1 //===--- ContinuationIndenter.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 /// This file implements the continuation indenter. 12 /// 13 //===----------------------------------------------------------------------===// 14 15 #include "ContinuationIndenter.h" 16 #include "BreakableToken.h" 17 #include "FormatInternal.h" 18 #include "WhitespaceManager.h" 19 #include "clang/Basic/OperatorPrecedence.h" 20 #include "clang/Basic/SourceManager.h" 21 #include "clang/Format/Format.h" 22 #include "llvm/Support/Debug.h" 23 24 #define DEBUG_TYPE "format-indenter" 25 26 namespace clang { 27 namespace format { 28 29 // Returns true if a TT_SelectorName should be indented when wrapped, 30 // false otherwise. 31 static bool shouldIndentWrappedSelectorName(const FormatStyle &Style, 32 LineType LineType) { 33 return Style.IndentWrappedFunctionNames || LineType == LT_ObjCMethodDecl; 34 } 35 36 // Returns the length of everything up to the first possible line break after 37 // the ), ], } or > matching \c Tok. 38 static unsigned getLengthToMatchingParen(const FormatToken &Tok, 39 const std::vector<ParenState> &Stack) { 40 // Normally whether or not a break before T is possible is calculated and 41 // stored in T.CanBreakBefore. Braces, array initializers and text proto 42 // messages like `key: < ... >` are an exception: a break is possible 43 // before a closing brace R if a break was inserted after the corresponding 44 // opening brace. The information about whether or not a break is needed 45 // before a closing brace R is stored in the ParenState field 46 // S.BreakBeforeClosingBrace where S is the state that R closes. 47 // 48 // In order to decide whether there can be a break before encountered right 49 // braces, this implementation iterates over the sequence of tokens and over 50 // the paren stack in lockstep, keeping track of the stack level which visited 51 // right braces correspond to in MatchingStackIndex. 52 // 53 // For example, consider: 54 // L. <- line number 55 // 1. { 56 // 2. {1}, 57 // 3. {2}, 58 // 4. {{3}}} 59 // ^ where we call this method with this token. 60 // The paren stack at this point contains 3 brace levels: 61 // 0. { at line 1, BreakBeforeClosingBrace: true 62 // 1. first { at line 4, BreakBeforeClosingBrace: false 63 // 2. second { at line 4, BreakBeforeClosingBrace: false, 64 // where there might be fake parens levels in-between these levels. 65 // The algorithm will start at the first } on line 4, which is the matching 66 // brace of the initial left brace and at level 2 of the stack. Then, 67 // examining BreakBeforeClosingBrace: false at level 2, it will continue to 68 // the second } on line 4, and will traverse the stack downwards until it 69 // finds the matching { on level 1. Then, examining BreakBeforeClosingBrace: 70 // false at level 1, it will continue to the third } on line 4 and will 71 // traverse the stack downwards until it finds the matching { on level 0. 72 // Then, examining BreakBeforeClosingBrace: true at level 0, the algorithm 73 // will stop and will use the second } on line 4 to determine the length to 74 // return, as in this example the range will include the tokens: {3}} 75 // 76 // The algorithm will only traverse the stack if it encounters braces, array 77 // initializer squares or text proto angle brackets. 78 if (!Tok.MatchingParen) 79 return 0; 80 FormatToken *End = Tok.MatchingParen; 81 // Maintains a stack level corresponding to the current End token. 82 int MatchingStackIndex = Stack.size() - 1; 83 // Traverses the stack downwards, looking for the level to which LBrace 84 // corresponds. Returns either a pointer to the matching level or nullptr if 85 // LParen is not found in the initial portion of the stack up to 86 // MatchingStackIndex. 87 auto FindParenState = [&](const FormatToken *LBrace) -> const ParenState * { 88 while (MatchingStackIndex >= 0 && Stack[MatchingStackIndex].Tok != LBrace) 89 --MatchingStackIndex; 90 return MatchingStackIndex >= 0 ? &Stack[MatchingStackIndex] : nullptr; 91 }; 92 for (; End->Next; End = End->Next) { 93 if (End->Next->CanBreakBefore) 94 break; 95 if (!End->Next->closesScope()) 96 continue; 97 if (End->Next->MatchingParen && 98 End->Next->MatchingParen->isOneOf( 99 tok::l_brace, TT_ArrayInitializerLSquare, tok::less)) { 100 const ParenState *State = FindParenState(End->Next->MatchingParen); 101 if (State && State->BreakBeforeClosingBrace) 102 break; 103 } 104 } 105 return End->TotalLength - Tok.TotalLength + 1; 106 } 107 108 static unsigned getLengthToNextOperator(const FormatToken &Tok) { 109 if (!Tok.NextOperator) 110 return 0; 111 return Tok.NextOperator->TotalLength - Tok.TotalLength; 112 } 113 114 // Returns \c true if \c Tok is the "." or "->" of a call and starts the next 115 // segment of a builder type call. 116 static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) { 117 return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope(); 118 } 119 120 // Returns \c true if \c Current starts a new parameter. 121 static bool startsNextParameter(const FormatToken &Current, 122 const FormatStyle &Style) { 123 const FormatToken &Previous = *Current.Previous; 124 if (Current.is(TT_CtorInitializerComma) && 125 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 126 return true; 127 if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName)) 128 return true; 129 return Previous.is(tok::comma) && !Current.isTrailingComment() && 130 ((Previous.isNot(TT_CtorInitializerComma) || 131 Style.BreakConstructorInitializers != 132 FormatStyle::BCIS_BeforeComma) && 133 (Previous.isNot(TT_InheritanceComma) || 134 Style.BreakInheritanceList != FormatStyle::BILS_BeforeComma)); 135 } 136 137 static bool opensProtoMessageField(const FormatToken &LessTok, 138 const FormatStyle &Style) { 139 if (LessTok.isNot(tok::less)) 140 return false; 141 return Style.Language == FormatStyle::LK_TextProto || 142 (Style.Language == FormatStyle::LK_Proto && 143 (LessTok.NestingLevel > 0 || 144 (LessTok.Previous && LessTok.Previous->is(tok::equal)))); 145 } 146 147 // Returns the delimiter of a raw string literal, or None if TokenText is not 148 // the text of a raw string literal. The delimiter could be the empty string. 149 // For example, the delimiter of R"deli(cont)deli" is deli. 150 static llvm::Optional<StringRef> getRawStringDelimiter(StringRef TokenText) { 151 if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'. 152 || !TokenText.startswith("R\"") || !TokenText.endswith("\"")) 153 return None; 154 155 // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has 156 // size at most 16 by the standard, so the first '(' must be among the first 157 // 19 bytes. 158 size_t LParenPos = TokenText.substr(0, 19).find_first_of('('); 159 if (LParenPos == StringRef::npos) 160 return None; 161 StringRef Delimiter = TokenText.substr(2, LParenPos - 2); 162 163 // Check that the string ends in ')Delimiter"'. 164 size_t RParenPos = TokenText.size() - Delimiter.size() - 2; 165 if (TokenText[RParenPos] != ')') 166 return None; 167 if (!TokenText.substr(RParenPos + 1).startswith(Delimiter)) 168 return None; 169 return Delimiter; 170 } 171 172 // Returns the canonical delimiter for \p Language, or the empty string if no 173 // canonical delimiter is specified. 174 static StringRef 175 getCanonicalRawStringDelimiter(const FormatStyle &Style, 176 FormatStyle::LanguageKind Language) { 177 for (const auto &Format : Style.RawStringFormats) { 178 if (Format.Language == Language) 179 return StringRef(Format.CanonicalDelimiter); 180 } 181 return ""; 182 } 183 184 RawStringFormatStyleManager::RawStringFormatStyleManager( 185 const FormatStyle &CodeStyle) { 186 for (const auto &RawStringFormat : CodeStyle.RawStringFormats) { 187 llvm::Optional<FormatStyle> LanguageStyle = 188 CodeStyle.GetLanguageStyle(RawStringFormat.Language); 189 if (!LanguageStyle) { 190 FormatStyle PredefinedStyle; 191 if (!getPredefinedStyle(RawStringFormat.BasedOnStyle, 192 RawStringFormat.Language, &PredefinedStyle)) { 193 PredefinedStyle = getLLVMStyle(); 194 PredefinedStyle.Language = RawStringFormat.Language; 195 } 196 LanguageStyle = PredefinedStyle; 197 } 198 LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit; 199 for (StringRef Delimiter : RawStringFormat.Delimiters) { 200 DelimiterStyle.insert({Delimiter, *LanguageStyle}); 201 } 202 for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions) { 203 EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle}); 204 } 205 } 206 } 207 208 llvm::Optional<FormatStyle> 209 RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const { 210 auto It = DelimiterStyle.find(Delimiter); 211 if (It == DelimiterStyle.end()) 212 return None; 213 return It->second; 214 } 215 216 llvm::Optional<FormatStyle> 217 RawStringFormatStyleManager::getEnclosingFunctionStyle( 218 StringRef EnclosingFunction) const { 219 auto It = EnclosingFunctionStyle.find(EnclosingFunction); 220 if (It == EnclosingFunctionStyle.end()) 221 return None; 222 return It->second; 223 } 224 225 ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style, 226 const AdditionalKeywords &Keywords, 227 const SourceManager &SourceMgr, 228 WhitespaceManager &Whitespaces, 229 encoding::Encoding Encoding, 230 bool BinPackInconclusiveFunctions) 231 : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr), 232 Whitespaces(Whitespaces), Encoding(Encoding), 233 BinPackInconclusiveFunctions(BinPackInconclusiveFunctions), 234 CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {} 235 236 LineState ContinuationIndenter::getInitialState(unsigned FirstIndent, 237 unsigned FirstStartColumn, 238 const AnnotatedLine *Line, 239 bool DryRun) { 240 LineState State; 241 State.FirstIndent = FirstIndent; 242 if (FirstStartColumn && Line->First->NewlinesBefore == 0) 243 State.Column = FirstStartColumn; 244 else 245 State.Column = FirstIndent; 246 // With preprocessor directive indentation, the line starts on column 0 247 // since it's indented after the hash, but FirstIndent is set to the 248 // preprocessor indent. 249 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 250 (Line->Type == LT_PreprocessorDirective || 251 Line->Type == LT_ImportStatement)) 252 State.Column = 0; 253 State.Line = Line; 254 State.NextToken = Line->First; 255 State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent, 256 /*AvoidBinPacking=*/false, 257 /*NoLineBreak=*/false)); 258 State.LineContainsContinuedForLoopSection = false; 259 State.NoContinuation = false; 260 State.StartOfStringLiteral = 0; 261 State.StartOfLineLevel = 0; 262 State.LowestLevelOnLine = 0; 263 State.IgnoreStackForComparison = false; 264 265 if (Style.Language == FormatStyle::LK_TextProto) { 266 // We need this in order to deal with the bin packing of text fields at 267 // global scope. 268 State.Stack.back().AvoidBinPacking = true; 269 State.Stack.back().BreakBeforeParameter = true; 270 State.Stack.back().AlignColons = false; 271 } 272 273 // The first token has already been indented and thus consumed. 274 moveStateToNextToken(State, DryRun, /*Newline=*/false); 275 return State; 276 } 277 278 bool ContinuationIndenter::canBreak(const LineState &State) { 279 const FormatToken &Current = *State.NextToken; 280 const FormatToken &Previous = *Current.Previous; 281 assert(&Previous == Current.Previous); 282 if (!Current.CanBreakBefore && !(State.Stack.back().BreakBeforeClosingBrace && 283 Current.closesBlockOrBlockTypeList(Style))) 284 return false; 285 // The opening "{" of a braced list has to be on the same line as the first 286 // element if it is nested in another braced init list or function call. 287 if (!Current.MustBreakBefore && Previous.is(tok::l_brace) && 288 Previous.isNot(TT_DictLiteral) && Previous.BlockKind == BK_BracedInit && 289 Previous.Previous && 290 Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma)) 291 return false; 292 // This prevents breaks like: 293 // ... 294 // SomeParameter, OtherParameter).DoSomething( 295 // ... 296 // As they hide "DoSomething" and are generally bad for readability. 297 if (Previous.opensScope() && Previous.isNot(tok::l_brace) && 298 State.LowestLevelOnLine < State.StartOfLineLevel && 299 State.LowestLevelOnLine < Current.NestingLevel) 300 return false; 301 if (Current.isMemberAccess() && State.Stack.back().ContainsUnwrappedBuilder) 302 return false; 303 304 // Don't create a 'hanging' indent if there are multiple blocks in a single 305 // statement. 306 if (Previous.is(tok::l_brace) && State.Stack.size() > 1 && 307 State.Stack[State.Stack.size() - 2].NestedBlockInlined && 308 State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks) 309 return false; 310 311 // Don't break after very short return types (e.g. "void") as that is often 312 // unexpected. 313 if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) { 314 if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None) 315 return false; 316 } 317 318 // If binary operators are moved to the next line (including commas for some 319 // styles of constructor initializers), that's always ok. 320 if (!Current.isOneOf(TT_BinaryOperator, tok::comma) && 321 State.Stack.back().NoLineBreakInOperand) 322 return false; 323 324 return !State.Stack.back().NoLineBreak; 325 } 326 327 bool ContinuationIndenter::mustBreak(const LineState &State) { 328 const FormatToken &Current = *State.NextToken; 329 const FormatToken &Previous = *Current.Previous; 330 if (Current.MustBreakBefore || Current.is(TT_InlineASMColon)) 331 return true; 332 if (State.Stack.back().BreakBeforeClosingBrace && 333 Current.closesBlockOrBlockTypeList(Style)) 334 return true; 335 if (Previous.is(tok::semi) && State.LineContainsContinuedForLoopSection) 336 return true; 337 if (Style.Language == FormatStyle::LK_ObjC && 338 Current.ObjCSelectorNameParts > 1 && 339 Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) { 340 return true; 341 } 342 if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) || 343 (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) && 344 Style.isCpp() && 345 // FIXME: This is a temporary workaround for the case where clang-format 346 // sets BreakBeforeParameter to avoid bin packing and this creates a 347 // completely unnecessary line break after a template type that isn't 348 // line-wrapped. 349 (Previous.NestingLevel == 1 || Style.BinPackParameters)) || 350 (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) && 351 Previous.isNot(tok::question)) || 352 (!Style.BreakBeforeTernaryOperators && 353 Previous.is(TT_ConditionalExpr))) && 354 State.Stack.back().BreakBeforeParameter && !Current.isTrailingComment() && 355 !Current.isOneOf(tok::r_paren, tok::r_brace)) 356 return true; 357 if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) || 358 (Previous.is(TT_ArrayInitializerLSquare) && 359 Previous.ParameterCount > 1) || 360 opensProtoMessageField(Previous, Style)) && 361 Style.ColumnLimit > 0 && 362 getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 > 363 getColumnLimit(State)) 364 return true; 365 366 const FormatToken &BreakConstructorInitializersToken = 367 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon 368 ? Previous 369 : Current; 370 if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) && 371 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength > 372 getColumnLimit(State) || 373 State.Stack.back().BreakBeforeParameter) && 374 (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All || 375 Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon || 376 Style.ColumnLimit != 0)) 377 return true; 378 379 if (Current.is(TT_ObjCMethodExpr) && !Previous.is(TT_SelectorName) && 380 State.Line->startsWith(TT_ObjCMethodSpecifier)) 381 return true; 382 if (Current.is(TT_SelectorName) && State.Stack.back().ObjCSelectorNameFound && 383 State.Stack.back().BreakBeforeParameter) 384 return true; 385 386 unsigned NewLineColumn = getNewLineColumn(State); 387 if (Current.isMemberAccess() && Style.ColumnLimit != 0 && 388 State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit && 389 (State.Column > NewLineColumn || 390 Current.NestingLevel < State.StartOfLineLevel)) 391 return true; 392 393 if (startsSegmentOfBuilderTypeCall(Current) && 394 (State.Stack.back().CallContinuation != 0 || 395 State.Stack.back().BreakBeforeParameter) && 396 // JavaScript is treated different here as there is a frequent pattern: 397 // SomeFunction(function() { 398 // ... 399 // }.bind(...)); 400 // FIXME: We should find a more generic solution to this problem. 401 !(State.Column <= NewLineColumn && 402 Style.Language == FormatStyle::LK_JavaScript)) 403 return true; 404 405 // If the template declaration spans multiple lines, force wrap before the 406 // function/class declaration 407 if (Previous.ClosesTemplateDeclaration && 408 State.Stack.back().BreakBeforeParameter && Current.CanBreakBefore) 409 return true; 410 411 if (State.Column <= NewLineColumn) 412 return false; 413 414 if (Style.AlwaysBreakBeforeMultilineStrings && 415 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth || 416 Previous.is(tok::comma) || Current.NestingLevel < 2) && 417 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at) && 418 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) && 419 nextIsMultilineString(State)) 420 return true; 421 422 // Using CanBreakBefore here and below takes care of the decision whether the 423 // current style uses wrapping before or after operators for the given 424 // operator. 425 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) { 426 // If we need to break somewhere inside the LHS of a binary expression, we 427 // should also break after the operator. Otherwise, the formatting would 428 // hide the operator precedence, e.g. in: 429 // if (aaaaaaaaaaaaaa == 430 // bbbbbbbbbbbbbb && c) {.. 431 // For comparisons, we only apply this rule, if the LHS is a binary 432 // expression itself as otherwise, the line breaks seem superfluous. 433 // We need special cases for ">>" which we have split into two ">" while 434 // lexing in order to make template parsing easier. 435 bool IsComparison = (Previous.getPrecedence() == prec::Relational || 436 Previous.getPrecedence() == prec::Equality || 437 Previous.getPrecedence() == prec::Spaceship) && 438 Previous.Previous && 439 Previous.Previous->isNot(TT_BinaryOperator); // For >>. 440 bool LHSIsBinaryExpr = 441 Previous.Previous && Previous.Previous->EndsBinaryExpression; 442 if ((!IsComparison || LHSIsBinaryExpr) && !Current.isTrailingComment() && 443 Previous.getPrecedence() != prec::Assignment && 444 State.Stack.back().BreakBeforeParameter) 445 return true; 446 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore && 447 State.Stack.back().BreakBeforeParameter) { 448 return true; 449 } 450 451 // Same as above, but for the first "<<" operator. 452 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) && 453 State.Stack.back().BreakBeforeParameter && 454 State.Stack.back().FirstLessLess == 0) 455 return true; 456 457 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) { 458 // Always break after "template <...>" and leading annotations. This is only 459 // for cases where the entire line does not fit on a single line as a 460 // different LineFormatter would be used otherwise. 461 if (Previous.ClosesTemplateDeclaration) 462 return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No; 463 if (Previous.is(TT_FunctionAnnotationRParen)) 464 return true; 465 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) && 466 Current.isNot(TT_LeadingJavaAnnotation)) 467 return true; 468 } 469 470 // If the return type spans multiple lines, wrap before the function name. 471 if ((Current.is(TT_FunctionDeclarationName) || 472 (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) && 473 !Previous.is(tok::kw_template) && State.Stack.back().BreakBeforeParameter) 474 return true; 475 476 // The following could be precomputed as they do not depend on the state. 477 // However, as they should take effect only if the UnwrappedLine does not fit 478 // into the ColumnLimit, they are checked here in the ContinuationIndenter. 479 if (Style.ColumnLimit != 0 && Previous.BlockKind == BK_Block && 480 Previous.is(tok::l_brace) && !Current.isOneOf(tok::r_brace, tok::comment)) 481 return true; 482 483 if (Current.is(tok::lessless) && 484 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") || 485 (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") || 486 Previous.TokenText == "\'\\n\'")))) 487 return true; 488 489 if (Previous.is(TT_BlockComment) && Previous.IsMultiline) 490 return true; 491 492 if (State.NoContinuation) 493 return true; 494 495 return false; 496 } 497 498 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline, 499 bool DryRun, 500 unsigned ExtraSpaces) { 501 const FormatToken &Current = *State.NextToken; 502 503 assert(!State.Stack.empty()); 504 State.NoContinuation = false; 505 506 if ((Current.is(TT_ImplicitStringLiteral) && 507 (Current.Previous->Tok.getIdentifierInfo() == nullptr || 508 Current.Previous->Tok.getIdentifierInfo()->getPPKeywordID() == 509 tok::pp_not_keyword))) { 510 unsigned EndColumn = 511 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd()); 512 if (Current.LastNewlineOffset != 0) { 513 // If there is a newline within this token, the final column will solely 514 // determined by the current end column. 515 State.Column = EndColumn; 516 } else { 517 unsigned StartColumn = 518 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin()); 519 assert(EndColumn >= StartColumn); 520 State.Column += EndColumn - StartColumn; 521 } 522 moveStateToNextToken(State, DryRun, /*Newline=*/false); 523 return 0; 524 } 525 526 unsigned Penalty = 0; 527 if (Newline) 528 Penalty = addTokenOnNewLine(State, DryRun); 529 else 530 addTokenOnCurrentLine(State, DryRun, ExtraSpaces); 531 532 return moveStateToNextToken(State, DryRun, Newline) + Penalty; 533 } 534 535 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun, 536 unsigned ExtraSpaces) { 537 FormatToken &Current = *State.NextToken; 538 const FormatToken &Previous = *State.NextToken->Previous; 539 if (Current.is(tok::equal) && 540 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) && 541 State.Stack.back().VariablePos == 0) { 542 State.Stack.back().VariablePos = State.Column; 543 // Move over * and & if they are bound to the variable name. 544 const FormatToken *Tok = &Previous; 545 while (Tok && State.Stack.back().VariablePos >= Tok->ColumnWidth) { 546 State.Stack.back().VariablePos -= Tok->ColumnWidth; 547 if (Tok->SpacesRequiredBefore != 0) 548 break; 549 Tok = Tok->Previous; 550 } 551 if (Previous.PartOfMultiVariableDeclStmt) 552 State.Stack.back().LastSpace = State.Stack.back().VariablePos; 553 } 554 555 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces; 556 557 // Indent preprocessor directives after the hash if required. 558 int PPColumnCorrection = 0; 559 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 560 Previous.is(tok::hash) && State.FirstIndent > 0 && 561 (State.Line->Type == LT_PreprocessorDirective || 562 State.Line->Type == LT_ImportStatement)) { 563 Spaces += State.FirstIndent; 564 565 // For preprocessor indent with tabs, State.Column will be 1 because of the 566 // hash. This causes second-level indents onward to have an extra space 567 // after the tabs. We avoid this misalignment by subtracting 1 from the 568 // column value passed to replaceWhitespace(). 569 if (Style.UseTab != FormatStyle::UT_Never) 570 PPColumnCorrection = -1; 571 } 572 573 if (!DryRun) 574 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces, 575 State.Column + Spaces + PPColumnCorrection); 576 577 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance 578 // declaration unless there is multiple inheritance. 579 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 580 Current.is(TT_InheritanceColon)) 581 State.Stack.back().NoLineBreak = true; 582 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon && 583 Previous.is(TT_InheritanceColon)) 584 State.Stack.back().NoLineBreak = true; 585 586 if (Current.is(TT_SelectorName) && 587 !State.Stack.back().ObjCSelectorNameFound) { 588 unsigned MinIndent = 589 std::max(State.FirstIndent + Style.ContinuationIndentWidth, 590 State.Stack.back().Indent); 591 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth; 592 if (Current.LongestObjCSelectorName == 0) 593 State.Stack.back().AlignColons = false; 594 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos) 595 State.Stack.back().ColonPos = MinIndent + Current.LongestObjCSelectorName; 596 else 597 State.Stack.back().ColonPos = FirstColonPos; 598 } 599 600 // In "AlwaysBreak" mode, enforce wrapping directly after the parenthesis by 601 // disallowing any further line breaks if there is no line break after the 602 // opening parenthesis. Don't break if it doesn't conserve columns. 603 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak && 604 Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) && 605 State.Column > getNewLineColumn(State) && 606 (!Previous.Previous || !Previous.Previous->isOneOf( 607 tok::kw_for, tok::kw_while, tok::kw_switch)) && 608 // Don't do this for simple (no expressions) one-argument function calls 609 // as that feels like needlessly wasting whitespace, e.g.: 610 // 611 // caaaaaaaaaaaall( 612 // caaaaaaaaaaaall( 613 // caaaaaaaaaaaall( 614 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa)))); 615 Current.FakeLParens.size() > 0 && 616 Current.FakeLParens.back() > prec::Unknown) 617 State.Stack.back().NoLineBreak = true; 618 if (Previous.is(TT_TemplateString) && Previous.opensScope()) 619 State.Stack.back().NoLineBreak = true; 620 621 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign && 622 Previous.opensScope() && Previous.isNot(TT_ObjCMethodExpr) && 623 (Current.isNot(TT_LineComment) || Previous.BlockKind == BK_BracedInit)) 624 State.Stack.back().Indent = State.Column + Spaces; 625 if (State.Stack.back().AvoidBinPacking && startsNextParameter(Current, Style)) 626 State.Stack.back().NoLineBreak = true; 627 if (startsSegmentOfBuilderTypeCall(Current) && 628 State.Column > getNewLineColumn(State)) 629 State.Stack.back().ContainsUnwrappedBuilder = true; 630 631 if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java) 632 State.Stack.back().NoLineBreak = true; 633 if (Current.isMemberAccess() && Previous.is(tok::r_paren) && 634 (Previous.MatchingParen && 635 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) 636 // If there is a function call with long parameters, break before trailing 637 // calls. This prevents things like: 638 // EXPECT_CALL(SomeLongParameter).Times( 639 // 2); 640 // We don't want to do this for short parameters as they can just be 641 // indexes. 642 State.Stack.back().NoLineBreak = true; 643 644 // Don't allow the RHS of an operator to be split over multiple lines unless 645 // there is a line-break right after the operator. 646 // Exclude relational operators, as there, it is always more desirable to 647 // have the LHS 'left' of the RHS. 648 const FormatToken *P = Current.getPreviousNonComment(); 649 if (!Current.is(tok::comment) && P && 650 (P->isOneOf(TT_BinaryOperator, tok::comma) || 651 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) && 652 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) && 653 P->getPrecedence() != prec::Assignment && 654 P->getPrecedence() != prec::Relational && 655 P->getPrecedence() != prec::Spaceship) { 656 bool BreakBeforeOperator = 657 P->MustBreakBefore || P->is(tok::lessless) || 658 (P->is(TT_BinaryOperator) && 659 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) || 660 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators); 661 // Don't do this if there are only two operands. In these cases, there is 662 // always a nice vertical separation between them and the extra line break 663 // does not help. 664 bool HasTwoOperands = 665 P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr); 666 if ((!BreakBeforeOperator && !(HasTwoOperands && Style.AlignOperands)) || 667 (!State.Stack.back().LastOperatorWrapped && BreakBeforeOperator)) 668 State.Stack.back().NoLineBreakInOperand = true; 669 } 670 671 State.Column += Spaces; 672 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) && 673 Previous.Previous && 674 (Previous.Previous->isOneOf(tok::kw_if, tok::kw_for) || 675 Previous.Previous->endsSequence(tok::kw_constexpr, tok::kw_if))) { 676 // Treat the condition inside an if as if it was a second function 677 // parameter, i.e. let nested calls have a continuation indent. 678 State.Stack.back().LastSpace = State.Column; 679 State.Stack.back().NestedBlockIndent = State.Column; 680 } else if (!Current.isOneOf(tok::comment, tok::caret) && 681 ((Previous.is(tok::comma) && 682 !Previous.is(TT_OverloadedOperator)) || 683 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) { 684 State.Stack.back().LastSpace = State.Column; 685 } else if (Previous.is(TT_CtorInitializerColon) && 686 Style.BreakConstructorInitializers == 687 FormatStyle::BCIS_AfterColon) { 688 State.Stack.back().Indent = State.Column; 689 State.Stack.back().LastSpace = State.Column; 690 } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 691 TT_CtorInitializerColon)) && 692 ((Previous.getPrecedence() != prec::Assignment && 693 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 || 694 Previous.NextOperator)) || 695 Current.StartsBinaryExpression)) { 696 // Indent relative to the RHS of the expression unless this is a simple 697 // assignment without binary expression on the RHS. Also indent relative to 698 // unary operators and the colons of constructor initializers. 699 State.Stack.back().LastSpace = State.Column; 700 } else if (Previous.is(TT_InheritanceColon)) { 701 State.Stack.back().Indent = State.Column; 702 State.Stack.back().LastSpace = State.Column; 703 } else if (Previous.opensScope()) { 704 // If a function has a trailing call, indent all parameters from the 705 // opening parenthesis. This avoids confusing indents like: 706 // OuterFunction(InnerFunctionCall( // break 707 // ParameterToInnerFunction)) // break 708 // .SecondInnerFunctionCall(); 709 bool HasTrailingCall = false; 710 if (Previous.MatchingParen) { 711 const FormatToken *Next = Previous.MatchingParen->getNextNonComment(); 712 HasTrailingCall = Next && Next->isMemberAccess(); 713 } 714 if (HasTrailingCall && State.Stack.size() > 1 && 715 State.Stack[State.Stack.size() - 2].CallContinuation == 0) 716 State.Stack.back().LastSpace = State.Column; 717 } 718 } 719 720 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State, 721 bool DryRun) { 722 FormatToken &Current = *State.NextToken; 723 const FormatToken &Previous = *State.NextToken->Previous; 724 725 // Extra penalty that needs to be added because of the way certain line 726 // breaks are chosen. 727 unsigned Penalty = 0; 728 729 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 730 const FormatToken *NextNonComment = Previous.getNextNonComment(); 731 if (!NextNonComment) 732 NextNonComment = &Current; 733 // The first line break on any NestingLevel causes an extra penalty in order 734 // prefer similar line breaks. 735 if (!State.Stack.back().ContainsLineBreak) 736 Penalty += 15; 737 State.Stack.back().ContainsLineBreak = true; 738 739 Penalty += State.NextToken->SplitPenalty; 740 741 // Breaking before the first "<<" is generally not desirable if the LHS is 742 // short. Also always add the penalty if the LHS is split over multiple lines 743 // to avoid unnecessary line breaks that just work around this penalty. 744 if (NextNonComment->is(tok::lessless) && 745 State.Stack.back().FirstLessLess == 0 && 746 (State.Column <= Style.ColumnLimit / 3 || 747 State.Stack.back().BreakBeforeParameter)) 748 Penalty += Style.PenaltyBreakFirstLessLess; 749 750 State.Column = getNewLineColumn(State); 751 752 // Indent nested blocks relative to this column, unless in a very specific 753 // JavaScript special case where: 754 // 755 // var loooooong_name = 756 // function() { 757 // // code 758 // } 759 // 760 // is common and should be formatted like a free-standing function. The same 761 // goes for wrapping before the lambda return type arrow. 762 if (!Current.is(TT_LambdaArrow) && 763 (Style.Language != FormatStyle::LK_JavaScript || 764 Current.NestingLevel != 0 || !PreviousNonComment || 765 !PreviousNonComment->is(tok::equal) || 766 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) 767 State.Stack.back().NestedBlockIndent = State.Column; 768 769 if (NextNonComment->isMemberAccess()) { 770 if (State.Stack.back().CallContinuation == 0) 771 State.Stack.back().CallContinuation = State.Column; 772 } else if (NextNonComment->is(TT_SelectorName)) { 773 if (!State.Stack.back().ObjCSelectorNameFound) { 774 if (NextNonComment->LongestObjCSelectorName == 0) { 775 State.Stack.back().AlignColons = false; 776 } else { 777 State.Stack.back().ColonPos = 778 (shouldIndentWrappedSelectorName(Style, State.Line->Type) 779 ? std::max(State.Stack.back().Indent, 780 State.FirstIndent + Style.ContinuationIndentWidth) 781 : State.Stack.back().Indent) + 782 std::max(NextNonComment->LongestObjCSelectorName, 783 NextNonComment->ColumnWidth); 784 } 785 } else if (State.Stack.back().AlignColons && 786 State.Stack.back().ColonPos <= NextNonComment->ColumnWidth) { 787 State.Stack.back().ColonPos = State.Column + NextNonComment->ColumnWidth; 788 } 789 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 790 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) { 791 // FIXME: This is hacky, find a better way. The problem is that in an ObjC 792 // method expression, the block should be aligned to the line starting it, 793 // e.g.: 794 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason 795 // ^(int *i) { 796 // // ... 797 // }]; 798 // Thus, we set LastSpace of the next higher NestingLevel, to which we move 799 // when we consume all of the "}"'s FakeRParens at the "{". 800 if (State.Stack.size() > 1) 801 State.Stack[State.Stack.size() - 2].LastSpace = 802 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 803 Style.ContinuationIndentWidth; 804 } 805 806 if ((PreviousNonComment && 807 PreviousNonComment->isOneOf(tok::comma, tok::semi) && 808 !State.Stack.back().AvoidBinPacking) || 809 Previous.is(TT_BinaryOperator)) 810 State.Stack.back().BreakBeforeParameter = false; 811 if (PreviousNonComment && 812 PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) && 813 Current.NestingLevel == 0) 814 State.Stack.back().BreakBeforeParameter = false; 815 if (NextNonComment->is(tok::question) || 816 (PreviousNonComment && PreviousNonComment->is(tok::question))) 817 State.Stack.back().BreakBeforeParameter = true; 818 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore) 819 State.Stack.back().BreakBeforeParameter = false; 820 821 if (!DryRun) { 822 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1; 823 if (Current.is(tok::r_brace) && Current.MatchingParen && 824 // Only strip trailing empty lines for l_braces that have children, i.e. 825 // for function expressions (lambdas, arrows, etc). 826 !Current.MatchingParen->Children.empty()) { 827 // lambdas and arrow functions are expressions, thus their r_brace is not 828 // on its own line, and thus not covered by UnwrappedLineFormatter's logic 829 // about removing empty lines on closing blocks. Special case them here. 830 MaxEmptyLinesToKeep = 1; 831 } 832 unsigned Newlines = std::max( 833 1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep)); 834 bool ContinuePPDirective = 835 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement; 836 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column, 837 ContinuePPDirective); 838 } 839 840 if (!Current.isTrailingComment()) 841 State.Stack.back().LastSpace = State.Column; 842 if (Current.is(tok::lessless)) 843 // If we are breaking before a "<<", we always want to indent relative to 844 // RHS. This is necessary only for "<<", as we special-case it and don't 845 // always indent relative to the RHS. 846 State.Stack.back().LastSpace += 3; // 3 -> width of "<< ". 847 848 State.StartOfLineLevel = Current.NestingLevel; 849 State.LowestLevelOnLine = Current.NestingLevel; 850 851 // Any break on this level means that the parent level has been broken 852 // and we need to avoid bin packing there. 853 bool NestedBlockSpecialCase = 854 !Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 && 855 State.Stack[State.Stack.size() - 2].NestedBlockInlined; 856 if (!NestedBlockSpecialCase) 857 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 858 State.Stack[i].BreakBeforeParameter = true; 859 860 if (PreviousNonComment && 861 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) && 862 (PreviousNonComment->isNot(TT_TemplateCloser) || 863 Current.NestingLevel != 0) && 864 !PreviousNonComment->isOneOf( 865 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation, 866 TT_LeadingJavaAnnotation) && 867 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope()) 868 State.Stack.back().BreakBeforeParameter = true; 869 870 // If we break after { or the [ of an array initializer, we should also break 871 // before the corresponding } or ]. 872 if (PreviousNonComment && 873 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 874 opensProtoMessageField(*PreviousNonComment, Style))) 875 State.Stack.back().BreakBeforeClosingBrace = true; 876 877 if (State.Stack.back().AvoidBinPacking) { 878 // If we are breaking after '(', '{', '<', this is not bin packing 879 // unless AllowAllParametersOfDeclarationOnNextLine is false or this is a 880 // dict/object literal. 881 if (!Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) || 882 (!Style.AllowAllParametersOfDeclarationOnNextLine && 883 State.Line->MustBeDeclaration) || 884 Previous.is(TT_DictLiteral)) 885 State.Stack.back().BreakBeforeParameter = true; 886 } 887 888 return Penalty; 889 } 890 891 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) { 892 if (!State.NextToken || !State.NextToken->Previous) 893 return 0; 894 FormatToken &Current = *State.NextToken; 895 const FormatToken &Previous = *Current.Previous; 896 // If we are continuing an expression, we want to use the continuation indent. 897 unsigned ContinuationIndent = 898 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 899 Style.ContinuationIndentWidth; 900 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 901 const FormatToken *NextNonComment = Previous.getNextNonComment(); 902 if (!NextNonComment) 903 NextNonComment = &Current; 904 905 // Java specific bits. 906 if (Style.Language == FormatStyle::LK_Java && 907 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) 908 return std::max(State.Stack.back().LastSpace, 909 State.Stack.back().Indent + Style.ContinuationIndentWidth); 910 911 if (NextNonComment->is(tok::l_brace) && NextNonComment->BlockKind == BK_Block) 912 return Current.NestingLevel == 0 ? State.FirstIndent 913 : State.Stack.back().Indent; 914 if ((Current.isOneOf(tok::r_brace, tok::r_square) || 915 (Current.is(tok::greater) && 916 (Style.Language == FormatStyle::LK_Proto || 917 Style.Language == FormatStyle::LK_TextProto))) && 918 State.Stack.size() > 1) { 919 if (Current.closesBlockOrBlockTypeList(Style)) 920 return State.Stack[State.Stack.size() - 2].NestedBlockIndent; 921 if (Current.MatchingParen && 922 Current.MatchingParen->BlockKind == BK_BracedInit) 923 return State.Stack[State.Stack.size() - 2].LastSpace; 924 return State.FirstIndent; 925 } 926 // Indent a closing parenthesis at the previous level if followed by a semi or 927 // opening brace. This allows indentations such as: 928 // foo( 929 // a, 930 // ); 931 // function foo( 932 // a, 933 // ) { 934 // code(); // 935 // } 936 if (Current.is(tok::r_paren) && State.Stack.size() > 1 && 937 (!Current.Next || Current.Next->isOneOf(tok::semi, tok::l_brace))) 938 return State.Stack[State.Stack.size() - 2].LastSpace; 939 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope()) 940 return State.Stack[State.Stack.size() - 2].LastSpace; 941 if (Current.is(tok::identifier) && Current.Next && 942 (Current.Next->is(TT_DictLiteral) || 943 ((Style.Language == FormatStyle::LK_Proto || 944 Style.Language == FormatStyle::LK_TextProto) && 945 Current.Next->isOneOf(tok::less, tok::l_brace)))) 946 return State.Stack.back().Indent; 947 if (NextNonComment->is(TT_ObjCStringLiteral) && 948 State.StartOfStringLiteral != 0) 949 return State.StartOfStringLiteral - 1; 950 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0) 951 return State.StartOfStringLiteral; 952 if (NextNonComment->is(tok::lessless) && 953 State.Stack.back().FirstLessLess != 0) 954 return State.Stack.back().FirstLessLess; 955 if (NextNonComment->isMemberAccess()) { 956 if (State.Stack.back().CallContinuation == 0) 957 return ContinuationIndent; 958 return State.Stack.back().CallContinuation; 959 } 960 if (State.Stack.back().QuestionColumn != 0 && 961 ((NextNonComment->is(tok::colon) && 962 NextNonComment->is(TT_ConditionalExpr)) || 963 Previous.is(TT_ConditionalExpr))) 964 return State.Stack.back().QuestionColumn; 965 if (Previous.is(tok::comma) && State.Stack.back().VariablePos != 0) 966 return State.Stack.back().VariablePos; 967 if ((PreviousNonComment && 968 (PreviousNonComment->ClosesTemplateDeclaration || 969 PreviousNonComment->isOneOf( 970 TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen, 971 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) || 972 (!Style.IndentWrappedFunctionNames && 973 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) 974 return std::max(State.Stack.back().LastSpace, State.Stack.back().Indent); 975 if (NextNonComment->is(TT_SelectorName)) { 976 if (!State.Stack.back().ObjCSelectorNameFound) { 977 unsigned MinIndent = State.Stack.back().Indent; 978 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) 979 MinIndent = std::max(MinIndent, 980 State.FirstIndent + Style.ContinuationIndentWidth); 981 // If LongestObjCSelectorName is 0, we are indenting the first 982 // part of an ObjC selector (or a selector component which is 983 // not colon-aligned due to block formatting). 984 // 985 // Otherwise, we are indenting a subsequent part of an ObjC 986 // selector which should be colon-aligned to the longest 987 // component of the ObjC selector. 988 // 989 // In either case, we want to respect Style.IndentWrappedFunctionNames. 990 return MinIndent + 991 std::max(NextNonComment->LongestObjCSelectorName, 992 NextNonComment->ColumnWidth) - 993 NextNonComment->ColumnWidth; 994 } 995 if (!State.Stack.back().AlignColons) 996 return State.Stack.back().Indent; 997 if (State.Stack.back().ColonPos > NextNonComment->ColumnWidth) 998 return State.Stack.back().ColonPos - NextNonComment->ColumnWidth; 999 return State.Stack.back().Indent; 1000 } 1001 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr)) 1002 return State.Stack.back().ColonPos; 1003 if (NextNonComment->is(TT_ArraySubscriptLSquare)) { 1004 if (State.Stack.back().StartOfArraySubscripts != 0) 1005 return State.Stack.back().StartOfArraySubscripts; 1006 return ContinuationIndent; 1007 } 1008 1009 // This ensure that we correctly format ObjC methods calls without inputs, 1010 // i.e. where the last element isn't selector like: [callee method]; 1011 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 && 1012 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) 1013 return State.Stack.back().Indent; 1014 1015 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) || 1016 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) 1017 return ContinuationIndent; 1018 if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 1019 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) 1020 return ContinuationIndent; 1021 if (NextNonComment->is(TT_CtorInitializerComma)) 1022 return State.Stack.back().Indent; 1023 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && 1024 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) 1025 return State.Stack.back().Indent; 1026 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) && 1027 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) 1028 return State.Stack.back().Indent; 1029 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon, 1030 TT_InheritanceComma)) 1031 return State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1032 if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() && 1033 !Current.isOneOf(tok::colon, tok::comment)) 1034 return ContinuationIndent; 1035 if (Current.is(TT_ProtoExtensionLSquare)) 1036 return State.Stack.back().Indent; 1037 if (State.Stack.back().Indent == State.FirstIndent && PreviousNonComment && 1038 PreviousNonComment->isNot(tok::r_brace)) 1039 // Ensure that we fall back to the continuation indent width instead of 1040 // just flushing continuations left. 1041 return State.Stack.back().Indent + Style.ContinuationIndentWidth; 1042 return State.Stack.back().Indent; 1043 } 1044 1045 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State, 1046 bool DryRun, bool Newline) { 1047 assert(State.Stack.size()); 1048 const FormatToken &Current = *State.NextToken; 1049 1050 if (Current.isOneOf(tok::comma, TT_BinaryOperator)) 1051 State.Stack.back().NoLineBreakInOperand = false; 1052 if (Current.is(TT_InheritanceColon)) 1053 State.Stack.back().AvoidBinPacking = true; 1054 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) { 1055 if (State.Stack.back().FirstLessLess == 0) 1056 State.Stack.back().FirstLessLess = State.Column; 1057 else 1058 State.Stack.back().LastOperatorWrapped = Newline; 1059 } 1060 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless)) 1061 State.Stack.back().LastOperatorWrapped = Newline; 1062 if (Current.is(TT_ConditionalExpr) && Current.Previous && 1063 !Current.Previous->is(TT_ConditionalExpr)) 1064 State.Stack.back().LastOperatorWrapped = Newline; 1065 if (Current.is(TT_ArraySubscriptLSquare) && 1066 State.Stack.back().StartOfArraySubscripts == 0) 1067 State.Stack.back().StartOfArraySubscripts = State.Column; 1068 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question)) 1069 State.Stack.back().QuestionColumn = State.Column; 1070 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) { 1071 const FormatToken *Previous = Current.Previous; 1072 while (Previous && Previous->isTrailingComment()) 1073 Previous = Previous->Previous; 1074 if (Previous && Previous->is(tok::question)) 1075 State.Stack.back().QuestionColumn = State.Column; 1076 } 1077 if (!Current.opensScope() && !Current.closesScope() && 1078 !Current.is(TT_PointerOrReference)) 1079 State.LowestLevelOnLine = 1080 std::min(State.LowestLevelOnLine, Current.NestingLevel); 1081 if (Current.isMemberAccess()) 1082 State.Stack.back().StartOfFunctionCall = 1083 !Current.NextOperator ? 0 : State.Column; 1084 if (Current.is(TT_SelectorName)) 1085 State.Stack.back().ObjCSelectorNameFound = true; 1086 if (Current.is(TT_CtorInitializerColon) && 1087 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) { 1088 // Indent 2 from the column, so: 1089 // SomeClass::SomeClass() 1090 // : First(...), ... 1091 // Next(...) 1092 // ^ line up here. 1093 State.Stack.back().Indent = 1094 State.Column + 1095 (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma 1096 ? 0 1097 : 2); 1098 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1099 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 1100 State.Stack.back().AvoidBinPacking = true; 1101 State.Stack.back().BreakBeforeParameter = false; 1102 } 1103 if (Current.is(TT_CtorInitializerColon) && 1104 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { 1105 State.Stack.back().Indent = 1106 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1107 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1108 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 1109 State.Stack.back().AvoidBinPacking = true; 1110 } 1111 if (Current.is(TT_InheritanceColon)) 1112 State.Stack.back().Indent = 1113 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1114 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline) 1115 State.Stack.back().NestedBlockIndent = 1116 State.Column + Current.ColumnWidth + 1; 1117 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow)) 1118 State.Stack.back().LastSpace = State.Column; 1119 1120 // Insert scopes created by fake parenthesis. 1121 const FormatToken *Previous = Current.getPreviousNonComment(); 1122 1123 // Add special behavior to support a format commonly used for JavaScript 1124 // closures: 1125 // SomeFunction(function() { 1126 // foo(); 1127 // bar(); 1128 // }, a, b, c); 1129 if (Current.isNot(tok::comment) && Previous && 1130 Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) && 1131 !Previous->is(TT_DictLiteral) && State.Stack.size() > 1) { 1132 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline) 1133 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 1134 State.Stack[i].NoLineBreak = true; 1135 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false; 1136 } 1137 if (Previous && 1138 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) || 1139 Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr)) && 1140 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 1141 State.Stack.back().NestedBlockInlined = 1142 !Newline && 1143 (Previous->isNot(tok::l_paren) || Previous->ParameterCount > 1); 1144 } 1145 1146 moveStatePastFakeLParens(State, Newline); 1147 moveStatePastScopeCloser(State); 1148 bool AllowBreak = !State.Stack.back().NoLineBreak && 1149 !State.Stack.back().NoLineBreakInOperand; 1150 moveStatePastScopeOpener(State, Newline); 1151 moveStatePastFakeRParens(State); 1152 1153 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0) 1154 State.StartOfStringLiteral = State.Column + 1; 1155 else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) 1156 State.StartOfStringLiteral = State.Column; 1157 else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) && 1158 !Current.isStringLiteral()) 1159 State.StartOfStringLiteral = 0; 1160 1161 State.Column += Current.ColumnWidth; 1162 State.NextToken = State.NextToken->Next; 1163 1164 unsigned Penalty = 1165 handleEndOfLine(Current, State, DryRun, AllowBreak); 1166 1167 if (Current.Role) 1168 Current.Role->formatFromToken(State, this, DryRun); 1169 // If the previous has a special role, let it consume tokens as appropriate. 1170 // It is necessary to start at the previous token for the only implemented 1171 // role (comma separated list). That way, the decision whether or not to break 1172 // after the "{" is already done and both options are tried and evaluated. 1173 // FIXME: This is ugly, find a better way. 1174 if (Previous && Previous->Role) 1175 Penalty += Previous->Role->formatAfterToken(State, this, DryRun); 1176 1177 return Penalty; 1178 } 1179 1180 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State, 1181 bool Newline) { 1182 const FormatToken &Current = *State.NextToken; 1183 const FormatToken *Previous = Current.getPreviousNonComment(); 1184 1185 // Don't add extra indentation for the first fake parenthesis after 1186 // 'return', assignments or opening <({[. The indentation for these cases 1187 // is special cased. 1188 bool SkipFirstExtraIndent = 1189 (Previous && (Previous->opensScope() || 1190 Previous->isOneOf(tok::semi, tok::kw_return) || 1191 (Previous->getPrecedence() == prec::Assignment && 1192 Style.AlignOperands) || 1193 Previous->is(TT_ObjCMethodExpr))); 1194 for (SmallVectorImpl<prec::Level>::const_reverse_iterator 1195 I = Current.FakeLParens.rbegin(), 1196 E = Current.FakeLParens.rend(); 1197 I != E; ++I) { 1198 ParenState NewParenState = State.Stack.back(); 1199 NewParenState.Tok = nullptr; 1200 NewParenState.ContainsLineBreak = false; 1201 NewParenState.LastOperatorWrapped = true; 1202 NewParenState.NoLineBreak = 1203 NewParenState.NoLineBreak || State.Stack.back().NoLineBreakInOperand; 1204 1205 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists. 1206 if (*I > prec::Comma) 1207 NewParenState.AvoidBinPacking = false; 1208 1209 // Indent from 'LastSpace' unless these are fake parentheses encapsulating 1210 // a builder type call after 'return' or, if the alignment after opening 1211 // brackets is disabled. 1212 if (!Current.isTrailingComment() && 1213 (Style.AlignOperands || *I < prec::Assignment) && 1214 (!Previous || Previous->isNot(tok::kw_return) || 1215 (Style.Language != FormatStyle::LK_Java && *I > 0)) && 1216 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign || 1217 *I != prec::Comma || Current.NestingLevel == 0)) 1218 NewParenState.Indent = 1219 std::max(std::max(State.Column, NewParenState.Indent), 1220 State.Stack.back().LastSpace); 1221 1222 // Do not indent relative to the fake parentheses inserted for "." or "->". 1223 // This is a special case to make the following to statements consistent: 1224 // OuterFunction(InnerFunctionCall( // break 1225 // ParameterToInnerFunction)); 1226 // OuterFunction(SomeObject.InnerFunctionCall( // break 1227 // ParameterToInnerFunction)); 1228 if (*I > prec::Unknown) 1229 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column); 1230 if (*I != prec::Conditional && !Current.is(TT_UnaryOperator) && 1231 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 1232 NewParenState.StartOfFunctionCall = State.Column; 1233 1234 // Always indent conditional expressions. Never indent expression where 1235 // the 'operator' is ',', ';' or an assignment (i.e. *I <= 1236 // prec::Assignment) as those have different indentation rules. Indent 1237 // other expression, unless the indentation needs to be skipped. 1238 if (*I == prec::Conditional || 1239 (!SkipFirstExtraIndent && *I > prec::Assignment && 1240 !Current.isTrailingComment())) 1241 NewParenState.Indent += Style.ContinuationIndentWidth; 1242 if ((Previous && !Previous->opensScope()) || *I != prec::Comma) 1243 NewParenState.BreakBeforeParameter = false; 1244 State.Stack.push_back(NewParenState); 1245 SkipFirstExtraIndent = false; 1246 } 1247 } 1248 1249 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) { 1250 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) { 1251 unsigned VariablePos = State.Stack.back().VariablePos; 1252 if (State.Stack.size() == 1) { 1253 // Do not pop the last element. 1254 break; 1255 } 1256 State.Stack.pop_back(); 1257 State.Stack.back().VariablePos = VariablePos; 1258 } 1259 } 1260 1261 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State, 1262 bool Newline) { 1263 const FormatToken &Current = *State.NextToken; 1264 if (!Current.opensScope()) 1265 return; 1266 1267 if (Current.MatchingParen && Current.BlockKind == BK_Block) { 1268 moveStateToNewBlock(State); 1269 return; 1270 } 1271 1272 unsigned NewIndent; 1273 unsigned LastSpace = State.Stack.back().LastSpace; 1274 bool AvoidBinPacking; 1275 bool BreakBeforeParameter = false; 1276 unsigned NestedBlockIndent = std::max(State.Stack.back().StartOfFunctionCall, 1277 State.Stack.back().NestedBlockIndent); 1278 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 1279 opensProtoMessageField(Current, Style)) { 1280 if (Current.opensBlockOrBlockTypeList(Style)) { 1281 NewIndent = Style.IndentWidth + 1282 std::min(State.Column, State.Stack.back().NestedBlockIndent); 1283 } else { 1284 NewIndent = State.Stack.back().LastSpace + Style.ContinuationIndentWidth; 1285 } 1286 const FormatToken *NextNoComment = Current.getNextNonComment(); 1287 bool EndsInComma = Current.MatchingParen && 1288 Current.MatchingParen->Previous && 1289 Current.MatchingParen->Previous->is(tok::comma); 1290 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) || 1291 Style.Language == FormatStyle::LK_Proto || 1292 Style.Language == FormatStyle::LK_TextProto || 1293 !Style.BinPackArguments || 1294 (NextNoComment && 1295 NextNoComment->isOneOf(TT_DesignatedInitializerPeriod, 1296 TT_DesignatedInitializerLSquare)); 1297 BreakBeforeParameter = EndsInComma; 1298 if (Current.ParameterCount > 1) 1299 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1); 1300 } else { 1301 NewIndent = Style.ContinuationIndentWidth + 1302 std::max(State.Stack.back().LastSpace, 1303 State.Stack.back().StartOfFunctionCall); 1304 1305 // Ensure that different different brackets force relative alignment, e.g.: 1306 // void SomeFunction(vector< // break 1307 // int> v); 1308 // FIXME: We likely want to do this for more combinations of brackets. 1309 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) { 1310 NewIndent = std::max(NewIndent, State.Stack.back().Indent); 1311 LastSpace = std::max(LastSpace, State.Stack.back().Indent); 1312 } 1313 1314 bool EndsInComma = 1315 Current.MatchingParen && 1316 Current.MatchingParen->getPreviousNonComment() && 1317 Current.MatchingParen->getPreviousNonComment()->is(tok::comma); 1318 1319 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters 1320 // for backwards compatibility. 1321 bool ObjCBinPackProtocolList = 1322 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto && 1323 Style.BinPackParameters) || 1324 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always; 1325 1326 bool BinPackDeclaration = 1327 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) || 1328 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList); 1329 1330 AvoidBinPacking = 1331 (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) || 1332 (State.Line->MustBeDeclaration && !BinPackDeclaration) || 1333 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) || 1334 (Style.ExperimentalAutoDetectBinPacking && 1335 (Current.PackingKind == PPK_OnePerLine || 1336 (!BinPackInconclusiveFunctions && 1337 Current.PackingKind == PPK_Inconclusive))); 1338 1339 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen) { 1340 if (Style.ColumnLimit) { 1341 // If this '[' opens an ObjC call, determine whether all parameters fit 1342 // into one line and put one per line if they don't. 1343 if (getLengthToMatchingParen(Current, State.Stack) + State.Column > 1344 getColumnLimit(State)) 1345 BreakBeforeParameter = true; 1346 } else { 1347 // For ColumnLimit = 0, we have to figure out whether there is or has to 1348 // be a line break within this call. 1349 for (const FormatToken *Tok = &Current; 1350 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) { 1351 if (Tok->MustBreakBefore || 1352 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) { 1353 BreakBeforeParameter = true; 1354 break; 1355 } 1356 } 1357 } 1358 } 1359 1360 if (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) 1361 BreakBeforeParameter = true; 1362 } 1363 // Generally inherit NoLineBreak from the current scope to nested scope. 1364 // However, don't do this for non-empty nested blocks, dict literals and 1365 // array literals as these follow different indentation rules. 1366 bool NoLineBreak = 1367 Current.Children.empty() && 1368 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) && 1369 (State.Stack.back().NoLineBreak || 1370 State.Stack.back().NoLineBreakInOperand || 1371 (Current.is(TT_TemplateOpener) && 1372 State.Stack.back().ContainsUnwrappedBuilder)); 1373 State.Stack.push_back( 1374 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak)); 1375 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1376 State.Stack.back().BreakBeforeParameter = BreakBeforeParameter; 1377 State.Stack.back().HasMultipleNestedBlocks = Current.BlockParameterCount > 1; 1378 State.Stack.back().IsInsideObjCArrayLiteral = 1379 Current.is(TT_ArrayInitializerLSquare) && Current.Previous && 1380 Current.Previous->is(tok::at); 1381 } 1382 1383 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) { 1384 const FormatToken &Current = *State.NextToken; 1385 if (!Current.closesScope()) 1386 return; 1387 1388 // If we encounter a closing ), ], } or >, we can remove a level from our 1389 // stacks. 1390 if (State.Stack.size() > 1 && 1391 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) || 1392 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) || 1393 State.NextToken->is(TT_TemplateCloser) || 1394 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) 1395 State.Stack.pop_back(); 1396 1397 if (Current.is(tok::r_square)) { 1398 // If this ends the array subscript expr, reset the corresponding value. 1399 const FormatToken *NextNonComment = Current.getNextNonComment(); 1400 if (NextNonComment && NextNonComment->isNot(tok::l_square)) 1401 State.Stack.back().StartOfArraySubscripts = 0; 1402 } 1403 } 1404 1405 void ContinuationIndenter::moveStateToNewBlock(LineState &State) { 1406 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent; 1407 // ObjC block sometimes follow special indentation rules. 1408 unsigned NewIndent = 1409 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace) 1410 ? Style.ObjCBlockIndentWidth 1411 : Style.IndentWidth); 1412 State.Stack.push_back(ParenState(State.NextToken, NewIndent, 1413 State.Stack.back().LastSpace, 1414 /*AvoidBinPacking=*/true, 1415 /*NoLineBreak=*/false)); 1416 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1417 State.Stack.back().BreakBeforeParameter = true; 1418 } 1419 1420 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn, 1421 unsigned TabWidth, 1422 encoding::Encoding Encoding) { 1423 size_t LastNewlinePos = Text.find_last_of("\n"); 1424 if (LastNewlinePos == StringRef::npos) { 1425 return StartColumn + 1426 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding); 1427 } else { 1428 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos), 1429 /*StartColumn=*/0, TabWidth, Encoding); 1430 } 1431 } 1432 1433 unsigned ContinuationIndenter::reformatRawStringLiteral( 1434 const FormatToken &Current, LineState &State, 1435 const FormatStyle &RawStringStyle, bool DryRun) { 1436 unsigned StartColumn = State.Column - Current.ColumnWidth; 1437 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText); 1438 StringRef NewDelimiter = 1439 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language); 1440 if (NewDelimiter.empty() || OldDelimiter.empty()) 1441 NewDelimiter = OldDelimiter; 1442 // The text of a raw string is between the leading 'R"delimiter(' and the 1443 // trailing 'delimiter)"'. 1444 unsigned OldPrefixSize = 3 + OldDelimiter.size(); 1445 unsigned OldSuffixSize = 2 + OldDelimiter.size(); 1446 // We create a virtual text environment which expects a null-terminated 1447 // string, so we cannot use StringRef. 1448 std::string RawText = 1449 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize); 1450 if (NewDelimiter != OldDelimiter) { 1451 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the 1452 // raw string. 1453 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str(); 1454 if (StringRef(RawText).contains(CanonicalDelimiterSuffix)) 1455 NewDelimiter = OldDelimiter; 1456 } 1457 1458 unsigned NewPrefixSize = 3 + NewDelimiter.size(); 1459 unsigned NewSuffixSize = 2 + NewDelimiter.size(); 1460 1461 // The first start column is the column the raw text starts after formatting. 1462 unsigned FirstStartColumn = StartColumn + NewPrefixSize; 1463 1464 // The next start column is the intended indentation a line break inside 1465 // the raw string at level 0. It is determined by the following rules: 1466 // - if the content starts on newline, it is one level more than the current 1467 // indent, and 1468 // - if the content does not start on a newline, it is the first start 1469 // column. 1470 // These rules have the advantage that the formatted content both does not 1471 // violate the rectangle rule and visually flows within the surrounding 1472 // source. 1473 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n'; 1474 unsigned NextStartColumn = 1475 ContentStartsOnNewline 1476 ? State.Stack.back().NestedBlockIndent + Style.IndentWidth 1477 : FirstStartColumn; 1478 1479 // The last start column is the column the raw string suffix starts if it is 1480 // put on a newline. 1481 // The last start column is the intended indentation of the raw string postfix 1482 // if it is put on a newline. It is determined by the following rules: 1483 // - if the raw string prefix starts on a newline, it is the column where 1484 // that raw string prefix starts, and 1485 // - if the raw string prefix does not start on a newline, it is the current 1486 // indent. 1487 unsigned LastStartColumn = Current.NewlinesBefore 1488 ? FirstStartColumn - NewPrefixSize 1489 : State.Stack.back().NestedBlockIndent; 1490 1491 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat( 1492 RawStringStyle, RawText, {tooling::Range(0, RawText.size())}, 1493 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>", 1494 /*Status=*/nullptr); 1495 1496 auto NewCode = applyAllReplacements(RawText, Fixes.first); 1497 tooling::Replacements NoFixes; 1498 if (!NewCode) { 1499 State.Column += Current.ColumnWidth; 1500 return 0; 1501 } 1502 if (!DryRun) { 1503 if (NewDelimiter != OldDelimiter) { 1504 // In 'R"delimiter(...', the delimiter starts 2 characters after the start 1505 // of the token. 1506 SourceLocation PrefixDelimiterStart = 1507 Current.Tok.getLocation().getLocWithOffset(2); 1508 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement( 1509 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1510 if (PrefixErr) { 1511 llvm::errs() 1512 << "Failed to update the prefix delimiter of a raw string: " 1513 << llvm::toString(std::move(PrefixErr)) << "\n"; 1514 } 1515 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at 1516 // position length - 1 - |delimiter|. 1517 SourceLocation SuffixDelimiterStart = 1518 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() - 1519 1 - OldDelimiter.size()); 1520 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement( 1521 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1522 if (SuffixErr) { 1523 llvm::errs() 1524 << "Failed to update the suffix delimiter of a raw string: " 1525 << llvm::toString(std::move(SuffixErr)) << "\n"; 1526 } 1527 } 1528 SourceLocation OriginLoc = 1529 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize); 1530 for (const tooling::Replacement &Fix : Fixes.first) { 1531 auto Err = Whitespaces.addReplacement(tooling::Replacement( 1532 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()), 1533 Fix.getLength(), Fix.getReplacementText())); 1534 if (Err) { 1535 llvm::errs() << "Failed to reformat raw string: " 1536 << llvm::toString(std::move(Err)) << "\n"; 1537 } 1538 } 1539 } 1540 unsigned RawLastLineEndColumn = getLastLineEndColumn( 1541 *NewCode, FirstStartColumn, Style.TabWidth, Encoding); 1542 State.Column = RawLastLineEndColumn + NewSuffixSize; 1543 // Since we're updating the column to after the raw string literal here, we 1544 // have to manually add the penalty for the prefix R"delim( over the column 1545 // limit. 1546 unsigned PrefixExcessCharacters = 1547 StartColumn + NewPrefixSize > Style.ColumnLimit ? 1548 StartColumn + NewPrefixSize - Style.ColumnLimit : 0; 1549 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter; 1550 } 1551 1552 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current, 1553 LineState &State) { 1554 // Break before further function parameters on all levels. 1555 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 1556 State.Stack[i].BreakBeforeParameter = true; 1557 1558 unsigned ColumnsUsed = State.Column; 1559 // We can only affect layout of the first and the last line, so the penalty 1560 // for all other lines is constant, and we ignore it. 1561 State.Column = Current.LastLineColumnWidth; 1562 1563 if (ColumnsUsed > getColumnLimit(State)) 1564 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State)); 1565 return 0; 1566 } 1567 1568 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current, 1569 LineState &State, bool DryRun, 1570 bool AllowBreak) { 1571 unsigned Penalty = 0; 1572 // Compute the raw string style to use in case this is a raw string literal 1573 // that can be reformatted. 1574 auto RawStringStyle = getRawStringStyle(Current, State); 1575 if (RawStringStyle && !Current.Finalized) { 1576 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun); 1577 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) { 1578 // Don't break multi-line tokens other than block comments and raw string 1579 // literals. Instead, just update the state. 1580 Penalty = addMultilineToken(Current, State); 1581 } else if (State.Line->Type != LT_ImportStatement) { 1582 // We generally don't break import statements. 1583 LineState OriginalState = State; 1584 1585 // Whether we force the reflowing algorithm to stay strictly within the 1586 // column limit. 1587 bool Strict = false; 1588 // Whether the first non-strict attempt at reflowing did intentionally 1589 // exceed the column limit. 1590 bool Exceeded = false; 1591 std::tie(Penalty, Exceeded) = breakProtrudingToken( 1592 Current, State, AllowBreak, /*DryRun=*/true, Strict); 1593 if (Exceeded) { 1594 // If non-strict reflowing exceeds the column limit, try whether strict 1595 // reflowing leads to an overall lower penalty. 1596 LineState StrictState = OriginalState; 1597 unsigned StrictPenalty = 1598 breakProtrudingToken(Current, StrictState, AllowBreak, 1599 /*DryRun=*/true, /*Strict=*/true) 1600 .first; 1601 Strict = StrictPenalty <= Penalty; 1602 if (Strict) { 1603 Penalty = StrictPenalty; 1604 State = StrictState; 1605 } 1606 } 1607 if (!DryRun) { 1608 // If we're not in dry-run mode, apply the changes with the decision on 1609 // strictness made above. 1610 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false, 1611 Strict); 1612 } 1613 } 1614 if (State.Column > getColumnLimit(State)) { 1615 unsigned ExcessCharacters = State.Column - getColumnLimit(State); 1616 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters; 1617 } 1618 return Penalty; 1619 } 1620 1621 // Returns the enclosing function name of a token, or the empty string if not 1622 // found. 1623 static StringRef getEnclosingFunctionName(const FormatToken &Current) { 1624 // Look for: 'function(' or 'function<templates>(' before Current. 1625 auto Tok = Current.getPreviousNonComment(); 1626 if (!Tok || !Tok->is(tok::l_paren)) 1627 return ""; 1628 Tok = Tok->getPreviousNonComment(); 1629 if (!Tok) 1630 return ""; 1631 if (Tok->is(TT_TemplateCloser)) { 1632 Tok = Tok->MatchingParen; 1633 if (Tok) 1634 Tok = Tok->getPreviousNonComment(); 1635 } 1636 if (!Tok || !Tok->is(tok::identifier)) 1637 return ""; 1638 return Tok->TokenText; 1639 } 1640 1641 llvm::Optional<FormatStyle> 1642 ContinuationIndenter::getRawStringStyle(const FormatToken &Current, 1643 const LineState &State) { 1644 if (!Current.isStringLiteral()) 1645 return None; 1646 auto Delimiter = getRawStringDelimiter(Current.TokenText); 1647 if (!Delimiter) 1648 return None; 1649 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter); 1650 if (!RawStringStyle) 1651 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle( 1652 getEnclosingFunctionName(Current)); 1653 if (!RawStringStyle) 1654 return None; 1655 RawStringStyle->ColumnLimit = getColumnLimit(State); 1656 return RawStringStyle; 1657 } 1658 1659 std::unique_ptr<BreakableToken> ContinuationIndenter::createBreakableToken( 1660 const FormatToken &Current, LineState &State, bool AllowBreak) { 1661 unsigned StartColumn = State.Column - Current.ColumnWidth; 1662 if (Current.isStringLiteral()) { 1663 // FIXME: String literal breaking is currently disabled for Java and JS, as 1664 // it requires strings to be merged using "+" which we don't support. 1665 if (Style.Language == FormatStyle::LK_Java || 1666 Style.Language == FormatStyle::LK_JavaScript || 1667 !Style.BreakStringLiterals || 1668 !AllowBreak) 1669 return nullptr; 1670 1671 // Don't break string literals inside preprocessor directives (except for 1672 // #define directives, as their contents are stored in separate lines and 1673 // are not affected by this check). 1674 // This way we avoid breaking code with line directives and unknown 1675 // preprocessor directives that contain long string literals. 1676 if (State.Line->Type == LT_PreprocessorDirective) 1677 return nullptr; 1678 // Exempts unterminated string literals from line breaking. The user will 1679 // likely want to terminate the string before any line breaking is done. 1680 if (Current.IsUnterminatedLiteral) 1681 return nullptr; 1682 // Don't break string literals inside Objective-C array literals (doing so 1683 // raises the warning -Wobjc-string-concatenation). 1684 if (State.Stack.back().IsInsideObjCArrayLiteral) { 1685 return nullptr; 1686 } 1687 1688 StringRef Text = Current.TokenText; 1689 StringRef Prefix; 1690 StringRef Postfix; 1691 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'. 1692 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to 1693 // reduce the overhead) for each FormatToken, which is a string, so that we 1694 // don't run multiple checks here on the hot path. 1695 if ((Text.endswith(Postfix = "\"") && 1696 (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") || 1697 Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") || 1698 Text.startswith(Prefix = "u8\"") || 1699 Text.startswith(Prefix = "L\""))) || 1700 (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) { 1701 // We need this to address the case where there is an unbreakable tail 1702 // only if certain other formatting decisions have been taken. The 1703 // UnbreakableTailLength of Current is an overapproximation is that case 1704 // and we need to be correct here. 1705 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State)) 1706 ? 0 1707 : Current.UnbreakableTailLength; 1708 return llvm::make_unique<BreakableStringLiteral>( 1709 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength, 1710 State.Line->InPPDirective, Encoding, Style); 1711 } 1712 } else if (Current.is(TT_BlockComment)) { 1713 if (!Style.ReflowComments || 1714 // If a comment token switches formatting, like 1715 // /* clang-format on */, we don't want to break it further, 1716 // but we may still want to adjust its indentation. 1717 switchesFormatting(Current)) { 1718 return nullptr; 1719 } 1720 return llvm::make_unique<BreakableBlockComment>( 1721 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1722 State.Line->InPPDirective, Encoding, Style); 1723 } else if (Current.is(TT_LineComment) && 1724 (Current.Previous == nullptr || 1725 Current.Previous->isNot(TT_ImplicitStringLiteral))) { 1726 if (!Style.ReflowComments || 1727 CommentPragmasRegex.match(Current.TokenText.substr(2)) || 1728 switchesFormatting(Current)) 1729 return nullptr; 1730 return llvm::make_unique<BreakableLineCommentSection>( 1731 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1732 /*InPPDirective=*/false, Encoding, Style); 1733 } 1734 return nullptr; 1735 } 1736 1737 std::pair<unsigned, bool> 1738 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current, 1739 LineState &State, bool AllowBreak, 1740 bool DryRun, bool Strict) { 1741 std::unique_ptr<const BreakableToken> Token = 1742 createBreakableToken(Current, State, AllowBreak); 1743 if (!Token) 1744 return {0, false}; 1745 assert(Token->getLineCount() > 0); 1746 unsigned ColumnLimit = getColumnLimit(State); 1747 if (Current.is(TT_LineComment)) { 1748 // We don't insert backslashes when breaking line comments. 1749 ColumnLimit = Style.ColumnLimit; 1750 } 1751 if (Current.UnbreakableTailLength >= ColumnLimit) 1752 return {0, false}; 1753 // ColumnWidth was already accounted into State.Column before calling 1754 // breakProtrudingToken. 1755 unsigned StartColumn = State.Column - Current.ColumnWidth; 1756 unsigned NewBreakPenalty = Current.isStringLiteral() 1757 ? Style.PenaltyBreakString 1758 : Style.PenaltyBreakComment; 1759 // Stores whether we intentionally decide to let a line exceed the column 1760 // limit. 1761 bool Exceeded = false; 1762 // Stores whether we introduce a break anywhere in the token. 1763 bool BreakInserted = Token->introducesBreakBeforeToken(); 1764 // Store whether we inserted a new line break at the end of the previous 1765 // logical line. 1766 bool NewBreakBefore = false; 1767 // We use a conservative reflowing strategy. Reflow starts after a line is 1768 // broken or the corresponding whitespace compressed. Reflow ends as soon as a 1769 // line that doesn't get reflown with the previous line is reached. 1770 bool Reflow = false; 1771 // Keep track of where we are in the token: 1772 // Where we are in the content of the current logical line. 1773 unsigned TailOffset = 0; 1774 // The column number we're currently at. 1775 unsigned ContentStartColumn = 1776 Token->getContentStartColumn(0, /*Break=*/false); 1777 // The number of columns left in the current logical line after TailOffset. 1778 unsigned RemainingTokenColumns = 1779 Token->getRemainingLength(0, TailOffset, ContentStartColumn); 1780 // Adapt the start of the token, for example indent. 1781 if (!DryRun) 1782 Token->adaptStartOfLine(0, Whitespaces); 1783 1784 unsigned Penalty = 0; 1785 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column " 1786 << StartColumn << ".\n"); 1787 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount(); 1788 LineIndex != EndIndex; ++LineIndex) { 1789 LLVM_DEBUG(llvm::dbgs() 1790 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n"); 1791 NewBreakBefore = false; 1792 // If we did reflow the previous line, we'll try reflowing again. Otherwise 1793 // we'll start reflowing if the current line is broken or whitespace is 1794 // compressed. 1795 bool TryReflow = Reflow; 1796 // Break the current token until we can fit the rest of the line. 1797 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 1798 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: " 1799 << (ContentStartColumn + RemainingTokenColumns) 1800 << ", space: " << ColumnLimit 1801 << ", reflown prefix: " << ContentStartColumn 1802 << ", offset in line: " << TailOffset << "\n"); 1803 // If the current token doesn't fit, find the latest possible split in the 1804 // current line so that breaking at it will be under the column limit. 1805 // FIXME: Use the earliest possible split while reflowing to correctly 1806 // compress whitespace within a line. 1807 BreakableToken::Split Split = 1808 Token->getSplit(LineIndex, TailOffset, ColumnLimit, 1809 ContentStartColumn, CommentPragmasRegex); 1810 if (Split.first == StringRef::npos) { 1811 // No break opportunity - update the penalty and continue with the next 1812 // logical line. 1813 if (LineIndex < EndIndex - 1) 1814 // The last line's penalty is handled in addNextStateToQueue(). 1815 Penalty += Style.PenaltyExcessCharacter * 1816 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 1817 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n"); 1818 break; 1819 } 1820 assert(Split.first != 0); 1821 1822 if (Token->supportsReflow()) { 1823 // Check whether the next natural split point after the current one can 1824 // still fit the line, either because we can compress away whitespace, 1825 // or because the penalty the excess characters introduce is lower than 1826 // the break penalty. 1827 // We only do this for tokens that support reflowing, and thus allow us 1828 // to change the whitespace arbitrarily (e.g. comments). 1829 // Other tokens, like string literals, can be broken on arbitrary 1830 // positions. 1831 1832 // First, compute the columns from TailOffset to the next possible split 1833 // position. 1834 // For example: 1835 // ColumnLimit: | 1836 // // Some text that breaks 1837 // ^ tail offset 1838 // ^-- split 1839 // ^-------- to split columns 1840 // ^--- next split 1841 // ^--------------- to next split columns 1842 unsigned ToSplitColumns = Token->getRangeLength( 1843 LineIndex, TailOffset, Split.first, ContentStartColumn); 1844 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n"); 1845 1846 BreakableToken::Split NextSplit = Token->getSplit( 1847 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit, 1848 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex); 1849 // Compute the columns necessary to fit the next non-breakable sequence 1850 // into the current line. 1851 unsigned ToNextSplitColumns = 0; 1852 if (NextSplit.first == StringRef::npos) { 1853 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset, 1854 ContentStartColumn); 1855 } else { 1856 ToNextSplitColumns = Token->getRangeLength( 1857 LineIndex, TailOffset, 1858 Split.first + Split.second + NextSplit.first, ContentStartColumn); 1859 } 1860 // Compress the whitespace between the break and the start of the next 1861 // unbreakable sequence. 1862 ToNextSplitColumns = 1863 Token->getLengthAfterCompression(ToNextSplitColumns, Split); 1864 LLVM_DEBUG(llvm::dbgs() 1865 << " ContentStartColumn: " << ContentStartColumn << "\n"); 1866 LLVM_DEBUG(llvm::dbgs() 1867 << " ToNextSplit: " << ToNextSplitColumns << "\n"); 1868 // If the whitespace compression makes us fit, continue on the current 1869 // line. 1870 bool ContinueOnLine = 1871 ContentStartColumn + ToNextSplitColumns <= ColumnLimit; 1872 unsigned ExcessCharactersPenalty = 0; 1873 if (!ContinueOnLine && !Strict) { 1874 // Similarly, if the excess characters' penalty is lower than the 1875 // penalty of introducing a new break, continue on the current line. 1876 ExcessCharactersPenalty = 1877 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) * 1878 Style.PenaltyExcessCharacter; 1879 LLVM_DEBUG(llvm::dbgs() 1880 << " Penalty excess: " << ExcessCharactersPenalty 1881 << "\n break : " << NewBreakPenalty << "\n"); 1882 if (ExcessCharactersPenalty < NewBreakPenalty) { 1883 Exceeded = true; 1884 ContinueOnLine = true; 1885 } 1886 } 1887 if (ContinueOnLine) { 1888 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n"); 1889 // The current line fits after compressing the whitespace - reflow 1890 // the next line into it if possible. 1891 TryReflow = true; 1892 if (!DryRun) 1893 Token->compressWhitespace(LineIndex, TailOffset, Split, 1894 Whitespaces); 1895 // When we continue on the same line, leave one space between content. 1896 ContentStartColumn += ToSplitColumns + 1; 1897 Penalty += ExcessCharactersPenalty; 1898 TailOffset += Split.first + Split.second; 1899 RemainingTokenColumns = Token->getRemainingLength( 1900 LineIndex, TailOffset, ContentStartColumn); 1901 continue; 1902 } 1903 } 1904 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n"); 1905 ContentStartColumn = 1906 Token->getContentStartColumn(LineIndex, /*Break=*/true); 1907 unsigned NewRemainingTokenColumns = Token->getRemainingLength( 1908 LineIndex, TailOffset + Split.first + Split.second, 1909 ContentStartColumn); 1910 1911 // When breaking before a tab character, it may be moved by a few columns, 1912 // but will still be expanded to the next tab stop, so we don't save any 1913 // columns. 1914 if (NewRemainingTokenColumns == RemainingTokenColumns) { 1915 // FIXME: Do we need to adjust the penalty? 1916 break; 1917 } 1918 assert(NewRemainingTokenColumns < RemainingTokenColumns); 1919 1920 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first 1921 << ", " << Split.second << "\n"); 1922 if (!DryRun) 1923 Token->insertBreak(LineIndex, TailOffset, Split, Whitespaces); 1924 1925 Penalty += NewBreakPenalty; 1926 TailOffset += Split.first + Split.second; 1927 RemainingTokenColumns = NewRemainingTokenColumns; 1928 BreakInserted = true; 1929 NewBreakBefore = true; 1930 } 1931 // In case there's another line, prepare the state for the start of the next 1932 // line. 1933 if (LineIndex + 1 != EndIndex) { 1934 unsigned NextLineIndex = LineIndex + 1; 1935 if (NewBreakBefore) 1936 // After breaking a line, try to reflow the next line into the current 1937 // one once RemainingTokenColumns fits. 1938 TryReflow = true; 1939 if (TryReflow) { 1940 // We decided that we want to try reflowing the next line into the 1941 // current one. 1942 // We will now adjust the state as if the reflow is successful (in 1943 // preparation for the next line), and see whether that works. If we 1944 // decide that we cannot reflow, we will later reset the state to the 1945 // start of the next line. 1946 Reflow = false; 1947 // As we did not continue breaking the line, RemainingTokenColumns is 1948 // known to fit after ContentStartColumn. Adapt ContentStartColumn to 1949 // the position at which we want to format the next line if we do 1950 // actually reflow. 1951 // When we reflow, we need to add a space between the end of the current 1952 // line and the next line's start column. 1953 ContentStartColumn += RemainingTokenColumns + 1; 1954 // Get the split that we need to reflow next logical line into the end 1955 // of the current one; the split will include any leading whitespace of 1956 // the next logical line. 1957 BreakableToken::Split SplitBeforeNext = 1958 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex); 1959 LLVM_DEBUG(llvm::dbgs() 1960 << " Size of reflown text: " << ContentStartColumn 1961 << "\n Potential reflow split: "); 1962 if (SplitBeforeNext.first != StringRef::npos) { 1963 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", " 1964 << SplitBeforeNext.second << "\n"); 1965 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second; 1966 // If the rest of the next line fits into the current line below the 1967 // column limit, we can safely reflow. 1968 RemainingTokenColumns = Token->getRemainingLength( 1969 NextLineIndex, TailOffset, ContentStartColumn); 1970 Reflow = true; 1971 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 1972 LLVM_DEBUG(llvm::dbgs() 1973 << " Over limit after reflow, need: " 1974 << (ContentStartColumn + RemainingTokenColumns) 1975 << ", space: " << ColumnLimit 1976 << ", reflown prefix: " << ContentStartColumn 1977 << ", offset in line: " << TailOffset << "\n"); 1978 // If the whole next line does not fit, try to find a point in 1979 // the next line at which we can break so that attaching the part 1980 // of the next line to that break point onto the current line is 1981 // below the column limit. 1982 BreakableToken::Split Split = 1983 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit, 1984 ContentStartColumn, CommentPragmasRegex); 1985 if (Split.first == StringRef::npos) { 1986 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n"); 1987 Reflow = false; 1988 } else { 1989 // Check whether the first split point gets us below the column 1990 // limit. Note that we will execute this split below as part of 1991 // the normal token breaking and reflow logic within the line. 1992 unsigned ToSplitColumns = Token->getRangeLength( 1993 NextLineIndex, TailOffset, Split.first, ContentStartColumn); 1994 if (ContentStartColumn + ToSplitColumns > ColumnLimit) { 1995 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: " 1996 << (ContentStartColumn + ToSplitColumns) 1997 << ", space: " << ColumnLimit); 1998 unsigned ExcessCharactersPenalty = 1999 (ContentStartColumn + ToSplitColumns - ColumnLimit) * 2000 Style.PenaltyExcessCharacter; 2001 if (NewBreakPenalty < ExcessCharactersPenalty) { 2002 Reflow = false; 2003 } 2004 } 2005 } 2006 } 2007 } else { 2008 LLVM_DEBUG(llvm::dbgs() << "not found.\n"); 2009 } 2010 } 2011 if (!Reflow) { 2012 // If we didn't reflow into the next line, the only space to consider is 2013 // the next logical line. Reset our state to match the start of the next 2014 // line. 2015 TailOffset = 0; 2016 ContentStartColumn = 2017 Token->getContentStartColumn(NextLineIndex, /*Break=*/false); 2018 RemainingTokenColumns = Token->getRemainingLength( 2019 NextLineIndex, TailOffset, ContentStartColumn); 2020 // Adapt the start of the token, for example indent. 2021 if (!DryRun) 2022 Token->adaptStartOfLine(NextLineIndex, Whitespaces); 2023 } else { 2024 // If we found a reflow split and have added a new break before the next 2025 // line, we are going to remove the line break at the start of the next 2026 // logical line. For example, here we'll add a new line break after 2027 // 'text', and subsequently delete the line break between 'that' and 2028 // 'reflows'. 2029 // // some text that 2030 // // reflows 2031 // -> 2032 // // some text 2033 // // that reflows 2034 // When adding the line break, we also added the penalty for it, so we 2035 // need to subtract that penalty again when we remove the line break due 2036 // to reflowing. 2037 if (NewBreakBefore) { 2038 assert(Penalty >= NewBreakPenalty); 2039 Penalty -= NewBreakPenalty; 2040 } 2041 if (!DryRun) 2042 Token->reflow(NextLineIndex, Whitespaces); 2043 } 2044 } 2045 } 2046 2047 BreakableToken::Split SplitAfterLastLine = 2048 Token->getSplitAfterLastLine(TailOffset); 2049 if (SplitAfterLastLine.first != StringRef::npos) { 2050 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n"); 2051 if (!DryRun) 2052 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine, 2053 Whitespaces); 2054 ContentStartColumn = 2055 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true); 2056 RemainingTokenColumns = Token->getRemainingLength( 2057 Token->getLineCount() - 1, 2058 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second, 2059 ContentStartColumn); 2060 } 2061 2062 State.Column = ContentStartColumn + RemainingTokenColumns - 2063 Current.UnbreakableTailLength; 2064 2065 if (BreakInserted) { 2066 // If we break the token inside a parameter list, we need to break before 2067 // the next parameter on all levels, so that the next parameter is clearly 2068 // visible. Line comments already introduce a break. 2069 if (Current.isNot(TT_LineComment)) { 2070 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 2071 State.Stack[i].BreakBeforeParameter = true; 2072 } 2073 2074 if (Current.is(TT_BlockComment)) 2075 State.NoContinuation = true; 2076 2077 State.Stack.back().LastSpace = StartColumn; 2078 } 2079 2080 Token->updateNextToken(State); 2081 2082 return {Penalty, Exceeded}; 2083 } 2084 2085 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const { 2086 // In preprocessor directives reserve two chars for trailing " \" 2087 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0); 2088 } 2089 2090 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) { 2091 const FormatToken &Current = *State.NextToken; 2092 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral)) 2093 return false; 2094 // We never consider raw string literals "multiline" for the purpose of 2095 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased 2096 // (see TokenAnnotator::mustBreakBefore(). 2097 if (Current.TokenText.startswith("R\"")) 2098 return false; 2099 if (Current.IsMultiline) 2100 return true; 2101 if (Current.getNextNonComment() && 2102 Current.getNextNonComment()->isStringLiteral()) 2103 return true; // Implicit concatenation. 2104 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals && 2105 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength > 2106 Style.ColumnLimit) 2107 return true; // String will be split. 2108 return false; 2109 } 2110 2111 } // namespace format 2112 } // namespace clang 2113