1 //===--- TokenAnnotator.cpp - Format C++ code -----------------------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 /// 10 /// \file 11 /// This file implements a token annotator, i.e. creates 12 /// \c AnnotatedTokens out of \c FormatTokens with required extra information. 13 /// 14 //===----------------------------------------------------------------------===// 15 16 #include "TokenAnnotator.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; 302 303 if (!HasMultipleLines) 304 Left->PackingKind = PPK_Inconclusive; 305 else if (HasMultipleParametersOnALine) 306 Left->PackingKind = PPK_BinPacked; 307 else 308 Left->PackingKind = PPK_OnePerLine; 309 310 next(); 311 return true; 312 } 313 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace)) 314 return false; 315 316 if (CurrentToken->is(tok::l_brace)) 317 Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen 318 if (CurrentToken->is(tok::comma) && CurrentToken->Next && 319 !CurrentToken->Next->HasUnescapedNewline && 320 !CurrentToken->Next->isTrailingComment()) 321 HasMultipleParametersOnALine = true; 322 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) || 323 CurrentToken->Previous->isSimpleTypeSpecifier()) && 324 !CurrentToken->is(tok::l_brace)) 325 Contexts.back().IsExpression = false; 326 if (CurrentToken->isOneOf(tok::semi, tok::colon)) { 327 MightBeObjCForRangeLoop = false; 328 if (PossibleObjCForInToken) { 329 PossibleObjCForInToken->Type = TT_Unknown; 330 PossibleObjCForInToken = nullptr; 331 } 332 } 333 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) { 334 PossibleObjCForInToken = CurrentToken; 335 PossibleObjCForInToken->Type = TT_ObjCForIn; 336 } 337 // When we discover a 'new', we set CanBeExpression to 'false' in order to 338 // parse the type correctly. Reset that after a comma. 339 if (CurrentToken->is(tok::comma)) 340 Contexts.back().CanBeExpression = true; 341 342 FormatToken *Tok = CurrentToken; 343 if (!consumeToken()) 344 return false; 345 updateParameterCount(Left, Tok); 346 if (CurrentToken && CurrentToken->HasUnescapedNewline) 347 HasMultipleLines = true; 348 } 349 return false; 350 } 351 352 bool isCpp11AttributeSpecifier(const FormatToken &Tok) { 353 if (!Style.isCpp() || !Tok.startsSequence(tok::l_square, tok::l_square)) 354 return false; 355 const FormatToken *AttrTok = Tok.Next->Next; 356 if (!AttrTok) 357 return false; 358 // C++17 '[[using ns: foo, bar(baz, blech)]]' 359 // We assume nobody will name an ObjC variable 'using'. 360 if (AttrTok->startsSequence(tok::kw_using, tok::identifier, tok::colon)) 361 return true; 362 if (AttrTok->isNot(tok::identifier)) 363 return false; 364 while (AttrTok && !AttrTok->startsSequence(tok::r_square, tok::r_square)) { 365 // ObjC message send. We assume nobody will use : in a C++11 attribute 366 // specifier parameter, although this is technically valid: 367 // [[foo(:)]] 368 if (AttrTok->is(tok::colon) || 369 AttrTok->startsSequence(tok::identifier, tok::identifier) || 370 AttrTok->startsSequence(tok::r_paren, tok::identifier)) 371 return false; 372 if (AttrTok->is(tok::ellipsis)) 373 return true; 374 AttrTok = AttrTok->Next; 375 } 376 return AttrTok && AttrTok->startsSequence(tok::r_square, tok::r_square); 377 } 378 379 bool parseSquare() { 380 if (!CurrentToken) 381 return false; 382 383 // A '[' could be an index subscript (after an identifier or after 384 // ')' or ']'), it could be the start of an Objective-C method 385 // expression, it could the start of an Objective-C array literal, 386 // or it could be a C++ attribute specifier [[foo::bar]]. 387 FormatToken *Left = CurrentToken->Previous; 388 Left->ParentBracket = Contexts.back().ContextKind; 389 FormatToken *Parent = Left->getPreviousNonComment(); 390 391 // Cases where '>' is followed by '['. 392 // In C++, this can happen either in array of templates (foo<int>[10]) 393 // or when array is a nested template type (unique_ptr<type1<type2>[]>). 394 bool CppArrayTemplates = 395 Style.isCpp() && Parent && Parent->is(TT_TemplateCloser) && 396 (Contexts.back().CanBeExpression || Contexts.back().IsExpression || 397 Contexts.back().InTemplateArgument); 398 399 bool IsCpp11AttributeSpecifier = isCpp11AttributeSpecifier(*Left) || 400 Contexts.back().InCpp11AttributeSpecifier; 401 402 bool StartsObjCMethodExpr = 403 !CppArrayTemplates && Style.isCpp() && !IsCpp11AttributeSpecifier && 404 Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) && 405 !CurrentToken->isOneOf(tok::l_brace, tok::r_square) && 406 (!Parent || 407 (Parent->is(tok::colon) && Parent->isNot(TT_InlineASMColon)) || 408 Parent->isOneOf(tok::l_square, tok::l_paren, tok::kw_return, 409 tok::kw_throw) || 410 Parent->isUnaryOperator() || 411 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 412 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) || 413 getBinOpPrecedence(Parent->Tok.getKind(), true, true) > prec::Unknown); 414 bool ColonFound = false; 415 416 unsigned BindingIncrease = 1; 417 if (Left->isCppStructuredBinding(Style)) { 418 Left->Type = TT_StructuredBindingLSquare; 419 } else if (Left->is(TT_Unknown)) { 420 if (StartsObjCMethodExpr) { 421 Left->Type = TT_ObjCMethodExpr; 422 } else if (IsCpp11AttributeSpecifier) { 423 Left->Type = TT_AttributeSquare; 424 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent && 425 Contexts.back().ContextKind == tok::l_brace && 426 Parent->isOneOf(tok::l_brace, tok::comma)) { 427 Left->Type = TT_JsComputedPropertyName; 428 } else if (Style.isCpp() && Contexts.back().ContextKind == tok::l_brace && 429 Parent && Parent->isOneOf(tok::l_brace, tok::comma)) { 430 Left->Type = TT_DesignatedInitializerLSquare; 431 } else if (CurrentToken->is(tok::r_square) && Parent && 432 Parent->is(TT_TemplateCloser)) { 433 Left->Type = TT_ArraySubscriptLSquare; 434 } else if (Style.Language == FormatStyle::LK_Proto || 435 Style.Language == FormatStyle::LK_TextProto) { 436 // Square braces in LK_Proto can either be message field attributes: 437 // 438 // optional Aaa aaa = 1 [ 439 // (aaa) = aaa 440 // ]; 441 // 442 // extensions 123 [ 443 // (aaa) = aaa 444 // ]; 445 // 446 // or text proto extensions (in options): 447 // 448 // option (Aaa.options) = { 449 // [type.type/type] { 450 // key: value 451 // } 452 // } 453 // 454 // or repeated fields (in options): 455 // 456 // option (Aaa.options) = { 457 // keys: [ 1, 2, 3 ] 458 // } 459 // 460 // In the first and the third case we want to spread the contents inside 461 // the square braces; in the second we want to keep them inline. 462 Left->Type = TT_ArrayInitializerLSquare; 463 if (!Left->endsSequence(tok::l_square, tok::numeric_constant, 464 tok::equal) && 465 !Left->endsSequence(tok::l_square, tok::numeric_constant, 466 tok::identifier) && 467 !Left->endsSequence(tok::l_square, tok::colon, TT_SelectorName)) { 468 Left->Type = TT_ProtoExtensionLSquare; 469 BindingIncrease = 10; 470 } 471 } else if (!CppArrayTemplates && Parent && 472 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at, 473 tok::comma, tok::l_paren, tok::l_square, 474 tok::question, tok::colon, tok::kw_return, 475 // Should only be relevant to JavaScript: 476 tok::kw_default)) { 477 Left->Type = TT_ArrayInitializerLSquare; 478 } else { 479 BindingIncrease = 10; 480 Left->Type = TT_ArraySubscriptLSquare; 481 } 482 } 483 484 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease); 485 Contexts.back().IsExpression = true; 486 if (Style.Language == FormatStyle::LK_JavaScript && Parent && 487 Parent->is(TT_JsTypeColon)) 488 Contexts.back().IsExpression = false; 489 490 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr; 491 Contexts.back().InCpp11AttributeSpecifier = IsCpp11AttributeSpecifier; 492 493 while (CurrentToken) { 494 if (CurrentToken->is(tok::r_square)) { 495 if (IsCpp11AttributeSpecifier) 496 CurrentToken->Type = TT_AttributeSquare; 497 else if (CurrentToken->Next && CurrentToken->Next->is(tok::l_paren) && 498 Left->is(TT_ObjCMethodExpr)) { 499 // An ObjC method call is rarely followed by an open parenthesis. 500 // FIXME: Do we incorrectly label ":" with this? 501 StartsObjCMethodExpr = false; 502 Left->Type = TT_Unknown; 503 } 504 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) { 505 CurrentToken->Type = TT_ObjCMethodExpr; 506 // If we haven't seen a colon yet, make sure the last identifier 507 // before the r_square is tagged as a selector name component. 508 if (!ColonFound && CurrentToken->Previous && 509 CurrentToken->Previous->is(TT_Unknown) && 510 canBeObjCSelectorComponent(*CurrentToken->Previous)) 511 CurrentToken->Previous->Type = TT_SelectorName; 512 // determineStarAmpUsage() thinks that '*' '[' is allocating an 513 // array of pointers, but if '[' starts a selector then '*' is a 514 // binary operator. 515 if (Parent && Parent->is(TT_PointerOrReference)) 516 Parent->Type = TT_BinaryOperator; 517 } 518 Left->MatchingParen = CurrentToken; 519 CurrentToken->MatchingParen = Left; 520 // FirstObjCSelectorName is set when a colon is found. This does 521 // not work, however, when the method has no parameters. 522 // Here, we set FirstObjCSelectorName when the end of the method call is 523 // reached, in case it was not set already. 524 if (!Contexts.back().FirstObjCSelectorName) { 525 FormatToken* Previous = CurrentToken->getPreviousNonComment(); 526 if (Previous && Previous->is(TT_SelectorName)) { 527 Previous->ObjCSelectorNameParts = 1; 528 Contexts.back().FirstObjCSelectorName = Previous; 529 } 530 } else { 531 Left->ParameterCount = 532 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 533 } 534 if (Contexts.back().FirstObjCSelectorName) { 535 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 536 Contexts.back().LongestObjCSelectorName; 537 if (Left->BlockParameterCount > 1) 538 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0; 539 } 540 next(); 541 return true; 542 } 543 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace)) 544 return false; 545 if (CurrentToken->is(tok::colon)) { 546 if (IsCpp11AttributeSpecifier && 547 CurrentToken->endsSequence(tok::colon, tok::identifier, 548 tok::kw_using)) { 549 // Remember that this is a [[using ns: foo]] C++ attribute, so we 550 // don't add a space before the colon (unlike other colons). 551 CurrentToken->Type = TT_AttributeColon; 552 } else if (Left->isOneOf(TT_ArraySubscriptLSquare, 553 TT_DesignatedInitializerLSquare)) { 554 Left->Type = TT_ObjCMethodExpr; 555 StartsObjCMethodExpr = true; 556 Contexts.back().ColonIsObjCMethodExpr = true; 557 if (Parent && Parent->is(tok::r_paren)) 558 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 559 Parent->Type = TT_CastRParen; 560 } 561 ColonFound = true; 562 } 563 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) && 564 !ColonFound) 565 Left->Type = TT_ArrayInitializerLSquare; 566 FormatToken *Tok = CurrentToken; 567 if (!consumeToken()) 568 return false; 569 updateParameterCount(Left, Tok); 570 } 571 return false; 572 } 573 574 bool parseBrace() { 575 if (CurrentToken) { 576 FormatToken *Left = CurrentToken->Previous; 577 Left->ParentBracket = Contexts.back().ContextKind; 578 579 if (Contexts.back().CaretFound) 580 Left->Type = TT_ObjCBlockLBrace; 581 Contexts.back().CaretFound = false; 582 583 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1); 584 Contexts.back().ColonIsDictLiteral = true; 585 if (Left->BlockKind == BK_BracedInit) 586 Contexts.back().IsExpression = true; 587 if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 588 Left->Previous->is(TT_JsTypeColon)) 589 Contexts.back().IsExpression = false; 590 591 while (CurrentToken) { 592 if (CurrentToken->is(tok::r_brace)) { 593 Left->MatchingParen = CurrentToken; 594 CurrentToken->MatchingParen = Left; 595 next(); 596 return true; 597 } 598 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 599 return false; 600 updateParameterCount(Left, CurrentToken); 601 if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) { 602 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 603 if (Previous->is(TT_JsTypeOptionalQuestion)) 604 Previous = Previous->getPreviousNonComment(); 605 if ((CurrentToken->is(tok::colon) && 606 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) || 607 Style.Language == FormatStyle::LK_Proto || 608 Style.Language == FormatStyle::LK_TextProto) { 609 Left->Type = TT_DictLiteral; 610 if (Previous->Tok.getIdentifierInfo() || 611 Previous->is(tok::string_literal)) 612 Previous->Type = TT_SelectorName; 613 } 614 if (CurrentToken->is(tok::colon) || 615 Style.Language == FormatStyle::LK_JavaScript) 616 Left->Type = TT_DictLiteral; 617 } 618 if (CurrentToken->is(tok::comma) && 619 Style.Language == FormatStyle::LK_JavaScript) 620 Left->Type = TT_DictLiteral; 621 if (!consumeToken()) 622 return false; 623 } 624 } 625 return true; 626 } 627 628 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 629 // For ObjC methods, the number of parameters is calculated differently as 630 // method declarations have a different structure (the parameters are not 631 // inside a bracket scope). 632 if (Current->is(tok::l_brace) && Current->BlockKind == BK_Block) 633 ++Left->BlockParameterCount; 634 if (Current->is(tok::comma)) { 635 ++Left->ParameterCount; 636 if (!Left->Role) 637 Left->Role.reset(new CommaSeparatedList(Style)); 638 Left->Role->CommaFound(Current); 639 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 640 Left->ParameterCount = 1; 641 } 642 } 643 644 bool parseConditional() { 645 while (CurrentToken) { 646 if (CurrentToken->is(tok::colon)) { 647 CurrentToken->Type = TT_ConditionalExpr; 648 next(); 649 return true; 650 } 651 if (!consumeToken()) 652 return false; 653 } 654 return false; 655 } 656 657 bool parseTemplateDeclaration() { 658 if (CurrentToken && CurrentToken->is(tok::less)) { 659 CurrentToken->Type = TT_TemplateOpener; 660 next(); 661 if (!parseAngle()) 662 return false; 663 if (CurrentToken) 664 CurrentToken->Previous->ClosesTemplateDeclaration = true; 665 return true; 666 } 667 return false; 668 } 669 670 bool consumeToken() { 671 FormatToken *Tok = CurrentToken; 672 next(); 673 switch (Tok->Tok.getKind()) { 674 case tok::plus: 675 case tok::minus: 676 if (!Tok->Previous && Line.MustBeDeclaration) 677 Tok->Type = TT_ObjCMethodSpecifier; 678 break; 679 case tok::colon: 680 if (!Tok->Previous) 681 return false; 682 // Colons from ?: are handled in parseConditional(). 683 if (Style.Language == FormatStyle::LK_JavaScript) { 684 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop 685 (Contexts.size() == 1 && // switch/case labels 686 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) || 687 Contexts.back().ContextKind == tok::l_paren || // function params 688 Contexts.back().ContextKind == tok::l_square || // array type 689 (!Contexts.back().IsExpression && 690 Contexts.back().ContextKind == tok::l_brace) || // object type 691 (Contexts.size() == 1 && 692 Line.MustBeDeclaration)) { // method/property declaration 693 Contexts.back().IsExpression = false; 694 Tok->Type = TT_JsTypeColon; 695 break; 696 } 697 } 698 if (Contexts.back().ColonIsDictLiteral || 699 Style.Language == FormatStyle::LK_Proto || 700 Style.Language == FormatStyle::LK_TextProto) { 701 Tok->Type = TT_DictLiteral; 702 if (Style.Language == FormatStyle::LK_TextProto) { 703 if (FormatToken *Previous = Tok->getPreviousNonComment()) 704 Previous->Type = TT_SelectorName; 705 } 706 } else if (Contexts.back().ColonIsObjCMethodExpr || 707 Line.startsWith(TT_ObjCMethodSpecifier)) { 708 Tok->Type = TT_ObjCMethodExpr; 709 const FormatToken *BeforePrevious = Tok->Previous->Previous; 710 // Ensure we tag all identifiers in method declarations as 711 // TT_SelectorName. 712 bool UnknownIdentifierInMethodDeclaration = 713 Line.startsWith(TT_ObjCMethodSpecifier) && 714 Tok->Previous->is(tok::identifier) && Tok->Previous->is(TT_Unknown); 715 if (!BeforePrevious || 716 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 717 !(BeforePrevious->is(TT_CastRParen) || 718 (BeforePrevious->is(TT_ObjCMethodExpr) && 719 BeforePrevious->is(tok::colon))) || 720 BeforePrevious->is(tok::r_square) || 721 Contexts.back().LongestObjCSelectorName == 0 || 722 UnknownIdentifierInMethodDeclaration) { 723 Tok->Previous->Type = TT_SelectorName; 724 if (!Contexts.back().FirstObjCSelectorName) 725 Contexts.back().FirstObjCSelectorName = Tok->Previous; 726 else if (Tok->Previous->ColumnWidth > 727 Contexts.back().LongestObjCSelectorName) 728 Contexts.back().LongestObjCSelectorName = 729 Tok->Previous->ColumnWidth; 730 Tok->Previous->ParameterIndex = 731 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 732 ++Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 733 } 734 } else if (Contexts.back().ColonIsForRangeExpr) { 735 Tok->Type = TT_RangeBasedForLoopColon; 736 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 737 Tok->Type = TT_BitFieldColon; 738 } else if (Contexts.size() == 1 && 739 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) { 740 if (Tok->getPreviousNonComment()->isOneOf(tok::r_paren, 741 tok::kw_noexcept)) 742 Tok->Type = TT_CtorInitializerColon; 743 else 744 Tok->Type = TT_InheritanceColon; 745 } else if (canBeObjCSelectorComponent(*Tok->Previous) && Tok->Next && 746 (Tok->Next->isOneOf(tok::r_paren, tok::comma) || 747 (canBeObjCSelectorComponent(*Tok->Next) && Tok->Next->Next && 748 Tok->Next->Next->is(tok::colon)))) { 749 // This handles a special macro in ObjC code where selectors including 750 // the colon are passed as macro arguments. 751 Tok->Type = TT_ObjCMethodExpr; 752 } else if (Contexts.back().ContextKind == tok::l_paren) { 753 Tok->Type = TT_InlineASMColon; 754 } 755 break; 756 case tok::pipe: 757 case tok::amp: 758 // | and & in declarations/type expressions represent union and 759 // intersection types, respectively. 760 if (Style.Language == FormatStyle::LK_JavaScript && 761 !Contexts.back().IsExpression) 762 Tok->Type = TT_JsTypeOperator; 763 break; 764 case tok::kw_if: 765 case tok::kw_while: 766 if (Tok->is(tok::kw_if) && CurrentToken && 767 CurrentToken->is(tok::kw_constexpr)) 768 next(); 769 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 770 next(); 771 if (!parseParens(/*LookForDecls=*/true)) 772 return false; 773 } 774 break; 775 case tok::kw_for: 776 if (Style.Language == FormatStyle::LK_JavaScript) { 777 // x.for and {for: ...} 778 if ((Tok->Previous && Tok->Previous->is(tok::period)) || 779 (Tok->Next && Tok->Next->is(tok::colon))) 780 break; 781 // JS' for await ( ... 782 if (CurrentToken && CurrentToken->is(Keywords.kw_await)) 783 next(); 784 } 785 Contexts.back().ColonIsForRangeExpr = true; 786 next(); 787 if (!parseParens()) 788 return false; 789 break; 790 case tok::l_paren: 791 // When faced with 'operator()()', the kw_operator handler incorrectly 792 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make 793 // the first two parens OverloadedOperators and the second l_paren an 794 // OverloadedOperatorLParen. 795 if (Tok->Previous && Tok->Previous->is(tok::r_paren) && 796 Tok->Previous->MatchingParen && 797 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) { 798 Tok->Previous->Type = TT_OverloadedOperator; 799 Tok->Previous->MatchingParen->Type = TT_OverloadedOperator; 800 Tok->Type = TT_OverloadedOperatorLParen; 801 } 802 803 if (!parseParens()) 804 return false; 805 if (Line.MustBeDeclaration && Contexts.size() == 1 && 806 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) && 807 (!Tok->Previous || 808 !Tok->Previous->isOneOf(tok::kw_decltype, tok::kw___attribute, 809 TT_LeadingJavaAnnotation))) 810 Line.MightBeFunctionDecl = true; 811 break; 812 case tok::l_square: 813 if (!parseSquare()) 814 return false; 815 break; 816 case tok::l_brace: 817 if (Style.Language == FormatStyle::LK_TextProto) { 818 FormatToken *Previous = Tok->getPreviousNonComment(); 819 if (Previous && Previous->Type != TT_DictLiteral) 820 Previous->Type = TT_SelectorName; 821 } 822 if (!parseBrace()) 823 return false; 824 break; 825 case tok::less: 826 if (parseAngle()) { 827 Tok->Type = TT_TemplateOpener; 828 // In TT_Proto, we must distignuish between: 829 // map<key, value> 830 // msg < item: data > 831 // msg: < item: data > 832 // In TT_TextProto, map<key, value> does not occur. 833 if (Style.Language == FormatStyle::LK_TextProto || 834 (Style.Language == FormatStyle::LK_Proto && Tok->Previous && 835 Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) { 836 Tok->Type = TT_DictLiteral; 837 FormatToken *Previous = Tok->getPreviousNonComment(); 838 if (Previous && Previous->Type != TT_DictLiteral) 839 Previous->Type = TT_SelectorName; 840 } 841 } else { 842 Tok->Type = TT_BinaryOperator; 843 NonTemplateLess.insert(Tok); 844 CurrentToken = Tok; 845 next(); 846 } 847 break; 848 case tok::r_paren: 849 case tok::r_square: 850 return false; 851 case tok::r_brace: 852 // Lines can start with '}'. 853 if (Tok->Previous) 854 return false; 855 break; 856 case tok::greater: 857 if (Style.Language != FormatStyle::LK_TextProto) 858 Tok->Type = TT_BinaryOperator; 859 break; 860 case tok::kw_operator: 861 if (Style.Language == FormatStyle::LK_TextProto || 862 Style.Language == FormatStyle::LK_Proto) 863 break; 864 while (CurrentToken && 865 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 866 if (CurrentToken->isOneOf(tok::star, tok::amp)) 867 CurrentToken->Type = TT_PointerOrReference; 868 consumeToken(); 869 if (CurrentToken && 870 CurrentToken->Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator, 871 tok::comma)) 872 CurrentToken->Previous->Type = TT_OverloadedOperator; 873 } 874 if (CurrentToken) { 875 CurrentToken->Type = TT_OverloadedOperatorLParen; 876 if (CurrentToken->Previous->is(TT_BinaryOperator)) 877 CurrentToken->Previous->Type = TT_OverloadedOperator; 878 } 879 break; 880 case tok::question: 881 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next && 882 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren, 883 tok::r_brace)) { 884 // Question marks before semicolons, colons, etc. indicate optional 885 // types (fields, parameters), e.g. 886 // function(x?: string, y?) {...} 887 // class X { y?; } 888 Tok->Type = TT_JsTypeOptionalQuestion; 889 break; 890 } 891 // Declarations cannot be conditional expressions, this can only be part 892 // of a type declaration. 893 if (Line.MustBeDeclaration && !Contexts.back().IsExpression && 894 Style.Language == FormatStyle::LK_JavaScript) 895 break; 896 parseConditional(); 897 break; 898 case tok::kw_template: 899 parseTemplateDeclaration(); 900 break; 901 case tok::comma: 902 if (Contexts.back().InCtorInitializer) 903 Tok->Type = TT_CtorInitializerComma; 904 else if (Contexts.back().InInheritanceList) 905 Tok->Type = TT_InheritanceComma; 906 else if (Contexts.back().FirstStartOfName && 907 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) { 908 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 909 Line.IsMultiVariableDeclStmt = true; 910 } 911 if (Contexts.back().IsForEachMacro) 912 Contexts.back().IsExpression = true; 913 break; 914 case tok::identifier: 915 if (Tok->isOneOf(Keywords.kw___has_include, 916 Keywords.kw___has_include_next)) { 917 parseHasInclude(); 918 } 919 break; 920 default: 921 break; 922 } 923 return true; 924 } 925 926 void parseIncludeDirective() { 927 if (CurrentToken && CurrentToken->is(tok::less)) { 928 next(); 929 while (CurrentToken) { 930 // Mark tokens up to the trailing line comments as implicit string 931 // literals. 932 if (CurrentToken->isNot(tok::comment) && 933 !CurrentToken->TokenText.startswith("//")) 934 CurrentToken->Type = TT_ImplicitStringLiteral; 935 next(); 936 } 937 } 938 } 939 940 void parseWarningOrError() { 941 next(); 942 // We still want to format the whitespace left of the first token of the 943 // warning or error. 944 next(); 945 while (CurrentToken) { 946 CurrentToken->Type = TT_ImplicitStringLiteral; 947 next(); 948 } 949 } 950 951 void parsePragma() { 952 next(); // Consume "pragma". 953 if (CurrentToken && 954 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 955 bool IsMark = CurrentToken->is(Keywords.kw_mark); 956 next(); // Consume "mark". 957 next(); // Consume first token (so we fix leading whitespace). 958 while (CurrentToken) { 959 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 960 CurrentToken->Type = TT_ImplicitStringLiteral; 961 next(); 962 } 963 } 964 } 965 966 void parseHasInclude() { 967 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 968 return; 969 next(); // '(' 970 parseIncludeDirective(); 971 next(); // ')' 972 } 973 974 LineType parsePreprocessorDirective() { 975 bool IsFirstToken = CurrentToken->IsFirst; 976 LineType Type = LT_PreprocessorDirective; 977 next(); 978 if (!CurrentToken) 979 return Type; 980 981 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) { 982 // JavaScript files can contain shebang lines of the form: 983 // #!/usr/bin/env node 984 // Treat these like C++ #include directives. 985 while (CurrentToken) { 986 // Tokens cannot be comments here. 987 CurrentToken->Type = TT_ImplicitStringLiteral; 988 next(); 989 } 990 return LT_ImportStatement; 991 } 992 993 if (CurrentToken->Tok.is(tok::numeric_constant)) { 994 CurrentToken->SpacesRequiredBefore = 1; 995 return Type; 996 } 997 // Hashes in the middle of a line can lead to any strange token 998 // sequence. 999 if (!CurrentToken->Tok.getIdentifierInfo()) 1000 return Type; 1001 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 1002 case tok::pp_include: 1003 case tok::pp_include_next: 1004 case tok::pp_import: 1005 next(); 1006 parseIncludeDirective(); 1007 Type = LT_ImportStatement; 1008 break; 1009 case tok::pp_error: 1010 case tok::pp_warning: 1011 parseWarningOrError(); 1012 break; 1013 case tok::pp_pragma: 1014 parsePragma(); 1015 break; 1016 case tok::pp_if: 1017 case tok::pp_elif: 1018 Contexts.back().IsExpression = true; 1019 parseLine(); 1020 break; 1021 default: 1022 break; 1023 } 1024 while (CurrentToken) { 1025 FormatToken *Tok = CurrentToken; 1026 next(); 1027 if (Tok->is(tok::l_paren)) 1028 parseParens(); 1029 else if (Tok->isOneOf(Keywords.kw___has_include, 1030 Keywords.kw___has_include_next)) 1031 parseHasInclude(); 1032 } 1033 return Type; 1034 } 1035 1036 public: 1037 LineType parseLine() { 1038 NonTemplateLess.clear(); 1039 if (CurrentToken->is(tok::hash)) 1040 return parsePreprocessorDirective(); 1041 1042 // Directly allow to 'import <string-literal>' to support protocol buffer 1043 // definitions (github.com/google/protobuf) or missing "#" (either way we 1044 // should not break the line). 1045 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 1046 if ((Style.Language == FormatStyle::LK_Java && 1047 CurrentToken->is(Keywords.kw_package)) || 1048 (Info && Info->getPPKeywordID() == tok::pp_import && 1049 CurrentToken->Next && 1050 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 1051 tok::kw_static))) { 1052 next(); 1053 parseIncludeDirective(); 1054 return LT_ImportStatement; 1055 } 1056 1057 // If this line starts and ends in '<' and '>', respectively, it is likely 1058 // part of "#define <a/b.h>". 1059 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 1060 parseIncludeDirective(); 1061 return LT_ImportStatement; 1062 } 1063 1064 // In .proto files, top-level options are very similar to import statements 1065 // and should not be line-wrapped. 1066 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 1067 CurrentToken->is(Keywords.kw_option)) { 1068 next(); 1069 if (CurrentToken && CurrentToken->is(tok::identifier)) 1070 return LT_ImportStatement; 1071 } 1072 1073 bool KeywordVirtualFound = false; 1074 bool ImportStatement = false; 1075 1076 // import {...} from '...'; 1077 if (Style.Language == FormatStyle::LK_JavaScript && 1078 CurrentToken->is(Keywords.kw_import)) 1079 ImportStatement = true; 1080 1081 while (CurrentToken) { 1082 if (CurrentToken->is(tok::kw_virtual)) 1083 KeywordVirtualFound = true; 1084 if (Style.Language == FormatStyle::LK_JavaScript) { 1085 // export {...} from '...'; 1086 // An export followed by "from 'some string';" is a re-export from 1087 // another module identified by a URI and is treated as a 1088 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 1089 // Just "export {...};" or "export class ..." should not be treated as 1090 // an import in this sense. 1091 if (Line.First->is(tok::kw_export) && 1092 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 1093 CurrentToken->Next->isStringLiteral()) 1094 ImportStatement = true; 1095 if (isClosureImportStatement(*CurrentToken)) 1096 ImportStatement = true; 1097 } 1098 if (!consumeToken()) 1099 return LT_Invalid; 1100 } 1101 if (KeywordVirtualFound) 1102 return LT_VirtualFunctionDecl; 1103 if (ImportStatement) 1104 return LT_ImportStatement; 1105 1106 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 1107 if (Contexts.back().FirstObjCSelectorName) 1108 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 1109 Contexts.back().LongestObjCSelectorName; 1110 return LT_ObjCMethodDecl; 1111 } 1112 1113 return LT_Other; 1114 } 1115 1116 private: 1117 bool isClosureImportStatement(const FormatToken &Tok) { 1118 // FIXME: Closure-library specific stuff should not be hard-coded but be 1119 // configurable. 1120 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 1121 Tok.Next->Next && 1122 (Tok.Next->Next->TokenText == "module" || 1123 Tok.Next->Next->TokenText == "provide" || 1124 Tok.Next->Next->TokenText == "require" || 1125 Tok.Next->Next->TokenText == "forwardDeclare") && 1126 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 1127 } 1128 1129 void resetTokenMetadata(FormatToken *Token) { 1130 if (!Token) 1131 return; 1132 1133 // Reset token type in case we have already looked at it and then 1134 // recovered from an error (e.g. failure to find the matching >). 1135 if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_ForEachMacro, 1136 TT_FunctionLBrace, TT_ImplicitStringLiteral, 1137 TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow, 1138 TT_OverloadedOperator, TT_RegexLiteral, 1139 TT_TemplateString, TT_ObjCStringLiteral)) 1140 CurrentToken->Type = TT_Unknown; 1141 CurrentToken->Role.reset(); 1142 CurrentToken->MatchingParen = nullptr; 1143 CurrentToken->FakeLParens.clear(); 1144 CurrentToken->FakeRParens = 0; 1145 } 1146 1147 void next() { 1148 if (CurrentToken) { 1149 CurrentToken->NestingLevel = Contexts.size() - 1; 1150 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1151 modifyContext(*CurrentToken); 1152 determineTokenType(*CurrentToken); 1153 CurrentToken = CurrentToken->Next; 1154 } 1155 1156 resetTokenMetadata(CurrentToken); 1157 } 1158 1159 /// A struct to hold information valid in a specific context, e.g. 1160 /// a pair of parenthesis. 1161 struct Context { 1162 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1163 bool IsExpression) 1164 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1165 IsExpression(IsExpression) {} 1166 1167 tok::TokenKind ContextKind; 1168 unsigned BindingStrength; 1169 bool IsExpression; 1170 unsigned LongestObjCSelectorName = 0; 1171 bool ColonIsForRangeExpr = false; 1172 bool ColonIsDictLiteral = false; 1173 bool ColonIsObjCMethodExpr = false; 1174 FormatToken *FirstObjCSelectorName = nullptr; 1175 FormatToken *FirstStartOfName = nullptr; 1176 bool CanBeExpression = true; 1177 bool InTemplateArgument = false; 1178 bool InCtorInitializer = false; 1179 bool InInheritanceList = false; 1180 bool CaretFound = false; 1181 bool IsForEachMacro = false; 1182 bool InCpp11AttributeSpecifier = false; 1183 }; 1184 1185 /// Puts a new \c Context onto the stack \c Contexts for the lifetime 1186 /// of each instance. 1187 struct ScopedContextCreator { 1188 AnnotatingParser &P; 1189 1190 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1191 unsigned Increase) 1192 : P(P) { 1193 P.Contexts.push_back(Context(ContextKind, 1194 P.Contexts.back().BindingStrength + Increase, 1195 P.Contexts.back().IsExpression)); 1196 } 1197 1198 ~ScopedContextCreator() { P.Contexts.pop_back(); } 1199 }; 1200 1201 void modifyContext(const FormatToken &Current) { 1202 if (Current.getPrecedence() == prec::Assignment && 1203 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 1204 // Type aliases use `type X = ...;` in TypeScript and can be exported 1205 // using `export type ...`. 1206 !(Style.Language == FormatStyle::LK_JavaScript && 1207 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1208 Line.startsWith(tok::kw_export, Keywords.kw_type, 1209 tok::identifier))) && 1210 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 1211 Contexts.back().IsExpression = true; 1212 if (!Line.startsWith(TT_UnaryOperator)) { 1213 for (FormatToken *Previous = Current.Previous; 1214 Previous && Previous->Previous && 1215 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1216 Previous = Previous->Previous) { 1217 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1218 Previous = Previous->MatchingParen; 1219 if (!Previous) 1220 break; 1221 } 1222 if (Previous->opensScope()) 1223 break; 1224 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1225 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1226 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1227 Previous->Type = TT_PointerOrReference; 1228 } 1229 } 1230 } else if (Current.is(tok::lessless) && 1231 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1232 Contexts.back().IsExpression = true; 1233 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1234 Contexts.back().IsExpression = true; 1235 } else if (Current.is(TT_TrailingReturnArrow)) { 1236 Contexts.back().IsExpression = false; 1237 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1238 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1239 } else if (Current.Previous && 1240 Current.Previous->is(TT_CtorInitializerColon)) { 1241 Contexts.back().IsExpression = true; 1242 Contexts.back().InCtorInitializer = true; 1243 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1244 Contexts.back().InInheritanceList = true; 1245 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1246 for (FormatToken *Previous = Current.Previous; 1247 Previous && Previous->isOneOf(tok::star, tok::amp); 1248 Previous = Previous->Previous) 1249 Previous->Type = TT_PointerOrReference; 1250 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 1251 Contexts.back().IsExpression = false; 1252 } else if (Current.is(tok::kw_new)) { 1253 Contexts.back().CanBeExpression = false; 1254 } else if (Current.isOneOf(tok::semi, tok::exclaim)) { 1255 // This should be the condition or increment in a for-loop. 1256 Contexts.back().IsExpression = true; 1257 } 1258 } 1259 1260 void determineTokenType(FormatToken &Current) { 1261 if (!Current.is(TT_Unknown)) 1262 // The token type is already known. 1263 return; 1264 1265 if (Style.Language == FormatStyle::LK_JavaScript) { 1266 if (Current.is(tok::exclaim)) { 1267 if (Current.Previous && 1268 (Current.Previous->isOneOf(tok::identifier, tok::kw_namespace, 1269 tok::r_paren, tok::r_square, 1270 tok::r_brace) || 1271 Current.Previous->Tok.isLiteral())) { 1272 Current.Type = TT_JsNonNullAssertion; 1273 return; 1274 } 1275 if (Current.Next && 1276 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1277 Current.Type = TT_JsNonNullAssertion; 1278 return; 1279 } 1280 } 1281 } 1282 1283 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1284 // function declaration have been found. In this case, 'Current' is a 1285 // trailing token of this declaration and thus cannot be a name. 1286 if (Current.is(Keywords.kw_instanceof)) { 1287 Current.Type = TT_BinaryOperator; 1288 } else if (isStartOfName(Current) && 1289 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1290 Contexts.back().FirstStartOfName = &Current; 1291 Current.Type = TT_StartOfName; 1292 } else if (Current.is(tok::semi)) { 1293 // Reset FirstStartOfName after finding a semicolon so that a for loop 1294 // with multiple increment statements is not confused with a for loop 1295 // having multiple variable declarations. 1296 Contexts.back().FirstStartOfName = nullptr; 1297 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1298 AutoFound = true; 1299 } else if (Current.is(tok::arrow) && 1300 Style.Language == FormatStyle::LK_Java) { 1301 Current.Type = TT_LambdaArrow; 1302 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1303 Current.NestingLevel == 0) { 1304 Current.Type = TT_TrailingReturnArrow; 1305 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1306 Current.Type = determineStarAmpUsage(Current, 1307 Contexts.back().CanBeExpression && 1308 Contexts.back().IsExpression, 1309 Contexts.back().InTemplateArgument); 1310 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1311 Current.Type = determinePlusMinusCaretUsage(Current); 1312 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1313 Contexts.back().CaretFound = true; 1314 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1315 Current.Type = determineIncrementUsage(Current); 1316 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1317 Current.Type = TT_UnaryOperator; 1318 } else if (Current.is(tok::question)) { 1319 if (Style.Language == FormatStyle::LK_JavaScript && 1320 Line.MustBeDeclaration && !Contexts.back().IsExpression) { 1321 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1322 // on the interface, not a ternary expression. 1323 Current.Type = TT_JsTypeOptionalQuestion; 1324 } else { 1325 Current.Type = TT_ConditionalExpr; 1326 } 1327 } else if (Current.isBinaryOperator() && 1328 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1329 (!Current.is(tok::greater) && 1330 Style.Language != FormatStyle::LK_TextProto)) { 1331 Current.Type = TT_BinaryOperator; 1332 } else if (Current.is(tok::comment)) { 1333 if (Current.TokenText.startswith("/*")) { 1334 if (Current.TokenText.endswith("*/")) 1335 Current.Type = TT_BlockComment; 1336 else 1337 // The lexer has for some reason determined a comment here. But we 1338 // cannot really handle it, if it isn't properly terminated. 1339 Current.Tok.setKind(tok::unknown); 1340 } else { 1341 Current.Type = TT_LineComment; 1342 } 1343 } else if (Current.is(tok::r_paren)) { 1344 if (rParenEndsCast(Current)) 1345 Current.Type = TT_CastRParen; 1346 if (Current.MatchingParen && Current.Next && 1347 !Current.Next->isBinaryOperator() && 1348 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1349 tok::comma, tok::period, tok::arrow, 1350 tok::coloncolon)) 1351 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1352 // Make sure this isn't the return type of an Obj-C block declaration 1353 if (AfterParen->Tok.isNot(tok::caret)) { 1354 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1355 if (BeforeParen->is(tok::identifier) && 1356 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1357 (!BeforeParen->Previous || 1358 BeforeParen->Previous->ClosesTemplateDeclaration)) 1359 Current.Type = TT_FunctionAnnotationRParen; 1360 } 1361 } 1362 } else if (Current.is(tok::at) && Current.Next && 1363 Style.Language != FormatStyle::LK_JavaScript && 1364 Style.Language != FormatStyle::LK_Java) { 1365 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1366 // marks declarations and properties that need special formatting. 1367 switch (Current.Next->Tok.getObjCKeywordID()) { 1368 case tok::objc_interface: 1369 case tok::objc_implementation: 1370 case tok::objc_protocol: 1371 Current.Type = TT_ObjCDecl; 1372 break; 1373 case tok::objc_property: 1374 Current.Type = TT_ObjCProperty; 1375 break; 1376 default: 1377 break; 1378 } 1379 } else if (Current.is(tok::period)) { 1380 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1381 if (PreviousNoComment && 1382 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1383 Current.Type = TT_DesignatedInitializerPeriod; 1384 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1385 Current.Previous->isOneOf(TT_JavaAnnotation, 1386 TT_LeadingJavaAnnotation)) { 1387 Current.Type = Current.Previous->Type; 1388 } 1389 } else if (canBeObjCSelectorComponent(Current) && 1390 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 1391 Current.Previous && Current.Previous->is(TT_CastRParen) && 1392 Current.Previous->MatchingParen && 1393 Current.Previous->MatchingParen->Previous && 1394 Current.Previous->MatchingParen->Previous->is( 1395 TT_ObjCMethodSpecifier)) { 1396 // This is the first part of an Objective-C selector name. (If there's no 1397 // colon after this, this is the only place which annotates the identifier 1398 // as a selector.) 1399 Current.Type = TT_SelectorName; 1400 } else if (Current.isOneOf(tok::identifier, tok::kw_const) && 1401 Current.Previous && 1402 !Current.Previous->isOneOf(tok::equal, tok::at) && 1403 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1404 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1405 // function declaration have been found. 1406 Current.Type = TT_TrailingAnnotation; 1407 } else if ((Style.Language == FormatStyle::LK_Java || 1408 Style.Language == FormatStyle::LK_JavaScript) && 1409 Current.Previous) { 1410 if (Current.Previous->is(tok::at) && 1411 Current.isNot(Keywords.kw_interface)) { 1412 const FormatToken &AtToken = *Current.Previous; 1413 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1414 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1415 Current.Type = TT_LeadingJavaAnnotation; 1416 else 1417 Current.Type = TT_JavaAnnotation; 1418 } else if (Current.Previous->is(tok::period) && 1419 Current.Previous->isOneOf(TT_JavaAnnotation, 1420 TT_LeadingJavaAnnotation)) { 1421 Current.Type = Current.Previous->Type; 1422 } 1423 } 1424 } 1425 1426 /// Take a guess at whether \p Tok starts a name of a function or 1427 /// variable declaration. 1428 /// 1429 /// This is a heuristic based on whether \p Tok is an identifier following 1430 /// something that is likely a type. 1431 bool isStartOfName(const FormatToken &Tok) { 1432 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1433 return false; 1434 1435 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1436 Keywords.kw_as)) 1437 return false; 1438 if (Style.Language == FormatStyle::LK_JavaScript && 1439 Tok.Previous->is(Keywords.kw_in)) 1440 return false; 1441 1442 // Skip "const" as it does not have an influence on whether this is a name. 1443 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1444 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1445 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1446 1447 if (!PreviousNotConst) 1448 return false; 1449 1450 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1451 PreviousNotConst->Previous && 1452 PreviousNotConst->Previous->is(tok::hash); 1453 1454 if (PreviousNotConst->is(TT_TemplateCloser)) 1455 return PreviousNotConst && PreviousNotConst->MatchingParen && 1456 PreviousNotConst->MatchingParen->Previous && 1457 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1458 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1459 1460 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen && 1461 PreviousNotConst->MatchingParen->Previous && 1462 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype)) 1463 return true; 1464 1465 return (!IsPPKeyword && 1466 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) || 1467 PreviousNotConst->is(TT_PointerOrReference) || 1468 PreviousNotConst->isSimpleTypeSpecifier(); 1469 } 1470 1471 /// Determine whether ')' is ending a cast. 1472 bool rParenEndsCast(const FormatToken &Tok) { 1473 // C-style casts are only used in C++ and Java. 1474 if (!Style.isCpp() && Style.Language != FormatStyle::LK_Java) 1475 return false; 1476 1477 // Empty parens aren't casts and there are no casts at the end of the line. 1478 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1479 return false; 1480 1481 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1482 if (LeftOfParens) { 1483 // If there is a closing parenthesis left of the current parentheses, 1484 // look past it as these might be chained casts. 1485 if (LeftOfParens->is(tok::r_paren)) { 1486 if (!LeftOfParens->MatchingParen || 1487 !LeftOfParens->MatchingParen->Previous) 1488 return false; 1489 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1490 } 1491 1492 // If there is an identifier (or with a few exceptions a keyword) right 1493 // before the parentheses, this is unlikely to be a cast. 1494 if (LeftOfParens->Tok.getIdentifierInfo() && 1495 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1496 tok::kw_delete)) 1497 return false; 1498 1499 // Certain other tokens right before the parentheses are also signals that 1500 // this cannot be a cast. 1501 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1502 TT_TemplateCloser, tok::ellipsis)) 1503 return false; 1504 } 1505 1506 if (Tok.Next->is(tok::question)) 1507 return false; 1508 1509 // As Java has no function types, a "(" after the ")" likely means that this 1510 // is a cast. 1511 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1512 return true; 1513 1514 // If a (non-string) literal follows, this is likely a cast. 1515 if (Tok.Next->isNot(tok::string_literal) && 1516 (Tok.Next->Tok.isLiteral() || 1517 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1518 return true; 1519 1520 // Heuristically try to determine whether the parentheses contain a type. 1521 bool ParensAreType = 1522 !Tok.Previous || 1523 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) || 1524 Tok.Previous->isSimpleTypeSpecifier(); 1525 bool ParensCouldEndDecl = 1526 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1527 if (ParensAreType && !ParensCouldEndDecl) 1528 return true; 1529 1530 // At this point, we heuristically assume that there are no casts at the 1531 // start of the line. We assume that we have found most cases where there 1532 // are by the logic above, e.g. "(void)x;". 1533 if (!LeftOfParens) 1534 return false; 1535 1536 // Certain token types inside the parentheses mean that this can't be a 1537 // cast. 1538 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 1539 Token = Token->Next) 1540 if (Token->is(TT_BinaryOperator)) 1541 return false; 1542 1543 // If the following token is an identifier or 'this', this is a cast. All 1544 // cases where this can be something else are handled above. 1545 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1546 return true; 1547 1548 if (!Tok.Next->Next) 1549 return false; 1550 1551 // If the next token after the parenthesis is a unary operator, assume 1552 // that this is cast, unless there are unexpected tokens inside the 1553 // parenthesis. 1554 bool NextIsUnary = 1555 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 1556 if (!NextIsUnary || Tok.Next->is(tok::plus) || 1557 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 1558 return false; 1559 // Search for unexpected tokens. 1560 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 1561 Prev = Prev->Previous) { 1562 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 1563 return false; 1564 } 1565 return true; 1566 } 1567 1568 /// Return the type of the given token assuming it is * or &. 1569 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 1570 bool InTemplateArgument) { 1571 if (Style.Language == FormatStyle::LK_JavaScript) 1572 return TT_BinaryOperator; 1573 1574 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1575 if (!PrevToken) 1576 return TT_UnaryOperator; 1577 1578 const FormatToken *NextToken = Tok.getNextNonComment(); 1579 if (!NextToken || 1580 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_const) || 1581 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 1582 return TT_PointerOrReference; 1583 1584 if (PrevToken->is(tok::coloncolon)) 1585 return TT_PointerOrReference; 1586 1587 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 1588 tok::comma, tok::semi, tok::kw_return, tok::colon, 1589 tok::equal, tok::kw_delete, tok::kw_sizeof, 1590 tok::kw_throw) || 1591 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 1592 TT_UnaryOperator, TT_CastRParen)) 1593 return TT_UnaryOperator; 1594 1595 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 1596 return TT_PointerOrReference; 1597 if (NextToken->is(tok::kw_operator) && !IsExpression) 1598 return TT_PointerOrReference; 1599 if (NextToken->isOneOf(tok::comma, tok::semi)) 1600 return TT_PointerOrReference; 1601 1602 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen) { 1603 FormatToken *TokenBeforeMatchingParen = 1604 PrevToken->MatchingParen->getPreviousNonComment(); 1605 if (TokenBeforeMatchingParen && 1606 TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype)) 1607 return TT_PointerOrReference; 1608 } 1609 1610 if (PrevToken->Tok.isLiteral() || 1611 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 1612 tok::kw_false, tok::r_brace) || 1613 NextToken->Tok.isLiteral() || 1614 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 1615 NextToken->isUnaryOperator() || 1616 // If we know we're in a template argument, there are no named 1617 // declarations. Thus, having an identifier on the right-hand side 1618 // indicates a binary operator. 1619 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 1620 return TT_BinaryOperator; 1621 1622 // "&&(" is quite unlikely to be two successive unary "&". 1623 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren)) 1624 return TT_BinaryOperator; 1625 1626 // This catches some cases where evaluation order is used as control flow: 1627 // aaa && aaa->f(); 1628 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 1629 if (NextNextToken && NextNextToken->is(tok::arrow)) 1630 return TT_BinaryOperator; 1631 1632 // It is very unlikely that we are going to find a pointer or reference type 1633 // definition on the RHS of an assignment. 1634 if (IsExpression && !Contexts.back().CaretFound) 1635 return TT_BinaryOperator; 1636 1637 return TT_PointerOrReference; 1638 } 1639 1640 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 1641 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1642 if (!PrevToken) 1643 return TT_UnaryOperator; 1644 1645 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 1646 // This must be a sequence of leading unary operators. 1647 return TT_UnaryOperator; 1648 1649 // Use heuristics to recognize unary operators. 1650 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 1651 tok::question, tok::colon, tok::kw_return, 1652 tok::kw_case, tok::at, tok::l_brace)) 1653 return TT_UnaryOperator; 1654 1655 // There can't be two consecutive binary operators. 1656 if (PrevToken->is(TT_BinaryOperator)) 1657 return TT_UnaryOperator; 1658 1659 // Fall back to marking the token as binary operator. 1660 return TT_BinaryOperator; 1661 } 1662 1663 /// Determine whether ++/-- are pre- or post-increments/-decrements. 1664 TokenType determineIncrementUsage(const FormatToken &Tok) { 1665 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1666 if (!PrevToken || PrevToken->is(TT_CastRParen)) 1667 return TT_UnaryOperator; 1668 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 1669 return TT_TrailingUnaryOperator; 1670 1671 return TT_UnaryOperator; 1672 } 1673 1674 SmallVector<Context, 8> Contexts; 1675 1676 const FormatStyle &Style; 1677 AnnotatedLine &Line; 1678 FormatToken *CurrentToken; 1679 bool AutoFound; 1680 const AdditionalKeywords &Keywords; 1681 1682 // Set of "<" tokens that do not open a template parameter list. If parseAngle 1683 // determines that a specific token can't be a template opener, it will make 1684 // same decision irrespective of the decisions for tokens leading up to it. 1685 // Store this information to prevent this from causing exponential runtime. 1686 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 1687 }; 1688 1689 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 1690 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 1691 1692 /// Parses binary expressions by inserting fake parenthesis based on 1693 /// operator precedence. 1694 class ExpressionParser { 1695 public: 1696 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 1697 AnnotatedLine &Line) 1698 : Style(Style), Keywords(Keywords), Current(Line.First) {} 1699 1700 /// Parse expressions with the given operator precedence. 1701 void parse(int Precedence = 0) { 1702 // Skip 'return' and ObjC selector colons as they are not part of a binary 1703 // expression. 1704 while (Current && (Current->is(tok::kw_return) || 1705 (Current->is(tok::colon) && 1706 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 1707 next(); 1708 1709 if (!Current || Precedence > PrecedenceArrowAndPeriod) 1710 return; 1711 1712 // Conditional expressions need to be parsed separately for proper nesting. 1713 if (Precedence == prec::Conditional) { 1714 parseConditionalExpr(); 1715 return; 1716 } 1717 1718 // Parse unary operators, which all have a higher precedence than binary 1719 // operators. 1720 if (Precedence == PrecedenceUnaryOperator) { 1721 parseUnaryOperator(); 1722 return; 1723 } 1724 1725 FormatToken *Start = Current; 1726 FormatToken *LatestOperator = nullptr; 1727 unsigned OperatorIndex = 0; 1728 1729 while (Current) { 1730 // Consume operators with higher precedence. 1731 parse(Precedence + 1); 1732 1733 int CurrentPrecedence = getCurrentPrecedence(); 1734 1735 if (Current && Current->is(TT_SelectorName) && 1736 Precedence == CurrentPrecedence) { 1737 if (LatestOperator) 1738 addFakeParenthesis(Start, prec::Level(Precedence)); 1739 Start = Current; 1740 } 1741 1742 // At the end of the line or when an operator with higher precedence is 1743 // found, insert fake parenthesis and return. 1744 if (!Current || 1745 (Current->closesScope() && 1746 (Current->MatchingParen || Current->is(TT_TemplateString))) || 1747 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 1748 (CurrentPrecedence == prec::Conditional && 1749 Precedence == prec::Assignment && Current->is(tok::colon))) { 1750 break; 1751 } 1752 1753 // Consume scopes: (), [], <> and {} 1754 if (Current->opensScope()) { 1755 // In fragment of a JavaScript template string can look like '}..${' and 1756 // thus close a scope and open a new one at the same time. 1757 while (Current && (!Current->closesScope() || Current->opensScope())) { 1758 next(); 1759 parse(); 1760 } 1761 next(); 1762 } else { 1763 // Operator found. 1764 if (CurrentPrecedence == Precedence) { 1765 if (LatestOperator) 1766 LatestOperator->NextOperator = Current; 1767 LatestOperator = Current; 1768 Current->OperatorIndex = OperatorIndex; 1769 ++OperatorIndex; 1770 } 1771 next(/*SkipPastLeadingComments=*/Precedence > 0); 1772 } 1773 } 1774 1775 if (LatestOperator && (Current || Precedence > 0)) { 1776 // LatestOperator->LastOperator = true; 1777 if (Precedence == PrecedenceArrowAndPeriod) { 1778 // Call expressions don't have a binary operator precedence. 1779 addFakeParenthesis(Start, prec::Unknown); 1780 } else { 1781 addFakeParenthesis(Start, prec::Level(Precedence)); 1782 } 1783 } 1784 } 1785 1786 private: 1787 /// Gets the precedence (+1) of the given token for binary operators 1788 /// and other tokens that we treat like binary operators. 1789 int getCurrentPrecedence() { 1790 if (Current) { 1791 const FormatToken *NextNonComment = Current->getNextNonComment(); 1792 if (Current->is(TT_ConditionalExpr)) 1793 return prec::Conditional; 1794 if (NextNonComment && Current->is(TT_SelectorName) && 1795 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 1796 ((Style.Language == FormatStyle::LK_Proto || 1797 Style.Language == FormatStyle::LK_TextProto) && 1798 NextNonComment->is(tok::less)))) 1799 return prec::Assignment; 1800 if (Current->is(TT_JsComputedPropertyName)) 1801 return prec::Assignment; 1802 if (Current->is(TT_LambdaArrow)) 1803 return prec::Comma; 1804 if (Current->is(TT_JsFatArrow)) 1805 return prec::Assignment; 1806 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 1807 (Current->is(tok::comment) && NextNonComment && 1808 NextNonComment->is(TT_SelectorName))) 1809 return 0; 1810 if (Current->is(TT_RangeBasedForLoopColon)) 1811 return prec::Comma; 1812 if ((Style.Language == FormatStyle::LK_Java || 1813 Style.Language == FormatStyle::LK_JavaScript) && 1814 Current->is(Keywords.kw_instanceof)) 1815 return prec::Relational; 1816 if (Style.Language == FormatStyle::LK_JavaScript && 1817 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 1818 return prec::Relational; 1819 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 1820 return Current->getPrecedence(); 1821 if (Current->isOneOf(tok::period, tok::arrow)) 1822 return PrecedenceArrowAndPeriod; 1823 if ((Style.Language == FormatStyle::LK_Java || 1824 Style.Language == FormatStyle::LK_JavaScript) && 1825 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 1826 Keywords.kw_throws)) 1827 return 0; 1828 } 1829 return -1; 1830 } 1831 1832 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 1833 Start->FakeLParens.push_back(Precedence); 1834 if (Precedence > prec::Unknown) 1835 Start->StartsBinaryExpression = true; 1836 if (Current) { 1837 FormatToken *Previous = Current->Previous; 1838 while (Previous->is(tok::comment) && Previous->Previous) 1839 Previous = Previous->Previous; 1840 ++Previous->FakeRParens; 1841 if (Precedence > prec::Unknown) 1842 Previous->EndsBinaryExpression = true; 1843 } 1844 } 1845 1846 /// Parse unary operator expressions and surround them with fake 1847 /// parentheses if appropriate. 1848 void parseUnaryOperator() { 1849 llvm::SmallVector<FormatToken *, 2> Tokens; 1850 while (Current && Current->is(TT_UnaryOperator)) { 1851 Tokens.push_back(Current); 1852 next(); 1853 } 1854 parse(PrecedenceArrowAndPeriod); 1855 for (FormatToken *Token : llvm::reverse(Tokens)) 1856 // The actual precedence doesn't matter. 1857 addFakeParenthesis(Token, prec::Unknown); 1858 } 1859 1860 void parseConditionalExpr() { 1861 while (Current && Current->isTrailingComment()) { 1862 next(); 1863 } 1864 FormatToken *Start = Current; 1865 parse(prec::LogicalOr); 1866 if (!Current || !Current->is(tok::question)) 1867 return; 1868 next(); 1869 parse(prec::Assignment); 1870 if (!Current || Current->isNot(TT_ConditionalExpr)) 1871 return; 1872 next(); 1873 parse(prec::Assignment); 1874 addFakeParenthesis(Start, prec::Conditional); 1875 } 1876 1877 void next(bool SkipPastLeadingComments = true) { 1878 if (Current) 1879 Current = Current->Next; 1880 while (Current && 1881 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 1882 Current->isTrailingComment()) 1883 Current = Current->Next; 1884 } 1885 1886 const FormatStyle &Style; 1887 const AdditionalKeywords &Keywords; 1888 FormatToken *Current; 1889 }; 1890 1891 } // end anonymous namespace 1892 1893 void TokenAnnotator::setCommentLineLevels( 1894 SmallVectorImpl<AnnotatedLine *> &Lines) { 1895 const AnnotatedLine *NextNonCommentLine = nullptr; 1896 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(), 1897 E = Lines.rend(); 1898 I != E; ++I) { 1899 bool CommentLine = true; 1900 for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) { 1901 if (!Tok->is(tok::comment)) { 1902 CommentLine = false; 1903 break; 1904 } 1905 } 1906 1907 // If the comment is currently aligned with the line immediately following 1908 // it, that's probably intentional and we should keep it. 1909 if (NextNonCommentLine && CommentLine && 1910 NextNonCommentLine->First->NewlinesBefore <= 1 && 1911 NextNonCommentLine->First->OriginalColumn == 1912 (*I)->First->OriginalColumn) { 1913 // Align comments for preprocessor lines with the # in column 0. 1914 // Otherwise, align with the next line. 1915 (*I)->Level = (NextNonCommentLine->Type == LT_PreprocessorDirective || 1916 NextNonCommentLine->Type == LT_ImportStatement) 1917 ? 0 1918 : NextNonCommentLine->Level; 1919 } else { 1920 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr; 1921 } 1922 1923 setCommentLineLevels((*I)->Children); 1924 } 1925 } 1926 1927 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 1928 unsigned Result = 0; 1929 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 1930 Result = std::max(Result, Tok->NestingLevel); 1931 return Result; 1932 } 1933 1934 void TokenAnnotator::annotate(AnnotatedLine &Line) { 1935 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 1936 E = Line.Children.end(); 1937 I != E; ++I) { 1938 annotate(**I); 1939 } 1940 AnnotatingParser Parser(Style, Line, Keywords); 1941 Line.Type = Parser.parseLine(); 1942 1943 // With very deep nesting, ExpressionParser uses lots of stack and the 1944 // formatting algorithm is very slow. We're not going to do a good job here 1945 // anyway - it's probably generated code being formatted by mistake. 1946 // Just skip the whole line. 1947 if (maxNestingDepth(Line) > 50) 1948 Line.Type = LT_Invalid; 1949 1950 if (Line.Type == LT_Invalid) 1951 return; 1952 1953 ExpressionParser ExprParser(Style, Keywords, Line); 1954 ExprParser.parse(); 1955 1956 if (Line.startsWith(TT_ObjCMethodSpecifier)) 1957 Line.Type = LT_ObjCMethodDecl; 1958 else if (Line.startsWith(TT_ObjCDecl)) 1959 Line.Type = LT_ObjCDecl; 1960 else if (Line.startsWith(TT_ObjCProperty)) 1961 Line.Type = LT_ObjCProperty; 1962 1963 Line.First->SpacesRequiredBefore = 1; 1964 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 1965 } 1966 1967 // This function heuristically determines whether 'Current' starts the name of a 1968 // function declaration. 1969 static bool isFunctionDeclarationName(const FormatToken &Current, 1970 const AnnotatedLine &Line) { 1971 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 1972 for (; Next; Next = Next->Next) { 1973 if (Next->is(TT_OverloadedOperatorLParen)) 1974 return Next; 1975 if (Next->is(TT_OverloadedOperator)) 1976 continue; 1977 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 1978 // For 'new[]' and 'delete[]'. 1979 if (Next->Next && Next->Next->is(tok::l_square) && Next->Next->Next && 1980 Next->Next->Next->is(tok::r_square)) 1981 Next = Next->Next->Next; 1982 continue; 1983 } 1984 1985 break; 1986 } 1987 return nullptr; 1988 }; 1989 1990 // Find parentheses of parameter list. 1991 const FormatToken *Next = Current.Next; 1992 if (Current.is(tok::kw_operator)) { 1993 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 1994 return false; 1995 Next = skipOperatorName(Next); 1996 } else { 1997 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 1998 return false; 1999 for (; Next; Next = Next->Next) { 2000 if (Next->is(TT_TemplateOpener)) { 2001 Next = Next->MatchingParen; 2002 } else if (Next->is(tok::coloncolon)) { 2003 Next = Next->Next; 2004 if (!Next) 2005 return false; 2006 if (Next->is(tok::kw_operator)) { 2007 Next = skipOperatorName(Next->Next); 2008 break; 2009 } 2010 if (!Next->is(tok::identifier)) 2011 return false; 2012 } else if (Next->is(tok::l_paren)) { 2013 break; 2014 } else { 2015 return false; 2016 } 2017 } 2018 } 2019 2020 // Check whether parameter list can belong to a function declaration. 2021 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 2022 return false; 2023 // If the lines ends with "{", this is likely an function definition. 2024 if (Line.Last->is(tok::l_brace)) 2025 return true; 2026 if (Next->Next == Next->MatchingParen) 2027 return true; // Empty parentheses. 2028 // If there is an &/&& after the r_paren, this is likely a function. 2029 if (Next->MatchingParen->Next && 2030 Next->MatchingParen->Next->is(TT_PointerOrReference)) 2031 return true; 2032 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 2033 Tok = Tok->Next) { 2034 if (Tok->is(tok::l_paren) && Tok->MatchingParen) { 2035 Tok = Tok->MatchingParen; 2036 continue; 2037 } 2038 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 2039 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 2040 return true; 2041 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 2042 Tok->Tok.isLiteral()) 2043 return false; 2044 } 2045 return false; 2046 } 2047 2048 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 2049 assert(Line.MightBeFunctionDecl); 2050 2051 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 2052 Style.AlwaysBreakAfterReturnType == 2053 FormatStyle::RTBS_TopLevelDefinitions) && 2054 Line.Level > 0) 2055 return false; 2056 2057 switch (Style.AlwaysBreakAfterReturnType) { 2058 case FormatStyle::RTBS_None: 2059 return false; 2060 case FormatStyle::RTBS_All: 2061 case FormatStyle::RTBS_TopLevel: 2062 return true; 2063 case FormatStyle::RTBS_AllDefinitions: 2064 case FormatStyle::RTBS_TopLevelDefinitions: 2065 return Line.mightBeFunctionDefinition(); 2066 } 2067 2068 return false; 2069 } 2070 2071 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 2072 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2073 E = Line.Children.end(); 2074 I != E; ++I) { 2075 calculateFormattingInformation(**I); 2076 } 2077 2078 Line.First->TotalLength = 2079 Line.First->IsMultiline ? Style.ColumnLimit 2080 : Line.FirstStartColumn + Line.First->ColumnWidth; 2081 FormatToken *Current = Line.First->Next; 2082 bool InFunctionDecl = Line.MightBeFunctionDecl; 2083 while (Current) { 2084 if (isFunctionDeclarationName(*Current, Line)) 2085 Current->Type = TT_FunctionDeclarationName; 2086 if (Current->is(TT_LineComment)) { 2087 if (Current->Previous->BlockKind == BK_BracedInit && 2088 Current->Previous->opensScope()) 2089 Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1; 2090 else 2091 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2092 2093 // If we find a trailing comment, iterate backwards to determine whether 2094 // it seems to relate to a specific parameter. If so, break before that 2095 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2096 // to the previous line in: 2097 // SomeFunction(a, 2098 // b, // comment 2099 // c); 2100 if (!Current->HasUnescapedNewline) { 2101 for (FormatToken *Parameter = Current->Previous; Parameter; 2102 Parameter = Parameter->Previous) { 2103 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 2104 break; 2105 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 2106 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 2107 Parameter->HasUnescapedNewline) 2108 Parameter->MustBreakBefore = true; 2109 break; 2110 } 2111 } 2112 } 2113 } else if (Current->SpacesRequiredBefore == 0 && 2114 spaceRequiredBefore(Line, *Current)) { 2115 Current->SpacesRequiredBefore = 1; 2116 } 2117 2118 Current->MustBreakBefore = 2119 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 2120 2121 if (!Current->MustBreakBefore && InFunctionDecl && 2122 Current->is(TT_FunctionDeclarationName)) 2123 Current->MustBreakBefore = mustBreakForReturnType(Line); 2124 2125 Current->CanBreakBefore = 2126 Current->MustBreakBefore || canBreakBefore(Line, *Current); 2127 unsigned ChildSize = 0; 2128 if (Current->Previous->Children.size() == 1) { 2129 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last; 2130 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 2131 : LastOfChild.TotalLength + 1; 2132 } 2133 const FormatToken *Prev = Current->Previous; 2134 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 2135 (Prev->Children.size() == 1 && 2136 Prev->Children[0]->First->MustBreakBefore) || 2137 Current->IsMultiline) 2138 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 2139 else 2140 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 2141 ChildSize + Current->SpacesRequiredBefore; 2142 2143 if (Current->is(TT_CtorInitializerColon)) 2144 InFunctionDecl = false; 2145 2146 // FIXME: Only calculate this if CanBreakBefore is true once static 2147 // initializers etc. are sorted out. 2148 // FIXME: Move magic numbers to a better place. 2149 2150 // Reduce penalty for aligning ObjC method arguments using the colon 2151 // alignment as this is the canonical way (still prefer fitting everything 2152 // into one line if possible). Trying to fit a whole expression into one 2153 // line should not force other line breaks (e.g. when ObjC method 2154 // expression is a part of other expression). 2155 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl); 2156 if (Style.Language == FormatStyle::LK_ObjC && 2157 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) { 2158 if (Current->ParameterIndex == 1) 2159 Current->SplitPenalty += 5 * Current->BindingStrength; 2160 } else { 2161 Current->SplitPenalty += 20 * Current->BindingStrength; 2162 } 2163 2164 Current = Current->Next; 2165 } 2166 2167 calculateUnbreakableTailLengths(Line); 2168 unsigned IndentLevel = Line.Level; 2169 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2170 if (Current->Role) 2171 Current->Role->precomputeFormattingInfos(Current); 2172 if (Current->MatchingParen && 2173 Current->MatchingParen->opensBlockOrBlockTypeList(Style)) { 2174 assert(IndentLevel > 0); 2175 --IndentLevel; 2176 } 2177 Current->IndentLevel = IndentLevel; 2178 if (Current->opensBlockOrBlockTypeList(Style)) 2179 ++IndentLevel; 2180 } 2181 2182 LLVM_DEBUG({ printDebugInfo(Line); }); 2183 } 2184 2185 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2186 unsigned UnbreakableTailLength = 0; 2187 FormatToken *Current = Line.Last; 2188 while (Current) { 2189 Current->UnbreakableTailLength = UnbreakableTailLength; 2190 if (Current->CanBreakBefore || 2191 Current->isOneOf(tok::comment, tok::string_literal)) { 2192 UnbreakableTailLength = 0; 2193 } else { 2194 UnbreakableTailLength += 2195 Current->ColumnWidth + Current->SpacesRequiredBefore; 2196 } 2197 Current = Current->Previous; 2198 } 2199 } 2200 2201 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2202 const FormatToken &Tok, 2203 bool InFunctionDecl) { 2204 const FormatToken &Left = *Tok.Previous; 2205 const FormatToken &Right = Tok; 2206 2207 if (Left.is(tok::semi)) 2208 return 0; 2209 2210 if (Style.Language == FormatStyle::LK_Java) { 2211 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2212 return 1; 2213 if (Right.is(Keywords.kw_implements)) 2214 return 2; 2215 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2216 return 3; 2217 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2218 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2219 return 100; 2220 if (Left.is(TT_JsTypeColon)) 2221 return 35; 2222 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2223 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2224 return 100; 2225 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2226 if (Left.opensScope() && Right.closesScope()) 2227 return 200; 2228 } 2229 2230 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2231 return 1; 2232 if (Right.is(tok::l_square)) { 2233 if (Style.Language == FormatStyle::LK_Proto) 2234 return 1; 2235 if (Left.is(tok::r_square)) 2236 return 200; 2237 // Slightly prefer formatting local lambda definitions like functions. 2238 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2239 return 35; 2240 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2241 TT_ArrayInitializerLSquare, 2242 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) 2243 return 500; 2244 } 2245 2246 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2247 Right.is(tok::kw_operator)) { 2248 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2249 return 3; 2250 if (Left.is(TT_StartOfName)) 2251 return 110; 2252 if (InFunctionDecl && Right.NestingLevel == 0) 2253 return Style.PenaltyReturnTypeOnItsOwnLine; 2254 return 200; 2255 } 2256 if (Right.is(TT_PointerOrReference)) 2257 return 190; 2258 if (Right.is(TT_LambdaArrow)) 2259 return 110; 2260 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2261 return 160; 2262 if (Left.is(TT_CastRParen)) 2263 return 100; 2264 if (Left.is(tok::coloncolon) || 2265 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2266 return 500; 2267 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2268 return 5000; 2269 if (Left.is(tok::comment)) 2270 return 1000; 2271 2272 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2273 TT_CtorInitializerColon)) 2274 return 2; 2275 2276 if (Right.isMemberAccess()) { 2277 // Breaking before the "./->" of a chained call/member access is reasonably 2278 // cheap, as formatting those with one call per line is generally 2279 // desirable. In particular, it should be cheaper to break before the call 2280 // than it is to break inside a call's parameters, which could lead to weird 2281 // "hanging" indents. The exception is the very last "./->" to support this 2282 // frequent pattern: 2283 // 2284 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2285 // dddddddd); 2286 // 2287 // which might otherwise be blown up onto many lines. Here, clang-format 2288 // won't produce "hanging" indents anyway as there is no other trailing 2289 // call. 2290 // 2291 // Also apply higher penalty is not a call as that might lead to a wrapping 2292 // like: 2293 // 2294 // aaaaaaa 2295 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2296 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2297 ? 150 2298 : 35; 2299 } 2300 2301 if (Right.is(TT_TrailingAnnotation) && 2302 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2303 // Moving trailing annotations to the next line is fine for ObjC method 2304 // declarations. 2305 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2306 return 10; 2307 // Generally, breaking before a trailing annotation is bad unless it is 2308 // function-like. It seems to be especially preferable to keep standard 2309 // annotations (i.e. "const", "final" and "override") on the same line. 2310 // Use a slightly higher penalty after ")" so that annotations like 2311 // "const override" are kept together. 2312 bool is_short_annotation = Right.TokenText.size() < 10; 2313 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 2314 } 2315 2316 // In for-loops, prefer breaking at ',' and ';'. 2317 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 2318 return 4; 2319 2320 // In Objective-C method expressions, prefer breaking before "param:" over 2321 // breaking after it. 2322 if (Right.is(TT_SelectorName)) 2323 return 0; 2324 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 2325 return Line.MightBeFunctionDecl ? 50 : 500; 2326 2327 // In Objective-C type declarations, avoid breaking after the category's 2328 // open paren (we'll prefer breaking after the protocol list's opening 2329 // angle bracket, if present). 2330 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous && 2331 Left.Previous->isOneOf(tok::identifier, tok::greater)) 2332 return 500; 2333 2334 if (Left.is(tok::l_paren) && InFunctionDecl && 2335 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 2336 return 100; 2337 if (Left.is(tok::l_paren) && Left.Previous && 2338 (Left.Previous->isOneOf(tok::kw_if, tok::kw_for) || 2339 Left.Previous->endsSequence(tok::kw_constexpr, tok::kw_if))) 2340 return 1000; 2341 if (Left.is(tok::equal) && InFunctionDecl) 2342 return 110; 2343 if (Right.is(tok::r_brace)) 2344 return 1; 2345 if (Left.is(TT_TemplateOpener)) 2346 return 100; 2347 if (Left.opensScope()) { 2348 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign) 2349 return 0; 2350 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle) 2351 return 19; 2352 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 2353 : 19; 2354 } 2355 if (Left.is(TT_JavaAnnotation)) 2356 return 50; 2357 2358 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 2359 Left.Previous->isLabelString() && 2360 (Left.NextOperator || Left.OperatorIndex != 0)) 2361 return 50; 2362 if (Right.is(tok::plus) && Left.isLabelString() && 2363 (Right.NextOperator || Right.OperatorIndex != 0)) 2364 return 25; 2365 if (Left.is(tok::comma)) 2366 return 1; 2367 if (Right.is(tok::lessless) && Left.isLabelString() && 2368 (Right.NextOperator || Right.OperatorIndex != 1)) 2369 return 25; 2370 if (Right.is(tok::lessless)) { 2371 // Breaking at a << is really cheap. 2372 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 2373 // Slightly prefer to break before the first one in log-like statements. 2374 return 2; 2375 return 1; 2376 } 2377 if (Left.ClosesTemplateDeclaration) 2378 return Style.PenaltyBreakTemplateDeclaration; 2379 if (Left.is(TT_ConditionalExpr)) 2380 return prec::Conditional; 2381 prec::Level Level = Left.getPrecedence(); 2382 if (Level == prec::Unknown) 2383 Level = Right.getPrecedence(); 2384 if (Level == prec::Assignment) 2385 return Style.PenaltyBreakAssignment; 2386 if (Level != prec::Unknown) 2387 return Level; 2388 2389 return 3; 2390 } 2391 2392 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 2393 const FormatToken &Left, 2394 const FormatToken &Right) { 2395 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 2396 return true; 2397 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 2398 return true; 2399 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 2400 Left.Tok.getObjCKeywordID() == tok::objc_property) 2401 return true; 2402 if (Right.is(tok::hashhash)) 2403 return Left.is(tok::hash); 2404 if (Left.isOneOf(tok::hashhash, tok::hash)) 2405 return Right.is(tok::hash); 2406 if (Left.is(tok::l_paren) && Right.is(tok::r_paren)) 2407 return Style.SpaceInEmptyParentheses; 2408 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 2409 return (Right.is(TT_CastRParen) || 2410 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 2411 ? Style.SpacesInCStyleCastParentheses 2412 : Style.SpacesInParentheses; 2413 if (Right.isOneOf(tok::semi, tok::comma)) 2414 return false; 2415 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) { 2416 bool IsLightweightGeneric = 2417 Right.MatchingParen && Right.MatchingParen->Next && 2418 Right.MatchingParen->Next->is(tok::colon); 2419 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList; 2420 } 2421 if (Right.is(tok::less) && Left.is(tok::kw_template)) 2422 return Style.SpaceAfterTemplateKeyword; 2423 if (Left.isOneOf(tok::exclaim, tok::tilde)) 2424 return false; 2425 if (Left.is(tok::at) && 2426 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 2427 tok::numeric_constant, tok::l_paren, tok::l_brace, 2428 tok::kw_true, tok::kw_false)) 2429 return false; 2430 if (Left.is(tok::colon)) 2431 return !Left.is(TT_ObjCMethodExpr); 2432 if (Left.is(tok::coloncolon)) 2433 return false; 2434 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 2435 if (Style.Language == FormatStyle::LK_TextProto || 2436 (Style.Language == FormatStyle::LK_Proto && 2437 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 2438 // Format empty list as `<>`. 2439 if (Left.is(tok::less) && Right.is(tok::greater)) 2440 return false; 2441 return !Style.Cpp11BracedListStyle; 2442 } 2443 return false; 2444 } 2445 if (Right.is(tok::ellipsis)) 2446 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 2447 Left.Previous->is(tok::kw_case)); 2448 if (Left.is(tok::l_square) && Right.is(tok::amp)) 2449 return false; 2450 if (Right.is(TT_PointerOrReference)) { 2451 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 2452 if (!Left.MatchingParen) 2453 return true; 2454 FormatToken *TokenBeforeMatchingParen = 2455 Left.MatchingParen->getPreviousNonComment(); 2456 if (!TokenBeforeMatchingParen || 2457 !TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype)) 2458 return true; 2459 } 2460 return (Left.Tok.isLiteral() || 2461 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 2462 (Style.PointerAlignment != FormatStyle::PAS_Left || 2463 (Line.IsMultiVariableDeclStmt && 2464 (Left.NestingLevel == 0 || 2465 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for))))))); 2466 } 2467 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 2468 (!Left.is(TT_PointerOrReference) || 2469 (Style.PointerAlignment != FormatStyle::PAS_Right && 2470 !Line.IsMultiVariableDeclStmt))) 2471 return true; 2472 if (Left.is(TT_PointerOrReference)) 2473 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) || 2474 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 2475 !Right.is(TT_StartOfName)) || 2476 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) || 2477 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 2478 tok::l_paren) && 2479 (Style.PointerAlignment != FormatStyle::PAS_Right && 2480 !Line.IsMultiVariableDeclStmt) && 2481 Left.Previous && 2482 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon)); 2483 if (Right.is(tok::star) && Left.is(tok::l_paren)) 2484 return false; 2485 const auto SpaceRequiredForArrayInitializerLSquare = 2486 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 2487 return Style.SpacesInContainerLiterals || 2488 ((Style.Language == FormatStyle::LK_Proto || 2489 Style.Language == FormatStyle::LK_TextProto) && 2490 !Style.Cpp11BracedListStyle && 2491 LSquareTok.endsSequence(tok::l_square, tok::colon, 2492 TT_SelectorName)); 2493 }; 2494 if (Left.is(tok::l_square)) 2495 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 2496 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 2497 (Left.isOneOf(TT_ArraySubscriptLSquare, 2498 TT_StructuredBindingLSquare) && 2499 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 2500 if (Right.is(tok::r_square)) 2501 return Right.MatchingParen && 2502 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 2503 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 2504 Style)) || 2505 (Style.SpacesInSquareBrackets && 2506 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 2507 TT_StructuredBindingLSquare)) || 2508 Right.MatchingParen->is(TT_AttributeParen)); 2509 if (Right.is(tok::l_square) && 2510 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2511 TT_DesignatedInitializerLSquare, 2512 TT_StructuredBindingLSquare, TT_AttributeSquare) && 2513 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral)) 2514 return false; 2515 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 2516 return !Left.Children.empty(); // No spaces in "{}". 2517 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) || 2518 (Right.is(tok::r_brace) && Right.MatchingParen && 2519 Right.MatchingParen->BlockKind != BK_Block)) 2520 return !Style.Cpp11BracedListStyle; 2521 if (Left.is(TT_BlockComment)) 2522 // No whitespace in x(/*foo=*/1), except for JavaScript. 2523 return Style.Language == FormatStyle::LK_JavaScript || 2524 !Left.TokenText.endswith("=*/"); 2525 if (Right.is(tok::l_paren)) { 2526 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 2527 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) 2528 return true; 2529 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) || 2530 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never && 2531 (Left.isOneOf(tok::kw_if, tok::pp_elif, tok::kw_for, tok::kw_while, 2532 tok::kw_switch, tok::kw_case, TT_ForEachMacro, 2533 TT_ObjCForIn) || 2534 Left.endsSequence(tok::kw_constexpr, tok::kw_if) || 2535 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch, 2536 tok::kw_new, tok::kw_delete) && 2537 (!Left.Previous || Left.Previous->isNot(tok::period))))) || 2538 (Style.SpaceBeforeParens == FormatStyle::SBPO_Always && 2539 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 2540 Left.is(tok::r_paren)) && 2541 Line.Type != LT_PreprocessorDirective); 2542 } 2543 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 2544 return false; 2545 if (Right.is(TT_UnaryOperator)) 2546 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 2547 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 2548 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 2549 tok::r_paren) || 2550 Left.isSimpleTypeSpecifier()) && 2551 Right.is(tok::l_brace) && Right.getNextNonComment() && 2552 Right.BlockKind != BK_Block) 2553 return false; 2554 if (Left.is(tok::period) || Right.is(tok::period)) 2555 return false; 2556 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L") 2557 return false; 2558 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 2559 Left.MatchingParen->Previous && 2560 (Left.MatchingParen->Previous->is(tok::period) || 2561 Left.MatchingParen->Previous->is(tok::coloncolon))) 2562 // Java call to generic function with explicit type: 2563 // A.<B<C<...>>>DoSomething(); 2564 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference. 2565 return false; 2566 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 2567 return false; 2568 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at)) 2569 // Objective-C dictionary literal -> no space after opening brace. 2570 return false; 2571 if (Right.is(tok::r_brace) && Right.MatchingParen && 2572 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at)) 2573 // Objective-C dictionary literal -> no space before closing brace. 2574 return false; 2575 return true; 2576 } 2577 2578 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 2579 const FormatToken &Right) { 2580 const FormatToken &Left = *Right.Previous; 2581 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 2582 return true; // Never ever merge two identifiers. 2583 if (Style.isCpp()) { 2584 if (Left.is(tok::kw_operator)) 2585 return Right.is(tok::coloncolon); 2586 if (Right.is(tok::l_brace) && Right.BlockKind == BK_BracedInit && 2587 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList) 2588 return true; 2589 } else if (Style.Language == FormatStyle::LK_Proto || 2590 Style.Language == FormatStyle::LK_TextProto) { 2591 if (Right.is(tok::period) && 2592 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 2593 Keywords.kw_repeated, Keywords.kw_extend)) 2594 return true; 2595 if (Right.is(tok::l_paren) && 2596 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 2597 return true; 2598 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 2599 return true; 2600 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 2601 if (Left.is(tok::slash) || Right.is(tok::slash)) 2602 return false; 2603 if (Left.MatchingParen && Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 2604 Right.isOneOf(tok::l_brace, tok::less)) 2605 return !Style.Cpp11BracedListStyle; 2606 // A percent is probably part of a formatting specification, such as %lld. 2607 if (Left.is(tok::percent)) 2608 return false; 2609 // Preserve the existence of a space before a percent for cases like 0x%04x 2610 // and "%d %d" 2611 if (Left.is(tok::numeric_constant) && Right.is(tok::percent)) 2612 return Right.WhitespaceRange.getEnd() != Right.WhitespaceRange.getBegin(); 2613 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2614 if (Left.is(TT_JsFatArrow)) 2615 return true; 2616 // for await ( ... 2617 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 2618 Left.Previous->is(tok::kw_for)) 2619 return true; 2620 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 2621 Right.MatchingParen) { 2622 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 2623 // An async arrow function, for example: `x = async () => foo();`, 2624 // as opposed to calling a function called async: `x = async();` 2625 if (Next && Next->is(TT_JsFatArrow)) 2626 return true; 2627 } 2628 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2629 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2630 return false; 2631 // In tagged template literals ("html`bar baz`"), there is no space between 2632 // the tag identifier and the template string. getIdentifierInfo makes sure 2633 // that the identifier is not a pseudo keyword like `yield`, either. 2634 if (Left.is(tok::identifier) && Keywords.IsJavaScriptIdentifier(Left) && 2635 Right.is(TT_TemplateString)) 2636 return false; 2637 if (Right.is(tok::star) && 2638 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 2639 return false; 2640 if (Right.isOneOf(tok::l_brace, tok::l_square) && 2641 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 2642 Keywords.kw_extends, Keywords.kw_implements)) 2643 return true; 2644 if (Right.is(tok::l_paren)) { 2645 // JS methods can use some keywords as names (e.g. `delete()`). 2646 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 2647 return false; 2648 // Valid JS method names can include keywords, e.g. `foo.delete()` or 2649 // `bar.instanceof()`. Recognize call positions by preceding period. 2650 if (Left.Previous && Left.Previous->is(tok::period) && 2651 Left.Tok.getIdentifierInfo()) 2652 return false; 2653 // Additional unary JavaScript operators that need a space after. 2654 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 2655 tok::kw_void)) 2656 return true; 2657 } 2658 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 2659 tok::kw_const) || 2660 // "of" is only a keyword if it appears after another identifier 2661 // (e.g. as "const x of y" in a for loop), or after a destructuring 2662 // operation (const [x, y] of z, const {a, b} of c). 2663 (Left.is(Keywords.kw_of) && Left.Previous && 2664 (Left.Previous->Tok.is(tok::identifier) || 2665 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 2666 (!Left.Previous || !Left.Previous->is(tok::period))) 2667 return true; 2668 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 2669 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 2670 return false; 2671 if (Left.is(Keywords.kw_as) && 2672 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 2673 return true; 2674 if (Left.is(tok::kw_default) && Left.Previous && 2675 Left.Previous->is(tok::kw_export)) 2676 return true; 2677 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 2678 return true; 2679 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 2680 return false; 2681 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 2682 return false; 2683 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 2684 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 2685 return false; 2686 if (Left.is(tok::ellipsis)) 2687 return false; 2688 if (Left.is(TT_TemplateCloser) && 2689 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 2690 Keywords.kw_implements, Keywords.kw_extends)) 2691 // Type assertions ('<type>expr') are not followed by whitespace. Other 2692 // locations that should have whitespace following are identified by the 2693 // above set of follower tokens. 2694 return false; 2695 if (Right.is(TT_JsNonNullAssertion)) 2696 return false; 2697 if (Left.is(TT_JsNonNullAssertion) && 2698 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 2699 return true; // "x! as string", "x! in y" 2700 } else if (Style.Language == FormatStyle::LK_Java) { 2701 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 2702 return true; 2703 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 2704 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never; 2705 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 2706 tok::kw_protected) || 2707 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 2708 Keywords.kw_native)) && 2709 Right.is(TT_TemplateOpener)) 2710 return true; 2711 } 2712 if (Left.is(TT_ImplicitStringLiteral)) 2713 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2714 if (Line.Type == LT_ObjCMethodDecl) { 2715 if (Left.is(TT_ObjCMethodSpecifier)) 2716 return true; 2717 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right)) 2718 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a 2719 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class 2720 // method declaration. 2721 return false; 2722 } 2723 if (Line.Type == LT_ObjCProperty && 2724 (Right.is(tok::equal) || Left.is(tok::equal))) 2725 return false; 2726 2727 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 2728 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 2729 return true; 2730 if (Right.is(TT_OverloadedOperatorLParen)) 2731 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always; 2732 if (Left.is(tok::comma)) 2733 return true; 2734 if (Right.is(tok::comma)) 2735 return false; 2736 if (Right.is(TT_ObjCBlockLParen)) 2737 return true; 2738 if (Right.is(TT_CtorInitializerColon)) 2739 return Style.SpaceBeforeCtorInitializerColon; 2740 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 2741 return false; 2742 if (Right.is(TT_RangeBasedForLoopColon) && 2743 !Style.SpaceBeforeRangeBasedForLoopColon) 2744 return false; 2745 if (Right.is(tok::colon)) { 2746 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) || 2747 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 2748 return false; 2749 if (Right.is(TT_ObjCMethodExpr)) 2750 return false; 2751 if (Left.is(tok::question)) 2752 return false; 2753 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 2754 return false; 2755 if (Right.is(TT_DictLiteral)) 2756 return Style.SpacesInContainerLiterals; 2757 if (Right.is(TT_AttributeColon)) 2758 return false; 2759 return true; 2760 } 2761 if (Left.is(TT_UnaryOperator)) 2762 return Right.is(TT_BinaryOperator); 2763 2764 // If the next token is a binary operator or a selector name, we have 2765 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 2766 if (Left.is(TT_CastRParen)) 2767 return Style.SpaceAfterCStyleCast || 2768 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 2769 2770 if (Left.is(tok::greater) && Right.is(tok::greater)) { 2771 if (Style.Language == FormatStyle::LK_TextProto || 2772 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 2773 return !Style.Cpp11BracedListStyle; 2774 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 2775 (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles); 2776 } 2777 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 2778 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 2779 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 2780 return false; 2781 if (!Style.SpaceBeforeAssignmentOperators && 2782 Right.getPrecedence() == prec::Assignment) 2783 return false; 2784 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) && 2785 (Left.is(tok::identifier) || Left.is(tok::kw_this))) 2786 return false; 2787 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 2788 // Generally don't remove existing spaces between an identifier and "::". 2789 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 2790 // this turns out to be too lenient, add analysis of the identifier itself. 2791 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2792 if (Right.is(tok::coloncolon) && !Left.isOneOf(tok::l_brace, tok::comment)) 2793 return (Left.is(TT_TemplateOpener) && 2794 Style.Standard == FormatStyle::LS_Cpp03) || 2795 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 2796 tok::kw___super, TT_TemplateCloser, 2797 TT_TemplateOpener)) || 2798 (Left.is(tok ::l_paren) && Style.SpacesInParentheses); 2799 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 2800 return Style.SpacesInAngles; 2801 // Space before TT_StructuredBindingLSquare. 2802 if (Right.is(TT_StructuredBindingLSquare)) 2803 return !Left.isOneOf(tok::amp, tok::ampamp) || 2804 Style.PointerAlignment != FormatStyle::PAS_Right; 2805 // Space before & or && following a TT_StructuredBindingLSquare. 2806 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 2807 Right.isOneOf(tok::amp, tok::ampamp)) 2808 return Style.PointerAlignment != FormatStyle::PAS_Left; 2809 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 2810 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 2811 !Right.is(tok::r_paren))) 2812 return true; 2813 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) && 2814 Right.isNot(TT_FunctionTypeLParen)) 2815 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always; 2816 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 2817 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 2818 return false; 2819 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 2820 Line.startsWith(tok::hash)) 2821 return true; 2822 if (Right.is(TT_TrailingUnaryOperator)) 2823 return false; 2824 if (Left.is(TT_RegexLiteral)) 2825 return false; 2826 return spaceRequiredBetween(Line, Left, Right); 2827 } 2828 2829 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 2830 static bool isAllmanBrace(const FormatToken &Tok) { 2831 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block && 2832 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral); 2833 } 2834 2835 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 2836 const FormatToken &Right) { 2837 const FormatToken &Left = *Right.Previous; 2838 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 2839 return true; 2840 2841 if (Style.Language == FormatStyle::LK_JavaScript) { 2842 // FIXME: This might apply to other languages and token kinds. 2843 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 2844 Left.Previous->is(tok::string_literal)) 2845 return true; 2846 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 2847 Left.Previous && Left.Previous->is(tok::equal) && 2848 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 2849 tok::kw_const) && 2850 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 2851 // above. 2852 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 2853 // Object literals on the top level of a file are treated as "enum-style". 2854 // Each key/value pair is put on a separate line, instead of bin-packing. 2855 return true; 2856 if (Left.is(tok::l_brace) && Line.Level == 0 && 2857 (Line.startsWith(tok::kw_enum) || 2858 Line.startsWith(tok::kw_const, tok::kw_enum) || 2859 Line.startsWith(tok::kw_export, tok::kw_enum) || 2860 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 2861 // JavaScript top-level enum key/value pairs are put on separate lines 2862 // instead of bin-packing. 2863 return true; 2864 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 2865 !Left.Children.empty()) 2866 // Support AllowShortFunctionsOnASingleLine for JavaScript. 2867 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 2868 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 2869 (Left.NestingLevel == 0 && Line.Level == 0 && 2870 Style.AllowShortFunctionsOnASingleLine & 2871 FormatStyle::SFS_InlineOnly); 2872 } else if (Style.Language == FormatStyle::LK_Java) { 2873 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 2874 Right.Next->is(tok::string_literal)) 2875 return true; 2876 } else if (Style.Language == FormatStyle::LK_Cpp || 2877 Style.Language == FormatStyle::LK_ObjC || 2878 Style.Language == FormatStyle::LK_Proto || 2879 Style.Language == FormatStyle::LK_TableGen || 2880 Style.Language == FormatStyle::LK_TextProto) { 2881 if (Left.isStringLiteral() && Right.isStringLiteral()) 2882 return true; 2883 } 2884 2885 // If the last token before a '}', ']', or ')' is a comma or a trailing 2886 // comment, the intention is to insert a line break after it in order to make 2887 // shuffling around entries easier. Import statements, especially in 2888 // JavaScript, can be an exception to this rule. 2889 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 2890 const FormatToken *BeforeClosingBrace = nullptr; 2891 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 2892 (Style.Language == FormatStyle::LK_JavaScript && 2893 Left.is(tok::l_paren))) && 2894 Left.BlockKind != BK_Block && Left.MatchingParen) 2895 BeforeClosingBrace = Left.MatchingParen->Previous; 2896 else if (Right.MatchingParen && 2897 (Right.MatchingParen->isOneOf(tok::l_brace, 2898 TT_ArrayInitializerLSquare) || 2899 (Style.Language == FormatStyle::LK_JavaScript && 2900 Right.MatchingParen->is(tok::l_paren)))) 2901 BeforeClosingBrace = &Left; 2902 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 2903 BeforeClosingBrace->isTrailingComment())) 2904 return true; 2905 } 2906 2907 if (Right.is(tok::comment)) 2908 return Left.BlockKind != BK_BracedInit && 2909 Left.isNot(TT_CtorInitializerColon) && 2910 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 2911 if (Left.isTrailingComment()) 2912 return true; 2913 if (Right.Previous->IsUnterminatedLiteral) 2914 return true; 2915 if (Right.is(tok::lessless) && Right.Next && 2916 Right.Previous->is(tok::string_literal) && 2917 Right.Next->is(tok::string_literal)) 2918 return true; 2919 if (Right.Previous->ClosesTemplateDeclaration && 2920 Right.Previous->MatchingParen && 2921 Right.Previous->MatchingParen->NestingLevel == 0 && 2922 Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes) 2923 return true; 2924 if (Right.is(TT_CtorInitializerComma) && 2925 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 2926 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 2927 return true; 2928 if (Right.is(TT_CtorInitializerColon) && 2929 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 2930 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 2931 return true; 2932 // Break only if we have multiple inheritance. 2933 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 2934 Right.is(TT_InheritanceComma)) 2935 return true; 2936 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 2937 // Multiline raw string literals are special wrt. line breaks. The author 2938 // has made a deliberate choice and might have aligned the contents of the 2939 // string literal accordingly. Thus, we try keep existing line breaks. 2940 return Right.IsMultiline && Right.NewlinesBefore > 0; 2941 if ((Right.Previous->is(tok::l_brace) || 2942 (Right.Previous->is(tok::less) && Right.Previous->Previous && 2943 Right.Previous->Previous->is(tok::equal))) && 2944 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 2945 // Don't put enums or option definitions onto single lines in protocol 2946 // buffers. 2947 return true; 2948 } 2949 if (Right.is(TT_InlineASMBrace)) 2950 return Right.HasUnescapedNewline; 2951 if (isAllmanBrace(Left) || isAllmanBrace(Right)) 2952 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) || 2953 (Line.startsWith(tok::kw_typedef, tok::kw_enum) && 2954 Style.BraceWrapping.AfterEnum) || 2955 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 2956 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 2957 if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine) 2958 return true; 2959 2960 if ((Style.Language == FormatStyle::LK_Java || 2961 Style.Language == FormatStyle::LK_JavaScript) && 2962 Left.is(TT_LeadingJavaAnnotation) && 2963 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 2964 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 2965 return true; 2966 2967 if (Right.is(TT_ProtoExtensionLSquare)) 2968 return true; 2969 2970 // In text proto instances if a submessage contains at least 2 entries and at 2971 // least one of them is a submessage, like A { ... B { ... } ... }, 2972 // put all of the entries of A on separate lines by forcing the selector of 2973 // the submessage B to be put on a newline. 2974 // 2975 // Example: these can stay on one line: 2976 // a { scalar_1: 1 scalar_2: 2 } 2977 // a { b { key: value } } 2978 // 2979 // and these entries need to be on a new line even if putting them all in one 2980 // line is under the column limit: 2981 // a { 2982 // scalar: 1 2983 // b { key: value } 2984 // } 2985 // 2986 // We enforce this by breaking before a submessage field that has previous 2987 // siblings, *and* breaking before a field that follows a submessage field. 2988 // 2989 // Be careful to exclude the case [proto.ext] { ... } since the `]` is 2990 // the TT_SelectorName there, but we don't want to break inside the brackets. 2991 // 2992 // Another edge case is @submessage { key: value }, which is a common 2993 // substitution placeholder. In this case we want to keep `@` and `submessage` 2994 // together. 2995 // 2996 // We ensure elsewhere that extensions are always on their own line. 2997 if ((Style.Language == FormatStyle::LK_Proto || 2998 Style.Language == FormatStyle::LK_TextProto) && 2999 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) { 3000 // Keep `@submessage` together in: 3001 // @submessage { key: value } 3002 if (Right.Previous && Right.Previous->is(tok::at)) 3003 return false; 3004 // Look for the scope opener after selector in cases like: 3005 // selector { ... 3006 // selector: { ... 3007 // selector: @base { ... 3008 FormatToken *LBrace = Right.Next; 3009 if (LBrace && LBrace->is(tok::colon)) { 3010 LBrace = LBrace->Next; 3011 if (LBrace && LBrace->is(tok::at)) { 3012 LBrace = LBrace->Next; 3013 if (LBrace) 3014 LBrace = LBrace->Next; 3015 } 3016 } 3017 if (LBrace && 3018 // The scope opener is one of {, [, <: 3019 // selector { ... } 3020 // selector [ ... ] 3021 // selector < ... > 3022 // 3023 // In case of selector { ... }, the l_brace is TT_DictLiteral. 3024 // In case of an empty selector {}, the l_brace is not TT_DictLiteral, 3025 // so we check for immediately following r_brace. 3026 ((LBrace->is(tok::l_brace) && 3027 (LBrace->is(TT_DictLiteral) || 3028 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) || 3029 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) { 3030 // If Left.ParameterCount is 0, then this submessage entry is not the 3031 // first in its parent submessage, and we want to break before this entry. 3032 // If Left.ParameterCount is greater than 0, then its parent submessage 3033 // might contain 1 or more entries and we want to break before this entry 3034 // if it contains at least 2 entries. We deal with this case later by 3035 // detecting and breaking before the next entry in the parent submessage. 3036 if (Left.ParameterCount == 0) 3037 return true; 3038 // However, if this submessage is the first entry in its parent 3039 // submessage, Left.ParameterCount might be 1 in some cases. 3040 // We deal with this case later by detecting an entry 3041 // following a closing paren of this submessage. 3042 } 3043 3044 // If this is an entry immediately following a submessage, it will be 3045 // preceded by a closing paren of that submessage, like in: 3046 // left---. .---right 3047 // v v 3048 // sub: { ... } key: value 3049 // If there was a comment between `}` an `key` above, then `key` would be 3050 // put on a new line anyways. 3051 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square)) 3052 return true; 3053 } 3054 3055 // Deal with lambda arguments in C++ - we want consistent line breaks whether 3056 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced 3057 // as aggressive line breaks are placed when the lambda is not the last arg. 3058 if ((Style.Language == FormatStyle::LK_Cpp || 3059 Style.Language == FormatStyle::LK_ObjC) && 3060 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 && 3061 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) { 3062 // Multiple lambdas in the same function call force line breaks. 3063 if (Left.BlockParameterCount > 1) 3064 return true; 3065 3066 // A lambda followed by another arg forces a line break. 3067 if (!Left.Role) 3068 return false; 3069 auto Comma = Left.Role->lastComma(); 3070 if (!Comma) 3071 return false; 3072 auto Next = Comma->getNextNonComment(); 3073 if (!Next) 3074 return false; 3075 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret)) 3076 return true; 3077 } 3078 3079 return false; 3080 } 3081 3082 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 3083 const FormatToken &Right) { 3084 const FormatToken &Left = *Right.Previous; 3085 3086 // Language-specific stuff. 3087 if (Style.Language == FormatStyle::LK_Java) { 3088 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3089 Keywords.kw_implements)) 3090 return false; 3091 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3092 Keywords.kw_implements)) 3093 return true; 3094 } else if (Style.Language == FormatStyle::LK_JavaScript) { 3095 const FormatToken *NonComment = Right.getPreviousNonComment(); 3096 if (NonComment && 3097 NonComment->isOneOf( 3098 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 3099 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 3100 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 3101 Keywords.kw_readonly, Keywords.kw_abstract, Keywords.kw_get, 3102 Keywords.kw_set, Keywords.kw_async, Keywords.kw_await)) 3103 return false; // Otherwise automatic semicolon insertion would trigger. 3104 if (Right.NestingLevel == 0 && 3105 (Left.Tok.getIdentifierInfo() || 3106 Left.isOneOf(tok::r_square, tok::r_paren)) && 3107 Right.isOneOf(tok::l_square, tok::l_paren)) 3108 return false; // Otherwise automatic semicolon insertion would trigger. 3109 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace)) 3110 return false; 3111 if (Left.is(TT_JsTypeColon)) 3112 return true; 3113 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;"). 3114 if (Left.is(tok::exclaim) && Right.is(tok::colon)) 3115 return false; 3116 if (Right.is(Keywords.kw_is)) 3117 return false; 3118 if (Left.is(Keywords.kw_in)) 3119 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 3120 if (Right.is(Keywords.kw_in)) 3121 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 3122 if (Right.is(Keywords.kw_as)) 3123 return false; // must not break before as in 'x as type' casts 3124 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) { 3125 // extends and infer can appear as keywords in conditional types: 3126 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types 3127 // do not break before them, as the expressions are subject to ASI. 3128 return false; 3129 } 3130 if (Left.is(Keywords.kw_as)) 3131 return true; 3132 if (Left.is(TT_JsNonNullAssertion)) 3133 return true; 3134 if (Left.is(Keywords.kw_declare) && 3135 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 3136 Keywords.kw_function, tok::kw_class, tok::kw_enum, 3137 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 3138 Keywords.kw_let, tok::kw_const)) 3139 // See grammar for 'declare' statements at: 3140 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10 3141 return false; 3142 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 3143 Right.isOneOf(tok::identifier, tok::string_literal)) 3144 return false; // must not break in "module foo { ...}" 3145 if (Right.is(TT_TemplateString) && Right.closesScope()) 3146 return false; 3147 if (Left.is(TT_TemplateString) && Left.opensScope()) 3148 return true; 3149 } 3150 3151 if (Left.is(tok::at)) 3152 return false; 3153 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 3154 return false; 3155 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 3156 return !Right.is(tok::l_paren); 3157 if (Right.is(TT_PointerOrReference)) 3158 return Line.IsMultiVariableDeclStmt || 3159 (Style.PointerAlignment == FormatStyle::PAS_Right && 3160 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 3161 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 3162 Right.is(tok::kw_operator)) 3163 return true; 3164 if (Left.is(TT_PointerOrReference)) 3165 return false; 3166 if (Right.isTrailingComment()) 3167 // We rely on MustBreakBefore being set correctly here as we should not 3168 // change the "binding" behavior of a comment. 3169 // The first comment in a braced lists is always interpreted as belonging to 3170 // the first list element. Otherwise, it should be placed outside of the 3171 // list. 3172 return Left.BlockKind == BK_BracedInit || 3173 (Left.is(TT_CtorInitializerColon) && 3174 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 3175 if (Left.is(tok::question) && Right.is(tok::colon)) 3176 return false; 3177 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 3178 return Style.BreakBeforeTernaryOperators; 3179 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 3180 return !Style.BreakBeforeTernaryOperators; 3181 if (Left.is(TT_InheritanceColon)) 3182 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon; 3183 if (Right.is(TT_InheritanceColon)) 3184 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon; 3185 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 3186 Left.isNot(TT_SelectorName)) 3187 return true; 3188 3189 if (Right.is(tok::colon) && 3190 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 3191 return false; 3192 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 3193 if (Style.Language == FormatStyle::LK_Proto || 3194 Style.Language == FormatStyle::LK_TextProto) { 3195 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 3196 return false; 3197 // Prevent cases like: 3198 // 3199 // submessage: 3200 // { key: valueeeeeeeeeeee } 3201 // 3202 // when the snippet does not fit into one line. 3203 // Prefer: 3204 // 3205 // submessage: { 3206 // key: valueeeeeeeeeeee 3207 // } 3208 // 3209 // instead, even if it is longer by one line. 3210 // 3211 // Note that this allows allows the "{" to go over the column limit 3212 // when the column limit is just between ":" and "{", but that does 3213 // not happen too often and alternative formattings in this case are 3214 // not much better. 3215 // 3216 // The code covers the cases: 3217 // 3218 // submessage: { ... } 3219 // submessage: < ... > 3220 // repeated: [ ... ] 3221 if (((Right.is(tok::l_brace) || Right.is(tok::less)) && 3222 Right.is(TT_DictLiteral)) || 3223 Right.is(TT_ArrayInitializerLSquare)) 3224 return false; 3225 } 3226 return true; 3227 } 3228 if (Right.is(tok::r_square) && Right.MatchingParen && 3229 Right.MatchingParen->is(TT_ProtoExtensionLSquare)) 3230 return false; 3231 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 3232 Right.Next->is(TT_ObjCMethodExpr))) 3233 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 3234 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 3235 return true; 3236 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 3237 return true; 3238 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 3239 TT_OverloadedOperator)) 3240 return false; 3241 if (Left.is(TT_RangeBasedForLoopColon)) 3242 return true; 3243 if (Right.is(TT_RangeBasedForLoopColon)) 3244 return false; 3245 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 3246 return true; 3247 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 3248 Left.is(tok::kw_operator)) 3249 return false; 3250 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 3251 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 3252 return false; 3253 if (Left.is(tok::equal) && Right.is(tok::l_brace) && 3254 !Style.Cpp11BracedListStyle) 3255 return false; 3256 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen)) 3257 return false; 3258 if (Left.is(tok::l_paren) && Left.Previous && 3259 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 3260 return false; 3261 if (Right.is(TT_ImplicitStringLiteral)) 3262 return false; 3263 3264 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser)) 3265 return false; 3266 if (Right.is(tok::r_square) && Right.MatchingParen && 3267 Right.MatchingParen->is(TT_LambdaLSquare)) 3268 return false; 3269 3270 // We only break before r_brace if there was a corresponding break before 3271 // the l_brace, which is tracked by BreakBeforeClosingBrace. 3272 if (Right.is(tok::r_brace)) 3273 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block; 3274 3275 // Allow breaking after a trailing annotation, e.g. after a method 3276 // declaration. 3277 if (Left.is(TT_TrailingAnnotation)) 3278 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 3279 tok::less, tok::coloncolon); 3280 3281 if (Right.is(tok::kw___attribute) || 3282 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) 3283 return true; 3284 3285 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 3286 return true; 3287 3288 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 3289 return true; 3290 3291 if (Left.is(TT_CtorInitializerColon)) 3292 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 3293 if (Right.is(TT_CtorInitializerColon)) 3294 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 3295 if (Left.is(TT_CtorInitializerComma) && 3296 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 3297 return false; 3298 if (Right.is(TT_CtorInitializerComma) && 3299 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 3300 return true; 3301 if (Left.is(TT_InheritanceComma) && 3302 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 3303 return false; 3304 if (Right.is(TT_InheritanceComma) && 3305 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 3306 return true; 3307 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 3308 (Left.is(tok::less) && Right.is(tok::less))) 3309 return false; 3310 if (Right.is(TT_BinaryOperator) && 3311 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 3312 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 3313 Right.getPrecedence() != prec::Assignment)) 3314 return true; 3315 if (Left.is(TT_ArrayInitializerLSquare)) 3316 return true; 3317 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 3318 return true; 3319 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 3320 !Left.isOneOf(tok::arrowstar, tok::lessless) && 3321 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 3322 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 3323 Left.getPrecedence() == prec::Assignment)) 3324 return true; 3325 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 3326 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) 3327 return false; 3328 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 3329 tok::kw_class, tok::kw_struct, tok::comment) || 3330 Right.isMemberAccess() || 3331 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 3332 tok::colon, tok::l_square, tok::at) || 3333 (Left.is(tok::r_paren) && 3334 Right.isOneOf(tok::identifier, tok::kw_const)) || 3335 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 3336 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 3337 } 3338 3339 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 3340 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 3341 const FormatToken *Tok = Line.First; 3342 while (Tok) { 3343 llvm::errs() << " M=" << Tok->MustBreakBefore 3344 << " C=" << Tok->CanBreakBefore 3345 << " T=" << getTokenTypeName(Tok->Type) 3346 << " S=" << Tok->SpacesRequiredBefore 3347 << " B=" << Tok->BlockParameterCount 3348 << " BK=" << Tok->BlockKind << " P=" << Tok->SplitPenalty 3349 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 3350 << " PPK=" << Tok->PackingKind << " FakeLParens="; 3351 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 3352 llvm::errs() << Tok->FakeLParens[i] << "/"; 3353 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 3354 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo(); 3355 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 3356 if (!Tok->Next) 3357 assert(Tok == Line.Last); 3358 Tok = Tok->Next; 3359 } 3360 llvm::errs() << "----\n"; 3361 } 3362 3363 } // namespace format 3364 } // namespace clang 3365