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 TT_RecordLBrace)) 1428 CurrentToken->setType(TT_Unknown); 1429 CurrentToken->Role.reset(); 1430 CurrentToken->MatchingParen = nullptr; 1431 CurrentToken->FakeLParens.clear(); 1432 CurrentToken->FakeRParens = 0; 1433 } 1434 1435 void next() { 1436 if (!CurrentToken) 1437 return; 1438 1439 CurrentToken->NestingLevel = Contexts.size() - 1; 1440 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1441 modifyContext(*CurrentToken); 1442 determineTokenType(*CurrentToken); 1443 CurrentToken = CurrentToken->Next; 1444 1445 resetTokenMetadata(); 1446 } 1447 1448 /// A struct to hold information valid in a specific context, e.g. 1449 /// a pair of parenthesis. 1450 struct Context { 1451 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1452 bool IsExpression) 1453 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1454 IsExpression(IsExpression) {} 1455 1456 tok::TokenKind ContextKind; 1457 unsigned BindingStrength; 1458 bool IsExpression; 1459 unsigned LongestObjCSelectorName = 0; 1460 bool ColonIsForRangeExpr = false; 1461 bool ColonIsDictLiteral = false; 1462 bool ColonIsObjCMethodExpr = false; 1463 FormatToken *FirstObjCSelectorName = nullptr; 1464 FormatToken *FirstStartOfName = nullptr; 1465 bool CanBeExpression = true; 1466 bool InTemplateArgument = false; 1467 bool InCtorInitializer = false; 1468 bool InInheritanceList = false; 1469 bool CaretFound = false; 1470 bool IsForEachMacro = false; 1471 bool InCpp11AttributeSpecifier = false; 1472 bool InCSharpAttributeSpecifier = false; 1473 bool InStructArrayInitializer = false; 1474 }; 1475 1476 /// Puts a new \c Context onto the stack \c Contexts for the lifetime 1477 /// of each instance. 1478 struct ScopedContextCreator { 1479 AnnotatingParser &P; 1480 1481 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1482 unsigned Increase) 1483 : P(P) { 1484 P.Contexts.push_back(Context(ContextKind, 1485 P.Contexts.back().BindingStrength + Increase, 1486 P.Contexts.back().IsExpression)); 1487 } 1488 1489 ~ScopedContextCreator() { 1490 if (P.Style.AlignArrayOfStructures != FormatStyle::AIAS_None) { 1491 if (P.Contexts.back().InStructArrayInitializer) { 1492 P.Contexts.pop_back(); 1493 P.Contexts.back().InStructArrayInitializer = true; 1494 return; 1495 } 1496 } 1497 P.Contexts.pop_back(); 1498 } 1499 }; 1500 1501 void modifyContext(const FormatToken &Current) { 1502 if (Current.getPrecedence() == prec::Assignment && 1503 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 1504 // Type aliases use `type X = ...;` in TypeScript and can be exported 1505 // using `export type ...`. 1506 !(Style.isJavaScript() && 1507 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1508 Line.startsWith(tok::kw_export, Keywords.kw_type, 1509 tok::identifier))) && 1510 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 1511 Contexts.back().IsExpression = true; 1512 if (!Line.startsWith(TT_UnaryOperator)) { 1513 for (FormatToken *Previous = Current.Previous; 1514 Previous && Previous->Previous && 1515 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1516 Previous = Previous->Previous) { 1517 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1518 Previous = Previous->MatchingParen; 1519 if (!Previous) 1520 break; 1521 } 1522 if (Previous->opensScope()) 1523 break; 1524 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1525 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1526 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1527 Previous->setType(TT_PointerOrReference); 1528 } 1529 } 1530 } else if (Current.is(tok::lessless) && 1531 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1532 Contexts.back().IsExpression = true; 1533 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1534 Contexts.back().IsExpression = true; 1535 } else if (Current.is(TT_TrailingReturnArrow)) { 1536 Contexts.back().IsExpression = false; 1537 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1538 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1539 } else if (Current.Previous && 1540 Current.Previous->is(TT_CtorInitializerColon)) { 1541 Contexts.back().IsExpression = true; 1542 Contexts.back().InCtorInitializer = true; 1543 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1544 Contexts.back().InInheritanceList = true; 1545 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1546 for (FormatToken *Previous = Current.Previous; 1547 Previous && Previous->isOneOf(tok::star, tok::amp); 1548 Previous = Previous->Previous) 1549 Previous->setType(TT_PointerOrReference); 1550 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 1551 Contexts.back().IsExpression = false; 1552 } else if (Current.is(tok::kw_new)) { 1553 Contexts.back().CanBeExpression = false; 1554 } else if (Current.is(tok::semi) || 1555 (Current.is(tok::exclaim) && Current.Previous && 1556 !Current.Previous->is(tok::kw_operator))) { 1557 // This should be the condition or increment in a for-loop. 1558 // But not operator !() (can't use TT_OverloadedOperator here as its not 1559 // been annotated yet). 1560 Contexts.back().IsExpression = true; 1561 } 1562 } 1563 1564 static FormatToken *untilMatchingParen(FormatToken *Current) { 1565 // Used when `MatchingParen` is not yet established. 1566 int ParenLevel = 0; 1567 while (Current) { 1568 if (Current->is(tok::l_paren)) 1569 ++ParenLevel; 1570 if (Current->is(tok::r_paren)) 1571 --ParenLevel; 1572 if (ParenLevel < 1) 1573 break; 1574 Current = Current->Next; 1575 } 1576 return Current; 1577 } 1578 1579 static bool isDeductionGuide(FormatToken &Current) { 1580 // Look for a deduction guide template<T> A(...) -> A<...>; 1581 if (Current.Previous && Current.Previous->is(tok::r_paren) && 1582 Current.startsSequence(tok::arrow, tok::identifier, tok::less)) { 1583 // Find the TemplateCloser. 1584 FormatToken *TemplateCloser = Current.Next->Next; 1585 int NestingLevel = 0; 1586 while (TemplateCloser) { 1587 // Skip over an expressions in parens A<(3 < 2)>; 1588 if (TemplateCloser->is(tok::l_paren)) { 1589 // No Matching Paren yet so skip to matching paren 1590 TemplateCloser = untilMatchingParen(TemplateCloser); 1591 if (!TemplateCloser) 1592 break; 1593 } 1594 if (TemplateCloser->is(tok::less)) 1595 ++NestingLevel; 1596 if (TemplateCloser->is(tok::greater)) 1597 --NestingLevel; 1598 if (NestingLevel < 1) 1599 break; 1600 TemplateCloser = TemplateCloser->Next; 1601 } 1602 // Assuming we have found the end of the template ensure its followed 1603 // with a semi-colon. 1604 if (TemplateCloser && TemplateCloser->Next && 1605 TemplateCloser->Next->is(tok::semi) && 1606 Current.Previous->MatchingParen) { 1607 // Determine if the identifier `A` prior to the A<..>; is the same as 1608 // prior to the A(..) 1609 FormatToken *LeadingIdentifier = 1610 Current.Previous->MatchingParen->Previous; 1611 1612 // Differentiate a deduction guide by seeing the 1613 // > of the template prior to the leading identifier. 1614 if (LeadingIdentifier) { 1615 FormatToken *PriorLeadingIdentifier = LeadingIdentifier->Previous; 1616 // Skip back past explicit decoration 1617 if (PriorLeadingIdentifier && 1618 PriorLeadingIdentifier->is(tok::kw_explicit)) 1619 PriorLeadingIdentifier = PriorLeadingIdentifier->Previous; 1620 1621 return (PriorLeadingIdentifier && 1622 PriorLeadingIdentifier->is(TT_TemplateCloser) && 1623 LeadingIdentifier->TokenText == Current.Next->TokenText); 1624 } 1625 } 1626 } 1627 return false; 1628 } 1629 1630 void determineTokenType(FormatToken &Current) { 1631 if (!Current.is(TT_Unknown)) 1632 // The token type is already known. 1633 return; 1634 1635 if ((Style.isJavaScript() || Style.isCSharp()) && 1636 Current.is(tok::exclaim)) { 1637 if (Current.Previous) { 1638 bool IsIdentifier = 1639 Style.isJavaScript() 1640 ? Keywords.IsJavaScriptIdentifier( 1641 *Current.Previous, /* AcceptIdentifierName= */ true) 1642 : Current.Previous->is(tok::identifier); 1643 if (IsIdentifier || 1644 Current.Previous->isOneOf( 1645 tok::kw_namespace, tok::r_paren, tok::r_square, tok::r_brace, 1646 tok::kw_false, tok::kw_true, Keywords.kw_type, Keywords.kw_get, 1647 Keywords.kw_set) || 1648 Current.Previous->Tok.isLiteral()) { 1649 Current.setType(TT_NonNullAssertion); 1650 return; 1651 } 1652 } 1653 if (Current.Next && 1654 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1655 Current.setType(TT_NonNullAssertion); 1656 return; 1657 } 1658 } 1659 1660 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1661 // function declaration have been found. In this case, 'Current' is a 1662 // trailing token of this declaration and thus cannot be a name. 1663 if (Current.is(Keywords.kw_instanceof)) { 1664 Current.setType(TT_BinaryOperator); 1665 } else if (isStartOfName(Current) && 1666 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1667 Contexts.back().FirstStartOfName = &Current; 1668 Current.setType(TT_StartOfName); 1669 } else if (Current.is(tok::semi)) { 1670 // Reset FirstStartOfName after finding a semicolon so that a for loop 1671 // with multiple increment statements is not confused with a for loop 1672 // having multiple variable declarations. 1673 Contexts.back().FirstStartOfName = nullptr; 1674 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1675 AutoFound = true; 1676 } else if (Current.is(tok::arrow) && 1677 Style.Language == FormatStyle::LK_Java) { 1678 Current.setType(TT_LambdaArrow); 1679 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1680 Current.NestingLevel == 0 && 1681 !Current.Previous->isOneOf(tok::kw_operator, tok::identifier)) { 1682 // not auto operator->() -> xxx; 1683 Current.setType(TT_TrailingReturnArrow); 1684 } else if (Current.is(tok::arrow) && Current.Previous && 1685 Current.Previous->is(tok::r_brace)) { 1686 // Concept implicit conversion constraint needs to be treated like 1687 // a trailing return type ... } -> <type>. 1688 Current.setType(TT_TrailingReturnArrow); 1689 } else if (isDeductionGuide(Current)) { 1690 // Deduction guides trailing arrow " A(...) -> A<T>;". 1691 Current.setType(TT_TrailingReturnArrow); 1692 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1693 Current.setType(determineStarAmpUsage( 1694 Current, 1695 Contexts.back().CanBeExpression && Contexts.back().IsExpression, 1696 Contexts.back().InTemplateArgument)); 1697 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1698 Current.setType(determinePlusMinusCaretUsage(Current)); 1699 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1700 Contexts.back().CaretFound = true; 1701 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1702 Current.setType(determineIncrementUsage(Current)); 1703 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1704 Current.setType(TT_UnaryOperator); 1705 } else if (Current.is(tok::question)) { 1706 if (Style.isJavaScript() && Line.MustBeDeclaration && 1707 !Contexts.back().IsExpression) { 1708 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1709 // on the interface, not a ternary expression. 1710 Current.setType(TT_JsTypeOptionalQuestion); 1711 } else { 1712 Current.setType(TT_ConditionalExpr); 1713 } 1714 } else if (Current.isBinaryOperator() && 1715 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1716 (!Current.is(tok::greater) && 1717 Style.Language != FormatStyle::LK_TextProto)) { 1718 Current.setType(TT_BinaryOperator); 1719 } else if (Current.is(tok::comment)) { 1720 if (Current.TokenText.startswith("/*")) { 1721 if (Current.TokenText.endswith("*/")) 1722 Current.setType(TT_BlockComment); 1723 else 1724 // The lexer has for some reason determined a comment here. But we 1725 // cannot really handle it, if it isn't properly terminated. 1726 Current.Tok.setKind(tok::unknown); 1727 } else { 1728 Current.setType(TT_LineComment); 1729 } 1730 } else if (Current.is(tok::r_paren)) { 1731 if (rParenEndsCast(Current)) 1732 Current.setType(TT_CastRParen); 1733 if (Current.MatchingParen && Current.Next && 1734 !Current.Next->isBinaryOperator() && 1735 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1736 tok::comma, tok::period, tok::arrow, 1737 tok::coloncolon)) 1738 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1739 // Make sure this isn't the return type of an Obj-C block declaration 1740 if (AfterParen->Tok.isNot(tok::caret)) { 1741 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1742 if (BeforeParen->is(tok::identifier) && 1743 !BeforeParen->is(TT_TypenameMacro) && 1744 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1745 (!BeforeParen->Previous || 1746 BeforeParen->Previous->ClosesTemplateDeclaration)) 1747 Current.setType(TT_FunctionAnnotationRParen); 1748 } 1749 } 1750 } else if (Current.is(tok::at) && Current.Next && !Style.isJavaScript() && 1751 Style.Language != FormatStyle::LK_Java) { 1752 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1753 // marks declarations and properties that need special formatting. 1754 switch (Current.Next->Tok.getObjCKeywordID()) { 1755 case tok::objc_interface: 1756 case tok::objc_implementation: 1757 case tok::objc_protocol: 1758 Current.setType(TT_ObjCDecl); 1759 break; 1760 case tok::objc_property: 1761 Current.setType(TT_ObjCProperty); 1762 break; 1763 default: 1764 break; 1765 } 1766 } else if (Current.is(tok::period)) { 1767 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1768 if (PreviousNoComment && 1769 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1770 Current.setType(TT_DesignatedInitializerPeriod); 1771 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1772 Current.Previous->isOneOf(TT_JavaAnnotation, 1773 TT_LeadingJavaAnnotation)) { 1774 Current.setType(Current.Previous->getType()); 1775 } 1776 } else if (canBeObjCSelectorComponent(Current) && 1777 // FIXME(bug 36976): ObjC return types shouldn't use 1778 // TT_CastRParen. 1779 Current.Previous && Current.Previous->is(TT_CastRParen) && 1780 Current.Previous->MatchingParen && 1781 Current.Previous->MatchingParen->Previous && 1782 Current.Previous->MatchingParen->Previous->is( 1783 TT_ObjCMethodSpecifier)) { 1784 // This is the first part of an Objective-C selector name. (If there's no 1785 // colon after this, this is the only place which annotates the identifier 1786 // as a selector.) 1787 Current.setType(TT_SelectorName); 1788 } else if (Current.isOneOf(tok::identifier, tok::kw_const, tok::kw_noexcept, 1789 tok::kw_requires) && 1790 Current.Previous && 1791 !Current.Previous->isOneOf(tok::equal, tok::at) && 1792 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1793 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1794 // function declaration have been found. 1795 Current.setType(TT_TrailingAnnotation); 1796 } else if ((Style.Language == FormatStyle::LK_Java || 1797 Style.isJavaScript()) && 1798 Current.Previous) { 1799 if (Current.Previous->is(tok::at) && 1800 Current.isNot(Keywords.kw_interface)) { 1801 const FormatToken &AtToken = *Current.Previous; 1802 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1803 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1804 Current.setType(TT_LeadingJavaAnnotation); 1805 else 1806 Current.setType(TT_JavaAnnotation); 1807 } else if (Current.Previous->is(tok::period) && 1808 Current.Previous->isOneOf(TT_JavaAnnotation, 1809 TT_LeadingJavaAnnotation)) { 1810 Current.setType(Current.Previous->getType()); 1811 } 1812 } 1813 } 1814 1815 /// Take a guess at whether \p Tok starts a name of a function or 1816 /// variable declaration. 1817 /// 1818 /// This is a heuristic based on whether \p Tok is an identifier following 1819 /// something that is likely a type. 1820 bool isStartOfName(const FormatToken &Tok) { 1821 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1822 return false; 1823 1824 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1825 Keywords.kw_as)) 1826 return false; 1827 if (Style.isJavaScript() && Tok.Previous->is(Keywords.kw_in)) 1828 return false; 1829 1830 // Skip "const" as it does not have an influence on whether this is a name. 1831 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1832 1833 // For javascript const can be like "let" or "var" 1834 if (!Style.isJavaScript()) 1835 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1836 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1837 1838 if (!PreviousNotConst) 1839 return false; 1840 1841 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1842 PreviousNotConst->Previous && 1843 PreviousNotConst->Previous->is(tok::hash); 1844 1845 if (PreviousNotConst->is(TT_TemplateCloser)) 1846 return PreviousNotConst && PreviousNotConst->MatchingParen && 1847 PreviousNotConst->MatchingParen->Previous && 1848 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1849 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1850 1851 if (PreviousNotConst->is(tok::r_paren) && 1852 PreviousNotConst->is(TT_TypeDeclarationParen)) 1853 return true; 1854 1855 // If is a preprocess keyword like #define. 1856 if (IsPPKeyword) 1857 return false; 1858 1859 // int a or auto a. 1860 if (PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) 1861 return true; 1862 1863 // *a or &a or &&a. 1864 if (PreviousNotConst->is(TT_PointerOrReference)) 1865 return true; 1866 1867 // MyClass a; 1868 if (PreviousNotConst->isSimpleTypeSpecifier()) 1869 return true; 1870 1871 // const a = in JavaScript. 1872 return (Style.isJavaScript() && PreviousNotConst->is(tok::kw_const)); 1873 } 1874 1875 /// Determine whether ')' is ending a cast. 1876 bool rParenEndsCast(const FormatToken &Tok) { 1877 // C-style casts are only used in C++, C# and Java. 1878 if (!Style.isCSharp() && !Style.isCpp() && 1879 Style.Language != FormatStyle::LK_Java) 1880 return false; 1881 1882 // Empty parens aren't casts and there are no casts at the end of the line. 1883 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1884 return false; 1885 1886 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1887 if (LeftOfParens) { 1888 // If there is a closing parenthesis left of the current 1889 // parentheses, look past it as these might be chained casts. 1890 if (LeftOfParens->is(tok::r_paren) && 1891 LeftOfParens->isNot(TT_CastRParen)) { 1892 if (!LeftOfParens->MatchingParen || 1893 !LeftOfParens->MatchingParen->Previous) 1894 return false; 1895 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1896 } 1897 1898 // The Condition directly below this one will see the operator arguments 1899 // as a (void *foo) cast. 1900 // void operator delete(void *foo) ATTRIB; 1901 if (LeftOfParens->Tok.getIdentifierInfo() && LeftOfParens->Previous && 1902 LeftOfParens->Previous->is(tok::kw_operator)) 1903 return false; 1904 1905 // If there is an identifier (or with a few exceptions a keyword) right 1906 // before the parentheses, this is unlikely to be a cast. 1907 if (LeftOfParens->Tok.getIdentifierInfo() && 1908 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1909 tok::kw_delete)) 1910 return false; 1911 1912 // Certain other tokens right before the parentheses are also signals that 1913 // this cannot be a cast. 1914 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1915 TT_TemplateCloser, tok::ellipsis)) 1916 return false; 1917 } 1918 1919 if (Tok.Next->is(tok::question)) 1920 return false; 1921 1922 // `foreach((A a, B b) in someList)` should not be seen as a cast. 1923 if (Tok.Next->is(Keywords.kw_in) && Style.isCSharp()) 1924 return false; 1925 1926 // Functions which end with decorations like volatile, noexcept are unlikely 1927 // to be casts. 1928 if (Tok.Next->isOneOf(tok::kw_noexcept, tok::kw_volatile, tok::kw_const, 1929 tok::kw_requires, tok::kw_throw, tok::arrow, 1930 Keywords.kw_override, Keywords.kw_final) || 1931 isCpp11AttributeSpecifier(*Tok.Next)) 1932 return false; 1933 1934 // As Java has no function types, a "(" after the ")" likely means that this 1935 // is a cast. 1936 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1937 return true; 1938 1939 // If a (non-string) literal follows, this is likely a cast. 1940 if (Tok.Next->isNot(tok::string_literal) && 1941 (Tok.Next->Tok.isLiteral() || 1942 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1943 return true; 1944 1945 // Heuristically try to determine whether the parentheses contain a type. 1946 auto IsQualifiedPointerOrReference = [](FormatToken *T) { 1947 // This is used to handle cases such as x = (foo *const)&y; 1948 assert(!T->isSimpleTypeSpecifier() && "Should have already been checked"); 1949 // Strip trailing qualifiers such as const or volatile when checking 1950 // whether the parens could be a cast to a pointer/reference type. 1951 while (T) { 1952 if (T->is(TT_AttributeParen)) { 1953 // Handle `x = (foo *__attribute__((foo)))&v;`: 1954 if (T->MatchingParen && T->MatchingParen->Previous && 1955 T->MatchingParen->Previous->is(tok::kw___attribute)) { 1956 T = T->MatchingParen->Previous->Previous; 1957 continue; 1958 } 1959 } else if (T->is(TT_AttributeSquare)) { 1960 // Handle `x = (foo *[[clang::foo]])&v;`: 1961 if (T->MatchingParen && T->MatchingParen->Previous) { 1962 T = T->MatchingParen->Previous; 1963 continue; 1964 } 1965 } else if (T->canBePointerOrReferenceQualifier()) { 1966 T = T->Previous; 1967 continue; 1968 } 1969 break; 1970 } 1971 return T && T->is(TT_PointerOrReference); 1972 }; 1973 bool ParensAreType = 1974 !Tok.Previous || 1975 Tok.Previous->isOneOf(TT_TemplateCloser, TT_TypeDeclarationParen) || 1976 Tok.Previous->isSimpleTypeSpecifier() || 1977 IsQualifiedPointerOrReference(Tok.Previous); 1978 bool ParensCouldEndDecl = 1979 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1980 if (ParensAreType && !ParensCouldEndDecl) 1981 return true; 1982 1983 // At this point, we heuristically assume that there are no casts at the 1984 // start of the line. We assume that we have found most cases where there 1985 // are by the logic above, e.g. "(void)x;". 1986 if (!LeftOfParens) 1987 return false; 1988 1989 // Certain token types inside the parentheses mean that this can't be a 1990 // cast. 1991 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 1992 Token = Token->Next) 1993 if (Token->is(TT_BinaryOperator)) 1994 return false; 1995 1996 // If the following token is an identifier or 'this', this is a cast. All 1997 // cases where this can be something else are handled above. 1998 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1999 return true; 2000 2001 // Look for a cast `( x ) (`. 2002 if (Tok.Next->is(tok::l_paren) && Tok.Previous && Tok.Previous->Previous) { 2003 if (Tok.Previous->is(tok::identifier) && 2004 Tok.Previous->Previous->is(tok::l_paren)) 2005 return true; 2006 } 2007 2008 if (!Tok.Next->Next) 2009 return false; 2010 2011 // If the next token after the parenthesis is a unary operator, assume 2012 // that this is cast, unless there are unexpected tokens inside the 2013 // parenthesis. 2014 bool NextIsUnary = 2015 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 2016 if (!NextIsUnary || Tok.Next->is(tok::plus) || 2017 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 2018 return false; 2019 // Search for unexpected tokens. 2020 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 2021 Prev = Prev->Previous) { 2022 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 2023 return false; 2024 } 2025 return true; 2026 } 2027 2028 /// Return the type of the given token assuming it is * or &. 2029 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 2030 bool InTemplateArgument) { 2031 if (Style.isJavaScript()) 2032 return TT_BinaryOperator; 2033 2034 // && in C# must be a binary operator. 2035 if (Style.isCSharp() && Tok.is(tok::ampamp)) 2036 return TT_BinaryOperator; 2037 2038 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2039 if (!PrevToken) 2040 return TT_UnaryOperator; 2041 2042 const FormatToken *NextToken = Tok.getNextNonComment(); 2043 if (!NextToken || 2044 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_noexcept) || 2045 NextToken->canBePointerOrReferenceQualifier() || 2046 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 2047 return TT_PointerOrReference; 2048 2049 if (PrevToken->is(tok::coloncolon)) 2050 return TT_PointerOrReference; 2051 2052 if (PrevToken->is(tok::r_paren) && PrevToken->is(TT_TypeDeclarationParen)) 2053 return TT_PointerOrReference; 2054 2055 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 2056 tok::comma, tok::semi, tok::kw_return, tok::colon, 2057 tok::kw_co_return, tok::kw_co_await, 2058 tok::kw_co_yield, tok::equal, tok::kw_delete, 2059 tok::kw_sizeof, tok::kw_throw, TT_BinaryOperator, 2060 TT_ConditionalExpr, TT_UnaryOperator, TT_CastRParen)) 2061 return TT_UnaryOperator; 2062 2063 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 2064 return TT_PointerOrReference; 2065 if (NextToken->is(tok::kw_operator) && !IsExpression) 2066 return TT_PointerOrReference; 2067 if (NextToken->isOneOf(tok::comma, tok::semi)) 2068 return TT_PointerOrReference; 2069 2070 if (PrevToken->Tok.isLiteral() || 2071 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 2072 tok::kw_false, tok::r_brace) || 2073 NextToken->Tok.isLiteral() || 2074 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 2075 NextToken->isUnaryOperator() || 2076 // If we know we're in a template argument, there are no named 2077 // declarations. Thus, having an identifier on the right-hand side 2078 // indicates a binary operator. 2079 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 2080 return TT_BinaryOperator; 2081 2082 // "&&(" is quite unlikely to be two successive unary "&". 2083 if (Tok.is(tok::ampamp) && NextToken->is(tok::l_paren)) 2084 return TT_BinaryOperator; 2085 2086 // This catches some cases where evaluation order is used as control flow: 2087 // aaa && aaa->f(); 2088 if (NextToken->Tok.isAnyIdentifier()) { 2089 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 2090 if (NextNextToken && NextNextToken->is(tok::arrow)) 2091 return TT_BinaryOperator; 2092 } 2093 2094 // It is very unlikely that we are going to find a pointer or reference type 2095 // definition on the RHS of an assignment. 2096 if (IsExpression && !Contexts.back().CaretFound) 2097 return TT_BinaryOperator; 2098 2099 return TT_PointerOrReference; 2100 } 2101 2102 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 2103 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2104 if (!PrevToken) 2105 return TT_UnaryOperator; 2106 2107 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 2108 // This must be a sequence of leading unary operators. 2109 return TT_UnaryOperator; 2110 2111 // Use heuristics to recognize unary operators. 2112 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 2113 tok::question, tok::colon, tok::kw_return, 2114 tok::kw_case, tok::at, tok::l_brace, tok::kw_throw, 2115 tok::kw_co_return, tok::kw_co_yield)) 2116 return TT_UnaryOperator; 2117 2118 // There can't be two consecutive binary operators. 2119 if (PrevToken->is(TT_BinaryOperator)) 2120 return TT_UnaryOperator; 2121 2122 // Fall back to marking the token as binary operator. 2123 return TT_BinaryOperator; 2124 } 2125 2126 /// Determine whether ++/-- are pre- or post-increments/-decrements. 2127 TokenType determineIncrementUsage(const FormatToken &Tok) { 2128 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2129 if (!PrevToken || PrevToken->is(TT_CastRParen)) 2130 return TT_UnaryOperator; 2131 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 2132 return TT_TrailingUnaryOperator; 2133 2134 return TT_UnaryOperator; 2135 } 2136 2137 SmallVector<Context, 8> Contexts; 2138 2139 const FormatStyle &Style; 2140 AnnotatedLine &Line; 2141 FormatToken *CurrentToken; 2142 bool AutoFound; 2143 const AdditionalKeywords &Keywords; 2144 2145 // Set of "<" tokens that do not open a template parameter list. If parseAngle 2146 // determines that a specific token can't be a template opener, it will make 2147 // same decision irrespective of the decisions for tokens leading up to it. 2148 // Store this information to prevent this from causing exponential runtime. 2149 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 2150 }; 2151 2152 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 2153 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 2154 2155 /// Parses binary expressions by inserting fake parenthesis based on 2156 /// operator precedence. 2157 class ExpressionParser { 2158 public: 2159 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 2160 AnnotatedLine &Line) 2161 : Style(Style), Keywords(Keywords), Current(Line.First) {} 2162 2163 /// Parse expressions with the given operator precedence. 2164 void parse(int Precedence = 0) { 2165 // Skip 'return' and ObjC selector colons as they are not part of a binary 2166 // expression. 2167 while (Current && (Current->is(tok::kw_return) || 2168 (Current->is(tok::colon) && 2169 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 2170 next(); 2171 2172 if (!Current || Precedence > PrecedenceArrowAndPeriod) 2173 return; 2174 2175 // Conditional expressions need to be parsed separately for proper nesting. 2176 if (Precedence == prec::Conditional) { 2177 parseConditionalExpr(); 2178 return; 2179 } 2180 2181 // Parse unary operators, which all have a higher precedence than binary 2182 // operators. 2183 if (Precedence == PrecedenceUnaryOperator) { 2184 parseUnaryOperator(); 2185 return; 2186 } 2187 2188 FormatToken *Start = Current; 2189 FormatToken *LatestOperator = nullptr; 2190 unsigned OperatorIndex = 0; 2191 2192 while (Current) { 2193 // Consume operators with higher precedence. 2194 parse(Precedence + 1); 2195 2196 int CurrentPrecedence = getCurrentPrecedence(); 2197 2198 if (Precedence == CurrentPrecedence && Current && 2199 Current->is(TT_SelectorName)) { 2200 if (LatestOperator) 2201 addFakeParenthesis(Start, prec::Level(Precedence)); 2202 Start = Current; 2203 } 2204 2205 // At the end of the line or when an operator with higher precedence is 2206 // found, insert fake parenthesis and return. 2207 if (!Current || 2208 (Current->closesScope() && 2209 (Current->MatchingParen || Current->is(TT_TemplateString))) || 2210 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 2211 (CurrentPrecedence == prec::Conditional && 2212 Precedence == prec::Assignment && Current->is(tok::colon))) { 2213 break; 2214 } 2215 2216 // Consume scopes: (), [], <> and {} 2217 if (Current->opensScope()) { 2218 // In fragment of a JavaScript template string can look like '}..${' and 2219 // thus close a scope and open a new one at the same time. 2220 while (Current && (!Current->closesScope() || Current->opensScope())) { 2221 next(); 2222 parse(); 2223 } 2224 next(); 2225 } else { 2226 // Operator found. 2227 if (CurrentPrecedence == Precedence) { 2228 if (LatestOperator) 2229 LatestOperator->NextOperator = Current; 2230 LatestOperator = Current; 2231 Current->OperatorIndex = OperatorIndex; 2232 ++OperatorIndex; 2233 } 2234 next(/*SkipPastLeadingComments=*/Precedence > 0); 2235 } 2236 } 2237 2238 if (LatestOperator && (Current || Precedence > 0)) { 2239 // LatestOperator->LastOperator = true; 2240 if (Precedence == PrecedenceArrowAndPeriod) { 2241 // Call expressions don't have a binary operator precedence. 2242 addFakeParenthesis(Start, prec::Unknown); 2243 } else { 2244 addFakeParenthesis(Start, prec::Level(Precedence)); 2245 } 2246 } 2247 } 2248 2249 private: 2250 /// Gets the precedence (+1) of the given token for binary operators 2251 /// and other tokens that we treat like binary operators. 2252 int getCurrentPrecedence() { 2253 if (Current) { 2254 const FormatToken *NextNonComment = Current->getNextNonComment(); 2255 if (Current->is(TT_ConditionalExpr)) 2256 return prec::Conditional; 2257 if (NextNonComment && Current->is(TT_SelectorName) && 2258 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 2259 ((Style.Language == FormatStyle::LK_Proto || 2260 Style.Language == FormatStyle::LK_TextProto) && 2261 NextNonComment->is(tok::less)))) 2262 return prec::Assignment; 2263 if (Current->is(TT_JsComputedPropertyName)) 2264 return prec::Assignment; 2265 if (Current->is(TT_LambdaArrow)) 2266 return prec::Comma; 2267 if (Current->is(TT_FatArrow)) 2268 return prec::Assignment; 2269 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 2270 (Current->is(tok::comment) && NextNonComment && 2271 NextNonComment->is(TT_SelectorName))) 2272 return 0; 2273 if (Current->is(TT_RangeBasedForLoopColon)) 2274 return prec::Comma; 2275 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2276 Current->is(Keywords.kw_instanceof)) 2277 return prec::Relational; 2278 if (Style.isJavaScript() && 2279 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 2280 return prec::Relational; 2281 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 2282 return Current->getPrecedence(); 2283 if (Current->isOneOf(tok::period, tok::arrow)) 2284 return PrecedenceArrowAndPeriod; 2285 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 2286 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 2287 Keywords.kw_throws)) 2288 return 0; 2289 } 2290 return -1; 2291 } 2292 2293 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 2294 Start->FakeLParens.push_back(Precedence); 2295 if (Precedence > prec::Unknown) 2296 Start->StartsBinaryExpression = true; 2297 if (Current) { 2298 FormatToken *Previous = Current->Previous; 2299 while (Previous->is(tok::comment) && Previous->Previous) 2300 Previous = Previous->Previous; 2301 ++Previous->FakeRParens; 2302 if (Precedence > prec::Unknown) 2303 Previous->EndsBinaryExpression = true; 2304 } 2305 } 2306 2307 /// Parse unary operator expressions and surround them with fake 2308 /// parentheses if appropriate. 2309 void parseUnaryOperator() { 2310 llvm::SmallVector<FormatToken *, 2> Tokens; 2311 while (Current && Current->is(TT_UnaryOperator)) { 2312 Tokens.push_back(Current); 2313 next(); 2314 } 2315 parse(PrecedenceArrowAndPeriod); 2316 for (FormatToken *Token : llvm::reverse(Tokens)) 2317 // The actual precedence doesn't matter. 2318 addFakeParenthesis(Token, prec::Unknown); 2319 } 2320 2321 void parseConditionalExpr() { 2322 while (Current && Current->isTrailingComment()) { 2323 next(); 2324 } 2325 FormatToken *Start = Current; 2326 parse(prec::LogicalOr); 2327 if (!Current || !Current->is(tok::question)) 2328 return; 2329 next(); 2330 parse(prec::Assignment); 2331 if (!Current || Current->isNot(TT_ConditionalExpr)) 2332 return; 2333 next(); 2334 parse(prec::Assignment); 2335 addFakeParenthesis(Start, prec::Conditional); 2336 } 2337 2338 void next(bool SkipPastLeadingComments = true) { 2339 if (Current) 2340 Current = Current->Next; 2341 while (Current && 2342 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 2343 Current->isTrailingComment()) 2344 Current = Current->Next; 2345 } 2346 2347 const FormatStyle &Style; 2348 const AdditionalKeywords &Keywords; 2349 FormatToken *Current; 2350 }; 2351 2352 } // end anonymous namespace 2353 2354 void TokenAnnotator::setCommentLineLevels( 2355 SmallVectorImpl<AnnotatedLine *> &Lines) { 2356 const AnnotatedLine *NextNonCommentLine = nullptr; 2357 for (AnnotatedLine *Line : llvm::reverse(Lines)) { 2358 assert(Line->First); 2359 bool CommentLine = true; 2360 for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) { 2361 if (!Tok->is(tok::comment)) { 2362 CommentLine = false; 2363 break; 2364 } 2365 } 2366 2367 // If the comment is currently aligned with the line immediately following 2368 // it, that's probably intentional and we should keep it. 2369 if (NextNonCommentLine && CommentLine && 2370 NextNonCommentLine->First->NewlinesBefore <= 1 && 2371 NextNonCommentLine->First->OriginalColumn == 2372 Line->First->OriginalColumn) { 2373 // Align comments for preprocessor lines with the # in column 0 if 2374 // preprocessor lines are not indented. Otherwise, align with the next 2375 // line. 2376 Line->Level = 2377 (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 2378 (NextNonCommentLine->Type == LT_PreprocessorDirective || 2379 NextNonCommentLine->Type == LT_ImportStatement)) 2380 ? 0 2381 : NextNonCommentLine->Level; 2382 } else { 2383 NextNonCommentLine = Line->First->isNot(tok::r_brace) ? Line : nullptr; 2384 } 2385 2386 setCommentLineLevels(Line->Children); 2387 } 2388 } 2389 2390 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 2391 unsigned Result = 0; 2392 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 2393 Result = std::max(Result, Tok->NestingLevel); 2394 return Result; 2395 } 2396 2397 void TokenAnnotator::annotate(AnnotatedLine &Line) { 2398 for (auto &Child : Line.Children) 2399 annotate(*Child); 2400 2401 AnnotatingParser Parser(Style, Line, Keywords); 2402 Line.Type = Parser.parseLine(); 2403 2404 // With very deep nesting, ExpressionParser uses lots of stack and the 2405 // formatting algorithm is very slow. We're not going to do a good job here 2406 // anyway - it's probably generated code being formatted by mistake. 2407 // Just skip the whole line. 2408 if (maxNestingDepth(Line) > 50) 2409 Line.Type = LT_Invalid; 2410 2411 if (Line.Type == LT_Invalid) 2412 return; 2413 2414 ExpressionParser ExprParser(Style, Keywords, Line); 2415 ExprParser.parse(); 2416 2417 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2418 Line.Type = LT_ObjCMethodDecl; 2419 else if (Line.startsWith(TT_ObjCDecl)) 2420 Line.Type = LT_ObjCDecl; 2421 else if (Line.startsWith(TT_ObjCProperty)) 2422 Line.Type = LT_ObjCProperty; 2423 2424 Line.First->SpacesRequiredBefore = 1; 2425 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 2426 } 2427 2428 // This function heuristically determines whether 'Current' starts the name of a 2429 // function declaration. 2430 static bool isFunctionDeclarationName(bool IsCpp, const FormatToken &Current, 2431 const AnnotatedLine &Line) { 2432 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 2433 for (; Next; Next = Next->Next) { 2434 if (Next->is(TT_OverloadedOperatorLParen)) 2435 return Next; 2436 if (Next->is(TT_OverloadedOperator)) 2437 continue; 2438 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 2439 // For 'new[]' and 'delete[]'. 2440 if (Next->Next && 2441 Next->Next->startsSequence(tok::l_square, tok::r_square)) 2442 Next = Next->Next->Next; 2443 continue; 2444 } 2445 if (Next->startsSequence(tok::l_square, tok::r_square)) { 2446 // For operator[](). 2447 Next = Next->Next; 2448 continue; 2449 } 2450 if ((Next->isSimpleTypeSpecifier() || Next->is(tok::identifier)) && 2451 Next->Next && Next->Next->isOneOf(tok::star, tok::amp, tok::ampamp)) { 2452 // For operator void*(), operator char*(), operator Foo*(). 2453 Next = Next->Next; 2454 continue; 2455 } 2456 if (Next->is(TT_TemplateOpener) && Next->MatchingParen) { 2457 Next = Next->MatchingParen; 2458 continue; 2459 } 2460 2461 break; 2462 } 2463 return nullptr; 2464 }; 2465 2466 // Find parentheses of parameter list. 2467 const FormatToken *Next = Current.Next; 2468 if (Current.is(tok::kw_operator)) { 2469 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 2470 return false; 2471 Next = skipOperatorName(Next); 2472 } else { 2473 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 2474 return false; 2475 for (; Next; Next = Next->Next) { 2476 if (Next->is(TT_TemplateOpener)) { 2477 Next = Next->MatchingParen; 2478 } else if (Next->is(tok::coloncolon)) { 2479 Next = Next->Next; 2480 if (!Next) 2481 return false; 2482 if (Next->is(tok::kw_operator)) { 2483 Next = skipOperatorName(Next->Next); 2484 break; 2485 } 2486 if (!Next->is(tok::identifier)) 2487 return false; 2488 } else if (Next->is(tok::l_paren)) { 2489 break; 2490 } else { 2491 return false; 2492 } 2493 } 2494 } 2495 2496 // Check whether parameter list can belong to a function declaration. 2497 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 2498 return false; 2499 // If the lines ends with "{", this is likely a function definition. 2500 if (Line.Last->is(tok::l_brace)) 2501 return true; 2502 if (Next->Next == Next->MatchingParen) 2503 return true; // Empty parentheses. 2504 // If there is an &/&& after the r_paren, this is likely a function. 2505 if (Next->MatchingParen->Next && 2506 Next->MatchingParen->Next->is(TT_PointerOrReference)) 2507 return true; 2508 2509 // Check for K&R C function definitions (and C++ function definitions with 2510 // unnamed parameters), e.g.: 2511 // int f(i) 2512 // { 2513 // return i + 1; 2514 // } 2515 // bool g(size_t = 0, bool b = false) 2516 // { 2517 // return !b; 2518 // } 2519 if (IsCpp && Next->Next && Next->Next->is(tok::identifier) && 2520 !Line.endsWith(tok::semi)) 2521 return true; 2522 2523 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 2524 Tok = Tok->Next) { 2525 if (Tok->is(TT_TypeDeclarationParen)) 2526 return true; 2527 if (Tok->isOneOf(tok::l_paren, TT_TemplateOpener) && Tok->MatchingParen) { 2528 Tok = Tok->MatchingParen; 2529 continue; 2530 } 2531 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 2532 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 2533 return true; 2534 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 2535 Tok->Tok.isLiteral()) 2536 return false; 2537 } 2538 return false; 2539 } 2540 2541 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 2542 assert(Line.MightBeFunctionDecl); 2543 2544 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 2545 Style.AlwaysBreakAfterReturnType == 2546 FormatStyle::RTBS_TopLevelDefinitions) && 2547 Line.Level > 0) 2548 return false; 2549 2550 switch (Style.AlwaysBreakAfterReturnType) { 2551 case FormatStyle::RTBS_None: 2552 return false; 2553 case FormatStyle::RTBS_All: 2554 case FormatStyle::RTBS_TopLevel: 2555 return true; 2556 case FormatStyle::RTBS_AllDefinitions: 2557 case FormatStyle::RTBS_TopLevelDefinitions: 2558 return Line.mightBeFunctionDefinition(); 2559 } 2560 2561 return false; 2562 } 2563 2564 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 2565 for (AnnotatedLine *ChildLine : Line.Children) 2566 calculateFormattingInformation(*ChildLine); 2567 2568 Line.First->TotalLength = 2569 Line.First->IsMultiline ? Style.ColumnLimit 2570 : Line.FirstStartColumn + Line.First->ColumnWidth; 2571 FormatToken *Current = Line.First->Next; 2572 bool InFunctionDecl = Line.MightBeFunctionDecl; 2573 bool AlignArrayOfStructures = 2574 (Style.AlignArrayOfStructures != FormatStyle::AIAS_None && 2575 Line.Type == LT_ArrayOfStructInitializer); 2576 if (AlignArrayOfStructures) 2577 calculateArrayInitializerColumnList(Line); 2578 2579 while (Current) { 2580 if (isFunctionDeclarationName(Style.isCpp(), *Current, Line)) 2581 Current->setType(TT_FunctionDeclarationName); 2582 const FormatToken *Prev = Current->Previous; 2583 if (Current->is(TT_LineComment)) { 2584 if (Prev->is(BK_BracedInit) && Prev->opensScope()) 2585 Current->SpacesRequiredBefore = 2586 (Style.Cpp11BracedListStyle && !Style.SpacesInParentheses) ? 0 : 1; 2587 else 2588 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2589 2590 // If we find a trailing comment, iterate backwards to determine whether 2591 // it seems to relate to a specific parameter. If so, break before that 2592 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2593 // to the previous line in: 2594 // SomeFunction(a, 2595 // b, // comment 2596 // c); 2597 if (!Current->HasUnescapedNewline) { 2598 for (FormatToken *Parameter = Current->Previous; Parameter; 2599 Parameter = Parameter->Previous) { 2600 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 2601 break; 2602 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 2603 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 2604 Parameter->HasUnescapedNewline) 2605 Parameter->MustBreakBefore = true; 2606 break; 2607 } 2608 } 2609 } 2610 } else if (Current->SpacesRequiredBefore == 0 && 2611 spaceRequiredBefore(Line, *Current)) { 2612 Current->SpacesRequiredBefore = 1; 2613 } 2614 2615 Current->MustBreakBefore = 2616 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 2617 2618 if (!Current->MustBreakBefore && InFunctionDecl && 2619 Current->is(TT_FunctionDeclarationName)) 2620 Current->MustBreakBefore = mustBreakForReturnType(Line); 2621 2622 Current->CanBreakBefore = 2623 Current->MustBreakBefore || canBreakBefore(Line, *Current); 2624 unsigned ChildSize = 0; 2625 if (Prev->Children.size() == 1) { 2626 FormatToken &LastOfChild = *Prev->Children[0]->Last; 2627 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 2628 : LastOfChild.TotalLength + 1; 2629 } 2630 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 2631 (Prev->Children.size() == 1 && 2632 Prev->Children[0]->First->MustBreakBefore) || 2633 Current->IsMultiline) 2634 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 2635 else 2636 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 2637 ChildSize + Current->SpacesRequiredBefore; 2638 2639 if (Current->is(TT_CtorInitializerColon)) 2640 InFunctionDecl = false; 2641 2642 // FIXME: Only calculate this if CanBreakBefore is true once static 2643 // initializers etc. are sorted out. 2644 // FIXME: Move magic numbers to a better place. 2645 2646 // Reduce penalty for aligning ObjC method arguments using the colon 2647 // alignment as this is the canonical way (still prefer fitting everything 2648 // into one line if possible). Trying to fit a whole expression into one 2649 // line should not force other line breaks (e.g. when ObjC method 2650 // expression is a part of other expression). 2651 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl); 2652 if (Style.Language == FormatStyle::LK_ObjC && 2653 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) { 2654 if (Current->ParameterIndex == 1) 2655 Current->SplitPenalty += 5 * Current->BindingStrength; 2656 } else { 2657 Current->SplitPenalty += 20 * Current->BindingStrength; 2658 } 2659 2660 Current = Current->Next; 2661 } 2662 2663 calculateUnbreakableTailLengths(Line); 2664 unsigned IndentLevel = Line.Level; 2665 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2666 if (Current->Role) 2667 Current->Role->precomputeFormattingInfos(Current); 2668 if (Current->MatchingParen && 2669 Current->MatchingParen->opensBlockOrBlockTypeList(Style) && 2670 IndentLevel > 0) { 2671 --IndentLevel; 2672 } 2673 Current->IndentLevel = IndentLevel; 2674 if (Current->opensBlockOrBlockTypeList(Style)) 2675 ++IndentLevel; 2676 } 2677 2678 LLVM_DEBUG({ printDebugInfo(Line); }); 2679 } 2680 2681 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2682 unsigned UnbreakableTailLength = 0; 2683 FormatToken *Current = Line.Last; 2684 while (Current) { 2685 Current->UnbreakableTailLength = UnbreakableTailLength; 2686 if (Current->CanBreakBefore || 2687 Current->isOneOf(tok::comment, tok::string_literal)) { 2688 UnbreakableTailLength = 0; 2689 } else { 2690 UnbreakableTailLength += 2691 Current->ColumnWidth + Current->SpacesRequiredBefore; 2692 } 2693 Current = Current->Previous; 2694 } 2695 } 2696 2697 void TokenAnnotator::calculateArrayInitializerColumnList(AnnotatedLine &Line) { 2698 if (Line.First == Line.Last) { 2699 return; 2700 } 2701 auto *CurrentToken = Line.First; 2702 CurrentToken->ArrayInitializerLineStart = true; 2703 unsigned Depth = 0; 2704 while (CurrentToken != nullptr && CurrentToken != Line.Last) { 2705 if (CurrentToken->is(tok::l_brace)) { 2706 CurrentToken->IsArrayInitializer = true; 2707 if (CurrentToken->Next != nullptr) 2708 CurrentToken->Next->MustBreakBefore = true; 2709 CurrentToken = 2710 calculateInitializerColumnList(Line, CurrentToken->Next, Depth + 1); 2711 } else { 2712 CurrentToken = CurrentToken->Next; 2713 } 2714 } 2715 } 2716 2717 FormatToken *TokenAnnotator::calculateInitializerColumnList( 2718 AnnotatedLine &Line, FormatToken *CurrentToken, unsigned Depth) { 2719 while (CurrentToken != nullptr && CurrentToken != Line.Last) { 2720 if (CurrentToken->is(tok::l_brace)) 2721 ++Depth; 2722 else if (CurrentToken->is(tok::r_brace)) 2723 --Depth; 2724 if (Depth == 2 && CurrentToken->isOneOf(tok::l_brace, tok::comma)) { 2725 CurrentToken = CurrentToken->Next; 2726 if (CurrentToken == nullptr) 2727 break; 2728 CurrentToken->StartsColumn = true; 2729 CurrentToken = CurrentToken->Previous; 2730 } 2731 CurrentToken = CurrentToken->Next; 2732 } 2733 return CurrentToken; 2734 } 2735 2736 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2737 const FormatToken &Tok, 2738 bool InFunctionDecl) { 2739 const FormatToken &Left = *Tok.Previous; 2740 const FormatToken &Right = Tok; 2741 2742 if (Left.is(tok::semi)) 2743 return 0; 2744 2745 if (Style.Language == FormatStyle::LK_Java) { 2746 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2747 return 1; 2748 if (Right.is(Keywords.kw_implements)) 2749 return 2; 2750 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2751 return 3; 2752 } else if (Style.isJavaScript()) { 2753 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2754 return 100; 2755 if (Left.is(TT_JsTypeColon)) 2756 return 35; 2757 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2758 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2759 return 100; 2760 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2761 if (Left.opensScope() && Right.closesScope()) 2762 return 200; 2763 } 2764 2765 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2766 return 1; 2767 if (Right.is(tok::l_square)) { 2768 if (Style.Language == FormatStyle::LK_Proto) 2769 return 1; 2770 if (Left.is(tok::r_square)) 2771 return 200; 2772 // Slightly prefer formatting local lambda definitions like functions. 2773 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2774 return 35; 2775 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2776 TT_ArrayInitializerLSquare, 2777 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) 2778 return 500; 2779 } 2780 2781 if (Left.is(tok::coloncolon) || 2782 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2783 return 500; 2784 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2785 Right.is(tok::kw_operator)) { 2786 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2787 return 3; 2788 if (Left.is(TT_StartOfName)) 2789 return 110; 2790 if (InFunctionDecl && Right.NestingLevel == 0) 2791 return Style.PenaltyReturnTypeOnItsOwnLine; 2792 return 200; 2793 } 2794 if (Right.is(TT_PointerOrReference)) 2795 return 190; 2796 if (Right.is(TT_LambdaArrow)) 2797 return 110; 2798 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2799 return 160; 2800 if (Left.is(TT_CastRParen)) 2801 return 100; 2802 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2803 return 5000; 2804 if (Left.is(tok::comment)) 2805 return 1000; 2806 2807 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2808 TT_CtorInitializerColon)) 2809 return 2; 2810 2811 if (Right.isMemberAccess()) { 2812 // Breaking before the "./->" of a chained call/member access is reasonably 2813 // cheap, as formatting those with one call per line is generally 2814 // desirable. In particular, it should be cheaper to break before the call 2815 // than it is to break inside a call's parameters, which could lead to weird 2816 // "hanging" indents. The exception is the very last "./->" to support this 2817 // frequent pattern: 2818 // 2819 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2820 // dddddddd); 2821 // 2822 // which might otherwise be blown up onto many lines. Here, clang-format 2823 // won't produce "hanging" indents anyway as there is no other trailing 2824 // call. 2825 // 2826 // Also apply higher penalty is not a call as that might lead to a wrapping 2827 // like: 2828 // 2829 // aaaaaaa 2830 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2831 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2832 ? 150 2833 : 35; 2834 } 2835 2836 if (Right.is(TT_TrailingAnnotation) && 2837 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2838 // Moving trailing annotations to the next line is fine for ObjC method 2839 // declarations. 2840 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2841 return 10; 2842 // Generally, breaking before a trailing annotation is bad unless it is 2843 // function-like. It seems to be especially preferable to keep standard 2844 // annotations (i.e. "const", "final" and "override") on the same line. 2845 // Use a slightly higher penalty after ")" so that annotations like 2846 // "const override" are kept together. 2847 bool is_short_annotation = Right.TokenText.size() < 10; 2848 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 2849 } 2850 2851 // In for-loops, prefer breaking at ',' and ';'. 2852 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 2853 return 4; 2854 2855 // In Objective-C method expressions, prefer breaking before "param:" over 2856 // breaking after it. 2857 if (Right.is(TT_SelectorName)) 2858 return 0; 2859 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 2860 return Line.MightBeFunctionDecl ? 50 : 500; 2861 2862 // In Objective-C type declarations, avoid breaking after the category's 2863 // open paren (we'll prefer breaking after the protocol list's opening 2864 // angle bracket, if present). 2865 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous && 2866 Left.Previous->isOneOf(tok::identifier, tok::greater)) 2867 return 500; 2868 2869 if (Left.is(tok::l_paren) && Style.PenaltyBreakOpenParenthesis != 0) 2870 return Style.PenaltyBreakOpenParenthesis; 2871 if (Left.is(tok::l_paren) && InFunctionDecl && 2872 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 2873 return 100; 2874 if (Left.is(tok::l_paren) && Left.Previous && 2875 (Left.Previous->is(tok::kw_for) || Left.Previous->isIf())) 2876 return 1000; 2877 if (Left.is(tok::equal) && InFunctionDecl) 2878 return 110; 2879 if (Right.is(tok::r_brace)) 2880 return 1; 2881 if (Left.is(TT_TemplateOpener)) 2882 return 100; 2883 if (Left.opensScope()) { 2884 // If we aren't aligning after opening parens/braces we can always break 2885 // here unless the style does not want us to place all arguments on the 2886 // next line. 2887 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign && 2888 (Left.ParameterCount <= 1 || Style.AllowAllArgumentsOnNextLine)) 2889 return 0; 2890 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle) 2891 return 19; 2892 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 2893 : 19; 2894 } 2895 if (Left.is(TT_JavaAnnotation)) 2896 return 50; 2897 2898 if (Left.is(TT_UnaryOperator)) 2899 return 60; 2900 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 2901 Left.Previous->isLabelString() && 2902 (Left.NextOperator || Left.OperatorIndex != 0)) 2903 return 50; 2904 if (Right.is(tok::plus) && Left.isLabelString() && 2905 (Right.NextOperator || Right.OperatorIndex != 0)) 2906 return 25; 2907 if (Left.is(tok::comma)) 2908 return 1; 2909 if (Right.is(tok::lessless) && Left.isLabelString() && 2910 (Right.NextOperator || Right.OperatorIndex != 1)) 2911 return 25; 2912 if (Right.is(tok::lessless)) { 2913 // Breaking at a << is really cheap. 2914 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 2915 // Slightly prefer to break before the first one in log-like statements. 2916 return 2; 2917 return 1; 2918 } 2919 if (Left.ClosesTemplateDeclaration) 2920 return Style.PenaltyBreakTemplateDeclaration; 2921 if (Left.is(TT_ConditionalExpr)) 2922 return prec::Conditional; 2923 prec::Level Level = Left.getPrecedence(); 2924 if (Level == prec::Unknown) 2925 Level = Right.getPrecedence(); 2926 if (Level == prec::Assignment) 2927 return Style.PenaltyBreakAssignment; 2928 if (Level != prec::Unknown) 2929 return Level; 2930 2931 return 3; 2932 } 2933 2934 bool TokenAnnotator::spaceRequiredBeforeParens(const FormatToken &Right) const { 2935 if (Style.SpaceBeforeParens == FormatStyle::SBPO_Always) 2936 return true; 2937 if (Right.is(TT_OverloadedOperatorLParen) && 2938 Style.SpaceBeforeParensOptions.AfterOverloadedOperator) 2939 return true; 2940 if (Style.SpaceBeforeParensOptions.BeforeNonEmptyParentheses && 2941 Right.ParameterCount > 0) 2942 return true; 2943 return false; 2944 } 2945 2946 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 2947 const FormatToken &Left, 2948 const FormatToken &Right) { 2949 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 2950 return true; 2951 if (Style.isJson() && Left.is(tok::string_literal) && Right.is(tok::colon)) 2952 return false; 2953 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 2954 return true; 2955 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 2956 Left.Tok.getObjCKeywordID() == tok::objc_property) 2957 return true; 2958 if (Right.is(tok::hashhash)) 2959 return Left.is(tok::hash); 2960 if (Left.isOneOf(tok::hashhash, tok::hash)) 2961 return Right.is(tok::hash); 2962 if ((Left.is(tok::l_paren) && Right.is(tok::r_paren)) || 2963 (Left.is(tok::l_brace) && Left.isNot(BK_Block) && 2964 Right.is(tok::r_brace) && Right.isNot(BK_Block))) 2965 return Style.SpaceInEmptyParentheses; 2966 if (Style.SpacesInConditionalStatement) { 2967 if (Left.is(tok::l_paren) && Left.Previous && 2968 isKeywordWithCondition(*Left.Previous)) 2969 return true; 2970 if (Right.is(tok::r_paren) && Right.MatchingParen && 2971 Right.MatchingParen->Previous && 2972 isKeywordWithCondition(*Right.MatchingParen->Previous)) 2973 return true; 2974 } 2975 2976 // auto{x} auto(x) 2977 if (Left.is(tok::kw_auto) && Right.isOneOf(tok::l_paren, tok::l_brace)) 2978 return false; 2979 2980 // operator co_await(x) 2981 if (Right.is(tok::l_paren) && Left.is(tok::kw_co_await) && Left.Previous && 2982 Left.Previous->is(tok::kw_operator)) 2983 return false; 2984 // co_await (x), co_yield (x), co_return (x) 2985 if (Left.isOneOf(tok::kw_co_await, tok::kw_co_yield, tok::kw_co_return) && 2986 Right.isNot(tok::semi)) 2987 return true; 2988 // requires clause Concept1<T> && Concept2<T> 2989 if (Left.is(TT_ConstraintJunctions) && Right.is(tok::identifier)) 2990 return true; 2991 2992 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 2993 return (Right.is(TT_CastRParen) || 2994 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 2995 ? Style.SpacesInCStyleCastParentheses 2996 : Style.SpacesInParentheses; 2997 if (Right.isOneOf(tok::semi, tok::comma)) 2998 return false; 2999 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) { 3000 bool IsLightweightGeneric = Right.MatchingParen && 3001 Right.MatchingParen->Next && 3002 Right.MatchingParen->Next->is(tok::colon); 3003 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList; 3004 } 3005 if (Right.is(tok::less) && Left.is(tok::kw_template)) 3006 return Style.SpaceAfterTemplateKeyword; 3007 if (Left.isOneOf(tok::exclaim, tok::tilde)) 3008 return false; 3009 if (Left.is(tok::at) && 3010 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 3011 tok::numeric_constant, tok::l_paren, tok::l_brace, 3012 tok::kw_true, tok::kw_false)) 3013 return false; 3014 if (Left.is(tok::colon)) 3015 return !Left.is(TT_ObjCMethodExpr); 3016 if (Left.is(tok::coloncolon)) 3017 return false; 3018 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 3019 if (Style.Language == FormatStyle::LK_TextProto || 3020 (Style.Language == FormatStyle::LK_Proto && 3021 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 3022 // Format empty list as `<>`. 3023 if (Left.is(tok::less) && Right.is(tok::greater)) 3024 return false; 3025 return !Style.Cpp11BracedListStyle; 3026 } 3027 return false; 3028 } 3029 if (Right.is(tok::ellipsis)) 3030 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 3031 Left.Previous->is(tok::kw_case)); 3032 if (Left.is(tok::l_square) && Right.is(tok::amp)) 3033 return Style.SpacesInSquareBrackets; 3034 if (Right.is(TT_PointerOrReference)) { 3035 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 3036 if (!Left.MatchingParen) 3037 return true; 3038 FormatToken *TokenBeforeMatchingParen = 3039 Left.MatchingParen->getPreviousNonComment(); 3040 if (!TokenBeforeMatchingParen || !Left.is(TT_TypeDeclarationParen)) 3041 return true; 3042 } 3043 // Add a space if the previous token is a pointer qualifier or the closing 3044 // parenthesis of __attribute__(()) expression and the style requires spaces 3045 // after pointer qualifiers. 3046 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_After || 3047 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 3048 (Left.is(TT_AttributeParen) || Left.canBePointerOrReferenceQualifier())) 3049 return true; 3050 if (Left.Tok.isLiteral()) 3051 return true; 3052 // for (auto a = 0, b = 0; const auto & c : {1, 2, 3}) 3053 if (Left.isTypeOrIdentifier() && Right.Next && Right.Next->Next && 3054 Right.Next->Next->is(TT_RangeBasedForLoopColon)) 3055 return getTokenPointerOrReferenceAlignment(Right) != 3056 FormatStyle::PAS_Left; 3057 return ( 3058 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 3059 (getTokenPointerOrReferenceAlignment(Right) != FormatStyle::PAS_Left || 3060 (Line.IsMultiVariableDeclStmt && 3061 (Left.NestingLevel == 0 || 3062 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for))))))); 3063 } 3064 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 3065 (!Left.is(TT_PointerOrReference) || 3066 (getTokenPointerOrReferenceAlignment(Left) != FormatStyle::PAS_Right && 3067 !Line.IsMultiVariableDeclStmt))) 3068 return true; 3069 if (Left.is(TT_PointerOrReference)) { 3070 // Add a space if the next token is a pointer qualifier and the style 3071 // requires spaces before pointer qualifiers. 3072 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Before || 3073 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 3074 Right.canBePointerOrReferenceQualifier()) 3075 return true; 3076 // & 1 3077 if (Right.Tok.isLiteral()) 3078 return true; 3079 // & /* comment 3080 if (Right.is(TT_BlockComment)) 3081 return true; 3082 // foo() -> const Bar * override/final 3083 if (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 3084 !Right.is(TT_StartOfName)) 3085 return true; 3086 // & { 3087 if (Right.is(tok::l_brace) && Right.is(BK_Block)) 3088 return true; 3089 // for (auto a = 0, b = 0; const auto& c : {1, 2, 3}) 3090 if (Left.Previous && Left.Previous->isTypeOrIdentifier() && Right.Next && 3091 Right.Next->is(TT_RangeBasedForLoopColon)) 3092 return getTokenPointerOrReferenceAlignment(Left) != 3093 FormatStyle::PAS_Right; 3094 return !Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 3095 tok::l_paren) && 3096 (getTokenPointerOrReferenceAlignment(Left) != 3097 FormatStyle::PAS_Right && 3098 !Line.IsMultiVariableDeclStmt) && 3099 Left.Previous && 3100 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon, 3101 tok::l_square); 3102 } 3103 // Ensure right pointer alignment with ellipsis e.g. int *...P 3104 if (Left.is(tok::ellipsis) && Left.Previous && 3105 Left.Previous->isOneOf(tok::star, tok::amp, tok::ampamp)) 3106 return Style.PointerAlignment != FormatStyle::PAS_Right; 3107 3108 if (Right.is(tok::star) && Left.is(tok::l_paren)) 3109 return false; 3110 if (Left.is(tok::star) && Right.isOneOf(tok::star, tok::amp, tok::ampamp)) 3111 return false; 3112 if (Right.isOneOf(tok::star, tok::amp, tok::ampamp)) { 3113 const FormatToken *Previous = &Left; 3114 while (Previous && !Previous->is(tok::kw_operator)) { 3115 if (Previous->is(tok::identifier) || Previous->isSimpleTypeSpecifier()) { 3116 Previous = Previous->getPreviousNonComment(); 3117 continue; 3118 } 3119 if (Previous->is(TT_TemplateCloser) && Previous->MatchingParen) { 3120 Previous = Previous->MatchingParen->getPreviousNonComment(); 3121 continue; 3122 } 3123 if (Previous->is(tok::coloncolon)) { 3124 Previous = Previous->getPreviousNonComment(); 3125 continue; 3126 } 3127 break; 3128 } 3129 // Space between the type and the * in: 3130 // operator void*() 3131 // operator char*() 3132 // operator void const*() 3133 // operator void volatile*() 3134 // operator /*comment*/ const char*() 3135 // operator volatile /*comment*/ char*() 3136 // operator Foo*() 3137 // operator C<T>*() 3138 // operator std::Foo*() 3139 // operator C<T>::D<U>*() 3140 // dependent on PointerAlignment style. 3141 if (Previous) { 3142 if (Previous->endsSequence(tok::kw_operator)) 3143 return (Style.PointerAlignment != FormatStyle::PAS_Left); 3144 if (Previous->is(tok::kw_const) || Previous->is(tok::kw_volatile)) 3145 return (Style.PointerAlignment != FormatStyle::PAS_Left) || 3146 (Style.SpaceAroundPointerQualifiers == 3147 FormatStyle::SAPQ_After) || 3148 (Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both); 3149 } 3150 } 3151 const auto SpaceRequiredForArrayInitializerLSquare = 3152 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 3153 return Style.SpacesInContainerLiterals || 3154 ((Style.Language == FormatStyle::LK_Proto || 3155 Style.Language == FormatStyle::LK_TextProto) && 3156 !Style.Cpp11BracedListStyle && 3157 LSquareTok.endsSequence(tok::l_square, tok::colon, 3158 TT_SelectorName)); 3159 }; 3160 if (Left.is(tok::l_square)) 3161 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 3162 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 3163 (Left.isOneOf(TT_ArraySubscriptLSquare, TT_StructuredBindingLSquare, 3164 TT_LambdaLSquare) && 3165 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 3166 if (Right.is(tok::r_square)) 3167 return Right.MatchingParen && 3168 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 3169 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 3170 Style)) || 3171 (Style.SpacesInSquareBrackets && 3172 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 3173 TT_StructuredBindingLSquare, 3174 TT_LambdaLSquare)) || 3175 Right.MatchingParen->is(TT_AttributeParen)); 3176 if (Right.is(tok::l_square) && 3177 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 3178 TT_DesignatedInitializerLSquare, 3179 TT_StructuredBindingLSquare, TT_AttributeSquare) && 3180 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral) && 3181 !(!Left.is(tok::r_square) && Style.SpaceBeforeSquareBrackets && 3182 Right.is(TT_ArraySubscriptLSquare))) 3183 return false; 3184 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 3185 return !Left.Children.empty(); // No spaces in "{}". 3186 if ((Left.is(tok::l_brace) && Left.isNot(BK_Block)) || 3187 (Right.is(tok::r_brace) && Right.MatchingParen && 3188 Right.MatchingParen->isNot(BK_Block))) 3189 return Style.Cpp11BracedListStyle ? Style.SpacesInParentheses : true; 3190 if (Left.is(TT_BlockComment)) 3191 // No whitespace in x(/*foo=*/1), except for JavaScript. 3192 return Style.isJavaScript() || !Left.TokenText.endswith("=*/"); 3193 3194 // Space between template and attribute. 3195 // e.g. template <typename T> [[nodiscard]] ... 3196 if (Left.is(TT_TemplateCloser) && Right.is(TT_AttributeSquare)) 3197 return true; 3198 // Space before parentheses common for all languages 3199 if (Right.is(tok::l_paren)) { 3200 if (Left.is(TT_TemplateCloser) && Right.isNot(TT_FunctionTypeLParen)) 3201 return spaceRequiredBeforeParens(Right); 3202 if (Left.is(tok::kw_requires)) 3203 return spaceRequiredBeforeParens(Right); 3204 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 3205 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) 3206 return true; 3207 if (Left.is(TT_ForEachMacro)) 3208 return (Style.SpaceBeforeParensOptions.AfterForeachMacros || 3209 spaceRequiredBeforeParens(Right)); 3210 if (Left.is(TT_IfMacro)) 3211 return (Style.SpaceBeforeParensOptions.AfterIfMacros || 3212 spaceRequiredBeforeParens(Right)); 3213 if (Line.Type == LT_ObjCDecl) 3214 return true; 3215 if (Left.is(tok::semi)) 3216 return true; 3217 if (Left.isOneOf(tok::pp_elif, tok::kw_for, tok::kw_while, tok::kw_switch, 3218 tok::kw_case, TT_ForEachMacro, TT_ObjCForIn)) 3219 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3220 spaceRequiredBeforeParens(Right); 3221 if (Left.isIf(Line.Type != LT_PreprocessorDirective)) 3222 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3223 spaceRequiredBeforeParens(Right); 3224 3225 // TODO add Operator overloading specific Options to 3226 // SpaceBeforeParensOptions 3227 if (Right.is(TT_OverloadedOperatorLParen)) 3228 return spaceRequiredBeforeParens(Right); 3229 // Function declaration or definition 3230 if (Line.MightBeFunctionDecl && (Left.is(TT_FunctionDeclarationName))) { 3231 if (Line.mightBeFunctionDefinition()) 3232 return Style.SpaceBeforeParensOptions.AfterFunctionDefinitionName || 3233 spaceRequiredBeforeParens(Right); 3234 else 3235 return Style.SpaceBeforeParensOptions.AfterFunctionDeclarationName || 3236 spaceRequiredBeforeParens(Right); 3237 } 3238 // Lambda 3239 if (Line.Type != LT_PreprocessorDirective && Left.is(tok::r_square) && 3240 Left.MatchingParen && Left.MatchingParen->is(TT_LambdaLSquare)) 3241 return Style.SpaceBeforeParensOptions.AfterFunctionDefinitionName || 3242 spaceRequiredBeforeParens(Right); 3243 if (!Left.Previous || Left.Previous->isNot(tok::period)) { 3244 if (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch)) 3245 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3246 spaceRequiredBeforeParens(Right); 3247 if (Left.isOneOf(tok::kw_new, tok::kw_delete)) 3248 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never || 3249 spaceRequiredBeforeParens(Right); 3250 } 3251 if (Line.Type != LT_PreprocessorDirective && 3252 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 3253 Left.is(tok::r_paren) || Left.isSimpleTypeSpecifier())) 3254 return spaceRequiredBeforeParens(Right); 3255 return false; 3256 } 3257 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 3258 return false; 3259 if (Right.is(TT_UnaryOperator)) 3260 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 3261 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 3262 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 3263 tok::r_paren) || 3264 Left.isSimpleTypeSpecifier()) && 3265 Right.is(tok::l_brace) && Right.getNextNonComment() && 3266 Right.isNot(BK_Block)) 3267 return false; 3268 if (Left.is(tok::period) || Right.is(tok::period)) 3269 return false; 3270 // u#str, U#str, L#str, u8#str 3271 // uR#str, UR#str, LR#str, u8R#str 3272 if (Right.is(tok::hash) && Left.is(tok::identifier) && 3273 (Left.TokenText == "L" || Left.TokenText == "u" || 3274 Left.TokenText == "U" || Left.TokenText == "u8" || 3275 Left.TokenText == "LR" || Left.TokenText == "uR" || 3276 Left.TokenText == "UR" || Left.TokenText == "u8R")) 3277 return false; 3278 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 3279 Left.MatchingParen->Previous && 3280 (Left.MatchingParen->Previous->is(tok::period) || 3281 Left.MatchingParen->Previous->is(tok::coloncolon))) 3282 // Java call to generic function with explicit type: 3283 // A.<B<C<...>>>DoSomething(); 3284 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference. 3285 return false; 3286 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 3287 return false; 3288 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at)) 3289 // Objective-C dictionary literal -> no space after opening brace. 3290 return false; 3291 if (Right.is(tok::r_brace) && Right.MatchingParen && 3292 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at)) 3293 // Objective-C dictionary literal -> no space before closing brace. 3294 return false; 3295 if (Right.getType() == TT_TrailingAnnotation && 3296 Right.isOneOf(tok::amp, tok::ampamp) && 3297 Left.isOneOf(tok::kw_const, tok::kw_volatile) && 3298 (!Right.Next || Right.Next->is(tok::semi))) 3299 // Match const and volatile ref-qualifiers without any additional 3300 // qualifiers such as 3301 // void Fn() const &; 3302 return getTokenReferenceAlignment(Right) != FormatStyle::PAS_Left; 3303 3304 return true; 3305 } 3306 3307 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 3308 const FormatToken &Right) { 3309 const FormatToken &Left = *Right.Previous; 3310 3311 // If the token is finalized don't touch it (as it could be in a 3312 // clang-format-off section). 3313 if (Left.Finalized) 3314 return Right.hasWhitespaceBefore(); 3315 3316 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 3317 return true; // Never ever merge two identifiers. 3318 3319 // Leave a space between * and /* to avoid C4138 `comment end` found outside 3320 // of comment. 3321 if (Left.is(tok::star) && Right.is(tok::comment)) 3322 return true; 3323 3324 if (Style.isCpp()) { 3325 // Space between import <iostream>. 3326 // or import .....; 3327 if (Left.is(Keywords.kw_import) && Right.isOneOf(tok::less, tok::ellipsis)) 3328 return true; 3329 // Space between `module :` and `import :`. 3330 if (Left.isOneOf(Keywords.kw_module, Keywords.kw_import) && 3331 Right.is(TT_ModulePartitionColon)) 3332 return true; 3333 // No space between import foo:bar but keep a space between import :bar; 3334 if (Left.is(tok::identifier) && Right.is(TT_ModulePartitionColon)) 3335 return false; 3336 // No space between :bar; 3337 if (Left.is(TT_ModulePartitionColon) && 3338 Right.isOneOf(tok::identifier, tok::kw_private)) 3339 return false; 3340 if (Left.is(tok::ellipsis) && Right.is(tok::identifier) && 3341 Line.First->is(Keywords.kw_import)) 3342 return false; 3343 // Space in __attribute__((attr)) ::type. 3344 if (Left.is(TT_AttributeParen) && Right.is(tok::coloncolon)) 3345 return true; 3346 3347 if (Left.is(tok::kw_operator)) 3348 return Right.is(tok::coloncolon); 3349 if (Right.is(tok::l_brace) && Right.is(BK_BracedInit) && 3350 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList) 3351 return true; 3352 if (Left.is(tok::less) && Left.is(TT_OverloadedOperator) && 3353 Right.is(TT_TemplateOpener)) 3354 return true; 3355 } else if (Style.Language == FormatStyle::LK_Proto || 3356 Style.Language == FormatStyle::LK_TextProto) { 3357 if (Right.is(tok::period) && 3358 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 3359 Keywords.kw_repeated, Keywords.kw_extend)) 3360 return true; 3361 if (Right.is(tok::l_paren) && 3362 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 3363 return true; 3364 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 3365 return true; 3366 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 3367 if (Left.is(tok::slash) || Right.is(tok::slash)) 3368 return false; 3369 if (Left.MatchingParen && 3370 Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 3371 Right.isOneOf(tok::l_brace, tok::less)) 3372 return !Style.Cpp11BracedListStyle; 3373 // A percent is probably part of a formatting specification, such as %lld. 3374 if (Left.is(tok::percent)) 3375 return false; 3376 // Preserve the existence of a space before a percent for cases like 0x%04x 3377 // and "%d %d" 3378 if (Left.is(tok::numeric_constant) && Right.is(tok::percent)) 3379 return Right.hasWhitespaceBefore(); 3380 } else if (Style.isJson()) { 3381 if (Right.is(tok::colon)) 3382 return false; 3383 } else if (Style.isCSharp()) { 3384 // Require spaces around '{' and before '}' unless they appear in 3385 // interpolated strings. Interpolated strings are merged into a single token 3386 // so cannot have spaces inserted by this function. 3387 3388 // No space between 'this' and '[' 3389 if (Left.is(tok::kw_this) && Right.is(tok::l_square)) 3390 return false; 3391 3392 // No space between 'new' and '(' 3393 if (Left.is(tok::kw_new) && Right.is(tok::l_paren)) 3394 return false; 3395 3396 // Space before { (including space within '{ {'). 3397 if (Right.is(tok::l_brace)) 3398 return true; 3399 3400 // Spaces inside braces. 3401 if (Left.is(tok::l_brace) && Right.isNot(tok::r_brace)) 3402 return true; 3403 3404 if (Left.isNot(tok::l_brace) && Right.is(tok::r_brace)) 3405 return true; 3406 3407 // Spaces around '=>'. 3408 if (Left.is(TT_FatArrow) || Right.is(TT_FatArrow)) 3409 return true; 3410 3411 // No spaces around attribute target colons 3412 if (Left.is(TT_AttributeColon) || Right.is(TT_AttributeColon)) 3413 return false; 3414 3415 // space between type and variable e.g. Dictionary<string,string> foo; 3416 if (Left.is(TT_TemplateCloser) && Right.is(TT_StartOfName)) 3417 return true; 3418 3419 // spaces inside square brackets. 3420 if (Left.is(tok::l_square) || Right.is(tok::r_square)) 3421 return Style.SpacesInSquareBrackets; 3422 3423 // No space before ? in nullable types. 3424 if (Right.is(TT_CSharpNullable)) 3425 return false; 3426 3427 // No space before null forgiving '!'. 3428 if (Right.is(TT_NonNullAssertion)) 3429 return false; 3430 3431 // No space between consecutive commas '[,,]'. 3432 if (Left.is(tok::comma) && Right.is(tok::comma)) 3433 return false; 3434 3435 // space after var in `var (key, value)` 3436 if (Left.is(Keywords.kw_var) && Right.is(tok::l_paren)) 3437 return true; 3438 3439 // space between keywords and paren e.g. "using (" 3440 if (Right.is(tok::l_paren)) 3441 if (Left.isOneOf(tok::kw_using, Keywords.kw_async, Keywords.kw_when, 3442 Keywords.kw_lock)) 3443 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3444 spaceRequiredBeforeParens(Right); 3445 3446 // space between method modifier and opening parenthesis of a tuple return 3447 // type 3448 if (Left.isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 3449 tok::kw_virtual, tok::kw_extern, tok::kw_static, 3450 Keywords.kw_internal, Keywords.kw_abstract, 3451 Keywords.kw_sealed, Keywords.kw_override, 3452 Keywords.kw_async, Keywords.kw_unsafe) && 3453 Right.is(tok::l_paren)) 3454 return true; 3455 } else if (Style.isJavaScript()) { 3456 if (Left.is(TT_FatArrow)) 3457 return true; 3458 // for await ( ... 3459 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 3460 Left.Previous->is(tok::kw_for)) 3461 return true; 3462 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 3463 Right.MatchingParen) { 3464 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 3465 // An async arrow function, for example: `x = async () => foo();`, 3466 // as opposed to calling a function called async: `x = async();` 3467 if (Next && Next->is(TT_FatArrow)) 3468 return true; 3469 } 3470 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 3471 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 3472 return false; 3473 // In tagged template literals ("html`bar baz`"), there is no space between 3474 // the tag identifier and the template string. 3475 if (Keywords.IsJavaScriptIdentifier(Left, 3476 /* AcceptIdentifierName= */ false) && 3477 Right.is(TT_TemplateString)) 3478 return false; 3479 if (Right.is(tok::star) && 3480 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 3481 return false; 3482 if (Right.isOneOf(tok::l_brace, tok::l_square) && 3483 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 3484 Keywords.kw_extends, Keywords.kw_implements)) 3485 return true; 3486 if (Right.is(tok::l_paren)) { 3487 // JS methods can use some keywords as names (e.g. `delete()`). 3488 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 3489 return false; 3490 // Valid JS method names can include keywords, e.g. `foo.delete()` or 3491 // `bar.instanceof()`. Recognize call positions by preceding period. 3492 if (Left.Previous && Left.Previous->is(tok::period) && 3493 Left.Tok.getIdentifierInfo()) 3494 return false; 3495 // Additional unary JavaScript operators that need a space after. 3496 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 3497 tok::kw_void)) 3498 return true; 3499 } 3500 // `foo as const;` casts into a const type. 3501 if (Left.endsSequence(tok::kw_const, Keywords.kw_as)) { 3502 return false; 3503 } 3504 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 3505 tok::kw_const) || 3506 // "of" is only a keyword if it appears after another identifier 3507 // (e.g. as "const x of y" in a for loop), or after a destructuring 3508 // operation (const [x, y] of z, const {a, b} of c). 3509 (Left.is(Keywords.kw_of) && Left.Previous && 3510 (Left.Previous->Tok.is(tok::identifier) || 3511 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 3512 (!Left.Previous || !Left.Previous->is(tok::period))) 3513 return true; 3514 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 3515 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 3516 return false; 3517 if (Left.is(Keywords.kw_as) && 3518 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 3519 return true; 3520 if (Left.is(tok::kw_default) && Left.Previous && 3521 Left.Previous->is(tok::kw_export)) 3522 return true; 3523 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 3524 return true; 3525 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 3526 return false; 3527 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 3528 return false; 3529 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 3530 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 3531 return false; 3532 if (Left.is(tok::ellipsis)) 3533 return false; 3534 if (Left.is(TT_TemplateCloser) && 3535 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 3536 Keywords.kw_implements, Keywords.kw_extends)) 3537 // Type assertions ('<type>expr') are not followed by whitespace. Other 3538 // locations that should have whitespace following are identified by the 3539 // above set of follower tokens. 3540 return false; 3541 if (Right.is(TT_NonNullAssertion)) 3542 return false; 3543 if (Left.is(TT_NonNullAssertion) && 3544 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 3545 return true; // "x! as string", "x! in y" 3546 } else if (Style.Language == FormatStyle::LK_Java) { 3547 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 3548 return true; 3549 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 3550 return Style.SpaceBeforeParensOptions.AfterControlStatements || 3551 spaceRequiredBeforeParens(Right); 3552 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 3553 tok::kw_protected) || 3554 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 3555 Keywords.kw_native)) && 3556 Right.is(TT_TemplateOpener)) 3557 return true; 3558 } 3559 if (Left.is(TT_ImplicitStringLiteral)) 3560 return Right.hasWhitespaceBefore(); 3561 if (Line.Type == LT_ObjCMethodDecl) { 3562 if (Left.is(TT_ObjCMethodSpecifier)) 3563 return true; 3564 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right)) 3565 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a 3566 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class 3567 // method declaration. 3568 return false; 3569 } 3570 if (Line.Type == LT_ObjCProperty && 3571 (Right.is(tok::equal) || Left.is(tok::equal))) 3572 return false; 3573 3574 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 3575 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 3576 return true; 3577 if (Left.is(tok::comma) && !Right.is(TT_OverloadedOperatorLParen)) 3578 return true; 3579 if (Right.is(tok::comma)) 3580 return false; 3581 if (Right.is(TT_ObjCBlockLParen)) 3582 return true; 3583 if (Right.is(TT_CtorInitializerColon)) 3584 return Style.SpaceBeforeCtorInitializerColon; 3585 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 3586 return false; 3587 if (Right.is(TT_RangeBasedForLoopColon) && 3588 !Style.SpaceBeforeRangeBasedForLoopColon) 3589 return false; 3590 if (Left.is(TT_BitFieldColon)) 3591 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 3592 Style.BitFieldColonSpacing == FormatStyle::BFCS_After; 3593 if (Right.is(tok::colon)) { 3594 if (Line.First->isOneOf(tok::kw_default, tok::kw_case)) 3595 return Style.SpaceBeforeCaseColon; 3596 if (!Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 3597 return false; 3598 if (Right.is(TT_ObjCMethodExpr)) 3599 return false; 3600 if (Left.is(tok::question)) 3601 return false; 3602 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 3603 return false; 3604 if (Right.is(TT_DictLiteral)) 3605 return Style.SpacesInContainerLiterals; 3606 if (Right.is(TT_AttributeColon)) 3607 return false; 3608 if (Right.is(TT_CSharpNamedArgumentColon)) 3609 return false; 3610 if (Right.is(TT_BitFieldColon)) 3611 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 3612 Style.BitFieldColonSpacing == FormatStyle::BFCS_Before; 3613 return true; 3614 } 3615 // Do not merge "- -" into "--". 3616 if ((Left.isOneOf(tok::minus, tok::minusminus) && 3617 Right.isOneOf(tok::minus, tok::minusminus)) || 3618 (Left.isOneOf(tok::plus, tok::plusplus) && 3619 Right.isOneOf(tok::plus, tok::plusplus))) 3620 return true; 3621 if (Left.is(TT_UnaryOperator)) { 3622 if (!Right.is(tok::l_paren)) { 3623 // The alternative operators for ~ and ! are "compl" and "not". 3624 // If they are used instead, we do not want to combine them with 3625 // the token to the right, unless that is a left paren. 3626 if (Left.is(tok::exclaim) && Left.TokenText == "not") 3627 return true; 3628 if (Left.is(tok::tilde) && Left.TokenText == "compl") 3629 return true; 3630 // Lambda captures allow for a lone &, so "&]" needs to be properly 3631 // handled. 3632 if (Left.is(tok::amp) && Right.is(tok::r_square)) 3633 return Style.SpacesInSquareBrackets; 3634 } 3635 return (Style.SpaceAfterLogicalNot && Left.is(tok::exclaim)) || 3636 Right.is(TT_BinaryOperator); 3637 } 3638 3639 // If the next token is a binary operator or a selector name, we have 3640 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 3641 if (Left.is(TT_CastRParen)) 3642 return Style.SpaceAfterCStyleCast || 3643 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 3644 3645 auto ShouldAddSpacesInAngles = [this, &Right]() { 3646 if (this->Style.SpacesInAngles == FormatStyle::SIAS_Always) 3647 return true; 3648 if (this->Style.SpacesInAngles == FormatStyle::SIAS_Leave) 3649 return Right.hasWhitespaceBefore(); 3650 return false; 3651 }; 3652 3653 if (Left.is(tok::greater) && Right.is(tok::greater)) { 3654 if (Style.Language == FormatStyle::LK_TextProto || 3655 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 3656 return !Style.Cpp11BracedListStyle; 3657 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 3658 ((Style.Standard < FormatStyle::LS_Cpp11) || 3659 ShouldAddSpacesInAngles()); 3660 } 3661 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 3662 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 3663 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 3664 return false; 3665 if (!Style.SpaceBeforeAssignmentOperators && Left.isNot(TT_TemplateCloser) && 3666 Right.getPrecedence() == prec::Assignment) 3667 return false; 3668 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) && 3669 (Left.is(tok::identifier) || Left.is(tok::kw_this))) 3670 return false; 3671 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 3672 // Generally don't remove existing spaces between an identifier and "::". 3673 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 3674 // this turns out to be too lenient, add analysis of the identifier itself. 3675 return Right.hasWhitespaceBefore(); 3676 if (Right.is(tok::coloncolon) && 3677 !Left.isOneOf(tok::l_brace, tok::comment, tok::l_paren)) 3678 // Put a space between < and :: in vector< ::std::string > 3679 return (Left.is(TT_TemplateOpener) && 3680 ((Style.Standard < FormatStyle::LS_Cpp11) || 3681 ShouldAddSpacesInAngles())) || 3682 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 3683 tok::kw___super, TT_TemplateOpener, 3684 TT_TemplateCloser)) || 3685 (Left.is(tok::l_paren) && Style.SpacesInParentheses); 3686 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 3687 return ShouldAddSpacesInAngles(); 3688 // Space before TT_StructuredBindingLSquare. 3689 if (Right.is(TT_StructuredBindingLSquare)) 3690 return !Left.isOneOf(tok::amp, tok::ampamp) || 3691 getTokenReferenceAlignment(Left) != FormatStyle::PAS_Right; 3692 // Space before & or && following a TT_StructuredBindingLSquare. 3693 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 3694 Right.isOneOf(tok::amp, tok::ampamp)) 3695 return getTokenReferenceAlignment(Right) != FormatStyle::PAS_Left; 3696 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 3697 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 3698 !Right.is(tok::r_paren))) 3699 return true; 3700 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 3701 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 3702 return false; 3703 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 3704 Line.startsWith(tok::hash)) 3705 return true; 3706 if (Right.is(TT_TrailingUnaryOperator)) 3707 return false; 3708 if (Left.is(TT_RegexLiteral)) 3709 return false; 3710 return spaceRequiredBetween(Line, Left, Right); 3711 } 3712 3713 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 3714 static bool isAllmanBrace(const FormatToken &Tok) { 3715 return Tok.is(tok::l_brace) && Tok.is(BK_Block) && 3716 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_LambdaLBrace, TT_DictLiteral); 3717 } 3718 3719 // Returns 'true' if 'Tok' is a function argument. 3720 static bool IsFunctionArgument(const FormatToken &Tok) { 3721 return Tok.MatchingParen && Tok.MatchingParen->Next && 3722 Tok.MatchingParen->Next->isOneOf(tok::comma, tok::r_paren); 3723 } 3724 3725 static bool 3726 isItAnEmptyLambdaAllowed(const FormatToken &Tok, 3727 FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3728 return Tok.Children.empty() && ShortLambdaOption != FormatStyle::SLS_None; 3729 } 3730 3731 static bool isAllmanLambdaBrace(const FormatToken &Tok) { 3732 return (Tok.is(tok::l_brace) && Tok.is(BK_Block) && 3733 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral)); 3734 } 3735 3736 // Returns the first token on the line that is not a comment. 3737 static const FormatToken *getFirstNonComment(const AnnotatedLine &Line) { 3738 const FormatToken *Next = Line.First; 3739 if (!Next) 3740 return Next; 3741 if (Next->is(tok::comment)) 3742 Next = Next->getNextNonComment(); 3743 return Next; 3744 } 3745 3746 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 3747 const FormatToken &Right) { 3748 const FormatToken &Left = *Right.Previous; 3749 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 3750 return true; 3751 3752 if (Style.isCSharp()) { 3753 if (Left.is(TT_FatArrow) && Right.is(tok::l_brace) && 3754 Style.BraceWrapping.AfterFunction) 3755 return true; 3756 if (Right.is(TT_CSharpNamedArgumentColon) || 3757 Left.is(TT_CSharpNamedArgumentColon)) 3758 return false; 3759 if (Right.is(TT_CSharpGenericTypeConstraint)) 3760 return true; 3761 if (Right.Next && Right.Next->is(TT_FatArrow) && 3762 (Right.is(tok::numeric_constant) || 3763 (Right.is(tok::identifier) && Right.TokenText == "_"))) 3764 return true; 3765 3766 // Break after C# [...] and before public/protected/private/internal. 3767 if (Left.is(TT_AttributeSquare) && Left.is(tok::r_square) && 3768 (Right.isAccessSpecifier(/*ColonRequired=*/false) || 3769 Right.is(Keywords.kw_internal))) 3770 return true; 3771 // Break between ] and [ but only when there are really 2 attributes. 3772 if (Left.is(TT_AttributeSquare) && Right.is(TT_AttributeSquare) && 3773 Left.is(tok::r_square) && Right.is(tok::l_square)) 3774 return true; 3775 3776 } else if (Style.isJavaScript()) { 3777 // FIXME: This might apply to other languages and token kinds. 3778 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 3779 Left.Previous->is(tok::string_literal)) 3780 return true; 3781 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 3782 Left.Previous && Left.Previous->is(tok::equal) && 3783 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 3784 tok::kw_const) && 3785 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 3786 // above. 3787 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 3788 // Object literals on the top level of a file are treated as "enum-style". 3789 // Each key/value pair is put on a separate line, instead of bin-packing. 3790 return true; 3791 if (Left.is(tok::l_brace) && Line.Level == 0 && 3792 (Line.startsWith(tok::kw_enum) || 3793 Line.startsWith(tok::kw_const, tok::kw_enum) || 3794 Line.startsWith(tok::kw_export, tok::kw_enum) || 3795 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 3796 // JavaScript top-level enum key/value pairs are put on separate lines 3797 // instead of bin-packing. 3798 return true; 3799 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && Left.Previous && 3800 Left.Previous->is(TT_FatArrow)) { 3801 // JS arrow function (=> {...}). 3802 switch (Style.AllowShortLambdasOnASingleLine) { 3803 case FormatStyle::SLS_All: 3804 return false; 3805 case FormatStyle::SLS_None: 3806 return true; 3807 case FormatStyle::SLS_Empty: 3808 return !Left.Children.empty(); 3809 case FormatStyle::SLS_Inline: 3810 // allow one-lining inline (e.g. in function call args) and empty arrow 3811 // functions. 3812 return (Left.NestingLevel == 0 && Line.Level == 0) && 3813 !Left.Children.empty(); 3814 } 3815 llvm_unreachable("Unknown FormatStyle::ShortLambdaStyle enum"); 3816 } 3817 3818 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 3819 !Left.Children.empty()) 3820 // Support AllowShortFunctionsOnASingleLine for JavaScript. 3821 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 3822 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 3823 (Left.NestingLevel == 0 && Line.Level == 0 && 3824 Style.AllowShortFunctionsOnASingleLine & 3825 FormatStyle::SFS_InlineOnly); 3826 } else if (Style.Language == FormatStyle::LK_Java) { 3827 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 3828 Right.Next->is(tok::string_literal)) 3829 return true; 3830 } else if (Style.Language == FormatStyle::LK_Cpp || 3831 Style.Language == FormatStyle::LK_ObjC || 3832 Style.Language == FormatStyle::LK_Proto || 3833 Style.Language == FormatStyle::LK_TableGen || 3834 Style.Language == FormatStyle::LK_TextProto) { 3835 if (Left.isStringLiteral() && Right.isStringLiteral()) 3836 return true; 3837 } 3838 3839 // Basic JSON newline processing. 3840 if (Style.isJson()) { 3841 // Always break after a JSON record opener. 3842 // { 3843 // } 3844 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace)) 3845 return true; 3846 // Always break after a JSON array opener. 3847 // [ 3848 // ] 3849 if (Left.is(TT_ArrayInitializerLSquare) && Left.is(tok::l_square) && 3850 !Right.is(tok::r_square)) 3851 return true; 3852 // Always break after successive entries. 3853 // 1, 3854 // 2 3855 if (Left.is(tok::comma)) 3856 return true; 3857 } 3858 3859 // If the last token before a '}', ']', or ')' is a comma or a trailing 3860 // comment, the intention is to insert a line break after it in order to make 3861 // shuffling around entries easier. Import statements, especially in 3862 // JavaScript, can be an exception to this rule. 3863 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 3864 const FormatToken *BeforeClosingBrace = nullptr; 3865 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 3866 (Style.isJavaScript() && Left.is(tok::l_paren))) && 3867 Left.isNot(BK_Block) && Left.MatchingParen) 3868 BeforeClosingBrace = Left.MatchingParen->Previous; 3869 else if (Right.MatchingParen && 3870 (Right.MatchingParen->isOneOf(tok::l_brace, 3871 TT_ArrayInitializerLSquare) || 3872 (Style.isJavaScript() && Right.MatchingParen->is(tok::l_paren)))) 3873 BeforeClosingBrace = &Left; 3874 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 3875 BeforeClosingBrace->isTrailingComment())) 3876 return true; 3877 } 3878 3879 if (Right.is(tok::comment)) 3880 return Left.isNot(BK_BracedInit) && Left.isNot(TT_CtorInitializerColon) && 3881 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 3882 if (Left.isTrailingComment()) 3883 return true; 3884 if (Left.IsUnterminatedLiteral) 3885 return true; 3886 if (Right.is(tok::lessless) && Right.Next && Left.is(tok::string_literal) && 3887 Right.Next->is(tok::string_literal)) 3888 return true; 3889 // Can break after template<> declaration 3890 if (Left.ClosesTemplateDeclaration && Left.MatchingParen && 3891 Left.MatchingParen->NestingLevel == 0) { 3892 // Put concepts on the next line e.g. 3893 // template<typename T> 3894 // concept ... 3895 if (Right.is(tok::kw_concept)) 3896 return Style.BreakBeforeConceptDeclarations; 3897 return (Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes); 3898 } 3899 if (Style.PackConstructorInitializers == FormatStyle::PCIS_Never) { 3900 if (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeColon && 3901 (Left.is(TT_CtorInitializerComma) || Right.is(TT_CtorInitializerColon))) 3902 return true; 3903 3904 if (Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon && 3905 Left.isOneOf(TT_CtorInitializerColon, TT_CtorInitializerComma)) 3906 return true; 3907 } 3908 if (Style.PackConstructorInitializers < FormatStyle::PCIS_CurrentLine && 3909 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3910 Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) 3911 return true; 3912 // Break only if we have multiple inheritance. 3913 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 3914 Right.is(TT_InheritanceComma)) 3915 return true; 3916 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterComma && 3917 Left.is(TT_InheritanceComma)) 3918 return true; 3919 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 3920 // Multiline raw string literals are special wrt. line breaks. The author 3921 // has made a deliberate choice and might have aligned the contents of the 3922 // string literal accordingly. Thus, we try keep existing line breaks. 3923 return Right.IsMultiline && Right.NewlinesBefore > 0; 3924 if ((Left.is(tok::l_brace) || (Left.is(tok::less) && Left.Previous && 3925 Left.Previous->is(tok::equal))) && 3926 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 3927 // Don't put enums or option definitions onto single lines in protocol 3928 // buffers. 3929 return true; 3930 } 3931 if (Right.is(TT_InlineASMBrace)) 3932 return Right.HasUnescapedNewline; 3933 3934 if (isAllmanBrace(Left) || isAllmanBrace(Right)) { 3935 auto FirstNonComment = getFirstNonComment(Line); 3936 bool AccessSpecifier = 3937 FirstNonComment && 3938 FirstNonComment->isOneOf(Keywords.kw_internal, tok::kw_public, 3939 tok::kw_private, tok::kw_protected); 3940 3941 if (Style.BraceWrapping.AfterEnum) { 3942 if (Line.startsWith(tok::kw_enum) || 3943 Line.startsWith(tok::kw_typedef, tok::kw_enum)) 3944 return true; 3945 // Ensure BraceWrapping for `public enum A {`. 3946 if (AccessSpecifier && FirstNonComment->Next && 3947 FirstNonComment->Next->is(tok::kw_enum)) 3948 return true; 3949 } 3950 3951 // Ensure BraceWrapping for `public interface A {`. 3952 if (Style.BraceWrapping.AfterClass && 3953 ((AccessSpecifier && FirstNonComment->Next && 3954 FirstNonComment->Next->is(Keywords.kw_interface)) || 3955 Line.startsWith(Keywords.kw_interface))) 3956 return true; 3957 3958 return (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 3959 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 3960 } 3961 3962 if (Left.is(TT_ObjCBlockLBrace) && 3963 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 3964 return true; 3965 3966 // Ensure wrapping after __attribute__((XX)) and @interface etc. 3967 if (Left.is(TT_AttributeParen) && Right.is(TT_ObjCDecl)) 3968 return true; 3969 3970 if (Left.is(TT_LambdaLBrace)) { 3971 if (IsFunctionArgument(Left) && 3972 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline) 3973 return false; 3974 3975 if (Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_None || 3976 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline || 3977 (!Left.Children.empty() && 3978 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Empty)) 3979 return true; 3980 } 3981 3982 if (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace) && 3983 Left.isOneOf(tok::star, tok::amp, tok::ampamp, TT_TemplateCloser)) { 3984 return true; 3985 } 3986 3987 // Put multiple Java annotation on a new line. 3988 if ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) && 3989 Left.is(TT_LeadingJavaAnnotation) && 3990 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 3991 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 3992 return true; 3993 3994 if (Right.is(TT_ProtoExtensionLSquare)) 3995 return true; 3996 3997 // In text proto instances if a submessage contains at least 2 entries and at 3998 // least one of them is a submessage, like A { ... B { ... } ... }, 3999 // put all of the entries of A on separate lines by forcing the selector of 4000 // the submessage B to be put on a newline. 4001 // 4002 // Example: these can stay on one line: 4003 // a { scalar_1: 1 scalar_2: 2 } 4004 // a { b { key: value } } 4005 // 4006 // and these entries need to be on a new line even if putting them all in one 4007 // line is under the column limit: 4008 // a { 4009 // scalar: 1 4010 // b { key: value } 4011 // } 4012 // 4013 // We enforce this by breaking before a submessage field that has previous 4014 // siblings, *and* breaking before a field that follows a submessage field. 4015 // 4016 // Be careful to exclude the case [proto.ext] { ... } since the `]` is 4017 // the TT_SelectorName there, but we don't want to break inside the brackets. 4018 // 4019 // Another edge case is @submessage { key: value }, which is a common 4020 // substitution placeholder. In this case we want to keep `@` and `submessage` 4021 // together. 4022 // 4023 // We ensure elsewhere that extensions are always on their own line. 4024 if ((Style.Language == FormatStyle::LK_Proto || 4025 Style.Language == FormatStyle::LK_TextProto) && 4026 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) { 4027 // Keep `@submessage` together in: 4028 // @submessage { key: value } 4029 if (Left.is(tok::at)) 4030 return false; 4031 // Look for the scope opener after selector in cases like: 4032 // selector { ... 4033 // selector: { ... 4034 // selector: @base { ... 4035 FormatToken *LBrace = Right.Next; 4036 if (LBrace && LBrace->is(tok::colon)) { 4037 LBrace = LBrace->Next; 4038 if (LBrace && LBrace->is(tok::at)) { 4039 LBrace = LBrace->Next; 4040 if (LBrace) 4041 LBrace = LBrace->Next; 4042 } 4043 } 4044 if (LBrace && 4045 // The scope opener is one of {, [, <: 4046 // selector { ... } 4047 // selector [ ... ] 4048 // selector < ... > 4049 // 4050 // In case of selector { ... }, the l_brace is TT_DictLiteral. 4051 // In case of an empty selector {}, the l_brace is not TT_DictLiteral, 4052 // so we check for immediately following r_brace. 4053 ((LBrace->is(tok::l_brace) && 4054 (LBrace->is(TT_DictLiteral) || 4055 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) || 4056 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) { 4057 // If Left.ParameterCount is 0, then this submessage entry is not the 4058 // first in its parent submessage, and we want to break before this entry. 4059 // If Left.ParameterCount is greater than 0, then its parent submessage 4060 // might contain 1 or more entries and we want to break before this entry 4061 // if it contains at least 2 entries. We deal with this case later by 4062 // detecting and breaking before the next entry in the parent submessage. 4063 if (Left.ParameterCount == 0) 4064 return true; 4065 // However, if this submessage is the first entry in its parent 4066 // submessage, Left.ParameterCount might be 1 in some cases. 4067 // We deal with this case later by detecting an entry 4068 // following a closing paren of this submessage. 4069 } 4070 4071 // If this is an entry immediately following a submessage, it will be 4072 // preceded by a closing paren of that submessage, like in: 4073 // left---. .---right 4074 // v v 4075 // sub: { ... } key: value 4076 // If there was a comment between `}` an `key` above, then `key` would be 4077 // put on a new line anyways. 4078 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square)) 4079 return true; 4080 } 4081 4082 // Deal with lambda arguments in C++ - we want consistent line breaks whether 4083 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced 4084 // as aggressive line breaks are placed when the lambda is not the last arg. 4085 if ((Style.Language == FormatStyle::LK_Cpp || 4086 Style.Language == FormatStyle::LK_ObjC) && 4087 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 && 4088 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) { 4089 // Multiple lambdas in the same function call force line breaks. 4090 if (Left.BlockParameterCount > 1) 4091 return true; 4092 4093 // A lambda followed by another arg forces a line break. 4094 if (!Left.Role) 4095 return false; 4096 auto Comma = Left.Role->lastComma(); 4097 if (!Comma) 4098 return false; 4099 auto Next = Comma->getNextNonComment(); 4100 if (!Next) 4101 return false; 4102 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret)) 4103 return true; 4104 } 4105 4106 return false; 4107 } 4108 4109 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 4110 const FormatToken &Right) { 4111 const FormatToken &Left = *Right.Previous; 4112 // Language-specific stuff. 4113 if (Style.isCSharp()) { 4114 if (Left.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon) || 4115 Right.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon)) 4116 return false; 4117 // Only break after commas for generic type constraints. 4118 if (Line.First->is(TT_CSharpGenericTypeConstraint)) 4119 return Left.is(TT_CSharpGenericTypeConstraintComma); 4120 // Keep nullable operators attached to their identifiers. 4121 if (Right.is(TT_CSharpNullable)) { 4122 return false; 4123 } 4124 } else if (Style.Language == FormatStyle::LK_Java) { 4125 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 4126 Keywords.kw_implements)) 4127 return false; 4128 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 4129 Keywords.kw_implements)) 4130 return true; 4131 } else if (Style.isJavaScript()) { 4132 const FormatToken *NonComment = Right.getPreviousNonComment(); 4133 if (NonComment && 4134 NonComment->isOneOf( 4135 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 4136 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 4137 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 4138 Keywords.kw_readonly, Keywords.kw_override, Keywords.kw_abstract, 4139 Keywords.kw_get, Keywords.kw_set, Keywords.kw_async, 4140 Keywords.kw_await)) 4141 return false; // Otherwise automatic semicolon insertion would trigger. 4142 if (Right.NestingLevel == 0 && 4143 (Left.Tok.getIdentifierInfo() || 4144 Left.isOneOf(tok::r_square, tok::r_paren)) && 4145 Right.isOneOf(tok::l_square, tok::l_paren)) 4146 return false; // Otherwise automatic semicolon insertion would trigger. 4147 if (NonComment && NonComment->is(tok::identifier) && 4148 NonComment->TokenText == "asserts") 4149 return false; 4150 if (Left.is(TT_FatArrow) && Right.is(tok::l_brace)) 4151 return false; 4152 if (Left.is(TT_JsTypeColon)) 4153 return true; 4154 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;"). 4155 if (Left.is(tok::exclaim) && Right.is(tok::colon)) 4156 return false; 4157 // Look for is type annotations like: 4158 // function f(): a is B { ... } 4159 // Do not break before is in these cases. 4160 if (Right.is(Keywords.kw_is)) { 4161 const FormatToken *Next = Right.getNextNonComment(); 4162 // If `is` is followed by a colon, it's likely that it's a dict key, so 4163 // ignore it for this check. 4164 // For example this is common in Polymer: 4165 // Polymer({ 4166 // is: 'name', 4167 // ... 4168 // }); 4169 if (!Next || !Next->is(tok::colon)) 4170 return false; 4171 } 4172 if (Left.is(Keywords.kw_in)) 4173 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 4174 if (Right.is(Keywords.kw_in)) 4175 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 4176 if (Right.is(Keywords.kw_as)) 4177 return false; // must not break before as in 'x as type' casts 4178 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) { 4179 // extends and infer can appear as keywords in conditional types: 4180 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types 4181 // do not break before them, as the expressions are subject to ASI. 4182 return false; 4183 } 4184 if (Left.is(Keywords.kw_as)) 4185 return true; 4186 if (Left.is(TT_NonNullAssertion)) 4187 return true; 4188 if (Left.is(Keywords.kw_declare) && 4189 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 4190 Keywords.kw_function, tok::kw_class, tok::kw_enum, 4191 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 4192 Keywords.kw_let, tok::kw_const)) 4193 // See grammar for 'declare' statements at: 4194 // https://github.com/Microsoft/TypeScript/blob/main/doc/spec-ARCHIVED.md#A.10 4195 return false; 4196 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 4197 Right.isOneOf(tok::identifier, tok::string_literal)) 4198 return false; // must not break in "module foo { ...}" 4199 if (Right.is(TT_TemplateString) && Right.closesScope()) 4200 return false; 4201 // Don't split tagged template literal so there is a break between the tag 4202 // identifier and template string. 4203 if (Left.is(tok::identifier) && Right.is(TT_TemplateString)) { 4204 return false; 4205 } 4206 if (Left.is(TT_TemplateString) && Left.opensScope()) 4207 return true; 4208 } 4209 4210 if (Left.is(tok::at)) 4211 return false; 4212 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 4213 return false; 4214 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 4215 return !Right.is(tok::l_paren); 4216 if (Right.is(TT_PointerOrReference)) 4217 return Line.IsMultiVariableDeclStmt || 4218 (getTokenPointerOrReferenceAlignment(Right) == 4219 FormatStyle::PAS_Right && 4220 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 4221 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 4222 Right.is(tok::kw_operator)) 4223 return true; 4224 if (Left.is(TT_PointerOrReference)) 4225 return false; 4226 if (Right.isTrailingComment()) 4227 // We rely on MustBreakBefore being set correctly here as we should not 4228 // change the "binding" behavior of a comment. 4229 // The first comment in a braced lists is always interpreted as belonging to 4230 // the first list element. Otherwise, it should be placed outside of the 4231 // list. 4232 return Left.is(BK_BracedInit) || 4233 (Left.is(TT_CtorInitializerColon) && 4234 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 4235 if (Left.is(tok::question) && Right.is(tok::colon)) 4236 return false; 4237 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 4238 return Style.BreakBeforeTernaryOperators; 4239 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 4240 return !Style.BreakBeforeTernaryOperators; 4241 if (Left.is(TT_InheritanceColon)) 4242 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon; 4243 if (Right.is(TT_InheritanceColon)) 4244 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon; 4245 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 4246 Left.isNot(TT_SelectorName)) 4247 return true; 4248 4249 if (Right.is(tok::colon) && 4250 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 4251 return false; 4252 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 4253 if (Style.Language == FormatStyle::LK_Proto || 4254 Style.Language == FormatStyle::LK_TextProto) { 4255 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 4256 return false; 4257 // Prevent cases like: 4258 // 4259 // submessage: 4260 // { key: valueeeeeeeeeeee } 4261 // 4262 // when the snippet does not fit into one line. 4263 // Prefer: 4264 // 4265 // submessage: { 4266 // key: valueeeeeeeeeeee 4267 // } 4268 // 4269 // instead, even if it is longer by one line. 4270 // 4271 // Note that this allows allows the "{" to go over the column limit 4272 // when the column limit is just between ":" and "{", but that does 4273 // not happen too often and alternative formattings in this case are 4274 // not much better. 4275 // 4276 // The code covers the cases: 4277 // 4278 // submessage: { ... } 4279 // submessage: < ... > 4280 // repeated: [ ... ] 4281 if (((Right.is(tok::l_brace) || Right.is(tok::less)) && 4282 Right.is(TT_DictLiteral)) || 4283 Right.is(TT_ArrayInitializerLSquare)) 4284 return false; 4285 } 4286 return true; 4287 } 4288 if (Right.is(tok::r_square) && Right.MatchingParen && 4289 Right.MatchingParen->is(TT_ProtoExtensionLSquare)) 4290 return false; 4291 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 4292 Right.Next->is(TT_ObjCMethodExpr))) 4293 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 4294 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 4295 return true; 4296 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 4297 return true; 4298 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 4299 TT_OverloadedOperator)) 4300 return false; 4301 if (Left.is(TT_RangeBasedForLoopColon)) 4302 return true; 4303 if (Right.is(TT_RangeBasedForLoopColon)) 4304 return false; 4305 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 4306 return true; 4307 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 4308 Left.is(tok::kw_operator)) 4309 return false; 4310 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 4311 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 4312 return false; 4313 if (Left.is(tok::equal) && Right.is(tok::l_brace) && 4314 !Style.Cpp11BracedListStyle) 4315 return false; 4316 if (Left.is(tok::l_paren) && 4317 Left.isOneOf(TT_AttributeParen, TT_TypeDeclarationParen)) 4318 return false; 4319 if (Left.is(tok::l_paren) && Left.Previous && 4320 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 4321 return false; 4322 if (Right.is(TT_ImplicitStringLiteral)) 4323 return false; 4324 4325 if (Right.is(TT_TemplateCloser)) 4326 return false; 4327 if (Right.is(tok::r_square) && Right.MatchingParen && 4328 Right.MatchingParen->is(TT_LambdaLSquare)) 4329 return false; 4330 4331 // We only break before r_brace if there was a corresponding break before 4332 // the l_brace, which is tracked by BreakBeforeClosingBrace. 4333 if (Right.is(tok::r_brace)) 4334 return Right.MatchingParen && Right.MatchingParen->is(BK_Block); 4335 4336 // We only break before r_paren if we're in a block indented context. 4337 if (Right.is(tok::r_paren)) { 4338 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent) { 4339 return Right.MatchingParen && 4340 !(Right.MatchingParen->Previous && 4341 (Right.MatchingParen->Previous->is(tok::kw_for) || 4342 Right.MatchingParen->Previous->isIf())); 4343 } 4344 4345 return false; 4346 } 4347 4348 // Allow breaking after a trailing annotation, e.g. after a method 4349 // declaration. 4350 if (Left.is(TT_TrailingAnnotation)) 4351 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 4352 tok::less, tok::coloncolon); 4353 4354 if (Right.is(tok::kw___attribute) || 4355 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) 4356 return !Left.is(TT_AttributeSquare); 4357 4358 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 4359 return true; 4360 4361 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 4362 return true; 4363 4364 if (Left.is(TT_CtorInitializerColon)) 4365 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 4366 if (Right.is(TT_CtorInitializerColon)) 4367 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 4368 if (Left.is(TT_CtorInitializerComma) && 4369 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 4370 return false; 4371 if (Right.is(TT_CtorInitializerComma) && 4372 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 4373 return true; 4374 if (Left.is(TT_InheritanceComma) && 4375 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 4376 return false; 4377 if (Right.is(TT_InheritanceComma) && 4378 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 4379 return true; 4380 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 4381 (Left.is(tok::less) && Right.is(tok::less))) 4382 return false; 4383 if (Right.is(TT_BinaryOperator) && 4384 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 4385 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 4386 Right.getPrecedence() != prec::Assignment)) 4387 return true; 4388 if (Left.is(TT_ArrayInitializerLSquare)) 4389 return true; 4390 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 4391 return true; 4392 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 4393 !Left.isOneOf(tok::arrowstar, tok::lessless) && 4394 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 4395 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 4396 Left.getPrecedence() == prec::Assignment)) 4397 return true; 4398 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 4399 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) 4400 return false; 4401 4402 auto ShortLambdaOption = Style.AllowShortLambdasOnASingleLine; 4403 if (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace)) { 4404 if (isAllmanLambdaBrace(Left)) 4405 return !isItAnEmptyLambdaAllowed(Left, ShortLambdaOption); 4406 if (isAllmanLambdaBrace(Right)) 4407 return !isItAnEmptyLambdaAllowed(Right, ShortLambdaOption); 4408 } 4409 4410 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 4411 tok::kw_class, tok::kw_struct, tok::comment) || 4412 Right.isMemberAccess() || 4413 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 4414 tok::colon, tok::l_square, tok::at) || 4415 (Left.is(tok::r_paren) && 4416 Right.isOneOf(tok::identifier, tok::kw_const)) || 4417 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 4418 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 4419 } 4420 4421 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 4422 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 4423 const FormatToken *Tok = Line.First; 4424 while (Tok) { 4425 llvm::errs() << " M=" << Tok->MustBreakBefore 4426 << " C=" << Tok->CanBreakBefore 4427 << " T=" << getTokenTypeName(Tok->getType()) 4428 << " S=" << Tok->SpacesRequiredBefore 4429 << " F=" << Tok->Finalized << " B=" << Tok->BlockParameterCount 4430 << " BK=" << Tok->getBlockKind() << " P=" << Tok->SplitPenalty 4431 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 4432 << " PPK=" << Tok->getPackingKind() << " FakeLParens="; 4433 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 4434 llvm::errs() << Tok->FakeLParens[i] << "/"; 4435 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 4436 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo(); 4437 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 4438 if (!Tok->Next) 4439 assert(Tok == Line.Last); 4440 Tok = Tok->Next; 4441 } 4442 llvm::errs() << "----\n"; 4443 } 4444 4445 FormatStyle::PointerAlignmentStyle 4446 TokenAnnotator::getTokenReferenceAlignment(const FormatToken &Reference) { 4447 assert(Reference.isOneOf(tok::amp, tok::ampamp)); 4448 switch (Style.ReferenceAlignment) { 4449 case FormatStyle::RAS_Pointer: 4450 return Style.PointerAlignment; 4451 case FormatStyle::RAS_Left: 4452 return FormatStyle::PAS_Left; 4453 case FormatStyle::RAS_Right: 4454 return FormatStyle::PAS_Right; 4455 case FormatStyle::RAS_Middle: 4456 return FormatStyle::PAS_Middle; 4457 } 4458 assert(0); //"Unhandled value of ReferenceAlignment" 4459 return Style.PointerAlignment; 4460 } 4461 4462 FormatStyle::PointerAlignmentStyle 4463 TokenAnnotator::getTokenPointerOrReferenceAlignment( 4464 const FormatToken &PointerOrReference) { 4465 if (PointerOrReference.isOneOf(tok::amp, tok::ampamp)) { 4466 switch (Style.ReferenceAlignment) { 4467 case FormatStyle::RAS_Pointer: 4468 return Style.PointerAlignment; 4469 case FormatStyle::RAS_Left: 4470 return FormatStyle::PAS_Left; 4471 case FormatStyle::RAS_Right: 4472 return FormatStyle::PAS_Right; 4473 case FormatStyle::RAS_Middle: 4474 return FormatStyle::PAS_Middle; 4475 } 4476 } 4477 assert(PointerOrReference.is(tok::star)); 4478 return Style.PointerAlignment; 4479 } 4480 4481 } // namespace format 4482 } // namespace clang 4483