->Previous->is(tok::l_paren)) 2004 return true; 2005 } 2006 2007 if (!Tok.Next->Next) 2008 return false; 2009 2010 // If the next token after the parenthesis is a unary operator, assume 2011 // that this is cast, unless there are unexpected tokens inside the 2012 // parenthesis. 2013 bool NextIsUnary = 2014 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 2015 if (!NextIsUnary || Tok.Next->is(tok::plus) || 2016 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 2017 return false; 2018 // Search for unexpected tokens. 2019 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 2020 Prev = Prev->Previous) { 2021 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 2022 return false; 2023 } 2024 return true; 2025 } 2026 2027 /// Return the type of the given token assuming it is * or &. 2028 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 2029 bool InTemplateArgument) { 2030 if (Style.isJavaScript()) 2031 return TT_BinaryOperator; 2032 2033 // && in C# must be a binary operator. 2034 if (Style.isCSharp() && Tok.is(tok::ampamp)) 2035 return TT_BinaryOperator; 2036 2037 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2038 if (!PrevToken) 2039 return TT_UnaryOperator; 2040 2041 const FormatToken *NextToken = Tok.getNextNonComment(); 2042 if (!NextToken || 2043 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_noexcept) || 2044 NextToken->canBePointerOrReferenceQualifier() || 2045 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 2046 return TT_PointerOrReference; 2047 2048 if (PrevToken->is(tok::coloncolon)) 2049 return TT_PointerOrReference; 2050 2051 if (PrevToken->is(tok::r_paren) && PrevToken->is(TT_TypeDeclarationParen)) 2052 return TT_PointerOrReference; 2053 2054 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 2055 tok::comma, tok::semi, tok::kw_return, tok::colon, 2056 tok::kw_co_return, tok::kw_co_await, 2057 tok::kw_co_yield, tok::equal, tok::kw_delete, 2058 tok::kw_sizeof, tok::kw_throw, TT_BinaryOperator, 2059 TT_ConditionalExpr, TT_UnaryOperator, TT_CastRParen)) 2060 return TT_UnaryOperator; 2061 2062 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 2063 return TT_PointerOrReference; 2064 if (NextToken->is(tok::kw_operator) && !IsExpression) 2065 return TT_PointerOrReference; 2066 if (NextToken->isOneOf(tok::comma, tok::semi)) 2067 return TT_PointerOrReference; 2068 2069 if (PrevToken->Tok.isLiteral() || 2070 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 2071 tok::kw_false, tok::r_brace) || 2072 NextToken->Tok.isLiteral() || 2073 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 2074 NextToken->isUnaryOperator() || 2075 // If we know we're in a template argument, there are no named 2076 // declarations. Thus, having an identifier on the right-hand side 2077 // indicates a binary operator. 2078 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 2079 return TT_BinaryOperator; 2080 2081 // "&&(" is quite unlikely to be two successive unary "&". 2082 if (Tok.is(tok::ampamp) && NextToken->is(tok::l_paren)) 2083 return TT_BinaryOperator; 2084 2085 // This catches some cases where evaluation order is used as control flow: 2086 // aaa && aaa->f(); 2087 if (NextToken->Tok.isAnyIdentifier()) { 2088 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 2089 if (NextNextToken && NextNextToken->is(tok::arrow)) 2090 return TT_BinaryOperator; 2091 } 2092 2093 // It is very unlikely that we are going to find a pointer or reference type 2094 // definition on the RHS of an assignment. 2095 if (IsExpression && !Contexts.back().CaretFound) 2096 return TT_BinaryOperator; 2097 2098 return TT_PointerOrReference; 2099 } 2100 2101 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 2102 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2103 if (!PrevToken) 2104 return TT_UnaryOperator; 2105 2106 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 2107 // This must be a sequence of leading unary operators. 2108 return TT_UnaryOperator; 2109 2110 // Use heuristics to recognize unary operators. 2111 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 2112 tok::question, tok::colon, tok::kw_return, 2113 tok::kw_case, tok::at, tok::l_brace, tok::kw_throw, 2114 tok::kw_co_return, tok::kw_co_yield)) 2115 return TT_UnaryOperator; 2116 2117 // There can't be two consecutive binary operators. 2118 if (PrevToken->is(TT_BinaryOperator)) 2119 return TT_UnaryOperator; 2120 2121 // Fall back to marking the token as binary operator. 2122 return TT_BinaryOperator; 2123 } 2124 2125 /// Determine whether ++/-- are pre- or post-increments/-decrements. 2126 TokenType determineIncrementUsage(const FormatToken &Tok) { 2127 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2128 if (!PrevToken || PrevToken->is(TT_CastRParen)) 2129 return TT_UnaryOperator; 2130 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 2131 return TT_TrailingUnaryOperator; 2132 2133 return TT_UnaryOperator; 2134 } 2135 2136 SmallVector<Context, 8> Contexts; 2137 2138 const FormatStyle &Style; 2139 AnnotatedLine &Line; 2140 FormatToken *CurrentToken; 2141 bool AutoFound; 2142 const AdditionalKeywords &Keywords; 2143 2144 // Set of "<" tokens that do not open a template parameter list. If parseAngle 2145 // determines that a specific token can't be a template opener, it will make 2146 // same decision irrespective of the decisions for tokens leading up to it. 2147 // Store this information to prevent this from causing exponential runtime. 2148 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 2149 }; 2150 2151 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 2152 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 2153 2154 /// Parses binary expressions by inserting fake parenthesis based on 2155 /// operator precedence. 2156 class ExpressionParser { 2157 public: 2158 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 2159 AnnotatedLine &Line) 2160 : Style(Style), Keywords(Keywords), Current(Line.First) {} 2161 2162 /// Parse expressions with the given operator precedence. 2163 void parse(int Precedence = 0) { 2164 // Skip 'return' and ObjC selector colons as they are not part of a binary 2165 // expression. 2166 while (Current && (Current->is(tok::kw_return) || 2167 (Current->is(tok::colon) && 2168 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 2169 next(); 2170 2171 if (!Current || Precedence > PrecedenceArrowAndPeriod) 2172 return; 2173 2174 // Conditional expressions need to be parsed separately for proper nesting. 2175 if (Precedence == prec::Conditional) { 2176 parseConditionalExpr(); 2177 return; 2178 } 2179 2180 // Parse unary operators, which all have a higher precedence than binary 2181 // operators. 2182 if (Precedence == PrecedenceUnaryOperator) { 2183 parseUnaryOperator(); 2184 return; 2185 } 2186 2187 FormatToken *Start = Current; 2188 FormatToken *LatestOperator = nullptr; 2189 unsigned OperatorIndex = 0; 2190 2191 while (Current) { 2192 // Consume operators with higher precedence. 2193 parse(Precedence + 1); 2194 2195 int CurrentPrecedence = getCurrentPrecedence(); 2196 2197 if (Precedence == CurrentPrecedence && Current && 2198 Current->is(TT_SelectorName)) { 2199 if (LatestOperator) 2200 addFakeParenthesis(Start, prec::Level(Precedence)); 2201 Start = Current; 2202 } 2203 2204 // At the end of the line or when an operator with higher precedence is 2205 // found, insert fake parenthesis and return. 2206 if (!Current || 2207 (Current->closesScope() && 2208 (Current->MatchingParen || Current->is(TT_TemplateString))) || 2209 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 2210 (CurrentPrecedence == prec::Conditional && 2211 Precedence == prec::Assignment && Current->is(tok::colon))) { 2212 break; 2213 } 2214 2215 // Consume scopes: (), [], <> and {} 2216 if (Current->opensScope()) { 2217 // In fragment of a JavaScript template string can look like '}..${' and 2218 // thus close a scope and open a new one at the same time. 2219 while (Current && (!Current->closesScope() || Current->opensScope())) { 2220 next(); 2221 parse(); 2222 } 2223 next(); 2224 } else { 2225 // Operator found. 2226 if (CurrentPrecedence == Precedence) { 2227 if (LatestOperator) 2228 LatestOperator->NextOperator = Current; 2229 LatestOperator = Current; 2230 Current->OperatorIndex = OperatorIndex; 2231 ++OperatorIndex; 2232 } 2233 next(/*SkipPastLeadingComments=*/Precedence > 0); 2234 } 2235 } 2236 2237 if (LatestOperator && (Current || Precedence > 0)) { 2238 // LatestOperator->LastOperator = true; 2239 if (Precedence == PrecedenceArrowAndPeriod) { 2240 // Call expressions don't have a binary operator precedence. 2241 addFakeParenthesis(Start, prec::Unknown); 2242 } else { 2243 addFakeParenthesis(Start, prec::Level(Precedence)); 2244 } 2245 } 2246 } 2247 2248 private: 2249 /// Gets the precedence (+1) of the given token for binary operators 2250 /// and other tokens that we treat like binary operators. 2251 int getCurrentPrecedence() { 2252 if (Current) { 2253 const FormatToken *NextNonComment = Current->getNextNonComment(); 2254 if (Current->is(TT_ConditionalExpr)) 2255 return prec::Conditional; 2256 if (NextNonComment && Current->is(TT_SelectorName) && 2257 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 2258 ((Style.Language == FormatStyle::LK_Proto || 2259 Style.Language == FormatStyle::LK_TextProto) && 2260 NextNonComment->is(tok::less)))) 2261 return prec::Assignment; 2262 if (Current->is(TT_JsComputedPropertyName)) 2263 return prec::Assignment; 2264 if (Current->is(TT_LambdaArrow)) 2265 return prec::Comma; 2266 if (Current->is(TT_FatArrow)) 2267 return prec::Assignment; 2268 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 2269 (Current->is(tok::comment) && NextNonComment && 2270 NextNonComment->is(TT_SelectorName))) 2271 return 0; 2272 if (Current->is(TT_RangeBasedForLoopColon)) 2273 return prec::Comma; 2274 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2275 Current->is(Keywords.kw_instanceof)) 2276 return prec::Relational; 2277 if (Style.isJavaScript() && 2278 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 2279 return prec::Relational; 2280 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 2281 return Current->getPrecedence(); 2282 if (Current->isOneOf(tok::period, tok::arrow)) 2283 return PrecedenceArrowAndPeriod; 2284 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2285 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 2286 Keywords.kw_throws)) 2287 return 0; 2288 } 2289 return -1; 2290 } 2291 2292 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 2293 Start->FakeLParens.push_back(Precedence); 2294 if (Precedence > prec::Unknown) 2295 Start->StartsBinaryExpression = true; 2296 if (Current) { 2297 FormatToken *Previous = Current->Previous; 2298 while (Previous->is(tok::comment) && Previous->Previous) 2299 Previous = Previous->Previous; 2300 ++Previous->FakeRParens; 2301 if (Precedence > prec::Unknown) 2302 Previous->EndsBinaryExpression = true; 2303 } 2304 } 2305 2306 /// Parse unary operator expressions and surround them with fake 2307 /// parentheses if appropriate. 2308 void parseUnaryOperator() { 2309 llvm::SmallVector<FormatToken *, 2> Tokens; 2310 while (Current && Current->is(TT_UnaryOperator)) { 2311 Tokens.push_back(Current); 2312 next(); 2313 } 2314 parse(PrecedenceArrowAndPeriod); 2315 for (FormatToken *Token : llvm::reverse(Tokens)) 2316 // The actual precedence doesn't matter. 2317 addFakeParenthesis(Token, prec::Unknown); 2318 } 2319 2320 void parseConditionalExpr() { 2321 while (Current && Current->isTrailingComment()) { 2322 next(); 2323 } 2324 FormatToken *Start = Current; 2325 parse(prec::LogicalOr); 2326 if (!Current || !Current->is(tok::question)) 2327 return; 2328 next(); 2329 parse(prec::Assignment); 2330 if (!Current || Current->isNot(TT_ConditionalExpr)) 2331 return; 2332 next(); 2333 parse(prec::Assignment); 2334 addFakeParenthesis(Start, prec::Conditional); 2335 } 2336 2337 void next(bool SkipPastLeadingComments = true) { 2338 if (Current) 2339 Current = Current->Next; 2340 while (Current && 2341 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 2342 Current->isTrailingComment()) 2343 Current = Current->Next; 2344 } 2345 2346 const FormatStyle &Style; 2347 const AdditionalKeywords &Keywords; 2348 FormatToken *Current; 2349 }; 2350 2351 } // end anonymous namespace 2352 2353 void TokenAnnotator::setCommentLineLevels( 2354 SmallVectorImpl<AnnotatedLine *> &Lines) { 2355 const AnnotatedLine *NextNonCommentLine = nullptr; 2356 for (AnnotatedLine *Line : llvm::reverse(Lines)) { 2357 assert(Line->First); 2358 bool CommentLine = true; 2359 for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) { 2360 if (!Tok->is(tok::comment)) { 2361 CommentLine = false; 2362 break; 2363 } 2364 } 2365 2366 // If the comment is currently aligned with the line immediately following 2367 // it, that's probably intentional and we should keep it. 2368 if (NextNonCommentLine && CommentLine && 2369 NextNonCommentLine->First->NewlinesBefore <= 1 && 2370 NextNonCommentLine->First->OriginalColumn == 2371 Line->First->OriginalColumn) { 2372 // Align comments for preprocessor lines with the # in column 0 if 2373 // preprocessor lines are not indented. Otherwise, align with the next 2374 // line. 2375 Line->Level = 2376 (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 2377 (NextNonCommentLine->Type == LT_PreprocessorDirective || 2378 NextNonCommentLine->Type == LT_ImportStatement)) 2379 ? 0 2380 : NextNonCommentLine->Level; 2381 } else { 2382 NextNonCommentLine = Line->First->isNot(tok::r_brace) ? Line : nullptr; 2383 } 2384 2385 setCommentLineLevels(Line->Children); 2386 } 2387 } 2388 2389 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 2390 unsigned Result = 0; 2391 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 2392 Result = std::max(Result, Tok->NestingLevel); 2393 return Result; 2394 } 2395 2396 void TokenAnnotator::annotate(AnnotatedLine &Line) { 2397 for (auto &Child : Line.Children) 2398 annotate(*Child); 2399 2400 AnnotatingParser Parser(Style, Line, Keywords); 2401 Line.Type = Parser.parseLine(); 2402 2403 // With very deep nesting, ExpressionParser uses lots of stack and the 2404 // formatting algorithm is very slow. We're not going to do a good job here 2405 // anyway - it's probably generated code being formatted by mistake. 2406 // Just skip the whole line. 2407 if (maxNestingDepth(Line) > 50) 2408 Line.Type = LT_Invalid; 2409 2410 if (Line.Type == LT_Invalid) 2411 return; 2412 2413 ExpressionParser ExprParser(Style, Keywords, Line); 2414 ExprParser.parse(); 2415 2416 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2417 Line.Type = LT_ObjCMethodDecl; 2418 else if (Line.startsWith(TT_ObjCDecl)) 2419 Line.Type = LT_ObjCDecl; 2420 else if (Line.startsWith(TT_ObjCProperty)) 2421 Line.Type = LT_ObjCProperty; 2422 2423 Line.First->SpacesRequiredBefore = 1; 2424 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 2425 } 2426 2427 // This function heuristically determines whether 'Current' starts the name of a 2428 // function declaration. 2429 static bool isFunctionDeclarationName(bool IsCpp, const FormatToken &Current, 2430 const AnnotatedLine &Line) { 2431 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 2432 for (; Next; Next = Next->Next) { 2433 if (Next->is(TT_OverloadedOperatorLParen)) 2434 return Next; 2435 if (Next->is(TT_OverloadedOperator)) 2436 continue; 2437 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 2438 // For 'new[]' and 'delete[]'. 2439 if (Next->Next && 2440 Next->Next->startsSequence(tok::l_square, tok::r_square)) 2441 Next = Next->Next->Next; 2442 continue; 2443 } 2444 if (Next->startsSequence(tok::l_square, tok::r_square)) { 2445 // For operator[](). 2446 Next = Next->Next; 2447 continue; 2448 } 2449 if ((Next->isSimpleTypeSpecifier() || Next->is(tok::identifier)) && 2450 Next->Next && Next->Next->isOneOf(tok::star, tok::amp, tok::ampamp)) { 2451 // For operator void*(), operator char*(), operator Foo*(). 2452 Next = Next->Next; 2453 continue; 2454 } 2455 if (Next->is(TT_TemplateOpener) && Next->MatchingParen) { 2456 Next = Next->MatchingParen; 2457 continue; 2458 } 2459 2460 break; 2461 } 2462 return nullptr; 2463 }; 2464 2465 // Find parentheses of parameter list. 2466 const FormatToken *Next = Current.Next; 2467 if (Current.is(tok::kw_operator)) { 2468 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 2469 return false; 2470 Next = skipOperatorName(Next); 2471 } else { 2472 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 2473 return false; 2474 for (; Next; Next = Next->Next) { 2475 if (Next->is(TT_TemplateOpener)) { 2476 Next = Next->MatchingParen; 2477 } else if (Next->is(tok::coloncolon)) { 2478 Next = Next->Next; 2479 if (!Next) 2480 return false; 2481 if (Next->is(tok::kw_operator)) { 2482 Next = skipOperatorName(Next->Next); 2483 break; 2484 } 2485 if (!Next->is(tok::identifier)) 2486 return false; 2487 } else if (Next->is(tok::l_paren)) { 2488 break; 2489 } else { 2490 return false; 2491 } 2492 } 2493 } 2494 2495 // Check whether parameter list can belong to a function declaration. 2496 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 2497 return false; 2498 // If the lines ends with "{", this is likely a function definition. 2499 if (Line.Last->is(tok::l_brace)) 2500 return true; 2501 if (Next->Next == Next->MatchingParen) 2502 return true; // Empty parentheses. 2503 // If there is an &/&& after the r_paren, this is likely a function. 2504 if (Next->MatchingParen->Next && 2505 Next->MatchingParen->Next->is(TT_PointerOrReference)) 2506 return true; 2507 2508 // Check for K&R C function definitions (and C++ function definitions with 2509 // unnamed parameters), e.g.: 2510 // int f(i) 2511 // { 2512 // return i + 1; 2513 // } 2514 // bool g(size_t = 0, bool b = false) 2515 // { 2516 // return !b; 2517 // } 2518 if (IsCpp && Next->Next && Next->Next->is(tok::identifier) && 2519 !Line.endsWith(tok::semi)) 2520 return true; 2521 2522 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 2523 Tok = Tok->Next) { 2524 if (Tok->is(TT_TypeDeclarationParen)) 2525 return true; 2526 if (Tok->isOneOf(tok::l_paren, TT_TemplateOpener) && Tok->MatchingParen) { 2527 Tok = Tok->MatchingParen; 2528 continue; 2529 } 2530 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 2531 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 2532 return true; 2533 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 2534 Tok->Tok.isLiteral()) 2535 return false; 2536 } 2537 return false; 2538 } 2539 2540 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 2541 assert(Line.MightBeFunctionDecl); 2542 2543 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 2544 Style.AlwaysBreakAfterReturnType == 2545 FormatStyle::RTBS_TopLevelDefinitions) && 2546 Line.Level > 0) 2547 return false; 2548 2549 switch (Style.AlwaysBreakAfterReturnType) { 2550 case FormatStyle::RTBS_None: 2551 return false; 2552 case FormatStyle::RTBS_All: 2553 case FormatStyle::RTBS_TopLevel: 2554 return true; 2555 case FormatStyle::RTBS_AllDefinitions: 2556 case FormatStyle::RTBS_TopLevelDefinitions: 2557 return Line.mightBeFunctionDefinition(); 2558 } 2559 2560 return false; 2561 } 2562 2563 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 2564 for (AnnotatedLine *ChildLine : Line.Children) 2565 calculateFormattingInformation(*ChildLine); 2566 2567 Line.First->TotalLength = 2568 Line.First->IsMultiline ? Style.ColumnLimit 2569 : Line.FirstStartColumn + Line.First->ColumnWidth; 2570 FormatToken *Current = Line.First->Next; 2571 bool InFunctionDecl = Line.MightBeFunctionDecl; 2572 bool AlignArrayOfStructures = 2573 (Style.AlignArrayOfStructures != FormatStyle::AIAS_None && 2574 Line.Type == LT_ArrayOfStructInitializer); 2575 if (AlignArrayOfStructures) 2576 calculateArrayInitializerColumnList(Line); 2577 2578 while (Current) { 2579 if (isFunctionDeclarationName(Style.isCpp(), *Current, Line)) 2580 Current->setType(TT_FunctionDeclarationName); 2581 const FormatToken *Prev = Current->Previous; 2582 if (Current->is(TT_LineComment)) { 2583 if (Prev->is(BK_BracedInit) && Prev->opensScope()) 2584 Current->SpacesRequiredBefore = 2585 (Style.Cpp11BracedListStyle && !Style.SpacesInParentheses) ? 0 : 1; 2586 else 2587 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2588 2589 // If we find a trailing comment, iterate backwards to determine whether 2590 // it seems to relate to a specific parameter. If so, break before that 2591 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2592 // to the previous line in: 2593 // SomeFunction(a, 2594 // b, // comment 2595 // c); 2596 if (!Current->HasUnescapedNewline) { 2597 for (FormatToken *Parameter = Current->Previous; Parameter; 2598 Parameter = Parameter->Previous) { 2599 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 2600 break; 2601 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 2602 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 2603 Parameter->HasUnescapedNewline) 2604 Parameter->MustBreakBefore = true; 2605 break; 2606 } 2607 } 2608 } 2609 } else if (Current->SpacesRequiredBefore == 0 && 2610 spaceRequiredBefore(Line, *Current)) { 2611 Current->SpacesRequiredBefore = 1; 2612 } 2613 2614 Current->MustBreakBefore = 2615 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 2616 2617 if (!Current->MustBreakBefore && InFunctionDecl && 2618 Current->is(TT_FunctionDeclarationName)) 2619 Current->MustBreakBefore = mustBreakForReturnType(Line); 2620 2621 Current->CanBreakBefore = 2622 Current->MustBreakBefore || canBreakBefore(Line, *Current); 2623 unsigned ChildSize = 0; 2624 if (Prev->Children.size() == 1) { 2625 FormatToken &LastOfChild = *Prev->Children[0]->Last; 2626 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 2627 : LastOfChild.TotalLength + 1; 2628 } 2629 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 2630 (Prev->Children.size() == 1 && 2631 Prev->Children[0]->First->MustBreakBefore) || 2632 Current->IsMultiline) 2633 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 2634 else 2635 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 2636 ChildSize + Current->SpacesRequiredBefore; 2637 2638 if (Current->is(TT_CtorInitializerColon)) 2639 InFunctionDecl = false; 2640 2641 // FIXME: Only calculate this if CanBreakBefore is true once static 2642 // initializers etc. are sorted out. 2643 // FIXME: Move magic numbers to a better place. 2644 2645 // Reduce penalty for aligning ObjC method arguments using the colon 2646 // alignment as this is the canonical way (still prefer fitting everything 2647 // into one line if possible). Trying to fit a whole expression into one 2648 // line should not force other line breaks (e.g. when ObjC method 2649 // expression is a part of other expression). 2650 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl); 2651 if (Style.Language == FormatStyle::LK_ObjC && 2652 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) { 2653 if (Current->ParameterIndex == 1) 2654 Current->SplitPenalty += 5 * Current->BindingStrength; 2655 } else { 2656 Current->SplitPenalty += 20 * Current->BindingStrength; 2657 } 2658 2659 Current = Current->Next; 2660 } 2661 2662 calculateUnbreakableTailLengths(Line); 2663 unsigned IndentLevel = Line.Level; 2664 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2665 if (Current->Role) 2666 Current->Role->precomputeFormattingInfos(Current); 2667 if (Current->MatchingParen && 2668 Current->MatchingParen->opensBlockOrBlockTypeList(Style) && 2669 IndentLevel > 0) { 2670 --IndentLevel; 2671 } 2672 Current->IndentLevel = IndentLevel; 2673 if (Current->opensBlockOrBlockTypeList(Style)) 2674 ++IndentLevel; 2675 } 2676 2677 LLVM_DEBUG({ printDebugInfo(Line); }); 2678 } 2679 2680 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2681 unsigned UnbreakableTailLength = 0; 2682 FormatToken *Current = Line.Last; 2683 while (Current) { 2684 Current->UnbreakableTailLength = UnbreakableTailLength; 2685 if (Current->CanBreakBefore || 2686 Current->isOneOf(tok::comment, tok::string_literal)) { 2687 UnbreakableTailLength = 0; 2688 } else { 2689 UnbreakableTailLength += 2690 Current->ColumnWidth + Current->SpacesRequiredBefore; 2691 } 2692 Current = Current->Previous; 2693 } 2694 } 2695 2696 void TokenAnnotator::calculateArrayInitializerColumnList(AnnotatedLine &Line) { 2697 if (Line.First == Line.Last) { 2698 return; 2699 } 2700 auto *CurrentToken = Line.First; 2701 CurrentToken->ArrayInitializerLineStart = true; 2702 unsigned Depth = 0; 2703 while (CurrentToken != nullptr && CurrentToken != Line.Last) { 2704 if (CurrentToken->is(tok::l_brace)) { 2705 CurrentToken->IsArrayInitializer = true; 2706 if (CurrentToken->Next != nullptr) 2707 CurrentToken->Next->MustBreakBefore = true; 2708 CurrentToken = 2709 calculateInitializerColumnList(Line, CurrentToken->Next, Depth + 1); 2710 } else { 2711 CurrentToken = CurrentToken->Next; 2712 } 2713 } 2714 } 2715 2716 FormatToken *TokenAnnotator::calculateInitializerColumnList( 2717 AnnotatedLine &Line, FormatToken *CurrentToken, unsigned Depth) { 2718 while (CurrentToken != nullptr && CurrentToken != Line.Last) { 2719 if (CurrentToken->is(tok::l_brace)) 2720 ++Depth; 2721 else if (CurrentToken->is(tok::r_brace)) 2722 --Depth; 2723 if (Depth == 2 && CurrentToken->isOneOf(tok::l_brace, tok::comma)) { 2724 CurrentToken = CurrentToken->Next; 2725 if (CurrentToken == nullptr) 2726 break; 2727 CurrentToken->StartsColumn = true; 2728 CurrentToken = CurrentToken->Previous; 2729 } 2730 CurrentToken = CurrentToken->Next; 2731 } 2732 return CurrentToken; 2733 } 2734 2735 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2736 const FormatToken &Tok, 2737 bool InFunctionDecl) { 2738 const FormatToken &Left = *Tok.Previous; 2739 const FormatToken &Right = Tok; 2740 2741 if (Left.is(tok::semi)) 2742 return 0; 2743 2744 if (Style.Language == FormatStyle::LK_Java) { 2745 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2746 return 1; 2747 if (Right.is(Keywords.kw_implements)) 2748 return 2; 2749 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2750 return 3; 2751 } else if (Style.isJavaScript()) { 2752 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2753 return 100; 2754 if (Left.is(TT_JsTypeColon)) 2755 return 35; 2756 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2757 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2758 return 100; 2759 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2760 if (Left.opensScope() && Right.closesScope()) 2761 return 200; 2762 } 2763 2764 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2765 return 1; 2766 if (Right.is(tok::l_square)) { 2767 if (Style.Language == FormatStyle::LK_Proto) 2768 return 1; 2769 if (Left.is(tok::r_square)) 2770 return 200; 2771 // Slightly prefer formatting local lambda definitions like functions. 2772 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2773 return 35; 2774 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2775 TT_ArrayInitializerLSquare, 2776 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) 2777 return 500; 2778 } 2779 2780 if (Left.is(tok::coloncolon) || 2781 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2782 return 500; 2783 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2784 Right.is(tok::kw_operator)) { 2785 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2786 return 3; 2787 if (Left.is(TT_StartOfName)) 2788 return 110; 2789 if (InFunctionDecl && Right.NestingLevel == 0) 2790 return Style.PenaltyReturnTypeOnItsOwnLine; 2791 return 200; 2792 } 2793 if (Right.is(TT_PointerOrReference)) 2794 return 190; 2795 if (Right.is(TT_LambdaArrow)) 2796 return 110; 2797 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2798 return 160; 2799 if (Left.is(TT_CastRParen)) 2800 return 100; 2801 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2802 return 5000; 2803 if (Left.is(tok::comment)) 2804 return 1000; 2805 2806 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2807 TT_CtorInitializerColon)) 2808 return 2; 2809 2810 if (Right.isMemberAccess()) { 2811 // Breaking before the "./->" of a chained call/member access is reasonably 2812 // cheap, as formatting those with one call per line is generally 2813 // desirable. In particular, it should be cheaper to break before the call 2814 // than it is to break inside a call's parameters, which could lead to weird 2815 // "hanging" indents. The exception is the very last "./->" to support this 2816 // frequent pattern: 2817 // 2818 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2819 // dddddddd); 2820 // 2821 // which might otherwise be blown up onto many lines. Here, clang-format 2822 // won't produce "hanging" indents anyway as there is no other trailing 2823 // call. 2824 // 2825 // Also apply higher penalty is not a call as that might lead to a wrapping 2826 // like: 2827 // 2828 // aaaaaaa 2829 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2830 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2831 ? 150 2832 : 35; 2833 } 2834 2835 if (Right.is(TT_TrailingAnnotation) && 2836 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2837 // Moving trailing annotations to the next line is fine for ObjC method 2838 // declarations. 2839 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2840 return 10; 2841 // Generally, breaking before a trailing annotation is bad unless it is 2842 // function-like. It seems to be especially preferable to keep standard 2843 // annotations (i.e. "const", "final" and "override") on the same line. 2844 // Use a slightly higher penalty after ")" so that annotations like 2845 // "const override" are kept together. 2846 bool is_short_annotation = Right.TokenText.size() < 10; 2847 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 2848 } 2849 2850 // In for-loops, prefer breaking at ',' and ';'. 2851 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 2852 return 4; 2853 2854 // In Objective-C method expressions, prefer breaking before "param:" over 2855 // breaking after it. 2856 if (Right.is(TT_SelectorName)) 2857 return 0; 2858 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 2859 return Line.MightBeFunctionDecl ? 50 : 500; 2860 2861 // In Objective-C type declarations, avoid breaking after the category's 2862 // open paren (we'll prefer breaking after the protocol list's opening 2863 // angle bracket, if present). 2864 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous && 2865 Left.Previous->isOneOf(tok::identifier, tok::greater)) 2866 return 500; 2867 2868 if (Left.is(tok::l_paren) && Style.PenaltyBreakOpenParenthesis != 0) 2869 return Style.PenaltyBreakOpenParenthesis; 2870 if (Left.is(tok::l_paren) && InFunctionDecl && 2871 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 2872 return 100; 2873 if (Left.is(tok::l_paren) && Left.Previous && 2874 (Left.Previous->is(tok::kw_for) || Left.Previous->isIf())) 2875 return 1000; 2876 if (Left.is(tok::equal) && InFunctionDecl) 2877 return 110; 2878 if (Right.is(tok::r_brace)) 2879 return 1; 2880 if (Left.is(TT_TemplateOpener)) 2881 return 100; 2882 if (Left.opensScope()) { 2883 // If we aren't aligning after opening parens/braces we can always break 2884 // here unless the style does not want us to place all arguments on the 2885 // next line. 2886 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign && 2887 (Left.ParameterCount <= 1 || Style.AllowAllArgumentsOnNextLine)) 2888 return 0; 2889 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle) 2890 return 19; 2891 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 2892 : 19; 2893 } 2894 if (Left.is(TT_JavaAnnotation)) 2895 return 50; 2896 2897 if (Left.is(TT_UnaryOperator)) 2898 return 60; 2899 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 2900 Left.Previous->isLabelString() && 2901 (Left.NextOperator || Left.OperatorIndex != 0)) 2902 return 50; 2903 if (Right.is(tok::plus) && Left.isLabelString() && 2904 (Right.NextOperator || Right.OperatorIndex != 0)) 2905 return 25; 2906 if (Left.is(tok::comma)) 2907 return 1; 2908 if (Right.is(tok::lessless) && Left.isLabelString() && 2909 (Right.NextOperator || Right.OperatorIndex != 1)) 2910 return 25; 2911 if (Right.is(tok::lessless)) { 2912 // Breaking at a << is really cheap. 2913 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 2914 // Slightly prefer to break before the first one in log-like statements. 2915 return 2; 2916 return 1; 2917 } 2918 if (Left.ClosesTemplateDeclaration) 2919 return Style.PenaltyBreakTemplateDeclaration; 2920 if (Left.is(TT_ConditionalExpr)) 2921 return prec::Conditional; 2922 prec::Level Level = Left.getPrecedence(); 2923 if (Level == prec::Unknown) 2924 Level = Right.getPrecedence(); 2925 if (Level == prec::Assignment) 2926 return Style.PenaltyBreakAssignment; 2927 if (Level != prec::Unknown) 2928 return Level; 2929 2930 return 3; 2931 } 2932 2933 bool TokenAnnotator::spaceRequiredBeforeParens(const FormatToken &Right) const { 2934 if (Style.SpaceBeforeParens == FormatStyle::SBPO_Always) 2935 return true; 2936 if (Right.is(TT_OverloadedOperatorLParen) && 2937 Style.SpaceBeforeParensOptions.AfterOverloadedOperator) 2938 return true; 2939 if (Style.SpaceBeforeParensOptions.BeforeNonEmptyParentheses && 2940 Right.ParameterCount > 0) 2941 return true; 2942 return false; 2943 } 2944 2945 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 2946 const FormatToken &Left, 2947 const FormatToken &Right) { 2948 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 2949 return true; 2950 if (Style.isJson() && Left.is(tok::string_literal) && Right.is(tok::colon)) 2951 return false; 2952 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 2953 return true; 2954 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 2955 Left.Tok.getObjCKeywordID() == tok::objc_property) 2956 return true; 2957 if (Right.is(tok::hashhash)) 2958 return Left.is(tok::hash); 2959 if (Left.isOneOf(tok::hashhash, tok::hash)) 2960 return Right.is(tok::hash); 2961 if ((Left.is(tok::l_paren) && Right.is(tok::r_paren)) || 2962 (Left.is(tok::l_brace) && Left.isNot(BK_Block) && 2963 Right.is(tok::r_brace) && Right.isNot(BK_Block))) 2964 return Style.SpaceInEmptyParentheses; 2965 if (Style.SpacesInConditionalStatement) { 2966 if (Left.is(tok::l_paren) && Left.Previous && 2967 isKeywordWithCondition(*Left.Previous)) 2968 return true; 2969 if (Right.is(tok::r_paren) && Right.MatchingParen && 2970 Right.MatchingParen->Previous && 2971 isKeywordWithCondition(*Right.MatchingParen->Previous)) 2972 return true; 2973 } 2974 2975 // auto{x} auto(x) 2976 if (Left.is(tok::kw_auto) && Right.isOneOf(tok::l_paren, tok::l_brace)) 2977 return false; 2978 2979 // operator co_await(x) 2980 if (Right.is(tok::l_paren) && Left.is(tok::kw_co_await) && Left.Previous && 2981 Left.Previous->is(tok::kw_operator)) 2982 return false; 2983 // co_await (x), co_yield (x), co_return (x) 2984 if (Left.isOneOf(tok::kw_co_await, tok::kw_co_yield, tok::kw_co_return) && 2985 Right.isNot(tok::semi)) 2986 return true; 2987 // requires clause Concept1<T> && Concept2<T> 2988 if (Left.is(TT_ConstraintJunctions) && Right.is(tok::identifier)) 2989 return true; 2990 2991 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 2992 return (Right.is(TT_CastRParen) || 2993 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 2994 ? Style.SpacesInCStyleCastParentheses 2995 : Style.SpacesInParentheses; 2996 if (Right.isOneOf(tok::semi, tok::comma)) 2997 return false; 2998 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) { 2999 bool IsLightweightGeneric = Right.MatchingParen && 3000 Right.MatchingParen->Next && 3001 Right.MatchingParen->Next->is(tok::colon); 3002 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList; 3003 } 3004 if (Right.is(tok::less) && Left.is(tok::kw_template)) 3005 return Style.SpaceAfterTemplateKeyword; 3006 if (Left.isOneOf(tok::exclaim, tok::tilde)) 3007 return false; 3008 if (Left.is(tok::at) && 3009 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 3010 tok::numeric_constant, tok::l_paren, tok::l_brace, 3011 tok::kw_true, tok::kw_false)) 3012 return false; 3013 if (Left.is(tok::colon)) 3014 return !Left.is(TT_ObjCMethodExpr); 3015 if (Left.is(tok::coloncolon)) 3016 return false; 3017 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 3018 if (Style.Language == FormatStyle::LK_TextProto || 3019 (Style.Language == FormatStyle::LK_Proto && 3020 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 3021 // Format empty list as `<>`. 3022 if (Left.is(tok::less) && Right.is(tok::greater)) 3023 return false; 3024 return !Style.Cpp11BracedListStyle; 3025 } 3026 return false; 3027 } 3028 if (Right.is(tok::ellipsis)) 3029 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 3030 Left.Previous->is(tok::kw_case)); 3031 if (Left.is(tok::l_square) && Right.is(tok::amp)) 3032 return Style.SpacesInSquareBrackets; 3033 if (Right.is(TT_PointerOrReference)) { 3034 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 3035 if (!Left.MatchingParen) 3036 return true; 3037 FormatToken *TokenBeforeMatchingParen = 3038 Left.MatchingParen->getPreviousNonComment(); 3039 if (!TokenBeforeMatchingParen || !Left.is(TT_TypeDeclarationParen)) 3040 return true; 3041 } 3042 // Add a space if the previous token is a pointer qualifier or the closing 3043 // parenthesis of __attribute__(()) expression and the style requires spaces 3044 // after pointer qualifiers. 3045 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_After || 3046 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 3047 (Left.is(TT_AttributeParen) || Left.canBePointerOrReferenceQualifier())) 3048 return true; 3049 if (Left.Tok.isLiteral()) 3050 return true; 3051 // for (auto a = 0, b = 0; const auto & c : {1, 2, 3}) 3052 if (Left.isTypeOrIdentifier() && Right.Next && Right.Next->Next && 3053 Right.Next->Next->is(TT_RangeBasedForLoopColon)) 3054 return getTokenPointerOrReferenceAlignment(Right) != 3055 FormatStyle::PAS_Left; 3056 return ( 3057 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 3058 (getTokenPointerOrReferenceAlignment(Right) != FormatStyle::PAS_Left || 3059 (Line.IsMultiVariableDeclStmt && 3060 (Left.NestingLevel == 0 || 3061 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for))))))); 3062 } 3063 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 3064 (!Left.is(TT_PointerOrReference) || 3065 (getTokenPointerOrReferenceAlignment(Left) != FormatStyle::PAS_Right && 3066 !Line.IsMultiVariableDeclStmt))) 3067 return true; 3068 if (Left.is(TT_PointerOrReference)) { 3069 // Add a space if the next token is a pointer qualifier and the style 3070 // requires spaces before pointer qualifiers. 3071 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Before || 3072 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 3073 Right.canBePointerOrReferenceQualifier()) 3074 return true; 3075 // & 1 3076 if (Right.Tok.isLiteral()) 3077 return true; 3078 // & /* comment 3079 if (Right.is(TT_BlockComment)) 3080 return true; 3081 // foo() -> const Bar * override/final 3082 if (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 3083 !Right.is(TT_StartOfName)) 3084 return true; 3085 // & { 3086 if (Right.is(tok::l_brace) && Right.is(BK_Block)) 3087 return true; 3088 // for (auto a = 0, b = 0; const auto& c : {1, 2, 3}) 3089 if (Left.Previous && Left.Previous->isTypeOrIdentifier() && Right.Next && 3090 Right.Next->is(TT_RangeBasedForLoopColon)) 3091 return getTokenPointerOrReferenceAlignment(Left) != 3092 FormatStyle::PAS_Right; 3093 return !Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 3094 tok::l_paren) && 3095 (getTokenPointerOrReferenceAlignment(Left) != 3096 FormatStyle::PAS_Right && 3097 !Line.IsMultiVariableDeclStmt) && 3098 Left.Previous && 3099 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon, 3100 tok::l_square); 3101 } 3102 // Ensure right pointer alignment with ellipsis e.g. int *...P 3103 if (Left.is(tok::ellipsis) && Left.Previous && 3104 Left.Previous->isOneOf(tok::star, tok::amp, tok::ampamp)) 3105 return Style.PointerAlignment != FormatStyle::PAS_Right; 3106 3107 if (Right.is(tok::star) && Left.is(tok::l_paren)) 3108 return false; 3109 if (Left.is(tok::star) && Right.isOneOf(tok::star, tok::amp, tok::ampamp)) 3110 return false; 3111 if (Right.isOneOf(tok::star, tok::amp, tok::ampamp)) { 3112 const FormatToken *Previous = &Left; 3113 while (Previous && !Previous->is(tok::kw_operator)) { 3114 if (Previous->is(tok::identifier) || Previous->isSimpleTypeSpecifier()) { 3115 Previous = Previous->getPreviousNonComment(); 3116 continue; 3117 } 3118 if (Previous->is(TT_TemplateCloser) && Previous->MatchingParen) { 3119 Previous = Previous->MatchingParen->getPreviousNonComment(); 3120 continue; 3121 } 3122 if (Previous->is(tok::coloncolon)) { 3123 Previous = Previous->getPreviousNonComment(); 3124 continue; 3125 } 3126 break; 3127 } 3128 // Space between the type and the * in: 3129 // operator void*() 3130 // operator char*() 3131 // operator void const*() 3132 // operator void volatile*() 3133 // operator /*comment*/ const char*() 3134 // operator volatile /*comment*/ char*() 3135 // operator Foo*() 3136 // operator C<T>*() 3137 // operator std::Foo*() 3138 // operator C<T>::D<U>*() 3139 // dependent on PointerAlignment style. 3140 if (Previous) { 3141 if (Previous->endsSequence(tok::kw_operator)) 3142 return (Style.PointerAlignment != FormatStyle::PAS_Left); 3143 if (Previous->is(tok::kw_const) || Previous->is(tok::kw_volatile)) 3144 return (Style.PointerAlignment != FormatStyle::PAS_Left) || 3145 (Style.SpaceAroundPointerQualifiers == 3146 FormatStyle::SAPQ_After) || 3147 (Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both); 3148 } 3149 } 3150 const auto SpaceRequiredForArrayInitializerLSquare = 3151 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 3152 return Style.SpacesInContainerLiterals || 3153 ((Style.Language == FormatStyle::LK_Proto || 3154 Style.Language == FormatStyle::LK_TextProto) && 3155 !Style.Cpp11BracedListStyle && 3156 LSquareTok.endsSequence(tok::l_square, tok::colon, 3157 TT_SelectorName)); 3158 }; 3159 if (Left.is(tok::l_square)) 3160 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 3161 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 3162 (Left.isOneOf(TT_ArraySubscriptLSquare, TT_StructuredBindingLSquare, 3163 TT_LambdaLSquare) && 3164 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 3165 if (Right.is(tok::r_square)) 3166 return Right.MatchingParen && 3167 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 3168 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 3169 Style)) || 3170 (Style.SpacesInSquareBrackets && 3171 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 3172 TT_StructuredBindingLSquare, 3173 TT_LambdaLSquare)) || 3174 Right.MatchingParen->is(TT_AttributeParen)); 3175 if (Right.is(tok::l_square) && 3176 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 3177 TT_DesignatedInitializerLSquare, 3178 TT_StructuredBindingLSquare, TT_AttributeSquare) && 3179 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral) && 3180 !(!Left.is(tok::r_square) && Style.SpaceBeforeSquareBrackets && 3181 Right.is(TT_ArraySubscriptLSquare))) 3182 return false; 3183 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 3184 return !Left.Children.empty(); // No spaces in "{}". 3185 if ((Left.is(tok::l_brace) && Left.isNot(BK_Block)) || 3186 (Right.is(tok::r_brace) && Right.MatchingParen && 3187 Right.MatchingParen->isNot(BK_Block))) 3188 return Style.Cpp11BracedListStyle ? Style.SpacesInParentheses : true; 3189 if (Left.is(TT_BlockComment)) 3190 // No whitespace in x(/*foo=*/1), except for JavaScript. 3191 return Style.isJavaScript() || !Left.TokenText.endswith("=*/"); 3192 3193 // Space between template and attribute. 3194 // e.g. template <typename T> [[nodiscard]] ... 3195 if (Left.is(TT_TemplateCloser) && Right.is(TT_AttributeSquare)) 3196 return true; 3197 // Space before parentheses common for all languages 3198 if (Right.is(tok::l_paren)) { 3199 if (Left.is(TT_TemplateCloser) && Right.isNot(TT_FunctionTypeLParen)) 3200 return spaceRequiredBeforeParens(Right); 3201 if (Left.is(tok::kw_requires)) 3202 return spaceRequiredBeforeParens(Right); 3203 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 3204 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) 3205 return true; 3206 if (Left.is(TT_ForEachMacro)) 3207 return (Style.SpaceBeforeParensOptions.AfterForeachMacros || 3208 spaceRequiredBeforeParens(Right)); 3209 if (Left.is(TT_IfMacro)) 3210 return (Style.SpaceBeforeParensOptions.AfterIfMacros || 3211 spaceRequiredBeforeParens(Right)); 3212 if (Line.Type == LT_ObjCDecl) 3213 return true; 3214 if (Left.is(tok::semi)) 3215 return true; 3216 if (Left.isOneOf(tok::pp_elif, tok::kw_for, tok::kw_while, tok::kw_switch, 3217 tok::kw_case, TT_ForEachMacro, TT_ObjCForIn)) 3218 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3219 spaceRequiredBeforeParens(Right); 3220 if (Left.isIf(Line.Type != LT_PreprocessorDirective)) 3221 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3222 spaceRequiredBeforeParens(Right); 3223 3224 // TODO add Operator overloading specific Options to 3225 // SpaceBeforeParensOptions 3226 if (Right.is(TT_OverloadedOperatorLParen)) 3227 return spaceRequiredBeforeParens(Right); 3228 // Function declaration or definition 3229 if (Line.MightBeFunctionDecl && (Left.is(TT_FunctionDeclarationName))) { 3230 if (Line.mightBeFunctionDefinition()) 3231 return Style.SpaceBeforeParensOptions.AfterFunctionDefinitionName || 3232 spaceRequiredBeforeParens(Right); 3233 else 3234 return Style.SpaceBeforeParensOptions.AfterFunctionDeclarationName || 3235 spaceRequiredBeforeParens(Right); 3236 } 3237 // Lambda 3238 if (Line.Type != LT_PreprocessorDirective && Left.is(tok::r_square) && 3239 Left.MatchingParen && Left.MatchingParen->is(TT_LambdaLSquare)) 3240 return Style.SpaceBeforeParensOptions.AfterFunctionDefinitionName || 3241 spaceRequiredBeforeParens(Right); 3242 if (!Left.Previous || Left.Previous->isNot(tok::period)) { 3243 if (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch)) 3244 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3245 spaceRequiredBeforeParens(Right); 3246 if (Left.isOneOf(tok::kw_new, tok::kw_delete)) 3247 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never || 3248 spaceRequiredBeforeParens(Right); 3249 } 3250 if (Line.Type != LT_PreprocessorDirective && 3251 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 3252 Left.is(tok::r_paren) || Left.isSimpleTypeSpecifier())) 3253 return spaceRequiredBeforeParens(Right); 3254 return false; 3255 } 3256 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 3257 return false; 3258 if (Right.is(TT_UnaryOperator)) 3259 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 3260 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 3261 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 3262 tok::r_paren) || 3263 Left.isSimpleTypeSpecifier()) && 3264 Right.is(tok::l_brace) && Right.getNextNonComment() && 3265 Right.isNot(BK_Block)) 3266 return false; 3267 if (Left.is(tok::period) || Right.is(tok::period)) 3268 return false; 3269 // u#str, U#str, L#str, u8#str 3270 // uR#str, UR#str, LR#str, u8R#str 3271 if (Right.is(tok::hash) && Left.is(tok::identifier) && 3272 (Left.TokenText == "L" || Left.TokenText == "u" || 3273 Left.TokenText == "U" || Left.TokenText == "u8" || 3274 Left.TokenText == "LR" || Left.TokenText == "uR" || 3275 Left.TokenText == "UR" || Left.TokenText == "u8R")) 3276 return false; 3277 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 3278 Left.MatchingParen->Previous && 3279 (Left.MatchingParen->Previous->is(tok::period) || 3280 Left.MatchingParen->Previous->is(tok::coloncolon))) 3281 // Java call to generic function with explicit type: 3282 // A.<B<C<...>>>DoSomething(); 3283 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference. 3284 return false; 3285 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 3286 return false; 3287 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at)) 3288 // Objective-C dictionary literal -> no space after opening brace. 3289 return false; 3290 if (Right.is(tok::r_brace) && Right.MatchingParen && 3291 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at)) 3292 // Objective-C dictionary literal -> no space before closing brace. 3293 return false; 3294 if (Right.getType() == TT_TrailingAnnotation && 3295 Right.isOneOf(tok::amp, tok::ampamp) && 3296 Left.isOneOf(tok::kw_const, tok::kw_volatile) && 3297 (!Right.Next || Right.Next->is(tok::semi))) 3298 // Match const and volatile ref-qualifiers without any additional 3299 // qualifiers such as 3300 // void Fn() const &; 3301 return getTokenReferenceAlignment(Right) != FormatStyle::PAS_Left; 3302 3303 return true; 3304 } 3305 3306 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 3307 const FormatToken &Right) { 3308 const FormatToken &Left = *Right.Previous; 3309 3310 // If the token is finalized don't touch it (as it could be in a 3311 // clang-format-off section). 3312 if (Left.Finalized) 3313 return Right.hasWhitespaceBefore(); 3314 3315 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 3316 return true; // Never ever merge two identifiers. 3317 3318 // Leave a space between * and /* to avoid C4138 `comment end` found outside 3319 // of comment. 3320 if (Left.is(tok::star) && Right.is(tok::comment)) 3321 return true; 3322 3323 if (Style.isCpp()) { 3324 // Space between import <iostream>. 3325 // or import .....; 3326 if (Left.is(Keywords.kw_import) && Right.isOneOf(tok::less, tok::ellipsis)) 3327 return true; 3328 // Space between `module :` and `import :`. 3329 if (Left.isOneOf(Keywords.kw_module, Keywords.kw_import) && 3330 Right.is(TT_ModulePartitionColon)) 3331 return true; 3332 // No space between import foo:bar but keep a space between import :bar; 3333 if (Left.is(tok::identifier) && Right.is(TT_ModulePartitionColon)) 3334 return false; 3335 // No space between :bar; 3336 if (Left.is(TT_ModulePartitionColon) && 3337 Right.isOneOf(tok::identifier, tok::kw_private)) 3338 return false; 3339 if (Left.is(tok::ellipsis) && Right.is(tok::identifier) && 3340 Line.First->is(Keywords.kw_import)) 3341 return false; 3342 // Space in __attribute__((attr)) ::type. 3343 if (Left.is(TT_AttributeParen) && Right.is(tok::coloncolon)) 3344 return true; 3345 3346 if (Left.is(tok::kw_operator)) 3347 return Right.is(tok::coloncolon); 3348 if (Right.is(tok::l_brace) && Right.is(BK_BracedInit) && 3349 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList) 3350 return true; 3351 if (Left.is(tok::less) && Left.is(TT_OverloadedOperator) && 3352 Right.is(TT_TemplateOpener)) 3353 return true; 3354 } else if (Style.Language == FormatStyle::LK_Proto || 3355 Style.Language == FormatStyle::LK_TextProto) { 3356 if (Right.is(tok::period) && 3357 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 3358 Keywords.kw_repeated, Keywords.kw_extend)) 3359 return true; 3360 if (Right.is(tok::l_paren) && 3361 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 3362 return true; 3363 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 3364 return true; 3365 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 3366 if (Left.is(tok::slash) || Right.is(tok::slash)) 3367 return false; 3368 if (Left.MatchingParen && 3369 Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 3370 Right.isOneOf(tok::l_brace, tok::less)) 3371 return !Style.Cpp11BracedListStyle; 3372 // A percent is probably part of a formatting specification, such as %lld. 3373 if (Left.is(tok::percent)) 3374 return false; 3375 // Preserve the existence of a space before a percent for cases like 0x%04x 3376 // and "%d %d" 3377 if (Left.is(tok::numeric_constant) && Right.is(tok::percent)) 3378 return Right.hasWhitespaceBefore(); 3379 } else if (Style.isJson()) { 3380 if (Right.is(tok::colon)) 3381 return false; 3382 } else if (Style.isCSharp()) { 3383 // Require spaces around '{' and before '}' unless they appear in 3384 // interpolated strings. Interpolated strings are merged into a single token 3385 // so cannot have spaces inserted by this function. 3386 3387 // No space between 'this' and '[' 3388 if (Left.is(tok::kw_this) && Right.is(tok::l_square)) 3389 return false; 3390 3391 // No space between 'new' and '(' 3392 if (Left.is(tok::kw_new) && Right.is(tok::l_paren)) 3393 return false; 3394 3395 // Space before { (including space within '{ {'). 3396 if (Right.is(tok::l_brace)) 3397 return true; 3398 3399 // Spaces inside braces. 3400 if (Left.is(tok::l_brace) && Right.isNot(tok::r_brace)) 3401 return true; 3402 3403 if (Left.isNot(tok::l_brace) && Right.is(tok::r_brace)) 3404 return true; 3405 3406 // Spaces around '=>'. 3407 if (Left.is(TT_FatArrow) || Right.is(TT_FatArrow)) 3408 return true; 3409 3410 // No spaces around attribute target colons 3411 if (Left.is(TT_AttributeColon) || Right.is(TT_AttributeColon)) 3412 return false; 3413 3414 // space between type and variable e.g. Dictionary<string,string> foo; 3415 if (Left.is(TT_TemplateCloser) && Right.is(TT_StartOfName)) 3416 return true; 3417 3418 // spaces inside square brackets. 3419 if (Left.is(tok::l_square) || Right.is(tok::r_square)) 3420 return Style.SpacesInSquareBrackets; 3421 3422 // No space before ? in nullable types. 3423 if (Right.is(TT_CSharpNullable)) 3424 return false; 3425 3426 // No space before null forgiving '!'. 3427 if (Right.is(TT_NonNullAssertion)) 3428 return false; 3429 3430 // No space between consecutive commas '[,,]'. 3431 if (Left.is(tok::comma) && Right.is(tok::comma)) 3432 return false; 3433 3434 // space after var in `var (key, value)` 3435 if (Left.is(Keywords.kw_var) && Right.is(tok::l_paren)) 3436 return true; 3437 3438 // space between keywords and paren e.g. "using (" 3439 if (Right.is(tok::l_paren)) 3440 if (Left.isOneOf(tok::kw_using, Keywords.kw_async, Keywords.kw_when, 3441 Keywords.kw_lock)) 3442 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3443 spaceRequiredBeforeParens(Right); 3444 3445 // space between method modifier and opening parenthesis of a tuple return 3446 // type 3447 if (Left.isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 3448 tok::kw_virtual, tok::kw_extern, tok::kw_static, 3449 Keywords.kw_internal, Keywords.kw_abstract, 3450 Keywords.kw_sealed, Keywords.kw_override, 3451 Keywords.kw_async, Keywords.kw_unsafe) && 3452 Right.is(tok::l_paren)) 3453 return true; 3454 } else if (Style.isJavaScript()) { 3455 if (Left.is(TT_FatArrow)) 3456 return true; 3457 // for await ( ... 3458 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 3459 Left.Previous->is(tok::kw_for)) 3460 return true; 3461 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 3462 Right.MatchingParen) { 3463 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 3464 // An async arrow function, for example: `x = async () => foo();`, 3465 // as opposed to calling a function called async: `x = async();` 3466 if (Next && Next->is(TT_FatArrow)) 3467 return true; 3468 } 3469 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 3470 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 3471 return false; 3472 // In tagged template literals ("html`bar baz`"), there is no space between 3473 // the tag identifier and the template string. 3474 if (Keywords.IsJavaScriptIdentifier(Left, 3475 /* AcceptIdentifierName= */ false) && 3476 Right.is(TT_TemplateString)) 3477 return false; 3478 if (Right.is(tok::star) && 3479 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 3480 return false; 3481 if (Right.isOneOf(tok::l_brace, tok::l_square) && 3482 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 3483 Keywords.kw_extends, Keywords.kw_implements)) 3484 return true; 3485 if (Right.is(tok::l_paren)) { 3486 // JS methods can use some keywords as names (e.g. `delete()`). 3487 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 3488 return false; 3489 // Valid JS method names can include keywords, e.g. `foo.delete()` or 3490 // `bar.instanceof()`. Recognize call positions by preceding period. 3491 if (Left.Previous && Left.Previous->is(tok::period) && 3492 Left.Tok.getIdentifierInfo()) 3493 return false; 3494 // Additional unary JavaScript operators that need a space after. 3495 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 3496 tok::kw_void)) 3497 return true; 3498 } 3499 // `foo as const;` casts into a const type. 3500 if (Left.endsSequence(tok::kw_const, Keywords.kw_as)) { 3501 return false; 3502 } 3503 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 3504 tok::kw_const) || 3505 // "of" is only a keyword if it appears after another identifier 3506 // (e.g. as "const x of y" in a for loop), or after a destructuring 3507 // operation (const [x, y] of z, const {a, b} of c). 3508 (Left.is(Keywords.kw_of) && Left.Previous && 3509 (Left.Previous->Tok.is(tok::identifier) || 3510 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 3511 (!Left.Previous || !Left.Previous->is(tok::period))) 3512 return true; 3513 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 3514 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 3515 return false; 3516 if (Left.is(Keywords.kw_as) && 3517 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 3518 return true; 3519 if (Left.is(tok::kw_default) && Left.Previous && 3520 Left.Previous->is(tok::kw_export)) 3521 return true; 3522 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 3523 return true; 3524 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 3525 return false; 3526 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 3527 return false; 3528 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 3529 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 3530 return false; 3531 if (Left.is(tok::ellipsis)) 3532 return false; 3533 if (Left.is(TT_TemplateCloser) && 3534 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 3535 Keywords.kw_implements, Keywords.kw_extends)) 3536 // Type assertions ('<type>expr') are not followed by whitespace. Other 3537 // locations that should have whitespace following are identified by the 3538 // above set of follower tokens. 3539 return false; 3540 if (Right.is(TT_NonNullAssertion)) 3541 return false; 3542 if (Left.is(TT_NonNullAssertion) && 3543 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 3544 return true; // "x! as string", "x! in y" 3545 } else if (Style.Language == FormatStyle::LK_Java) { 3546 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 3547 return true; 3548 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 3549 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3550 spaceRequiredBeforeParens(Right); 3551 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 3552 tok::kw_protected) || 3553 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 3554 Keywords.kw_native)) && 3555 Right.is(TT_TemplateOpener)) 3556 return true; 3557 } 3558 if (Left.is(TT_ImplicitStringLiteral)) 3559 return Right.hasWhitespaceBefore(); 3560 if (Line.Type == LT_ObjCMethodDecl) { 3561 if (Left.is(TT_ObjCMethodSpecifier)) 3562 return true; 3563 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right)) 3564 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a 3565 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class 3566 // method declaration. 3567 return false; 3568 } 3569 if (Line.Type == LT_ObjCProperty && 3570 (Right.is(tok::equal) || Left.is(tok::equal))) 3571 return false; 3572 3573 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 3574 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 3575 return true; 3576 if (Left.is(tok::comma) && !Right.is(TT_OverloadedOperatorLParen)) 3577 return true; 3578 if (Right.is(tok::comma)) 3579 return false; 3580 if (Right.is(TT_ObjCBlockLParen)) 3581 return true; 3582 if (Right.is(TT_CtorInitializerColon)) 3583 return Style.SpaceBeforeCtorInitializerColon; 3584 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 3585 return false; 3586 if (Right.is(TT_RangeBasedForLoopColon) && 3587 !Style.SpaceBeforeRangeBasedForLoopColon) 3588 return false; 3589 if (Left.is(TT_BitFieldColon)) 3590 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 3591 Style.BitFieldColonSpacing == FormatStyle::BFCS_After; 3592 if (Right.is(tok::colon)) { 3593 if (Line.First->isOneOf(tok::kw_default, tok::kw_case)) 3594 return Style.SpaceBeforeCaseColon; 3595 if (!Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 3596 return false; 3597 if (Right.is(TT_ObjCMethodExpr)) 3598 return false; 3599 if (Left.is(tok::question)) 3600 return false; 3601 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 3602 return false; 3603 if (Right.is(TT_DictLiteral)) 3604 return Style.SpacesInContainerLiterals; 3605 if (Right.is(TT_AttributeColon)) 3606 return false; 3607 if (Right.is(TT_CSharpNamedArgumentColon)) 3608 return false; 3609 if (Right.is(TT_BitFieldColon)) 3610 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 3611 Style.BitFieldColonSpacing == FormatStyle::BFCS_Before; 3612 return true; 3613 } 3614 // Do not merge "- -" into "--". 3615 if ((Left.isOneOf(tok::minus, tok::minusminus) && 3616 Right.isOneOf(tok::minus, tok::minusminus)) || 3617 (Left.isOneOf(tok::plus, tok::plusplus) && 3618 Right.isOneOf(tok::plus, tok::plusplus))) 3619 return true; 3620 if (Left.is(TT_UnaryOperator)) { 3621 if (!Right.is(tok::l_paren)) { 3622 // The alternative operators for ~ and ! are "compl" and "not". 3623 // If they are used instead, we do not want to combine them with 3624 // the token to the right, unless that is a left paren. 3625 if (Left.is(tok::exclaim) && Left.TokenText == "not") 3626 return true; 3627 if (Left.is(tok::tilde) && Left.TokenText == "compl") 3628 return true; 3629 // Lambda captures allow for a lone &, so "&]" needs to be properly 3630 // handled. 3631 if (Left.is(tok::amp) && Right.is(tok::r_square)) 3632 return Style.SpacesInSquareBrackets; 3633 } 3634 return (Style.SpaceAfterLogicalNot && Left.is(tok::exclaim)) || 3635 Right.is(TT_BinaryOperator); 3636 } 3637 3638 // If the next token is a binary operator or a selector name, we have 3639 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 3640 if (Left.is(TT_CastRParen)) 3641 return Style.SpaceAfterCStyleCast || 3642 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 3643 3644 auto ShouldAddSpacesInAngles = [this, &Right]() { 3645 if (this->Style.SpacesInAngles == FormatStyle::SIAS_Always) 3646 return true; 3647 if (this->Style.SpacesInAngles == FormatStyle::SIAS_Leave) 3648 return Right.hasWhitespaceBefore(); 3649 return false; 3650 }; 3651 3652 if (Left.is(tok::greater) && Right.is(tok::greater)) { 3653 if (Style.Language == FormatStyle::LK_TextProto || 3654 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 3655 return !Style.Cpp11BracedListStyle; 3656 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 3657 ((Style.Standard < FormatStyle::LS_Cpp11) || 3658 ShouldAddSpacesInAngles()); 3659 } 3660 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 3661 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 3662 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 3663 return false; 3664 if (!Style.SpaceBeforeAssignmentOperators && Left.isNot(TT_TemplateCloser) && 3665 Right.getPrecedence() == prec::Assignment) 3666 return false; 3667 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) && 3668 (Left.is(tok::identifier) || Left.is(tok::kw_this))) 3669 return false; 3670 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 3671 // Generally don't remove existing spaces between an identifier and "::". 3672 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 3673 // this turns out to be too lenient, add analysis of the identifier itself. 3674 return Right.hasWhitespaceBefore(); 3675 if (Right.is(tok::coloncolon) && 3676 !Left.isOneOf(tok::l_brace, tok::comment, tok::l_paren)) 3677 // Put a space between < and :: in vector< ::std::string > 3678 return (Left.is(TT_TemplateOpener) && 3679 ((Style.Standard < FormatStyle::LS_Cpp11) || 3680 ShouldAddSpacesInAngles())) || 3681 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 3682 tok::kw___super, TT_TemplateOpener, 3683 TT_TemplateCloser)) || 3684 (Left.is(tok::l_paren) && Style.SpacesInParentheses); 3685 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 3686 return ShouldAddSpacesInAngles(); 3687 // Space before TT_StructuredBindingLSquare. 3688 if (Right.is(TT_StructuredBindingLSquare)) 3689 return !Left.isOneOf(tok::amp, tok::ampamp) || 3690 getTokenReferenceAlignment(Left) != FormatStyle::PAS_Right; 3691 // Space before & or && following a TT_StructuredBindingLSquare. 3692 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 3693 Right.isOneOf(tok::amp, tok::ampamp)) 3694 return getTokenReferenceAlignment(Right) != FormatStyle::PAS_Left; 3695 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 3696 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 3697 !Right.is(tok::r_paren))) 3698 return true; 3699 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 3700 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 3701 return false; 3702 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 3703 Line.startsWith(tok::hash)) 3704 return true; 3705 if (Right.is(TT_TrailingUnaryOperator)) 3706 return false; 3707 if (Left.is(TT_RegexLiteral)) 3708 return false; 3709 return spaceRequiredBetween(Line, Left, Right); 3710 } 3711 3712 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 3713 static bool isAllmanBrace(const FormatToken &Tok) { 3714 return Tok.is(tok::l_brace) && Tok.is(BK_Block) && 3715 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_LambdaLBrace, TT_DictLiteral); 3716 } 3717 3718 // Returns 'true' if 'Tok' is a function argument. 3719 static bool IsFunctionArgument(const FormatToken &Tok) { 3720 return Tok.MatchingParen && Tok.MatchingParen->Next && 3721 Tok.MatchingParen->Next->isOneOf(tok::comma, tok::r_paren); 3722 } 3723 3724 static bool 3725 isItAnEmptyLambdaAllowed(const FormatToken &Tok, 3726 FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3727 return Tok.Children.empty() && ShortLambdaOption != FormatStyle::SLS_None; 3728 } 3729 3730 static bool isAllmanLambdaBrace(const FormatToken &Tok) { 3731 return (Tok.is(tok::l_brace) && Tok.is(BK_Block) && 3732 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral)); 3733 } 3734 3735 // Returns the first token on the line that is not a comment. 3736 static const FormatToken *getFirstNonComment(const AnnotatedLine &Line) { 3737 const FormatToken *Next = Line.First; 3738 if (!Next) 3739 return Next; 3740 if (Next->is(tok::comment)) 3741 Next = Next->getNextNonComment(); 3742 return Next; 3743 } 3744 3745 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 3746 const FormatToken &Right) { 3747 const FormatToken &Left = *Right.Previous; 3748 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 3749 return true; 3750 3751 if (Style.isCSharp()) { 3752 if (Left.is(TT_FatArrow) && Right.is(tok::l_brace) && 3753 Style.BraceWrapping.AfterFunction) 3754 return true; 3755 if (Right.is(TT_CSharpNamedArgumentColon) || 3756 Left.is(TT_CSharpNamedArgumentColon)) 3757 return false; 3758 if (Right.is(TT_CSharpGenericTypeConstraint)) 3759 return true; 3760 if (Right.Next && Right.Next->is(TT_FatArrow) && 3761 (Right.is(tok::numeric_constant) || 3762 (Right.is(tok::identifier) && Right.TokenText == "_"))) 3763 return true; 3764 3765 // Break after C# [...] and before public/protected/private/internal. 3766 if (Left.is(TT_AttributeSquare) && Left.is(tok::r_square) && 3767 (Right.isAccessSpecifier(/*ColonRequired=*/false) || 3768 Right.is(Keywords.kw_internal))) 3769 return true; 3770 // Break between ] and [ but only when there are really 2 attributes. 3771 if (Left.is(TT_AttributeSquare) && Right.is(TT_AttributeSquare) && 3772 Left.is(tok::r_square) && Right.is(tok::l_square)) 3773 return true; 3774 3775 } else if (Style.isJavaScript()) { 3776 // FIXME: This might apply to other languages and token kinds. 3777 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 3778 Left.Previous->is(tok::string_literal)) 3779 return true; 3780 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 3781 Left.Previous && Left.Previous->is(tok::equal) && 3782 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 3783 tok::kw_const) && 3784 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 3785 // above. 3786 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 3787 // Object literals on the top level of a file are treated as "enum-style". 3788 // Each key/value pair is put on a separate line, instead of bin-packing. 3789 return true; 3790 if (Left.is(tok::l_brace) && Line.Level == 0 && 3791 (Line.startsWith(tok::kw_enum) || 3792 Line.startsWith(tok::kw_const, tok::kw_enum) || 3793 Line.startsWith(tok::kw_export, tok::kw_enum) || 3794 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 3795 // JavaScript top-level enum key/value pairs are put on separate lines 3796 // instead of bin-packing. 3797 return true; 3798 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && Left.Previous && 3799 Left.Previous->is(TT_FatArrow)) { 3800 // JS arrow function (=> {...}). 3801 switch (Style.AllowShortLambdasOnASingleLine) { 3802 case FormatStyle::SLS_All: 3803 return false; 3804 case FormatStyle::SLS_None: 3805 return true; 3806 case FormatStyle::SLS_Empty: 3807 return !Left.Children.empty(); 3808 case FormatStyle::SLS_Inline: 3809 // allow one-lining inline (e.g. in function call args) and empty arrow 3810 // functions. 3811 return (Left.NestingLevel == 0 && Line.Level == 0) && 3812 !Left.Children.empty(); 3813 } 3814 llvm_unreachable("Unknown FormatStyle::ShortLambdaStyle enum"); 3815 } 3816 3817 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 3818 !Left.Children.empty()) 3819 // Support AllowShortFunctionsOnASingleLine for JavaScript. 3820 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 3821 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 3822 (Left.NestingLevel == 0 && Line.Level == 0 && 3823 Style.AllowShortFunctionsOnASingleLine & 3824 FormatStyle::SFS_InlineOnly); 3825 } else if (Style.Language == FormatStyle::LK_Java) { 3826 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 3827 Right.Next->is(tok::string_literal)) 3828 return true; 3829 } else if (Style.Language == FormatStyle::LK_Cpp || 3830 Style.Language == FormatStyle::LK_ObjC || 3831 Style.Language == FormatStyle::LK_Proto || 3832 Style.Language == FormatStyle::LK_TableGen || 3833 Style.Language == FormatStyle::LK_TextProto) { 3834 if (Left.isStringLiteral() && Right.isStringLiteral()) 3835 return true; 3836 } 3837 3838 // Basic JSON newline processing. 3839 if (Style.isJson()) { 3840 // Always break after a JSON record opener. 3841 // { 3842 // } 3843 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace)) 3844 return true; 3845 // Always break after a JSON array opener. 3846 // [ 3847 // ] 3848 if (Left.is(TT_ArrayInitializerLSquare) && Left.is(tok::l_square) && 3849 !Right.is(tok::r_square)) 3850 return true; 3851 // Always break after successive entries. 3852 // 1, 3853 // 2 3854 if (Left.is(tok::comma)) 3855 return true; 3856 } 3857 3858 // If the last token before a '}', ']', or ')' is a comma or a trailing 3859 // comment, the intention is to insert a line break after it in order to make 3860 // shuffling around entries easier. Import statements, especially in 3861 // JavaScript, can be an exception to this rule. 3862 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 3863 const FormatToken *BeforeClosingBrace = nullptr; 3864 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 3865 (Style.isJavaScript() && Left.is(tok::l_paren))) && 3866 Left.isNot(BK_Block) && Left.MatchingParen) 3867 BeforeClosingBrace = Left.MatchingParen->Previous; 3868 else if (Right.MatchingParen && 3869 (Right.MatchingParen->isOneOf(tok::l_brace, 3870 TT_ArrayInitializerLSquare) || 3871 (Style.isJavaScript() && Right.MatchingParen->is(tok::l_paren)))) 3872 BeforeClosingBrace = &Left; 3873 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 3874 BeforeClosingBrace->isTrailingComment())) 3875 return true; 3876 } 3877 3878 if (Right.is(tok::comment)) 3879 return Left.isNot(BK_BracedInit) && Left.isNot(TT_CtorInitializerColon) && 3880 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 3881 if (Left.isTrailingComment()) 3882 return true; 3883 if (Left.IsUnterminatedLiteral) 3884 return true; 3885 if (Right.is(tok::lessless) && Right.Next && Left.is(tok::string_literal) && 3886 Right.Next->is(tok::string_literal)) 3887 return true; 3888 // Can break after template<> declaration 3889 if (Left.ClosesTemplateDeclaration && Left.MatchingParen && 3890 Left.MatchingParen->NestingLevel == 0) { 3891 // Put concepts on the next line e.g. 3892 // template<typename T> 3893 // concept ... 3894 if (Right.is(tok::kw_concept)) 3895 return Style.BreakBeforeConceptDeclarations; 3896 return (Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes); 3897 } 3898 if (Style.PackConstructorInitializers == FormatStyle::PCIS_Never) { 3899 if (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeColon && 3900 (Left.is(TT_CtorInitializerComma) || Right.is(TT_CtorInitializerColon))) 3901 return true; 3902 3903 if (Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon && 3904 Left.isOneOf(TT_CtorInitializerColon, TT_CtorInitializerComma)) 3905 return true; 3906 } 3907 if (Style.PackConstructorInitializers < FormatStyle::PCIS_CurrentLine && 3908 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3909 Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) 3910 return true; 3911 // Break only if we have multiple inheritance. 3912 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 3913 Right.is(TT_InheritanceComma)) 3914 return true; 3915 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterComma && 3916 Left.is(TT_InheritanceComma)) 3917 return true; 3918 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 3919 // Multiline raw string literals are special wrt. line breaks. The author 3920 // has made a deliberate choice and might have aligned the contents of the 3921 // string literal accordingly. Thus, we try keep existing line breaks. 3922 return Right.IsMultiline && Right.NewlinesBefore > 0; 3923 if ((Left.is(tok::l_brace) || (Left.is(tok::less) && Left.Previous && 3924 Left.Previous->is(tok::equal))) && 3925 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 3926 // Don't put enums or option definitions onto single lines in protocol 3927 // buffers. 3928 return true; 3929 } 3930 if (Right.is(TT_InlineASMBrace)) 3931 return Right.HasUnescapedNewline; 3932 3933 if (isAllmanBrace(Left) || isAllmanBrace(Right)) { 3934 auto FirstNonComment = getFirstNonComment(Line); 3935 bool AccessSpecifier = 3936 FirstNonComment && 3937 FirstNonComment->isOneOf(Keywords.kw_internal, tok::kw_public, 3938 tok::kw_private, tok::kw_protected); 3939 3940 if (Style.BraceWrapping.AfterEnum) { 3941 if (Line.startsWith(tok::kw_enum) || 3942 Line.startsWith(tok::kw_typedef, tok::kw_enum)) 3943 return true; 3944 // Ensure BraceWrapping for `public enum A {`. 3945 if (AccessSpecifier && FirstNonComment->Next && 3946 FirstNonComment->Next->is(tok::kw_enum)) 3947 return true; 3948 } 3949 3950 // Ensure BraceWrapping for `public interface A {`. 3951 if (Style.BraceWrapping.AfterClass && 3952 ((AccessSpecifier && FirstNonComment->Next && 3953 FirstNonComment->Next->is(Keywords.kw_interface)) || 3954 Line.startsWith(Keywords.kw_interface))) 3955 return true; 3956 3957 return (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 3958 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 3959 } 3960 3961 if (Left.is(TT_ObjCBlockLBrace) && 3962 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 3963 return true; 3964 3965 // Ensure wrapping after __attribute__((XX)) and @interface etc. 3966 if (Left.is(TT_AttributeParen) && Right.is(TT_ObjCDecl)) 3967 return true; 3968 3969 if (Left.is(TT_LambdaLBrace)) { 3970 if (IsFunctionArgument(Left) && 3971 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline) 3972 return false; 3973 3974 if (Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_None || 3975 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline || 3976 (!Left.Children.empty() && 3977 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Empty)) 3978 return true; 3979 } 3980 3981 if (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace) && 3982 Left.isOneOf(tok::star, tok::amp, tok::ampamp, TT_TemplateCloser)) { 3983 return true; 3984 } 3985 3986 // Put multiple Java annotation on a new line. 3987 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 3988 Left.is(TT_LeadingJavaAnnotation) && 3989 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 3990 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 3991 return true; 3992 3993 if (Right.is(TT_ProtoExtensionLSquare)) 3994 return true; 3995 3996 // In text proto instances if a submessage contains at least 2 entries and at 3997 // least one of them is a submessage, like A { ... B { ... } ... }, 3998 // put all of the entries of A on separate lines by forcing the selector of 3999 // the submessage B to be put on a newline. 4000 // 4001 // Example: these can stay on one line: 4002 // a { scalar_1: 1 scalar_2: 2 } 4003 // a { b { key: value } } 4004 // 4005 // and these entries need to be on a new line even if putting them all in one 4006 // line is under the column limit: 4007 // a { 4008 // scalar: 1 4009 // b { key: value } 4010 // } 4011 // 4012 // We enforce this by breaking before a submessage field that has previous 4013 // siblings, *and* breaking before a field that follows a submessage field. 4014 // 4015 // Be careful to exclude the case [proto.ext] { ... } since the `]` is 4016 // the TT_SelectorName there, but we don't want to break inside the brackets. 4017 // 4018 // Another edge case is @submessage { key: value }, which is a common 4019 // substitution placeholder. In this case we want to keep `@` and `submessage` 4020 // together. 4021 // 4022 // We ensure elsewhere that extensions are always on their own line. 4023 if ((Style.Language == FormatStyle::LK_Proto || 4024 Style.Language == FormatStyle::LK_TextProto) && 4025 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) { 4026 // Keep `@submessage` together in: 4027 // @submessage { key: value } 4028 if (Left.is(tok::at)) 4029 return false; 4030 // Look for the scope opener after selector in cases like: 4031 // selector { ... 4032 // selector: { ... 4033 // selector: @base { ... 4034 FormatToken *LBrace = Right.Next; 4035 if (LBrace && LBrace->is(tok::colon)) { 4036 LBrace = LBrace->Next; 4037 if (LBrace && LBrace->is(tok::at)) { 4038 LBrace = LBrace->Next; 4039 if (LBrace) 4040 LBrace = LBrace->Next; 4041 } 4042 } 4043 if (LBrace && 4044 // The scope opener is one of {, [, <: 4045 // selector { ... } 4046 // selector [ ... ] 4047 // selector < ... > 4048 // 4049 // In case of selector { ... }, the l_brace is TT_DictLiteral. 4050 // In case of an empty selector {}, the l_brace is not TT_DictLiteral, 4051 // so we check for immediately following r_brace. 4052 ((LBrace->is(tok::l_brace) && 4053 (LBrace->is(TT_DictLiteral) || 4054 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) || 4055 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) { 4056 // If Left.ParameterCount is 0, then this submessage entry is not the 4057 // first in its parent submessage, and we want to break before this entry. 4058 // If Left.ParameterCount is greater than 0, then its parent submessage 4059 // might contain 1 or more entries and we want to break before this entry 4060 // if it contains at least 2 entries. We deal with this case later by 4061 // detecting and breaking before the next entry in the parent submessage. 4062 if (Left.ParameterCount == 0) 4063 return true; 4064 // However, if this submessage is the first entry in its parent 4065 // submessage, Left.ParameterCount might be 1 in some cases. 4066 // We deal with this case later by detecting an entry 4067 // following a closing paren of this submessage. 4068 } 4069 4070 // If this is an entry immediately following a submessage, it will be 4071 // preceded by a closing paren of that submessage, like in: 4072 // left---. .---right 4073 // v v 4074 // sub: { ... } key: value 4075 // If there was a comment between `}` an `key` above, then `key` would be 4076 // put on a new line anyways. 4077 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square)) 4078 return true; 4079 } 4080 4081 // Deal with lambda arguments in C++ - we want consistent line breaks whether 4082 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced 4083 // as aggressive line breaks are placed when the lambda is not the last arg. 4084 if ((Style.Language == FormatStyle::LK_Cpp || 4085 Style.Language == FormatStyle::LK_ObjC) && 4086 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 && 4087 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) { 4088 // Multiple lambdas in the same function call force line breaks. 4089 if (Left.BlockParameterCount > 1) 4090 return true; 4091 4092 // A lambda followed by another arg forces a line break. 4093 if (!Left.Role) 4094 return false; 4095 auto Comma = Left.Role->lastComma(); 4096 if (!Comma) 4097 return false; 4098 auto Next = Comma->getNextNonComment(); 4099 if (!Next) 4100 return false; 4101 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret)) 4102 return true; 4103 } 4104 4105 return false; 4106 } 4107 4108 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 4109 const FormatToken &Right) { 4110 const FormatToken &Left = *Right.Previous; 4111 // Language-specific stuff. 4112 if (Style.isCSharp()) { 4113 if (Left.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon) || 4114 Right.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon)) 4115 return false; 4116 // Only break after commas for generic type constraints. 4117 if (Line.First->is(TT_CSharpGenericTypeConstraint)) 4118 return Left.is(TT_CSharpGenericTypeConstraintComma); 4119 // Keep nullable operators attached to their identifiers. 4120 if (Right.is(TT_CSharpNullable)) { 4121 return false; 4122 } 4123 } else if (Style.Language == FormatStyle::LK_Java) { 4124 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 4125 Keywords.kw_implements)) 4126 return false; 4127 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 4128 Keywords.kw_implements)) 4129 return true; 4130 } else if (Style.isJavaScript()) { 4131 const FormatToken *NonComment = Right.getPreviousNonComment(); 4132 if (NonComment && 4133 NonComment->isOneOf( 4134 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 4135 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 4136 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 4137 Keywords.kw_readonly, Keywords.kw_override, Keywords.kw_abstract, 4138 Keywords.kw_get, Keywords.kw_set, Keywords.kw_async, 4139 Keywords.kw_await)) 4140 return false; // Otherwise automatic semicolon insertion would trigger. 4141 if (Right.NestingLevel == 0 && 4142 (Left.Tok.getIdentifierInfo() || 4143 Left.isOneOf(tok::r_square, tok::r_paren)) && 4144 Right.isOneOf(tok::l_square, tok::l_paren)) 4145 return false; // Otherwise automatic semicolon insertion would trigger. 4146 if (NonComment && NonComment->is(tok::identifier) && 4147 NonComment->TokenText == "asserts") 4148 return false; 4149 if (Left.is(TT_FatArrow) && Right.is(tok::l_brace)) 4150 return false; 4151 if (Left.is(TT_JsTypeColon)) 4152 return true; 4153 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;"). 4154 if (Left.is(tok::exclaim) && Right.is(tok::colon)) 4155 return false; 4156 // Look for is type annotations like: 4157 // function f(): a is B { ... } 4158 // Do not break before is in these cases. 4159 if (Right.is(Keywords.kw_is)) { 4160 const FormatToken *Next = Right.getNextNonComment(); 4161 // If `is` is followed by a colon, it's likely that it's a dict key, so 4162 // ignore it for this check. 4163 // For example this is common in Polymer: 4164 // Polymer({ 4165 // is: 'name', 4166 // ... 4167 // }); 4168 if (!Next || !Next->is(tok::colon)) 4169 return false; 4170 } 4171 if (Left.is(Keywords.kw_in)) 4172 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 4173 if (Right.is(Keywords.kw_in)) 4174 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 4175 if (Right.is(Keywords.kw_as)) 4176 return false; // must not break before as in 'x as type' casts 4177 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) { 4178 // extends and infer can appear as keywords in conditional types: 4179 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types 4180 // do not break before them, as the expressions are subject to ASI. 4181 return false; 4182 } 4183 if (Left.is(Keywords.kw_as)) 4184 return true; 4185 if (Left.is(TT_NonNullAssertion)) 4186 return true; 4187 if (Left.is(Keywords.kw_declare) && 4188 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 4189 Keywords.kw_function, tok::kw_class, tok::kw_enum, 4190 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 4191 Keywords.kw_let, tok::kw_const)) 4192 // See grammar for 'declare' statements at: 4193 // https://github.com/Microsoft/TypeScript/blob/main/doc/spec-ARCHIVED.md#A.10 4194 return false; 4195 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 4196 Right.isOneOf(tok::identifier, tok::string_literal)) 4197 return false; // must not break in "module foo { ...}" 4198 if (Right.is(TT_TemplateString) && Right.closesScope()) 4199 return false; 4200 // Don't split tagged template literal so there is a break between the tag 4201 // identifier and template string. 4202 if (Left.is(tok::identifier) && Right.is(TT_TemplateString)) { 4203 return false; 4204 } 4205 if (Left.is(TT_TemplateString) && Left.opensScope()) 4206 return true; 4207 } 4208 4209 if (Left.is(tok::at)) 4210 return false; 4211 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 4212 return false; 4213 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 4214 return !Right.is(tok::l_paren); 4215 if (Right.is(TT_PointerOrReference)) 4216 return Line.IsMultiVariableDeclStmt || 4217 (getTokenPointerOrReferenceAlignment(Right) == 4218 FormatStyle::PAS_Right && 4219 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 4220 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 4221 Right.is(tok::kw_operator)) 4222 return true; 4223 if (Left.is(TT_PointerOrReference)) 4224 return false; 4225 if (Right.isTrailingComment()) 4226 // We rely on MustBreakBefore being set correctly here as we should not 4227 // change the "binding" behavior of a comment. 4228 // The first comment in a braced lists is always interpreted as belonging to 4229 // the first list element. Otherwise, it should be placed outside of the 4230 // list. 4231 return Left.is(BK_BracedInit) || 4232 (Left.is(TT_CtorInitializerColon) && 4233 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 4234 if (Left.is(tok::question) && Right.is(tok::colon)) 4235 return false; 4236 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 4237 return Style.BreakBeforeTernaryOperators; 4238 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 4239 return !Style.BreakBeforeTernaryOperators; 4240 if (Left.is(TT_InheritanceColon)) 4241 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon; 4242 if (Right.is(TT_InheritanceColon)) 4243 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon; 4244 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 4245 Left.isNot(TT_SelectorName)) 4246 return true; 4247 4248 if (Right.is(tok::colon) && 4249 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 4250 return false; 4251 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 4252 if (Style.Language == FormatStyle::LK_Proto || 4253 Style.Language == FormatStyle::LK_TextProto) { 4254 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 4255 return false; 4256 // Prevent cases like: 4257 // 4258 // submessage: 4259 // { key: valueeeeeeeeeeee } 4260 // 4261 // when the snippet does not fit into one line. 4262 // Prefer: 4263 // 4264 // submessage: { 4265 // key: valueeeeeeeeeeee 4266 // } 4267 // 4268 // instead, even if it is longer by one line. 4269 // 4270 // Note that this allows allows the "{" to go over the column limit 4271 // when the column limit is just between ":" and "{", but that does 4272 // not happen too often and alternative formattings in this case are 4273 // not much better. 4274 // 4275 // The code covers the cases: 4276 // 4277 // submessage: { ... } 4278 // submessage: < ... > 4279 // repeated: [ ... ] 4280 if (((Right.is(tok::l_brace) || Right.is(tok::less)) && 4281 Right.is(TT_DictLiteral)) || 4282 Right.is(TT_ArrayInitializerLSquare)) 4283 return false; 4284 } 4285 return true; 4286 } 4287 if (Right.is(tok::r_square) && Right.MatchingParen && 4288 Right.MatchingParen->is(TT_ProtoExtensionLSquare)) 4289 return false; 4290 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 4291 Right.Next->is(TT_ObjCMethodExpr))) 4292 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 4293 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 4294 return true; 4295 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 4296 return true; 4297 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 4298 TT_OverloadedOperator)) 4299 return false; 4300 if (Left.is(TT_RangeBasedForLoopColon)) 4301 return true; 4302 if (Right.is(TT_RangeBasedForLoopColon)) 4303 return false; 4304 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 4305 return true; 4306 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 4307 Left.is(tok::kw_operator)) 4308 return false; 4309 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 4310 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 4311 return false; 4312 if (Left.is(tok::equal) && Right.is(tok::l_brace) && 4313 !Style.Cpp11BracedListStyle) 4314 return false; 4315 if (Left.is(tok::l_paren) && 4316 Left.isOneOf(TT_AttributeParen, TT_TypeDeclarationParen)) 4317 return false; 4318 if (Left.is(tok::l_paren) && Left.Previous && 4319 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 4320 return false; 4321 if (Right.is(TT_ImplicitStringLiteral)) 4322 return false; 4323 4324 if (Right.is(TT_TemplateCloser)) 4325 return false; 4326 if (Right.is(tok::r_square) && Right.MatchingParen && 4327 Right.MatchingParen->is(TT_LambdaLSquare)) 4328 return false; 4329 4330 // We only break before r_brace if there was a corresponding break before 4331 // the l_brace, which is tracked by BreakBeforeClosingBrace. 4332 if (Right.is(tok::r_brace)) 4333 return Right.MatchingParen && Right.MatchingParen->is(BK_Block); 4334 4335 // We only break before r_paren if we're in a block indented context. 4336 if (Right.is(tok::r_paren)) { 4337 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent) { 4338 return Right.MatchingParen && 4339 !(Right.MatchingParen->Previous && 4340 (Right.MatchingParen->Previous->is(tok::kw_for) || 4341 Right.MatchingParen->Previous->isIf())); 4342 } 4343 4344 return false; 4345 } 4346 4347 // Allow breaking after a trailing annotation, e.g. after a method 4348 // declaration. 4349 if (Left.is(TT_TrailingAnnotation)) 4350 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 4351 tok::less, tok::coloncolon); 4352 4353 if (Right.is(tok::kw___attribute) || 4354 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) 4355 return !Left.is(TT_AttributeSquare); 4356 4357 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 4358 return true; 4359 4360 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 4361 return true; 4362 4363 if (Left.is(TT_CtorInitializerColon)) 4364 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 4365 if (Right.is(TT_CtorInitializerColon)) 4366 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 4367 if (Left.is(TT_CtorInitializerComma) && 4368 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 4369 return false; 4370 if (Right.is(TT_CtorInitializerComma) && 4371 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 4372 return true; 4373 if (Left.is(TT_InheritanceComma) && 4374 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 4375 return false; 4376 if (Right.is(TT_InheritanceComma) && 4377 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 4378 return true; 4379 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 4380 (Left.is(tok::less) && Right.is(tok::less))) 4381 return false; 4382 if (Right.is(TT_BinaryOperator) && 4383 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 4384 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 4385 Right.getPrecedence() != prec::Assignment)) 4386 return true; 4387 if (Left.is(TT_ArrayInitializerLSquare)) 4388 return true; 4389 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 4390 return true; 4391 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 4392 !Left.isOneOf(tok::arrowstar, tok::lessless) && 4393 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 4394 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 4395 Left.getPrecedence() == prec::Assignment)) 4396 return true; 4397 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 4398 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) 4399 return false; 4400 4401 auto ShortLambdaOption = Style.AllowShortLambdasOnASingleLine; 4402 if (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace)) { 4403 if (isAllmanLambdaBrace(Left)) 4404 return !isItAnEmptyLambdaAllowed(Left, ShortLambdaOption); 4405 if (isAllmanLambdaBrace(Right)) 4406 return !isItAnEmptyLambdaAllowed(Right, ShortLambdaOption); 4407 } 4408 4409 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 4410 tok::kw_class, tok::kw_struct, tok::comment) || 4411 Right.isMemberAccess() || 4412 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 4413 tok::colon, tok::l_square, tok::at) || 4414 (Left.is(tok::r_paren) && 4415 Right.isOneOf(tok::identifier, tok::kw_const)) || 4416 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 4417 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 4418 } 4419 4420 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 4421 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 4422 const FormatToken *Tok = Line.First; 4423 while (Tok) { 4424 llvm::errs() << " M=" << Tok->MustBreakBefore 4425 << " C=" << Tok->CanBreakBefore 4426 << " T=" << getTokenTypeName(Tok->getType()) 4427 << " S=" << Tok->SpacesRequiredBefore 4428 << " F=" << Tok->Finalized << " B=" << Tok->BlockParameterCount 4429 << " BK=" << Tok->getBlockKind() << " P=" << Tok->SplitPenalty 4430 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 4431 << " PPK=" << Tok->getPackingKind() << " FakeLParens="; 4432 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 4433 llvm::errs() << Tok->FakeLParens[i] << "/"; 4434 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 4435 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo(); 4436 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 4437 if (!Tok->Next) 4438 assert(Tok == Line.Last); 4439 Tok = Tok->Next; 4440 } 4441 llvm::errs() << "----\n"; 4442 } 4443 4444 FormatStyle::PointerAlignmentStyle 4445 TokenAnnotator::getTokenReferenceAlignment(const FormatToken &Reference) { 4446 assert(Reference.isOneOf(tok::amp, tok::ampamp)); 4447 switch (Style.ReferenceAlignment) { 4448 case FormatStyle::RAS_Pointer: 4449 return Style.PointerAlignment; 4450 case FormatStyle::RAS_Left: 4451 return FormatStyle::PAS_Left; 4452 case FormatStyle::RAS_Right: 4453 return FormatStyle::PAS_Right; 4454 case FormatStyle::RAS_Middle: 4455 return FormatStyle::PAS_Middle; 4456 } 4457 assert(0); //"Unhandled value of ReferenceAlignment" 4458 return Style.PointerAlignment; 4459 } 4460 4461 FormatStyle::PointerAlignmentStyle 4462 TokenAnnotator::getTokenPointerOrReferenceAlignment( 4463 const FormatToken &PointerOrReference) { 4464 if (PointerOrReference.isOneOf(tok::amp, tok::ampamp)) { 4465 switch (Style.ReferenceAlignment) { 4466 case FormatStyle::RAS_Pointer: 4467 return Style.PointerAlignment; 4468 case FormatStyle::RAS_Left: 4469 return FormatStyle::PAS_Left; 4470 case FormatStyle::RAS_Right: 4471 return FormatStyle::PAS_Right; 4472 case FormatStyle::RAS_Middle: 4473 return FormatStyle::PAS_Middle; 4474 } 4475 } 4476 assert(PointerOrReference.is(tok::star)); 4477 return Style.PointerAlignment; 4478 } 4479 4480 } // namespace format 4481 } // namespace clang 4482