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 "llvm/ADT/SmallPtrSet.h" 19 #include "llvm/Support/Debug.h" 20 21 #define DEBUG_TYPE "format-token-annotator" 22 23 namespace clang { 24 namespace format { 25 26 namespace { 27 28 /// Returns \c true if the token can be used as an identifier in 29 /// an Objective-C \c @selector, \c false otherwise. 30 /// 31 /// Because getFormattingLangOpts() always lexes source code as 32 /// Objective-C++, C++ keywords like \c new and \c delete are 33 /// lexed as tok::kw_*, not tok::identifier, even for Objective-C. 34 /// 35 /// For Objective-C and Objective-C++, both identifiers and keywords 36 /// are valid inside @selector(...) (or a macro which 37 /// invokes @selector(...)). So, we allow treat any identifier or 38 /// keyword as a potential Objective-C selector component. 39 static bool canBeObjCSelectorComponent(const FormatToken &Tok) { 40 return Tok.Tok.getIdentifierInfo() != nullptr; 41 } 42 43 /// A parser that gathers additional information about tokens. 44 /// 45 /// The \c TokenAnnotator tries to match parenthesis and square brakets and 46 /// store a parenthesis levels. It also tries to resolve matching "<" and ">" 47 /// into template parameter lists. 48 class AnnotatingParser { 49 public: 50 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line, 51 const AdditionalKeywords &Keywords) 52 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false), 53 Keywords(Keywords) { 54 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false)); 55 resetTokenMetadata(CurrentToken); 56 } 57 58 private: 59 bool parseAngle() { 60 if (!CurrentToken || !CurrentToken->Previous) 61 return false; 62 if (NonTemplateLess.count(CurrentToken->Previous)) 63 return false; 64 65 const FormatToken &Previous = *CurrentToken->Previous; // The '<'. 66 if (Previous.Previous) { 67 if (Previous.Previous->Tok.isLiteral()) 68 return false; 69 if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 && 70 (!Previous.Previous->MatchingParen || 71 !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen))) 72 return false; 73 } 74 75 FormatToken *Left = CurrentToken->Previous; 76 Left->ParentBracket = Contexts.back().ContextKind; 77 ScopedContextCreator ContextCreator(*this, tok::less, 12); 78 79 // If this angle is in the context of an expression, we need to be more 80 // hesitant to detect it as opening template parameters. 81 bool InExprContext = Contexts.back().IsExpression; 82 83 Contexts.back().IsExpression = false; 84 // If there's a template keyword before the opening angle bracket, this is a 85 // template parameter, not an argument. 86 Contexts.back().InTemplateArgument = 87 Left->Previous && Left->Previous->Tok.isNot(tok::kw_template); 88 89 if (Style.Language == FormatStyle::LK_Java && 90 CurrentToken->is(tok::question)) 91 next(); 92 93 while (CurrentToken) { 94 if (CurrentToken->is(tok::greater)) { 95 Left->MatchingParen = CurrentToken; 96 CurrentToken->MatchingParen = Left; 97 // In TT_Proto, we must distignuish between: 98 // map<key, value> 99 // msg < item: data > 100 // msg: < item: data > 101 // In TT_TextProto, map<key, value> does not occur. 102 if (Style.Language == FormatStyle::LK_TextProto || 103 (Style.Language == FormatStyle::LK_Proto && Left->Previous && 104 Left->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) 105 CurrentToken->Type = TT_DictLiteral; 106 else 107 CurrentToken->Type = TT_TemplateCloser; 108 next(); 109 return true; 110 } 111 if (CurrentToken->is(tok::question) && 112 Style.Language == FormatStyle::LK_Java) { 113 next(); 114 continue; 115 } 116 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) || 117 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext && 118 Style.Language != FormatStyle::LK_Proto && 119 Style.Language != FormatStyle::LK_TextProto)) 120 return false; 121 // If a && or || is found and interpreted as a binary operator, this set 122 // of angles is likely part of something like "a < b && c > d". If the 123 // angles are inside an expression, the ||/&& might also be a binary 124 // operator that was misinterpreted because we are parsing template 125 // parameters. 126 // FIXME: This is getting out of hand, write a decent parser. 127 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) && 128 CurrentToken->Previous->is(TT_BinaryOperator) && 129 Contexts[Contexts.size() - 2].IsExpression && 130 !Line.startsWith(tok::kw_template)) 131 return false; 132 updateParameterCount(Left, CurrentToken); 133 if (Style.Language == FormatStyle::LK_Proto) { 134 if (FormatToken *Previous = CurrentToken->getPreviousNonComment()) { 135 if (CurrentToken->is(tok::colon) || 136 (CurrentToken->isOneOf(tok::l_brace, tok::less) && 137 Previous->isNot(tok::colon))) 138 Previous->Type = TT_SelectorName; 139 } 140 } 141 if (!consumeToken()) 142 return false; 143 } 144 return false; 145 } 146 147 bool parseParens(bool LookForDecls = false) { 148 if (!CurrentToken) 149 return false; 150 FormatToken *Left = CurrentToken->Previous; 151 Left->ParentBracket = Contexts.back().ContextKind; 152 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1); 153 154 // FIXME: This is a bit of a hack. Do better. 155 Contexts.back().ColonIsForRangeExpr = 156 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr; 157 158 bool StartsObjCMethodExpr = false; 159 if (FormatToken *MaybeSel = Left->Previous) { 160 // @selector( starts a selector. 161 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous && 162 MaybeSel->Previous->is(tok::at)) { 163 StartsObjCMethodExpr = true; 164 } 165 } 166 167 if (Left->is(TT_OverloadedOperatorLParen)) { 168 Contexts.back().IsExpression = false; 169 } else if (Style.Language == FormatStyle::LK_JavaScript && 170 (Line.startsWith(Keywords.kw_type, tok::identifier) || 171 Line.startsWith(tok::kw_export, Keywords.kw_type, 172 tok::identifier))) { 173 // type X = (...); 174 // export type X = (...); 175 Contexts.back().IsExpression = false; 176 } else if (Left->Previous && 177 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_decltype, 178 tok::kw_if, tok::kw_while, tok::l_paren, 179 tok::comma) || 180 Left->Previous->endsSequence(tok::kw_constexpr, tok::kw_if) || 181 Left->Previous->is(TT_BinaryOperator))) { 182 // static_assert, if and while usually contain expressions. 183 Contexts.back().IsExpression = true; 184 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 185 (Left->Previous->is(Keywords.kw_function) || 186 (Left->Previous->endsSequence(tok::identifier, 187 Keywords.kw_function)))) { 188 // function(...) or function f(...) 189 Contexts.back().IsExpression = false; 190 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 191 Left->Previous->is(TT_JsTypeColon)) { 192 // let x: (SomeType); 193 Contexts.back().IsExpression = false; 194 } else if (Left->Previous && Left->Previous->is(tok::r_square) && 195 Left->Previous->MatchingParen && 196 Left->Previous->MatchingParen->is(TT_LambdaLSquare)) { 197 // This is a parameter list of a lambda expression. 198 Contexts.back().IsExpression = false; 199 } else if (Line.InPPDirective && 200 (!Left->Previous || !Left->Previous->is(tok::identifier))) { 201 Contexts.back().IsExpression = true; 202 } else if (Contexts[Contexts.size() - 2].CaretFound) { 203 // This is the parameter list of an ObjC block. 204 Contexts.back().IsExpression = false; 205 } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) { 206 Left->Type = TT_AttributeParen; 207 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) { 208 // The first argument to a foreach macro is a declaration. 209 Contexts.back().IsForEachMacro = true; 210 Contexts.back().IsExpression = false; 211 } else if (Left->Previous && Left->Previous->MatchingParen && 212 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) { 213 Contexts.back().IsExpression = false; 214 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) { 215 bool IsForOrCatch = 216 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch); 217 Contexts.back().IsExpression = !IsForOrCatch; 218 } 219 220 if (StartsObjCMethodExpr) { 221 Contexts.back().ColonIsObjCMethodExpr = true; 222 Left->Type = TT_ObjCMethodExpr; 223 } 224 225 // MightBeFunctionType and ProbablyFunctionType are used for 226 // function pointer and reference types as well as Objective-C 227 // block types: 228 // 229 // void (*FunctionPointer)(void); 230 // void (&FunctionReference)(void); 231 // void (^ObjCBlock)(void); 232 bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression; 233 bool ProbablyFunctionType = 234 CurrentToken->isOneOf(tok::star, tok::amp, tok::caret); 235 bool HasMultipleLines = false; 236 bool HasMultipleParametersOnALine = false; 237 bool MightBeObjCForRangeLoop = 238 Left->Previous && Left->Previous->is(tok::kw_for); 239 FormatToken *PossibleObjCForInToken = nullptr; 240 while (CurrentToken) { 241 // LookForDecls is set when "if (" has been seen. Check for 242 // 'identifier' '*' 'identifier' followed by not '=' -- this 243 // '*' has to be a binary operator but determineStarAmpUsage() will 244 // categorize it as an unary operator, so set the right type here. 245 if (LookForDecls && CurrentToken->Next) { 246 FormatToken *Prev = CurrentToken->getPreviousNonComment(); 247 if (Prev) { 248 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 249 FormatToken *Next = CurrentToken->Next; 250 if (PrevPrev && PrevPrev->is(tok::identifier) && 251 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) && 252 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) { 253 Prev->Type = TT_BinaryOperator; 254 LookForDecls = false; 255 } 256 } 257 } 258 259 if (CurrentToken->Previous->is(TT_PointerOrReference) && 260 CurrentToken->Previous->Previous->isOneOf(tok::l_paren, 261 tok::coloncolon)) 262 ProbablyFunctionType = true; 263 if (CurrentToken->is(tok::comma)) 264 MightBeFunctionType = false; 265 if (CurrentToken->Previous->is(TT_BinaryOperator)) 266 Contexts.back().IsExpression = true; 267 if (CurrentToken->is(tok::r_paren)) { 268 if (MightBeFunctionType && ProbablyFunctionType && CurrentToken->Next && 269 (CurrentToken->Next->is(tok::l_paren) || 270 (CurrentToken->Next->is(tok::l_square) && Line.MustBeDeclaration))) 271 Left->Type = Left->Next->is(tok::caret) ? TT_ObjCBlockLParen 272 : TT_FunctionTypeLParen; 273 Left->MatchingParen = CurrentToken; 274 CurrentToken->MatchingParen = Left; 275 276 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) && 277 Left->Previous && Left->Previous->is(tok::l_paren)) { 278 // Detect the case where macros are used to generate lambdas or 279 // function bodies, e.g.: 280 // auto my_lambda = MARCO((Type *type, int i) { .. body .. }); 281 for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) { 282 if (Tok->is(TT_BinaryOperator) && 283 Tok->isOneOf(tok::star, tok::amp, tok::ampamp)) 284 Tok->Type = TT_PointerOrReference; 285 } 286 } 287 288 if (StartsObjCMethodExpr) { 289 CurrentToken->Type = TT_ObjCMethodExpr; 290 if (Contexts.back().FirstObjCSelectorName) { 291 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 292 Contexts.back().LongestObjCSelectorName; 293 } 294 } 295 296 if (Left->is(TT_AttributeParen)) 297 CurrentToken->Type = TT_AttributeParen; 298 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation)) 299 CurrentToken->Type = TT_JavaAnnotation; 300 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation)) 301 CurrentToken->Type = TT_LeadingJavaAnnotation; 302 if (Left->Previous && Left->Previous->is(TT_AttributeSquare)) 303 CurrentToken->Type = TT_AttributeSquare; 304 305 if (!HasMultipleLines) 306 Left->PackingKind = PPK_Inconclusive; 307 else if (HasMultipleParametersOnALine) 308 Left->PackingKind = PPK_BinPacked; 309 else 310 Left->PackingKind = PPK_OnePerLine; 311 312 next(); 313 return true; 314 } 315 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace)) 316 return false; 317 318 if (CurrentToken->is(tok::l_brace)) 319 Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen 320 if (CurrentToken->is(tok::comma) && CurrentToken->Next && 321 !CurrentToken->Next->HasUnescapedNewline && 322 !CurrentToken->Next->isTrailingComment()) 323 HasMultipleParametersOnALine = true; 324 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) || 325 CurrentToken->Previous->isSimpleTypeSpecifier()) && 326 !CurrentToken->is(tok::l_brace)) 327 Contexts.back().IsExpression = false; 328 if (CurrentToken->isOneOf(tok::semi, tok::colon)) { 329 MightBeObjCForRangeLoop = false; 330 if (PossibleObjCForInToken) { 331 PossibleObjCForInToken->Type = TT_Unknown; 332 PossibleObjCForInToken = nullptr; 333 } 334 } 335 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) { 336 PossibleObjCForInToken = CurrentToken; 337 PossibleObjCForInToken->Type = TT_ObjCForIn; 338 } 339 // When we discover a 'new', we set CanBeExpression to 'false' in order to 340 // parse the type correctly. Reset that after a comma. 341 if (CurrentToken->is(tok::comma)) 342 Contexts.back().CanBeExpression = true; 343 344 FormatToken *Tok = CurrentToken; 345 if (!consumeToken()) 346 return false; 347 updateParameterCount(Left, Tok); 348 if (CurrentToken && CurrentToken->HasUnescapedNewline) 349 HasMultipleLines = true; 350 } 351 return false; 352 } 353 354 bool isCSharpAttributeSpecifier(const FormatToken &Tok) { 355 if (!Style.isCSharp()) 356 return false; 357 358 const FormatToken *AttrTok = Tok.Next; 359 if (!AttrTok) 360 return false; 361 362 // Just an empty declaration e.g. string []. 363 if (AttrTok->is(tok::r_square)) 364 return false; 365 366 // Move along the tokens inbetween the '[' and ']' e.g. [STAThread]. 367 while (AttrTok && AttrTok->isNot(tok::r_square)) { 368 AttrTok = AttrTok->Next; 369 } 370 371 if (!AttrTok) 372 return false; 373 374 // Move past the end of ']'. 375 AttrTok = AttrTok->Next; 376 if (!AttrTok) 377 return false; 378 379 // Limit this to being an access modifier that follows. 380 if (AttrTok->isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 381 tok::kw_class, tok::kw_static, tok::l_square, 382 Keywords.kw_internal)) { 383 return true; 384 } 385 return false; 386 } 387 388 bool isCpp11AttributeSpecifier(const FormatToken &Tok) { 389 if (!Style.isCpp() || !Tok.startsSequence(tok::l_square, tok::l_square)) 390 return false; 391 const FormatToken *AttrTok = Tok.Next->Next; 392 if (!AttrTok) 393 return false; 394 // C++17 '[[using ns: foo, bar(baz, blech)]]' 395 // We assume nobody will name an ObjC variable 'using'. 396 if (AttrTok->startsSequence(tok::kw_using, tok::identifier, tok::colon)) 397 return true; 398 if (AttrTok->isNot(tok::identifier)) 399 return false; 400 while (AttrTok && !AttrTok->startsSequence(tok::r_square, tok::r_square)) { 401 // ObjC message send. We assume nobody will use : in a C++11 attribute 402 // specifier parameter, although this is technically valid: 403 // [[foo(:)]] 404 if (AttrTok->is(tok::colon) || 405 AttrTok->startsSequence(tok::identifier, tok::identifier) || 406 AttrTok->startsSequence(tok::r_paren, tok::identifier)) 407 return false; 408 if (AttrTok->is(tok::ellipsis)) 409 return true; 410 AttrTok = AttrTok->Next; 411 } 412 return AttrTok && AttrTok->startsSequence(tok::r_square, tok::r_square); 413 } 414 415 bool parseSquare() { 416 if (!CurrentToken) 417 return false; 418 419 // A '[' could be an index subscript (after an identifier or after 420 // ')' or ']'), it could be the start of an Objective-C method 421 // expression, it could the start of an Objective-C array literal, 422 // or it could be a C++ attribute specifier [[foo::bar]]. 423 FormatToken *Left = CurrentToken->Previous; 424 Left->ParentBracket = Contexts.back().ContextKind; 425 FormatToken *Parent = Left->getPreviousNonComment(); 426 427 // Cases where '>' is followed by '['. 428 // In C++, this can happen either in array of templates (foo<int>[10]) 429 // or when array is a nested template type (unique_ptr<type1<type2>[]>). 430 bool CppArrayTemplates = 431 Style.isCpp() && Parent && Parent->is(TT_TemplateCloser) && 432 (Contexts.back().CanBeExpression || Contexts.back().IsExpression || 433 Contexts.back().InTemplateArgument); 434 435 bool IsCpp11AttributeSpecifier = isCpp11AttributeSpecifier(*Left) || 436 Contexts.back().InCpp11AttributeSpecifier; 437 438 // Treat C# Attributes [STAThread] much like C++ attributes [[...]]. 439 bool IsCSharp11AttributeSpecifier = 440 isCSharpAttributeSpecifier(*Left) || 441 Contexts.back().InCSharpAttributeSpecifier; 442 443 bool InsideInlineASM = Line.startsWith(tok::kw_asm); 444 bool IsCppStructuredBinding = Left->isCppStructuredBinding(Style); 445 bool StartsObjCMethodExpr = 446 !IsCppStructuredBinding && !InsideInlineASM && !CppArrayTemplates && 447 Style.isCpp() && !IsCpp11AttributeSpecifier && 448 Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) && 449 !CurrentToken->isOneOf(tok::l_brace, tok::r_square) && 450 (!Parent || 451 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren, 452 tok::kw_return, tok::kw_throw) || 453 Parent->isUnaryOperator() || 454 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 455 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) || 456 (getBinOpPrecedence(Parent->Tok.getKind(), true, true) > 457 prec::Unknown)); 458 bool ColonFound = false; 459 460 unsigned BindingIncrease = 1; 461 if (IsCppStructuredBinding) { 462 Left->Type = TT_StructuredBindingLSquare; 463 } else if (Left->is(TT_Unknown)) { 464 if (StartsObjCMethodExpr) { 465 Left->Type = TT_ObjCMethodExpr; 466 } else if (IsCpp11AttributeSpecifier) { 467 Left->Type = TT_AttributeSquare; 468 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent && 469 Contexts.back().ContextKind == tok::l_brace && 470 Parent->isOneOf(tok::l_brace, tok::comma)) { 471 Left->Type = TT_JsComputedPropertyName; 472 } else if (Style.isCpp() && Contexts.back().ContextKind == tok::l_brace && 473 Parent && Parent->isOneOf(tok::l_brace, tok::comma)) { 474 Left->Type = TT_DesignatedInitializerLSquare; 475 } else if (CurrentToken->is(tok::r_square) && Parent && 476 Parent->is(TT_TemplateCloser)) { 477 Left->Type = TT_ArraySubscriptLSquare; 478 } else if (Style.Language == FormatStyle::LK_Proto || 479 Style.Language == FormatStyle::LK_TextProto) { 480 // Square braces in LK_Proto can either be message field attributes: 481 // 482 // optional Aaa aaa = 1 [ 483 // (aaa) = aaa 484 // ]; 485 // 486 // extensions 123 [ 487 // (aaa) = aaa 488 // ]; 489 // 490 // or text proto extensions (in options): 491 // 492 // option (Aaa.options) = { 493 // [type.type/type] { 494 // key: value 495 // } 496 // } 497 // 498 // or repeated fields (in options): 499 // 500 // option (Aaa.options) = { 501 // keys: [ 1, 2, 3 ] 502 // } 503 // 504 // In the first and the third case we want to spread the contents inside 505 // the square braces; in the second we want to keep them inline. 506 Left->Type = TT_ArrayInitializerLSquare; 507 if (!Left->endsSequence(tok::l_square, tok::numeric_constant, 508 tok::equal) && 509 !Left->endsSequence(tok::l_square, tok::numeric_constant, 510 tok::identifier) && 511 !Left->endsSequence(tok::l_square, tok::colon, TT_SelectorName)) { 512 Left->Type = TT_ProtoExtensionLSquare; 513 BindingIncrease = 10; 514 } 515 } else if (!CppArrayTemplates && Parent && 516 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at, 517 tok::comma, tok::l_paren, tok::l_square, 518 tok::question, tok::colon, tok::kw_return, 519 // Should only be relevant to JavaScript: 520 tok::kw_default)) { 521 Left->Type = TT_ArrayInitializerLSquare; 522 } else if (IsCSharp11AttributeSpecifier) { 523 Left->Type = TT_AttributeSquare; 524 } else { 525 BindingIncrease = 10; 526 Left->Type = TT_ArraySubscriptLSquare; 527 } 528 } 529 530 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease); 531 Contexts.back().IsExpression = true; 532 if (Style.Language == FormatStyle::LK_JavaScript && Parent && 533 Parent->is(TT_JsTypeColon)) 534 Contexts.back().IsExpression = false; 535 536 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr; 537 Contexts.back().InCpp11AttributeSpecifier = IsCpp11AttributeSpecifier; 538 Contexts.back().InCSharpAttributeSpecifier = IsCSharp11AttributeSpecifier; 539 540 while (CurrentToken) { 541 if (CurrentToken->is(tok::r_square)) { 542 if (IsCpp11AttributeSpecifier) 543 CurrentToken->Type = TT_AttributeSquare; 544 if (IsCSharp11AttributeSpecifier) 545 CurrentToken->Type = TT_AttributeSquare; 546 else if (((CurrentToken->Next && 547 CurrentToken->Next->is(tok::l_paren)) || 548 (CurrentToken->Previous && 549 CurrentToken->Previous->Previous == Left)) && 550 Left->is(TT_ObjCMethodExpr)) { 551 // An ObjC method call is rarely followed by an open parenthesis. It 552 // also can't be composed of just one token, unless it's a macro that 553 // will be expanded to more tokens. 554 // FIXME: Do we incorrectly label ":" with this? 555 StartsObjCMethodExpr = false; 556 Left->Type = TT_Unknown; 557 } 558 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) { 559 CurrentToken->Type = TT_ObjCMethodExpr; 560 // If we haven't seen a colon yet, make sure the last identifier 561 // before the r_square is tagged as a selector name component. 562 if (!ColonFound && CurrentToken->Previous && 563 CurrentToken->Previous->is(TT_Unknown) && 564 canBeObjCSelectorComponent(*CurrentToken->Previous)) 565 CurrentToken->Previous->Type = TT_SelectorName; 566 // determineStarAmpUsage() thinks that '*' '[' is allocating an 567 // array of pointers, but if '[' starts a selector then '*' is a 568 // binary operator. 569 if (Parent && Parent->is(TT_PointerOrReference)) 570 Parent->Type = TT_BinaryOperator; 571 } 572 // An arrow after an ObjC method expression is not a lambda arrow. 573 if (CurrentToken->Type == TT_ObjCMethodExpr && CurrentToken->Next && 574 CurrentToken->Next->is(TT_LambdaArrow)) 575 CurrentToken->Next->Type = TT_Unknown; 576 Left->MatchingParen = CurrentToken; 577 CurrentToken->MatchingParen = Left; 578 // FirstObjCSelectorName is set when a colon is found. This does 579 // not work, however, when the method has no parameters. 580 // Here, we set FirstObjCSelectorName when the end of the method call is 581 // reached, in case it was not set already. 582 if (!Contexts.back().FirstObjCSelectorName) { 583 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 584 if (Previous && Previous->is(TT_SelectorName)) { 585 Previous->ObjCSelectorNameParts = 1; 586 Contexts.back().FirstObjCSelectorName = Previous; 587 } 588 } else { 589 Left->ParameterCount = 590 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 591 } 592 if (Contexts.back().FirstObjCSelectorName) { 593 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 594 Contexts.back().LongestObjCSelectorName; 595 if (Left->BlockParameterCount > 1) 596 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0; 597 } 598 next(); 599 return true; 600 } 601 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace)) 602 return false; 603 if (CurrentToken->is(tok::colon)) { 604 if (IsCpp11AttributeSpecifier && 605 CurrentToken->endsSequence(tok::colon, tok::identifier, 606 tok::kw_using)) { 607 // Remember that this is a [[using ns: foo]] C++ attribute, so we 608 // don't add a space before the colon (unlike other colons). 609 CurrentToken->Type = TT_AttributeColon; 610 } else if (Left->isOneOf(TT_ArraySubscriptLSquare, 611 TT_DesignatedInitializerLSquare)) { 612 Left->Type = TT_ObjCMethodExpr; 613 StartsObjCMethodExpr = true; 614 Contexts.back().ColonIsObjCMethodExpr = true; 615 if (Parent && Parent->is(tok::r_paren)) 616 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 617 Parent->Type = TT_CastRParen; 618 } 619 ColonFound = true; 620 } 621 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) && 622 !ColonFound) 623 Left->Type = TT_ArrayInitializerLSquare; 624 FormatToken *Tok = CurrentToken; 625 if (!consumeToken()) 626 return false; 627 updateParameterCount(Left, Tok); 628 } 629 return false; 630 } 631 632 bool parseBrace() { 633 if (CurrentToken) { 634 FormatToken *Left = CurrentToken->Previous; 635 Left->ParentBracket = Contexts.back().ContextKind; 636 637 if (Contexts.back().CaretFound) 638 Left->Type = TT_ObjCBlockLBrace; 639 Contexts.back().CaretFound = false; 640 641 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1); 642 Contexts.back().ColonIsDictLiteral = true; 643 if (Left->BlockKind == BK_BracedInit) 644 Contexts.back().IsExpression = true; 645 if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 646 Left->Previous->is(TT_JsTypeColon)) 647 Contexts.back().IsExpression = false; 648 649 while (CurrentToken) { 650 if (CurrentToken->is(tok::r_brace)) { 651 Left->MatchingParen = CurrentToken; 652 CurrentToken->MatchingParen = Left; 653 next(); 654 return true; 655 } 656 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 657 return false; 658 updateParameterCount(Left, CurrentToken); 659 if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) { 660 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 661 if (Previous->is(TT_JsTypeOptionalQuestion)) 662 Previous = Previous->getPreviousNonComment(); 663 if ((CurrentToken->is(tok::colon) && 664 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) || 665 Style.Language == FormatStyle::LK_Proto || 666 Style.Language == FormatStyle::LK_TextProto) { 667 Left->Type = TT_DictLiteral; 668 if (Previous->Tok.getIdentifierInfo() || 669 Previous->is(tok::string_literal)) 670 Previous->Type = TT_SelectorName; 671 } 672 if (CurrentToken->is(tok::colon) || 673 Style.Language == FormatStyle::LK_JavaScript) 674 Left->Type = TT_DictLiteral; 675 } 676 if (CurrentToken->is(tok::comma) && 677 Style.Language == FormatStyle::LK_JavaScript) 678 Left->Type = TT_DictLiteral; 679 if (!consumeToken()) 680 return false; 681 } 682 } 683 return true; 684 } 685 686 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 687 // For ObjC methods, the number of parameters is calculated differently as 688 // method declarations have a different structure (the parameters are not 689 // inside a bracket scope). 690 if (Current->is(tok::l_brace) && Current->BlockKind == BK_Block) 691 ++Left->BlockParameterCount; 692 if (Current->is(tok::comma)) { 693 ++Left->ParameterCount; 694 if (!Left->Role) 695 Left->Role.reset(new CommaSeparatedList(Style)); 696 Left->Role->CommaFound(Current); 697 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 698 Left->ParameterCount = 1; 699 } 700 } 701 702 bool parseConditional() { 703 while (CurrentToken) { 704 if (CurrentToken->is(tok::colon)) { 705 CurrentToken->Type = TT_ConditionalExpr; 706 next(); 707 return true; 708 } 709 if (!consumeToken()) 710 return false; 711 } 712 return false; 713 } 714 715 bool parseTemplateDeclaration() { 716 if (CurrentToken && CurrentToken->is(tok::less)) { 717 CurrentToken->Type = TT_TemplateOpener; 718 next(); 719 if (!parseAngle()) 720 return false; 721 if (CurrentToken) 722 CurrentToken->Previous->ClosesTemplateDeclaration = true; 723 return true; 724 } 725 return false; 726 } 727 728 bool consumeToken() { 729 FormatToken *Tok = CurrentToken; 730 next(); 731 switch (Tok->Tok.getKind()) { 732 case tok::plus: 733 case tok::minus: 734 if (!Tok->Previous && Line.MustBeDeclaration) 735 Tok->Type = TT_ObjCMethodSpecifier; 736 break; 737 case tok::colon: 738 if (!Tok->Previous) 739 return false; 740 // Colons from ?: are handled in parseConditional(). 741 if (Style.Language == FormatStyle::LK_JavaScript) { 742 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop 743 (Contexts.size() == 1 && // switch/case labels 744 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) || 745 Contexts.back().ContextKind == tok::l_paren || // function params 746 Contexts.back().ContextKind == tok::l_square || // array type 747 (!Contexts.back().IsExpression && 748 Contexts.back().ContextKind == tok::l_brace) || // object type 749 (Contexts.size() == 1 && 750 Line.MustBeDeclaration)) { // method/property declaration 751 Contexts.back().IsExpression = false; 752 Tok->Type = TT_JsTypeColon; 753 break; 754 } 755 } 756 if (Contexts.back().ColonIsDictLiteral || 757 Style.Language == FormatStyle::LK_Proto || 758 Style.Language == FormatStyle::LK_TextProto) { 759 Tok->Type = TT_DictLiteral; 760 if (Style.Language == FormatStyle::LK_TextProto) { 761 if (FormatToken *Previous = Tok->getPreviousNonComment()) 762 Previous->Type = TT_SelectorName; 763 } 764 } else if (Contexts.back().ColonIsObjCMethodExpr || 765 Line.startsWith(TT_ObjCMethodSpecifier)) { 766 Tok->Type = TT_ObjCMethodExpr; 767 const FormatToken *BeforePrevious = Tok->Previous->Previous; 768 // Ensure we tag all identifiers in method declarations as 769 // TT_SelectorName. 770 bool UnknownIdentifierInMethodDeclaration = 771 Line.startsWith(TT_ObjCMethodSpecifier) && 772 Tok->Previous->is(tok::identifier) && Tok->Previous->is(TT_Unknown); 773 if (!BeforePrevious || 774 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 775 !(BeforePrevious->is(TT_CastRParen) || 776 (BeforePrevious->is(TT_ObjCMethodExpr) && 777 BeforePrevious->is(tok::colon))) || 778 BeforePrevious->is(tok::r_square) || 779 Contexts.back().LongestObjCSelectorName == 0 || 780 UnknownIdentifierInMethodDeclaration) { 781 Tok->Previous->Type = TT_SelectorName; 782 if (!Contexts.back().FirstObjCSelectorName) 783 Contexts.back().FirstObjCSelectorName = Tok->Previous; 784 else if (Tok->Previous->ColumnWidth > 785 Contexts.back().LongestObjCSelectorName) 786 Contexts.back().LongestObjCSelectorName = 787 Tok->Previous->ColumnWidth; 788 Tok->Previous->ParameterIndex = 789 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 790 ++Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 791 } 792 } else if (Contexts.back().ColonIsForRangeExpr) { 793 Tok->Type = TT_RangeBasedForLoopColon; 794 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 795 Tok->Type = TT_BitFieldColon; 796 } else if (Contexts.size() == 1 && 797 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) { 798 if (Tok->getPreviousNonComment()->isOneOf(tok::r_paren, 799 tok::kw_noexcept)) 800 Tok->Type = TT_CtorInitializerColon; 801 else 802 Tok->Type = TT_InheritanceColon; 803 } else if (canBeObjCSelectorComponent(*Tok->Previous) && Tok->Next && 804 (Tok->Next->isOneOf(tok::r_paren, tok::comma) || 805 (canBeObjCSelectorComponent(*Tok->Next) && Tok->Next->Next && 806 Tok->Next->Next->is(tok::colon)))) { 807 // This handles a special macro in ObjC code where selectors including 808 // the colon are passed as macro arguments. 809 Tok->Type = TT_ObjCMethodExpr; 810 } else if (Contexts.back().ContextKind == tok::l_paren) { 811 Tok->Type = TT_InlineASMColon; 812 } 813 break; 814 case tok::pipe: 815 case tok::amp: 816 // | and & in declarations/type expressions represent union and 817 // intersection types, respectively. 818 if (Style.Language == FormatStyle::LK_JavaScript && 819 !Contexts.back().IsExpression) 820 Tok->Type = TT_JsTypeOperator; 821 break; 822 case tok::kw_if: 823 case tok::kw_while: 824 if (Tok->is(tok::kw_if) && CurrentToken && 825 CurrentToken->is(tok::kw_constexpr)) 826 next(); 827 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 828 next(); 829 if (!parseParens(/*LookForDecls=*/true)) 830 return false; 831 } 832 break; 833 case tok::kw_for: 834 if (Style.Language == FormatStyle::LK_JavaScript) { 835 // x.for and {for: ...} 836 if ((Tok->Previous && Tok->Previous->is(tok::period)) || 837 (Tok->Next && Tok->Next->is(tok::colon))) 838 break; 839 // JS' for await ( ... 840 if (CurrentToken && CurrentToken->is(Keywords.kw_await)) 841 next(); 842 } 843 Contexts.back().ColonIsForRangeExpr = true; 844 next(); 845 if (!parseParens()) 846 return false; 847 break; 848 case tok::l_paren: 849 // When faced with 'operator()()', the kw_operator handler incorrectly 850 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make 851 // the first two parens OverloadedOperators and the second l_paren an 852 // OverloadedOperatorLParen. 853 if (Tok->Previous && Tok->Previous->is(tok::r_paren) && 854 Tok->Previous->MatchingParen && 855 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) { 856 Tok->Previous->Type = TT_OverloadedOperator; 857 Tok->Previous->MatchingParen->Type = TT_OverloadedOperator; 858 Tok->Type = TT_OverloadedOperatorLParen; 859 } 860 861 if (!parseParens()) 862 return false; 863 if (Line.MustBeDeclaration && Contexts.size() == 1 && 864 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) && 865 (!Tok->Previous || 866 !Tok->Previous->isOneOf(tok::kw_decltype, tok::kw___attribute, 867 TT_LeadingJavaAnnotation))) 868 Line.MightBeFunctionDecl = true; 869 break; 870 case tok::l_square: 871 if (!parseSquare()) 872 return false; 873 break; 874 case tok::l_brace: 875 if (Style.Language == FormatStyle::LK_TextProto) { 876 FormatToken *Previous = Tok->getPreviousNonComment(); 877 if (Previous && Previous->Type != TT_DictLiteral) 878 Previous->Type = TT_SelectorName; 879 } 880 if (!parseBrace()) 881 return false; 882 break; 883 case tok::less: 884 if (parseAngle()) { 885 Tok->Type = TT_TemplateOpener; 886 // In TT_Proto, we must distignuish between: 887 // map<key, value> 888 // msg < item: data > 889 // msg: < item: data > 890 // In TT_TextProto, map<key, value> does not occur. 891 if (Style.Language == FormatStyle::LK_TextProto || 892 (Style.Language == FormatStyle::LK_Proto && Tok->Previous && 893 Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) { 894 Tok->Type = TT_DictLiteral; 895 FormatToken *Previous = Tok->getPreviousNonComment(); 896 if (Previous && Previous->Type != TT_DictLiteral) 897 Previous->Type = TT_SelectorName; 898 } 899 } else { 900 Tok->Type = TT_BinaryOperator; 901 NonTemplateLess.insert(Tok); 902 CurrentToken = Tok; 903 next(); 904 } 905 break; 906 case tok::r_paren: 907 case tok::r_square: 908 return false; 909 case tok::r_brace: 910 // Lines can start with '}'. 911 if (Tok->Previous) 912 return false; 913 break; 914 case tok::greater: 915 if (Style.Language != FormatStyle::LK_TextProto) 916 Tok->Type = TT_BinaryOperator; 917 break; 918 case tok::kw_operator: 919 if (Style.Language == FormatStyle::LK_TextProto || 920 Style.Language == FormatStyle::LK_Proto) 921 break; 922 while (CurrentToken && 923 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 924 if (CurrentToken->isOneOf(tok::star, tok::amp)) 925 CurrentToken->Type = TT_PointerOrReference; 926 consumeToken(); 927 if (CurrentToken && 928 CurrentToken->Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator, 929 tok::comma)) 930 CurrentToken->Previous->Type = TT_OverloadedOperator; 931 } 932 if (CurrentToken) { 933 CurrentToken->Type = TT_OverloadedOperatorLParen; 934 if (CurrentToken->Previous->is(TT_BinaryOperator)) 935 CurrentToken->Previous->Type = TT_OverloadedOperator; 936 } 937 break; 938 case tok::question: 939 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next && 940 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren, 941 tok::r_brace)) { 942 // Question marks before semicolons, colons, etc. indicate optional 943 // types (fields, parameters), e.g. 944 // function(x?: string, y?) {...} 945 // class X { y?; } 946 Tok->Type = TT_JsTypeOptionalQuestion; 947 break; 948 } 949 // Declarations cannot be conditional expressions, this can only be part 950 // of a type declaration. 951 if (Line.MustBeDeclaration && !Contexts.back().IsExpression && 952 Style.Language == FormatStyle::LK_JavaScript) 953 break; 954 parseConditional(); 955 break; 956 case tok::kw_template: 957 parseTemplateDeclaration(); 958 break; 959 case tok::comma: 960 if (Contexts.back().InCtorInitializer) 961 Tok->Type = TT_CtorInitializerComma; 962 else if (Contexts.back().InInheritanceList) 963 Tok->Type = TT_InheritanceComma; 964 else if (Contexts.back().FirstStartOfName && 965 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) { 966 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 967 Line.IsMultiVariableDeclStmt = true; 968 } 969 if (Contexts.back().IsForEachMacro) 970 Contexts.back().IsExpression = true; 971 break; 972 case tok::identifier: 973 if (Tok->isOneOf(Keywords.kw___has_include, 974 Keywords.kw___has_include_next)) { 975 parseHasInclude(); 976 } 977 break; 978 default: 979 break; 980 } 981 return true; 982 } 983 984 void parseIncludeDirective() { 985 if (CurrentToken && CurrentToken->is(tok::less)) { 986 next(); 987 while (CurrentToken) { 988 // Mark tokens up to the trailing line comments as implicit string 989 // literals. 990 if (CurrentToken->isNot(tok::comment) && 991 !CurrentToken->TokenText.startswith("//")) 992 CurrentToken->Type = TT_ImplicitStringLiteral; 993 next(); 994 } 995 } 996 } 997 998 void parseWarningOrError() { 999 next(); 1000 // We still want to format the whitespace left of the first token of the 1001 // warning or error. 1002 next(); 1003 while (CurrentToken) { 1004 CurrentToken->Type = TT_ImplicitStringLiteral; 1005 next(); 1006 } 1007 } 1008 1009 void parsePragma() { 1010 next(); // Consume "pragma". 1011 if (CurrentToken && 1012 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 1013 bool IsMark = CurrentToken->is(Keywords.kw_mark); 1014 next(); // Consume "mark". 1015 next(); // Consume first token (so we fix leading whitespace). 1016 while (CurrentToken) { 1017 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 1018 CurrentToken->Type = TT_ImplicitStringLiteral; 1019 next(); 1020 } 1021 } 1022 } 1023 1024 void parseHasInclude() { 1025 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 1026 return; 1027 next(); // '(' 1028 parseIncludeDirective(); 1029 next(); // ')' 1030 } 1031 1032 LineType parsePreprocessorDirective() { 1033 bool IsFirstToken = CurrentToken->IsFirst; 1034 LineType Type = LT_PreprocessorDirective; 1035 next(); 1036 if (!CurrentToken) 1037 return Type; 1038 1039 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) { 1040 // JavaScript files can contain shebang lines of the form: 1041 // #!/usr/bin/env node 1042 // Treat these like C++ #include directives. 1043 while (CurrentToken) { 1044 // Tokens cannot be comments here. 1045 CurrentToken->Type = TT_ImplicitStringLiteral; 1046 next(); 1047 } 1048 return LT_ImportStatement; 1049 } 1050 1051 if (CurrentToken->Tok.is(tok::numeric_constant)) { 1052 CurrentToken->SpacesRequiredBefore = 1; 1053 return Type; 1054 } 1055 // Hashes in the middle of a line can lead to any strange token 1056 // sequence. 1057 if (!CurrentToken->Tok.getIdentifierInfo()) 1058 return Type; 1059 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 1060 case tok::pp_include: 1061 case tok::pp_include_next: 1062 case tok::pp_import: 1063 next(); 1064 parseIncludeDirective(); 1065 Type = LT_ImportStatement; 1066 break; 1067 case tok::pp_error: 1068 case tok::pp_warning: 1069 parseWarningOrError(); 1070 break; 1071 case tok::pp_pragma: 1072 parsePragma(); 1073 break; 1074 case tok::pp_if: 1075 case tok::pp_elif: 1076 Contexts.back().IsExpression = true; 1077 parseLine(); 1078 break; 1079 default: 1080 break; 1081 } 1082 while (CurrentToken) { 1083 FormatToken *Tok = CurrentToken; 1084 next(); 1085 if (Tok->is(tok::l_paren)) 1086 parseParens(); 1087 else if (Tok->isOneOf(Keywords.kw___has_include, 1088 Keywords.kw___has_include_next)) 1089 parseHasInclude(); 1090 } 1091 return Type; 1092 } 1093 1094 public: 1095 LineType parseLine() { 1096 NonTemplateLess.clear(); 1097 if (CurrentToken->is(tok::hash)) 1098 return parsePreprocessorDirective(); 1099 1100 // Directly allow to 'import <string-literal>' to support protocol buffer 1101 // definitions (github.com/google/protobuf) or missing "#" (either way we 1102 // should not break the line). 1103 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 1104 if ((Style.Language == FormatStyle::LK_Java && 1105 CurrentToken->is(Keywords.kw_package)) || 1106 (Info && Info->getPPKeywordID() == tok::pp_import && 1107 CurrentToken->Next && 1108 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 1109 tok::kw_static))) { 1110 next(); 1111 parseIncludeDirective(); 1112 return LT_ImportStatement; 1113 } 1114 1115 // If this line starts and ends in '<' and '>', respectively, it is likely 1116 // part of "#define <a/b.h>". 1117 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 1118 parseIncludeDirective(); 1119 return LT_ImportStatement; 1120 } 1121 1122 // In .proto files, top-level options and package statements are very 1123 // similar to import statements and should not be line-wrapped. 1124 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 1125 CurrentToken->isOneOf(Keywords.kw_option, Keywords.kw_package)) { 1126 next(); 1127 if (CurrentToken && CurrentToken->is(tok::identifier)) { 1128 while (CurrentToken) 1129 next(); 1130 return LT_ImportStatement; 1131 } 1132 } 1133 1134 bool KeywordVirtualFound = false; 1135 bool ImportStatement = false; 1136 1137 // import {...} from '...'; 1138 if (Style.Language == FormatStyle::LK_JavaScript && 1139 CurrentToken->is(Keywords.kw_import)) 1140 ImportStatement = true; 1141 1142 while (CurrentToken) { 1143 if (CurrentToken->is(tok::kw_virtual)) 1144 KeywordVirtualFound = true; 1145 if (Style.Language == FormatStyle::LK_JavaScript) { 1146 // export {...} from '...'; 1147 // An export followed by "from 'some string';" is a re-export from 1148 // another module identified by a URI and is treated as a 1149 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 1150 // Just "export {...};" or "export class ..." should not be treated as 1151 // an import in this sense. 1152 if (Line.First->is(tok::kw_export) && 1153 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 1154 CurrentToken->Next->isStringLiteral()) 1155 ImportStatement = true; 1156 if (isClosureImportStatement(*CurrentToken)) 1157 ImportStatement = true; 1158 } 1159 if (!consumeToken()) 1160 return LT_Invalid; 1161 } 1162 if (KeywordVirtualFound) 1163 return LT_VirtualFunctionDecl; 1164 if (ImportStatement) 1165 return LT_ImportStatement; 1166 1167 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 1168 if (Contexts.back().FirstObjCSelectorName) 1169 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 1170 Contexts.back().LongestObjCSelectorName; 1171 return LT_ObjCMethodDecl; 1172 } 1173 1174 return LT_Other; 1175 } 1176 1177 private: 1178 bool isClosureImportStatement(const FormatToken &Tok) { 1179 // FIXME: Closure-library specific stuff should not be hard-coded but be 1180 // configurable. 1181 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 1182 Tok.Next->Next && 1183 (Tok.Next->Next->TokenText == "module" || 1184 Tok.Next->Next->TokenText == "provide" || 1185 Tok.Next->Next->TokenText == "require" || 1186 Tok.Next->Next->TokenText == "requireType" || 1187 Tok.Next->Next->TokenText == "forwardDeclare") && 1188 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 1189 } 1190 1191 void resetTokenMetadata(FormatToken *Token) { 1192 if (!Token) 1193 return; 1194 1195 // Reset token type in case we have already looked at it and then 1196 // recovered from an error (e.g. failure to find the matching >). 1197 if (!CurrentToken->isOneOf( 1198 TT_LambdaLSquare, TT_LambdaLBrace, TT_ForEachMacro, 1199 TT_TypenameMacro, TT_FunctionLBrace, TT_ImplicitStringLiteral, 1200 TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow, TT_NamespaceMacro, 1201 TT_OverloadedOperator, TT_RegexLiteral, TT_TemplateString, 1202 TT_ObjCStringLiteral)) 1203 CurrentToken->Type = TT_Unknown; 1204 CurrentToken->Role.reset(); 1205 CurrentToken->MatchingParen = nullptr; 1206 CurrentToken->FakeLParens.clear(); 1207 CurrentToken->FakeRParens = 0; 1208 } 1209 1210 void next() { 1211 if (CurrentToken) { 1212 CurrentToken->NestingLevel = Contexts.size() - 1; 1213 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1214 modifyContext(*CurrentToken); 1215 determineTokenType(*CurrentToken); 1216 CurrentToken = CurrentToken->Next; 1217 } 1218 1219 resetTokenMetadata(CurrentToken); 1220 } 1221 1222 /// A struct to hold information valid in a specific context, e.g. 1223 /// a pair of parenthesis. 1224 struct Context { 1225 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1226 bool IsExpression) 1227 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1228 IsExpression(IsExpression) {} 1229 1230 tok::TokenKind ContextKind; 1231 unsigned BindingStrength; 1232 bool IsExpression; 1233 unsigned LongestObjCSelectorName = 0; 1234 bool ColonIsForRangeExpr = false; 1235 bool ColonIsDictLiteral = false; 1236 bool ColonIsObjCMethodExpr = false; 1237 FormatToken *FirstObjCSelectorName = nullptr; 1238 FormatToken *FirstStartOfName = nullptr; 1239 bool CanBeExpression = true; 1240 bool InTemplateArgument = false; 1241 bool InCtorInitializer = false; 1242 bool InInheritanceList = false; 1243 bool CaretFound = false; 1244 bool IsForEachMacro = false; 1245 bool InCpp11AttributeSpecifier = false; 1246 bool InCSharpAttributeSpecifier = false; 1247 }; 1248 1249 /// Puts a new \c Context onto the stack \c Contexts for the lifetime 1250 /// of each instance. 1251 struct ScopedContextCreator { 1252 AnnotatingParser &P; 1253 1254 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1255 unsigned Increase) 1256 : P(P) { 1257 P.Contexts.push_back(Context(ContextKind, 1258 P.Contexts.back().BindingStrength + Increase, 1259 P.Contexts.back().IsExpression)); 1260 } 1261 1262 ~ScopedContextCreator() { P.Contexts.pop_back(); } 1263 }; 1264 1265 void modifyContext(const FormatToken &Current) { 1266 if (Current.getPrecedence() == prec::Assignment && 1267 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 1268 // Type aliases use `type X = ...;` in TypeScript and can be exported 1269 // using `export type ...`. 1270 !(Style.Language == FormatStyle::LK_JavaScript && 1271 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1272 Line.startsWith(tok::kw_export, Keywords.kw_type, 1273 tok::identifier))) && 1274 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 1275 Contexts.back().IsExpression = true; 1276 if (!Line.startsWith(TT_UnaryOperator)) { 1277 for (FormatToken *Previous = Current.Previous; 1278 Previous && Previous->Previous && 1279 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1280 Previous = Previous->Previous) { 1281 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1282 Previous = Previous->MatchingParen; 1283 if (!Previous) 1284 break; 1285 } 1286 if (Previous->opensScope()) 1287 break; 1288 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1289 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1290 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1291 Previous->Type = TT_PointerOrReference; 1292 } 1293 } 1294 } else if (Current.is(tok::lessless) && 1295 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1296 Contexts.back().IsExpression = true; 1297 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1298 Contexts.back().IsExpression = true; 1299 } else if (Current.is(TT_TrailingReturnArrow)) { 1300 Contexts.back().IsExpression = false; 1301 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1302 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1303 } else if (Current.Previous && 1304 Current.Previous->is(TT_CtorInitializerColon)) { 1305 Contexts.back().IsExpression = true; 1306 Contexts.back().InCtorInitializer = true; 1307 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1308 Contexts.back().InInheritanceList = true; 1309 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1310 for (FormatToken *Previous = Current.Previous; 1311 Previous && Previous->isOneOf(tok::star, tok::amp); 1312 Previous = Previous->Previous) 1313 Previous->Type = TT_PointerOrReference; 1314 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 1315 Contexts.back().IsExpression = false; 1316 } else if (Current.is(tok::kw_new)) { 1317 Contexts.back().CanBeExpression = false; 1318 } else if (Current.isOneOf(tok::semi, tok::exclaim)) { 1319 // This should be the condition or increment in a for-loop. 1320 Contexts.back().IsExpression = true; 1321 } 1322 } 1323 1324 void determineTokenType(FormatToken &Current) { 1325 if (!Current.is(TT_Unknown)) 1326 // The token type is already known. 1327 return; 1328 1329 if (Style.Language == FormatStyle::LK_JavaScript) { 1330 if (Current.is(tok::exclaim)) { 1331 if (Current.Previous && 1332 (Current.Previous->isOneOf(tok::identifier, tok::kw_namespace, 1333 tok::r_paren, tok::r_square, 1334 tok::r_brace) || 1335 Current.Previous->Tok.isLiteral())) { 1336 Current.Type = TT_JsNonNullAssertion; 1337 return; 1338 } 1339 if (Current.Next && 1340 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1341 Current.Type = TT_JsNonNullAssertion; 1342 return; 1343 } 1344 } 1345 } 1346 1347 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1348 // function declaration have been found. In this case, 'Current' is a 1349 // trailing token of this declaration and thus cannot be a name. 1350 if (Current.is(Keywords.kw_instanceof)) { 1351 Current.Type = TT_BinaryOperator; 1352 } else if (isStartOfName(Current) && 1353 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1354 Contexts.back().FirstStartOfName = &Current; 1355 Current.Type = TT_StartOfName; 1356 } else if (Current.is(tok::semi)) { 1357 // Reset FirstStartOfName after finding a semicolon so that a for loop 1358 // with multiple increment statements is not confused with a for loop 1359 // having multiple variable declarations. 1360 Contexts.back().FirstStartOfName = nullptr; 1361 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1362 AutoFound = true; 1363 } else if (Current.is(tok::arrow) && 1364 Style.Language == FormatStyle::LK_Java) { 1365 Current.Type = TT_LambdaArrow; 1366 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1367 Current.NestingLevel == 0) { 1368 Current.Type = TT_TrailingReturnArrow; 1369 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1370 Current.Type = determineStarAmpUsage(Current, 1371 Contexts.back().CanBeExpression && 1372 Contexts.back().IsExpression, 1373 Contexts.back().InTemplateArgument); 1374 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1375 Current.Type = determinePlusMinusCaretUsage(Current); 1376 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1377 Contexts.back().CaretFound = true; 1378 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1379 Current.Type = determineIncrementUsage(Current); 1380 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1381 Current.Type = TT_UnaryOperator; 1382 } else if (Current.is(tok::question)) { 1383 if (Style.Language == FormatStyle::LK_JavaScript && 1384 Line.MustBeDeclaration && !Contexts.back().IsExpression) { 1385 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1386 // on the interface, not a ternary expression. 1387 Current.Type = TT_JsTypeOptionalQuestion; 1388 } else { 1389 Current.Type = TT_ConditionalExpr; 1390 } 1391 } else if (Current.isBinaryOperator() && 1392 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1393 (!Current.is(tok::greater) && 1394 Style.Language != FormatStyle::LK_TextProto)) { 1395 Current.Type = TT_BinaryOperator; 1396 } else if (Current.is(tok::comment)) { 1397 if (Current.TokenText.startswith("/*")) { 1398 if (Current.TokenText.endswith("*/")) 1399 Current.Type = TT_BlockComment; 1400 else 1401 // The lexer has for some reason determined a comment here. But we 1402 // cannot really handle it, if it isn't properly terminated. 1403 Current.Tok.setKind(tok::unknown); 1404 } else { 1405 Current.Type = TT_LineComment; 1406 } 1407 } else if (Current.is(tok::r_paren)) { 1408 if (rParenEndsCast(Current)) 1409 Current.Type = TT_CastRParen; 1410 if (Current.MatchingParen && Current.Next && 1411 !Current.Next->isBinaryOperator() && 1412 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1413 tok::comma, tok::period, tok::arrow, 1414 tok::coloncolon)) 1415 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1416 // Make sure this isn't the return type of an Obj-C block declaration 1417 if (AfterParen->Tok.isNot(tok::caret)) { 1418 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1419 if (BeforeParen->is(tok::identifier) && 1420 !BeforeParen->is(TT_TypenameMacro) && 1421 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1422 (!BeforeParen->Previous || 1423 BeforeParen->Previous->ClosesTemplateDeclaration)) 1424 Current.Type = TT_FunctionAnnotationRParen; 1425 } 1426 } 1427 } else if (Current.is(tok::at) && Current.Next && 1428 Style.Language != FormatStyle::LK_JavaScript && 1429 Style.Language != FormatStyle::LK_Java) { 1430 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1431 // marks declarations and properties that need special formatting. 1432 switch (Current.Next->Tok.getObjCKeywordID()) { 1433 case tok::objc_interface: 1434 case tok::objc_implementation: 1435 case tok::objc_protocol: 1436 Current.Type = TT_ObjCDecl; 1437 break; 1438 case tok::objc_property: 1439 Current.Type = TT_ObjCProperty; 1440 break; 1441 default: 1442 break; 1443 } 1444 } else if (Current.is(tok::period)) { 1445 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1446 if (PreviousNoComment && 1447 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1448 Current.Type = TT_DesignatedInitializerPeriod; 1449 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1450 Current.Previous->isOneOf(TT_JavaAnnotation, 1451 TT_LeadingJavaAnnotation)) { 1452 Current.Type = Current.Previous->Type; 1453 } 1454 } else if (canBeObjCSelectorComponent(Current) && 1455 // FIXME(bug 36976): ObjC return types shouldn't use 1456 // TT_CastRParen. 1457 Current.Previous && Current.Previous->is(TT_CastRParen) && 1458 Current.Previous->MatchingParen && 1459 Current.Previous->MatchingParen->Previous && 1460 Current.Previous->MatchingParen->Previous->is( 1461 TT_ObjCMethodSpecifier)) { 1462 // This is the first part of an Objective-C selector name. (If there's no 1463 // colon after this, this is the only place which annotates the identifier 1464 // as a selector.) 1465 Current.Type = TT_SelectorName; 1466 } else if (Current.isOneOf(tok::identifier, tok::kw_const) && 1467 Current.Previous && 1468 !Current.Previous->isOneOf(tok::equal, tok::at) && 1469 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1470 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1471 // function declaration have been found. 1472 Current.Type = TT_TrailingAnnotation; 1473 } else if ((Style.Language == FormatStyle::LK_Java || 1474 Style.Language == FormatStyle::LK_JavaScript) && 1475 Current.Previous) { 1476 if (Current.Previous->is(tok::at) && 1477 Current.isNot(Keywords.kw_interface)) { 1478 const FormatToken &AtToken = *Current.Previous; 1479 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1480 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1481 Current.Type = TT_LeadingJavaAnnotation; 1482 else 1483 Current.Type = TT_JavaAnnotation; 1484 } else if (Current.Previous->is(tok::period) && 1485 Current.Previous->isOneOf(TT_JavaAnnotation, 1486 TT_LeadingJavaAnnotation)) { 1487 Current.Type = Current.Previous->Type; 1488 } 1489 } 1490 } 1491 1492 /// Take a guess at whether \p Tok starts a name of a function or 1493 /// variable declaration. 1494 /// 1495 /// This is a heuristic based on whether \p Tok is an identifier following 1496 /// something that is likely a type. 1497 bool isStartOfName(const FormatToken &Tok) { 1498 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1499 return false; 1500 1501 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1502 Keywords.kw_as)) 1503 return false; 1504 if (Style.Language == FormatStyle::LK_JavaScript && 1505 Tok.Previous->is(Keywords.kw_in)) 1506 return false; 1507 1508 // Skip "const" as it does not have an influence on whether this is a name. 1509 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1510 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1511 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1512 1513 if (!PreviousNotConst) 1514 return false; 1515 1516 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1517 PreviousNotConst->Previous && 1518 PreviousNotConst->Previous->is(tok::hash); 1519 1520 if (PreviousNotConst->is(TT_TemplateCloser)) 1521 return PreviousNotConst && PreviousNotConst->MatchingParen && 1522 PreviousNotConst->MatchingParen->Previous && 1523 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1524 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1525 1526 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen && 1527 PreviousNotConst->MatchingParen->Previous && 1528 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype)) 1529 return true; 1530 1531 return (!IsPPKeyword && 1532 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) || 1533 PreviousNotConst->is(TT_PointerOrReference) || 1534 PreviousNotConst->isSimpleTypeSpecifier(); 1535 } 1536 1537 /// Determine whether ')' is ending a cast. 1538 bool rParenEndsCast(const FormatToken &Tok) { 1539 // C-style casts are only used in C++ and Java. 1540 if (!Style.isCpp() && Style.Language != FormatStyle::LK_Java) 1541 return false; 1542 1543 // Empty parens aren't casts and there are no casts at the end of the line. 1544 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1545 return false; 1546 1547 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1548 if (LeftOfParens) { 1549 // If there is a closing parenthesis left of the current parentheses, 1550 // look past it as these might be chained casts. 1551 if (LeftOfParens->is(tok::r_paren)) { 1552 if (!LeftOfParens->MatchingParen || 1553 !LeftOfParens->MatchingParen->Previous) 1554 return false; 1555 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1556 } 1557 1558 // If there is an identifier (or with a few exceptions a keyword) right 1559 // before the parentheses, this is unlikely to be a cast. 1560 if (LeftOfParens->Tok.getIdentifierInfo() && 1561 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1562 tok::kw_delete)) 1563 return false; 1564 1565 // Certain other tokens right before the parentheses are also signals that 1566 // this cannot be a cast. 1567 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1568 TT_TemplateCloser, tok::ellipsis)) 1569 return false; 1570 } 1571 1572 if (Tok.Next->is(tok::question)) 1573 return false; 1574 1575 // As Java has no function types, a "(" after the ")" likely means that this 1576 // is a cast. 1577 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1578 return true; 1579 1580 // If a (non-string) literal follows, this is likely a cast. 1581 if (Tok.Next->isNot(tok::string_literal) && 1582 (Tok.Next->Tok.isLiteral() || 1583 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1584 return true; 1585 1586 // Heuristically try to determine whether the parentheses contain a type. 1587 bool ParensAreType = 1588 !Tok.Previous || 1589 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) || 1590 Tok.Previous->isSimpleTypeSpecifier(); 1591 bool ParensCouldEndDecl = 1592 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1593 if (ParensAreType && !ParensCouldEndDecl) 1594 return true; 1595 1596 // At this point, we heuristically assume that there are no casts at the 1597 // start of the line. We assume that we have found most cases where there 1598 // are by the logic above, e.g. "(void)x;". 1599 if (!LeftOfParens) 1600 return false; 1601 1602 // Certain token types inside the parentheses mean that this can't be a 1603 // cast. 1604 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 1605 Token = Token->Next) 1606 if (Token->is(TT_BinaryOperator)) 1607 return false; 1608 1609 // If the following token is an identifier or 'this', this is a cast. All 1610 // cases where this can be something else are handled above. 1611 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1612 return true; 1613 1614 if (!Tok.Next->Next) 1615 return false; 1616 1617 // If the next token after the parenthesis is a unary operator, assume 1618 // that this is cast, unless there are unexpected tokens inside the 1619 // parenthesis. 1620 bool NextIsUnary = 1621 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 1622 if (!NextIsUnary || Tok.Next->is(tok::plus) || 1623 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 1624 return false; 1625 // Search for unexpected tokens. 1626 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 1627 Prev = Prev->Previous) { 1628 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 1629 return false; 1630 } 1631 return true; 1632 } 1633 1634 /// Return the type of the given token assuming it is * or &. 1635 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 1636 bool InTemplateArgument) { 1637 if (Style.Language == FormatStyle::LK_JavaScript) 1638 return TT_BinaryOperator; 1639 1640 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1641 if (!PrevToken) 1642 return TT_UnaryOperator; 1643 1644 const FormatToken *NextToken = Tok.getNextNonComment(); 1645 if (!NextToken || 1646 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_const) || 1647 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 1648 return TT_PointerOrReference; 1649 1650 if (PrevToken->is(tok::coloncolon)) 1651 return TT_PointerOrReference; 1652 1653 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 1654 tok::comma, tok::semi, tok::kw_return, tok::colon, 1655 tok::equal, tok::kw_delete, tok::kw_sizeof, 1656 tok::kw_throw) || 1657 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 1658 TT_UnaryOperator, TT_CastRParen)) 1659 return TT_UnaryOperator; 1660 1661 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 1662 return TT_PointerOrReference; 1663 if (NextToken->is(tok::kw_operator) && !IsExpression) 1664 return TT_PointerOrReference; 1665 if (NextToken->isOneOf(tok::comma, tok::semi)) 1666 return TT_PointerOrReference; 1667 1668 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen) { 1669 FormatToken *TokenBeforeMatchingParen = 1670 PrevToken->MatchingParen->getPreviousNonComment(); 1671 if (TokenBeforeMatchingParen && 1672 TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype, 1673 TT_TypenameMacro)) 1674 return TT_PointerOrReference; 1675 } 1676 1677 if (PrevToken->Tok.isLiteral() || 1678 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 1679 tok::kw_false, tok::r_brace) || 1680 NextToken->Tok.isLiteral() || 1681 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 1682 NextToken->isUnaryOperator() || 1683 // If we know we're in a template argument, there are no named 1684 // declarations. Thus, having an identifier on the right-hand side 1685 // indicates a binary operator. 1686 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 1687 return TT_BinaryOperator; 1688 1689 // "&&(" is quite unlikely to be two successive unary "&". 1690 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren)) 1691 return TT_BinaryOperator; 1692 1693 // This catches some cases where evaluation order is used as control flow: 1694 // aaa && aaa->f(); 1695 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 1696 if (NextNextToken && NextNextToken->is(tok::arrow)) 1697 return TT_BinaryOperator; 1698 1699 // It is very unlikely that we are going to find a pointer or reference type 1700 // definition on the RHS of an assignment. 1701 if (IsExpression && !Contexts.back().CaretFound) 1702 return TT_BinaryOperator; 1703 1704 return TT_PointerOrReference; 1705 } 1706 1707 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 1708 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1709 if (!PrevToken) 1710 return TT_UnaryOperator; 1711 1712 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 1713 // This must be a sequence of leading unary operators. 1714 return TT_UnaryOperator; 1715 1716 // Use heuristics to recognize unary operators. 1717 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 1718 tok::question, tok::colon, tok::kw_return, 1719 tok::kw_case, tok::at, tok::l_brace)) 1720 return TT_UnaryOperator; 1721 1722 // There can't be two consecutive binary operators. 1723 if (PrevToken->is(TT_BinaryOperator)) 1724 return TT_UnaryOperator; 1725 1726 // Fall back to marking the token as binary operator. 1727 return TT_BinaryOperator; 1728 } 1729 1730 /// Determine whether ++/-- are pre- or post-increments/-decrements. 1731 TokenType determineIncrementUsage(const FormatToken &Tok) { 1732 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1733 if (!PrevToken || PrevToken->is(TT_CastRParen)) 1734 return TT_UnaryOperator; 1735 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 1736 return TT_TrailingUnaryOperator; 1737 1738 return TT_UnaryOperator; 1739 } 1740 1741 SmallVector<Context, 8> Contexts; 1742 1743 const FormatStyle &Style; 1744 AnnotatedLine &Line; 1745 FormatToken *CurrentToken; 1746 bool AutoFound; 1747 const AdditionalKeywords &Keywords; 1748 1749 // Set of "<" tokens that do not open a template parameter list. If parseAngle 1750 // determines that a specific token can't be a template opener, it will make 1751 // same decision irrespective of the decisions for tokens leading up to it. 1752 // Store this information to prevent this from causing exponential runtime. 1753 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 1754 }; 1755 1756 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 1757 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 1758 1759 /// Parses binary expressions by inserting fake parenthesis based on 1760 /// operator precedence. 1761 class ExpressionParser { 1762 public: 1763 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 1764 AnnotatedLine &Line) 1765 : Style(Style), Keywords(Keywords), Current(Line.First) {} 1766 1767 /// Parse expressions with the given operator precedence. 1768 void parse(int Precedence = 0) { 1769 // Skip 'return' and ObjC selector colons as they are not part of a binary 1770 // expression. 1771 while (Current && (Current->is(tok::kw_return) || 1772 (Current->is(tok::colon) && 1773 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 1774 next(); 1775 1776 if (!Current || Precedence > PrecedenceArrowAndPeriod) 1777 return; 1778 1779 // Conditional expressions need to be parsed separately for proper nesting. 1780 if (Precedence == prec::Conditional) { 1781 parseConditionalExpr(); 1782 return; 1783 } 1784 1785 // Parse unary operators, which all have a higher precedence than binary 1786 // operators. 1787 if (Precedence == PrecedenceUnaryOperator) { 1788 parseUnaryOperator(); 1789 return; 1790 } 1791 1792 FormatToken *Start = Current; 1793 FormatToken *LatestOperator = nullptr; 1794 unsigned OperatorIndex = 0; 1795 1796 while (Current) { 1797 // Consume operators with higher precedence. 1798 parse(Precedence + 1); 1799 1800 int CurrentPrecedence = getCurrentPrecedence(); 1801 1802 if (Current && Current->is(TT_SelectorName) && 1803 Precedence == CurrentPrecedence) { 1804 if (LatestOperator) 1805 addFakeParenthesis(Start, prec::Level(Precedence)); 1806 Start = Current; 1807 } 1808 1809 // At the end of the line or when an operator with higher precedence is 1810 // found, insert fake parenthesis and return. 1811 if (!Current || 1812 (Current->closesScope() && 1813 (Current->MatchingParen || Current->is(TT_TemplateString))) || 1814 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 1815 (CurrentPrecedence == prec::Conditional && 1816 Precedence == prec::Assignment && Current->is(tok::colon))) { 1817 break; 1818 } 1819 1820 // Consume scopes: (), [], <> and {} 1821 if (Current->opensScope()) { 1822 // In fragment of a JavaScript template string can look like '}..${' and 1823 // thus close a scope and open a new one at the same time. 1824 while (Current && (!Current->closesScope() || Current->opensScope())) { 1825 next(); 1826 parse(); 1827 } 1828 next(); 1829 } else { 1830 // Operator found. 1831 if (CurrentPrecedence == Precedence) { 1832 if (LatestOperator) 1833 LatestOperator->NextOperator = Current; 1834 LatestOperator = Current; 1835 Current->OperatorIndex = OperatorIndex; 1836 ++OperatorIndex; 1837 } 1838 next(/*SkipPastLeadingComments=*/Precedence > 0); 1839 } 1840 } 1841 1842 if (LatestOperator && (Current || Precedence > 0)) { 1843 // LatestOperator->LastOperator = true; 1844 if (Precedence == PrecedenceArrowAndPeriod) { 1845 // Call expressions don't have a binary operator precedence. 1846 addFakeParenthesis(Start, prec::Unknown); 1847 } else { 1848 addFakeParenthesis(Start, prec::Level(Precedence)); 1849 } 1850 } 1851 } 1852 1853 private: 1854 /// Gets the precedence (+1) of the given token for binary operators 1855 /// and other tokens that we treat like binary operators. 1856 int getCurrentPrecedence() { 1857 if (Current) { 1858 const FormatToken *NextNonComment = Current->getNextNonComment(); 1859 if (Current->is(TT_ConditionalExpr)) 1860 return prec::Conditional; 1861 if (NextNonComment && Current->is(TT_SelectorName) && 1862 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 1863 ((Style.Language == FormatStyle::LK_Proto || 1864 Style.Language == FormatStyle::LK_TextProto) && 1865 NextNonComment->is(tok::less)))) 1866 return prec::Assignment; 1867 if (Current->is(TT_JsComputedPropertyName)) 1868 return prec::Assignment; 1869 if (Current->is(TT_LambdaArrow)) 1870 return prec::Comma; 1871 if (Current->is(TT_JsFatArrow)) 1872 return prec::Assignment; 1873 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 1874 (Current->is(tok::comment) && NextNonComment && 1875 NextNonComment->is(TT_SelectorName))) 1876 return 0; 1877 if (Current->is(TT_RangeBasedForLoopColon)) 1878 return prec::Comma; 1879 if ((Style.Language == FormatStyle::LK_Java || 1880 Style.Language == FormatStyle::LK_JavaScript) && 1881 Current->is(Keywords.kw_instanceof)) 1882 return prec::Relational; 1883 if (Style.Language == FormatStyle::LK_JavaScript && 1884 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 1885 return prec::Relational; 1886 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 1887 return Current->getPrecedence(); 1888 if (Current->isOneOf(tok::period, tok::arrow)) 1889 return PrecedenceArrowAndPeriod; 1890 if ((Style.Language == FormatStyle::LK_Java || 1891 Style.Language == FormatStyle::LK_JavaScript) && 1892 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 1893 Keywords.kw_throws)) 1894 return 0; 1895 } 1896 return -1; 1897 } 1898 1899 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 1900 Start->FakeLParens.push_back(Precedence); 1901 if (Precedence > prec::Unknown) 1902 Start->StartsBinaryExpression = true; 1903 if (Current) { 1904 FormatToken *Previous = Current->Previous; 1905 while (Previous->is(tok::comment) && Previous->Previous) 1906 Previous = Previous->Previous; 1907 ++Previous->FakeRParens; 1908 if (Precedence > prec::Unknown) 1909 Previous->EndsBinaryExpression = true; 1910 } 1911 } 1912 1913 /// Parse unary operator expressions and surround them with fake 1914 /// parentheses if appropriate. 1915 void parseUnaryOperator() { 1916 llvm::SmallVector<FormatToken *, 2> Tokens; 1917 while (Current && Current->is(TT_UnaryOperator)) { 1918 Tokens.push_back(Current); 1919 next(); 1920 } 1921 parse(PrecedenceArrowAndPeriod); 1922 for (FormatToken *Token : llvm::reverse(Tokens)) 1923 // The actual precedence doesn't matter. 1924 addFakeParenthesis(Token, prec::Unknown); 1925 } 1926 1927 void parseConditionalExpr() { 1928 while (Current && Current->isTrailingComment()) { 1929 next(); 1930 } 1931 FormatToken *Start = Current; 1932 parse(prec::LogicalOr); 1933 if (!Current || !Current->is(tok::question)) 1934 return; 1935 next(); 1936 parse(prec::Assignment); 1937 if (!Current || Current->isNot(TT_ConditionalExpr)) 1938 return; 1939 next(); 1940 parse(prec::Assignment); 1941 addFakeParenthesis(Start, prec::Conditional); 1942 } 1943 1944 void next(bool SkipPastLeadingComments = true) { 1945 if (Current) 1946 Current = Current->Next; 1947 while (Current && 1948 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 1949 Current->isTrailingComment()) 1950 Current = Current->Next; 1951 } 1952 1953 const FormatStyle &Style; 1954 const AdditionalKeywords &Keywords; 1955 FormatToken *Current; 1956 }; 1957 1958 } // end anonymous namespace 1959 1960 void TokenAnnotator::setCommentLineLevels( 1961 SmallVectorImpl<AnnotatedLine *> &Lines) { 1962 const AnnotatedLine *NextNonCommentLine = nullptr; 1963 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(), 1964 E = Lines.rend(); 1965 I != E; ++I) { 1966 bool CommentLine = true; 1967 for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) { 1968 if (!Tok->is(tok::comment)) { 1969 CommentLine = false; 1970 break; 1971 } 1972 } 1973 1974 // If the comment is currently aligned with the line immediately following 1975 // it, that's probably intentional and we should keep it. 1976 if (NextNonCommentLine && CommentLine && 1977 NextNonCommentLine->First->NewlinesBefore <= 1 && 1978 NextNonCommentLine->First->OriginalColumn == 1979 (*I)->First->OriginalColumn) { 1980 // Align comments for preprocessor lines with the # in column 0 if 1981 // preprocessor lines are not indented. Otherwise, align with the next 1982 // line. 1983 (*I)->Level = 1984 (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 1985 (NextNonCommentLine->Type == LT_PreprocessorDirective || 1986 NextNonCommentLine->Type == LT_ImportStatement)) 1987 ? 0 1988 : NextNonCommentLine->Level; 1989 } else { 1990 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr; 1991 } 1992 1993 setCommentLineLevels((*I)->Children); 1994 } 1995 } 1996 1997 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 1998 unsigned Result = 0; 1999 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 2000 Result = std::max(Result, Tok->NestingLevel); 2001 return Result; 2002 } 2003 2004 void TokenAnnotator::annotate(AnnotatedLine &Line) { 2005 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2006 E = Line.Children.end(); 2007 I != E; ++I) { 2008 annotate(**I); 2009 } 2010 AnnotatingParser Parser(Style, Line, Keywords); 2011 Line.Type = Parser.parseLine(); 2012 2013 // With very deep nesting, ExpressionParser uses lots of stack and the 2014 // formatting algorithm is very slow. We're not going to do a good job here 2015 // anyway - it's probably generated code being formatted by mistake. 2016 // Just skip the whole line. 2017 if (maxNestingDepth(Line) > 50) 2018 Line.Type = LT_Invalid; 2019 2020 if (Line.Type == LT_Invalid) 2021 return; 2022 2023 ExpressionParser ExprParser(Style, Keywords, Line); 2024 ExprParser.parse(); 2025 2026 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2027 Line.Type = LT_ObjCMethodDecl; 2028 else if (Line.startsWith(TT_ObjCDecl)) 2029 Line.Type = LT_ObjCDecl; 2030 else if (Line.startsWith(TT_ObjCProperty)) 2031 Line.Type = LT_ObjCProperty; 2032 2033 Line.First->SpacesRequiredBefore = 1; 2034 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 2035 } 2036 2037 // This function heuristically determines whether 'Current' starts the name of a 2038 // function declaration. 2039 static bool isFunctionDeclarationName(const FormatToken &Current, 2040 const AnnotatedLine &Line) { 2041 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 2042 for (; Next; Next = Next->Next) { 2043 if (Next->is(TT_OverloadedOperatorLParen)) 2044 return Next; 2045 if (Next->is(TT_OverloadedOperator)) 2046 continue; 2047 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 2048 // For 'new[]' and 'delete[]'. 2049 if (Next->Next && Next->Next->is(tok::l_square) && Next->Next->Next && 2050 Next->Next->Next->is(tok::r_square)) 2051 Next = Next->Next->Next; 2052 continue; 2053 } 2054 2055 break; 2056 } 2057 return nullptr; 2058 }; 2059 2060 // Find parentheses of parameter list. 2061 const FormatToken *Next = Current.Next; 2062 if (Current.is(tok::kw_operator)) { 2063 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 2064 return false; 2065 Next = skipOperatorName(Next); 2066 } else { 2067 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 2068 return false; 2069 for (; Next; Next = Next->Next) { 2070 if (Next->is(TT_TemplateOpener)) { 2071 Next = Next->MatchingParen; 2072 } else if (Next->is(tok::coloncolon)) { 2073 Next = Next->Next; 2074 if (!Next) 2075 return false; 2076 if (Next->is(tok::kw_operator)) { 2077 Next = skipOperatorName(Next->Next); 2078 break; 2079 } 2080 if (!Next->is(tok::identifier)) 2081 return false; 2082 } else if (Next->is(tok::l_paren)) { 2083 break; 2084 } else { 2085 return false; 2086 } 2087 } 2088 } 2089 2090 // Check whether parameter list can belong to a function declaration. 2091 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 2092 return false; 2093 // If the lines ends with "{", this is likely an function definition. 2094 if (Line.Last->is(tok::l_brace)) 2095 return true; 2096 if (Next->Next == Next->MatchingParen) 2097 return true; // Empty parentheses. 2098 // If there is an &/&& after the r_paren, this is likely a function. 2099 if (Next->MatchingParen->Next && 2100 Next->MatchingParen->Next->is(TT_PointerOrReference)) 2101 return true; 2102 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 2103 Tok = Tok->Next) { 2104 if (Tok->isOneOf(tok::l_paren, TT_TemplateOpener) && Tok->MatchingParen) { 2105 Tok = Tok->MatchingParen; 2106 continue; 2107 } 2108 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 2109 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 2110 return true; 2111 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 2112 Tok->Tok.isLiteral()) 2113 return false; 2114 } 2115 return false; 2116 } 2117 2118 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 2119 assert(Line.MightBeFunctionDecl); 2120 2121 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 2122 Style.AlwaysBreakAfterReturnType == 2123 FormatStyle::RTBS_TopLevelDefinitions) && 2124 Line.Level > 0) 2125 return false; 2126 2127 switch (Style.AlwaysBreakAfterReturnType) { 2128 case FormatStyle::RTBS_None: 2129 return false; 2130 case FormatStyle::RTBS_All: 2131 case FormatStyle::RTBS_TopLevel: 2132 return true; 2133 case FormatStyle::RTBS_AllDefinitions: 2134 case FormatStyle::RTBS_TopLevelDefinitions: 2135 return Line.mightBeFunctionDefinition(); 2136 } 2137 2138 return false; 2139 } 2140 2141 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 2142 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2143 E = Line.Children.end(); 2144 I != E; ++I) { 2145 calculateFormattingInformation(**I); 2146 } 2147 2148 Line.First->TotalLength = 2149 Line.First->IsMultiline ? Style.ColumnLimit 2150 : Line.FirstStartColumn + Line.First->ColumnWidth; 2151 FormatToken *Current = Line.First->Next; 2152 bool InFunctionDecl = Line.MightBeFunctionDecl; 2153 while (Current) { 2154 if (isFunctionDeclarationName(*Current, Line)) 2155 Current->Type = TT_FunctionDeclarationName; 2156 if (Current->is(TT_LineComment)) { 2157 if (Current->Previous->BlockKind == BK_BracedInit && 2158 Current->Previous->opensScope()) 2159 Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1; 2160 else 2161 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2162 2163 // If we find a trailing comment, iterate backwards to determine whether 2164 // it seems to relate to a specific parameter. If so, break before that 2165 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2166 // to the previous line in: 2167 // SomeFunction(a, 2168 // b, // comment 2169 // c); 2170 if (!Current->HasUnescapedNewline) { 2171 for (FormatToken *Parameter = Current->Previous; Parameter; 2172 Parameter = Parameter->Previous) { 2173 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 2174 break; 2175 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 2176 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 2177 Parameter->HasUnescapedNewline) 2178 Parameter->MustBreakBefore = true; 2179 break; 2180 } 2181 } 2182 } 2183 } else if (Current->SpacesRequiredBefore == 0 && 2184 spaceRequiredBefore(Line, *Current)) { 2185 Current->SpacesRequiredBefore = 1; 2186 } 2187 2188 Current->MustBreakBefore = 2189 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 2190 2191 if (!Current->MustBreakBefore && InFunctionDecl && 2192 Current->is(TT_FunctionDeclarationName)) 2193 Current->MustBreakBefore = mustBreakForReturnType(Line); 2194 2195 Current->CanBreakBefore = 2196 Current->MustBreakBefore || canBreakBefore(Line, *Current); 2197 unsigned ChildSize = 0; 2198 if (Current->Previous->Children.size() == 1) { 2199 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last; 2200 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 2201 : LastOfChild.TotalLength + 1; 2202 } 2203 const FormatToken *Prev = Current->Previous; 2204 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 2205 (Prev->Children.size() == 1 && 2206 Prev->Children[0]->First->MustBreakBefore) || 2207 Current->IsMultiline) 2208 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 2209 else 2210 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 2211 ChildSize + Current->SpacesRequiredBefore; 2212 2213 if (Current->is(TT_CtorInitializerColon)) 2214 InFunctionDecl = false; 2215 2216 // FIXME: Only calculate this if CanBreakBefore is true once static 2217 // initializers etc. are sorted out. 2218 // FIXME: Move magic numbers to a better place. 2219 2220 // Reduce penalty for aligning ObjC method arguments using the colon 2221 // alignment as this is the canonical way (still prefer fitting everything 2222 // into one line if possible). Trying to fit a whole expression into one 2223 // line should not force other line breaks (e.g. when ObjC method 2224 // expression is a part of other expression). 2225 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl); 2226 if (Style.Language == FormatStyle::LK_ObjC && 2227 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) { 2228 if (Current->ParameterIndex == 1) 2229 Current->SplitPenalty += 5 * Current->BindingStrength; 2230 } else { 2231 Current->SplitPenalty += 20 * Current->BindingStrength; 2232 } 2233 2234 Current = Current->Next; 2235 } 2236 2237 calculateUnbreakableTailLengths(Line); 2238 unsigned IndentLevel = Line.Level; 2239 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2240 if (Current->Role) 2241 Current->Role->precomputeFormattingInfos(Current); 2242 if (Current->MatchingParen && 2243 Current->MatchingParen->opensBlockOrBlockTypeList(Style)) { 2244 assert(IndentLevel > 0); 2245 --IndentLevel; 2246 } 2247 Current->IndentLevel = IndentLevel; 2248 if (Current->opensBlockOrBlockTypeList(Style)) 2249 ++IndentLevel; 2250 } 2251 2252 LLVM_DEBUG({ printDebugInfo(Line); }); 2253 } 2254 2255 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2256 unsigned UnbreakableTailLength = 0; 2257 FormatToken *Current = Line.Last; 2258 while (Current) { 2259 Current->UnbreakableTailLength = UnbreakableTailLength; 2260 if (Current->CanBreakBefore || 2261 Current->isOneOf(tok::comment, tok::string_literal)) { 2262 UnbreakableTailLength = 0; 2263 } else { 2264 UnbreakableTailLength += 2265 Current->ColumnWidth + Current->SpacesRequiredBefore; 2266 } 2267 Current = Current->Previous; 2268 } 2269 } 2270 2271 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2272 const FormatToken &Tok, 2273 bool InFunctionDecl) { 2274 const FormatToken &Left = *Tok.Previous; 2275 const FormatToken &Right = Tok; 2276 2277 if (Left.is(tok::semi)) 2278 return 0; 2279 2280 if (Style.Language == FormatStyle::LK_Java) { 2281 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2282 return 1; 2283 if (Right.is(Keywords.kw_implements)) 2284 return 2; 2285 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2286 return 3; 2287 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2288 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2289 return 100; 2290 if (Left.is(TT_JsTypeColon)) 2291 return 35; 2292 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2293 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2294 return 100; 2295 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2296 if (Left.opensScope() && Right.closesScope()) 2297 return 200; 2298 } 2299 2300 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2301 return 1; 2302 if (Right.is(tok::l_square)) { 2303 if (Style.Language == FormatStyle::LK_Proto) 2304 return 1; 2305 if (Left.is(tok::r_square)) 2306 return 200; 2307 // Slightly prefer formatting local lambda definitions like functions. 2308 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2309 return 35; 2310 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2311 TT_ArrayInitializerLSquare, 2312 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) 2313 return 500; 2314 } 2315 2316 if (Left.is(tok::coloncolon) || 2317 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2318 return 500; 2319 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2320 Right.is(tok::kw_operator)) { 2321 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2322 return 3; 2323 if (Left.is(TT_StartOfName)) 2324 return 110; 2325 if (InFunctionDecl && Right.NestingLevel == 0) 2326 return Style.PenaltyReturnTypeOnItsOwnLine; 2327 return 200; 2328 } 2329 if (Right.is(TT_PointerOrReference)) 2330 return 190; 2331 if (Right.is(TT_LambdaArrow)) 2332 return 110; 2333 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2334 return 160; 2335 if (Left.is(TT_CastRParen)) 2336 return 100; 2337 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2338 return 5000; 2339 if (Left.is(tok::comment)) 2340 return 1000; 2341 2342 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2343 TT_CtorInitializerColon)) 2344 return 2; 2345 2346 if (Right.isMemberAccess()) { 2347 // Breaking before the "./->" of a chained call/member access is reasonably 2348 // cheap, as formatting those with one call per line is generally 2349 // desirable. In particular, it should be cheaper to break before the call 2350 // than it is to break inside a call's parameters, which could lead to weird 2351 // "hanging" indents. The exception is the very last "./->" to support this 2352 // frequent pattern: 2353 // 2354 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2355 // dddddddd); 2356 // 2357 // which might otherwise be blown up onto many lines. Here, clang-format 2358 // won't produce "hanging" indents anyway as there is no other trailing 2359 // call. 2360 // 2361 // Also apply higher penalty is not a call as that might lead to a wrapping 2362 // like: 2363 // 2364 // aaaaaaa 2365 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2366 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2367 ? 150 2368 : 35; 2369 } 2370 2371 if (Right.is(TT_TrailingAnnotation) && 2372 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2373 // Moving trailing annotations to the next line is fine for ObjC method 2374 // declarations. 2375 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2376 return 10; 2377 // Generally, breaking before a trailing annotation is bad unless it is 2378 // function-like. It seems to be especially preferable to keep standard 2379 // annotations (i.e. "const", "final" and "override") on the same line. 2380 // Use a slightly higher penalty after ")" so that annotations like 2381 // "const override" are kept together. 2382 bool is_short_annotation = Right.TokenText.size() < 10; 2383 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 2384 } 2385 2386 // In for-loops, prefer breaking at ',' and ';'. 2387 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 2388 return 4; 2389 2390 // In Objective-C method expressions, prefer breaking before "param:" over 2391 // breaking after it. 2392 if (Right.is(TT_SelectorName)) 2393 return 0; 2394 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 2395 return Line.MightBeFunctionDecl ? 50 : 500; 2396 2397 // In Objective-C type declarations, avoid breaking after the category's 2398 // open paren (we'll prefer breaking after the protocol list's opening 2399 // angle bracket, if present). 2400 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous && 2401 Left.Previous->isOneOf(tok::identifier, tok::greater)) 2402 return 500; 2403 2404 if (Left.is(tok::l_paren) && InFunctionDecl && 2405 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 2406 return 100; 2407 if (Left.is(tok::l_paren) && Left.Previous && 2408 (Left.Previous->isOneOf(tok::kw_if, tok::kw_for) || 2409 Left.Previous->endsSequence(tok::kw_constexpr, tok::kw_if))) 2410 return 1000; 2411 if (Left.is(tok::equal) && InFunctionDecl) 2412 return 110; 2413 if (Right.is(tok::r_brace)) 2414 return 1; 2415 if (Left.is(TT_TemplateOpener)) 2416 return 100; 2417 if (Left.opensScope()) { 2418 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign) 2419 return 0; 2420 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle) 2421 return 19; 2422 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 2423 : 19; 2424 } 2425 if (Left.is(TT_JavaAnnotation)) 2426 return 50; 2427 2428 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 2429 Left.Previous->isLabelString() && 2430 (Left.NextOperator || Left.OperatorIndex != 0)) 2431 return 50; 2432 if (Right.is(tok::plus) && Left.isLabelString() && 2433 (Right.NextOperator || Right.OperatorIndex != 0)) 2434 return 25; 2435 if (Left.is(tok::comma)) 2436 return 1; 2437 if (Right.is(tok::lessless) && Left.isLabelString() && 2438 (Right.NextOperator || Right.OperatorIndex != 1)) 2439 return 25; 2440 if (Right.is(tok::lessless)) { 2441 // Breaking at a << is really cheap. 2442 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 2443 // Slightly prefer to break before the first one in log-like statements. 2444 return 2; 2445 return 1; 2446 } 2447 if (Left.ClosesTemplateDeclaration) 2448 return Style.PenaltyBreakTemplateDeclaration; 2449 if (Left.is(TT_ConditionalExpr)) 2450 return prec::Conditional; 2451 prec::Level Level = Left.getPrecedence(); 2452 if (Level == prec::Unknown) 2453 Level = Right.getPrecedence(); 2454 if (Level == prec::Assignment) 2455 return Style.PenaltyBreakAssignment; 2456 if (Level != prec::Unknown) 2457 return Level; 2458 2459 return 3; 2460 } 2461 2462 bool TokenAnnotator::spaceRequiredBeforeParens(const FormatToken &Right) const { 2463 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always || 2464 (Style.SpaceBeforeParens == FormatStyle::SBPO_NonEmptyParentheses && 2465 Right.ParameterCount > 0); 2466 } 2467 2468 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 2469 const FormatToken &Left, 2470 const FormatToken &Right) { 2471 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 2472 return true; 2473 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 2474 return true; 2475 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 2476 Left.Tok.getObjCKeywordID() == tok::objc_property) 2477 return true; 2478 if (Right.is(tok::hashhash)) 2479 return Left.is(tok::hash); 2480 if (Left.isOneOf(tok::hashhash, tok::hash)) 2481 return Right.is(tok::hash); 2482 if (Left.is(tok::l_paren) && Right.is(tok::r_paren)) 2483 return Style.SpaceInEmptyParentheses; 2484 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 2485 return (Right.is(TT_CastRParen) || 2486 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 2487 ? Style.SpacesInCStyleCastParentheses 2488 : Style.SpacesInParentheses; 2489 if (Right.isOneOf(tok::semi, tok::comma)) 2490 return false; 2491 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) { 2492 bool IsLightweightGeneric = Right.MatchingParen && 2493 Right.MatchingParen->Next && 2494 Right.MatchingParen->Next->is(tok::colon); 2495 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList; 2496 } 2497 if (Right.is(tok::less) && Left.is(tok::kw_template)) 2498 return Style.SpaceAfterTemplateKeyword; 2499 if (Left.isOneOf(tok::exclaim, tok::tilde)) 2500 return false; 2501 if (Left.is(tok::at) && 2502 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 2503 tok::numeric_constant, tok::l_paren, tok::l_brace, 2504 tok::kw_true, tok::kw_false)) 2505 return false; 2506 if (Left.is(tok::colon)) 2507 return !Left.is(TT_ObjCMethodExpr); 2508 if (Left.is(tok::coloncolon)) 2509 return false; 2510 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 2511 if (Style.Language == FormatStyle::LK_TextProto || 2512 (Style.Language == FormatStyle::LK_Proto && 2513 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 2514 // Format empty list as `<>`. 2515 if (Left.is(tok::less) && Right.is(tok::greater)) 2516 return false; 2517 return !Style.Cpp11BracedListStyle; 2518 } 2519 return false; 2520 } 2521 if (Right.is(tok::ellipsis)) 2522 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 2523 Left.Previous->is(tok::kw_case)); 2524 if (Left.is(tok::l_square) && Right.is(tok::amp)) 2525 return false; 2526 if (Right.is(TT_PointerOrReference)) { 2527 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 2528 if (!Left.MatchingParen) 2529 return true; 2530 FormatToken *TokenBeforeMatchingParen = 2531 Left.MatchingParen->getPreviousNonComment(); 2532 if (!TokenBeforeMatchingParen || 2533 !TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype, 2534 TT_TypenameMacro)) 2535 return true; 2536 } 2537 return (Left.Tok.isLiteral() || 2538 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 2539 (Style.PointerAlignment != FormatStyle::PAS_Left || 2540 (Line.IsMultiVariableDeclStmt && 2541 (Left.NestingLevel == 0 || 2542 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for))))))); 2543 } 2544 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 2545 (!Left.is(TT_PointerOrReference) || 2546 (Style.PointerAlignment != FormatStyle::PAS_Right && 2547 !Line.IsMultiVariableDeclStmt))) 2548 return true; 2549 if (Left.is(TT_PointerOrReference)) 2550 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) || 2551 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 2552 !Right.is(TT_StartOfName)) || 2553 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) || 2554 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 2555 tok::l_paren) && 2556 (Style.PointerAlignment != FormatStyle::PAS_Right && 2557 !Line.IsMultiVariableDeclStmt) && 2558 Left.Previous && 2559 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon)); 2560 if (Right.is(tok::star) && Left.is(tok::l_paren)) 2561 return false; 2562 const auto SpaceRequiredForArrayInitializerLSquare = 2563 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 2564 return Style.SpacesInContainerLiterals || 2565 ((Style.Language == FormatStyle::LK_Proto || 2566 Style.Language == FormatStyle::LK_TextProto) && 2567 !Style.Cpp11BracedListStyle && 2568 LSquareTok.endsSequence(tok::l_square, tok::colon, 2569 TT_SelectorName)); 2570 }; 2571 if (Left.is(tok::l_square)) 2572 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 2573 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 2574 (Left.isOneOf(TT_ArraySubscriptLSquare, 2575 TT_StructuredBindingLSquare) && 2576 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 2577 if (Right.is(tok::r_square)) 2578 return Right.MatchingParen && 2579 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 2580 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 2581 Style)) || 2582 (Style.SpacesInSquareBrackets && 2583 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 2584 TT_StructuredBindingLSquare)) || 2585 Right.MatchingParen->is(TT_AttributeParen)); 2586 if (Right.is(tok::l_square) && 2587 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2588 TT_DesignatedInitializerLSquare, 2589 TT_StructuredBindingLSquare, TT_AttributeSquare) && 2590 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral)) 2591 return false; 2592 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 2593 return !Left.Children.empty(); // No spaces in "{}". 2594 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) || 2595 (Right.is(tok::r_brace) && Right.MatchingParen && 2596 Right.MatchingParen->BlockKind != BK_Block)) 2597 return !Style.Cpp11BracedListStyle; 2598 if (Left.is(TT_BlockComment)) 2599 // No whitespace in x(/*foo=*/1), except for JavaScript. 2600 return Style.Language == FormatStyle::LK_JavaScript || 2601 !Left.TokenText.endswith("=*/"); 2602 if (Right.is(tok::l_paren)) { 2603 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 2604 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) 2605 return true; 2606 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) || 2607 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never && 2608 (Left.isOneOf(tok::kw_if, tok::pp_elif, tok::kw_for, tok::kw_while, 2609 tok::kw_switch, tok::kw_case, TT_ForEachMacro, 2610 TT_ObjCForIn) || 2611 Left.endsSequence(tok::kw_constexpr, tok::kw_if) || 2612 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch, 2613 tok::kw_new, tok::kw_delete) && 2614 (!Left.Previous || Left.Previous->isNot(tok::period))))) || 2615 (spaceRequiredBeforeParens(Right) && 2616 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 2617 Left.is(tok::r_paren) || Left.isSimpleTypeSpecifier() || 2618 (Left.is(tok::r_square) && Left.MatchingParen && 2619 Left.MatchingParen->is(TT_LambdaLSquare))) && 2620 Line.Type != LT_PreprocessorDirective); 2621 } 2622 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 2623 return false; 2624 if (Right.is(TT_UnaryOperator)) 2625 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 2626 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 2627 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 2628 tok::r_paren) || 2629 Left.isSimpleTypeSpecifier()) && 2630 Right.is(tok::l_brace) && Right.getNextNonComment() && 2631 Right.BlockKind != BK_Block) 2632 return false; 2633 if (Left.is(tok::period) || Right.is(tok::period)) 2634 return false; 2635 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L") 2636 return false; 2637 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 2638 Left.MatchingParen->Previous && 2639 (Left.MatchingParen->Previous->is(tok::period) || 2640 Left.MatchingParen->Previous->is(tok::coloncolon))) 2641 // Java call to generic function with explicit type: 2642 // A.<B<C<...>>>DoSomething(); 2643 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference. 2644 return false; 2645 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 2646 return false; 2647 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at)) 2648 // Objective-C dictionary literal -> no space after opening brace. 2649 return false; 2650 if (Right.is(tok::r_brace) && Right.MatchingParen && 2651 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at)) 2652 // Objective-C dictionary literal -> no space before closing brace. 2653 return false; 2654 return true; 2655 } 2656 2657 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 2658 const FormatToken &Right) { 2659 const FormatToken &Left = *Right.Previous; 2660 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 2661 return true; // Never ever merge two identifiers. 2662 if (Style.isCpp()) { 2663 if (Left.is(tok::kw_operator)) 2664 return Right.is(tok::coloncolon); 2665 if (Right.is(tok::l_brace) && Right.BlockKind == BK_BracedInit && 2666 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList) 2667 return true; 2668 } else if (Style.Language == FormatStyle::LK_Proto || 2669 Style.Language == FormatStyle::LK_TextProto) { 2670 if (Right.is(tok::period) && 2671 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 2672 Keywords.kw_repeated, Keywords.kw_extend)) 2673 return true; 2674 if (Right.is(tok::l_paren) && 2675 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 2676 return true; 2677 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 2678 return true; 2679 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 2680 if (Left.is(tok::slash) || Right.is(tok::slash)) 2681 return false; 2682 if (Left.MatchingParen && 2683 Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 2684 Right.isOneOf(tok::l_brace, tok::less)) 2685 return !Style.Cpp11BracedListStyle; 2686 // A percent is probably part of a formatting specification, such as %lld. 2687 if (Left.is(tok::percent)) 2688 return false; 2689 // Preserve the existence of a space before a percent for cases like 0x%04x 2690 // and "%d %d" 2691 if (Left.is(tok::numeric_constant) && Right.is(tok::percent)) 2692 return Right.WhitespaceRange.getEnd() != Right.WhitespaceRange.getBegin(); 2693 } else if (Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp()) { 2694 if (Left.is(TT_JsFatArrow)) 2695 return true; 2696 // for await ( ... 2697 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 2698 Left.Previous->is(tok::kw_for)) 2699 return true; 2700 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 2701 Right.MatchingParen) { 2702 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 2703 // An async arrow function, for example: `x = async () => foo();`, 2704 // as opposed to calling a function called async: `x = async();` 2705 if (Next && Next->is(TT_JsFatArrow)) 2706 return true; 2707 } 2708 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2709 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2710 return false; 2711 // In tagged template literals ("html`bar baz`"), there is no space between 2712 // the tag identifier and the template string. getIdentifierInfo makes sure 2713 // that the identifier is not a pseudo keyword like `yield`, either. 2714 if (Left.is(tok::identifier) && Keywords.IsJavaScriptIdentifier(Left) && 2715 Right.is(TT_TemplateString)) 2716 return false; 2717 if (Right.is(tok::star) && 2718 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 2719 return false; 2720 if (Right.isOneOf(tok::l_brace, tok::l_square) && 2721 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 2722 Keywords.kw_extends, Keywords.kw_implements)) 2723 return true; 2724 if (Right.is(tok::l_paren)) { 2725 // JS methods can use some keywords as names (e.g. `delete()`). 2726 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 2727 return false; 2728 // Valid JS method names can include keywords, e.g. `foo.delete()` or 2729 // `bar.instanceof()`. Recognize call positions by preceding period. 2730 if (Left.Previous && Left.Previous->is(tok::period) && 2731 Left.Tok.getIdentifierInfo()) 2732 return false; 2733 // Additional unary JavaScript operators that need a space after. 2734 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 2735 tok::kw_void)) 2736 return true; 2737 } 2738 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 2739 tok::kw_const) || 2740 // "of" is only a keyword if it appears after another identifier 2741 // (e.g. as "const x of y" in a for loop), or after a destructuring 2742 // operation (const [x, y] of z, const {a, b} of c). 2743 (Left.is(Keywords.kw_of) && Left.Previous && 2744 (Left.Previous->Tok.is(tok::identifier) || 2745 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 2746 (!Left.Previous || !Left.Previous->is(tok::period))) 2747 return true; 2748 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 2749 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 2750 return false; 2751 if (Left.is(Keywords.kw_as) && 2752 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 2753 return true; 2754 if (Left.is(tok::kw_default) && Left.Previous && 2755 Left.Previous->is(tok::kw_export)) 2756 return true; 2757 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 2758 return true; 2759 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 2760 return false; 2761 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 2762 return false; 2763 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 2764 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 2765 return false; 2766 if (Left.is(tok::ellipsis)) 2767 return false; 2768 if (Left.is(TT_TemplateCloser) && 2769 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 2770 Keywords.kw_implements, Keywords.kw_extends)) 2771 // Type assertions ('<type>expr') are not followed by whitespace. Other 2772 // locations that should have whitespace following are identified by the 2773 // above set of follower tokens. 2774 return false; 2775 if (Right.is(TT_JsNonNullAssertion)) 2776 return false; 2777 if (Left.is(TT_JsNonNullAssertion) && 2778 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 2779 return true; // "x! as string", "x! in y" 2780 } else if (Style.Language == FormatStyle::LK_Java) { 2781 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 2782 return true; 2783 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 2784 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never; 2785 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 2786 tok::kw_protected) || 2787 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 2788 Keywords.kw_native)) && 2789 Right.is(TT_TemplateOpener)) 2790 return true; 2791 } 2792 if (Left.is(TT_ImplicitStringLiteral)) 2793 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2794 if (Line.Type == LT_ObjCMethodDecl) { 2795 if (Left.is(TT_ObjCMethodSpecifier)) 2796 return true; 2797 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right)) 2798 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a 2799 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class 2800 // method declaration. 2801 return false; 2802 } 2803 if (Line.Type == LT_ObjCProperty && 2804 (Right.is(tok::equal) || Left.is(tok::equal))) 2805 return false; 2806 2807 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 2808 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 2809 return true; 2810 if (Right.is(TT_OverloadedOperatorLParen)) 2811 return spaceRequiredBeforeParens(Right); 2812 if (Left.is(tok::comma)) 2813 return true; 2814 if (Right.is(tok::comma)) 2815 return false; 2816 if (Right.is(TT_ObjCBlockLParen)) 2817 return true; 2818 if (Right.is(TT_CtorInitializerColon)) 2819 return Style.SpaceBeforeCtorInitializerColon; 2820 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 2821 return false; 2822 if (Right.is(TT_RangeBasedForLoopColon) && 2823 !Style.SpaceBeforeRangeBasedForLoopColon) 2824 return false; 2825 if (Right.is(tok::colon)) { 2826 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) || 2827 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 2828 return false; 2829 if (Right.is(TT_ObjCMethodExpr)) 2830 return false; 2831 if (Left.is(tok::question)) 2832 return false; 2833 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 2834 return false; 2835 if (Right.is(TT_DictLiteral)) 2836 return Style.SpacesInContainerLiterals; 2837 if (Right.is(TT_AttributeColon)) 2838 return false; 2839 return true; 2840 } 2841 if (Left.is(TT_UnaryOperator)) 2842 return (Style.SpaceAfterLogicalNot && Left.is(tok::exclaim)) || 2843 Right.is(TT_BinaryOperator); 2844 2845 // If the next token is a binary operator or a selector name, we have 2846 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 2847 if (Left.is(TT_CastRParen)) 2848 return Style.SpaceAfterCStyleCast || 2849 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 2850 2851 if (Left.is(tok::greater) && Right.is(tok::greater)) { 2852 if (Style.Language == FormatStyle::LK_TextProto || 2853 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 2854 return !Style.Cpp11BracedListStyle; 2855 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 2856 (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles); 2857 } 2858 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 2859 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 2860 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 2861 return false; 2862 if (!Style.SpaceBeforeAssignmentOperators && 2863 Right.getPrecedence() == prec::Assignment) 2864 return false; 2865 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) && 2866 (Left.is(tok::identifier) || Left.is(tok::kw_this))) 2867 return false; 2868 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 2869 // Generally don't remove existing spaces between an identifier and "::". 2870 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 2871 // this turns out to be too lenient, add analysis of the identifier itself. 2872 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2873 if (Right.is(tok::coloncolon) && !Left.isOneOf(tok::l_brace, tok::comment)) 2874 return (Left.is(TT_TemplateOpener) && 2875 Style.Standard == FormatStyle::LS_Cpp03) || 2876 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 2877 tok::kw___super, TT_TemplateCloser, 2878 TT_TemplateOpener)) || 2879 (Left.is(tok ::l_paren) && Style.SpacesInParentheses); 2880 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 2881 return Style.SpacesInAngles; 2882 // Space before TT_StructuredBindingLSquare. 2883 if (Right.is(TT_StructuredBindingLSquare)) 2884 return !Left.isOneOf(tok::amp, tok::ampamp) || 2885 Style.PointerAlignment != FormatStyle::PAS_Right; 2886 // Space before & or && following a TT_StructuredBindingLSquare. 2887 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 2888 Right.isOneOf(tok::amp, tok::ampamp)) 2889 return Style.PointerAlignment != FormatStyle::PAS_Left; 2890 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 2891 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 2892 !Right.is(tok::r_paren))) 2893 return true; 2894 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) && 2895 Right.isNot(TT_FunctionTypeLParen)) 2896 return spaceRequiredBeforeParens(Right); 2897 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 2898 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 2899 return false; 2900 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 2901 Line.startsWith(tok::hash)) 2902 return true; 2903 if (Right.is(TT_TrailingUnaryOperator)) 2904 return false; 2905 if (Left.is(TT_RegexLiteral)) 2906 return false; 2907 return spaceRequiredBetween(Line, Left, Right); 2908 } 2909 2910 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 2911 static bool isAllmanBrace(const FormatToken &Tok) { 2912 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block && 2913 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_LambdaLBrace, TT_DictLiteral); 2914 } 2915 2916 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 2917 const FormatToken &Right) { 2918 const FormatToken &Left = *Right.Previous; 2919 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 2920 return true; 2921 2922 if (Style.Language == FormatStyle::LK_JavaScript) { 2923 // FIXME: This might apply to other languages and token kinds. 2924 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 2925 Left.Previous->is(tok::string_literal)) 2926 return true; 2927 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 2928 Left.Previous && Left.Previous->is(tok::equal) && 2929 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 2930 tok::kw_const) && 2931 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 2932 // above. 2933 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 2934 // Object literals on the top level of a file are treated as "enum-style". 2935 // Each key/value pair is put on a separate line, instead of bin-packing. 2936 return true; 2937 if (Left.is(tok::l_brace) && Line.Level == 0 && 2938 (Line.startsWith(tok::kw_enum) || 2939 Line.startsWith(tok::kw_const, tok::kw_enum) || 2940 Line.startsWith(tok::kw_export, tok::kw_enum) || 2941 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 2942 // JavaScript top-level enum key/value pairs are put on separate lines 2943 // instead of bin-packing. 2944 return true; 2945 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 2946 !Left.Children.empty()) 2947 // Support AllowShortFunctionsOnASingleLine for JavaScript. 2948 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 2949 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 2950 (Left.NestingLevel == 0 && Line.Level == 0 && 2951 Style.AllowShortFunctionsOnASingleLine & 2952 FormatStyle::SFS_InlineOnly); 2953 } else if (Style.Language == FormatStyle::LK_Java) { 2954 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 2955 Right.Next->is(tok::string_literal)) 2956 return true; 2957 } else if (Style.Language == FormatStyle::LK_Cpp || 2958 Style.Language == FormatStyle::LK_ObjC || 2959 Style.Language == FormatStyle::LK_Proto || 2960 Style.Language == FormatStyle::LK_TableGen || 2961 Style.Language == FormatStyle::LK_TextProto) { 2962 if (Left.isStringLiteral() && Right.isStringLiteral()) 2963 return true; 2964 } 2965 2966 // If the last token before a '}', ']', or ')' is a comma or a trailing 2967 // comment, the intention is to insert a line break after it in order to make 2968 // shuffling around entries easier. Import statements, especially in 2969 // JavaScript, can be an exception to this rule. 2970 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 2971 const FormatToken *BeforeClosingBrace = nullptr; 2972 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 2973 (Style.Language == FormatStyle::LK_JavaScript && 2974 Left.is(tok::l_paren))) && 2975 Left.BlockKind != BK_Block && Left.MatchingParen) 2976 BeforeClosingBrace = Left.MatchingParen->Previous; 2977 else if (Right.MatchingParen && 2978 (Right.MatchingParen->isOneOf(tok::l_brace, 2979 TT_ArrayInitializerLSquare) || 2980 (Style.Language == FormatStyle::LK_JavaScript && 2981 Right.MatchingParen->is(tok::l_paren)))) 2982 BeforeClosingBrace = &Left; 2983 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 2984 BeforeClosingBrace->isTrailingComment())) 2985 return true; 2986 } 2987 2988 if (Right.is(tok::comment)) 2989 return Left.BlockKind != BK_BracedInit && 2990 Left.isNot(TT_CtorInitializerColon) && 2991 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 2992 if (Left.isTrailingComment()) 2993 return true; 2994 if (Right.Previous->IsUnterminatedLiteral) 2995 return true; 2996 if (Right.is(tok::lessless) && Right.Next && 2997 Right.Previous->is(tok::string_literal) && 2998 Right.Next->is(tok::string_literal)) 2999 return true; 3000 if (Right.Previous->ClosesTemplateDeclaration && 3001 Right.Previous->MatchingParen && 3002 Right.Previous->MatchingParen->NestingLevel == 0 && 3003 Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes) 3004 return true; 3005 if (Right.is(TT_CtorInitializerComma) && 3006 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3007 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 3008 return true; 3009 if (Right.is(TT_CtorInitializerColon) && 3010 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3011 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 3012 return true; 3013 // Break only if we have multiple inheritance. 3014 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 3015 Right.is(TT_InheritanceComma)) 3016 return true; 3017 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 3018 // Multiline raw string literals are special wrt. line breaks. The author 3019 // has made a deliberate choice and might have aligned the contents of the 3020 // string literal accordingly. Thus, we try keep existing line breaks. 3021 return Right.IsMultiline && Right.NewlinesBefore > 0; 3022 if ((Right.Previous->is(tok::l_brace) || 3023 (Right.Previous->is(tok::less) && Right.Previous->Previous && 3024 Right.Previous->Previous->is(tok::equal))) && 3025 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 3026 // Don't put enums or option definitions onto single lines in protocol 3027 // buffers. 3028 return true; 3029 } 3030 if (Right.is(TT_InlineASMBrace)) 3031 return Right.HasUnescapedNewline; 3032 if (isAllmanBrace(Left) || isAllmanBrace(Right)) 3033 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) || 3034 (Line.startsWith(tok::kw_typedef, tok::kw_enum) && 3035 Style.BraceWrapping.AfterEnum) || 3036 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 3037 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 3038 if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine) 3039 return true; 3040 3041 if (Left.is(TT_LambdaLBrace)) { 3042 if (Left.MatchingParen && Left.MatchingParen->Next && 3043 Left.MatchingParen->Next->isOneOf(tok::comma, tok::r_paren) && 3044 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline) 3045 return false; 3046 3047 if (Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_None || 3048 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline || 3049 (!Left.Children.empty() && 3050 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Empty)) 3051 return true; 3052 } 3053 3054 // Put multiple C# attributes on a new line. 3055 if (Style.isCSharp() && 3056 ((Left.is(TT_AttributeSquare) && Left.is(tok::r_square)) || 3057 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare) && 3058 Right.is(tok::l_square)))) 3059 return true; 3060 3061 // Put multiple Java annotation on a new line. 3062 if ((Style.Language == FormatStyle::LK_Java || 3063 Style.Language == FormatStyle::LK_JavaScript) && 3064 Left.is(TT_LeadingJavaAnnotation) && 3065 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 3066 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 3067 return true; 3068 3069 if (Right.is(TT_ProtoExtensionLSquare)) 3070 return true; 3071 3072 // In text proto instances if a submessage contains at least 2 entries and at 3073 // least one of them is a submessage, like A { ... B { ... } ... }, 3074 // put all of the entries of A on separate lines by forcing the selector of 3075 // the submessage B to be put on a newline. 3076 // 3077 // Example: these can stay on one line: 3078 // a { scalar_1: 1 scalar_2: 2 } 3079 // a { b { key: value } } 3080 // 3081 // and these entries need to be on a new line even if putting them all in one 3082 // line is under the column limit: 3083 // a { 3084 // scalar: 1 3085 // b { key: value } 3086 // } 3087 // 3088 // We enforce this by breaking before a submessage field that has previous 3089 // siblings, *and* breaking before a field that follows a submessage field. 3090 // 3091 // Be careful to exclude the case [proto.ext] { ... } since the `]` is 3092 // the TT_SelectorName there, but we don't want to break inside the brackets. 3093 // 3094 // Another edge case is @submessage { key: value }, which is a common 3095 // substitution placeholder. In this case we want to keep `@` and `submessage` 3096 // together. 3097 // 3098 // We ensure elsewhere that extensions are always on their own line. 3099 if ((Style.Language == FormatStyle::LK_Proto || 3100 Style.Language == FormatStyle::LK_TextProto) && 3101 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) { 3102 // Keep `@submessage` together in: 3103 // @submessage { key: value } 3104 if (Right.Previous && Right.Previous->is(tok::at)) 3105 return false; 3106 // Look for the scope opener after selector in cases like: 3107 // selector { ... 3108 // selector: { ... 3109 // selector: @base { ... 3110 FormatToken *LBrace = Right.Next; 3111 if (LBrace && LBrace->is(tok::colon)) { 3112 LBrace = LBrace->Next; 3113 if (LBrace && LBrace->is(tok::at)) { 3114 LBrace = LBrace->Next; 3115 if (LBrace) 3116 LBrace = LBrace->Next; 3117 } 3118 } 3119 if (LBrace && 3120 // The scope opener is one of {, [, <: 3121 // selector { ... } 3122 // selector [ ... ] 3123 // selector < ... > 3124 // 3125 // In case of selector { ... }, the l_brace is TT_DictLiteral. 3126 // In case of an empty selector {}, the l_brace is not TT_DictLiteral, 3127 // so we check for immediately following r_brace. 3128 ((LBrace->is(tok::l_brace) && 3129 (LBrace->is(TT_DictLiteral) || 3130 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) || 3131 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) { 3132 // If Left.ParameterCount is 0, then this submessage entry is not the 3133 // first in its parent submessage, and we want to break before this entry. 3134 // If Left.ParameterCount is greater than 0, then its parent submessage 3135 // might contain 1 or more entries and we want to break before this entry 3136 // if it contains at least 2 entries. We deal with this case later by 3137 // detecting and breaking before the next entry in the parent submessage. 3138 if (Left.ParameterCount == 0) 3139 return true; 3140 // However, if this submessage is the first entry in its parent 3141 // submessage, Left.ParameterCount might be 1 in some cases. 3142 // We deal with this case later by detecting an entry 3143 // following a closing paren of this submessage. 3144 } 3145 3146 // If this is an entry immediately following a submessage, it will be 3147 // preceded by a closing paren of that submessage, like in: 3148 // left---. .---right 3149 // v v 3150 // sub: { ... } key: value 3151 // If there was a comment between `}` an `key` above, then `key` would be 3152 // put on a new line anyways. 3153 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square)) 3154 return true; 3155 } 3156 3157 // Deal with lambda arguments in C++ - we want consistent line breaks whether 3158 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced 3159 // as aggressive line breaks are placed when the lambda is not the last arg. 3160 if ((Style.Language == FormatStyle::LK_Cpp || 3161 Style.Language == FormatStyle::LK_ObjC) && 3162 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 && 3163 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) { 3164 // Multiple lambdas in the same function call force line breaks. 3165 if (Left.BlockParameterCount > 1) 3166 return true; 3167 3168 // A lambda followed by another arg forces a line break. 3169 if (!Left.Role) 3170 return false; 3171 auto Comma = Left.Role->lastComma(); 3172 if (!Comma) 3173 return false; 3174 auto Next = Comma->getNextNonComment(); 3175 if (!Next) 3176 return false; 3177 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret)) 3178 return true; 3179 } 3180 3181 return false; 3182 } 3183 3184 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 3185 const FormatToken &Right) { 3186 const FormatToken &Left = *Right.Previous; 3187 3188 // Language-specific stuff. 3189 if (Style.Language == FormatStyle::LK_Java) { 3190 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3191 Keywords.kw_implements)) 3192 return false; 3193 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3194 Keywords.kw_implements)) 3195 return true; 3196 } else if (Style.Language == FormatStyle::LK_JavaScript) { 3197 const FormatToken *NonComment = Right.getPreviousNonComment(); 3198 if (NonComment && 3199 NonComment->isOneOf( 3200 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 3201 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 3202 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 3203 Keywords.kw_readonly, Keywords.kw_abstract, Keywords.kw_get, 3204 Keywords.kw_set, Keywords.kw_async, Keywords.kw_await)) 3205 return false; // Otherwise automatic semicolon insertion would trigger. 3206 if (Right.NestingLevel == 0 && 3207 (Left.Tok.getIdentifierInfo() || 3208 Left.isOneOf(tok::r_square, tok::r_paren)) && 3209 Right.isOneOf(tok::l_square, tok::l_paren)) 3210 return false; // Otherwise automatic semicolon insertion would trigger. 3211 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace)) 3212 return false; 3213 if (Left.is(TT_JsTypeColon)) 3214 return true; 3215 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;"). 3216 if (Left.is(tok::exclaim) && Right.is(tok::colon)) 3217 return false; 3218 // Look for is type annotations like: 3219 // function f(): a is B { ... } 3220 // Do not break before is in these cases. 3221 if (Right.is(Keywords.kw_is)) { 3222 const FormatToken *Next = Right.getNextNonComment(); 3223 // If `is` is followed by a colon, it's likely that it's a dict key, so 3224 // ignore it for this check. 3225 // For example this is common in Polymer: 3226 // Polymer({ 3227 // is: 'name', 3228 // ... 3229 // }); 3230 if (!Next || !Next->is(tok::colon)) 3231 return false; 3232 } 3233 if (Left.is(Keywords.kw_in)) 3234 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 3235 if (Right.is(Keywords.kw_in)) 3236 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 3237 if (Right.is(Keywords.kw_as)) 3238 return false; // must not break before as in 'x as type' casts 3239 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) { 3240 // extends and infer can appear as keywords in conditional types: 3241 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types 3242 // do not break before them, as the expressions are subject to ASI. 3243 return false; 3244 } 3245 if (Left.is(Keywords.kw_as)) 3246 return true; 3247 if (Left.is(TT_JsNonNullAssertion)) 3248 return true; 3249 if (Left.is(Keywords.kw_declare) && 3250 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 3251 Keywords.kw_function, tok::kw_class, tok::kw_enum, 3252 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 3253 Keywords.kw_let, tok::kw_const)) 3254 // See grammar for 'declare' statements at: 3255 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10 3256 return false; 3257 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 3258 Right.isOneOf(tok::identifier, tok::string_literal)) 3259 return false; // must not break in "module foo { ...}" 3260 if (Right.is(TT_TemplateString) && Right.closesScope()) 3261 return false; 3262 // Don't split tagged template literal so there is a break between the tag 3263 // identifier and template string. 3264 if (Left.is(tok::identifier) && Right.is(TT_TemplateString)) { 3265 return false; 3266 } 3267 if (Left.is(TT_TemplateString) && Left.opensScope()) 3268 return true; 3269 } 3270 3271 if (Left.is(tok::at)) 3272 return false; 3273 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 3274 return false; 3275 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 3276 return !Right.is(tok::l_paren); 3277 if (Right.is(TT_PointerOrReference)) 3278 return Line.IsMultiVariableDeclStmt || 3279 (Style.PointerAlignment == FormatStyle::PAS_Right && 3280 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 3281 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 3282 Right.is(tok::kw_operator)) 3283 return true; 3284 if (Left.is(TT_PointerOrReference)) 3285 return false; 3286 if (Right.isTrailingComment()) 3287 // We rely on MustBreakBefore being set correctly here as we should not 3288 // change the "binding" behavior of a comment. 3289 // The first comment in a braced lists is always interpreted as belonging to 3290 // the first list element. Otherwise, it should be placed outside of the 3291 // list. 3292 return Left.BlockKind == BK_BracedInit || 3293 (Left.is(TT_CtorInitializerColon) && 3294 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 3295 if (Left.is(tok::question) && Right.is(tok::colon)) 3296 return false; 3297 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 3298 return Style.BreakBeforeTernaryOperators; 3299 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 3300 return !Style.BreakBeforeTernaryOperators; 3301 if (Left.is(TT_InheritanceColon)) 3302 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon; 3303 if (Right.is(TT_InheritanceColon)) 3304 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon; 3305 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 3306 Left.isNot(TT_SelectorName)) 3307 return true; 3308 3309 if (Right.is(tok::colon) && 3310 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 3311 return false; 3312 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 3313 if (Style.Language == FormatStyle::LK_Proto || 3314 Style.Language == FormatStyle::LK_TextProto) { 3315 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 3316 return false; 3317 // Prevent cases like: 3318 // 3319 // submessage: 3320 // { key: valueeeeeeeeeeee } 3321 // 3322 // when the snippet does not fit into one line. 3323 // Prefer: 3324 // 3325 // submessage: { 3326 // key: valueeeeeeeeeeee 3327 // } 3328 // 3329 // instead, even if it is longer by one line. 3330 // 3331 // Note that this allows allows the "{" to go over the column limit 3332 // when the column limit is just between ":" and "{", but that does 3333 // not happen too often and alternative formattings in this case are 3334 // not much better. 3335 // 3336 // The code covers the cases: 3337 // 3338 // submessage: { ... } 3339 // submessage: < ... > 3340 // repeated: [ ... ] 3341 if (((Right.is(tok::l_brace) || Right.is(tok::less)) && 3342 Right.is(TT_DictLiteral)) || 3343 Right.is(TT_ArrayInitializerLSquare)) 3344 return false; 3345 } 3346 return true; 3347 } 3348 if (Right.is(tok::r_square) && Right.MatchingParen && 3349 Right.MatchingParen->is(TT_ProtoExtensionLSquare)) 3350 return false; 3351 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 3352 Right.Next->is(TT_ObjCMethodExpr))) 3353 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 3354 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 3355 return true; 3356 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 3357 return true; 3358 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 3359 TT_OverloadedOperator)) 3360 return false; 3361 if (Left.is(TT_RangeBasedForLoopColon)) 3362 return true; 3363 if (Right.is(TT_RangeBasedForLoopColon)) 3364 return false; 3365 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 3366 return true; 3367 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 3368 Left.is(tok::kw_operator)) 3369 return false; 3370 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 3371 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 3372 return false; 3373 if (Left.is(tok::equal) && Right.is(tok::l_brace) && 3374 !Style.Cpp11BracedListStyle) 3375 return false; 3376 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen)) 3377 return false; 3378 if (Left.is(tok::l_paren) && Left.Previous && 3379 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 3380 return false; 3381 if (Right.is(TT_ImplicitStringLiteral)) 3382 return false; 3383 3384 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser)) 3385 return false; 3386 if (Right.is(tok::r_square) && Right.MatchingParen && 3387 Right.MatchingParen->is(TT_LambdaLSquare)) 3388 return false; 3389 3390 // We only break before r_brace if there was a corresponding break before 3391 // the l_brace, which is tracked by BreakBeforeClosingBrace. 3392 if (Right.is(tok::r_brace)) 3393 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block; 3394 3395 // Allow breaking after a trailing annotation, e.g. after a method 3396 // declaration. 3397 if (Left.is(TT_TrailingAnnotation)) 3398 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 3399 tok::less, tok::coloncolon); 3400 3401 if (Right.is(tok::kw___attribute) || 3402 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) 3403 return true; 3404 3405 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 3406 return true; 3407 3408 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 3409 return true; 3410 3411 if (Left.is(TT_CtorInitializerColon)) 3412 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 3413 if (Right.is(TT_CtorInitializerColon)) 3414 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 3415 if (Left.is(TT_CtorInitializerComma) && 3416 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 3417 return false; 3418 if (Right.is(TT_CtorInitializerComma) && 3419 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 3420 return true; 3421 if (Left.is(TT_InheritanceComma) && 3422 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 3423 return false; 3424 if (Right.is(TT_InheritanceComma) && 3425 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 3426 return true; 3427 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 3428 (Left.is(tok::less) && Right.is(tok::less))) 3429 return false; 3430 if (Right.is(TT_BinaryOperator) && 3431 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 3432 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 3433 Right.getPrecedence() != prec::Assignment)) 3434 return true; 3435 if (Left.is(TT_ArrayInitializerLSquare)) 3436 return true; 3437 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 3438 return true; 3439 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 3440 !Left.isOneOf(tok::arrowstar, tok::lessless) && 3441 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 3442 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 3443 Left.getPrecedence() == prec::Assignment)) 3444 return true; 3445 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 3446 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) 3447 return false; 3448 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 3449 tok::kw_class, tok::kw_struct, tok::comment) || 3450 Right.isMemberAccess() || 3451 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 3452 tok::colon, tok::l_square, tok::at) || 3453 (Left.is(tok::r_paren) && 3454 Right.isOneOf(tok::identifier, tok::kw_const)) || 3455 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 3456 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 3457 } 3458 3459 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 3460 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 3461 const FormatToken *Tok = Line.First; 3462 while (Tok) { 3463 llvm::errs() << " M=" << Tok->MustBreakBefore 3464 << " C=" << Tok->CanBreakBefore 3465 << " T=" << getTokenTypeName(Tok->Type) 3466 << " S=" << Tok->SpacesRequiredBefore 3467 << " B=" << Tok->BlockParameterCount 3468 << " BK=" << Tok->BlockKind << " P=" << Tok->SplitPenalty 3469 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 3470 << " PPK=" << Tok->PackingKind << " FakeLParens="; 3471 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 3472 llvm::errs() << Tok->FakeLParens[i] << "/"; 3473 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 3474 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo(); 3475 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 3476 if (!Tok->Next) 3477 assert(Tok == Line.Last); 3478 Tok = Tok->Next; 3479 } 3480 llvm::errs() << "----\n"; 3481 } 3482 3483 } // namespace format 3484 } // namespace clang 3485