1 //===--- TokenAnnotator.cpp - Format C++ code -----------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// 9 /// \file 10 /// This file implements a token annotator, i.e. creates 11 /// \c AnnotatedTokens out of \c FormatTokens with required extra information. 12 /// 13 //===----------------------------------------------------------------------===// 14 15 #include "TokenAnnotator.h" 16 #include "FormatToken.h" 17 #include "clang/Basic/SourceManager.h" 18 #include "llvm/ADT/SmallPtrSet.h" 19 #include "llvm/Support/Debug.h" 20 21 #define DEBUG_TYPE "format-token-annotator" 22 23 namespace clang { 24 namespace format { 25 26 namespace { 27 28 /// Returns \c true if the token can be used as an identifier in 29 /// an Objective-C \c @selector, \c false otherwise. 30 /// 31 /// Because getFormattingLangOpts() always lexes source code as 32 /// Objective-C++, C++ keywords like \c new and \c delete are 33 /// lexed as tok::kw_*, not tok::identifier, even for Objective-C. 34 /// 35 /// For Objective-C and Objective-C++, both identifiers and keywords 36 /// are valid inside @selector(...) (or a macro which 37 /// invokes @selector(...)). So, we allow treat any identifier or 38 /// keyword as a potential Objective-C selector component. 39 static bool canBeObjCSelectorComponent(const FormatToken &Tok) { 40 return Tok.Tok.getIdentifierInfo() != nullptr; 41 } 42 43 /// A parser that gathers additional information about tokens. 44 /// 45 /// The \c TokenAnnotator tries to match parenthesis and square brakets and 46 /// store a parenthesis levels. It also tries to resolve matching "<" and ">" 47 /// into template parameter lists. 48 class AnnotatingParser { 49 public: 50 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line, 51 const AdditionalKeywords &Keywords) 52 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false), 53 Keywords(Keywords) { 54 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false)); 55 resetTokenMetadata(CurrentToken); 56 } 57 58 private: 59 bool parseAngle() { 60 if (!CurrentToken || !CurrentToken->Previous) 61 return false; 62 if (NonTemplateLess.count(CurrentToken->Previous)) 63 return false; 64 65 const FormatToken &Previous = *CurrentToken->Previous; // The '<'. 66 if (Previous.Previous) { 67 if (Previous.Previous->Tok.isLiteral()) 68 return false; 69 if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 && 70 (!Previous.Previous->MatchingParen || 71 !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen))) 72 return false; 73 } 74 75 FormatToken *Left = CurrentToken->Previous; 76 Left->ParentBracket = Contexts.back().ContextKind; 77 ScopedContextCreator ContextCreator(*this, tok::less, 12); 78 79 // If this angle is in the context of an expression, we need to be more 80 // hesitant to detect it as opening template parameters. 81 bool InExprContext = Contexts.back().IsExpression; 82 83 Contexts.back().IsExpression = false; 84 // If there's a template keyword before the opening angle bracket, this is a 85 // template parameter, not an argument. 86 Contexts.back().InTemplateArgument = 87 Left->Previous && Left->Previous->Tok.isNot(tok::kw_template); 88 89 if (Style.Language == FormatStyle::LK_Java && 90 CurrentToken->is(tok::question)) 91 next(); 92 93 while (CurrentToken) { 94 if (CurrentToken->is(tok::greater)) { 95 Left->MatchingParen = CurrentToken; 96 CurrentToken->MatchingParen = Left; 97 // In TT_Proto, we must distignuish between: 98 // map<key, value> 99 // msg < item: data > 100 // msg: < item: data > 101 // In TT_TextProto, map<key, value> does not occur. 102 if (Style.Language == FormatStyle::LK_TextProto || 103 (Style.Language == FormatStyle::LK_Proto && Left->Previous && 104 Left->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) 105 CurrentToken->Type = TT_DictLiteral; 106 else 107 CurrentToken->Type = TT_TemplateCloser; 108 next(); 109 return true; 110 } 111 if (CurrentToken->is(tok::question) && 112 Style.Language == FormatStyle::LK_Java) { 113 next(); 114 continue; 115 } 116 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) || 117 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext && 118 Style.Language != FormatStyle::LK_Proto && 119 Style.Language != FormatStyle::LK_TextProto)) 120 return false; 121 // If a && or || is found and interpreted as a binary operator, this set 122 // of angles is likely part of something like "a < b && c > d". If the 123 // angles are inside an expression, the ||/&& might also be a binary 124 // operator that was misinterpreted because we are parsing template 125 // parameters. 126 // FIXME: This is getting out of hand, write a decent parser. 127 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) && 128 CurrentToken->Previous->is(TT_BinaryOperator) && 129 Contexts[Contexts.size() - 2].IsExpression && 130 !Line.startsWith(tok::kw_template)) 131 return false; 132 updateParameterCount(Left, CurrentToken); 133 if (Style.Language == FormatStyle::LK_Proto) { 134 if (FormatToken *Previous = CurrentToken->getPreviousNonComment()) { 135 if (CurrentToken->is(tok::colon) || 136 (CurrentToken->isOneOf(tok::l_brace, tok::less) && 137 Previous->isNot(tok::colon))) 138 Previous->Type = TT_SelectorName; 139 } 140 } 141 if (!consumeToken()) 142 return false; 143 } 144 return false; 145 } 146 147 bool parseParens(bool LookForDecls = false) { 148 if (!CurrentToken) 149 return false; 150 FormatToken *Left = CurrentToken->Previous; 151 Left->ParentBracket = Contexts.back().ContextKind; 152 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1); 153 154 // FIXME: This is a bit of a hack. Do better. 155 Contexts.back().ColonIsForRangeExpr = 156 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr; 157 158 bool StartsObjCMethodExpr = false; 159 if (FormatToken *MaybeSel = Left->Previous) { 160 // @selector( starts a selector. 161 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous && 162 MaybeSel->Previous->is(tok::at)) { 163 StartsObjCMethodExpr = true; 164 } 165 } 166 167 if (Left->is(TT_OverloadedOperatorLParen)) { 168 Contexts.back().IsExpression = false; 169 } else if (Style.Language == FormatStyle::LK_JavaScript && 170 (Line.startsWith(Keywords.kw_type, tok::identifier) || 171 Line.startsWith(tok::kw_export, Keywords.kw_type, 172 tok::identifier))) { 173 // type X = (...); 174 // export type X = (...); 175 Contexts.back().IsExpression = false; 176 } else if (Left->Previous && 177 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_decltype, 178 tok::kw_while, tok::l_paren, 179 tok::comma) || 180 Left->Previous->isIf() || 181 Left->Previous->is(TT_BinaryOperator))) { 182 // static_assert, if and while usually contain expressions. 183 Contexts.back().IsExpression = true; 184 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 185 (Left->Previous->is(Keywords.kw_function) || 186 (Left->Previous->endsSequence(tok::identifier, 187 Keywords.kw_function)))) { 188 // function(...) or function f(...) 189 Contexts.back().IsExpression = false; 190 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 191 Left->Previous->is(TT_JsTypeColon)) { 192 // let x: (SomeType); 193 Contexts.back().IsExpression = false; 194 } else if (Left->Previous && Left->Previous->is(tok::r_square) && 195 Left->Previous->MatchingParen && 196 Left->Previous->MatchingParen->is(TT_LambdaLSquare)) { 197 // This is a parameter list of a lambda expression. 198 Contexts.back().IsExpression = false; 199 } else if (Line.InPPDirective && 200 (!Left->Previous || !Left->Previous->is(tok::identifier))) { 201 Contexts.back().IsExpression = true; 202 } else if (Contexts[Contexts.size() - 2].CaretFound) { 203 // This is the parameter list of an ObjC block. 204 Contexts.back().IsExpression = false; 205 } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) { 206 Left->Type = TT_AttributeParen; 207 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) { 208 // The first argument to a foreach macro is a declaration. 209 Contexts.back().IsForEachMacro = true; 210 Contexts.back().IsExpression = false; 211 } else if (Left->Previous && Left->Previous->MatchingParen && 212 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) { 213 Contexts.back().IsExpression = false; 214 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) { 215 bool IsForOrCatch = 216 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch); 217 Contexts.back().IsExpression = !IsForOrCatch; 218 } 219 220 if (StartsObjCMethodExpr) { 221 Contexts.back().ColonIsObjCMethodExpr = true; 222 Left->Type = TT_ObjCMethodExpr; 223 } 224 225 // MightBeFunctionType and ProbablyFunctionType are used for 226 // function pointer and reference types as well as Objective-C 227 // block types: 228 // 229 // void (*FunctionPointer)(void); 230 // void (&FunctionReference)(void); 231 // void (^ObjCBlock)(void); 232 bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression; 233 bool ProbablyFunctionType = 234 CurrentToken->isOneOf(tok::star, tok::amp, tok::caret); 235 bool HasMultipleLines = false; 236 bool HasMultipleParametersOnALine = false; 237 bool MightBeObjCForRangeLoop = 238 Left->Previous && Left->Previous->is(tok::kw_for); 239 FormatToken *PossibleObjCForInToken = nullptr; 240 while (CurrentToken) { 241 // LookForDecls is set when "if (" has been seen. Check for 242 // 'identifier' '*' 'identifier' followed by not '=' -- this 243 // '*' has to be a binary operator but determineStarAmpUsage() will 244 // categorize it as an unary operator, so set the right type here. 245 if (LookForDecls && CurrentToken->Next) { 246 FormatToken *Prev = CurrentToken->getPreviousNonComment(); 247 if (Prev) { 248 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 249 FormatToken *Next = CurrentToken->Next; 250 if (PrevPrev && PrevPrev->is(tok::identifier) && 251 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) && 252 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) { 253 Prev->Type = TT_BinaryOperator; 254 LookForDecls = false; 255 } 256 } 257 } 258 259 if (CurrentToken->Previous->is(TT_PointerOrReference) && 260 CurrentToken->Previous->Previous->isOneOf(tok::l_paren, 261 tok::coloncolon)) 262 ProbablyFunctionType = true; 263 if (CurrentToken->is(tok::comma)) 264 MightBeFunctionType = false; 265 if (CurrentToken->Previous->is(TT_BinaryOperator)) 266 Contexts.back().IsExpression = true; 267 if (CurrentToken->is(tok::r_paren)) { 268 if (MightBeFunctionType && ProbablyFunctionType && CurrentToken->Next && 269 (CurrentToken->Next->is(tok::l_paren) || 270 (CurrentToken->Next->is(tok::l_square) && Line.MustBeDeclaration))) 271 Left->Type = Left->Next->is(tok::caret) ? TT_ObjCBlockLParen 272 : TT_FunctionTypeLParen; 273 Left->MatchingParen = CurrentToken; 274 CurrentToken->MatchingParen = Left; 275 276 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) && 277 Left->Previous && Left->Previous->is(tok::l_paren)) { 278 // Detect the case where macros are used to generate lambdas or 279 // function bodies, e.g.: 280 // auto my_lambda = MARCO((Type *type, int i) { .. body .. }); 281 for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) { 282 if (Tok->is(TT_BinaryOperator) && 283 Tok->isOneOf(tok::star, tok::amp, tok::ampamp)) 284 Tok->Type = TT_PointerOrReference; 285 } 286 } 287 288 if (StartsObjCMethodExpr) { 289 CurrentToken->Type = TT_ObjCMethodExpr; 290 if (Contexts.back().FirstObjCSelectorName) { 291 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 292 Contexts.back().LongestObjCSelectorName; 293 } 294 } 295 296 if (Left->is(TT_AttributeParen)) 297 CurrentToken->Type = TT_AttributeParen; 298 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation)) 299 CurrentToken->Type = TT_JavaAnnotation; 300 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation)) 301 CurrentToken->Type = TT_LeadingJavaAnnotation; 302 if (Left->Previous && Left->Previous->is(TT_AttributeSquare)) 303 CurrentToken->Type = TT_AttributeSquare; 304 305 if (!HasMultipleLines) 306 Left->PackingKind = PPK_Inconclusive; 307 else if (HasMultipleParametersOnALine) 308 Left->PackingKind = PPK_BinPacked; 309 else 310 Left->PackingKind = PPK_OnePerLine; 311 312 next(); 313 return true; 314 } 315 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace)) 316 return false; 317 318 if (CurrentToken->is(tok::l_brace)) 319 Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen 320 if (CurrentToken->is(tok::comma) && CurrentToken->Next && 321 !CurrentToken->Next->HasUnescapedNewline && 322 !CurrentToken->Next->isTrailingComment()) 323 HasMultipleParametersOnALine = true; 324 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) || 325 CurrentToken->Previous->isSimpleTypeSpecifier()) && 326 !CurrentToken->is(tok::l_brace)) 327 Contexts.back().IsExpression = false; 328 if (CurrentToken->isOneOf(tok::semi, tok::colon)) { 329 MightBeObjCForRangeLoop = false; 330 if (PossibleObjCForInToken) { 331 PossibleObjCForInToken->Type = TT_Unknown; 332 PossibleObjCForInToken = nullptr; 333 } 334 } 335 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) { 336 PossibleObjCForInToken = CurrentToken; 337 PossibleObjCForInToken->Type = TT_ObjCForIn; 338 } 339 // When we discover a 'new', we set CanBeExpression to 'false' in order to 340 // parse the type correctly. Reset that after a comma. 341 if (CurrentToken->is(tok::comma)) 342 Contexts.back().CanBeExpression = true; 343 344 FormatToken *Tok = CurrentToken; 345 if (!consumeToken()) 346 return false; 347 updateParameterCount(Left, Tok); 348 if (CurrentToken && CurrentToken->HasUnescapedNewline) 349 HasMultipleLines = true; 350 } 351 return false; 352 } 353 354 bool isCSharpAttributeSpecifier(const FormatToken &Tok) { 355 if (!Style.isCSharp()) 356 return false; 357 358 const FormatToken *AttrTok = Tok.Next; 359 if (!AttrTok) 360 return false; 361 362 // Just an empty declaration e.g. string []. 363 if (AttrTok->is(tok::r_square)) 364 return false; 365 366 // Move along the tokens inbetween the '[' and ']' e.g. [STAThread]. 367 while (AttrTok && AttrTok->isNot(tok::r_square)) { 368 AttrTok = AttrTok->Next; 369 } 370 371 if (!AttrTok) 372 return false; 373 374 // Move past the end of ']'. 375 AttrTok = AttrTok->Next; 376 if (!AttrTok) 377 return false; 378 379 // Limit this to being an access modifier that follows. 380 if (AttrTok->isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 381 tok::kw_class, tok::kw_static, tok::l_square, 382 Keywords.kw_internal)) { 383 return true; 384 } 385 return false; 386 } 387 388 bool isCpp11AttributeSpecifier(const FormatToken &Tok) { 389 if (!Style.isCpp() || !Tok.startsSequence(tok::l_square, tok::l_square)) 390 return false; 391 // The first square bracket is part of an ObjC array literal 392 if (Tok.Previous && Tok.Previous->is(tok::at)) { 393 return false; 394 } 395 const FormatToken *AttrTok = Tok.Next->Next; 396 if (!AttrTok) 397 return false; 398 // C++17 '[[using ns: foo, bar(baz, blech)]]' 399 // We assume nobody will name an ObjC variable 'using'. 400 if (AttrTok->startsSequence(tok::kw_using, tok::identifier, tok::colon)) 401 return true; 402 if (AttrTok->isNot(tok::identifier)) 403 return false; 404 while (AttrTok && !AttrTok->startsSequence(tok::r_square, tok::r_square)) { 405 // ObjC message send. We assume nobody will use : in a C++11 attribute 406 // specifier parameter, although this is technically valid: 407 // [[foo(:)]]. 408 if (AttrTok->is(tok::colon) || 409 AttrTok->startsSequence(tok::identifier, tok::identifier) || 410 AttrTok->startsSequence(tok::r_paren, tok::identifier)) 411 return false; 412 if (AttrTok->is(tok::ellipsis)) 413 return true; 414 AttrTok = AttrTok->Next; 415 } 416 return AttrTok && AttrTok->startsSequence(tok::r_square, tok::r_square); 417 } 418 419 bool parseSquare() { 420 if (!CurrentToken) 421 return false; 422 423 // A '[' could be an index subscript (after an identifier or after 424 // ')' or ']'), it could be the start of an Objective-C method 425 // expression, it could the start of an Objective-C array literal, 426 // or it could be a C++ attribute specifier [[foo::bar]]. 427 FormatToken *Left = CurrentToken->Previous; 428 Left->ParentBracket = Contexts.back().ContextKind; 429 FormatToken *Parent = Left->getPreviousNonComment(); 430 431 // Cases where '>' is followed by '['. 432 // In C++, this can happen either in array of templates (foo<int>[10]) 433 // or when array is a nested template type (unique_ptr<type1<type2>[]>). 434 bool CppArrayTemplates = 435 Style.isCpp() && Parent && Parent->is(TT_TemplateCloser) && 436 (Contexts.back().CanBeExpression || Contexts.back().IsExpression || 437 Contexts.back().InTemplateArgument); 438 439 bool IsCpp11AttributeSpecifier = isCpp11AttributeSpecifier(*Left) || 440 Contexts.back().InCpp11AttributeSpecifier; 441 442 // Treat C# Attributes [STAThread] much like C++ attributes [[...]]. 443 bool IsCSharp11AttributeSpecifier = 444 isCSharpAttributeSpecifier(*Left) || 445 Contexts.back().InCSharpAttributeSpecifier; 446 447 bool InsideInlineASM = Line.startsWith(tok::kw_asm); 448 bool IsCppStructuredBinding = Left->isCppStructuredBinding(Style); 449 bool StartsObjCMethodExpr = 450 !IsCppStructuredBinding && !InsideInlineASM && !CppArrayTemplates && 451 Style.isCpp() && !IsCpp11AttributeSpecifier && 452 Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) && 453 !CurrentToken->isOneOf(tok::l_brace, tok::r_square) && 454 (!Parent || 455 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren, 456 tok::kw_return, tok::kw_throw) || 457 Parent->isUnaryOperator() || 458 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 459 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) || 460 (getBinOpPrecedence(Parent->Tok.getKind(), true, true) > 461 prec::Unknown)); 462 bool ColonFound = false; 463 464 unsigned BindingIncrease = 1; 465 if (IsCppStructuredBinding) { 466 Left->Type = TT_StructuredBindingLSquare; 467 } else if (Left->is(TT_Unknown)) { 468 if (StartsObjCMethodExpr) { 469 Left->Type = TT_ObjCMethodExpr; 470 } else if (IsCpp11AttributeSpecifier) { 471 Left->Type = TT_AttributeSquare; 472 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent && 473 Contexts.back().ContextKind == tok::l_brace && 474 Parent->isOneOf(tok::l_brace, tok::comma)) { 475 Left->Type = TT_JsComputedPropertyName; 476 } else if (Style.isCpp() && Contexts.back().ContextKind == tok::l_brace && 477 Parent && Parent->isOneOf(tok::l_brace, tok::comma)) { 478 Left->Type = TT_DesignatedInitializerLSquare; 479 } else if (CurrentToken->is(tok::r_square) && Parent && 480 Parent->is(TT_TemplateCloser)) { 481 Left->Type = TT_ArraySubscriptLSquare; 482 } else if (Style.Language == FormatStyle::LK_Proto || 483 Style.Language == FormatStyle::LK_TextProto) { 484 // Square braces in LK_Proto can either be message field attributes: 485 // 486 // optional Aaa aaa = 1 [ 487 // (aaa) = aaa 488 // ]; 489 // 490 // extensions 123 [ 491 // (aaa) = aaa 492 // ]; 493 // 494 // or text proto extensions (in options): 495 // 496 // option (Aaa.options) = { 497 // [type.type/type] { 498 // key: value 499 // } 500 // } 501 // 502 // or repeated fields (in options): 503 // 504 // option (Aaa.options) = { 505 // keys: [ 1, 2, 3 ] 506 // } 507 // 508 // In the first and the third case we want to spread the contents inside 509 // the square braces; in the second we want to keep them inline. 510 Left->Type = TT_ArrayInitializerLSquare; 511 if (!Left->endsSequence(tok::l_square, tok::numeric_constant, 512 tok::equal) && 513 !Left->endsSequence(tok::l_square, tok::numeric_constant, 514 tok::identifier) && 515 !Left->endsSequence(tok::l_square, tok::colon, TT_SelectorName)) { 516 Left->Type = TT_ProtoExtensionLSquare; 517 BindingIncrease = 10; 518 } 519 } else if (!CppArrayTemplates && Parent && 520 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at, 521 tok::comma, tok::l_paren, tok::l_square, 522 tok::question, tok::colon, tok::kw_return, 523 // Should only be relevant to JavaScript: 524 tok::kw_default)) { 525 Left->Type = TT_ArrayInitializerLSquare; 526 } else if (IsCSharp11AttributeSpecifier) { 527 Left->Type = TT_AttributeSquare; 528 } else { 529 BindingIncrease = 10; 530 Left->Type = TT_ArraySubscriptLSquare; 531 } 532 } 533 534 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease); 535 Contexts.back().IsExpression = true; 536 if (Style.Language == FormatStyle::LK_JavaScript && Parent && 537 Parent->is(TT_JsTypeColon)) 538 Contexts.back().IsExpression = false; 539 540 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr; 541 Contexts.back().InCpp11AttributeSpecifier = IsCpp11AttributeSpecifier; 542 Contexts.back().InCSharpAttributeSpecifier = IsCSharp11AttributeSpecifier; 543 544 while (CurrentToken) { 545 if (CurrentToken->is(tok::r_square)) { 546 if (IsCpp11AttributeSpecifier) 547 CurrentToken->Type = TT_AttributeSquare; 548 if (IsCSharp11AttributeSpecifier) 549 CurrentToken->Type = TT_AttributeSquare; 550 else if (((CurrentToken->Next && 551 CurrentToken->Next->is(tok::l_paren)) || 552 (CurrentToken->Previous && 553 CurrentToken->Previous->Previous == Left)) && 554 Left->is(TT_ObjCMethodExpr)) { 555 // An ObjC method call is rarely followed by an open parenthesis. It 556 // also can't be composed of just one token, unless it's a macro that 557 // will be expanded to more tokens. 558 // FIXME: Do we incorrectly label ":" with this? 559 StartsObjCMethodExpr = false; 560 Left->Type = TT_Unknown; 561 } 562 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) { 563 CurrentToken->Type = TT_ObjCMethodExpr; 564 // If we haven't seen a colon yet, make sure the last identifier 565 // before the r_square is tagged as a selector name component. 566 if (!ColonFound && CurrentToken->Previous && 567 CurrentToken->Previous->is(TT_Unknown) && 568 canBeObjCSelectorComponent(*CurrentToken->Previous)) 569 CurrentToken->Previous->Type = TT_SelectorName; 570 // determineStarAmpUsage() thinks that '*' '[' is allocating an 571 // array of pointers, but if '[' starts a selector then '*' is a 572 // binary operator. 573 if (Parent && Parent->is(TT_PointerOrReference)) 574 Parent->Type = TT_BinaryOperator; 575 } 576 // An arrow after an ObjC method expression is not a lambda arrow. 577 if (CurrentToken->Type == TT_ObjCMethodExpr && CurrentToken->Next && 578 CurrentToken->Next->is(TT_LambdaArrow)) 579 CurrentToken->Next->Type = TT_Unknown; 580 Left->MatchingParen = CurrentToken; 581 CurrentToken->MatchingParen = Left; 582 // FirstObjCSelectorName is set when a colon is found. This does 583 // not work, however, when the method has no parameters. 584 // Here, we set FirstObjCSelectorName when the end of the method call is 585 // reached, in case it was not set already. 586 if (!Contexts.back().FirstObjCSelectorName) { 587 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 588 if (Previous && Previous->is(TT_SelectorName)) { 589 Previous->ObjCSelectorNameParts = 1; 590 Contexts.back().FirstObjCSelectorName = Previous; 591 } 592 } else { 593 Left->ParameterCount = 594 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 595 } 596 if (Contexts.back().FirstObjCSelectorName) { 597 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 598 Contexts.back().LongestObjCSelectorName; 599 if (Left->BlockParameterCount > 1) 600 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0; 601 } 602 next(); 603 return true; 604 } 605 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace)) 606 return false; 607 if (CurrentToken->is(tok::colon)) { 608 if (IsCpp11AttributeSpecifier && 609 CurrentToken->endsSequence(tok::colon, tok::identifier, 610 tok::kw_using)) { 611 // Remember that this is a [[using ns: foo]] C++ attribute, so we 612 // don't add a space before the colon (unlike other colons). 613 CurrentToken->Type = TT_AttributeColon; 614 } else if (Left->isOneOf(TT_ArraySubscriptLSquare, 615 TT_DesignatedInitializerLSquare)) { 616 Left->Type = TT_ObjCMethodExpr; 617 StartsObjCMethodExpr = true; 618 Contexts.back().ColonIsObjCMethodExpr = true; 619 if (Parent && Parent->is(tok::r_paren)) 620 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 621 Parent->Type = TT_CastRParen; 622 } 623 ColonFound = true; 624 } 625 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) && 626 !ColonFound) 627 Left->Type = TT_ArrayInitializerLSquare; 628 FormatToken *Tok = CurrentToken; 629 if (!consumeToken()) 630 return false; 631 updateParameterCount(Left, Tok); 632 } 633 return false; 634 } 635 636 bool parseBrace() { 637 if (CurrentToken) { 638 FormatToken *Left = CurrentToken->Previous; 639 Left->ParentBracket = Contexts.back().ContextKind; 640 641 if (Contexts.back().CaretFound) 642 Left->Type = TT_ObjCBlockLBrace; 643 Contexts.back().CaretFound = false; 644 645 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1); 646 Contexts.back().ColonIsDictLiteral = true; 647 if (Left->BlockKind == BK_BracedInit) 648 Contexts.back().IsExpression = true; 649 if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 650 Left->Previous->is(TT_JsTypeColon)) 651 Contexts.back().IsExpression = false; 652 653 while (CurrentToken) { 654 if (CurrentToken->is(tok::r_brace)) { 655 Left->MatchingParen = CurrentToken; 656 CurrentToken->MatchingParen = Left; 657 next(); 658 return true; 659 } 660 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 661 return false; 662 updateParameterCount(Left, CurrentToken); 663 if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) { 664 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 665 if (Previous->is(TT_JsTypeOptionalQuestion)) 666 Previous = Previous->getPreviousNonComment(); 667 if ((CurrentToken->is(tok::colon) && 668 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) || 669 Style.Language == FormatStyle::LK_Proto || 670 Style.Language == FormatStyle::LK_TextProto) { 671 Left->Type = TT_DictLiteral; 672 if (Previous->Tok.getIdentifierInfo() || 673 Previous->is(tok::string_literal)) 674 Previous->Type = TT_SelectorName; 675 } 676 if (CurrentToken->is(tok::colon) || 677 Style.Language == FormatStyle::LK_JavaScript) 678 Left->Type = TT_DictLiteral; 679 } 680 if (CurrentToken->is(tok::comma) && 681 Style.Language == FormatStyle::LK_JavaScript) 682 Left->Type = TT_DictLiteral; 683 if (!consumeToken()) 684 return false; 685 } 686 } 687 return true; 688 } 689 690 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 691 // For ObjC methods, the number of parameters is calculated differently as 692 // method declarations have a different structure (the parameters are not 693 // inside a bracket scope). 694 if (Current->is(tok::l_brace) && Current->BlockKind == BK_Block) 695 ++Left->BlockParameterCount; 696 if (Current->is(tok::comma)) { 697 ++Left->ParameterCount; 698 if (!Left->Role) 699 Left->Role.reset(new CommaSeparatedList(Style)); 700 Left->Role->CommaFound(Current); 701 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 702 Left->ParameterCount = 1; 703 } 704 } 705 706 bool parseConditional() { 707 while (CurrentToken) { 708 if (CurrentToken->is(tok::colon)) { 709 CurrentToken->Type = TT_ConditionalExpr; 710 next(); 711 return true; 712 } 713 if (!consumeToken()) 714 return false; 715 } 716 return false; 717 } 718 719 bool parseTemplateDeclaration() { 720 if (CurrentToken && CurrentToken->is(tok::less)) { 721 CurrentToken->Type = TT_TemplateOpener; 722 next(); 723 if (!parseAngle()) 724 return false; 725 if (CurrentToken) 726 CurrentToken->Previous->ClosesTemplateDeclaration = true; 727 return true; 728 } 729 return false; 730 } 731 732 bool consumeToken() { 733 FormatToken *Tok = CurrentToken; 734 next(); 735 switch (Tok->Tok.getKind()) { 736 case tok::plus: 737 case tok::minus: 738 if (!Tok->Previous && Line.MustBeDeclaration) 739 Tok->Type = TT_ObjCMethodSpecifier; 740 break; 741 case tok::colon: 742 if (!Tok->Previous) 743 return false; 744 // Colons from ?: are handled in parseConditional(). 745 if (Style.Language == FormatStyle::LK_JavaScript) { 746 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop 747 (Contexts.size() == 1 && // switch/case labels 748 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) || 749 Contexts.back().ContextKind == tok::l_paren || // function params 750 Contexts.back().ContextKind == tok::l_square || // array type 751 (!Contexts.back().IsExpression && 752 Contexts.back().ContextKind == tok::l_brace) || // object type 753 (Contexts.size() == 1 && 754 Line.MustBeDeclaration)) { // method/property declaration 755 Contexts.back().IsExpression = false; 756 Tok->Type = TT_JsTypeColon; 757 break; 758 } 759 } 760 if (Contexts.back().ColonIsDictLiteral || 761 Style.Language == FormatStyle::LK_Proto || 762 Style.Language == FormatStyle::LK_TextProto) { 763 Tok->Type = TT_DictLiteral; 764 if (Style.Language == FormatStyle::LK_TextProto) { 765 if (FormatToken *Previous = Tok->getPreviousNonComment()) 766 Previous->Type = TT_SelectorName; 767 } 768 } else if (Contexts.back().ColonIsObjCMethodExpr || 769 Line.startsWith(TT_ObjCMethodSpecifier)) { 770 Tok->Type = TT_ObjCMethodExpr; 771 const FormatToken *BeforePrevious = Tok->Previous->Previous; 772 // Ensure we tag all identifiers in method declarations as 773 // TT_SelectorName. 774 bool UnknownIdentifierInMethodDeclaration = 775 Line.startsWith(TT_ObjCMethodSpecifier) && 776 Tok->Previous->is(tok::identifier) && Tok->Previous->is(TT_Unknown); 777 if (!BeforePrevious || 778 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 779 !(BeforePrevious->is(TT_CastRParen) || 780 (BeforePrevious->is(TT_ObjCMethodExpr) && 781 BeforePrevious->is(tok::colon))) || 782 BeforePrevious->is(tok::r_square) || 783 Contexts.back().LongestObjCSelectorName == 0 || 784 UnknownIdentifierInMethodDeclaration) { 785 Tok->Previous->Type = TT_SelectorName; 786 if (!Contexts.back().FirstObjCSelectorName) 787 Contexts.back().FirstObjCSelectorName = Tok->Previous; 788 else if (Tok->Previous->ColumnWidth > 789 Contexts.back().LongestObjCSelectorName) 790 Contexts.back().LongestObjCSelectorName = 791 Tok->Previous->ColumnWidth; 792 Tok->Previous->ParameterIndex = 793 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 794 ++Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 795 } 796 } else if (Contexts.back().ColonIsForRangeExpr) { 797 Tok->Type = TT_RangeBasedForLoopColon; 798 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 799 Tok->Type = TT_BitFieldColon; 800 } else if (Contexts.size() == 1 && 801 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) { 802 if (Tok->getPreviousNonComment()->isOneOf(tok::r_paren, 803 tok::kw_noexcept)) 804 Tok->Type = TT_CtorInitializerColon; 805 else 806 Tok->Type = TT_InheritanceColon; 807 } else if (canBeObjCSelectorComponent(*Tok->Previous) && Tok->Next && 808 (Tok->Next->isOneOf(tok::r_paren, tok::comma) || 809 (canBeObjCSelectorComponent(*Tok->Next) && Tok->Next->Next && 810 Tok->Next->Next->is(tok::colon)))) { 811 // This handles a special macro in ObjC code where selectors including 812 // the colon are passed as macro arguments. 813 Tok->Type = TT_ObjCMethodExpr; 814 } else if (Contexts.back().ContextKind == tok::l_paren) { 815 Tok->Type = TT_InlineASMColon; 816 } 817 break; 818 case tok::pipe: 819 case tok::amp: 820 // | and & in declarations/type expressions represent union and 821 // intersection types, respectively. 822 if (Style.Language == FormatStyle::LK_JavaScript && 823 !Contexts.back().IsExpression) 824 Tok->Type = TT_JsTypeOperator; 825 break; 826 case tok::kw_if: 827 case tok::kw_while: 828 assert(!Line.startsWith(tok::hash)); 829 if (Tok->is(tok::kw_if) && CurrentToken && 830 CurrentToken->isOneOf(tok::kw_constexpr, tok::identifier)) 831 next(); 832 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 833 next(); 834 if (!parseParens(/*LookForDecls=*/true)) 835 return false; 836 } 837 break; 838 case tok::kw_for: 839 if (Style.Language == FormatStyle::LK_JavaScript) { 840 // x.for and {for: ...} 841 if ((Tok->Previous && Tok->Previous->is(tok::period)) || 842 (Tok->Next && Tok->Next->is(tok::colon))) 843 break; 844 // JS' for await ( ... 845 if (CurrentToken && CurrentToken->is(Keywords.kw_await)) 846 next(); 847 } 848 Contexts.back().ColonIsForRangeExpr = true; 849 next(); 850 if (!parseParens()) 851 return false; 852 break; 853 case tok::l_paren: 854 // When faced with 'operator()()', the kw_operator handler incorrectly 855 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make 856 // the first two parens OverloadedOperators and the second l_paren an 857 // OverloadedOperatorLParen. 858 if (Tok->Previous && Tok->Previous->is(tok::r_paren) && 859 Tok->Previous->MatchingParen && 860 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) { 861 Tok->Previous->Type = TT_OverloadedOperator; 862 Tok->Previous->MatchingParen->Type = TT_OverloadedOperator; 863 Tok->Type = TT_OverloadedOperatorLParen; 864 } 865 866 if (!parseParens()) 867 return false; 868 if (Line.MustBeDeclaration && Contexts.size() == 1 && 869 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) && 870 (!Tok->Previous || 871 !Tok->Previous->isOneOf(tok::kw_decltype, tok::kw___attribute, 872 TT_LeadingJavaAnnotation))) 873 Line.MightBeFunctionDecl = true; 874 break; 875 case tok::l_square: 876 if (!parseSquare()) 877 return false; 878 break; 879 case tok::l_brace: 880 if (Style.Language == FormatStyle::LK_TextProto) { 881 FormatToken *Previous = Tok->getPreviousNonComment(); 882 if (Previous && Previous->Type != TT_DictLiteral) 883 Previous->Type = TT_SelectorName; 884 } 885 if (!parseBrace()) 886 return false; 887 break; 888 case tok::less: 889 if (parseAngle()) { 890 Tok->Type = TT_TemplateOpener; 891 // In TT_Proto, we must distignuish between: 892 // map<key, value> 893 // msg < item: data > 894 // msg: < item: data > 895 // In TT_TextProto, map<key, value> does not occur. 896 if (Style.Language == FormatStyle::LK_TextProto || 897 (Style.Language == FormatStyle::LK_Proto && Tok->Previous && 898 Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) { 899 Tok->Type = TT_DictLiteral; 900 FormatToken *Previous = Tok->getPreviousNonComment(); 901 if (Previous && Previous->Type != TT_DictLiteral) 902 Previous->Type = TT_SelectorName; 903 } 904 } else { 905 Tok->Type = TT_BinaryOperator; 906 NonTemplateLess.insert(Tok); 907 CurrentToken = Tok; 908 next(); 909 } 910 break; 911 case tok::r_paren: 912 case tok::r_square: 913 return false; 914 case tok::r_brace: 915 // Lines can start with '}'. 916 if (Tok->Previous) 917 return false; 918 break; 919 case tok::greater: 920 if (Style.Language != FormatStyle::LK_TextProto) 921 Tok->Type = TT_BinaryOperator; 922 if (Tok->Previous && Tok->Previous->is(TT_TemplateCloser)) 923 Tok->SpacesRequiredBefore = 1; 924 break; 925 case tok::kw_operator: 926 if (Style.Language == FormatStyle::LK_TextProto || 927 Style.Language == FormatStyle::LK_Proto) 928 break; 929 while (CurrentToken && 930 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 931 if (CurrentToken->isOneOf(tok::star, tok::amp)) 932 CurrentToken->Type = TT_PointerOrReference; 933 consumeToken(); 934 if (CurrentToken && 935 CurrentToken->Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator, 936 tok::comma)) 937 CurrentToken->Previous->Type = TT_OverloadedOperator; 938 } 939 if (CurrentToken) { 940 CurrentToken->Type = TT_OverloadedOperatorLParen; 941 if (CurrentToken->Previous->is(TT_BinaryOperator)) 942 CurrentToken->Previous->Type = TT_OverloadedOperator; 943 } 944 break; 945 case tok::question: 946 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next && 947 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren, 948 tok::r_brace)) { 949 // Question marks before semicolons, colons, etc. indicate optional 950 // types (fields, parameters), e.g. 951 // function(x?: string, y?) {...} 952 // class X { y?; } 953 Tok->Type = TT_JsTypeOptionalQuestion; 954 break; 955 } 956 // Declarations cannot be conditional expressions, this can only be part 957 // of a type declaration. 958 if (Line.MustBeDeclaration && !Contexts.back().IsExpression && 959 Style.Language == FormatStyle::LK_JavaScript) 960 break; 961 parseConditional(); 962 break; 963 case tok::kw_template: 964 parseTemplateDeclaration(); 965 break; 966 case tok::comma: 967 if (Contexts.back().InCtorInitializer) 968 Tok->Type = TT_CtorInitializerComma; 969 else if (Contexts.back().InInheritanceList) 970 Tok->Type = TT_InheritanceComma; 971 else if (Contexts.back().FirstStartOfName && 972 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) { 973 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 974 Line.IsMultiVariableDeclStmt = true; 975 } 976 if (Contexts.back().IsForEachMacro) 977 Contexts.back().IsExpression = true; 978 break; 979 case tok::identifier: 980 if (Tok->isOneOf(Keywords.kw___has_include, 981 Keywords.kw___has_include_next)) { 982 parseHasInclude(); 983 } 984 break; 985 default: 986 break; 987 } 988 return true; 989 } 990 991 void parseIncludeDirective() { 992 if (CurrentToken && CurrentToken->is(tok::less)) { 993 next(); 994 while (CurrentToken) { 995 // Mark tokens up to the trailing line comments as implicit string 996 // literals. 997 if (CurrentToken->isNot(tok::comment) && 998 !CurrentToken->TokenText.startswith("//")) 999 CurrentToken->Type = TT_ImplicitStringLiteral; 1000 next(); 1001 } 1002 } 1003 } 1004 1005 void parseWarningOrError() { 1006 next(); 1007 // We still want to format the whitespace left of the first token of the 1008 // warning or error. 1009 next(); 1010 while (CurrentToken) { 1011 CurrentToken->Type = TT_ImplicitStringLiteral; 1012 next(); 1013 } 1014 } 1015 1016 void parsePragma() { 1017 next(); // Consume "pragma". 1018 if (CurrentToken && 1019 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 1020 bool IsMark = CurrentToken->is(Keywords.kw_mark); 1021 next(); // Consume "mark". 1022 next(); // Consume first token (so we fix leading whitespace). 1023 while (CurrentToken) { 1024 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 1025 CurrentToken->Type = TT_ImplicitStringLiteral; 1026 next(); 1027 } 1028 } 1029 } 1030 1031 void parseHasInclude() { 1032 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 1033 return; 1034 next(); // '(' 1035 parseIncludeDirective(); 1036 next(); // ')' 1037 } 1038 1039 LineType parsePreprocessorDirective() { 1040 bool IsFirstToken = CurrentToken->IsFirst; 1041 LineType Type = LT_PreprocessorDirective; 1042 next(); 1043 if (!CurrentToken) 1044 return Type; 1045 1046 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) { 1047 // JavaScript files can contain shebang lines of the form: 1048 // #!/usr/bin/env node 1049 // Treat these like C++ #include directives. 1050 while (CurrentToken) { 1051 // Tokens cannot be comments here. 1052 CurrentToken->Type = TT_ImplicitStringLiteral; 1053 next(); 1054 } 1055 return LT_ImportStatement; 1056 } 1057 1058 if (CurrentToken->Tok.is(tok::numeric_constant)) { 1059 CurrentToken->SpacesRequiredBefore = 1; 1060 return Type; 1061 } 1062 // Hashes in the middle of a line can lead to any strange token 1063 // sequence. 1064 if (!CurrentToken->Tok.getIdentifierInfo()) 1065 return Type; 1066 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 1067 case tok::pp_include: 1068 case tok::pp_include_next: 1069 case tok::pp_import: 1070 next(); 1071 parseIncludeDirective(); 1072 Type = LT_ImportStatement; 1073 break; 1074 case tok::pp_error: 1075 case tok::pp_warning: 1076 parseWarningOrError(); 1077 break; 1078 case tok::pp_pragma: 1079 parsePragma(); 1080 break; 1081 case tok::pp_if: 1082 case tok::pp_elif: 1083 Contexts.back().IsExpression = true; 1084 next(); 1085 parseLine(); 1086 break; 1087 default: 1088 break; 1089 } 1090 while (CurrentToken) { 1091 FormatToken *Tok = CurrentToken; 1092 next(); 1093 if (Tok->is(tok::l_paren)) 1094 parseParens(); 1095 else if (Tok->isOneOf(Keywords.kw___has_include, 1096 Keywords.kw___has_include_next)) 1097 parseHasInclude(); 1098 } 1099 return Type; 1100 } 1101 1102 public: 1103 LineType parseLine() { 1104 if (!CurrentToken) 1105 return LT_Invalid; 1106 NonTemplateLess.clear(); 1107 if (CurrentToken->is(tok::hash)) 1108 return parsePreprocessorDirective(); 1109 1110 // Directly allow to 'import <string-literal>' to support protocol buffer 1111 // definitions (github.com/google/protobuf) or missing "#" (either way we 1112 // should not break the line). 1113 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 1114 if ((Style.Language == FormatStyle::LK_Java && 1115 CurrentToken->is(Keywords.kw_package)) || 1116 (Info && Info->getPPKeywordID() == tok::pp_import && 1117 CurrentToken->Next && 1118 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 1119 tok::kw_static))) { 1120 next(); 1121 parseIncludeDirective(); 1122 return LT_ImportStatement; 1123 } 1124 1125 // If this line starts and ends in '<' and '>', respectively, it is likely 1126 // part of "#define <a/b.h>". 1127 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 1128 parseIncludeDirective(); 1129 return LT_ImportStatement; 1130 } 1131 1132 // In .proto files, top-level options and package statements are very 1133 // similar to import statements and should not be line-wrapped. 1134 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 1135 CurrentToken->isOneOf(Keywords.kw_option, Keywords.kw_package)) { 1136 next(); 1137 if (CurrentToken && CurrentToken->is(tok::identifier)) { 1138 while (CurrentToken) 1139 next(); 1140 return LT_ImportStatement; 1141 } 1142 } 1143 1144 bool KeywordVirtualFound = false; 1145 bool ImportStatement = false; 1146 1147 // import {...} from '...'; 1148 if (Style.Language == FormatStyle::LK_JavaScript && 1149 CurrentToken->is(Keywords.kw_import)) 1150 ImportStatement = true; 1151 1152 while (CurrentToken) { 1153 if (CurrentToken->is(tok::kw_virtual)) 1154 KeywordVirtualFound = true; 1155 if (Style.Language == FormatStyle::LK_JavaScript) { 1156 // export {...} from '...'; 1157 // An export followed by "from 'some string';" is a re-export from 1158 // another module identified by a URI and is treated as a 1159 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 1160 // Just "export {...};" or "export class ..." should not be treated as 1161 // an import in this sense. 1162 if (Line.First->is(tok::kw_export) && 1163 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 1164 CurrentToken->Next->isStringLiteral()) 1165 ImportStatement = true; 1166 if (isClosureImportStatement(*CurrentToken)) 1167 ImportStatement = true; 1168 } 1169 if (!consumeToken()) 1170 return LT_Invalid; 1171 } 1172 if (KeywordVirtualFound) 1173 return LT_VirtualFunctionDecl; 1174 if (ImportStatement) 1175 return LT_ImportStatement; 1176 1177 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 1178 if (Contexts.back().FirstObjCSelectorName) 1179 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 1180 Contexts.back().LongestObjCSelectorName; 1181 return LT_ObjCMethodDecl; 1182 } 1183 1184 return LT_Other; 1185 } 1186 1187 private: 1188 bool isClosureImportStatement(const FormatToken &Tok) { 1189 // FIXME: Closure-library specific stuff should not be hard-coded but be 1190 // configurable. 1191 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 1192 Tok.Next->Next && 1193 (Tok.Next->Next->TokenText == "module" || 1194 Tok.Next->Next->TokenText == "provide" || 1195 Tok.Next->Next->TokenText == "require" || 1196 Tok.Next->Next->TokenText == "requireType" || 1197 Tok.Next->Next->TokenText == "forwardDeclare") && 1198 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 1199 } 1200 1201 void resetTokenMetadata(FormatToken *Token) { 1202 if (!Token) 1203 return; 1204 1205 // Reset token type in case we have already looked at it and then 1206 // recovered from an error (e.g. failure to find the matching >). 1207 if (!CurrentToken->isOneOf( 1208 TT_LambdaLSquare, TT_LambdaLBrace, TT_ForEachMacro, 1209 TT_TypenameMacro, TT_FunctionLBrace, TT_ImplicitStringLiteral, 1210 TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow, TT_NamespaceMacro, 1211 TT_OverloadedOperator, TT_RegexLiteral, TT_TemplateString, 1212 TT_ObjCStringLiteral)) 1213 CurrentToken->Type = TT_Unknown; 1214 CurrentToken->Role.reset(); 1215 CurrentToken->MatchingParen = nullptr; 1216 CurrentToken->FakeLParens.clear(); 1217 CurrentToken->FakeRParens = 0; 1218 } 1219 1220 void next() { 1221 if (CurrentToken) { 1222 CurrentToken->NestingLevel = Contexts.size() - 1; 1223 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1224 modifyContext(*CurrentToken); 1225 determineTokenType(*CurrentToken); 1226 CurrentToken = CurrentToken->Next; 1227 } 1228 1229 resetTokenMetadata(CurrentToken); 1230 } 1231 1232 /// A struct to hold information valid in a specific context, e.g. 1233 /// a pair of parenthesis. 1234 struct Context { 1235 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1236 bool IsExpression) 1237 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1238 IsExpression(IsExpression) {} 1239 1240 tok::TokenKind ContextKind; 1241 unsigned BindingStrength; 1242 bool IsExpression; 1243 unsigned LongestObjCSelectorName = 0; 1244 bool ColonIsForRangeExpr = false; 1245 bool ColonIsDictLiteral = false; 1246 bool ColonIsObjCMethodExpr = false; 1247 FormatToken *FirstObjCSelectorName = nullptr; 1248 FormatToken *FirstStartOfName = nullptr; 1249 bool CanBeExpression = true; 1250 bool InTemplateArgument = false; 1251 bool InCtorInitializer = false; 1252 bool InInheritanceList = false; 1253 bool CaretFound = false; 1254 bool IsForEachMacro = false; 1255 bool InCpp11AttributeSpecifier = false; 1256 bool InCSharpAttributeSpecifier = false; 1257 }; 1258 1259 /// Puts a new \c Context onto the stack \c Contexts for the lifetime 1260 /// of each instance. 1261 struct ScopedContextCreator { 1262 AnnotatingParser &P; 1263 1264 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1265 unsigned Increase) 1266 : P(P) { 1267 P.Contexts.push_back(Context(ContextKind, 1268 P.Contexts.back().BindingStrength + Increase, 1269 P.Contexts.back().IsExpression)); 1270 } 1271 1272 ~ScopedContextCreator() { P.Contexts.pop_back(); } 1273 }; 1274 1275 void modifyContext(const FormatToken &Current) { 1276 if (Current.getPrecedence() == prec::Assignment && 1277 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 1278 // Type aliases use `type X = ...;` in TypeScript and can be exported 1279 // using `export type ...`. 1280 !(Style.Language == FormatStyle::LK_JavaScript && 1281 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1282 Line.startsWith(tok::kw_export, Keywords.kw_type, 1283 tok::identifier))) && 1284 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 1285 Contexts.back().IsExpression = true; 1286 if (!Line.startsWith(TT_UnaryOperator)) { 1287 for (FormatToken *Previous = Current.Previous; 1288 Previous && Previous->Previous && 1289 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1290 Previous = Previous->Previous) { 1291 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1292 Previous = Previous->MatchingParen; 1293 if (!Previous) 1294 break; 1295 } 1296 if (Previous->opensScope()) 1297 break; 1298 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1299 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1300 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1301 Previous->Type = TT_PointerOrReference; 1302 } 1303 } 1304 } else if (Current.is(tok::lessless) && 1305 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1306 Contexts.back().IsExpression = true; 1307 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1308 Contexts.back().IsExpression = true; 1309 } else if (Current.is(TT_TrailingReturnArrow)) { 1310 Contexts.back().IsExpression = false; 1311 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1312 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1313 } else if (Current.Previous && 1314 Current.Previous->is(TT_CtorInitializerColon)) { 1315 Contexts.back().IsExpression = true; 1316 Contexts.back().InCtorInitializer = true; 1317 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1318 Contexts.back().InInheritanceList = true; 1319 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1320 for (FormatToken *Previous = Current.Previous; 1321 Previous && Previous->isOneOf(tok::star, tok::amp); 1322 Previous = Previous->Previous) 1323 Previous->Type = TT_PointerOrReference; 1324 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 1325 Contexts.back().IsExpression = false; 1326 } else if (Current.is(tok::kw_new)) { 1327 Contexts.back().CanBeExpression = false; 1328 } else if (Current.isOneOf(tok::semi, tok::exclaim)) { 1329 // This should be the condition or increment in a for-loop. 1330 Contexts.back().IsExpression = true; 1331 } 1332 } 1333 1334 void determineTokenType(FormatToken &Current) { 1335 if (!Current.is(TT_Unknown)) 1336 // The token type is already known. 1337 return; 1338 1339 if (Style.Language == FormatStyle::LK_JavaScript) { 1340 if (Current.is(tok::exclaim)) { 1341 if (Current.Previous && 1342 (Current.Previous->isOneOf(tok::identifier, tok::kw_namespace, 1343 tok::r_paren, tok::r_square, 1344 tok::r_brace) || 1345 Current.Previous->Tok.isLiteral())) { 1346 Current.Type = TT_JsNonNullAssertion; 1347 return; 1348 } 1349 if (Current.Next && 1350 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1351 Current.Type = TT_JsNonNullAssertion; 1352 return; 1353 } 1354 } 1355 } 1356 1357 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1358 // function declaration have been found. In this case, 'Current' is a 1359 // trailing token of this declaration and thus cannot be a name. 1360 if (Current.is(Keywords.kw_instanceof)) { 1361 Current.Type = TT_BinaryOperator; 1362 } else if (isStartOfName(Current) && 1363 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1364 Contexts.back().FirstStartOfName = &Current; 1365 Current.Type = TT_StartOfName; 1366 } else if (Current.is(tok::semi)) { 1367 // Reset FirstStartOfName after finding a semicolon so that a for loop 1368 // with multiple increment statements is not confused with a for loop 1369 // having multiple variable declarations. 1370 Contexts.back().FirstStartOfName = nullptr; 1371 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1372 AutoFound = true; 1373 } else if (Current.is(tok::arrow) && 1374 Style.Language == FormatStyle::LK_Java) { 1375 Current.Type = TT_LambdaArrow; 1376 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1377 Current.NestingLevel == 0) { 1378 Current.Type = TT_TrailingReturnArrow; 1379 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1380 Current.Type = determineStarAmpUsage(Current, 1381 Contexts.back().CanBeExpression && 1382 Contexts.back().IsExpression, 1383 Contexts.back().InTemplateArgument); 1384 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1385 Current.Type = determinePlusMinusCaretUsage(Current); 1386 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1387 Contexts.back().CaretFound = true; 1388 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1389 Current.Type = determineIncrementUsage(Current); 1390 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1391 Current.Type = TT_UnaryOperator; 1392 } else if (Current.is(tok::question)) { 1393 if (Style.Language == FormatStyle::LK_JavaScript && 1394 Line.MustBeDeclaration && !Contexts.back().IsExpression) { 1395 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1396 // on the interface, not a ternary expression. 1397 Current.Type = TT_JsTypeOptionalQuestion; 1398 } else { 1399 Current.Type = TT_ConditionalExpr; 1400 } 1401 } else if (Current.isBinaryOperator() && 1402 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1403 (!Current.is(tok::greater) && 1404 Style.Language != FormatStyle::LK_TextProto)) { 1405 Current.Type = TT_BinaryOperator; 1406 } else if (Current.is(tok::comment)) { 1407 if (Current.TokenText.startswith("/*")) { 1408 if (Current.TokenText.endswith("*/")) 1409 Current.Type = TT_BlockComment; 1410 else 1411 // The lexer has for some reason determined a comment here. But we 1412 // cannot really handle it, if it isn't properly terminated. 1413 Current.Tok.setKind(tok::unknown); 1414 } else { 1415 Current.Type = TT_LineComment; 1416 } 1417 } else if (Current.is(tok::r_paren)) { 1418 if (rParenEndsCast(Current)) 1419 Current.Type = TT_CastRParen; 1420 if (Current.MatchingParen && Current.Next && 1421 !Current.Next->isBinaryOperator() && 1422 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1423 tok::comma, tok::period, tok::arrow, 1424 tok::coloncolon)) 1425 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1426 // Make sure this isn't the return type of an Obj-C block declaration 1427 if (AfterParen->Tok.isNot(tok::caret)) { 1428 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1429 if (BeforeParen->is(tok::identifier) && 1430 !BeforeParen->is(TT_TypenameMacro) && 1431 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1432 (!BeforeParen->Previous || 1433 BeforeParen->Previous->ClosesTemplateDeclaration)) 1434 Current.Type = TT_FunctionAnnotationRParen; 1435 } 1436 } 1437 } else if (Current.is(tok::at) && Current.Next && 1438 Style.Language != FormatStyle::LK_JavaScript && 1439 Style.Language != FormatStyle::LK_Java) { 1440 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1441 // marks declarations and properties that need special formatting. 1442 switch (Current.Next->Tok.getObjCKeywordID()) { 1443 case tok::objc_interface: 1444 case tok::objc_implementation: 1445 case tok::objc_protocol: 1446 Current.Type = TT_ObjCDecl; 1447 break; 1448 case tok::objc_property: 1449 Current.Type = TT_ObjCProperty; 1450 break; 1451 default: 1452 break; 1453 } 1454 } else if (Current.is(tok::period)) { 1455 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1456 if (PreviousNoComment && 1457 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1458 Current.Type = TT_DesignatedInitializerPeriod; 1459 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1460 Current.Previous->isOneOf(TT_JavaAnnotation, 1461 TT_LeadingJavaAnnotation)) { 1462 Current.Type = Current.Previous->Type; 1463 } 1464 } else if (canBeObjCSelectorComponent(Current) && 1465 // FIXME(bug 36976): ObjC return types shouldn't use 1466 // TT_CastRParen. 1467 Current.Previous && Current.Previous->is(TT_CastRParen) && 1468 Current.Previous->MatchingParen && 1469 Current.Previous->MatchingParen->Previous && 1470 Current.Previous->MatchingParen->Previous->is( 1471 TT_ObjCMethodSpecifier)) { 1472 // This is the first part of an Objective-C selector name. (If there's no 1473 // colon after this, this is the only place which annotates the identifier 1474 // as a selector.) 1475 Current.Type = TT_SelectorName; 1476 } else if (Current.isOneOf(tok::identifier, tok::kw_const) && 1477 Current.Previous && 1478 !Current.Previous->isOneOf(tok::equal, tok::at) && 1479 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1480 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1481 // function declaration have been found. 1482 Current.Type = TT_TrailingAnnotation; 1483 } else if ((Style.Language == FormatStyle::LK_Java || 1484 Style.Language == FormatStyle::LK_JavaScript) && 1485 Current.Previous) { 1486 if (Current.Previous->is(tok::at) && 1487 Current.isNot(Keywords.kw_interface)) { 1488 const FormatToken &AtToken = *Current.Previous; 1489 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1490 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1491 Current.Type = TT_LeadingJavaAnnotation; 1492 else 1493 Current.Type = TT_JavaAnnotation; 1494 } else if (Current.Previous->is(tok::period) && 1495 Current.Previous->isOneOf(TT_JavaAnnotation, 1496 TT_LeadingJavaAnnotation)) { 1497 Current.Type = Current.Previous->Type; 1498 } 1499 } 1500 } 1501 1502 /// Take a guess at whether \p Tok starts a name of a function or 1503 /// variable declaration. 1504 /// 1505 /// This is a heuristic based on whether \p Tok is an identifier following 1506 /// something that is likely a type. 1507 bool isStartOfName(const FormatToken &Tok) { 1508 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1509 return false; 1510 1511 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1512 Keywords.kw_as)) 1513 return false; 1514 if (Style.Language == FormatStyle::LK_JavaScript && 1515 Tok.Previous->is(Keywords.kw_in)) 1516 return false; 1517 1518 // Skip "const" as it does not have an influence on whether this is a name. 1519 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1520 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1521 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1522 1523 if (!PreviousNotConst) 1524 return false; 1525 1526 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1527 PreviousNotConst->Previous && 1528 PreviousNotConst->Previous->is(tok::hash); 1529 1530 if (PreviousNotConst->is(TT_TemplateCloser)) 1531 return PreviousNotConst && PreviousNotConst->MatchingParen && 1532 PreviousNotConst->MatchingParen->Previous && 1533 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1534 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1535 1536 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen && 1537 PreviousNotConst->MatchingParen->Previous && 1538 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype)) 1539 return true; 1540 1541 return (!IsPPKeyword && 1542 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) || 1543 PreviousNotConst->is(TT_PointerOrReference) || 1544 PreviousNotConst->isSimpleTypeSpecifier(); 1545 } 1546 1547 /// Determine whether ')' is ending a cast. 1548 bool rParenEndsCast(const FormatToken &Tok) { 1549 // C-style casts are only used in C++ and Java. 1550 if (!Style.isCpp() && Style.Language != FormatStyle::LK_Java) 1551 return false; 1552 1553 // Empty parens aren't casts and there are no casts at the end of the line. 1554 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1555 return false; 1556 1557 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1558 if (LeftOfParens) { 1559 // If there is a closing parenthesis left of the current parentheses, 1560 // look past it as these might be chained casts. 1561 if (LeftOfParens->is(tok::r_paren)) { 1562 if (!LeftOfParens->MatchingParen || 1563 !LeftOfParens->MatchingParen->Previous) 1564 return false; 1565 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1566 } 1567 1568 // If there is an identifier (or with a few exceptions a keyword) right 1569 // before the parentheses, this is unlikely to be a cast. 1570 if (LeftOfParens->Tok.getIdentifierInfo() && 1571 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1572 tok::kw_delete)) 1573 return false; 1574 1575 // Certain other tokens right before the parentheses are also signals that 1576 // this cannot be a cast. 1577 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1578 TT_TemplateCloser, tok::ellipsis)) 1579 return false; 1580 } 1581 1582 if (Tok.Next->is(tok::question)) 1583 return false; 1584 1585 // As Java has no function types, a "(" after the ")" likely means that this 1586 // is a cast. 1587 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1588 return true; 1589 1590 // If a (non-string) literal follows, this is likely a cast. 1591 if (Tok.Next->isNot(tok::string_literal) && 1592 (Tok.Next->Tok.isLiteral() || 1593 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1594 return true; 1595 1596 // Heuristically try to determine whether the parentheses contain a type. 1597 bool ParensAreType = 1598 !Tok.Previous || 1599 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) || 1600 Tok.Previous->isSimpleTypeSpecifier(); 1601 bool ParensCouldEndDecl = 1602 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1603 if (ParensAreType && !ParensCouldEndDecl) 1604 return true; 1605 1606 // At this point, we heuristically assume that there are no casts at the 1607 // start of the line. We assume that we have found most cases where there 1608 // are by the logic above, e.g. "(void)x;". 1609 if (!LeftOfParens) 1610 return false; 1611 1612 // Certain token types inside the parentheses mean that this can't be a 1613 // cast. 1614 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 1615 Token = Token->Next) 1616 if (Token->is(TT_BinaryOperator)) 1617 return false; 1618 1619 // If the following token is an identifier or 'this', this is a cast. All 1620 // cases where this can be something else are handled above. 1621 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1622 return true; 1623 1624 if (!Tok.Next->Next) 1625 return false; 1626 1627 // If the next token after the parenthesis is a unary operator, assume 1628 // that this is cast, unless there are unexpected tokens inside the 1629 // parenthesis. 1630 bool NextIsUnary = 1631 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 1632 if (!NextIsUnary || Tok.Next->is(tok::plus) || 1633 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 1634 return false; 1635 // Search for unexpected tokens. 1636 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 1637 Prev = Prev->Previous) { 1638 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 1639 return false; 1640 } 1641 return true; 1642 } 1643 1644 /// Return the type of the given token assuming it is * or &. 1645 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 1646 bool InTemplateArgument) { 1647 if (Style.Language == FormatStyle::LK_JavaScript) 1648 return TT_BinaryOperator; 1649 1650 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1651 if (!PrevToken) 1652 return TT_UnaryOperator; 1653 1654 const FormatToken *NextToken = Tok.getNextNonComment(); 1655 if (!NextToken || 1656 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_const) || 1657 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 1658 return TT_PointerOrReference; 1659 1660 if (PrevToken->is(tok::coloncolon)) 1661 return TT_PointerOrReference; 1662 1663 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 1664 tok::comma, tok::semi, tok::kw_return, tok::colon, 1665 tok::equal, tok::kw_delete, tok::kw_sizeof, 1666 tok::kw_throw) || 1667 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 1668 TT_UnaryOperator, TT_CastRParen)) 1669 return TT_UnaryOperator; 1670 1671 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 1672 return TT_PointerOrReference; 1673 if (NextToken->is(tok::kw_operator) && !IsExpression) 1674 return TT_PointerOrReference; 1675 if (NextToken->isOneOf(tok::comma, tok::semi)) 1676 return TT_PointerOrReference; 1677 1678 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen) { 1679 FormatToken *TokenBeforeMatchingParen = 1680 PrevToken->MatchingParen->getPreviousNonComment(); 1681 if (TokenBeforeMatchingParen && 1682 TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype, 1683 TT_TypenameMacro)) 1684 return TT_PointerOrReference; 1685 } 1686 1687 if (PrevToken->Tok.isLiteral() || 1688 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 1689 tok::kw_false, tok::r_brace) || 1690 NextToken->Tok.isLiteral() || 1691 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 1692 NextToken->isUnaryOperator() || 1693 // If we know we're in a template argument, there are no named 1694 // declarations. Thus, having an identifier on the right-hand side 1695 // indicates a binary operator. 1696 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 1697 return TT_BinaryOperator; 1698 1699 // "&&(" is quite unlikely to be two successive unary "&". 1700 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren)) 1701 return TT_BinaryOperator; 1702 1703 // This catches some cases where evaluation order is used as control flow: 1704 // aaa && aaa->f(); 1705 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 1706 if (NextNextToken && NextNextToken->is(tok::arrow)) 1707 return TT_BinaryOperator; 1708 1709 // It is very unlikely that we are going to find a pointer or reference type 1710 // definition on the RHS of an assignment. 1711 if (IsExpression && !Contexts.back().CaretFound) 1712 return TT_BinaryOperator; 1713 1714 return TT_PointerOrReference; 1715 } 1716 1717 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 1718 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1719 if (!PrevToken) 1720 return TT_UnaryOperator; 1721 1722 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 1723 // This must be a sequence of leading unary operators. 1724 return TT_UnaryOperator; 1725 1726 // Use heuristics to recognize unary operators. 1727 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 1728 tok::question, tok::colon, tok::kw_return, 1729 tok::kw_case, tok::at, tok::l_brace)) 1730 return TT_UnaryOperator; 1731 1732 // There can't be two consecutive binary operators. 1733 if (PrevToken->is(TT_BinaryOperator)) 1734 return TT_UnaryOperator; 1735 1736 // Fall back to marking the token as binary operator. 1737 return TT_BinaryOperator; 1738 } 1739 1740 /// Determine whether ++/-- are pre- or post-increments/-decrements. 1741 TokenType determineIncrementUsage(const FormatToken &Tok) { 1742 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1743 if (!PrevToken || PrevToken->is(TT_CastRParen)) 1744 return TT_UnaryOperator; 1745 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 1746 return TT_TrailingUnaryOperator; 1747 1748 return TT_UnaryOperator; 1749 } 1750 1751 SmallVector<Context, 8> Contexts; 1752 1753 const FormatStyle &Style; 1754 AnnotatedLine &Line; 1755 FormatToken *CurrentToken; 1756 bool AutoFound; 1757 const AdditionalKeywords &Keywords; 1758 1759 // Set of "<" tokens that do not open a template parameter list. If parseAngle 1760 // determines that a specific token can't be a template opener, it will make 1761 // same decision irrespective of the decisions for tokens leading up to it. 1762 // Store this information to prevent this from causing exponential runtime. 1763 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 1764 }; 1765 1766 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 1767 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 1768 1769 /// Parses binary expressions by inserting fake parenthesis based on 1770 /// operator precedence. 1771 class ExpressionParser { 1772 public: 1773 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 1774 AnnotatedLine &Line) 1775 : Style(Style), Keywords(Keywords), Current(Line.First) {} 1776 1777 /// Parse expressions with the given operator precedence. 1778 void parse(int Precedence = 0) { 1779 // Skip 'return' and ObjC selector colons as they are not part of a binary 1780 // expression. 1781 while (Current && (Current->is(tok::kw_return) || 1782 (Current->is(tok::colon) && 1783 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 1784 next(); 1785 1786 if (!Current || Precedence > PrecedenceArrowAndPeriod) 1787 return; 1788 1789 // Conditional expressions need to be parsed separately for proper nesting. 1790 if (Precedence == prec::Conditional) { 1791 parseConditionalExpr(); 1792 return; 1793 } 1794 1795 // Parse unary operators, which all have a higher precedence than binary 1796 // operators. 1797 if (Precedence == PrecedenceUnaryOperator) { 1798 parseUnaryOperator(); 1799 return; 1800 } 1801 1802 FormatToken *Start = Current; 1803 FormatToken *LatestOperator = nullptr; 1804 unsigned OperatorIndex = 0; 1805 1806 while (Current) { 1807 // Consume operators with higher precedence. 1808 parse(Precedence + 1); 1809 1810 int CurrentPrecedence = getCurrentPrecedence(); 1811 1812 if (Current && Current->is(TT_SelectorName) && 1813 Precedence == CurrentPrecedence) { 1814 if (LatestOperator) 1815 addFakeParenthesis(Start, prec::Level(Precedence)); 1816 Start = Current; 1817 } 1818 1819 // At the end of the line or when an operator with higher precedence is 1820 // found, insert fake parenthesis and return. 1821 if (!Current || 1822 (Current->closesScope() && 1823 (Current->MatchingParen || Current->is(TT_TemplateString))) || 1824 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 1825 (CurrentPrecedence == prec::Conditional && 1826 Precedence == prec::Assignment && Current->is(tok::colon))) { 1827 break; 1828 } 1829 1830 // Consume scopes: (), [], <> and {} 1831 if (Current->opensScope()) { 1832 // In fragment of a JavaScript template string can look like '}..${' and 1833 // thus close a scope and open a new one at the same time. 1834 while (Current && (!Current->closesScope() || Current->opensScope())) { 1835 next(); 1836 parse(); 1837 } 1838 next(); 1839 } else { 1840 // Operator found. 1841 if (CurrentPrecedence == Precedence) { 1842 if (LatestOperator) 1843 LatestOperator->NextOperator = Current; 1844 LatestOperator = Current; 1845 Current->OperatorIndex = OperatorIndex; 1846 ++OperatorIndex; 1847 } 1848 next(/*SkipPastLeadingComments=*/Precedence > 0); 1849 } 1850 } 1851 1852 if (LatestOperator && (Current || Precedence > 0)) { 1853 // LatestOperator->LastOperator = true; 1854 if (Precedence == PrecedenceArrowAndPeriod) { 1855 // Call expressions don't have a binary operator precedence. 1856 addFakeParenthesis(Start, prec::Unknown); 1857 } else { 1858 addFakeParenthesis(Start, prec::Level(Precedence)); 1859 } 1860 } 1861 } 1862 1863 private: 1864 /// Gets the precedence (+1) of the given token for binary operators 1865 /// and other tokens that we treat like binary operators. 1866 int getCurrentPrecedence() { 1867 if (Current) { 1868 const FormatToken *NextNonComment = Current->getNextNonComment(); 1869 if (Current->is(TT_ConditionalExpr)) 1870 return prec::Conditional; 1871 if (NextNonComment && Current->is(TT_SelectorName) && 1872 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 1873 ((Style.Language == FormatStyle::LK_Proto || 1874 Style.Language == FormatStyle::LK_TextProto) && 1875 NextNonComment->is(tok::less)))) 1876 return prec::Assignment; 1877 if (Current->is(TT_JsComputedPropertyName)) 1878 return prec::Assignment; 1879 if (Current->is(TT_LambdaArrow)) 1880 return prec::Comma; 1881 if (Current->is(TT_JsFatArrow)) 1882 return prec::Assignment; 1883 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 1884 (Current->is(tok::comment) && NextNonComment && 1885 NextNonComment->is(TT_SelectorName))) 1886 return 0; 1887 if (Current->is(TT_RangeBasedForLoopColon)) 1888 return prec::Comma; 1889 if ((Style.Language == FormatStyle::LK_Java || 1890 Style.Language == FormatStyle::LK_JavaScript) && 1891 Current->is(Keywords.kw_instanceof)) 1892 return prec::Relational; 1893 if (Style.Language == FormatStyle::LK_JavaScript && 1894 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 1895 return prec::Relational; 1896 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 1897 return Current->getPrecedence(); 1898 if (Current->isOneOf(tok::period, tok::arrow)) 1899 return PrecedenceArrowAndPeriod; 1900 if ((Style.Language == FormatStyle::LK_Java || 1901 Style.Language == FormatStyle::LK_JavaScript) && 1902 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 1903 Keywords.kw_throws)) 1904 return 0; 1905 } 1906 return -1; 1907 } 1908 1909 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 1910 Start->FakeLParens.push_back(Precedence); 1911 if (Precedence > prec::Unknown) 1912 Start->StartsBinaryExpression = true; 1913 if (Current) { 1914 FormatToken *Previous = Current->Previous; 1915 while (Previous->is(tok::comment) && Previous->Previous) 1916 Previous = Previous->Previous; 1917 ++Previous->FakeRParens; 1918 if (Precedence > prec::Unknown) 1919 Previous->EndsBinaryExpression = true; 1920 } 1921 } 1922 1923 /// Parse unary operator expressions and surround them with fake 1924 /// parentheses if appropriate. 1925 void parseUnaryOperator() { 1926 llvm::SmallVector<FormatToken *, 2> Tokens; 1927 while (Current && Current->is(TT_UnaryOperator)) { 1928 Tokens.push_back(Current); 1929 next(); 1930 } 1931 parse(PrecedenceArrowAndPeriod); 1932 for (FormatToken *Token : llvm::reverse(Tokens)) 1933 // The actual precedence doesn't matter. 1934 addFakeParenthesis(Token, prec::Unknown); 1935 } 1936 1937 void parseConditionalExpr() { 1938 while (Current && Current->isTrailingComment()) { 1939 next(); 1940 } 1941 FormatToken *Start = Current; 1942 parse(prec::LogicalOr); 1943 if (!Current || !Current->is(tok::question)) 1944 return; 1945 next(); 1946 parse(prec::Assignment); 1947 if (!Current || Current->isNot(TT_ConditionalExpr)) 1948 return; 1949 next(); 1950 parse(prec::Assignment); 1951 addFakeParenthesis(Start, prec::Conditional); 1952 } 1953 1954 void next(bool SkipPastLeadingComments = true) { 1955 if (Current) 1956 Current = Current->Next; 1957 while (Current && 1958 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 1959 Current->isTrailingComment()) 1960 Current = Current->Next; 1961 } 1962 1963 const FormatStyle &Style; 1964 const AdditionalKeywords &Keywords; 1965 FormatToken *Current; 1966 }; 1967 1968 } // end anonymous namespace 1969 1970 void TokenAnnotator::setCommentLineLevels( 1971 SmallVectorImpl<AnnotatedLine *> &Lines) { 1972 const AnnotatedLine *NextNonCommentLine = nullptr; 1973 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(), 1974 E = Lines.rend(); 1975 I != E; ++I) { 1976 bool CommentLine = true; 1977 for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) { 1978 if (!Tok->is(tok::comment)) { 1979 CommentLine = false; 1980 break; 1981 } 1982 } 1983 1984 // If the comment is currently aligned with the line immediately following 1985 // it, that's probably intentional and we should keep it. 1986 if (NextNonCommentLine && CommentLine && 1987 NextNonCommentLine->First->NewlinesBefore <= 1 && 1988 NextNonCommentLine->First->OriginalColumn == 1989 (*I)->First->OriginalColumn) { 1990 // Align comments for preprocessor lines with the # in column 0 if 1991 // preprocessor lines are not indented. Otherwise, align with the next 1992 // line. 1993 (*I)->Level = 1994 (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 1995 (NextNonCommentLine->Type == LT_PreprocessorDirective || 1996 NextNonCommentLine->Type == LT_ImportStatement)) 1997 ? 0 1998 : NextNonCommentLine->Level; 1999 } else { 2000 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr; 2001 } 2002 2003 setCommentLineLevels((*I)->Children); 2004 } 2005 } 2006 2007 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 2008 unsigned Result = 0; 2009 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 2010 Result = std::max(Result, Tok->NestingLevel); 2011 return Result; 2012 } 2013 2014 void TokenAnnotator::annotate(AnnotatedLine &Line) { 2015 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2016 E = Line.Children.end(); 2017 I != E; ++I) { 2018 annotate(**I); 2019 } 2020 AnnotatingParser Parser(Style, Line, Keywords); 2021 Line.Type = Parser.parseLine(); 2022 2023 // With very deep nesting, ExpressionParser uses lots of stack and the 2024 // formatting algorithm is very slow. We're not going to do a good job here 2025 // anyway - it's probably generated code being formatted by mistake. 2026 // Just skip the whole line. 2027 if (maxNestingDepth(Line) > 50) 2028 Line.Type = LT_Invalid; 2029 2030 if (Line.Type == LT_Invalid) 2031 return; 2032 2033 ExpressionParser ExprParser(Style, Keywords, Line); 2034 ExprParser.parse(); 2035 2036 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2037 Line.Type = LT_ObjCMethodDecl; 2038 else if (Line.startsWith(TT_ObjCDecl)) 2039 Line.Type = LT_ObjCDecl; 2040 else if (Line.startsWith(TT_ObjCProperty)) 2041 Line.Type = LT_ObjCProperty; 2042 2043 Line.First->SpacesRequiredBefore = 1; 2044 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 2045 } 2046 2047 // This function heuristically determines whether 'Current' starts the name of a 2048 // function declaration. 2049 static bool isFunctionDeclarationName(const FormatToken &Current, 2050 const AnnotatedLine &Line) { 2051 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 2052 for (; Next; Next = Next->Next) { 2053 if (Next->is(TT_OverloadedOperatorLParen)) 2054 return Next; 2055 if (Next->is(TT_OverloadedOperator)) 2056 continue; 2057 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 2058 // For 'new[]' and 'delete[]'. 2059 if (Next->Next && Next->Next->is(tok::l_square) && Next->Next->Next && 2060 Next->Next->Next->is(tok::r_square)) 2061 Next = Next->Next->Next; 2062 continue; 2063 } 2064 2065 break; 2066 } 2067 return nullptr; 2068 }; 2069 2070 // Find parentheses of parameter list. 2071 const FormatToken *Next = Current.Next; 2072 if (Current.is(tok::kw_operator)) { 2073 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 2074 return false; 2075 Next = skipOperatorName(Next); 2076 } else { 2077 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 2078 return false; 2079 for (; Next; Next = Next->Next) { 2080 if (Next->is(TT_TemplateOpener)) { 2081 Next = Next->MatchingParen; 2082 } else if (Next->is(tok::coloncolon)) { 2083 Next = Next->Next; 2084 if (!Next) 2085 return false; 2086 if (Next->is(tok::kw_operator)) { 2087 Next = skipOperatorName(Next->Next); 2088 break; 2089 } 2090 if (!Next->is(tok::identifier)) 2091 return false; 2092 } else if (Next->is(tok::l_paren)) { 2093 break; 2094 } else { 2095 return false; 2096 } 2097 } 2098 } 2099 2100 // Check whether parameter list can belong to a function declaration. 2101 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 2102 return false; 2103 // If the lines ends with "{", this is likely an function definition. 2104 if (Line.Last->is(tok::l_brace)) 2105 return true; 2106 if (Next->Next == Next->MatchingParen) 2107 return true; // Empty parentheses. 2108 // If there is an &/&& after the r_paren, this is likely a function. 2109 if (Next->MatchingParen->Next && 2110 Next->MatchingParen->Next->is(TT_PointerOrReference)) 2111 return true; 2112 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 2113 Tok = Tok->Next) { 2114 if (Tok->isOneOf(tok::l_paren, TT_TemplateOpener) && Tok->MatchingParen) { 2115 Tok = Tok->MatchingParen; 2116 continue; 2117 } 2118 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 2119 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 2120 return true; 2121 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 2122 Tok->Tok.isLiteral()) 2123 return false; 2124 } 2125 return false; 2126 } 2127 2128 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 2129 assert(Line.MightBeFunctionDecl); 2130 2131 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 2132 Style.AlwaysBreakAfterReturnType == 2133 FormatStyle::RTBS_TopLevelDefinitions) && 2134 Line.Level > 0) 2135 return false; 2136 2137 switch (Style.AlwaysBreakAfterReturnType) { 2138 case FormatStyle::RTBS_None: 2139 return false; 2140 case FormatStyle::RTBS_All: 2141 case FormatStyle::RTBS_TopLevel: 2142 return true; 2143 case FormatStyle::RTBS_AllDefinitions: 2144 case FormatStyle::RTBS_TopLevelDefinitions: 2145 return Line.mightBeFunctionDefinition(); 2146 } 2147 2148 return false; 2149 } 2150 2151 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 2152 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2153 E = Line.Children.end(); 2154 I != E; ++I) { 2155 calculateFormattingInformation(**I); 2156 } 2157 2158 Line.First->TotalLength = 2159 Line.First->IsMultiline ? Style.ColumnLimit 2160 : Line.FirstStartColumn + Line.First->ColumnWidth; 2161 FormatToken *Current = Line.First->Next; 2162 bool InFunctionDecl = Line.MightBeFunctionDecl; 2163 while (Current) { 2164 if (isFunctionDeclarationName(*Current, Line)) 2165 Current->Type = TT_FunctionDeclarationName; 2166 if (Current->is(TT_LineComment)) { 2167 if (Current->Previous->BlockKind == BK_BracedInit && 2168 Current->Previous->opensScope()) 2169 Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1; 2170 else 2171 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2172 2173 // If we find a trailing comment, iterate backwards to determine whether 2174 // it seems to relate to a specific parameter. If so, break before that 2175 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2176 // to the previous line in: 2177 // SomeFunction(a, 2178 // b, // comment 2179 // c); 2180 if (!Current->HasUnescapedNewline) { 2181 for (FormatToken *Parameter = Current->Previous; Parameter; 2182 Parameter = Parameter->Previous) { 2183 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 2184 break; 2185 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 2186 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 2187 Parameter->HasUnescapedNewline) 2188 Parameter->MustBreakBefore = true; 2189 break; 2190 } 2191 } 2192 } 2193 } else if (Current->SpacesRequiredBefore == 0 && 2194 spaceRequiredBefore(Line, *Current)) { 2195 Current->SpacesRequiredBefore = 1; 2196 } 2197 2198 Current->MustBreakBefore = 2199 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 2200 2201 if (!Current->MustBreakBefore && InFunctionDecl && 2202 Current->is(TT_FunctionDeclarationName)) 2203 Current->MustBreakBefore = mustBreakForReturnType(Line); 2204 2205 Current->CanBreakBefore = 2206 Current->MustBreakBefore || canBreakBefore(Line, *Current); 2207 unsigned ChildSize = 0; 2208 if (Current->Previous->Children.size() == 1) { 2209 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last; 2210 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 2211 : LastOfChild.TotalLength + 1; 2212 } 2213 const FormatToken *Prev = Current->Previous; 2214 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 2215 (Prev->Children.size() == 1 && 2216 Prev->Children[0]->First->MustBreakBefore) || 2217 Current->IsMultiline) 2218 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 2219 else 2220 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 2221 ChildSize + Current->SpacesRequiredBefore; 2222 2223 if (Current->is(TT_CtorInitializerColon)) 2224 InFunctionDecl = false; 2225 2226 // FIXME: Only calculate this if CanBreakBefore is true once static 2227 // initializers etc. are sorted out. 2228 // FIXME: Move magic numbers to a better place. 2229 2230 // Reduce penalty for aligning ObjC method arguments using the colon 2231 // alignment as this is the canonical way (still prefer fitting everything 2232 // into one line if possible). Trying to fit a whole expression into one 2233 // line should not force other line breaks (e.g. when ObjC method 2234 // expression is a part of other expression). 2235 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl); 2236 if (Style.Language == FormatStyle::LK_ObjC && 2237 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) { 2238 if (Current->ParameterIndex == 1) 2239 Current->SplitPenalty += 5 * Current->BindingStrength; 2240 } else { 2241 Current->SplitPenalty += 20 * Current->BindingStrength; 2242 } 2243 2244 Current = Current->Next; 2245 } 2246 2247 calculateUnbreakableTailLengths(Line); 2248 unsigned IndentLevel = Line.Level; 2249 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2250 if (Current->Role) 2251 Current->Role->precomputeFormattingInfos(Current); 2252 if (Current->MatchingParen && 2253 Current->MatchingParen->opensBlockOrBlockTypeList(Style)) { 2254 assert(IndentLevel > 0); 2255 --IndentLevel; 2256 } 2257 Current->IndentLevel = IndentLevel; 2258 if (Current->opensBlockOrBlockTypeList(Style)) 2259 ++IndentLevel; 2260 } 2261 2262 LLVM_DEBUG({ printDebugInfo(Line); }); 2263 } 2264 2265 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2266 unsigned UnbreakableTailLength = 0; 2267 FormatToken *Current = Line.Last; 2268 while (Current) { 2269 Current->UnbreakableTailLength = UnbreakableTailLength; 2270 if (Current->CanBreakBefore || 2271 Current->isOneOf(tok::comment, tok::string_literal)) { 2272 UnbreakableTailLength = 0; 2273 } else { 2274 UnbreakableTailLength += 2275 Current->ColumnWidth + Current->SpacesRequiredBefore; 2276 } 2277 Current = Current->Previous; 2278 } 2279 } 2280 2281 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2282 const FormatToken &Tok, 2283 bool InFunctionDecl) { 2284 const FormatToken &Left = *Tok.Previous; 2285 const FormatToken &Right = Tok; 2286 2287 if (Left.is(tok::semi)) 2288 return 0; 2289 2290 if (Style.Language == FormatStyle::LK_Java) { 2291 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2292 return 1; 2293 if (Right.is(Keywords.kw_implements)) 2294 return 2; 2295 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2296 return 3; 2297 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2298 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2299 return 100; 2300 if (Left.is(TT_JsTypeColon)) 2301 return 35; 2302 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2303 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2304 return 100; 2305 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2306 if (Left.opensScope() && Right.closesScope()) 2307 return 200; 2308 } 2309 2310 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2311 return 1; 2312 if (Right.is(tok::l_square)) { 2313 if (Style.Language == FormatStyle::LK_Proto) 2314 return 1; 2315 if (Left.is(tok::r_square)) 2316 return 200; 2317 // Slightly prefer formatting local lambda definitions like functions. 2318 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2319 return 35; 2320 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2321 TT_ArrayInitializerLSquare, 2322 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) 2323 return 500; 2324 } 2325 2326 if (Left.is(tok::coloncolon) || 2327 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2328 return 500; 2329 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2330 Right.is(tok::kw_operator)) { 2331 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2332 return 3; 2333 if (Left.is(TT_StartOfName)) 2334 return 110; 2335 if (InFunctionDecl && Right.NestingLevel == 0) 2336 return Style.PenaltyReturnTypeOnItsOwnLine; 2337 return 200; 2338 } 2339 if (Right.is(TT_PointerOrReference)) 2340 return 190; 2341 if (Right.is(TT_LambdaArrow)) 2342 return 110; 2343 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2344 return 160; 2345 if (Left.is(TT_CastRParen)) 2346 return 100; 2347 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2348 return 5000; 2349 if (Left.is(tok::comment)) 2350 return 1000; 2351 2352 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2353 TT_CtorInitializerColon)) 2354 return 2; 2355 2356 if (Right.isMemberAccess()) { 2357 // Breaking before the "./->" of a chained call/member access is reasonably 2358 // cheap, as formatting those with one call per line is generally 2359 // desirable. In particular, it should be cheaper to break before the call 2360 // than it is to break inside a call's parameters, which could lead to weird 2361 // "hanging" indents. The exception is the very last "./->" to support this 2362 // frequent pattern: 2363 // 2364 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2365 // dddddddd); 2366 // 2367 // which might otherwise be blown up onto many lines. Here, clang-format 2368 // won't produce "hanging" indents anyway as there is no other trailing 2369 // call. 2370 // 2371 // Also apply higher penalty is not a call as that might lead to a wrapping 2372 // like: 2373 // 2374 // aaaaaaa 2375 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2376 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2377 ? 150 2378 : 35; 2379 } 2380 2381 if (Right.is(TT_TrailingAnnotation) && 2382 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2383 // Moving trailing annotations to the next line is fine for ObjC method 2384 // declarations. 2385 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2386 return 10; 2387 // Generally, breaking before a trailing annotation is bad unless it is 2388 // function-like. It seems to be especially preferable to keep standard 2389 // annotations (i.e. "const", "final" and "override") on the same line. 2390 // Use a slightly higher penalty after ")" so that annotations like 2391 // "const override" are kept together. 2392 bool is_short_annotation = Right.TokenText.size() < 10; 2393 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 2394 } 2395 2396 // In for-loops, prefer breaking at ',' and ';'. 2397 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 2398 return 4; 2399 2400 // In Objective-C method expressions, prefer breaking before "param:" over 2401 // breaking after it. 2402 if (Right.is(TT_SelectorName)) 2403 return 0; 2404 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 2405 return Line.MightBeFunctionDecl ? 50 : 500; 2406 2407 // In Objective-C type declarations, avoid breaking after the category's 2408 // open paren (we'll prefer breaking after the protocol list's opening 2409 // angle bracket, if present). 2410 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous && 2411 Left.Previous->isOneOf(tok::identifier, tok::greater)) 2412 return 500; 2413 2414 if (Left.is(tok::l_paren) && InFunctionDecl && 2415 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 2416 return 100; 2417 if (Left.is(tok::l_paren) && Left.Previous && 2418 (Left.Previous->is(tok::kw_for) || Left.Previous->isIf())) 2419 return 1000; 2420 if (Left.is(tok::equal) && InFunctionDecl) 2421 return 110; 2422 if (Right.is(tok::r_brace)) 2423 return 1; 2424 if (Left.is(TT_TemplateOpener)) 2425 return 100; 2426 if (Left.opensScope()) { 2427 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign) 2428 return 0; 2429 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle) 2430 return 19; 2431 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 2432 : 19; 2433 } 2434 if (Left.is(TT_JavaAnnotation)) 2435 return 50; 2436 2437 if (Left.is(TT_UnaryOperator)) 2438 return 60; 2439 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 2440 Left.Previous->isLabelString() && 2441 (Left.NextOperator || Left.OperatorIndex != 0)) 2442 return 50; 2443 if (Right.is(tok::plus) && Left.isLabelString() && 2444 (Right.NextOperator || Right.OperatorIndex != 0)) 2445 return 25; 2446 if (Left.is(tok::comma)) 2447 return 1; 2448 if (Right.is(tok::lessless) && Left.isLabelString() && 2449 (Right.NextOperator || Right.OperatorIndex != 1)) 2450 return 25; 2451 if (Right.is(tok::lessless)) { 2452 // Breaking at a << is really cheap. 2453 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 2454 // Slightly prefer to break before the first one in log-like statements. 2455 return 2; 2456 return 1; 2457 } 2458 if (Left.ClosesTemplateDeclaration) 2459 return Style.PenaltyBreakTemplateDeclaration; 2460 if (Left.is(TT_ConditionalExpr)) 2461 return prec::Conditional; 2462 prec::Level Level = Left.getPrecedence(); 2463 if (Level == prec::Unknown) 2464 Level = Right.getPrecedence(); 2465 if (Level == prec::Assignment) 2466 return Style.PenaltyBreakAssignment; 2467 if (Level != prec::Unknown) 2468 return Level; 2469 2470 return 3; 2471 } 2472 2473 bool TokenAnnotator::spaceRequiredBeforeParens(const FormatToken &Right) const { 2474 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always || 2475 (Style.SpaceBeforeParens == FormatStyle::SBPO_NonEmptyParentheses && 2476 Right.ParameterCount > 0); 2477 } 2478 2479 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 2480 const FormatToken &Left, 2481 const FormatToken &Right) { 2482 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 2483 return true; 2484 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 2485 return true; 2486 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 2487 Left.Tok.getObjCKeywordID() == tok::objc_property) 2488 return true; 2489 if (Right.is(tok::hashhash)) 2490 return Left.is(tok::hash); 2491 if (Left.isOneOf(tok::hashhash, tok::hash)) 2492 return Right.is(tok::hash); 2493 if (Left.is(tok::l_paren) && Right.is(tok::r_paren)) 2494 return Style.SpaceInEmptyParentheses; 2495 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 2496 return (Right.is(TT_CastRParen) || 2497 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 2498 ? Style.SpacesInCStyleCastParentheses 2499 : Style.SpacesInParentheses; 2500 if (Right.isOneOf(tok::semi, tok::comma)) 2501 return false; 2502 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) { 2503 bool IsLightweightGeneric = Right.MatchingParen && 2504 Right.MatchingParen->Next && 2505 Right.MatchingParen->Next->is(tok::colon); 2506 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList; 2507 } 2508 if (Right.is(tok::less) && Left.is(tok::kw_template)) 2509 return Style.SpaceAfterTemplateKeyword; 2510 if (Left.isOneOf(tok::exclaim, tok::tilde)) 2511 return false; 2512 if (Left.is(tok::at) && 2513 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 2514 tok::numeric_constant, tok::l_paren, tok::l_brace, 2515 tok::kw_true, tok::kw_false)) 2516 return false; 2517 if (Left.is(tok::colon)) 2518 return !Left.is(TT_ObjCMethodExpr); 2519 if (Left.is(tok::coloncolon)) 2520 return false; 2521 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 2522 if (Style.Language == FormatStyle::LK_TextProto || 2523 (Style.Language == FormatStyle::LK_Proto && 2524 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 2525 // Format empty list as `<>`. 2526 if (Left.is(tok::less) && Right.is(tok::greater)) 2527 return false; 2528 return !Style.Cpp11BracedListStyle; 2529 } 2530 return false; 2531 } 2532 if (Right.is(tok::ellipsis)) 2533 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 2534 Left.Previous->is(tok::kw_case)); 2535 if (Left.is(tok::l_square) && Right.is(tok::amp)) 2536 return false; 2537 if (Right.is(TT_PointerOrReference)) { 2538 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 2539 if (!Left.MatchingParen) 2540 return true; 2541 FormatToken *TokenBeforeMatchingParen = 2542 Left.MatchingParen->getPreviousNonComment(); 2543 if (!TokenBeforeMatchingParen || 2544 !TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype, 2545 TT_TypenameMacro)) 2546 return true; 2547 } 2548 return (Left.Tok.isLiteral() || 2549 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 2550 (Style.PointerAlignment != FormatStyle::PAS_Left || 2551 (Line.IsMultiVariableDeclStmt && 2552 (Left.NestingLevel == 0 || 2553 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for))))))); 2554 } 2555 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 2556 (!Left.is(TT_PointerOrReference) || 2557 (Style.PointerAlignment != FormatStyle::PAS_Right && 2558 !Line.IsMultiVariableDeclStmt))) 2559 return true; 2560 if (Left.is(TT_PointerOrReference)) 2561 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) || 2562 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 2563 !Right.is(TT_StartOfName)) || 2564 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) || 2565 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 2566 tok::l_paren) && 2567 (Style.PointerAlignment != FormatStyle::PAS_Right && 2568 !Line.IsMultiVariableDeclStmt) && 2569 Left.Previous && 2570 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon)); 2571 if (Right.is(tok::star) && Left.is(tok::l_paren)) 2572 return false; 2573 const auto SpaceRequiredForArrayInitializerLSquare = 2574 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 2575 return Style.SpacesInContainerLiterals || 2576 ((Style.Language == FormatStyle::LK_Proto || 2577 Style.Language == FormatStyle::LK_TextProto) && 2578 !Style.Cpp11BracedListStyle && 2579 LSquareTok.endsSequence(tok::l_square, tok::colon, 2580 TT_SelectorName)); 2581 }; 2582 if (Left.is(tok::l_square)) 2583 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 2584 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 2585 (Left.isOneOf(TT_ArraySubscriptLSquare, 2586 TT_StructuredBindingLSquare) && 2587 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 2588 if (Right.is(tok::r_square)) 2589 return Right.MatchingParen && 2590 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 2591 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 2592 Style)) || 2593 (Style.SpacesInSquareBrackets && 2594 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 2595 TT_StructuredBindingLSquare)) || 2596 Right.MatchingParen->is(TT_AttributeParen)); 2597 if (Right.is(tok::l_square) && 2598 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2599 TT_DesignatedInitializerLSquare, 2600 TT_StructuredBindingLSquare, TT_AttributeSquare) && 2601 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral)) 2602 return false; 2603 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 2604 return !Left.Children.empty(); // No spaces in "{}". 2605 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) || 2606 (Right.is(tok::r_brace) && Right.MatchingParen && 2607 Right.MatchingParen->BlockKind != BK_Block)) 2608 return !Style.Cpp11BracedListStyle; 2609 if (Left.is(TT_BlockComment)) 2610 // No whitespace in x(/*foo=*/1), except for JavaScript. 2611 return Style.Language == FormatStyle::LK_JavaScript || 2612 !Left.TokenText.endswith("=*/"); 2613 if (Right.is(tok::l_paren)) { 2614 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 2615 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) 2616 return true; 2617 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) || 2618 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never && 2619 (Left.isOneOf(tok::pp_elif, tok::kw_for, tok::kw_while, 2620 tok::kw_switch, tok::kw_case, TT_ForEachMacro, 2621 TT_ObjCForIn) || 2622 Left.isIf(Line.Type != LT_PreprocessorDirective) || 2623 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch, 2624 tok::kw_new, tok::kw_delete) && 2625 (!Left.Previous || Left.Previous->isNot(tok::period))))) || 2626 (spaceRequiredBeforeParens(Right) && 2627 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 2628 Left.is(tok::r_paren) || Left.isSimpleTypeSpecifier() || 2629 (Left.is(tok::r_square) && Left.MatchingParen && 2630 Left.MatchingParen->is(TT_LambdaLSquare))) && 2631 Line.Type != LT_PreprocessorDirective); 2632 } 2633 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 2634 return false; 2635 if (Right.is(TT_UnaryOperator)) 2636 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 2637 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 2638 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 2639 tok::r_paren) || 2640 Left.isSimpleTypeSpecifier()) && 2641 Right.is(tok::l_brace) && Right.getNextNonComment() && 2642 Right.BlockKind != BK_Block) 2643 return false; 2644 if (Left.is(tok::period) || Right.is(tok::period)) 2645 return false; 2646 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L") 2647 return false; 2648 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 2649 Left.MatchingParen->Previous && 2650 (Left.MatchingParen->Previous->is(tok::period) || 2651 Left.MatchingParen->Previous->is(tok::coloncolon))) 2652 // Java call to generic function with explicit type: 2653 // A.<B<C<...>>>DoSomething(); 2654 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference. 2655 return false; 2656 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 2657 return false; 2658 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at)) 2659 // Objective-C dictionary literal -> no space after opening brace. 2660 return false; 2661 if (Right.is(tok::r_brace) && Right.MatchingParen && 2662 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at)) 2663 // Objective-C dictionary literal -> no space before closing brace. 2664 return false; 2665 return true; 2666 } 2667 2668 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 2669 const FormatToken &Right) { 2670 const FormatToken &Left = *Right.Previous; 2671 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 2672 return true; // Never ever merge two identifiers. 2673 if (Style.isCpp()) { 2674 if (Left.is(tok::kw_operator)) 2675 return Right.is(tok::coloncolon); 2676 if (Right.is(tok::l_brace) && Right.BlockKind == BK_BracedInit && 2677 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList) 2678 return true; 2679 } else if (Style.Language == FormatStyle::LK_Proto || 2680 Style.Language == FormatStyle::LK_TextProto) { 2681 if (Right.is(tok::period) && 2682 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 2683 Keywords.kw_repeated, Keywords.kw_extend)) 2684 return true; 2685 if (Right.is(tok::l_paren) && 2686 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 2687 return true; 2688 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 2689 return true; 2690 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 2691 if (Left.is(tok::slash) || Right.is(tok::slash)) 2692 return false; 2693 if (Left.MatchingParen && 2694 Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 2695 Right.isOneOf(tok::l_brace, tok::less)) 2696 return !Style.Cpp11BracedListStyle; 2697 // A percent is probably part of a formatting specification, such as %lld. 2698 if (Left.is(tok::percent)) 2699 return false; 2700 // Preserve the existence of a space before a percent for cases like 0x%04x 2701 // and "%d %d" 2702 if (Left.is(tok::numeric_constant) && Right.is(tok::percent)) 2703 return Right.WhitespaceRange.getEnd() != Right.WhitespaceRange.getBegin(); 2704 } else if (Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp()) { 2705 if (Left.is(TT_JsFatArrow)) 2706 return true; 2707 // for await ( ... 2708 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 2709 Left.Previous->is(tok::kw_for)) 2710 return true; 2711 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 2712 Right.MatchingParen) { 2713 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 2714 // An async arrow function, for example: `x = async () => foo();`, 2715 // as opposed to calling a function called async: `x = async();` 2716 if (Next && Next->is(TT_JsFatArrow)) 2717 return true; 2718 } 2719 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2720 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2721 return false; 2722 // In tagged template literals ("html`bar baz`"), there is no space between 2723 // the tag identifier and the template string. getIdentifierInfo makes sure 2724 // that the identifier is not a pseudo keyword like `yield`, either. 2725 if (Left.is(tok::identifier) && Keywords.IsJavaScriptIdentifier(Left) && 2726 Right.is(TT_TemplateString)) 2727 return false; 2728 if (Right.is(tok::star) && 2729 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 2730 return false; 2731 if (Right.isOneOf(tok::l_brace, tok::l_square) && 2732 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 2733 Keywords.kw_extends, Keywords.kw_implements)) 2734 return true; 2735 if (Right.is(tok::l_paren)) { 2736 // JS methods can use some keywords as names (e.g. `delete()`). 2737 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 2738 return false; 2739 // Valid JS method names can include keywords, e.g. `foo.delete()` or 2740 // `bar.instanceof()`. Recognize call positions by preceding period. 2741 if (Left.Previous && Left.Previous->is(tok::period) && 2742 Left.Tok.getIdentifierInfo()) 2743 return false; 2744 // Additional unary JavaScript operators that need a space after. 2745 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 2746 tok::kw_void)) 2747 return true; 2748 } 2749 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 2750 tok::kw_const) || 2751 // "of" is only a keyword if it appears after another identifier 2752 // (e.g. as "const x of y" in a for loop), or after a destructuring 2753 // operation (const [x, y] of z, const {a, b} of c). 2754 (Left.is(Keywords.kw_of) && Left.Previous && 2755 (Left.Previous->Tok.is(tok::identifier) || 2756 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 2757 (!Left.Previous || !Left.Previous->is(tok::period))) 2758 return true; 2759 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 2760 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 2761 return false; 2762 if (Left.is(Keywords.kw_as) && 2763 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 2764 return true; 2765 if (Left.is(tok::kw_default) && Left.Previous && 2766 Left.Previous->is(tok::kw_export)) 2767 return true; 2768 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 2769 return true; 2770 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 2771 return false; 2772 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 2773 return false; 2774 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 2775 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 2776 return false; 2777 if (Left.is(tok::ellipsis)) 2778 return false; 2779 if (Left.is(TT_TemplateCloser) && 2780 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 2781 Keywords.kw_implements, Keywords.kw_extends)) 2782 // Type assertions ('<type>expr') are not followed by whitespace. Other 2783 // locations that should have whitespace following are identified by the 2784 // above set of follower tokens. 2785 return false; 2786 if (Right.is(TT_JsNonNullAssertion)) 2787 return false; 2788 if (Left.is(TT_JsNonNullAssertion) && 2789 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 2790 return true; // "x! as string", "x! in y" 2791 } else if (Style.Language == FormatStyle::LK_Java) { 2792 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 2793 return true; 2794 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 2795 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never; 2796 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 2797 tok::kw_protected) || 2798 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 2799 Keywords.kw_native)) && 2800 Right.is(TT_TemplateOpener)) 2801 return true; 2802 } 2803 if (Left.is(TT_ImplicitStringLiteral)) 2804 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2805 if (Line.Type == LT_ObjCMethodDecl) { 2806 if (Left.is(TT_ObjCMethodSpecifier)) 2807 return true; 2808 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right)) 2809 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a 2810 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class 2811 // method declaration. 2812 return false; 2813 } 2814 if (Line.Type == LT_ObjCProperty && 2815 (Right.is(tok::equal) || Left.is(tok::equal))) 2816 return false; 2817 2818 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 2819 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 2820 return true; 2821 if (Right.is(TT_OverloadedOperatorLParen)) 2822 return spaceRequiredBeforeParens(Right); 2823 if (Left.is(tok::comma)) 2824 return true; 2825 if (Right.is(tok::comma)) 2826 return false; 2827 if (Right.is(TT_ObjCBlockLParen)) 2828 return true; 2829 if (Right.is(TT_CtorInitializerColon)) 2830 return Style.SpaceBeforeCtorInitializerColon; 2831 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 2832 return false; 2833 if (Right.is(TT_RangeBasedForLoopColon) && 2834 !Style.SpaceBeforeRangeBasedForLoopColon) 2835 return false; 2836 if (Right.is(tok::colon)) { 2837 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) || 2838 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 2839 return false; 2840 if (Right.is(TT_ObjCMethodExpr)) 2841 return false; 2842 if (Left.is(tok::question)) 2843 return false; 2844 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 2845 return false; 2846 if (Right.is(TT_DictLiteral)) 2847 return Style.SpacesInContainerLiterals; 2848 if (Right.is(TT_AttributeColon)) 2849 return false; 2850 return true; 2851 } 2852 if (Left.is(TT_UnaryOperator)) 2853 return (Style.SpaceAfterLogicalNot && Left.is(tok::exclaim)) || 2854 Right.is(TT_BinaryOperator); 2855 2856 // If the next token is a binary operator or a selector name, we have 2857 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 2858 if (Left.is(TT_CastRParen)) 2859 return Style.SpaceAfterCStyleCast || 2860 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 2861 2862 if (Left.is(tok::greater) && Right.is(tok::greater)) { 2863 if (Style.Language == FormatStyle::LK_TextProto || 2864 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 2865 return !Style.Cpp11BracedListStyle; 2866 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 2867 (Style.Standard < FormatStyle::LS_Cpp11 || Style.SpacesInAngles); 2868 } 2869 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 2870 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 2871 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 2872 return false; 2873 if (!Style.SpaceBeforeAssignmentOperators && Left.isNot(TT_TemplateCloser) && 2874 Right.getPrecedence() == prec::Assignment) 2875 return false; 2876 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) && 2877 (Left.is(tok::identifier) || Left.is(tok::kw_this))) 2878 return false; 2879 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 2880 // Generally don't remove existing spaces between an identifier and "::". 2881 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 2882 // this turns out to be too lenient, add analysis of the identifier itself. 2883 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2884 if (Right.is(tok::coloncolon) && !Left.isOneOf(tok::l_brace, tok::comment)) 2885 return (Left.is(TT_TemplateOpener) && 2886 Style.Standard < FormatStyle::LS_Cpp11) || 2887 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 2888 tok::kw___super, TT_TemplateCloser, 2889 TT_TemplateOpener)) || 2890 (Left.is(tok ::l_paren) && Style.SpacesInParentheses); 2891 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 2892 return Style.SpacesInAngles; 2893 // Space before TT_StructuredBindingLSquare. 2894 if (Right.is(TT_StructuredBindingLSquare)) 2895 return !Left.isOneOf(tok::amp, tok::ampamp) || 2896 Style.PointerAlignment != FormatStyle::PAS_Right; 2897 // Space before & or && following a TT_StructuredBindingLSquare. 2898 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 2899 Right.isOneOf(tok::amp, tok::ampamp)) 2900 return Style.PointerAlignment != FormatStyle::PAS_Left; 2901 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 2902 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 2903 !Right.is(tok::r_paren))) 2904 return true; 2905 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) && 2906 Right.isNot(TT_FunctionTypeLParen)) 2907 return spaceRequiredBeforeParens(Right); 2908 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 2909 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 2910 return false; 2911 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 2912 Line.startsWith(tok::hash)) 2913 return true; 2914 if (Right.is(TT_TrailingUnaryOperator)) 2915 return false; 2916 if (Left.is(TT_RegexLiteral)) 2917 return false; 2918 return spaceRequiredBetween(Line, Left, Right); 2919 } 2920 2921 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 2922 static bool isAllmanBrace(const FormatToken &Tok) { 2923 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block && 2924 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_LambdaLBrace, TT_DictLiteral); 2925 } 2926 2927 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 2928 const FormatToken &Right) { 2929 const FormatToken &Left = *Right.Previous; 2930 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 2931 return true; 2932 2933 if (Style.Language == FormatStyle::LK_JavaScript) { 2934 // FIXME: This might apply to other languages and token kinds. 2935 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 2936 Left.Previous->is(tok::string_literal)) 2937 return true; 2938 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 2939 Left.Previous && Left.Previous->is(tok::equal) && 2940 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 2941 tok::kw_const) && 2942 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 2943 // above. 2944 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 2945 // Object literals on the top level of a file are treated as "enum-style". 2946 // Each key/value pair is put on a separate line, instead of bin-packing. 2947 return true; 2948 if (Left.is(tok::l_brace) && Line.Level == 0 && 2949 (Line.startsWith(tok::kw_enum) || 2950 Line.startsWith(tok::kw_const, tok::kw_enum) || 2951 Line.startsWith(tok::kw_export, tok::kw_enum) || 2952 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 2953 // JavaScript top-level enum key/value pairs are put on separate lines 2954 // instead of bin-packing. 2955 return true; 2956 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 2957 !Left.Children.empty()) 2958 // Support AllowShortFunctionsOnASingleLine for JavaScript. 2959 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 2960 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 2961 (Left.NestingLevel == 0 && Line.Level == 0 && 2962 Style.AllowShortFunctionsOnASingleLine & 2963 FormatStyle::SFS_InlineOnly); 2964 } else if (Style.Language == FormatStyle::LK_Java) { 2965 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 2966 Right.Next->is(tok::string_literal)) 2967 return true; 2968 } else if (Style.Language == FormatStyle::LK_Cpp || 2969 Style.Language == FormatStyle::LK_ObjC || 2970 Style.Language == FormatStyle::LK_Proto || 2971 Style.Language == FormatStyle::LK_TableGen || 2972 Style.Language == FormatStyle::LK_TextProto) { 2973 if (Left.isStringLiteral() && Right.isStringLiteral()) 2974 return true; 2975 } 2976 2977 // If the last token before a '}', ']', or ')' is a comma or a trailing 2978 // comment, the intention is to insert a line break after it in order to make 2979 // shuffling around entries easier. Import statements, especially in 2980 // JavaScript, can be an exception to this rule. 2981 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 2982 const FormatToken *BeforeClosingBrace = nullptr; 2983 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 2984 (Style.Language == FormatStyle::LK_JavaScript && 2985 Left.is(tok::l_paren))) && 2986 Left.BlockKind != BK_Block && Left.MatchingParen) 2987 BeforeClosingBrace = Left.MatchingParen->Previous; 2988 else if (Right.MatchingParen && 2989 (Right.MatchingParen->isOneOf(tok::l_brace, 2990 TT_ArrayInitializerLSquare) || 2991 (Style.Language == FormatStyle::LK_JavaScript && 2992 Right.MatchingParen->is(tok::l_paren)))) 2993 BeforeClosingBrace = &Left; 2994 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 2995 BeforeClosingBrace->isTrailingComment())) 2996 return true; 2997 } 2998 2999 if (Right.is(tok::comment)) 3000 return Left.BlockKind != BK_BracedInit && 3001 Left.isNot(TT_CtorInitializerColon) && 3002 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 3003 if (Left.isTrailingComment()) 3004 return true; 3005 if (Right.Previous->IsUnterminatedLiteral) 3006 return true; 3007 if (Right.is(tok::lessless) && Right.Next && 3008 Right.Previous->is(tok::string_literal) && 3009 Right.Next->is(tok::string_literal)) 3010 return true; 3011 if (Right.Previous->ClosesTemplateDeclaration && 3012 Right.Previous->MatchingParen && 3013 Right.Previous->MatchingParen->NestingLevel == 0 && 3014 Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes) 3015 return true; 3016 if (Right.is(TT_CtorInitializerComma) && 3017 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3018 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 3019 return true; 3020 if (Right.is(TT_CtorInitializerColon) && 3021 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3022 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 3023 return true; 3024 // Break only if we have multiple inheritance. 3025 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 3026 Right.is(TT_InheritanceComma)) 3027 return true; 3028 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 3029 // Multiline raw string literals are special wrt. line breaks. The author 3030 // has made a deliberate choice and might have aligned the contents of the 3031 // string literal accordingly. Thus, we try keep existing line breaks. 3032 return Right.IsMultiline && Right.NewlinesBefore > 0; 3033 if ((Right.Previous->is(tok::l_brace) || 3034 (Right.Previous->is(tok::less) && Right.Previous->Previous && 3035 Right.Previous->Previous->is(tok::equal))) && 3036 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 3037 // Don't put enums or option definitions onto single lines in protocol 3038 // buffers. 3039 return true; 3040 } 3041 if (Right.is(TT_InlineASMBrace)) 3042 return Right.HasUnescapedNewline; 3043 if (isAllmanBrace(Left) || isAllmanBrace(Right)) 3044 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) || 3045 (Line.startsWith(tok::kw_typedef, tok::kw_enum) && 3046 Style.BraceWrapping.AfterEnum) || 3047 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 3048 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 3049 if (Left.is(TT_ObjCBlockLBrace) && 3050 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 3051 return true; 3052 3053 if (Left.is(TT_LambdaLBrace)) { 3054 if (Left.MatchingParen && Left.MatchingParen->Next && 3055 Left.MatchingParen->Next->isOneOf(tok::comma, tok::r_paren) && 3056 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline) 3057 return false; 3058 3059 if (Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_None || 3060 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline || 3061 (!Left.Children.empty() && 3062 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Empty)) 3063 return true; 3064 } 3065 3066 // Put multiple C# attributes on a new line. 3067 if (Style.isCSharp() && 3068 ((Left.is(TT_AttributeSquare) && Left.is(tok::r_square)) || 3069 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare) && 3070 Right.is(tok::l_square)))) 3071 return true; 3072 3073 // Put multiple Java annotation on a new line. 3074 if ((Style.Language == FormatStyle::LK_Java || 3075 Style.Language == FormatStyle::LK_JavaScript) && 3076 Left.is(TT_LeadingJavaAnnotation) && 3077 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 3078 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 3079 return true; 3080 3081 if (Right.is(TT_ProtoExtensionLSquare)) 3082 return true; 3083 3084 // In text proto instances if a submessage contains at least 2 entries and at 3085 // least one of them is a submessage, like A { ... B { ... } ... }, 3086 // put all of the entries of A on separate lines by forcing the selector of 3087 // the submessage B to be put on a newline. 3088 // 3089 // Example: these can stay on one line: 3090 // a { scalar_1: 1 scalar_2: 2 } 3091 // a { b { key: value } } 3092 // 3093 // and these entries need to be on a new line even if putting them all in one 3094 // line is under the column limit: 3095 // a { 3096 // scalar: 1 3097 // b { key: value } 3098 // } 3099 // 3100 // We enforce this by breaking before a submessage field that has previous 3101 // siblings, *and* breaking before a field that follows a submessage field. 3102 // 3103 // Be careful to exclude the case [proto.ext] { ... } since the `]` is 3104 // the TT_SelectorName there, but we don't want to break inside the brackets. 3105 // 3106 // Another edge case is @submessage { key: value }, which is a common 3107 // substitution placeholder. In this case we want to keep `@` and `submessage` 3108 // together. 3109 // 3110 // We ensure elsewhere that extensions are always on their own line. 3111 if ((Style.Language == FormatStyle::LK_Proto || 3112 Style.Language == FormatStyle::LK_TextProto) && 3113 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) { 3114 // Keep `@submessage` together in: 3115 // @submessage { key: value } 3116 if (Right.Previous && Right.Previous->is(tok::at)) 3117 return false; 3118 // Look for the scope opener after selector in cases like: 3119 // selector { ... 3120 // selector: { ... 3121 // selector: @base { ... 3122 FormatToken *LBrace = Right.Next; 3123 if (LBrace && LBrace->is(tok::colon)) { 3124 LBrace = LBrace->Next; 3125 if (LBrace && LBrace->is(tok::at)) { 3126 LBrace = LBrace->Next; 3127 if (LBrace) 3128 LBrace = LBrace->Next; 3129 } 3130 } 3131 if (LBrace && 3132 // The scope opener is one of {, [, <: 3133 // selector { ... } 3134 // selector [ ... ] 3135 // selector < ... > 3136 // 3137 // In case of selector { ... }, the l_brace is TT_DictLiteral. 3138 // In case of an empty selector {}, the l_brace is not TT_DictLiteral, 3139 // so we check for immediately following r_brace. 3140 ((LBrace->is(tok::l_brace) && 3141 (LBrace->is(TT_DictLiteral) || 3142 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) || 3143 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) { 3144 // If Left.ParameterCount is 0, then this submessage entry is not the 3145 // first in its parent submessage, and we want to break before this entry. 3146 // If Left.ParameterCount is greater than 0, then its parent submessage 3147 // might contain 1 or more entries and we want to break before this entry 3148 // if it contains at least 2 entries. We deal with this case later by 3149 // detecting and breaking before the next entry in the parent submessage. 3150 if (Left.ParameterCount == 0) 3151 return true; 3152 // However, if this submessage is the first entry in its parent 3153 // submessage, Left.ParameterCount might be 1 in some cases. 3154 // We deal with this case later by detecting an entry 3155 // following a closing paren of this submessage. 3156 } 3157 3158 // If this is an entry immediately following a submessage, it will be 3159 // preceded by a closing paren of that submessage, like in: 3160 // left---. .---right 3161 // v v 3162 // sub: { ... } key: value 3163 // If there was a comment between `}` an `key` above, then `key` would be 3164 // put on a new line anyways. 3165 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square)) 3166 return true; 3167 } 3168 3169 // Deal with lambda arguments in C++ - we want consistent line breaks whether 3170 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced 3171 // as aggressive line breaks are placed when the lambda is not the last arg. 3172 if ((Style.Language == FormatStyle::LK_Cpp || 3173 Style.Language == FormatStyle::LK_ObjC) && 3174 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 && 3175 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) { 3176 // Multiple lambdas in the same function call force line breaks. 3177 if (Left.BlockParameterCount > 1) 3178 return true; 3179 3180 // A lambda followed by another arg forces a line break. 3181 if (!Left.Role) 3182 return false; 3183 auto Comma = Left.Role->lastComma(); 3184 if (!Comma) 3185 return false; 3186 auto Next = Comma->getNextNonComment(); 3187 if (!Next) 3188 return false; 3189 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret)) 3190 return true; 3191 } 3192 3193 return false; 3194 } 3195 3196 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 3197 const FormatToken &Right) { 3198 const FormatToken &Left = *Right.Previous; 3199 3200 // Language-specific stuff. 3201 if (Style.Language == FormatStyle::LK_Java) { 3202 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3203 Keywords.kw_implements)) 3204 return false; 3205 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3206 Keywords.kw_implements)) 3207 return true; 3208 } else if (Style.Language == FormatStyle::LK_JavaScript) { 3209 const FormatToken *NonComment = Right.getPreviousNonComment(); 3210 if (NonComment && 3211 NonComment->isOneOf( 3212 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 3213 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 3214 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 3215 Keywords.kw_readonly, Keywords.kw_abstract, Keywords.kw_get, 3216 Keywords.kw_set, Keywords.kw_async, Keywords.kw_await)) 3217 return false; // Otherwise automatic semicolon insertion would trigger. 3218 if (Right.NestingLevel == 0 && 3219 (Left.Tok.getIdentifierInfo() || 3220 Left.isOneOf(tok::r_square, tok::r_paren)) && 3221 Right.isOneOf(tok::l_square, tok::l_paren)) 3222 return false; // Otherwise automatic semicolon insertion would trigger. 3223 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace)) 3224 return false; 3225 if (Left.is(TT_JsTypeColon)) 3226 return true; 3227 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;"). 3228 if (Left.is(tok::exclaim) && Right.is(tok::colon)) 3229 return false; 3230 // Look for is type annotations like: 3231 // function f(): a is B { ... } 3232 // Do not break before is in these cases. 3233 if (Right.is(Keywords.kw_is)) { 3234 const FormatToken *Next = Right.getNextNonComment(); 3235 // If `is` is followed by a colon, it's likely that it's a dict key, so 3236 // ignore it for this check. 3237 // For example this is common in Polymer: 3238 // Polymer({ 3239 // is: 'name', 3240 // ... 3241 // }); 3242 if (!Next || !Next->is(tok::colon)) 3243 return false; 3244 } 3245 if (Left.is(Keywords.kw_in)) 3246 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 3247 if (Right.is(Keywords.kw_in)) 3248 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 3249 if (Right.is(Keywords.kw_as)) 3250 return false; // must not break before as in 'x as type' casts 3251 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) { 3252 // extends and infer can appear as keywords in conditional types: 3253 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types 3254 // do not break before them, as the expressions are subject to ASI. 3255 return false; 3256 } 3257 if (Left.is(Keywords.kw_as)) 3258 return true; 3259 if (Left.is(TT_JsNonNullAssertion)) 3260 return true; 3261 if (Left.is(Keywords.kw_declare) && 3262 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 3263 Keywords.kw_function, tok::kw_class, tok::kw_enum, 3264 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 3265 Keywords.kw_let, tok::kw_const)) 3266 // See grammar for 'declare' statements at: 3267 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10 3268 return false; 3269 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 3270 Right.isOneOf(tok::identifier, tok::string_literal)) 3271 return false; // must not break in "module foo { ...}" 3272 if (Right.is(TT_TemplateString) && Right.closesScope()) 3273 return false; 3274 // Don't split tagged template literal so there is a break between the tag 3275 // identifier and template string. 3276 if (Left.is(tok::identifier) && Right.is(TT_TemplateString)) { 3277 return false; 3278 } 3279 if (Left.is(TT_TemplateString) && Left.opensScope()) 3280 return true; 3281 } 3282 3283 if (Left.is(tok::at)) 3284 return false; 3285 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 3286 return false; 3287 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 3288 return !Right.is(tok::l_paren); 3289 if (Right.is(TT_PointerOrReference)) 3290 return Line.IsMultiVariableDeclStmt || 3291 (Style.PointerAlignment == FormatStyle::PAS_Right && 3292 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 3293 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 3294 Right.is(tok::kw_operator)) 3295 return true; 3296 if (Left.is(TT_PointerOrReference)) 3297 return false; 3298 if (Right.isTrailingComment()) 3299 // We rely on MustBreakBefore being set correctly here as we should not 3300 // change the "binding" behavior of a comment. 3301 // The first comment in a braced lists is always interpreted as belonging to 3302 // the first list element. Otherwise, it should be placed outside of the 3303 // list. 3304 return Left.BlockKind == BK_BracedInit || 3305 (Left.is(TT_CtorInitializerColon) && 3306 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 3307 if (Left.is(tok::question) && Right.is(tok::colon)) 3308 return false; 3309 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 3310 return Style.BreakBeforeTernaryOperators; 3311 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 3312 return !Style.BreakBeforeTernaryOperators; 3313 if (Left.is(TT_InheritanceColon)) 3314 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon; 3315 if (Right.is(TT_InheritanceColon)) 3316 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon; 3317 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 3318 Left.isNot(TT_SelectorName)) 3319 return true; 3320 3321 if (Right.is(tok::colon) && 3322 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 3323 return false; 3324 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 3325 if (Style.Language == FormatStyle::LK_Proto || 3326 Style.Language == FormatStyle::LK_TextProto) { 3327 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 3328 return false; 3329 // Prevent cases like: 3330 // 3331 // submessage: 3332 // { key: valueeeeeeeeeeee } 3333 // 3334 // when the snippet does not fit into one line. 3335 // Prefer: 3336 // 3337 // submessage: { 3338 // key: valueeeeeeeeeeee 3339 // } 3340 // 3341 // instead, even if it is longer by one line. 3342 // 3343 // Note that this allows allows the "{" to go over the column limit 3344 // when the column limit is just between ":" and "{", but that does 3345 // not happen too often and alternative formattings in this case are 3346 // not much better. 3347 // 3348 // The code covers the cases: 3349 // 3350 // submessage: { ... } 3351 // submessage: < ... > 3352 // repeated: [ ... ] 3353 if (((Right.is(tok::l_brace) || Right.is(tok::less)) && 3354 Right.is(TT_DictLiteral)) || 3355 Right.is(TT_ArrayInitializerLSquare)) 3356 return false; 3357 } 3358 return true; 3359 } 3360 if (Right.is(tok::r_square) && Right.MatchingParen && 3361 Right.MatchingParen->is(TT_ProtoExtensionLSquare)) 3362 return false; 3363 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 3364 Right.Next->is(TT_ObjCMethodExpr))) 3365 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 3366 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 3367 return true; 3368 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 3369 return true; 3370 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 3371 TT_OverloadedOperator)) 3372 return false; 3373 if (Left.is(TT_RangeBasedForLoopColon)) 3374 return true; 3375 if (Right.is(TT_RangeBasedForLoopColon)) 3376 return false; 3377 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 3378 return true; 3379 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 3380 Left.is(tok::kw_operator)) 3381 return false; 3382 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 3383 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 3384 return false; 3385 if (Left.is(tok::equal) && Right.is(tok::l_brace) && 3386 !Style.Cpp11BracedListStyle) 3387 return false; 3388 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen)) 3389 return false; 3390 if (Left.is(tok::l_paren) && Left.Previous && 3391 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 3392 return false; 3393 if (Right.is(TT_ImplicitStringLiteral)) 3394 return false; 3395 3396 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser)) 3397 return false; 3398 if (Right.is(tok::r_square) && Right.MatchingParen && 3399 Right.MatchingParen->is(TT_LambdaLSquare)) 3400 return false; 3401 3402 // We only break before r_brace if there was a corresponding break before 3403 // the l_brace, which is tracked by BreakBeforeClosingBrace. 3404 if (Right.is(tok::r_brace)) 3405 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block; 3406 3407 // Allow breaking after a trailing annotation, e.g. after a method 3408 // declaration. 3409 if (Left.is(TT_TrailingAnnotation)) 3410 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 3411 tok::less, tok::coloncolon); 3412 3413 if (Right.is(tok::kw___attribute) || 3414 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) 3415 return true; 3416 3417 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 3418 return true; 3419 3420 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 3421 return true; 3422 3423 if (Left.is(TT_CtorInitializerColon)) 3424 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 3425 if (Right.is(TT_CtorInitializerColon)) 3426 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 3427 if (Left.is(TT_CtorInitializerComma) && 3428 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 3429 return false; 3430 if (Right.is(TT_CtorInitializerComma) && 3431 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 3432 return true; 3433 if (Left.is(TT_InheritanceComma) && 3434 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 3435 return false; 3436 if (Right.is(TT_InheritanceComma) && 3437 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 3438 return true; 3439 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 3440 (Left.is(tok::less) && Right.is(tok::less))) 3441 return false; 3442 if (Right.is(TT_BinaryOperator) && 3443 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 3444 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 3445 Right.getPrecedence() != prec::Assignment)) 3446 return true; 3447 if (Left.is(TT_ArrayInitializerLSquare)) 3448 return true; 3449 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 3450 return true; 3451 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 3452 !Left.isOneOf(tok::arrowstar, tok::lessless) && 3453 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 3454 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 3455 Left.getPrecedence() == prec::Assignment)) 3456 return true; 3457 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 3458 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) 3459 return false; 3460 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 3461 tok::kw_class, tok::kw_struct, tok::comment) || 3462 Right.isMemberAccess() || 3463 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 3464 tok::colon, tok::l_square, tok::at) || 3465 (Left.is(tok::r_paren) && 3466 Right.isOneOf(tok::identifier, tok::kw_const)) || 3467 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 3468 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 3469 } 3470 3471 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 3472 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 3473 const FormatToken *Tok = Line.First; 3474 while (Tok) { 3475 llvm::errs() << " M=" << Tok->MustBreakBefore 3476 << " C=" << Tok->CanBreakBefore 3477 << " T=" << getTokenTypeName(Tok->Type) 3478 << " S=" << Tok->SpacesRequiredBefore 3479 << " B=" << Tok->BlockParameterCount 3480 << " BK=" << Tok->BlockKind << " P=" << Tok->SplitPenalty 3481 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 3482 << " PPK=" << Tok->PackingKind << " FakeLParens="; 3483 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 3484 llvm::errs() << Tok->FakeLParens[i] << "/"; 3485 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 3486 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo(); 3487 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 3488 if (!Tok->Next) 3489 assert(Tok == Line.Last); 3490 Tok = Tok->Next; 3491 } 3492 llvm::errs() << "----\n"; 3493 } 3494 3495 } // namespace format 3496 } // namespace clang 3497