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