1 //===--- TokenAnnotator.cpp - Format C++ code -----------------------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 /// 10 /// \file 11 /// \brief This file implements a token annotator, i.e. creates 12 /// \c AnnotatedTokens out of \c FormatTokens with required extra information. 13 /// 14 //===----------------------------------------------------------------------===// 15 16 #include "TokenAnnotator.h" 17 #include "clang/Basic/SourceManager.h" 18 #include "llvm/ADT/SmallPtrSet.h" 19 #include "llvm/Support/Debug.h" 20 21 #define DEBUG_TYPE "format-token-annotator" 22 23 namespace clang { 24 namespace format { 25 26 namespace { 27 28 /// \brief A parser that gathers additional information about tokens. 29 /// 30 /// The \c TokenAnnotator tries to match parenthesis and square brakets and 31 /// store a parenthesis levels. It also tries to resolve matching "<" and ">" 32 /// into template parameter lists. 33 class AnnotatingParser { 34 public: 35 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line, 36 const AdditionalKeywords &Keywords) 37 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false), 38 Keywords(Keywords) { 39 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false)); 40 resetTokenMetadata(CurrentToken); 41 } 42 43 private: 44 bool parseAngle() { 45 if (!CurrentToken || !CurrentToken->Previous) 46 return false; 47 if (NonTemplateLess.count(CurrentToken->Previous)) 48 return false; 49 50 const FormatToken &Previous = *CurrentToken->Previous; // The '<'. 51 if (Previous.Previous) { 52 if (Previous.Previous->Tok.isLiteral()) 53 return false; 54 if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 && 55 (!Previous.Previous->MatchingParen || 56 !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen))) 57 return false; 58 } 59 60 FormatToken *Left = CurrentToken->Previous; 61 Left->ParentBracket = Contexts.back().ContextKind; 62 ScopedContextCreator ContextCreator(*this, tok::less, 12); 63 64 // If this angle is in the context of an expression, we need to be more 65 // hesitant to detect it as opening template parameters. 66 bool InExprContext = Contexts.back().IsExpression; 67 68 Contexts.back().IsExpression = false; 69 // If there's a template keyword before the opening angle bracket, this is a 70 // template parameter, not an argument. 71 Contexts.back().InTemplateArgument = 72 Left->Previous && Left->Previous->Tok.isNot(tok::kw_template); 73 74 if (Style.Language == FormatStyle::LK_Java && 75 CurrentToken->is(tok::question)) 76 next(); 77 78 while (CurrentToken) { 79 if (CurrentToken->is(tok::greater)) { 80 Left->MatchingParen = CurrentToken; 81 CurrentToken->MatchingParen = Left; 82 // In TT_Proto, we must distignuish between: 83 // map<key, value> 84 // msg < item: data > 85 // msg: < item: data > 86 // In TT_TextProto, map<key, value> does not occur. 87 if (Style.Language == FormatStyle::LK_TextProto || 88 (Style.Language == FormatStyle::LK_Proto && Left->Previous && 89 Left->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) 90 CurrentToken->Type = TT_DictLiteral; 91 else 92 CurrentToken->Type = TT_TemplateCloser; 93 next(); 94 return true; 95 } 96 if (CurrentToken->is(tok::question) && 97 Style.Language == FormatStyle::LK_Java) { 98 next(); 99 continue; 100 } 101 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) || 102 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext && 103 Style.Language != FormatStyle::LK_Proto && 104 Style.Language != FormatStyle::LK_TextProto)) 105 return false; 106 // If a && or || is found and interpreted as a binary operator, this set 107 // of angles is likely part of something like "a < b && c > d". If the 108 // angles are inside an expression, the ||/&& might also be a binary 109 // operator that was misinterpreted because we are parsing template 110 // parameters. 111 // FIXME: This is getting out of hand, write a decent parser. 112 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) && 113 CurrentToken->Previous->is(TT_BinaryOperator) && 114 Contexts[Contexts.size() - 2].IsExpression && 115 !Line.startsWith(tok::kw_template)) 116 return false; 117 updateParameterCount(Left, CurrentToken); 118 if (Style.Language == FormatStyle::LK_Proto) { 119 if (FormatToken *Previous = CurrentToken->getPreviousNonComment()) { 120 if (CurrentToken->is(tok::colon) || 121 (CurrentToken->isOneOf(tok::l_brace, tok::less) && 122 Previous->isNot(tok::colon))) 123 Previous->Type = TT_SelectorName; 124 } 125 } 126 if (!consumeToken()) 127 return false; 128 } 129 return false; 130 } 131 132 bool parseParens(bool LookForDecls = false) { 133 if (!CurrentToken) 134 return false; 135 FormatToken *Left = CurrentToken->Previous; 136 Left->ParentBracket = Contexts.back().ContextKind; 137 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1); 138 139 // FIXME: This is a bit of a hack. Do better. 140 Contexts.back().ColonIsForRangeExpr = 141 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr; 142 143 bool StartsObjCMethodExpr = false; 144 if (FormatToken *MaybeSel = Left->Previous) { 145 // @selector( starts a selector. 146 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous && 147 MaybeSel->Previous->is(tok::at)) { 148 StartsObjCMethodExpr = true; 149 } 150 } 151 152 if (Left->is(TT_OverloadedOperatorLParen)) { 153 Contexts.back().IsExpression = false; 154 } else if (Style.Language == FormatStyle::LK_JavaScript && 155 (Line.startsWith(Keywords.kw_type, tok::identifier) || 156 Line.startsWith(tok::kw_export, Keywords.kw_type, 157 tok::identifier))) { 158 // type X = (...); 159 // export type X = (...); 160 Contexts.back().IsExpression = false; 161 } else if (Left->Previous && 162 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_decltype, 163 tok::kw_if, tok::kw_while, tok::l_paren, 164 tok::comma) || 165 Left->Previous->endsSequence(tok::kw_constexpr, tok::kw_if) || 166 Left->Previous->is(TT_BinaryOperator))) { 167 // static_assert, if and while usually contain expressions. 168 Contexts.back().IsExpression = true; 169 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 170 (Left->Previous->is(Keywords.kw_function) || 171 (Left->Previous->endsSequence(tok::identifier, 172 Keywords.kw_function)))) { 173 // function(...) or function f(...) 174 Contexts.back().IsExpression = false; 175 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 176 Left->Previous->is(TT_JsTypeColon)) { 177 // let x: (SomeType); 178 Contexts.back().IsExpression = false; 179 } else if (Left->Previous && Left->Previous->is(tok::r_square) && 180 Left->Previous->MatchingParen && 181 Left->Previous->MatchingParen->is(TT_LambdaLSquare)) { 182 // This is a parameter list of a lambda expression. 183 Contexts.back().IsExpression = false; 184 } else if (Line.InPPDirective && 185 (!Left->Previous || !Left->Previous->is(tok::identifier))) { 186 Contexts.back().IsExpression = true; 187 } else if (Contexts[Contexts.size() - 2].CaretFound) { 188 // This is the parameter list of an ObjC block. 189 Contexts.back().IsExpression = false; 190 } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) { 191 Left->Type = TT_AttributeParen; 192 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) { 193 // The first argument to a foreach macro is a declaration. 194 Contexts.back().IsForEachMacro = true; 195 Contexts.back().IsExpression = false; 196 } else if (Left->Previous && Left->Previous->MatchingParen && 197 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) { 198 Contexts.back().IsExpression = false; 199 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) { 200 bool IsForOrCatch = 201 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch); 202 Contexts.back().IsExpression = !IsForOrCatch; 203 } 204 205 if (StartsObjCMethodExpr) { 206 Contexts.back().ColonIsObjCMethodExpr = true; 207 Left->Type = TT_ObjCMethodExpr; 208 } 209 210 // MightBeFunctionType and ProbablyFunctionType are used for 211 // function pointer and reference types as well as Objective-C 212 // block types: 213 // 214 // void (*FunctionPointer)(void); 215 // void (&FunctionReference)(void); 216 // void (^ObjCBlock)(void); 217 bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression; 218 bool ProbablyFunctionType = 219 CurrentToken->isOneOf(tok::star, tok::amp, tok::caret); 220 bool HasMultipleLines = false; 221 bool HasMultipleParametersOnALine = false; 222 bool MightBeObjCForRangeLoop = 223 Left->Previous && Left->Previous->is(tok::kw_for); 224 FormatToken *PossibleObjCForInToken = nullptr; 225 while (CurrentToken) { 226 // LookForDecls is set when "if (" has been seen. Check for 227 // 'identifier' '*' 'identifier' followed by not '=' -- this 228 // '*' has to be a binary operator but determineStarAmpUsage() will 229 // categorize it as an unary operator, so set the right type here. 230 if (LookForDecls && CurrentToken->Next) { 231 FormatToken *Prev = CurrentToken->getPreviousNonComment(); 232 if (Prev) { 233 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 234 FormatToken *Next = CurrentToken->Next; 235 if (PrevPrev && PrevPrev->is(tok::identifier) && 236 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) && 237 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) { 238 Prev->Type = TT_BinaryOperator; 239 LookForDecls = false; 240 } 241 } 242 } 243 244 if (CurrentToken->Previous->is(TT_PointerOrReference) && 245 CurrentToken->Previous->Previous->isOneOf(tok::l_paren, 246 tok::coloncolon)) 247 ProbablyFunctionType = true; 248 if (CurrentToken->is(tok::comma)) 249 MightBeFunctionType = false; 250 if (CurrentToken->Previous->is(TT_BinaryOperator)) 251 Contexts.back().IsExpression = true; 252 if (CurrentToken->is(tok::r_paren)) { 253 if (MightBeFunctionType && ProbablyFunctionType && CurrentToken->Next && 254 (CurrentToken->Next->is(tok::l_paren) || 255 (CurrentToken->Next->is(tok::l_square) && Line.MustBeDeclaration))) 256 Left->Type = Left->Next->is(tok::caret) ? TT_ObjCBlockLParen 257 : TT_FunctionTypeLParen; 258 Left->MatchingParen = CurrentToken; 259 CurrentToken->MatchingParen = Left; 260 261 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) && 262 Left->Previous && Left->Previous->is(tok::l_paren)) { 263 // Detect the case where macros are used to generate lambdas or 264 // function bodies, e.g.: 265 // auto my_lambda = MARCO((Type *type, int i) { .. body .. }); 266 for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) { 267 if (Tok->is(TT_BinaryOperator) && 268 Tok->isOneOf(tok::star, tok::amp, tok::ampamp)) 269 Tok->Type = TT_PointerOrReference; 270 } 271 } 272 273 if (StartsObjCMethodExpr) { 274 CurrentToken->Type = TT_ObjCMethodExpr; 275 if (Contexts.back().FirstObjCSelectorName) { 276 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 277 Contexts.back().LongestObjCSelectorName; 278 } 279 } 280 281 if (Left->is(TT_AttributeParen)) 282 CurrentToken->Type = TT_AttributeParen; 283 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation)) 284 CurrentToken->Type = TT_JavaAnnotation; 285 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation)) 286 CurrentToken->Type = TT_LeadingJavaAnnotation; 287 288 if (!HasMultipleLines) 289 Left->PackingKind = PPK_Inconclusive; 290 else if (HasMultipleParametersOnALine) 291 Left->PackingKind = PPK_BinPacked; 292 else 293 Left->PackingKind = PPK_OnePerLine; 294 295 next(); 296 return true; 297 } 298 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace)) 299 return false; 300 301 if (CurrentToken->is(tok::l_brace)) 302 Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen 303 if (CurrentToken->is(tok::comma) && CurrentToken->Next && 304 !CurrentToken->Next->HasUnescapedNewline && 305 !CurrentToken->Next->isTrailingComment()) 306 HasMultipleParametersOnALine = true; 307 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) || 308 CurrentToken->Previous->isSimpleTypeSpecifier()) && 309 !CurrentToken->is(tok::l_brace)) 310 Contexts.back().IsExpression = false; 311 if (CurrentToken->isOneOf(tok::semi, tok::colon)) { 312 MightBeObjCForRangeLoop = false; 313 if (PossibleObjCForInToken) { 314 PossibleObjCForInToken->Type = TT_Unknown; 315 PossibleObjCForInToken = nullptr; 316 } 317 } 318 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) { 319 PossibleObjCForInToken = CurrentToken; 320 PossibleObjCForInToken->Type = TT_ObjCForIn; 321 } 322 // When we discover a 'new', we set CanBeExpression to 'false' in order to 323 // parse the type correctly. Reset that after a comma. 324 if (CurrentToken->is(tok::comma)) 325 Contexts.back().CanBeExpression = true; 326 327 FormatToken *Tok = CurrentToken; 328 if (!consumeToken()) 329 return false; 330 updateParameterCount(Left, Tok); 331 if (CurrentToken && CurrentToken->HasUnescapedNewline) 332 HasMultipleLines = true; 333 } 334 return false; 335 } 336 337 bool isCpp11AttributeSpecifier(const FormatToken &Tok) { 338 if (!Style.isCpp() || !Tok.startsSequence(tok::l_square, tok::l_square)) 339 return false; 340 const FormatToken *AttrTok = Tok.Next->Next; 341 if (!AttrTok) 342 return false; 343 // C++17 '[[using ns: foo, bar(baz, blech)]]' 344 // We assume nobody will name an ObjC variable 'using'. 345 if (AttrTok->startsSequence(tok::kw_using, tok::identifier, tok::colon)) 346 return true; 347 if (AttrTok->isNot(tok::identifier)) 348 return false; 349 while (AttrTok && !AttrTok->startsSequence(tok::r_square, tok::r_square)) { 350 // ObjC message send. We assume nobody will use : in a C++11 attribute 351 // specifier parameter, although this is technically valid: 352 // [[foo(:)]] 353 if (AttrTok->is(tok::colon) || 354 AttrTok->startsSequence(tok::identifier, tok::identifier)) 355 return false; 356 if (AttrTok->is(tok::ellipsis)) 357 return true; 358 AttrTok = AttrTok->Next; 359 } 360 return AttrTok && AttrTok->startsSequence(tok::r_square, tok::r_square); 361 } 362 363 bool parseSquare() { 364 if (!CurrentToken) 365 return false; 366 367 // A '[' could be an index subscript (after an identifier or after 368 // ')' or ']'), it could be the start of an Objective-C method 369 // expression, it could the start of an Objective-C array literal, 370 // or it could be a C++ attribute specifier [[foo::bar]]. 371 FormatToken *Left = CurrentToken->Previous; 372 Left->ParentBracket = Contexts.back().ContextKind; 373 FormatToken *Parent = Left->getPreviousNonComment(); 374 375 // Cases where '>' is followed by '['. 376 // In C++, this can happen either in array of templates (foo<int>[10]) 377 // or when array is a nested template type (unique_ptr<type1<type2>[]>). 378 bool CppArrayTemplates = 379 Style.isCpp() && Parent && Parent->is(TT_TemplateCloser) && 380 (Contexts.back().CanBeExpression || Contexts.back().IsExpression || 381 Contexts.back().InTemplateArgument); 382 383 bool IsCpp11AttributeSpecifier = isCpp11AttributeSpecifier(*Left) || 384 Contexts.back().InCpp11AttributeSpecifier; 385 386 bool StartsObjCMethodExpr = 387 !CppArrayTemplates && Style.isCpp() && !IsCpp11AttributeSpecifier && 388 Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) && 389 CurrentToken->isNot(tok::l_brace) && 390 (!Parent || 391 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren, 392 tok::kw_return, tok::kw_throw) || 393 Parent->isUnaryOperator() || 394 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) || 395 getBinOpPrecedence(Parent->Tok.getKind(), true, true) > prec::Unknown); 396 bool ColonFound = false; 397 398 unsigned BindingIncrease = 1; 399 if (Left->isCppStructuredBinding(Style)) { 400 Left->Type = TT_StructuredBindingLSquare; 401 } else if (Left->is(TT_Unknown)) { 402 if (StartsObjCMethodExpr) { 403 Left->Type = TT_ObjCMethodExpr; 404 } else if (IsCpp11AttributeSpecifier) { 405 Left->Type = TT_AttributeSquare; 406 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent && 407 Contexts.back().ContextKind == tok::l_brace && 408 Parent->isOneOf(tok::l_brace, tok::comma)) { 409 Left->Type = TT_JsComputedPropertyName; 410 } else if (Style.isCpp() && Contexts.back().ContextKind == tok::l_brace && 411 Parent && Parent->isOneOf(tok::l_brace, tok::comma)) { 412 Left->Type = TT_DesignatedInitializerLSquare; 413 } else if (CurrentToken->is(tok::r_square) && Parent && 414 Parent->is(TT_TemplateCloser)) { 415 Left->Type = TT_ArraySubscriptLSquare; 416 } else if (Style.Language == FormatStyle::LK_Proto || 417 Style.Language == FormatStyle::LK_TextProto) { 418 // Square braces in LK_Proto can either be message field attributes: 419 // 420 // optional Aaa aaa = 1 [ 421 // (aaa) = aaa 422 // ]; 423 // 424 // extensions 123 [ 425 // (aaa) = aaa 426 // ]; 427 // 428 // or text proto extensions (in options): 429 // 430 // option (Aaa.options) = { 431 // [type.type/type] { 432 // key: value 433 // } 434 // } 435 // 436 // or repeated fields (in options): 437 // 438 // option (Aaa.options) = { 439 // keys: [ 1, 2, 3 ] 440 // } 441 // 442 // In the first and the third case we want to spread the contents inside 443 // the square braces; in the second we want to keep them inline. 444 Left->Type = TT_ArrayInitializerLSquare; 445 if (!Left->endsSequence(tok::l_square, tok::numeric_constant, 446 tok::equal) && 447 !Left->endsSequence(tok::l_square, tok::numeric_constant, 448 tok::identifier) && 449 !Left->endsSequence(tok::l_square, tok::colon, TT_SelectorName)) { 450 Left->Type = TT_ProtoExtensionLSquare; 451 BindingIncrease = 10; 452 } 453 } else if (!CppArrayTemplates && Parent && 454 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at, 455 tok::comma, tok::l_paren, tok::l_square, 456 tok::question, tok::colon, tok::kw_return, 457 // Should only be relevant to JavaScript: 458 tok::kw_default)) { 459 Left->Type = TT_ArrayInitializerLSquare; 460 } else { 461 BindingIncrease = 10; 462 Left->Type = TT_ArraySubscriptLSquare; 463 } 464 } 465 466 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease); 467 Contexts.back().IsExpression = true; 468 if (Style.Language == FormatStyle::LK_JavaScript && Parent && 469 Parent->is(TT_JsTypeColon)) 470 Contexts.back().IsExpression = false; 471 472 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr; 473 Contexts.back().InCpp11AttributeSpecifier = IsCpp11AttributeSpecifier; 474 475 while (CurrentToken) { 476 if (CurrentToken->is(tok::r_square)) { 477 if (IsCpp11AttributeSpecifier) 478 CurrentToken->Type = TT_AttributeSquare; 479 else if (CurrentToken->Next && CurrentToken->Next->is(tok::l_paren) && 480 Left->is(TT_ObjCMethodExpr)) { 481 // An ObjC method call is rarely followed by an open parenthesis. 482 // FIXME: Do we incorrectly label ":" with this? 483 StartsObjCMethodExpr = false; 484 Left->Type = TT_Unknown; 485 } 486 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) { 487 CurrentToken->Type = TT_ObjCMethodExpr; 488 // determineStarAmpUsage() thinks that '*' '[' is allocating an 489 // array of pointers, but if '[' starts a selector then '*' is a 490 // binary operator. 491 if (Parent && Parent->is(TT_PointerOrReference)) 492 Parent->Type = TT_BinaryOperator; 493 } 494 Left->MatchingParen = CurrentToken; 495 CurrentToken->MatchingParen = Left; 496 if (Contexts.back().FirstObjCSelectorName) { 497 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 498 Contexts.back().LongestObjCSelectorName; 499 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts = 500 Left->ParameterCount; 501 if (Left->BlockParameterCount > 1) 502 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0; 503 } 504 next(); 505 return true; 506 } 507 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace)) 508 return false; 509 if (CurrentToken->is(tok::colon)) { 510 if (IsCpp11AttributeSpecifier && 511 CurrentToken->endsSequence(tok::colon, tok::identifier, 512 tok::kw_using)) { 513 // Remember that this is a [[using ns: foo]] C++ attribute, so we 514 // don't add a space before the colon (unlike other colons). 515 CurrentToken->Type = TT_AttributeColon; 516 } else if (Left->isOneOf(TT_ArraySubscriptLSquare, 517 TT_DesignatedInitializerLSquare)) { 518 Left->Type = TT_ObjCMethodExpr; 519 StartsObjCMethodExpr = true; 520 // ParameterCount might have been set to 1 before expression was 521 // recognized as ObjCMethodExpr (as '1 + number of commas' formula is 522 // used for other expression types). Parameter counter has to be, 523 // therefore, reset to 0. 524 Left->ParameterCount = 0; 525 Contexts.back().ColonIsObjCMethodExpr = true; 526 if (Parent && Parent->is(tok::r_paren)) 527 Parent->Type = TT_CastRParen; 528 } 529 ColonFound = true; 530 } 531 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) && 532 !ColonFound) 533 Left->Type = TT_ArrayInitializerLSquare; 534 FormatToken *Tok = CurrentToken; 535 if (!consumeToken()) 536 return false; 537 updateParameterCount(Left, Tok); 538 } 539 return false; 540 } 541 542 bool parseBrace() { 543 if (CurrentToken) { 544 FormatToken *Left = CurrentToken->Previous; 545 Left->ParentBracket = Contexts.back().ContextKind; 546 547 if (Contexts.back().CaretFound) 548 Left->Type = TT_ObjCBlockLBrace; 549 Contexts.back().CaretFound = false; 550 551 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1); 552 Contexts.back().ColonIsDictLiteral = true; 553 if (Left->BlockKind == BK_BracedInit) 554 Contexts.back().IsExpression = true; 555 if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 556 Left->Previous->is(TT_JsTypeColon)) 557 Contexts.back().IsExpression = false; 558 559 while (CurrentToken) { 560 if (CurrentToken->is(tok::r_brace)) { 561 Left->MatchingParen = CurrentToken; 562 CurrentToken->MatchingParen = Left; 563 next(); 564 return true; 565 } 566 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 567 return false; 568 updateParameterCount(Left, CurrentToken); 569 if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) { 570 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 571 if (Previous->is(TT_JsTypeOptionalQuestion)) 572 Previous = Previous->getPreviousNonComment(); 573 if ((CurrentToken->is(tok::colon) && 574 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) || 575 Style.Language == FormatStyle::LK_Proto || 576 Style.Language == FormatStyle::LK_TextProto) { 577 Left->Type = TT_DictLiteral; 578 if (Previous->Tok.getIdentifierInfo() || 579 Previous->is(tok::string_literal)) 580 Previous->Type = TT_SelectorName; 581 } 582 if (CurrentToken->is(tok::colon) || 583 Style.Language == FormatStyle::LK_JavaScript) 584 Left->Type = TT_DictLiteral; 585 } 586 if (CurrentToken->is(tok::comma) && 587 Style.Language == FormatStyle::LK_JavaScript) 588 Left->Type = TT_DictLiteral; 589 if (!consumeToken()) 590 return false; 591 } 592 } 593 return true; 594 } 595 596 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 597 if (Current->is(tok::l_brace) && Current->BlockKind == BK_Block) 598 ++Left->BlockParameterCount; 599 if (Left->Type == TT_ObjCMethodExpr) { 600 if (Current->is(tok::colon)) 601 ++Left->ParameterCount; 602 } else if (Current->is(tok::comma)) { 603 ++Left->ParameterCount; 604 if (!Left->Role) 605 Left->Role.reset(new CommaSeparatedList(Style)); 606 Left->Role->CommaFound(Current); 607 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 608 Left->ParameterCount = 1; 609 } 610 } 611 612 bool parseConditional() { 613 while (CurrentToken) { 614 if (CurrentToken->is(tok::colon)) { 615 CurrentToken->Type = TT_ConditionalExpr; 616 next(); 617 return true; 618 } 619 if (!consumeToken()) 620 return false; 621 } 622 return false; 623 } 624 625 bool parseTemplateDeclaration() { 626 if (CurrentToken && CurrentToken->is(tok::less)) { 627 CurrentToken->Type = TT_TemplateOpener; 628 next(); 629 if (!parseAngle()) 630 return false; 631 if (CurrentToken) 632 CurrentToken->Previous->ClosesTemplateDeclaration = true; 633 return true; 634 } 635 return false; 636 } 637 638 bool consumeToken() { 639 FormatToken *Tok = CurrentToken; 640 next(); 641 switch (Tok->Tok.getKind()) { 642 case tok::plus: 643 case tok::minus: 644 if (!Tok->Previous && Line.MustBeDeclaration) 645 Tok->Type = TT_ObjCMethodSpecifier; 646 break; 647 case tok::colon: 648 if (!Tok->Previous) 649 return false; 650 // Colons from ?: are handled in parseConditional(). 651 if (Style.Language == FormatStyle::LK_JavaScript) { 652 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop 653 (Contexts.size() == 1 && // switch/case labels 654 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) || 655 Contexts.back().ContextKind == tok::l_paren || // function params 656 Contexts.back().ContextKind == tok::l_square || // array type 657 (!Contexts.back().IsExpression && 658 Contexts.back().ContextKind == tok::l_brace) || // object type 659 (Contexts.size() == 1 && 660 Line.MustBeDeclaration)) { // method/property declaration 661 Contexts.back().IsExpression = false; 662 Tok->Type = TT_JsTypeColon; 663 break; 664 } 665 } 666 if (Contexts.back().ColonIsDictLiteral || 667 Style.Language == FormatStyle::LK_Proto || 668 Style.Language == FormatStyle::LK_TextProto) { 669 Tok->Type = TT_DictLiteral; 670 if (Style.Language == FormatStyle::LK_TextProto) { 671 if (FormatToken *Previous = Tok->getPreviousNonComment()) 672 Previous->Type = TT_SelectorName; 673 } 674 } else if (Contexts.back().ColonIsObjCMethodExpr || 675 Line.startsWith(TT_ObjCMethodSpecifier)) { 676 Tok->Type = TT_ObjCMethodExpr; 677 const FormatToken *BeforePrevious = Tok->Previous->Previous; 678 if (!BeforePrevious || 679 !(BeforePrevious->is(TT_CastRParen) || 680 (BeforePrevious->is(TT_ObjCMethodExpr) && 681 BeforePrevious->is(tok::colon))) || 682 BeforePrevious->is(tok::r_square) || 683 Contexts.back().LongestObjCSelectorName == 0) { 684 Tok->Previous->Type = TT_SelectorName; 685 if (!Contexts.back().FirstObjCSelectorName) 686 Contexts.back().FirstObjCSelectorName = Tok->Previous; 687 else if (Tok->Previous->ColumnWidth > 688 Contexts.back().LongestObjCSelectorName) 689 Contexts.back().LongestObjCSelectorName = 690 Tok->Previous->ColumnWidth; 691 } 692 } else if (Contexts.back().ColonIsForRangeExpr) { 693 Tok->Type = TT_RangeBasedForLoopColon; 694 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 695 Tok->Type = TT_BitFieldColon; 696 } else if (Contexts.size() == 1 && 697 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) { 698 if (Tok->getPreviousNonComment()->isOneOf(tok::r_paren, 699 tok::kw_noexcept)) 700 Tok->Type = TT_CtorInitializerColon; 701 else 702 Tok->Type = TT_InheritanceColon; 703 } else if (Tok->Previous->is(tok::identifier) && Tok->Next && 704 Tok->Next->isOneOf(tok::r_paren, tok::comma)) { 705 // This handles a special macro in ObjC code where selectors including 706 // the colon are passed as macro arguments. 707 Tok->Type = TT_ObjCMethodExpr; 708 } else if (Contexts.back().ContextKind == tok::l_paren) { 709 Tok->Type = TT_InlineASMColon; 710 } 711 break; 712 case tok::pipe: 713 case tok::amp: 714 // | and & in declarations/type expressions represent union and 715 // intersection types, respectively. 716 if (Style.Language == FormatStyle::LK_JavaScript && 717 !Contexts.back().IsExpression) 718 Tok->Type = TT_JsTypeOperator; 719 break; 720 case tok::kw_if: 721 case tok::kw_while: 722 if (Tok->is(tok::kw_if) && CurrentToken && 723 CurrentToken->is(tok::kw_constexpr)) 724 next(); 725 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 726 next(); 727 if (!parseParens(/*LookForDecls=*/true)) 728 return false; 729 } 730 break; 731 case tok::kw_for: 732 if (Style.Language == FormatStyle::LK_JavaScript) { 733 // x.for and {for: ...} 734 if ((Tok->Previous && Tok->Previous->is(tok::period)) || 735 (Tok->Next && Tok->Next->is(tok::colon))) 736 break; 737 // JS' for await ( ... 738 if (CurrentToken && CurrentToken->is(Keywords.kw_await)) 739 next(); 740 } 741 Contexts.back().ColonIsForRangeExpr = true; 742 next(); 743 if (!parseParens()) 744 return false; 745 break; 746 case tok::l_paren: 747 // When faced with 'operator()()', the kw_operator handler incorrectly 748 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make 749 // the first two parens OverloadedOperators and the second l_paren an 750 // OverloadedOperatorLParen. 751 if (Tok->Previous && Tok->Previous->is(tok::r_paren) && 752 Tok->Previous->MatchingParen && 753 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) { 754 Tok->Previous->Type = TT_OverloadedOperator; 755 Tok->Previous->MatchingParen->Type = TT_OverloadedOperator; 756 Tok->Type = TT_OverloadedOperatorLParen; 757 } 758 759 if (!parseParens()) 760 return false; 761 if (Line.MustBeDeclaration && Contexts.size() == 1 && 762 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) && 763 (!Tok->Previous || 764 !Tok->Previous->isOneOf(tok::kw_decltype, tok::kw___attribute, 765 TT_LeadingJavaAnnotation))) 766 Line.MightBeFunctionDecl = true; 767 break; 768 case tok::l_square: 769 if (!parseSquare()) 770 return false; 771 break; 772 case tok::l_brace: 773 if (Style.Language == FormatStyle::LK_TextProto) { 774 FormatToken *Previous = Tok->getPreviousNonComment(); 775 if (Previous && Previous->Type != TT_DictLiteral) 776 Previous->Type = TT_SelectorName; 777 } 778 if (!parseBrace()) 779 return false; 780 break; 781 case tok::less: 782 if (parseAngle()) { 783 Tok->Type = TT_TemplateOpener; 784 // In TT_Proto, we must distignuish between: 785 // map<key, value> 786 // msg < item: data > 787 // msg: < item: data > 788 // In TT_TextProto, map<key, value> does not occur. 789 if (Style.Language == FormatStyle::LK_TextProto || 790 (Style.Language == FormatStyle::LK_Proto && Tok->Previous && 791 Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) { 792 Tok->Type = TT_DictLiteral; 793 FormatToken *Previous = Tok->getPreviousNonComment(); 794 if (Previous && Previous->Type != TT_DictLiteral) 795 Previous->Type = TT_SelectorName; 796 } 797 } else { 798 Tok->Type = TT_BinaryOperator; 799 NonTemplateLess.insert(Tok); 800 CurrentToken = Tok; 801 next(); 802 } 803 break; 804 case tok::r_paren: 805 case tok::r_square: 806 return false; 807 case tok::r_brace: 808 // Lines can start with '}'. 809 if (Tok->Previous) 810 return false; 811 break; 812 case tok::greater: 813 if (Style.Language != FormatStyle::LK_TextProto) 814 Tok->Type = TT_BinaryOperator; 815 break; 816 case tok::kw_operator: 817 if (Style.Language == FormatStyle::LK_TextProto || 818 Style.Language == FormatStyle::LK_Proto) 819 break; 820 while (CurrentToken && 821 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 822 if (CurrentToken->isOneOf(tok::star, tok::amp)) 823 CurrentToken->Type = TT_PointerOrReference; 824 consumeToken(); 825 if (CurrentToken && 826 CurrentToken->Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator, 827 tok::comma)) 828 CurrentToken->Previous->Type = TT_OverloadedOperator; 829 } 830 if (CurrentToken) { 831 CurrentToken->Type = TT_OverloadedOperatorLParen; 832 if (CurrentToken->Previous->is(TT_BinaryOperator)) 833 CurrentToken->Previous->Type = TT_OverloadedOperator; 834 } 835 break; 836 case tok::question: 837 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next && 838 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren, 839 tok::r_brace)) { 840 // Question marks before semicolons, colons, etc. indicate optional 841 // types (fields, parameters), e.g. 842 // function(x?: string, y?) {...} 843 // class X { y?; } 844 Tok->Type = TT_JsTypeOptionalQuestion; 845 break; 846 } 847 // Declarations cannot be conditional expressions, this can only be part 848 // of a type declaration. 849 if (Line.MustBeDeclaration && !Contexts.back().IsExpression && 850 Style.Language == FormatStyle::LK_JavaScript) 851 break; 852 parseConditional(); 853 break; 854 case tok::kw_template: 855 parseTemplateDeclaration(); 856 break; 857 case tok::comma: 858 if (Contexts.back().InCtorInitializer) 859 Tok->Type = TT_CtorInitializerComma; 860 else if (Contexts.back().InInheritanceList) 861 Tok->Type = TT_InheritanceComma; 862 else if (Contexts.back().FirstStartOfName && 863 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) { 864 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 865 Line.IsMultiVariableDeclStmt = true; 866 } 867 if (Contexts.back().IsForEachMacro) 868 Contexts.back().IsExpression = true; 869 break; 870 case tok::identifier: 871 if (Tok->isOneOf(Keywords.kw___has_include, 872 Keywords.kw___has_include_next)) { 873 parseHasInclude(); 874 } 875 break; 876 default: 877 break; 878 } 879 return true; 880 } 881 882 void parseIncludeDirective() { 883 if (CurrentToken && CurrentToken->is(tok::less)) { 884 next(); 885 while (CurrentToken) { 886 // Mark tokens up to the trailing line comments as implicit string 887 // literals. 888 if (CurrentToken->isNot(tok::comment) && 889 !CurrentToken->TokenText.startswith("//")) 890 CurrentToken->Type = TT_ImplicitStringLiteral; 891 next(); 892 } 893 } 894 } 895 896 void parseWarningOrError() { 897 next(); 898 // We still want to format the whitespace left of the first token of the 899 // warning or error. 900 next(); 901 while (CurrentToken) { 902 CurrentToken->Type = TT_ImplicitStringLiteral; 903 next(); 904 } 905 } 906 907 void parsePragma() { 908 next(); // Consume "pragma". 909 if (CurrentToken && 910 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 911 bool IsMark = CurrentToken->is(Keywords.kw_mark); 912 next(); // Consume "mark". 913 next(); // Consume first token (so we fix leading whitespace). 914 while (CurrentToken) { 915 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 916 CurrentToken->Type = TT_ImplicitStringLiteral; 917 next(); 918 } 919 } 920 } 921 922 void parseHasInclude() { 923 if (!CurrentToken || !CurrentToken->is(tok::l_paren)) 924 return; 925 next(); // '(' 926 parseIncludeDirective(); 927 next(); // ')' 928 } 929 930 LineType parsePreprocessorDirective() { 931 bool IsFirstToken = CurrentToken->IsFirst; 932 LineType Type = LT_PreprocessorDirective; 933 next(); 934 if (!CurrentToken) 935 return Type; 936 937 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) { 938 // JavaScript files can contain shebang lines of the form: 939 // #!/usr/bin/env node 940 // Treat these like C++ #include directives. 941 while (CurrentToken) { 942 // Tokens cannot be comments here. 943 CurrentToken->Type = TT_ImplicitStringLiteral; 944 next(); 945 } 946 return LT_ImportStatement; 947 } 948 949 if (CurrentToken->Tok.is(tok::numeric_constant)) { 950 CurrentToken->SpacesRequiredBefore = 1; 951 return Type; 952 } 953 // Hashes in the middle of a line can lead to any strange token 954 // sequence. 955 if (!CurrentToken->Tok.getIdentifierInfo()) 956 return Type; 957 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 958 case tok::pp_include: 959 case tok::pp_include_next: 960 case tok::pp_import: 961 next(); 962 parseIncludeDirective(); 963 Type = LT_ImportStatement; 964 break; 965 case tok::pp_error: 966 case tok::pp_warning: 967 parseWarningOrError(); 968 break; 969 case tok::pp_pragma: 970 parsePragma(); 971 break; 972 case tok::pp_if: 973 case tok::pp_elif: 974 Contexts.back().IsExpression = true; 975 parseLine(); 976 break; 977 default: 978 break; 979 } 980 while (CurrentToken) { 981 FormatToken *Tok = CurrentToken; 982 next(); 983 if (Tok->is(tok::l_paren)) 984 parseParens(); 985 else if (Tok->isOneOf(Keywords.kw___has_include, 986 Keywords.kw___has_include_next)) 987 parseHasInclude(); 988 } 989 return Type; 990 } 991 992 public: 993 LineType parseLine() { 994 NonTemplateLess.clear(); 995 if (CurrentToken->is(tok::hash)) 996 return parsePreprocessorDirective(); 997 998 // Directly allow to 'import <string-literal>' to support protocol buffer 999 // definitions (github.com/google/protobuf) or missing "#" (either way we 1000 // should not break the line). 1001 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 1002 if ((Style.Language == FormatStyle::LK_Java && 1003 CurrentToken->is(Keywords.kw_package)) || 1004 (Info && Info->getPPKeywordID() == tok::pp_import && 1005 CurrentToken->Next && 1006 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 1007 tok::kw_static))) { 1008 next(); 1009 parseIncludeDirective(); 1010 return LT_ImportStatement; 1011 } 1012 1013 // If this line starts and ends in '<' and '>', respectively, it is likely 1014 // part of "#define <a/b.h>". 1015 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 1016 parseIncludeDirective(); 1017 return LT_ImportStatement; 1018 } 1019 1020 // In .proto files, top-level options are very similar to import statements 1021 // and should not be line-wrapped. 1022 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 1023 CurrentToken->is(Keywords.kw_option)) { 1024 next(); 1025 if (CurrentToken && CurrentToken->is(tok::identifier)) 1026 return LT_ImportStatement; 1027 } 1028 1029 bool KeywordVirtualFound = false; 1030 bool ImportStatement = false; 1031 1032 // import {...} from '...'; 1033 if (Style.Language == FormatStyle::LK_JavaScript && 1034 CurrentToken->is(Keywords.kw_import)) 1035 ImportStatement = true; 1036 1037 while (CurrentToken) { 1038 if (CurrentToken->is(tok::kw_virtual)) 1039 KeywordVirtualFound = true; 1040 if (Style.Language == FormatStyle::LK_JavaScript) { 1041 // export {...} from '...'; 1042 // An export followed by "from 'some string';" is a re-export from 1043 // another module identified by a URI and is treated as a 1044 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 1045 // Just "export {...};" or "export class ..." should not be treated as 1046 // an import in this sense. 1047 if (Line.First->is(tok::kw_export) && 1048 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 1049 CurrentToken->Next->isStringLiteral()) 1050 ImportStatement = true; 1051 if (isClosureImportStatement(*CurrentToken)) 1052 ImportStatement = true; 1053 } 1054 if (!consumeToken()) 1055 return LT_Invalid; 1056 } 1057 if (KeywordVirtualFound) 1058 return LT_VirtualFunctionDecl; 1059 if (ImportStatement) 1060 return LT_ImportStatement; 1061 1062 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 1063 if (Contexts.back().FirstObjCSelectorName) 1064 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 1065 Contexts.back().LongestObjCSelectorName; 1066 return LT_ObjCMethodDecl; 1067 } 1068 1069 return LT_Other; 1070 } 1071 1072 private: 1073 bool isClosureImportStatement(const FormatToken &Tok) { 1074 // FIXME: Closure-library specific stuff should not be hard-coded but be 1075 // configurable. 1076 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 1077 Tok.Next->Next && 1078 (Tok.Next->Next->TokenText == "module" || 1079 Tok.Next->Next->TokenText == "provide" || 1080 Tok.Next->Next->TokenText == "require" || 1081 Tok.Next->Next->TokenText == "forwardDeclare") && 1082 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 1083 } 1084 1085 void resetTokenMetadata(FormatToken *Token) { 1086 if (!Token) 1087 return; 1088 1089 // Reset token type in case we have already looked at it and then 1090 // recovered from an error (e.g. failure to find the matching >). 1091 if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_ForEachMacro, 1092 TT_FunctionLBrace, TT_ImplicitStringLiteral, 1093 TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow, 1094 TT_OverloadedOperator, TT_RegexLiteral, 1095 TT_TemplateString, TT_ObjCStringLiteral)) 1096 CurrentToken->Type = TT_Unknown; 1097 CurrentToken->Role.reset(); 1098 CurrentToken->MatchingParen = nullptr; 1099 CurrentToken->FakeLParens.clear(); 1100 CurrentToken->FakeRParens = 0; 1101 } 1102 1103 void next() { 1104 if (CurrentToken) { 1105 CurrentToken->NestingLevel = Contexts.size() - 1; 1106 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 1107 modifyContext(*CurrentToken); 1108 determineTokenType(*CurrentToken); 1109 CurrentToken = CurrentToken->Next; 1110 } 1111 1112 resetTokenMetadata(CurrentToken); 1113 } 1114 1115 /// \brief A struct to hold information valid in a specific context, e.g. 1116 /// a pair of parenthesis. 1117 struct Context { 1118 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 1119 bool IsExpression) 1120 : ContextKind(ContextKind), BindingStrength(BindingStrength), 1121 IsExpression(IsExpression) {} 1122 1123 tok::TokenKind ContextKind; 1124 unsigned BindingStrength; 1125 bool IsExpression; 1126 unsigned LongestObjCSelectorName = 0; 1127 bool ColonIsForRangeExpr = false; 1128 bool ColonIsDictLiteral = false; 1129 bool ColonIsObjCMethodExpr = false; 1130 FormatToken *FirstObjCSelectorName = nullptr; 1131 FormatToken *FirstStartOfName = nullptr; 1132 bool CanBeExpression = true; 1133 bool InTemplateArgument = false; 1134 bool InCtorInitializer = false; 1135 bool InInheritanceList = false; 1136 bool CaretFound = false; 1137 bool IsForEachMacro = false; 1138 bool InCpp11AttributeSpecifier = false; 1139 }; 1140 1141 /// \brief Puts a new \c Context onto the stack \c Contexts for the lifetime 1142 /// of each instance. 1143 struct ScopedContextCreator { 1144 AnnotatingParser &P; 1145 1146 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 1147 unsigned Increase) 1148 : P(P) { 1149 P.Contexts.push_back(Context(ContextKind, 1150 P.Contexts.back().BindingStrength + Increase, 1151 P.Contexts.back().IsExpression)); 1152 } 1153 1154 ~ScopedContextCreator() { P.Contexts.pop_back(); } 1155 }; 1156 1157 void modifyContext(const FormatToken &Current) { 1158 if (Current.getPrecedence() == prec::Assignment && 1159 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 1160 // Type aliases use `type X = ...;` in TypeScript and can be exported 1161 // using `export type ...`. 1162 !(Style.Language == FormatStyle::LK_JavaScript && 1163 (Line.startsWith(Keywords.kw_type, tok::identifier) || 1164 Line.startsWith(tok::kw_export, Keywords.kw_type, 1165 tok::identifier))) && 1166 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 1167 Contexts.back().IsExpression = true; 1168 if (!Line.startsWith(TT_UnaryOperator)) { 1169 for (FormatToken *Previous = Current.Previous; 1170 Previous && Previous->Previous && 1171 !Previous->Previous->isOneOf(tok::comma, tok::semi); 1172 Previous = Previous->Previous) { 1173 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 1174 Previous = Previous->MatchingParen; 1175 if (!Previous) 1176 break; 1177 } 1178 if (Previous->opensScope()) 1179 break; 1180 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 1181 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 1182 Previous->Previous && Previous->Previous->isNot(tok::equal)) 1183 Previous->Type = TT_PointerOrReference; 1184 } 1185 } 1186 } else if (Current.is(tok::lessless) && 1187 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 1188 Contexts.back().IsExpression = true; 1189 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 1190 Contexts.back().IsExpression = true; 1191 } else if (Current.is(TT_TrailingReturnArrow)) { 1192 Contexts.back().IsExpression = false; 1193 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 1194 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 1195 } else if (Current.Previous && 1196 Current.Previous->is(TT_CtorInitializerColon)) { 1197 Contexts.back().IsExpression = true; 1198 Contexts.back().InCtorInitializer = true; 1199 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) { 1200 Contexts.back().InInheritanceList = true; 1201 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 1202 for (FormatToken *Previous = Current.Previous; 1203 Previous && Previous->isOneOf(tok::star, tok::amp); 1204 Previous = Previous->Previous) 1205 Previous->Type = TT_PointerOrReference; 1206 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 1207 Contexts.back().IsExpression = false; 1208 } else if (Current.is(tok::kw_new)) { 1209 Contexts.back().CanBeExpression = false; 1210 } else if (Current.isOneOf(tok::semi, tok::exclaim)) { 1211 // This should be the condition or increment in a for-loop. 1212 Contexts.back().IsExpression = true; 1213 } 1214 } 1215 1216 void determineTokenType(FormatToken &Current) { 1217 if (!Current.is(TT_Unknown)) 1218 // The token type is already known. 1219 return; 1220 1221 if (Style.Language == FormatStyle::LK_JavaScript) { 1222 if (Current.is(tok::exclaim)) { 1223 if (Current.Previous && 1224 (Current.Previous->isOneOf(tok::identifier, tok::kw_namespace, 1225 tok::r_paren, tok::r_square, 1226 tok::r_brace) || 1227 Current.Previous->Tok.isLiteral())) { 1228 Current.Type = TT_JsNonNullAssertion; 1229 return; 1230 } 1231 if (Current.Next && 1232 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) { 1233 Current.Type = TT_JsNonNullAssertion; 1234 return; 1235 } 1236 } 1237 } 1238 1239 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1240 // function declaration have been found. In this case, 'Current' is a 1241 // trailing token of this declaration and thus cannot be a name. 1242 if (Current.is(Keywords.kw_instanceof)) { 1243 Current.Type = TT_BinaryOperator; 1244 } else if (isStartOfName(Current) && 1245 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 1246 Contexts.back().FirstStartOfName = &Current; 1247 Current.Type = TT_StartOfName; 1248 } else if (Current.is(tok::semi)) { 1249 // Reset FirstStartOfName after finding a semicolon so that a for loop 1250 // with multiple increment statements is not confused with a for loop 1251 // having multiple variable declarations. 1252 Contexts.back().FirstStartOfName = nullptr; 1253 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 1254 AutoFound = true; 1255 } else if (Current.is(tok::arrow) && 1256 Style.Language == FormatStyle::LK_Java) { 1257 Current.Type = TT_LambdaArrow; 1258 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 1259 Current.NestingLevel == 0) { 1260 Current.Type = TT_TrailingReturnArrow; 1261 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 1262 Current.Type = determineStarAmpUsage(Current, 1263 Contexts.back().CanBeExpression && 1264 Contexts.back().IsExpression, 1265 Contexts.back().InTemplateArgument); 1266 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 1267 Current.Type = determinePlusMinusCaretUsage(Current); 1268 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 1269 Contexts.back().CaretFound = true; 1270 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 1271 Current.Type = determineIncrementUsage(Current); 1272 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 1273 Current.Type = TT_UnaryOperator; 1274 } else if (Current.is(tok::question)) { 1275 if (Style.Language == FormatStyle::LK_JavaScript && 1276 Line.MustBeDeclaration && !Contexts.back().IsExpression) { 1277 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 1278 // on the interface, not a ternary expression. 1279 Current.Type = TT_JsTypeOptionalQuestion; 1280 } else { 1281 Current.Type = TT_ConditionalExpr; 1282 } 1283 } else if (Current.isBinaryOperator() && 1284 (!Current.Previous || Current.Previous->isNot(tok::l_square)) && 1285 (!Current.is(tok::greater) && 1286 Style.Language != FormatStyle::LK_TextProto)) { 1287 Current.Type = TT_BinaryOperator; 1288 } else if (Current.is(tok::comment)) { 1289 if (Current.TokenText.startswith("/*")) { 1290 if (Current.TokenText.endswith("*/")) 1291 Current.Type = TT_BlockComment; 1292 else 1293 // The lexer has for some reason determined a comment here. But we 1294 // cannot really handle it, if it isn't properly terminated. 1295 Current.Tok.setKind(tok::unknown); 1296 } else { 1297 Current.Type = TT_LineComment; 1298 } 1299 } else if (Current.is(tok::r_paren)) { 1300 if (rParenEndsCast(Current)) 1301 Current.Type = TT_CastRParen; 1302 if (Current.MatchingParen && Current.Next && 1303 !Current.Next->isBinaryOperator() && 1304 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1305 tok::comma, tok::period, tok::arrow, 1306 tok::coloncolon)) 1307 if (FormatToken *AfterParen = Current.MatchingParen->Next) { 1308 // Make sure this isn't the return type of an Obj-C block declaration 1309 if (AfterParen->Tok.isNot(tok::caret)) { 1310 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1311 if (BeforeParen->is(tok::identifier) && 1312 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1313 (!BeforeParen->Previous || 1314 BeforeParen->Previous->ClosesTemplateDeclaration)) 1315 Current.Type = TT_FunctionAnnotationRParen; 1316 } 1317 } 1318 } else if (Current.is(tok::at) && Current.Next && 1319 Style.Language != FormatStyle::LK_JavaScript && 1320 Style.Language != FormatStyle::LK_Java) { 1321 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it 1322 // marks declarations and properties that need special formatting. 1323 switch (Current.Next->Tok.getObjCKeywordID()) { 1324 case tok::objc_interface: 1325 case tok::objc_implementation: 1326 case tok::objc_protocol: 1327 Current.Type = TT_ObjCDecl; 1328 break; 1329 case tok::objc_property: 1330 Current.Type = TT_ObjCProperty; 1331 break; 1332 default: 1333 break; 1334 } 1335 } else if (Current.is(tok::period)) { 1336 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1337 if (PreviousNoComment && 1338 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1339 Current.Type = TT_DesignatedInitializerPeriod; 1340 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1341 Current.Previous->isOneOf(TT_JavaAnnotation, 1342 TT_LeadingJavaAnnotation)) { 1343 Current.Type = Current.Previous->Type; 1344 } 1345 } else if (Current.isOneOf(tok::identifier, tok::kw_const) && 1346 Current.Previous && 1347 !Current.Previous->isOneOf(tok::equal, tok::at) && 1348 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1349 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1350 // function declaration have been found. 1351 Current.Type = TT_TrailingAnnotation; 1352 } else if ((Style.Language == FormatStyle::LK_Java || 1353 Style.Language == FormatStyle::LK_JavaScript) && 1354 Current.Previous) { 1355 if (Current.Previous->is(tok::at) && 1356 Current.isNot(Keywords.kw_interface)) { 1357 const FormatToken &AtToken = *Current.Previous; 1358 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1359 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1360 Current.Type = TT_LeadingJavaAnnotation; 1361 else 1362 Current.Type = TT_JavaAnnotation; 1363 } else if (Current.Previous->is(tok::period) && 1364 Current.Previous->isOneOf(TT_JavaAnnotation, 1365 TT_LeadingJavaAnnotation)) { 1366 Current.Type = Current.Previous->Type; 1367 } 1368 } 1369 } 1370 1371 /// \brief Take a guess at whether \p Tok starts a name of a function or 1372 /// variable declaration. 1373 /// 1374 /// This is a heuristic based on whether \p Tok is an identifier following 1375 /// something that is likely a type. 1376 bool isStartOfName(const FormatToken &Tok) { 1377 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1378 return false; 1379 1380 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof, 1381 Keywords.kw_as)) 1382 return false; 1383 if (Style.Language == FormatStyle::LK_JavaScript && 1384 Tok.Previous->is(Keywords.kw_in)) 1385 return false; 1386 1387 // Skip "const" as it does not have an influence on whether this is a name. 1388 FormatToken *PreviousNotConst = Tok.getPreviousNonComment(); 1389 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1390 PreviousNotConst = PreviousNotConst->getPreviousNonComment(); 1391 1392 if (!PreviousNotConst) 1393 return false; 1394 1395 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1396 PreviousNotConst->Previous && 1397 PreviousNotConst->Previous->is(tok::hash); 1398 1399 if (PreviousNotConst->is(TT_TemplateCloser)) 1400 return PreviousNotConst && PreviousNotConst->MatchingParen && 1401 PreviousNotConst->MatchingParen->Previous && 1402 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1403 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1404 1405 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen && 1406 PreviousNotConst->MatchingParen->Previous && 1407 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype)) 1408 return true; 1409 1410 return (!IsPPKeyword && 1411 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) || 1412 PreviousNotConst->is(TT_PointerOrReference) || 1413 PreviousNotConst->isSimpleTypeSpecifier(); 1414 } 1415 1416 /// \brief Determine whether ')' is ending a cast. 1417 bool rParenEndsCast(const FormatToken &Tok) { 1418 // C-style casts are only used in C++ and Java. 1419 if (!Style.isCpp() && Style.Language != FormatStyle::LK_Java) 1420 return false; 1421 1422 // Empty parens aren't casts and there are no casts at the end of the line. 1423 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1424 return false; 1425 1426 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1427 if (LeftOfParens) { 1428 // If there is a closing parenthesis left of the current parentheses, 1429 // look past it as these might be chained casts. 1430 if (LeftOfParens->is(tok::r_paren)) { 1431 if (!LeftOfParens->MatchingParen || 1432 !LeftOfParens->MatchingParen->Previous) 1433 return false; 1434 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1435 } 1436 1437 // If there is an identifier (or with a few exceptions a keyword) right 1438 // before the parentheses, this is unlikely to be a cast. 1439 if (LeftOfParens->Tok.getIdentifierInfo() && 1440 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1441 tok::kw_delete)) 1442 return false; 1443 1444 // Certain other tokens right before the parentheses are also signals that 1445 // this cannot be a cast. 1446 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1447 TT_TemplateCloser, tok::ellipsis)) 1448 return false; 1449 } 1450 1451 if (Tok.Next->is(tok::question)) 1452 return false; 1453 1454 // As Java has no function types, a "(" after the ")" likely means that this 1455 // is a cast. 1456 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1457 return true; 1458 1459 // If a (non-string) literal follows, this is likely a cast. 1460 if (Tok.Next->isNot(tok::string_literal) && 1461 (Tok.Next->Tok.isLiteral() || 1462 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1463 return true; 1464 1465 // Heuristically try to determine whether the parentheses contain a type. 1466 bool ParensAreType = 1467 !Tok.Previous || 1468 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) || 1469 Tok.Previous->isSimpleTypeSpecifier(); 1470 bool ParensCouldEndDecl = 1471 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1472 if (ParensAreType && !ParensCouldEndDecl) 1473 return true; 1474 1475 // At this point, we heuristically assume that there are no casts at the 1476 // start of the line. We assume that we have found most cases where there 1477 // are by the logic above, e.g. "(void)x;". 1478 if (!LeftOfParens) 1479 return false; 1480 1481 // Certain token types inside the parentheses mean that this can't be a 1482 // cast. 1483 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok; 1484 Token = Token->Next) 1485 if (Token->is(TT_BinaryOperator)) 1486 return false; 1487 1488 // If the following token is an identifier or 'this', this is a cast. All 1489 // cases where this can be something else are handled above. 1490 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1491 return true; 1492 1493 if (!Tok.Next->Next) 1494 return false; 1495 1496 // If the next token after the parenthesis is a unary operator, assume 1497 // that this is cast, unless there are unexpected tokens inside the 1498 // parenthesis. 1499 bool NextIsUnary = 1500 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 1501 if (!NextIsUnary || Tok.Next->is(tok::plus) || 1502 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 1503 return false; 1504 // Search for unexpected tokens. 1505 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 1506 Prev = Prev->Previous) { 1507 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 1508 return false; 1509 } 1510 return true; 1511 } 1512 1513 /// \brief Return the type of the given token assuming it is * or &. 1514 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 1515 bool InTemplateArgument) { 1516 if (Style.Language == FormatStyle::LK_JavaScript) 1517 return TT_BinaryOperator; 1518 1519 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1520 if (!PrevToken) 1521 return TT_UnaryOperator; 1522 1523 const FormatToken *NextToken = Tok.getNextNonComment(); 1524 if (!NextToken || 1525 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_const) || 1526 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 1527 return TT_PointerOrReference; 1528 1529 if (PrevToken->is(tok::coloncolon)) 1530 return TT_PointerOrReference; 1531 1532 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 1533 tok::comma, tok::semi, tok::kw_return, tok::colon, 1534 tok::equal, tok::kw_delete, tok::kw_sizeof, 1535 tok::kw_throw) || 1536 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 1537 TT_UnaryOperator, TT_CastRParen)) 1538 return TT_UnaryOperator; 1539 1540 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 1541 return TT_PointerOrReference; 1542 if (NextToken->is(tok::kw_operator) && !IsExpression) 1543 return TT_PointerOrReference; 1544 if (NextToken->isOneOf(tok::comma, tok::semi)) 1545 return TT_PointerOrReference; 1546 1547 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen) { 1548 FormatToken *TokenBeforeMatchingParen = 1549 PrevToken->MatchingParen->getPreviousNonComment(); 1550 if (TokenBeforeMatchingParen && 1551 TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype)) 1552 return TT_PointerOrReference; 1553 } 1554 1555 if (PrevToken->Tok.isLiteral() || 1556 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 1557 tok::kw_false, tok::r_brace) || 1558 NextToken->Tok.isLiteral() || 1559 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 1560 NextToken->isUnaryOperator() || 1561 // If we know we're in a template argument, there are no named 1562 // declarations. Thus, having an identifier on the right-hand side 1563 // indicates a binary operator. 1564 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 1565 return TT_BinaryOperator; 1566 1567 // "&&(" is quite unlikely to be two successive unary "&". 1568 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren)) 1569 return TT_BinaryOperator; 1570 1571 // This catches some cases where evaluation order is used as control flow: 1572 // aaa && aaa->f(); 1573 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 1574 if (NextNextToken && NextNextToken->is(tok::arrow)) 1575 return TT_BinaryOperator; 1576 1577 // It is very unlikely that we are going to find a pointer or reference type 1578 // definition on the RHS of an assignment. 1579 if (IsExpression && !Contexts.back().CaretFound) 1580 return TT_BinaryOperator; 1581 1582 return TT_PointerOrReference; 1583 } 1584 1585 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 1586 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1587 if (!PrevToken) 1588 return TT_UnaryOperator; 1589 1590 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator)) 1591 // This must be a sequence of leading unary operators. 1592 return TT_UnaryOperator; 1593 1594 // Use heuristics to recognize unary operators. 1595 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 1596 tok::question, tok::colon, tok::kw_return, 1597 tok::kw_case, tok::at, tok::l_brace)) 1598 return TT_UnaryOperator; 1599 1600 // There can't be two consecutive binary operators. 1601 if (PrevToken->is(TT_BinaryOperator)) 1602 return TT_UnaryOperator; 1603 1604 // Fall back to marking the token as binary operator. 1605 return TT_BinaryOperator; 1606 } 1607 1608 /// \brief Determine whether ++/-- are pre- or post-increments/-decrements. 1609 TokenType determineIncrementUsage(const FormatToken &Tok) { 1610 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1611 if (!PrevToken || PrevToken->is(TT_CastRParen)) 1612 return TT_UnaryOperator; 1613 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 1614 return TT_TrailingUnaryOperator; 1615 1616 return TT_UnaryOperator; 1617 } 1618 1619 SmallVector<Context, 8> Contexts; 1620 1621 const FormatStyle &Style; 1622 AnnotatedLine &Line; 1623 FormatToken *CurrentToken; 1624 bool AutoFound; 1625 const AdditionalKeywords &Keywords; 1626 1627 // Set of "<" tokens that do not open a template parameter list. If parseAngle 1628 // determines that a specific token can't be a template opener, it will make 1629 // same decision irrespective of the decisions for tokens leading up to it. 1630 // Store this information to prevent this from causing exponential runtime. 1631 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 1632 }; 1633 1634 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 1635 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 1636 1637 /// \brief Parses binary expressions by inserting fake parenthesis based on 1638 /// operator precedence. 1639 class ExpressionParser { 1640 public: 1641 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 1642 AnnotatedLine &Line) 1643 : Style(Style), Keywords(Keywords), Current(Line.First) {} 1644 1645 /// \brief Parse expressions with the given operator precedence. 1646 void parse(int Precedence = 0) { 1647 // Skip 'return' and ObjC selector colons as they are not part of a binary 1648 // expression. 1649 while (Current && (Current->is(tok::kw_return) || 1650 (Current->is(tok::colon) && 1651 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 1652 next(); 1653 1654 if (!Current || Precedence > PrecedenceArrowAndPeriod) 1655 return; 1656 1657 // Conditional expressions need to be parsed separately for proper nesting. 1658 if (Precedence == prec::Conditional) { 1659 parseConditionalExpr(); 1660 return; 1661 } 1662 1663 // Parse unary operators, which all have a higher precedence than binary 1664 // operators. 1665 if (Precedence == PrecedenceUnaryOperator) { 1666 parseUnaryOperator(); 1667 return; 1668 } 1669 1670 FormatToken *Start = Current; 1671 FormatToken *LatestOperator = nullptr; 1672 unsigned OperatorIndex = 0; 1673 1674 while (Current) { 1675 // Consume operators with higher precedence. 1676 parse(Precedence + 1); 1677 1678 int CurrentPrecedence = getCurrentPrecedence(); 1679 1680 if (Current && Current->is(TT_SelectorName) && 1681 Precedence == CurrentPrecedence) { 1682 if (LatestOperator) 1683 addFakeParenthesis(Start, prec::Level(Precedence)); 1684 Start = Current; 1685 } 1686 1687 // At the end of the line or when an operator with higher precedence is 1688 // found, insert fake parenthesis and return. 1689 if (!Current || 1690 (Current->closesScope() && 1691 (Current->MatchingParen || Current->is(TT_TemplateString))) || 1692 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 1693 (CurrentPrecedence == prec::Conditional && 1694 Precedence == prec::Assignment && Current->is(tok::colon))) { 1695 break; 1696 } 1697 1698 // Consume scopes: (), [], <> and {} 1699 if (Current->opensScope()) { 1700 // In fragment of a JavaScript template string can look like '}..${' and 1701 // thus close a scope and open a new one at the same time. 1702 while (Current && (!Current->closesScope() || Current->opensScope())) { 1703 next(); 1704 parse(); 1705 } 1706 next(); 1707 } else { 1708 // Operator found. 1709 if (CurrentPrecedence == Precedence) { 1710 if (LatestOperator) 1711 LatestOperator->NextOperator = Current; 1712 LatestOperator = Current; 1713 Current->OperatorIndex = OperatorIndex; 1714 ++OperatorIndex; 1715 } 1716 next(/*SkipPastLeadingComments=*/Precedence > 0); 1717 } 1718 } 1719 1720 if (LatestOperator && (Current || Precedence > 0)) { 1721 // LatestOperator->LastOperator = true; 1722 if (Precedence == PrecedenceArrowAndPeriod) { 1723 // Call expressions don't have a binary operator precedence. 1724 addFakeParenthesis(Start, prec::Unknown); 1725 } else { 1726 addFakeParenthesis(Start, prec::Level(Precedence)); 1727 } 1728 } 1729 } 1730 1731 private: 1732 /// \brief Gets the precedence (+1) of the given token for binary operators 1733 /// and other tokens that we treat like binary operators. 1734 int getCurrentPrecedence() { 1735 if (Current) { 1736 const FormatToken *NextNonComment = Current->getNextNonComment(); 1737 if (Current->is(TT_ConditionalExpr)) 1738 return prec::Conditional; 1739 if (NextNonComment && Current->is(TT_SelectorName) && 1740 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) || 1741 ((Style.Language == FormatStyle::LK_Proto || 1742 Style.Language == FormatStyle::LK_TextProto) && 1743 NextNonComment->is(tok::less)))) 1744 return prec::Assignment; 1745 if (Current->is(TT_JsComputedPropertyName)) 1746 return prec::Assignment; 1747 if (Current->is(TT_LambdaArrow)) 1748 return prec::Comma; 1749 if (Current->is(TT_JsFatArrow)) 1750 return prec::Assignment; 1751 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) || 1752 (Current->is(tok::comment) && NextNonComment && 1753 NextNonComment->is(TT_SelectorName))) 1754 return 0; 1755 if (Current->is(TT_RangeBasedForLoopColon)) 1756 return prec::Comma; 1757 if ((Style.Language == FormatStyle::LK_Java || 1758 Style.Language == FormatStyle::LK_JavaScript) && 1759 Current->is(Keywords.kw_instanceof)) 1760 return prec::Relational; 1761 if (Style.Language == FormatStyle::LK_JavaScript && 1762 Current->isOneOf(Keywords.kw_in, Keywords.kw_as)) 1763 return prec::Relational; 1764 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 1765 return Current->getPrecedence(); 1766 if (Current->isOneOf(tok::period, tok::arrow)) 1767 return PrecedenceArrowAndPeriod; 1768 if ((Style.Language == FormatStyle::LK_Java || 1769 Style.Language == FormatStyle::LK_JavaScript) && 1770 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 1771 Keywords.kw_throws)) 1772 return 0; 1773 } 1774 return -1; 1775 } 1776 1777 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 1778 Start->FakeLParens.push_back(Precedence); 1779 if (Precedence > prec::Unknown) 1780 Start->StartsBinaryExpression = true; 1781 if (Current) { 1782 FormatToken *Previous = Current->Previous; 1783 while (Previous->is(tok::comment) && Previous->Previous) 1784 Previous = Previous->Previous; 1785 ++Previous->FakeRParens; 1786 if (Precedence > prec::Unknown) 1787 Previous->EndsBinaryExpression = true; 1788 } 1789 } 1790 1791 /// \brief Parse unary operator expressions and surround them with fake 1792 /// parentheses if appropriate. 1793 void parseUnaryOperator() { 1794 llvm::SmallVector<FormatToken *, 2> Tokens; 1795 while (Current && Current->is(TT_UnaryOperator)) { 1796 Tokens.push_back(Current); 1797 next(); 1798 } 1799 parse(PrecedenceArrowAndPeriod); 1800 for (FormatToken *Token : llvm::reverse(Tokens)) 1801 // The actual precedence doesn't matter. 1802 addFakeParenthesis(Token, prec::Unknown); 1803 } 1804 1805 void parseConditionalExpr() { 1806 while (Current && Current->isTrailingComment()) { 1807 next(); 1808 } 1809 FormatToken *Start = Current; 1810 parse(prec::LogicalOr); 1811 if (!Current || !Current->is(tok::question)) 1812 return; 1813 next(); 1814 parse(prec::Assignment); 1815 if (!Current || Current->isNot(TT_ConditionalExpr)) 1816 return; 1817 next(); 1818 parse(prec::Assignment); 1819 addFakeParenthesis(Start, prec::Conditional); 1820 } 1821 1822 void next(bool SkipPastLeadingComments = true) { 1823 if (Current) 1824 Current = Current->Next; 1825 while (Current && 1826 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 1827 Current->isTrailingComment()) 1828 Current = Current->Next; 1829 } 1830 1831 const FormatStyle &Style; 1832 const AdditionalKeywords &Keywords; 1833 FormatToken *Current; 1834 }; 1835 1836 } // end anonymous namespace 1837 1838 void TokenAnnotator::setCommentLineLevels( 1839 SmallVectorImpl<AnnotatedLine *> &Lines) { 1840 const AnnotatedLine *NextNonCommentLine = nullptr; 1841 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(), 1842 E = Lines.rend(); 1843 I != E; ++I) { 1844 bool CommentLine = true; 1845 for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) { 1846 if (!Tok->is(tok::comment)) { 1847 CommentLine = false; 1848 break; 1849 } 1850 } 1851 1852 // If the comment is currently aligned with the line immediately following 1853 // it, that's probably intentional and we should keep it. 1854 if (NextNonCommentLine && CommentLine && 1855 NextNonCommentLine->First->NewlinesBefore <= 1 && 1856 NextNonCommentLine->First->OriginalColumn == 1857 (*I)->First->OriginalColumn) { 1858 // Align comments for preprocessor lines with the # in column 0. 1859 // Otherwise, align with the next line. 1860 (*I)->Level = (NextNonCommentLine->Type == LT_PreprocessorDirective || 1861 NextNonCommentLine->Type == LT_ImportStatement) 1862 ? 0 1863 : NextNonCommentLine->Level; 1864 } else { 1865 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr; 1866 } 1867 1868 setCommentLineLevels((*I)->Children); 1869 } 1870 } 1871 1872 static unsigned maxNestingDepth(const AnnotatedLine &Line) { 1873 unsigned Result = 0; 1874 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) 1875 Result = std::max(Result, Tok->NestingLevel); 1876 return Result; 1877 } 1878 1879 void TokenAnnotator::annotate(AnnotatedLine &Line) { 1880 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 1881 E = Line.Children.end(); 1882 I != E; ++I) { 1883 annotate(**I); 1884 } 1885 AnnotatingParser Parser(Style, Line, Keywords); 1886 Line.Type = Parser.parseLine(); 1887 1888 // With very deep nesting, ExpressionParser uses lots of stack and the 1889 // formatting algorithm is very slow. We're not going to do a good job here 1890 // anyway - it's probably generated code being formatted by mistake. 1891 // Just skip the whole line. 1892 if (maxNestingDepth(Line) > 50) 1893 Line.Type = LT_Invalid; 1894 1895 if (Line.Type == LT_Invalid) 1896 return; 1897 1898 ExpressionParser ExprParser(Style, Keywords, Line); 1899 ExprParser.parse(); 1900 1901 if (Line.startsWith(TT_ObjCMethodSpecifier)) 1902 Line.Type = LT_ObjCMethodDecl; 1903 else if (Line.startsWith(TT_ObjCDecl)) 1904 Line.Type = LT_ObjCDecl; 1905 else if (Line.startsWith(TT_ObjCProperty)) 1906 Line.Type = LT_ObjCProperty; 1907 1908 Line.First->SpacesRequiredBefore = 1; 1909 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 1910 } 1911 1912 // This function heuristically determines whether 'Current' starts the name of a 1913 // function declaration. 1914 static bool isFunctionDeclarationName(const FormatToken &Current, 1915 const AnnotatedLine &Line) { 1916 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * { 1917 for (; Next; Next = Next->Next) { 1918 if (Next->is(TT_OverloadedOperatorLParen)) 1919 return Next; 1920 if (Next->is(TT_OverloadedOperator)) 1921 continue; 1922 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 1923 // For 'new[]' and 'delete[]'. 1924 if (Next->Next && Next->Next->is(tok::l_square) && Next->Next->Next && 1925 Next->Next->Next->is(tok::r_square)) 1926 Next = Next->Next->Next; 1927 continue; 1928 } 1929 1930 break; 1931 } 1932 return nullptr; 1933 }; 1934 1935 // Find parentheses of parameter list. 1936 const FormatToken *Next = Current.Next; 1937 if (Current.is(tok::kw_operator)) { 1938 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 1939 return false; 1940 Next = skipOperatorName(Next); 1941 } else { 1942 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 1943 return false; 1944 for (; Next; Next = Next->Next) { 1945 if (Next->is(TT_TemplateOpener)) { 1946 Next = Next->MatchingParen; 1947 } else if (Next->is(tok::coloncolon)) { 1948 Next = Next->Next; 1949 if (!Next) 1950 return false; 1951 if (Next->is(tok::kw_operator)) { 1952 Next = skipOperatorName(Next->Next); 1953 break; 1954 } 1955 if (!Next->is(tok::identifier)) 1956 return false; 1957 } else if (Next->is(tok::l_paren)) { 1958 break; 1959 } else { 1960 return false; 1961 } 1962 } 1963 } 1964 1965 // Check whether parameter list can belong to a function declaration. 1966 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 1967 return false; 1968 // If the lines ends with "{", this is likely an function definition. 1969 if (Line.Last->is(tok::l_brace)) 1970 return true; 1971 if (Next->Next == Next->MatchingParen) 1972 return true; // Empty parentheses. 1973 // If there is an &/&& after the r_paren, this is likely a function. 1974 if (Next->MatchingParen->Next && 1975 Next->MatchingParen->Next->is(TT_PointerOrReference)) 1976 return true; 1977 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 1978 Tok = Tok->Next) { 1979 if (Tok->is(tok::l_paren) && Tok->MatchingParen) { 1980 Tok = Tok->MatchingParen; 1981 continue; 1982 } 1983 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 1984 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 1985 return true; 1986 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 1987 Tok->Tok.isLiteral()) 1988 return false; 1989 } 1990 return false; 1991 } 1992 1993 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 1994 assert(Line.MightBeFunctionDecl); 1995 1996 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 1997 Style.AlwaysBreakAfterReturnType == 1998 FormatStyle::RTBS_TopLevelDefinitions) && 1999 Line.Level > 0) 2000 return false; 2001 2002 switch (Style.AlwaysBreakAfterReturnType) { 2003 case FormatStyle::RTBS_None: 2004 return false; 2005 case FormatStyle::RTBS_All: 2006 case FormatStyle::RTBS_TopLevel: 2007 return true; 2008 case FormatStyle::RTBS_AllDefinitions: 2009 case FormatStyle::RTBS_TopLevelDefinitions: 2010 return Line.mightBeFunctionDefinition(); 2011 } 2012 2013 return false; 2014 } 2015 2016 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 2017 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 2018 E = Line.Children.end(); 2019 I != E; ++I) { 2020 calculateFormattingInformation(**I); 2021 } 2022 2023 Line.First->TotalLength = 2024 Line.First->IsMultiline ? Style.ColumnLimit 2025 : Line.FirstStartColumn + Line.First->ColumnWidth; 2026 FormatToken *Current = Line.First->Next; 2027 bool InFunctionDecl = Line.MightBeFunctionDecl; 2028 while (Current) { 2029 if (isFunctionDeclarationName(*Current, Line)) 2030 Current->Type = TT_FunctionDeclarationName; 2031 if (Current->is(TT_LineComment)) { 2032 if (Current->Previous->BlockKind == BK_BracedInit && 2033 Current->Previous->opensScope()) 2034 Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1; 2035 else 2036 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 2037 2038 // If we find a trailing comment, iterate backwards to determine whether 2039 // it seems to relate to a specific parameter. If so, break before that 2040 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 2041 // to the previous line in: 2042 // SomeFunction(a, 2043 // b, // comment 2044 // c); 2045 if (!Current->HasUnescapedNewline) { 2046 for (FormatToken *Parameter = Current->Previous; Parameter; 2047 Parameter = Parameter->Previous) { 2048 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 2049 break; 2050 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 2051 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 2052 Parameter->HasUnescapedNewline) 2053 Parameter->MustBreakBefore = true; 2054 break; 2055 } 2056 } 2057 } 2058 } else if (Current->SpacesRequiredBefore == 0 && 2059 spaceRequiredBefore(Line, *Current)) { 2060 Current->SpacesRequiredBefore = 1; 2061 } 2062 2063 Current->MustBreakBefore = 2064 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 2065 2066 if (!Current->MustBreakBefore && InFunctionDecl && 2067 Current->is(TT_FunctionDeclarationName)) 2068 Current->MustBreakBefore = mustBreakForReturnType(Line); 2069 2070 Current->CanBreakBefore = 2071 Current->MustBreakBefore || canBreakBefore(Line, *Current); 2072 unsigned ChildSize = 0; 2073 if (Current->Previous->Children.size() == 1) { 2074 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last; 2075 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 2076 : LastOfChild.TotalLength + 1; 2077 } 2078 const FormatToken *Prev = Current->Previous; 2079 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 2080 (Prev->Children.size() == 1 && 2081 Prev->Children[0]->First->MustBreakBefore) || 2082 Current->IsMultiline) 2083 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 2084 else 2085 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 2086 ChildSize + Current->SpacesRequiredBefore; 2087 2088 if (Current->is(TT_CtorInitializerColon)) 2089 InFunctionDecl = false; 2090 2091 // FIXME: Only calculate this if CanBreakBefore is true once static 2092 // initializers etc. are sorted out. 2093 // FIXME: Move magic numbers to a better place. 2094 Current->SplitPenalty = 20 * Current->BindingStrength + 2095 splitPenalty(Line, *Current, InFunctionDecl); 2096 2097 Current = Current->Next; 2098 } 2099 2100 calculateUnbreakableTailLengths(Line); 2101 unsigned IndentLevel = Line.Level; 2102 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 2103 if (Current->Role) 2104 Current->Role->precomputeFormattingInfos(Current); 2105 if (Current->MatchingParen && 2106 Current->MatchingParen->opensBlockOrBlockTypeList(Style)) { 2107 assert(IndentLevel > 0); 2108 --IndentLevel; 2109 } 2110 Current->IndentLevel = IndentLevel; 2111 if (Current->opensBlockOrBlockTypeList(Style)) 2112 ++IndentLevel; 2113 } 2114 2115 DEBUG({ printDebugInfo(Line); }); 2116 } 2117 2118 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 2119 unsigned UnbreakableTailLength = 0; 2120 FormatToken *Current = Line.Last; 2121 while (Current) { 2122 Current->UnbreakableTailLength = UnbreakableTailLength; 2123 if (Current->CanBreakBefore || 2124 Current->isOneOf(tok::comment, tok::string_literal)) { 2125 UnbreakableTailLength = 0; 2126 } else { 2127 UnbreakableTailLength += 2128 Current->ColumnWidth + Current->SpacesRequiredBefore; 2129 } 2130 Current = Current->Previous; 2131 } 2132 } 2133 2134 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 2135 const FormatToken &Tok, 2136 bool InFunctionDecl) { 2137 const FormatToken &Left = *Tok.Previous; 2138 const FormatToken &Right = Tok; 2139 2140 if (Left.is(tok::semi)) 2141 return 0; 2142 2143 if (Style.Language == FormatStyle::LK_Java) { 2144 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 2145 return 1; 2146 if (Right.is(Keywords.kw_implements)) 2147 return 2; 2148 if (Left.is(tok::comma) && Left.NestingLevel == 0) 2149 return 3; 2150 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2151 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 2152 return 100; 2153 if (Left.is(TT_JsTypeColon)) 2154 return 35; 2155 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2156 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2157 return 100; 2158 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()". 2159 if (Left.opensScope() && Right.closesScope()) 2160 return 200; 2161 } 2162 2163 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2164 return 1; 2165 if (Right.is(tok::l_square)) { 2166 if (Style.Language == FormatStyle::LK_Proto) 2167 return 1; 2168 if (Left.is(tok::r_square)) 2169 return 200; 2170 // Slightly prefer formatting local lambda definitions like functions. 2171 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 2172 return 35; 2173 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2174 TT_ArrayInitializerLSquare, 2175 TT_DesignatedInitializerLSquare, TT_AttributeSquare)) 2176 return 500; 2177 } 2178 2179 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2180 Right.is(tok::kw_operator)) { 2181 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 2182 return 3; 2183 if (Left.is(TT_StartOfName)) 2184 return 110; 2185 if (InFunctionDecl && Right.NestingLevel == 0) 2186 return Style.PenaltyReturnTypeOnItsOwnLine; 2187 return 200; 2188 } 2189 if (Right.is(TT_PointerOrReference)) 2190 return 190; 2191 if (Right.is(TT_LambdaArrow)) 2192 return 110; 2193 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 2194 return 160; 2195 if (Left.is(TT_CastRParen)) 2196 return 100; 2197 if (Left.is(tok::coloncolon) || 2198 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 2199 return 500; 2200 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 2201 return 5000; 2202 if (Left.is(tok::comment)) 2203 return 1000; 2204 2205 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon, 2206 TT_CtorInitializerColon)) 2207 return 2; 2208 2209 if (Right.isMemberAccess()) { 2210 // Breaking before the "./->" of a chained call/member access is reasonably 2211 // cheap, as formatting those with one call per line is generally 2212 // desirable. In particular, it should be cheaper to break before the call 2213 // than it is to break inside a call's parameters, which could lead to weird 2214 // "hanging" indents. The exception is the very last "./->" to support this 2215 // frequent pattern: 2216 // 2217 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 2218 // dddddddd); 2219 // 2220 // which might otherwise be blown up onto many lines. Here, clang-format 2221 // won't produce "hanging" indents anyway as there is no other trailing 2222 // call. 2223 // 2224 // Also apply higher penalty is not a call as that might lead to a wrapping 2225 // like: 2226 // 2227 // aaaaaaa 2228 // .aaaaaaaaa.bbbbbbbb(cccccccc); 2229 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 2230 ? 150 2231 : 35; 2232 } 2233 2234 if (Right.is(TT_TrailingAnnotation) && 2235 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 2236 // Moving trailing annotations to the next line is fine for ObjC method 2237 // declarations. 2238 if (Line.startsWith(TT_ObjCMethodSpecifier)) 2239 return 10; 2240 // Generally, breaking before a trailing annotation is bad unless it is 2241 // function-like. It seems to be especially preferable to keep standard 2242 // annotations (i.e. "const", "final" and "override") on the same line. 2243 // Use a slightly higher penalty after ")" so that annotations like 2244 // "const override" are kept together. 2245 bool is_short_annotation = Right.TokenText.size() < 10; 2246 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 2247 } 2248 2249 // In for-loops, prefer breaking at ',' and ';'. 2250 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 2251 return 4; 2252 2253 // In Objective-C method expressions, prefer breaking before "param:" over 2254 // breaking after it. 2255 if (Right.is(TT_SelectorName)) 2256 return 0; 2257 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 2258 return Line.MightBeFunctionDecl ? 50 : 500; 2259 2260 if (Left.is(tok::l_paren) && InFunctionDecl && 2261 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 2262 return 100; 2263 if (Left.is(tok::l_paren) && Left.Previous && 2264 (Left.Previous->isOneOf(tok::kw_if, tok::kw_for) || 2265 Left.Previous->endsSequence(tok::kw_constexpr, tok::kw_if))) 2266 return 1000; 2267 if (Left.is(tok::equal) && InFunctionDecl) 2268 return 110; 2269 if (Right.is(tok::r_brace)) 2270 return 1; 2271 if (Left.is(TT_TemplateOpener)) 2272 return 100; 2273 if (Left.opensScope()) { 2274 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign) 2275 return 0; 2276 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 2277 : 19; 2278 } 2279 if (Left.is(TT_JavaAnnotation)) 2280 return 50; 2281 2282 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous && 2283 Left.Previous->isLabelString() && 2284 (Left.NextOperator || Left.OperatorIndex != 0)) 2285 return 50; 2286 if (Right.is(tok::plus) && Left.isLabelString() && 2287 (Right.NextOperator || Right.OperatorIndex != 0)) 2288 return 25; 2289 if (Left.is(tok::comma)) 2290 return 1; 2291 if (Right.is(tok::lessless) && Left.isLabelString() && 2292 (Right.NextOperator || Right.OperatorIndex != 1)) 2293 return 25; 2294 if (Right.is(tok::lessless)) { 2295 // Breaking at a << is really cheap. 2296 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0) 2297 // Slightly prefer to break before the first one in log-like statements. 2298 return 2; 2299 return 1; 2300 } 2301 if (Left.is(TT_ConditionalExpr)) 2302 return prec::Conditional; 2303 prec::Level Level = Left.getPrecedence(); 2304 if (Level == prec::Unknown) 2305 Level = Right.getPrecedence(); 2306 if (Level == prec::Assignment) 2307 return Style.PenaltyBreakAssignment; 2308 if (Level != prec::Unknown) 2309 return Level; 2310 2311 return 3; 2312 } 2313 2314 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 2315 const FormatToken &Left, 2316 const FormatToken &Right) { 2317 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 2318 return true; 2319 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java) 2320 return true; 2321 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 2322 Left.Tok.getObjCKeywordID() == tok::objc_property) 2323 return true; 2324 if (Right.is(tok::hashhash)) 2325 return Left.is(tok::hash); 2326 if (Left.isOneOf(tok::hashhash, tok::hash)) 2327 return Right.is(tok::hash); 2328 if (Left.is(tok::l_paren) && Right.is(tok::r_paren)) 2329 return Style.SpaceInEmptyParentheses; 2330 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 2331 return (Right.is(TT_CastRParen) || 2332 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 2333 ? Style.SpacesInCStyleCastParentheses 2334 : Style.SpacesInParentheses; 2335 if (Right.isOneOf(tok::semi, tok::comma)) 2336 return false; 2337 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl && 2338 Style.ObjCSpaceBeforeProtocolList) 2339 return true; 2340 if (Right.is(tok::less) && Left.is(tok::kw_template)) 2341 return Style.SpaceAfterTemplateKeyword; 2342 if (Left.isOneOf(tok::exclaim, tok::tilde)) 2343 return false; 2344 if (Left.is(tok::at) && 2345 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 2346 tok::numeric_constant, tok::l_paren, tok::l_brace, 2347 tok::kw_true, tok::kw_false)) 2348 return false; 2349 if (Left.is(tok::colon)) 2350 return !Left.is(TT_ObjCMethodExpr); 2351 if (Left.is(tok::coloncolon)) 2352 return false; 2353 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) { 2354 if (Style.Language == FormatStyle::LK_TextProto || 2355 (Style.Language == FormatStyle::LK_Proto && 2356 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) { 2357 // Format empty list as `<>`. 2358 if (Left.is(tok::less) && Right.is(tok::greater)) 2359 return false; 2360 return !Style.Cpp11BracedListStyle; 2361 } 2362 return false; 2363 } 2364 if (Right.is(tok::ellipsis)) 2365 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous && 2366 Left.Previous->is(tok::kw_case)); 2367 if (Left.is(tok::l_square) && Right.is(tok::amp)) 2368 return false; 2369 if (Right.is(TT_PointerOrReference)) { 2370 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) { 2371 if (!Left.MatchingParen) 2372 return true; 2373 FormatToken *TokenBeforeMatchingParen = 2374 Left.MatchingParen->getPreviousNonComment(); 2375 if (!TokenBeforeMatchingParen || 2376 !TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype)) 2377 return true; 2378 } 2379 return (Left.Tok.isLiteral() || 2380 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 2381 (Style.PointerAlignment != FormatStyle::PAS_Left || 2382 (Line.IsMultiVariableDeclStmt && 2383 (Left.NestingLevel == 0 || 2384 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for))))))); 2385 } 2386 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 2387 (!Left.is(TT_PointerOrReference) || 2388 (Style.PointerAlignment != FormatStyle::PAS_Right && 2389 !Line.IsMultiVariableDeclStmt))) 2390 return true; 2391 if (Left.is(TT_PointerOrReference)) 2392 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) || 2393 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) && 2394 !Right.is(TT_StartOfName)) || 2395 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) || 2396 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 2397 tok::l_paren) && 2398 (Style.PointerAlignment != FormatStyle::PAS_Right && 2399 !Line.IsMultiVariableDeclStmt) && 2400 Left.Previous && 2401 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon)); 2402 if (Right.is(tok::star) && Left.is(tok::l_paren)) 2403 return false; 2404 const auto SpaceRequiredForArrayInitializerLSquare = 2405 [](const FormatToken &LSquareTok, const FormatStyle &Style) { 2406 return Style.SpacesInContainerLiterals || 2407 ((Style.Language == FormatStyle::LK_Proto || 2408 Style.Language == FormatStyle::LK_TextProto) && 2409 !Style.Cpp11BracedListStyle && 2410 LSquareTok.endsSequence(tok::l_square, tok::colon, 2411 TT_SelectorName)); 2412 }; 2413 if (Left.is(tok::l_square)) 2414 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) && 2415 SpaceRequiredForArrayInitializerLSquare(Left, Style)) || 2416 (Left.isOneOf(TT_ArraySubscriptLSquare, 2417 TT_StructuredBindingLSquare) && 2418 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square)); 2419 if (Right.is(tok::r_square)) 2420 return Right.MatchingParen && 2421 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) && 2422 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen, 2423 Style)) || 2424 (Style.SpacesInSquareBrackets && 2425 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare, 2426 TT_StructuredBindingLSquare)) || 2427 Right.MatchingParen->is(TT_AttributeParen)); 2428 if (Right.is(tok::l_square) && 2429 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 2430 TT_DesignatedInitializerLSquare, 2431 TT_StructuredBindingLSquare, TT_AttributeSquare) && 2432 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral)) 2433 return false; 2434 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 2435 return !Left.Children.empty(); // No spaces in "{}". 2436 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) || 2437 (Right.is(tok::r_brace) && Right.MatchingParen && 2438 Right.MatchingParen->BlockKind != BK_Block)) 2439 return !Style.Cpp11BracedListStyle; 2440 if (Left.is(TT_BlockComment)) 2441 return !Left.TokenText.endswith("=*/"); 2442 if (Right.is(tok::l_paren)) { 2443 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) || 2444 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare))) 2445 return true; 2446 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) || 2447 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never && 2448 (Left.isOneOf(tok::kw_if, tok::pp_elif, tok::kw_for, tok::kw_while, 2449 tok::kw_switch, tok::kw_case, TT_ForEachMacro, 2450 TT_ObjCForIn) || 2451 Left.endsSequence(tok::kw_constexpr, tok::kw_if) || 2452 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch, 2453 tok::kw_new, tok::kw_delete) && 2454 (!Left.Previous || Left.Previous->isNot(tok::period))))) || 2455 (Style.SpaceBeforeParens == FormatStyle::SBPO_Always && 2456 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 2457 Left.is(tok::r_paren)) && 2458 Line.Type != LT_PreprocessorDirective); 2459 } 2460 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 2461 return false; 2462 if (Right.is(TT_UnaryOperator)) 2463 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 2464 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 2465 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 2466 tok::r_paren) || 2467 Left.isSimpleTypeSpecifier()) && 2468 Right.is(tok::l_brace) && Right.getNextNonComment() && 2469 Right.BlockKind != BK_Block) 2470 return false; 2471 if (Left.is(tok::period) || Right.is(tok::period)) 2472 return false; 2473 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L") 2474 return false; 2475 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 2476 Left.MatchingParen->Previous && 2477 Left.MatchingParen->Previous->is(tok::period)) 2478 // A.<B<C<...>>>DoSomething(); 2479 return false; 2480 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 2481 return false; 2482 return true; 2483 } 2484 2485 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 2486 const FormatToken &Right) { 2487 const FormatToken &Left = *Right.Previous; 2488 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 2489 return true; // Never ever merge two identifiers. 2490 if (Style.isCpp()) { 2491 if (Left.is(tok::kw_operator)) 2492 return Right.is(tok::coloncolon); 2493 } else if (Style.Language == FormatStyle::LK_Proto || 2494 Style.Language == FormatStyle::LK_TextProto) { 2495 if (Right.is(tok::period) && 2496 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 2497 Keywords.kw_repeated, Keywords.kw_extend)) 2498 return true; 2499 if (Right.is(tok::l_paren) && 2500 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 2501 return true; 2502 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName)) 2503 return true; 2504 // Slashes occur in text protocol extension syntax: [type/type] { ... }. 2505 if (Left.is(tok::slash) || Right.is(tok::slash)) 2506 return false; 2507 if (Left.MatchingParen && Left.MatchingParen->is(TT_ProtoExtensionLSquare) && 2508 Right.isOneOf(tok::l_brace, tok::less)) 2509 return !Style.Cpp11BracedListStyle; 2510 // A percent is probably part of a formatting specification, such as %lld. 2511 if (Left.is(tok::percent)) 2512 return false; 2513 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2514 if (Left.is(TT_JsFatArrow)) 2515 return true; 2516 // for await ( ... 2517 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous && 2518 Left.Previous->is(tok::kw_for)) 2519 return true; 2520 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) && 2521 Right.MatchingParen) { 2522 const FormatToken *Next = Right.MatchingParen->getNextNonComment(); 2523 // An async arrow function, for example: `x = async () => foo();`, 2524 // as opposed to calling a function called async: `x = async();` 2525 if (Next && Next->is(TT_JsFatArrow)) 2526 return true; 2527 } 2528 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) || 2529 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}"))) 2530 return false; 2531 // In tagged template literals ("html`bar baz`"), there is no space between 2532 // the tag identifier and the template string. getIdentifierInfo makes sure 2533 // that the identifier is not a pseudo keyword like `yield`, either. 2534 if (Left.is(tok::identifier) && Keywords.IsJavaScriptIdentifier(Left) && 2535 Right.is(TT_TemplateString)) 2536 return false; 2537 if (Right.is(tok::star) && 2538 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 2539 return false; 2540 if (Right.isOneOf(tok::l_brace, tok::l_square) && 2541 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield, 2542 Keywords.kw_extends, Keywords.kw_implements)) 2543 return true; 2544 if (Right.is(tok::l_paren)) { 2545 // JS methods can use some keywords as names (e.g. `delete()`). 2546 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo()) 2547 return false; 2548 // Valid JS method names can include keywords, e.g. `foo.delete()` or 2549 // `bar.instanceof()`. Recognize call positions by preceding period. 2550 if (Left.Previous && Left.Previous->is(tok::period) && 2551 Left.Tok.getIdentifierInfo()) 2552 return false; 2553 // Additional unary JavaScript operators that need a space after. 2554 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof, 2555 tok::kw_void)) 2556 return true; 2557 } 2558 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 2559 tok::kw_const) || 2560 // "of" is only a keyword if it appears after another identifier 2561 // (e.g. as "const x of y" in a for loop), or after a destructuring 2562 // operation (const [x, y] of z, const {a, b} of c). 2563 (Left.is(Keywords.kw_of) && Left.Previous && 2564 (Left.Previous->Tok.is(tok::identifier) || 2565 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) && 2566 (!Left.Previous || !Left.Previous->is(tok::period))) 2567 return true; 2568 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous && 2569 Left.Previous->is(tok::period) && Right.is(tok::l_paren)) 2570 return false; 2571 if (Left.is(Keywords.kw_as) && 2572 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren)) 2573 return true; 2574 if (Left.is(tok::kw_default) && Left.Previous && 2575 Left.Previous->is(tok::kw_export)) 2576 return true; 2577 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 2578 return true; 2579 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 2580 return false; 2581 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 2582 return false; 2583 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 2584 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 2585 return false; 2586 if (Left.is(tok::ellipsis)) 2587 return false; 2588 if (Left.is(TT_TemplateCloser) && 2589 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 2590 Keywords.kw_implements, Keywords.kw_extends)) 2591 // Type assertions ('<type>expr') are not followed by whitespace. Other 2592 // locations that should have whitespace following are identified by the 2593 // above set of follower tokens. 2594 return false; 2595 if (Right.is(TT_JsNonNullAssertion)) 2596 return false; 2597 if (Left.is(TT_JsNonNullAssertion) && 2598 Right.isOneOf(Keywords.kw_as, Keywords.kw_in)) 2599 return true; // "x! as string", "x! in y" 2600 } else if (Style.Language == FormatStyle::LK_Java) { 2601 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 2602 return true; 2603 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 2604 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never; 2605 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 2606 tok::kw_protected) || 2607 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 2608 Keywords.kw_native)) && 2609 Right.is(TT_TemplateOpener)) 2610 return true; 2611 } 2612 if (Left.is(TT_ImplicitStringLiteral)) 2613 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2614 if (Line.Type == LT_ObjCMethodDecl) { 2615 if (Left.is(TT_ObjCMethodSpecifier)) 2616 return true; 2617 if (Left.is(tok::r_paren) && Right.is(tok::identifier)) 2618 // Don't space between ')' and <id> 2619 return false; 2620 } 2621 if (Line.Type == LT_ObjCProperty && 2622 (Right.is(tok::equal) || Left.is(tok::equal))) 2623 return false; 2624 2625 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 2626 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 2627 return true; 2628 if (Right.is(TT_OverloadedOperatorLParen)) 2629 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always; 2630 if (Left.is(tok::comma)) 2631 return true; 2632 if (Right.is(tok::comma)) 2633 return false; 2634 if (Right.is(TT_ObjCBlockLParen)) 2635 return true; 2636 if (Right.is(TT_CtorInitializerColon)) 2637 return Style.SpaceBeforeCtorInitializerColon; 2638 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon) 2639 return false; 2640 if (Right.is(TT_RangeBasedForLoopColon) && 2641 !Style.SpaceBeforeRangeBasedForLoopColon) 2642 return false; 2643 if (Right.is(tok::colon)) { 2644 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) || 2645 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 2646 return false; 2647 if (Right.is(TT_ObjCMethodExpr)) 2648 return false; 2649 if (Left.is(tok::question)) 2650 return false; 2651 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 2652 return false; 2653 if (Right.is(TT_DictLiteral)) 2654 return Style.SpacesInContainerLiterals; 2655 if (Right.is(TT_AttributeColon)) 2656 return false; 2657 return true; 2658 } 2659 if (Left.is(TT_UnaryOperator)) 2660 return Right.is(TT_BinaryOperator); 2661 2662 // If the next token is a binary operator or a selector name, we have 2663 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 2664 if (Left.is(TT_CastRParen)) 2665 return Style.SpaceAfterCStyleCast || 2666 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 2667 2668 if (Left.is(tok::greater) && Right.is(tok::greater)) { 2669 if (Style.Language == FormatStyle::LK_TextProto || 2670 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral))) 2671 return !Style.Cpp11BracedListStyle; 2672 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 2673 (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles); 2674 } 2675 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) || 2676 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 2677 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod))) 2678 return false; 2679 if (!Style.SpaceBeforeAssignmentOperators && 2680 Right.getPrecedence() == prec::Assignment) 2681 return false; 2682 if (Right.is(tok::coloncolon) && Left.is(tok::identifier)) 2683 // Generally don't remove existing spaces between an identifier and "::". 2684 // The identifier might actually be a macro name such as ALWAYS_INLINE. If 2685 // this turns out to be too lenient, add analysis of the identifier itself. 2686 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2687 if (Right.is(tok::coloncolon) && !Left.isOneOf(tok::l_brace, tok::comment)) 2688 return (Left.is(TT_TemplateOpener) && 2689 Style.Standard == FormatStyle::LS_Cpp03) || 2690 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square, 2691 tok::kw___super, TT_TemplateCloser, 2692 TT_TemplateOpener)); 2693 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 2694 return Style.SpacesInAngles; 2695 // Space before TT_StructuredBindingLSquare. 2696 if (Right.is(TT_StructuredBindingLSquare)) 2697 return !Left.isOneOf(tok::amp, tok::ampamp) || 2698 Style.PointerAlignment != FormatStyle::PAS_Right; 2699 // Space before & or && following a TT_StructuredBindingLSquare. 2700 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) && 2701 Right.isOneOf(tok::amp, tok::ampamp)) 2702 return Style.PointerAlignment != FormatStyle::PAS_Left; 2703 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 2704 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 2705 !Right.is(tok::r_paren))) 2706 return true; 2707 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) && 2708 Right.isNot(TT_FunctionTypeLParen)) 2709 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always; 2710 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 2711 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 2712 return false; 2713 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 2714 Line.startsWith(tok::hash)) 2715 return true; 2716 if (Right.is(TT_TrailingUnaryOperator)) 2717 return false; 2718 if (Left.is(TT_RegexLiteral)) 2719 return false; 2720 return spaceRequiredBetween(Line, Left, Right); 2721 } 2722 2723 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 2724 static bool isAllmanBrace(const FormatToken &Tok) { 2725 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block && 2726 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral); 2727 } 2728 2729 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 2730 const FormatToken &Right) { 2731 const FormatToken &Left = *Right.Previous; 2732 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 2733 return true; 2734 2735 if (Style.Language == FormatStyle::LK_JavaScript) { 2736 // FIXME: This might apply to other languages and token kinds. 2737 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 2738 Left.Previous->is(tok::string_literal)) 2739 return true; 2740 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 2741 Left.Previous && Left.Previous->is(tok::equal) && 2742 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 2743 tok::kw_const) && 2744 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 2745 // above. 2746 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 2747 // Object literals on the top level of a file are treated as "enum-style". 2748 // Each key/value pair is put on a separate line, instead of bin-packing. 2749 return true; 2750 if (Left.is(tok::l_brace) && Line.Level == 0 && 2751 (Line.startsWith(tok::kw_enum) || 2752 Line.startsWith(tok::kw_const, tok::kw_enum) || 2753 Line.startsWith(tok::kw_export, tok::kw_enum) || 2754 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum))) 2755 // JavaScript top-level enum key/value pairs are put on separate lines 2756 // instead of bin-packing. 2757 return true; 2758 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 2759 !Left.Children.empty()) 2760 // Support AllowShortFunctionsOnASingleLine for JavaScript. 2761 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 2762 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 2763 (Left.NestingLevel == 0 && Line.Level == 0 && 2764 Style.AllowShortFunctionsOnASingleLine & 2765 FormatStyle::SFS_InlineOnly); 2766 } else if (Style.Language == FormatStyle::LK_Java) { 2767 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 2768 Right.Next->is(tok::string_literal)) 2769 return true; 2770 } else if (Style.Language == FormatStyle::LK_Cpp || 2771 Style.Language == FormatStyle::LK_ObjC || 2772 Style.Language == FormatStyle::LK_Proto || 2773 Style.Language == FormatStyle::LK_TextProto) { 2774 if (Left.isStringLiteral() && Right.isStringLiteral()) 2775 return true; 2776 } 2777 2778 // If the last token before a '}', ']', or ')' is a comma or a trailing 2779 // comment, the intention is to insert a line break after it in order to make 2780 // shuffling around entries easier. Import statements, especially in 2781 // JavaScript, can be an exception to this rule. 2782 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) { 2783 const FormatToken *BeforeClosingBrace = nullptr; 2784 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 2785 (Style.Language == FormatStyle::LK_JavaScript && 2786 Left.is(tok::l_paren))) && 2787 Left.BlockKind != BK_Block && Left.MatchingParen) 2788 BeforeClosingBrace = Left.MatchingParen->Previous; 2789 else if (Right.MatchingParen && 2790 (Right.MatchingParen->isOneOf(tok::l_brace, 2791 TT_ArrayInitializerLSquare) || 2792 (Style.Language == FormatStyle::LK_JavaScript && 2793 Right.MatchingParen->is(tok::l_paren)))) 2794 BeforeClosingBrace = &Left; 2795 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 2796 BeforeClosingBrace->isTrailingComment())) 2797 return true; 2798 } 2799 2800 if (Right.is(tok::comment)) 2801 return Left.BlockKind != BK_BracedInit && 2802 Left.isNot(TT_CtorInitializerColon) && 2803 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 2804 if (Left.isTrailingComment()) 2805 return true; 2806 if (Right.Previous->IsUnterminatedLiteral) 2807 return true; 2808 if (Right.is(tok::lessless) && Right.Next && 2809 Right.Previous->is(tok::string_literal) && 2810 Right.Next->is(tok::string_literal)) 2811 return true; 2812 if (Right.Previous->ClosesTemplateDeclaration && 2813 Right.Previous->MatchingParen && 2814 Right.Previous->MatchingParen->NestingLevel == 0 && 2815 Style.AlwaysBreakTemplateDeclarations) 2816 return true; 2817 if (Right.is(TT_CtorInitializerComma) && 2818 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 2819 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 2820 return true; 2821 if (Right.is(TT_CtorInitializerColon) && 2822 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma && 2823 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 2824 return true; 2825 // Break only if we have multiple inheritance. 2826 if (Style.BreakBeforeInheritanceComma && Right.is(TT_InheritanceComma)) 2827 return true; 2828 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 2829 // Raw string literals are special wrt. line breaks. The author has made a 2830 // deliberate choice and might have aligned the contents of the string 2831 // literal accordingly. Thus, we try keep existing line breaks. 2832 return Right.NewlinesBefore > 0; 2833 if ((Right.Previous->is(tok::l_brace) || 2834 (Right.Previous->is(tok::less) && Right.Previous->Previous && 2835 Right.Previous->Previous->is(tok::equal))) && 2836 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) { 2837 // Don't put enums or option definitions onto single lines in protocol 2838 // buffers. 2839 return true; 2840 } 2841 if (Right.is(TT_InlineASMBrace)) 2842 return Right.HasUnescapedNewline; 2843 if (isAllmanBrace(Left) || isAllmanBrace(Right)) 2844 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) || 2845 (Line.startsWith(tok::kw_typedef, tok::kw_enum) && 2846 Style.BraceWrapping.AfterEnum) || 2847 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 2848 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 2849 if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine) 2850 return true; 2851 2852 if ((Style.Language == FormatStyle::LK_Java || 2853 Style.Language == FormatStyle::LK_JavaScript) && 2854 Left.is(TT_LeadingJavaAnnotation) && 2855 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 2856 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 2857 return true; 2858 2859 if (Right.is(TT_ProtoExtensionLSquare)) 2860 return true; 2861 2862 return false; 2863 } 2864 2865 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 2866 const FormatToken &Right) { 2867 const FormatToken &Left = *Right.Previous; 2868 2869 // Language-specific stuff. 2870 if (Style.Language == FormatStyle::LK_Java) { 2871 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 2872 Keywords.kw_implements)) 2873 return false; 2874 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 2875 Keywords.kw_implements)) 2876 return true; 2877 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2878 const FormatToken *NonComment = Right.getPreviousNonComment(); 2879 if (NonComment && 2880 NonComment->isOneOf( 2881 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break, 2882 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type, 2883 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected, 2884 Keywords.kw_readonly, Keywords.kw_abstract, Keywords.kw_get, 2885 Keywords.kw_set, Keywords.kw_async, Keywords.kw_await)) 2886 return false; // Otherwise automatic semicolon insertion would trigger. 2887 if (Right.NestingLevel == 0 && 2888 (Left.Tok.getIdentifierInfo() || 2889 Left.isOneOf(tok::r_square, tok::r_paren)) && 2890 Right.isOneOf(tok::l_square, tok::l_paren)) 2891 return false; // Otherwise automatic semicolon insertion would trigger. 2892 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace)) 2893 return false; 2894 if (Left.is(TT_JsTypeColon)) 2895 return true; 2896 if (Right.NestingLevel == 0 && Right.is(Keywords.kw_is)) 2897 return false; 2898 if (Left.is(Keywords.kw_in)) 2899 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 2900 if (Right.is(Keywords.kw_in)) 2901 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 2902 if (Right.is(Keywords.kw_as)) 2903 return false; // must not break before as in 'x as type' casts 2904 if (Left.is(Keywords.kw_as)) 2905 return true; 2906 if (Left.is(TT_JsNonNullAssertion)) 2907 return true; 2908 if (Left.is(Keywords.kw_declare) && 2909 Right.isOneOf(Keywords.kw_module, tok::kw_namespace, 2910 Keywords.kw_function, tok::kw_class, tok::kw_enum, 2911 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var, 2912 Keywords.kw_let, tok::kw_const)) 2913 // See grammar for 'declare' statements at: 2914 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10 2915 return false; 2916 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) && 2917 Right.isOneOf(tok::identifier, tok::string_literal)) 2918 return false; // must not break in "module foo { ...}" 2919 if (Right.is(TT_TemplateString) && Right.closesScope()) 2920 return false; 2921 if (Left.is(TT_TemplateString) && Left.opensScope()) 2922 return true; 2923 } 2924 2925 if (Left.is(tok::at)) 2926 return false; 2927 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 2928 return false; 2929 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 2930 return !Right.is(tok::l_paren); 2931 if (Right.is(TT_PointerOrReference)) 2932 return Line.IsMultiVariableDeclStmt || 2933 (Style.PointerAlignment == FormatStyle::PAS_Right && 2934 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 2935 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2936 Right.is(tok::kw_operator)) 2937 return true; 2938 if (Left.is(TT_PointerOrReference)) 2939 return false; 2940 if (Right.isTrailingComment()) 2941 // We rely on MustBreakBefore being set correctly here as we should not 2942 // change the "binding" behavior of a comment. 2943 // The first comment in a braced lists is always interpreted as belonging to 2944 // the first list element. Otherwise, it should be placed outside of the 2945 // list. 2946 return Left.BlockKind == BK_BracedInit || 2947 (Left.is(TT_CtorInitializerColon) && 2948 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon); 2949 if (Left.is(tok::question) && Right.is(tok::colon)) 2950 return false; 2951 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 2952 return Style.BreakBeforeTernaryOperators; 2953 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 2954 return !Style.BreakBeforeTernaryOperators; 2955 if (Right.is(TT_InheritanceColon)) 2956 return true; 2957 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) && 2958 Left.isNot(TT_SelectorName)) 2959 return true; 2960 2961 if (Right.is(tok::colon) && 2962 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 2963 return false; 2964 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 2965 if ((Style.Language == FormatStyle::LK_Proto || 2966 Style.Language == FormatStyle::LK_TextProto) && 2967 !Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral()) 2968 return false; 2969 return true; 2970 } 2971 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 2972 Right.Next->is(TT_ObjCMethodExpr))) 2973 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 2974 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 2975 return true; 2976 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 2977 return true; 2978 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 2979 TT_OverloadedOperator)) 2980 return false; 2981 if (Left.is(TT_RangeBasedForLoopColon)) 2982 return true; 2983 if (Right.is(TT_RangeBasedForLoopColon)) 2984 return false; 2985 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener)) 2986 return true; 2987 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 2988 Left.is(tok::kw_operator)) 2989 return false; 2990 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 2991 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 2992 return false; 2993 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen)) 2994 return false; 2995 if (Left.is(tok::l_paren) && Left.Previous && 2996 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 2997 return false; 2998 if (Right.is(TT_ImplicitStringLiteral)) 2999 return false; 3000 3001 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser)) 3002 return false; 3003 if (Right.is(tok::r_square) && Right.MatchingParen && 3004 Right.MatchingParen->is(TT_LambdaLSquare)) 3005 return false; 3006 3007 // We only break before r_brace if there was a corresponding break before 3008 // the l_brace, which is tracked by BreakBeforeClosingBrace. 3009 if (Right.is(tok::r_brace)) 3010 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block; 3011 3012 // Allow breaking after a trailing annotation, e.g. after a method 3013 // declaration. 3014 if (Left.is(TT_TrailingAnnotation)) 3015 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 3016 tok::less, tok::coloncolon); 3017 3018 if (Right.is(tok::kw___attribute) || 3019 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare))) 3020 return true; 3021 3022 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 3023 return true; 3024 3025 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 3026 return true; 3027 3028 if (Left.is(TT_CtorInitializerColon)) 3029 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 3030 if (Right.is(TT_CtorInitializerColon)) 3031 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon; 3032 if (Left.is(TT_CtorInitializerComma) && 3033 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 3034 return false; 3035 if (Right.is(TT_CtorInitializerComma) && 3036 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 3037 return true; 3038 if (Left.is(TT_InheritanceComma) && Style.BreakBeforeInheritanceComma) 3039 return false; 3040 if (Right.is(TT_InheritanceComma) && Style.BreakBeforeInheritanceComma) 3041 return true; 3042 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 3043 (Left.is(tok::less) && Right.is(tok::less))) 3044 return false; 3045 if (Right.is(TT_BinaryOperator) && 3046 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 3047 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 3048 Right.getPrecedence() != prec::Assignment)) 3049 return true; 3050 if (Left.is(TT_ArrayInitializerLSquare)) 3051 return true; 3052 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 3053 return true; 3054 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 3055 !Left.isOneOf(tok::arrowstar, tok::lessless) && 3056 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 3057 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 3058 Left.getPrecedence() == prec::Assignment)) 3059 return true; 3060 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) || 3061 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare))) 3062 return false; 3063 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 3064 tok::kw_class, tok::kw_struct, tok::comment) || 3065 Right.isMemberAccess() || 3066 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 3067 tok::colon, tok::l_square, tok::at) || 3068 (Left.is(tok::r_paren) && 3069 Right.isOneOf(tok::identifier, tok::kw_const)) || 3070 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) || 3071 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser)); 3072 } 3073 3074 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 3075 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n"; 3076 const FormatToken *Tok = Line.First; 3077 while (Tok) { 3078 llvm::errs() << " M=" << Tok->MustBreakBefore 3079 << " C=" << Tok->CanBreakBefore 3080 << " T=" << getTokenTypeName(Tok->Type) 3081 << " S=" << Tok->SpacesRequiredBefore 3082 << " B=" << Tok->BlockParameterCount 3083 << " BK=" << Tok->BlockKind << " P=" << Tok->SplitPenalty 3084 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength 3085 << " PPK=" << Tok->PackingKind << " FakeLParens="; 3086 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 3087 llvm::errs() << Tok->FakeLParens[i] << "/"; 3088 llvm::errs() << " FakeRParens=" << Tok->FakeRParens; 3089 llvm::errs() << " Text='" << Tok->TokenText << "'\n"; 3090 if (!Tok->Next) 3091 assert(Tok == Line.Last); 3092 Tok = Tok->Next; 3093 } 3094 llvm::errs() << "----\n"; 3095 } 3096 3097 } // namespace format 3098 } // namespace clang 3099