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; 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 CurrentToken->Type = TT_TemplateCloser; 83 next(); 84 return true; 85 } 86 if (CurrentToken->is(tok::question) && 87 Style.Language == FormatStyle::LK_Java) { 88 next(); 89 continue; 90 } 91 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) || 92 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext)) 93 return false; 94 // If a && or || is found and interpreted as a binary operator, this set 95 // of angles is likely part of something like "a < b && c > d". If the 96 // angles are inside an expression, the ||/&& might also be a binary 97 // operator that was misinterpreted because we are parsing template 98 // parameters. 99 // FIXME: This is getting out of hand, write a decent parser. 100 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) && 101 CurrentToken->Previous->is(TT_BinaryOperator) && 102 Contexts[Contexts.size() - 2].IsExpression && 103 !Line.startsWith(tok::kw_template)) 104 return false; 105 updateParameterCount(Left, CurrentToken); 106 if (!consumeToken()) 107 return false; 108 } 109 return false; 110 } 111 112 bool parseParens(bool LookForDecls = false) { 113 if (!CurrentToken) 114 return false; 115 FormatToken *Left = CurrentToken->Previous; 116 Left->ParentBracket = Contexts.back().ContextKind; 117 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1); 118 119 // FIXME: This is a bit of a hack. Do better. 120 Contexts.back().ColonIsForRangeExpr = 121 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr; 122 123 bool StartsObjCMethodExpr = false; 124 if (CurrentToken->is(tok::caret)) { 125 // (^ can start a block type. 126 Left->Type = TT_ObjCBlockLParen; 127 } else if (FormatToken *MaybeSel = Left->Previous) { 128 // @selector( starts a selector. 129 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous && 130 MaybeSel->Previous->is(tok::at)) { 131 StartsObjCMethodExpr = true; 132 } 133 } 134 135 if (Left->is(TT_OverloadedOperatorLParen)) { 136 Contexts.back().IsExpression = false; 137 } else if (Left->Previous && 138 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_decltype, 139 tok::kw_if, tok::kw_while, tok::l_paren, 140 tok::comma) || 141 Left->Previous->is(TT_BinaryOperator))) { 142 // static_assert, if and while usually contain expressions. 143 Contexts.back().IsExpression = true; 144 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous && 145 Left->Previous->is(Keywords.kw_function)) { 146 Contexts.back().IsExpression = false; 147 } else if (Left->Previous && Left->Previous->is(tok::r_square) && 148 Left->Previous->MatchingParen && 149 Left->Previous->MatchingParen->is(TT_LambdaLSquare)) { 150 // This is a parameter list of a lambda expression. 151 Contexts.back().IsExpression = false; 152 } else if (Line.InPPDirective && 153 (!Left->Previous || !Left->Previous->is(tok::identifier))) { 154 Contexts.back().IsExpression = true; 155 } else if (Contexts[Contexts.size() - 2].CaretFound) { 156 // This is the parameter list of an ObjC block. 157 Contexts.back().IsExpression = false; 158 } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) { 159 Left->Type = TT_AttributeParen; 160 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) { 161 // The first argument to a foreach macro is a declaration. 162 Contexts.back().IsForEachMacro = true; 163 Contexts.back().IsExpression = false; 164 } else if (Left->Previous && Left->Previous->MatchingParen && 165 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) { 166 Contexts.back().IsExpression = false; 167 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) { 168 bool IsForOrCatch = 169 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch); 170 Contexts.back().IsExpression = !IsForOrCatch; 171 } 172 173 if (StartsObjCMethodExpr) { 174 Contexts.back().ColonIsObjCMethodExpr = true; 175 Left->Type = TT_ObjCMethodExpr; 176 } 177 178 bool MightBeFunctionType = CurrentToken->isOneOf(tok::star, tok::amp) && 179 !Contexts[Contexts.size() - 2].IsExpression; 180 bool HasMultipleLines = false; 181 bool HasMultipleParametersOnALine = false; 182 bool MightBeObjCForRangeLoop = 183 Left->Previous && Left->Previous->is(tok::kw_for); 184 while (CurrentToken) { 185 // LookForDecls is set when "if (" has been seen. Check for 186 // 'identifier' '*' 'identifier' followed by not '=' -- this 187 // '*' has to be a binary operator but determineStarAmpUsage() will 188 // categorize it as an unary operator, so set the right type here. 189 if (LookForDecls && CurrentToken->Next) { 190 FormatToken *Prev = CurrentToken->getPreviousNonComment(); 191 if (Prev) { 192 FormatToken *PrevPrev = Prev->getPreviousNonComment(); 193 FormatToken *Next = CurrentToken->Next; 194 if (PrevPrev && PrevPrev->is(tok::identifier) && 195 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) && 196 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) { 197 Prev->Type = TT_BinaryOperator; 198 LookForDecls = false; 199 } 200 } 201 } 202 203 if (CurrentToken->Previous->is(TT_PointerOrReference) && 204 CurrentToken->Previous->Previous->isOneOf(tok::l_paren, 205 tok::coloncolon)) 206 MightBeFunctionType = true; 207 if (CurrentToken->Previous->is(TT_BinaryOperator)) 208 Contexts.back().IsExpression = true; 209 if (CurrentToken->is(tok::r_paren)) { 210 if (MightBeFunctionType && CurrentToken->Next && 211 (CurrentToken->Next->is(tok::l_paren) || 212 (CurrentToken->Next->is(tok::l_square) && 213 Line.MustBeDeclaration))) 214 Left->Type = TT_FunctionTypeLParen; 215 Left->MatchingParen = CurrentToken; 216 CurrentToken->MatchingParen = Left; 217 218 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) && 219 Left->Previous && Left->Previous->is(tok::l_paren)) { 220 // Detect the case where macros are used to generate lambdas or 221 // function bodies, e.g.: 222 // auto my_lambda = MARCO((Type *type, int i) { .. body .. }); 223 for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) { 224 if (Tok->is(TT_BinaryOperator) && 225 Tok->isOneOf(tok::star, tok::amp, tok::ampamp)) 226 Tok->Type = TT_PointerOrReference; 227 } 228 } 229 230 if (StartsObjCMethodExpr) { 231 CurrentToken->Type = TT_ObjCMethodExpr; 232 if (Contexts.back().FirstObjCSelectorName) { 233 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 234 Contexts.back().LongestObjCSelectorName; 235 } 236 } 237 238 if (Left->is(TT_AttributeParen)) 239 CurrentToken->Type = TT_AttributeParen; 240 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation)) 241 CurrentToken->Type = TT_JavaAnnotation; 242 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation)) 243 CurrentToken->Type = TT_LeadingJavaAnnotation; 244 245 if (!HasMultipleLines) 246 Left->PackingKind = PPK_Inconclusive; 247 else if (HasMultipleParametersOnALine) 248 Left->PackingKind = PPK_BinPacked; 249 else 250 Left->PackingKind = PPK_OnePerLine; 251 252 next(); 253 return true; 254 } 255 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace)) 256 return false; 257 258 if (CurrentToken->is(tok::l_brace)) 259 Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen 260 if (CurrentToken->is(tok::comma) && CurrentToken->Next && 261 !CurrentToken->Next->HasUnescapedNewline && 262 !CurrentToken->Next->isTrailingComment()) 263 HasMultipleParametersOnALine = true; 264 if (CurrentToken->isOneOf(tok::kw_const, tok::kw_auto) || 265 CurrentToken->isSimpleTypeSpecifier()) 266 Contexts.back().IsExpression = false; 267 if (CurrentToken->isOneOf(tok::semi, tok::colon)) 268 MightBeObjCForRangeLoop = false; 269 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) 270 CurrentToken->Type = TT_ObjCForIn; 271 // When we discover a 'new', we set CanBeExpression to 'false' in order to 272 // parse the type correctly. Reset that after a comma. 273 if (CurrentToken->is(tok::comma)) 274 Contexts.back().CanBeExpression = true; 275 276 FormatToken *Tok = CurrentToken; 277 if (!consumeToken()) 278 return false; 279 updateParameterCount(Left, Tok); 280 if (CurrentToken && CurrentToken->HasUnescapedNewline) 281 HasMultipleLines = true; 282 } 283 return false; 284 } 285 286 bool parseSquare() { 287 if (!CurrentToken) 288 return false; 289 290 // A '[' could be an index subscript (after an identifier or after 291 // ')' or ']'), it could be the start of an Objective-C method 292 // expression, or it could the start of an Objective-C array literal. 293 FormatToken *Left = CurrentToken->Previous; 294 Left->ParentBracket = Contexts.back().ContextKind; 295 FormatToken *Parent = Left->getPreviousNonComment(); 296 bool StartsObjCMethodExpr = 297 Style.Language == FormatStyle::LK_Cpp && 298 Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) && 299 CurrentToken->isNot(tok::l_brace) && 300 (!Parent || 301 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren, 302 tok::kw_return, tok::kw_throw) || 303 Parent->isUnaryOperator() || 304 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) || 305 getBinOpPrecedence(Parent->Tok.getKind(), true, true) > prec::Unknown); 306 bool ColonFound = false; 307 308 unsigned BindingIncrease = 1; 309 if (Left->is(TT_Unknown)) { 310 if (StartsObjCMethodExpr) { 311 Left->Type = TT_ObjCMethodExpr; 312 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent && 313 Contexts.back().ContextKind == tok::l_brace && 314 Parent->isOneOf(tok::l_brace, tok::comma)) { 315 Left->Type = TT_JsComputedPropertyName; 316 } else if (Style.Language == FormatStyle::LK_Proto || 317 (Parent && 318 Parent->isOneOf(TT_BinaryOperator, tok::at, tok::comma, 319 tok::l_paren, tok::l_square, tok::question, 320 tok::colon, tok::kw_return, 321 // Should only be relevant to JavaScript: 322 tok::kw_default))) { 323 Left->Type = TT_ArrayInitializerLSquare; 324 } else { 325 BindingIncrease = 10; 326 Left->Type = TT_ArraySubscriptLSquare; 327 } 328 } 329 330 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease); 331 Contexts.back().IsExpression = true; 332 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr; 333 334 while (CurrentToken) { 335 if (CurrentToken->is(tok::r_square)) { 336 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_paren) && 337 Left->is(TT_ObjCMethodExpr)) { 338 // An ObjC method call is rarely followed by an open parenthesis. 339 // FIXME: Do we incorrectly label ":" with this? 340 StartsObjCMethodExpr = false; 341 Left->Type = TT_Unknown; 342 } 343 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) { 344 CurrentToken->Type = TT_ObjCMethodExpr; 345 // determineStarAmpUsage() thinks that '*' '[' is allocating an 346 // array of pointers, but if '[' starts a selector then '*' is a 347 // binary operator. 348 if (Parent && Parent->is(TT_PointerOrReference)) 349 Parent->Type = TT_BinaryOperator; 350 } 351 Left->MatchingParen = CurrentToken; 352 CurrentToken->MatchingParen = Left; 353 if (Contexts.back().FirstObjCSelectorName) { 354 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 355 Contexts.back().LongestObjCSelectorName; 356 if (Left->BlockParameterCount > 1) 357 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0; 358 } 359 next(); 360 return true; 361 } 362 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace)) 363 return false; 364 if (CurrentToken->is(tok::colon)) { 365 if (Left->is(TT_ArraySubscriptLSquare)) { 366 Left->Type = TT_ObjCMethodExpr; 367 StartsObjCMethodExpr = true; 368 Contexts.back().ColonIsObjCMethodExpr = true; 369 if (Parent && Parent->is(tok::r_paren)) 370 Parent->Type = TT_CastRParen; 371 } 372 ColonFound = true; 373 } 374 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) && 375 !ColonFound) 376 Left->Type = TT_ArrayInitializerLSquare; 377 FormatToken *Tok = CurrentToken; 378 if (!consumeToken()) 379 return false; 380 updateParameterCount(Left, Tok); 381 } 382 return false; 383 } 384 385 bool parseBrace() { 386 if (CurrentToken) { 387 FormatToken *Left = CurrentToken->Previous; 388 Left->ParentBracket = Contexts.back().ContextKind; 389 390 if (Contexts.back().CaretFound) 391 Left->Type = TT_ObjCBlockLBrace; 392 Contexts.back().CaretFound = false; 393 394 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1); 395 Contexts.back().ColonIsDictLiteral = true; 396 if (Left->BlockKind == BK_BracedInit) 397 Contexts.back().IsExpression = true; 398 399 while (CurrentToken) { 400 if (CurrentToken->is(tok::r_brace)) { 401 Left->MatchingParen = CurrentToken; 402 CurrentToken->MatchingParen = Left; 403 next(); 404 return true; 405 } 406 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square)) 407 return false; 408 updateParameterCount(Left, CurrentToken); 409 if (CurrentToken->isOneOf(tok::colon, tok::l_brace)) { 410 FormatToken *Previous = CurrentToken->getPreviousNonComment(); 411 if (((CurrentToken->is(tok::colon) && 412 (!Contexts.back().ColonIsDictLiteral || 413 Style.Language != FormatStyle::LK_Cpp)) || 414 Style.Language == FormatStyle::LK_Proto) && 415 (Previous->Tok.getIdentifierInfo() || 416 Previous->is(tok::string_literal))) 417 Previous->Type = TT_SelectorName; 418 if (CurrentToken->is(tok::colon) || 419 Style.Language == FormatStyle::LK_JavaScript) 420 Left->Type = TT_DictLiteral; 421 } 422 if (!consumeToken()) 423 return false; 424 } 425 } 426 return true; 427 } 428 429 void updateParameterCount(FormatToken *Left, FormatToken *Current) { 430 if (Current->is(tok::l_brace) && Current->BlockKind == BK_Block) 431 ++Left->BlockParameterCount; 432 if (Current->is(tok::comma)) { 433 ++Left->ParameterCount; 434 if (!Left->Role) 435 Left->Role.reset(new CommaSeparatedList(Style)); 436 Left->Role->CommaFound(Current); 437 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) { 438 Left->ParameterCount = 1; 439 } 440 } 441 442 bool parseConditional() { 443 while (CurrentToken) { 444 if (CurrentToken->is(tok::colon)) { 445 CurrentToken->Type = TT_ConditionalExpr; 446 next(); 447 return true; 448 } 449 if (!consumeToken()) 450 return false; 451 } 452 return false; 453 } 454 455 bool parseTemplateDeclaration() { 456 if (CurrentToken && CurrentToken->is(tok::less)) { 457 CurrentToken->Type = TT_TemplateOpener; 458 next(); 459 if (!parseAngle()) 460 return false; 461 if (CurrentToken) 462 CurrentToken->Previous->ClosesTemplateDeclaration = true; 463 return true; 464 } 465 return false; 466 } 467 468 bool consumeToken() { 469 FormatToken *Tok = CurrentToken; 470 next(); 471 switch (Tok->Tok.getKind()) { 472 case tok::plus: 473 case tok::minus: 474 if (!Tok->Previous && Line.MustBeDeclaration) 475 Tok->Type = TT_ObjCMethodSpecifier; 476 break; 477 case tok::colon: 478 if (!Tok->Previous) 479 return false; 480 // Colons from ?: are handled in parseConditional(). 481 if (Style.Language == FormatStyle::LK_JavaScript) { 482 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop 483 (Contexts.size() == 1 && // switch/case labels 484 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) || 485 Contexts.back().ContextKind == tok::l_paren || // function params 486 Contexts.back().ContextKind == tok::l_square || // array type 487 (Contexts.size() == 1 && 488 Line.MustBeDeclaration)) { // method/property declaration 489 Tok->Type = TT_JsTypeColon; 490 break; 491 } 492 } 493 if (Contexts.back().ColonIsDictLiteral || 494 Style.Language == FormatStyle::LK_Proto) { 495 Tok->Type = TT_DictLiteral; 496 } else if (Contexts.back().ColonIsObjCMethodExpr || 497 Line.startsWith(TT_ObjCMethodSpecifier)) { 498 Tok->Type = TT_ObjCMethodExpr; 499 Tok->Previous->Type = TT_SelectorName; 500 if (Tok->Previous->ColumnWidth > 501 Contexts.back().LongestObjCSelectorName) 502 Contexts.back().LongestObjCSelectorName = Tok->Previous->ColumnWidth; 503 if (!Contexts.back().FirstObjCSelectorName) 504 Contexts.back().FirstObjCSelectorName = Tok->Previous; 505 } else if (Contexts.back().ColonIsForRangeExpr) { 506 Tok->Type = TT_RangeBasedForLoopColon; 507 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) { 508 Tok->Type = TT_BitFieldColon; 509 } else if (Contexts.size() == 1 && 510 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) { 511 if (Tok->Previous->isOneOf(tok::r_paren, tok::kw_noexcept)) 512 Tok->Type = TT_CtorInitializerColon; 513 else 514 Tok->Type = TT_InheritanceColon; 515 } else if (Tok->Previous->is(tok::identifier) && Tok->Next && 516 Tok->Next->isOneOf(tok::r_paren, tok::comma)) { 517 // This handles a special macro in ObjC code where selectors including 518 // the colon are passed as macro arguments. 519 Tok->Type = TT_ObjCMethodExpr; 520 } else if (Contexts.back().ContextKind == tok::l_paren) { 521 Tok->Type = TT_InlineASMColon; 522 } 523 break; 524 case tok::pipe: 525 case tok::amp: 526 // | and & in declarations/type expressions represent union and 527 // intersection types, respectively. 528 if (Style.Language == FormatStyle::LK_JavaScript && 529 !Contexts.back().IsExpression) 530 Tok->Type = TT_JsTypeOperator; 531 break; 532 case tok::kw_if: 533 case tok::kw_while: 534 if (CurrentToken && CurrentToken->is(tok::l_paren)) { 535 next(); 536 if (!parseParens(/*LookForDecls=*/true)) 537 return false; 538 } 539 break; 540 case tok::kw_for: 541 Contexts.back().ColonIsForRangeExpr = true; 542 next(); 543 if (!parseParens()) 544 return false; 545 break; 546 case tok::l_paren: 547 // When faced with 'operator()()', the kw_operator handler incorrectly 548 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make 549 // the first two parens OverloadedOperators and the second l_paren an 550 // OverloadedOperatorLParen. 551 if (Tok->Previous && 552 Tok->Previous->is(tok::r_paren) && 553 Tok->Previous->MatchingParen && 554 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) { 555 Tok->Previous->Type = TT_OverloadedOperator; 556 Tok->Previous->MatchingParen->Type = TT_OverloadedOperator; 557 Tok->Type = TT_OverloadedOperatorLParen; 558 } 559 560 if (!parseParens()) 561 return false; 562 if (Line.MustBeDeclaration && Contexts.size() == 1 && 563 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) && 564 (!Tok->Previous || 565 !Tok->Previous->isOneOf(tok::kw_decltype, tok::kw___attribute, 566 TT_LeadingJavaAnnotation))) 567 Line.MightBeFunctionDecl = true; 568 break; 569 case tok::l_square: 570 if (!parseSquare()) 571 return false; 572 break; 573 case tok::l_brace: 574 if (!parseBrace()) 575 return false; 576 break; 577 case tok::less: 578 if (parseAngle()) { 579 Tok->Type = TT_TemplateOpener; 580 } else { 581 Tok->Type = TT_BinaryOperator; 582 NonTemplateLess.insert(Tok); 583 CurrentToken = Tok; 584 next(); 585 } 586 break; 587 case tok::r_paren: 588 case tok::r_square: 589 return false; 590 case tok::r_brace: 591 // Lines can start with '}'. 592 if (Tok->Previous) 593 return false; 594 break; 595 case tok::greater: 596 Tok->Type = TT_BinaryOperator; 597 break; 598 case tok::kw_operator: 599 while (CurrentToken && 600 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) { 601 if (CurrentToken->isOneOf(tok::star, tok::amp)) 602 CurrentToken->Type = TT_PointerOrReference; 603 consumeToken(); 604 if (CurrentToken && 605 CurrentToken->Previous->isOneOf(TT_BinaryOperator, tok::comma)) 606 CurrentToken->Previous->Type = TT_OverloadedOperator; 607 } 608 if (CurrentToken) { 609 CurrentToken->Type = TT_OverloadedOperatorLParen; 610 if (CurrentToken->Previous->is(TT_BinaryOperator)) 611 CurrentToken->Previous->Type = TT_OverloadedOperator; 612 } 613 break; 614 case tok::question: 615 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next && 616 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren, 617 tok::r_brace)) { 618 // Question marks before semicolons, colons, etc. indicate optional 619 // types (fields, parameters), e.g. 620 // function(x?: string, y?) {...} 621 // class X { y?; } 622 Tok->Type = TT_JsTypeOptionalQuestion; 623 break; 624 } 625 // Declarations cannot be conditional expressions, this can only be part 626 // of a type declaration. 627 if (Line.MustBeDeclaration && 628 Style.Language == FormatStyle::LK_JavaScript) 629 break; 630 parseConditional(); 631 break; 632 case tok::kw_template: 633 parseTemplateDeclaration(); 634 break; 635 case tok::comma: 636 if (Contexts.back().InCtorInitializer) 637 Tok->Type = TT_CtorInitializerComma; 638 else if (Contexts.back().FirstStartOfName && 639 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) { 640 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true; 641 Line.IsMultiVariableDeclStmt = true; 642 } 643 if (Contexts.back().IsForEachMacro) 644 Contexts.back().IsExpression = true; 645 break; 646 default: 647 break; 648 } 649 return true; 650 } 651 652 void parseIncludeDirective() { 653 if (CurrentToken && CurrentToken->is(tok::less)) { 654 next(); 655 while (CurrentToken) { 656 if (CurrentToken->isNot(tok::comment) || CurrentToken->Next) 657 CurrentToken->Type = TT_ImplicitStringLiteral; 658 next(); 659 } 660 } 661 } 662 663 void parseWarningOrError() { 664 next(); 665 // We still want to format the whitespace left of the first token of the 666 // warning or error. 667 next(); 668 while (CurrentToken) { 669 CurrentToken->Type = TT_ImplicitStringLiteral; 670 next(); 671 } 672 } 673 674 void parsePragma() { 675 next(); // Consume "pragma". 676 if (CurrentToken && 677 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) { 678 bool IsMark = CurrentToken->is(Keywords.kw_mark); 679 next(); // Consume "mark". 680 next(); // Consume first token (so we fix leading whitespace). 681 while (CurrentToken) { 682 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator)) 683 CurrentToken->Type = TT_ImplicitStringLiteral; 684 next(); 685 } 686 } 687 } 688 689 LineType parsePreprocessorDirective() { 690 LineType Type = LT_PreprocessorDirective; 691 next(); 692 if (!CurrentToken) 693 return Type; 694 if (CurrentToken->Tok.is(tok::numeric_constant)) { 695 CurrentToken->SpacesRequiredBefore = 1; 696 return Type; 697 } 698 // Hashes in the middle of a line can lead to any strange token 699 // sequence. 700 if (!CurrentToken->Tok.getIdentifierInfo()) 701 return Type; 702 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) { 703 case tok::pp_include: 704 case tok::pp_include_next: 705 case tok::pp_import: 706 next(); 707 parseIncludeDirective(); 708 Type = LT_ImportStatement; 709 break; 710 case tok::pp_error: 711 case tok::pp_warning: 712 parseWarningOrError(); 713 break; 714 case tok::pp_pragma: 715 parsePragma(); 716 break; 717 case tok::pp_if: 718 case tok::pp_elif: 719 Contexts.back().IsExpression = true; 720 parseLine(); 721 break; 722 default: 723 break; 724 } 725 while (CurrentToken) 726 next(); 727 return Type; 728 } 729 730 public: 731 LineType parseLine() { 732 NonTemplateLess.clear(); 733 if (CurrentToken->is(tok::hash)) 734 return parsePreprocessorDirective(); 735 736 // Directly allow to 'import <string-literal>' to support protocol buffer 737 // definitions (code.google.com/p/protobuf) or missing "#" (either way we 738 // should not break the line). 739 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo(); 740 if ((Style.Language == FormatStyle::LK_Java && 741 CurrentToken->is(Keywords.kw_package)) || 742 (Info && Info->getPPKeywordID() == tok::pp_import && 743 CurrentToken->Next && 744 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier, 745 tok::kw_static))) { 746 next(); 747 parseIncludeDirective(); 748 return LT_ImportStatement; 749 } 750 751 // If this line starts and ends in '<' and '>', respectively, it is likely 752 // part of "#define <a/b.h>". 753 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) { 754 parseIncludeDirective(); 755 return LT_ImportStatement; 756 } 757 758 // In .proto files, top-level options are very similar to import statements 759 // and should not be line-wrapped. 760 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 && 761 CurrentToken->is(Keywords.kw_option)) { 762 next(); 763 if (CurrentToken && CurrentToken->is(tok::identifier)) 764 return LT_ImportStatement; 765 } 766 767 // import {...} from '...'; 768 if (Style.Language == FormatStyle::LK_JavaScript && 769 CurrentToken->is(Keywords.kw_import)) 770 return LT_ImportStatement; 771 772 bool KeywordVirtualFound = false; 773 bool ImportStatement = false; 774 while (CurrentToken) { 775 if (CurrentToken->is(tok::kw_virtual)) 776 KeywordVirtualFound = true; 777 if (Style.Language == FormatStyle::LK_JavaScript) { 778 // export {...} from '...'; 779 // An export followed by "from 'some string';" is a re-export from 780 // another module identified by a URI and is treated as a 781 // LT_ImportStatement (i.e. prevent wraps on it for long URIs). 782 // Just "export {...};" or "export class ..." should not be treated as 783 // an import in this sense. 784 if (Line.First->is(tok::kw_export) && 785 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next && 786 CurrentToken->Next->isStringLiteral()) 787 ImportStatement = true; 788 if (isClosureImportStatement(*CurrentToken)) 789 ImportStatement = true; 790 } 791 if (!consumeToken()) 792 return LT_Invalid; 793 } 794 if (KeywordVirtualFound) 795 return LT_VirtualFunctionDecl; 796 if (ImportStatement) 797 return LT_ImportStatement; 798 799 if (Line.startsWith(TT_ObjCMethodSpecifier)) { 800 if (Contexts.back().FirstObjCSelectorName) 801 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 802 Contexts.back().LongestObjCSelectorName; 803 return LT_ObjCMethodDecl; 804 } 805 806 return LT_Other; 807 } 808 809 private: 810 bool isClosureImportStatement(const FormatToken &Tok) { 811 // FIXME: Closure-library specific stuff should not be hard-coded but be 812 // configurable. 813 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) && 814 Tok.Next->Next && (Tok.Next->Next->TokenText == "module" || 815 Tok.Next->Next->TokenText == "provide" || 816 Tok.Next->Next->TokenText == "require" || 817 Tok.Next->Next->TokenText == "setTestOnly" || 818 Tok.Next->Next->TokenText == "forwardDeclare") && 819 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren); 820 } 821 822 void resetTokenMetadata(FormatToken *Token) { 823 if (!Token) 824 return; 825 826 // Reset token type in case we have already looked at it and then 827 // recovered from an error (e.g. failure to find the matching >). 828 if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_ForEachMacro, 829 TT_FunctionLBrace, TT_ImplicitStringLiteral, 830 TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow, 831 TT_RegexLiteral)) 832 CurrentToken->Type = TT_Unknown; 833 CurrentToken->Role.reset(); 834 CurrentToken->MatchingParen = nullptr; 835 CurrentToken->FakeLParens.clear(); 836 CurrentToken->FakeRParens = 0; 837 } 838 839 void next() { 840 if (CurrentToken) { 841 CurrentToken->NestingLevel = Contexts.size() - 1; 842 CurrentToken->BindingStrength = Contexts.back().BindingStrength; 843 modifyContext(*CurrentToken); 844 determineTokenType(*CurrentToken); 845 CurrentToken = CurrentToken->Next; 846 } 847 848 resetTokenMetadata(CurrentToken); 849 } 850 851 /// \brief A struct to hold information valid in a specific context, e.g. 852 /// a pair of parenthesis. 853 struct Context { 854 Context(tok::TokenKind ContextKind, unsigned BindingStrength, 855 bool IsExpression) 856 : ContextKind(ContextKind), BindingStrength(BindingStrength), 857 IsExpression(IsExpression) {} 858 859 tok::TokenKind ContextKind; 860 unsigned BindingStrength; 861 bool IsExpression; 862 unsigned LongestObjCSelectorName = 0; 863 bool ColonIsForRangeExpr = false; 864 bool ColonIsDictLiteral = false; 865 bool ColonIsObjCMethodExpr = false; 866 FormatToken *FirstObjCSelectorName = nullptr; 867 FormatToken *FirstStartOfName = nullptr; 868 bool CanBeExpression = true; 869 bool InTemplateArgument = false; 870 bool InCtorInitializer = false; 871 bool CaretFound = false; 872 bool IsForEachMacro = false; 873 }; 874 875 /// \brief Puts a new \c Context onto the stack \c Contexts for the lifetime 876 /// of each instance. 877 struct ScopedContextCreator { 878 AnnotatingParser &P; 879 880 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind, 881 unsigned Increase) 882 : P(P) { 883 P.Contexts.push_back(Context(ContextKind, 884 P.Contexts.back().BindingStrength + Increase, 885 P.Contexts.back().IsExpression)); 886 } 887 888 ~ScopedContextCreator() { P.Contexts.pop_back(); } 889 }; 890 891 void modifyContext(const FormatToken &Current) { 892 if (Current.getPrecedence() == prec::Assignment && 893 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) && 894 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) { 895 Contexts.back().IsExpression = true; 896 if (!Line.startsWith(TT_UnaryOperator)) { 897 for (FormatToken *Previous = Current.Previous; 898 Previous && Previous->Previous && 899 !Previous->Previous->isOneOf(tok::comma, tok::semi); 900 Previous = Previous->Previous) { 901 if (Previous->isOneOf(tok::r_square, tok::r_paren)) { 902 Previous = Previous->MatchingParen; 903 if (!Previous) 904 break; 905 } 906 if (Previous->opensScope()) 907 break; 908 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) && 909 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) && 910 Previous->Previous && Previous->Previous->isNot(tok::equal)) 911 Previous->Type = TT_PointerOrReference; 912 } 913 } 914 } else if (Current.is(tok::lessless) && 915 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) { 916 Contexts.back().IsExpression = true; 917 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) { 918 Contexts.back().IsExpression = true; 919 } else if (Current.is(TT_TrailingReturnArrow)) { 920 Contexts.back().IsExpression = false; 921 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) { 922 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java; 923 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) { 924 for (FormatToken *Previous = Current.Previous; 925 Previous && Previous->isOneOf(tok::star, tok::amp); 926 Previous = Previous->Previous) 927 Previous->Type = TT_PointerOrReference; 928 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer) 929 Contexts.back().IsExpression = false; 930 } else if (Current.Previous && 931 Current.Previous->is(TT_CtorInitializerColon)) { 932 Contexts.back().IsExpression = true; 933 Contexts.back().InCtorInitializer = true; 934 } else if (Current.is(tok::kw_new)) { 935 Contexts.back().CanBeExpression = false; 936 } else if (Current.isOneOf(tok::semi, tok::exclaim)) { 937 // This should be the condition or increment in a for-loop. 938 Contexts.back().IsExpression = true; 939 } 940 } 941 942 void determineTokenType(FormatToken &Current) { 943 if (!Current.is(TT_Unknown)) 944 // The token type is already known. 945 return; 946 947 // Line.MightBeFunctionDecl can only be true after the parentheses of a 948 // function declaration have been found. In this case, 'Current' is a 949 // trailing token of this declaration and thus cannot be a name. 950 if (Current.is(Keywords.kw_instanceof)) { 951 Current.Type = TT_BinaryOperator; 952 } else if (isStartOfName(Current) && 953 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) { 954 Contexts.back().FirstStartOfName = &Current; 955 Current.Type = TT_StartOfName; 956 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) { 957 AutoFound = true; 958 } else if (Current.is(tok::arrow) && 959 Style.Language == FormatStyle::LK_Java) { 960 Current.Type = TT_LambdaArrow; 961 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration && 962 Current.NestingLevel == 0) { 963 Current.Type = TT_TrailingReturnArrow; 964 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) { 965 Current.Type = 966 determineStarAmpUsage(Current, Contexts.back().CanBeExpression && 967 Contexts.back().IsExpression, 968 Contexts.back().InTemplateArgument); 969 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) { 970 Current.Type = determinePlusMinusCaretUsage(Current); 971 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret)) 972 Contexts.back().CaretFound = true; 973 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) { 974 Current.Type = determineIncrementUsage(Current); 975 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) { 976 Current.Type = TT_UnaryOperator; 977 } else if (Current.is(tok::question)) { 978 if (Style.Language == FormatStyle::LK_JavaScript && 979 Line.MustBeDeclaration) { 980 // In JavaScript, `interface X { foo?(): bar; }` is an optional method 981 // on the interface, not a ternary expression. 982 Current.Type = TT_JsTypeOptionalQuestion; 983 } else { 984 Current.Type = TT_ConditionalExpr; 985 } 986 } else if (Current.isBinaryOperator() && 987 (!Current.Previous || Current.Previous->isNot(tok::l_square))) { 988 Current.Type = TT_BinaryOperator; 989 } else if (Current.is(tok::comment)) { 990 if (Current.TokenText.startswith("/*")) { 991 if (Current.TokenText.endswith("*/")) 992 Current.Type = TT_BlockComment; 993 else 994 // The lexer has for some reason determined a comment here. But we 995 // cannot really handle it, if it isn't properly terminated. 996 Current.Tok.setKind(tok::unknown); 997 } else { 998 Current.Type = TT_LineComment; 999 } 1000 } else if (Current.is(tok::r_paren)) { 1001 if (rParenEndsCast(Current)) 1002 Current.Type = TT_CastRParen; 1003 if (Current.MatchingParen && Current.Next && 1004 !Current.Next->isBinaryOperator() && 1005 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace, 1006 tok::period, tok::arrow, tok::coloncolon)) 1007 if (FormatToken *BeforeParen = Current.MatchingParen->Previous) 1008 if (BeforeParen->is(tok::identifier) && 1009 BeforeParen->TokenText == BeforeParen->TokenText.upper() && 1010 (!BeforeParen->Previous || 1011 BeforeParen->Previous->ClosesTemplateDeclaration)) 1012 Current.Type = TT_FunctionAnnotationRParen; 1013 } else if (Current.is(tok::at) && Current.Next) { 1014 if (Current.Next->isStringLiteral()) { 1015 Current.Type = TT_ObjCStringLiteral; 1016 } else { 1017 switch (Current.Next->Tok.getObjCKeywordID()) { 1018 case tok::objc_interface: 1019 case tok::objc_implementation: 1020 case tok::objc_protocol: 1021 Current.Type = TT_ObjCDecl; 1022 break; 1023 case tok::objc_property: 1024 Current.Type = TT_ObjCProperty; 1025 break; 1026 default: 1027 break; 1028 } 1029 } 1030 } else if (Current.is(tok::period)) { 1031 FormatToken *PreviousNoComment = Current.getPreviousNonComment(); 1032 if (PreviousNoComment && 1033 PreviousNoComment->isOneOf(tok::comma, tok::l_brace)) 1034 Current.Type = TT_DesignatedInitializerPeriod; 1035 else if (Style.Language == FormatStyle::LK_Java && Current.Previous && 1036 Current.Previous->isOneOf(TT_JavaAnnotation, 1037 TT_LeadingJavaAnnotation)) { 1038 Current.Type = Current.Previous->Type; 1039 } 1040 } else if (Current.isOneOf(tok::identifier, tok::kw_const) && 1041 Current.Previous && 1042 !Current.Previous->isOneOf(tok::equal, tok::at) && 1043 Line.MightBeFunctionDecl && Contexts.size() == 1) { 1044 // Line.MightBeFunctionDecl can only be true after the parentheses of a 1045 // function declaration have been found. 1046 Current.Type = TT_TrailingAnnotation; 1047 } else if ((Style.Language == FormatStyle::LK_Java || 1048 Style.Language == FormatStyle::LK_JavaScript) && 1049 Current.Previous) { 1050 if (Current.Previous->is(tok::at) && 1051 Current.isNot(Keywords.kw_interface)) { 1052 const FormatToken &AtToken = *Current.Previous; 1053 const FormatToken *Previous = AtToken.getPreviousNonComment(); 1054 if (!Previous || Previous->is(TT_LeadingJavaAnnotation)) 1055 Current.Type = TT_LeadingJavaAnnotation; 1056 else 1057 Current.Type = TT_JavaAnnotation; 1058 } else if (Current.Previous->is(tok::period) && 1059 Current.Previous->isOneOf(TT_JavaAnnotation, 1060 TT_LeadingJavaAnnotation)) { 1061 Current.Type = Current.Previous->Type; 1062 } 1063 } 1064 } 1065 1066 /// \brief Take a guess at whether \p Tok starts a name of a function or 1067 /// variable declaration. 1068 /// 1069 /// This is a heuristic based on whether \p Tok is an identifier following 1070 /// something that is likely a type. 1071 bool isStartOfName(const FormatToken &Tok) { 1072 if (Tok.isNot(tok::identifier) || !Tok.Previous) 1073 return false; 1074 1075 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof)) 1076 return false; 1077 if (Style.Language == FormatStyle::LK_JavaScript && 1078 Tok.Previous->is(Keywords.kw_in)) 1079 return false; 1080 1081 // Skip "const" as it does not have an influence on whether this is a name. 1082 FormatToken *PreviousNotConst = Tok.Previous; 1083 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const)) 1084 PreviousNotConst = PreviousNotConst->Previous; 1085 1086 if (!PreviousNotConst) 1087 return false; 1088 1089 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) && 1090 PreviousNotConst->Previous && 1091 PreviousNotConst->Previous->is(tok::hash); 1092 1093 if (PreviousNotConst->is(TT_TemplateCloser)) 1094 return PreviousNotConst && PreviousNotConst->MatchingParen && 1095 PreviousNotConst->MatchingParen->Previous && 1096 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) && 1097 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template); 1098 1099 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen && 1100 PreviousNotConst->MatchingParen->Previous && 1101 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype)) 1102 return true; 1103 1104 return (!IsPPKeyword && 1105 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) || 1106 PreviousNotConst->is(TT_PointerOrReference) || 1107 PreviousNotConst->isSimpleTypeSpecifier(); 1108 } 1109 1110 /// \brief Determine whether ')' is ending a cast. 1111 bool rParenEndsCast(const FormatToken &Tok) { 1112 // C-style casts are only used in C++ and Java. 1113 if (Style.Language != FormatStyle::LK_Cpp && 1114 Style.Language != FormatStyle::LK_Java) 1115 return false; 1116 1117 // Empty parens aren't casts and there are no casts at the end of the line. 1118 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen) 1119 return false; 1120 1121 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment(); 1122 if (LeftOfParens) { 1123 // If there is an opening parenthesis left of the current parentheses, 1124 // look past it as these might be chained casts. 1125 if (LeftOfParens->is(tok::r_paren)) { 1126 if (!LeftOfParens->MatchingParen || 1127 !LeftOfParens->MatchingParen->Previous) 1128 return false; 1129 LeftOfParens = LeftOfParens->MatchingParen->Previous; 1130 } 1131 1132 // If there is an identifier (or with a few exceptions a keyword) right 1133 // before the parentheses, this is unlikely to be a cast. 1134 if (LeftOfParens->Tok.getIdentifierInfo() && 1135 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case, 1136 tok::kw_delete)) 1137 return false; 1138 1139 // Certain other tokens right before the parentheses are also signals that 1140 // this cannot be a cast. 1141 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator, 1142 TT_TemplateCloser)) 1143 return false; 1144 } 1145 1146 if (Tok.Next->is(tok::question)) 1147 return false; 1148 1149 // As Java has no function types, a "(" after the ")" likely means that this 1150 // is a cast. 1151 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren)) 1152 return true; 1153 1154 // If a (non-string) literal follows, this is likely a cast. 1155 if (Tok.Next->isNot(tok::string_literal) && 1156 (Tok.Next->Tok.isLiteral() || 1157 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof))) 1158 return true; 1159 1160 // Heuristically try to determine whether the parentheses contain a type. 1161 bool ParensAreType = 1162 !Tok.Previous || 1163 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) || 1164 Tok.Previous->isSimpleTypeSpecifier(); 1165 bool ParensCouldEndDecl = 1166 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater); 1167 if (ParensAreType && !ParensCouldEndDecl) 1168 return true; 1169 1170 // At this point, we heuristically assume that there are no casts at the 1171 // start of the line. We assume that we have found most cases where there 1172 // are by the logic above, e.g. "(void)x;". 1173 if (!LeftOfParens) 1174 return false; 1175 1176 // If the following token is an identifier or 'this', this is a cast. All 1177 // cases where this can be something else are handled above. 1178 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this)) 1179 return true; 1180 1181 if (!Tok.Next->Next) 1182 return false; 1183 1184 // If the next token after the parenthesis is a unary operator, assume 1185 // that this is cast, unless there are unexpected tokens inside the 1186 // parenthesis. 1187 bool NextIsUnary = 1188 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star); 1189 if (!NextIsUnary || Tok.Next->is(tok::plus) || 1190 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant)) 1191 return false; 1192 // Search for unexpected tokens. 1193 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen; 1194 Prev = Prev->Previous) { 1195 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) 1196 return false; 1197 } 1198 return true; 1199 } 1200 1201 /// \brief Return the type of the given token assuming it is * or &. 1202 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression, 1203 bool InTemplateArgument) { 1204 if (Style.Language == FormatStyle::LK_JavaScript) 1205 return TT_BinaryOperator; 1206 1207 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1208 if (!PrevToken) 1209 return TT_UnaryOperator; 1210 1211 const FormatToken *NextToken = Tok.getNextNonComment(); 1212 if (!NextToken || 1213 NextToken->isOneOf(tok::arrow, Keywords.kw_final, 1214 Keywords.kw_override) || 1215 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment())) 1216 return TT_PointerOrReference; 1217 1218 if (PrevToken->is(tok::coloncolon)) 1219 return TT_PointerOrReference; 1220 1221 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace, 1222 tok::comma, tok::semi, tok::kw_return, tok::colon, 1223 tok::equal, tok::kw_delete, tok::kw_sizeof) || 1224 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 1225 TT_UnaryOperator, TT_CastRParen)) 1226 return TT_UnaryOperator; 1227 1228 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare)) 1229 return TT_PointerOrReference; 1230 if (NextToken->is(tok::kw_operator) && !IsExpression) 1231 return TT_PointerOrReference; 1232 if (NextToken->isOneOf(tok::comma, tok::semi)) 1233 return TT_PointerOrReference; 1234 1235 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen && 1236 PrevToken->MatchingParen->Previous && 1237 PrevToken->MatchingParen->Previous->isOneOf(tok::kw_typeof, 1238 tok::kw_decltype)) 1239 return TT_PointerOrReference; 1240 1241 if (PrevToken->Tok.isLiteral() || 1242 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true, 1243 tok::kw_false, tok::r_brace) || 1244 NextToken->Tok.isLiteral() || 1245 NextToken->isOneOf(tok::kw_true, tok::kw_false) || 1246 NextToken->isUnaryOperator() || 1247 // If we know we're in a template argument, there are no named 1248 // declarations. Thus, having an identifier on the right-hand side 1249 // indicates a binary operator. 1250 (InTemplateArgument && NextToken->Tok.isAnyIdentifier())) 1251 return TT_BinaryOperator; 1252 1253 // "&&(" is quite unlikely to be two successive unary "&". 1254 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren)) 1255 return TT_BinaryOperator; 1256 1257 // This catches some cases where evaluation order is used as control flow: 1258 // aaa && aaa->f(); 1259 const FormatToken *NextNextToken = NextToken->getNextNonComment(); 1260 if (NextNextToken && NextNextToken->is(tok::arrow)) 1261 return TT_BinaryOperator; 1262 1263 // It is very unlikely that we are going to find a pointer or reference type 1264 // definition on the RHS of an assignment. 1265 if (IsExpression && !Contexts.back().CaretFound) 1266 return TT_BinaryOperator; 1267 1268 return TT_PointerOrReference; 1269 } 1270 1271 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) { 1272 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1273 if (!PrevToken || PrevToken->is(TT_CastRParen)) 1274 return TT_UnaryOperator; 1275 1276 // Use heuristics to recognize unary operators. 1277 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square, 1278 tok::question, tok::colon, tok::kw_return, 1279 tok::kw_case, tok::at, tok::l_brace)) 1280 return TT_UnaryOperator; 1281 1282 // There can't be two consecutive binary operators. 1283 if (PrevToken->is(TT_BinaryOperator)) 1284 return TT_UnaryOperator; 1285 1286 // Fall back to marking the token as binary operator. 1287 return TT_BinaryOperator; 1288 } 1289 1290 /// \brief Determine whether ++/-- are pre- or post-increments/-decrements. 1291 TokenType determineIncrementUsage(const FormatToken &Tok) { 1292 const FormatToken *PrevToken = Tok.getPreviousNonComment(); 1293 if (!PrevToken || PrevToken->is(TT_CastRParen)) 1294 return TT_UnaryOperator; 1295 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier)) 1296 return TT_TrailingUnaryOperator; 1297 1298 return TT_UnaryOperator; 1299 } 1300 1301 SmallVector<Context, 8> Contexts; 1302 1303 const FormatStyle &Style; 1304 AnnotatedLine &Line; 1305 FormatToken *CurrentToken; 1306 bool AutoFound; 1307 const AdditionalKeywords &Keywords; 1308 1309 // Set of "<" tokens that do not open a template parameter list. If parseAngle 1310 // determines that a specific token can't be a template opener, it will make 1311 // same decision irrespective of the decisions for tokens leading up to it. 1312 // Store this information to prevent this from causing exponential runtime. 1313 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess; 1314 }; 1315 1316 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1; 1317 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2; 1318 1319 /// \brief Parses binary expressions by inserting fake parenthesis based on 1320 /// operator precedence. 1321 class ExpressionParser { 1322 public: 1323 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords, 1324 AnnotatedLine &Line) 1325 : Style(Style), Keywords(Keywords), Current(Line.First) {} 1326 1327 /// \brief Parse expressions with the given operatore precedence. 1328 void parse(int Precedence = 0) { 1329 // Skip 'return' and ObjC selector colons as they are not part of a binary 1330 // expression. 1331 while (Current && (Current->is(tok::kw_return) || 1332 (Current->is(tok::colon) && 1333 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)))) 1334 next(); 1335 1336 if (!Current || Precedence > PrecedenceArrowAndPeriod) 1337 return; 1338 1339 // Conditional expressions need to be parsed separately for proper nesting. 1340 if (Precedence == prec::Conditional) { 1341 parseConditionalExpr(); 1342 return; 1343 } 1344 1345 // Parse unary operators, which all have a higher precedence than binary 1346 // operators. 1347 if (Precedence == PrecedenceUnaryOperator) { 1348 parseUnaryOperator(); 1349 return; 1350 } 1351 1352 FormatToken *Start = Current; 1353 FormatToken *LatestOperator = nullptr; 1354 unsigned OperatorIndex = 0; 1355 1356 while (Current) { 1357 // Consume operators with higher precedence. 1358 parse(Precedence + 1); 1359 1360 int CurrentPrecedence = getCurrentPrecedence(); 1361 1362 if (Current && Current->is(TT_SelectorName) && 1363 Precedence == CurrentPrecedence) { 1364 if (LatestOperator) 1365 addFakeParenthesis(Start, prec::Level(Precedence)); 1366 Start = Current; 1367 } 1368 1369 // At the end of the line or when an operator with higher precedence is 1370 // found, insert fake parenthesis and return. 1371 if (!Current || (Current->closesScope() && Current->MatchingParen) || 1372 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) || 1373 (CurrentPrecedence == prec::Conditional && 1374 Precedence == prec::Assignment && Current->is(tok::colon))) { 1375 break; 1376 } 1377 1378 // Consume scopes: (), [], <> and {} 1379 if (Current->opensScope()) { 1380 while (Current && !Current->closesScope()) { 1381 next(); 1382 parse(); 1383 } 1384 next(); 1385 } else { 1386 // Operator found. 1387 if (CurrentPrecedence == Precedence) { 1388 if (LatestOperator) 1389 LatestOperator->NextOperator = Current; 1390 LatestOperator = Current; 1391 Current->OperatorIndex = OperatorIndex; 1392 ++OperatorIndex; 1393 } 1394 next(/*SkipPastLeadingComments=*/Precedence > 0); 1395 } 1396 } 1397 1398 if (LatestOperator && (Current || Precedence > 0)) { 1399 // LatestOperator->LastOperator = true; 1400 if (Precedence == PrecedenceArrowAndPeriod) { 1401 // Call expressions don't have a binary operator precedence. 1402 addFakeParenthesis(Start, prec::Unknown); 1403 } else { 1404 addFakeParenthesis(Start, prec::Level(Precedence)); 1405 } 1406 } 1407 } 1408 1409 private: 1410 /// \brief Gets the precedence (+1) of the given token for binary operators 1411 /// and other tokens that we treat like binary operators. 1412 int getCurrentPrecedence() { 1413 if (Current) { 1414 const FormatToken *NextNonComment = Current->getNextNonComment(); 1415 if (Current->is(TT_ConditionalExpr)) 1416 return prec::Conditional; 1417 if (NextNonComment && NextNonComment->is(tok::colon) && 1418 NextNonComment->is(TT_DictLiteral)) 1419 return prec::Comma; 1420 if (Current->is(TT_LambdaArrow)) 1421 return prec::Comma; 1422 if (Current->is(TT_JsFatArrow)) 1423 return prec::Assignment; 1424 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName, 1425 TT_JsComputedPropertyName) || 1426 (Current->is(tok::comment) && NextNonComment && 1427 NextNonComment->is(TT_SelectorName))) 1428 return 0; 1429 if (Current->is(TT_RangeBasedForLoopColon)) 1430 return prec::Comma; 1431 if ((Style.Language == FormatStyle::LK_Java || 1432 Style.Language == FormatStyle::LK_JavaScript) && 1433 Current->is(Keywords.kw_instanceof)) 1434 return prec::Relational; 1435 if (Style.Language == FormatStyle::LK_JavaScript && 1436 Current->is(Keywords.kw_in)) 1437 return prec::Relational; 1438 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma)) 1439 return Current->getPrecedence(); 1440 if (Current->isOneOf(tok::period, tok::arrow)) 1441 return PrecedenceArrowAndPeriod; 1442 if (Style.Language == FormatStyle::LK_Java && 1443 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements, 1444 Keywords.kw_throws)) 1445 return 0; 1446 } 1447 return -1; 1448 } 1449 1450 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) { 1451 Start->FakeLParens.push_back(Precedence); 1452 if (Precedence > prec::Unknown) 1453 Start->StartsBinaryExpression = true; 1454 if (Current) { 1455 FormatToken *Previous = Current->Previous; 1456 while (Previous->is(tok::comment) && Previous->Previous) 1457 Previous = Previous->Previous; 1458 ++Previous->FakeRParens; 1459 if (Precedence > prec::Unknown) 1460 Previous->EndsBinaryExpression = true; 1461 } 1462 } 1463 1464 /// \brief Parse unary operator expressions and surround them with fake 1465 /// parentheses if appropriate. 1466 void parseUnaryOperator() { 1467 if (!Current || Current->isNot(TT_UnaryOperator)) { 1468 parse(PrecedenceArrowAndPeriod); 1469 return; 1470 } 1471 1472 FormatToken *Start = Current; 1473 next(); 1474 parseUnaryOperator(); 1475 1476 // The actual precedence doesn't matter. 1477 addFakeParenthesis(Start, prec::Unknown); 1478 } 1479 1480 void parseConditionalExpr() { 1481 while (Current && Current->isTrailingComment()) { 1482 next(); 1483 } 1484 FormatToken *Start = Current; 1485 parse(prec::LogicalOr); 1486 if (!Current || !Current->is(tok::question)) 1487 return; 1488 next(); 1489 parse(prec::Assignment); 1490 if (!Current || Current->isNot(TT_ConditionalExpr)) 1491 return; 1492 next(); 1493 parse(prec::Assignment); 1494 addFakeParenthesis(Start, prec::Conditional); 1495 } 1496 1497 void next(bool SkipPastLeadingComments = true) { 1498 if (Current) 1499 Current = Current->Next; 1500 while (Current && 1501 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) && 1502 Current->isTrailingComment()) 1503 Current = Current->Next; 1504 } 1505 1506 const FormatStyle &Style; 1507 const AdditionalKeywords &Keywords; 1508 FormatToken *Current; 1509 }; 1510 1511 } // end anonymous namespace 1512 1513 void TokenAnnotator::setCommentLineLevels( 1514 SmallVectorImpl<AnnotatedLine *> &Lines) { 1515 const AnnotatedLine *NextNonCommentLine = nullptr; 1516 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(), 1517 E = Lines.rend(); 1518 I != E; ++I) { 1519 if (NextNonCommentLine && (*I)->First->is(tok::comment) && 1520 (*I)->First->Next == nullptr) 1521 (*I)->Level = NextNonCommentLine->Level; 1522 else 1523 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr; 1524 1525 setCommentLineLevels((*I)->Children); 1526 } 1527 } 1528 1529 void TokenAnnotator::annotate(AnnotatedLine &Line) { 1530 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 1531 E = Line.Children.end(); 1532 I != E; ++I) { 1533 annotate(**I); 1534 } 1535 AnnotatingParser Parser(Style, Line, Keywords); 1536 Line.Type = Parser.parseLine(); 1537 if (Line.Type == LT_Invalid) 1538 return; 1539 1540 ExpressionParser ExprParser(Style, Keywords, Line); 1541 ExprParser.parse(); 1542 1543 if (Line.startsWith(TT_ObjCMethodSpecifier)) 1544 Line.Type = LT_ObjCMethodDecl; 1545 else if (Line.startsWith(TT_ObjCDecl)) 1546 Line.Type = LT_ObjCDecl; 1547 else if (Line.startsWith(TT_ObjCProperty)) 1548 Line.Type = LT_ObjCProperty; 1549 1550 Line.First->SpacesRequiredBefore = 1; 1551 Line.First->CanBreakBefore = Line.First->MustBreakBefore; 1552 } 1553 1554 // This function heuristically determines whether 'Current' starts the name of a 1555 // function declaration. 1556 static bool isFunctionDeclarationName(const FormatToken &Current, 1557 const AnnotatedLine &Line) { 1558 auto skipOperatorName = [](const FormatToken* Next) -> const FormatToken* { 1559 for (; Next; Next = Next->Next) { 1560 if (Next->is(TT_OverloadedOperatorLParen)) 1561 return Next; 1562 if (Next->is(TT_OverloadedOperator)) 1563 continue; 1564 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) { 1565 // For 'new[]' and 'delete[]'. 1566 if (Next->Next && Next->Next->is(tok::l_square) && 1567 Next->Next->Next && Next->Next->Next->is(tok::r_square)) 1568 Next = Next->Next->Next; 1569 continue; 1570 } 1571 1572 break; 1573 } 1574 return nullptr; 1575 }; 1576 1577 // Find parentheses of parameter list. 1578 const FormatToken *Next = Current.Next; 1579 if (Current.is(tok::kw_operator)) { 1580 if (Current.Previous && Current.Previous->is(tok::coloncolon)) 1581 return false; 1582 Next = skipOperatorName(Next); 1583 } else { 1584 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0) 1585 return false; 1586 for (; Next; Next = Next->Next) { 1587 if (Next->is(TT_TemplateOpener)) { 1588 Next = Next->MatchingParen; 1589 } else if (Next->is(tok::coloncolon)) { 1590 Next = Next->Next; 1591 if (!Next) 1592 return false; 1593 if (Next->is(tok::kw_operator)) { 1594 Next = skipOperatorName(Next->Next); 1595 break; 1596 } 1597 if (!Next->is(tok::identifier)) 1598 return false; 1599 } else if (Next->is(tok::l_paren)) { 1600 break; 1601 } else { 1602 return false; 1603 } 1604 } 1605 } 1606 1607 // Check whether parameter list can be long to a function declaration. 1608 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen) 1609 return false; 1610 // If the lines ends with "{", this is likely an function definition. 1611 if (Line.Last->is(tok::l_brace)) 1612 return true; 1613 if (Next->Next == Next->MatchingParen) 1614 return true; // Empty parentheses. 1615 // If there is an &/&& after the r_paren, this is likely a function. 1616 if (Next->MatchingParen->Next && 1617 Next->MatchingParen->Next->is(TT_PointerOrReference)) 1618 return true; 1619 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen; 1620 Tok = Tok->Next) { 1621 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() || 1622 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis)) 1623 return true; 1624 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) || 1625 Tok->Tok.isLiteral()) 1626 return false; 1627 } 1628 return false; 1629 } 1630 1631 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const { 1632 assert(Line.MightBeFunctionDecl); 1633 1634 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel || 1635 Style.AlwaysBreakAfterReturnType == 1636 FormatStyle::RTBS_TopLevelDefinitions) && 1637 Line.Level > 0) 1638 return false; 1639 1640 switch (Style.AlwaysBreakAfterReturnType) { 1641 case FormatStyle::RTBS_None: 1642 return false; 1643 case FormatStyle::RTBS_All: 1644 case FormatStyle::RTBS_TopLevel: 1645 return true; 1646 case FormatStyle::RTBS_AllDefinitions: 1647 case FormatStyle::RTBS_TopLevelDefinitions: 1648 return Line.mightBeFunctionDefinition(); 1649 } 1650 1651 return false; 1652 } 1653 1654 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) { 1655 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(), 1656 E = Line.Children.end(); 1657 I != E; ++I) { 1658 calculateFormattingInformation(**I); 1659 } 1660 1661 Line.First->TotalLength = 1662 Line.First->IsMultiline ? Style.ColumnLimit : Line.First->ColumnWidth; 1663 if (!Line.First->Next) 1664 return; 1665 FormatToken *Current = Line.First->Next; 1666 bool InFunctionDecl = Line.MightBeFunctionDecl; 1667 while (Current) { 1668 if (isFunctionDeclarationName(*Current, Line)) 1669 Current->Type = TT_FunctionDeclarationName; 1670 if (Current->is(TT_LineComment)) { 1671 if (Current->Previous->BlockKind == BK_BracedInit && 1672 Current->Previous->opensScope()) 1673 Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1; 1674 else 1675 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments; 1676 1677 // If we find a trailing comment, iterate backwards to determine whether 1678 // it seems to relate to a specific parameter. If so, break before that 1679 // parameter to avoid changing the comment's meaning. E.g. don't move 'b' 1680 // to the previous line in: 1681 // SomeFunction(a, 1682 // b, // comment 1683 // c); 1684 if (!Current->HasUnescapedNewline) { 1685 for (FormatToken *Parameter = Current->Previous; Parameter; 1686 Parameter = Parameter->Previous) { 1687 if (Parameter->isOneOf(tok::comment, tok::r_brace)) 1688 break; 1689 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) { 1690 if (!Parameter->Previous->is(TT_CtorInitializerComma) && 1691 Parameter->HasUnescapedNewline) 1692 Parameter->MustBreakBefore = true; 1693 break; 1694 } 1695 } 1696 } 1697 } else if (Current->SpacesRequiredBefore == 0 && 1698 spaceRequiredBefore(Line, *Current)) { 1699 Current->SpacesRequiredBefore = 1; 1700 } 1701 1702 Current->MustBreakBefore = 1703 Current->MustBreakBefore || mustBreakBefore(Line, *Current); 1704 1705 if (!Current->MustBreakBefore && InFunctionDecl && 1706 Current->is(TT_FunctionDeclarationName)) 1707 Current->MustBreakBefore = mustBreakForReturnType(Line); 1708 1709 Current->CanBreakBefore = 1710 Current->MustBreakBefore || canBreakBefore(Line, *Current); 1711 unsigned ChildSize = 0; 1712 if (Current->Previous->Children.size() == 1) { 1713 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last; 1714 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit 1715 : LastOfChild.TotalLength + 1; 1716 } 1717 const FormatToken *Prev = Current->Previous; 1718 if (Current->MustBreakBefore || Prev->Children.size() > 1 || 1719 (Prev->Children.size() == 1 && 1720 Prev->Children[0]->First->MustBreakBefore) || 1721 Current->IsMultiline) 1722 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit; 1723 else 1724 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth + 1725 ChildSize + Current->SpacesRequiredBefore; 1726 1727 if (Current->is(TT_CtorInitializerColon)) 1728 InFunctionDecl = false; 1729 1730 // FIXME: Only calculate this if CanBreakBefore is true once static 1731 // initializers etc. are sorted out. 1732 // FIXME: Move magic numbers to a better place. 1733 Current->SplitPenalty = 20 * Current->BindingStrength + 1734 splitPenalty(Line, *Current, InFunctionDecl); 1735 1736 Current = Current->Next; 1737 } 1738 1739 calculateUnbreakableTailLengths(Line); 1740 for (Current = Line.First; Current != nullptr; Current = Current->Next) { 1741 if (Current->Role) 1742 Current->Role->precomputeFormattingInfos(Current); 1743 } 1744 1745 DEBUG({ printDebugInfo(Line); }); 1746 } 1747 1748 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) { 1749 unsigned UnbreakableTailLength = 0; 1750 FormatToken *Current = Line.Last; 1751 while (Current) { 1752 Current->UnbreakableTailLength = UnbreakableTailLength; 1753 if (Current->CanBreakBefore || 1754 Current->isOneOf(tok::comment, tok::string_literal)) { 1755 UnbreakableTailLength = 0; 1756 } else { 1757 UnbreakableTailLength += 1758 Current->ColumnWidth + Current->SpacesRequiredBefore; 1759 } 1760 Current = Current->Previous; 1761 } 1762 } 1763 1764 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line, 1765 const FormatToken &Tok, 1766 bool InFunctionDecl) { 1767 const FormatToken &Left = *Tok.Previous; 1768 const FormatToken &Right = Tok; 1769 1770 if (Left.is(tok::semi)) 1771 return 0; 1772 1773 if (Style.Language == FormatStyle::LK_Java) { 1774 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws)) 1775 return 1; 1776 if (Right.is(Keywords.kw_implements)) 1777 return 2; 1778 if (Left.is(tok::comma) && Left.NestingLevel == 0) 1779 return 3; 1780 } else if (Style.Language == FormatStyle::LK_JavaScript) { 1781 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma)) 1782 return 100; 1783 if (Left.is(TT_JsTypeColon)) 1784 return 35; 1785 } 1786 1787 if (Left.is(tok::comma) || (Right.is(tok::identifier) && Right.Next && 1788 Right.Next->is(TT_DictLiteral))) 1789 return 1; 1790 if (Right.is(tok::l_square)) { 1791 if (Style.Language == FormatStyle::LK_Proto) 1792 return 1; 1793 if (Left.is(tok::r_square)) 1794 return 200; 1795 // Slightly prefer formatting local lambda definitions like functions. 1796 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal)) 1797 return 35; 1798 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare, 1799 TT_ArrayInitializerLSquare)) 1800 return 500; 1801 } 1802 1803 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 1804 Right.is(tok::kw_operator)) { 1805 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt) 1806 return 3; 1807 if (Left.is(TT_StartOfName)) 1808 return 110; 1809 if (InFunctionDecl && Right.NestingLevel == 0) 1810 return Style.PenaltyReturnTypeOnItsOwnLine; 1811 return 200; 1812 } 1813 if (Right.is(TT_PointerOrReference)) 1814 return 190; 1815 if (Right.is(TT_LambdaArrow)) 1816 return 110; 1817 if (Left.is(tok::equal) && Right.is(tok::l_brace)) 1818 return 150; 1819 if (Left.is(TT_CastRParen)) 1820 return 100; 1821 if (Left.is(tok::coloncolon) || 1822 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto)) 1823 return 500; 1824 if (Left.isOneOf(tok::kw_class, tok::kw_struct)) 1825 return 5000; 1826 1827 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon)) 1828 return 2; 1829 1830 if (Right.isMemberAccess()) { 1831 // Breaking before the "./->" of a chained call/member access is reasonably 1832 // cheap, as formatting those with one call per line is generally 1833 // desirable. In particular, it should be cheaper to break before the call 1834 // than it is to break inside a call's parameters, which could lead to weird 1835 // "hanging" indents. The exception is the very last "./->" to support this 1836 // frequent pattern: 1837 // 1838 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc( 1839 // dddddddd); 1840 // 1841 // which might otherwise be blown up onto many lines. Here, clang-format 1842 // won't produce "hanging" indents anyway as there is no other trailing 1843 // call. 1844 // 1845 // Also apply higher penalty is not a call as that might lead to a wrapping 1846 // like: 1847 // 1848 // aaaaaaa 1849 // .aaaaaaaaa.bbbbbbbb(cccccccc); 1850 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope() 1851 ? 150 1852 : 35; 1853 } 1854 1855 if (Right.is(TT_TrailingAnnotation) && 1856 (!Right.Next || Right.Next->isNot(tok::l_paren))) { 1857 // Moving trailing annotations to the next line is fine for ObjC method 1858 // declarations. 1859 if (Line.startsWith(TT_ObjCMethodSpecifier)) 1860 return 10; 1861 // Generally, breaking before a trailing annotation is bad unless it is 1862 // function-like. It seems to be especially preferable to keep standard 1863 // annotations (i.e. "const", "final" and "override") on the same line. 1864 // Use a slightly higher penalty after ")" so that annotations like 1865 // "const override" are kept together. 1866 bool is_short_annotation = Right.TokenText.size() < 10; 1867 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0); 1868 } 1869 1870 // In for-loops, prefer breaking at ',' and ';'. 1871 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal)) 1872 return 4; 1873 1874 // In Objective-C method expressions, prefer breaking before "param:" over 1875 // breaking after it. 1876 if (Right.is(TT_SelectorName)) 1877 return 0; 1878 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr)) 1879 return Line.MightBeFunctionDecl ? 50 : 500; 1880 1881 if (Left.is(tok::l_paren) && InFunctionDecl && 1882 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 1883 return 100; 1884 if (Left.is(tok::l_paren) && Left.Previous && 1885 Left.Previous->isOneOf(tok::kw_if, tok::kw_for)) 1886 return 1000; 1887 if (Left.is(tok::equal) && InFunctionDecl) 1888 return 110; 1889 if (Right.is(tok::r_brace)) 1890 return 1; 1891 if (Left.is(TT_TemplateOpener)) 1892 return 100; 1893 if (Left.opensScope()) { 1894 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign) 1895 return 0; 1896 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter 1897 : 19; 1898 } 1899 if (Left.is(TT_JavaAnnotation)) 1900 return 50; 1901 1902 if (Right.is(tok::lessless)) { 1903 if (Left.is(tok::string_literal) && 1904 (Right.NextOperator || Right.OperatorIndex != 1)) { 1905 StringRef Content = Left.TokenText; 1906 if (Content.startswith("\"")) 1907 Content = Content.drop_front(1); 1908 if (Content.endswith("\"")) 1909 Content = Content.drop_back(1); 1910 Content = Content.trim(); 1911 if (Content.size() > 1 && 1912 (Content.back() == ':' || Content.back() == '=')) 1913 return 25; 1914 } 1915 return 1; // Breaking at a << is really cheap. 1916 } 1917 if (Left.is(TT_ConditionalExpr)) 1918 return prec::Conditional; 1919 prec::Level Level = Left.getPrecedence(); 1920 if (Level != prec::Unknown) 1921 return Level; 1922 Level = Right.getPrecedence(); 1923 if (Level != prec::Unknown) 1924 return Level; 1925 1926 return 3; 1927 } 1928 1929 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line, 1930 const FormatToken &Left, 1931 const FormatToken &Right) { 1932 if (Left.is(tok::kw_return) && Right.isNot(tok::semi)) 1933 return true; 1934 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty && 1935 Left.Tok.getObjCKeywordID() == tok::objc_property) 1936 return true; 1937 if (Right.is(tok::hashhash)) 1938 return Left.is(tok::hash); 1939 if (Left.isOneOf(tok::hashhash, tok::hash)) 1940 return Right.is(tok::hash); 1941 if (Left.is(tok::l_paren) && Right.is(tok::r_paren)) 1942 return Style.SpaceInEmptyParentheses; 1943 if (Left.is(tok::l_paren) || Right.is(tok::r_paren)) 1944 return (Right.is(TT_CastRParen) || 1945 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen))) 1946 ? Style.SpacesInCStyleCastParentheses 1947 : Style.SpacesInParentheses; 1948 if (Right.isOneOf(tok::semi, tok::comma)) 1949 return false; 1950 if (Right.is(tok::less) && 1951 (Left.is(tok::kw_template) || 1952 (Line.Type == LT_ObjCDecl && Style.ObjCSpaceBeforeProtocolList))) 1953 return true; 1954 if (Left.isOneOf(tok::exclaim, tok::tilde)) 1955 return false; 1956 if (Left.is(tok::at) && 1957 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant, 1958 tok::numeric_constant, tok::l_paren, tok::l_brace, 1959 tok::kw_true, tok::kw_false)) 1960 return false; 1961 if (Left.is(tok::colon)) 1962 return !Left.is(TT_ObjCMethodExpr); 1963 if (Left.is(tok::coloncolon)) 1964 return false; 1965 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) 1966 return false; 1967 if (Right.is(tok::ellipsis)) 1968 return Left.Tok.isLiteral(); 1969 if (Left.is(tok::l_square) && Right.is(tok::amp)) 1970 return false; 1971 if (Right.is(TT_PointerOrReference)) 1972 return (Left.is(tok::r_paren) && Left.MatchingParen && 1973 (Left.MatchingParen->is(TT_OverloadedOperatorLParen) || 1974 (Left.MatchingParen->Previous && 1975 Left.MatchingParen->Previous->is(TT_FunctionDeclarationName)))) || 1976 (Left.Tok.isLiteral() || 1977 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) && 1978 (Style.PointerAlignment != FormatStyle::PAS_Left || 1979 Line.IsMultiVariableDeclStmt))); 1980 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) && 1981 (!Left.is(TT_PointerOrReference) || 1982 (Style.PointerAlignment != FormatStyle::PAS_Right && 1983 !Line.IsMultiVariableDeclStmt))) 1984 return true; 1985 if (Left.is(TT_PointerOrReference)) 1986 return Right.Tok.isLiteral() || 1987 Right.isOneOf(TT_BlockComment, Keywords.kw_final, 1988 Keywords.kw_override) || 1989 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) || 1990 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare, 1991 tok::l_paren) && 1992 (Style.PointerAlignment != FormatStyle::PAS_Right && 1993 !Line.IsMultiVariableDeclStmt) && 1994 Left.Previous && 1995 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon)); 1996 if (Right.is(tok::star) && Left.is(tok::l_paren)) 1997 return false; 1998 if (Left.is(tok::l_square)) 1999 return (Left.is(TT_ArrayInitializerLSquare) && 2000 Style.SpacesInContainerLiterals && Right.isNot(tok::r_square)) || 2001 (Left.is(TT_ArraySubscriptLSquare) && Style.SpacesInSquareBrackets && 2002 Right.isNot(tok::r_square)); 2003 if (Right.is(tok::r_square)) 2004 return Right.MatchingParen && 2005 ((Style.SpacesInContainerLiterals && 2006 Right.MatchingParen->is(TT_ArrayInitializerLSquare)) || 2007 (Style.SpacesInSquareBrackets && 2008 Right.MatchingParen->is(TT_ArraySubscriptLSquare))); 2009 if (Right.is(tok::l_square) && 2010 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare) && 2011 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral)) 2012 return false; 2013 if (Left.is(tok::l_brace) && Right.is(tok::r_brace)) 2014 return !Left.Children.empty(); // No spaces in "{}". 2015 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) || 2016 (Right.is(tok::r_brace) && Right.MatchingParen && 2017 Right.MatchingParen->BlockKind != BK_Block)) 2018 return !Style.Cpp11BracedListStyle; 2019 if (Left.is(TT_BlockComment)) 2020 return !Left.TokenText.endswith("=*/"); 2021 if (Right.is(tok::l_paren)) { 2022 if (Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) 2023 return true; 2024 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) || 2025 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never && 2026 (Left.isOneOf(tok::kw_if, tok::pp_elif, tok::kw_for, tok::kw_while, 2027 tok::kw_switch, tok::kw_case, TT_ForEachMacro, 2028 TT_ObjCForIn) || 2029 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch, 2030 tok::kw_new, tok::kw_delete) && 2031 (!Left.Previous || Left.Previous->isNot(tok::period))))) || 2032 (Style.SpaceBeforeParens == FormatStyle::SBPO_Always && 2033 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || 2034 Left.is(tok::r_paren)) && 2035 Line.Type != LT_PreprocessorDirective); 2036 } 2037 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword) 2038 return false; 2039 if (Right.is(TT_UnaryOperator)) 2040 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) && 2041 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr)); 2042 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square, 2043 tok::r_paren) || 2044 Left.isSimpleTypeSpecifier()) && 2045 Right.is(tok::l_brace) && Right.getNextNonComment() && 2046 Right.BlockKind != BK_Block) 2047 return false; 2048 if (Left.is(tok::period) || Right.is(tok::period)) 2049 return false; 2050 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L") 2051 return false; 2052 if (Left.is(TT_TemplateCloser) && Left.MatchingParen && 2053 Left.MatchingParen->Previous && 2054 Left.MatchingParen->Previous->is(tok::period)) 2055 // A.<B>DoSomething(); 2056 return false; 2057 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square)) 2058 return false; 2059 return true; 2060 } 2061 2062 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line, 2063 const FormatToken &Right) { 2064 const FormatToken &Left = *Right.Previous; 2065 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo()) 2066 return true; // Never ever merge two identifiers. 2067 if (Style.Language == FormatStyle::LK_Cpp) { 2068 if (Left.is(tok::kw_operator)) 2069 return Right.is(tok::coloncolon); 2070 } else if (Style.Language == FormatStyle::LK_Proto) { 2071 if (Right.is(tok::period) && 2072 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required, 2073 Keywords.kw_repeated, Keywords.kw_extend)) 2074 return true; 2075 if (Right.is(tok::l_paren) && 2076 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option)) 2077 return true; 2078 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2079 if (Left.is(TT_JsFatArrow)) 2080 return true; 2081 if (Right.is(tok::star) && 2082 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield)) 2083 return false; 2084 if (Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in, 2085 Keywords.kw_of) && 2086 (!Left.Previous || !Left.Previous->is(tok::period))) 2087 return true; 2088 if (Left.is(tok::kw_default) && Left.Previous && 2089 Left.Previous->is(tok::kw_export)) 2090 return true; 2091 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace)) 2092 return true; 2093 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion)) 2094 return false; 2095 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator)) 2096 return false; 2097 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) && 2098 Line.First->isOneOf(Keywords.kw_import, tok::kw_export)) 2099 return false; 2100 if (Left.is(tok::ellipsis)) 2101 return false; 2102 if (Left.is(TT_TemplateCloser) && 2103 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square, 2104 Keywords.kw_implements, Keywords.kw_extends)) 2105 // Type assertions ('<type>expr') are not followed by whitespace. Other 2106 // locations that should have whitespace following are identified by the 2107 // above set of follower tokens. 2108 return false; 2109 } else if (Style.Language == FormatStyle::LK_Java) { 2110 if (Left.is(tok::r_square) && Right.is(tok::l_brace)) 2111 return true; 2112 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren)) 2113 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never; 2114 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private, 2115 tok::kw_protected) || 2116 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract, 2117 Keywords.kw_native)) && 2118 Right.is(TT_TemplateOpener)) 2119 return true; 2120 } 2121 if (Left.is(TT_ImplicitStringLiteral)) 2122 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd(); 2123 if (Line.Type == LT_ObjCMethodDecl) { 2124 if (Left.is(TT_ObjCMethodSpecifier)) 2125 return true; 2126 if (Left.is(tok::r_paren) && Right.is(tok::identifier)) 2127 // Don't space between ')' and <id> 2128 return false; 2129 } 2130 if (Line.Type == LT_ObjCProperty && 2131 (Right.is(tok::equal) || Left.is(tok::equal))) 2132 return false; 2133 2134 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) || 2135 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow)) 2136 return true; 2137 if (Right.is(TT_OverloadedOperatorLParen)) 2138 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always; 2139 if (Left.is(tok::comma)) 2140 return true; 2141 if (Right.is(tok::comma)) 2142 return false; 2143 if (Right.isOneOf(TT_CtorInitializerColon, TT_ObjCBlockLParen)) 2144 return true; 2145 if (Right.is(tok::colon)) { 2146 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) || 2147 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi)) 2148 return false; 2149 if (Right.is(TT_ObjCMethodExpr)) 2150 return false; 2151 if (Left.is(tok::question)) 2152 return false; 2153 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) 2154 return false; 2155 if (Right.is(TT_DictLiteral)) 2156 return Style.SpacesInContainerLiterals; 2157 return true; 2158 } 2159 if (Left.is(TT_UnaryOperator)) 2160 return Right.is(TT_BinaryOperator); 2161 2162 // If the next token is a binary operator or a selector name, we have 2163 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly. 2164 if (Left.is(TT_CastRParen)) 2165 return Style.SpaceAfterCStyleCast || 2166 Right.isOneOf(TT_BinaryOperator, TT_SelectorName); 2167 2168 if (Left.is(tok::greater) && Right.is(tok::greater)) 2169 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) && 2170 (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles); 2171 if (Right.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) || 2172 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar)) 2173 return false; 2174 if (!Style.SpaceBeforeAssignmentOperators && 2175 Right.getPrecedence() == prec::Assignment) 2176 return false; 2177 if (Right.is(tok::coloncolon) && Left.isNot(tok::l_brace)) 2178 return (Left.is(TT_TemplateOpener) && 2179 Style.Standard == FormatStyle::LS_Cpp03) || 2180 !(Left.isOneOf(tok::identifier, tok::l_paren, tok::r_paren, 2181 tok::l_square) || 2182 Left.isOneOf(TT_TemplateCloser, TT_TemplateOpener)); 2183 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser))) 2184 return Style.SpacesInAngles; 2185 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) || 2186 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && 2187 !Right.is(tok::r_paren))) 2188 return true; 2189 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) && 2190 Right.isNot(TT_FunctionTypeLParen)) 2191 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always; 2192 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) && 2193 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen)) 2194 return false; 2195 if (Right.is(tok::less) && Left.isNot(tok::l_paren) && 2196 Line.startsWith(tok::hash)) 2197 return true; 2198 if (Right.is(TT_TrailingUnaryOperator)) 2199 return false; 2200 if (Left.is(TT_RegexLiteral)) 2201 return false; 2202 return spaceRequiredBetween(Line, Left, Right); 2203 } 2204 2205 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style. 2206 static bool isAllmanBrace(const FormatToken &Tok) { 2207 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block && 2208 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral); 2209 } 2210 2211 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line, 2212 const FormatToken &Right) { 2213 const FormatToken &Left = *Right.Previous; 2214 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0) 2215 return true; 2216 2217 if (Style.Language == FormatStyle::LK_JavaScript) { 2218 // FIXME: This might apply to other languages and token kinds. 2219 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous && 2220 Left.Previous->is(tok::string_literal)) 2221 return true; 2222 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 && 2223 Left.Previous && Left.Previous->is(tok::equal) && 2224 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export, 2225 tok::kw_const) && 2226 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match 2227 // above. 2228 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let)) 2229 // Object literals on the top level of a file are treated as "enum-style". 2230 // Each key/value pair is put on a separate line, instead of bin-packing. 2231 return true; 2232 if (Left.is(tok::l_brace) && Line.Level == 0 && 2233 (Line.startsWith(tok::kw_enum) || 2234 Line.startsWith(tok::kw_export, tok::kw_enum))) 2235 // JavaScript top-level enum key/value pairs are put on separate lines 2236 // instead of bin-packing. 2237 return true; 2238 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && 2239 !Left.Children.empty()) 2240 // Support AllowShortFunctionsOnASingleLine for JavaScript. 2241 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None || 2242 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty || 2243 (Left.NestingLevel == 0 && Line.Level == 0 && 2244 Style.AllowShortFunctionsOnASingleLine == 2245 FormatStyle::SFS_Inline); 2246 } else if (Style.Language == FormatStyle::LK_Java) { 2247 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next && 2248 Right.Next->is(tok::string_literal)) 2249 return true; 2250 } 2251 2252 // If the last token before a '}' is a comma or a trailing comment, the 2253 // intention is to insert a line break after it in order to make shuffling 2254 // around entries easier. 2255 const FormatToken *BeforeClosingBrace = nullptr; 2256 if (Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) && 2257 Left.BlockKind != BK_Block && Left.MatchingParen) 2258 BeforeClosingBrace = Left.MatchingParen->Previous; 2259 else if (Right.MatchingParen && 2260 Right.MatchingParen->isOneOf(tok::l_brace, 2261 TT_ArrayInitializerLSquare)) 2262 BeforeClosingBrace = &Left; 2263 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) || 2264 BeforeClosingBrace->isTrailingComment())) 2265 return true; 2266 2267 if (Right.is(tok::comment)) 2268 return Left.BlockKind != BK_BracedInit && 2269 Left.isNot(TT_CtorInitializerColon) && 2270 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline); 2271 if (Left.isTrailingComment()) 2272 return true; 2273 if (Left.isStringLiteral() && 2274 (Right.isStringLiteral() || Right.is(TT_ObjCStringLiteral))) 2275 return true; 2276 if (Right.Previous->IsUnterminatedLiteral) 2277 return true; 2278 if (Right.is(tok::lessless) && Right.Next && 2279 Right.Previous->is(tok::string_literal) && 2280 Right.Next->is(tok::string_literal)) 2281 return true; 2282 if (Right.Previous->ClosesTemplateDeclaration && 2283 Right.Previous->MatchingParen && 2284 Right.Previous->MatchingParen->NestingLevel == 0 && 2285 Style.AlwaysBreakTemplateDeclarations) 2286 return true; 2287 if ((Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) && 2288 Style.BreakConstructorInitializersBeforeComma && 2289 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 2290 return true; 2291 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\"")) 2292 // Raw string literals are special wrt. line breaks. The author has made a 2293 // deliberate choice and might have aligned the contents of the string 2294 // literal accordingly. Thus, we try keep existing line breaks. 2295 return Right.NewlinesBefore > 0; 2296 if (Right.Previous->is(tok::l_brace) && Right.NestingLevel == 1 && 2297 Style.Language == FormatStyle::LK_Proto) 2298 // Don't put enums onto single lines in protocol buffers. 2299 return true; 2300 if (Right.is(TT_InlineASMBrace)) 2301 return Right.HasUnescapedNewline; 2302 if (isAllmanBrace(Left) || isAllmanBrace(Right)) 2303 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) || 2304 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) || 2305 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct); 2306 if (Style.Language == FormatStyle::LK_Proto && Left.isNot(tok::l_brace) && 2307 Right.is(TT_SelectorName)) 2308 return true; 2309 if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine) 2310 return true; 2311 2312 if ((Style.Language == FormatStyle::LK_Java || 2313 Style.Language == FormatStyle::LK_JavaScript) && 2314 Left.is(TT_LeadingJavaAnnotation) && 2315 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) && 2316 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations)) 2317 return true; 2318 2319 return false; 2320 } 2321 2322 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line, 2323 const FormatToken &Right) { 2324 const FormatToken &Left = *Right.Previous; 2325 2326 // Language-specific stuff. 2327 if (Style.Language == FormatStyle::LK_Java) { 2328 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 2329 Keywords.kw_implements)) 2330 return false; 2331 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends, 2332 Keywords.kw_implements)) 2333 return true; 2334 } else if (Style.Language == FormatStyle::LK_JavaScript) { 2335 if (Left.is(tok::kw_return)) 2336 return false; // Otherwise a semicolon is inserted. 2337 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace)) 2338 return false; 2339 if (Left.is(TT_JsTypeColon)) 2340 return true; 2341 if (Right.NestingLevel == 0 && Right.is(Keywords.kw_is)) 2342 return false; 2343 if (Left.is(Keywords.kw_in)) 2344 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None; 2345 if (Right.is(Keywords.kw_in)) 2346 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None; 2347 } 2348 2349 if (Left.is(tok::at)) 2350 return false; 2351 if (Left.Tok.getObjCKeywordID() == tok::objc_interface) 2352 return false; 2353 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation)) 2354 return !Right.is(tok::l_paren); 2355 if (Right.is(TT_PointerOrReference)) 2356 return Line.IsMultiVariableDeclStmt || 2357 (Style.PointerAlignment == FormatStyle::PAS_Right && 2358 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName))); 2359 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) || 2360 Right.is(tok::kw_operator)) 2361 return true; 2362 if (Left.is(TT_PointerOrReference)) 2363 return false; 2364 if (Right.isTrailingComment()) 2365 // We rely on MustBreakBefore being set correctly here as we should not 2366 // change the "binding" behavior of a comment. 2367 // The first comment in a braced lists is always interpreted as belonging to 2368 // the first list element. Otherwise, it should be placed outside of the 2369 // list. 2370 return Left.BlockKind == BK_BracedInit; 2371 if (Left.is(tok::question) && Right.is(tok::colon)) 2372 return false; 2373 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question)) 2374 return Style.BreakBeforeTernaryOperators; 2375 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question)) 2376 return !Style.BreakBeforeTernaryOperators; 2377 if (Right.is(TT_InheritanceColon)) 2378 return true; 2379 if (Right.is(tok::colon) && 2380 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon)) 2381 return false; 2382 if (Left.is(tok::colon) && (Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr))) 2383 return true; 2384 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next && 2385 Right.Next->is(TT_ObjCMethodExpr))) 2386 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls. 2387 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty) 2388 return true; 2389 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen)) 2390 return true; 2391 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen, 2392 TT_OverloadedOperator)) 2393 return false; 2394 if (Left.is(TT_RangeBasedForLoopColon)) 2395 return true; 2396 if (Right.is(TT_RangeBasedForLoopColon)) 2397 return false; 2398 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) || 2399 Left.is(tok::kw_operator)) 2400 return false; 2401 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) && 2402 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0) 2403 return false; 2404 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen)) 2405 return false; 2406 if (Left.is(tok::l_paren) && Left.Previous && 2407 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen))) 2408 return false; 2409 if (Right.is(TT_ImplicitStringLiteral)) 2410 return false; 2411 2412 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser)) 2413 return false; 2414 if (Right.is(tok::r_square) && Right.MatchingParen && 2415 Right.MatchingParen->is(TT_LambdaLSquare)) 2416 return false; 2417 2418 // We only break before r_brace if there was a corresponding break before 2419 // the l_brace, which is tracked by BreakBeforeClosingBrace. 2420 if (Right.is(tok::r_brace)) 2421 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block; 2422 2423 // Allow breaking after a trailing annotation, e.g. after a method 2424 // declaration. 2425 if (Left.is(TT_TrailingAnnotation)) 2426 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren, 2427 tok::less, tok::coloncolon); 2428 2429 if (Right.is(tok::kw___attribute)) 2430 return true; 2431 2432 if (Left.is(tok::identifier) && Right.is(tok::string_literal)) 2433 return true; 2434 2435 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral)) 2436 return true; 2437 2438 if (Left.is(TT_CtorInitializerComma) && 2439 Style.BreakConstructorInitializersBeforeComma) 2440 return false; 2441 if (Right.is(TT_CtorInitializerComma) && 2442 Style.BreakConstructorInitializersBeforeComma) 2443 return true; 2444 if ((Left.is(tok::greater) && Right.is(tok::greater)) || 2445 (Left.is(tok::less) && Right.is(tok::less))) 2446 return false; 2447 if (Right.is(TT_BinaryOperator) && 2448 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None && 2449 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All || 2450 Right.getPrecedence() != prec::Assignment)) 2451 return true; 2452 if (Left.is(TT_ArrayInitializerLSquare)) 2453 return true; 2454 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const)) 2455 return true; 2456 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) && 2457 !Left.isOneOf(tok::arrowstar, tok::lessless) && 2458 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All && 2459 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None || 2460 Left.getPrecedence() == prec::Assignment)) 2461 return true; 2462 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace, 2463 tok::kw_class, tok::kw_struct) || 2464 Right.isMemberAccess() || 2465 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless, 2466 tok::colon, tok::l_square, tok::at) || 2467 (Left.is(tok::r_paren) && 2468 Right.isOneOf(tok::identifier, tok::kw_const)) || 2469 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)); 2470 } 2471 2472 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) { 2473 llvm::errs() << "AnnotatedTokens:\n"; 2474 const FormatToken *Tok = Line.First; 2475 while (Tok) { 2476 llvm::errs() << " M=" << Tok->MustBreakBefore 2477 << " C=" << Tok->CanBreakBefore 2478 << " T=" << getTokenTypeName(Tok->Type) 2479 << " S=" << Tok->SpacesRequiredBefore 2480 << " B=" << Tok->BlockParameterCount 2481 << " P=" << Tok->SplitPenalty << " Name=" << Tok->Tok.getName() 2482 << " L=" << Tok->TotalLength << " PPK=" << Tok->PackingKind 2483 << " FakeLParens="; 2484 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i) 2485 llvm::errs() << Tok->FakeLParens[i] << "/"; 2486 llvm::errs() << " FakeRParens=" << Tok->FakeRParens << "\n"; 2487 if (!Tok->Next) 2488 assert(Tok == Line.Last); 2489 Tok = Tok->Next; 2490 } 2491 llvm::errs() << "----\n"; 2492 } 2493 2494 } // namespace format 2495 } // namespace clang 2496