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