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 break; 1051 default: 1052 break; 1053 } 1054 return true; 1055 } 1056 1057 void parseIncludeDirective() { 1058 if (CurrentToken && CurrentToken->is(tok::less)) { 1059 next(); 1060 while (CurrentToken) { 1061 // Mark tokens up to the trailing line comments as implicit string 1062 // literals. 1063 if (CurrentToken->isNot(tok::comment) && 1064 !CurrentToken->TokenText.startswith("//")) 1065 CurrentToken->Type = TT_ImplicitStringLiteral; 1066 next(); 1067 } 1068 } 1069 } 1070 1071 void parseWarningOrError() { 1072 next(); 1073 // We still want to format the whitespace left of the first token of the 1074 // warning or error. 1075 next(); 1076 while (CurrentToken) { 1077 CurrentToken->Type = TT_ImplicitStringLiteral; 1078 next(); 1079 } 1080 } 1081 1082 void parsePragma() { 1083 next(); // Consume "pragma". 1084 if (CurrentToken && 1085 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 1086 bool IsMark = CurrentToken->is(Keywords.kw_mark); 1087 next(); // Consume "mark". 1088 next(); // Consume first token (so we fix leading whitespace). 1089 while (CurrentToken) { 1090 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 1091 CurrentToken->Type = TT_ImplicitStringLiteral; 1092 next(); 1093 } 1094 } 1095 } 1096 1097 void parseHasInclude() { 1098 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 1099 return; 1100 next(); // '(' 1101 parseIncludeDirective(); 1102 next(); // ')' 1103 } 1104 1105 LineType parsePreprocessorDirective() { 1106 bool IsFirstToken = CurrentToken->IsFirst; 1107 LineType Type = LT_PreprocessorDirective; 1108 next(); 1109 if (!CurrentToken) 1110 return Type; 1111 1112 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) { 1113 // JavaScript files can contain shebang lines of the form: 1114 // #!/usr/bin/env node 1115 // Treat these like C++ #include directives. 1116 while (CurrentToken) { 1117 // Tokens cannot be comments here. 1118 CurrentToken->Type = TT_ImplicitStringLiteral; 1119 next(); 1120 } 1121 return LT_ImportStatement; 1122 } 1123 1124 if (CurrentToken->Tok.is(tok::numeric_constant)) { 1125 CurrentToken->SpacesRequiredBefore = 1; 1126 return Type; 1127 } 1128 // Hashes in the middle of a line can lead to any strange token 1129 // sequence. 1130 if (!CurrentToken->Tok.getIdentifierInfo()) 1131 return Type; 1132 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 1133 case tok::pp_include: 1134 case tok::pp_include_next: 1135 case tok::pp_import: 1136 next(); 1137 parseIncludeDirective(); 1138 Type = LT_ImportStatement; 1139 break; 1140 case tok::pp_error: 1141 case tok::pp_warning: 1142 parseWarningOrError(); 1143 break; 1144 case tok::pp_pragma: 1145 parsePragma(); 1146 break; 1147 case tok::pp_if: 1148 case tok::pp_elif: 1149 Contexts.back().IsExpression = true; 1150 next(); 1151 parseLine(); 1152 break; 1153 default: 1154 break; 1155 } 1156 while (CurrentToken) { 1157 FormatToken *Tok = CurrentToken; 1158 next(); 1159 if (Tok->is(tok::l_paren)) 1160 parseParens(); 1161 else if (Tok->isOneOf(Keywords.kw___has_include, 1162 Keywords.kw___has_include_next)) 1163 parseHasInclude(); 1164 } 1165 return Type; 1166 } 1167 1168 public: 1169 LineType parseLine() { 1170 if (!CurrentToken) 1171 return LT_Invalid; 1172 NonTemplateLess.clear(); 1173 if (CurrentToken->is(tok::hash)) 1174 return parsePreprocessorDirective(); 1175 1176 // Directly allow to 'import <string-literal>' to support protocol buffer 1177 // definitions (github.com/google/protobuf) or missing "#" (either way we 1178 // should not break the line). 1179 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 1180 if ((Style.Language == FormatStyle::LK_Java && 1181 CurrentToken->is(Keywords.kw_package)) || 1182 (Info && Info->getPPKeywordID() == tok::pp_import && 1183 CurrentToken->Next && 1184 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 1185 tok::kw_static))) { 1186 next(); 1187 parseIncludeDirective(); 1188 return LT_ImportStatement; 1189 } 1190 1191 // If this line starts and ends in '<' and '>', respectively, it is likely 1192 // part of "#define <a/b.h>". 1193 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 1194 parseIncludeDirective(); 1195 return LT_ImportStatement; 1196 } 1197 1198 // In .proto files, top-level options and package statements are very 1199 // similar to import statements and should not be line-wrapped. 1200 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 1201 CurrentToken->isOneOf(Keywords.kw_option, Keywords.kw_package)) { 1202 next(); 1203 if (CurrentToken && CurrentToken->is(tok::identifier)) { 1204 while (CurrentToken) 1205 next(); 1206 return LT_ImportStatement; 1207 } 1208 } 1209 1210 bool KeywordVirtualFound = false; 1211 bool ImportStatement = false; 1212 1213 // import {...} from '...'; 1214 if (Style.Language == FormatStyle::LK_JavaScript && 1215 CurrentToken->is(Keywords.kw_import)) 1216 ImportStatement = true; 1217 1218 while (CurrentToken) { 1219 if (CurrentToken->is(tok::kw_virtual)) 1220 KeywordVirtualFound = true; 1221 if (Style.Language == FormatStyle::LK_JavaScript) { 1222 // export {...} from '...'; 1223 // An export followed by "from 'some string';" is a re-export from 1224 // another module identified by a URI and is treated as a 1225 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 1226 // Just "export {...};" or "export class ..." should not be treated as 1227 // an import in this sense. 1228 if (Line.First->is(tok::kw_export) && 1229 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 1230 CurrentToken->Next->isStringLiteral()) 1231 ImportStatement = true; 1232 if (isClosureImportStatement(*CurrentToken)) 1233 ImportStatement = true; 1234 } 1235 if (!consumeToken()) 1236 return LT_Invalid; 1237 } 1238 if (KeywordVirtualFound) 1239 return LT_VirtualFunctionDecl; 1240 if (ImportStatement) 1241 return LT_ImportStatement; 1242 1243 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 1244 if (Contexts.back().FirstObjCSelectorName) 1245 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 1246 Contexts.back().LongestObjCSelectorName; 1247 return LT_ObjCMethodDecl; 1248 } 1249 1250 return LT_Other; 1251 } 1252 1253 private: 1254 bool isClosureImportStatement(const FormatToken &Tok) { 1255 // FIXME: Closure-library specific stuff should not be hard-coded but be 1256 // configurable. 1257 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 1258 Tok.Next->Next && 1259 (Tok.Next->Next->TokenText == "module" || 1260 Tok.Next->Next->TokenText == "provide" || 1261 Tok.Next->Next->TokenText == "require" || 1262 Tok.Next->Next->TokenText == "requireType" || 1263 Tok.Next->Next->TokenText == "forwardDeclare") && 1264 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 1265 } 1266 1267 void resetTokenMetadata(FormatToken *Token) { 1268 if (!Token) 1269 return; 1270 1271 // Reset token type in case we have already looked at it and then 1272 // recovered from an error (e.g. failure to find the matching >). 1273 if (!CurrentToken->isOneOf( 1274 TT_LambdaLSquare, TT_LambdaLBrace, TT_ForEachMacro, 1275 TT_TypenameMacro, TT_FunctionLBrace, TT_ImplicitStringLiteral, 1276 TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow, TT_NamespaceMacro, 1277 TT_OverloadedOperator, TT_RegexLiteral, TT_TemplateString, 1278 TT_ObjCStringLiteral)) 1279 CurrentToken->Type = TT_Unknown; 1280 CurrentToken->Role.reset(); 1281 CurrentToken->MatchingParen = nullptr; 1282 CurrentToken->FakeLParens.clear(); 1283 CurrentToken->FakeRParens = 0; 1284 } 1285 1286 void next() { 1287 if (CurrentToken) { 1288 CurrentToken->NestingLevel = Contexts.size() - 1; 1289 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1290 modifyContext(*CurrentToken); 1291 determineTokenType(*CurrentToken); 1292 CurrentToken = CurrentToken->Next; 1293 } 1294 1295 resetTokenMetadata(CurrentToken); 1296 } 1297 1298 /// A struct to hold information valid in a specific context, e.g. 1299 /// a pair of parenthesis. 1300 struct Context { 1301 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1302 bool IsExpression) 1303 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1304 IsExpression(IsExpression) {} 1305 1306 tok::TokenKind ContextKind; 1307 unsigned BindingStrength; 1308 bool IsExpression; 1309 unsigned LongestObjCSelectorName = 0; 1310 bool ColonIsForRangeExpr = false; 1311 bool ColonIsDictLiteral = false; 1312 bool ColonIsObjCMethodExpr = false; 1313 FormatToken *FirstObjCSelectorName = nullptr; 1314 FormatToken *FirstStartOfName = nullptr; 1315 bool CanBeExpression = true; 1316 bool InTemplateArgument = false; 1317 bool InCtorInitializer = false; 1318 bool InInheritanceList = false; 1319 bool CaretFound = false; 1320 bool IsForEachMacro = false; 1321 bool InCpp11AttributeSpecifier = false; 1322 bool InCSharpAttributeSpecifier = false; 1323 }; 1324 1325 /// Puts a new \c Context onto the stack \c Contexts for the lifetime 1326 /// of each instance. 1327 struct ScopedContextCreator { 1328 AnnotatingParser &P; 1329 1330 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1331 unsigned Increase) 1332 : P(P) { 1333 P.Contexts.push_back(Context(ContextKind, 1334 P.Contexts.back().BindingStrength + Increase, 1335 P.Contexts.back().IsExpression)); 1336 } 1337 1338 ~ScopedContextCreator() { P.Contexts.pop_back(); } 1339 }; 1340 1341 void modifyContext(const FormatToken &Current) { 1342 if (Current.getPrecedence() == prec::Assignment && 1343 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 1344 // Type aliases use `type X = ...;` in TypeScript and can be exported 1345 // using `export type ...`. 1346 !(Style.Language == FormatStyle::LK_JavaScript && 1347 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1348 Line.startsWith(tok::kw_export, Keywords.kw_type, 1349 tok::identifier))) && 1350 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 1351 Contexts.back().IsExpression = true; 1352 if (!Line.startsWith(TT_UnaryOperator)) { 1353 for (FormatToken *Previous = Current.Previous; 1354 Previous && Previous->Previous && 1355 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1356 Previous = Previous->Previous) { 1357 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1358 Previous = Previous->MatchingParen; 1359 if (!Previous) 1360 break; 1361 } 1362 if (Previous->opensScope()) 1363 break; 1364 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1365 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1366 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1367 Previous->Type = TT_PointerOrReference; 1368 } 1369 } 1370 } else if (Current.is(tok::lessless) && 1371 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1372 Contexts.back().IsExpression = true; 1373 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1374 Contexts.back().IsExpression = true; 1375 } else if (Current.is(TT_TrailingReturnArrow)) { 1376 Contexts.back().IsExpression = false; 1377 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1378 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1379 } else if (Current.Previous && 1380 Current.Previous->is(TT_CtorInitializerColon)) { 1381 Contexts.back().IsExpression = true; 1382 Contexts.back().InCtorInitializer = true; 1383 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1384 Contexts.back().InInheritanceList = true; 1385 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1386 for (FormatToken *Previous = Current.Previous; 1387 Previous && Previous->isOneOf(tok::star, tok::amp); 1388 Previous = Previous->Previous) 1389 Previous->Type = TT_PointerOrReference; 1390 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 1391 Contexts.back().IsExpression = false; 1392 } else if (Current.is(tok::kw_new)) { 1393 Contexts.back().CanBeExpression = false; 1394 } else if (Current.is(tok::semi) || 1395 (Current.is(tok::exclaim) && Current.Previous && 1396 !Current.Previous->is(tok::kw_operator))) { 1397 // This should be the condition or increment in a for-loop. 1398 // But not operator !() (can't use TT_OverloadedOperator here as its not 1399 // been annotated yet). 1400 Contexts.back().IsExpression = true; 1401 } 1402 } 1403 1404 static FormatToken *untilMatchingParen(FormatToken *Current) { 1405 // Used when `MatchingParen` is not yet established. 1406 int ParenLevel = 0; 1407 while (Current) { 1408 if (Current->is(tok::l_paren)) 1409 ParenLevel++; 1410 if (Current->is(tok::r_paren)) 1411 ParenLevel--; 1412 if (ParenLevel < 1) 1413 break; 1414 Current = Current->Next; 1415 } 1416 return Current; 1417 } 1418 1419 static bool isDeductionGuide(FormatToken &Current) { 1420 // Look for a deduction guide template<T> A(...) -> A<...>; 1421 if (Current.Previous && Current.Previous->is(tok::r_paren) && 1422 Current.startsSequence(tok::arrow, tok::identifier, tok::less)) { 1423 // Find the TemplateCloser. 1424 FormatToken *TemplateCloser = Current.Next->Next; 1425 int NestingLevel = 0; 1426 while (TemplateCloser) { 1427 // Skip over an expressions in parens A<(3 < 2)>; 1428 if (TemplateCloser->is(tok::l_paren)) { 1429 // No Matching Paren yet so skip to matching paren 1430 TemplateCloser = untilMatchingParen(TemplateCloser); 1431 } 1432 if (TemplateCloser->is(tok::less)) 1433 NestingLevel++; 1434 if (TemplateCloser->is(tok::greater)) 1435 NestingLevel--; 1436 if (NestingLevel < 1) 1437 break; 1438 TemplateCloser = TemplateCloser->Next; 1439 } 1440 // Assuming we have found the end of the template ensure its followed 1441 // with a semi-colon. 1442 if (TemplateCloser && TemplateCloser->Next && 1443 TemplateCloser->Next->is(tok::semi) && 1444 Current.Previous->MatchingParen) { 1445 // Determine if the identifier `A` prior to the A<..>; is the same as 1446 // prior to the A(..) 1447 FormatToken *LeadingIdentifier = 1448 Current.Previous->MatchingParen->Previous; 1449 1450 // Differentiate a deduction guide by seeing the 1451 // > of the template prior to the leading identifier. 1452 if (LeadingIdentifier) { 1453 FormatToken *PriorLeadingIdentifier = LeadingIdentifier->Previous; 1454 // Skip back past explicit decoration 1455 if (PriorLeadingIdentifier && 1456 PriorLeadingIdentifier->is(tok::kw_explicit)) 1457 PriorLeadingIdentifier = PriorLeadingIdentifier->Previous; 1458 1459 return (PriorLeadingIdentifier && 1460 PriorLeadingIdentifier->is(TT_TemplateCloser) && 1461 LeadingIdentifier->TokenText == Current.Next->TokenText); 1462 } 1463 } 1464 } 1465 return false; 1466 } 1467 1468 void determineTokenType(FormatToken &Current) { 1469 if (!Current.is(TT_Unknown)) 1470 // The token type is already known. 1471 return; 1472 1473 if (Style.isCSharp() && CurrentToken->is(tok::question)) { 1474 if (CurrentToken->TokenText == "??") { 1475 Current.Type = TT_CSharpNullCoalescing; 1476 return; 1477 } 1478 if (CurrentToken->TokenText == "?.") { 1479 Current.Type = TT_CSharpNullConditional; 1480 return; 1481 } 1482 if (CurrentToken->TokenText == "?[") { 1483 Current.Type = TT_CSharpNullConditionalLSquare; 1484 return; 1485 } 1486 } 1487 1488 if (Style.Language == FormatStyle::LK_JavaScript) { 1489 if (Current.is(tok::exclaim)) { 1490 if (Current.Previous && 1491 (Current.Previous->isOneOf(tok::identifier, tok::kw_namespace, 1492 tok::r_paren, tok::r_square, 1493 tok::r_brace) || 1494 Current.Previous->Tok.isLiteral())) { 1495 Current.Type = TT_JsNonNullAssertion; 1496 return; 1497 } 1498 if (Current.Next && 1499 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1500 Current.Type = TT_JsNonNullAssertion; 1501 return; 1502 } 1503 } 1504 } 1505 1506 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1507 // function declaration have been found. In this case, 'Current' is a 1508 // trailing token of this declaration and thus cannot be a name. 1509 if (Current.is(Keywords.kw_instanceof)) { 1510 Current.Type = TT_BinaryOperator; 1511 } else if (isStartOfName(Current) && 1512 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1513 Contexts.back().FirstStartOfName = &Current; 1514 Current.Type = TT_StartOfName; 1515 } else if (Current.is(tok::semi)) { 1516 // Reset FirstStartOfName after finding a semicolon so that a for loop 1517 // with multiple increment statements is not confused with a for loop 1518 // having multiple variable declarations. 1519 Contexts.back().FirstStartOfName = nullptr; 1520 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1521 AutoFound = true; 1522 } else if (Current.is(tok::arrow) && 1523 Style.Language == FormatStyle::LK_Java) { 1524 Current.Type = TT_LambdaArrow; 1525 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1526 Current.NestingLevel == 0 && 1527 !Current.Previous->is(tok::kw_operator)) { 1528 // not auto operator->() -> xxx; 1529 Current.Type = TT_TrailingReturnArrow; 1530 1531 } else if (isDeductionGuide(Current)) { 1532 // Deduction guides trailing arrow " A(...) -> A<T>;". 1533 Current.Type = TT_TrailingReturnArrow; 1534 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1535 Current.Type = determineStarAmpUsage(Current, 1536 Contexts.back().CanBeExpression && 1537 Contexts.back().IsExpression, 1538 Contexts.back().InTemplateArgument); 1539 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1540 Current.Type = determinePlusMinusCaretUsage(Current); 1541 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1542 Contexts.back().CaretFound = true; 1543 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1544 Current.Type = determineIncrementUsage(Current); 1545 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1546 Current.Type = TT_UnaryOperator; 1547 } else if (Current.is(tok::question)) { 1548 if (Style.Language == FormatStyle::LK_JavaScript && 1549 Line.MustBeDeclaration && !Contexts.back().IsExpression) { 1550 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1551 // on the interface, not a ternary expression. 1552 Current.Type = TT_JsTypeOptionalQuestion; 1553 } else { 1554 Current.Type = TT_ConditionalExpr; 1555 } 1556 } else if (Current.isBinaryOperator() && 1557 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1558 (!Current.is(tok::greater) && 1559 Style.Language != FormatStyle::LK_TextProto)) { 1560 Current.Type = TT_BinaryOperator; 1561 } else if (Current.is(tok::comment)) { 1562 if (Current.TokenText.startswith("/*")) { 1563 if (Current.TokenText.endswith("*/")) 1564 Current.Type = TT_BlockComment; 1565 else 1566 // The lexer has for some reason determined a comment here. But we 1567 // cannot really handle it, if it isn't properly terminated. 1568 Current.Tok.setKind(tok::unknown); 1569 } else { 1570 Current.Type = TT_LineComment; 1571 } 1572 } else if (Current.is(tok::r_paren)) { 1573 if (rParenEndsCast(Current)) 1574 Current.Type = TT_CastRParen; 1575 if (Current.MatchingParen && Current.Next && 1576 !Current.Next->isBinaryOperator() && 1577 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1578 tok::comma, tok::period, tok::arrow, 1579 tok::coloncolon)) 1580 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1581 // Make sure this isn't the return type of an Obj-C block declaration 1582 if (AfterParen->Tok.isNot(tok::caret)) { 1583 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1584 if (BeforeParen->is(tok::identifier) && 1585 !BeforeParen->is(TT_TypenameMacro) && 1586 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1587 (!BeforeParen->Previous || 1588 BeforeParen->Previous->ClosesTemplateDeclaration)) 1589 Current.Type = TT_FunctionAnnotationRParen; 1590 } 1591 } 1592 } else if (Current.is(tok::at) && Current.Next && 1593 Style.Language != FormatStyle::LK_JavaScript && 1594 Style.Language != FormatStyle::LK_Java) { 1595 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1596 // marks declarations and properties that need special formatting. 1597 switch (Current.Next->Tok.getObjCKeywordID()) { 1598 case tok::objc_interface: 1599 case tok::objc_implementation: 1600 case tok::objc_protocol: 1601 Current.Type = TT_ObjCDecl; 1602 break; 1603 case tok::objc_property: 1604 Current.Type = TT_ObjCProperty; 1605 break; 1606 default: 1607 break; 1608 } 1609 } else if (Current.is(tok::period)) { 1610 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1611 if (PreviousNoComment && 1612 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1613 Current.Type = TT_DesignatedInitializerPeriod; 1614 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1615 Current.Previous->isOneOf(TT_JavaAnnotation, 1616 TT_LeadingJavaAnnotation)) { 1617 Current.Type = Current.Previous->Type; 1618 } 1619 } else if (canBeObjCSelectorComponent(Current) && 1620 // FIXME(bug 36976): ObjC return types shouldn't use 1621 // TT_CastRParen. 1622 Current.Previous && Current.Previous->is(TT_CastRParen) && 1623 Current.Previous->MatchingParen && 1624 Current.Previous->MatchingParen->Previous && 1625 Current.Previous->MatchingParen->Previous->is( 1626 TT_ObjCMethodSpecifier)) { 1627 // This is the first part of an Objective-C selector name. (If there's no 1628 // colon after this, this is the only place which annotates the identifier 1629 // as a selector.) 1630 Current.Type = TT_SelectorName; 1631 } else if (Current.isOneOf(tok::identifier, tok::kw_const, 1632 tok::kw_noexcept) && 1633 Current.Previous && 1634 !Current.Previous->isOneOf(tok::equal, tok::at) && 1635 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1636 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1637 // function declaration have been found. 1638 Current.Type = TT_TrailingAnnotation; 1639 } else if ((Style.Language == FormatStyle::LK_Java || 1640 Style.Language == FormatStyle::LK_JavaScript) && 1641 Current.Previous) { 1642 if (Current.Previous->is(tok::at) && 1643 Current.isNot(Keywords.kw_interface)) { 1644 const FormatToken &AtToken = *Current.Previous; 1645 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1646 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1647 Current.Type = TT_LeadingJavaAnnotation; 1648 else 1649 Current.Type = TT_JavaAnnotation; 1650 } else if (Current.Previous->is(tok::period) && 1651 Current.Previous->isOneOf(TT_JavaAnnotation, 1652 TT_LeadingJavaAnnotation)) { 1653 Current.Type = Current.Previous->Type; 1654 } 1655 } 1656 } 1657 1658 /// Take a guess at whether \p Tok starts a name of a function or 1659 /// variable declaration. 1660 /// 1661 /// This is a heuristic based on whether \p Tok is an identifier following 1662 /// something that is likely a type. 1663 bool isStartOfName(const FormatToken &Tok) { 1664 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1665 return false; 1666 1667 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1668 Keywords.kw_as)) 1669 return false; 1670 if (Style.Language == FormatStyle::LK_JavaScript && 1671 Tok.Previous->is(Keywords.kw_in)) 1672 return false; 1673 1674 // Skip "const" as it does not have an influence on whether this is a name. 1675 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1676 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1677 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1678 1679 if (!PreviousNotConst) 1680 return false; 1681 1682 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1683 PreviousNotConst->Previous && 1684 PreviousNotConst->Previous->is(tok::hash); 1685 1686 if (PreviousNotConst->is(TT_TemplateCloser)) 1687 return PreviousNotConst && PreviousNotConst->MatchingParen && 1688 PreviousNotConst->MatchingParen->Previous && 1689 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1690 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1691 1692 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen && 1693 PreviousNotConst->MatchingParen->Previous && 1694 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype)) 1695 return true; 1696 1697 return (!IsPPKeyword && 1698 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) || 1699 PreviousNotConst->is(TT_PointerOrReference) || 1700 PreviousNotConst->isSimpleTypeSpecifier(); 1701 } 1702 1703 /// Determine whether ')' is ending a cast. 1704 bool rParenEndsCast(const FormatToken &Tok) { 1705 // C-style casts are only used in C++, C# and Java. 1706 if (!Style.isCSharp() && !Style.isCpp() && 1707 Style.Language != FormatStyle::LK_Java) 1708 return false; 1709 1710 // Empty parens aren't casts and there are no casts at the end of the line. 1711 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1712 return false; 1713 1714 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1715 if (LeftOfParens) { 1716 // If there is a closing parenthesis left of the current parentheses, 1717 // look past it as these might be chained casts. 1718 if (LeftOfParens->is(tok::r_paren)) { 1719 if (!LeftOfParens->MatchingParen || 1720 !LeftOfParens->MatchingParen->Previous) 1721 return false; 1722 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1723 } 1724 1725 // If there is an identifier (or with a few exceptions a keyword) right 1726 // before the parentheses, this is unlikely to be a cast. 1727 if (LeftOfParens->Tok.getIdentifierInfo() && 1728 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1729 tok::kw_delete)) 1730 return false; 1731 1732 // Certain other tokens right before the parentheses are also signals that 1733 // this cannot be a cast. 1734 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1735 TT_TemplateCloser, tok::ellipsis)) 1736 return false; 1737 } 1738 1739 if (Tok.Next->is(tok::question)) 1740 return false; 1741 1742 // Functions which end with decorations like volatile, noexcept are unlikely 1743 // to be casts. 1744 if (Tok.Next->isOneOf(tok::kw_noexcept, tok::kw_volatile, tok::kw_const, 1745 tok::kw_throw, tok::arrow, Keywords.kw_override, 1746 Keywords.kw_final) || 1747 isCpp11AttributeSpecifier(*Tok.Next)) 1748 return false; 1749 1750 // As Java has no function types, a "(" after the ")" likely means that this 1751 // is a cast. 1752 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1753 return true; 1754 1755 // If a (non-string) literal follows, this is likely a cast. 1756 if (Tok.Next->isNot(tok::string_literal) && 1757 (Tok.Next->Tok.isLiteral() || 1758 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1759 return true; 1760 1761 // Heuristically try to determine whether the parentheses contain a type. 1762 bool ParensAreType = 1763 !Tok.Previous || 1764 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) || 1765 Tok.Previous->isSimpleTypeSpecifier(); 1766 bool ParensCouldEndDecl = 1767 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1768 if (ParensAreType && !ParensCouldEndDecl) 1769 return true; 1770 1771 // At this point, we heuristically assume that there are no casts at the 1772 // start of the line. We assume that we have found most cases where there 1773 // are by the logic above, e.g. "(void)x;". 1774 if (!LeftOfParens) 1775 return false; 1776 1777 // Certain token types inside the parentheses mean that this can't be a 1778 // cast. 1779 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 1780 Token = Token->Next) 1781 if (Token->is(TT_BinaryOperator)) 1782 return false; 1783 1784 // If the following token is an identifier or 'this', this is a cast. All 1785 // cases where this can be something else are handled above. 1786 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1787 return true; 1788 1789 if (!Tok.Next->Next) 1790 return false; 1791 1792 // If the next token after the parenthesis is a unary operator, assume 1793 // that this is cast, unless there are unexpected tokens inside the 1794 // parenthesis. 1795 bool NextIsUnary = 1796 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 1797 if (!NextIsUnary || Tok.Next->is(tok::plus) || 1798 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 1799 return false; 1800 // Search for unexpected tokens. 1801 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 1802 Prev = Prev->Previous) { 1803 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 1804 return false; 1805 } 1806 return true; 1807 } 1808 1809 /// Return the type of the given token assuming it is * or &. 1810 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 1811 bool InTemplateArgument) { 1812 if (Style.Language == FormatStyle::LK_JavaScript) 1813 return TT_BinaryOperator; 1814 1815 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1816 if (!PrevToken) 1817 return TT_UnaryOperator; 1818 1819 const FormatToken *NextToken = Tok.getNextNonComment(); 1820 if (!NextToken || 1821 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_const, 1822 tok::kw_noexcept) || 1823 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 1824 return TT_PointerOrReference; 1825 1826 if (PrevToken->is(tok::coloncolon)) 1827 return TT_PointerOrReference; 1828 1829 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 1830 tok::comma, tok::semi, tok::kw_return, tok::colon, 1831 tok::equal, tok::kw_delete, tok::kw_sizeof, 1832 tok::kw_throw) || 1833 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 1834 TT_UnaryOperator, TT_CastRParen)) 1835 return TT_UnaryOperator; 1836 1837 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 1838 return TT_PointerOrReference; 1839 if (NextToken->is(tok::kw_operator) && !IsExpression) 1840 return TT_PointerOrReference; 1841 if (NextToken->isOneOf(tok::comma, tok::semi)) 1842 return TT_PointerOrReference; 1843 1844 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen) { 1845 FormatToken *TokenBeforeMatchingParen = 1846 PrevToken->MatchingParen->getPreviousNonComment(); 1847 if (TokenBeforeMatchingParen && 1848 TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype, 1849 TT_TypenameMacro)) 1850 return TT_PointerOrReference; 1851 } 1852 1853 if (PrevToken->Tok.isLiteral() || 1854 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 1855 tok::kw_false, tok::r_brace) || 1856 NextToken->Tok.isLiteral() || 1857 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 1858 NextToken->isUnaryOperator() || 1859 // If we know we're in a template argument, there are no named 1860 // declarations. Thus, having an identifier on the right-hand side 1861 // indicates a binary operator. 1862 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 1863 return TT_BinaryOperator; 1864 1865 // "&&(" is quite unlikely to be two successive unary "&". 1866 if (Tok.is(tok::ampamp) && NextToken->is(tok::l_paren)) 1867 return TT_BinaryOperator; 1868 1869 // This catches some cases where evaluation order is used as control flow: 1870 // aaa && aaa->f(); 1871 if (NextToken->Tok.isAnyIdentifier()) { 1872 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 1873 if (NextNextToken && NextNextToken->is(tok::arrow)) 1874 return TT_BinaryOperator; 1875 } 1876 1877 // It is very unlikely that we are going to find a pointer or reference type 1878 // definition on the RHS of an assignment. 1879 if (IsExpression && !Contexts.back().CaretFound) 1880 return TT_BinaryOperator; 1881 1882 return TT_PointerOrReference; 1883 } 1884 1885 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 1886 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1887 if (!PrevToken) 1888 return TT_UnaryOperator; 1889 1890 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 1891 // This must be a sequence of leading unary operators. 1892 return TT_UnaryOperator; 1893 1894 // Use heuristics to recognize unary operators. 1895 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 1896 tok::question, tok::colon, tok::kw_return, 1897 tok::kw_case, tok::at, tok::l_brace, tok::kw_throw, 1898 tok::kw_co_return, tok::kw_co_yield)) 1899 return TT_UnaryOperator; 1900 1901 // There can't be two consecutive binary operators. 1902 if (PrevToken->is(TT_BinaryOperator)) 1903 return TT_UnaryOperator; 1904 1905 // Fall back to marking the token as binary operator. 1906 return TT_BinaryOperator; 1907 } 1908 1909 /// Determine whether ++/-- are pre- or post-increments/-decrements. 1910 TokenType determineIncrementUsage(const FormatToken &Tok) { 1911 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1912 if (!PrevToken || PrevToken->is(TT_CastRParen)) 1913 return TT_UnaryOperator; 1914 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 1915 return TT_TrailingUnaryOperator; 1916 1917 return TT_UnaryOperator; 1918 } 1919 1920 SmallVector<Context, 8> Contexts; 1921 1922 const FormatStyle &Style; 1923 AnnotatedLine &Line; 1924 FormatToken *CurrentToken; 1925 bool AutoFound; 1926 const AdditionalKeywords &Keywords; 1927 1928 // Set of "<" tokens that do not open a template parameter list. If parseAngle 1929 // determines that a specific token can't be a template opener, it will make 1930 // same decision irrespective of the decisions for tokens leading up to it. 1931 // Store this information to prevent this from causing exponential runtime. 1932 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 1933 }; 1934 1935 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 1936 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 1937 1938 /// Parses binary expressions by inserting fake parenthesis based on 1939 /// operator precedence. 1940 class ExpressionParser { 1941 public: 1942 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 1943 AnnotatedLine &Line) 1944 : Style(Style), Keywords(Keywords), Current(Line.First) {} 1945 1946 /// Parse expressions with the given operator precedence. 1947 void parse(int Precedence = 0) { 1948 // Skip 'return' and ObjC selector colons as they are not part of a binary 1949 // expression. 1950 while (Current && (Current->is(tok::kw_return) || 1951 (Current->is(tok::colon) && 1952 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 1953 next(); 1954 1955 if (!Current || Precedence > PrecedenceArrowAndPeriod) 1956 return; 1957 1958 // Conditional expressions need to be parsed separately for proper nesting. 1959 if (Precedence == prec::Conditional) { 1960 parseConditionalExpr(); 1961 return; 1962 } 1963 1964 // Parse unary operators, which all have a higher precedence than binary 1965 // operators. 1966 if (Precedence == PrecedenceUnaryOperator) { 1967 parseUnaryOperator(); 1968 return; 1969 } 1970 1971 FormatToken *Start = Current; 1972 FormatToken *LatestOperator = nullptr; 1973 unsigned OperatorIndex = 0; 1974 1975 while (Current) { 1976 // Consume operators with higher precedence. 1977 parse(Precedence + 1); 1978 1979 int CurrentPrecedence = getCurrentPrecedence(); 1980 1981 if (Current && Current->is(TT_SelectorName) && 1982 Precedence == CurrentPrecedence) { 1983 if (LatestOperator) 1984 addFakeParenthesis(Start, prec::Level(Precedence)); 1985 Start = Current; 1986 } 1987 1988 // At the end of the line or when an operator with higher precedence is 1989 // found, insert fake parenthesis and return. 1990 if (!Current || 1991 (Current->closesScope() && 1992 (Current->MatchingParen || Current->is(TT_TemplateString))) || 1993 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 1994 (CurrentPrecedence == prec::Conditional && 1995 Precedence == prec::Assignment && Current->is(tok::colon))) { 1996 break; 1997 } 1998 1999 // Consume scopes: (), [], <> and {} 2000 if (Current->opensScope()) { 2001 // In fragment of a JavaScript template string can look like '}..${' and 2002 // thus close a scope and open a new one at the same time. 2003 while (Current && (!Current->closesScope() || Current->opensScope())) { 2004 next(); 2005 parse(); 2006 } 2007 next(); 2008 } else { 2009 // Operator found. 2010 if (CurrentPrecedence == Precedence) { 2011 if (LatestOperator) 2012 LatestOperator->NextOperator = Current; 2013 LatestOperator = Current; 2014 Current->OperatorIndex = OperatorIndex; 2015 ++OperatorIndex; 2016 } 2017 next(/*SkipPastLeadingComments=*/Precedence > 0); 2018 } 2019 } 2020 2021 if (LatestOperator && (Current || Precedence > 0)) { 2022 // LatestOperator->LastOperator = true; 2023 if (Precedence == PrecedenceArrowAndPeriod) { 2024 // Call expressions don't have a binary operator precedence. 2025 addFakeParenthesis(Start, prec::Unknown); 2026 } else { 2027 addFakeParenthesis(Start, prec::Level(Precedence)); 2028 } 2029 } 2030 } 2031 2032 private: 2033 /// Gets the precedence (+1) of the given token for binary operators 2034 /// and other tokens that we treat like binary operators. 2035 int getCurrentPrecedence() { 2036 if (Current) { 2037 const FormatToken *NextNonComment = Current->getNextNonComment(); 2038 if (Current->is(TT_ConditionalExpr)) 2039 return prec::Conditional; 2040 if (NextNonComment && Current->is(TT_SelectorName) && 2041 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 2042 ((Style.Language == FormatStyle::LK_Proto || 2043 Style.Language == FormatStyle::LK_TextProto) && 2044 NextNonComment->is(tok::less)))) 2045 return prec::Assignment; 2046 if (Current->is(TT_JsComputedPropertyName)) 2047 return prec::Assignment; 2048 if (Current->is(TT_LambdaArrow)) 2049 return prec::Comma; 2050 if (Current->is(TT_JsFatArrow)) 2051 return prec::Assignment; 2052 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 2053 (Current->is(tok::comment) && NextNonComment && 2054 NextNonComment->is(TT_SelectorName))) 2055 return 0; 2056 if (Current->is(TT_RangeBasedForLoopColon)) 2057 return prec::Comma; 2058 if ((Style.Language == FormatStyle::LK_Java || 2059 Style.Language == FormatStyle::LK_JavaScript) && 2060 Current->is(Keywords.kw_instanceof)) 2061 return prec::Relational; 2062 if (Style.Language == FormatStyle::LK_JavaScript && 2063 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 2064 return prec::Relational; 2065 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 2066 return Current->getPrecedence(); 2067 if (Current->isOneOf(tok::period, tok::arrow)) 2068 return PrecedenceArrowAndPeriod; 2069 if ((Style.Language == FormatStyle::LK_Java || 2070 Style.Language == FormatStyle::LK_JavaScript) && 2071 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 2072 Keywords.kw_throws)) 2073 return 0; 2074 } 2075 return -1; 2076 } 2077 2078 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 2079 Start->FakeLParens.push_back(Precedence); 2080 if (Precedence > prec::Unknown) 2081 Start->StartsBinaryExpression = true; 2082 if (Current) { 2083 FormatToken *Previous = Current->Previous; 2084 while (Previous->is(tok::comment) && Previous->Previous) 2085 Previous = Previous->Previous; 2086 ++Previous->FakeRParens; 2087 if (Precedence > prec::Unknown) 2088 Previous->EndsBinaryExpression = true; 2089 } 2090 } 2091 2092 /// Parse unary operator expressions and surround them with fake 2093 /// parentheses if appropriate. 2094 void parseUnaryOperator() { 2095 llvm::SmallVector<FormatToken *, 2> Tokens; 2096 while (Current && Current->is(TT_UnaryOperator)) { 2097 Tokens.push_back(Current); 2098 next(); 2099 } 2100 parse(PrecedenceArrowAndPeriod); 2101 for (FormatToken *Token : llvm::reverse(Tokens)) 2102 // The actual precedence doesn't matter. 2103 addFakeParenthesis(Token, prec::Unknown); 2104 } 2105 2106 void parseConditionalExpr() { 2107 while (Current && Current->isTrailingComment()) { 2108 next(); 2109 } 2110 FormatToken *Start = Current; 2111 parse(prec::LogicalOr); 2112 if (!Current || !Current->is(tok::question)) 2113 return; 2114 next(); 2115 parse(prec::Assignment); 2116 if (!Current || Current->isNot(TT_ConditionalExpr)) 2117 return; 2118 next(); 2119 parse(prec::Assignment); 2120 addFakeParenthesis(Start, prec::Conditional); 2121 } 2122 2123 void next(bool SkipPastLeadingComments = true) { 2124 if (Current) 2125 Current = Current->Next; 2126 while (Current && 2127 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 2128 Current->isTrailingComment()) 2129 Current = Current->Next; 2130 } 2131 2132 const FormatStyle &Style; 2133 const AdditionalKeywords &Keywords; 2134 FormatToken *Current; 2135 }; 2136 2137 } // end anonymous namespace 2138 2139 void TokenAnnotator::setCommentLineLevels( 2140 SmallVectorImpl<AnnotatedLine *> &Lines) { 2141 const AnnotatedLine *NextNonCommentLine = nullptr; 2142 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(), 2143 E = Lines.rend(); 2144 I != E; ++I) { 2145 bool CommentLine = true; 2146 for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) { 2147 if (!Tok->is(tok::comment)) { 2148 CommentLine = false; 2149 break; 2150 } 2151 } 2152 2153 // If the comment is currently aligned with the line immediately following 2154 // it, that's probably intentional and we should keep it. 2155 if (NextNonCommentLine && CommentLine && 2156 NextNonCommentLine->First->NewlinesBefore <= 1 && 2157 NextNonCommentLine->First->OriginalColumn == 2158 (*I)->First->OriginalColumn) { 2159 // Align comments for preprocessor lines with the # in column 0 if 2160 // preprocessor lines are not indented. Otherwise, align with the next 2161 // line. 2162 (*I)->Level = 2163 (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 2164 (NextNonCommentLine->Type == LT_PreprocessorDirective || 2165 NextNonCommentLine->Type == LT_ImportStatement)) 2166 ? 0 2167 : NextNonCommentLine->Level; 2168 } else { 2169 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr; 2170 } 2171 2172 setCommentLineLevels((*I)->Children); 2173 } 2174 } 2175 2176 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 2177 unsigned Result = 0; 2178 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 2179 Result = std::max(Result, Tok->NestingLevel); 2180 return Result; 2181 } 2182 2183 void TokenAnnotator::annotate(AnnotatedLine &Line) { 2184 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2185 E = Line.Children.end(); 2186 I != E; ++I) { 2187 annotate(**I); 2188 } 2189 AnnotatingParser Parser(Style, Line, Keywords); 2190 Line.Type = Parser.parseLine(); 2191 2192 // With very deep nesting, ExpressionParser uses lots of stack and the 2193 // formatting algorithm is very slow. We're not going to do a good job here 2194 // anyway - it's probably generated code being formatted by mistake. 2195 // Just skip the whole line. 2196 if (maxNestingDepth(Line) > 50) 2197 Line.Type = LT_Invalid; 2198 2199 if (Line.Type == LT_Invalid) 2200 return; 2201 2202 ExpressionParser ExprParser(Style, Keywords, Line); 2203 ExprParser.parse(); 2204 2205 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2206 Line.Type = LT_ObjCMethodDecl; 2207 else if (Line.startsWith(TT_ObjCDecl)) 2208 Line.Type = LT_ObjCDecl; 2209 else if (Line.startsWith(TT_ObjCProperty)) 2210 Line.Type = LT_ObjCProperty; 2211 2212 Line.First->SpacesRequiredBefore = 1; 2213 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 2214 } 2215 2216 // This function heuristically determines whether 'Current' starts the name of a 2217 // function declaration. 2218 static bool isFunctionDeclarationName(const FormatToken &Current, 2219 const AnnotatedLine &Line) { 2220 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 2221 for (; Next; Next = Next->Next) { 2222 if (Next->is(TT_OverloadedOperatorLParen)) 2223 return Next; 2224 if (Next->is(TT_OverloadedOperator)) 2225 continue; 2226 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 2227 // For 'new[]' and 'delete[]'. 2228 if (Next->Next && 2229 Next->Next->startsSequence(tok::l_square, tok::r_square)) 2230 Next = Next->Next->Next; 2231 continue; 2232 } 2233 if (Next->startsSequence(tok::l_square, tok::r_square)) { 2234 // For operator[](). 2235 Next = Next->Next; 2236 continue; 2237 } 2238 if ((Next->isSimpleTypeSpecifier() || Next->is(tok::identifier)) && 2239 Next->Next && Next->Next->isOneOf(tok::star, tok::amp, tok::ampamp)) { 2240 // For operator void*(), operator char*(), operator Foo*(). 2241 Next = Next->Next; 2242 continue; 2243 } 2244 2245 break; 2246 } 2247 return nullptr; 2248 }; 2249 2250 // Find parentheses of parameter list. 2251 const FormatToken *Next = Current.Next; 2252 if (Current.is(tok::kw_operator)) { 2253 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 2254 return false; 2255 Next = skipOperatorName(Next); 2256 } else { 2257 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 2258 return false; 2259 for (; Next; Next = Next->Next) { 2260 if (Next->is(TT_TemplateOpener)) { 2261 Next = Next->MatchingParen; 2262 } else if (Next->is(tok::coloncolon)) { 2263 Next = Next->Next; 2264 if (!Next) 2265 return false; 2266 if (Next->is(tok::kw_operator)) { 2267 Next = skipOperatorName(Next->Next); 2268 break; 2269 } 2270 if (!Next->is(tok::identifier)) 2271 return false; 2272 } else if (Next->is(tok::l_paren)) { 2273 break; 2274 } else { 2275 return false; 2276 } 2277 } 2278 } 2279 2280 // Check whether parameter list can belong to a function declaration. 2281 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 2282 return false; 2283 // If the lines ends with "{", this is likely an function definition. 2284 if (Line.Last->is(tok::l_brace)) 2285 return true; 2286 if (Next->Next == Next->MatchingParen) 2287 return true; // Empty parentheses. 2288 // If there is an &/&& after the r_paren, this is likely a function. 2289 if (Next->MatchingParen->Next && 2290 Next->MatchingParen->Next->is(TT_PointerOrReference)) 2291 return true; 2292 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 2293 Tok = Tok->Next) { 2294 if (Tok->isOneOf(tok::l_paren, TT_TemplateOpener) && Tok->MatchingParen) { 2295 Tok = Tok->MatchingParen; 2296 continue; 2297 } 2298 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 2299 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 2300 return true; 2301 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 2302 Tok->Tok.isLiteral()) 2303 return false; 2304 } 2305 return false; 2306 } 2307 2308 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 2309 assert(Line.MightBeFunctionDecl); 2310 2311 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 2312 Style.AlwaysBreakAfterReturnType == 2313 FormatStyle::RTBS_TopLevelDefinitions) && 2314 Line.Level > 0) 2315 return false; 2316 2317 switch (Style.AlwaysBreakAfterReturnType) { 2318 case FormatStyle::RTBS_None: 2319 return false; 2320 case FormatStyle::RTBS_All: 2321 case FormatStyle::RTBS_TopLevel: 2322 return true; 2323 case FormatStyle::RTBS_AllDefinitions: 2324 case FormatStyle::RTBS_TopLevelDefinitions: 2325 return Line.mightBeFunctionDefinition(); 2326 } 2327 2328 return false; 2329 } 2330 2331 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 2332 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2333 E = Line.Children.end(); 2334 I != E; ++I) { 2335 calculateFormattingInformation(**I); 2336 } 2337 2338 Line.First->TotalLength = 2339 Line.First->IsMultiline ? Style.ColumnLimit 2340 : Line.FirstStartColumn + Line.First->ColumnWidth; 2341 FormatToken *Current = Line.First->Next; 2342 bool InFunctionDecl = Line.MightBeFunctionDecl; 2343 while (Current) { 2344 if (isFunctionDeclarationName(*Current, Line)) 2345 Current->Type = TT_FunctionDeclarationName; 2346 if (Current->is(TT_LineComment)) { 2347 if (Current->Previous->BlockKind == BK_BracedInit && 2348 Current->Previous->opensScope()) 2349 Current->SpacesRequiredBefore = 2350 (Style.Cpp11BracedListStyle && !Style.SpacesInParentheses) ? 0 : 1; 2351 else 2352 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2353 2354 // If we find a trailing comment, iterate backwards to determine whether 2355 // it seems to relate to a specific parameter. If so, break before that 2356 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2357 // to the previous line in: 2358 // SomeFunction(a, 2359 // b, // comment 2360 // c); 2361 if (!Current->HasUnescapedNewline) { 2362 for (FormatToken *Parameter = Current->Previous; Parameter; 2363 Parameter = Parameter->Previous) { 2364 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 2365 break; 2366 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 2367 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 2368 Parameter->HasUnescapedNewline) 2369 Parameter->MustBreakBefore = true; 2370 break; 2371 } 2372 } 2373 } 2374 } else if (Current->SpacesRequiredBefore == 0 && 2375 spaceRequiredBefore(Line, *Current)) { 2376 Current->SpacesRequiredBefore = 1; 2377 } 2378 2379 Current->MustBreakBefore = 2380 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 2381 2382 if (!Current->MustBreakBefore && InFunctionDecl && 2383 Current->is(TT_FunctionDeclarationName)) 2384 Current->MustBreakBefore = mustBreakForReturnType(Line); 2385 2386 Current->CanBreakBefore = 2387 Current->MustBreakBefore || canBreakBefore(Line, *Current); 2388 unsigned ChildSize = 0; 2389 if (Current->Previous->Children.size() == 1) { 2390 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last; 2391 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 2392 : LastOfChild.TotalLength + 1; 2393 } 2394 const FormatToken *Prev = Current->Previous; 2395 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 2396 (Prev->Children.size() == 1 && 2397 Prev->Children[0]->First->MustBreakBefore) || 2398 Current->IsMultiline) 2399 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 2400 else 2401 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 2402 ChildSize + Current->SpacesRequiredBefore; 2403 2404 if (Current->is(TT_CtorInitializerColon)) 2405 InFunctionDecl = false; 2406 2407 // FIXME: Only calculate this if CanBreakBefore is true once static 2408 // initializers etc. are sorted out. 2409 // FIXME: Move magic numbers to a better place. 2410 2411 // Reduce penalty for aligning ObjC method arguments using the colon 2412 // alignment as this is the canonical way (still prefer fitting everything 2413 // into one line if possible). Trying to fit a whole expression into one 2414 // line should not force other line breaks (e.g. when ObjC method 2415 // expression is a part of other expression). 2416 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl); 2417 if (Style.Language == FormatStyle::LK_ObjC && 2418 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) { 2419 if (Current->ParameterIndex == 1) 2420 Current->SplitPenalty += 5 * Current->BindingStrength; 2421 } else { 2422 Current->SplitPenalty += 20 * Current->BindingStrength; 2423 } 2424 2425 Current = Current->Next; 2426 } 2427 2428 calculateUnbreakableTailLengths(Line); 2429 unsigned IndentLevel = Line.Level; 2430 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2431 if (Current->Role) 2432 Current->Role->precomputeFormattingInfos(Current); 2433 if (Current->MatchingParen && 2434 Current->MatchingParen->opensBlockOrBlockTypeList(Style)) { 2435 assert(IndentLevel > 0); 2436 --IndentLevel; 2437 } 2438 Current->IndentLevel = IndentLevel; 2439 if (Current->opensBlockOrBlockTypeList(Style)) 2440 ++IndentLevel; 2441 } 2442 2443 LLVM_DEBUG({ printDebugInfo(Line); }); 2444 } 2445 2446 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2447 unsigned UnbreakableTailLength = 0; 2448 FormatToken *Current = Line.Last; 2449 while (Current) { 2450 Current->UnbreakableTailLength = UnbreakableTailLength; 2451 if (Current->CanBreakBefore || 2452 Current->isOneOf(tok::comment, tok::string_literal)) { 2453 UnbreakableTailLength = 0; 2454 } else { 2455 UnbreakableTailLength += 2456 Current->ColumnWidth + Current->SpacesRequiredBefore; 2457 } 2458 Current = Current->Previous; 2459 } 2460 } 2461 2462 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2463 const FormatToken &Tok, 2464 bool InFunctionDecl) { 2465 const FormatToken &Left = *Tok.Previous; 2466 const FormatToken &Right = Tok; 2467 2468 if (Left.is(tok::semi)) 2469 return 0; 2470 2471 if (Style.Language == FormatStyle::LK_Java) { 2472 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2473 return 1; 2474 if (Right.is(Keywords.kw_implements)) 2475 return 2; 2476 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2477 return 3; 2478 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2479 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2480 return 100; 2481 if (Left.is(TT_JsTypeColon)) 2482 return 35; 2483 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2484 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2485 return 100; 2486 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2487 if (Left.opensScope() && Right.closesScope()) 2488 return 200; 2489 } 2490 2491 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2492 return 1; 2493 if (Right.is(tok::l_square)) { 2494 if (Style.Language == FormatStyle::LK_Proto) 2495 return 1; 2496 if (Left.is(tok::r_square)) 2497 return 200; 2498 // Slightly prefer formatting local lambda definitions like functions. 2499 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2500 return 35; 2501 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2502 TT_ArrayInitializerLSquare, 2503 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) 2504 return 500; 2505 } 2506 2507 if (Left.is(tok::coloncolon) || 2508 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2509 return 500; 2510 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2511 Right.is(tok::kw_operator)) { 2512 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2513 return 3; 2514 if (Left.is(TT_StartOfName)) 2515 return 110; 2516 if (InFunctionDecl && Right.NestingLevel == 0) 2517 return Style.PenaltyReturnTypeOnItsOwnLine; 2518 return 200; 2519 } 2520 if (Right.is(TT_PointerOrReference)) 2521 return 190; 2522 if (Right.is(TT_LambdaArrow)) 2523 return 110; 2524 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2525 return 160; 2526 if (Left.is(TT_CastRParen)) 2527 return 100; 2528 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2529 return 5000; 2530 if (Left.is(tok::comment)) 2531 return 1000; 2532 2533 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2534 TT_CtorInitializerColon)) 2535 return 2; 2536 2537 if (Right.isMemberAccess()) { 2538 // Breaking before the "./->" of a chained call/member access is reasonably 2539 // cheap, as formatting those with one call per line is generally 2540 // desirable. In particular, it should be cheaper to break before the call 2541 // than it is to break inside a call's parameters, which could lead to weird 2542 // "hanging" indents. The exception is the very last "./->" to support this 2543 // frequent pattern: 2544 // 2545 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2546 // dddddddd); 2547 // 2548 // which might otherwise be blown up onto many lines. Here, clang-format 2549 // won't produce "hanging" indents anyway as there is no other trailing 2550 // call. 2551 // 2552 // Also apply higher penalty is not a call as that might lead to a wrapping 2553 // like: 2554 // 2555 // aaaaaaa 2556 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2557 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2558 ? 150 2559 : 35; 2560 } 2561 2562 if (Right.is(TT_TrailingAnnotation) && 2563 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2564 // Moving trailing annotations to the next line is fine for ObjC method 2565 // declarations. 2566 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2567 return 10; 2568 // Generally, breaking before a trailing annotation is bad unless it is 2569 // function-like. It seems to be especially preferable to keep standard 2570 // annotations (i.e. "const", "final" and "override") on the same line. 2571 // Use a slightly higher penalty after ")" so that annotations like 2572 // "const override" are kept together. 2573 bool is_short_annotation = Right.TokenText.size() < 10; 2574 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 2575 } 2576 2577 // In for-loops, prefer breaking at ',' and ';'. 2578 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 2579 return 4; 2580 2581 // In Objective-C method expressions, prefer breaking before "param:" over 2582 // breaking after it. 2583 if (Right.is(TT_SelectorName)) 2584 return 0; 2585 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 2586 return Line.MightBeFunctionDecl ? 50 : 500; 2587 2588 // In Objective-C type declarations, avoid breaking after the category's 2589 // open paren (we'll prefer breaking after the protocol list's opening 2590 // angle bracket, if present). 2591 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous && 2592 Left.Previous->isOneOf(tok::identifier, tok::greater)) 2593 return 500; 2594 2595 if (Left.is(tok::l_paren) && InFunctionDecl && 2596 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 2597 return 100; 2598 if (Left.is(tok::l_paren) && Left.Previous && 2599 (Left.Previous->is(tok::kw_for) || Left.Previous->isIf())) 2600 return 1000; 2601 if (Left.is(tok::equal) && InFunctionDecl) 2602 return 110; 2603 if (Right.is(tok::r_brace)) 2604 return 1; 2605 if (Left.is(TT_TemplateOpener)) 2606 return 100; 2607 if (Left.opensScope()) { 2608 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign) 2609 return 0; 2610 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle) 2611 return 19; 2612 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 2613 : 19; 2614 } 2615 if (Left.is(TT_JavaAnnotation)) 2616 return 50; 2617 2618 if (Left.is(TT_UnaryOperator)) 2619 return 60; 2620 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 2621 Left.Previous->isLabelString() && 2622 (Left.NextOperator || Left.OperatorIndex != 0)) 2623 return 50; 2624 if (Right.is(tok::plus) && Left.isLabelString() && 2625 (Right.NextOperator || Right.OperatorIndex != 0)) 2626 return 25; 2627 if (Left.is(tok::comma)) 2628 return 1; 2629 if (Right.is(tok::lessless) && Left.isLabelString() && 2630 (Right.NextOperator || Right.OperatorIndex != 1)) 2631 return 25; 2632 if (Right.is(tok::lessless)) { 2633 // Breaking at a << is really cheap. 2634 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 2635 // Slightly prefer to break before the first one in log-like statements. 2636 return 2; 2637 return 1; 2638 } 2639 if (Left.ClosesTemplateDeclaration) 2640 return Style.PenaltyBreakTemplateDeclaration; 2641 if (Left.is(TT_ConditionalExpr)) 2642 return prec::Conditional; 2643 prec::Level Level = Left.getPrecedence(); 2644 if (Level == prec::Unknown) 2645 Level = Right.getPrecedence(); 2646 if (Level == prec::Assignment) 2647 return Style.PenaltyBreakAssignment; 2648 if (Level != prec::Unknown) 2649 return Level; 2650 2651 return 3; 2652 } 2653 2654 bool TokenAnnotator::spaceRequiredBeforeParens(const FormatToken &Right) const { 2655 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always || 2656 (Style.SpaceBeforeParens == FormatStyle::SBPO_NonEmptyParentheses && 2657 Right.ParameterCount > 0); 2658 } 2659 2660 /// Returns \c true if the token is followed by a boolean condition, \c false 2661 /// otherwise. 2662 static bool isKeywordWithCondition(const FormatToken &Tok) { 2663 return Tok.isOneOf(tok::kw_if, tok::kw_for, tok::kw_while, tok::kw_switch, 2664 tok::kw_constexpr, tok::kw_catch); 2665 } 2666 2667 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 2668 const FormatToken &Left, 2669 const FormatToken &Right) { 2670 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 2671 return true; 2672 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 2673 return true; 2674 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 2675 Left.Tok.getObjCKeywordID() == tok::objc_property) 2676 return true; 2677 if (Right.is(tok::hashhash)) 2678 return Left.is(tok::hash); 2679 if (Left.isOneOf(tok::hashhash, tok::hash)) 2680 return Right.is(tok::hash); 2681 if ((Left.is(tok::l_paren) && Right.is(tok::r_paren)) || 2682 (Left.is(tok::l_brace) && Left.BlockKind != BK_Block && 2683 Right.is(tok::r_brace) && Right.BlockKind != BK_Block)) 2684 return Style.SpaceInEmptyParentheses; 2685 if (Style.SpacesInConditionalStatement) { 2686 if (Left.is(tok::l_paren) && Left.Previous && 2687 isKeywordWithCondition(*Left.Previous)) 2688 return true; 2689 if (Right.is(tok::r_paren) && Right.MatchingParen && 2690 Right.MatchingParen->Previous && 2691 isKeywordWithCondition(*Right.MatchingParen->Previous)) 2692 return true; 2693 } 2694 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 2695 return (Right.is(TT_CastRParen) || 2696 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 2697 ? Style.SpacesInCStyleCastParentheses 2698 : Style.SpacesInParentheses; 2699 if (Right.isOneOf(tok::semi, tok::comma)) 2700 return false; 2701 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) { 2702 bool IsLightweightGeneric = Right.MatchingParen && 2703 Right.MatchingParen->Next && 2704 Right.MatchingParen->Next->is(tok::colon); 2705 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList; 2706 } 2707 if (Right.is(tok::less) && Left.is(tok::kw_template)) 2708 return Style.SpaceAfterTemplateKeyword; 2709 if (Left.isOneOf(tok::exclaim, tok::tilde)) 2710 return false; 2711 if (Left.is(tok::at) && 2712 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 2713 tok::numeric_constant, tok::l_paren, tok::l_brace, 2714 tok::kw_true, tok::kw_false)) 2715 return false; 2716 if (Left.is(tok::colon)) 2717 return !Left.is(TT_ObjCMethodExpr); 2718 if (Left.is(tok::coloncolon)) 2719 return false; 2720 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 2721 if (Style.Language == FormatStyle::LK_TextProto || 2722 (Style.Language == FormatStyle::LK_Proto && 2723 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 2724 // Format empty list as `<>`. 2725 if (Left.is(tok::less) && Right.is(tok::greater)) 2726 return false; 2727 return !Style.Cpp11BracedListStyle; 2728 } 2729 return false; 2730 } 2731 if (Right.is(tok::ellipsis)) 2732 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 2733 Left.Previous->is(tok::kw_case)); 2734 if (Left.is(tok::l_square) && Right.is(tok::amp)) 2735 return Style.SpacesInSquareBrackets; 2736 if (Right.is(TT_PointerOrReference)) { 2737 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 2738 if (!Left.MatchingParen) 2739 return true; 2740 FormatToken *TokenBeforeMatchingParen = 2741 Left.MatchingParen->getPreviousNonComment(); 2742 if (!TokenBeforeMatchingParen || 2743 !TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype, 2744 TT_TypenameMacro)) 2745 return true; 2746 } 2747 return (Left.Tok.isLiteral() || 2748 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 2749 (Style.PointerAlignment != FormatStyle::PAS_Left || 2750 (Line.IsMultiVariableDeclStmt && 2751 (Left.NestingLevel == 0 || 2752 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for))))))); 2753 } 2754 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 2755 (!Left.is(TT_PointerOrReference) || 2756 (Style.PointerAlignment != FormatStyle::PAS_Right && 2757 !Line.IsMultiVariableDeclStmt))) 2758 return true; 2759 if (Left.is(TT_PointerOrReference)) 2760 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) || 2761 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 2762 !Right.is(TT_StartOfName)) || 2763 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) || 2764 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 2765 tok::l_paren) && 2766 (Style.PointerAlignment != FormatStyle::PAS_Right && 2767 !Line.IsMultiVariableDeclStmt) && 2768 Left.Previous && 2769 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon, 2770 tok::l_square)); 2771 if (Right.is(tok::star) && Left.is(tok::l_paren)) 2772 return false; 2773 if (Right.isOneOf(tok::star, tok::amp, tok::ampamp) && 2774 (Left.is(tok::identifier) || Left.isSimpleTypeSpecifier()) && 2775 // Space between the type and the * in: 2776 // operator void*() 2777 // operator char*() 2778 // operator /*comment*/ const char*() 2779 // operator volatile /*comment*/ char*() 2780 // operator Foo*() 2781 // dependent on PointerAlignment style. 2782 Left.Previous && 2783 (Left.Previous->endsSequence(tok::kw_operator) || 2784 Left.Previous->endsSequence(tok::kw_const, tok::kw_operator) || 2785 Left.Previous->endsSequence(tok::kw_volatile, tok::kw_operator))) 2786 return (Style.PointerAlignment != FormatStyle::PAS_Left); 2787 const auto SpaceRequiredForArrayInitializerLSquare = 2788 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 2789 return Style.SpacesInContainerLiterals || 2790 ((Style.Language == FormatStyle::LK_Proto || 2791 Style.Language == FormatStyle::LK_TextProto) && 2792 !Style.Cpp11BracedListStyle && 2793 LSquareTok.endsSequence(tok::l_square, tok::colon, 2794 TT_SelectorName)); 2795 }; 2796 if (Left.is(tok::l_square)) 2797 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 2798 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 2799 (Left.isOneOf(TT_ArraySubscriptLSquare, TT_StructuredBindingLSquare, 2800 TT_LambdaLSquare) && 2801 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 2802 if (Right.is(tok::r_square)) 2803 return Right.MatchingParen && 2804 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 2805 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 2806 Style)) || 2807 (Style.SpacesInSquareBrackets && 2808 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 2809 TT_StructuredBindingLSquare, 2810 TT_LambdaLSquare)) || 2811 Right.MatchingParen->is(TT_AttributeParen)); 2812 if (Right.is(tok::l_square) && 2813 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2814 TT_DesignatedInitializerLSquare, 2815 TT_StructuredBindingLSquare, TT_AttributeSquare) && 2816 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral) && 2817 !(!Left.is(tok::r_square) && Style.SpaceBeforeSquareBrackets && 2818 Right.is(TT_ArraySubscriptLSquare))) 2819 return false; 2820 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 2821 return !Left.Children.empty(); // No spaces in "{}". 2822 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) || 2823 (Right.is(tok::r_brace) && Right.MatchingParen && 2824 Right.MatchingParen->BlockKind != BK_Block)) 2825 return Style.Cpp11BracedListStyle ? Style.SpacesInParentheses : true; 2826 if (Left.is(TT_BlockComment)) 2827 // No whitespace in x(/*foo=*/1), except for JavaScript. 2828 return Style.Language == FormatStyle::LK_JavaScript || 2829 !Left.TokenText.endswith("=*/"); 2830 if (Right.is(tok::l_paren)) { 2831 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 2832 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) 2833 return true; 2834 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) || 2835 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never && 2836 (Left.isOneOf(tok::pp_elif, tok::kw_for, tok::kw_while, 2837 tok::kw_switch, tok::kw_case, TT_ForEachMacro, 2838 TT_ObjCForIn) || 2839 Left.isIf(Line.Type != LT_PreprocessorDirective) || 2840 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch, 2841 tok::kw_new, tok::kw_delete) && 2842 (!Left.Previous || Left.Previous->isNot(tok::period))))) || 2843 (spaceRequiredBeforeParens(Right) && 2844 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 2845 Left.is(tok::r_paren) || Left.isSimpleTypeSpecifier() || 2846 (Left.is(tok::r_square) && Left.MatchingParen && 2847 Left.MatchingParen->is(TT_LambdaLSquare))) && 2848 Line.Type != LT_PreprocessorDirective); 2849 } 2850 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 2851 return false; 2852 if (Right.is(TT_UnaryOperator)) 2853 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 2854 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 2855 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 2856 tok::r_paren) || 2857 Left.isSimpleTypeSpecifier()) && 2858 Right.is(tok::l_brace) && Right.getNextNonComment() && 2859 Right.BlockKind != BK_Block) 2860 return false; 2861 if (Left.is(tok::period) || Right.is(tok::period)) 2862 return false; 2863 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L") 2864 return false; 2865 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 2866 Left.MatchingParen->Previous && 2867 (Left.MatchingParen->Previous->is(tok::period) || 2868 Left.MatchingParen->Previous->is(tok::coloncolon))) 2869 // Java call to generic function with explicit type: 2870 // A.<B<C<...>>>DoSomething(); 2871 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference. 2872 return false; 2873 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 2874 return false; 2875 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at)) 2876 // Objective-C dictionary literal -> no space after opening brace. 2877 return false; 2878 if (Right.is(tok::r_brace) && Right.MatchingParen && 2879 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at)) 2880 // Objective-C dictionary literal -> no space before closing brace. 2881 return false; 2882 if (Right.Type == TT_TrailingAnnotation && 2883 Right.isOneOf(tok::amp, tok::ampamp) && 2884 Left.isOneOf(tok::kw_const, tok::kw_volatile) && 2885 (!Right.Next || Right.Next->is(tok::semi))) 2886 // Match const and volatile ref-qualifiers without any additional 2887 // qualifiers such as 2888 // void Fn() const &; 2889 return Style.PointerAlignment != FormatStyle::PAS_Left; 2890 return true; 2891 } 2892 2893 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 2894 const FormatToken &Right) { 2895 const FormatToken &Left = *Right.Previous; 2896 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 2897 return true; // Never ever merge two identifiers. 2898 if (Style.isCpp()) { 2899 if (Left.is(tok::kw_operator)) 2900 return Right.is(tok::coloncolon); 2901 if (Right.is(tok::l_brace) && Right.BlockKind == BK_BracedInit && 2902 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList) 2903 return true; 2904 } else if (Style.Language == FormatStyle::LK_Proto || 2905 Style.Language == FormatStyle::LK_TextProto) { 2906 if (Right.is(tok::period) && 2907 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 2908 Keywords.kw_repeated, Keywords.kw_extend)) 2909 return true; 2910 if (Right.is(tok::l_paren) && 2911 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 2912 return true; 2913 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 2914 return true; 2915 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 2916 if (Left.is(tok::slash) || Right.is(tok::slash)) 2917 return false; 2918 if (Left.MatchingParen && 2919 Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 2920 Right.isOneOf(tok::l_brace, tok::less)) 2921 return !Style.Cpp11BracedListStyle; 2922 // A percent is probably part of a formatting specification, such as %lld. 2923 if (Left.is(tok::percent)) 2924 return false; 2925 // Preserve the existence of a space before a percent for cases like 0x%04x 2926 // and "%d %d" 2927 if (Left.is(tok::numeric_constant) && Right.is(tok::percent)) 2928 return Right.WhitespaceRange.getEnd() != Right.WhitespaceRange.getBegin(); 2929 } else if (Style.isCSharp()) { 2930 // Require spaces around '{' and before '}' unless they appear in 2931 // interpolated strings. Interpolated strings are merged into a single token 2932 // so cannot have spaces inserted by this function. 2933 2934 // No space between 'this' and '[' 2935 if (Left.is(tok::kw_this) && Right.is(tok::l_square)) 2936 return false; 2937 2938 // No space between 'new' and '(' 2939 if (Left.is(tok::kw_new) && Right.is(tok::l_paren)) 2940 return false; 2941 2942 // Space before { (including space within '{ {'). 2943 if (Right.is(tok::l_brace)) 2944 return true; 2945 2946 // Spaces inside braces. 2947 if (Left.is(tok::l_brace) && Right.isNot(tok::r_brace)) 2948 return true; 2949 2950 if (Left.isNot(tok::l_brace) && Right.is(tok::r_brace)) 2951 return true; 2952 2953 // Spaces around '=>'. 2954 if (Left.is(TT_JsFatArrow) || Right.is(TT_JsFatArrow)) 2955 return true; 2956 2957 // No spaces around attribute target colons 2958 if (Left.is(TT_AttributeColon) || Right.is(TT_AttributeColon)) 2959 return false; 2960 2961 // space between type and variable e.g. Dictionary<string,string> foo; 2962 if (Left.is(TT_TemplateCloser) && Right.is(TT_StartOfName)) 2963 return true; 2964 2965 // spaces inside square brackets. 2966 if (Left.is(tok::l_square) || Right.is(tok::r_square)) 2967 return Style.SpacesInSquareBrackets; 2968 2969 // No space before ? in nullable types. 2970 if (Right.is(TT_CSharpNullable)) 2971 return false; 2972 2973 // Require space after ? in nullable types except in generics and casts. 2974 if (Left.is(TT_CSharpNullable)) 2975 return !Right.isOneOf(TT_TemplateCloser, tok::r_paren); 2976 2977 // No space before or after '?.'. 2978 if (Left.is(TT_CSharpNullConditional) || Right.is(TT_CSharpNullConditional)) 2979 return false; 2980 2981 // Space before and after '??'. 2982 if (Left.is(TT_CSharpNullCoalescing) || Right.is(TT_CSharpNullCoalescing)) 2983 return true; 2984 2985 // No space before '?['. 2986 if (Right.is(TT_CSharpNullConditionalLSquare)) 2987 return false; 2988 2989 // Possible space inside `?[ 0 ]`. 2990 if (Left.is(TT_CSharpNullConditionalLSquare)) 2991 return Style.SpacesInSquareBrackets; 2992 2993 // space between keywords and paren e.g. "using (" 2994 if (Right.is(tok::l_paren)) 2995 if (Left.isOneOf(tok::kw_using, Keywords.kw_async, Keywords.kw_when)) 2996 return Style.SpaceBeforeParens == FormatStyle::SBPO_ControlStatements || 2997 spaceRequiredBeforeParens(Right); 2998 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2999 if (Left.is(TT_JsFatArrow)) 3000 return true; 3001 // for await ( ... 3002 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 3003 Left.Previous->is(tok::kw_for)) 3004 return true; 3005 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 3006 Right.MatchingParen) { 3007 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 3008 // An async arrow function, for example: `x = async () => foo();`, 3009 // as opposed to calling a function called async: `x = async();` 3010 if (Next && Next->is(TT_JsFatArrow)) 3011 return true; 3012 } 3013 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 3014 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 3015 return false; 3016 // In tagged template literals ("html`bar baz`"), there is no space between 3017 // the tag identifier and the template string. getIdentifierInfo makes sure 3018 // that the identifier is not a pseudo keyword like `yield`, either. 3019 if (Left.is(tok::identifier) && Keywords.IsJavaScriptIdentifier(Left) && 3020 Right.is(TT_TemplateString)) 3021 return false; 3022 if (Right.is(tok::star) && 3023 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 3024 return false; 3025 if (Right.isOneOf(tok::l_brace, tok::l_square) && 3026 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 3027 Keywords.kw_extends, Keywords.kw_implements)) 3028 return true; 3029 if (Right.is(tok::l_paren)) { 3030 // JS methods can use some keywords as names (e.g. `delete()`). 3031 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 3032 return false; 3033 // Valid JS method names can include keywords, e.g. `foo.delete()` or 3034 // `bar.instanceof()`. Recognize call positions by preceding period. 3035 if (Left.Previous && Left.Previous->is(tok::period) && 3036 Left.Tok.getIdentifierInfo()) 3037 return false; 3038 // Additional unary JavaScript operators that need a space after. 3039 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 3040 tok::kw_void)) 3041 return true; 3042 } 3043 // `foo as const;` casts into a const type. 3044 if (Left.endsSequence(tok::kw_const, Keywords.kw_as)) { 3045 return false; 3046 } 3047 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 3048 tok::kw_const) || 3049 // "of" is only a keyword if it appears after another identifier 3050 // (e.g. as "const x of y" in a for loop), or after a destructuring 3051 // operation (const [x, y] of z, const {a, b} of c). 3052 (Left.is(Keywords.kw_of) && Left.Previous && 3053 (Left.Previous->Tok.is(tok::identifier) || 3054 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 3055 (!Left.Previous || !Left.Previous->is(tok::period))) 3056 return true; 3057 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 3058 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 3059 return false; 3060 if (Left.is(Keywords.kw_as) && 3061 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 3062 return true; 3063 if (Left.is(tok::kw_default) && Left.Previous && 3064 Left.Previous->is(tok::kw_export)) 3065 return true; 3066 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 3067 return true; 3068 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 3069 return false; 3070 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 3071 return false; 3072 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 3073 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 3074 return false; 3075 if (Left.is(tok::ellipsis)) 3076 return false; 3077 if (Left.is(TT_TemplateCloser) && 3078 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 3079 Keywords.kw_implements, Keywords.kw_extends)) 3080 // Type assertions ('<type>expr') are not followed by whitespace. Other 3081 // locations that should have whitespace following are identified by the 3082 // above set of follower tokens. 3083 return false; 3084 if (Right.is(TT_JsNonNullAssertion)) 3085 return false; 3086 if (Left.is(TT_JsNonNullAssertion) && 3087 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 3088 return true; // "x! as string", "x! in y" 3089 } else if (Style.Language == FormatStyle::LK_Java) { 3090 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 3091 return true; 3092 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 3093 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never; 3094 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 3095 tok::kw_protected) || 3096 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 3097 Keywords.kw_native)) && 3098 Right.is(TT_TemplateOpener)) 3099 return true; 3100 } 3101 if (Left.is(TT_ImplicitStringLiteral)) 3102 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 3103 if (Line.Type == LT_ObjCMethodDecl) { 3104 if (Left.is(TT_ObjCMethodSpecifier)) 3105 return true; 3106 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right)) 3107 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a 3108 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class 3109 // method declaration. 3110 return false; 3111 } 3112 if (Line.Type == LT_ObjCProperty && 3113 (Right.is(tok::equal) || Left.is(tok::equal))) 3114 return false; 3115 3116 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 3117 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 3118 return true; 3119 if (Right.is(TT_OverloadedOperatorLParen)) 3120 return spaceRequiredBeforeParens(Right); 3121 if (Left.is(tok::comma)) 3122 return true; 3123 if (Right.is(tok::comma)) 3124 return false; 3125 if (Right.is(TT_ObjCBlockLParen)) 3126 return true; 3127 if (Right.is(TT_CtorInitializerColon)) 3128 return Style.SpaceBeforeCtorInitializerColon; 3129 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 3130 return false; 3131 if (Right.is(TT_RangeBasedForLoopColon) && 3132 !Style.SpaceBeforeRangeBasedForLoopColon) 3133 return false; 3134 if (Right.is(tok::colon)) { 3135 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) || 3136 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 3137 return false; 3138 if (Right.is(TT_ObjCMethodExpr)) 3139 return false; 3140 if (Left.is(tok::question)) 3141 return false; 3142 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 3143 return false; 3144 if (Right.is(TT_DictLiteral)) 3145 return Style.SpacesInContainerLiterals; 3146 if (Right.is(TT_AttributeColon)) 3147 return false; 3148 if (Right.is(TT_CSharpNamedArgumentColon)) 3149 return false; 3150 return true; 3151 } 3152 if (Left.is(TT_UnaryOperator)) { 3153 if (!Right.is(tok::l_paren)) { 3154 // The alternative operators for ~ and ! are "compl" and "not". 3155 // If they are used instead, we do not want to combine them with 3156 // the token to the right, unless that is a left paren. 3157 if (Left.is(tok::exclaim) && Left.TokenText == "not") 3158 return true; 3159 if (Left.is(tok::tilde) && Left.TokenText == "compl") 3160 return true; 3161 // Lambda captures allow for a lone &, so "&]" needs to be properly 3162 // handled. 3163 if (Left.is(tok::amp) && Right.is(tok::r_square)) 3164 return Style.SpacesInSquareBrackets; 3165 } 3166 return (Style.SpaceAfterLogicalNot && Left.is(tok::exclaim)) || 3167 Right.is(TT_BinaryOperator); 3168 } 3169 3170 // If the next token is a binary operator or a selector name, we have 3171 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 3172 if (Left.is(TT_CastRParen)) 3173 return Style.SpaceAfterCStyleCast || 3174 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 3175 3176 if (Left.is(tok::greater) && Right.is(tok::greater)) { 3177 if (Style.Language == FormatStyle::LK_TextProto || 3178 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 3179 return !Style.Cpp11BracedListStyle; 3180 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 3181 (Style.Standard < FormatStyle::LS_Cpp11 || Style.SpacesInAngles); 3182 } 3183 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 3184 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 3185 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 3186 return false; 3187 if (!Style.SpaceBeforeAssignmentOperators && Left.isNot(TT_TemplateCloser) && 3188 Right.getPrecedence() == prec::Assignment) 3189 return false; 3190 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) && 3191 (Left.is(tok::identifier) || Left.is(tok::kw_this))) 3192 return false; 3193 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 3194 // Generally don't remove existing spaces between an identifier and "::". 3195 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 3196 // this turns out to be too lenient, add analysis of the identifier itself. 3197 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 3198 if (Right.is(tok::coloncolon) && 3199 !Left.isOneOf(tok::l_brace, tok::comment, tok::l_paren)) 3200 return (Left.is(TT_TemplateOpener) && 3201 Style.Standard < FormatStyle::LS_Cpp11) || 3202 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 3203 tok::kw___super, TT_TemplateCloser, 3204 TT_TemplateOpener)) || 3205 (Left.is(tok ::l_paren) && Style.SpacesInParentheses); 3206 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 3207 return Style.SpacesInAngles; 3208 // Space before TT_StructuredBindingLSquare. 3209 if (Right.is(TT_StructuredBindingLSquare)) 3210 return !Left.isOneOf(tok::amp, tok::ampamp) || 3211 Style.PointerAlignment != FormatStyle::PAS_Right; 3212 // Space before & or && following a TT_StructuredBindingLSquare. 3213 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 3214 Right.isOneOf(tok::amp, tok::ampamp)) 3215 return Style.PointerAlignment != FormatStyle::PAS_Left; 3216 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 3217 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 3218 !Right.is(tok::r_paren))) 3219 return true; 3220 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) && 3221 Right.isNot(TT_FunctionTypeLParen)) 3222 return spaceRequiredBeforeParens(Right); 3223 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 3224 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 3225 return false; 3226 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 3227 Line.startsWith(tok::hash)) 3228 return true; 3229 if (Right.is(TT_TrailingUnaryOperator)) 3230 return false; 3231 if (Left.is(TT_RegexLiteral)) 3232 return false; 3233 return spaceRequiredBetween(Line, Left, Right); 3234 } 3235 3236 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 3237 static bool isAllmanBrace(const FormatToken &Tok) { 3238 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block && 3239 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_LambdaLBrace, TT_DictLiteral); 3240 } 3241 3242 // Returns 'true' if 'Tok' is an function argument. 3243 static bool IsFunctionArgument(const FormatToken &Tok) { 3244 return Tok.MatchingParen && Tok.MatchingParen->Next && 3245 Tok.MatchingParen->Next->isOneOf(tok::comma, tok::r_paren); 3246 } 3247 3248 static bool 3249 isItAnEmptyLambdaAllowed(const FormatToken &Tok, 3250 FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3251 return Tok.Children.empty() && ShortLambdaOption != FormatStyle::SLS_None; 3252 } 3253 3254 static bool 3255 isItAInlineLambdaAllowed(const FormatToken &Tok, 3256 FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3257 return (ShortLambdaOption == FormatStyle::SLS_Inline && 3258 IsFunctionArgument(Tok)) || 3259 (ShortLambdaOption == FormatStyle::SLS_All); 3260 } 3261 3262 static bool isOneChildWithoutMustBreakBefore(const FormatToken &Tok) { 3263 if (Tok.Children.size() != 1) 3264 return false; 3265 FormatToken *curElt = Tok.Children[0]->First; 3266 while (curElt) { 3267 if (curElt->MustBreakBefore) 3268 return false; 3269 curElt = curElt->Next; 3270 } 3271 return true; 3272 } 3273 static bool 3274 isAllmanLambdaBrace(const FormatToken &Tok) { 3275 return (Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block && 3276 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral)); 3277 } 3278 3279 static bool 3280 isAllmanBraceIncludedBreakableLambda(const FormatToken &Tok, 3281 FormatStyle::ShortLambdaStyle ShortLambdaOption) { 3282 if (!isAllmanLambdaBrace(Tok)) 3283 return false; 3284 3285 if (isItAnEmptyLambdaAllowed(Tok, ShortLambdaOption)) 3286 return false; 3287 3288 return !isItAInlineLambdaAllowed(Tok, ShortLambdaOption) || 3289 !isOneChildWithoutMustBreakBefore(Tok); 3290 } 3291 3292 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 3293 const FormatToken &Right) { 3294 const FormatToken &Left = *Right.Previous; 3295 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 3296 return true; 3297 3298 if (Style.isCSharp()) { 3299 if (Right.is(TT_CSharpNamedArgumentColon) || 3300 Left.is(TT_CSharpNamedArgumentColon)) 3301 return false; 3302 } else if (Style.Language == FormatStyle::LK_JavaScript) { 3303 // FIXME: This might apply to other languages and token kinds. 3304 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 3305 Left.Previous->is(tok::string_literal)) 3306 return true; 3307 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 3308 Left.Previous && Left.Previous->is(tok::equal) && 3309 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 3310 tok::kw_const) && 3311 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 3312 // above. 3313 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 3314 // Object literals on the top level of a file are treated as "enum-style". 3315 // Each key/value pair is put on a separate line, instead of bin-packing. 3316 return true; 3317 if (Left.is(tok::l_brace) && Line.Level == 0 && 3318 (Line.startsWith(tok::kw_enum) || 3319 Line.startsWith(tok::kw_const, tok::kw_enum) || 3320 Line.startsWith(tok::kw_export, tok::kw_enum) || 3321 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 3322 // JavaScript top-level enum key/value pairs are put on separate lines 3323 // instead of bin-packing. 3324 return true; 3325 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && Left.Previous && 3326 Left.Previous->is(TT_JsFatArrow)) { 3327 // JS arrow function (=> {...}). 3328 switch (Style.AllowShortLambdasOnASingleLine) { 3329 case FormatStyle::SLS_All: 3330 return false; 3331 case FormatStyle::SLS_None: 3332 return true; 3333 case FormatStyle::SLS_Empty: 3334 return !Left.Children.empty(); 3335 case FormatStyle::SLS_Inline: 3336 // allow one-lining inline (e.g. in function call args) and empty arrow 3337 // functions. 3338 return (Left.NestingLevel == 0 && Line.Level == 0) && 3339 !Left.Children.empty(); 3340 } 3341 llvm_unreachable("Unknown FormatStyle::ShortLambdaStyle enum"); 3342 } 3343 3344 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 3345 !Left.Children.empty()) 3346 // Support AllowShortFunctionsOnASingleLine for JavaScript. 3347 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 3348 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 3349 (Left.NestingLevel == 0 && Line.Level == 0 && 3350 Style.AllowShortFunctionsOnASingleLine & 3351 FormatStyle::SFS_InlineOnly); 3352 } else if (Style.Language == FormatStyle::LK_Java) { 3353 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 3354 Right.Next->is(tok::string_literal)) 3355 return true; 3356 } else if (Style.Language == FormatStyle::LK_Cpp || 3357 Style.Language == FormatStyle::LK_ObjC || 3358 Style.Language == FormatStyle::LK_Proto || 3359 Style.Language == FormatStyle::LK_TableGen || 3360 Style.Language == FormatStyle::LK_TextProto) { 3361 if (Left.isStringLiteral() && Right.isStringLiteral()) 3362 return true; 3363 } 3364 3365 // If the last token before a '}', ']', or ')' is a comma or a trailing 3366 // comment, the intention is to insert a line break after it in order to make 3367 // shuffling around entries easier. Import statements, especially in 3368 // JavaScript, can be an exception to this rule. 3369 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 3370 const FormatToken *BeforeClosingBrace = nullptr; 3371 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 3372 (Style.Language == FormatStyle::LK_JavaScript && 3373 Left.is(tok::l_paren))) && 3374 Left.BlockKind != BK_Block && Left.MatchingParen) 3375 BeforeClosingBrace = Left.MatchingParen->Previous; 3376 else if (Right.MatchingParen && 3377 (Right.MatchingParen->isOneOf(tok::l_brace, 3378 TT_ArrayInitializerLSquare) || 3379 (Style.Language == FormatStyle::LK_JavaScript && 3380 Right.MatchingParen->is(tok::l_paren)))) 3381 BeforeClosingBrace = &Left; 3382 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 3383 BeforeClosingBrace->isTrailingComment())) 3384 return true; 3385 } 3386 3387 if (Right.is(tok::comment)) 3388 return Left.BlockKind != BK_BracedInit && 3389 Left.isNot(TT_CtorInitializerColon) && 3390 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 3391 if (Left.isTrailingComment()) 3392 return true; 3393 if (Right.Previous->IsUnterminatedLiteral) 3394 return true; 3395 if (Right.is(tok::lessless) && Right.Next && 3396 Right.Previous->is(tok::string_literal) && 3397 Right.Next->is(tok::string_literal)) 3398 return true; 3399 if (Right.Previous->ClosesTemplateDeclaration && 3400 Right.Previous->MatchingParen && 3401 Right.Previous->MatchingParen->NestingLevel == 0 && 3402 Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes) 3403 return true; 3404 if (Right.is(TT_CtorInitializerComma) && 3405 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3406 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 3407 return true; 3408 if (Right.is(TT_CtorInitializerColon) && 3409 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3410 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 3411 return true; 3412 // Break only if we have multiple inheritance. 3413 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 3414 Right.is(TT_InheritanceComma)) 3415 return true; 3416 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 3417 // Multiline raw string literals are special wrt. line breaks. The author 3418 // has made a deliberate choice and might have aligned the contents of the 3419 // string literal accordingly. Thus, we try keep existing line breaks. 3420 return Right.IsMultiline && Right.NewlinesBefore > 0; 3421 if ((Right.Previous->is(tok::l_brace) || 3422 (Right.Previous->is(tok::less) && Right.Previous->Previous && 3423 Right.Previous->Previous->is(tok::equal))) && 3424 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 3425 // Don't put enums or option definitions onto single lines in protocol 3426 // buffers. 3427 return true; 3428 } 3429 if (Right.is(TT_InlineASMBrace)) 3430 return Right.HasUnescapedNewline; 3431 3432 auto ShortLambdaOption = Style.AllowShortLambdasOnASingleLine; 3433 if (Style.BraceWrapping.BeforeLambdaBody && 3434 (isAllmanBraceIncludedBreakableLambda(Left, ShortLambdaOption) || 3435 isAllmanBraceIncludedBreakableLambda(Right, ShortLambdaOption))) { 3436 return true; 3437 } 3438 3439 if (isAllmanBrace(Left) || isAllmanBrace(Right)) 3440 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) || 3441 (Line.startsWith(tok::kw_typedef, tok::kw_enum) && 3442 Style.BraceWrapping.AfterEnum) || 3443 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 3444 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 3445 if (Left.is(TT_ObjCBlockLBrace) && 3446 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 3447 return true; 3448 3449 if (Left.is(TT_LambdaLBrace)) { 3450 if (IsFunctionArgument(Left) && 3451 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline) 3452 return false; 3453 3454 if (Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_None || 3455 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline || 3456 (!Left.Children.empty() && 3457 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Empty)) 3458 return true; 3459 } 3460 3461 // Put multiple Java annotation on a new line. 3462 if ((Style.Language == FormatStyle::LK_Java || 3463 Style.Language == FormatStyle::LK_JavaScript) && 3464 Left.is(TT_LeadingJavaAnnotation) && 3465 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 3466 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 3467 return true; 3468 3469 if (Right.is(TT_ProtoExtensionLSquare)) 3470 return true; 3471 3472 // In text proto instances if a submessage contains at least 2 entries and at 3473 // least one of them is a submessage, like A { ... B { ... } ... }, 3474 // put all of the entries of A on separate lines by forcing the selector of 3475 // the submessage B to be put on a newline. 3476 // 3477 // Example: these can stay on one line: 3478 // a { scalar_1: 1 scalar_2: 2 } 3479 // a { b { key: value } } 3480 // 3481 // and these entries need to be on a new line even if putting them all in one 3482 // line is under the column limit: 3483 // a { 3484 // scalar: 1 3485 // b { key: value } 3486 // } 3487 // 3488 // We enforce this by breaking before a submessage field that has previous 3489 // siblings, *and* breaking before a field that follows a submessage field. 3490 // 3491 // Be careful to exclude the case [proto.ext] { ... } since the `]` is 3492 // the TT_SelectorName there, but we don't want to break inside the brackets. 3493 // 3494 // Another edge case is @submessage { key: value }, which is a common 3495 // substitution placeholder. In this case we want to keep `@` and `submessage` 3496 // together. 3497 // 3498 // We ensure elsewhere that extensions are always on their own line. 3499 if ((Style.Language == FormatStyle::LK_Proto || 3500 Style.Language == FormatStyle::LK_TextProto) && 3501 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) { 3502 // Keep `@submessage` together in: 3503 // @submessage { key: value } 3504 if (Right.Previous && Right.Previous->is(tok::at)) 3505 return false; 3506 // Look for the scope opener after selector in cases like: 3507 // selector { ... 3508 // selector: { ... 3509 // selector: @base { ... 3510 FormatToken *LBrace = Right.Next; 3511 if (LBrace && LBrace->is(tok::colon)) { 3512 LBrace = LBrace->Next; 3513 if (LBrace && LBrace->is(tok::at)) { 3514 LBrace = LBrace->Next; 3515 if (LBrace) 3516 LBrace = LBrace->Next; 3517 } 3518 } 3519 if (LBrace && 3520 // The scope opener is one of {, [, <: 3521 // selector { ... } 3522 // selector [ ... ] 3523 // selector < ... > 3524 // 3525 // In case of selector { ... }, the l_brace is TT_DictLiteral. 3526 // In case of an empty selector {}, the l_brace is not TT_DictLiteral, 3527 // so we check for immediately following r_brace. 3528 ((LBrace->is(tok::l_brace) && 3529 (LBrace->is(TT_DictLiteral) || 3530 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) || 3531 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) { 3532 // If Left.ParameterCount is 0, then this submessage entry is not the 3533 // first in its parent submessage, and we want to break before this entry. 3534 // If Left.ParameterCount is greater than 0, then its parent submessage 3535 // might contain 1 or more entries and we want to break before this entry 3536 // if it contains at least 2 entries. We deal with this case later by 3537 // detecting and breaking before the next entry in the parent submessage. 3538 if (Left.ParameterCount == 0) 3539 return true; 3540 // However, if this submessage is the first entry in its parent 3541 // submessage, Left.ParameterCount might be 1 in some cases. 3542 // We deal with this case later by detecting an entry 3543 // following a closing paren of this submessage. 3544 } 3545 3546 // If this is an entry immediately following a submessage, it will be 3547 // preceded by a closing paren of that submessage, like in: 3548 // left---. .---right 3549 // v v 3550 // sub: { ... } key: value 3551 // If there was a comment between `}` an `key` above, then `key` would be 3552 // put on a new line anyways. 3553 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square)) 3554 return true; 3555 } 3556 3557 // Deal with lambda arguments in C++ - we want consistent line breaks whether 3558 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced 3559 // as aggressive line breaks are placed when the lambda is not the last arg. 3560 if ((Style.Language == FormatStyle::LK_Cpp || 3561 Style.Language == FormatStyle::LK_ObjC) && 3562 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 && 3563 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) { 3564 // Multiple lambdas in the same function call force line breaks. 3565 if (Left.BlockParameterCount > 1) 3566 return true; 3567 3568 // A lambda followed by another arg forces a line break. 3569 if (!Left.Role) 3570 return false; 3571 auto Comma = Left.Role->lastComma(); 3572 if (!Comma) 3573 return false; 3574 auto Next = Comma->getNextNonComment(); 3575 if (!Next) 3576 return false; 3577 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret)) 3578 return true; 3579 } 3580 3581 return false; 3582 } 3583 3584 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 3585 const FormatToken &Right) { 3586 const FormatToken &Left = *Right.Previous; 3587 // Language-specific stuff. 3588 if (Style.isCSharp()) { 3589 if (Left.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon) || 3590 Right.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon)) 3591 return false; 3592 } else if (Style.Language == FormatStyle::LK_Java) { 3593 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3594 Keywords.kw_implements)) 3595 return false; 3596 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3597 Keywords.kw_implements)) 3598 return true; 3599 } else if (Style.Language == FormatStyle::LK_JavaScript) { 3600 const FormatToken *NonComment = Right.getPreviousNonComment(); 3601 if (NonComment && 3602 NonComment->isOneOf( 3603 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 3604 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 3605 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 3606 Keywords.kw_readonly, Keywords.kw_abstract, Keywords.kw_get, 3607 Keywords.kw_set, Keywords.kw_async, Keywords.kw_await)) 3608 return false; // Otherwise automatic semicolon insertion would trigger. 3609 if (Right.NestingLevel == 0 && 3610 (Left.Tok.getIdentifierInfo() || 3611 Left.isOneOf(tok::r_square, tok::r_paren)) && 3612 Right.isOneOf(tok::l_square, tok::l_paren)) 3613 return false; // Otherwise automatic semicolon insertion would trigger. 3614 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace)) 3615 return false; 3616 if (Left.is(TT_JsTypeColon)) 3617 return true; 3618 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;"). 3619 if (Left.is(tok::exclaim) && Right.is(tok::colon)) 3620 return false; 3621 // Look for is type annotations like: 3622 // function f(): a is B { ... } 3623 // Do not break before is in these cases. 3624 if (Right.is(Keywords.kw_is)) { 3625 const FormatToken *Next = Right.getNextNonComment(); 3626 // If `is` is followed by a colon, it's likely that it's a dict key, so 3627 // ignore it for this check. 3628 // For example this is common in Polymer: 3629 // Polymer({ 3630 // is: 'name', 3631 // ... 3632 // }); 3633 if (!Next || !Next->is(tok::colon)) 3634 return false; 3635 } 3636 if (Left.is(Keywords.kw_in)) 3637 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 3638 if (Right.is(Keywords.kw_in)) 3639 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 3640 if (Right.is(Keywords.kw_as)) 3641 return false; // must not break before as in 'x as type' casts 3642 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) { 3643 // extends and infer can appear as keywords in conditional types: 3644 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types 3645 // do not break before them, as the expressions are subject to ASI. 3646 return false; 3647 } 3648 if (Left.is(Keywords.kw_as)) 3649 return true; 3650 if (Left.is(TT_JsNonNullAssertion)) 3651 return true; 3652 if (Left.is(Keywords.kw_declare) && 3653 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 3654 Keywords.kw_function, tok::kw_class, tok::kw_enum, 3655 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 3656 Keywords.kw_let, tok::kw_const)) 3657 // See grammar for 'declare' statements at: 3658 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10 3659 return false; 3660 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 3661 Right.isOneOf(tok::identifier, tok::string_literal)) 3662 return false; // must not break in "module foo { ...}" 3663 if (Right.is(TT_TemplateString) && Right.closesScope()) 3664 return false; 3665 // Don't split tagged template literal so there is a break between the tag 3666 // identifier and template string. 3667 if (Left.is(tok::identifier) && Right.is(TT_TemplateString)) { 3668 return false; 3669 } 3670 if (Left.is(TT_TemplateString) && Left.opensScope()) 3671 return true; 3672 } 3673 3674 if (Left.is(tok::at)) 3675 return false; 3676 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 3677 return false; 3678 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 3679 return !Right.is(tok::l_paren); 3680 if (Right.is(TT_PointerOrReference)) 3681 return Line.IsMultiVariableDeclStmt || 3682 (Style.PointerAlignment == FormatStyle::PAS_Right && 3683 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 3684 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 3685 Right.is(tok::kw_operator)) 3686 return true; 3687 if (Left.is(TT_PointerOrReference)) 3688 return false; 3689 if (Right.isTrailingComment()) 3690 // We rely on MustBreakBefore being set correctly here as we should not 3691 // change the "binding" behavior of a comment. 3692 // The first comment in a braced lists is always interpreted as belonging to 3693 // the first list element. Otherwise, it should be placed outside of the 3694 // list. 3695 return Left.BlockKind == BK_BracedInit || 3696 (Left.is(TT_CtorInitializerColon) && 3697 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 3698 if (Left.is(tok::question) && Right.is(tok::colon)) 3699 return false; 3700 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 3701 return Style.BreakBeforeTernaryOperators; 3702 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 3703 return !Style.BreakBeforeTernaryOperators; 3704 if (Left.is(TT_InheritanceColon)) 3705 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon; 3706 if (Right.is(TT_InheritanceColon)) 3707 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon; 3708 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 3709 Left.isNot(TT_SelectorName)) 3710 return true; 3711 3712 if (Right.is(tok::colon) && 3713 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 3714 return false; 3715 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 3716 if (Style.Language == FormatStyle::LK_Proto || 3717 Style.Language == FormatStyle::LK_TextProto) { 3718 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 3719 return false; 3720 // Prevent cases like: 3721 // 3722 // submessage: 3723 // { key: valueeeeeeeeeeee } 3724 // 3725 // when the snippet does not fit into one line. 3726 // Prefer: 3727 // 3728 // submessage: { 3729 // key: valueeeeeeeeeeee 3730 // } 3731 // 3732 // instead, even if it is longer by one line. 3733 // 3734 // Note that this allows allows the "{" to go over the column limit 3735 // when the column limit is just between ":" and "{", but that does 3736 // not happen too often and alternative formattings in this case are 3737 // not much better. 3738 // 3739 // The code covers the cases: 3740 // 3741 // submessage: { ... } 3742 // submessage: < ... > 3743 // repeated: [ ... ] 3744 if (((Right.is(tok::l_brace) || Right.is(tok::less)) && 3745 Right.is(TT_DictLiteral)) || 3746 Right.is(TT_ArrayInitializerLSquare)) 3747 return false; 3748 } 3749 return true; 3750 } 3751 if (Right.is(tok::r_square) && Right.MatchingParen && 3752 Right.MatchingParen->is(TT_ProtoExtensionLSquare)) 3753 return false; 3754 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 3755 Right.Next->is(TT_ObjCMethodExpr))) 3756 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 3757 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 3758 return true; 3759 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 3760 return true; 3761 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 3762 TT_OverloadedOperator)) 3763 return false; 3764 if (Left.is(TT_RangeBasedForLoopColon)) 3765 return true; 3766 if (Right.is(TT_RangeBasedForLoopColon)) 3767 return false; 3768 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 3769 return true; 3770 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 3771 Left.is(tok::kw_operator)) 3772 return false; 3773 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 3774 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 3775 return false; 3776 if (Left.is(tok::equal) && Right.is(tok::l_brace) && 3777 !Style.Cpp11BracedListStyle) 3778 return false; 3779 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen)) 3780 return false; 3781 if (Left.is(tok::l_paren) && Left.Previous && 3782 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 3783 return false; 3784 if (Right.is(TT_ImplicitStringLiteral)) 3785 return false; 3786 3787 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser)) 3788 return false; 3789 if (Right.is(tok::r_square) && Right.MatchingParen && 3790 Right.MatchingParen->is(TT_LambdaLSquare)) 3791 return false; 3792 3793 // We only break before r_brace if there was a corresponding break before 3794 // the l_brace, which is tracked by BreakBeforeClosingBrace. 3795 if (Right.is(tok::r_brace)) 3796 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block; 3797 3798 // Allow breaking after a trailing annotation, e.g. after a method 3799 // declaration. 3800 if (Left.is(TT_TrailingAnnotation)) 3801 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 3802 tok::less, tok::coloncolon); 3803 3804 if (Right.is(tok::kw___attribute) || 3805 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) 3806 return true; 3807 3808 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 3809 return true; 3810 3811 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 3812 return true; 3813 3814 if (Left.is(TT_CtorInitializerColon)) 3815 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 3816 if (Right.is(TT_CtorInitializerColon)) 3817 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 3818 if (Left.is(TT_CtorInitializerComma) && 3819 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 3820 return false; 3821 if (Right.is(TT_CtorInitializerComma) && 3822 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 3823 return true; 3824 if (Left.is(TT_InheritanceComma) && 3825 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 3826 return false; 3827 if (Right.is(TT_InheritanceComma) && 3828 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 3829 return true; 3830 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 3831 (Left.is(tok::less) && Right.is(tok::less))) 3832 return false; 3833 if (Right.is(TT_BinaryOperator) && 3834 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 3835 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 3836 Right.getPrecedence() != prec::Assignment)) 3837 return true; 3838 if (Left.is(TT_ArrayInitializerLSquare)) 3839 return true; 3840 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 3841 return true; 3842 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 3843 !Left.isOneOf(tok::arrowstar, tok::lessless) && 3844 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 3845 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 3846 Left.getPrecedence() == prec::Assignment)) 3847 return true; 3848 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 3849 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) 3850 return false; 3851 3852 auto ShortLambdaOption = Style.AllowShortLambdasOnASingleLine; 3853 if (Style.BraceWrapping.BeforeLambdaBody) { 3854 if (isAllmanLambdaBrace(Left)) 3855 return !isItAnEmptyLambdaAllowed(Left, ShortLambdaOption); 3856 if (isAllmanLambdaBrace(Right)) 3857 return !isItAnEmptyLambdaAllowed(Right, ShortLambdaOption); 3858 } 3859 3860 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 3861 tok::kw_class, tok::kw_struct, tok::comment) || 3862 Right.isMemberAccess() || 3863 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 3864 tok::colon, tok::l_square, tok::at) || 3865 (Style.BraceWrapping.BeforeLambdaBody && Right.is(tok::l_brace)) || 3866 (Left.is(tok::r_paren) && 3867 Right.isOneOf(tok::identifier, tok::kw_const)) || 3868 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 3869 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 3870 } 3871 3872 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 3873 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 3874 const FormatToken *Tok = Line.First; 3875 while (Tok) { 3876 llvm::errs() << " M=" << Tok->MustBreakBefore 3877 << " C=" << Tok->CanBreakBefore 3878 << " T=" << getTokenTypeName(Tok->Type) 3879 << " S=" << Tok->SpacesRequiredBefore 3880 << " B=" << Tok->BlockParameterCount 3881 << " BK=" << Tok->BlockKind << " P=" << Tok->SplitPenalty 3882 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 3883 << " PPK=" << Tok->PackingKind << " FakeLParens="; 3884 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 3885 llvm::errs() << Tok->FakeLParens[i] << "/"; 3886 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 3887 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo(); 3888 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 3889 if (!Tok->Next) 3890 assert(Tok == Line.Last); 3891 Tok = Tok->Next; 3892 } 3893 llvm::errs() << "----\n"; 3894 } 3895 3896 } // namespace format 3897 } // namespace clang 3898