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(CurrentToken); 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. 120 if (CurrentToken->Next && CurrentToken->Next->is(tok::greater) && 121 Left->ParentBracket != tok::less && 122 isKeywordWithCondition(*Line.First)) 123 return false; 124 Left->MatchingParen = CurrentToken; 125 CurrentToken->MatchingParen = Left; 126 // In TT_Proto, we must distignuish between: 127 // map<key, value> 128 // msg < item: data > 129 // msg: < item: data > 130 // In TT_TextProto, map<key, value> does not occur. 131 if (Style.Language == FormatStyle::LK_TextProto || 132 (Style.Language == FormatStyle::LK_Proto && Left->Previous && 133 Left->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) 134 CurrentToken->setType(TT_DictLiteral); 135 else 136 CurrentToken->setType(TT_TemplateCloser); 137 next(); 138 return true; 139 } 140 if (CurrentToken->is(tok::question) && 141 Style.Language == FormatStyle::LK_Java) { 142 next(); 143 continue; 144 } 145 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) || 146 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext && 147 !Style.isCSharp() && Style.Language != FormatStyle::LK_Proto && 148 Style.Language != FormatStyle::LK_TextProto)) 149 return false; 150 // If a && or || is found and interpreted as a binary operator, this set 151 // of angles is likely part of something like "a < b && c > d". If the 152 // angles are inside an expression, the ||/&& might also be a binary 153 // operator that was misinterpreted because we are parsing template 154 // parameters. 155 // FIXME: This is getting out of hand, write a decent parser. 156 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) && 157 CurrentToken->Previous->is(TT_BinaryOperator) && 158 Contexts[Contexts.size() - 2].IsExpression && 159 !Line.startsWith(tok::kw_template)) 160 return false; 161 updateParameterCount(Left, CurrentToken); 162 if (Style.Language == FormatStyle::LK_Proto) { 163 if (FormatToken *Previous = CurrentToken->getPreviousNonComment()) { 164 if (CurrentToken->is(tok::colon) || 165 (CurrentToken->isOneOf(tok::l_brace, tok::less) && 166 Previous->isNot(tok::colon))) 167 Previous->setType(TT_SelectorName); 168 } 169 } 170 if (!consumeToken()) 171 return false; 172 } 173 return false; 174 } 175 176 bool parseUntouchableParens() { 177 while (CurrentToken) { 178 CurrentToken->Finalized = true; 179 switch (CurrentToken->Tok.getKind()) { 180 case tok::l_paren: 181 next(); 182 if (!parseUntouchableParens()) 183 return false; 184 continue; 185 case tok::r_paren: 186 next(); 187 return true; 188 default: 189 // no-op 190 break; 191 } 192 next(); 193 } 194 return false; 195 } 196 197 bool parseParens(bool LookForDecls = false) { 198 if (!CurrentToken) 199 return false; 200 FormatToken *Left = CurrentToken->Previous; 201 assert(Left && "Unknown previous token"); 202 FormatToken *PrevNonComment = Left->getPreviousNonComment(); 203 Left->ParentBracket = Contexts.back().ContextKind; 204 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1); 205 206 // FIXME: This is a bit of a hack. Do better. 207 Contexts.back().ColonIsForRangeExpr = 208 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr; 209 210 if (Left->Previous && Left->Previous->is(TT_UntouchableMacroFunc)) { 211 Left->Finalized = true; 212 return parseUntouchableParens(); 213 } 214 215 bool StartsObjCMethodExpr = false; 216 if (FormatToken *MaybeSel = Left->Previous) { 217 // @selector( starts a selector. 218 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous && 219 MaybeSel->Previous->is(tok::at)) { 220 StartsObjCMethodExpr = true; 221 } 222 } 223 224 if (Left->is(TT_OverloadedOperatorLParen)) { 225 Contexts.back().IsExpression = false; 226 } else if (Style.Language == FormatStyle::LK_JavaScript && 227 (Line.startsWith(Keywords.kw_type, tok::identifier) || 228 Line.startsWith(tok::kw_export, Keywords.kw_type, 229 tok::identifier))) { 230 // type X = (...); 231 // export type X = (...); 232 Contexts.back().IsExpression = false; 233 } else if (Left->Previous && 234 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_while, 235 tok::l_paren, tok::comma) || 236 Left->Previous->isIf() || 237 Left->Previous->is(TT_BinaryOperator))) { 238 // static_assert, if and while usually contain expressions. 239 Contexts.back().IsExpression = true; 240 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 241 (Left->Previous->is(Keywords.kw_function) || 242 (Left->Previous->endsSequence(tok::identifier, 243 Keywords.kw_function)))) { 244 // function(...) or function f(...) 245 Contexts.back().IsExpression = false; 246 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 247 Left->Previous->is(TT_JsTypeColon)) { 248 // let x: (SomeType); 249 Contexts.back().IsExpression = false; 250 } else if (isLambdaParameterList(Left)) { 251 // This is a parameter list of a lambda expression. 252 Contexts.back().IsExpression = false; 253 } else if (Line.InPPDirective && 254 (!Left->Previous || !Left->Previous->is(tok::identifier))) { 255 Contexts.back().IsExpression = true; 256 } else if (Contexts[Contexts.size() - 2].CaretFound) { 257 // This is the parameter list of an ObjC block. 258 Contexts.back().IsExpression = false; 259 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) { 260 // The first argument to a foreach macro is a declaration. 261 Contexts.back().IsForEachMacro = true; 262 Contexts.back().IsExpression = false; 263 } else if (Left->Previous && Left->Previous->MatchingParen && 264 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) { 265 Contexts.back().IsExpression = false; 266 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) { 267 bool IsForOrCatch = 268 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch); 269 Contexts.back().IsExpression = !IsForOrCatch; 270 } 271 272 // Infer the role of the l_paren based on the previous token if we haven't 273 // detected one one yet. 274 if (PrevNonComment && Left->is(TT_Unknown)) { 275 if (PrevNonComment->is(tok::kw___attribute)) { 276 Left->setType(TT_AttributeParen); 277 } else if (PrevNonComment->isOneOf(TT_TypenameMacro, tok::kw_decltype, 278 tok::kw_typeof, tok::kw__Atomic, 279 tok::kw___underlying_type)) { 280 Left->setType(TT_TypeDeclarationParen); 281 // decltype() and typeof() usually contain expressions. 282 if (PrevNonComment->isOneOf(tok::kw_decltype, tok::kw_typeof)) 283 Contexts.back().IsExpression = true; 284 } 285 } 286 287 if (StartsObjCMethodExpr) { 288 Contexts.back().ColonIsObjCMethodExpr = true; 289 Left->setType(TT_ObjCMethodExpr); 290 } 291 292 // MightBeFunctionType and ProbablyFunctionType are used for 293 // function pointer and reference types as well as Objective-C 294 // block types: 295 // 296 // void (*FunctionPointer)(void); 297 // void (&FunctionReference)(void); 298 // void (^ObjCBlock)(void); 299 bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression; 300 bool ProbablyFunctionType = 301 CurrentToken->isOneOf(tok::star, tok::amp, tok::caret); 302 bool HasMultipleLines = false; 303 bool HasMultipleParametersOnALine = false; 304 bool MightBeObjCForRangeLoop = 305 Left->Previous && Left->Previous->is(tok::kw_for); 306 FormatToken *PossibleObjCForInToken = nullptr; 307 while (CurrentToken) { 308 // LookForDecls is set when "if (" has been seen. Check for 309 // 'identifier' '*' 'identifier' followed by not '=' -- this 310 // '*' has to be a binary operator but determineStarAmpUsage() will 311 // categorize it as an unary operator, so set the right type here. 312 if (LookForDecls && CurrentToken->Next) { 313 FormatToken *Prev = CurrentToken->getPreviousNonComment(); 314 if (Prev) { 315 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 316 FormatToken *Next = CurrentToken->Next; 317 if (PrevPrev && PrevPrev->is(tok::identifier) && 318 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) && 319 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) { 320 Prev->setType(TT_BinaryOperator); 321 LookForDecls = false; 322 } 323 } 324 } 325 326 if (CurrentToken->Previous->is(TT_PointerOrReference) && 327 CurrentToken->Previous->Previous->isOneOf(tok::l_paren, 328 tok::coloncolon)) 329 ProbablyFunctionType = true; 330 if (CurrentToken->is(tok::comma)) 331 MightBeFunctionType = false; 332 if (CurrentToken->Previous->is(TT_BinaryOperator)) 333 Contexts.back().IsExpression = true; 334 if (CurrentToken->is(tok::r_paren)) { 335 if (MightBeFunctionType && ProbablyFunctionType && CurrentToken->Next && 336 (CurrentToken->Next->is(tok::l_paren) || 337 (CurrentToken->Next->is(tok::l_square) && Line.MustBeDeclaration))) 338 Left->setType(Left->Next->is(tok::caret) ? TT_ObjCBlockLParen 339 : TT_FunctionTypeLParen); 340 Left->MatchingParen = CurrentToken; 341 CurrentToken->MatchingParen = Left; 342 343 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) && 344 Left->Previous && Left->Previous->is(tok::l_paren)) { 345 // Detect the case where macros are used to generate lambdas or 346 // function bodies, e.g.: 347 // auto my_lambda = MARCO((Type *type, int i) { .. body .. }); 348 for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) { 349 if (Tok->is(TT_BinaryOperator) && 350 Tok->isOneOf(tok::star, tok::amp, tok::ampamp)) 351 Tok->setType(TT_PointerOrReference); 352 } 353 } 354 355 if (StartsObjCMethodExpr) { 356 CurrentToken->setType(TT_ObjCMethodExpr); 357 if (Contexts.back().FirstObjCSelectorName) { 358 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 359 Contexts.back().LongestObjCSelectorName; 360 } 361 } 362 363 if (Left->is(TT_AttributeParen)) 364 CurrentToken->setType(TT_AttributeParen); 365 if (Left->is(TT_TypeDeclarationParen)) 366 CurrentToken->setType(TT_TypeDeclarationParen); 367 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation)) 368 CurrentToken->setType(TT_JavaAnnotation); 369 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation)) 370 CurrentToken->setType(TT_LeadingJavaAnnotation); 371 if (Left->Previous && Left->Previous->is(TT_AttributeSquare)) 372 CurrentToken->setType(TT_AttributeSquare); 373 374 if (!HasMultipleLines) 375 Left->setPackingKind(PPK_Inconclusive); 376 else if (HasMultipleParametersOnALine) 377 Left->setPackingKind(PPK_BinPacked); 378 else 379 Left->setPackingKind(PPK_OnePerLine); 380 381 next(); 382 return true; 383 } 384 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace)) 385 return false; 386 387 if (CurrentToken->is(tok::l_brace)) 388 Left->setType(TT_Unknown); // Not TT_ObjCBlockLParen 389 if (CurrentToken->is(tok::comma) && CurrentToken->Next && 390 !CurrentToken->Next->HasUnescapedNewline && 391 !CurrentToken->Next->isTrailingComment()) 392 HasMultipleParametersOnALine = true; 393 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) || 394 CurrentToken->Previous->isSimpleTypeSpecifier()) && 395 !CurrentToken->is(tok::l_brace)) 396 Contexts.back().IsExpression = false; 397 if (CurrentToken->isOneOf(tok::semi, tok::colon)) { 398 MightBeObjCForRangeLoop = false; 399 if (PossibleObjCForInToken) { 400 PossibleObjCForInToken->setType(TT_Unknown); 401 PossibleObjCForInToken = nullptr; 402 } 403 } 404 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) { 405 PossibleObjCForInToken = CurrentToken; 406 PossibleObjCForInToken->setType(TT_ObjCForIn); 407 } 408 // When we discover a 'new', we set CanBeExpression to 'false' in order to 409 // parse the type correctly. Reset that after a comma. 410 if (CurrentToken->is(tok::comma)) 411 Contexts.back().CanBeExpression = true; 412 413 FormatToken *Tok = CurrentToken; 414 if (!consumeToken()) 415 return false; 416 updateParameterCount(Left, Tok); 417 if (CurrentToken && CurrentToken->HasUnescapedNewline) 418 HasMultipleLines = true; 419 } 420 return false; 421 } 422 423 bool isCSharpAttributeSpecifier(const FormatToken &Tok) { 424 if (!Style.isCSharp()) 425 return false; 426 427 // `identifier[i]` is not an attribute. 428 if (Tok.Previous && Tok.Previous->is(tok::identifier)) 429 return false; 430 431 // Chains of [] in `identifier[i][j][k]` are not attributes. 432 if (Tok.Previous && Tok.Previous->is(tok::r_square)) { 433 auto *MatchingParen = Tok.Previous->MatchingParen; 434 if (!MatchingParen || MatchingParen->is(TT_ArraySubscriptLSquare)) 435 return false; 436 } 437 438 const FormatToken *AttrTok = Tok.Next; 439 if (!AttrTok) 440 return false; 441 442 // Just an empty declaration e.g. string []. 443 if (AttrTok->is(tok::r_square)) 444 return false; 445 446 // Move along the tokens inbetween the '[' and ']' e.g. [STAThread]. 447 while (AttrTok && AttrTok->isNot(tok::r_square)) { 448 AttrTok = AttrTok->Next; 449 } 450 451 if (!AttrTok) 452 return false; 453 454 // Allow an attribute to be the only content of a file. 455 AttrTok = AttrTok->Next; 456 if (!AttrTok) 457 return true; 458 459 // Limit this to being an access modifier that follows. 460 if (AttrTok->isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 461 tok::comment, tok::kw_class, tok::kw_static, 462 tok::l_square, Keywords.kw_internal)) { 463 return true; 464 } 465 466 // incase its a [XXX] retval func(.... 467 if (AttrTok->Next && 468 AttrTok->Next->startsSequence(tok::identifier, tok::l_paren)) 469 return true; 470 471 return false; 472 } 473 474 bool isCpp11AttributeSpecifier(const FormatToken &Tok) { 475 if (!Style.isCpp() || !Tok.startsSequence(tok::l_square, tok::l_square)) 476 return false; 477 // The first square bracket is part of an ObjC array literal 478 if (Tok.Previous && Tok.Previous->is(tok::at)) { 479 return false; 480 } 481 const FormatToken *AttrTok = Tok.Next->Next; 482 if (!AttrTok) 483 return false; 484 // C++17 '[[using ns: foo, bar(baz, blech)]]' 485 // We assume nobody will name an ObjC variable 'using'. 486 if (AttrTok->startsSequence(tok::kw_using, tok::identifier, tok::colon)) 487 return true; 488 if (AttrTok->isNot(tok::identifier)) 489 return false; 490 while (AttrTok && !AttrTok->startsSequence(tok::r_square, tok::r_square)) { 491 // ObjC message send. We assume nobody will use : in a C++11 attribute 492 // specifier parameter, although this is technically valid: 493 // [[foo(:)]]. 494 if (AttrTok->is(tok::colon) || 495 AttrTok->startsSequence(tok::identifier, tok::identifier) || 496 AttrTok->startsSequence(tok::r_paren, tok::identifier)) 497 return false; 498 if (AttrTok->is(tok::ellipsis)) 499 return true; 500 AttrTok = AttrTok->Next; 501 } 502 return AttrTok && AttrTok->startsSequence(tok::r_square, tok::r_square); 503 } 504 505 bool parseSquare() { 506 if (!CurrentToken) 507 return false; 508 509 // A '[' could be an index subscript (after an identifier or after 510 // ')' or ']'), it could be the start of an Objective-C method 511 // expression, it could the start of an Objective-C array literal, 512 // or it could be a C++ attribute specifier [[foo::bar]]. 513 FormatToken *Left = CurrentToken->Previous; 514 Left->ParentBracket = Contexts.back().ContextKind; 515 FormatToken *Parent = Left->getPreviousNonComment(); 516 517 // Cases where '>' is followed by '['. 518 // In C++, this can happen either in array of templates (foo<int>[10]) 519 // or when array is a nested template type (unique_ptr<type1<type2>[]>). 520 bool CppArrayTemplates = 521 Style.isCpp() && Parent && Parent->is(TT_TemplateCloser) && 522 (Contexts.back().CanBeExpression || Contexts.back().IsExpression || 523 Contexts.back().InTemplateArgument); 524 525 bool IsCpp11AttributeSpecifier = isCpp11AttributeSpecifier(*Left) || 526 Contexts.back().InCpp11AttributeSpecifier; 527 528 // Treat C# Attributes [STAThread] much like C++ attributes [[...]]. 529 bool IsCSharpAttributeSpecifier = 530 isCSharpAttributeSpecifier(*Left) || 531 Contexts.back().InCSharpAttributeSpecifier; 532 533 bool InsideInlineASM = Line.startsWith(tok::kw_asm); 534 bool IsCppStructuredBinding = Left->isCppStructuredBinding(Style); 535 bool StartsObjCMethodExpr = 536 !IsCppStructuredBinding && !InsideInlineASM && !CppArrayTemplates && 537 Style.isCpp() && !IsCpp11AttributeSpecifier && 538 !IsCSharpAttributeSpecifier && Contexts.back().CanBeExpression && 539 Left->isNot(TT_LambdaLSquare) && 540 !CurrentToken->isOneOf(tok::l_brace, tok::r_square) && 541 (!Parent || 542 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren, 543 tok::kw_return, tok::kw_throw) || 544 Parent->isUnaryOperator() || 545 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 546 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) || 547 (getBinOpPrecedence(Parent->Tok.getKind(), true, true) > 548 prec::Unknown)); 549 bool ColonFound = false; 550 551 unsigned BindingIncrease = 1; 552 if (IsCppStructuredBinding) { 553 Left->setType(TT_StructuredBindingLSquare); 554 } else if (Left->is(TT_Unknown)) { 555 if (StartsObjCMethodExpr) { 556 Left->setType(TT_ObjCMethodExpr); 557 } else if (InsideInlineASM) { 558 Left->setType(TT_InlineASMSymbolicNameLSquare); 559 } else if (IsCpp11AttributeSpecifier) { 560 Left->setType(TT_AttributeSquare); 561 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent && 562 Contexts.back().ContextKind == tok::l_brace && 563 Parent->isOneOf(tok::l_brace, tok::comma)) { 564 Left->setType(TT_JsComputedPropertyName); 565 } else if (Style.isCpp() && Contexts.back().ContextKind == tok::l_brace && 566 Parent && Parent->isOneOf(tok::l_brace, tok::comma)) { 567 Left->setType(TT_DesignatedInitializerLSquare); 568 } else if (IsCSharpAttributeSpecifier) { 569 Left->setType(TT_AttributeSquare); 570 } else if (CurrentToken->is(tok::r_square) && Parent && 571 Parent->is(TT_TemplateCloser)) { 572 Left->setType(TT_ArraySubscriptLSquare); 573 } else if (Style.Language == FormatStyle::LK_Proto || 574 Style.Language == FormatStyle::LK_TextProto) { 575 // Square braces in LK_Proto can either be message field attributes: 576 // 577 // optional Aaa aaa = 1 [ 578 // (aaa) = aaa 579 // ]; 580 // 581 // extensions 123 [ 582 // (aaa) = aaa 583 // ]; 584 // 585 // or text proto extensions (in options): 586 // 587 // option (Aaa.options) = { 588 // [type.type/type] { 589 // key: value 590 // } 591 // } 592 // 593 // or repeated fields (in options): 594 // 595 // option (Aaa.options) = { 596 // keys: [ 1, 2, 3 ] 597 // } 598 // 599 // In the first and the third case we want to spread the contents inside 600 // the square braces; in the second we want to keep them inline. 601 Left->setType(TT_ArrayInitializerLSquare); 602 if (!Left->endsSequence(tok::l_square, tok::numeric_constant, 603 tok::equal) && 604 !Left->endsSequence(tok::l_square, tok::numeric_constant, 605 tok::identifier) && 606 !Left->endsSequence(tok::l_square, tok::colon, TT_SelectorName)) { 607 Left->setType(TT_ProtoExtensionLSquare); 608 BindingIncrease = 10; 609 } 610 } else if (!CppArrayTemplates && Parent && 611 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at, 612 tok::comma, tok::l_paren, tok::l_square, 613 tok::question, tok::colon, tok::kw_return, 614 // Should only be relevant to JavaScript: 615 tok::kw_default)) { 616 Left->setType(TT_ArrayInitializerLSquare); 617 } else { 618 BindingIncrease = 10; 619 Left->setType(TT_ArraySubscriptLSquare); 620 } 621 } 622 623 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease); 624 Contexts.back().IsExpression = true; 625 if (Style.Language == FormatStyle::LK_JavaScript && Parent && 626 Parent->is(TT_JsTypeColon)) 627 Contexts.back().IsExpression = false; 628 629 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr; 630 Contexts.back().InCpp11AttributeSpecifier = IsCpp11AttributeSpecifier; 631 Contexts.back().InCSharpAttributeSpecifier = IsCSharpAttributeSpecifier; 632 633 while (CurrentToken) { 634 if (CurrentToken->is(tok::r_square)) { 635 if (IsCpp11AttributeSpecifier) 636 CurrentToken->setType(TT_AttributeSquare); 637 if (IsCSharpAttributeSpecifier) 638 CurrentToken->setType(TT_AttributeSquare); 639 else if (((CurrentToken->Next && 640 CurrentToken->Next->is(tok::l_paren)) || 641 (CurrentToken->Previous && 642 CurrentToken->Previous->Previous == Left)) && 643 Left->is(TT_ObjCMethodExpr)) { 644 // An ObjC method call is rarely followed by an open parenthesis. It 645 // also can't be composed of just one token, unless it's a macro that 646 // will be expanded to more tokens. 647 // FIXME: Do we incorrectly label ":" with this? 648 StartsObjCMethodExpr = false; 649 Left->setType(TT_Unknown); 650 } 651 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) { 652 CurrentToken->setType(TT_ObjCMethodExpr); 653 // If we haven't seen a colon yet, make sure the last identifier 654 // before the r_square is tagged as a selector name component. 655 if (!ColonFound && CurrentToken->Previous && 656 CurrentToken->Previous->is(TT_Unknown) && 657 canBeObjCSelectorComponent(*CurrentToken->Previous)) 658 CurrentToken->Previous->setType(TT_SelectorName); 659 // determineStarAmpUsage() thinks that '*' '[' is allocating an 660 // array of pointers, but if '[' starts a selector then '*' is a 661 // binary operator. 662 if (Parent && Parent->is(TT_PointerOrReference)) 663 Parent->setType(TT_BinaryOperator); 664 } 665 // An arrow after an ObjC method expression is not a lambda arrow. 666 if (CurrentToken->getType() == TT_ObjCMethodExpr && 667 CurrentToken->Next && CurrentToken->Next->is(TT_LambdaArrow)) 668 CurrentToken->Next->setType(TT_Unknown); 669 Left->MatchingParen = CurrentToken; 670 CurrentToken->MatchingParen = Left; 671 // FirstObjCSelectorName is set when a colon is found. This does 672 // not work, however, when the method has no parameters. 673 // Here, we set FirstObjCSelectorName when the end of the method call is 674 // reached, in case it was not set already. 675 if (!Contexts.back().FirstObjCSelectorName) { 676 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 677 if (Previous && Previous->is(TT_SelectorName)) { 678 Previous->ObjCSelectorNameParts = 1; 679 Contexts.back().FirstObjCSelectorName = Previous; 680 } 681 } else { 682 Left->ParameterCount = 683 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 684 } 685 if (Contexts.back().FirstObjCSelectorName) { 686 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 687 Contexts.back().LongestObjCSelectorName; 688 if (Left->BlockParameterCount > 1) 689 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0; 690 } 691 next(); 692 return true; 693 } 694 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace)) 695 return false; 696 if (CurrentToken->is(tok::colon)) { 697 if (IsCpp11AttributeSpecifier && 698 CurrentToken->endsSequence(tok::colon, tok::identifier, 699 tok::kw_using)) { 700 // Remember that this is a [[using ns: foo]] C++ attribute, so we 701 // don't add a space before the colon (unlike other colons). 702 CurrentToken->setType(TT_AttributeColon); 703 } else if (Left->isOneOf(TT_ArraySubscriptLSquare, 704 TT_DesignatedInitializerLSquare)) { 705 Left->setType(TT_ObjCMethodExpr); 706 StartsObjCMethodExpr = true; 707 Contexts.back().ColonIsObjCMethodExpr = true; 708 if (Parent && Parent->is(tok::r_paren)) 709 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 710 Parent->setType(TT_CastRParen); 711 } 712 ColonFound = true; 713 } 714 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) && 715 !ColonFound) 716 Left->setType(TT_ArrayInitializerLSquare); 717 FormatToken *Tok = CurrentToken; 718 if (!consumeToken()) 719 return false; 720 updateParameterCount(Left, Tok); 721 } 722 return false; 723 } 724 725 bool parseBrace() { 726 if (CurrentToken) { 727 FormatToken *Left = CurrentToken->Previous; 728 Left->ParentBracket = Contexts.back().ContextKind; 729 730 if (Contexts.back().CaretFound) 731 Left->setType(TT_ObjCBlockLBrace); 732 Contexts.back().CaretFound = false; 733 734 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1); 735 Contexts.back().ColonIsDictLiteral = true; 736 if (Left->is(BK_BracedInit)) 737 Contexts.back().IsExpression = true; 738 if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 739 Left->Previous->is(TT_JsTypeColon)) 740 Contexts.back().IsExpression = false; 741 742 while (CurrentToken) { 743 if (CurrentToken->is(tok::r_brace)) { 744 Left->MatchingParen = CurrentToken; 745 CurrentToken->MatchingParen = Left; 746 next(); 747 return true; 748 } 749 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 750 return false; 751 updateParameterCount(Left, CurrentToken); 752 if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) { 753 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 754 if (Previous->is(TT_JsTypeOptionalQuestion)) 755 Previous = Previous->getPreviousNonComment(); 756 if ((CurrentToken->is(tok::colon) && 757 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) || 758 Style.Language == FormatStyle::LK_Proto || 759 Style.Language == FormatStyle::LK_TextProto) { 760 Left->setType(TT_DictLiteral); 761 if (Previous->Tok.getIdentifierInfo() || 762 Previous->is(tok::string_literal)) 763 Previous->setType(TT_SelectorName); 764 } 765 if (CurrentToken->is(tok::colon) || 766 Style.Language == FormatStyle::LK_JavaScript) 767 Left->setType(TT_DictLiteral); 768 } 769 if (CurrentToken->is(tok::comma) && 770 Style.Language == FormatStyle::LK_JavaScript) 771 Left->setType(TT_DictLiteral); 772 if (!consumeToken()) 773 return false; 774 } 775 } 776 return true; 777 } 778 779 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 780 // For ObjC methods, the number of parameters is calculated differently as 781 // method declarations have a different structure (the parameters are not 782 // inside a bracket scope). 783 if (Current->is(tok::l_brace) && Current->is(BK_Block)) 784 ++Left->BlockParameterCount; 785 if (Current->is(tok::comma)) { 786 ++Left->ParameterCount; 787 if (!Left->Role) 788 Left->Role.reset(new CommaSeparatedList(Style)); 789 Left->Role->CommaFound(Current); 790 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 791 Left->ParameterCount = 1; 792 } 793 } 794 795 bool parseConditional() { 796 while (CurrentToken) { 797 if (CurrentToken->is(tok::colon)) { 798 CurrentToken->setType(TT_ConditionalExpr); 799 next(); 800 return true; 801 } 802 if (!consumeToken()) 803 return false; 804 } 805 return false; 806 } 807 808 bool parseTemplateDeclaration() { 809 if (CurrentToken && CurrentToken->is(tok::less)) { 810 CurrentToken->setType(TT_TemplateOpener); 811 next(); 812 if (!parseAngle()) 813 return false; 814 if (CurrentToken) 815 CurrentToken->Previous->ClosesTemplateDeclaration = true; 816 return true; 817 } 818 return false; 819 } 820 821 bool consumeToken() { 822 FormatToken *Tok = CurrentToken; 823 next(); 824 switch (Tok->Tok.getKind()) { 825 case tok::plus: 826 case tok::minus: 827 if (!Tok->Previous && Line.MustBeDeclaration) 828 Tok->setType(TT_ObjCMethodSpecifier); 829 break; 830 case tok::colon: 831 if (!Tok->Previous) 832 return false; 833 // Colons from ?: are handled in parseConditional(). 834 if (Style.Language == FormatStyle::LK_JavaScript) { 835 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop 836 (Contexts.size() == 1 && // switch/case labels 837 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) || 838 Contexts.back().ContextKind == tok::l_paren || // function params 839 Contexts.back().ContextKind == tok::l_square || // array type 840 (!Contexts.back().IsExpression && 841 Contexts.back().ContextKind == tok::l_brace) || // object type 842 (Contexts.size() == 1 && 843 Line.MustBeDeclaration)) { // method/property declaration 844 Contexts.back().IsExpression = false; 845 Tok->setType(TT_JsTypeColon); 846 break; 847 } 848 } else if (Style.isCSharp()) { 849 if (Contexts.back().InCSharpAttributeSpecifier) { 850 Tok->setType(TT_AttributeColon); 851 break; 852 } 853 if (Contexts.back().ContextKind == tok::l_paren) { 854 Tok->setType(TT_CSharpNamedArgumentColon); 855 break; 856 } 857 } 858 if (Contexts.back().ColonIsDictLiteral || 859 Style.Language == FormatStyle::LK_Proto || 860 Style.Language == FormatStyle::LK_TextProto) { 861 Tok->setType(TT_DictLiteral); 862 if (Style.Language == FormatStyle::LK_TextProto) { 863 if (FormatToken *Previous = Tok->getPreviousNonComment()) 864 Previous->setType(TT_SelectorName); 865 } 866 } else if (Contexts.back().ColonIsObjCMethodExpr || 867 Line.startsWith(TT_ObjCMethodSpecifier)) { 868 Tok->setType(TT_ObjCMethodExpr); 869 const FormatToken *BeforePrevious = Tok->Previous->Previous; 870 // Ensure we tag all identifiers in method declarations as 871 // TT_SelectorName. 872 bool UnknownIdentifierInMethodDeclaration = 873 Line.startsWith(TT_ObjCMethodSpecifier) && 874 Tok->Previous->is(tok::identifier) && Tok->Previous->is(TT_Unknown); 875 if (!BeforePrevious || 876 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 877 !(BeforePrevious->is(TT_CastRParen) || 878 (BeforePrevious->is(TT_ObjCMethodExpr) && 879 BeforePrevious->is(tok::colon))) || 880 BeforePrevious->is(tok::r_square) || 881 Contexts.back().LongestObjCSelectorName == 0 || 882 UnknownIdentifierInMethodDeclaration) { 883 Tok->Previous->setType(TT_SelectorName); 884 if (!Contexts.back().FirstObjCSelectorName) 885 Contexts.back().FirstObjCSelectorName = Tok->Previous; 886 else if (Tok->Previous->ColumnWidth > 887 Contexts.back().LongestObjCSelectorName) 888 Contexts.back().LongestObjCSelectorName = 889 Tok->Previous->ColumnWidth; 890 Tok->Previous->ParameterIndex = 891 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 892 ++Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 893 } 894 } else if (Contexts.back().ColonIsForRangeExpr) { 895 Tok->setType(TT_RangeBasedForLoopColon); 896 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 897 Tok->setType(TT_BitFieldColon); 898 } else if (Contexts.size() == 1 && 899 !Line.First->isOneOf(tok::kw_enum, tok::kw_case, 900 tok::kw_default)) { 901 FormatToken *Prev = Tok->getPreviousNonComment(); 902 if (Prev->isOneOf(tok::r_paren, tok::kw_noexcept)) 903 Tok->setType(TT_CtorInitializerColon); 904 else if (Prev->is(tok::kw_try)) { 905 // Member initializer list within function try block. 906 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 907 if (PrevPrev && PrevPrev->isOneOf(tok::r_paren, tok::kw_noexcept)) 908 Tok->setType(TT_CtorInitializerColon); 909 } else 910 Tok->setType(TT_InheritanceColon); 911 } else if (canBeObjCSelectorComponent(*Tok->Previous) && Tok->Next && 912 (Tok->Next->isOneOf(tok::r_paren, tok::comma) || 913 (canBeObjCSelectorComponent(*Tok->Next) && Tok->Next->Next && 914 Tok->Next->Next->is(tok::colon)))) { 915 // This handles a special macro in ObjC code where selectors including 916 // the colon are passed as macro arguments. 917 Tok->setType(TT_ObjCMethodExpr); 918 } else if (Contexts.back().ContextKind == tok::l_paren) { 919 Tok->setType(TT_InlineASMColon); 920 } 921 break; 922 case tok::pipe: 923 case tok::amp: 924 // | and & in declarations/type expressions represent union and 925 // intersection types, respectively. 926 if (Style.Language == FormatStyle::LK_JavaScript && 927 !Contexts.back().IsExpression) 928 Tok->setType(TT_JsTypeOperator); 929 break; 930 case tok::kw_if: 931 case tok::kw_while: 932 if (Tok->is(tok::kw_if) && CurrentToken && 933 CurrentToken->isOneOf(tok::kw_constexpr, tok::identifier)) 934 next(); 935 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 936 next(); 937 if (!parseParens(/*LookForDecls=*/true)) 938 return false; 939 } 940 break; 941 case tok::kw_for: 942 if (Style.Language == FormatStyle::LK_JavaScript) { 943 // x.for and {for: ...} 944 if ((Tok->Previous && Tok->Previous->is(tok::period)) || 945 (Tok->Next && Tok->Next->is(tok::colon))) 946 break; 947 // JS' for await ( ... 948 if (CurrentToken && CurrentToken->is(Keywords.kw_await)) 949 next(); 950 } 951 Contexts.back().ColonIsForRangeExpr = true; 952 next(); 953 if (!parseParens()) 954 return false; 955 break; 956 case tok::l_paren: 957 // When faced with 'operator()()', the kw_operator handler incorrectly 958 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make 959 // the first two parens OverloadedOperators and the second l_paren an 960 // OverloadedOperatorLParen. 961 if (Tok->Previous && Tok->Previous->is(tok::r_paren) && 962 Tok->Previous->MatchingParen && 963 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) { 964 Tok->Previous->setType(TT_OverloadedOperator); 965 Tok->Previous->MatchingParen->setType(TT_OverloadedOperator); 966 Tok->setType(TT_OverloadedOperatorLParen); 967 } 968 969 if (!parseParens()) 970 return false; 971 if (Line.MustBeDeclaration && Contexts.size() == 1 && 972 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) && 973 !Tok->is(TT_TypeDeclarationParen) && 974 (!Tok->Previous || !Tok->Previous->isOneOf(tok::kw___attribute, 975 TT_LeadingJavaAnnotation))) 976 Line.MightBeFunctionDecl = true; 977 break; 978 case tok::l_square: 979 if (!parseSquare()) 980 return false; 981 break; 982 case tok::l_brace: 983 if (Style.Language == FormatStyle::LK_TextProto) { 984 FormatToken *Previous = Tok->getPreviousNonComment(); 985 if (Previous && Previous->getType() != TT_DictLiteral) 986 Previous->setType(TT_SelectorName); 987 } 988 if (!parseBrace()) 989 return false; 990 break; 991 case tok::less: 992 if (parseAngle()) { 993 Tok->setType(TT_TemplateOpener); 994 // In TT_Proto, we must distignuish between: 995 // map<key, value> 996 // msg < item: data > 997 // msg: < item: data > 998 // In TT_TextProto, map<key, value> does not occur. 999 if (Style.Language == FormatStyle::LK_TextProto || 1000 (Style.Language == FormatStyle::LK_Proto && Tok->Previous && 1001 Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) { 1002 Tok->setType(TT_DictLiteral); 1003 FormatToken *Previous = Tok->getPreviousNonComment(); 1004 if (Previous && Previous->getType() != TT_DictLiteral) 1005 Previous->setType(TT_SelectorName); 1006 } 1007 } else { 1008 Tok->setType(TT_BinaryOperator); 1009 NonTemplateLess.insert(Tok); 1010 CurrentToken = Tok; 1011 next(); 1012 } 1013 break; 1014 case tok::r_paren: 1015 case tok::r_square: 1016 return false; 1017 case tok::r_brace: 1018 // Lines can start with '}'. 1019 if (Tok->Previous) 1020 return false; 1021 break; 1022 case tok::greater: 1023 if (Style.Language != FormatStyle::LK_TextProto) 1024 Tok->setType(TT_BinaryOperator); 1025 if (Tok->Previous && Tok->Previous->is(TT_TemplateCloser)) 1026 Tok->SpacesRequiredBefore = 1; 1027 break; 1028 case tok::kw_operator: 1029 if (Style.Language == FormatStyle::LK_TextProto || 1030 Style.Language == FormatStyle::LK_Proto) 1031 break; 1032 while (CurrentToken && 1033 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 1034 if (CurrentToken->isOneOf(tok::star, tok::amp)) 1035 CurrentToken->setType(TT_PointerOrReference); 1036 consumeToken(); 1037 if (CurrentToken && CurrentToken->is(tok::comma) && 1038 CurrentToken->Previous->isNot(tok::kw_operator)) 1039 break; 1040 if (CurrentToken && CurrentToken->Previous->isOneOf( 1041 TT_BinaryOperator, TT_UnaryOperator, tok::comma, 1042 tok::star, tok::arrow, tok::amp, tok::ampamp)) 1043 CurrentToken->Previous->setType(TT_OverloadedOperator); 1044 } 1045 if (CurrentToken && CurrentToken->is(tok::l_paren)) 1046 CurrentToken->setType(TT_OverloadedOperatorLParen); 1047 if (CurrentToken && CurrentToken->Previous->is(TT_BinaryOperator)) 1048 CurrentToken->Previous->setType(TT_OverloadedOperator); 1049 break; 1050 case tok::question: 1051 if (Tok->is(TT_CSharpNullConditionalLSquare)) { 1052 if (!parseSquare()) 1053 return false; 1054 break; 1055 } 1056 if (Tok->isOneOf(TT_CSharpNullConditional, TT_CSharpNullCoalescing)) 1057 break; 1058 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next && 1059 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren, 1060 tok::r_brace)) { 1061 // Question marks before semicolons, colons, etc. indicate optional 1062 // types (fields, parameters), e.g. 1063 // function(x?: string, y?) {...} 1064 // class X { y?; } 1065 Tok->setType(TT_JsTypeOptionalQuestion); 1066 break; 1067 } 1068 // Declarations cannot be conditional expressions, this can only be part 1069 // of a type declaration. 1070 if (Line.MustBeDeclaration && !Contexts.back().IsExpression && 1071 Style.Language == FormatStyle::LK_JavaScript) 1072 break; 1073 if (Style.isCSharp()) { 1074 // `Type?)`, `Type?>`, `Type? name;` and `Type? name =` can only be 1075 // nullable types. 1076 // Line.MustBeDeclaration will be true for `Type? name;`. 1077 if ((!Contexts.back().IsExpression && Line.MustBeDeclaration) || 1078 (Tok->Next && Tok->Next->isOneOf(tok::r_paren, tok::greater)) || 1079 (Tok->Next && Tok->Next->is(tok::identifier) && Tok->Next->Next && 1080 Tok->Next->Next->is(tok::equal))) { 1081 Tok->setType(TT_CSharpNullable); 1082 break; 1083 } 1084 } 1085 parseConditional(); 1086 break; 1087 case tok::kw_template: 1088 parseTemplateDeclaration(); 1089 break; 1090 case tok::comma: 1091 if (Contexts.back().InCtorInitializer) 1092 Tok->setType(TT_CtorInitializerComma); 1093 else if (Contexts.back().InInheritanceList) 1094 Tok->setType(TT_InheritanceComma); 1095 else if (Contexts.back().FirstStartOfName && 1096 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) { 1097 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 1098 Line.IsMultiVariableDeclStmt = true; 1099 } 1100 if (Contexts.back().IsForEachMacro) 1101 Contexts.back().IsExpression = true; 1102 break; 1103 case tok::identifier: 1104 if (Tok->isOneOf(Keywords.kw___has_include, 1105 Keywords.kw___has_include_next)) { 1106 parseHasInclude(); 1107 } 1108 if (Style.isCSharp() && Tok->is(Keywords.kw_where) && Tok->Next && 1109 Tok->Next->isNot(tok::l_paren)) { 1110 Tok->setType(TT_CSharpGenericTypeConstraint); 1111 parseCSharpGenericTypeConstraint(); 1112 } 1113 break; 1114 default: 1115 break; 1116 } 1117 return true; 1118 } 1119 1120 void parseCSharpGenericTypeConstraint() { 1121 int OpenAngleBracketsCount = 0; 1122 while (CurrentToken) { 1123 if (CurrentToken->is(tok::less)) { 1124 // parseAngle is too greedy and will consume the whole line. 1125 CurrentToken->setType(TT_TemplateOpener); 1126 ++OpenAngleBracketsCount; 1127 next(); 1128 } else if (CurrentToken->is(tok::greater)) { 1129 CurrentToken->setType(TT_TemplateCloser); 1130 --OpenAngleBracketsCount; 1131 next(); 1132 } else if (CurrentToken->is(tok::comma) && OpenAngleBracketsCount == 0) { 1133 // We allow line breaks after GenericTypeConstraintComma's 1134 // so do not flag commas in Generics as GenericTypeConstraintComma's. 1135 CurrentToken->setType(TT_CSharpGenericTypeConstraintComma); 1136 next(); 1137 } else if (CurrentToken->is(Keywords.kw_where)) { 1138 CurrentToken->setType(TT_CSharpGenericTypeConstraint); 1139 next(); 1140 } else if (CurrentToken->is(tok::colon)) { 1141 CurrentToken->setType(TT_CSharpGenericTypeConstraintColon); 1142 next(); 1143 } else { 1144 next(); 1145 } 1146 } 1147 } 1148 1149 void parseIncludeDirective() { 1150 if (CurrentToken && CurrentToken->is(tok::less)) { 1151 next(); 1152 while (CurrentToken) { 1153 // Mark tokens up to the trailing line comments as implicit string 1154 // literals. 1155 if (CurrentToken->isNot(tok::comment) && 1156 !CurrentToken->TokenText.startswith("//")) 1157 CurrentToken->setType(TT_ImplicitStringLiteral); 1158 next(); 1159 } 1160 } 1161 } 1162 1163 void parseWarningOrError() { 1164 next(); 1165 // We still want to format the whitespace left of the first token of the 1166 // warning or error. 1167 next(); 1168 while (CurrentToken) { 1169 CurrentToken->setType(TT_ImplicitStringLiteral); 1170 next(); 1171 } 1172 } 1173 1174 void parsePragma() { 1175 next(); // Consume "pragma". 1176 if (CurrentToken && 1177 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 1178 bool IsMark = CurrentToken->is(Keywords.kw_mark); 1179 next(); // Consume "mark". 1180 next(); // Consume first token (so we fix leading whitespace). 1181 while (CurrentToken) { 1182 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 1183 CurrentToken->setType(TT_ImplicitStringLiteral); 1184 next(); 1185 } 1186 } 1187 } 1188 1189 void parseHasInclude() { 1190 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 1191 return; 1192 next(); // '(' 1193 parseIncludeDirective(); 1194 next(); // ')' 1195 } 1196 1197 LineType parsePreprocessorDirective() { 1198 bool IsFirstToken = CurrentToken->IsFirst; 1199 LineType Type = LT_PreprocessorDirective; 1200 next(); 1201 if (!CurrentToken) 1202 return Type; 1203 1204 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) { 1205 // JavaScript files can contain shebang lines of the form: 1206 // #!/usr/bin/env node 1207 // Treat these like C++ #include directives. 1208 while (CurrentToken) { 1209 // Tokens cannot be comments here. 1210 CurrentToken->setType(TT_ImplicitStringLiteral); 1211 next(); 1212 } 1213 return LT_ImportStatement; 1214 } 1215 1216 if (CurrentToken->Tok.is(tok::numeric_constant)) { 1217 CurrentToken->SpacesRequiredBefore = 1; 1218 return Type; 1219 } 1220 // Hashes in the middle of a line can lead to any strange token 1221 // sequence. 1222 if (!CurrentToken->Tok.getIdentifierInfo()) 1223 return Type; 1224 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 1225 case tok::pp_include: 1226 case tok::pp_include_next: 1227 case tok::pp_import: 1228 next(); 1229 parseIncludeDirective(); 1230 Type = LT_ImportStatement; 1231 break; 1232 case tok::pp_error: 1233 case tok::pp_warning: 1234 parseWarningOrError(); 1235 break; 1236 case tok::pp_pragma: 1237 parsePragma(); 1238 break; 1239 case tok::pp_if: 1240 case tok::pp_elif: 1241 Contexts.back().IsExpression = true; 1242 next(); 1243 parseLine(); 1244 break; 1245 default: 1246 break; 1247 } 1248 while (CurrentToken) { 1249 FormatToken *Tok = CurrentToken; 1250 next(); 1251 if (Tok->is(tok::l_paren)) 1252 parseParens(); 1253 else if (Tok->isOneOf(Keywords.kw___has_include, 1254 Keywords.kw___has_include_next)) 1255 parseHasInclude(); 1256 } 1257 return Type; 1258 } 1259 1260 public: 1261 LineType parseLine() { 1262 if (!CurrentToken) 1263 return LT_Invalid; 1264 NonTemplateLess.clear(); 1265 if (CurrentToken->is(tok::hash)) 1266 return parsePreprocessorDirective(); 1267 1268 // Directly allow to 'import <string-literal>' to support protocol buffer 1269 // definitions (github.com/google/protobuf) or missing "#" (either way we 1270 // should not break the line). 1271 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 1272 if ((Style.Language == FormatStyle::LK_Java && 1273 CurrentToken->is(Keywords.kw_package)) || 1274 (Info && Info->getPPKeywordID() == tok::pp_import && 1275 CurrentToken->Next && 1276 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 1277 tok::kw_static))) { 1278 next(); 1279 parseIncludeDirective(); 1280 return LT_ImportStatement; 1281 } 1282 1283 // If this line starts and ends in '<' and '>', respectively, it is likely 1284 // part of "#define <a/b.h>". 1285 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 1286 parseIncludeDirective(); 1287 return LT_ImportStatement; 1288 } 1289 1290 // In .proto files, top-level options and package statements are very 1291 // similar to import statements and should not be line-wrapped. 1292 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 1293 CurrentToken->isOneOf(Keywords.kw_option, Keywords.kw_package)) { 1294 next(); 1295 if (CurrentToken && CurrentToken->is(tok::identifier)) { 1296 while (CurrentToken) 1297 next(); 1298 return LT_ImportStatement; 1299 } 1300 } 1301 1302 bool KeywordVirtualFound = false; 1303 bool ImportStatement = false; 1304 1305 // import {...} from '...'; 1306 if (Style.Language == FormatStyle::LK_JavaScript && 1307 CurrentToken->is(Keywords.kw_import)) 1308 ImportStatement = true; 1309 1310 while (CurrentToken) { 1311 if (CurrentToken->is(tok::kw_virtual)) 1312 KeywordVirtualFound = true; 1313 if (Style.Language == FormatStyle::LK_JavaScript) { 1314 // export {...} from '...'; 1315 // An export followed by "from 'some string';" is a re-export from 1316 // another module identified by a URI and is treated as a 1317 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 1318 // Just "export {...};" or "export class ..." should not be treated as 1319 // an import in this sense. 1320 if (Line.First->is(tok::kw_export) && 1321 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 1322 CurrentToken->Next->isStringLiteral()) 1323 ImportStatement = true; 1324 if (isClosureImportStatement(*CurrentToken)) 1325 ImportStatement = true; 1326 } 1327 if (!consumeToken()) 1328 return LT_Invalid; 1329 } 1330 if (KeywordVirtualFound) 1331 return LT_VirtualFunctionDecl; 1332 if (ImportStatement) 1333 return LT_ImportStatement; 1334 1335 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 1336 if (Contexts.back().FirstObjCSelectorName) 1337 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 1338 Contexts.back().LongestObjCSelectorName; 1339 return LT_ObjCMethodDecl; 1340 } 1341 1342 return LT_Other; 1343 } 1344 1345 private: 1346 bool isClosureImportStatement(const FormatToken &Tok) { 1347 // FIXME: Closure-library specific stuff should not be hard-coded but be 1348 // configurable. 1349 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 1350 Tok.Next->Next && 1351 (Tok.Next->Next->TokenText == "module" || 1352 Tok.Next->Next->TokenText == "provide" || 1353 Tok.Next->Next->TokenText == "require" || 1354 Tok.Next->Next->TokenText == "requireType" || 1355 Tok.Next->Next->TokenText == "forwardDeclare") && 1356 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 1357 } 1358 1359 void resetTokenMetadata(FormatToken *Token) { 1360 if (!Token) 1361 return; 1362 1363 // Reset token type in case we have already looked at it and then 1364 // recovered from an error (e.g. failure to find the matching >). 1365 if (!CurrentToken->isOneOf( 1366 TT_LambdaLSquare, TT_LambdaLBrace, TT_AttributeMacro, 1367 TT_ForEachMacro, TT_TypenameMacro, TT_FunctionLBrace, 1368 TT_ImplicitStringLiteral, TT_InlineASMBrace, TT_JsFatArrow, 1369 TT_LambdaArrow, TT_NamespaceMacro, TT_OverloadedOperator, 1370 TT_RegexLiteral, TT_TemplateString, TT_ObjCStringLiteral, 1371 TT_UntouchableMacroFunc, TT_ConstraintJunctions, 1372 TT_StatementAttributeLikeMacro)) 1373 CurrentToken->setType(TT_Unknown); 1374 CurrentToken->Role.reset(); 1375 CurrentToken->MatchingParen = nullptr; 1376 CurrentToken->FakeLParens.clear(); 1377 CurrentToken->FakeRParens = 0; 1378 } 1379 1380 void next() { 1381 if (CurrentToken) { 1382 CurrentToken->NestingLevel = Contexts.size() - 1; 1383 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1384 modifyContext(*CurrentToken); 1385 determineTokenType(*CurrentToken); 1386 CurrentToken = CurrentToken->Next; 1387 } 1388 1389 resetTokenMetadata(CurrentToken); 1390 } 1391 1392 /// A struct to hold information valid in a specific context, e.g. 1393 /// a pair of parenthesis. 1394 struct Context { 1395 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1396 bool IsExpression) 1397 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1398 IsExpression(IsExpression) {} 1399 1400 tok::TokenKind ContextKind; 1401 unsigned BindingStrength; 1402 bool IsExpression; 1403 unsigned LongestObjCSelectorName = 0; 1404 bool ColonIsForRangeExpr = false; 1405 bool ColonIsDictLiteral = false; 1406 bool ColonIsObjCMethodExpr = false; 1407 FormatToken *FirstObjCSelectorName = nullptr; 1408 FormatToken *FirstStartOfName = nullptr; 1409 bool CanBeExpression = true; 1410 bool InTemplateArgument = false; 1411 bool InCtorInitializer = false; 1412 bool InInheritanceList = false; 1413 bool CaretFound = false; 1414 bool IsForEachMacro = false; 1415 bool InCpp11AttributeSpecifier = false; 1416 bool InCSharpAttributeSpecifier = false; 1417 }; 1418 1419 /// Puts a new \c Context onto the stack \c Contexts for the lifetime 1420 /// of each instance. 1421 struct ScopedContextCreator { 1422 AnnotatingParser &P; 1423 1424 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1425 unsigned Increase) 1426 : P(P) { 1427 P.Contexts.push_back(Context(ContextKind, 1428 P.Contexts.back().BindingStrength + Increase, 1429 P.Contexts.back().IsExpression)); 1430 } 1431 1432 ~ScopedContextCreator() { P.Contexts.pop_back(); } 1433 }; 1434 1435 void modifyContext(const FormatToken &Current) { 1436 if (Current.getPrecedence() == prec::Assignment && 1437 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 1438 // Type aliases use `type X = ...;` in TypeScript and can be exported 1439 // using `export type ...`. 1440 !(Style.Language == FormatStyle::LK_JavaScript && 1441 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1442 Line.startsWith(tok::kw_export, Keywords.kw_type, 1443 tok::identifier))) && 1444 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 1445 Contexts.back().IsExpression = true; 1446 if (!Line.startsWith(TT_UnaryOperator)) { 1447 for (FormatToken *Previous = Current.Previous; 1448 Previous && Previous->Previous && 1449 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1450 Previous = Previous->Previous) { 1451 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1452 Previous = Previous->MatchingParen; 1453 if (!Previous) 1454 break; 1455 } 1456 if (Previous->opensScope()) 1457 break; 1458 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1459 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1460 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1461 Previous->setType(TT_PointerOrReference); 1462 } 1463 } 1464 } else if (Current.is(tok::lessless) && 1465 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1466 Contexts.back().IsExpression = true; 1467 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1468 Contexts.back().IsExpression = true; 1469 } else if (Current.is(TT_TrailingReturnArrow)) { 1470 Contexts.back().IsExpression = false; 1471 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1472 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1473 } else if (Current.Previous && 1474 Current.Previous->is(TT_CtorInitializerColon)) { 1475 Contexts.back().IsExpression = true; 1476 Contexts.back().InCtorInitializer = true; 1477 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1478 Contexts.back().InInheritanceList = true; 1479 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1480 for (FormatToken *Previous = Current.Previous; 1481 Previous && Previous->isOneOf(tok::star, tok::amp); 1482 Previous = Previous->Previous) 1483 Previous->setType(TT_PointerOrReference); 1484 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 1485 Contexts.back().IsExpression = false; 1486 } else if (Current.is(tok::kw_new)) { 1487 Contexts.back().CanBeExpression = false; 1488 } else if (Current.is(tok::semi) || 1489 (Current.is(tok::exclaim) && Current.Previous && 1490 !Current.Previous->is(tok::kw_operator))) { 1491 // This should be the condition or increment in a for-loop. 1492 // But not operator !() (can't use TT_OverloadedOperator here as its not 1493 // been annotated yet). 1494 Contexts.back().IsExpression = true; 1495 } 1496 } 1497 1498 static FormatToken *untilMatchingParen(FormatToken *Current) { 1499 // Used when `MatchingParen` is not yet established. 1500 int ParenLevel = 0; 1501 while (Current) { 1502 if (Current->is(tok::l_paren)) 1503 ParenLevel++; 1504 if (Current->is(tok::r_paren)) 1505 ParenLevel--; 1506 if (ParenLevel < 1) 1507 break; 1508 Current = Current->Next; 1509 } 1510 return Current; 1511 } 1512 1513 static bool isDeductionGuide(FormatToken &Current) { 1514 // Look for a deduction guide template<T> A(...) -> A<...>; 1515 if (Current.Previous && Current.Previous->is(tok::r_paren) && 1516 Current.startsSequence(tok::arrow, tok::identifier, tok::less)) { 1517 // Find the TemplateCloser. 1518 FormatToken *TemplateCloser = Current.Next->Next; 1519 int NestingLevel = 0; 1520 while (TemplateCloser) { 1521 // Skip over an expressions in parens A<(3 < 2)>; 1522 if (TemplateCloser->is(tok::l_paren)) { 1523 // No Matching Paren yet so skip to matching paren 1524 TemplateCloser = untilMatchingParen(TemplateCloser); 1525 } 1526 if (TemplateCloser->is(tok::less)) 1527 NestingLevel++; 1528 if (TemplateCloser->is(tok::greater)) 1529 NestingLevel--; 1530 if (NestingLevel < 1) 1531 break; 1532 TemplateCloser = TemplateCloser->Next; 1533 } 1534 // Assuming we have found the end of the template ensure its followed 1535 // with a semi-colon. 1536 if (TemplateCloser && TemplateCloser->Next && 1537 TemplateCloser->Next->is(tok::semi) && 1538 Current.Previous->MatchingParen) { 1539 // Determine if the identifier `A` prior to the A<..>; is the same as 1540 // prior to the A(..) 1541 FormatToken *LeadingIdentifier = 1542 Current.Previous->MatchingParen->Previous; 1543 1544 // Differentiate a deduction guide by seeing the 1545 // > of the template prior to the leading identifier. 1546 if (LeadingIdentifier) { 1547 FormatToken *PriorLeadingIdentifier = LeadingIdentifier->Previous; 1548 // Skip back past explicit decoration 1549 if (PriorLeadingIdentifier && 1550 PriorLeadingIdentifier->is(tok::kw_explicit)) 1551 PriorLeadingIdentifier = PriorLeadingIdentifier->Previous; 1552 1553 return (PriorLeadingIdentifier && 1554 PriorLeadingIdentifier->is(TT_TemplateCloser) && 1555 LeadingIdentifier->TokenText == Current.Next->TokenText); 1556 } 1557 } 1558 } 1559 return false; 1560 } 1561 1562 void determineTokenType(FormatToken &Current) { 1563 if (!Current.is(TT_Unknown)) 1564 // The token type is already known. 1565 return; 1566 1567 if (Style.isCSharp() && CurrentToken->is(tok::question)) { 1568 if (CurrentToken->TokenText == "??") { 1569 Current.setType(TT_CSharpNullCoalescing); 1570 return; 1571 } 1572 if (CurrentToken->TokenText == "?.") { 1573 Current.setType(TT_CSharpNullConditional); 1574 return; 1575 } 1576 if (CurrentToken->TokenText == "?[") { 1577 Current.setType(TT_CSharpNullConditionalLSquare); 1578 return; 1579 } 1580 } 1581 1582 if (Style.Language == FormatStyle::LK_JavaScript) { 1583 if (Current.is(tok::exclaim)) { 1584 if (Current.Previous && 1585 (Keywords.IsJavaScriptIdentifier( 1586 *Current.Previous, /* AcceptIdentifierName= */ true) || 1587 Current.Previous->isOneOf( 1588 tok::kw_namespace, tok::r_paren, tok::r_square, tok::r_brace, 1589 Keywords.kw_type, Keywords.kw_get, Keywords.kw_set) || 1590 Current.Previous->Tok.isLiteral())) { 1591 Current.setType(TT_JsNonNullAssertion); 1592 return; 1593 } 1594 if (Current.Next && 1595 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1596 Current.setType(TT_JsNonNullAssertion); 1597 return; 1598 } 1599 } 1600 } 1601 1602 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1603 // function declaration have been found. In this case, 'Current' is a 1604 // trailing token of this declaration and thus cannot be a name. 1605 if (Current.is(Keywords.kw_instanceof)) { 1606 Current.setType(TT_BinaryOperator); 1607 } else if (isStartOfName(Current) && 1608 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1609 Contexts.back().FirstStartOfName = &Current; 1610 Current.setType(TT_StartOfName); 1611 } else if (Current.is(tok::semi)) { 1612 // Reset FirstStartOfName after finding a semicolon so that a for loop 1613 // with multiple increment statements is not confused with a for loop 1614 // having multiple variable declarations. 1615 Contexts.back().FirstStartOfName = nullptr; 1616 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1617 AutoFound = true; 1618 } else if (Current.is(tok::arrow) && 1619 Style.Language == FormatStyle::LK_Java) { 1620 Current.setType(TT_LambdaArrow); 1621 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1622 Current.NestingLevel == 0 && 1623 !Current.Previous->is(tok::kw_operator)) { 1624 // not auto operator->() -> xxx; 1625 Current.setType(TT_TrailingReturnArrow); 1626 } else if (Current.is(tok::arrow) && Current.Previous && 1627 Current.Previous->is(tok::r_brace)) { 1628 // Concept implicit conversion contraint needs to be treated like 1629 // a trailing return type ... } -> <type>. 1630 Current.setType(TT_TrailingReturnArrow); 1631 } else if (isDeductionGuide(Current)) { 1632 // Deduction guides trailing arrow " A(...) -> A<T>;". 1633 Current.setType(TT_TrailingReturnArrow); 1634 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1635 Current.setType(determineStarAmpUsage( 1636 Current, 1637 Contexts.back().CanBeExpression && Contexts.back().IsExpression, 1638 Contexts.back().InTemplateArgument)); 1639 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1640 Current.setType(determinePlusMinusCaretUsage(Current)); 1641 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1642 Contexts.back().CaretFound = true; 1643 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1644 Current.setType(determineIncrementUsage(Current)); 1645 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1646 Current.setType(TT_UnaryOperator); 1647 } else if (Current.is(tok::question)) { 1648 if (Style.Language == FormatStyle::LK_JavaScript && 1649 Line.MustBeDeclaration && !Contexts.back().IsExpression) { 1650 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1651 // on the interface, not a ternary expression. 1652 Current.setType(TT_JsTypeOptionalQuestion); 1653 } else { 1654 Current.setType(TT_ConditionalExpr); 1655 } 1656 } else if (Current.isBinaryOperator() && 1657 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1658 (!Current.is(tok::greater) && 1659 Style.Language != FormatStyle::LK_TextProto)) { 1660 Current.setType(TT_BinaryOperator); 1661 } else if (Current.is(tok::comment)) { 1662 if (Current.TokenText.startswith("/*")) { 1663 if (Current.TokenText.endswith("*/")) 1664 Current.setType(TT_BlockComment); 1665 else 1666 // The lexer has for some reason determined a comment here. But we 1667 // cannot really handle it, if it isn't properly terminated. 1668 Current.Tok.setKind(tok::unknown); 1669 } else { 1670 Current.setType(TT_LineComment); 1671 } 1672 } else if (Current.is(tok::r_paren)) { 1673 if (rParenEndsCast(Current)) 1674 Current.setType(TT_CastRParen); 1675 if (Current.MatchingParen && Current.Next && 1676 !Current.Next->isBinaryOperator() && 1677 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1678 tok::comma, tok::period, tok::arrow, 1679 tok::coloncolon)) 1680 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1681 // Make sure this isn't the return type of an Obj-C block declaration 1682 if (AfterParen->Tok.isNot(tok::caret)) { 1683 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1684 if (BeforeParen->is(tok::identifier) && 1685 !BeforeParen->is(TT_TypenameMacro) && 1686 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1687 (!BeforeParen->Previous || 1688 BeforeParen->Previous->ClosesTemplateDeclaration)) 1689 Current.setType(TT_FunctionAnnotationRParen); 1690 } 1691 } 1692 } else if (Current.is(tok::at) && Current.Next && 1693 Style.Language != FormatStyle::LK_JavaScript && 1694 Style.Language != FormatStyle::LK_Java) { 1695 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1696 // marks declarations and properties that need special formatting. 1697 switch (Current.Next->Tok.getObjCKeywordID()) { 1698 case tok::objc_interface: 1699 case tok::objc_implementation: 1700 case tok::objc_protocol: 1701 Current.setType(TT_ObjCDecl); 1702 break; 1703 case tok::objc_property: 1704 Current.setType(TT_ObjCProperty); 1705 break; 1706 default: 1707 break; 1708 } 1709 } else if (Current.is(tok::period)) { 1710 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1711 if (PreviousNoComment && 1712 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1713 Current.setType(TT_DesignatedInitializerPeriod); 1714 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1715 Current.Previous->isOneOf(TT_JavaAnnotation, 1716 TT_LeadingJavaAnnotation)) { 1717 Current.setType(Current.Previous->getType()); 1718 } 1719 } else if (canBeObjCSelectorComponent(Current) && 1720 // FIXME(bug 36976): ObjC return types shouldn't use 1721 // TT_CastRParen. 1722 Current.Previous && Current.Previous->is(TT_CastRParen) && 1723 Current.Previous->MatchingParen && 1724 Current.Previous->MatchingParen->Previous && 1725 Current.Previous->MatchingParen->Previous->is( 1726 TT_ObjCMethodSpecifier)) { 1727 // This is the first part of an Objective-C selector name. (If there's no 1728 // colon after this, this is the only place which annotates the identifier 1729 // as a selector.) 1730 Current.setType(TT_SelectorName); 1731 } else if (Current.isOneOf(tok::identifier, tok::kw_const, tok::kw_noexcept, 1732 tok::kw_requires) && 1733 Current.Previous && 1734 !Current.Previous->isOneOf(tok::equal, tok::at) && 1735 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1736 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1737 // function declaration have been found. 1738 Current.setType(TT_TrailingAnnotation); 1739 } else if ((Style.Language == FormatStyle::LK_Java || 1740 Style.Language == FormatStyle::LK_JavaScript) && 1741 Current.Previous) { 1742 if (Current.Previous->is(tok::at) && 1743 Current.isNot(Keywords.kw_interface)) { 1744 const FormatToken &AtToken = *Current.Previous; 1745 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1746 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1747 Current.setType(TT_LeadingJavaAnnotation); 1748 else 1749 Current.setType(TT_JavaAnnotation); 1750 } else if (Current.Previous->is(tok::period) && 1751 Current.Previous->isOneOf(TT_JavaAnnotation, 1752 TT_LeadingJavaAnnotation)) { 1753 Current.setType(Current.Previous->getType()); 1754 } 1755 } 1756 } 1757 1758 /// Take a guess at whether \p Tok starts a name of a function or 1759 /// variable declaration. 1760 /// 1761 /// This is a heuristic based on whether \p Tok is an identifier following 1762 /// something that is likely a type. 1763 bool isStartOfName(const FormatToken &Tok) { 1764 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1765 return false; 1766 1767 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1768 Keywords.kw_as)) 1769 return false; 1770 if (Style.Language == FormatStyle::LK_JavaScript && 1771 Tok.Previous->is(Keywords.kw_in)) 1772 return false; 1773 1774 // Skip "const" as it does not have an influence on whether this is a name. 1775 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1776 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1777 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1778 1779 if (!PreviousNotConst) 1780 return false; 1781 1782 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1783 PreviousNotConst->Previous && 1784 PreviousNotConst->Previous->is(tok::hash); 1785 1786 if (PreviousNotConst->is(TT_TemplateCloser)) 1787 return PreviousNotConst && PreviousNotConst->MatchingParen && 1788 PreviousNotConst->MatchingParen->Previous && 1789 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1790 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1791 1792 if (PreviousNotConst->is(tok::r_paren) && 1793 PreviousNotConst->is(TT_TypeDeclarationParen)) 1794 return true; 1795 1796 return (!IsPPKeyword && 1797 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) || 1798 PreviousNotConst->is(TT_PointerOrReference) || 1799 PreviousNotConst->isSimpleTypeSpecifier(); 1800 } 1801 1802 /// Determine whether ')' is ending a cast. 1803 bool rParenEndsCast(const FormatToken &Tok) { 1804 // C-style casts are only used in C++, C# and Java. 1805 if (!Style.isCSharp() && !Style.isCpp() && 1806 Style.Language != FormatStyle::LK_Java) 1807 return false; 1808 1809 // Empty parens aren't casts and there are no casts at the end of the line. 1810 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1811 return false; 1812 1813 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1814 if (LeftOfParens) { 1815 // If there is a closing parenthesis left of the current parentheses, 1816 // look past it as these might be chained casts. 1817 if (LeftOfParens->is(tok::r_paren)) { 1818 if (!LeftOfParens->MatchingParen || 1819 !LeftOfParens->MatchingParen->Previous) 1820 return false; 1821 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1822 } 1823 1824 // If there is an identifier (or with a few exceptions a keyword) right 1825 // before the parentheses, this is unlikely to be a cast. 1826 if (LeftOfParens->Tok.getIdentifierInfo() && 1827 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1828 tok::kw_delete)) 1829 return false; 1830 1831 // Certain other tokens right before the parentheses are also signals that 1832 // this cannot be a cast. 1833 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1834 TT_TemplateCloser, tok::ellipsis)) 1835 return false; 1836 } 1837 1838 if (Tok.Next->is(tok::question)) 1839 return false; 1840 1841 // `foreach((A a, B b) in someList)` should not be seen as a cast. 1842 if (Tok.Next->is(Keywords.kw_in) && Style.isCSharp()) 1843 return false; 1844 1845 // Functions which end with decorations like volatile, noexcept are unlikely 1846 // to be casts. 1847 if (Tok.Next->isOneOf(tok::kw_noexcept, tok::kw_volatile, tok::kw_const, 1848 tok::kw_requires, tok::kw_throw, tok::arrow, 1849 Keywords.kw_override, Keywords.kw_final) || 1850 isCpp11AttributeSpecifier(*Tok.Next)) 1851 return false; 1852 1853 // As Java has no function types, a "(" after the ")" likely means that this 1854 // is a cast. 1855 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1856 return true; 1857 1858 // If a (non-string) literal follows, this is likely a cast. 1859 if (Tok.Next->isNot(tok::string_literal) && 1860 (Tok.Next->Tok.isLiteral() || 1861 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1862 return true; 1863 1864 // Heuristically try to determine whether the parentheses contain a type. 1865 auto IsQualifiedPointerOrReference = [](FormatToken *T) { 1866 // This is used to handle cases such as x = (foo *const)&y; 1867 assert(!T->isSimpleTypeSpecifier() && "Should have already been checked"); 1868 // Strip trailing qualifiers such as const or volatile when checking 1869 // whether the parens could be a cast to a pointer/reference type. 1870 while (T) { 1871 if (T->is(TT_AttributeParen)) { 1872 // Handle `x = (foo *__attribute__((foo)))&v;`: 1873 if (T->MatchingParen && T->MatchingParen->Previous && 1874 T->MatchingParen->Previous->is(tok::kw___attribute)) { 1875 T = T->MatchingParen->Previous->Previous; 1876 continue; 1877 } 1878 } else if (T->is(TT_AttributeSquare)) { 1879 // Handle `x = (foo *[[clang::foo]])&v;`: 1880 if (T->MatchingParen && T->MatchingParen->Previous) { 1881 T = T->MatchingParen->Previous; 1882 continue; 1883 } 1884 } else if (T->canBePointerOrReferenceQualifier()) { 1885 T = T->Previous; 1886 continue; 1887 } 1888 break; 1889 } 1890 return T && T->is(TT_PointerOrReference); 1891 }; 1892 bool ParensAreType = 1893 !Tok.Previous || 1894 Tok.Previous->isOneOf(TT_TemplateCloser, TT_TypeDeclarationParen) || 1895 Tok.Previous->isSimpleTypeSpecifier() || 1896 IsQualifiedPointerOrReference(Tok.Previous); 1897 bool ParensCouldEndDecl = 1898 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1899 if (ParensAreType && !ParensCouldEndDecl) 1900 return true; 1901 1902 // At this point, we heuristically assume that there are no casts at the 1903 // start of the line. We assume that we have found most cases where there 1904 // are by the logic above, e.g. "(void)x;". 1905 if (!LeftOfParens) 1906 return false; 1907 1908 // Certain token types inside the parentheses mean that this can't be a 1909 // cast. 1910 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 1911 Token = Token->Next) 1912 if (Token->is(TT_BinaryOperator)) 1913 return false; 1914 1915 // If the following token is an identifier or 'this', this is a cast. All 1916 // cases where this can be something else are handled above. 1917 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1918 return true; 1919 1920 // Look for a cast `( x ) (`. 1921 if (Tok.Next->is(tok::l_paren) && Tok.Previous && Tok.Previous->Previous) { 1922 if (Tok.Previous->is(tok::identifier) && 1923 Tok.Previous->Previous->is(tok::l_paren)) 1924 return true; 1925 } 1926 1927 if (!Tok.Next->Next) 1928 return false; 1929 1930 // If the next token after the parenthesis is a unary operator, assume 1931 // that this is cast, unless there are unexpected tokens inside the 1932 // parenthesis. 1933 bool NextIsUnary = 1934 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 1935 if (!NextIsUnary || Tok.Next->is(tok::plus) || 1936 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 1937 return false; 1938 // Search for unexpected tokens. 1939 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 1940 Prev = Prev->Previous) { 1941 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 1942 return false; 1943 } 1944 return true; 1945 } 1946 1947 /// Return the type of the given token assuming it is * or &. 1948 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 1949 bool InTemplateArgument) { 1950 if (Style.Language == FormatStyle::LK_JavaScript) 1951 return TT_BinaryOperator; 1952 1953 // && in C# must be a binary operator. 1954 if (Style.isCSharp() && Tok.is(tok::ampamp)) 1955 return TT_BinaryOperator; 1956 1957 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1958 if (!PrevToken) 1959 return TT_UnaryOperator; 1960 1961 const FormatToken *NextToken = Tok.getNextNonComment(); 1962 if (!NextToken || 1963 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_noexcept) || 1964 NextToken->canBePointerOrReferenceQualifier() || 1965 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 1966 return TT_PointerOrReference; 1967 1968 if (PrevToken->is(tok::coloncolon)) 1969 return TT_PointerOrReference; 1970 1971 if (PrevToken->is(tok::r_paren) && PrevToken->is(TT_TypeDeclarationParen)) 1972 return TT_PointerOrReference; 1973 1974 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 1975 tok::comma, tok::semi, tok::kw_return, tok::colon, 1976 tok::kw_co_return, tok::kw_co_await, 1977 tok::kw_co_yield, tok::equal, tok::kw_delete, 1978 tok::kw_sizeof, tok::kw_throw) || 1979 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 1980 TT_UnaryOperator, TT_CastRParen)) 1981 return TT_UnaryOperator; 1982 1983 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 1984 return TT_PointerOrReference; 1985 if (NextToken->is(tok::kw_operator) && !IsExpression) 1986 return TT_PointerOrReference; 1987 if (NextToken->isOneOf(tok::comma, tok::semi)) 1988 return TT_PointerOrReference; 1989 1990 if (PrevToken->Tok.isLiteral() || 1991 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 1992 tok::kw_false, tok::r_brace) || 1993 NextToken->Tok.isLiteral() || 1994 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 1995 NextToken->isUnaryOperator() || 1996 // If we know we're in a template argument, there are no named 1997 // declarations. Thus, having an identifier on the right-hand side 1998 // indicates a binary operator. 1999 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 2000 return TT_BinaryOperator; 2001 2002 // "&&(" is quite unlikely to be two successive unary "&". 2003 if (Tok.is(tok::ampamp) && NextToken->is(tok::l_paren)) 2004 return TT_BinaryOperator; 2005 2006 // This catches some cases where evaluation order is used as control flow: 2007 // aaa && aaa->f(); 2008 if (NextToken->Tok.isAnyIdentifier()) { 2009 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 2010 if (NextNextToken && NextNextToken->is(tok::arrow)) 2011 return TT_BinaryOperator; 2012 } 2013 2014 // It is very unlikely that we are going to find a pointer or reference type 2015 // definition on the RHS of an assignment. 2016 if (IsExpression && !Contexts.back().CaretFound) 2017 return TT_BinaryOperator; 2018 2019 return TT_PointerOrReference; 2020 } 2021 2022 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 2023 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2024 if (!PrevToken) 2025 return TT_UnaryOperator; 2026 2027 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 2028 // This must be a sequence of leading unary operators. 2029 return TT_UnaryOperator; 2030 2031 // Use heuristics to recognize unary operators. 2032 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 2033 tok::question, tok::colon, tok::kw_return, 2034 tok::kw_case, tok::at, tok::l_brace, tok::kw_throw, 2035 tok::kw_co_return, tok::kw_co_yield)) 2036 return TT_UnaryOperator; 2037 2038 // There can't be two consecutive binary operators. 2039 if (PrevToken->is(TT_BinaryOperator)) 2040 return TT_UnaryOperator; 2041 2042 // Fall back to marking the token as binary operator. 2043 return TT_BinaryOperator; 2044 } 2045 2046 /// Determine whether ++/-- are pre- or post-increments/-decrements. 2047 TokenType determineIncrementUsage(const FormatToken &Tok) { 2048 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2049 if (!PrevToken || PrevToken->is(TT_CastRParen)) 2050 return TT_UnaryOperator; 2051 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 2052 return TT_TrailingUnaryOperator; 2053 2054 return TT_UnaryOperator; 2055 } 2056 2057 SmallVector<Context, 8> Contexts; 2058 2059 const FormatStyle &Style; 2060 AnnotatedLine &Line; 2061 FormatToken *CurrentToken; 2062 bool AutoFound; 2063 const AdditionalKeywords &Keywords; 2064 2065 // Set of "<" tokens that do not open a template parameter list. If parseAngle 2066 // determines that a specific token can't be a template opener, it will make 2067 // same decision irrespective of the decisions for tokens leading up to it. 2068 // Store this information to prevent this from causing exponential runtime. 2069 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 2070 }; 2071 2072 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 2073 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 2074 2075 /// Parses binary expressions by inserting fake parenthesis based on 2076 /// operator precedence. 2077 class ExpressionParser { 2078 public: 2079 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 2080 AnnotatedLine &Line) 2081 : Style(Style), Keywords(Keywords), Current(Line.First) {} 2082 2083 /// Parse expressions with the given operator precedence. 2084 void parse(int Precedence = 0) { 2085 // Skip 'return' and ObjC selector colons as they are not part of a binary 2086 // expression. 2087 while (Current && (Current->is(tok::kw_return) || 2088 (Current->is(tok::colon) && 2089 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 2090 next(); 2091 2092 if (!Current || Precedence > PrecedenceArrowAndPeriod) 2093 return; 2094 2095 // Conditional expressions need to be parsed separately for proper nesting. 2096 if (Precedence == prec::Conditional) { 2097 parseConditionalExpr(); 2098 return; 2099 } 2100 2101 // Parse unary operators, which all have a higher precedence than binary 2102 // operators. 2103 if (Precedence == PrecedenceUnaryOperator) { 2104 parseUnaryOperator(); 2105 return; 2106 } 2107 2108 FormatToken *Start = Current; 2109 FormatToken *LatestOperator = nullptr; 2110 unsigned OperatorIndex = 0; 2111 2112 while (Current) { 2113 // Consume operators with higher precedence. 2114 parse(Precedence + 1); 2115 2116 int CurrentPrecedence = getCurrentPrecedence(); 2117 2118 if (Current && Current->is(TT_SelectorName) && 2119 Precedence == CurrentPrecedence) { 2120 if (LatestOperator) 2121 addFakeParenthesis(Start, prec::Level(Precedence)); 2122 Start = Current; 2123 } 2124 2125 // At the end of the line or when an operator with higher precedence is 2126 // found, insert fake parenthesis and return. 2127 if (!Current || 2128 (Current->closesScope() && 2129 (Current->MatchingParen || Current->is(TT_TemplateString))) || 2130 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 2131 (CurrentPrecedence == prec::Conditional && 2132 Precedence == prec::Assignment && Current->is(tok::colon))) { 2133 break; 2134 } 2135 2136 // Consume scopes: (), [], <> and {} 2137 if (Current->opensScope()) { 2138 // In fragment of a JavaScript template string can look like '}..${' and 2139 // thus close a scope and open a new one at the same time. 2140 while (Current && (!Current->closesScope() || Current->opensScope())) { 2141 next(); 2142 parse(); 2143 } 2144 next(); 2145 } else { 2146 // Operator found. 2147 if (CurrentPrecedence == Precedence) { 2148 if (LatestOperator) 2149 LatestOperator->NextOperator = Current; 2150 LatestOperator = Current; 2151 Current->OperatorIndex = OperatorIndex; 2152 ++OperatorIndex; 2153 } 2154 next(/*SkipPastLeadingComments=*/Precedence > 0); 2155 } 2156 } 2157 2158 if (LatestOperator && (Current || Precedence > 0)) { 2159 // LatestOperator->LastOperator = true; 2160 if (Precedence == PrecedenceArrowAndPeriod) { 2161 // Call expressions don't have a binary operator precedence. 2162 addFakeParenthesis(Start, prec::Unknown); 2163 } else { 2164 addFakeParenthesis(Start, prec::Level(Precedence)); 2165 } 2166 } 2167 } 2168 2169 private: 2170 /// Gets the precedence (+1) of the given token for binary operators 2171 /// and other tokens that we treat like binary operators. 2172 int getCurrentPrecedence() { 2173 if (Current) { 2174 const FormatToken *NextNonComment = Current->getNextNonComment(); 2175 if (Current->is(TT_ConditionalExpr)) 2176 return prec::Conditional; 2177 if (NextNonComment && Current->is(TT_SelectorName) && 2178 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 2179 ((Style.Language == FormatStyle::LK_Proto || 2180 Style.Language == FormatStyle::LK_TextProto) && 2181 NextNonComment->is(tok::less)))) 2182 return prec::Assignment; 2183 if (Current->is(TT_JsComputedPropertyName)) 2184 return prec::Assignment; 2185 if (Current->is(TT_LambdaArrow)) 2186 return prec::Comma; 2187 if (Current->is(TT_JsFatArrow)) 2188 return prec::Assignment; 2189 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 2190 (Current->is(tok::comment) && NextNonComment && 2191 NextNonComment->is(TT_SelectorName))) 2192 return 0; 2193 if (Current->is(TT_RangeBasedForLoopColon)) 2194 return prec::Comma; 2195 if ((Style.Language == FormatStyle::LK_Java || 2196 Style.Language == FormatStyle::LK_JavaScript) && 2197 Current->is(Keywords.kw_instanceof)) 2198 return prec::Relational; 2199 if (Style.Language == FormatStyle::LK_JavaScript && 2200 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 2201 return prec::Relational; 2202 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 2203 return Current->getPrecedence(); 2204 if (Current->isOneOf(tok::period, tok::arrow)) 2205 return PrecedenceArrowAndPeriod; 2206 if ((Style.Language == FormatStyle::LK_Java || 2207 Style.Language == FormatStyle::LK_JavaScript) && 2208 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 2209 Keywords.kw_throws)) 2210 return 0; 2211 } 2212 return -1; 2213 } 2214 2215 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 2216 Start->FakeLParens.push_back(Precedence); 2217 if (Precedence > prec::Unknown) 2218 Start->StartsBinaryExpression = true; 2219 if (Current) { 2220 FormatToken *Previous = Current->Previous; 2221 while (Previous->is(tok::comment) && Previous->Previous) 2222 Previous = Previous->Previous; 2223 ++Previous->FakeRParens; 2224 if (Precedence > prec::Unknown) 2225 Previous->EndsBinaryExpression = true; 2226 } 2227 } 2228 2229 /// Parse unary operator expressions and surround them with fake 2230 /// parentheses if appropriate. 2231 void parseUnaryOperator() { 2232 llvm::SmallVector<FormatToken *, 2> Tokens; 2233 while (Current && Current->is(TT_UnaryOperator)) { 2234 Tokens.push_back(Current); 2235 next(); 2236 } 2237 parse(PrecedenceArrowAndPeriod); 2238 for (FormatToken *Token : llvm::reverse(Tokens)) 2239 // The actual precedence doesn't matter. 2240 addFakeParenthesis(Token, prec::Unknown); 2241 } 2242 2243 void parseConditionalExpr() { 2244 while (Current && Current->isTrailingComment()) { 2245 next(); 2246 } 2247 FormatToken *Start = Current; 2248 parse(prec::LogicalOr); 2249 if (!Current || !Current->is(tok::question)) 2250 return; 2251 next(); 2252 parse(prec::Assignment); 2253 if (!Current || Current->isNot(TT_ConditionalExpr)) 2254 return; 2255 next(); 2256 parse(prec::Assignment); 2257 addFakeParenthesis(Start, prec::Conditional); 2258 } 2259 2260 void next(bool SkipPastLeadingComments = true) { 2261 if (Current) 2262 Current = Current->Next; 2263 while (Current && 2264 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 2265 Current->isTrailingComment()) 2266 Current = Current->Next; 2267 } 2268 2269 const FormatStyle &Style; 2270 const AdditionalKeywords &Keywords; 2271 FormatToken *Current; 2272 }; 2273 2274 } // end anonymous namespace 2275 2276 void TokenAnnotator::setCommentLineLevels( 2277 SmallVectorImpl<AnnotatedLine *> &Lines) { 2278 const AnnotatedLine *NextNonCommentLine = nullptr; 2279 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(), 2280 E = Lines.rend(); 2281 I != E; ++I) { 2282 bool CommentLine = true; 2283 for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) { 2284 if (!Tok->is(tok::comment)) { 2285 CommentLine = false; 2286 break; 2287 } 2288 } 2289 2290 // If the comment is currently aligned with the line immediately following 2291 // it, that's probably intentional and we should keep it. 2292 if (NextNonCommentLine && CommentLine && 2293 NextNonCommentLine->First->NewlinesBefore <= 1 && 2294 NextNonCommentLine->First->OriginalColumn == 2295 (*I)->First->OriginalColumn) { 2296 // Align comments for preprocessor lines with the # in column 0 if 2297 // preprocessor lines are not indented. Otherwise, align with the next 2298 // line. 2299 (*I)->Level = 2300 (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 2301 (NextNonCommentLine->Type == LT_PreprocessorDirective || 2302 NextNonCommentLine->Type == LT_ImportStatement)) 2303 ? 0 2304 : NextNonCommentLine->Level; 2305 } else { 2306 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr; 2307 } 2308 2309 setCommentLineLevels((*I)->Children); 2310 } 2311 } 2312 2313 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 2314 unsigned Result = 0; 2315 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 2316 Result = std::max(Result, Tok->NestingLevel); 2317 return Result; 2318 } 2319 2320 void TokenAnnotator::annotate(AnnotatedLine &Line) { 2321 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2322 E = Line.Children.end(); 2323 I != E; ++I) { 2324 annotate(**I); 2325 } 2326 AnnotatingParser Parser(Style, Line, Keywords); 2327 Line.Type = Parser.parseLine(); 2328 2329 // With very deep nesting, ExpressionParser uses lots of stack and the 2330 // formatting algorithm is very slow. We're not going to do a good job here 2331 // anyway - it's probably generated code being formatted by mistake. 2332 // Just skip the whole line. 2333 if (maxNestingDepth(Line) > 50) 2334 Line.Type = LT_Invalid; 2335 2336 if (Line.Type == LT_Invalid) 2337 return; 2338 2339 ExpressionParser ExprParser(Style, Keywords, Line); 2340 ExprParser.parse(); 2341 2342 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2343 Line.Type = LT_ObjCMethodDecl; 2344 else if (Line.startsWith(TT_ObjCDecl)) 2345 Line.Type = LT_ObjCDecl; 2346 else if (Line.startsWith(TT_ObjCProperty)) 2347 Line.Type = LT_ObjCProperty; 2348 2349 Line.First->SpacesRequiredBefore = 1; 2350 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 2351 } 2352 2353 // This function heuristically determines whether 'Current' starts the name of a 2354 // function declaration. 2355 static bool isFunctionDeclarationName(const FormatToken &Current, 2356 const AnnotatedLine &Line) { 2357 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 2358 for (; Next; Next = Next->Next) { 2359 if (Next->is(TT_OverloadedOperatorLParen)) 2360 return Next; 2361 if (Next->is(TT_OverloadedOperator)) 2362 continue; 2363 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 2364 // For 'new[]' and 'delete[]'. 2365 if (Next->Next && 2366 Next->Next->startsSequence(tok::l_square, tok::r_square)) 2367 Next = Next->Next->Next; 2368 continue; 2369 } 2370 if (Next->startsSequence(tok::l_square, tok::r_square)) { 2371 // For operator[](). 2372 Next = Next->Next; 2373 continue; 2374 } 2375 if ((Next->isSimpleTypeSpecifier() || Next->is(tok::identifier)) && 2376 Next->Next && Next->Next->isOneOf(tok::star, tok::amp, tok::ampamp)) { 2377 // For operator void*(), operator char*(), operator Foo*(). 2378 Next = Next->Next; 2379 continue; 2380 } 2381 if (Next->is(TT_TemplateOpener) && Next->MatchingParen) { 2382 Next = Next->MatchingParen; 2383 continue; 2384 } 2385 2386 break; 2387 } 2388 return nullptr; 2389 }; 2390 2391 // Find parentheses of parameter list. 2392 const FormatToken *Next = Current.Next; 2393 if (Current.is(tok::kw_operator)) { 2394 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 2395 return false; 2396 Next = skipOperatorName(Next); 2397 } else { 2398 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 2399 return false; 2400 for (; Next; Next = Next->Next) { 2401 if (Next->is(TT_TemplateOpener)) { 2402 Next = Next->MatchingParen; 2403 } else if (Next->is(tok::coloncolon)) { 2404 Next = Next->Next; 2405 if (!Next) 2406 return false; 2407 if (Next->is(tok::kw_operator)) { 2408 Next = skipOperatorName(Next->Next); 2409 break; 2410 } 2411 if (!Next->is(tok::identifier)) 2412 return false; 2413 } else if (Next->is(tok::l_paren)) { 2414 break; 2415 } else { 2416 return false; 2417 } 2418 } 2419 } 2420 2421 // Check whether parameter list can belong to a function declaration. 2422 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 2423 return false; 2424 // If the lines ends with "{", this is likely an function definition. 2425 if (Line.Last->is(tok::l_brace)) 2426 return true; 2427 if (Next->Next == Next->MatchingParen) 2428 return true; // Empty parentheses. 2429 // If there is an &/&& after the r_paren, this is likely a function. 2430 if (Next->MatchingParen->Next && 2431 Next->MatchingParen->Next->is(TT_PointerOrReference)) 2432 return true; 2433 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 2434 Tok = Tok->Next) { 2435 if (Tok->is(TT_TypeDeclarationParen)) 2436 return true; 2437 if (Tok->isOneOf(tok::l_paren, TT_TemplateOpener) && Tok->MatchingParen) { 2438 Tok = Tok->MatchingParen; 2439 continue; 2440 } 2441 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 2442 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 2443 return true; 2444 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 2445 Tok->Tok.isLiteral()) 2446 return false; 2447 } 2448 return false; 2449 } 2450 2451 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 2452 assert(Line.MightBeFunctionDecl); 2453 2454 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 2455 Style.AlwaysBreakAfterReturnType == 2456 FormatStyle::RTBS_TopLevelDefinitions) && 2457 Line.Level > 0) 2458 return false; 2459 2460 switch (Style.AlwaysBreakAfterReturnType) { 2461 case FormatStyle::RTBS_None: 2462 return false; 2463 case FormatStyle::RTBS_All: 2464 case FormatStyle::RTBS_TopLevel: 2465 return true; 2466 case FormatStyle::RTBS_AllDefinitions: 2467 case FormatStyle::RTBS_TopLevelDefinitions: 2468 return Line.mightBeFunctionDefinition(); 2469 } 2470 2471 return false; 2472 } 2473 2474 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 2475 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2476 E = Line.Children.end(); 2477 I != E; ++I) { 2478 calculateFormattingInformation(**I); 2479 } 2480 2481 Line.First->TotalLength = 2482 Line.First->IsMultiline ? Style.ColumnLimit 2483 : Line.FirstStartColumn + Line.First->ColumnWidth; 2484 FormatToken *Current = Line.First->Next; 2485 bool InFunctionDecl = Line.MightBeFunctionDecl; 2486 while (Current) { 2487 if (isFunctionDeclarationName(*Current, Line)) 2488 Current->setType(TT_FunctionDeclarationName); 2489 if (Current->is(TT_LineComment)) { 2490 if (Current->Previous->is(BK_BracedInit) && 2491 Current->Previous->opensScope()) 2492 Current->SpacesRequiredBefore = 2493 (Style.Cpp11BracedListStyle && !Style.SpacesInParentheses) ? 0 : 1; 2494 else 2495 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2496 2497 // If we find a trailing comment, iterate backwards to determine whether 2498 // it seems to relate to a specific parameter. If so, break before that 2499 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2500 // to the previous line in: 2501 // SomeFunction(a, 2502 // b, // comment 2503 // c); 2504 if (!Current->HasUnescapedNewline) { 2505 for (FormatToken *Parameter = Current->Previous; Parameter; 2506 Parameter = Parameter->Previous) { 2507 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 2508 break; 2509 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 2510 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 2511 Parameter->HasUnescapedNewline) 2512 Parameter->MustBreakBefore = true; 2513 break; 2514 } 2515 } 2516 } 2517 } else if (Current->SpacesRequiredBefore == 0 && 2518 spaceRequiredBefore(Line, *Current)) { 2519 Current->SpacesRequiredBefore = 1; 2520 } 2521 2522 Current->MustBreakBefore = 2523 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 2524 2525 if (!Current->MustBreakBefore && InFunctionDecl && 2526 Current->is(TT_FunctionDeclarationName)) 2527 Current->MustBreakBefore = mustBreakForReturnType(Line); 2528 2529 Current->CanBreakBefore = 2530 Current->MustBreakBefore || canBreakBefore(Line, *Current); 2531 unsigned ChildSize = 0; 2532 if (Current->Previous->Children.size() == 1) { 2533 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last; 2534 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 2535 : LastOfChild.TotalLength + 1; 2536 } 2537 const FormatToken *Prev = Current->Previous; 2538 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 2539 (Prev->Children.size() == 1 && 2540 Prev->Children[0]->First->MustBreakBefore) || 2541 Current->IsMultiline) 2542 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 2543 else 2544 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 2545 ChildSize + Current->SpacesRequiredBefore; 2546 2547 if (Current->is(TT_CtorInitializerColon)) 2548 InFunctionDecl = false; 2549 2550 // FIXME: Only calculate this if CanBreakBefore is true once static 2551 // initializers etc. are sorted out. 2552 // FIXME: Move magic numbers to a better place. 2553 2554 // Reduce penalty for aligning ObjC method arguments using the colon 2555 // alignment as this is the canonical way (still prefer fitting everything 2556 // into one line if possible). Trying to fit a whole expression into one 2557 // line should not force other line breaks (e.g. when ObjC method 2558 // expression is a part of other expression). 2559 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl); 2560 if (Style.Language == FormatStyle::LK_ObjC && 2561 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) { 2562 if (Current->ParameterIndex == 1) 2563 Current->SplitPenalty += 5 * Current->BindingStrength; 2564 } else { 2565 Current->SplitPenalty += 20 * Current->BindingStrength; 2566 } 2567 2568 Current = Current->Next; 2569 } 2570 2571 calculateUnbreakableTailLengths(Line); 2572 unsigned IndentLevel = Line.Level; 2573 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2574 if (Current->Role) 2575 Current->Role->precomputeFormattingInfos(Current); 2576 if (Current->MatchingParen && 2577 Current->MatchingParen->opensBlockOrBlockTypeList(Style)) { 2578 assert(IndentLevel > 0); 2579 --IndentLevel; 2580 } 2581 Current->IndentLevel = IndentLevel; 2582 if (Current->opensBlockOrBlockTypeList(Style)) 2583 ++IndentLevel; 2584 } 2585 2586 LLVM_DEBUG({ printDebugInfo(Line); }); 2587 } 2588 2589 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2590 unsigned UnbreakableTailLength = 0; 2591 FormatToken *Current = Line.Last; 2592 while (Current) { 2593 Current->UnbreakableTailLength = UnbreakableTailLength; 2594 if (Current->CanBreakBefore || 2595 Current->isOneOf(tok::comment, tok::string_literal)) { 2596 UnbreakableTailLength = 0; 2597 } else { 2598 UnbreakableTailLength += 2599 Current->ColumnWidth + Current->SpacesRequiredBefore; 2600 } 2601 Current = Current->Previous; 2602 } 2603 } 2604 2605 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2606 const FormatToken &Tok, 2607 bool InFunctionDecl) { 2608 const FormatToken &Left = *Tok.Previous; 2609 const FormatToken &Right = Tok; 2610 2611 if (Left.is(tok::semi)) 2612 return 0; 2613 2614 if (Style.Language == FormatStyle::LK_Java) { 2615 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2616 return 1; 2617 if (Right.is(Keywords.kw_implements)) 2618 return 2; 2619 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2620 return 3; 2621 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2622 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2623 return 100; 2624 if (Left.is(TT_JsTypeColon)) 2625 return 35; 2626 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2627 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2628 return 100; 2629 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2630 if (Left.opensScope() && Right.closesScope()) 2631 return 200; 2632 } 2633 2634 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2635 return 1; 2636 if (Right.is(tok::l_square)) { 2637 if (Style.Language == FormatStyle::LK_Proto) 2638 return 1; 2639 if (Left.is(tok::r_square)) 2640 return 200; 2641 // Slightly prefer formatting local lambda definitions like functions. 2642 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2643 return 35; 2644 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2645 TT_ArrayInitializerLSquare, 2646 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) 2647 return 500; 2648 } 2649 2650 if (Left.is(tok::coloncolon) || 2651 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2652 return 500; 2653 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2654 Right.is(tok::kw_operator)) { 2655 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2656 return 3; 2657 if (Left.is(TT_StartOfName)) 2658 return 110; 2659 if (InFunctionDecl && Right.NestingLevel == 0) 2660 return Style.PenaltyReturnTypeOnItsOwnLine; 2661 return 200; 2662 } 2663 if (Right.is(TT_PointerOrReference)) 2664 return 190; 2665 if (Right.is(TT_LambdaArrow)) 2666 return 110; 2667 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2668 return 160; 2669 if (Left.is(TT_CastRParen)) 2670 return 100; 2671 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2672 return 5000; 2673 if (Left.is(tok::comment)) 2674 return 1000; 2675 2676 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2677 TT_CtorInitializerColon)) 2678 return 2; 2679 2680 if (Right.isMemberAccess()) { 2681 // Breaking before the "./->" of a chained call/member access is reasonably 2682 // cheap, as formatting those with one call per line is generally 2683 // desirable. In particular, it should be cheaper to break before the call 2684 // than it is to break inside a call's parameters, which could lead to weird 2685 // "hanging" indents. The exception is the very last "./->" to support this 2686 // frequent pattern: 2687 // 2688 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2689 // dddddddd); 2690 // 2691 // which might otherwise be blown up onto many lines. Here, clang-format 2692 // won't produce "hanging" indents anyway as there is no other trailing 2693 // call. 2694 // 2695 // Also apply higher penalty is not a call as that might lead to a wrapping 2696 // like: 2697 // 2698 // aaaaaaa 2699 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2700 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2701 ? 150 2702 : 35; 2703 } 2704 2705 if (Right.is(TT_TrailingAnnotation) && 2706 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2707 // Moving trailing annotations to the next line is fine for ObjC method 2708 // declarations. 2709 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2710 return 10; 2711 // Generally, breaking before a trailing annotation is bad unless it is 2712 // function-like. It seems to be especially preferable to keep standard 2713 // annotations (i.e. "const", "final" and "override") on the same line. 2714 // Use a slightly higher penalty after ")" so that annotations like 2715 // "const override" are kept together. 2716 bool is_short_annotation = Right.TokenText.size() < 10; 2717 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 2718 } 2719 2720 // In for-loops, prefer breaking at ',' and ';'. 2721 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 2722 return 4; 2723 2724 // In Objective-C method expressions, prefer breaking before "param:" over 2725 // breaking after it. 2726 if (Right.is(TT_SelectorName)) 2727 return 0; 2728 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 2729 return Line.MightBeFunctionDecl ? 50 : 500; 2730 2731 // In Objective-C type declarations, avoid breaking after the category's 2732 // open paren (we'll prefer breaking after the protocol list's opening 2733 // angle bracket, if present). 2734 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous && 2735 Left.Previous->isOneOf(tok::identifier, tok::greater)) 2736 return 500; 2737 2738 if (Left.is(tok::l_paren) && InFunctionDecl && 2739 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 2740 return 100; 2741 if (Left.is(tok::l_paren) && Left.Previous && 2742 (Left.Previous->is(tok::kw_for) || Left.Previous->isIf())) 2743 return 1000; 2744 if (Left.is(tok::equal) && InFunctionDecl) 2745 return 110; 2746 if (Right.is(tok::r_brace)) 2747 return 1; 2748 if (Left.is(TT_TemplateOpener)) 2749 return 100; 2750 if (Left.opensScope()) { 2751 // If we aren't aligning after opening parens/braces we can always break 2752 // here unless the style does not want us to place all arguments on the 2753 // next line. 2754 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign && 2755 (Left.ParameterCount <= 1 || Style.AllowAllArgumentsOnNextLine)) 2756 return 0; 2757 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle) 2758 return 19; 2759 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 2760 : 19; 2761 } 2762 if (Left.is(TT_JavaAnnotation)) 2763 return 50; 2764 2765 if (Left.is(TT_UnaryOperator)) 2766 return 60; 2767 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 2768 Left.Previous->isLabelString() && 2769 (Left.NextOperator || Left.OperatorIndex != 0)) 2770 return 50; 2771 if (Right.is(tok::plus) && Left.isLabelString() && 2772 (Right.NextOperator || Right.OperatorIndex != 0)) 2773 return 25; 2774 if (Left.is(tok::comma)) 2775 return 1; 2776 if (Right.is(tok::lessless) && Left.isLabelString() && 2777 (Right.NextOperator || Right.OperatorIndex != 1)) 2778 return 25; 2779 if (Right.is(tok::lessless)) { 2780 // Breaking at a << is really cheap. 2781 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 2782 // Slightly prefer to break before the first one in log-like statements. 2783 return 2; 2784 return 1; 2785 } 2786 if (Left.ClosesTemplateDeclaration) 2787 return Style.PenaltyBreakTemplateDeclaration; 2788 if (Left.is(TT_ConditionalExpr)) 2789 return prec::Conditional; 2790 prec::Level Level = Left.getPrecedence(); 2791 if (Level == prec::Unknown) 2792 Level = Right.getPrecedence(); 2793 if (Level == prec::Assignment) 2794 return Style.PenaltyBreakAssignment; 2795 if (Level != prec::Unknown) 2796 return Level; 2797 2798 return 3; 2799 } 2800 2801 bool TokenAnnotator::spaceRequiredBeforeParens(const FormatToken &Right) const { 2802 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always || 2803 (Style.SpaceBeforeParens == FormatStyle::SBPO_NonEmptyParentheses && 2804 Right.ParameterCount > 0); 2805 } 2806 2807 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 2808 const FormatToken &Left, 2809 const FormatToken &Right) { 2810 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 2811 return true; 2812 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 2813 return true; 2814 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 2815 Left.Tok.getObjCKeywordID() == tok::objc_property) 2816 return true; 2817 if (Right.is(tok::hashhash)) 2818 return Left.is(tok::hash); 2819 if (Left.isOneOf(tok::hashhash, tok::hash)) 2820 return Right.is(tok::hash); 2821 if ((Left.is(tok::l_paren) && Right.is(tok::r_paren)) || 2822 (Left.is(tok::l_brace) && Left.isNot(BK_Block) && 2823 Right.is(tok::r_brace) && Right.isNot(BK_Block))) 2824 return Style.SpaceInEmptyParentheses; 2825 if (Style.SpacesInConditionalStatement) { 2826 if (Left.is(tok::l_paren) && Left.Previous && 2827 isKeywordWithCondition(*Left.Previous)) 2828 return true; 2829 if (Right.is(tok::r_paren) && Right.MatchingParen && 2830 Right.MatchingParen->Previous && 2831 isKeywordWithCondition(*Right.MatchingParen->Previous)) 2832 return true; 2833 } 2834 2835 // requires ( or requires( 2836 if (Right.is(tok::l_paren) && Left.is(tok::kw_requires)) 2837 return spaceRequiredBeforeParens(Right); 2838 // requires clause Concept1<T> && Concept2<T> 2839 if (Left.is(TT_ConstraintJunctions) && Right.is(tok::identifier)) 2840 return true; 2841 2842 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 2843 return (Right.is(TT_CastRParen) || 2844 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 2845 ? Style.SpacesInCStyleCastParentheses 2846 : Style.SpacesInParentheses; 2847 if (Right.isOneOf(tok::semi, tok::comma)) 2848 return false; 2849 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) { 2850 bool IsLightweightGeneric = Right.MatchingParen && 2851 Right.MatchingParen->Next && 2852 Right.MatchingParen->Next->is(tok::colon); 2853 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList; 2854 } 2855 if (Right.is(tok::less) && Left.is(tok::kw_template)) 2856 return Style.SpaceAfterTemplateKeyword; 2857 if (Left.isOneOf(tok::exclaim, tok::tilde)) 2858 return false; 2859 if (Left.is(tok::at) && 2860 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 2861 tok::numeric_constant, tok::l_paren, tok::l_brace, 2862 tok::kw_true, tok::kw_false)) 2863 return false; 2864 if (Left.is(tok::colon)) 2865 return !Left.is(TT_ObjCMethodExpr); 2866 if (Left.is(tok::coloncolon)) 2867 return false; 2868 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 2869 if (Style.Language == FormatStyle::LK_TextProto || 2870 (Style.Language == FormatStyle::LK_Proto && 2871 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 2872 // Format empty list as `<>`. 2873 if (Left.is(tok::less) && Right.is(tok::greater)) 2874 return false; 2875 return !Style.Cpp11BracedListStyle; 2876 } 2877 return false; 2878 } 2879 if (Right.is(tok::ellipsis)) 2880 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 2881 Left.Previous->is(tok::kw_case)); 2882 if (Left.is(tok::l_square) && Right.is(tok::amp)) 2883 return Style.SpacesInSquareBrackets; 2884 if (Right.is(TT_PointerOrReference)) { 2885 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 2886 if (!Left.MatchingParen) 2887 return true; 2888 FormatToken *TokenBeforeMatchingParen = 2889 Left.MatchingParen->getPreviousNonComment(); 2890 if (!TokenBeforeMatchingParen || !Left.is(TT_TypeDeclarationParen)) 2891 return true; 2892 } 2893 // Add a space if the previous token is a pointer qualifer or the closing 2894 // parenthesis of __attribute__(()) expression and the style requires spaces 2895 // after pointer qualifiers. 2896 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_After || 2897 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 2898 (Left.is(TT_AttributeParen) || Left.canBePointerOrReferenceQualifier())) 2899 return true; 2900 return (Left.Tok.isLiteral() || 2901 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 2902 (Style.PointerAlignment != FormatStyle::PAS_Left || 2903 (Line.IsMultiVariableDeclStmt && 2904 (Left.NestingLevel == 0 || 2905 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for))))))); 2906 } 2907 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 2908 (!Left.is(TT_PointerOrReference) || 2909 (Style.PointerAlignment != FormatStyle::PAS_Right && 2910 !Line.IsMultiVariableDeclStmt))) 2911 return true; 2912 if (Left.is(TT_PointerOrReference)) { 2913 // Add a space if the next token is a pointer qualifer and the style 2914 // requires spaces before pointer qualifiers. 2915 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Before || 2916 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 2917 Right.canBePointerOrReferenceQualifier()) 2918 return true; 2919 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) || 2920 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 2921 !Right.is(TT_StartOfName)) || 2922 (Right.is(tok::l_brace) && Right.is(BK_Block)) || 2923 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 2924 tok::l_paren) && 2925 (Style.PointerAlignment != FormatStyle::PAS_Right && 2926 !Line.IsMultiVariableDeclStmt) && 2927 Left.Previous && 2928 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon, 2929 tok::l_square)); 2930 } 2931 // Ensure right pointer alignement with ellipsis e.g. int *...P 2932 if (Left.is(tok::ellipsis) && Left.Previous && 2933 Left.Previous->isOneOf(tok::star, tok::amp, tok::ampamp)) 2934 return Style.PointerAlignment != FormatStyle::PAS_Right; 2935 2936 if (Right.is(tok::star) && Left.is(tok::l_paren)) 2937 return false; 2938 if (Left.is(tok::star) && Right.isOneOf(tok::star, tok::amp, tok::ampamp)) 2939 return false; 2940 if (Right.isOneOf(tok::star, tok::amp, tok::ampamp)) { 2941 const FormatToken *Previous = &Left; 2942 while (Previous && !Previous->is(tok::kw_operator)) { 2943 if (Previous->is(tok::identifier) || Previous->isSimpleTypeSpecifier()) { 2944 Previous = Previous->getPreviousNonComment(); 2945 continue; 2946 } 2947 if (Previous->is(TT_TemplateCloser) && Previous->MatchingParen) { 2948 Previous = Previous->MatchingParen->getPreviousNonComment(); 2949 continue; 2950 } 2951 if (Previous->is(tok::coloncolon)) { 2952 Previous = Previous->getPreviousNonComment(); 2953 continue; 2954 } 2955 break; 2956 } 2957 // Space between the type and the * in: 2958 // operator void*() 2959 // operator char*() 2960 // operator /*comment*/ const char*() 2961 // operator volatile /*comment*/ char*() 2962 // operator Foo*() 2963 // operator C<T>*() 2964 // operator std::Foo*() 2965 // operator C<T>::D<U>*() 2966 // dependent on PointerAlignment style. 2967 if (Previous && 2968 (Previous->endsSequence(tok::kw_operator) || 2969 Previous->endsSequence(tok::kw_const, tok::kw_operator) || 2970 Previous->endsSequence(tok::kw_volatile, tok::kw_operator))) 2971 return (Style.PointerAlignment != FormatStyle::PAS_Left); 2972 } 2973 const auto SpaceRequiredForArrayInitializerLSquare = 2974 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 2975 return Style.SpacesInContainerLiterals || 2976 ((Style.Language == FormatStyle::LK_Proto || 2977 Style.Language == FormatStyle::LK_TextProto) && 2978 !Style.Cpp11BracedListStyle && 2979 LSquareTok.endsSequence(tok::l_square, tok::colon, 2980 TT_SelectorName)); 2981 }; 2982 if (Left.is(tok::l_square)) 2983 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 2984 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 2985 (Left.isOneOf(TT_ArraySubscriptLSquare, TT_StructuredBindingLSquare, 2986 TT_LambdaLSquare) && 2987 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 2988 if (Right.is(tok::r_square)) 2989 return Right.MatchingParen && 2990 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 2991 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 2992 Style)) || 2993 (Style.SpacesInSquareBrackets && 2994 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 2995 TT_StructuredBindingLSquare, 2996 TT_LambdaLSquare)) || 2997 Right.MatchingParen->is(TT_AttributeParen)); 2998 if (Right.is(tok::l_square) && 2999 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 3000 TT_DesignatedInitializerLSquare, 3001 TT_StructuredBindingLSquare, TT_AttributeSquare) && 3002 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral) && 3003 !(!Left.is(tok::r_square) && Style.SpaceBeforeSquareBrackets && 3004 Right.is(TT_ArraySubscriptLSquare))) 3005 return false; 3006 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 3007 return !Left.Children.empty(); // No spaces in "{}". 3008 if ((Left.is(tok::l_brace) && Left.isNot(BK_Block)) || 3009 (Right.is(tok::r_brace) && Right.MatchingParen && 3010 Right.MatchingParen->isNot(BK_Block))) 3011 return Style.Cpp11BracedListStyle ? Style.SpacesInParentheses : true; 3012 if (Left.is(TT_BlockComment)) 3013 // No whitespace in x(/*foo=*/1), except for JavaScript. 3014 return Style.Language == FormatStyle::LK_JavaScript || 3015 !Left.TokenText.endswith("=*/"); 3016 3017 // Space between template and attribute. 3018 // e.g. template <typename T> [[nodiscard]] ... 3019 if (Left.is(TT_TemplateCloser) && Right.is(TT_AttributeSquare)) 3020 return true; 3021 if (Right.is(tok::l_paren)) { 3022 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 3023 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) 3024 return true; 3025 if (Style.SpaceBeforeParens == 3026 FormatStyle::SBPO_ControlStatementsExceptForEachMacros && 3027 Left.is(TT_ForEachMacro)) 3028 return false; 3029 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) || 3030 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never && 3031 (Left.isOneOf(tok::pp_elif, tok::kw_for, tok::kw_while, 3032 tok::kw_switch, tok::kw_case, TT_ForEachMacro, 3033 TT_ObjCForIn) || 3034 Left.isIf(Line.Type != LT_PreprocessorDirective) || 3035 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch, 3036 tok::kw_new, tok::kw_delete) && 3037 (!Left.Previous || Left.Previous->isNot(tok::period))))) || 3038 (spaceRequiredBeforeParens(Right) && 3039 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 3040 Left.is(tok::r_paren) || Left.isSimpleTypeSpecifier() || 3041 (Left.is(tok::r_square) && Left.MatchingParen && 3042 Left.MatchingParen->is(TT_LambdaLSquare))) && 3043 Line.Type != LT_PreprocessorDirective); 3044 } 3045 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 3046 return false; 3047 if (Right.is(TT_UnaryOperator)) 3048 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 3049 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 3050 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 3051 tok::r_paren) || 3052 Left.isSimpleTypeSpecifier()) && 3053 Right.is(tok::l_brace) && Right.getNextNonComment() && 3054 Right.isNot(BK_Block)) 3055 return false; 3056 if (Left.is(tok::period) || Right.is(tok::period)) 3057 return false; 3058 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L") 3059 return false; 3060 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 3061 Left.MatchingParen->Previous && 3062 (Left.MatchingParen->Previous->is(tok::period) || 3063 Left.MatchingParen->Previous->is(tok::coloncolon))) 3064 // Java call to generic function with explicit type: 3065 // A.<B<C<...>>>DoSomething(); 3066 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference. 3067 return false; 3068 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 3069 return false; 3070 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at)) 3071 // Objective-C dictionary literal -> no space after opening brace. 3072 return false; 3073 if (Right.is(tok::r_brace) && Right.MatchingParen && 3074 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at)) 3075 // Objective-C dictionary literal -> no space before closing brace. 3076 return false; 3077 if (Right.getType() == TT_TrailingAnnotation && 3078 Right.isOneOf(tok::amp, tok::ampamp) && 3079 Left.isOneOf(tok::kw_const, tok::kw_volatile) && 3080 (!Right.Next || Right.Next->is(tok::semi))) 3081 // Match const and volatile ref-qualifiers without any additional 3082 // qualifiers such as 3083 // void Fn() const &; 3084 return Style.PointerAlignment != FormatStyle::PAS_Left; 3085 return true; 3086 } 3087 3088 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 3089 const FormatToken &Right) { 3090 const FormatToken &Left = *Right.Previous; 3091 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 3092 return true; // Never ever merge two identifiers. 3093 if (Style.isCpp()) { 3094 if (Left.is(tok::kw_operator)) 3095 return Right.is(tok::coloncolon); 3096 if (Right.is(tok::l_brace) && Right.is(BK_BracedInit) && 3097 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList) 3098 return true; 3099 } else if (Style.Language == FormatStyle::LK_Proto || 3100 Style.Language == FormatStyle::LK_TextProto) { 3101 if (Right.is(tok::period) && 3102 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 3103 Keywords.kw_repeated, Keywords.kw_extend)) 3104 return true; 3105 if (Right.is(tok::l_paren) && 3106 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 3107 return true; 3108 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 3109 return true; 3110 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 3111 if (Left.is(tok::slash) || Right.is(tok::slash)) 3112 return false; 3113 if (Left.MatchingParen && 3114 Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 3115 Right.isOneOf(tok::l_brace, tok::less)) 3116 return !Style.Cpp11BracedListStyle; 3117 // A percent is probably part of a formatting specification, such as %lld. 3118 if (Left.is(tok::percent)) 3119 return false; 3120 // Preserve the existence of a space before a percent for cases like 0x%04x 3121 // and "%d %d" 3122 if (Left.is(tok::numeric_constant) && Right.is(tok::percent)) 3123 return Right.WhitespaceRange.getEnd() != Right.WhitespaceRange.getBegin(); 3124 } else if (Style.isCSharp()) { 3125 // Require spaces around '{' and before '}' unless they appear in 3126 // interpolated strings. Interpolated strings are merged into a single token 3127 // so cannot have spaces inserted by this function. 3128 3129 // No space between 'this' and '[' 3130 if (Left.is(tok::kw_this) && Right.is(tok::l_square)) 3131 return false; 3132 3133 // No space between 'new' and '(' 3134 if (Left.is(tok::kw_new) && Right.is(tok::l_paren)) 3135 return false; 3136 3137 // Space before { (including space within '{ {'). 3138 if (Right.is(tok::l_brace)) 3139 return true; 3140 3141 // Spaces inside braces. 3142 if (Left.is(tok::l_brace) && Right.isNot(tok::r_brace)) 3143 return true; 3144 3145 if (Left.isNot(tok::l_brace) && Right.is(tok::r_brace)) 3146 return true; 3147 3148 // Spaces around '=>'. 3149 if (Left.is(TT_JsFatArrow) || Right.is(TT_JsFatArrow)) 3150 return true; 3151 3152 // No spaces around attribute target colons 3153 if (Left.is(TT_AttributeColon) || Right.is(TT_AttributeColon)) 3154 return false; 3155 3156 // space between type and variable e.g. Dictionary<string,string> foo; 3157 if (Left.is(TT_TemplateCloser) && Right.is(TT_StartOfName)) 3158 return true; 3159 3160 // spaces inside square brackets. 3161 if (Left.is(tok::l_square) || Right.is(tok::r_square)) 3162 return Style.SpacesInSquareBrackets; 3163 3164 // No space before ? in nullable types. 3165 if (Right.is(TT_CSharpNullable)) 3166 return false; 3167 3168 // Require space after ? in nullable types except in generics and casts. 3169 if (Left.is(TT_CSharpNullable)) 3170 return !Right.isOneOf(TT_TemplateCloser, tok::r_paren); 3171 3172 // No space before or after '?.'. 3173 if (Left.is(TT_CSharpNullConditional) || Right.is(TT_CSharpNullConditional)) 3174 return false; 3175 3176 // Space before and after '??'. 3177 if (Left.is(TT_CSharpNullCoalescing) || Right.is(TT_CSharpNullCoalescing)) 3178 return true; 3179 3180 // No space before '?['. 3181 if (Right.is(TT_CSharpNullConditionalLSquare)) 3182 return false; 3183 3184 // No space between consecutive commas '[,,]'. 3185 if (Left.is(tok::comma) && Right.is(tok::comma)) 3186 return false; 3187 3188 // Possible space inside `?[ 0 ]`. 3189 if (Left.is(TT_CSharpNullConditionalLSquare)) 3190 return Style.SpacesInSquareBrackets; 3191 3192 // space after var in `var (key, value)` 3193 if (Left.is(Keywords.kw_var) && Right.is(tok::l_paren)) 3194 return true; 3195 3196 // space between keywords and paren e.g. "using (" 3197 if (Right.is(tok::l_paren)) 3198 if (Left.isOneOf(tok::kw_using, Keywords.kw_async, Keywords.kw_when, 3199 Keywords.kw_lock)) 3200 return Style.SpaceBeforeParens == FormatStyle::SBPO_ControlStatements || 3201 spaceRequiredBeforeParens(Right); 3202 3203 // space between method modifier and opening parenthesis of a tuple return 3204 // type 3205 if (Left.isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 3206 tok::kw_virtual, tok::kw_extern, tok::kw_static, 3207 Keywords.kw_internal, Keywords.kw_abstract, 3208 Keywords.kw_sealed, Keywords.kw_override, 3209 Keywords.kw_async, Keywords.kw_unsafe) && 3210 Right.is(tok::l_paren)) 3211 return true; 3212 } else if (Style.Language == FormatStyle::LK_JavaScript) { 3213 if (Left.is(TT_JsFatArrow)) 3214 return true; 3215 // for await ( ... 3216 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 3217 Left.Previous->is(tok::kw_for)) 3218 return true; 3219 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 3220 Right.MatchingParen) { 3221 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 3222 // An async arrow function, for example: `x = async () => foo();`, 3223 // as opposed to calling a function called async: `x = async();` 3224 if (Next && Next->is(TT_JsFatArrow)) 3225 return true; 3226 } 3227 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 3228 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 3229 return false; 3230 // In tagged template literals ("html`bar baz`"), there is no space between 3231 // the tag identifier and the template string. 3232 if (Keywords.IsJavaScriptIdentifier(Left, 3233 /* AcceptIdentifierName= */ false) && 3234 Right.is(TT_TemplateString)) 3235 return false; 3236 if (Right.is(tok::star) && 3237 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 3238 return false; 3239 if (Right.isOneOf(tok::l_brace, tok::l_square) && 3240 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 3241 Keywords.kw_extends, Keywords.kw_implements)) 3242 return true; 3243 if (Right.is(tok::l_paren)) { 3244 // JS methods can use some keywords as names (e.g. `delete()`). 3245 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 3246 return false; 3247 // Valid JS method names can include keywords, e.g. `foo.delete()` or 3248 // `bar.instanceof()`. Recognize call positions by preceding period. 3249 if (Left.Previous && Left.Previous->is(tok::period) && 3250 Left.Tok.getIdentifierInfo()) 3251 return false; 3252 // Additional unary JavaScript operators that need a space after. 3253 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 3254 tok::kw_void)) 3255 return true; 3256 } 3257 // `foo as const;` casts into a const type. 3258 if (Left.endsSequence(tok::kw_const, Keywords.kw_as)) { 3259 return false; 3260 } 3261 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 3262 tok::kw_const) || 3263 // "of" is only a keyword if it appears after another identifier 3264 // (e.g. as "const x of y" in a for loop), or after a destructuring 3265 // operation (const [x, y] of z, const {a, b} of c). 3266 (Left.is(Keywords.kw_of) && Left.Previous && 3267 (Left.Previous->Tok.is(tok::identifier) || 3268 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 3269 (!Left.Previous || !Left.Previous->is(tok::period))) 3270 return true; 3271 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 3272 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 3273 return false; 3274 if (Left.is(Keywords.kw_as) && 3275 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 3276 return true; 3277 if (Left.is(tok::kw_default) && Left.Previous && 3278 Left.Previous->is(tok::kw_export)) 3279 return true; 3280 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 3281 return true; 3282 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 3283 return false; 3284 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 3285 return false; 3286 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 3287 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 3288 return false; 3289 if (Left.is(tok::ellipsis)) 3290 return false; 3291 if (Left.is(TT_TemplateCloser) && 3292 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 3293 Keywords.kw_implements, Keywords.kw_extends)) 3294 // Type assertions ('<type>expr') are not followed by whitespace. Other 3295 // locations that should have whitespace following are identified by the 3296 // above set of follower tokens. 3297 return false; 3298 if (Right.is(TT_JsNonNullAssertion)) 3299 return false; 3300 if (Left.is(TT_JsNonNullAssertion) && 3301 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 3302 return true; // "x! as string", "x! in y" 3303 } else if (Style.Language == FormatStyle::LK_Java) { 3304 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 3305 return true; 3306 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 3307 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never; 3308 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 3309 tok::kw_protected) || 3310 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 3311 Keywords.kw_native)) && 3312 Right.is(TT_TemplateOpener)) 3313 return true; 3314 } 3315 if (Left.is(TT_ImplicitStringLiteral)) 3316 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 3317 if (Line.Type == LT_ObjCMethodDecl) { 3318 if (Left.is(TT_ObjCMethodSpecifier)) 3319 return true; 3320 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right)) 3321 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a 3322 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class 3323 // method declaration. 3324 return false; 3325 } 3326 if (Line.Type == LT_ObjCProperty && 3327 (Right.is(tok::equal) || Left.is(tok::equal))) 3328 return false; 3329 3330 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 3331 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 3332 return true; 3333 if (Right.is(TT_OverloadedOperatorLParen)) 3334 return spaceRequiredBeforeParens(Right); 3335 if (Left.is(tok::comma)) 3336 return true; 3337 if (Right.is(tok::comma)) 3338 return false; 3339 if (Right.is(TT_ObjCBlockLParen)) 3340 return true; 3341 if (Right.is(TT_CtorInitializerColon)) 3342 return Style.SpaceBeforeCtorInitializerColon; 3343 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 3344 return false; 3345 if (Right.is(TT_RangeBasedForLoopColon) && 3346 !Style.SpaceBeforeRangeBasedForLoopColon) 3347 return false; 3348 if (Left.is(TT_BitFieldColon)) 3349 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 3350 Style.BitFieldColonSpacing == FormatStyle::BFCS_After; 3351 if (Right.is(tok::colon)) { 3352 if (Line.First->isOneOf(tok::kw_default, tok::kw_case)) 3353 return Style.SpaceBeforeCaseColon; 3354 if (!Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 3355 return false; 3356 if (Right.is(TT_ObjCMethodExpr)) 3357 return false; 3358 if (Left.is(tok::question)) 3359 return false; 3360 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 3361 return false; 3362 if (Right.is(TT_DictLiteral)) 3363 return Style.SpacesInContainerLiterals; 3364 if (Right.is(TT_AttributeColon)) 3365 return false; 3366 if (Right.is(TT_CSharpNamedArgumentColon)) 3367 return false; 3368 if (Right.is(TT_BitFieldColon)) 3369 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 3370 Style.BitFieldColonSpacing == FormatStyle::BFCS_Before; 3371 return true; 3372 } 3373 if (Left.is(TT_UnaryOperator)) { 3374 if (!Right.is(tok::l_paren)) { 3375 // The alternative operators for ~ and ! are "compl" and "not". 3376 // If they are used instead, we do not want to combine them with 3377 // the token to the right, unless that is a left paren. 3378 if (Left.is(tok::exclaim) && Left.TokenText == "not") 3379 return true; 3380 if (Left.is(tok::tilde) && Left.TokenText == "compl") 3381 return true; 3382 // Lambda captures allow for a lone &, so "&]" needs to be properly 3383 // handled. 3384 if (Left.is(tok::amp) && Right.is(tok::r_square)) 3385 return Style.SpacesInSquareBrackets; 3386 } 3387 return (Style.SpaceAfterLogicalNot && Left.is(tok::exclaim)) || 3388 Right.is(TT_BinaryOperator); 3389 } 3390 3391 // If the next token is a binary operator or a selector name, we have 3392 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 3393 if (Left.is(TT_CastRParen)) 3394 return Style.SpaceAfterCStyleCast || 3395 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 3396 3397 if (Left.is(tok::greater) && Right.is(tok::greater)) { 3398 if (Style.Language == FormatStyle::LK_TextProto || 3399 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 3400 return !Style.Cpp11BracedListStyle; 3401 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 3402 (Style.Standard < FormatStyle::LS_Cpp11 || Style.SpacesInAngles); 3403 } 3404 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 3405 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 3406 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 3407 return false; 3408 if (!Style.SpaceBeforeAssignmentOperators && Left.isNot(TT_TemplateCloser) && 3409 Right.getPrecedence() == prec::Assignment) 3410 return false; 3411 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) && 3412 (Left.is(tok::identifier) || Left.is(tok::kw_this))) 3413 return false; 3414 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 3415 // Generally don't remove existing spaces between an identifier and "::". 3416 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 3417 // this turns out to be too lenient, add analysis of the identifier itself. 3418 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 3419 if (Right.is(tok::coloncolon) && 3420 !Left.isOneOf(tok::l_brace, tok::comment, tok::l_paren)) 3421 // Put a space between < and :: in vector< ::std::string > 3422 return (Left.is(TT_TemplateOpener) && 3423 (Style.Standard < FormatStyle::LS_Cpp11 || Style.SpacesInAngles)) || 3424 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 3425 tok::kw___super, TT_TemplateOpener, 3426 TT_TemplateCloser)) || 3427 (Left.is(tok::l_paren) && Style.SpacesInParentheses); 3428 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 3429 return Style.SpacesInAngles; 3430 // Space before TT_StructuredBindingLSquare. 3431 if (Right.is(TT_StructuredBindingLSquare)) 3432 return !Left.isOneOf(tok::amp, tok::ampamp) || 3433 Style.PointerAlignment != FormatStyle::PAS_Right; 3434 // Space before & or && following a TT_StructuredBindingLSquare. 3435 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 3436 Right.isOneOf(tok::amp, tok::ampamp)) 3437 return Style.PointerAlignment != FormatStyle::PAS_Left; 3438 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 3439 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 3440 !Right.is(tok::r_paren))) 3441 return true; 3442 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) && 3443 Right.isNot(TT_FunctionTypeLParen)) 3444 return spaceRequiredBeforeParens(Right); 3445 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 3446 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 3447 return false; 3448 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 3449 Line.startsWith(tok::hash)) 3450 return true; 3451 if (Right.is(TT_TrailingUnaryOperator)) 3452 return false; 3453 if (Left.is(TT_RegexLiteral)) 3454 return false; 3455 return spaceRequiredBetween(Line, Left, Right); 3456 } 3457 3458 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 3459 static bool isAllmanBrace(const FormatToken &Tok) { 3460 return Tok.is(tok::l_brace) && Tok.is(BK_Block) && 3461 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_LambdaLBrace, TT_DictLiteral); 3462 } 3463 3464 // Returns 'true' if 'Tok' is an function argument. 3465 static bool IsFunctionArgument(const FormatToken &Tok) { 3466 return Tok.MatchingParen && Tok.MatchingParen->Next && 3467 Tok.MatchingParen->Next->isOneOf(tok::comma, tok::r_paren); 3468 } 3469 3470 static bool 3471 isItAnEmptyLambdaAllowed(const FormatToken &Tok, 3472 FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3473 return Tok.Children.empty() && ShortLambdaOption != FormatStyle::SLS_None; 3474 } 3475 3476 static bool 3477 isItAInlineLambdaAllowed(const FormatToken &Tok, 3478 FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3479 return (ShortLambdaOption == FormatStyle::SLS_Inline && 3480 IsFunctionArgument(Tok)) || 3481 (ShortLambdaOption == FormatStyle::SLS_All); 3482 } 3483 3484 static bool isOneChildWithoutMustBreakBefore(const FormatToken &Tok) { 3485 if (Tok.Children.size() != 1) 3486 return false; 3487 FormatToken *curElt = Tok.Children[0]->First; 3488 while (curElt) { 3489 if (curElt->MustBreakBefore) 3490 return false; 3491 curElt = curElt->Next; 3492 } 3493 return true; 3494 } 3495 static bool isAllmanLambdaBrace(const FormatToken &Tok) { 3496 return (Tok.is(tok::l_brace) && Tok.is(BK_Block) && 3497 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral)); 3498 } 3499 3500 static bool isAllmanBraceIncludedBreakableLambda( 3501 const FormatToken &Tok, FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3502 if (!isAllmanLambdaBrace(Tok)) 3503 return false; 3504 3505 if (isItAnEmptyLambdaAllowed(Tok, ShortLambdaOption)) 3506 return false; 3507 3508 return !isItAInlineLambdaAllowed(Tok, ShortLambdaOption) || 3509 !isOneChildWithoutMustBreakBefore(Tok); 3510 } 3511 3512 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 3513 const FormatToken &Right) { 3514 const FormatToken &Left = *Right.Previous; 3515 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 3516 return true; 3517 3518 if (Style.isCSharp()) { 3519 if (Right.is(TT_CSharpNamedArgumentColon) || 3520 Left.is(TT_CSharpNamedArgumentColon)) 3521 return false; 3522 if (Right.is(TT_CSharpGenericTypeConstraint)) 3523 return true; 3524 } else if (Style.Language == FormatStyle::LK_JavaScript) { 3525 // FIXME: This might apply to other languages and token kinds. 3526 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 3527 Left.Previous->is(tok::string_literal)) 3528 return true; 3529 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 3530 Left.Previous && Left.Previous->is(tok::equal) && 3531 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 3532 tok::kw_const) && 3533 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 3534 // above. 3535 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 3536 // Object literals on the top level of a file are treated as "enum-style". 3537 // Each key/value pair is put on a separate line, instead of bin-packing. 3538 return true; 3539 if (Left.is(tok::l_brace) && Line.Level == 0 && 3540 (Line.startsWith(tok::kw_enum) || 3541 Line.startsWith(tok::kw_const, tok::kw_enum) || 3542 Line.startsWith(tok::kw_export, tok::kw_enum) || 3543 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 3544 // JavaScript top-level enum key/value pairs are put on separate lines 3545 // instead of bin-packing. 3546 return true; 3547 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && Left.Previous && 3548 Left.Previous->is(TT_JsFatArrow)) { 3549 // JS arrow function (=> {...}). 3550 switch (Style.AllowShortLambdasOnASingleLine) { 3551 case FormatStyle::SLS_All: 3552 return false; 3553 case FormatStyle::SLS_None: 3554 return true; 3555 case FormatStyle::SLS_Empty: 3556 return !Left.Children.empty(); 3557 case FormatStyle::SLS_Inline: 3558 // allow one-lining inline (e.g. in function call args) and empty arrow 3559 // functions. 3560 return (Left.NestingLevel == 0 && Line.Level == 0) && 3561 !Left.Children.empty(); 3562 } 3563 llvm_unreachable("Unknown FormatStyle::ShortLambdaStyle enum"); 3564 } 3565 3566 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 3567 !Left.Children.empty()) 3568 // Support AllowShortFunctionsOnASingleLine for JavaScript. 3569 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 3570 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 3571 (Left.NestingLevel == 0 && Line.Level == 0 && 3572 Style.AllowShortFunctionsOnASingleLine & 3573 FormatStyle::SFS_InlineOnly); 3574 } else if (Style.Language == FormatStyle::LK_Java) { 3575 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 3576 Right.Next->is(tok::string_literal)) 3577 return true; 3578 } else if (Style.Language == FormatStyle::LK_Cpp || 3579 Style.Language == FormatStyle::LK_ObjC || 3580 Style.Language == FormatStyle::LK_Proto || 3581 Style.Language == FormatStyle::LK_TableGen || 3582 Style.Language == FormatStyle::LK_TextProto) { 3583 if (Left.isStringLiteral() && Right.isStringLiteral()) 3584 return true; 3585 } 3586 3587 // If the last token before a '}', ']', or ')' is a comma or a trailing 3588 // comment, the intention is to insert a line break after it in order to make 3589 // shuffling around entries easier. Import statements, especially in 3590 // JavaScript, can be an exception to this rule. 3591 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 3592 const FormatToken *BeforeClosingBrace = nullptr; 3593 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 3594 (Style.Language == FormatStyle::LK_JavaScript && 3595 Left.is(tok::l_paren))) && 3596 Left.isNot(BK_Block) && Left.MatchingParen) 3597 BeforeClosingBrace = Left.MatchingParen->Previous; 3598 else if (Right.MatchingParen && 3599 (Right.MatchingParen->isOneOf(tok::l_brace, 3600 TT_ArrayInitializerLSquare) || 3601 (Style.Language == FormatStyle::LK_JavaScript && 3602 Right.MatchingParen->is(tok::l_paren)))) 3603 BeforeClosingBrace = &Left; 3604 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 3605 BeforeClosingBrace->isTrailingComment())) 3606 return true; 3607 } 3608 3609 if (Right.is(tok::comment)) 3610 return Left.isNot(BK_BracedInit) && Left.isNot(TT_CtorInitializerColon) && 3611 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 3612 if (Left.isTrailingComment()) 3613 return true; 3614 if (Right.Previous->IsUnterminatedLiteral) 3615 return true; 3616 if (Right.is(tok::lessless) && Right.Next && 3617 Right.Previous->is(tok::string_literal) && 3618 Right.Next->is(tok::string_literal)) 3619 return true; 3620 // Can break after template<> declaration 3621 if (Right.Previous->ClosesTemplateDeclaration && 3622 Right.Previous->MatchingParen && 3623 Right.Previous->MatchingParen->NestingLevel == 0) { 3624 // Put concepts on the next line e.g. 3625 // template<typename T> 3626 // concept ... 3627 if (Right.is(tok::kw_concept)) 3628 return Style.BreakBeforeConceptDeclarations; 3629 return (Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes); 3630 } 3631 if (Right.is(TT_CtorInitializerComma) && 3632 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3633 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 3634 return true; 3635 if (Right.is(TT_CtorInitializerColon) && 3636 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3637 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 3638 return true; 3639 // Break only if we have multiple inheritance. 3640 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 3641 Right.is(TT_InheritanceComma)) 3642 return true; 3643 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 3644 // Multiline raw string literals are special wrt. line breaks. The author 3645 // has made a deliberate choice and might have aligned the contents of the 3646 // string literal accordingly. Thus, we try keep existing line breaks. 3647 return Right.IsMultiline && Right.NewlinesBefore > 0; 3648 if ((Right.Previous->is(tok::l_brace) || 3649 (Right.Previous->is(tok::less) && Right.Previous->Previous && 3650 Right.Previous->Previous->is(tok::equal))) && 3651 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 3652 // Don't put enums or option definitions onto single lines in protocol 3653 // buffers. 3654 return true; 3655 } 3656 if (Right.is(TT_InlineASMBrace)) 3657 return Right.HasUnescapedNewline; 3658 3659 auto ShortLambdaOption = Style.AllowShortLambdasOnASingleLine; 3660 if (Style.BraceWrapping.BeforeLambdaBody && 3661 (isAllmanBraceIncludedBreakableLambda(Left, ShortLambdaOption) || 3662 isAllmanBraceIncludedBreakableLambda(Right, ShortLambdaOption))) { 3663 return true; 3664 } 3665 3666 if (isAllmanBrace(Left) || isAllmanBrace(Right)) 3667 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) || 3668 (Line.startsWith(tok::kw_typedef, tok::kw_enum) && 3669 Style.BraceWrapping.AfterEnum) || 3670 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 3671 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 3672 if (Left.is(TT_ObjCBlockLBrace) && 3673 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 3674 return true; 3675 3676 if (Left.is(TT_LambdaLBrace)) { 3677 if (IsFunctionArgument(Left) && 3678 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline) 3679 return false; 3680 3681 if (Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_None || 3682 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline || 3683 (!Left.Children.empty() && 3684 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Empty)) 3685 return true; 3686 } 3687 3688 // Put multiple Java annotation on a new line. 3689 if ((Style.Language == FormatStyle::LK_Java || 3690 Style.Language == FormatStyle::LK_JavaScript) && 3691 Left.is(TT_LeadingJavaAnnotation) && 3692 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 3693 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 3694 return true; 3695 3696 if (Right.is(TT_ProtoExtensionLSquare)) 3697 return true; 3698 3699 // In text proto instances if a submessage contains at least 2 entries and at 3700 // least one of them is a submessage, like A { ... B { ... } ... }, 3701 // put all of the entries of A on separate lines by forcing the selector of 3702 // the submessage B to be put on a newline. 3703 // 3704 // Example: these can stay on one line: 3705 // a { scalar_1: 1 scalar_2: 2 } 3706 // a { b { key: value } } 3707 // 3708 // and these entries need to be on a new line even if putting them all in one 3709 // line is under the column limit: 3710 // a { 3711 // scalar: 1 3712 // b { key: value } 3713 // } 3714 // 3715 // We enforce this by breaking before a submessage field that has previous 3716 // siblings, *and* breaking before a field that follows a submessage field. 3717 // 3718 // Be careful to exclude the case [proto.ext] { ... } since the `]` is 3719 // the TT_SelectorName there, but we don't want to break inside the brackets. 3720 // 3721 // Another edge case is @submessage { key: value }, which is a common 3722 // substitution placeholder. In this case we want to keep `@` and `submessage` 3723 // together. 3724 // 3725 // We ensure elsewhere that extensions are always on their own line. 3726 if ((Style.Language == FormatStyle::LK_Proto || 3727 Style.Language == FormatStyle::LK_TextProto) && 3728 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) { 3729 // Keep `@submessage` together in: 3730 // @submessage { key: value } 3731 if (Right.Previous && Right.Previous->is(tok::at)) 3732 return false; 3733 // Look for the scope opener after selector in cases like: 3734 // selector { ... 3735 // selector: { ... 3736 // selector: @base { ... 3737 FormatToken *LBrace = Right.Next; 3738 if (LBrace && LBrace->is(tok::colon)) { 3739 LBrace = LBrace->Next; 3740 if (LBrace && LBrace->is(tok::at)) { 3741 LBrace = LBrace->Next; 3742 if (LBrace) 3743 LBrace = LBrace->Next; 3744 } 3745 } 3746 if (LBrace && 3747 // The scope opener is one of {, [, <: 3748 // selector { ... } 3749 // selector [ ... ] 3750 // selector < ... > 3751 // 3752 // In case of selector { ... }, the l_brace is TT_DictLiteral. 3753 // In case of an empty selector {}, the l_brace is not TT_DictLiteral, 3754 // so we check for immediately following r_brace. 3755 ((LBrace->is(tok::l_brace) && 3756 (LBrace->is(TT_DictLiteral) || 3757 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) || 3758 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) { 3759 // If Left.ParameterCount is 0, then this submessage entry is not the 3760 // first in its parent submessage, and we want to break before this entry. 3761 // If Left.ParameterCount is greater than 0, then its parent submessage 3762 // might contain 1 or more entries and we want to break before this entry 3763 // if it contains at least 2 entries. We deal with this case later by 3764 // detecting and breaking before the next entry in the parent submessage. 3765 if (Left.ParameterCount == 0) 3766 return true; 3767 // However, if this submessage is the first entry in its parent 3768 // submessage, Left.ParameterCount might be 1 in some cases. 3769 // We deal with this case later by detecting an entry 3770 // following a closing paren of this submessage. 3771 } 3772 3773 // If this is an entry immediately following a submessage, it will be 3774 // preceded by a closing paren of that submessage, like in: 3775 // left---. .---right 3776 // v v 3777 // sub: { ... } key: value 3778 // If there was a comment between `}` an `key` above, then `key` would be 3779 // put on a new line anyways. 3780 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square)) 3781 return true; 3782 } 3783 3784 // Deal with lambda arguments in C++ - we want consistent line breaks whether 3785 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced 3786 // as aggressive line breaks are placed when the lambda is not the last arg. 3787 if ((Style.Language == FormatStyle::LK_Cpp || 3788 Style.Language == FormatStyle::LK_ObjC) && 3789 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 && 3790 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) { 3791 // Multiple lambdas in the same function call force line breaks. 3792 if (Left.BlockParameterCount > 1) 3793 return true; 3794 3795 // A lambda followed by another arg forces a line break. 3796 if (!Left.Role) 3797 return false; 3798 auto Comma = Left.Role->lastComma(); 3799 if (!Comma) 3800 return false; 3801 auto Next = Comma->getNextNonComment(); 3802 if (!Next) 3803 return false; 3804 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret)) 3805 return true; 3806 } 3807 3808 return false; 3809 } 3810 3811 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 3812 const FormatToken &Right) { 3813 const FormatToken &Left = *Right.Previous; 3814 // Language-specific stuff. 3815 if (Style.isCSharp()) { 3816 if (Left.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon) || 3817 Right.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon)) 3818 return false; 3819 // Only break after commas for generic type constraints. 3820 if (Line.First->is(TT_CSharpGenericTypeConstraint)) 3821 return Left.is(TT_CSharpGenericTypeConstraintComma); 3822 } else if (Style.Language == FormatStyle::LK_Java) { 3823 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3824 Keywords.kw_implements)) 3825 return false; 3826 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3827 Keywords.kw_implements)) 3828 return true; 3829 } else if (Style.Language == FormatStyle::LK_JavaScript) { 3830 const FormatToken *NonComment = Right.getPreviousNonComment(); 3831 if (NonComment && 3832 NonComment->isOneOf( 3833 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 3834 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 3835 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 3836 Keywords.kw_readonly, Keywords.kw_abstract, Keywords.kw_get, 3837 Keywords.kw_set, Keywords.kw_async, Keywords.kw_await)) 3838 return false; // Otherwise automatic semicolon insertion would trigger. 3839 if (Right.NestingLevel == 0 && 3840 (Left.Tok.getIdentifierInfo() || 3841 Left.isOneOf(tok::r_square, tok::r_paren)) && 3842 Right.isOneOf(tok::l_square, tok::l_paren)) 3843 return false; // Otherwise automatic semicolon insertion would trigger. 3844 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace)) 3845 return false; 3846 if (Left.is(TT_JsTypeColon)) 3847 return true; 3848 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;"). 3849 if (Left.is(tok::exclaim) && Right.is(tok::colon)) 3850 return false; 3851 // Look for is type annotations like: 3852 // function f(): a is B { ... } 3853 // Do not break before is in these cases. 3854 if (Right.is(Keywords.kw_is)) { 3855 const FormatToken *Next = Right.getNextNonComment(); 3856 // If `is` is followed by a colon, it's likely that it's a dict key, so 3857 // ignore it for this check. 3858 // For example this is common in Polymer: 3859 // Polymer({ 3860 // is: 'name', 3861 // ... 3862 // }); 3863 if (!Next || !Next->is(tok::colon)) 3864 return false; 3865 } 3866 if (Left.is(Keywords.kw_in)) 3867 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 3868 if (Right.is(Keywords.kw_in)) 3869 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 3870 if (Right.is(Keywords.kw_as)) 3871 return false; // must not break before as in 'x as type' casts 3872 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) { 3873 // extends and infer can appear as keywords in conditional types: 3874 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types 3875 // do not break before them, as the expressions are subject to ASI. 3876 return false; 3877 } 3878 if (Left.is(Keywords.kw_as)) 3879 return true; 3880 if (Left.is(TT_JsNonNullAssertion)) 3881 return true; 3882 if (Left.is(Keywords.kw_declare) && 3883 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 3884 Keywords.kw_function, tok::kw_class, tok::kw_enum, 3885 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 3886 Keywords.kw_let, tok::kw_const)) 3887 // See grammar for 'declare' statements at: 3888 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10 3889 return false; 3890 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 3891 Right.isOneOf(tok::identifier, tok::string_literal)) 3892 return false; // must not break in "module foo { ...}" 3893 if (Right.is(TT_TemplateString) && Right.closesScope()) 3894 return false; 3895 // Don't split tagged template literal so there is a break between the tag 3896 // identifier and template string. 3897 if (Left.is(tok::identifier) && Right.is(TT_TemplateString)) { 3898 return false; 3899 } 3900 if (Left.is(TT_TemplateString) && Left.opensScope()) 3901 return true; 3902 } 3903 3904 if (Left.is(tok::at)) 3905 return false; 3906 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 3907 return false; 3908 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 3909 return !Right.is(tok::l_paren); 3910 if (Right.is(TT_PointerOrReference)) 3911 return Line.IsMultiVariableDeclStmt || 3912 (Style.PointerAlignment == FormatStyle::PAS_Right && 3913 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 3914 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 3915 Right.is(tok::kw_operator)) 3916 return true; 3917 if (Left.is(TT_PointerOrReference)) 3918 return false; 3919 if (Right.isTrailingComment()) 3920 // We rely on MustBreakBefore being set correctly here as we should not 3921 // change the "binding" behavior of a comment. 3922 // The first comment in a braced lists is always interpreted as belonging to 3923 // the first list element. Otherwise, it should be placed outside of the 3924 // list. 3925 return Left.is(BK_BracedInit) || 3926 (Left.is(TT_CtorInitializerColon) && 3927 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 3928 if (Left.is(tok::question) && Right.is(tok::colon)) 3929 return false; 3930 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 3931 return Style.BreakBeforeTernaryOperators; 3932 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 3933 return !Style.BreakBeforeTernaryOperators; 3934 if (Left.is(TT_InheritanceColon)) 3935 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon; 3936 if (Right.is(TT_InheritanceColon)) 3937 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon; 3938 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 3939 Left.isNot(TT_SelectorName)) 3940 return true; 3941 3942 if (Right.is(tok::colon) && 3943 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 3944 return false; 3945 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 3946 if (Style.Language == FormatStyle::LK_Proto || 3947 Style.Language == FormatStyle::LK_TextProto) { 3948 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 3949 return false; 3950 // Prevent cases like: 3951 // 3952 // submessage: 3953 // { key: valueeeeeeeeeeee } 3954 // 3955 // when the snippet does not fit into one line. 3956 // Prefer: 3957 // 3958 // submessage: { 3959 // key: valueeeeeeeeeeee 3960 // } 3961 // 3962 // instead, even if it is longer by one line. 3963 // 3964 // Note that this allows allows the "{" to go over the column limit 3965 // when the column limit is just between ":" and "{", but that does 3966 // not happen too often and alternative formattings in this case are 3967 // not much better. 3968 // 3969 // The code covers the cases: 3970 // 3971 // submessage: { ... } 3972 // submessage: < ... > 3973 // repeated: [ ... ] 3974 if (((Right.is(tok::l_brace) || Right.is(tok::less)) && 3975 Right.is(TT_DictLiteral)) || 3976 Right.is(TT_ArrayInitializerLSquare)) 3977 return false; 3978 } 3979 return true; 3980 } 3981 if (Right.is(tok::r_square) && Right.MatchingParen && 3982 Right.MatchingParen->is(TT_ProtoExtensionLSquare)) 3983 return false; 3984 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 3985 Right.Next->is(TT_ObjCMethodExpr))) 3986 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 3987 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 3988 return true; 3989 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 3990 return true; 3991 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 3992 TT_OverloadedOperator)) 3993 return false; 3994 if (Left.is(TT_RangeBasedForLoopColon)) 3995 return true; 3996 if (Right.is(TT_RangeBasedForLoopColon)) 3997 return false; 3998 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 3999 return true; 4000 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 4001 Left.is(tok::kw_operator)) 4002 return false; 4003 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 4004 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 4005 return false; 4006 if (Left.is(tok::equal) && Right.is(tok::l_brace) && 4007 !Style.Cpp11BracedListStyle) 4008 return false; 4009 if (Left.is(tok::l_paren) && 4010 Left.isOneOf(TT_AttributeParen, TT_TypeDeclarationParen)) 4011 return false; 4012 if (Left.is(tok::l_paren) && Left.Previous && 4013 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 4014 return false; 4015 if (Right.is(TT_ImplicitStringLiteral)) 4016 return false; 4017 4018 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser)) 4019 return false; 4020 if (Right.is(tok::r_square) && Right.MatchingParen && 4021 Right.MatchingParen->is(TT_LambdaLSquare)) 4022 return false; 4023 4024 // We only break before r_brace if there was a corresponding break before 4025 // the l_brace, which is tracked by BreakBeforeClosingBrace. 4026 if (Right.is(tok::r_brace)) 4027 return Right.MatchingParen && Right.MatchingParen->is(BK_Block); 4028 4029 // Allow breaking after a trailing annotation, e.g. after a method 4030 // declaration. 4031 if (Left.is(TT_TrailingAnnotation)) 4032 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 4033 tok::less, tok::coloncolon); 4034 4035 if (Right.is(tok::kw___attribute) || 4036 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) 4037 return !Left.is(TT_AttributeSquare); 4038 4039 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 4040 return true; 4041 4042 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 4043 return true; 4044 4045 if (Left.is(TT_CtorInitializerColon)) 4046 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 4047 if (Right.is(TT_CtorInitializerColon)) 4048 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 4049 if (Left.is(TT_CtorInitializerComma) && 4050 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 4051 return false; 4052 if (Right.is(TT_CtorInitializerComma) && 4053 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 4054 return true; 4055 if (Left.is(TT_InheritanceComma) && 4056 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 4057 return false; 4058 if (Right.is(TT_InheritanceComma) && 4059 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 4060 return true; 4061 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 4062 (Left.is(tok::less) && Right.is(tok::less))) 4063 return false; 4064 if (Right.is(TT_BinaryOperator) && 4065 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 4066 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 4067 Right.getPrecedence() != prec::Assignment)) 4068 return true; 4069 if (Left.is(TT_ArrayInitializerLSquare)) 4070 return true; 4071 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 4072 return true; 4073 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 4074 !Left.isOneOf(tok::arrowstar, tok::lessless) && 4075 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 4076 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 4077 Left.getPrecedence() == prec::Assignment)) 4078 return true; 4079 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 4080 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) 4081 return false; 4082 4083 auto ShortLambdaOption = Style.AllowShortLambdasOnASingleLine; 4084 if (Style.BraceWrapping.BeforeLambdaBody) { 4085 if (isAllmanLambdaBrace(Left)) 4086 return !isItAnEmptyLambdaAllowed(Left, ShortLambdaOption); 4087 if (isAllmanLambdaBrace(Right)) 4088 return !isItAnEmptyLambdaAllowed(Right, ShortLambdaOption); 4089 } 4090 4091 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 4092 tok::kw_class, tok::kw_struct, tok::comment) || 4093 Right.isMemberAccess() || 4094 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 4095 tok::colon, tok::l_square, tok::at) || 4096 (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace)) || 4097 (Left.is(tok::r_paren) && 4098 Right.isOneOf(tok::identifier, tok::kw_const)) || 4099 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 4100 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 4101 } 4102 4103 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 4104 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 4105 const FormatToken *Tok = Line.First; 4106 while (Tok) { 4107 llvm::errs() << " M=" << Tok->MustBreakBefore 4108 << " C=" << Tok->CanBreakBefore 4109 << " T=" << getTokenTypeName(Tok->getType()) 4110 << " S=" << Tok->SpacesRequiredBefore 4111 << " F=" << Tok->Finalized << " B=" << Tok->BlockParameterCount 4112 << " BK=" << Tok->getBlockKind() << " P=" << Tok->SplitPenalty 4113 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 4114 << " PPK=" << Tok->getPackingKind() << " FakeLParens="; 4115 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 4116 llvm::errs() << Tok->FakeLParens[i] << "/"; 4117 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 4118 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo(); 4119 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 4120 if (!Tok->Next) 4121 assert(Tok == Line.Last); 4122 Tok = Tok->Next; 4123 } 4124 llvm::errs() << "----\n"; 4125 } 4126 4127 } // namespace format 4128 } // namespace clang 4129