1 //===--- TokenAnnotator.cpp - Format C++ code -----------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// 9 /// \file 10 /// This file implements a token annotator, i.e. creates 11 /// \c AnnotatedTokens out of \c FormatTokens with required extra information. 12 /// 13 //===----------------------------------------------------------------------===// 14 15 #include "TokenAnnotator.h" 16 #include "FormatToken.h" 17 #include "clang/Basic/SourceManager.h" 18 #include "clang/Basic/TokenKinds.h" 19 #include "llvm/ADT/SmallPtrSet.h" 20 #include "llvm/Support/Debug.h" 21 22 #define DEBUG_TYPE "format-token-annotator" 23 24 namespace clang { 25 namespace format { 26 27 namespace { 28 29 /// Returns \c true if the token can be used as an identifier in 30 /// an Objective-C \c @selector, \c false otherwise. 31 /// 32 /// Because getFormattingLangOpts() always lexes source code as 33 /// Objective-C++, C++ keywords like \c new and \c delete are 34 /// lexed as tok::kw_*, not tok::identifier, even for Objective-C. 35 /// 36 /// For Objective-C and Objective-C++, both identifiers and keywords 37 /// are valid inside @selector(...) (or a macro which 38 /// invokes @selector(...)). So, we allow treat any identifier or 39 /// keyword as a potential Objective-C selector component. 40 static bool canBeObjCSelectorComponent(const FormatToken &Tok) { 41 return Tok.Tok.getIdentifierInfo() != nullptr; 42 } 43 44 /// With `Left` being '(', check if we're at either `[...](` or 45 /// `[...]<...>(`, where the [ opens a lambda capture list. 46 static bool isLambdaParameterList(const FormatToken *Left) { 47 // Skip <...> if present. 48 if (Left->Previous && Left->Previous->is(tok::greater) && 49 Left->Previous->MatchingParen && 50 Left->Previous->MatchingParen->is(TT_TemplateOpener)) 51 Left = Left->Previous->MatchingParen; 52 53 // Check for `[...]`. 54 return Left->Previous && Left->Previous->is(tok::r_square) && 55 Left->Previous->MatchingParen && 56 Left->Previous->MatchingParen->is(TT_LambdaLSquare); 57 } 58 59 /// A parser that gathers additional information about tokens. 60 /// 61 /// The \c TokenAnnotator tries to match parenthesis and square brakets and 62 /// store a parenthesis levels. It also tries to resolve matching "<" and ">" 63 /// into template parameter lists. 64 class AnnotatingParser { 65 public: 66 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line, 67 const AdditionalKeywords &Keywords) 68 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false), 69 Keywords(Keywords) { 70 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false)); 71 resetTokenMetadata(CurrentToken); 72 } 73 74 private: 75 bool parseAngle() { 76 if (!CurrentToken || !CurrentToken->Previous) 77 return false; 78 if (NonTemplateLess.count(CurrentToken->Previous)) 79 return false; 80 81 const FormatToken &Previous = *CurrentToken->Previous; // The '<'. 82 if (Previous.Previous) { 83 if (Previous.Previous->Tok.isLiteral()) 84 return false; 85 if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 && 86 (!Previous.Previous->MatchingParen || 87 !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen))) 88 return false; 89 } 90 91 FormatToken *Left = CurrentToken->Previous; 92 Left->ParentBracket = Contexts.back().ContextKind; 93 ScopedContextCreator ContextCreator(*this, tok::less, 12); 94 95 // If this angle is in the context of an expression, we need to be more 96 // hesitant to detect it as opening template parameters. 97 bool InExprContext = Contexts.back().IsExpression; 98 99 Contexts.back().IsExpression = false; 100 // If there's a template keyword before the opening angle bracket, this is a 101 // template parameter, not an argument. 102 Contexts.back().InTemplateArgument = 103 Left->Previous && Left->Previous->Tok.isNot(tok::kw_template); 104 105 if (Style.Language == FormatStyle::LK_Java && 106 CurrentToken->is(tok::question)) 107 next(); 108 109 while (CurrentToken) { 110 if (CurrentToken->is(tok::greater)) { 111 Left->MatchingParen = CurrentToken; 112 CurrentToken->MatchingParen = Left; 113 // In TT_Proto, we must distignuish between: 114 // map<key, value> 115 // msg < item: data > 116 // msg: < item: data > 117 // In TT_TextProto, map<key, value> does not occur. 118 if (Style.Language == FormatStyle::LK_TextProto || 119 (Style.Language == FormatStyle::LK_Proto && Left->Previous && 120 Left->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) 121 CurrentToken->Type = TT_DictLiteral; 122 else 123 CurrentToken->Type = TT_TemplateCloser; 124 next(); 125 return true; 126 } 127 if (CurrentToken->is(tok::question) && 128 Style.Language == FormatStyle::LK_Java) { 129 next(); 130 continue; 131 } 132 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) || 133 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext && 134 Style.Language != FormatStyle::LK_Proto && 135 Style.Language != FormatStyle::LK_TextProto)) 136 return false; 137 // If a && or || is found and interpreted as a binary operator, this set 138 // of angles is likely part of something like "a < b && c > d". If the 139 // angles are inside an expression, the ||/&& might also be a binary 140 // operator that was misinterpreted because we are parsing template 141 // parameters. 142 // FIXME: This is getting out of hand, write a decent parser. 143 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) && 144 CurrentToken->Previous->is(TT_BinaryOperator) && 145 Contexts[Contexts.size() - 2].IsExpression && 146 !Line.startsWith(tok::kw_template)) 147 return false; 148 updateParameterCount(Left, CurrentToken); 149 if (Style.Language == FormatStyle::LK_Proto) { 150 if (FormatToken *Previous = CurrentToken->getPreviousNonComment()) { 151 if (CurrentToken->is(tok::colon) || 152 (CurrentToken->isOneOf(tok::l_brace, tok::less) && 153 Previous->isNot(tok::colon))) 154 Previous->Type = TT_SelectorName; 155 } 156 } 157 if (!consumeToken()) 158 return false; 159 } 160 return false; 161 } 162 163 bool parseParens(bool LookForDecls = false) { 164 if (!CurrentToken) 165 return false; 166 FormatToken *Left = CurrentToken->Previous; 167 Left->ParentBracket = Contexts.back().ContextKind; 168 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1); 169 170 // FIXME: This is a bit of a hack. Do better. 171 Contexts.back().ColonIsForRangeExpr = 172 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr; 173 174 bool StartsObjCMethodExpr = false; 175 if (FormatToken *MaybeSel = Left->Previous) { 176 // @selector( starts a selector. 177 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous && 178 MaybeSel->Previous->is(tok::at)) { 179 StartsObjCMethodExpr = true; 180 } 181 } 182 183 if (Left->is(TT_OverloadedOperatorLParen)) { 184 Contexts.back().IsExpression = false; 185 } else if (Style.Language == FormatStyle::LK_JavaScript && 186 (Line.startsWith(Keywords.kw_type, tok::identifier) || 187 Line.startsWith(tok::kw_export, Keywords.kw_type, 188 tok::identifier))) { 189 // type X = (...); 190 // export type X = (...); 191 Contexts.back().IsExpression = false; 192 } else if (Left->Previous && 193 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_decltype, 194 tok::kw_while, tok::l_paren, 195 tok::comma) || 196 Left->Previous->isIf() || 197 Left->Previous->is(TT_BinaryOperator))) { 198 // static_assert, if and while usually contain expressions. 199 Contexts.back().IsExpression = true; 200 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 201 (Left->Previous->is(Keywords.kw_function) || 202 (Left->Previous->endsSequence(tok::identifier, 203 Keywords.kw_function)))) { 204 // function(...) or function f(...) 205 Contexts.back().IsExpression = false; 206 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 207 Left->Previous->is(TT_JsTypeColon)) { 208 // let x: (SomeType); 209 Contexts.back().IsExpression = false; 210 } else if (isLambdaParameterList(Left)) { 211 // This is a parameter list of a lambda expression. 212 Contexts.back().IsExpression = false; 213 } else if (Line.InPPDirective && 214 (!Left->Previous || !Left->Previous->is(tok::identifier))) { 215 Contexts.back().IsExpression = true; 216 } else if (Contexts[Contexts.size() - 2].CaretFound) { 217 // This is the parameter list of an ObjC block. 218 Contexts.back().IsExpression = false; 219 } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) { 220 Left->Type = TT_AttributeParen; 221 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) { 222 // The first argument to a foreach macro is a declaration. 223 Contexts.back().IsForEachMacro = true; 224 Contexts.back().IsExpression = false; 225 } else if (Left->Previous && Left->Previous->MatchingParen && 226 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) { 227 Contexts.back().IsExpression = false; 228 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) { 229 bool IsForOrCatch = 230 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch); 231 Contexts.back().IsExpression = !IsForOrCatch; 232 } 233 234 if (StartsObjCMethodExpr) { 235 Contexts.back().ColonIsObjCMethodExpr = true; 236 Left->Type = TT_ObjCMethodExpr; 237 } 238 239 // MightBeFunctionType and ProbablyFunctionType are used for 240 // function pointer and reference types as well as Objective-C 241 // block types: 242 // 243 // void (*FunctionPointer)(void); 244 // void (&FunctionReference)(void); 245 // void (^ObjCBlock)(void); 246 bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression; 247 bool ProbablyFunctionType = 248 CurrentToken->isOneOf(tok::star, tok::amp, tok::caret); 249 bool HasMultipleLines = false; 250 bool HasMultipleParametersOnALine = false; 251 bool MightBeObjCForRangeLoop = 252 Left->Previous && Left->Previous->is(tok::kw_for); 253 FormatToken *PossibleObjCForInToken = nullptr; 254 while (CurrentToken) { 255 // LookForDecls is set when "if (" has been seen. Check for 256 // 'identifier' '*' 'identifier' followed by not '=' -- this 257 // '*' has to be a binary operator but determineStarAmpUsage() will 258 // categorize it as an unary operator, so set the right type here. 259 if (LookForDecls && CurrentToken->Next) { 260 FormatToken *Prev = CurrentToken->getPreviousNonComment(); 261 if (Prev) { 262 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 263 FormatToken *Next = CurrentToken->Next; 264 if (PrevPrev && PrevPrev->is(tok::identifier) && 265 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) && 266 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) { 267 Prev->Type = TT_BinaryOperator; 268 LookForDecls = false; 269 } 270 } 271 } 272 273 if (CurrentToken->Previous->is(TT_PointerOrReference) && 274 CurrentToken->Previous->Previous->isOneOf(tok::l_paren, 275 tok::coloncolon)) 276 ProbablyFunctionType = true; 277 if (CurrentToken->is(tok::comma)) 278 MightBeFunctionType = false; 279 if (CurrentToken->Previous->is(TT_BinaryOperator)) 280 Contexts.back().IsExpression = true; 281 if (CurrentToken->is(tok::r_paren)) { 282 if (MightBeFunctionType && ProbablyFunctionType && CurrentToken->Next && 283 (CurrentToken->Next->is(tok::l_paren) || 284 (CurrentToken->Next->is(tok::l_square) && Line.MustBeDeclaration))) 285 Left->Type = Left->Next->is(tok::caret) ? TT_ObjCBlockLParen 286 : TT_FunctionTypeLParen; 287 Left->MatchingParen = CurrentToken; 288 CurrentToken->MatchingParen = Left; 289 290 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) && 291 Left->Previous && Left->Previous->is(tok::l_paren)) { 292 // Detect the case where macros are used to generate lambdas or 293 // function bodies, e.g.: 294 // auto my_lambda = MARCO((Type *type, int i) { .. body .. }); 295 for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) { 296 if (Tok->is(TT_BinaryOperator) && 297 Tok->isOneOf(tok::star, tok::amp, tok::ampamp)) 298 Tok->Type = TT_PointerOrReference; 299 } 300 } 301 302 if (StartsObjCMethodExpr) { 303 CurrentToken->Type = TT_ObjCMethodExpr; 304 if (Contexts.back().FirstObjCSelectorName) { 305 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 306 Contexts.back().LongestObjCSelectorName; 307 } 308 } 309 310 if (Left->is(TT_AttributeParen)) 311 CurrentToken->Type = TT_AttributeParen; 312 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation)) 313 CurrentToken->Type = TT_JavaAnnotation; 314 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation)) 315 CurrentToken->Type = TT_LeadingJavaAnnotation; 316 if (Left->Previous && Left->Previous->is(TT_AttributeSquare)) 317 CurrentToken->Type = TT_AttributeSquare; 318 319 if (!HasMultipleLines) 320 Left->PackingKind = PPK_Inconclusive; 321 else if (HasMultipleParametersOnALine) 322 Left->PackingKind = PPK_BinPacked; 323 else 324 Left->PackingKind = PPK_OnePerLine; 325 326 next(); 327 return true; 328 } 329 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace)) 330 return false; 331 332 if (CurrentToken->is(tok::l_brace)) 333 Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen 334 if (CurrentToken->is(tok::comma) && CurrentToken->Next && 335 !CurrentToken->Next->HasUnescapedNewline && 336 !CurrentToken->Next->isTrailingComment()) 337 HasMultipleParametersOnALine = true; 338 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) || 339 CurrentToken->Previous->isSimpleTypeSpecifier()) && 340 !CurrentToken->is(tok::l_brace)) 341 Contexts.back().IsExpression = false; 342 if (CurrentToken->isOneOf(tok::semi, tok::colon)) { 343 MightBeObjCForRangeLoop = false; 344 if (PossibleObjCForInToken) { 345 PossibleObjCForInToken->Type = TT_Unknown; 346 PossibleObjCForInToken = nullptr; 347 } 348 } 349 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) { 350 PossibleObjCForInToken = CurrentToken; 351 PossibleObjCForInToken->Type = TT_ObjCForIn; 352 } 353 // When we discover a 'new', we set CanBeExpression to 'false' in order to 354 // parse the type correctly. Reset that after a comma. 355 if (CurrentToken->is(tok::comma)) 356 Contexts.back().CanBeExpression = true; 357 358 FormatToken *Tok = CurrentToken; 359 if (!consumeToken()) 360 return false; 361 updateParameterCount(Left, Tok); 362 if (CurrentToken && CurrentToken->HasUnescapedNewline) 363 HasMultipleLines = true; 364 } 365 return false; 366 } 367 368 bool isCSharpAttributeSpecifier(const FormatToken &Tok) { 369 if (!Style.isCSharp()) 370 return false; 371 372 const FormatToken *AttrTok = Tok.Next; 373 if (!AttrTok) 374 return false; 375 376 // Just an empty declaration e.g. string []. 377 if (AttrTok->is(tok::r_square)) 378 return false; 379 380 // Move along the tokens inbetween the '[' and ']' e.g. [STAThread]. 381 while (AttrTok && AttrTok->isNot(tok::r_square)) { 382 AttrTok = AttrTok->Next; 383 } 384 385 if (!AttrTok) 386 return false; 387 388 // Move past the end of ']'. 389 AttrTok = AttrTok->Next; 390 if (!AttrTok) 391 return false; 392 393 // Limit this to being an access modifier that follows. 394 if (AttrTok->isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected, 395 tok::kw_class, tok::kw_static, tok::l_square, 396 Keywords.kw_internal)) { 397 return true; 398 } 399 400 // incase its a [XXX] retval func(.... 401 if (AttrTok->Next && 402 AttrTok->Next->startsSequence(tok::identifier, tok::l_paren)) 403 return true; 404 405 return false; 406 } 407 408 bool isCpp11AttributeSpecifier(const FormatToken &Tok) { 409 if (!Style.isCpp() || !Tok.startsSequence(tok::l_square, tok::l_square)) 410 return false; 411 // The first square bracket is part of an ObjC array literal 412 if (Tok.Previous && Tok.Previous->is(tok::at)) { 413 return false; 414 } 415 const FormatToken *AttrTok = Tok.Next->Next; 416 if (!AttrTok) 417 return false; 418 // C++17 '[[using ns: foo, bar(baz, blech)]]' 419 // We assume nobody will name an ObjC variable 'using'. 420 if (AttrTok->startsSequence(tok::kw_using, tok::identifier, tok::colon)) 421 return true; 422 if (AttrTok->isNot(tok::identifier)) 423 return false; 424 while (AttrTok && !AttrTok->startsSequence(tok::r_square, tok::r_square)) { 425 // ObjC message send. We assume nobody will use : in a C++11 attribute 426 // specifier parameter, although this is technically valid: 427 // [[foo(:)]]. 428 if (AttrTok->is(tok::colon) || 429 AttrTok->startsSequence(tok::identifier, tok::identifier) || 430 AttrTok->startsSequence(tok::r_paren, tok::identifier)) 431 return false; 432 if (AttrTok->is(tok::ellipsis)) 433 return true; 434 AttrTok = AttrTok->Next; 435 } 436 return AttrTok && AttrTok->startsSequence(tok::r_square, tok::r_square); 437 } 438 439 bool parseSquare() { 440 if (!CurrentToken) 441 return false; 442 443 // A '[' could be an index subscript (after an identifier or after 444 // ')' or ']'), it could be the start of an Objective-C method 445 // expression, it could the start of an Objective-C array literal, 446 // or it could be a C++ attribute specifier [[foo::bar]]. 447 FormatToken *Left = CurrentToken->Previous; 448 Left->ParentBracket = Contexts.back().ContextKind; 449 FormatToken *Parent = Left->getPreviousNonComment(); 450 451 // Cases where '>' is followed by '['. 452 // In C++, this can happen either in array of templates (foo<int>[10]) 453 // or when array is a nested template type (unique_ptr<type1<type2>[]>). 454 bool CppArrayTemplates = 455 Style.isCpp() && Parent && Parent->is(TT_TemplateCloser) && 456 (Contexts.back().CanBeExpression || Contexts.back().IsExpression || 457 Contexts.back().InTemplateArgument); 458 459 bool IsCpp11AttributeSpecifier = isCpp11AttributeSpecifier(*Left) || 460 Contexts.back().InCpp11AttributeSpecifier; 461 462 // Treat C# Attributes [STAThread] much like C++ attributes [[...]]. 463 bool IsCSharp11AttributeSpecifier = 464 isCSharpAttributeSpecifier(*Left) || 465 Contexts.back().InCSharpAttributeSpecifier; 466 467 bool InsideInlineASM = Line.startsWith(tok::kw_asm); 468 bool IsCppStructuredBinding = Left->isCppStructuredBinding(Style); 469 bool StartsObjCMethodExpr = 470 !IsCppStructuredBinding && !InsideInlineASM && !CppArrayTemplates && 471 Style.isCpp() && !IsCpp11AttributeSpecifier && 472 Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) && 473 !CurrentToken->isOneOf(tok::l_brace, tok::r_square) && 474 (!Parent || 475 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren, 476 tok::kw_return, tok::kw_throw) || 477 Parent->isUnaryOperator() || 478 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 479 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) || 480 (getBinOpPrecedence(Parent->Tok.getKind(), true, true) > 481 prec::Unknown)); 482 bool ColonFound = false; 483 484 unsigned BindingIncrease = 1; 485 if (IsCppStructuredBinding) { 486 Left->Type = TT_StructuredBindingLSquare; 487 } else if (Left->is(TT_Unknown)) { 488 if (StartsObjCMethodExpr) { 489 Left->Type = TT_ObjCMethodExpr; 490 } else if (IsCpp11AttributeSpecifier) { 491 Left->Type = TT_AttributeSquare; 492 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent && 493 Contexts.back().ContextKind == tok::l_brace && 494 Parent->isOneOf(tok::l_brace, tok::comma)) { 495 Left->Type = TT_JsComputedPropertyName; 496 } else if (Style.isCpp() && Contexts.back().ContextKind == tok::l_brace && 497 Parent && Parent->isOneOf(tok::l_brace, tok::comma)) { 498 Left->Type = TT_DesignatedInitializerLSquare; 499 } else if (IsCSharp11AttributeSpecifier) { 500 Left->Type = TT_AttributeSquare; 501 } else if (CurrentToken->is(tok::r_square) && Parent && 502 Parent->is(TT_TemplateCloser)) { 503 Left->Type = TT_ArraySubscriptLSquare; 504 } else if (Style.Language == FormatStyle::LK_Proto || 505 Style.Language == FormatStyle::LK_TextProto) { 506 // Square braces in LK_Proto can either be message field attributes: 507 // 508 // optional Aaa aaa = 1 [ 509 // (aaa) = aaa 510 // ]; 511 // 512 // extensions 123 [ 513 // (aaa) = aaa 514 // ]; 515 // 516 // or text proto extensions (in options): 517 // 518 // option (Aaa.options) = { 519 // [type.type/type] { 520 // key: value 521 // } 522 // } 523 // 524 // or repeated fields (in options): 525 // 526 // option (Aaa.options) = { 527 // keys: [ 1, 2, 3 ] 528 // } 529 // 530 // In the first and the third case we want to spread the contents inside 531 // the square braces; in the second we want to keep them inline. 532 Left->Type = TT_ArrayInitializerLSquare; 533 if (!Left->endsSequence(tok::l_square, tok::numeric_constant, 534 tok::equal) && 535 !Left->endsSequence(tok::l_square, tok::numeric_constant, 536 tok::identifier) && 537 !Left->endsSequence(tok::l_square, tok::colon, TT_SelectorName)) { 538 Left->Type = TT_ProtoExtensionLSquare; 539 BindingIncrease = 10; 540 } 541 } else if (!CppArrayTemplates && Parent && 542 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at, 543 tok::comma, tok::l_paren, tok::l_square, 544 tok::question, tok::colon, tok::kw_return, 545 // Should only be relevant to JavaScript: 546 tok::kw_default)) { 547 Left->Type = TT_ArrayInitializerLSquare; 548 } else { 549 BindingIncrease = 10; 550 Left->Type = TT_ArraySubscriptLSquare; 551 } 552 } 553 554 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease); 555 Contexts.back().IsExpression = true; 556 if (Style.Language == FormatStyle::LK_JavaScript && Parent && 557 Parent->is(TT_JsTypeColon)) 558 Contexts.back().IsExpression = false; 559 560 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr; 561 Contexts.back().InCpp11AttributeSpecifier = IsCpp11AttributeSpecifier; 562 Contexts.back().InCSharpAttributeSpecifier = IsCSharp11AttributeSpecifier; 563 564 while (CurrentToken) { 565 if (CurrentToken->is(tok::r_square)) { 566 if (IsCpp11AttributeSpecifier) 567 CurrentToken->Type = TT_AttributeSquare; 568 if (IsCSharp11AttributeSpecifier) 569 CurrentToken->Type = TT_AttributeSquare; 570 else if (((CurrentToken->Next && 571 CurrentToken->Next->is(tok::l_paren)) || 572 (CurrentToken->Previous && 573 CurrentToken->Previous->Previous == Left)) && 574 Left->is(TT_ObjCMethodExpr)) { 575 // An ObjC method call is rarely followed by an open parenthesis. It 576 // also can't be composed of just one token, unless it's a macro that 577 // will be expanded to more tokens. 578 // FIXME: Do we incorrectly label ":" with this? 579 StartsObjCMethodExpr = false; 580 Left->Type = TT_Unknown; 581 } 582 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) { 583 CurrentToken->Type = TT_ObjCMethodExpr; 584 // If we haven't seen a colon yet, make sure the last identifier 585 // before the r_square is tagged as a selector name component. 586 if (!ColonFound && CurrentToken->Previous && 587 CurrentToken->Previous->is(TT_Unknown) && 588 canBeObjCSelectorComponent(*CurrentToken->Previous)) 589 CurrentToken->Previous->Type = TT_SelectorName; 590 // determineStarAmpUsage() thinks that '*' '[' is allocating an 591 // array of pointers, but if '[' starts a selector then '*' is a 592 // binary operator. 593 if (Parent && Parent->is(TT_PointerOrReference)) 594 Parent->Type = TT_BinaryOperator; 595 } 596 // An arrow after an ObjC method expression is not a lambda arrow. 597 if (CurrentToken->Type == TT_ObjCMethodExpr && CurrentToken->Next && 598 CurrentToken->Next->is(TT_LambdaArrow)) 599 CurrentToken->Next->Type = TT_Unknown; 600 Left->MatchingParen = CurrentToken; 601 CurrentToken->MatchingParen = Left; 602 // FirstObjCSelectorName is set when a colon is found. This does 603 // not work, however, when the method has no parameters. 604 // Here, we set FirstObjCSelectorName when the end of the method call is 605 // reached, in case it was not set already. 606 if (!Contexts.back().FirstObjCSelectorName) { 607 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 608 if (Previous && Previous->is(TT_SelectorName)) { 609 Previous->ObjCSelectorNameParts = 1; 610 Contexts.back().FirstObjCSelectorName = Previous; 611 } 612 } else { 613 Left->ParameterCount = 614 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 615 } 616 if (Contexts.back().FirstObjCSelectorName) { 617 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 618 Contexts.back().LongestObjCSelectorName; 619 if (Left->BlockParameterCount > 1) 620 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0; 621 } 622 next(); 623 return true; 624 } 625 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace)) 626 return false; 627 if (CurrentToken->is(tok::colon)) { 628 if (IsCpp11AttributeSpecifier && 629 CurrentToken->endsSequence(tok::colon, tok::identifier, 630 tok::kw_using)) { 631 // Remember that this is a [[using ns: foo]] C++ attribute, so we 632 // don't add a space before the colon (unlike other colons). 633 CurrentToken->Type = TT_AttributeColon; 634 } else if (Left->isOneOf(TT_ArraySubscriptLSquare, 635 TT_DesignatedInitializerLSquare)) { 636 Left->Type = TT_ObjCMethodExpr; 637 StartsObjCMethodExpr = true; 638 Contexts.back().ColonIsObjCMethodExpr = true; 639 if (Parent && Parent->is(tok::r_paren)) 640 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 641 Parent->Type = TT_CastRParen; 642 } 643 ColonFound = true; 644 } 645 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) && 646 !ColonFound) 647 Left->Type = TT_ArrayInitializerLSquare; 648 FormatToken *Tok = CurrentToken; 649 if (!consumeToken()) 650 return false; 651 updateParameterCount(Left, Tok); 652 } 653 return false; 654 } 655 656 bool parseBrace() { 657 if (CurrentToken) { 658 FormatToken *Left = CurrentToken->Previous; 659 Left->ParentBracket = Contexts.back().ContextKind; 660 661 if (Contexts.back().CaretFound) 662 Left->Type = TT_ObjCBlockLBrace; 663 Contexts.back().CaretFound = false; 664 665 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1); 666 Contexts.back().ColonIsDictLiteral = true; 667 if (Left->BlockKind == BK_BracedInit) 668 Contexts.back().IsExpression = true; 669 if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 670 Left->Previous->is(TT_JsTypeColon)) 671 Contexts.back().IsExpression = false; 672 673 while (CurrentToken) { 674 if (CurrentToken->is(tok::r_brace)) { 675 Left->MatchingParen = CurrentToken; 676 CurrentToken->MatchingParen = Left; 677 next(); 678 return true; 679 } 680 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 681 return false; 682 updateParameterCount(Left, CurrentToken); 683 if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) { 684 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 685 if (Previous->is(TT_JsTypeOptionalQuestion)) 686 Previous = Previous->getPreviousNonComment(); 687 if ((CurrentToken->is(tok::colon) && 688 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) || 689 Style.Language == FormatStyle::LK_Proto || 690 Style.Language == FormatStyle::LK_TextProto) { 691 Left->Type = TT_DictLiteral; 692 if (Previous->Tok.getIdentifierInfo() || 693 Previous->is(tok::string_literal)) 694 Previous->Type = TT_SelectorName; 695 } 696 if (CurrentToken->is(tok::colon) || 697 Style.Language == FormatStyle::LK_JavaScript) 698 Left->Type = TT_DictLiteral; 699 } 700 if (CurrentToken->is(tok::comma) && 701 Style.Language == FormatStyle::LK_JavaScript) 702 Left->Type = TT_DictLiteral; 703 if (!consumeToken()) 704 return false; 705 } 706 } 707 return true; 708 } 709 710 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 711 // For ObjC methods, the number of parameters is calculated differently as 712 // method declarations have a different structure (the parameters are not 713 // inside a bracket scope). 714 if (Current->is(tok::l_brace) && Current->BlockKind == BK_Block) 715 ++Left->BlockParameterCount; 716 if (Current->is(tok::comma)) { 717 ++Left->ParameterCount; 718 if (!Left->Role) 719 Left->Role.reset(new CommaSeparatedList(Style)); 720 Left->Role->CommaFound(Current); 721 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 722 Left->ParameterCount = 1; 723 } 724 } 725 726 bool parseConditional() { 727 while (CurrentToken) { 728 if (CurrentToken->is(tok::colon)) { 729 CurrentToken->Type = TT_ConditionalExpr; 730 next(); 731 return true; 732 } 733 if (!consumeToken()) 734 return false; 735 } 736 return false; 737 } 738 739 bool parseTemplateDeclaration() { 740 if (CurrentToken && CurrentToken->is(tok::less)) { 741 CurrentToken->Type = TT_TemplateOpener; 742 next(); 743 if (!parseAngle()) 744 return false; 745 if (CurrentToken) 746 CurrentToken->Previous->ClosesTemplateDeclaration = true; 747 return true; 748 } 749 return false; 750 } 751 752 bool consumeToken() { 753 FormatToken *Tok = CurrentToken; 754 next(); 755 switch (Tok->Tok.getKind()) { 756 case tok::plus: 757 case tok::minus: 758 if (!Tok->Previous && Line.MustBeDeclaration) 759 Tok->Type = TT_ObjCMethodSpecifier; 760 break; 761 case tok::colon: 762 if (!Tok->Previous) 763 return false; 764 // Colons from ?: are handled in parseConditional(). 765 if (Style.Language == FormatStyle::LK_JavaScript) { 766 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop 767 (Contexts.size() == 1 && // switch/case labels 768 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) || 769 Contexts.back().ContextKind == tok::l_paren || // function params 770 Contexts.back().ContextKind == tok::l_square || // array type 771 (!Contexts.back().IsExpression && 772 Contexts.back().ContextKind == tok::l_brace) || // object type 773 (Contexts.size() == 1 && 774 Line.MustBeDeclaration)) { // method/property declaration 775 Contexts.back().IsExpression = false; 776 Tok->Type = TT_JsTypeColon; 777 break; 778 } 779 } 780 if (Contexts.back().ColonIsDictLiteral || 781 Style.Language == FormatStyle::LK_Proto || 782 Style.Language == FormatStyle::LK_TextProto) { 783 Tok->Type = TT_DictLiteral; 784 if (Style.Language == FormatStyle::LK_TextProto) { 785 if (FormatToken *Previous = Tok->getPreviousNonComment()) 786 Previous->Type = TT_SelectorName; 787 } 788 } else if (Contexts.back().ColonIsObjCMethodExpr || 789 Line.startsWith(TT_ObjCMethodSpecifier)) { 790 Tok->Type = TT_ObjCMethodExpr; 791 const FormatToken *BeforePrevious = Tok->Previous->Previous; 792 // Ensure we tag all identifiers in method declarations as 793 // TT_SelectorName. 794 bool UnknownIdentifierInMethodDeclaration = 795 Line.startsWith(TT_ObjCMethodSpecifier) && 796 Tok->Previous->is(tok::identifier) && Tok->Previous->is(TT_Unknown); 797 if (!BeforePrevious || 798 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen. 799 !(BeforePrevious->is(TT_CastRParen) || 800 (BeforePrevious->is(TT_ObjCMethodExpr) && 801 BeforePrevious->is(tok::colon))) || 802 BeforePrevious->is(tok::r_square) || 803 Contexts.back().LongestObjCSelectorName == 0 || 804 UnknownIdentifierInMethodDeclaration) { 805 Tok->Previous->Type = TT_SelectorName; 806 if (!Contexts.back().FirstObjCSelectorName) 807 Contexts.back().FirstObjCSelectorName = Tok->Previous; 808 else if (Tok->Previous->ColumnWidth > 809 Contexts.back().LongestObjCSelectorName) 810 Contexts.back().LongestObjCSelectorName = 811 Tok->Previous->ColumnWidth; 812 Tok->Previous->ParameterIndex = 813 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 814 ++Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts; 815 } 816 } else if (Contexts.back().ColonIsForRangeExpr) { 817 Tok->Type = TT_RangeBasedForLoopColon; 818 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 819 Tok->Type = TT_BitFieldColon; 820 } else if (Contexts.size() == 1 && 821 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) { 822 if (Tok->getPreviousNonComment()->isOneOf(tok::r_paren, 823 tok::kw_noexcept)) 824 Tok->Type = TT_CtorInitializerColon; 825 else 826 Tok->Type = TT_InheritanceColon; 827 } else if (canBeObjCSelectorComponent(*Tok->Previous) && Tok->Next && 828 (Tok->Next->isOneOf(tok::r_paren, tok::comma) || 829 (canBeObjCSelectorComponent(*Tok->Next) && Tok->Next->Next && 830 Tok->Next->Next->is(tok::colon)))) { 831 // This handles a special macro in ObjC code where selectors including 832 // the colon are passed as macro arguments. 833 Tok->Type = TT_ObjCMethodExpr; 834 } else if (Contexts.back().ContextKind == tok::l_paren) { 835 Tok->Type = TT_InlineASMColon; 836 } 837 break; 838 case tok::pipe: 839 case tok::amp: 840 // | and & in declarations/type expressions represent union and 841 // intersection types, respectively. 842 if (Style.Language == FormatStyle::LK_JavaScript && 843 !Contexts.back().IsExpression) 844 Tok->Type = TT_JsTypeOperator; 845 break; 846 case tok::kw_if: 847 case tok::kw_while: 848 if (Tok->is(tok::kw_if) && CurrentToken && 849 CurrentToken->isOneOf(tok::kw_constexpr, tok::identifier)) 850 next(); 851 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 852 next(); 853 if (!parseParens(/*LookForDecls=*/true)) 854 return false; 855 } 856 break; 857 case tok::kw_for: 858 if (Style.Language == FormatStyle::LK_JavaScript) { 859 // x.for and {for: ...} 860 if ((Tok->Previous && Tok->Previous->is(tok::period)) || 861 (Tok->Next && Tok->Next->is(tok::colon))) 862 break; 863 // JS' for await ( ... 864 if (CurrentToken && CurrentToken->is(Keywords.kw_await)) 865 next(); 866 } 867 Contexts.back().ColonIsForRangeExpr = true; 868 next(); 869 if (!parseParens()) 870 return false; 871 break; 872 case tok::l_paren: 873 // When faced with 'operator()()', the kw_operator handler incorrectly 874 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make 875 // the first two parens OverloadedOperators and the second l_paren an 876 // OverloadedOperatorLParen. 877 if (Tok->Previous && Tok->Previous->is(tok::r_paren) && 878 Tok->Previous->MatchingParen && 879 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) { 880 Tok->Previous->Type = TT_OverloadedOperator; 881 Tok->Previous->MatchingParen->Type = TT_OverloadedOperator; 882 Tok->Type = TT_OverloadedOperatorLParen; 883 } 884 885 if (!parseParens()) 886 return false; 887 if (Line.MustBeDeclaration && Contexts.size() == 1 && 888 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) && 889 (!Tok->Previous || 890 !Tok->Previous->isOneOf(tok::kw_decltype, tok::kw___attribute, 891 TT_LeadingJavaAnnotation))) 892 Line.MightBeFunctionDecl = true; 893 break; 894 case tok::l_square: 895 if (!parseSquare()) 896 return false; 897 break; 898 case tok::l_brace: 899 if (Style.Language == FormatStyle::LK_TextProto) { 900 FormatToken *Previous = Tok->getPreviousNonComment(); 901 if (Previous && Previous->Type != TT_DictLiteral) 902 Previous->Type = TT_SelectorName; 903 } 904 if (!parseBrace()) 905 return false; 906 break; 907 case tok::less: 908 if (parseAngle()) { 909 Tok->Type = TT_TemplateOpener; 910 // In TT_Proto, we must distignuish between: 911 // map<key, value> 912 // msg < item: data > 913 // msg: < item: data > 914 // In TT_TextProto, map<key, value> does not occur. 915 if (Style.Language == FormatStyle::LK_TextProto || 916 (Style.Language == FormatStyle::LK_Proto && Tok->Previous && 917 Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) { 918 Tok->Type = TT_DictLiteral; 919 FormatToken *Previous = Tok->getPreviousNonComment(); 920 if (Previous && Previous->Type != TT_DictLiteral) 921 Previous->Type = TT_SelectorName; 922 } 923 } else { 924 Tok->Type = TT_BinaryOperator; 925 NonTemplateLess.insert(Tok); 926 CurrentToken = Tok; 927 next(); 928 } 929 break; 930 case tok::r_paren: 931 case tok::r_square: 932 return false; 933 case tok::r_brace: 934 // Lines can start with '}'. 935 if (Tok->Previous) 936 return false; 937 break; 938 case tok::greater: 939 if (Style.Language != FormatStyle::LK_TextProto) 940 Tok->Type = TT_BinaryOperator; 941 if (Tok->Previous && Tok->Previous->is(TT_TemplateCloser)) 942 Tok->SpacesRequiredBefore = 1; 943 break; 944 case tok::kw_operator: 945 if (Style.Language == FormatStyle::LK_TextProto || 946 Style.Language == FormatStyle::LK_Proto) 947 break; 948 while (CurrentToken && 949 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 950 if (CurrentToken->isOneOf(tok::star, tok::amp)) 951 CurrentToken->Type = TT_PointerOrReference; 952 consumeToken(); 953 if (CurrentToken && 954 CurrentToken->Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator, 955 tok::comma)) 956 CurrentToken->Previous->Type = TT_OverloadedOperator; 957 } 958 if (CurrentToken) { 959 CurrentToken->Type = TT_OverloadedOperatorLParen; 960 if (CurrentToken->Previous->is(TT_BinaryOperator)) 961 CurrentToken->Previous->Type = TT_OverloadedOperator; 962 } 963 break; 964 case tok::question: 965 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next && 966 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren, 967 tok::r_brace)) { 968 // Question marks before semicolons, colons, etc. indicate optional 969 // types (fields, parameters), e.g. 970 // function(x?: string, y?) {...} 971 // class X { y?; } 972 Tok->Type = TT_JsTypeOptionalQuestion; 973 break; 974 } 975 // Declarations cannot be conditional expressions, this can only be part 976 // of a type declaration. 977 if (Line.MustBeDeclaration && !Contexts.back().IsExpression && 978 Style.Language == FormatStyle::LK_JavaScript) 979 break; 980 parseConditional(); 981 break; 982 case tok::kw_template: 983 parseTemplateDeclaration(); 984 break; 985 case tok::comma: 986 if (Contexts.back().InCtorInitializer) 987 Tok->Type = TT_CtorInitializerComma; 988 else if (Contexts.back().InInheritanceList) 989 Tok->Type = TT_InheritanceComma; 990 else if (Contexts.back().FirstStartOfName && 991 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) { 992 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 993 Line.IsMultiVariableDeclStmt = true; 994 } 995 if (Contexts.back().IsForEachMacro) 996 Contexts.back().IsExpression = true; 997 break; 998 case tok::identifier: 999 if (Tok->isOneOf(Keywords.kw___has_include, 1000 Keywords.kw___has_include_next)) { 1001 parseHasInclude(); 1002 } 1003 break; 1004 default: 1005 break; 1006 } 1007 return true; 1008 } 1009 1010 void parseIncludeDirective() { 1011 if (CurrentToken && CurrentToken->is(tok::less)) { 1012 next(); 1013 while (CurrentToken) { 1014 // Mark tokens up to the trailing line comments as implicit string 1015 // literals. 1016 if (CurrentToken->isNot(tok::comment) && 1017 !CurrentToken->TokenText.startswith("//")) 1018 CurrentToken->Type = TT_ImplicitStringLiteral; 1019 next(); 1020 } 1021 } 1022 } 1023 1024 void parseWarningOrError() { 1025 next(); 1026 // We still want to format the whitespace left of the first token of the 1027 // warning or error. 1028 next(); 1029 while (CurrentToken) { 1030 CurrentToken->Type = TT_ImplicitStringLiteral; 1031 next(); 1032 } 1033 } 1034 1035 void parsePragma() { 1036 next(); // Consume "pragma". 1037 if (CurrentToken && 1038 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 1039 bool IsMark = CurrentToken->is(Keywords.kw_mark); 1040 next(); // Consume "mark". 1041 next(); // Consume first token (so we fix leading whitespace). 1042 while (CurrentToken) { 1043 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 1044 CurrentToken->Type = TT_ImplicitStringLiteral; 1045 next(); 1046 } 1047 } 1048 } 1049 1050 void parseHasInclude() { 1051 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 1052 return; 1053 next(); // '(' 1054 parseIncludeDirective(); 1055 next(); // ')' 1056 } 1057 1058 LineType parsePreprocessorDirective() { 1059 bool IsFirstToken = CurrentToken->IsFirst; 1060 LineType Type = LT_PreprocessorDirective; 1061 next(); 1062 if (!CurrentToken) 1063 return Type; 1064 1065 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) { 1066 // JavaScript files can contain shebang lines of the form: 1067 // #!/usr/bin/env node 1068 // Treat these like C++ #include directives. 1069 while (CurrentToken) { 1070 // Tokens cannot be comments here. 1071 CurrentToken->Type = TT_ImplicitStringLiteral; 1072 next(); 1073 } 1074 return LT_ImportStatement; 1075 } 1076 1077 if (CurrentToken->Tok.is(tok::numeric_constant)) { 1078 CurrentToken->SpacesRequiredBefore = 1; 1079 return Type; 1080 } 1081 // Hashes in the middle of a line can lead to any strange token 1082 // sequence. 1083 if (!CurrentToken->Tok.getIdentifierInfo()) 1084 return Type; 1085 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 1086 case tok::pp_include: 1087 case tok::pp_include_next: 1088 case tok::pp_import: 1089 next(); 1090 parseIncludeDirective(); 1091 Type = LT_ImportStatement; 1092 break; 1093 case tok::pp_error: 1094 case tok::pp_warning: 1095 parseWarningOrError(); 1096 break; 1097 case tok::pp_pragma: 1098 parsePragma(); 1099 break; 1100 case tok::pp_if: 1101 case tok::pp_elif: 1102 Contexts.back().IsExpression = true; 1103 next(); 1104 parseLine(); 1105 break; 1106 default: 1107 break; 1108 } 1109 while (CurrentToken) { 1110 FormatToken *Tok = CurrentToken; 1111 next(); 1112 if (Tok->is(tok::l_paren)) 1113 parseParens(); 1114 else if (Tok->isOneOf(Keywords.kw___has_include, 1115 Keywords.kw___has_include_next)) 1116 parseHasInclude(); 1117 } 1118 return Type; 1119 } 1120 1121 public: 1122 LineType parseLine() { 1123 if (!CurrentToken) 1124 return LT_Invalid; 1125 NonTemplateLess.clear(); 1126 if (CurrentToken->is(tok::hash)) 1127 return parsePreprocessorDirective(); 1128 1129 // Directly allow to 'import <string-literal>' to support protocol buffer 1130 // definitions (github.com/google/protobuf) or missing "#" (either way we 1131 // should not break the line). 1132 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 1133 if ((Style.Language == FormatStyle::LK_Java && 1134 CurrentToken->is(Keywords.kw_package)) || 1135 (Info && Info->getPPKeywordID() == tok::pp_import && 1136 CurrentToken->Next && 1137 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 1138 tok::kw_static))) { 1139 next(); 1140 parseIncludeDirective(); 1141 return LT_ImportStatement; 1142 } 1143 1144 // If this line starts and ends in '<' and '>', respectively, it is likely 1145 // part of "#define <a/b.h>". 1146 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 1147 parseIncludeDirective(); 1148 return LT_ImportStatement; 1149 } 1150 1151 // In .proto files, top-level options and package statements are very 1152 // similar to import statements and should not be line-wrapped. 1153 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 1154 CurrentToken->isOneOf(Keywords.kw_option, Keywords.kw_package)) { 1155 next(); 1156 if (CurrentToken && CurrentToken->is(tok::identifier)) { 1157 while (CurrentToken) 1158 next(); 1159 return LT_ImportStatement; 1160 } 1161 } 1162 1163 bool KeywordVirtualFound = false; 1164 bool ImportStatement = false; 1165 1166 // import {...} from '...'; 1167 if (Style.Language == FormatStyle::LK_JavaScript && 1168 CurrentToken->is(Keywords.kw_import)) 1169 ImportStatement = true; 1170 1171 while (CurrentToken) { 1172 if (CurrentToken->is(tok::kw_virtual)) 1173 KeywordVirtualFound = true; 1174 if (Style.Language == FormatStyle::LK_JavaScript) { 1175 // export {...} from '...'; 1176 // An export followed by "from 'some string';" is a re-export from 1177 // another module identified by a URI and is treated as a 1178 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 1179 // Just "export {...};" or "export class ..." should not be treated as 1180 // an import in this sense. 1181 if (Line.First->is(tok::kw_export) && 1182 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 1183 CurrentToken->Next->isStringLiteral()) 1184 ImportStatement = true; 1185 if (isClosureImportStatement(*CurrentToken)) 1186 ImportStatement = true; 1187 } 1188 if (!consumeToken()) 1189 return LT_Invalid; 1190 } 1191 if (KeywordVirtualFound) 1192 return LT_VirtualFunctionDecl; 1193 if (ImportStatement) 1194 return LT_ImportStatement; 1195 1196 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 1197 if (Contexts.back().FirstObjCSelectorName) 1198 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 1199 Contexts.back().LongestObjCSelectorName; 1200 return LT_ObjCMethodDecl; 1201 } 1202 1203 return LT_Other; 1204 } 1205 1206 private: 1207 bool isClosureImportStatement(const FormatToken &Tok) { 1208 // FIXME: Closure-library specific stuff should not be hard-coded but be 1209 // configurable. 1210 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 1211 Tok.Next->Next && 1212 (Tok.Next->Next->TokenText == "module" || 1213 Tok.Next->Next->TokenText == "provide" || 1214 Tok.Next->Next->TokenText == "require" || 1215 Tok.Next->Next->TokenText == "requireType" || 1216 Tok.Next->Next->TokenText == "forwardDeclare") && 1217 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 1218 } 1219 1220 void resetTokenMetadata(FormatToken *Token) { 1221 if (!Token) 1222 return; 1223 1224 // Reset token type in case we have already looked at it and then 1225 // recovered from an error (e.g. failure to find the matching >). 1226 if (!CurrentToken->isOneOf( 1227 TT_LambdaLSquare, TT_LambdaLBrace, TT_ForEachMacro, 1228 TT_TypenameMacro, TT_FunctionLBrace, TT_ImplicitStringLiteral, 1229 TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow, TT_NamespaceMacro, 1230 TT_OverloadedOperator, TT_RegexLiteral, TT_TemplateString, 1231 TT_ObjCStringLiteral)) 1232 CurrentToken->Type = TT_Unknown; 1233 CurrentToken->Role.reset(); 1234 CurrentToken->MatchingParen = nullptr; 1235 CurrentToken->FakeLParens.clear(); 1236 CurrentToken->FakeRParens = 0; 1237 } 1238 1239 void next() { 1240 if (CurrentToken) { 1241 CurrentToken->NestingLevel = Contexts.size() - 1; 1242 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1243 modifyContext(*CurrentToken); 1244 determineTokenType(*CurrentToken); 1245 CurrentToken = CurrentToken->Next; 1246 } 1247 1248 resetTokenMetadata(CurrentToken); 1249 } 1250 1251 /// A struct to hold information valid in a specific context, e.g. 1252 /// a pair of parenthesis. 1253 struct Context { 1254 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1255 bool IsExpression) 1256 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1257 IsExpression(IsExpression) {} 1258 1259 tok::TokenKind ContextKind; 1260 unsigned BindingStrength; 1261 bool IsExpression; 1262 unsigned LongestObjCSelectorName = 0; 1263 bool ColonIsForRangeExpr = false; 1264 bool ColonIsDictLiteral = false; 1265 bool ColonIsObjCMethodExpr = false; 1266 FormatToken *FirstObjCSelectorName = nullptr; 1267 FormatToken *FirstStartOfName = nullptr; 1268 bool CanBeExpression = true; 1269 bool InTemplateArgument = false; 1270 bool InCtorInitializer = false; 1271 bool InInheritanceList = false; 1272 bool CaretFound = false; 1273 bool IsForEachMacro = false; 1274 bool InCpp11AttributeSpecifier = false; 1275 bool InCSharpAttributeSpecifier = false; 1276 }; 1277 1278 /// Puts a new \c Context onto the stack \c Contexts for the lifetime 1279 /// of each instance. 1280 struct ScopedContextCreator { 1281 AnnotatingParser &P; 1282 1283 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1284 unsigned Increase) 1285 : P(P) { 1286 P.Contexts.push_back(Context(ContextKind, 1287 P.Contexts.back().BindingStrength + Increase, 1288 P.Contexts.back().IsExpression)); 1289 } 1290 1291 ~ScopedContextCreator() { P.Contexts.pop_back(); } 1292 }; 1293 1294 void modifyContext(const FormatToken &Current) { 1295 if (Current.getPrecedence() == prec::Assignment && 1296 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 1297 // Type aliases use `type X = ...;` in TypeScript and can be exported 1298 // using `export type ...`. 1299 !(Style.Language == FormatStyle::LK_JavaScript && 1300 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1301 Line.startsWith(tok::kw_export, Keywords.kw_type, 1302 tok::identifier))) && 1303 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 1304 Contexts.back().IsExpression = true; 1305 if (!Line.startsWith(TT_UnaryOperator)) { 1306 for (FormatToken *Previous = Current.Previous; 1307 Previous && Previous->Previous && 1308 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1309 Previous = Previous->Previous) { 1310 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1311 Previous = Previous->MatchingParen; 1312 if (!Previous) 1313 break; 1314 } 1315 if (Previous->opensScope()) 1316 break; 1317 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1318 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1319 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1320 Previous->Type = TT_PointerOrReference; 1321 } 1322 } 1323 } else if (Current.is(tok::lessless) && 1324 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1325 Contexts.back().IsExpression = true; 1326 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1327 Contexts.back().IsExpression = true; 1328 } else if (Current.is(TT_TrailingReturnArrow)) { 1329 Contexts.back().IsExpression = false; 1330 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1331 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1332 } else if (Current.Previous && 1333 Current.Previous->is(TT_CtorInitializerColon)) { 1334 Contexts.back().IsExpression = true; 1335 Contexts.back().InCtorInitializer = true; 1336 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1337 Contexts.back().InInheritanceList = true; 1338 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1339 for (FormatToken *Previous = Current.Previous; 1340 Previous && Previous->isOneOf(tok::star, tok::amp); 1341 Previous = Previous->Previous) 1342 Previous->Type = TT_PointerOrReference; 1343 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 1344 Contexts.back().IsExpression = false; 1345 } else if (Current.is(tok::kw_new)) { 1346 Contexts.back().CanBeExpression = false; 1347 } else if (Current.isOneOf(tok::semi, tok::exclaim)) { 1348 // This should be the condition or increment in a for-loop. 1349 Contexts.back().IsExpression = true; 1350 } 1351 } 1352 1353 void determineTokenType(FormatToken &Current) { 1354 if (!Current.is(TT_Unknown)) 1355 // The token type is already known. 1356 return; 1357 1358 if (Style.Language == FormatStyle::LK_JavaScript) { 1359 if (Current.is(tok::exclaim)) { 1360 if (Current.Previous && 1361 (Current.Previous->isOneOf(tok::identifier, tok::kw_namespace, 1362 tok::r_paren, tok::r_square, 1363 tok::r_brace) || 1364 Current.Previous->Tok.isLiteral())) { 1365 Current.Type = TT_JsNonNullAssertion; 1366 return; 1367 } 1368 if (Current.Next && 1369 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1370 Current.Type = TT_JsNonNullAssertion; 1371 return; 1372 } 1373 } 1374 } 1375 1376 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1377 // function declaration have been found. In this case, 'Current' is a 1378 // trailing token of this declaration and thus cannot be a name. 1379 if (Current.is(Keywords.kw_instanceof)) { 1380 Current.Type = TT_BinaryOperator; 1381 } else if (isStartOfName(Current) && 1382 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1383 Contexts.back().FirstStartOfName = &Current; 1384 Current.Type = TT_StartOfName; 1385 } else if (Current.is(tok::semi)) { 1386 // Reset FirstStartOfName after finding a semicolon so that a for loop 1387 // with multiple increment statements is not confused with a for loop 1388 // having multiple variable declarations. 1389 Contexts.back().FirstStartOfName = nullptr; 1390 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1391 AutoFound = true; 1392 } else if (Current.is(tok::arrow) && 1393 Style.Language == FormatStyle::LK_Java) { 1394 Current.Type = TT_LambdaArrow; 1395 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1396 Current.NestingLevel == 0 && 1397 !Current.Previous->is(tok::kw_operator)) { 1398 // not auto operator->() -> xxx; 1399 Current.Type = TT_TrailingReturnArrow; 1400 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1401 Current.Type = determineStarAmpUsage(Current, 1402 Contexts.back().CanBeExpression && 1403 Contexts.back().IsExpression, 1404 Contexts.back().InTemplateArgument); 1405 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1406 Current.Type = determinePlusMinusCaretUsage(Current); 1407 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1408 Contexts.back().CaretFound = true; 1409 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1410 Current.Type = determineIncrementUsage(Current); 1411 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1412 Current.Type = TT_UnaryOperator; 1413 } else if (Current.is(tok::question)) { 1414 if (Style.Language == FormatStyle::LK_JavaScript && 1415 Line.MustBeDeclaration && !Contexts.back().IsExpression) { 1416 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1417 // on the interface, not a ternary expression. 1418 Current.Type = TT_JsTypeOptionalQuestion; 1419 } else { 1420 Current.Type = TT_ConditionalExpr; 1421 } 1422 } else if (Current.isBinaryOperator() && 1423 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1424 (!Current.is(tok::greater) && 1425 Style.Language != FormatStyle::LK_TextProto)) { 1426 Current.Type = TT_BinaryOperator; 1427 } else if (Current.is(tok::comment)) { 1428 if (Current.TokenText.startswith("/*")) { 1429 if (Current.TokenText.endswith("*/")) 1430 Current.Type = TT_BlockComment; 1431 else 1432 // The lexer has for some reason determined a comment here. But we 1433 // cannot really handle it, if it isn't properly terminated. 1434 Current.Tok.setKind(tok::unknown); 1435 } else { 1436 Current.Type = TT_LineComment; 1437 } 1438 } else if (Current.is(tok::r_paren)) { 1439 if (rParenEndsCast(Current)) 1440 Current.Type = TT_CastRParen; 1441 if (Current.MatchingParen && Current.Next && 1442 !Current.Next->isBinaryOperator() && 1443 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1444 tok::comma, tok::period, tok::arrow, 1445 tok::coloncolon)) 1446 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1447 // Make sure this isn't the return type of an Obj-C block declaration 1448 if (AfterParen->Tok.isNot(tok::caret)) { 1449 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1450 if (BeforeParen->is(tok::identifier) && 1451 !BeforeParen->is(TT_TypenameMacro) && 1452 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1453 (!BeforeParen->Previous || 1454 BeforeParen->Previous->ClosesTemplateDeclaration)) 1455 Current.Type = TT_FunctionAnnotationRParen; 1456 } 1457 } 1458 } else if (Current.is(tok::at) && Current.Next && 1459 Style.Language != FormatStyle::LK_JavaScript && 1460 Style.Language != FormatStyle::LK_Java) { 1461 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1462 // marks declarations and properties that need special formatting. 1463 switch (Current.Next->Tok.getObjCKeywordID()) { 1464 case tok::objc_interface: 1465 case tok::objc_implementation: 1466 case tok::objc_protocol: 1467 Current.Type = TT_ObjCDecl; 1468 break; 1469 case tok::objc_property: 1470 Current.Type = TT_ObjCProperty; 1471 break; 1472 default: 1473 break; 1474 } 1475 } else if (Current.is(tok::period)) { 1476 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1477 if (PreviousNoComment && 1478 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1479 Current.Type = TT_DesignatedInitializerPeriod; 1480 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1481 Current.Previous->isOneOf(TT_JavaAnnotation, 1482 TT_LeadingJavaAnnotation)) { 1483 Current.Type = Current.Previous->Type; 1484 } 1485 } else if (canBeObjCSelectorComponent(Current) && 1486 // FIXME(bug 36976): ObjC return types shouldn't use 1487 // TT_CastRParen. 1488 Current.Previous && Current.Previous->is(TT_CastRParen) && 1489 Current.Previous->MatchingParen && 1490 Current.Previous->MatchingParen->Previous && 1491 Current.Previous->MatchingParen->Previous->is( 1492 TT_ObjCMethodSpecifier)) { 1493 // This is the first part of an Objective-C selector name. (If there's no 1494 // colon after this, this is the only place which annotates the identifier 1495 // as a selector.) 1496 Current.Type = TT_SelectorName; 1497 } else if (Current.isOneOf(tok::identifier, tok::kw_const, 1498 tok::kw_noexcept) && 1499 Current.Previous && 1500 !Current.Previous->isOneOf(tok::equal, tok::at) && 1501 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1502 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1503 // function declaration have been found. 1504 Current.Type = TT_TrailingAnnotation; 1505 } else if ((Style.Language == FormatStyle::LK_Java || 1506 Style.Language == FormatStyle::LK_JavaScript) && 1507 Current.Previous) { 1508 if (Current.Previous->is(tok::at) && 1509 Current.isNot(Keywords.kw_interface)) { 1510 const FormatToken &AtToken = *Current.Previous; 1511 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1512 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1513 Current.Type = TT_LeadingJavaAnnotation; 1514 else 1515 Current.Type = TT_JavaAnnotation; 1516 } else if (Current.Previous->is(tok::period) && 1517 Current.Previous->isOneOf(TT_JavaAnnotation, 1518 TT_LeadingJavaAnnotation)) { 1519 Current.Type = Current.Previous->Type; 1520 } 1521 } 1522 } 1523 1524 /// Take a guess at whether \p Tok starts a name of a function or 1525 /// variable declaration. 1526 /// 1527 /// This is a heuristic based on whether \p Tok is an identifier following 1528 /// something that is likely a type. 1529 bool isStartOfName(const FormatToken &Tok) { 1530 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1531 return false; 1532 1533 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1534 Keywords.kw_as)) 1535 return false; 1536 if (Style.Language == FormatStyle::LK_JavaScript && 1537 Tok.Previous->is(Keywords.kw_in)) 1538 return false; 1539 1540 // Skip "const" as it does not have an influence on whether this is a name. 1541 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1542 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1543 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1544 1545 if (!PreviousNotConst) 1546 return false; 1547 1548 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1549 PreviousNotConst->Previous && 1550 PreviousNotConst->Previous->is(tok::hash); 1551 1552 if (PreviousNotConst->is(TT_TemplateCloser)) 1553 return PreviousNotConst && PreviousNotConst->MatchingParen && 1554 PreviousNotConst->MatchingParen->Previous && 1555 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1556 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1557 1558 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen && 1559 PreviousNotConst->MatchingParen->Previous && 1560 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype)) 1561 return true; 1562 1563 return (!IsPPKeyword && 1564 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) || 1565 PreviousNotConst->is(TT_PointerOrReference) || 1566 PreviousNotConst->isSimpleTypeSpecifier(); 1567 } 1568 1569 /// Determine whether ')' is ending a cast. 1570 bool rParenEndsCast(const FormatToken &Tok) { 1571 // C-style casts are only used in C++ and Java. 1572 if (!Style.isCpp() && Style.Language != FormatStyle::LK_Java) 1573 return false; 1574 1575 // Empty parens aren't casts and there are no casts at the end of the line. 1576 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1577 return false; 1578 1579 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1580 if (LeftOfParens) { 1581 // If there is a closing parenthesis left of the current parentheses, 1582 // look past it as these might be chained casts. 1583 if (LeftOfParens->is(tok::r_paren)) { 1584 if (!LeftOfParens->MatchingParen || 1585 !LeftOfParens->MatchingParen->Previous) 1586 return false; 1587 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1588 } 1589 1590 // If there is an identifier (or with a few exceptions a keyword) right 1591 // before the parentheses, this is unlikely to be a cast. 1592 if (LeftOfParens->Tok.getIdentifierInfo() && 1593 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1594 tok::kw_delete)) 1595 return false; 1596 1597 // Certain other tokens right before the parentheses are also signals that 1598 // this cannot be a cast. 1599 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1600 TT_TemplateCloser, tok::ellipsis)) 1601 return false; 1602 } 1603 1604 if (Tok.Next->is(tok::question)) 1605 return false; 1606 1607 // Functions which end with decorations like volatile, noexcept are unlikely 1608 // to be casts. 1609 if (Tok.Next->isOneOf(tok::kw_noexcept, tok::kw_volatile, tok::kw_const, 1610 tok::kw_throw, tok::arrow, Keywords.kw_override, 1611 Keywords.kw_final) || 1612 isCpp11AttributeSpecifier(*Tok.Next)) 1613 return false; 1614 1615 // As Java has no function types, a "(" after the ")" likely means that this 1616 // is a cast. 1617 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1618 return true; 1619 1620 // If a (non-string) literal follows, this is likely a cast. 1621 if (Tok.Next->isNot(tok::string_literal) && 1622 (Tok.Next->Tok.isLiteral() || 1623 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1624 return true; 1625 1626 // Heuristically try to determine whether the parentheses contain a type. 1627 bool ParensAreType = 1628 !Tok.Previous || 1629 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) || 1630 Tok.Previous->isSimpleTypeSpecifier(); 1631 bool ParensCouldEndDecl = 1632 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1633 if (ParensAreType && !ParensCouldEndDecl) 1634 return true; 1635 1636 // At this point, we heuristically assume that there are no casts at the 1637 // start of the line. We assume that we have found most cases where there 1638 // are by the logic above, e.g. "(void)x;". 1639 if (!LeftOfParens) 1640 return false; 1641 1642 // Certain token types inside the parentheses mean that this can't be a 1643 // cast. 1644 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 1645 Token = Token->Next) 1646 if (Token->is(TT_BinaryOperator)) 1647 return false; 1648 1649 // If the following token is an identifier or 'this', this is a cast. All 1650 // cases where this can be something else are handled above. 1651 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1652 return true; 1653 1654 if (!Tok.Next->Next) 1655 return false; 1656 1657 // If the next token after the parenthesis is a unary operator, assume 1658 // that this is cast, unless there are unexpected tokens inside the 1659 // parenthesis. 1660 bool NextIsUnary = 1661 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 1662 if (!NextIsUnary || Tok.Next->is(tok::plus) || 1663 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 1664 return false; 1665 // Search for unexpected tokens. 1666 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 1667 Prev = Prev->Previous) { 1668 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 1669 return false; 1670 } 1671 return true; 1672 } 1673 1674 /// Return the type of the given token assuming it is * or &. 1675 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 1676 bool InTemplateArgument) { 1677 if (Style.Language == FormatStyle::LK_JavaScript) 1678 return TT_BinaryOperator; 1679 1680 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1681 if (!PrevToken) 1682 return TT_UnaryOperator; 1683 1684 const FormatToken *NextToken = Tok.getNextNonComment(); 1685 if (!NextToken || 1686 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_const, 1687 tok::kw_noexcept) || 1688 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 1689 return TT_PointerOrReference; 1690 1691 if (PrevToken->is(tok::coloncolon)) 1692 return TT_PointerOrReference; 1693 1694 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 1695 tok::comma, tok::semi, tok::kw_return, tok::colon, 1696 tok::equal, tok::kw_delete, tok::kw_sizeof, 1697 tok::kw_throw) || 1698 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 1699 TT_UnaryOperator, TT_CastRParen)) 1700 return TT_UnaryOperator; 1701 1702 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 1703 return TT_PointerOrReference; 1704 if (NextToken->is(tok::kw_operator) && !IsExpression) 1705 return TT_PointerOrReference; 1706 if (NextToken->isOneOf(tok::comma, tok::semi)) 1707 return TT_PointerOrReference; 1708 1709 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen) { 1710 FormatToken *TokenBeforeMatchingParen = 1711 PrevToken->MatchingParen->getPreviousNonComment(); 1712 if (TokenBeforeMatchingParen && 1713 TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype, 1714 TT_TypenameMacro)) 1715 return TT_PointerOrReference; 1716 } 1717 1718 if (PrevToken->Tok.isLiteral() || 1719 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 1720 tok::kw_false, tok::r_brace) || 1721 NextToken->Tok.isLiteral() || 1722 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 1723 NextToken->isUnaryOperator() || 1724 // If we know we're in a template argument, there are no named 1725 // declarations. Thus, having an identifier on the right-hand side 1726 // indicates a binary operator. 1727 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 1728 return TT_BinaryOperator; 1729 1730 // "&&(" is quite unlikely to be two successive unary "&". 1731 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren)) 1732 return TT_BinaryOperator; 1733 1734 // This catches some cases where evaluation order is used as control flow: 1735 // aaa && aaa->f(); 1736 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 1737 if (NextNextToken && NextNextToken->is(tok::arrow)) 1738 return TT_BinaryOperator; 1739 1740 // It is very unlikely that we are going to find a pointer or reference type 1741 // definition on the RHS of an assignment. 1742 if (IsExpression && !Contexts.back().CaretFound) 1743 return TT_BinaryOperator; 1744 1745 return TT_PointerOrReference; 1746 } 1747 1748 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 1749 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1750 if (!PrevToken) 1751 return TT_UnaryOperator; 1752 1753 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 1754 // This must be a sequence of leading unary operators. 1755 return TT_UnaryOperator; 1756 1757 // Use heuristics to recognize unary operators. 1758 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 1759 tok::question, tok::colon, tok::kw_return, 1760 tok::kw_case, tok::at, tok::l_brace, tok::kw_throw)) 1761 return TT_UnaryOperator; 1762 1763 // There can't be two consecutive binary operators. 1764 if (PrevToken->is(TT_BinaryOperator)) 1765 return TT_UnaryOperator; 1766 1767 // Fall back to marking the token as binary operator. 1768 return TT_BinaryOperator; 1769 } 1770 1771 /// Determine whether ++/-- are pre- or post-increments/-decrements. 1772 TokenType determineIncrementUsage(const FormatToken &Tok) { 1773 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1774 if (!PrevToken || PrevToken->is(TT_CastRParen)) 1775 return TT_UnaryOperator; 1776 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 1777 return TT_TrailingUnaryOperator; 1778 1779 return TT_UnaryOperator; 1780 } 1781 1782 SmallVector<Context, 8> Contexts; 1783 1784 const FormatStyle &Style; 1785 AnnotatedLine &Line; 1786 FormatToken *CurrentToken; 1787 bool AutoFound; 1788 const AdditionalKeywords &Keywords; 1789 1790 // Set of "<" tokens that do not open a template parameter list. If parseAngle 1791 // determines that a specific token can't be a template opener, it will make 1792 // same decision irrespective of the decisions for tokens leading up to it. 1793 // Store this information to prevent this from causing exponential runtime. 1794 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 1795 }; 1796 1797 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 1798 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 1799 1800 /// Parses binary expressions by inserting fake parenthesis based on 1801 /// operator precedence. 1802 class ExpressionParser { 1803 public: 1804 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 1805 AnnotatedLine &Line) 1806 : Style(Style), Keywords(Keywords), Current(Line.First) {} 1807 1808 /// Parse expressions with the given operator precedence. 1809 void parse(int Precedence = 0) { 1810 // Skip 'return' and ObjC selector colons as they are not part of a binary 1811 // expression. 1812 while (Current && (Current->is(tok::kw_return) || 1813 (Current->is(tok::colon) && 1814 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 1815 next(); 1816 1817 if (!Current || Precedence > PrecedenceArrowAndPeriod) 1818 return; 1819 1820 // Conditional expressions need to be parsed separately for proper nesting. 1821 if (Precedence == prec::Conditional) { 1822 parseConditionalExpr(); 1823 return; 1824 } 1825 1826 // Parse unary operators, which all have a higher precedence than binary 1827 // operators. 1828 if (Precedence == PrecedenceUnaryOperator) { 1829 parseUnaryOperator(); 1830 return; 1831 } 1832 1833 FormatToken *Start = Current; 1834 FormatToken *LatestOperator = nullptr; 1835 unsigned OperatorIndex = 0; 1836 1837 while (Current) { 1838 // Consume operators with higher precedence. 1839 parse(Precedence + 1); 1840 1841 int CurrentPrecedence = getCurrentPrecedence(); 1842 1843 if (Current && Current->is(TT_SelectorName) && 1844 Precedence == CurrentPrecedence) { 1845 if (LatestOperator) 1846 addFakeParenthesis(Start, prec::Level(Precedence)); 1847 Start = Current; 1848 } 1849 1850 // At the end of the line or when an operator with higher precedence is 1851 // found, insert fake parenthesis and return. 1852 if (!Current || 1853 (Current->closesScope() && 1854 (Current->MatchingParen || Current->is(TT_TemplateString))) || 1855 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 1856 (CurrentPrecedence == prec::Conditional && 1857 Precedence == prec::Assignment && Current->is(tok::colon))) { 1858 break; 1859 } 1860 1861 // Consume scopes: (), [], <> and {} 1862 if (Current->opensScope()) { 1863 // In fragment of a JavaScript template string can look like '}..${' and 1864 // thus close a scope and open a new one at the same time. 1865 while (Current && (!Current->closesScope() || Current->opensScope())) { 1866 next(); 1867 parse(); 1868 } 1869 next(); 1870 } else { 1871 // Operator found. 1872 if (CurrentPrecedence == Precedence) { 1873 if (LatestOperator) 1874 LatestOperator->NextOperator = Current; 1875 LatestOperator = Current; 1876 Current->OperatorIndex = OperatorIndex; 1877 ++OperatorIndex; 1878 } 1879 next(/*SkipPastLeadingComments=*/Precedence > 0); 1880 } 1881 } 1882 1883 if (LatestOperator && (Current || Precedence > 0)) { 1884 // LatestOperator->LastOperator = true; 1885 if (Precedence == PrecedenceArrowAndPeriod) { 1886 // Call expressions don't have a binary operator precedence. 1887 addFakeParenthesis(Start, prec::Unknown); 1888 } else { 1889 addFakeParenthesis(Start, prec::Level(Precedence)); 1890 } 1891 } 1892 } 1893 1894 private: 1895 /// Gets the precedence (+1) of the given token for binary operators 1896 /// and other tokens that we treat like binary operators. 1897 int getCurrentPrecedence() { 1898 if (Current) { 1899 const FormatToken *NextNonComment = Current->getNextNonComment(); 1900 if (Current->is(TT_ConditionalExpr)) 1901 return prec::Conditional; 1902 if (NextNonComment && Current->is(TT_SelectorName) && 1903 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 1904 ((Style.Language == FormatStyle::LK_Proto || 1905 Style.Language == FormatStyle::LK_TextProto) && 1906 NextNonComment->is(tok::less)))) 1907 return prec::Assignment; 1908 if (Current->is(TT_JsComputedPropertyName)) 1909 return prec::Assignment; 1910 if (Current->is(TT_LambdaArrow)) 1911 return prec::Comma; 1912 if (Current->is(TT_JsFatArrow)) 1913 return prec::Assignment; 1914 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 1915 (Current->is(tok::comment) && NextNonComment && 1916 NextNonComment->is(TT_SelectorName))) 1917 return 0; 1918 if (Current->is(TT_RangeBasedForLoopColon)) 1919 return prec::Comma; 1920 if ((Style.Language == FormatStyle::LK_Java || 1921 Style.Language == FormatStyle::LK_JavaScript) && 1922 Current->is(Keywords.kw_instanceof)) 1923 return prec::Relational; 1924 if (Style.Language == FormatStyle::LK_JavaScript && 1925 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 1926 return prec::Relational; 1927 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 1928 return Current->getPrecedence(); 1929 if (Current->isOneOf(tok::period, tok::arrow)) 1930 return PrecedenceArrowAndPeriod; 1931 if ((Style.Language == FormatStyle::LK_Java || 1932 Style.Language == FormatStyle::LK_JavaScript) && 1933 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 1934 Keywords.kw_throws)) 1935 return 0; 1936 } 1937 return -1; 1938 } 1939 1940 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 1941 Start->FakeLParens.push_back(Precedence); 1942 if (Precedence > prec::Unknown) 1943 Start->StartsBinaryExpression = true; 1944 if (Current) { 1945 FormatToken *Previous = Current->Previous; 1946 while (Previous->is(tok::comment) && Previous->Previous) 1947 Previous = Previous->Previous; 1948 ++Previous->FakeRParens; 1949 if (Precedence > prec::Unknown) 1950 Previous->EndsBinaryExpression = true; 1951 } 1952 } 1953 1954 /// Parse unary operator expressions and surround them with fake 1955 /// parentheses if appropriate. 1956 void parseUnaryOperator() { 1957 llvm::SmallVector<FormatToken *, 2> Tokens; 1958 while (Current && Current->is(TT_UnaryOperator)) { 1959 Tokens.push_back(Current); 1960 next(); 1961 } 1962 parse(PrecedenceArrowAndPeriod); 1963 for (FormatToken *Token : llvm::reverse(Tokens)) 1964 // The actual precedence doesn't matter. 1965 addFakeParenthesis(Token, prec::Unknown); 1966 } 1967 1968 void parseConditionalExpr() { 1969 while (Current && Current->isTrailingComment()) { 1970 next(); 1971 } 1972 FormatToken *Start = Current; 1973 parse(prec::LogicalOr); 1974 if (!Current || !Current->is(tok::question)) 1975 return; 1976 next(); 1977 parse(prec::Assignment); 1978 if (!Current || Current->isNot(TT_ConditionalExpr)) 1979 return; 1980 next(); 1981 parse(prec::Assignment); 1982 addFakeParenthesis(Start, prec::Conditional); 1983 } 1984 1985 void next(bool SkipPastLeadingComments = true) { 1986 if (Current) 1987 Current = Current->Next; 1988 while (Current && 1989 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 1990 Current->isTrailingComment()) 1991 Current = Current->Next; 1992 } 1993 1994 const FormatStyle &Style; 1995 const AdditionalKeywords &Keywords; 1996 FormatToken *Current; 1997 }; 1998 1999 } // end anonymous namespace 2000 2001 void TokenAnnotator::setCommentLineLevels( 2002 SmallVectorImpl<AnnotatedLine *> &Lines) { 2003 const AnnotatedLine *NextNonCommentLine = nullptr; 2004 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(), 2005 E = Lines.rend(); 2006 I != E; ++I) { 2007 bool CommentLine = true; 2008 for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) { 2009 if (!Tok->is(tok::comment)) { 2010 CommentLine = false; 2011 break; 2012 } 2013 } 2014 2015 // If the comment is currently aligned with the line immediately following 2016 // it, that's probably intentional and we should keep it. 2017 if (NextNonCommentLine && CommentLine && 2018 NextNonCommentLine->First->NewlinesBefore <= 1 && 2019 NextNonCommentLine->First->OriginalColumn == 2020 (*I)->First->OriginalColumn) { 2021 // Align comments for preprocessor lines with the # in column 0 if 2022 // preprocessor lines are not indented. Otherwise, align with the next 2023 // line. 2024 (*I)->Level = 2025 (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 2026 (NextNonCommentLine->Type == LT_PreprocessorDirective || 2027 NextNonCommentLine->Type == LT_ImportStatement)) 2028 ? 0 2029 : NextNonCommentLine->Level; 2030 } else { 2031 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr; 2032 } 2033 2034 setCommentLineLevels((*I)->Children); 2035 } 2036 } 2037 2038 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 2039 unsigned Result = 0; 2040 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 2041 Result = std::max(Result, Tok->NestingLevel); 2042 return Result; 2043 } 2044 2045 void TokenAnnotator::annotate(AnnotatedLine &Line) { 2046 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2047 E = Line.Children.end(); 2048 I != E; ++I) { 2049 annotate(**I); 2050 } 2051 AnnotatingParser Parser(Style, Line, Keywords); 2052 Line.Type = Parser.parseLine(); 2053 2054 // With very deep nesting, ExpressionParser uses lots of stack and the 2055 // formatting algorithm is very slow. We're not going to do a good job here 2056 // anyway - it's probably generated code being formatted by mistake. 2057 // Just skip the whole line. 2058 if (maxNestingDepth(Line) > 50) 2059 Line.Type = LT_Invalid; 2060 2061 if (Line.Type == LT_Invalid) 2062 return; 2063 2064 ExpressionParser ExprParser(Style, Keywords, Line); 2065 ExprParser.parse(); 2066 2067 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2068 Line.Type = LT_ObjCMethodDecl; 2069 else if (Line.startsWith(TT_ObjCDecl)) 2070 Line.Type = LT_ObjCDecl; 2071 else if (Line.startsWith(TT_ObjCProperty)) 2072 Line.Type = LT_ObjCProperty; 2073 2074 Line.First->SpacesRequiredBefore = 1; 2075 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 2076 } 2077 2078 // This function heuristically determines whether 'Current' starts the name of a 2079 // function declaration. 2080 static bool isFunctionDeclarationName(const FormatToken &Current, 2081 const AnnotatedLine &Line) { 2082 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 2083 for (; Next; Next = Next->Next) { 2084 if (Next->is(TT_OverloadedOperatorLParen)) 2085 return Next; 2086 if (Next->is(TT_OverloadedOperator)) 2087 continue; 2088 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 2089 // For 'new[]' and 'delete[]'. 2090 if (Next->Next && Next->Next->is(tok::l_square) && Next->Next->Next && 2091 Next->Next->Next->is(tok::r_square)) 2092 Next = Next->Next->Next; 2093 continue; 2094 } 2095 2096 break; 2097 } 2098 return nullptr; 2099 }; 2100 2101 // Find parentheses of parameter list. 2102 const FormatToken *Next = Current.Next; 2103 if (Current.is(tok::kw_operator)) { 2104 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 2105 return false; 2106 Next = skipOperatorName(Next); 2107 } else { 2108 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 2109 return false; 2110 for (; Next; Next = Next->Next) { 2111 if (Next->is(TT_TemplateOpener)) { 2112 Next = Next->MatchingParen; 2113 } else if (Next->is(tok::coloncolon)) { 2114 Next = Next->Next; 2115 if (!Next) 2116 return false; 2117 if (Next->is(tok::kw_operator)) { 2118 Next = skipOperatorName(Next->Next); 2119 break; 2120 } 2121 if (!Next->is(tok::identifier)) 2122 return false; 2123 } else if (Next->is(tok::l_paren)) { 2124 break; 2125 } else { 2126 return false; 2127 } 2128 } 2129 } 2130 2131 // Check whether parameter list can belong to a function declaration. 2132 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 2133 return false; 2134 // If the lines ends with "{", this is likely an function definition. 2135 if (Line.Last->is(tok::l_brace)) 2136 return true; 2137 if (Next->Next == Next->MatchingParen) 2138 return true; // Empty parentheses. 2139 // If there is an &/&& after the r_paren, this is likely a function. 2140 if (Next->MatchingParen->Next && 2141 Next->MatchingParen->Next->is(TT_PointerOrReference)) 2142 return true; 2143 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 2144 Tok = Tok->Next) { 2145 if (Tok->isOneOf(tok::l_paren, TT_TemplateOpener) && Tok->MatchingParen) { 2146 Tok = Tok->MatchingParen; 2147 continue; 2148 } 2149 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 2150 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 2151 return true; 2152 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 2153 Tok->Tok.isLiteral()) 2154 return false; 2155 } 2156 return false; 2157 } 2158 2159 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 2160 assert(Line.MightBeFunctionDecl); 2161 2162 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 2163 Style.AlwaysBreakAfterReturnType == 2164 FormatStyle::RTBS_TopLevelDefinitions) && 2165 Line.Level > 0) 2166 return false; 2167 2168 switch (Style.AlwaysBreakAfterReturnType) { 2169 case FormatStyle::RTBS_None: 2170 return false; 2171 case FormatStyle::RTBS_All: 2172 case FormatStyle::RTBS_TopLevel: 2173 return true; 2174 case FormatStyle::RTBS_AllDefinitions: 2175 case FormatStyle::RTBS_TopLevelDefinitions: 2176 return Line.mightBeFunctionDefinition(); 2177 } 2178 2179 return false; 2180 } 2181 2182 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 2183 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2184 E = Line.Children.end(); 2185 I != E; ++I) { 2186 calculateFormattingInformation(**I); 2187 } 2188 2189 Line.First->TotalLength = 2190 Line.First->IsMultiline ? Style.ColumnLimit 2191 : Line.FirstStartColumn + Line.First->ColumnWidth; 2192 FormatToken *Current = Line.First->Next; 2193 bool InFunctionDecl = Line.MightBeFunctionDecl; 2194 while (Current) { 2195 if (isFunctionDeclarationName(*Current, Line)) 2196 Current->Type = TT_FunctionDeclarationName; 2197 if (Current->is(TT_LineComment)) { 2198 if (Current->Previous->BlockKind == BK_BracedInit && 2199 Current->Previous->opensScope()) 2200 Current->SpacesRequiredBefore = 2201 (Style.Cpp11BracedListStyle && !Style.SpacesInParentheses) ? 0 : 1; 2202 else 2203 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2204 2205 // If we find a trailing comment, iterate backwards to determine whether 2206 // it seems to relate to a specific parameter. If so, break before that 2207 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2208 // to the previous line in: 2209 // SomeFunction(a, 2210 // b, // comment 2211 // c); 2212 if (!Current->HasUnescapedNewline) { 2213 for (FormatToken *Parameter = Current->Previous; Parameter; 2214 Parameter = Parameter->Previous) { 2215 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 2216 break; 2217 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 2218 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 2219 Parameter->HasUnescapedNewline) 2220 Parameter->MustBreakBefore = true; 2221 break; 2222 } 2223 } 2224 } 2225 } else if (Current->SpacesRequiredBefore == 0 && 2226 spaceRequiredBefore(Line, *Current)) { 2227 Current->SpacesRequiredBefore = 1; 2228 } 2229 2230 Current->MustBreakBefore = 2231 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 2232 2233 if (!Current->MustBreakBefore && InFunctionDecl && 2234 Current->is(TT_FunctionDeclarationName)) 2235 Current->MustBreakBefore = mustBreakForReturnType(Line); 2236 2237 Current->CanBreakBefore = 2238 Current->MustBreakBefore || canBreakBefore(Line, *Current); 2239 unsigned ChildSize = 0; 2240 if (Current->Previous->Children.size() == 1) { 2241 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last; 2242 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 2243 : LastOfChild.TotalLength + 1; 2244 } 2245 const FormatToken *Prev = Current->Previous; 2246 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 2247 (Prev->Children.size() == 1 && 2248 Prev->Children[0]->First->MustBreakBefore) || 2249 Current->IsMultiline) 2250 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 2251 else 2252 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 2253 ChildSize + Current->SpacesRequiredBefore; 2254 2255 if (Current->is(TT_CtorInitializerColon)) 2256 InFunctionDecl = false; 2257 2258 // FIXME: Only calculate this if CanBreakBefore is true once static 2259 // initializers etc. are sorted out. 2260 // FIXME: Move magic numbers to a better place. 2261 2262 // Reduce penalty for aligning ObjC method arguments using the colon 2263 // alignment as this is the canonical way (still prefer fitting everything 2264 // into one line if possible). Trying to fit a whole expression into one 2265 // line should not force other line breaks (e.g. when ObjC method 2266 // expression is a part of other expression). 2267 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl); 2268 if (Style.Language == FormatStyle::LK_ObjC && 2269 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) { 2270 if (Current->ParameterIndex == 1) 2271 Current->SplitPenalty += 5 * Current->BindingStrength; 2272 } else { 2273 Current->SplitPenalty += 20 * Current->BindingStrength; 2274 } 2275 2276 Current = Current->Next; 2277 } 2278 2279 calculateUnbreakableTailLengths(Line); 2280 unsigned IndentLevel = Line.Level; 2281 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2282 if (Current->Role) 2283 Current->Role->precomputeFormattingInfos(Current); 2284 if (Current->MatchingParen && 2285 Current->MatchingParen->opensBlockOrBlockTypeList(Style)) { 2286 assert(IndentLevel > 0); 2287 --IndentLevel; 2288 } 2289 Current->IndentLevel = IndentLevel; 2290 if (Current->opensBlockOrBlockTypeList(Style)) 2291 ++IndentLevel; 2292 } 2293 2294 LLVM_DEBUG({ printDebugInfo(Line); }); 2295 } 2296 2297 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2298 unsigned UnbreakableTailLength = 0; 2299 FormatToken *Current = Line.Last; 2300 while (Current) { 2301 Current->UnbreakableTailLength = UnbreakableTailLength; 2302 if (Current->CanBreakBefore || 2303 Current->isOneOf(tok::comment, tok::string_literal)) { 2304 UnbreakableTailLength = 0; 2305 } else { 2306 UnbreakableTailLength += 2307 Current->ColumnWidth + Current->SpacesRequiredBefore; 2308 } 2309 Current = Current->Previous; 2310 } 2311 } 2312 2313 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2314 const FormatToken &Tok, 2315 bool InFunctionDecl) { 2316 const FormatToken &Left = *Tok.Previous; 2317 const FormatToken &Right = Tok; 2318 2319 if (Left.is(tok::semi)) 2320 return 0; 2321 2322 if (Style.Language == FormatStyle::LK_Java) { 2323 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2324 return 1; 2325 if (Right.is(Keywords.kw_implements)) 2326 return 2; 2327 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2328 return 3; 2329 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2330 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2331 return 100; 2332 if (Left.is(TT_JsTypeColon)) 2333 return 35; 2334 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2335 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2336 return 100; 2337 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2338 if (Left.opensScope() && Right.closesScope()) 2339 return 200; 2340 } 2341 2342 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2343 return 1; 2344 if (Right.is(tok::l_square)) { 2345 if (Style.Language == FormatStyle::LK_Proto) 2346 return 1; 2347 if (Left.is(tok::r_square)) 2348 return 200; 2349 // Slightly prefer formatting local lambda definitions like functions. 2350 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2351 return 35; 2352 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2353 TT_ArrayInitializerLSquare, 2354 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) 2355 return 500; 2356 } 2357 2358 if (Left.is(tok::coloncolon) || 2359 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2360 return 500; 2361 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2362 Right.is(tok::kw_operator)) { 2363 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2364 return 3; 2365 if (Left.is(TT_StartOfName)) 2366 return 110; 2367 if (InFunctionDecl && Right.NestingLevel == 0) 2368 return Style.PenaltyReturnTypeOnItsOwnLine; 2369 return 200; 2370 } 2371 if (Right.is(TT_PointerOrReference)) 2372 return 190; 2373 if (Right.is(TT_LambdaArrow)) 2374 return 110; 2375 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2376 return 160; 2377 if (Left.is(TT_CastRParen)) 2378 return 100; 2379 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2380 return 5000; 2381 if (Left.is(tok::comment)) 2382 return 1000; 2383 2384 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2385 TT_CtorInitializerColon)) 2386 return 2; 2387 2388 if (Right.isMemberAccess()) { 2389 // Breaking before the "./->" of a chained call/member access is reasonably 2390 // cheap, as formatting those with one call per line is generally 2391 // desirable. In particular, it should be cheaper to break before the call 2392 // than it is to break inside a call's parameters, which could lead to weird 2393 // "hanging" indents. The exception is the very last "./->" to support this 2394 // frequent pattern: 2395 // 2396 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2397 // dddddddd); 2398 // 2399 // which might otherwise be blown up onto many lines. Here, clang-format 2400 // won't produce "hanging" indents anyway as there is no other trailing 2401 // call. 2402 // 2403 // Also apply higher penalty is not a call as that might lead to a wrapping 2404 // like: 2405 // 2406 // aaaaaaa 2407 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2408 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2409 ? 150 2410 : 35; 2411 } 2412 2413 if (Right.is(TT_TrailingAnnotation) && 2414 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2415 // Moving trailing annotations to the next line is fine for ObjC method 2416 // declarations. 2417 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2418 return 10; 2419 // Generally, breaking before a trailing annotation is bad unless it is 2420 // function-like. It seems to be especially preferable to keep standard 2421 // annotations (i.e. "const", "final" and "override") on the same line. 2422 // Use a slightly higher penalty after ")" so that annotations like 2423 // "const override" are kept together. 2424 bool is_short_annotation = Right.TokenText.size() < 10; 2425 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 2426 } 2427 2428 // In for-loops, prefer breaking at ',' and ';'. 2429 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 2430 return 4; 2431 2432 // In Objective-C method expressions, prefer breaking before "param:" over 2433 // breaking after it. 2434 if (Right.is(TT_SelectorName)) 2435 return 0; 2436 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 2437 return Line.MightBeFunctionDecl ? 50 : 500; 2438 2439 // In Objective-C type declarations, avoid breaking after the category's 2440 // open paren (we'll prefer breaking after the protocol list's opening 2441 // angle bracket, if present). 2442 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous && 2443 Left.Previous->isOneOf(tok::identifier, tok::greater)) 2444 return 500; 2445 2446 if (Left.is(tok::l_paren) && InFunctionDecl && 2447 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 2448 return 100; 2449 if (Left.is(tok::l_paren) && Left.Previous && 2450 (Left.Previous->is(tok::kw_for) || Left.Previous->isIf())) 2451 return 1000; 2452 if (Left.is(tok::equal) && InFunctionDecl) 2453 return 110; 2454 if (Right.is(tok::r_brace)) 2455 return 1; 2456 if (Left.is(TT_TemplateOpener)) 2457 return 100; 2458 if (Left.opensScope()) { 2459 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign) 2460 return 0; 2461 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle) 2462 return 19; 2463 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 2464 : 19; 2465 } 2466 if (Left.is(TT_JavaAnnotation)) 2467 return 50; 2468 2469 if (Left.is(TT_UnaryOperator)) 2470 return 60; 2471 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 2472 Left.Previous->isLabelString() && 2473 (Left.NextOperator || Left.OperatorIndex != 0)) 2474 return 50; 2475 if (Right.is(tok::plus) && Left.isLabelString() && 2476 (Right.NextOperator || Right.OperatorIndex != 0)) 2477 return 25; 2478 if (Left.is(tok::comma)) 2479 return 1; 2480 if (Right.is(tok::lessless) && Left.isLabelString() && 2481 (Right.NextOperator || Right.OperatorIndex != 1)) 2482 return 25; 2483 if (Right.is(tok::lessless)) { 2484 // Breaking at a << is really cheap. 2485 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 2486 // Slightly prefer to break before the first one in log-like statements. 2487 return 2; 2488 return 1; 2489 } 2490 if (Left.ClosesTemplateDeclaration) 2491 return Style.PenaltyBreakTemplateDeclaration; 2492 if (Left.is(TT_ConditionalExpr)) 2493 return prec::Conditional; 2494 prec::Level Level = Left.getPrecedence(); 2495 if (Level == prec::Unknown) 2496 Level = Right.getPrecedence(); 2497 if (Level == prec::Assignment) 2498 return Style.PenaltyBreakAssignment; 2499 if (Level != prec::Unknown) 2500 return Level; 2501 2502 return 3; 2503 } 2504 2505 bool TokenAnnotator::spaceRequiredBeforeParens(const FormatToken &Right) const { 2506 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always || 2507 (Style.SpaceBeforeParens == FormatStyle::SBPO_NonEmptyParentheses && 2508 Right.ParameterCount > 0); 2509 } 2510 2511 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 2512 const FormatToken &Left, 2513 const FormatToken &Right) { 2514 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 2515 return true; 2516 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 2517 return true; 2518 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 2519 Left.Tok.getObjCKeywordID() == tok::objc_property) 2520 return true; 2521 if (Right.is(tok::hashhash)) 2522 return Left.is(tok::hash); 2523 if (Left.isOneOf(tok::hashhash, tok::hash)) 2524 return Right.is(tok::hash); 2525 if ((Left.is(tok::l_paren) && Right.is(tok::r_paren)) || 2526 (Left.is(tok::l_brace) && Left.BlockKind != BK_Block && 2527 Right.is(tok::r_brace) && Right.BlockKind != BK_Block)) 2528 return Style.SpaceInEmptyParentheses; 2529 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 2530 return (Right.is(TT_CastRParen) || 2531 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 2532 ? Style.SpacesInCStyleCastParentheses 2533 : Style.SpacesInParentheses; 2534 if (Right.isOneOf(tok::semi, tok::comma)) 2535 return false; 2536 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) { 2537 bool IsLightweightGeneric = Right.MatchingParen && 2538 Right.MatchingParen->Next && 2539 Right.MatchingParen->Next->is(tok::colon); 2540 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList; 2541 } 2542 if (Right.is(tok::less) && Left.is(tok::kw_template)) 2543 return Style.SpaceAfterTemplateKeyword; 2544 if (Left.isOneOf(tok::exclaim, tok::tilde)) 2545 return false; 2546 if (Left.is(tok::at) && 2547 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 2548 tok::numeric_constant, tok::l_paren, tok::l_brace, 2549 tok::kw_true, tok::kw_false)) 2550 return false; 2551 if (Left.is(tok::colon)) 2552 return !Left.is(TT_ObjCMethodExpr); 2553 if (Left.is(tok::coloncolon)) 2554 return false; 2555 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 2556 if (Style.Language == FormatStyle::LK_TextProto || 2557 (Style.Language == FormatStyle::LK_Proto && 2558 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 2559 // Format empty list as `<>`. 2560 if (Left.is(tok::less) && Right.is(tok::greater)) 2561 return false; 2562 return !Style.Cpp11BracedListStyle; 2563 } 2564 return false; 2565 } 2566 if (Right.is(tok::ellipsis)) 2567 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 2568 Left.Previous->is(tok::kw_case)); 2569 if (Left.is(tok::l_square) && Right.is(tok::amp)) 2570 return false; 2571 if (Right.is(TT_PointerOrReference)) { 2572 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 2573 if (!Left.MatchingParen) 2574 return true; 2575 FormatToken *TokenBeforeMatchingParen = 2576 Left.MatchingParen->getPreviousNonComment(); 2577 if (!TokenBeforeMatchingParen || 2578 !TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype, 2579 TT_TypenameMacro)) 2580 return true; 2581 } 2582 return (Left.Tok.isLiteral() || 2583 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 2584 (Style.PointerAlignment != FormatStyle::PAS_Left || 2585 (Line.IsMultiVariableDeclStmt && 2586 (Left.NestingLevel == 0 || 2587 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for))))))); 2588 } 2589 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 2590 (!Left.is(TT_PointerOrReference) || 2591 (Style.PointerAlignment != FormatStyle::PAS_Right && 2592 !Line.IsMultiVariableDeclStmt))) 2593 return true; 2594 if (Left.is(TT_PointerOrReference)) 2595 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) || 2596 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 2597 !Right.is(TT_StartOfName)) || 2598 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) || 2599 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 2600 tok::l_paren) && 2601 (Style.PointerAlignment != FormatStyle::PAS_Right && 2602 !Line.IsMultiVariableDeclStmt) && 2603 Left.Previous && 2604 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon, 2605 tok::l_square)); 2606 if (Right.is(tok::star) && Left.is(tok::l_paren)) 2607 return false; 2608 const auto SpaceRequiredForArrayInitializerLSquare = 2609 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 2610 return Style.SpacesInContainerLiterals || 2611 ((Style.Language == FormatStyle::LK_Proto || 2612 Style.Language == FormatStyle::LK_TextProto) && 2613 !Style.Cpp11BracedListStyle && 2614 LSquareTok.endsSequence(tok::l_square, tok::colon, 2615 TT_SelectorName)); 2616 }; 2617 if (Left.is(tok::l_square)) 2618 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 2619 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 2620 (Left.isOneOf(TT_ArraySubscriptLSquare, TT_StructuredBindingLSquare, 2621 TT_LambdaLSquare) && 2622 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 2623 if (Right.is(tok::r_square)) 2624 return Right.MatchingParen && 2625 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 2626 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 2627 Style)) || 2628 (Style.SpacesInSquareBrackets && 2629 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 2630 TT_StructuredBindingLSquare, 2631 TT_LambdaLSquare)) || 2632 Right.MatchingParen->is(TT_AttributeParen)); 2633 if (Right.is(tok::l_square) && 2634 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2635 TT_DesignatedInitializerLSquare, 2636 TT_StructuredBindingLSquare, TT_AttributeSquare) && 2637 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral)) 2638 return false; 2639 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 2640 return !Left.Children.empty(); // No spaces in "{}". 2641 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) || 2642 (Right.is(tok::r_brace) && Right.MatchingParen && 2643 Right.MatchingParen->BlockKind != BK_Block)) 2644 return Style.Cpp11BracedListStyle ? Style.SpacesInParentheses : true; 2645 if (Left.is(TT_BlockComment)) 2646 // No whitespace in x(/*foo=*/1), except for JavaScript. 2647 return Style.Language == FormatStyle::LK_JavaScript || 2648 !Left.TokenText.endswith("=*/"); 2649 if (Right.is(tok::l_paren)) { 2650 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 2651 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) 2652 return true; 2653 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) || 2654 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never && 2655 (Left.isOneOf(tok::pp_elif, tok::kw_for, tok::kw_while, 2656 tok::kw_switch, tok::kw_case, TT_ForEachMacro, 2657 TT_ObjCForIn) || 2658 Left.isIf(Line.Type != LT_PreprocessorDirective) || 2659 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch, 2660 tok::kw_new, tok::kw_delete) && 2661 (!Left.Previous || Left.Previous->isNot(tok::period))))) || 2662 (spaceRequiredBeforeParens(Right) && 2663 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 2664 Left.is(tok::r_paren) || Left.isSimpleTypeSpecifier() || 2665 (Left.is(tok::r_square) && Left.MatchingParen && 2666 Left.MatchingParen->is(TT_LambdaLSquare))) && 2667 Line.Type != LT_PreprocessorDirective); 2668 } 2669 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 2670 return false; 2671 if (Right.is(TT_UnaryOperator)) 2672 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 2673 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 2674 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 2675 tok::r_paren) || 2676 Left.isSimpleTypeSpecifier()) && 2677 Right.is(tok::l_brace) && Right.getNextNonComment() && 2678 Right.BlockKind != BK_Block) 2679 return false; 2680 if (Left.is(tok::period) || Right.is(tok::period)) 2681 return false; 2682 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L") 2683 return false; 2684 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 2685 Left.MatchingParen->Previous && 2686 (Left.MatchingParen->Previous->is(tok::period) || 2687 Left.MatchingParen->Previous->is(tok::coloncolon))) 2688 // Java call to generic function with explicit type: 2689 // A.<B<C<...>>>DoSomething(); 2690 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference. 2691 return false; 2692 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 2693 return false; 2694 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at)) 2695 // Objective-C dictionary literal -> no space after opening brace. 2696 return false; 2697 if (Right.is(tok::r_brace) && Right.MatchingParen && 2698 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at)) 2699 // Objective-C dictionary literal -> no space before closing brace. 2700 return false; 2701 if (Right.Type == TT_TrailingAnnotation && 2702 Right.isOneOf(tok::amp, tok::ampamp) && 2703 Left.isOneOf(tok::kw_const, tok::kw_volatile) && 2704 (!Right.Next || Right.Next->is(tok::semi))) 2705 // Match const and volatile ref-qualifiers without any additional 2706 // qualifiers such as 2707 // void Fn() const &; 2708 return Style.PointerAlignment != FormatStyle::PAS_Left; 2709 return true; 2710 } 2711 2712 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 2713 const FormatToken &Right) { 2714 const FormatToken &Left = *Right.Previous; 2715 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 2716 return true; // Never ever merge two identifiers. 2717 if (Style.isCpp()) { 2718 if (Left.is(tok::kw_operator)) 2719 return Right.is(tok::coloncolon); 2720 if (Right.is(tok::l_brace) && Right.BlockKind == BK_BracedInit && 2721 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList) 2722 return true; 2723 } else if (Style.Language == FormatStyle::LK_Proto || 2724 Style.Language == FormatStyle::LK_TextProto) { 2725 if (Right.is(tok::period) && 2726 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 2727 Keywords.kw_repeated, Keywords.kw_extend)) 2728 return true; 2729 if (Right.is(tok::l_paren) && 2730 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 2731 return true; 2732 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 2733 return true; 2734 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 2735 if (Left.is(tok::slash) || Right.is(tok::slash)) 2736 return false; 2737 if (Left.MatchingParen && 2738 Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 2739 Right.isOneOf(tok::l_brace, tok::less)) 2740 return !Style.Cpp11BracedListStyle; 2741 // A percent is probably part of a formatting specification, such as %lld. 2742 if (Left.is(tok::percent)) 2743 return false; 2744 // Preserve the existence of a space before a percent for cases like 0x%04x 2745 // and "%d %d" 2746 if (Left.is(tok::numeric_constant) && Right.is(tok::percent)) 2747 return Right.WhitespaceRange.getEnd() != Right.WhitespaceRange.getBegin(); 2748 } else if (Style.isCSharp()) { 2749 // space between type and variable e.g. Dictionary<string,string> foo; 2750 if (Left.is(TT_TemplateCloser) && Right.is(TT_StartOfName)) 2751 return true; 2752 // space between keywords and paren e.g. "using (" 2753 if (Right.is(tok::l_paren)) 2754 if (Left.is(tok::kw_using)) 2755 return spaceRequiredBeforeParens(Left); 2756 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2757 if (Left.is(TT_JsFatArrow)) 2758 return true; 2759 // for await ( ... 2760 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 2761 Left.Previous->is(tok::kw_for)) 2762 return true; 2763 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 2764 Right.MatchingParen) { 2765 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 2766 // An async arrow function, for example: `x = async () => foo();`, 2767 // as opposed to calling a function called async: `x = async();` 2768 if (Next && Next->is(TT_JsFatArrow)) 2769 return true; 2770 } 2771 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2772 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2773 return false; 2774 // In tagged template literals ("html`bar baz`"), there is no space between 2775 // the tag identifier and the template string. getIdentifierInfo makes sure 2776 // that the identifier is not a pseudo keyword like `yield`, either. 2777 if (Left.is(tok::identifier) && Keywords.IsJavaScriptIdentifier(Left) && 2778 Right.is(TT_TemplateString)) 2779 return false; 2780 if (Right.is(tok::star) && 2781 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 2782 return false; 2783 if (Right.isOneOf(tok::l_brace, tok::l_square) && 2784 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 2785 Keywords.kw_extends, Keywords.kw_implements)) 2786 return true; 2787 if (Right.is(tok::l_paren)) { 2788 // JS methods can use some keywords as names (e.g. `delete()`). 2789 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 2790 return false; 2791 // Valid JS method names can include keywords, e.g. `foo.delete()` or 2792 // `bar.instanceof()`. Recognize call positions by preceding period. 2793 if (Left.Previous && Left.Previous->is(tok::period) && 2794 Left.Tok.getIdentifierInfo()) 2795 return false; 2796 // Additional unary JavaScript operators that need a space after. 2797 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 2798 tok::kw_void)) 2799 return true; 2800 } 2801 // `foo as const;` casts into a const type. 2802 if (Left.endsSequence(tok::kw_const, Keywords.kw_as)) { 2803 return false; 2804 } 2805 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 2806 tok::kw_const) || 2807 // "of" is only a keyword if it appears after another identifier 2808 // (e.g. as "const x of y" in a for loop), or after a destructuring 2809 // operation (const [x, y] of z, const {a, b} of c). 2810 (Left.is(Keywords.kw_of) && Left.Previous && 2811 (Left.Previous->Tok.is(tok::identifier) || 2812 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 2813 (!Left.Previous || !Left.Previous->is(tok::period))) 2814 return true; 2815 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 2816 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 2817 return false; 2818 if (Left.is(Keywords.kw_as) && 2819 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 2820 return true; 2821 if (Left.is(tok::kw_default) && Left.Previous && 2822 Left.Previous->is(tok::kw_export)) 2823 return true; 2824 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 2825 return true; 2826 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 2827 return false; 2828 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 2829 return false; 2830 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 2831 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 2832 return false; 2833 if (Left.is(tok::ellipsis)) 2834 return false; 2835 if (Left.is(TT_TemplateCloser) && 2836 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 2837 Keywords.kw_implements, Keywords.kw_extends)) 2838 // Type assertions ('<type>expr') are not followed by whitespace. Other 2839 // locations that should have whitespace following are identified by the 2840 // above set of follower tokens. 2841 return false; 2842 if (Right.is(TT_JsNonNullAssertion)) 2843 return false; 2844 if (Left.is(TT_JsNonNullAssertion) && 2845 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 2846 return true; // "x! as string", "x! in y" 2847 } else if (Style.Language == FormatStyle::LK_Java) { 2848 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 2849 return true; 2850 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 2851 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never; 2852 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 2853 tok::kw_protected) || 2854 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 2855 Keywords.kw_native)) && 2856 Right.is(TT_TemplateOpener)) 2857 return true; 2858 } 2859 if (Left.is(TT_ImplicitStringLiteral)) 2860 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2861 if (Line.Type == LT_ObjCMethodDecl) { 2862 if (Left.is(TT_ObjCMethodSpecifier)) 2863 return true; 2864 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right)) 2865 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a 2866 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class 2867 // method declaration. 2868 return false; 2869 } 2870 if (Line.Type == LT_ObjCProperty && 2871 (Right.is(tok::equal) || Left.is(tok::equal))) 2872 return false; 2873 2874 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 2875 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 2876 return true; 2877 if (Right.is(TT_OverloadedOperatorLParen)) 2878 return spaceRequiredBeforeParens(Right); 2879 if (Left.is(tok::comma)) 2880 return true; 2881 if (Right.is(tok::comma)) 2882 return false; 2883 if (Right.is(TT_ObjCBlockLParen)) 2884 return true; 2885 if (Right.is(TT_CtorInitializerColon)) 2886 return Style.SpaceBeforeCtorInitializerColon; 2887 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 2888 return false; 2889 if (Right.is(TT_RangeBasedForLoopColon) && 2890 !Style.SpaceBeforeRangeBasedForLoopColon) 2891 return false; 2892 if (Right.is(tok::colon)) { 2893 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) || 2894 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 2895 return false; 2896 if (Right.is(TT_ObjCMethodExpr)) 2897 return false; 2898 if (Left.is(tok::question)) 2899 return false; 2900 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 2901 return false; 2902 if (Right.is(TT_DictLiteral)) 2903 return Style.SpacesInContainerLiterals; 2904 if (Right.is(TT_AttributeColon)) 2905 return false; 2906 return true; 2907 } 2908 if (Left.is(TT_UnaryOperator)) { 2909 // The alternative operators for ~ and ! are "compl" and "not". 2910 // If they are used instead, we do not want to combine them with 2911 // the token to the right, unless that is a left paren. 2912 if (!Right.is(tok::l_paren)) { 2913 if (Left.is(tok::exclaim) && Left.TokenText == "not") 2914 return true; 2915 if (Left.is(tok::tilde) && Left.TokenText == "compl") 2916 return true; 2917 } 2918 return (Style.SpaceAfterLogicalNot && Left.is(tok::exclaim)) || 2919 Right.is(TT_BinaryOperator); 2920 } 2921 2922 // If the next token is a binary operator or a selector name, we have 2923 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 2924 if (Left.is(TT_CastRParen)) 2925 return Style.SpaceAfterCStyleCast || 2926 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 2927 2928 if (Left.is(tok::greater) && Right.is(tok::greater)) { 2929 if (Style.Language == FormatStyle::LK_TextProto || 2930 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 2931 return !Style.Cpp11BracedListStyle; 2932 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 2933 (Style.Standard < FormatStyle::LS_Cpp11 || Style.SpacesInAngles); 2934 } 2935 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 2936 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 2937 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 2938 return false; 2939 if (!Style.SpaceBeforeAssignmentOperators && Left.isNot(TT_TemplateCloser) && 2940 Right.getPrecedence() == prec::Assignment) 2941 return false; 2942 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) && 2943 (Left.is(tok::identifier) || Left.is(tok::kw_this))) 2944 return false; 2945 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 2946 // Generally don't remove existing spaces between an identifier and "::". 2947 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 2948 // this turns out to be too lenient, add analysis of the identifier itself. 2949 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2950 if (Right.is(tok::coloncolon) && !Left.isOneOf(tok::l_brace, tok::comment)) 2951 return (Left.is(TT_TemplateOpener) && 2952 Style.Standard < FormatStyle::LS_Cpp11) || 2953 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 2954 tok::kw___super, TT_TemplateCloser, 2955 TT_TemplateOpener)) || 2956 (Left.is(tok ::l_paren) && Style.SpacesInParentheses); 2957 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 2958 return Style.SpacesInAngles; 2959 // Space before TT_StructuredBindingLSquare. 2960 if (Right.is(TT_StructuredBindingLSquare)) 2961 return !Left.isOneOf(tok::amp, tok::ampamp) || 2962 Style.PointerAlignment != FormatStyle::PAS_Right; 2963 // Space before & or && following a TT_StructuredBindingLSquare. 2964 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 2965 Right.isOneOf(tok::amp, tok::ampamp)) 2966 return Style.PointerAlignment != FormatStyle::PAS_Left; 2967 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 2968 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 2969 !Right.is(tok::r_paren))) 2970 return true; 2971 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) && 2972 Right.isNot(TT_FunctionTypeLParen)) 2973 return spaceRequiredBeforeParens(Right); 2974 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 2975 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 2976 return false; 2977 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 2978 Line.startsWith(tok::hash)) 2979 return true; 2980 if (Right.is(TT_TrailingUnaryOperator)) 2981 return false; 2982 if (Left.is(TT_RegexLiteral)) 2983 return false; 2984 return spaceRequiredBetween(Line, Left, Right); 2985 } 2986 2987 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 2988 static bool isAllmanBrace(const FormatToken &Tok) { 2989 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block && 2990 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_LambdaLBrace, TT_DictLiteral); 2991 } 2992 2993 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 2994 const FormatToken &Right) { 2995 const FormatToken &Left = *Right.Previous; 2996 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 2997 return true; 2998 2999 if (Style.Language == FormatStyle::LK_JavaScript) { 3000 // FIXME: This might apply to other languages and token kinds. 3001 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 3002 Left.Previous->is(tok::string_literal)) 3003 return true; 3004 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 3005 Left.Previous && Left.Previous->is(tok::equal) && 3006 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 3007 tok::kw_const) && 3008 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 3009 // above. 3010 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 3011 // Object literals on the top level of a file are treated as "enum-style". 3012 // Each key/value pair is put on a separate line, instead of bin-packing. 3013 return true; 3014 if (Left.is(tok::l_brace) && Line.Level == 0 && 3015 (Line.startsWith(tok::kw_enum) || 3016 Line.startsWith(tok::kw_const, tok::kw_enum) || 3017 Line.startsWith(tok::kw_export, tok::kw_enum) || 3018 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 3019 // JavaScript top-level enum key/value pairs are put on separate lines 3020 // instead of bin-packing. 3021 return true; 3022 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 3023 !Left.Children.empty()) 3024 // Support AllowShortFunctionsOnASingleLine for JavaScript. 3025 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 3026 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 3027 (Left.NestingLevel == 0 && Line.Level == 0 && 3028 Style.AllowShortFunctionsOnASingleLine & 3029 FormatStyle::SFS_InlineOnly); 3030 } else if (Style.Language == FormatStyle::LK_Java) { 3031 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 3032 Right.Next->is(tok::string_literal)) 3033 return true; 3034 } else if (Style.Language == FormatStyle::LK_Cpp || 3035 Style.Language == FormatStyle::LK_ObjC || 3036 Style.Language == FormatStyle::LK_Proto || 3037 Style.Language == FormatStyle::LK_TableGen || 3038 Style.Language == FormatStyle::LK_TextProto) { 3039 if (Left.isStringLiteral() && Right.isStringLiteral()) 3040 return true; 3041 } 3042 3043 // If the last token before a '}', ']', or ')' is a comma or a trailing 3044 // comment, the intention is to insert a line break after it in order to make 3045 // shuffling around entries easier. Import statements, especially in 3046 // JavaScript, can be an exception to this rule. 3047 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 3048 const FormatToken *BeforeClosingBrace = nullptr; 3049 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 3050 (Style.Language == FormatStyle::LK_JavaScript && 3051 Left.is(tok::l_paren))) && 3052 Left.BlockKind != BK_Block && Left.MatchingParen) 3053 BeforeClosingBrace = Left.MatchingParen->Previous; 3054 else if (Right.MatchingParen && 3055 (Right.MatchingParen->isOneOf(tok::l_brace, 3056 TT_ArrayInitializerLSquare) || 3057 (Style.Language == FormatStyle::LK_JavaScript && 3058 Right.MatchingParen->is(tok::l_paren)))) 3059 BeforeClosingBrace = &Left; 3060 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 3061 BeforeClosingBrace->isTrailingComment())) 3062 return true; 3063 } 3064 3065 if (Right.is(tok::comment)) 3066 return Left.BlockKind != BK_BracedInit && 3067 Left.isNot(TT_CtorInitializerColon) && 3068 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 3069 if (Left.isTrailingComment()) 3070 return true; 3071 if (Right.Previous->IsUnterminatedLiteral) 3072 return true; 3073 if (Right.is(tok::lessless) && Right.Next && 3074 Right.Previous->is(tok::string_literal) && 3075 Right.Next->is(tok::string_literal)) 3076 return true; 3077 if (Right.Previous->ClosesTemplateDeclaration && 3078 Right.Previous->MatchingParen && 3079 Right.Previous->MatchingParen->NestingLevel == 0 && 3080 Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes) 3081 return true; 3082 if (Right.is(TT_CtorInitializerComma) && 3083 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3084 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 3085 return true; 3086 if (Right.is(TT_CtorInitializerColon) && 3087 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 3088 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 3089 return true; 3090 // Break only if we have multiple inheritance. 3091 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 3092 Right.is(TT_InheritanceComma)) 3093 return true; 3094 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 3095 // Multiline raw string literals are special wrt. line breaks. The author 3096 // has made a deliberate choice and might have aligned the contents of the 3097 // string literal accordingly. Thus, we try keep existing line breaks. 3098 return Right.IsMultiline && Right.NewlinesBefore > 0; 3099 if ((Right.Previous->is(tok::l_brace) || 3100 (Right.Previous->is(tok::less) && Right.Previous->Previous && 3101 Right.Previous->Previous->is(tok::equal))) && 3102 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 3103 // Don't put enums or option definitions onto single lines in protocol 3104 // buffers. 3105 return true; 3106 } 3107 if (Right.is(TT_InlineASMBrace)) 3108 return Right.HasUnescapedNewline; 3109 if (isAllmanBrace(Left) || isAllmanBrace(Right)) 3110 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) || 3111 (Line.startsWith(tok::kw_typedef, tok::kw_enum) && 3112 Style.BraceWrapping.AfterEnum) || 3113 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 3114 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 3115 if (Left.is(TT_ObjCBlockLBrace) && 3116 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 3117 return true; 3118 3119 if (Left.is(TT_LambdaLBrace)) { 3120 if (Left.MatchingParen && Left.MatchingParen->Next && 3121 Left.MatchingParen->Next->isOneOf(tok::comma, tok::r_paren) && 3122 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline) 3123 return false; 3124 3125 if (Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_None || 3126 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline || 3127 (!Left.Children.empty() && 3128 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Empty)) 3129 return true; 3130 } 3131 3132 // Put multiple C# attributes on a new line. 3133 if (Style.isCSharp() && 3134 ((Left.is(TT_AttributeSquare) && Left.is(tok::r_square)) || 3135 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare) && 3136 Right.is(tok::l_square)))) 3137 return true; 3138 3139 // Put multiple Java annotation on a new line. 3140 if ((Style.Language == FormatStyle::LK_Java || 3141 Style.Language == FormatStyle::LK_JavaScript) && 3142 Left.is(TT_LeadingJavaAnnotation) && 3143 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 3144 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 3145 return true; 3146 3147 if (Right.is(TT_ProtoExtensionLSquare)) 3148 return true; 3149 3150 // In text proto instances if a submessage contains at least 2 entries and at 3151 // least one of them is a submessage, like A { ... B { ... } ... }, 3152 // put all of the entries of A on separate lines by forcing the selector of 3153 // the submessage B to be put on a newline. 3154 // 3155 // Example: these can stay on one line: 3156 // a { scalar_1: 1 scalar_2: 2 } 3157 // a { b { key: value } } 3158 // 3159 // and these entries need to be on a new line even if putting them all in one 3160 // line is under the column limit: 3161 // a { 3162 // scalar: 1 3163 // b { key: value } 3164 // } 3165 // 3166 // We enforce this by breaking before a submessage field that has previous 3167 // siblings, *and* breaking before a field that follows a submessage field. 3168 // 3169 // Be careful to exclude the case [proto.ext] { ... } since the `]` is 3170 // the TT_SelectorName there, but we don't want to break inside the brackets. 3171 // 3172 // Another edge case is @submessage { key: value }, which is a common 3173 // substitution placeholder. In this case we want to keep `@` and `submessage` 3174 // together. 3175 // 3176 // We ensure elsewhere that extensions are always on their own line. 3177 if ((Style.Language == FormatStyle::LK_Proto || 3178 Style.Language == FormatStyle::LK_TextProto) && 3179 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) { 3180 // Keep `@submessage` together in: 3181 // @submessage { key: value } 3182 if (Right.Previous && Right.Previous->is(tok::at)) 3183 return false; 3184 // Look for the scope opener after selector in cases like: 3185 // selector { ... 3186 // selector: { ... 3187 // selector: @base { ... 3188 FormatToken *LBrace = Right.Next; 3189 if (LBrace && LBrace->is(tok::colon)) { 3190 LBrace = LBrace->Next; 3191 if (LBrace && LBrace->is(tok::at)) { 3192 LBrace = LBrace->Next; 3193 if (LBrace) 3194 LBrace = LBrace->Next; 3195 } 3196 } 3197 if (LBrace && 3198 // The scope opener is one of {, [, <: 3199 // selector { ... } 3200 // selector [ ... ] 3201 // selector < ... > 3202 // 3203 // In case of selector { ... }, the l_brace is TT_DictLiteral. 3204 // In case of an empty selector {}, the l_brace is not TT_DictLiteral, 3205 // so we check for immediately following r_brace. 3206 ((LBrace->is(tok::l_brace) && 3207 (LBrace->is(TT_DictLiteral) || 3208 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) || 3209 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) { 3210 // If Left.ParameterCount is 0, then this submessage entry is not the 3211 // first in its parent submessage, and we want to break before this entry. 3212 // If Left.ParameterCount is greater than 0, then its parent submessage 3213 // might contain 1 or more entries and we want to break before this entry 3214 // if it contains at least 2 entries. We deal with this case later by 3215 // detecting and breaking before the next entry in the parent submessage. 3216 if (Left.ParameterCount == 0) 3217 return true; 3218 // However, if this submessage is the first entry in its parent 3219 // submessage, Left.ParameterCount might be 1 in some cases. 3220 // We deal with this case later by detecting an entry 3221 // following a closing paren of this submessage. 3222 } 3223 3224 // If this is an entry immediately following a submessage, it will be 3225 // preceded by a closing paren of that submessage, like in: 3226 // left---. .---right 3227 // v v 3228 // sub: { ... } key: value 3229 // If there was a comment between `}` an `key` above, then `key` would be 3230 // put on a new line anyways. 3231 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square)) 3232 return true; 3233 } 3234 3235 // Deal with lambda arguments in C++ - we want consistent line breaks whether 3236 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced 3237 // as aggressive line breaks are placed when the lambda is not the last arg. 3238 if ((Style.Language == FormatStyle::LK_Cpp || 3239 Style.Language == FormatStyle::LK_ObjC) && 3240 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 && 3241 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) { 3242 // Multiple lambdas in the same function call force line breaks. 3243 if (Left.BlockParameterCount > 1) 3244 return true; 3245 3246 // A lambda followed by another arg forces a line break. 3247 if (!Left.Role) 3248 return false; 3249 auto Comma = Left.Role->lastComma(); 3250 if (!Comma) 3251 return false; 3252 auto Next = Comma->getNextNonComment(); 3253 if (!Next) 3254 return false; 3255 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret)) 3256 return true; 3257 } 3258 3259 return false; 3260 } 3261 3262 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 3263 const FormatToken &Right) { 3264 const FormatToken &Left = *Right.Previous; 3265 3266 // Language-specific stuff. 3267 if (Style.Language == FormatStyle::LK_Java) { 3268 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3269 Keywords.kw_implements)) 3270 return false; 3271 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 3272 Keywords.kw_implements)) 3273 return true; 3274 } else if (Style.Language == FormatStyle::LK_JavaScript) { 3275 const FormatToken *NonComment = Right.getPreviousNonComment(); 3276 if (NonComment && 3277 NonComment->isOneOf( 3278 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 3279 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 3280 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 3281 Keywords.kw_readonly, Keywords.kw_abstract, Keywords.kw_get, 3282 Keywords.kw_set, Keywords.kw_async, Keywords.kw_await)) 3283 return false; // Otherwise automatic semicolon insertion would trigger. 3284 if (Right.NestingLevel == 0 && 3285 (Left.Tok.getIdentifierInfo() || 3286 Left.isOneOf(tok::r_square, tok::r_paren)) && 3287 Right.isOneOf(tok::l_square, tok::l_paren)) 3288 return false; // Otherwise automatic semicolon insertion would trigger. 3289 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace)) 3290 return false; 3291 if (Left.is(TT_JsTypeColon)) 3292 return true; 3293 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;"). 3294 if (Left.is(tok::exclaim) && Right.is(tok::colon)) 3295 return false; 3296 // Look for is type annotations like: 3297 // function f(): a is B { ... } 3298 // Do not break before is in these cases. 3299 if (Right.is(Keywords.kw_is)) { 3300 const FormatToken *Next = Right.getNextNonComment(); 3301 // If `is` is followed by a colon, it's likely that it's a dict key, so 3302 // ignore it for this check. 3303 // For example this is common in Polymer: 3304 // Polymer({ 3305 // is: 'name', 3306 // ... 3307 // }); 3308 if (!Next || !Next->is(tok::colon)) 3309 return false; 3310 } 3311 if (Left.is(Keywords.kw_in)) 3312 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 3313 if (Right.is(Keywords.kw_in)) 3314 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 3315 if (Right.is(Keywords.kw_as)) 3316 return false; // must not break before as in 'x as type' casts 3317 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) { 3318 // extends and infer can appear as keywords in conditional types: 3319 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types 3320 // do not break before them, as the expressions are subject to ASI. 3321 return false; 3322 } 3323 if (Left.is(Keywords.kw_as)) 3324 return true; 3325 if (Left.is(TT_JsNonNullAssertion)) 3326 return true; 3327 if (Left.is(Keywords.kw_declare) && 3328 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 3329 Keywords.kw_function, tok::kw_class, tok::kw_enum, 3330 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 3331 Keywords.kw_let, tok::kw_const)) 3332 // See grammar for 'declare' statements at: 3333 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10 3334 return false; 3335 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 3336 Right.isOneOf(tok::identifier, tok::string_literal)) 3337 return false; // must not break in "module foo { ...}" 3338 if (Right.is(TT_TemplateString) && Right.closesScope()) 3339 return false; 3340 // Don't split tagged template literal so there is a break between the tag 3341 // identifier and template string. 3342 if (Left.is(tok::identifier) && Right.is(TT_TemplateString)) { 3343 return false; 3344 } 3345 if (Left.is(TT_TemplateString) && Left.opensScope()) 3346 return true; 3347 } 3348 3349 if (Left.is(tok::at)) 3350 return false; 3351 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 3352 return false; 3353 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 3354 return !Right.is(tok::l_paren); 3355 if (Right.is(TT_PointerOrReference)) 3356 return Line.IsMultiVariableDeclStmt || 3357 (Style.PointerAlignment == FormatStyle::PAS_Right && 3358 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 3359 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 3360 Right.is(tok::kw_operator)) 3361 return true; 3362 if (Left.is(TT_PointerOrReference)) 3363 return false; 3364 if (Right.isTrailingComment()) 3365 // We rely on MustBreakBefore being set correctly here as we should not 3366 // change the "binding" behavior of a comment. 3367 // The first comment in a braced lists is always interpreted as belonging to 3368 // the first list element. Otherwise, it should be placed outside of the 3369 // list. 3370 return Left.BlockKind == BK_BracedInit || 3371 (Left.is(TT_CtorInitializerColon) && 3372 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 3373 if (Left.is(tok::question) && Right.is(tok::colon)) 3374 return false; 3375 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 3376 return Style.BreakBeforeTernaryOperators; 3377 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 3378 return !Style.BreakBeforeTernaryOperators; 3379 if (Left.is(TT_InheritanceColon)) 3380 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon; 3381 if (Right.is(TT_InheritanceColon)) 3382 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon; 3383 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 3384 Left.isNot(TT_SelectorName)) 3385 return true; 3386 3387 if (Right.is(tok::colon) && 3388 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 3389 return false; 3390 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 3391 if (Style.Language == FormatStyle::LK_Proto || 3392 Style.Language == FormatStyle::LK_TextProto) { 3393 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 3394 return false; 3395 // Prevent cases like: 3396 // 3397 // submessage: 3398 // { key: valueeeeeeeeeeee } 3399 // 3400 // when the snippet does not fit into one line. 3401 // Prefer: 3402 // 3403 // submessage: { 3404 // key: valueeeeeeeeeeee 3405 // } 3406 // 3407 // instead, even if it is longer by one line. 3408 // 3409 // Note that this allows allows the "{" to go over the column limit 3410 // when the column limit is just between ":" and "{", but that does 3411 // not happen too often and alternative formattings in this case are 3412 // not much better. 3413 // 3414 // The code covers the cases: 3415 // 3416 // submessage: { ... } 3417 // submessage: < ... > 3418 // repeated: [ ... ] 3419 if (((Right.is(tok::l_brace) || Right.is(tok::less)) && 3420 Right.is(TT_DictLiteral)) || 3421 Right.is(TT_ArrayInitializerLSquare)) 3422 return false; 3423 } 3424 return true; 3425 } 3426 if (Right.is(tok::r_square) && Right.MatchingParen && 3427 Right.MatchingParen->is(TT_ProtoExtensionLSquare)) 3428 return false; 3429 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 3430 Right.Next->is(TT_ObjCMethodExpr))) 3431 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 3432 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 3433 return true; 3434 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 3435 return true; 3436 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 3437 TT_OverloadedOperator)) 3438 return false; 3439 if (Left.is(TT_RangeBasedForLoopColon)) 3440 return true; 3441 if (Right.is(TT_RangeBasedForLoopColon)) 3442 return false; 3443 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 3444 return true; 3445 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 3446 Left.is(tok::kw_operator)) 3447 return false; 3448 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 3449 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 3450 return false; 3451 if (Left.is(tok::equal) && Right.is(tok::l_brace) && 3452 !Style.Cpp11BracedListStyle) 3453 return false; 3454 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen)) 3455 return false; 3456 if (Left.is(tok::l_paren) && Left.Previous && 3457 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 3458 return false; 3459 if (Right.is(TT_ImplicitStringLiteral)) 3460 return false; 3461 3462 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser)) 3463 return false; 3464 if (Right.is(tok::r_square) && Right.MatchingParen && 3465 Right.MatchingParen->is(TT_LambdaLSquare)) 3466 return false; 3467 3468 // We only break before r_brace if there was a corresponding break before 3469 // the l_brace, which is tracked by BreakBeforeClosingBrace. 3470 if (Right.is(tok::r_brace)) 3471 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block; 3472 3473 // Allow breaking after a trailing annotation, e.g. after a method 3474 // declaration. 3475 if (Left.is(TT_TrailingAnnotation)) 3476 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 3477 tok::less, tok::coloncolon); 3478 3479 if (Right.is(tok::kw___attribute) || 3480 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) 3481 return true; 3482 3483 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 3484 return true; 3485 3486 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 3487 return true; 3488 3489 if (Left.is(TT_CtorInitializerColon)) 3490 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 3491 if (Right.is(TT_CtorInitializerColon)) 3492 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 3493 if (Left.is(TT_CtorInitializerComma) && 3494 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 3495 return false; 3496 if (Right.is(TT_CtorInitializerComma) && 3497 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 3498 return true; 3499 if (Left.is(TT_InheritanceComma) && 3500 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 3501 return false; 3502 if (Right.is(TT_InheritanceComma) && 3503 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma) 3504 return true; 3505 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 3506 (Left.is(tok::less) && Right.is(tok::less))) 3507 return false; 3508 if (Right.is(TT_BinaryOperator) && 3509 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 3510 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 3511 Right.getPrecedence() != prec::Assignment)) 3512 return true; 3513 if (Left.is(TT_ArrayInitializerLSquare)) 3514 return true; 3515 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 3516 return true; 3517 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 3518 !Left.isOneOf(tok::arrowstar, tok::lessless) && 3519 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 3520 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 3521 Left.getPrecedence() == prec::Assignment)) 3522 return true; 3523 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 3524 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) 3525 return false; 3526 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 3527 tok::kw_class, tok::kw_struct, tok::comment) || 3528 Right.isMemberAccess() || 3529 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 3530 tok::colon, tok::l_square, tok::at) || 3531 (Left.is(tok::r_paren) && 3532 Right.isOneOf(tok::identifier, tok::kw_const)) || 3533 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 3534 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 3535 } 3536 3537 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 3538 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 3539 const FormatToken *Tok = Line.First; 3540 while (Tok) { 3541 llvm::errs() << " M=" << Tok->MustBreakBefore 3542 << " C=" << Tok->CanBreakBefore 3543 << " T=" << getTokenTypeName(Tok->Type) 3544 << " S=" << Tok->SpacesRequiredBefore 3545 << " B=" << Tok->BlockParameterCount 3546 << " BK=" << Tok->BlockKind << " P=" << Tok->SplitPenalty 3547 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 3548 << " PPK=" << Tok->PackingKind << " FakeLParens="; 3549 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 3550 llvm::errs() << Tok->FakeLParens[i] << "/"; 3551 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 3552 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo(); 3553 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 3554 if (!Tok->Next) 3555 assert(Tok == Line.Last); 3556 Tok = Tok->Next; 3557 } 3558 llvm::errs() << "----\n"; 3559 } 3560 3561 } // namespace format 3562 } // namespace clang 3563