1 //===--- TokenAnnotator.cpp - Format C++ code -----------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// 9 /// \file 10 /// This file implements a token annotator, i.e. creates 11 /// \c AnnotatedTokens out of \c FormatTokens with required extra information. 12 /// 13 //===----------------------------------------------------------------------===// 14 15 #include "TokenAnnotator.h" 16 #include "FormatToken.h" 17 #include "clang/Basic/SourceManager.h" 18 #include "clang/Basic/TokenKinds.h" 19 #include "llvm/ADT/SmallPtrSet.h" 20 #include "llvm/Support/Debug.h" 21 22 #define DEBUG_TYPE "format-token-annotator" 23 24 namespace clang { 25 namespace format { 26 27 namespace { 28 29 /// Returns \c true if the token can be used as an identifier in 30 /// an Objective-C \c \@selector, \c false otherwise. 31 /// 32 /// Because getFormattingLangOpts() always lexes source code as 33 /// Objective-C++, C++ keywords like \c new and \c delete are 34 /// lexed as tok::kw_*, not tok::identifier, even for Objective-C. 35 /// 36 /// For Objective-C and Objective-C++, both identifiers and keywords 37 /// are valid inside @selector(...) (or a macro which 38 /// invokes @selector(...)). So, we allow treat any identifier or 39 /// keyword as a potential Objective-C selector component. 40 static bool canBeObjCSelectorComponent(const FormatToken &Tok) { 41 return Tok.Tok.getIdentifierInfo() != nullptr; 42 } 43 44 /// With `Left` being '(', check if we're at either `[...](` or 45 /// `[...]<...>(`, where the [ opens a lambda capture list. 46 static bool isLambdaParameterList(const FormatToken *Left) { 47 // Skip <...> if present. 48 if (Left->Previous && Left->Previous->is(tok::greater) && 49 Left->Previous->MatchingParen && 50 Left->Previous->MatchingParen->is(TT_TemplateOpener)) 51 Left = Left->Previous->MatchingParen; 52 53 // Check for `[...]`. 54 return Left->Previous && Left->Previous->is(tok::r_square) && 55 Left->Previous->MatchingParen && 56 Left->Previous->MatchingParen->is(TT_LambdaLSquare); 57 } 58 59 /// Returns \c true if the token is followed by a boolean condition, \c false 60 /// otherwise. 61 static bool isKeywordWithCondition(const FormatToken &Tok) { 62 return Tok.isOneOf(tok::kw_if, tok::kw_for, tok::kw_while, tok::kw_switch, 63 tok::kw_constexpr, tok::kw_catch); 64 } 65 66 /// A parser that gathers additional information about tokens. 67 /// 68 /// The \c TokenAnnotator tries to match parenthesis and square brakets and 69 /// store a parenthesis levels. It also tries to resolve matching "<" and ">" 70 /// into template parameter lists. 71 class AnnotatingParser { 72 public: 73 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line, 74 const AdditionalKeywords &Keywords) 75 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false), 76 Keywords(Keywords) { 77 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false)); 78 resetTokenMetadata(CurrentToken); 79 } 80 81 private: 82 bool parseAngle() { 83 if (!CurrentToken || !CurrentToken->Previous) 84 return false; 85 if (NonTemplateLess.count(CurrentToken->Previous)) 86 return false; 87 88 const FormatToken &Previous = *CurrentToken->Previous; // The '<'. 89 if (Previous.Previous) { 90 if (Previous.Previous->Tok.isLiteral()) 91 return false; 92 if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 && 93 (!Previous.Previous->MatchingParen || 94 !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen))) 95 return false; 96 } 97 98 FormatToken *Left = CurrentToken->Previous; 99 Left->ParentBracket = Contexts.back().ContextKind; 100 ScopedContextCreator ContextCreator(*this, tok::less, 12); 101 102 // If this angle is in the context of an expression, we need to be more 103 // hesitant to detect it as opening template parameters. 104 bool InExprContext = Contexts.back().IsExpression; 105 106 Contexts.back().IsExpression = false; 107 // If there's a template keyword before the opening angle bracket, this is a 108 // template parameter, not an argument. 109 Contexts.back().InTemplateArgument = 110 Left->Previous && Left->Previous->Tok.isNot(tok::kw_template); 111 112 if (Style.Language == FormatStyle::LK_Java && 113 CurrentToken->is(tok::question)) 114 next(); 115 116 while (CurrentToken) { 117 if (CurrentToken->is(tok::greater)) { 118 // Try to do a better job at looking for ">>" within the condition of 119 // a statement. 120 if (CurrentToken->Next && CurrentToken->Next->is(tok::greater) && 121 Left->ParentBracket != tok::less && 122 isKeywordWithCondition(*Line.First)) 123 return false; 124 Left->MatchingParen = CurrentToken; 125 CurrentToken->MatchingParen = Left; 126 // In TT_Proto, we must distignuish between: 127 // map<key, value> 128 // msg < item: data > 129 // msg: < item: data > 130 // In TT_TextProto, map<key, value> does not occur. 131 if (Style.Language == FormatStyle::LK_TextProto || 132 (Style.Language == FormatStyle::LK_Proto && Left->Previous && 133 Left->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) 134 CurrentToken->setType(TT_DictLiteral); 135 else 136 CurrentToken->setType(TT_TemplateCloser); 137 next(); 138 return true; 139 } 140 if (CurrentToken->is(tok::question) && 141 Style.Language == FormatStyle::LK_Java) { 142 next(); 143 continue; 144 } 145 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) || 146 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext && 147 !Style.isCSharp() && Style.Language != FormatStyle::LK_Proto && 148 Style.Language != FormatStyle::LK_TextProto)) 149 return false; 150 // If a && or || is found and interpreted as a binary operator, this set 151 // of angles is likely part of something like "a < b && c > d". If the 152 // angles are inside an expression, the ||/&& might also be a binary 153 // operator that was misinterpreted because we are parsing template 154 // parameters. 155 // FIXME: This is getting out of hand, write a decent parser. 156 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) && 157 CurrentToken->Previous->is(TT_BinaryOperator) && 158 Contexts[Contexts.size() - 2].IsExpression && 159 !Line.startsWith(tok::kw_template)) 160 return false; 161 updateParameterCount(Left, CurrentToken); 162 if (Style.Language == FormatStyle::LK_Proto) { 163 if (FormatToken *Previous = CurrentToken->getPreviousNonComment()) { 164 if (CurrentToken->is(tok::colon) || 165 (CurrentToken->isOneOf(tok::l_brace, tok::less) && 166 Previous->isNot(tok::colon))) 167 Previous->setType(TT_SelectorName); 168 } 169 } 170 if (!consumeToken()) 171 return false; 172 } 173 return false; 174 } 175 176 bool parseUntouchableParens() { 177 while (CurrentToken) { 178 CurrentToken->Finalized = true; 179 switch (CurrentToken->Tok.getKind()) { 180 case tok::l_paren: 181 next(); 182 if (!parseUntouchableParens()) 183 return false; 184 continue; 185 case tok::r_paren: 186 next(); 187 return true; 188 default: 189 // no-op 190 break; 191 } 192 next(); 193 } 194 return false; 195 } 196 197 bool parseParens(bool LookForDecls = false) { 198 if (!CurrentToken) 199 return false; 200 FormatToken *Left = CurrentToken->Previous; 201 assert(Left && "Unknown previous token"); 202 FormatToken *PrevNonComment = Left->getPreviousNonComment(); 203 Left->ParentBracket = Contexts.back().ContextKind; 204 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1); 205 206 // FIXME: This is a bit of a hack. Do better. 207 Contexts.back().ColonIsForRangeExpr = 208 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr; 209 210 if (Left->Previous && Left->Previous->is(TT_UntouchableMacroFunc)) { 211 Left->Finalized = true; 212 return parseUntouchableParens(); 213 } 214 215 bool StartsObjCMethodExpr = false; 216 if (FormatToken *MaybeSel = Left->Previous) { 217 // @selector( starts a selector. 218 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous && 219 MaybeSel->Previous->is(tok::at)) { 220 StartsObjCMethodExpr = true; 221 } 222 } 223 224 if (Left->is(TT_OverloadedOperatorLParen)) { 225 Contexts.back().IsExpression = false; 226 } else if (Style.Language == FormatStyle::LK_JavaScript && 227 (Line.startsWith(Keywords.kw_type, tok::identifier) || 228 Line.startsWith(tok::kw_export, Keywords.kw_type, 229 tok::identifier))) { 230 // type X = (...); 231 // export type X = (...); 232 Contexts.back().IsExpression = false; 233 } else if (Left->Previous && 234 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_while, 235 tok::l_paren, tok::comma) || 236 Left->Previous->isIf() || 237 Left->Previous->is(TT_BinaryOperator))) { 238 // static_assert, if and while usually contain expressions. 239 Contexts.back().IsExpression = true; 240 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 241 (Left->Previous->is(Keywords.kw_function) || 242 (Left->Previous->endsSequence(tok::identifier, 243 Keywords.kw_function)))) { 244 // function(...) or function f(...) 245 Contexts.back().IsExpression = false; 246 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 247 Left->Previous->is(TT_JsTypeColon)) { 248 // let x: (SomeType); 249 Contexts.back().IsExpression = false; 250 } else if (isLambdaParameterList(Left)) { 251 // This is a parameter list of a lambda expression. 252 Contexts.back().IsExpression = false; 253 } else if (Line.InPPDirective && 254 (!Left->Previous || !Left->Previous->is(tok::identifier))) { 255 Contexts.back().IsExpression = true; 256 } else if (Contexts[Contexts.size() - 2].CaretFound) { 257 // This is the parameter list of an ObjC block. 258 Contexts.back().IsExpression = false; 259 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) { 260 // The first argument to a foreach macro is a declaration. 261 Contexts.back().IsForEachMacro = true; 262 Contexts.back().IsExpression = false; 263 } else if (Left->Previous && Left->Previous->MatchingParen && 264 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) { 265 Contexts.back().IsExpression = false; 266 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) { 267 bool IsForOrCatch = 268 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch); 269 Contexts.back().IsExpression = !IsForOrCatch; 270 } 271 272 // Infer the role of the l_paren based on the previous token if we haven't 273 // detected one one yet. 274 if (PrevNonComment && Left->is(TT_Unknown)) { 275 if (PrevNonComment->is(tok::kw___attribute)) { 276 Left->setType(TT_AttributeParen); 277 } else if (PrevNonComment->isOneOf(TT_TypenameMacro, tok::kw_decltype, 278 tok::kw_typeof, tok::kw__Atomic, 279 tok::kw___underlying_type)) { 280 Left->setType(TT_TypeDeclarationParen); 281 // decltype() and typeof() usually contain expressions. 282 if (PrevNonComment->isOneOf(tok::kw_decltype, tok::kw_typeof)) 283 Contexts.back().IsExpression = true; 284 } 285 } 286 287 if (StartsObjCMethodExpr) { 288 Contexts.back().ColonIsObjCMethodExpr = true; 289 Left->setType(TT_ObjCMethodExpr); 290 } 291 292 // MightBeFunctionType and ProbablyFunctionType are used for 293 // function pointer and reference types as well as Objective-C 294 // block types: 295 // 296 // void (*FunctionPointer)(void); 297 // void (&FunctionReference)(void); 298 // void (^ObjCBlock)(void); 299 bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression; 300 bool ProbablyFunctionType = 301 CurrentToken->isOneOf(tok::star, tok::amp, tok::caret); 302 bool HasMultipleLines = false; 303 bool HasMultipleParametersOnALine = false; 304 bool MightBeObjCForRangeLoop = 305 Left->Previous && Left->Previous->is(tok::kw_for); 306 FormatToken *PossibleObjCForInToken = nullptr; 307 while (CurrentToken) { 308 // LookForDecls is set when "if (" has been seen. Check for 309 // 'identifier' '*' 'identifier' followed by not '=' -- this 310 // '*' has to be a binary operator but determineStarAmpUsage() will 311 // categorize it as an unary operator, so set the right type here. 312 if (LookForDecls && CurrentToken->Next) { 313 FormatToken *Prev = CurrentToken->getPreviousNonComment(); 314 if (Prev) { 315 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 316 FormatToken *Next = CurrentToken->Next; 317 if (PrevPrev && PrevPrev->is(tok::identifier) && 318 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) && 319 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) { 320 Prev->setType(TT_BinaryOperator); 321 LookForDecls = false; 322 } 323 } 324 } 325 326 if (CurrentToken->Previous->is(TT_PointerOrReference) && 327 CurrentToken->Previous->Previous->isOneOf(tok::l_paren, 328 tok::coloncolon)) 329 ProbablyFunctionType = true; 330 if (CurrentToken->is(tok::comma)) 331 MightBeFunctionType = false; 332 if (CurrentToken->Previous->is(TT_BinaryOperator)) 333 Contexts.back().IsExpression = true; 334 if (CurrentToken->is(tok::r_paren)) { 335 if (MightBeFunctionType && ProbablyFunctionType && CurrentToken->Next && 336 (CurrentToken->Next->is(tok::l_paren) || 337 (CurrentToken->Next->is(tok::l_square) && Line.MustBeDeclaration))) 338 Left->setType(Left->Next->is(tok::caret) ? TT_ObjCBlockLParen 339 : TT_FunctionTypeLParen); 340 Left->MatchingParen = CurrentToken; 341 CurrentToken->MatchingParen = Left; 342 343 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) && 344 Left->Previous && Left->Previous->is(tok::l_paren)) { 345 // Detect the case where macros are used to generate lambdas or 346 // function bodies, e.g.: 347 // auto my_lambda = MARCO((Type *type, int i) { .. body .. }); 348 for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) { 349 if (Tok->is(TT_BinaryOperator) && 350 Tok->isOneOf(tok::star, tok::amp, tok::ampamp)) 351 Tok->setType(TT_PointerOrReference); 352 } 353 } 354 355 if (StartsObjCMethodExpr) { 356 CurrentToken->setType(TT_ObjCMethodExpr); 357 if (Contexts.back().FirstObjCSelectorName) { 358 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 359 Contexts.back().LongestObjCSelectorName; 360 } 361 } 362 363 if (Left->is(TT_AttributeParen)) 364 CurrentToken->setType(TT_AttributeParen); 365 if (Left->is(TT_TypeDeclarationParen)) 366 CurrentToken->setType(TT_TypeDeclarationParen); 367 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation)) 368 CurrentToken->setType(TT_JavaAnnotation); 369 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation)) 370 CurrentToken->setType(TT_LeadingJavaAnnotation); 371 if (Left->Previous && Left->Previous->is(TT_AttributeSquare)) 372 CurrentToken->setType(TT_AttributeSquare); 373 374 if (!HasMultipleLines) 375 Left->setPackingKind(PPK_Inconclusive); 376 else if (HasMultipleParametersOnALine) 377 Left->setPackingKind(PPK_BinPacked); 378 else 379 Left->setPackingKind(PPK_OnePerLine); 380 381 next(); 382 return true; 383 } 384 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace)) 385 return false; 386 387 if (CurrentToken->is(tok::l_brace)) 388 Left->setType(TT_Unknown); // Not TT_ObjCBlockLParen 389 if (CurrentToken->is(tok::comma) && CurrentToken->Next && 390 !CurrentToken->Next->HasUnescapedNewline && 391 !CurrentToken->Next->isTrailingComment()) 392 HasMultipleParametersOnALine = true; 393 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) || 394 CurrentToken->Previous->isSimpleTypeSpecifier()) && 395 !CurrentToken->is(tok::l_brace)) 396 Contexts.back().IsExpression = false; 397 if (CurrentToken->isOneOf(tok::semi, tok::colon)) { 398 MightBeObjCForRangeLoop = false; 399 if (PossibleObjCForInToken) { 400 PossibleObjCForInToken->setType(TT_Unknown); 401 PossibleObjCForInToken = nullptr; 402 } 403 } 404 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) { 405 PossibleObjCForInToken = CurrentToken; 406 PossibleObjCForInToken->setType(TT_ObjCForIn); 407 } 408 // When we discover a 'new', we set CanBeExpression to 'false' in order to 409 // parse the type correctly. Reset that after a comma. 410 if (CurrentToken->is(tok::comma)) 411 Contexts.back().CanBeExpression = true; 412 413 FormatToken *Tok = CurrentToken; 414 if (!consumeToken()) 415 return false; 416 updateParameterCount(Left, Tok); 417 if (CurrentToken && CurrentToken->HasUnescapedNewline) 418 HasMultipleLines = true; 419 } 420 return false; 421 } 422 423 bool isCSharpAttributeSpecifier(const FormatToken &Tok) { 424 if (!Style.isCSharp()) 425 return false; 426 427 // `identifier[i]` is not an attribute. 428 if (Tok.Previous && Tok.Previous->is(tok::identifier)) 429 return false; 430 431 // Chains of [] in `identifier[i][j][k]` are not attributes. 432 if (Tok.Previous && Tok.Previous->is(tok::r_square)) { 433 auto *MatchingParen = Tok.Previous->MatchingParen; 434 if (!MatchingParen || MatchingParen->is(TT_ArraySubscriptLSquare)) 435 return false; 436 } 437 438 const FormatToken *AttrTok = Tok.Next; 439 if (!AttrTok) 440 return false; 441 442 // Just an empty declaration e.g. string []. 443 if (AttrTok->is(tok::r_square)) 444 return false; 445 446 // Move along the tokens inbetween the '[' and ']' e.g. [STAThread]. 447 while (AttrTok && AttrTok->isNot(tok::r_square)) { 448 AttrTok = AttrTok->Next; 449 } 450 451 if (!AttrTok) 452 return false; 453 454 // Allow an attribute to be the only content of a file. 455 AttrTok = AttrTok->Next; 456 if (!AttrTok) 457 return true; 458 459 // Limit this to being an access modifier that follows. 460 if (AttrTok->isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 461 tok::comment, tok::kw_class, tok::kw_static, 462 tok::l_square, Keywords.kw_internal)) { 463 return true; 464 } 465 466 // incase its a [XXX] retval func(.... 467 if (AttrTok->Next && 468 AttrTok->Next->startsSequence(tok::identifier, tok::l_paren)) 469 return true; 470 471 return false; 472 } 473 474 bool isCpp11AttributeSpecifier(const FormatToken &Tok) { 475 if (!Style.isCpp() || !Tok.startsSequence(tok::l_square, tok::l_square)) 476 return false; 477 // The first square bracket is part of an ObjC array literal 478 if (Tok.Previous && Tok.Previous->is(tok::at)) { 479 return false; 480 } 481 const FormatToken *AttrTok = Tok.Next->Next; 482 if (!AttrTok) 483 return false; 484 // C++17 '[[using ns: foo, bar(baz, blech)]]' 485 // We assume nobody will name an ObjC variable 'using'. 486 if (AttrTok->startsSequence(tok::kw_using, tok::identifier, tok::colon)) 487 return true; 488 if (AttrTok->isNot(tok::identifier)) 489 return false; 490 while (AttrTok && !AttrTok->startsSequence(tok::r_square, tok::r_square)) { 491 // ObjC message send. We assume nobody will use : in a C++11 attribute 492 // specifier parameter, although this is technically valid: 493 // [[foo(:)]]. 494 if (AttrTok->is(tok::colon) || 495 AttrTok->startsSequence(tok::identifier, tok::identifier) || 496 AttrTok->startsSequence(tok::r_paren, tok::identifier)) 497 return false; 498 if (AttrTok->is(tok::ellipsis)) 499 return true; 500 AttrTok = AttrTok->Next; 501 } 502 return AttrTok && AttrTok->startsSequence(tok::r_square, tok::r_square); 503 } 504 505 bool parseSquare() { 506 if (!CurrentToken) 507 return false; 508 509 // A '[' could be an index subscript (after an identifier or after 510 // ')' or ']'), it could be the start of an Objective-C method 511 // expression, it could the start of an Objective-C array literal, 512 // or it could be a C++ attribute specifier [[foo::bar]]. 513 FormatToken *Left = CurrentToken->Previous; 514 Left->ParentBracket = Contexts.back().ContextKind; 515 FormatToken *Parent = Left->getPreviousNonComment(); 516 517 // Cases where '>' is followed by '['. 518 // In C++, this can happen either in array of templates (foo<int>[10]) 519 // or when array is a nested template type (unique_ptr<type1<type2>[]>). 520 bool CppArrayTemplates = 521 Style.isCpp() && Parent && Parent->is(TT_TemplateCloser) && 522 (Contexts.back().CanBeExpression || Contexts.back().IsExpression || 523 Contexts.back().InTemplateArgument); 524 525 bool IsCpp11AttributeSpecifier = isCpp11AttributeSpecifier(*Left) || 526 Contexts.back().InCpp11AttributeSpecifier; 527 528 // Treat C# Attributes [STAThread] much like C++ attributes [[...]]. 529 bool IsCSharpAttributeSpecifier = 530 isCSharpAttributeSpecifier(*Left) || 531 Contexts.back().InCSharpAttributeSpecifier; 532 533 bool InsideInlineASM = Line.startsWith(tok::kw_asm); 534 bool IsCppStructuredBinding = Left->isCppStructuredBinding(Style); 535 bool StartsObjCMethodExpr = 536 !IsCppStructuredBinding && !InsideInlineASM && !CppArrayTemplates && 537 Style.isCpp() && !IsCpp11AttributeSpecifier && 538 !IsCSharpAttributeSpecifier && Contexts.back().CanBeExpression && 539 Left->isNot(TT_LambdaLSquare) && 540 !CurrentToken->isOneOf(tok::l_brace, tok::r_square) && 541 (!Parent || 542 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren, 543 tok::kw_return, tok::kw_throw) || 544 Parent->isUnaryOperator() || 545 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 546 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) || 547 (getBinOpPrecedence(Parent->Tok.getKind(), true, true) > 548 prec::Unknown)); 549 bool ColonFound = false; 550 551 unsigned BindingIncrease = 1; 552 if (IsCppStructuredBinding) { 553 Left->setType(TT_StructuredBindingLSquare); 554 } else if (Left->is(TT_Unknown)) { 555 if (StartsObjCMethodExpr) { 556 Left->setType(TT_ObjCMethodExpr); 557 } else if (InsideInlineASM) { 558 Left->setType(TT_InlineASMSymbolicNameLSquare); 559 } else if (IsCpp11AttributeSpecifier) { 560 Left->setType(TT_AttributeSquare); 561 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent && 562 Contexts.back().ContextKind == tok::l_brace && 563 Parent->isOneOf(tok::l_brace, tok::comma)) { 564 Left->setType(TT_JsComputedPropertyName); 565 } else if (Style.isCpp() && Contexts.back().ContextKind == tok::l_brace && 566 Parent && Parent->isOneOf(tok::l_brace, tok::comma)) { 567 Left->setType(TT_DesignatedInitializerLSquare); 568 } else if (IsCSharpAttributeSpecifier) { 569 Left->setType(TT_AttributeSquare); 570 } else if (CurrentToken->is(tok::r_square) && Parent && 571 Parent->is(TT_TemplateCloser)) { 572 Left->setType(TT_ArraySubscriptLSquare); 573 } else if (Style.Language == FormatStyle::LK_Proto || 574 Style.Language == FormatStyle::LK_TextProto) { 575 // Square braces in LK_Proto can either be message field attributes: 576 // 577 // optional Aaa aaa = 1 [ 578 // (aaa) = aaa 579 // ]; 580 // 581 // extensions 123 [ 582 // (aaa) = aaa 583 // ]; 584 // 585 // or text proto extensions (in options): 586 // 587 // option (Aaa.options) = { 588 // [type.type/type] { 589 // key: value 590 // } 591 // } 592 // 593 // or repeated fields (in options): 594 // 595 // option (Aaa.options) = { 596 // keys: [ 1, 2, 3 ] 597 // } 598 // 599 // In the first and the third case we want to spread the contents inside 600 // the square braces; in the second we want to keep them inline. 601 Left->setType(TT_ArrayInitializerLSquare); 602 if (!Left->endsSequence(tok::l_square, tok::numeric_constant, 603 tok::equal) && 604 !Left->endsSequence(tok::l_square, tok::numeric_constant, 605 tok::identifier) && 606 !Left->endsSequence(tok::l_square, tok::colon, TT_SelectorName)) { 607 Left->setType(TT_ProtoExtensionLSquare); 608 BindingIncrease = 10; 609 } 610 } else if (!CppArrayTemplates && Parent && 611 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at, 612 tok::comma, tok::l_paren, tok::l_square, 613 tok::question, tok::colon, tok::kw_return, 614 // Should only be relevant to JavaScript: 615 tok::kw_default)) { 616 Left->setType(TT_ArrayInitializerLSquare); 617 } else { 618 BindingIncrease = 10; 619 Left->setType(TT_ArraySubscriptLSquare); 620 } 621 } 622 623 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease); 624 Contexts.back().IsExpression = true; 625 if (Style.Language == FormatStyle::LK_JavaScript && Parent && 626 Parent->is(TT_JsTypeColon)) 627 Contexts.back().IsExpression = false; 628 629 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr; 630 Contexts.back().InCpp11AttributeSpecifier = IsCpp11AttributeSpecifier; 631 Contexts.back().InCSharpAttributeSpecifier = IsCSharpAttributeSpecifier; 632 633 while (CurrentToken) { 634 if (CurrentToken->is(tok::r_square)) { 635 if (IsCpp11AttributeSpecifier) 636 CurrentToken->setType(TT_AttributeSquare); 637 if (IsCSharpAttributeSpecifier) 638 CurrentToken->setType(TT_AttributeSquare); 639 else if (((CurrentToken->Next && 640 CurrentToken->Next->is(tok::l_paren)) || 641 (CurrentToken->Previous && 642 CurrentToken->Previous->Previous == Left)) && 643 Left->is(TT_ObjCMethodExpr)) { 644 // An ObjC method call is rarely followed by an open parenthesis. It 645 // also can't be composed of just one token, unless it's a macro that 646 // will be expanded to more tokens. 647 // FIXME: Do we incorrectly label ":" with this? 648 StartsObjCMethodExpr = false; 649 Left->setType(TT_Unknown); 650 } 651 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) { 652 CurrentToken->setType(TT_ObjCMethodExpr); 653 // If we haven't seen a colon yet, make sure the last identifier 654 // before the r_square is tagged as a selector name component. 655 if (!ColonFound && CurrentToken->Previous && 656 CurrentToken->Previous->is(TT_Unknown) && 657 canBeObjCSelectorComponent(*CurrentToken->Previous)) 658 CurrentToken->Previous->setType(TT_SelectorName); 659 // determineStarAmpUsage() thinks that '*' '[' is allocating an 660 // array of pointers, but if '[' starts a selector then '*' is a 661 // binary operator. 662 if (Parent && Parent->is(TT_PointerOrReference)) 663 Parent->setType(TT_BinaryOperator); 664 } 665 // An arrow after an ObjC method expression is not a lambda arrow. 666 if (CurrentToken->getType() == TT_ObjCMethodExpr && 667 CurrentToken->Next && CurrentToken->Next->is(TT_LambdaArrow)) 668 CurrentToken->Next->setType(TT_Unknown); 669 Left->MatchingParen = CurrentToken; 670 CurrentToken->MatchingParen = Left; 671 // FirstObjCSelectorName is set when a colon is found. This does 672 // not work, however, when the method has no parameters. 673 // Here, we set FirstObjCSelectorName when the end of the method call is 674 // reached, in case it was not set already. 675 if (!Contexts.back().FirstObjCSelectorName) { 676 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 677 if (Previous && Previous->is(TT_SelectorName)) { 678 Previous->ObjCSelectorNameParts = 1; 679 Contexts.back().FirstObjCSelectorName = Previous; 680 } 681 } else { 682 Left->ParameterCount = 683 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 684 } 685 if (Contexts.back().FirstObjCSelectorName) { 686 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 687 Contexts.back().LongestObjCSelectorName; 688 if (Left->BlockParameterCount > 1) 689 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0; 690 } 691 next(); 692 return true; 693 } 694 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace)) 695 return false; 696 if (CurrentToken->is(tok::colon)) { 697 if (IsCpp11AttributeSpecifier && 698 CurrentToken->endsSequence(tok::colon, tok::identifier, 699 tok::kw_using)) { 700 // Remember that this is a [[using ns: foo]] C++ attribute, so we 701 // don't add a space before the colon (unlike other colons). 702 CurrentToken->setType(TT_AttributeColon); 703 } else if (Left->isOneOf(TT_ArraySubscriptLSquare, 704 TT_DesignatedInitializerLSquare)) { 705 Left->setType(TT_ObjCMethodExpr); 706 StartsObjCMethodExpr = true; 707 Contexts.back().ColonIsObjCMethodExpr = true; 708 if (Parent && Parent->is(tok::r_paren)) 709 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 710 Parent->setType(TT_CastRParen); 711 } 712 ColonFound = true; 713 } 714 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) && 715 !ColonFound) 716 Left->setType(TT_ArrayInitializerLSquare); 717 FormatToken *Tok = CurrentToken; 718 if (!consumeToken()) 719 return false; 720 updateParameterCount(Left, Tok); 721 } 722 return false; 723 } 724 725 bool parseBrace() { 726 if (CurrentToken) { 727 FormatToken *Left = CurrentToken->Previous; 728 Left->ParentBracket = Contexts.back().ContextKind; 729 730 if (Contexts.back().CaretFound) 731 Left->setType(TT_ObjCBlockLBrace); 732 Contexts.back().CaretFound = false; 733 734 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1); 735 Contexts.back().ColonIsDictLiteral = true; 736 if (Left->is(BK_BracedInit)) 737 Contexts.back().IsExpression = true; 738 if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 739 Left->Previous->is(TT_JsTypeColon)) 740 Contexts.back().IsExpression = false; 741 742 while (CurrentToken) { 743 if (CurrentToken->is(tok::r_brace)) { 744 Left->MatchingParen = CurrentToken; 745 CurrentToken->MatchingParen = Left; 746 next(); 747 return true; 748 } 749 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 750 return false; 751 updateParameterCount(Left, CurrentToken); 752 if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) { 753 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 754 if (Previous->is(TT_JsTypeOptionalQuestion)) 755 Previous = Previous->getPreviousNonComment(); 756 if ((CurrentToken->is(tok::colon) && 757 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) || 758 Style.Language == FormatStyle::LK_Proto || 759 Style.Language == FormatStyle::LK_TextProto) { 760 Left->setType(TT_DictLiteral); 761 if (Previous->Tok.getIdentifierInfo() || 762 Previous->is(tok::string_literal)) 763 Previous->setType(TT_SelectorName); 764 } 765 if (CurrentToken->is(tok::colon) || 766 Style.Language == FormatStyle::LK_JavaScript) 767 Left->setType(TT_DictLiteral); 768 } 769 if (CurrentToken->is(tok::comma) && 770 Style.Language == FormatStyle::LK_JavaScript) 771 Left->setType(TT_DictLiteral); 772 if (!consumeToken()) 773 return false; 774 } 775 } 776 return true; 777 } 778 779 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 780 // For ObjC methods, the number of parameters is calculated differently as 781 // method declarations have a different structure (the parameters are not 782 // inside a bracket scope). 783 if (Current->is(tok::l_brace) && Current->is(BK_Block)) 784 ++Left->BlockParameterCount; 785 if (Current->is(tok::comma)) { 786 ++Left->ParameterCount; 787 if (!Left->Role) 788 Left->Role.reset(new CommaSeparatedList(Style)); 789 Left->Role->CommaFound(Current); 790 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 791 Left->ParameterCount = 1; 792 } 793 } 794 795 bool parseConditional() { 796 while (CurrentToken) { 797 if (CurrentToken->is(tok::colon)) { 798 CurrentToken->setType(TT_ConditionalExpr); 799 next(); 800 return true; 801 } 802 if (!consumeToken()) 803 return false; 804 } 805 return false; 806 } 807 808 bool parseTemplateDeclaration() { 809 if (CurrentToken && CurrentToken->is(tok::less)) { 810 CurrentToken->setType(TT_TemplateOpener); 811 next(); 812 if (!parseAngle()) 813 return false; 814 if (CurrentToken) 815 CurrentToken->Previous->ClosesTemplateDeclaration = true; 816 return true; 817 } 818 return false; 819 } 820 821 bool consumeToken() { 822 FormatToken *Tok = CurrentToken; 823 next(); 824 switch (Tok->Tok.getKind()) { 825 case tok::plus: 826 case tok::minus: 827 if (!Tok->Previous && Line.MustBeDeclaration) 828 Tok->setType(TT_ObjCMethodSpecifier); 829 break; 830 case tok::colon: 831 if (!Tok->Previous) 832 return false; 833 // Colons from ?: are handled in parseConditional(). 834 if (Style.Language == FormatStyle::LK_JavaScript) { 835 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop 836 (Contexts.size() == 1 && // switch/case labels 837 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) || 838 Contexts.back().ContextKind == tok::l_paren || // function params 839 Contexts.back().ContextKind == tok::l_square || // array type 840 (!Contexts.back().IsExpression && 841 Contexts.back().ContextKind == tok::l_brace) || // object type 842 (Contexts.size() == 1 && 843 Line.MustBeDeclaration)) { // method/property declaration 844 Contexts.back().IsExpression = false; 845 Tok->setType(TT_JsTypeColon); 846 break; 847 } 848 } else if (Style.isCSharp()) { 849 if (Contexts.back().InCSharpAttributeSpecifier) { 850 Tok->setType(TT_AttributeColon); 851 break; 852 } 853 if (Contexts.back().ContextKind == tok::l_paren) { 854 Tok->setType(TT_CSharpNamedArgumentColon); 855 break; 856 } 857 } 858 if (Contexts.back().ColonIsDictLiteral || 859 Style.Language == FormatStyle::LK_Proto || 860 Style.Language == FormatStyle::LK_TextProto) { 861 Tok->setType(TT_DictLiteral); 862 if (Style.Language == FormatStyle::LK_TextProto) { 863 if (FormatToken *Previous = Tok->getPreviousNonComment()) 864 Previous->setType(TT_SelectorName); 865 } 866 } else if (Contexts.back().ColonIsObjCMethodExpr || 867 Line.startsWith(TT_ObjCMethodSpecifier)) { 868 Tok->setType(TT_ObjCMethodExpr); 869 const FormatToken *BeforePrevious = Tok->Previous->Previous; 870 // Ensure we tag all identifiers in method declarations as 871 // TT_SelectorName. 872 bool UnknownIdentifierInMethodDeclaration = 873 Line.startsWith(TT_ObjCMethodSpecifier) && 874 Tok->Previous->is(tok::identifier) && Tok->Previous->is(TT_Unknown); 875 if (!BeforePrevious || 876 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 877 !(BeforePrevious->is(TT_CastRParen) || 878 (BeforePrevious->is(TT_ObjCMethodExpr) && 879 BeforePrevious->is(tok::colon))) || 880 BeforePrevious->is(tok::r_square) || 881 Contexts.back().LongestObjCSelectorName == 0 || 882 UnknownIdentifierInMethodDeclaration) { 883 Tok->Previous->setType(TT_SelectorName); 884 if (!Contexts.back().FirstObjCSelectorName) 885 Contexts.back().FirstObjCSelectorName = Tok->Previous; 886 else if (Tok->Previous->ColumnWidth > 887 Contexts.back().LongestObjCSelectorName) 888 Contexts.back().LongestObjCSelectorName = 889 Tok->Previous->ColumnWidth; 890 Tok->Previous->ParameterIndex = 891 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 892 ++Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 893 } 894 } else if (Contexts.back().ColonIsForRangeExpr) { 895 Tok->setType(TT_RangeBasedForLoopColon); 896 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 897 Tok->setType(TT_BitFieldColon); 898 } else if (Contexts.size() == 1 && 899 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) { 900 FormatToken *Prev = Tok->getPreviousNonComment(); 901 if (Prev->isOneOf(tok::r_paren, tok::kw_noexcept)) 902 Tok->setType(TT_CtorInitializerColon); 903 else if (Prev->is(tok::kw_try)) { 904 // Member initializer list within function try block. 905 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 906 if (PrevPrev && PrevPrev->isOneOf(tok::r_paren, tok::kw_noexcept)) 907 Tok->setType(TT_CtorInitializerColon); 908 } else 909 Tok->setType(TT_InheritanceColon); 910 } else if (canBeObjCSelectorComponent(*Tok->Previous) && Tok->Next && 911 (Tok->Next->isOneOf(tok::r_paren, tok::comma) || 912 (canBeObjCSelectorComponent(*Tok->Next) && Tok->Next->Next && 913 Tok->Next->Next->is(tok::colon)))) { 914 // This handles a special macro in ObjC code where selectors including 915 // the colon are passed as macro arguments. 916 Tok->setType(TT_ObjCMethodExpr); 917 } else if (Contexts.back().ContextKind == tok::l_paren) { 918 Tok->setType(TT_InlineASMColon); 919 } 920 break; 921 case tok::pipe: 922 case tok::amp: 923 // | and & in declarations/type expressions represent union and 924 // intersection types, respectively. 925 if (Style.Language == FormatStyle::LK_JavaScript && 926 !Contexts.back().IsExpression) 927 Tok->setType(TT_JsTypeOperator); 928 break; 929 case tok::kw_if: 930 case tok::kw_while: 931 if (Tok->is(tok::kw_if) && CurrentToken && 932 CurrentToken->isOneOf(tok::kw_constexpr, tok::identifier)) 933 next(); 934 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 935 next(); 936 if (!parseParens(/*LookForDecls=*/true)) 937 return false; 938 } 939 break; 940 case tok::kw_for: 941 if (Style.Language == FormatStyle::LK_JavaScript) { 942 // x.for and {for: ...} 943 if ((Tok->Previous && Tok->Previous->is(tok::period)) || 944 (Tok->Next && Tok->Next->is(tok::colon))) 945 break; 946 // JS' for await ( ... 947 if (CurrentToken && CurrentToken->is(Keywords.kw_await)) 948 next(); 949 } 950 Contexts.back().ColonIsForRangeExpr = true; 951 next(); 952 if (!parseParens()) 953 return false; 954 break; 955 case tok::l_paren: 956 // When faced with 'operator()()', the kw_operator handler incorrectly 957 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make 958 // the first two parens OverloadedOperators and the second l_paren an 959 // OverloadedOperatorLParen. 960 if (Tok->Previous && Tok->Previous->is(tok::r_paren) && 961 Tok->Previous->MatchingParen && 962 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) { 963 Tok->Previous->setType(TT_OverloadedOperator); 964 Tok->Previous->MatchingParen->setType(TT_OverloadedOperator); 965 Tok->setType(TT_OverloadedOperatorLParen); 966 } 967 968 if (!parseParens()) 969 return false; 970 if (Line.MustBeDeclaration && Contexts.size() == 1 && 971 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) && 972 !Tok->is(TT_TypeDeclarationParen) && 973 (!Tok->Previous || !Tok->Previous->isOneOf(tok::kw___attribute, 974 TT_LeadingJavaAnnotation))) 975 Line.MightBeFunctionDecl = true; 976 break; 977 case tok::l_square: 978 if (!parseSquare()) 979 return false; 980 break; 981 case tok::l_brace: 982 if (Style.Language == FormatStyle::LK_TextProto) { 983 FormatToken *Previous = Tok->getPreviousNonComment(); 984 if (Previous && Previous->getType() != TT_DictLiteral) 985 Previous->setType(TT_SelectorName); 986 } 987 if (!parseBrace()) 988 return false; 989 break; 990 case tok::less: 991 if (parseAngle()) { 992 Tok->setType(TT_TemplateOpener); 993 // In TT_Proto, we must distignuish between: 994 // map<key, value> 995 // msg < item: data > 996 // msg: < item: data > 997 // In TT_TextProto, map<key, value> does not occur. 998 if (Style.Language == FormatStyle::LK_TextProto || 999 (Style.Language == FormatStyle::LK_Proto && Tok->Previous && 1000 Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) { 1001 Tok->setType(TT_DictLiteral); 1002 FormatToken *Previous = Tok->getPreviousNonComment(); 1003 if (Previous && Previous->getType() != TT_DictLiteral) 1004 Previous->setType(TT_SelectorName); 1005 } 1006 } else { 1007 Tok->setType(TT_BinaryOperator); 1008 NonTemplateLess.insert(Tok); 1009 CurrentToken = Tok; 1010 next(); 1011 } 1012 break; 1013 case tok::r_paren: 1014 case tok::r_square: 1015 return false; 1016 case tok::r_brace: 1017 // Lines can start with '}'. 1018 if (Tok->Previous) 1019 return false; 1020 break; 1021 case tok::greater: 1022 if (Style.Language != FormatStyle::LK_TextProto) 1023 Tok->setType(TT_BinaryOperator); 1024 if (Tok->Previous && Tok->Previous->is(TT_TemplateCloser)) 1025 Tok->SpacesRequiredBefore = 1; 1026 break; 1027 case tok::kw_operator: 1028 if (Style.Language == FormatStyle::LK_TextProto || 1029 Style.Language == FormatStyle::LK_Proto) 1030 break; 1031 while (CurrentToken && 1032 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 1033 if (CurrentToken->isOneOf(tok::star, tok::amp)) 1034 CurrentToken->setType(TT_PointerOrReference); 1035 consumeToken(); 1036 if (CurrentToken && CurrentToken->is(tok::comma) && 1037 CurrentToken->Previous->isNot(tok::kw_operator)) 1038 break; 1039 if (CurrentToken && CurrentToken->Previous->isOneOf( 1040 TT_BinaryOperator, TT_UnaryOperator, tok::comma, 1041 tok::star, tok::arrow, tok::amp, tok::ampamp)) 1042 CurrentToken->Previous->setType(TT_OverloadedOperator); 1043 } 1044 if (CurrentToken && CurrentToken->is(tok::l_paren)) 1045 CurrentToken->setType(TT_OverloadedOperatorLParen); 1046 if (CurrentToken && CurrentToken->Previous->is(TT_BinaryOperator)) 1047 CurrentToken->Previous->setType(TT_OverloadedOperator); 1048 break; 1049 case tok::question: 1050 if (Tok->is(TT_CSharpNullConditionalLSquare)) { 1051 if (!parseSquare()) 1052 return false; 1053 break; 1054 } 1055 if (Tok->isOneOf(TT_CSharpNullConditional, TT_CSharpNullCoalescing)) 1056 break; 1057 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next && 1058 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren, 1059 tok::r_brace)) { 1060 // Question marks before semicolons, colons, etc. indicate optional 1061 // types (fields, parameters), e.g. 1062 // function(x?: string, y?) {...} 1063 // class X { y?; } 1064 Tok->setType(TT_JsTypeOptionalQuestion); 1065 break; 1066 } 1067 // Declarations cannot be conditional expressions, this can only be part 1068 // of a type declaration. 1069 if (Line.MustBeDeclaration && !Contexts.back().IsExpression && 1070 Style.Language == FormatStyle::LK_JavaScript) 1071 break; 1072 if (Style.isCSharp()) { 1073 // `Type?)`, `Type?>`, `Type? name;` and `Type? name =` can only be 1074 // nullable types. 1075 // Line.MustBeDeclaration will be true for `Type? name;`. 1076 if ((!Contexts.back().IsExpression && Line.MustBeDeclaration) || 1077 (Tok->Next && Tok->Next->isOneOf(tok::r_paren, tok::greater)) || 1078 (Tok->Next && Tok->Next->is(tok::identifier) && Tok->Next->Next && 1079 Tok->Next->Next->is(tok::equal))) { 1080 Tok->setType(TT_CSharpNullable); 1081 break; 1082 } 1083 } 1084 parseConditional(); 1085 break; 1086 case tok::kw_template: 1087 parseTemplateDeclaration(); 1088 break; 1089 case tok::comma: 1090 if (Contexts.back().InCtorInitializer) 1091 Tok->setType(TT_CtorInitializerComma); 1092 else if (Contexts.back().InInheritanceList) 1093 Tok->setType(TT_InheritanceComma); 1094 else if (Contexts.back().FirstStartOfName && 1095 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) { 1096 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 1097 Line.IsMultiVariableDeclStmt = true; 1098 } 1099 if (Contexts.back().IsForEachMacro) 1100 Contexts.back().IsExpression = true; 1101 break; 1102 case tok::identifier: 1103 if (Tok->isOneOf(Keywords.kw___has_include, 1104 Keywords.kw___has_include_next)) { 1105 parseHasInclude(); 1106 } 1107 if (Style.isCSharp() && Tok->is(Keywords.kw_where) && Tok->Next && 1108 Tok->Next->isNot(tok::l_paren)) { 1109 Tok->setType(TT_CSharpGenericTypeConstraint); 1110 parseCSharpGenericTypeConstraint(); 1111 } 1112 break; 1113 default: 1114 break; 1115 } 1116 return true; 1117 } 1118 1119 void parseCSharpGenericTypeConstraint() { 1120 int OpenAngleBracketsCount = 0; 1121 while (CurrentToken) { 1122 if (CurrentToken->is(tok::less)) { 1123 // parseAngle is too greedy and will consume the whole line. 1124 CurrentToken->setType(TT_TemplateOpener); 1125 ++OpenAngleBracketsCount; 1126 next(); 1127 } else if (CurrentToken->is(tok::greater)) { 1128 CurrentToken->setType(TT_TemplateCloser); 1129 --OpenAngleBracketsCount; 1130 next(); 1131 } else if (CurrentToken->is(tok::comma) && OpenAngleBracketsCount == 0) { 1132 // We allow line breaks after GenericTypeConstraintComma's 1133 // so do not flag commas in Generics as GenericTypeConstraintComma's. 1134 CurrentToken->setType(TT_CSharpGenericTypeConstraintComma); 1135 next(); 1136 } else if (CurrentToken->is(Keywords.kw_where)) { 1137 CurrentToken->setType(TT_CSharpGenericTypeConstraint); 1138 next(); 1139 } else if (CurrentToken->is(tok::colon)) { 1140 CurrentToken->setType(TT_CSharpGenericTypeConstraintColon); 1141 next(); 1142 } else { 1143 next(); 1144 } 1145 } 1146 } 1147 1148 void parseIncludeDirective() { 1149 if (CurrentToken && CurrentToken->is(tok::less)) { 1150 next(); 1151 while (CurrentToken) { 1152 // Mark tokens up to the trailing line comments as implicit string 1153 // literals. 1154 if (CurrentToken->isNot(tok::comment) && 1155 !CurrentToken->TokenText.startswith("//")) 1156 CurrentToken->setType(TT_ImplicitStringLiteral); 1157 next(); 1158 } 1159 } 1160 } 1161 1162 void parseWarningOrError() { 1163 next(); 1164 // We still want to format the whitespace left of the first token of the 1165 // warning or error. 1166 next(); 1167 while (CurrentToken) { 1168 CurrentToken->setType(TT_ImplicitStringLiteral); 1169 next(); 1170 } 1171 } 1172 1173 void parsePragma() { 1174 next(); // Consume "pragma". 1175 if (CurrentToken && 1176 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 1177 bool IsMark = CurrentToken->is(Keywords.kw_mark); 1178 next(); // Consume "mark". 1179 next(); // Consume first token (so we fix leading whitespace). 1180 while (CurrentToken) { 1181 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 1182 CurrentToken->setType(TT_ImplicitStringLiteral); 1183 next(); 1184 } 1185 } 1186 } 1187 1188 void parseHasInclude() { 1189 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 1190 return; 1191 next(); // '(' 1192 parseIncludeDirective(); 1193 next(); // ')' 1194 } 1195 1196 LineType parsePreprocessorDirective() { 1197 bool IsFirstToken = CurrentToken->IsFirst; 1198 LineType Type = LT_PreprocessorDirective; 1199 next(); 1200 if (!CurrentToken) 1201 return Type; 1202 1203 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) { 1204 // JavaScript files can contain shebang lines of the form: 1205 // #!/usr/bin/env node 1206 // Treat these like C++ #include directives. 1207 while (CurrentToken) { 1208 // Tokens cannot be comments here. 1209 CurrentToken->setType(TT_ImplicitStringLiteral); 1210 next(); 1211 } 1212 return LT_ImportStatement; 1213 } 1214 1215 if (CurrentToken->Tok.is(tok::numeric_constant)) { 1216 CurrentToken->SpacesRequiredBefore = 1; 1217 return Type; 1218 } 1219 // Hashes in the middle of a line can lead to any strange token 1220 // sequence. 1221 if (!CurrentToken->Tok.getIdentifierInfo()) 1222 return Type; 1223 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 1224 case tok::pp_include: 1225 case tok::pp_include_next: 1226 case tok::pp_import: 1227 next(); 1228 parseIncludeDirective(); 1229 Type = LT_ImportStatement; 1230 break; 1231 case tok::pp_error: 1232 case tok::pp_warning: 1233 parseWarningOrError(); 1234 break; 1235 case tok::pp_pragma: 1236 parsePragma(); 1237 break; 1238 case tok::pp_if: 1239 case tok::pp_elif: 1240 Contexts.back().IsExpression = true; 1241 next(); 1242 parseLine(); 1243 break; 1244 default: 1245 break; 1246 } 1247 while (CurrentToken) { 1248 FormatToken *Tok = CurrentToken; 1249 next(); 1250 if (Tok->is(tok::l_paren)) 1251 parseParens(); 1252 else if (Tok->isOneOf(Keywords.kw___has_include, 1253 Keywords.kw___has_include_next)) 1254 parseHasInclude(); 1255 } 1256 return Type; 1257 } 1258 1259 public: 1260 LineType parseLine() { 1261 if (!CurrentToken) 1262 return LT_Invalid; 1263 NonTemplateLess.clear(); 1264 if (CurrentToken->is(tok::hash)) 1265 return parsePreprocessorDirective(); 1266 1267 // Directly allow to 'import <string-literal>' to support protocol buffer 1268 // definitions (github.com/google/protobuf) or missing "#" (either way we 1269 // should not break the line). 1270 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 1271 if ((Style.Language == FormatStyle::LK_Java && 1272 CurrentToken->is(Keywords.kw_package)) || 1273 (Info && Info->getPPKeywordID() == tok::pp_import && 1274 CurrentToken->Next && 1275 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 1276 tok::kw_static))) { 1277 next(); 1278 parseIncludeDirective(); 1279 return LT_ImportStatement; 1280 } 1281 1282 // If this line starts and ends in '<' and '>', respectively, it is likely 1283 // part of "#define <a/b.h>". 1284 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 1285 parseIncludeDirective(); 1286 return LT_ImportStatement; 1287 } 1288 1289 // In .proto files, top-level options and package statements are very 1290 // similar to import statements and should not be line-wrapped. 1291 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 1292 CurrentToken->isOneOf(Keywords.kw_option, Keywords.kw_package)) { 1293 next(); 1294 if (CurrentToken && CurrentToken->is(tok::identifier)) { 1295 while (CurrentToken) 1296 next(); 1297 return LT_ImportStatement; 1298 } 1299 } 1300 1301 bool KeywordVirtualFound = false; 1302 bool ImportStatement = false; 1303 1304 // import {...} from '...'; 1305 if (Style.Language == FormatStyle::LK_JavaScript && 1306 CurrentToken->is(Keywords.kw_import)) 1307 ImportStatement = true; 1308 1309 while (CurrentToken) { 1310 if (CurrentToken->is(tok::kw_virtual)) 1311 KeywordVirtualFound = true; 1312 if (Style.Language == FormatStyle::LK_JavaScript) { 1313 // export {...} from '...'; 1314 // An export followed by "from 'some string';" is a re-export from 1315 // another module identified by a URI and is treated as a 1316 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 1317 // Just "export {...};" or "export class ..." should not be treated as 1318 // an import in this sense. 1319 if (Line.First->is(tok::kw_export) && 1320 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 1321 CurrentToken->Next->isStringLiteral()) 1322 ImportStatement = true; 1323 if (isClosureImportStatement(*CurrentToken)) 1324 ImportStatement = true; 1325 } 1326 if (!consumeToken()) 1327 return LT_Invalid; 1328 } 1329 if (KeywordVirtualFound) 1330 return LT_VirtualFunctionDecl; 1331 if (ImportStatement) 1332 return LT_ImportStatement; 1333 1334 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 1335 if (Contexts.back().FirstObjCSelectorName) 1336 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 1337 Contexts.back().LongestObjCSelectorName; 1338 return LT_ObjCMethodDecl; 1339 } 1340 1341 return LT_Other; 1342 } 1343 1344 private: 1345 bool isClosureImportStatement(const FormatToken &Tok) { 1346 // FIXME: Closure-library specific stuff should not be hard-coded but be 1347 // configurable. 1348 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 1349 Tok.Next->Next && 1350 (Tok.Next->Next->TokenText == "module" || 1351 Tok.Next->Next->TokenText == "provide" || 1352 Tok.Next->Next->TokenText == "require" || 1353 Tok.Next->Next->TokenText == "requireType" || 1354 Tok.Next->Next->TokenText == "forwardDeclare") && 1355 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 1356 } 1357 1358 void resetTokenMetadata(FormatToken *Token) { 1359 if (!Token) 1360 return; 1361 1362 // Reset token type in case we have already looked at it and then 1363 // recovered from an error (e.g. failure to find the matching >). 1364 if (!CurrentToken->isOneOf( 1365 TT_LambdaLSquare, TT_LambdaLBrace, TT_AttributeMacro, 1366 TT_ForEachMacro, TT_TypenameMacro, TT_FunctionLBrace, 1367 TT_ImplicitStringLiteral, TT_InlineASMBrace, TT_JsFatArrow, 1368 TT_LambdaArrow, TT_NamespaceMacro, TT_OverloadedOperator, 1369 TT_RegexLiteral, TT_TemplateString, TT_ObjCStringLiteral, 1370 TT_UntouchableMacroFunc)) 1371 CurrentToken->setType(TT_Unknown); 1372 CurrentToken->Role.reset(); 1373 CurrentToken->MatchingParen = nullptr; 1374 CurrentToken->FakeLParens.clear(); 1375 CurrentToken->FakeRParens = 0; 1376 } 1377 1378 void next() { 1379 if (CurrentToken) { 1380 CurrentToken->NestingLevel = Contexts.size() - 1; 1381 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1382 modifyContext(*CurrentToken); 1383 determineTokenType(*CurrentToken); 1384 CurrentToken = CurrentToken->Next; 1385 } 1386 1387 resetTokenMetadata(CurrentToken); 1388 } 1389 1390 /// A struct to hold information valid in a specific context, e.g. 1391 /// a pair of parenthesis. 1392 struct Context { 1393 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1394 bool IsExpression) 1395 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1396 IsExpression(IsExpression) {} 1397 1398 tok::TokenKind ContextKind; 1399 unsigned BindingStrength; 1400 bool IsExpression; 1401 unsigned LongestObjCSelectorName = 0; 1402 bool ColonIsForRangeExpr = false; 1403 bool ColonIsDictLiteral = false; 1404 bool ColonIsObjCMethodExpr = false; 1405 FormatToken *FirstObjCSelectorName = nullptr; 1406 FormatToken *FirstStartOfName = nullptr; 1407 bool CanBeExpression = true; 1408 bool InTemplateArgument = false; 1409 bool InCtorInitializer = false; 1410 bool InInheritanceList = false; 1411 bool CaretFound = false; 1412 bool IsForEachMacro = false; 1413 bool InCpp11AttributeSpecifier = false; 1414 bool InCSharpAttributeSpecifier = false; 1415 }; 1416 1417 /// Puts a new \c Context onto the stack \c Contexts for the lifetime 1418 /// of each instance. 1419 struct ScopedContextCreator { 1420 AnnotatingParser &P; 1421 1422 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1423 unsigned Increase) 1424 : P(P) { 1425 P.Contexts.push_back(Context(ContextKind, 1426 P.Contexts.back().BindingStrength + Increase, 1427 P.Contexts.back().IsExpression)); 1428 } 1429 1430 ~ScopedContextCreator() { P.Contexts.pop_back(); } 1431 }; 1432 1433 void modifyContext(const FormatToken &Current) { 1434 if (Current.getPrecedence() == prec::Assignment && 1435 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 1436 // Type aliases use `type X = ...;` in TypeScript and can be exported 1437 // using `export type ...`. 1438 !(Style.Language == FormatStyle::LK_JavaScript && 1439 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1440 Line.startsWith(tok::kw_export, Keywords.kw_type, 1441 tok::identifier))) && 1442 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 1443 Contexts.back().IsExpression = true; 1444 if (!Line.startsWith(TT_UnaryOperator)) { 1445 for (FormatToken *Previous = Current.Previous; 1446 Previous && Previous->Previous && 1447 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1448 Previous = Previous->Previous) { 1449 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1450 Previous = Previous->MatchingParen; 1451 if (!Previous) 1452 break; 1453 } 1454 if (Previous->opensScope()) 1455 break; 1456 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1457 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1458 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1459 Previous->setType(TT_PointerOrReference); 1460 } 1461 } 1462 } else if (Current.is(tok::lessless) && 1463 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1464 Contexts.back().IsExpression = true; 1465 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1466 Contexts.back().IsExpression = true; 1467 } else if (Current.is(TT_TrailingReturnArrow)) { 1468 Contexts.back().IsExpression = false; 1469 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1470 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1471 } else if (Current.Previous && 1472 Current.Previous->is(TT_CtorInitializerColon)) { 1473 Contexts.back().IsExpression = true; 1474 Contexts.back().InCtorInitializer = true; 1475 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1476 Contexts.back().InInheritanceList = true; 1477 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1478 for (FormatToken *Previous = Current.Previous; 1479 Previous && Previous->isOneOf(tok::star, tok::amp); 1480 Previous = Previous->Previous) 1481 Previous->setType(TT_PointerOrReference); 1482 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 1483 Contexts.back().IsExpression = false; 1484 } else if (Current.is(tok::kw_new)) { 1485 Contexts.back().CanBeExpression = false; 1486 } else if (Current.is(tok::semi) || 1487 (Current.is(tok::exclaim) && Current.Previous && 1488 !Current.Previous->is(tok::kw_operator))) { 1489 // This should be the condition or increment in a for-loop. 1490 // But not operator !() (can't use TT_OverloadedOperator here as its not 1491 // been annotated yet). 1492 Contexts.back().IsExpression = true; 1493 } 1494 } 1495 1496 static FormatToken *untilMatchingParen(FormatToken *Current) { 1497 // Used when `MatchingParen` is not yet established. 1498 int ParenLevel = 0; 1499 while (Current) { 1500 if (Current->is(tok::l_paren)) 1501 ParenLevel++; 1502 if (Current->is(tok::r_paren)) 1503 ParenLevel--; 1504 if (ParenLevel < 1) 1505 break; 1506 Current = Current->Next; 1507 } 1508 return Current; 1509 } 1510 1511 static bool isDeductionGuide(FormatToken &Current) { 1512 // Look for a deduction guide template<T> A(...) -> A<...>; 1513 if (Current.Previous && Current.Previous->is(tok::r_paren) && 1514 Current.startsSequence(tok::arrow, tok::identifier, tok::less)) { 1515 // Find the TemplateCloser. 1516 FormatToken *TemplateCloser = Current.Next->Next; 1517 int NestingLevel = 0; 1518 while (TemplateCloser) { 1519 // Skip over an expressions in parens A<(3 < 2)>; 1520 if (TemplateCloser->is(tok::l_paren)) { 1521 // No Matching Paren yet so skip to matching paren 1522 TemplateCloser = untilMatchingParen(TemplateCloser); 1523 } 1524 if (TemplateCloser->is(tok::less)) 1525 NestingLevel++; 1526 if (TemplateCloser->is(tok::greater)) 1527 NestingLevel--; 1528 if (NestingLevel < 1) 1529 break; 1530 TemplateCloser = TemplateCloser->Next; 1531 } 1532 // Assuming we have found the end of the template ensure its followed 1533 // with a semi-colon. 1534 if (TemplateCloser && TemplateCloser->Next && 1535 TemplateCloser->Next->is(tok::semi) && 1536 Current.Previous->MatchingParen) { 1537 // Determine if the identifier `A` prior to the A<..>; is the same as 1538 // prior to the A(..) 1539 FormatToken *LeadingIdentifier = 1540 Current.Previous->MatchingParen->Previous; 1541 1542 // Differentiate a deduction guide by seeing the 1543 // > of the template prior to the leading identifier. 1544 if (LeadingIdentifier) { 1545 FormatToken *PriorLeadingIdentifier = LeadingIdentifier->Previous; 1546 // Skip back past explicit decoration 1547 if (PriorLeadingIdentifier && 1548 PriorLeadingIdentifier->is(tok::kw_explicit)) 1549 PriorLeadingIdentifier = PriorLeadingIdentifier->Previous; 1550 1551 return (PriorLeadingIdentifier && 1552 PriorLeadingIdentifier->is(TT_TemplateCloser) && 1553 LeadingIdentifier->TokenText == Current.Next->TokenText); 1554 } 1555 } 1556 } 1557 return false; 1558 } 1559 1560 void determineTokenType(FormatToken &Current) { 1561 if (!Current.is(TT_Unknown)) 1562 // The token type is already known. 1563 return; 1564 1565 if (Style.isCSharp() && CurrentToken->is(tok::question)) { 1566 if (CurrentToken->TokenText == "??") { 1567 Current.setType(TT_CSharpNullCoalescing); 1568 return; 1569 } 1570 if (CurrentToken->TokenText == "?.") { 1571 Current.setType(TT_CSharpNullConditional); 1572 return; 1573 } 1574 if (CurrentToken->TokenText == "?[") { 1575 Current.setType(TT_CSharpNullConditionalLSquare); 1576 return; 1577 } 1578 } 1579 1580 if (Style.Language == FormatStyle::LK_JavaScript) { 1581 if (Current.is(tok::exclaim)) { 1582 if (Current.Previous && 1583 (Keywords.IsJavaScriptIdentifier( 1584 *Current.Previous, /* AcceptIdentifierName= */ true) || 1585 Current.Previous->isOneOf( 1586 tok::kw_namespace, tok::r_paren, tok::r_square, tok::r_brace, 1587 Keywords.kw_type, Keywords.kw_get, Keywords.kw_set) || 1588 Current.Previous->Tok.isLiteral())) { 1589 Current.setType(TT_JsNonNullAssertion); 1590 return; 1591 } 1592 if (Current.Next && 1593 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1594 Current.setType(TT_JsNonNullAssertion); 1595 return; 1596 } 1597 } 1598 } 1599 1600 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1601 // function declaration have been found. In this case, 'Current' is a 1602 // trailing token of this declaration and thus cannot be a name. 1603 if (Current.is(Keywords.kw_instanceof)) { 1604 Current.setType(TT_BinaryOperator); 1605 } else if (isStartOfName(Current) && 1606 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1607 Contexts.back().FirstStartOfName = &Current; 1608 Current.setType(TT_StartOfName); 1609 } else if (Current.is(tok::semi)) { 1610 // Reset FirstStartOfName after finding a semicolon so that a for loop 1611 // with multiple increment statements is not confused with a for loop 1612 // having multiple variable declarations. 1613 Contexts.back().FirstStartOfName = nullptr; 1614 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1615 AutoFound = true; 1616 } else if (Current.is(tok::arrow) && 1617 Style.Language == FormatStyle::LK_Java) { 1618 Current.setType(TT_LambdaArrow); 1619 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1620 Current.NestingLevel == 0 && 1621 !Current.Previous->is(tok::kw_operator)) { 1622 // not auto operator->() -> xxx; 1623 Current.setType(TT_TrailingReturnArrow); 1624 1625 } else if (isDeductionGuide(Current)) { 1626 // Deduction guides trailing arrow " A(...) -> A<T>;". 1627 Current.setType(TT_TrailingReturnArrow); 1628 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1629 Current.setType(determineStarAmpUsage( 1630 Current, 1631 Contexts.back().CanBeExpression && Contexts.back().IsExpression, 1632 Contexts.back().InTemplateArgument)); 1633 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1634 Current.setType(determinePlusMinusCaretUsage(Current)); 1635 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1636 Contexts.back().CaretFound = true; 1637 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1638 Current.setType(determineIncrementUsage(Current)); 1639 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1640 Current.setType(TT_UnaryOperator); 1641 } else if (Current.is(tok::question)) { 1642 if (Style.Language == FormatStyle::LK_JavaScript && 1643 Line.MustBeDeclaration && !Contexts.back().IsExpression) { 1644 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1645 // on the interface, not a ternary expression. 1646 Current.setType(TT_JsTypeOptionalQuestion); 1647 } else { 1648 Current.setType(TT_ConditionalExpr); 1649 } 1650 } else if (Current.isBinaryOperator() && 1651 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1652 (!Current.is(tok::greater) && 1653 Style.Language != FormatStyle::LK_TextProto)) { 1654 Current.setType(TT_BinaryOperator); 1655 } else if (Current.is(tok::comment)) { 1656 if (Current.TokenText.startswith("/*")) { 1657 if (Current.TokenText.endswith("*/")) 1658 Current.setType(TT_BlockComment); 1659 else 1660 // The lexer has for some reason determined a comment here. But we 1661 // cannot really handle it, if it isn't properly terminated. 1662 Current.Tok.setKind(tok::unknown); 1663 } else { 1664 Current.setType(TT_LineComment); 1665 } 1666 } else if (Current.is(tok::r_paren)) { 1667 if (rParenEndsCast(Current)) 1668 Current.setType(TT_CastRParen); 1669 if (Current.MatchingParen && Current.Next && 1670 !Current.Next->isBinaryOperator() && 1671 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1672 tok::comma, tok::period, tok::arrow, 1673 tok::coloncolon)) 1674 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1675 // Make sure this isn't the return type of an Obj-C block declaration 1676 if (AfterParen->Tok.isNot(tok::caret)) { 1677 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1678 if (BeforeParen->is(tok::identifier) && 1679 !BeforeParen->is(TT_TypenameMacro) && 1680 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1681 (!BeforeParen->Previous || 1682 BeforeParen->Previous->ClosesTemplateDeclaration)) 1683 Current.setType(TT_FunctionAnnotationRParen); 1684 } 1685 } 1686 } else if (Current.is(tok::at) && Current.Next && 1687 Style.Language != FormatStyle::LK_JavaScript && 1688 Style.Language != FormatStyle::LK_Java) { 1689 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1690 // marks declarations and properties that need special formatting. 1691 switch (Current.Next->Tok.getObjCKeywordID()) { 1692 case tok::objc_interface: 1693 case tok::objc_implementation: 1694 case tok::objc_protocol: 1695 Current.setType(TT_ObjCDecl); 1696 break; 1697 case tok::objc_property: 1698 Current.setType(TT_ObjCProperty); 1699 break; 1700 default: 1701 break; 1702 } 1703 } else if (Current.is(tok::period)) { 1704 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1705 if (PreviousNoComment && 1706 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1707 Current.setType(TT_DesignatedInitializerPeriod); 1708 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1709 Current.Previous->isOneOf(TT_JavaAnnotation, 1710 TT_LeadingJavaAnnotation)) { 1711 Current.setType(Current.Previous->getType()); 1712 } 1713 } else if (canBeObjCSelectorComponent(Current) && 1714 // FIXME(bug 36976): ObjC return types shouldn't use 1715 // TT_CastRParen. 1716 Current.Previous && Current.Previous->is(TT_CastRParen) && 1717 Current.Previous->MatchingParen && 1718 Current.Previous->MatchingParen->Previous && 1719 Current.Previous->MatchingParen->Previous->is( 1720 TT_ObjCMethodSpecifier)) { 1721 // This is the first part of an Objective-C selector name. (If there's no 1722 // colon after this, this is the only place which annotates the identifier 1723 // as a selector.) 1724 Current.setType(TT_SelectorName); 1725 } else if (Current.isOneOf(tok::identifier, tok::kw_const, 1726 tok::kw_noexcept) && 1727 Current.Previous && 1728 !Current.Previous->isOneOf(tok::equal, tok::at) && 1729 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1730 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1731 // function declaration have been found. 1732 Current.setType(TT_TrailingAnnotation); 1733 } else if ((Style.Language == FormatStyle::LK_Java || 1734 Style.Language == FormatStyle::LK_JavaScript) && 1735 Current.Previous) { 1736 if (Current.Previous->is(tok::at) && 1737 Current.isNot(Keywords.kw_interface)) { 1738 const FormatToken &AtToken = *Current.Previous; 1739 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1740 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1741 Current.setType(TT_LeadingJavaAnnotation); 1742 else 1743 Current.setType(TT_JavaAnnotation); 1744 } else if (Current.Previous->is(tok::period) && 1745 Current.Previous->isOneOf(TT_JavaAnnotation, 1746 TT_LeadingJavaAnnotation)) { 1747 Current.setType(Current.Previous->getType()); 1748 } 1749 } 1750 } 1751 1752 /// Take a guess at whether \p Tok starts a name of a function or 1753 /// variable declaration. 1754 /// 1755 /// This is a heuristic based on whether \p Tok is an identifier following 1756 /// something that is likely a type. 1757 bool isStartOfName(const FormatToken &Tok) { 1758 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1759 return false; 1760 1761 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1762 Keywords.kw_as)) 1763 return false; 1764 if (Style.Language == FormatStyle::LK_JavaScript && 1765 Tok.Previous->is(Keywords.kw_in)) 1766 return false; 1767 1768 // Skip "const" as it does not have an influence on whether this is a name. 1769 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1770 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1771 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1772 1773 if (!PreviousNotConst) 1774 return false; 1775 1776 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1777 PreviousNotConst->Previous && 1778 PreviousNotConst->Previous->is(tok::hash); 1779 1780 if (PreviousNotConst->is(TT_TemplateCloser)) 1781 return PreviousNotConst && PreviousNotConst->MatchingParen && 1782 PreviousNotConst->MatchingParen->Previous && 1783 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1784 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1785 1786 if (PreviousNotConst->is(tok::r_paren) && 1787 PreviousNotConst->is(TT_TypeDeclarationParen)) 1788 return true; 1789 1790 return (!IsPPKeyword && 1791 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) || 1792 PreviousNotConst->is(TT_PointerOrReference) || 1793 PreviousNotConst->isSimpleTypeSpecifier(); 1794 } 1795 1796 /// Determine whether ')' is ending a cast. 1797 bool rParenEndsCast(const FormatToken &Tok) { 1798 // C-style casts are only used in C++, C# and Java. 1799 if (!Style.isCSharp() && !Style.isCpp() && 1800 Style.Language != FormatStyle::LK_Java) 1801 return false; 1802 1803 // Empty parens aren't casts and there are no casts at the end of the line. 1804 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1805 return false; 1806 1807 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1808 if (LeftOfParens) { 1809 // If there is a closing parenthesis left of the current parentheses, 1810 // look past it as these might be chained casts. 1811 if (LeftOfParens->is(tok::r_paren)) { 1812 if (!LeftOfParens->MatchingParen || 1813 !LeftOfParens->MatchingParen->Previous) 1814 return false; 1815 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1816 } 1817 1818 // If there is an identifier (or with a few exceptions a keyword) right 1819 // before the parentheses, this is unlikely to be a cast. 1820 if (LeftOfParens->Tok.getIdentifierInfo() && 1821 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1822 tok::kw_delete)) 1823 return false; 1824 1825 // Certain other tokens right before the parentheses are also signals that 1826 // this cannot be a cast. 1827 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1828 TT_TemplateCloser, tok::ellipsis)) 1829 return false; 1830 } 1831 1832 if (Tok.Next->is(tok::question)) 1833 return false; 1834 1835 // `foreach((A a, B b) in someList)` should not be seen as a cast. 1836 if (Tok.Next->is(Keywords.kw_in) && Style.isCSharp()) 1837 return false; 1838 1839 // Functions which end with decorations like volatile, noexcept are unlikely 1840 // to be casts. 1841 if (Tok.Next->isOneOf(tok::kw_noexcept, tok::kw_volatile, tok::kw_const, 1842 tok::kw_throw, tok::arrow, Keywords.kw_override, 1843 Keywords.kw_final) || 1844 isCpp11AttributeSpecifier(*Tok.Next)) 1845 return false; 1846 1847 // As Java has no function types, a "(" after the ")" likely means that this 1848 // is a cast. 1849 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1850 return true; 1851 1852 // If a (non-string) literal follows, this is likely a cast. 1853 if (Tok.Next->isNot(tok::string_literal) && 1854 (Tok.Next->Tok.isLiteral() || 1855 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1856 return true; 1857 1858 // Heuristically try to determine whether the parentheses contain a type. 1859 auto IsQualifiedPointerOrReference = [](FormatToken *T) { 1860 // This is used to handle cases such as x = (foo *const)&y; 1861 assert(!T->isSimpleTypeSpecifier() && "Should have already been checked"); 1862 // Strip trailing qualifiers such as const or volatile when checking 1863 // whether the parens could be a cast to a pointer/reference type. 1864 while (T) { 1865 if (T->is(TT_AttributeParen)) { 1866 // Handle `x = (foo *__attribute__((foo)))&v;`: 1867 if (T->MatchingParen && T->MatchingParen->Previous && 1868 T->MatchingParen->Previous->is(tok::kw___attribute)) { 1869 T = T->MatchingParen->Previous->Previous; 1870 continue; 1871 } 1872 } else if (T->is(TT_AttributeSquare)) { 1873 // Handle `x = (foo *[[clang::foo]])&v;`: 1874 if (T->MatchingParen && T->MatchingParen->Previous) { 1875 T = T->MatchingParen->Previous; 1876 continue; 1877 } 1878 } else if (T->canBePointerOrReferenceQualifier()) { 1879 T = T->Previous; 1880 continue; 1881 } 1882 break; 1883 } 1884 return T && T->is(TT_PointerOrReference); 1885 }; 1886 bool ParensAreType = 1887 !Tok.Previous || 1888 Tok.Previous->isOneOf(TT_TemplateCloser, TT_TypeDeclarationParen) || 1889 Tok.Previous->isSimpleTypeSpecifier() || 1890 IsQualifiedPointerOrReference(Tok.Previous); 1891 bool ParensCouldEndDecl = 1892 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1893 if (ParensAreType && !ParensCouldEndDecl) 1894 return true; 1895 1896 // At this point, we heuristically assume that there are no casts at the 1897 // start of the line. We assume that we have found most cases where there 1898 // are by the logic above, e.g. "(void)x;". 1899 if (!LeftOfParens) 1900 return false; 1901 1902 // Certain token types inside the parentheses mean that this can't be a 1903 // cast. 1904 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 1905 Token = Token->Next) 1906 if (Token->is(TT_BinaryOperator)) 1907 return false; 1908 1909 // If the following token is an identifier or 'this', this is a cast. All 1910 // cases where this can be something else are handled above. 1911 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1912 return true; 1913 1914 if (!Tok.Next->Next) 1915 return false; 1916 1917 // If the next token after the parenthesis is a unary operator, assume 1918 // that this is cast, unless there are unexpected tokens inside the 1919 // parenthesis. 1920 bool NextIsUnary = 1921 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 1922 if (!NextIsUnary || Tok.Next->is(tok::plus) || 1923 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 1924 return false; 1925 // Search for unexpected tokens. 1926 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 1927 Prev = Prev->Previous) { 1928 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 1929 return false; 1930 } 1931 return true; 1932 } 1933 1934 /// Return the type of the given token assuming it is * or &. 1935 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 1936 bool InTemplateArgument) { 1937 if (Style.Language == FormatStyle::LK_JavaScript) 1938 return TT_BinaryOperator; 1939 1940 // && in C# must be a binary operator. 1941 if (Style.isCSharp() && Tok.is(tok::ampamp)) 1942 return TT_BinaryOperator; 1943 1944 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1945 if (!PrevToken) 1946 return TT_UnaryOperator; 1947 1948 const FormatToken *NextToken = Tok.getNextNonComment(); 1949 if (!NextToken || 1950 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_noexcept) || 1951 NextToken->canBePointerOrReferenceQualifier() || 1952 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 1953 return TT_PointerOrReference; 1954 1955 if (PrevToken->is(tok::coloncolon)) 1956 return TT_PointerOrReference; 1957 1958 if (PrevToken->is(tok::r_paren) && PrevToken->is(TT_TypeDeclarationParen)) 1959 return TT_PointerOrReference; 1960 1961 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 1962 tok::comma, tok::semi, tok::kw_return, tok::colon, 1963 tok::equal, tok::kw_delete, tok::kw_sizeof, 1964 tok::kw_throw) || 1965 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 1966 TT_UnaryOperator, TT_CastRParen)) 1967 return TT_UnaryOperator; 1968 1969 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 1970 return TT_PointerOrReference; 1971 if (NextToken->is(tok::kw_operator) && !IsExpression) 1972 return TT_PointerOrReference; 1973 if (NextToken->isOneOf(tok::comma, tok::semi)) 1974 return TT_PointerOrReference; 1975 1976 if (PrevToken->Tok.isLiteral() || 1977 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 1978 tok::kw_false, tok::r_brace) || 1979 NextToken->Tok.isLiteral() || 1980 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 1981 NextToken->isUnaryOperator() || 1982 // If we know we're in a template argument, there are no named 1983 // declarations. Thus, having an identifier on the right-hand side 1984 // indicates a binary operator. 1985 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 1986 return TT_BinaryOperator; 1987 1988 // "&&(" is quite unlikely to be two successive unary "&". 1989 if (Tok.is(tok::ampamp) && NextToken->is(tok::l_paren)) 1990 return TT_BinaryOperator; 1991 1992 // This catches some cases where evaluation order is used as control flow: 1993 // aaa && aaa->f(); 1994 if (NextToken->Tok.isAnyIdentifier()) { 1995 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 1996 if (NextNextToken && NextNextToken->is(tok::arrow)) 1997 return TT_BinaryOperator; 1998 } 1999 2000 // It is very unlikely that we are going to find a pointer or reference type 2001 // definition on the RHS of an assignment. 2002 if (IsExpression && !Contexts.back().CaretFound) 2003 return TT_BinaryOperator; 2004 2005 return TT_PointerOrReference; 2006 } 2007 2008 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 2009 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2010 if (!PrevToken) 2011 return TT_UnaryOperator; 2012 2013 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 2014 // This must be a sequence of leading unary operators. 2015 return TT_UnaryOperator; 2016 2017 // Use heuristics to recognize unary operators. 2018 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 2019 tok::question, tok::colon, tok::kw_return, 2020 tok::kw_case, tok::at, tok::l_brace, tok::kw_throw, 2021 tok::kw_co_return, tok::kw_co_yield)) 2022 return TT_UnaryOperator; 2023 2024 // There can't be two consecutive binary operators. 2025 if (PrevToken->is(TT_BinaryOperator)) 2026 return TT_UnaryOperator; 2027 2028 // Fall back to marking the token as binary operator. 2029 return TT_BinaryOperator; 2030 } 2031 2032 /// Determine whether ++/-- are pre- or post-increments/-decrements. 2033 TokenType determineIncrementUsage(const FormatToken &Tok) { 2034 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 2035 if (!PrevToken || PrevToken->is(TT_CastRParen)) 2036 return TT_UnaryOperator; 2037 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 2038 return TT_TrailingUnaryOperator; 2039 2040 return TT_UnaryOperator; 2041 } 2042 2043 SmallVector<Context, 8> Contexts; 2044 2045 const FormatStyle &Style; 2046 AnnotatedLine &Line; 2047 FormatToken *CurrentToken; 2048 bool AutoFound; 2049 const AdditionalKeywords &Keywords; 2050 2051 // Set of "<" tokens that do not open a template parameter list. If parseAngle 2052 // determines that a specific token can't be a template opener, it will make 2053 // same decision irrespective of the decisions for tokens leading up to it. 2054 // Store this information to prevent this from causing exponential runtime. 2055 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 2056 }; 2057 2058 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 2059 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 2060 2061 /// Parses binary expressions by inserting fake parenthesis based on 2062 /// operator precedence. 2063 class ExpressionParser { 2064 public: 2065 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 2066 AnnotatedLine &Line) 2067 : Style(Style), Keywords(Keywords), Current(Line.First) {} 2068 2069 /// Parse expressions with the given operator precedence. 2070 void parse(int Precedence = 0) { 2071 // Skip 'return' and ObjC selector colons as they are not part of a binary 2072 // expression. 2073 while (Current && (Current->is(tok::kw_return) || 2074 (Current->is(tok::colon) && 2075 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 2076 next(); 2077 2078 if (!Current || Precedence > PrecedenceArrowAndPeriod) 2079 return; 2080 2081 // Conditional expressions need to be parsed separately for proper nesting. 2082 if (Precedence == prec::Conditional) { 2083 parseConditionalExpr(); 2084 return; 2085 } 2086 2087 // Parse unary operators, which all have a higher precedence than binary 2088 // operators. 2089 if (Precedence == PrecedenceUnaryOperator) { 2090 parseUnaryOperator(); 2091 return; 2092 } 2093 2094 FormatToken *Start = Current; 2095 FormatToken *LatestOperator = nullptr; 2096 unsigned OperatorIndex = 0; 2097 2098 while (Current) { 2099 // Consume operators with higher precedence. 2100 parse(Precedence + 1); 2101 2102 int CurrentPrecedence = getCurrentPrecedence(); 2103 2104 if (Current && Current->is(TT_SelectorName) && 2105 Precedence == CurrentPrecedence) { 2106 if (LatestOperator) 2107 addFakeParenthesis(Start, prec::Level(Precedence)); 2108 Start = Current; 2109 } 2110 2111 // At the end of the line or when an operator with higher precedence is 2112 // found, insert fake parenthesis and return. 2113 if (!Current || 2114 (Current->closesScope() && 2115 (Current->MatchingParen || Current->is(TT_TemplateString))) || 2116 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 2117 (CurrentPrecedence == prec::Conditional && 2118 Precedence == prec::Assignment && Current->is(tok::colon))) { 2119 break; 2120 } 2121 2122 // Consume scopes: (), [], <> and {} 2123 if (Current->opensScope()) { 2124 // In fragment of a JavaScript template string can look like '}..${' and 2125 // thus close a scope and open a new one at the same time. 2126 while (Current && (!Current->closesScope() || Current->opensScope())) { 2127 next(); 2128 parse(); 2129 } 2130 next(); 2131 } else { 2132 // Operator found. 2133 if (CurrentPrecedence == Precedence) { 2134 if (LatestOperator) 2135 LatestOperator->NextOperator = Current; 2136 LatestOperator = Current; 2137 Current->OperatorIndex = OperatorIndex; 2138 ++OperatorIndex; 2139 } 2140 next(/*SkipPastLeadingComments=*/Precedence > 0); 2141 } 2142 } 2143 2144 if (LatestOperator && (Current || Precedence > 0)) { 2145 // LatestOperator->LastOperator = true; 2146 if (Precedence == PrecedenceArrowAndPeriod) { 2147 // Call expressions don't have a binary operator precedence. 2148 addFakeParenthesis(Start, prec::Unknown); 2149 } else { 2150 addFakeParenthesis(Start, prec::Level(Precedence)); 2151 } 2152 } 2153 } 2154 2155 private: 2156 /// Gets the precedence (+1) of the given token for binary operators 2157 /// and other tokens that we treat like binary operators. 2158 int getCurrentPrecedence() { 2159 if (Current) { 2160 const FormatToken *NextNonComment = Current->getNextNonComment(); 2161 if (Current->is(TT_ConditionalExpr)) 2162 return prec::Conditional; 2163 if (NextNonComment && Current->is(TT_SelectorName) && 2164 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 2165 ((Style.Language == FormatStyle::LK_Proto || 2166 Style.Language == FormatStyle::LK_TextProto) && 2167 NextNonComment->is(tok::less)))) 2168 return prec::Assignment; 2169 if (Current->is(TT_JsComputedPropertyName)) 2170 return prec::Assignment; 2171 if (Current->is(TT_LambdaArrow)) 2172 return prec::Comma; 2173 if (Current->is(TT_JsFatArrow)) 2174 return prec::Assignment; 2175 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 2176 (Current->is(tok::comment) && NextNonComment && 2177 NextNonComment->is(TT_SelectorName))) 2178 return 0; 2179 if (Current->is(TT_RangeBasedForLoopColon)) 2180 return prec::Comma; 2181 if ((Style.Language == FormatStyle::LK_Java || 2182 Style.Language == FormatStyle::LK_JavaScript) && 2183 Current->is(Keywords.kw_instanceof)) 2184 return prec::Relational; 2185 if (Style.Language == FormatStyle::LK_JavaScript && 2186 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 2187 return prec::Relational; 2188 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 2189 return Current->getPrecedence(); 2190 if (Current->isOneOf(tok::period, tok::arrow)) 2191 return PrecedenceArrowAndPeriod; 2192 if ((Style.Language == FormatStyle::LK_Java || 2193 Style.Language == FormatStyle::LK_JavaScript) && 2194 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 2195 Keywords.kw_throws)) 2196 return 0; 2197 } 2198 return -1; 2199 } 2200 2201 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 2202 Start->FakeLParens.push_back(Precedence); 2203 if (Precedence > prec::Unknown) 2204 Start->StartsBinaryExpression = true; 2205 if (Current) { 2206 FormatToken *Previous = Current->Previous; 2207 while (Previous->is(tok::comment) && Previous->Previous) 2208 Previous = Previous->Previous; 2209 ++Previous->FakeRParens; 2210 if (Precedence > prec::Unknown) 2211 Previous->EndsBinaryExpression = true; 2212 } 2213 } 2214 2215 /// Parse unary operator expressions and surround them with fake 2216 /// parentheses if appropriate. 2217 void parseUnaryOperator() { 2218 llvm::SmallVector<FormatToken *, 2> Tokens; 2219 while (Current && Current->is(TT_UnaryOperator)) { 2220 Tokens.push_back(Current); 2221 next(); 2222 } 2223 parse(PrecedenceArrowAndPeriod); 2224 for (FormatToken *Token : llvm::reverse(Tokens)) 2225 // The actual precedence doesn't matter. 2226 addFakeParenthesis(Token, prec::Unknown); 2227 } 2228 2229 void parseConditionalExpr() { 2230 while (Current && Current->isTrailingComment()) { 2231 next(); 2232 } 2233 FormatToken *Start = Current; 2234 parse(prec::LogicalOr); 2235 if (!Current || !Current->is(tok::question)) 2236 return; 2237 next(); 2238 parse(prec::Assignment); 2239 if (!Current || Current->isNot(TT_ConditionalExpr)) 2240 return; 2241 next(); 2242 parse(prec::Assignment); 2243 addFakeParenthesis(Start, prec::Conditional); 2244 } 2245 2246 void next(bool SkipPastLeadingComments = true) { 2247 if (Current) 2248 Current = Current->Next; 2249 while (Current && 2250 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 2251 Current->isTrailingComment()) 2252 Current = Current->Next; 2253 } 2254 2255 const FormatStyle &Style; 2256 const AdditionalKeywords &Keywords; 2257 FormatToken *Current; 2258 }; 2259 2260 } // end anonymous namespace 2261 2262 void TokenAnnotator::setCommentLineLevels( 2263 SmallVectorImpl<AnnotatedLine *> &Lines) { 2264 const AnnotatedLine *NextNonCommentLine = nullptr; 2265 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(), 2266 E = Lines.rend(); 2267 I != E; ++I) { 2268 bool CommentLine = true; 2269 for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) { 2270 if (!Tok->is(tok::comment)) { 2271 CommentLine = false; 2272 break; 2273 } 2274 } 2275 2276 // If the comment is currently aligned with the line immediately following 2277 // it, that's probably intentional and we should keep it. 2278 if (NextNonCommentLine && CommentLine && 2279 NextNonCommentLine->First->NewlinesBefore <= 1 && 2280 NextNonCommentLine->First->OriginalColumn == 2281 (*I)->First->OriginalColumn) { 2282 // Align comments for preprocessor lines with the # in column 0 if 2283 // preprocessor lines are not indented. Otherwise, align with the next 2284 // line. 2285 (*I)->Level = 2286 (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 2287 (NextNonCommentLine->Type == LT_PreprocessorDirective || 2288 NextNonCommentLine->Type == LT_ImportStatement)) 2289 ? 0 2290 : NextNonCommentLine->Level; 2291 } else { 2292 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr; 2293 } 2294 2295 setCommentLineLevels((*I)->Children); 2296 } 2297 } 2298 2299 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 2300 unsigned Result = 0; 2301 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 2302 Result = std::max(Result, Tok->NestingLevel); 2303 return Result; 2304 } 2305 2306 void TokenAnnotator::annotate(AnnotatedLine &Line) { 2307 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2308 E = Line.Children.end(); 2309 I != E; ++I) { 2310 annotate(**I); 2311 } 2312 AnnotatingParser Parser(Style, Line, Keywords); 2313 Line.Type = Parser.parseLine(); 2314 2315 // With very deep nesting, ExpressionParser uses lots of stack and the 2316 // formatting algorithm is very slow. We're not going to do a good job here 2317 // anyway - it's probably generated code being formatted by mistake. 2318 // Just skip the whole line. 2319 if (maxNestingDepth(Line) > 50) 2320 Line.Type = LT_Invalid; 2321 2322 if (Line.Type == LT_Invalid) 2323 return; 2324 2325 ExpressionParser ExprParser(Style, Keywords, Line); 2326 ExprParser.parse(); 2327 2328 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2329 Line.Type = LT_ObjCMethodDecl; 2330 else if (Line.startsWith(TT_ObjCDecl)) 2331 Line.Type = LT_ObjCDecl; 2332 else if (Line.startsWith(TT_ObjCProperty)) 2333 Line.Type = LT_ObjCProperty; 2334 2335 Line.First->SpacesRequiredBefore = 1; 2336 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 2337 } 2338 2339 // This function heuristically determines whether 'Current' starts the name of a 2340 // function declaration. 2341 static bool isFunctionDeclarationName(const FormatToken &Current, 2342 const AnnotatedLine &Line) { 2343 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 2344 for (; Next; Next = Next->Next) { 2345 if (Next->is(TT_OverloadedOperatorLParen)) 2346 return Next; 2347 if (Next->is(TT_OverloadedOperator)) 2348 continue; 2349 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 2350 // For 'new[]' and 'delete[]'. 2351 if (Next->Next && 2352 Next->Next->startsSequence(tok::l_square, tok::r_square)) 2353 Next = Next->Next->Next; 2354 continue; 2355 } 2356 if (Next->startsSequence(tok::l_square, tok::r_square)) { 2357 // For operator[](). 2358 Next = Next->Next; 2359 continue; 2360 } 2361 if ((Next->isSimpleTypeSpecifier() || Next->is(tok::identifier)) && 2362 Next->Next && Next->Next->isOneOf(tok::star, tok::amp, tok::ampamp)) { 2363 // For operator void*(), operator char*(), operator Foo*(). 2364 Next = Next->Next; 2365 continue; 2366 } 2367 if (Next->is(TT_TemplateOpener) && Next->MatchingParen) { 2368 Next = Next->MatchingParen; 2369 continue; 2370 } 2371 2372 break; 2373 } 2374 return nullptr; 2375 }; 2376 2377 // Find parentheses of parameter list. 2378 const FormatToken *Next = Current.Next; 2379 if (Current.is(tok::kw_operator)) { 2380 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 2381 return false; 2382 Next = skipOperatorName(Next); 2383 } else { 2384 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 2385 return false; 2386 for (; Next; Next = Next->Next) { 2387 if (Next->is(TT_TemplateOpener)) { 2388 Next = Next->MatchingParen; 2389 } else if (Next->is(tok::coloncolon)) { 2390 Next = Next->Next; 2391 if (!Next) 2392 return false; 2393 if (Next->is(tok::kw_operator)) { 2394 Next = skipOperatorName(Next->Next); 2395 break; 2396 } 2397 if (!Next->is(tok::identifier)) 2398 return false; 2399 } else if (Next->is(tok::l_paren)) { 2400 break; 2401 } else { 2402 return false; 2403 } 2404 } 2405 } 2406 2407 // Check whether parameter list can belong to a function declaration. 2408 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 2409 return false; 2410 // If the lines ends with "{", this is likely an function definition. 2411 if (Line.Last->is(tok::l_brace)) 2412 return true; 2413 if (Next->Next == Next->MatchingParen) 2414 return true; // Empty parentheses. 2415 // If there is an &/&& after the r_paren, this is likely a function. 2416 if (Next->MatchingParen->Next && 2417 Next->MatchingParen->Next->is(TT_PointerOrReference)) 2418 return true; 2419 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 2420 Tok = Tok->Next) { 2421 if (Tok->is(TT_TypeDeclarationParen)) 2422 return true; 2423 if (Tok->isOneOf(tok::l_paren, TT_TemplateOpener) && Tok->MatchingParen) { 2424 Tok = Tok->MatchingParen; 2425 continue; 2426 } 2427 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 2428 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 2429 return true; 2430 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 2431 Tok->Tok.isLiteral()) 2432 return false; 2433 } 2434 return false; 2435 } 2436 2437 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 2438 assert(Line.MightBeFunctionDecl); 2439 2440 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 2441 Style.AlwaysBreakAfterReturnType == 2442 FormatStyle::RTBS_TopLevelDefinitions) && 2443 Line.Level > 0) 2444 return false; 2445 2446 switch (Style.AlwaysBreakAfterReturnType) { 2447 case FormatStyle::RTBS_None: 2448 return false; 2449 case FormatStyle::RTBS_All: 2450 case FormatStyle::RTBS_TopLevel: 2451 return true; 2452 case FormatStyle::RTBS_AllDefinitions: 2453 case FormatStyle::RTBS_TopLevelDefinitions: 2454 return Line.mightBeFunctionDefinition(); 2455 } 2456 2457 return false; 2458 } 2459 2460 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 2461 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2462 E = Line.Children.end(); 2463 I != E; ++I) { 2464 calculateFormattingInformation(**I); 2465 } 2466 2467 Line.First->TotalLength = 2468 Line.First->IsMultiline ? Style.ColumnLimit 2469 : Line.FirstStartColumn + Line.First->ColumnWidth; 2470 FormatToken *Current = Line.First->Next; 2471 bool InFunctionDecl = Line.MightBeFunctionDecl; 2472 while (Current) { 2473 if (isFunctionDeclarationName(*Current, Line)) 2474 Current->setType(TT_FunctionDeclarationName); 2475 if (Current->is(TT_LineComment)) { 2476 if (Current->Previous->is(BK_BracedInit) && 2477 Current->Previous->opensScope()) 2478 Current->SpacesRequiredBefore = 2479 (Style.Cpp11BracedListStyle && !Style.SpacesInParentheses) ? 0 : 1; 2480 else 2481 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2482 2483 // If we find a trailing comment, iterate backwards to determine whether 2484 // it seems to relate to a specific parameter. If so, break before that 2485 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2486 // to the previous line in: 2487 // SomeFunction(a, 2488 // b, // comment 2489 // c); 2490 if (!Current->HasUnescapedNewline) { 2491 for (FormatToken *Parameter = Current->Previous; Parameter; 2492 Parameter = Parameter->Previous) { 2493 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 2494 break; 2495 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 2496 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 2497 Parameter->HasUnescapedNewline) 2498 Parameter->MustBreakBefore = true; 2499 break; 2500 } 2501 } 2502 } 2503 } else if (Current->SpacesRequiredBefore == 0 && 2504 spaceRequiredBefore(Line, *Current)) { 2505 Current->SpacesRequiredBefore = 1; 2506 } 2507 2508 Current->MustBreakBefore = 2509 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 2510 2511 if (!Current->MustBreakBefore && InFunctionDecl && 2512 Current->is(TT_FunctionDeclarationName)) 2513 Current->MustBreakBefore = mustBreakForReturnType(Line); 2514 2515 Current->CanBreakBefore = 2516 Current->MustBreakBefore || canBreakBefore(Line, *Current); 2517 unsigned ChildSize = 0; 2518 if (Current->Previous->Children.size() == 1) { 2519 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last; 2520 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 2521 : LastOfChild.TotalLength + 1; 2522 } 2523 const FormatToken *Prev = Current->Previous; 2524 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 2525 (Prev->Children.size() == 1 && 2526 Prev->Children[0]->First->MustBreakBefore) || 2527 Current->IsMultiline) 2528 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 2529 else 2530 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 2531 ChildSize + Current->SpacesRequiredBefore; 2532 2533 if (Current->is(TT_CtorInitializerColon)) 2534 InFunctionDecl = false; 2535 2536 // FIXME: Only calculate this if CanBreakBefore is true once static 2537 // initializers etc. are sorted out. 2538 // FIXME: Move magic numbers to a better place. 2539 2540 // Reduce penalty for aligning ObjC method arguments using the colon 2541 // alignment as this is the canonical way (still prefer fitting everything 2542 // into one line if possible). Trying to fit a whole expression into one 2543 // line should not force other line breaks (e.g. when ObjC method 2544 // expression is a part of other expression). 2545 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl); 2546 if (Style.Language == FormatStyle::LK_ObjC && 2547 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) { 2548 if (Current->ParameterIndex == 1) 2549 Current->SplitPenalty += 5 * Current->BindingStrength; 2550 } else { 2551 Current->SplitPenalty += 20 * Current->BindingStrength; 2552 } 2553 2554 Current = Current->Next; 2555 } 2556 2557 calculateUnbreakableTailLengths(Line); 2558 unsigned IndentLevel = Line.Level; 2559 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2560 if (Current->Role) 2561 Current->Role->precomputeFormattingInfos(Current); 2562 if (Current->MatchingParen && 2563 Current->MatchingParen->opensBlockOrBlockTypeList(Style)) { 2564 assert(IndentLevel > 0); 2565 --IndentLevel; 2566 } 2567 Current->IndentLevel = IndentLevel; 2568 if (Current->opensBlockOrBlockTypeList(Style)) 2569 ++IndentLevel; 2570 } 2571 2572 LLVM_DEBUG({ printDebugInfo(Line); }); 2573 } 2574 2575 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2576 unsigned UnbreakableTailLength = 0; 2577 FormatToken *Current = Line.Last; 2578 while (Current) { 2579 Current->UnbreakableTailLength = UnbreakableTailLength; 2580 if (Current->CanBreakBefore || 2581 Current->isOneOf(tok::comment, tok::string_literal)) { 2582 UnbreakableTailLength = 0; 2583 } else { 2584 UnbreakableTailLength += 2585 Current->ColumnWidth + Current->SpacesRequiredBefore; 2586 } 2587 Current = Current->Previous; 2588 } 2589 } 2590 2591 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2592 const FormatToken &Tok, 2593 bool InFunctionDecl) { 2594 const FormatToken &Left = *Tok.Previous; 2595 const FormatToken &Right = Tok; 2596 2597 if (Left.is(tok::semi)) 2598 return 0; 2599 2600 if (Style.Language == FormatStyle::LK_Java) { 2601 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2602 return 1; 2603 if (Right.is(Keywords.kw_implements)) 2604 return 2; 2605 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2606 return 3; 2607 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2608 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2609 return 100; 2610 if (Left.is(TT_JsTypeColon)) 2611 return 35; 2612 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2613 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2614 return 100; 2615 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2616 if (Left.opensScope() && Right.closesScope()) 2617 return 200; 2618 } 2619 2620 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2621 return 1; 2622 if (Right.is(tok::l_square)) { 2623 if (Style.Language == FormatStyle::LK_Proto) 2624 return 1; 2625 if (Left.is(tok::r_square)) 2626 return 200; 2627 // Slightly prefer formatting local lambda definitions like functions. 2628 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2629 return 35; 2630 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2631 TT_ArrayInitializerLSquare, 2632 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) 2633 return 500; 2634 } 2635 2636 if (Left.is(tok::coloncolon) || 2637 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2638 return 500; 2639 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2640 Right.is(tok::kw_operator)) { 2641 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2642 return 3; 2643 if (Left.is(TT_StartOfName)) 2644 return 110; 2645 if (InFunctionDecl && Right.NestingLevel == 0) 2646 return Style.PenaltyReturnTypeOnItsOwnLine; 2647 return 200; 2648 } 2649 if (Right.is(TT_PointerOrReference)) 2650 return 190; 2651 if (Right.is(TT_LambdaArrow)) 2652 return 110; 2653 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2654 return 160; 2655 if (Left.is(TT_CastRParen)) 2656 return 100; 2657 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2658 return 5000; 2659 if (Left.is(tok::comment)) 2660 return 1000; 2661 2662 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2663 TT_CtorInitializerColon)) 2664 return 2; 2665 2666 if (Right.isMemberAccess()) { 2667 // Breaking before the "./->" of a chained call/member access is reasonably 2668 // cheap, as formatting those with one call per line is generally 2669 // desirable. In particular, it should be cheaper to break before the call 2670 // than it is to break inside a call's parameters, which could lead to weird 2671 // "hanging" indents. The exception is the very last "./->" to support this 2672 // frequent pattern: 2673 // 2674 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2675 // dddddddd); 2676 // 2677 // which might otherwise be blown up onto many lines. Here, clang-format 2678 // won't produce "hanging" indents anyway as there is no other trailing 2679 // call. 2680 // 2681 // Also apply higher penalty is not a call as that might lead to a wrapping 2682 // like: 2683 // 2684 // aaaaaaa 2685 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2686 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2687 ? 150 2688 : 35; 2689 } 2690 2691 if (Right.is(TT_TrailingAnnotation) && 2692 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2693 // Moving trailing annotations to the next line is fine for ObjC method 2694 // declarations. 2695 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2696 return 10; 2697 // Generally, breaking before a trailing annotation is bad unless it is 2698 // function-like. It seems to be especially preferable to keep standard 2699 // annotations (i.e. "const", "final" and "override") on the same line. 2700 // Use a slightly higher penalty after ")" so that annotations like 2701 // "const override" are kept together. 2702 bool is_short_annotation = Right.TokenText.size() < 10; 2703 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 2704 } 2705 2706 // In for-loops, prefer breaking at ',' and ';'. 2707 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 2708 return 4; 2709 2710 // In Objective-C method expressions, prefer breaking before "param:" over 2711 // breaking after it. 2712 if (Right.is(TT_SelectorName)) 2713 return 0; 2714 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 2715 return Line.MightBeFunctionDecl ? 50 : 500; 2716 2717 // In Objective-C type declarations, avoid breaking after the category's 2718 // open paren (we'll prefer breaking after the protocol list's opening 2719 // angle bracket, if present). 2720 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous && 2721 Left.Previous->isOneOf(tok::identifier, tok::greater)) 2722 return 500; 2723 2724 if (Left.is(tok::l_paren) && InFunctionDecl && 2725 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 2726 return 100; 2727 if (Left.is(tok::l_paren) && Left.Previous && 2728 (Left.Previous->is(tok::kw_for) || Left.Previous->isIf())) 2729 return 1000; 2730 if (Left.is(tok::equal) && InFunctionDecl) 2731 return 110; 2732 if (Right.is(tok::r_brace)) 2733 return 1; 2734 if (Left.is(TT_TemplateOpener)) 2735 return 100; 2736 if (Left.opensScope()) { 2737 // If we aren't aligning after opening parens/braces we can always break 2738 // here unless the style does not want us to place all arguments on the 2739 // next line. 2740 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign && 2741 (Left.ParameterCount <= 1 || Style.AllowAllArgumentsOnNextLine)) 2742 return 0; 2743 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle) 2744 return 19; 2745 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 2746 : 19; 2747 } 2748 if (Left.is(TT_JavaAnnotation)) 2749 return 50; 2750 2751 if (Left.is(TT_UnaryOperator)) 2752 return 60; 2753 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 2754 Left.Previous->isLabelString() && 2755 (Left.NextOperator || Left.OperatorIndex != 0)) 2756 return 50; 2757 if (Right.is(tok::plus) && Left.isLabelString() && 2758 (Right.NextOperator || Right.OperatorIndex != 0)) 2759 return 25; 2760 if (Left.is(tok::comma)) 2761 return 1; 2762 if (Right.is(tok::lessless) && Left.isLabelString() && 2763 (Right.NextOperator || Right.OperatorIndex != 1)) 2764 return 25; 2765 if (Right.is(tok::lessless)) { 2766 // Breaking at a << is really cheap. 2767 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 2768 // Slightly prefer to break before the first one in log-like statements. 2769 return 2; 2770 return 1; 2771 } 2772 if (Left.ClosesTemplateDeclaration) 2773 return Style.PenaltyBreakTemplateDeclaration; 2774 if (Left.is(TT_ConditionalExpr)) 2775 return prec::Conditional; 2776 prec::Level Level = Left.getPrecedence(); 2777 if (Level == prec::Unknown) 2778 Level = Right.getPrecedence(); 2779 if (Level == prec::Assignment) 2780 return Style.PenaltyBreakAssignment; 2781 if (Level != prec::Unknown) 2782 return Level; 2783 2784 return 3; 2785 } 2786 2787 bool TokenAnnotator::spaceRequiredBeforeParens(const FormatToken &Right) const { 2788 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always || 2789 (Style.SpaceBeforeParens == FormatStyle::SBPO_NonEmptyParentheses && 2790 Right.ParameterCount > 0); 2791 } 2792 2793 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 2794 const FormatToken &Left, 2795 const FormatToken &Right) { 2796 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 2797 return true; 2798 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 2799 return true; 2800 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 2801 Left.Tok.getObjCKeywordID() == tok::objc_property) 2802 return true; 2803 if (Right.is(tok::hashhash)) 2804 return Left.is(tok::hash); 2805 if (Left.isOneOf(tok::hashhash, tok::hash)) 2806 return Right.is(tok::hash); 2807 if ((Left.is(tok::l_paren) && Right.is(tok::r_paren)) || 2808 (Left.is(tok::l_brace) && Left.isNot(BK_Block) && 2809 Right.is(tok::r_brace) && Right.isNot(BK_Block))) 2810 return Style.SpaceInEmptyParentheses; 2811 if (Style.SpacesInConditionalStatement) { 2812 if (Left.is(tok::l_paren) && Left.Previous && 2813 isKeywordWithCondition(*Left.Previous)) 2814 return true; 2815 if (Right.is(tok::r_paren) && Right.MatchingParen && 2816 Right.MatchingParen->Previous && 2817 isKeywordWithCondition(*Right.MatchingParen->Previous)) 2818 return true; 2819 } 2820 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 2821 return (Right.is(TT_CastRParen) || 2822 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 2823 ? Style.SpacesInCStyleCastParentheses 2824 : Style.SpacesInParentheses; 2825 if (Right.isOneOf(tok::semi, tok::comma)) 2826 return false; 2827 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) { 2828 bool IsLightweightGeneric = Right.MatchingParen && 2829 Right.MatchingParen->Next && 2830 Right.MatchingParen->Next->is(tok::colon); 2831 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList; 2832 } 2833 if (Right.is(tok::less) && Left.is(tok::kw_template)) 2834 return Style.SpaceAfterTemplateKeyword; 2835 if (Left.isOneOf(tok::exclaim, tok::tilde)) 2836 return false; 2837 if (Left.is(tok::at) && 2838 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 2839 tok::numeric_constant, tok::l_paren, tok::l_brace, 2840 tok::kw_true, tok::kw_false)) 2841 return false; 2842 if (Left.is(tok::colon)) 2843 return !Left.is(TT_ObjCMethodExpr); 2844 if (Left.is(tok::coloncolon)) 2845 return false; 2846 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 2847 if (Style.Language == FormatStyle::LK_TextProto || 2848 (Style.Language == FormatStyle::LK_Proto && 2849 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 2850 // Format empty list as `<>`. 2851 if (Left.is(tok::less) && Right.is(tok::greater)) 2852 return false; 2853 return !Style.Cpp11BracedListStyle; 2854 } 2855 return false; 2856 } 2857 if (Right.is(tok::ellipsis)) 2858 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 2859 Left.Previous->is(tok::kw_case)); 2860 if (Left.is(tok::l_square) && Right.is(tok::amp)) 2861 return Style.SpacesInSquareBrackets; 2862 if (Right.is(TT_PointerOrReference)) { 2863 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 2864 if (!Left.MatchingParen) 2865 return true; 2866 FormatToken *TokenBeforeMatchingParen = 2867 Left.MatchingParen->getPreviousNonComment(); 2868 if (!TokenBeforeMatchingParen || !Left.is(TT_TypeDeclarationParen)) 2869 return true; 2870 } 2871 // Add a space if the previous token is a pointer qualifer or the closing 2872 // parenthesis of __attribute__(()) expression and the style requires spaces 2873 // after pointer qualifiers. 2874 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_After || 2875 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 2876 (Left.is(TT_AttributeParen) || Left.canBePointerOrReferenceQualifier())) 2877 return true; 2878 return (Left.Tok.isLiteral() || 2879 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 2880 (Style.PointerAlignment != FormatStyle::PAS_Left || 2881 (Line.IsMultiVariableDeclStmt && 2882 (Left.NestingLevel == 0 || 2883 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for))))))); 2884 } 2885 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 2886 (!Left.is(TT_PointerOrReference) || 2887 (Style.PointerAlignment != FormatStyle::PAS_Right && 2888 !Line.IsMultiVariableDeclStmt))) 2889 return true; 2890 if (Left.is(TT_PointerOrReference)) { 2891 // Add a space if the next token is a pointer qualifer and the style 2892 // requires spaces before pointer qualifiers. 2893 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Before || 2894 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) && 2895 Right.canBePointerOrReferenceQualifier()) 2896 return true; 2897 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) || 2898 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 2899 !Right.is(TT_StartOfName)) || 2900 (Right.is(tok::l_brace) && Right.is(BK_Block)) || 2901 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 2902 tok::l_paren) && 2903 (Style.PointerAlignment != FormatStyle::PAS_Right && 2904 !Line.IsMultiVariableDeclStmt) && 2905 Left.Previous && 2906 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon, 2907 tok::l_square)); 2908 } 2909 // Ensure right pointer alignement with ellipsis e.g. int *...P 2910 if (Left.is(tok::ellipsis) && Left.Previous && 2911 Left.Previous->isOneOf(tok::star, tok::amp, tok::ampamp)) 2912 return Style.PointerAlignment != FormatStyle::PAS_Right; 2913 2914 if (Right.is(tok::star) && Left.is(tok::l_paren)) 2915 return false; 2916 if (Left.is(tok::star) && Right.isOneOf(tok::star, tok::amp, tok::ampamp)) 2917 return false; 2918 if (Right.isOneOf(tok::star, tok::amp, tok::ampamp)) { 2919 const FormatToken *Previous = &Left; 2920 while (Previous && !Previous->is(tok::kw_operator)) { 2921 if (Previous->is(tok::identifier) || Previous->isSimpleTypeSpecifier()) { 2922 Previous = Previous->getPreviousNonComment(); 2923 continue; 2924 } 2925 if (Previous->is(TT_TemplateCloser) && Previous->MatchingParen) { 2926 Previous = Previous->MatchingParen->getPreviousNonComment(); 2927 continue; 2928 } 2929 if (Previous->is(tok::coloncolon)) { 2930 Previous = Previous->getPreviousNonComment(); 2931 continue; 2932 } 2933 break; 2934 } 2935 // Space between the type and the * in: 2936 // operator void*() 2937 // operator char*() 2938 // operator /*comment*/ const char*() 2939 // operator volatile /*comment*/ char*() 2940 // operator Foo*() 2941 // operator C<T>*() 2942 // operator std::Foo*() 2943 // operator C<T>::D<U>*() 2944 // dependent on PointerAlignment style. 2945 if (Previous && 2946 (Previous->endsSequence(tok::kw_operator) || 2947 Previous->endsSequence(tok::kw_const, tok::kw_operator) || 2948 Previous->endsSequence(tok::kw_volatile, tok::kw_operator))) 2949 return (Style.PointerAlignment != FormatStyle::PAS_Left); 2950 } 2951 const auto SpaceRequiredForArrayInitializerLSquare = 2952 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 2953 return Style.SpacesInContainerLiterals || 2954 ((Style.Language == FormatStyle::LK_Proto || 2955 Style.Language == FormatStyle::LK_TextProto) && 2956 !Style.Cpp11BracedListStyle && 2957 LSquareTok.endsSequence(tok::l_square, tok::colon, 2958 TT_SelectorName)); 2959 }; 2960 if (Left.is(tok::l_square)) 2961 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 2962 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 2963 (Left.isOneOf(TT_ArraySubscriptLSquare, TT_StructuredBindingLSquare, 2964 TT_LambdaLSquare) && 2965 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 2966 if (Right.is(tok::r_square)) 2967 return Right.MatchingParen && 2968 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 2969 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 2970 Style)) || 2971 (Style.SpacesInSquareBrackets && 2972 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 2973 TT_StructuredBindingLSquare, 2974 TT_LambdaLSquare)) || 2975 Right.MatchingParen->is(TT_AttributeParen)); 2976 if (Right.is(tok::l_square) && 2977 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2978 TT_DesignatedInitializerLSquare, 2979 TT_StructuredBindingLSquare, TT_AttributeSquare) && 2980 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral) && 2981 !(!Left.is(tok::r_square) && Style.SpaceBeforeSquareBrackets && 2982 Right.is(TT_ArraySubscriptLSquare))) 2983 return false; 2984 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 2985 return !Left.Children.empty(); // No spaces in "{}". 2986 if ((Left.is(tok::l_brace) && Left.isNot(BK_Block)) || 2987 (Right.is(tok::r_brace) && Right.MatchingParen && 2988 Right.MatchingParen->isNot(BK_Block))) 2989 return Style.Cpp11BracedListStyle ? Style.SpacesInParentheses : true; 2990 if (Left.is(TT_BlockComment)) 2991 // No whitespace in x(/*foo=*/1), except for JavaScript. 2992 return Style.Language == FormatStyle::LK_JavaScript || 2993 !Left.TokenText.endswith("=*/"); 2994 2995 // Space between template and attribute. 2996 // e.g. template <typename T> [[nodiscard]] ... 2997 if (Left.is(TT_TemplateCloser) && Right.is(TT_AttributeSquare)) 2998 return true; 2999 if (Right.is(tok::l_paren)) { 3000 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 3001 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) 3002 return true; 3003 if (Style.SpaceBeforeParens == 3004 FormatStyle::SBPO_ControlStatementsExceptForEachMacros && 3005 Left.is(TT_ForEachMacro)) 3006 return false; 3007 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) || 3008 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never && 3009 (Left.isOneOf(tok::pp_elif, tok::kw_for, tok::kw_while, 3010 tok::kw_switch, tok::kw_case, TT_ForEachMacro, 3011 TT_ObjCForIn) || 3012 Left.isIf(Line.Type != LT_PreprocessorDirective) || 3013 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch, 3014 tok::kw_new, tok::kw_delete) && 3015 (!Left.Previous || Left.Previous->isNot(tok::period))))) || 3016 (spaceRequiredBeforeParens(Right) && 3017 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 3018 Left.is(tok::r_paren) || Left.isSimpleTypeSpecifier() || 3019 (Left.is(tok::r_square) && Left.MatchingParen && 3020 Left.MatchingParen->is(TT_LambdaLSquare))) && 3021 Line.Type != LT_PreprocessorDirective); 3022 } 3023 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 3024 return false; 3025 if (Right.is(TT_UnaryOperator)) 3026 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 3027 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 3028 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 3029 tok::r_paren) || 3030 Left.isSimpleTypeSpecifier()) && 3031 Right.is(tok::l_brace) && Right.getNextNonComment() && 3032 Right.isNot(BK_Block)) 3033 return false; 3034 if (Left.is(tok::period) || Right.is(tok::period)) 3035 return false; 3036 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L") 3037 return false; 3038 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 3039 Left.MatchingParen->Previous && 3040 (Left.MatchingParen->Previous->is(tok::period) || 3041 Left.MatchingParen->Previous->is(tok::coloncolon))) 3042 // Java call to generic function with explicit type: 3043 // A.<B<C<...>>>DoSomething(); 3044 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference. 3045 return false; 3046 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 3047 return false; 3048 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at)) 3049 // Objective-C dictionary literal -> no space after opening brace. 3050 return false; 3051 if (Right.is(tok::r_brace) && Right.MatchingParen && 3052 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at)) 3053 // Objective-C dictionary literal -> no space before closing brace. 3054 return false; 3055 if (Right.getType() == TT_TrailingAnnotation && 3056 Right.isOneOf(tok::amp, tok::ampamp) && 3057 Left.isOneOf(tok::kw_const, tok::kw_volatile) && 3058 (!Right.Next || Right.Next->is(tok::semi))) 3059 // Match const and volatile ref-qualifiers without any additional 3060 // qualifiers such as 3061 // void Fn() const &; 3062 return Style.PointerAlignment != FormatStyle::PAS_Left; 3063 return true; 3064 } 3065 3066 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 3067 const FormatToken &Right) { 3068 const FormatToken &Left = *Right.Previous; 3069 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 3070 return true; // Never ever merge two identifiers. 3071 if (Style.isCpp()) { 3072 if (Left.is(tok::kw_operator)) 3073 return Right.is(tok::coloncolon); 3074 if (Right.is(tok::l_brace) && Right.is(BK_BracedInit) && 3075 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList) 3076 return true; 3077 } else if (Style.Language == FormatStyle::LK_Proto || 3078 Style.Language == FormatStyle::LK_TextProto) { 3079 if (Right.is(tok::period) && 3080 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 3081 Keywords.kw_repeated, Keywords.kw_extend)) 3082 return true; 3083 if (Right.is(tok::l_paren) && 3084 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 3085 return true; 3086 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 3087 return true; 3088 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 3089 if (Left.is(tok::slash) || Right.is(tok::slash)) 3090 return false; 3091 if (Left.MatchingParen && 3092 Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 3093 Right.isOneOf(tok::l_brace, tok::less)) 3094 return !Style.Cpp11BracedListStyle; 3095 // A percent is probably part of a formatting specification, such as %lld. 3096 if (Left.is(tok::percent)) 3097 return false; 3098 // Preserve the existence of a space before a percent for cases like 0x%04x 3099 // and "%d %d" 3100 if (Left.is(tok::numeric_constant) && Right.is(tok::percent)) 3101 return Right.WhitespaceRange.getEnd() != Right.WhitespaceRange.getBegin(); 3102 } else if (Style.isCSharp()) { 3103 // Require spaces around '{' and before '}' unless they appear in 3104 // interpolated strings. Interpolated strings are merged into a single token 3105 // so cannot have spaces inserted by this function. 3106 3107 // No space between 'this' and '[' 3108 if (Left.is(tok::kw_this) && Right.is(tok::l_square)) 3109 return false; 3110 3111 // No space between 'new' and '(' 3112 if (Left.is(tok::kw_new) && Right.is(tok::l_paren)) 3113 return false; 3114 3115 // Space before { (including space within '{ {'). 3116 if (Right.is(tok::l_brace)) 3117 return true; 3118 3119 // Spaces inside braces. 3120 if (Left.is(tok::l_brace) && Right.isNot(tok::r_brace)) 3121 return true; 3122 3123 if (Left.isNot(tok::l_brace) && Right.is(tok::r_brace)) 3124 return true; 3125 3126 // Spaces around '=>'. 3127 if (Left.is(TT_JsFatArrow) || Right.is(TT_JsFatArrow)) 3128 return true; 3129 3130 // No spaces around attribute target colons 3131 if (Left.is(TT_AttributeColon) || Right.is(TT_AttributeColon)) 3132 return false; 3133 3134 // space between type and variable e.g. Dictionary<string,string> foo; 3135 if (Left.is(TT_TemplateCloser) && Right.is(TT_StartOfName)) 3136 return true; 3137 3138 // spaces inside square brackets. 3139 if (Left.is(tok::l_square) || Right.is(tok::r_square)) 3140 return Style.SpacesInSquareBrackets; 3141 3142 // No space before ? in nullable types. 3143 if (Right.is(TT_CSharpNullable)) 3144 return false; 3145 3146 // Require space after ? in nullable types except in generics and casts. 3147 if (Left.is(TT_CSharpNullable)) 3148 return !Right.isOneOf(TT_TemplateCloser, tok::r_paren); 3149 3150 // No space before or after '?.'. 3151 if (Left.is(TT_CSharpNullConditional) || Right.is(TT_CSharpNullConditional)) 3152 return false; 3153 3154 // Space before and after '??'. 3155 if (Left.is(TT_CSharpNullCoalescing) || Right.is(TT_CSharpNullCoalescing)) 3156 return true; 3157 3158 // No space before '?['. 3159 if (Right.is(TT_CSharpNullConditionalLSquare)) 3160 return false; 3161 3162 // No space between consecutive commas '[,,]'. 3163 if (Left.is(tok::comma) && Right.is(tok::comma)) 3164 return false; 3165 3166 // Possible space inside `?[ 0 ]`. 3167 if (Left.is(TT_CSharpNullConditionalLSquare)) 3168 return Style.SpacesInSquareBrackets; 3169 3170 // space after var in `var (key, value)` 3171 if (Left.is(Keywords.kw_var) && Right.is(tok::l_paren)) 3172 return true; 3173 3174 // space between keywords and paren e.g. "using (" 3175 if (Right.is(tok::l_paren)) 3176 if (Left.isOneOf(tok::kw_using, Keywords.kw_async, Keywords.kw_when, 3177 Keywords.kw_lock)) 3178 return Style.SpaceBeforeParens == FormatStyle::SBPO_ControlStatements || 3179 spaceRequiredBeforeParens(Right); 3180 3181 // space between method modifier and opening parenthesis of a tuple return 3182 // type 3183 if (Left.isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 3184 tok::kw_virtual, tok::kw_extern, tok::kw_static, 3185 Keywords.kw_internal, Keywords.kw_abstract, 3186 Keywords.kw_sealed, Keywords.kw_override, 3187 Keywords.kw_async, Keywords.kw_unsafe) && 3188 Right.is(tok::l_paren)) 3189 return true; 3190 } else if (Style.Language == FormatStyle::LK_JavaScript) { 3191 if (Left.is(TT_JsFatArrow)) 3192 return true; 3193 // for await ( ... 3194 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 3195 Left.Previous->is(tok::kw_for)) 3196 return true; 3197 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 3198 Right.MatchingParen) { 3199 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 3200 // An async arrow function, for example: `x = async () => foo();`, 3201 // as opposed to calling a function called async: `x = async();` 3202 if (Next && Next->is(TT_JsFatArrow)) 3203 return true; 3204 } 3205 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 3206 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 3207 return false; 3208 // In tagged template literals ("html`bar baz`"), there is no space between 3209 // the tag identifier and the template string. 3210 if (Keywords.IsJavaScriptIdentifier(Left, 3211 /* AcceptIdentifierName= */ false) && 3212 Right.is(TT_TemplateString)) 3213 return false; 3214 if (Right.is(tok::star) && 3215 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 3216 return false; 3217 if (Right.isOneOf(tok::l_brace, tok::l_square) && 3218 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 3219 Keywords.kw_extends, Keywords.kw_implements)) 3220 return true; 3221 if (Right.is(tok::l_paren)) { 3222 // JS methods can use some keywords as names (e.g. `delete()`). 3223 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 3224 return false; 3225 // Valid JS method names can include keywords, e.g. `foo.delete()` or 3226 // `bar.instanceof()`. Recognize call positions by preceding period. 3227 if (Left.Previous && Left.Previous->is(tok::period) && 3228 Left.Tok.getIdentifierInfo()) 3229 return false; 3230 // Additional unary JavaScript operators that need a space after. 3231 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 3232 tok::kw_void)) 3233 return true; 3234 } 3235 // `foo as const;` casts into a const type. 3236 if (Left.endsSequence(tok::kw_const, Keywords.kw_as)) { 3237 return false; 3238 } 3239 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 3240 tok::kw_const) || 3241 // "of" is only a keyword if it appears after another identifier 3242 // (e.g. as "const x of y" in a for loop), or after a destructuring 3243 // operation (const [x, y] of z, const {a, b} of c). 3244 (Left.is(Keywords.kw_of) && Left.Previous && 3245 (Left.Previous->Tok.is(tok::identifier) || 3246 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 3247 (!Left.Previous || !Left.Previous->is(tok::period))) 3248 return true; 3249 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 3250 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 3251 return false; 3252 if (Left.is(Keywords.kw_as) && 3253 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 3254 return true; 3255 if (Left.is(tok::kw_default) && Left.Previous && 3256 Left.Previous->is(tok::kw_export)) 3257 return true; 3258 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 3259 return true; 3260 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 3261 return false; 3262 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 3263 return false; 3264 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 3265 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 3266 return false; 3267 if (Left.is(tok::ellipsis)) 3268 return false; 3269 if (Left.is(TT_TemplateCloser) && 3270 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 3271 Keywords.kw_implements, Keywords.kw_extends)) 3272 // Type assertions ('<type>expr') are not followed by whitespace. Other 3273 // locations that should have whitespace following are identified by the 3274 // above set of follower tokens. 3275 return false; 3276 if (Right.is(TT_JsNonNullAssertion)) 3277 return false; 3278 if (Left.is(TT_JsNonNullAssertion) && 3279 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 3280 return true; // "x! as string", "x! in y" 3281 } else if (Style.Language == FormatStyle::LK_Java) { 3282 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 3283 return true; 3284 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 3285 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never; 3286 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 3287 tok::kw_protected) || 3288 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 3289 Keywords.kw_native)) && 3290 Right.is(TT_TemplateOpener)) 3291 return true; 3292 } 3293 if (Left.is(TT_ImplicitStringLiteral)) 3294 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 3295 if (Line.Type == LT_ObjCMethodDecl) { 3296 if (Left.is(TT_ObjCMethodSpecifier)) 3297 return true; 3298 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right)) 3299 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a 3300 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class 3301 // method declaration. 3302 return false; 3303 } 3304 if (Line.Type == LT_ObjCProperty && 3305 (Right.is(tok::equal) || Left.is(tok::equal))) 3306 return false; 3307 3308 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 3309 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 3310 return true; 3311 if (Right.is(TT_OverloadedOperatorLParen)) 3312 return spaceRequiredBeforeParens(Right); 3313 if (Left.is(tok::comma)) 3314 return true; 3315 if (Right.is(tok::comma)) 3316 return false; 3317 if (Right.is(TT_ObjCBlockLParen)) 3318 return true; 3319 if (Right.is(TT_CtorInitializerColon)) 3320 return Style.SpaceBeforeCtorInitializerColon; 3321 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 3322 return false; 3323 if (Right.is(TT_RangeBasedForLoopColon) && 3324 !Style.SpaceBeforeRangeBasedForLoopColon) 3325 return false; 3326 if (Left.is(TT_BitFieldColon)) 3327 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 3328 Style.BitFieldColonSpacing == FormatStyle::BFCS_After; 3329 if (Right.is(tok::colon)) { 3330 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) || 3331 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 3332 return false; 3333 if (Right.is(TT_ObjCMethodExpr)) 3334 return false; 3335 if (Left.is(tok::question)) 3336 return false; 3337 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 3338 return false; 3339 if (Right.is(TT_DictLiteral)) 3340 return Style.SpacesInContainerLiterals; 3341 if (Right.is(TT_AttributeColon)) 3342 return false; 3343 if (Right.is(TT_CSharpNamedArgumentColon)) 3344 return false; 3345 if (Right.is(TT_BitFieldColon)) 3346 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both || 3347 Style.BitFieldColonSpacing == FormatStyle::BFCS_Before; 3348 return true; 3349 } 3350 if (Left.is(TT_UnaryOperator)) { 3351 if (!Right.is(tok::l_paren)) { 3352 // The alternative operators for ~ and ! are "compl" and "not". 3353 // If they are used instead, we do not want to combine them with 3354 // the token to the right, unless that is a left paren. 3355 if (Left.is(tok::exclaim) && Left.TokenText == "not") 3356 return true; 3357 if (Left.is(tok::tilde) && Left.TokenText == "compl") 3358 return true; 3359 // Lambda captures allow for a lone &, so "&]" needs to be properly 3360 // handled. 3361 if (Left.is(tok::amp) && Right.is(tok::r_square)) 3362 return Style.SpacesInSquareBrackets; 3363 } 3364 return (Style.SpaceAfterLogicalNot && Left.is(tok::exclaim)) || 3365 Right.is(TT_BinaryOperator); 3366 } 3367 3368 // If the next token is a binary operator or a selector name, we have 3369 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 3370 if (Left.is(TT_CastRParen)) 3371 return Style.SpaceAfterCStyleCast || 3372 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 3373 3374 if (Left.is(tok::greater) && Right.is(tok::greater)) { 3375 if (Style.Language == FormatStyle::LK_TextProto || 3376 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 3377 return !Style.Cpp11BracedListStyle; 3378 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 3379 (Style.Standard < FormatStyle::LS_Cpp11 || Style.SpacesInAngles); 3380 } 3381 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 3382 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 3383 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 3384 return false; 3385 if (!Style.SpaceBeforeAssignmentOperators && Left.isNot(TT_TemplateCloser) && 3386 Right.getPrecedence() == prec::Assignment) 3387 return false; 3388 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) && 3389 (Left.is(tok::identifier) || Left.is(tok::kw_this))) 3390 return false; 3391 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 3392 // Generally don't remove existing spaces between an identifier and "::". 3393 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 3394 // this turns out to be too lenient, add analysis of the identifier itself. 3395 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 3396 if (Right.is(tok::coloncolon) && 3397 !Left.isOneOf(tok::l_brace, tok::comment, tok::l_paren)) 3398 // Put a space between < and :: in vector< ::std::string > 3399 return (Left.is(TT_TemplateOpener) && 3400 (Style.Standard < FormatStyle::LS_Cpp11 || Style.SpacesInAngles)) || 3401 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 3402 tok::kw___super, TT_TemplateOpener, 3403 TT_TemplateCloser)) || 3404 (Left.is(tok::l_paren) && Style.SpacesInParentheses); 3405 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 3406 return Style.SpacesInAngles; 3407 // Space before TT_StructuredBindingLSquare. 3408 if (Right.is(TT_StructuredBindingLSquare)) 3409 return !Left.isOneOf(tok::amp, tok::ampamp) || 3410 Style.PointerAlignment != FormatStyle::PAS_Right; 3411 // Space before & or && following a TT_StructuredBindingLSquare. 3412 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 3413 Right.isOneOf(tok::amp, tok::ampamp)) 3414 return Style.PointerAlignment != FormatStyle::PAS_Left; 3415 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 3416 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 3417 !Right.is(tok::r_paren))) 3418 return true; 3419 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) && 3420 Right.isNot(TT_FunctionTypeLParen)) 3421 return spaceRequiredBeforeParens(Right); 3422 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 3423 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 3424 return false; 3425 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 3426 Line.startsWith(tok::hash)) 3427 return true; 3428 if (Right.is(TT_TrailingUnaryOperator)) 3429 return false; 3430 if (Left.is(TT_RegexLiteral)) 3431 return false; 3432 return spaceRequiredBetween(Line, Left, Right); 3433 } 3434 3435 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 3436 static bool isAllmanBrace(const FormatToken &Tok) { 3437 return Tok.is(tok::l_brace) && Tok.is(BK_Block) && 3438 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_LambdaLBrace, TT_DictLiteral); 3439 } 3440 3441 // Returns 'true' if 'Tok' is an function argument. 3442 static bool IsFunctionArgument(const FormatToken &Tok) { 3443 return Tok.MatchingParen && Tok.MatchingParen->Next && 3444 Tok.MatchingParen->Next->isOneOf(tok::comma, tok::r_paren); 3445 } 3446 3447 static bool 3448 isItAnEmptyLambdaAllowed(const FormatToken &Tok, 3449 FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3450 return Tok.Children.empty() && ShortLambdaOption != FormatStyle::SLS_None; 3451 } 3452 3453 static bool 3454 isItAInlineLambdaAllowed(const FormatToken &Tok, 3455 FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3456 return (ShortLambdaOption == FormatStyle::SLS_Inline && 3457 IsFunctionArgument(Tok)) || 3458 (ShortLambdaOption == FormatStyle::SLS_All); 3459 } 3460 3461 static bool isOneChildWithoutMustBreakBefore(const FormatToken &Tok) { 3462 if (Tok.Children.size() != 1) 3463 return false; 3464 FormatToken *curElt = Tok.Children[0]->First; 3465 while (curElt) { 3466 if (curElt->MustBreakBefore) 3467 return false; 3468 curElt = curElt->Next; 3469 } 3470 return true; 3471 } 3472 static bool isAllmanLambdaBrace(const FormatToken &Tok) { 3473 return (Tok.is(tok::l_brace) && Tok.is(BK_Block) && 3474 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral)); 3475 } 3476 3477 static bool isAllmanBraceIncludedBreakableLambda( 3478 const FormatToken &Tok, FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3479 if (!isAllmanLambdaBrace(Tok)) 3480 return false; 3481 3482 if (isItAnEmptyLambdaAllowed(Tok, ShortLambdaOption)) 3483 return false; 3484 3485 return !isItAInlineLambdaAllowed(Tok, ShortLambdaOption) || 3486 !isOneChildWithoutMustBreakBefore(Tok); 3487 } 3488 3489 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 3490 const FormatToken &Right) { 3491 const FormatToken &Left = *Right.Previous; 3492 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 3493 return true; 3494 3495 if (Style.isCSharp()) { 3496 if (Right.is(TT_CSharpNamedArgumentColon) || 3497 Left.is(TT_CSharpNamedArgumentColon)) 3498 return false; 3499 if (Right.is(TT_CSharpGenericTypeConstraint)) 3500 return true; 3501 } else if (Style.Language == FormatStyle::LK_JavaScript) { 3502 // FIXME: This might apply to other languages and token kinds. 3503 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 3504 Left.Previous->is(tok::string_literal)) 3505 return true; 3506 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 3507 Left.Previous && Left.Previous->is(tok::equal) && 3508 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 3509 tok::kw_const) && 3510 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 3511 // above. 3512 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 3513 // Object literals on the top level of a file are treated as "enum-style". 3514 // Each key/value pair is put on a separate line, instead of bin-packing. 3515 return true; 3516 if (Left.is(tok::l_brace) && Line.Level == 0 && 3517 (Line.startsWith(tok::kw_enum) || 3518 Line.startsWith(tok::kw_const, tok::kw_enum) || 3519 Line.startsWith(tok::kw_export, tok::kw_enum) || 3520 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 3521 // JavaScript top-level enum key/value pairs are put on separate lines 3522 // instead of bin-packing. 3523 return true; 3524 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && Left.Previous && 3525 Left.Previous->is(TT_JsFatArrow)) { 3526 // JS arrow function (=> {...}). 3527 switch (Style.AllowShortLambdasOnASingleLine) { 3528 case FormatStyle::SLS_All: 3529 return false; 3530 case FormatStyle::SLS_None: 3531 return true; 3532 case FormatStyle::SLS_Empty: 3533 return !Left.Children.empty(); 3534 case FormatStyle::SLS_Inline: 3535 // allow one-lining inline (e.g. in function call args) and empty arrow 3536 // functions. 3537 return (Left.NestingLevel == 0 && Line.Level == 0) && 3538 !Left.Children.empty(); 3539 } 3540 llvm_unreachable("Unknown FormatStyle::ShortLambdaStyle enum"); 3541 } 3542 3543 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 3544 !Left.Children.empty()) 3545 // Support AllowShortFunctionsOnASingleLine for JavaScript. 3546 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 3547 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 3548 (Left.NestingLevel == 0 && Line.Level == 0 && 3549 Style.AllowShortFunctionsOnASingleLine & 3550 FormatStyle::SFS_InlineOnly); 3551 } else if (Style.Language == FormatStyle::LK_Java) { 3552 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 3553 Right.Next->is(tok::string_literal)) 3554 return true; 3555 } else if (Style.Language == FormatStyle::LK_Cpp || 3556 Style.Language == FormatStyle::LK_ObjC || 3557 Style.Language == FormatStyle::LK_Proto || 3558 Style.Language == FormatStyle::LK_TableGen || 3559 Style.Language == FormatStyle::LK_TextProto) { 3560 if (Left.isStringLiteral() && Right.isStringLiteral()) 3561 return true; 3562 } 3563 3564 // If the last token before a '}', ']', or ')' is a comma or a trailing 3565 // comment, the intention is to insert a line break after it in order to make 3566 // shuffling around entries easier. Import statements, especially in 3567 // JavaScript, can be an exception to this rule. 3568 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 3569 const FormatToken *BeforeClosingBrace = nullptr; 3570 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 3571 (Style.Language == FormatStyle::LK_JavaScript && 3572 Left.is(tok::l_paren))) && 3573 Left.isNot(BK_Block) && Left.MatchingParen) 3574 BeforeClosingBrace = Left.MatchingParen->Previous; 3575 else if (Right.MatchingParen && 3576 (Right.MatchingParen->isOneOf(tok::l_brace, 3577 TT_ArrayInitializerLSquare) || 3578 (Style.Language == FormatStyle::LK_JavaScript && 3579 Right.MatchingParen->is(tok::l_paren)))) 3580 BeforeClosingBrace = &Left; 3581 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 3582 BeforeClosingBrace->isTrailingComment())) 3583 return true; 3584 } 3585 3586 if (Right.is(tok::comment)) 3587 return Left.isNot(BK_BracedInit) && Left.isNot(TT_CtorInitializerColon) && 3588 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 3589 if (Left.isTrailingComment()) 3590 return true; 3591 if (Right.Previous->IsUnterminatedLiteral) 3592 return true; 3593 if (Right.is(tok::lessless) && Right.Next && 3594 Right.Previous->is(tok::string_literal) && 3595 Right.Next->is(tok::string_literal)) 3596 return true; 3597 if (Right.Previous->ClosesTemplateDeclaration && 3598 Right.Previous->MatchingParen && 3599 Right.Previous->MatchingParen->NestingLevel == 0 && 3600 Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes) 3601 return true; 3602 if (Right.is(TT_CtorInitializerComma) && 3603 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3604 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 3605 return true; 3606 if (Right.is(TT_CtorInitializerColon) && 3607 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3608 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 3609 return true; 3610 // Break only if we have multiple inheritance. 3611 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 3612 Right.is(TT_InheritanceComma)) 3613 return true; 3614 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 3615 // Multiline raw string literals are special wrt. line breaks. The author 3616 // has made a deliberate choice and might have aligned the contents of the 3617 // string literal accordingly. Thus, we try keep existing line breaks. 3618 return Right.IsMultiline && Right.NewlinesBefore > 0; 3619 if ((Right.Previous->is(tok::l_brace) || 3620 (Right.Previous->is(tok::less) && Right.Previous->Previous && 3621 Right.Previous->Previous->is(tok::equal))) && 3622 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 3623 // Don't put enums or option definitions onto single lines in protocol 3624 // buffers. 3625 return true; 3626 } 3627 if (Right.is(TT_InlineASMBrace)) 3628 return Right.HasUnescapedNewline; 3629 3630 auto ShortLambdaOption = Style.AllowShortLambdasOnASingleLine; 3631 if (Style.BraceWrapping.BeforeLambdaBody && 3632 (isAllmanBraceIncludedBreakableLambda(Left, ShortLambdaOption) || 3633 isAllmanBraceIncludedBreakableLambda(Right, ShortLambdaOption))) { 3634 return true; 3635 } 3636 3637 if (isAllmanBrace(Left) || isAllmanBrace(Right)) 3638 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) || 3639 (Line.startsWith(tok::kw_typedef, tok::kw_enum) && 3640 Style.BraceWrapping.AfterEnum) || 3641 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 3642 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 3643 if (Left.is(TT_ObjCBlockLBrace) && 3644 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 3645 return true; 3646 3647 if (Left.is(TT_LambdaLBrace)) { 3648 if (IsFunctionArgument(Left) && 3649 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline) 3650 return false; 3651 3652 if (Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_None || 3653 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline || 3654 (!Left.Children.empty() && 3655 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Empty)) 3656 return true; 3657 } 3658 3659 // Put multiple Java annotation on a new line. 3660 if ((Style.Language == FormatStyle::LK_Java || 3661 Style.Language == FormatStyle::LK_JavaScript) && 3662 Left.is(TT_LeadingJavaAnnotation) && 3663 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 3664 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 3665 return true; 3666 3667 if (Right.is(TT_ProtoExtensionLSquare)) 3668 return true; 3669 3670 // In text proto instances if a submessage contains at least 2 entries and at 3671 // least one of them is a submessage, like A { ... B { ... } ... }, 3672 // put all of the entries of A on separate lines by forcing the selector of 3673 // the submessage B to be put on a newline. 3674 // 3675 // Example: these can stay on one line: 3676 // a { scalar_1: 1 scalar_2: 2 } 3677 // a { b { key: value } } 3678 // 3679 // and these entries need to be on a new line even if putting them all in one 3680 // line is under the column limit: 3681 // a { 3682 // scalar: 1 3683 // b { key: value } 3684 // } 3685 // 3686 // We enforce this by breaking before a submessage field that has previous 3687 // siblings, *and* breaking before a field that follows a submessage field. 3688 // 3689 // Be careful to exclude the case [proto.ext] { ... } since the `]` is 3690 // the TT_SelectorName there, but we don't want to break inside the brackets. 3691 // 3692 // Another edge case is @submessage { key: value }, which is a common 3693 // substitution placeholder. In this case we want to keep `@` and `submessage` 3694 // together. 3695 // 3696 // We ensure elsewhere that extensions are always on their own line. 3697 if ((Style.Language == FormatStyle::LK_Proto || 3698 Style.Language == FormatStyle::LK_TextProto) && 3699 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) { 3700 // Keep `@submessage` together in: 3701 // @submessage { key: value } 3702 if (Right.Previous && Right.Previous->is(tok::at)) 3703 return false; 3704 // Look for the scope opener after selector in cases like: 3705 // selector { ... 3706 // selector: { ... 3707 // selector: @base { ... 3708 FormatToken *LBrace = Right.Next; 3709 if (LBrace && LBrace->is(tok::colon)) { 3710 LBrace = LBrace->Next; 3711 if (LBrace && LBrace->is(tok::at)) { 3712 LBrace = LBrace->Next; 3713 if (LBrace) 3714 LBrace = LBrace->Next; 3715 } 3716 } 3717 if (LBrace && 3718 // The scope opener is one of {, [, <: 3719 // selector { ... } 3720 // selector [ ... ] 3721 // selector < ... > 3722 // 3723 // In case of selector { ... }, the l_brace is TT_DictLiteral. 3724 // In case of an empty selector {}, the l_brace is not TT_DictLiteral, 3725 // so we check for immediately following r_brace. 3726 ((LBrace->is(tok::l_brace) && 3727 (LBrace->is(TT_DictLiteral) || 3728 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) || 3729 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) { 3730 // If Left.ParameterCount is 0, then this submessage entry is not the 3731 // first in its parent submessage, and we want to break before this entry. 3732 // If Left.ParameterCount is greater than 0, then its parent submessage 3733 // might contain 1 or more entries and we want to break before this entry 3734 // if it contains at least 2 entries. We deal with this case later by 3735 // detecting and breaking before the next entry in the parent submessage. 3736 if (Left.ParameterCount == 0) 3737 return true; 3738 // However, if this submessage is the first entry in its parent 3739 // submessage, Left.ParameterCount might be 1 in some cases. 3740 // We deal with this case later by detecting an entry 3741 // following a closing paren of this submessage. 3742 } 3743 3744 // If this is an entry immediately following a submessage, it will be 3745 // preceded by a closing paren of that submessage, like in: 3746 // left---. .---right 3747 // v v 3748 // sub: { ... } key: value 3749 // If there was a comment between `}` an `key` above, then `key` would be 3750 // put on a new line anyways. 3751 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square)) 3752 return true; 3753 } 3754 3755 // Deal with lambda arguments in C++ - we want consistent line breaks whether 3756 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced 3757 // as aggressive line breaks are placed when the lambda is not the last arg. 3758 if ((Style.Language == FormatStyle::LK_Cpp || 3759 Style.Language == FormatStyle::LK_ObjC) && 3760 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 && 3761 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) { 3762 // Multiple lambdas in the same function call force line breaks. 3763 if (Left.BlockParameterCount > 1) 3764 return true; 3765 3766 // A lambda followed by another arg forces a line break. 3767 if (!Left.Role) 3768 return false; 3769 auto Comma = Left.Role->lastComma(); 3770 if (!Comma) 3771 return false; 3772 auto Next = Comma->getNextNonComment(); 3773 if (!Next) 3774 return false; 3775 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret)) 3776 return true; 3777 } 3778 3779 return false; 3780 } 3781 3782 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 3783 const FormatToken &Right) { 3784 const FormatToken &Left = *Right.Previous; 3785 // Language-specific stuff. 3786 if (Style.isCSharp()) { 3787 if (Left.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon) || 3788 Right.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon)) 3789 return false; 3790 // Only break after commas for generic type constraints. 3791 if (Line.First->is(TT_CSharpGenericTypeConstraint)) 3792 return Left.is(TT_CSharpGenericTypeConstraintComma); 3793 } else if (Style.Language == FormatStyle::LK_Java) { 3794 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3795 Keywords.kw_implements)) 3796 return false; 3797 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3798 Keywords.kw_implements)) 3799 return true; 3800 } else if (Style.Language == FormatStyle::LK_JavaScript) { 3801 const FormatToken *NonComment = Right.getPreviousNonComment(); 3802 if (NonComment && 3803 NonComment->isOneOf( 3804 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 3805 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 3806 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 3807 Keywords.kw_readonly, Keywords.kw_abstract, Keywords.kw_get, 3808 Keywords.kw_set, Keywords.kw_async, Keywords.kw_await)) 3809 return false; // Otherwise automatic semicolon insertion would trigger. 3810 if (Right.NestingLevel == 0 && 3811 (Left.Tok.getIdentifierInfo() || 3812 Left.isOneOf(tok::r_square, tok::r_paren)) && 3813 Right.isOneOf(tok::l_square, tok::l_paren)) 3814 return false; // Otherwise automatic semicolon insertion would trigger. 3815 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace)) 3816 return false; 3817 if (Left.is(TT_JsTypeColon)) 3818 return true; 3819 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;"). 3820 if (Left.is(tok::exclaim) && Right.is(tok::colon)) 3821 return false; 3822 // Look for is type annotations like: 3823 // function f(): a is B { ... } 3824 // Do not break before is in these cases. 3825 if (Right.is(Keywords.kw_is)) { 3826 const FormatToken *Next = Right.getNextNonComment(); 3827 // If `is` is followed by a colon, it's likely that it's a dict key, so 3828 // ignore it for this check. 3829 // For example this is common in Polymer: 3830 // Polymer({ 3831 // is: 'name', 3832 // ... 3833 // }); 3834 if (!Next || !Next->is(tok::colon)) 3835 return false; 3836 } 3837 if (Left.is(Keywords.kw_in)) 3838 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 3839 if (Right.is(Keywords.kw_in)) 3840 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 3841 if (Right.is(Keywords.kw_as)) 3842 return false; // must not break before as in 'x as type' casts 3843 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) { 3844 // extends and infer can appear as keywords in conditional types: 3845 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types 3846 // do not break before them, as the expressions are subject to ASI. 3847 return false; 3848 } 3849 if (Left.is(Keywords.kw_as)) 3850 return true; 3851 if (Left.is(TT_JsNonNullAssertion)) 3852 return true; 3853 if (Left.is(Keywords.kw_declare) && 3854 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 3855 Keywords.kw_function, tok::kw_class, tok::kw_enum, 3856 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 3857 Keywords.kw_let, tok::kw_const)) 3858 // See grammar for 'declare' statements at: 3859 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10 3860 return false; 3861 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 3862 Right.isOneOf(tok::identifier, tok::string_literal)) 3863 return false; // must not break in "module foo { ...}" 3864 if (Right.is(TT_TemplateString) && Right.closesScope()) 3865 return false; 3866 // Don't split tagged template literal so there is a break between the tag 3867 // identifier and template string. 3868 if (Left.is(tok::identifier) && Right.is(TT_TemplateString)) { 3869 return false; 3870 } 3871 if (Left.is(TT_TemplateString) && Left.opensScope()) 3872 return true; 3873 } 3874 3875 if (Left.is(tok::at)) 3876 return false; 3877 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 3878 return false; 3879 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 3880 return !Right.is(tok::l_paren); 3881 if (Right.is(TT_PointerOrReference)) 3882 return Line.IsMultiVariableDeclStmt || 3883 (Style.PointerAlignment == FormatStyle::PAS_Right && 3884 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 3885 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 3886 Right.is(tok::kw_operator)) 3887 return true; 3888 if (Left.is(TT_PointerOrReference)) 3889 return false; 3890 if (Right.isTrailingComment()) 3891 // We rely on MustBreakBefore being set correctly here as we should not 3892 // change the "binding" behavior of a comment. 3893 // The first comment in a braced lists is always interpreted as belonging to 3894 // the first list element. Otherwise, it should be placed outside of the 3895 // list. 3896 return Left.is(BK_BracedInit) || 3897 (Left.is(TT_CtorInitializerColon) && 3898 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 3899 if (Left.is(tok::question) && Right.is(tok::colon)) 3900 return false; 3901 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 3902 return Style.BreakBeforeTernaryOperators; 3903 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 3904 return !Style.BreakBeforeTernaryOperators; 3905 if (Left.is(TT_InheritanceColon)) 3906 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon; 3907 if (Right.is(TT_InheritanceColon)) 3908 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon; 3909 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 3910 Left.isNot(TT_SelectorName)) 3911 return true; 3912 3913 if (Right.is(tok::colon) && 3914 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 3915 return false; 3916 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 3917 if (Style.Language == FormatStyle::LK_Proto || 3918 Style.Language == FormatStyle::LK_TextProto) { 3919 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 3920 return false; 3921 // Prevent cases like: 3922 // 3923 // submessage: 3924 // { key: valueeeeeeeeeeee } 3925 // 3926 // when the snippet does not fit into one line. 3927 // Prefer: 3928 // 3929 // submessage: { 3930 // key: valueeeeeeeeeeee 3931 // } 3932 // 3933 // instead, even if it is longer by one line. 3934 // 3935 // Note that this allows allows the "{" to go over the column limit 3936 // when the column limit is just between ":" and "{", but that does 3937 // not happen too often and alternative formattings in this case are 3938 // not much better. 3939 // 3940 // The code covers the cases: 3941 // 3942 // submessage: { ... } 3943 // submessage: < ... > 3944 // repeated: [ ... ] 3945 if (((Right.is(tok::l_brace) || Right.is(tok::less)) && 3946 Right.is(TT_DictLiteral)) || 3947 Right.is(TT_ArrayInitializerLSquare)) 3948 return false; 3949 } 3950 return true; 3951 } 3952 if (Right.is(tok::r_square) && Right.MatchingParen && 3953 Right.MatchingParen->is(TT_ProtoExtensionLSquare)) 3954 return false; 3955 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 3956 Right.Next->is(TT_ObjCMethodExpr))) 3957 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 3958 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 3959 return true; 3960 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 3961 return true; 3962 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 3963 TT_OverloadedOperator)) 3964 return false; 3965 if (Left.is(TT_RangeBasedForLoopColon)) 3966 return true; 3967 if (Right.is(TT_RangeBasedForLoopColon)) 3968 return false; 3969 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 3970 return true; 3971 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 3972 Left.is(tok::kw_operator)) 3973 return false; 3974 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 3975 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 3976 return false; 3977 if (Left.is(tok::equal) && Right.is(tok::l_brace) && 3978 !Style.Cpp11BracedListStyle) 3979 return false; 3980 if (Left.is(tok::l_paren) && 3981 Left.isOneOf(TT_AttributeParen, TT_TypeDeclarationParen)) 3982 return false; 3983 if (Left.is(tok::l_paren) && Left.Previous && 3984 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 3985 return false; 3986 if (Right.is(TT_ImplicitStringLiteral)) 3987 return false; 3988 3989 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser)) 3990 return false; 3991 if (Right.is(tok::r_square) && Right.MatchingParen && 3992 Right.MatchingParen->is(TT_LambdaLSquare)) 3993 return false; 3994 3995 // We only break before r_brace if there was a corresponding break before 3996 // the l_brace, which is tracked by BreakBeforeClosingBrace. 3997 if (Right.is(tok::r_brace)) 3998 return Right.MatchingParen && Right.MatchingParen->is(BK_Block); 3999 4000 // Allow breaking after a trailing annotation, e.g. after a method 4001 // declaration. 4002 if (Left.is(TT_TrailingAnnotation)) 4003 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 4004 tok::less, tok::coloncolon); 4005 4006 if (Right.is(tok::kw___attribute) || 4007 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) 4008 return !Left.is(TT_AttributeSquare); 4009 4010 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 4011 return true; 4012 4013 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 4014 return true; 4015 4016 if (Left.is(TT_CtorInitializerColon)) 4017 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 4018 if (Right.is(TT_CtorInitializerColon)) 4019 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 4020 if (Left.is(TT_CtorInitializerComma) && 4021 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 4022 return false; 4023 if (Right.is(TT_CtorInitializerComma) && 4024 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 4025 return true; 4026 if (Left.is(TT_InheritanceComma) && 4027 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 4028 return false; 4029 if (Right.is(TT_InheritanceComma) && 4030 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 4031 return true; 4032 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 4033 (Left.is(tok::less) && Right.is(tok::less))) 4034 return false; 4035 if (Right.is(TT_BinaryOperator) && 4036 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 4037 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 4038 Right.getPrecedence() != prec::Assignment)) 4039 return true; 4040 if (Left.is(TT_ArrayInitializerLSquare)) 4041 return true; 4042 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 4043 return true; 4044 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 4045 !Left.isOneOf(tok::arrowstar, tok::lessless) && 4046 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 4047 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 4048 Left.getPrecedence() == prec::Assignment)) 4049 return true; 4050 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 4051 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) 4052 return false; 4053 4054 auto ShortLambdaOption = Style.AllowShortLambdasOnASingleLine; 4055 if (Style.BraceWrapping.BeforeLambdaBody) { 4056 if (isAllmanLambdaBrace(Left)) 4057 return !isItAnEmptyLambdaAllowed(Left, ShortLambdaOption); 4058 if (isAllmanLambdaBrace(Right)) 4059 return !isItAnEmptyLambdaAllowed(Right, ShortLambdaOption); 4060 } 4061 4062 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 4063 tok::kw_class, tok::kw_struct, tok::comment) || 4064 Right.isMemberAccess() || 4065 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 4066 tok::colon, tok::l_square, tok::at) || 4067 (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace)) || 4068 (Left.is(tok::r_paren) && 4069 Right.isOneOf(tok::identifier, tok::kw_const)) || 4070 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 4071 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 4072 } 4073 4074 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 4075 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 4076 const FormatToken *Tok = Line.First; 4077 while (Tok) { 4078 llvm::errs() << " M=" << Tok->MustBreakBefore 4079 << " C=" << Tok->CanBreakBefore 4080 << " T=" << getTokenTypeName(Tok->getType()) 4081 << " S=" << Tok->SpacesRequiredBefore 4082 << " F=" << Tok->Finalized << " B=" << Tok->BlockParameterCount 4083 << " BK=" << Tok->getBlockKind() << " P=" << Tok->SplitPenalty 4084 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 4085 << " PPK=" << Tok->getPackingKind() << " FakeLParens="; 4086 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 4087 llvm::errs() << Tok->FakeLParens[i] << "/"; 4088 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 4089 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo(); 4090 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 4091 if (!Tok->Next) 4092 assert(Tok == Line.Last); 4093 Tok = Tok->Next; 4094 } 4095 llvm::errs() << "----\n"; 4096 } 4097 4098 } // namespace format 4099 } // namespace clang 4100