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