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 if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr)) 325 return false; 326 327 return !State.Stack.back().NoLineBreak; 328 } 329 330 bool ContinuationIndenter::mustBreak(const LineState &State) { 331 const FormatToken &Current = *State.NextToken; 332 const FormatToken &Previous = *Current.Previous; 333 if (Current.MustBreakBefore || Current.is(TT_InlineASMColon)) 334 return true; 335 if (State.Stack.back().BreakBeforeClosingBrace && 336 Current.closesBlockOrBlockTypeList(Style)) 337 return true; 338 if (Previous.is(tok::semi) && State.LineContainsContinuedForLoopSection) 339 return true; 340 if (Style.Language == FormatStyle::LK_ObjC && 341 Current.ObjCSelectorNameParts > 1 && 342 Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) { 343 return true; 344 } 345 if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) || 346 (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) && 347 Style.isCpp() && 348 // FIXME: This is a temporary workaround for the case where clang-format 349 // sets BreakBeforeParameter to avoid bin packing and this creates a 350 // completely unnecessary line break after a template type that isn't 351 // line-wrapped. 352 (Previous.NestingLevel == 1 || Style.BinPackParameters)) || 353 (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) && 354 Previous.isNot(tok::question)) || 355 (!Style.BreakBeforeTernaryOperators && 356 Previous.is(TT_ConditionalExpr))) && 357 State.Stack.back().BreakBeforeParameter && !Current.isTrailingComment() && 358 !Current.isOneOf(tok::r_paren, tok::r_brace)) 359 return true; 360 if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) || 361 (Previous.is(TT_ArrayInitializerLSquare) && 362 Previous.ParameterCount > 1) || 363 opensProtoMessageField(Previous, Style)) && 364 Style.ColumnLimit > 0 && 365 getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 > 366 getColumnLimit(State)) 367 return true; 368 369 const FormatToken &BreakConstructorInitializersToken = 370 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon 371 ? Previous 372 : Current; 373 if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) && 374 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength > 375 getColumnLimit(State) || 376 State.Stack.back().BreakBeforeParameter) && 377 (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All || 378 Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon || 379 Style.ColumnLimit != 0)) 380 return true; 381 382 if (Current.is(TT_ObjCMethodExpr) && !Previous.is(TT_SelectorName) && 383 State.Line->startsWith(TT_ObjCMethodSpecifier)) 384 return true; 385 if (Current.is(TT_SelectorName) && !Previous.is(tok::at) && 386 State.Stack.back().ObjCSelectorNameFound && 387 State.Stack.back().BreakBeforeParameter) 388 return true; 389 390 unsigned NewLineColumn = getNewLineColumn(State); 391 if (Current.isMemberAccess() && Style.ColumnLimit != 0 && 392 State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit && 393 (State.Column > NewLineColumn || 394 Current.NestingLevel < State.StartOfLineLevel)) 395 return true; 396 397 if (startsSegmentOfBuilderTypeCall(Current) && 398 (State.Stack.back().CallContinuation != 0 || 399 State.Stack.back().BreakBeforeParameter) && 400 // JavaScript is treated different here as there is a frequent pattern: 401 // SomeFunction(function() { 402 // ... 403 // }.bind(...)); 404 // FIXME: We should find a more generic solution to this problem. 405 !(State.Column <= NewLineColumn && 406 Style.Language == FormatStyle::LK_JavaScript)) 407 return true; 408 409 // If the template declaration spans multiple lines, force wrap before the 410 // function/class declaration 411 if (Previous.ClosesTemplateDeclaration && 412 State.Stack.back().BreakBeforeParameter && Current.CanBreakBefore) 413 return true; 414 415 if (State.Column <= NewLineColumn) 416 return false; 417 418 if (Style.AlwaysBreakBeforeMultilineStrings && 419 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth || 420 Previous.is(tok::comma) || Current.NestingLevel < 2) && 421 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at) && 422 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) && 423 nextIsMultilineString(State)) 424 return true; 425 426 // Using CanBreakBefore here and below takes care of the decision whether the 427 // current style uses wrapping before or after operators for the given 428 // operator. 429 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) { 430 // If we need to break somewhere inside the LHS of a binary expression, we 431 // should also break after the operator. Otherwise, the formatting would 432 // hide the operator precedence, e.g. in: 433 // if (aaaaaaaaaaaaaa == 434 // bbbbbbbbbbbbbb && c) {.. 435 // For comparisons, we only apply this rule, if the LHS is a binary 436 // expression itself as otherwise, the line breaks seem superfluous. 437 // We need special cases for ">>" which we have split into two ">" while 438 // lexing in order to make template parsing easier. 439 bool IsComparison = (Previous.getPrecedence() == prec::Relational || 440 Previous.getPrecedence() == prec::Equality || 441 Previous.getPrecedence() == prec::Spaceship) && 442 Previous.Previous && 443 Previous.Previous->isNot(TT_BinaryOperator); // For >>. 444 bool LHSIsBinaryExpr = 445 Previous.Previous && Previous.Previous->EndsBinaryExpression; 446 if ((!IsComparison || LHSIsBinaryExpr) && !Current.isTrailingComment() && 447 Previous.getPrecedence() != prec::Assignment && 448 State.Stack.back().BreakBeforeParameter) 449 return true; 450 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore && 451 State.Stack.back().BreakBeforeParameter) { 452 return true; 453 } 454 455 // Same as above, but for the first "<<" operator. 456 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) && 457 State.Stack.back().BreakBeforeParameter && 458 State.Stack.back().FirstLessLess == 0) 459 return true; 460 461 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) { 462 // Always break after "template <...>" and leading annotations. This is only 463 // for cases where the entire line does not fit on a single line as a 464 // different LineFormatter would be used otherwise. 465 if (Previous.ClosesTemplateDeclaration) 466 return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No; 467 if (Previous.is(TT_FunctionAnnotationRParen)) 468 return true; 469 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) && 470 Current.isNot(TT_LeadingJavaAnnotation)) 471 return true; 472 } 473 474 // If the return type spans multiple lines, wrap before the function name. 475 if ((Current.is(TT_FunctionDeclarationName) || 476 (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) && 477 !Previous.is(tok::kw_template) && State.Stack.back().BreakBeforeParameter) 478 return true; 479 480 // The following could be precomputed as they do not depend on the state. 481 // However, as they should take effect only if the UnwrappedLine does not fit 482 // into the ColumnLimit, they are checked here in the ContinuationIndenter. 483 if (Style.ColumnLimit != 0 && Previous.BlockKind == BK_Block && 484 Previous.is(tok::l_brace) && !Current.isOneOf(tok::r_brace, tok::comment)) 485 return true; 486 487 if (Current.is(tok::lessless) && 488 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") || 489 (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") || 490 Previous.TokenText == "\'\\n\'")))) 491 return true; 492 493 if (Previous.is(TT_BlockComment) && Previous.IsMultiline) 494 return true; 495 496 if (State.NoContinuation) 497 return true; 498 499 return false; 500 } 501 502 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline, 503 bool DryRun, 504 unsigned ExtraSpaces) { 505 const FormatToken &Current = *State.NextToken; 506 507 assert(!State.Stack.empty()); 508 State.NoContinuation = false; 509 510 if ((Current.is(TT_ImplicitStringLiteral) && 511 (Current.Previous->Tok.getIdentifierInfo() == nullptr || 512 Current.Previous->Tok.getIdentifierInfo()->getPPKeywordID() == 513 tok::pp_not_keyword))) { 514 unsigned EndColumn = 515 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd()); 516 if (Current.LastNewlineOffset != 0) { 517 // If there is a newline within this token, the final column will solely 518 // determined by the current end column. 519 State.Column = EndColumn; 520 } else { 521 unsigned StartColumn = 522 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin()); 523 assert(EndColumn >= StartColumn); 524 State.Column += EndColumn - StartColumn; 525 } 526 moveStateToNextToken(State, DryRun, /*Newline=*/false); 527 return 0; 528 } 529 530 unsigned Penalty = 0; 531 if (Newline) 532 Penalty = addTokenOnNewLine(State, DryRun); 533 else 534 addTokenOnCurrentLine(State, DryRun, ExtraSpaces); 535 536 return moveStateToNextToken(State, DryRun, Newline) + Penalty; 537 } 538 539 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun, 540 unsigned ExtraSpaces) { 541 FormatToken &Current = *State.NextToken; 542 const FormatToken &Previous = *State.NextToken->Previous; 543 if (Current.is(tok::equal) && 544 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) && 545 State.Stack.back().VariablePos == 0) { 546 State.Stack.back().VariablePos = State.Column; 547 // Move over * and & if they are bound to the variable name. 548 const FormatToken *Tok = &Previous; 549 while (Tok && State.Stack.back().VariablePos >= Tok->ColumnWidth) { 550 State.Stack.back().VariablePos -= Tok->ColumnWidth; 551 if (Tok->SpacesRequiredBefore != 0) 552 break; 553 Tok = Tok->Previous; 554 } 555 if (Previous.PartOfMultiVariableDeclStmt) 556 State.Stack.back().LastSpace = State.Stack.back().VariablePos; 557 } 558 559 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces; 560 561 // Indent preprocessor directives after the hash if required. 562 int PPColumnCorrection = 0; 563 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 564 Previous.is(tok::hash) && State.FirstIndent > 0 && 565 (State.Line->Type == LT_PreprocessorDirective || 566 State.Line->Type == LT_ImportStatement)) { 567 Spaces += State.FirstIndent; 568 569 // For preprocessor indent with tabs, State.Column will be 1 because of the 570 // hash. This causes second-level indents onward to have an extra space 571 // after the tabs. We avoid this misalignment by subtracting 1 from the 572 // column value passed to replaceWhitespace(). 573 if (Style.UseTab != FormatStyle::UT_Never) 574 PPColumnCorrection = -1; 575 } 576 577 if (!DryRun) 578 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces, 579 State.Column + Spaces + PPColumnCorrection); 580 581 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance 582 // declaration unless there is multiple inheritance. 583 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 584 Current.is(TT_InheritanceColon)) 585 State.Stack.back().NoLineBreak = true; 586 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon && 587 Previous.is(TT_InheritanceColon)) 588 State.Stack.back().NoLineBreak = true; 589 590 if (Current.is(TT_SelectorName) && 591 !State.Stack.back().ObjCSelectorNameFound) { 592 unsigned MinIndent = 593 std::max(State.FirstIndent + Style.ContinuationIndentWidth, 594 State.Stack.back().Indent); 595 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth; 596 if (Current.LongestObjCSelectorName == 0) 597 State.Stack.back().AlignColons = false; 598 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos) 599 State.Stack.back().ColonPos = MinIndent + Current.LongestObjCSelectorName; 600 else 601 State.Stack.back().ColonPos = FirstColonPos; 602 } 603 604 // In "AlwaysBreak" mode, enforce wrapping directly after the parenthesis by 605 // disallowing any further line breaks if there is no line break after the 606 // opening parenthesis. Don't break if it doesn't conserve columns. 607 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak && 608 Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) && 609 State.Column > getNewLineColumn(State) && 610 (!Previous.Previous || !Previous.Previous->isOneOf( 611 tok::kw_for, tok::kw_while, tok::kw_switch)) && 612 // Don't do this for simple (no expressions) one-argument function calls 613 // as that feels like needlessly wasting whitespace, e.g.: 614 // 615 // caaaaaaaaaaaall( 616 // caaaaaaaaaaaall( 617 // caaaaaaaaaaaall( 618 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa)))); 619 Current.FakeLParens.size() > 0 && 620 Current.FakeLParens.back() > prec::Unknown) 621 State.Stack.back().NoLineBreak = true; 622 if (Previous.is(TT_TemplateString) && Previous.opensScope()) 623 State.Stack.back().NoLineBreak = true; 624 625 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign && 626 Previous.opensScope() && Previous.isNot(TT_ObjCMethodExpr) && 627 (Current.isNot(TT_LineComment) || Previous.BlockKind == BK_BracedInit)) 628 State.Stack.back().Indent = State.Column + Spaces; 629 if (State.Stack.back().AvoidBinPacking && startsNextParameter(Current, Style)) 630 State.Stack.back().NoLineBreak = true; 631 if (startsSegmentOfBuilderTypeCall(Current) && 632 State.Column > getNewLineColumn(State)) 633 State.Stack.back().ContainsUnwrappedBuilder = true; 634 635 if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java) 636 State.Stack.back().NoLineBreak = true; 637 if (Current.isMemberAccess() && Previous.is(tok::r_paren) && 638 (Previous.MatchingParen && 639 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) 640 // If there is a function call with long parameters, break before trailing 641 // calls. This prevents things like: 642 // EXPECT_CALL(SomeLongParameter).Times( 643 // 2); 644 // We don't want to do this for short parameters as they can just be 645 // indexes. 646 State.Stack.back().NoLineBreak = true; 647 648 // Don't allow the RHS of an operator to be split over multiple lines unless 649 // there is a line-break right after the operator. 650 // Exclude relational operators, as there, it is always more desirable to 651 // have the LHS 'left' of the RHS. 652 const FormatToken *P = Current.getPreviousNonComment(); 653 if (!Current.is(tok::comment) && P && 654 (P->isOneOf(TT_BinaryOperator, tok::comma) || 655 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) && 656 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) && 657 P->getPrecedence() != prec::Assignment && 658 P->getPrecedence() != prec::Relational && 659 P->getPrecedence() != prec::Spaceship) { 660 bool BreakBeforeOperator = 661 P->MustBreakBefore || P->is(tok::lessless) || 662 (P->is(TT_BinaryOperator) && 663 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) || 664 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators); 665 // Don't do this if there are only two operands. In these cases, there is 666 // always a nice vertical separation between them and the extra line break 667 // does not help. 668 bool HasTwoOperands = 669 P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr); 670 if ((!BreakBeforeOperator && !(HasTwoOperands && Style.AlignOperands)) || 671 (!State.Stack.back().LastOperatorWrapped && BreakBeforeOperator)) 672 State.Stack.back().NoLineBreakInOperand = true; 673 } 674 675 State.Column += Spaces; 676 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) && 677 Previous.Previous && 678 (Previous.Previous->isOneOf(tok::kw_if, tok::kw_for) || 679 Previous.Previous->endsSequence(tok::kw_constexpr, tok::kw_if))) { 680 // Treat the condition inside an if as if it was a second function 681 // parameter, i.e. let nested calls have a continuation indent. 682 State.Stack.back().LastSpace = State.Column; 683 State.Stack.back().NestedBlockIndent = State.Column; 684 } else if (!Current.isOneOf(tok::comment, tok::caret) && 685 ((Previous.is(tok::comma) && 686 !Previous.is(TT_OverloadedOperator)) || 687 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) { 688 State.Stack.back().LastSpace = State.Column; 689 } else if (Previous.is(TT_CtorInitializerColon) && 690 Style.BreakConstructorInitializers == 691 FormatStyle::BCIS_AfterColon) { 692 State.Stack.back().Indent = State.Column; 693 State.Stack.back().LastSpace = State.Column; 694 } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 695 TT_CtorInitializerColon)) && 696 ((Previous.getPrecedence() != prec::Assignment && 697 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 || 698 Previous.NextOperator)) || 699 Current.StartsBinaryExpression)) { 700 // Indent relative to the RHS of the expression unless this is a simple 701 // assignment without binary expression on the RHS. Also indent relative to 702 // unary operators and the colons of constructor initializers. 703 State.Stack.back().LastSpace = State.Column; 704 } else if (Previous.is(TT_InheritanceColon)) { 705 State.Stack.back().Indent = State.Column; 706 State.Stack.back().LastSpace = State.Column; 707 } else if (Previous.opensScope()) { 708 // If a function has a trailing call, indent all parameters from the 709 // opening parenthesis. This avoids confusing indents like: 710 // OuterFunction(InnerFunctionCall( // break 711 // ParameterToInnerFunction)) // break 712 // .SecondInnerFunctionCall(); 713 bool HasTrailingCall = false; 714 if (Previous.MatchingParen) { 715 const FormatToken *Next = Previous.MatchingParen->getNextNonComment(); 716 HasTrailingCall = Next && Next->isMemberAccess(); 717 } 718 if (HasTrailingCall && State.Stack.size() > 1 && 719 State.Stack[State.Stack.size() - 2].CallContinuation == 0) 720 State.Stack.back().LastSpace = State.Column; 721 } 722 } 723 724 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State, 725 bool DryRun) { 726 FormatToken &Current = *State.NextToken; 727 const FormatToken &Previous = *State.NextToken->Previous; 728 729 // Extra penalty that needs to be added because of the way certain line 730 // breaks are chosen. 731 unsigned Penalty = 0; 732 733 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 734 const FormatToken *NextNonComment = Previous.getNextNonComment(); 735 if (!NextNonComment) 736 NextNonComment = &Current; 737 // The first line break on any NestingLevel causes an extra penalty in order 738 // prefer similar line breaks. 739 if (!State.Stack.back().ContainsLineBreak) 740 Penalty += 15; 741 State.Stack.back().ContainsLineBreak = true; 742 743 Penalty += State.NextToken->SplitPenalty; 744 745 // Breaking before the first "<<" is generally not desirable if the LHS is 746 // short. Also always add the penalty if the LHS is split over multiple lines 747 // to avoid unnecessary line breaks that just work around this penalty. 748 if (NextNonComment->is(tok::lessless) && 749 State.Stack.back().FirstLessLess == 0 && 750 (State.Column <= Style.ColumnLimit / 3 || 751 State.Stack.back().BreakBeforeParameter)) 752 Penalty += Style.PenaltyBreakFirstLessLess; 753 754 State.Column = getNewLineColumn(State); 755 756 // Indent nested blocks relative to this column, unless in a very specific 757 // JavaScript special case where: 758 // 759 // var loooooong_name = 760 // function() { 761 // // code 762 // } 763 // 764 // is common and should be formatted like a free-standing function. The same 765 // goes for wrapping before the lambda return type arrow. 766 if (!Current.is(TT_LambdaArrow) && 767 (Style.Language != FormatStyle::LK_JavaScript || 768 Current.NestingLevel != 0 || !PreviousNonComment || 769 !PreviousNonComment->is(tok::equal) || 770 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) 771 State.Stack.back().NestedBlockIndent = State.Column; 772 773 if (NextNonComment->isMemberAccess()) { 774 if (State.Stack.back().CallContinuation == 0) 775 State.Stack.back().CallContinuation = State.Column; 776 } else if (NextNonComment->is(TT_SelectorName)) { 777 if (!State.Stack.back().ObjCSelectorNameFound) { 778 if (NextNonComment->LongestObjCSelectorName == 0) { 779 State.Stack.back().AlignColons = false; 780 } else { 781 State.Stack.back().ColonPos = 782 (shouldIndentWrappedSelectorName(Style, State.Line->Type) 783 ? std::max(State.Stack.back().Indent, 784 State.FirstIndent + Style.ContinuationIndentWidth) 785 : State.Stack.back().Indent) + 786 std::max(NextNonComment->LongestObjCSelectorName, 787 NextNonComment->ColumnWidth); 788 } 789 } else if (State.Stack.back().AlignColons && 790 State.Stack.back().ColonPos <= NextNonComment->ColumnWidth) { 791 State.Stack.back().ColonPos = State.Column + NextNonComment->ColumnWidth; 792 } 793 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 794 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) { 795 // FIXME: This is hacky, find a better way. The problem is that in an ObjC 796 // method expression, the block should be aligned to the line starting it, 797 // e.g.: 798 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason 799 // ^(int *i) { 800 // // ... 801 // }]; 802 // Thus, we set LastSpace of the next higher NestingLevel, to which we move 803 // when we consume all of the "}"'s FakeRParens at the "{". 804 if (State.Stack.size() > 1) 805 State.Stack[State.Stack.size() - 2].LastSpace = 806 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 807 Style.ContinuationIndentWidth; 808 } 809 810 if ((PreviousNonComment && 811 PreviousNonComment->isOneOf(tok::comma, tok::semi) && 812 !State.Stack.back().AvoidBinPacking) || 813 Previous.is(TT_BinaryOperator)) 814 State.Stack.back().BreakBeforeParameter = false; 815 if (PreviousNonComment && 816 PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) && 817 Current.NestingLevel == 0) 818 State.Stack.back().BreakBeforeParameter = false; 819 if (NextNonComment->is(tok::question) || 820 (PreviousNonComment && PreviousNonComment->is(tok::question))) 821 State.Stack.back().BreakBeforeParameter = true; 822 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore) 823 State.Stack.back().BreakBeforeParameter = false; 824 825 if (!DryRun) { 826 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1; 827 if (Current.is(tok::r_brace) && Current.MatchingParen && 828 // Only strip trailing empty lines for l_braces that have children, i.e. 829 // for function expressions (lambdas, arrows, etc). 830 !Current.MatchingParen->Children.empty()) { 831 // lambdas and arrow functions are expressions, thus their r_brace is not 832 // on its own line, and thus not covered by UnwrappedLineFormatter's logic 833 // about removing empty lines on closing blocks. Special case them here. 834 MaxEmptyLinesToKeep = 1; 835 } 836 unsigned Newlines = std::max( 837 1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep)); 838 bool ContinuePPDirective = 839 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement; 840 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column, 841 ContinuePPDirective); 842 } 843 844 if (!Current.isTrailingComment()) 845 State.Stack.back().LastSpace = State.Column; 846 if (Current.is(tok::lessless)) 847 // If we are breaking before a "<<", we always want to indent relative to 848 // RHS. This is necessary only for "<<", as we special-case it and don't 849 // always indent relative to the RHS. 850 State.Stack.back().LastSpace += 3; // 3 -> width of "<< ". 851 852 State.StartOfLineLevel = Current.NestingLevel; 853 State.LowestLevelOnLine = Current.NestingLevel; 854 855 // Any break on this level means that the parent level has been broken 856 // and we need to avoid bin packing there. 857 bool NestedBlockSpecialCase = 858 !Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 && 859 State.Stack[State.Stack.size() - 2].NestedBlockInlined; 860 if (!NestedBlockSpecialCase) 861 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 862 State.Stack[i].BreakBeforeParameter = true; 863 864 if (PreviousNonComment && 865 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) && 866 (PreviousNonComment->isNot(TT_TemplateCloser) || 867 Current.NestingLevel != 0) && 868 !PreviousNonComment->isOneOf( 869 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation, 870 TT_LeadingJavaAnnotation) && 871 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope()) 872 State.Stack.back().BreakBeforeParameter = true; 873 874 // If we break after { or the [ of an array initializer, we should also break 875 // before the corresponding } or ]. 876 if (PreviousNonComment && 877 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 878 opensProtoMessageField(*PreviousNonComment, Style))) 879 State.Stack.back().BreakBeforeClosingBrace = true; 880 881 if (State.Stack.back().AvoidBinPacking) { 882 // If we are breaking after '(', '{', '<', this is not bin packing 883 // unless AllowAllParametersOfDeclarationOnNextLine is false or this is a 884 // dict/object literal. 885 if (!Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) || 886 (!Style.AllowAllParametersOfDeclarationOnNextLine && 887 State.Line->MustBeDeclaration) || 888 Previous.is(TT_DictLiteral)) 889 State.Stack.back().BreakBeforeParameter = true; 890 } 891 892 return Penalty; 893 } 894 895 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) { 896 if (!State.NextToken || !State.NextToken->Previous) 897 return 0; 898 FormatToken &Current = *State.NextToken; 899 const FormatToken &Previous = *Current.Previous; 900 // If we are continuing an expression, we want to use the continuation indent. 901 unsigned ContinuationIndent = 902 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 903 Style.ContinuationIndentWidth; 904 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 905 const FormatToken *NextNonComment = Previous.getNextNonComment(); 906 if (!NextNonComment) 907 NextNonComment = &Current; 908 909 // Java specific bits. 910 if (Style.Language == FormatStyle::LK_Java && 911 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) 912 return std::max(State.Stack.back().LastSpace, 913 State.Stack.back().Indent + Style.ContinuationIndentWidth); 914 915 if (NextNonComment->is(tok::l_brace) && NextNonComment->BlockKind == BK_Block) 916 return Current.NestingLevel == 0 ? State.FirstIndent 917 : State.Stack.back().Indent; 918 if ((Current.isOneOf(tok::r_brace, tok::r_square) || 919 (Current.is(tok::greater) && 920 (Style.Language == FormatStyle::LK_Proto || 921 Style.Language == FormatStyle::LK_TextProto))) && 922 State.Stack.size() > 1) { 923 if (Current.closesBlockOrBlockTypeList(Style)) 924 return State.Stack[State.Stack.size() - 2].NestedBlockIndent; 925 if (Current.MatchingParen && 926 Current.MatchingParen->BlockKind == BK_BracedInit) 927 return State.Stack[State.Stack.size() - 2].LastSpace; 928 return State.FirstIndent; 929 } 930 // Indent a closing parenthesis at the previous level if followed by a semi or 931 // opening brace. This allows indentations such as: 932 // foo( 933 // a, 934 // ); 935 // function foo( 936 // a, 937 // ) { 938 // code(); // 939 // } 940 if (Current.is(tok::r_paren) && State.Stack.size() > 1 && 941 (!Current.Next || Current.Next->isOneOf(tok::semi, tok::l_brace))) 942 return State.Stack[State.Stack.size() - 2].LastSpace; 943 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope()) 944 return State.Stack[State.Stack.size() - 2].LastSpace; 945 if (Current.is(tok::identifier) && Current.Next && 946 (Current.Next->is(TT_DictLiteral) || 947 ((Style.Language == FormatStyle::LK_Proto || 948 Style.Language == FormatStyle::LK_TextProto) && 949 Current.Next->isOneOf(tok::less, tok::l_brace)))) 950 return State.Stack.back().Indent; 951 if (NextNonComment->is(TT_ObjCStringLiteral) && 952 State.StartOfStringLiteral != 0) 953 return State.StartOfStringLiteral - 1; 954 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0) 955 return State.StartOfStringLiteral; 956 if (NextNonComment->is(tok::lessless) && 957 State.Stack.back().FirstLessLess != 0) 958 return State.Stack.back().FirstLessLess; 959 if (NextNonComment->isMemberAccess()) { 960 if (State.Stack.back().CallContinuation == 0) 961 return ContinuationIndent; 962 return State.Stack.back().CallContinuation; 963 } 964 if (State.Stack.back().QuestionColumn != 0 && 965 ((NextNonComment->is(tok::colon) && 966 NextNonComment->is(TT_ConditionalExpr)) || 967 Previous.is(TT_ConditionalExpr))) 968 return State.Stack.back().QuestionColumn; 969 if (Previous.is(tok::comma) && State.Stack.back().VariablePos != 0) 970 return State.Stack.back().VariablePos; 971 if ((PreviousNonComment && 972 (PreviousNonComment->ClosesTemplateDeclaration || 973 PreviousNonComment->isOneOf( 974 TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen, 975 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) || 976 (!Style.IndentWrappedFunctionNames && 977 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) 978 return std::max(State.Stack.back().LastSpace, State.Stack.back().Indent); 979 if (NextNonComment->is(TT_SelectorName)) { 980 if (!State.Stack.back().ObjCSelectorNameFound) { 981 unsigned MinIndent = State.Stack.back().Indent; 982 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) 983 MinIndent = std::max(MinIndent, 984 State.FirstIndent + Style.ContinuationIndentWidth); 985 // If LongestObjCSelectorName is 0, we are indenting the first 986 // part of an ObjC selector (or a selector component which is 987 // not colon-aligned due to block formatting). 988 // 989 // Otherwise, we are indenting a subsequent part of an ObjC 990 // selector which should be colon-aligned to the longest 991 // component of the ObjC selector. 992 // 993 // In either case, we want to respect Style.IndentWrappedFunctionNames. 994 return MinIndent + 995 std::max(NextNonComment->LongestObjCSelectorName, 996 NextNonComment->ColumnWidth) - 997 NextNonComment->ColumnWidth; 998 } 999 if (!State.Stack.back().AlignColons) 1000 return State.Stack.back().Indent; 1001 if (State.Stack.back().ColonPos > NextNonComment->ColumnWidth) 1002 return State.Stack.back().ColonPos - NextNonComment->ColumnWidth; 1003 return State.Stack.back().Indent; 1004 } 1005 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr)) 1006 return State.Stack.back().ColonPos; 1007 if (NextNonComment->is(TT_ArraySubscriptLSquare)) { 1008 if (State.Stack.back().StartOfArraySubscripts != 0) 1009 return State.Stack.back().StartOfArraySubscripts; 1010 return ContinuationIndent; 1011 } 1012 1013 // This ensure that we correctly format ObjC methods calls without inputs, 1014 // i.e. where the last element isn't selector like: [callee method]; 1015 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 && 1016 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) 1017 return State.Stack.back().Indent; 1018 1019 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) || 1020 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) 1021 return ContinuationIndent; 1022 if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 1023 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) 1024 return ContinuationIndent; 1025 if (NextNonComment->is(TT_CtorInitializerComma)) 1026 return State.Stack.back().Indent; 1027 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && 1028 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) 1029 return State.Stack.back().Indent; 1030 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) && 1031 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) 1032 return State.Stack.back().Indent; 1033 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon, 1034 TT_InheritanceComma)) 1035 return State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1036 if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() && 1037 !Current.isOneOf(tok::colon, tok::comment)) 1038 return ContinuationIndent; 1039 if (Current.is(TT_ProtoExtensionLSquare)) 1040 return State.Stack.back().Indent; 1041 if (State.Stack.back().Indent == State.FirstIndent && PreviousNonComment && 1042 PreviousNonComment->isNot(tok::r_brace)) 1043 // Ensure that we fall back to the continuation indent width instead of 1044 // just flushing continuations left. 1045 return State.Stack.back().Indent + Style.ContinuationIndentWidth; 1046 return State.Stack.back().Indent; 1047 } 1048 1049 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State, 1050 bool DryRun, bool Newline) { 1051 assert(State.Stack.size()); 1052 const FormatToken &Current = *State.NextToken; 1053 1054 if (Current.isOneOf(tok::comma, TT_BinaryOperator)) 1055 State.Stack.back().NoLineBreakInOperand = false; 1056 if (Current.is(TT_InheritanceColon)) 1057 State.Stack.back().AvoidBinPacking = true; 1058 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) { 1059 if (State.Stack.back().FirstLessLess == 0) 1060 State.Stack.back().FirstLessLess = State.Column; 1061 else 1062 State.Stack.back().LastOperatorWrapped = Newline; 1063 } 1064 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless)) 1065 State.Stack.back().LastOperatorWrapped = Newline; 1066 if (Current.is(TT_ConditionalExpr) && Current.Previous && 1067 !Current.Previous->is(TT_ConditionalExpr)) 1068 State.Stack.back().LastOperatorWrapped = Newline; 1069 if (Current.is(TT_ArraySubscriptLSquare) && 1070 State.Stack.back().StartOfArraySubscripts == 0) 1071 State.Stack.back().StartOfArraySubscripts = State.Column; 1072 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question)) 1073 State.Stack.back().QuestionColumn = State.Column; 1074 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) { 1075 const FormatToken *Previous = Current.Previous; 1076 while (Previous && Previous->isTrailingComment()) 1077 Previous = Previous->Previous; 1078 if (Previous && Previous->is(tok::question)) 1079 State.Stack.back().QuestionColumn = State.Column; 1080 } 1081 if (!Current.opensScope() && !Current.closesScope() && 1082 !Current.is(TT_PointerOrReference)) 1083 State.LowestLevelOnLine = 1084 std::min(State.LowestLevelOnLine, Current.NestingLevel); 1085 if (Current.isMemberAccess()) 1086 State.Stack.back().StartOfFunctionCall = 1087 !Current.NextOperator ? 0 : State.Column; 1088 if (Current.is(TT_SelectorName)) 1089 State.Stack.back().ObjCSelectorNameFound = true; 1090 if (Current.is(TT_CtorInitializerColon) && 1091 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) { 1092 // Indent 2 from the column, so: 1093 // SomeClass::SomeClass() 1094 // : First(...), ... 1095 // Next(...) 1096 // ^ line up here. 1097 State.Stack.back().Indent = 1098 State.Column + 1099 (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma 1100 ? 0 1101 : 2); 1102 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1103 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 1104 State.Stack.back().AvoidBinPacking = true; 1105 State.Stack.back().BreakBeforeParameter = false; 1106 } 1107 if (Current.is(TT_CtorInitializerColon) && 1108 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { 1109 State.Stack.back().Indent = 1110 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1111 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1112 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 1113 State.Stack.back().AvoidBinPacking = true; 1114 } 1115 if (Current.is(TT_InheritanceColon)) 1116 State.Stack.back().Indent = 1117 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1118 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline) 1119 State.Stack.back().NestedBlockIndent = 1120 State.Column + Current.ColumnWidth + 1; 1121 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow)) 1122 State.Stack.back().LastSpace = State.Column; 1123 1124 // Insert scopes created by fake parenthesis. 1125 const FormatToken *Previous = Current.getPreviousNonComment(); 1126 1127 // Add special behavior to support a format commonly used for JavaScript 1128 // closures: 1129 // SomeFunction(function() { 1130 // foo(); 1131 // bar(); 1132 // }, a, b, c); 1133 if (Current.isNot(tok::comment) && Previous && 1134 Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) && 1135 !Previous->is(TT_DictLiteral) && State.Stack.size() > 1) { 1136 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline) 1137 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 1138 State.Stack[i].NoLineBreak = true; 1139 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false; 1140 } 1141 if (Previous && 1142 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) || 1143 Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr)) && 1144 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 1145 State.Stack.back().NestedBlockInlined = 1146 !Newline && 1147 (Previous->isNot(tok::l_paren) || Previous->ParameterCount > 1); 1148 } 1149 1150 moveStatePastFakeLParens(State, Newline); 1151 moveStatePastScopeCloser(State); 1152 bool AllowBreak = !State.Stack.back().NoLineBreak && 1153 !State.Stack.back().NoLineBreakInOperand; 1154 moveStatePastScopeOpener(State, Newline); 1155 moveStatePastFakeRParens(State); 1156 1157 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0) 1158 State.StartOfStringLiteral = State.Column + 1; 1159 else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) 1160 State.StartOfStringLiteral = State.Column; 1161 else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) && 1162 !Current.isStringLiteral()) 1163 State.StartOfStringLiteral = 0; 1164 1165 State.Column += Current.ColumnWidth; 1166 State.NextToken = State.NextToken->Next; 1167 1168 unsigned Penalty = 1169 handleEndOfLine(Current, State, DryRun, AllowBreak); 1170 1171 if (Current.Role) 1172 Current.Role->formatFromToken(State, this, DryRun); 1173 // If the previous has a special role, let it consume tokens as appropriate. 1174 // It is necessary to start at the previous token for the only implemented 1175 // role (comma separated list). That way, the decision whether or not to break 1176 // after the "{" is already done and both options are tried and evaluated. 1177 // FIXME: This is ugly, find a better way. 1178 if (Previous && Previous->Role) 1179 Penalty += Previous->Role->formatAfterToken(State, this, DryRun); 1180 1181 return Penalty; 1182 } 1183 1184 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State, 1185 bool Newline) { 1186 const FormatToken &Current = *State.NextToken; 1187 const FormatToken *Previous = Current.getPreviousNonComment(); 1188 1189 // Don't add extra indentation for the first fake parenthesis after 1190 // 'return', assignments or opening <({[. The indentation for these cases 1191 // is special cased. 1192 bool SkipFirstExtraIndent = 1193 (Previous && (Previous->opensScope() || 1194 Previous->isOneOf(tok::semi, tok::kw_return) || 1195 (Previous->getPrecedence() == prec::Assignment && 1196 Style.AlignOperands) || 1197 Previous->is(TT_ObjCMethodExpr))); 1198 for (SmallVectorImpl<prec::Level>::const_reverse_iterator 1199 I = Current.FakeLParens.rbegin(), 1200 E = Current.FakeLParens.rend(); 1201 I != E; ++I) { 1202 ParenState NewParenState = State.Stack.back(); 1203 NewParenState.Tok = nullptr; 1204 NewParenState.ContainsLineBreak = false; 1205 NewParenState.LastOperatorWrapped = true; 1206 NewParenState.NoLineBreak = 1207 NewParenState.NoLineBreak || State.Stack.back().NoLineBreakInOperand; 1208 1209 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists. 1210 if (*I > prec::Comma) 1211 NewParenState.AvoidBinPacking = false; 1212 1213 // Indent from 'LastSpace' unless these are fake parentheses encapsulating 1214 // a builder type call after 'return' or, if the alignment after opening 1215 // brackets is disabled. 1216 if (!Current.isTrailingComment() && 1217 (Style.AlignOperands || *I < prec::Assignment) && 1218 (!Previous || Previous->isNot(tok::kw_return) || 1219 (Style.Language != FormatStyle::LK_Java && *I > 0)) && 1220 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign || 1221 *I != prec::Comma || Current.NestingLevel == 0)) 1222 NewParenState.Indent = 1223 std::max(std::max(State.Column, NewParenState.Indent), 1224 State.Stack.back().LastSpace); 1225 1226 // Do not indent relative to the fake parentheses inserted for "." or "->". 1227 // This is a special case to make the following to statements consistent: 1228 // OuterFunction(InnerFunctionCall( // break 1229 // ParameterToInnerFunction)); 1230 // OuterFunction(SomeObject.InnerFunctionCall( // break 1231 // ParameterToInnerFunction)); 1232 if (*I > prec::Unknown) 1233 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column); 1234 if (*I != prec::Conditional && !Current.is(TT_UnaryOperator) && 1235 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 1236 NewParenState.StartOfFunctionCall = State.Column; 1237 1238 // Always indent conditional expressions. Never indent expression where 1239 // the 'operator' is ',', ';' or an assignment (i.e. *I <= 1240 // prec::Assignment) as those have different indentation rules. Indent 1241 // other expression, unless the indentation needs to be skipped. 1242 if (*I == prec::Conditional || 1243 (!SkipFirstExtraIndent && *I > prec::Assignment && 1244 !Current.isTrailingComment())) 1245 NewParenState.Indent += Style.ContinuationIndentWidth; 1246 if ((Previous && !Previous->opensScope()) || *I != prec::Comma) 1247 NewParenState.BreakBeforeParameter = false; 1248 State.Stack.push_back(NewParenState); 1249 SkipFirstExtraIndent = false; 1250 } 1251 } 1252 1253 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) { 1254 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) { 1255 unsigned VariablePos = State.Stack.back().VariablePos; 1256 if (State.Stack.size() == 1) { 1257 // Do not pop the last element. 1258 break; 1259 } 1260 State.Stack.pop_back(); 1261 State.Stack.back().VariablePos = VariablePos; 1262 } 1263 } 1264 1265 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State, 1266 bool Newline) { 1267 const FormatToken &Current = *State.NextToken; 1268 if (!Current.opensScope()) 1269 return; 1270 1271 if (Current.MatchingParen && Current.BlockKind == BK_Block) { 1272 moveStateToNewBlock(State); 1273 return; 1274 } 1275 1276 unsigned NewIndent; 1277 unsigned LastSpace = State.Stack.back().LastSpace; 1278 bool AvoidBinPacking; 1279 bool BreakBeforeParameter = false; 1280 unsigned NestedBlockIndent = std::max(State.Stack.back().StartOfFunctionCall, 1281 State.Stack.back().NestedBlockIndent); 1282 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 1283 opensProtoMessageField(Current, Style)) { 1284 if (Current.opensBlockOrBlockTypeList(Style)) { 1285 NewIndent = Style.IndentWidth + 1286 std::min(State.Column, State.Stack.back().NestedBlockIndent); 1287 } else { 1288 NewIndent = State.Stack.back().LastSpace + Style.ContinuationIndentWidth; 1289 } 1290 const FormatToken *NextNoComment = Current.getNextNonComment(); 1291 bool EndsInComma = Current.MatchingParen && 1292 Current.MatchingParen->Previous && 1293 Current.MatchingParen->Previous->is(tok::comma); 1294 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) || 1295 Style.Language == FormatStyle::LK_Proto || 1296 Style.Language == FormatStyle::LK_TextProto || 1297 !Style.BinPackArguments || 1298 (NextNoComment && 1299 NextNoComment->isOneOf(TT_DesignatedInitializerPeriod, 1300 TT_DesignatedInitializerLSquare)); 1301 BreakBeforeParameter = EndsInComma; 1302 if (Current.ParameterCount > 1) 1303 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1); 1304 } else { 1305 NewIndent = Style.ContinuationIndentWidth + 1306 std::max(State.Stack.back().LastSpace, 1307 State.Stack.back().StartOfFunctionCall); 1308 1309 // Ensure that different different brackets force relative alignment, e.g.: 1310 // void SomeFunction(vector< // break 1311 // int> v); 1312 // FIXME: We likely want to do this for more combinations of brackets. 1313 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) { 1314 NewIndent = std::max(NewIndent, State.Stack.back().Indent); 1315 LastSpace = std::max(LastSpace, State.Stack.back().Indent); 1316 } 1317 1318 bool EndsInComma = 1319 Current.MatchingParen && 1320 Current.MatchingParen->getPreviousNonComment() && 1321 Current.MatchingParen->getPreviousNonComment()->is(tok::comma); 1322 1323 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters 1324 // for backwards compatibility. 1325 bool ObjCBinPackProtocolList = 1326 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto && 1327 Style.BinPackParameters) || 1328 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always; 1329 1330 bool BinPackDeclaration = 1331 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) || 1332 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList); 1333 1334 AvoidBinPacking = 1335 (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) || 1336 (State.Line->MustBeDeclaration && !BinPackDeclaration) || 1337 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) || 1338 (Style.ExperimentalAutoDetectBinPacking && 1339 (Current.PackingKind == PPK_OnePerLine || 1340 (!BinPackInconclusiveFunctions && 1341 Current.PackingKind == PPK_Inconclusive))); 1342 1343 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen) { 1344 if (Style.ColumnLimit) { 1345 // If this '[' opens an ObjC call, determine whether all parameters fit 1346 // into one line and put one per line if they don't. 1347 if (getLengthToMatchingParen(Current, State.Stack) + State.Column > 1348 getColumnLimit(State)) 1349 BreakBeforeParameter = true; 1350 } else { 1351 // For ColumnLimit = 0, we have to figure out whether there is or has to 1352 // be a line break within this call. 1353 for (const FormatToken *Tok = &Current; 1354 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) { 1355 if (Tok->MustBreakBefore || 1356 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) { 1357 BreakBeforeParameter = true; 1358 break; 1359 } 1360 } 1361 } 1362 } 1363 1364 if (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) 1365 BreakBeforeParameter = true; 1366 } 1367 // Generally inherit NoLineBreak from the current scope to nested scope. 1368 // However, don't do this for non-empty nested blocks, dict literals and 1369 // array literals as these follow different indentation rules. 1370 bool NoLineBreak = 1371 Current.Children.empty() && 1372 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) && 1373 (State.Stack.back().NoLineBreak || 1374 State.Stack.back().NoLineBreakInOperand || 1375 (Current.is(TT_TemplateOpener) && 1376 State.Stack.back().ContainsUnwrappedBuilder)); 1377 State.Stack.push_back( 1378 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak)); 1379 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1380 State.Stack.back().BreakBeforeParameter = BreakBeforeParameter; 1381 State.Stack.back().HasMultipleNestedBlocks = Current.BlockParameterCount > 1; 1382 State.Stack.back().IsInsideObjCArrayLiteral = 1383 Current.is(TT_ArrayInitializerLSquare) && Current.Previous && 1384 Current.Previous->is(tok::at); 1385 } 1386 1387 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) { 1388 const FormatToken &Current = *State.NextToken; 1389 if (!Current.closesScope()) 1390 return; 1391 1392 // If we encounter a closing ), ], } or >, we can remove a level from our 1393 // stacks. 1394 if (State.Stack.size() > 1 && 1395 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) || 1396 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) || 1397 State.NextToken->is(TT_TemplateCloser) || 1398 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) 1399 State.Stack.pop_back(); 1400 1401 // Reevaluate whether ObjC message arguments fit into one line. 1402 // If a receiver spans multiple lines, e.g.: 1403 // [[object block:^{ 1404 // return 42; 1405 // }] a:42 b:42]; 1406 // BreakBeforeParameter is calculated based on an incorrect assumption 1407 // (it is checked whether the whole expression fits into one line without 1408 // considering a line break inside a message receiver). 1409 // We check whether arguements fit after receiver scope closer (into the same 1410 // line). 1411 if (State.Stack.back().BreakBeforeParameter && Current.MatchingParen && 1412 Current.MatchingParen->Previous) { 1413 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous; 1414 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) && 1415 CurrentScopeOpener.MatchingParen) { 1416 int NecessarySpaceInLine = 1417 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) + 1418 CurrentScopeOpener.TotalLength - Current.TotalLength - 1; 1419 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <= 1420 Style.ColumnLimit) 1421 State.Stack.back().BreakBeforeParameter = false; 1422 } 1423 } 1424 1425 if (Current.is(tok::r_square)) { 1426 // If this ends the array subscript expr, reset the corresponding value. 1427 const FormatToken *NextNonComment = Current.getNextNonComment(); 1428 if (NextNonComment && NextNonComment->isNot(tok::l_square)) 1429 State.Stack.back().StartOfArraySubscripts = 0; 1430 } 1431 } 1432 1433 void ContinuationIndenter::moveStateToNewBlock(LineState &State) { 1434 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent; 1435 // ObjC block sometimes follow special indentation rules. 1436 unsigned NewIndent = 1437 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace) 1438 ? Style.ObjCBlockIndentWidth 1439 : Style.IndentWidth); 1440 State.Stack.push_back(ParenState(State.NextToken, NewIndent, 1441 State.Stack.back().LastSpace, 1442 /*AvoidBinPacking=*/true, 1443 /*NoLineBreak=*/false)); 1444 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1445 State.Stack.back().BreakBeforeParameter = true; 1446 } 1447 1448 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn, 1449 unsigned TabWidth, 1450 encoding::Encoding Encoding) { 1451 size_t LastNewlinePos = Text.find_last_of("\n"); 1452 if (LastNewlinePos == StringRef::npos) { 1453 return StartColumn + 1454 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding); 1455 } else { 1456 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos), 1457 /*StartColumn=*/0, TabWidth, Encoding); 1458 } 1459 } 1460 1461 unsigned ContinuationIndenter::reformatRawStringLiteral( 1462 const FormatToken &Current, LineState &State, 1463 const FormatStyle &RawStringStyle, bool DryRun) { 1464 unsigned StartColumn = State.Column - Current.ColumnWidth; 1465 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText); 1466 StringRef NewDelimiter = 1467 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language); 1468 if (NewDelimiter.empty() || OldDelimiter.empty()) 1469 NewDelimiter = OldDelimiter; 1470 // The text of a raw string is between the leading 'R"delimiter(' and the 1471 // trailing 'delimiter)"'. 1472 unsigned OldPrefixSize = 3 + OldDelimiter.size(); 1473 unsigned OldSuffixSize = 2 + OldDelimiter.size(); 1474 // We create a virtual text environment which expects a null-terminated 1475 // string, so we cannot use StringRef. 1476 std::string RawText = 1477 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize); 1478 if (NewDelimiter != OldDelimiter) { 1479 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the 1480 // raw string. 1481 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str(); 1482 if (StringRef(RawText).contains(CanonicalDelimiterSuffix)) 1483 NewDelimiter = OldDelimiter; 1484 } 1485 1486 unsigned NewPrefixSize = 3 + NewDelimiter.size(); 1487 unsigned NewSuffixSize = 2 + NewDelimiter.size(); 1488 1489 // The first start column is the column the raw text starts after formatting. 1490 unsigned FirstStartColumn = StartColumn + NewPrefixSize; 1491 1492 // The next start column is the intended indentation a line break inside 1493 // the raw string at level 0. It is determined by the following rules: 1494 // - if the content starts on newline, it is one level more than the current 1495 // indent, and 1496 // - if the content does not start on a newline, it is the first start 1497 // column. 1498 // These rules have the advantage that the formatted content both does not 1499 // violate the rectangle rule and visually flows within the surrounding 1500 // source. 1501 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n'; 1502 unsigned NextStartColumn = 1503 ContentStartsOnNewline 1504 ? State.Stack.back().NestedBlockIndent + Style.IndentWidth 1505 : FirstStartColumn; 1506 1507 // The last start column is the column the raw string suffix starts if it is 1508 // put on a newline. 1509 // The last start column is the intended indentation of the raw string postfix 1510 // if it is put on a newline. It is determined by the following rules: 1511 // - if the raw string prefix starts on a newline, it is the column where 1512 // that raw string prefix starts, and 1513 // - if the raw string prefix does not start on a newline, it is the current 1514 // indent. 1515 unsigned LastStartColumn = Current.NewlinesBefore 1516 ? FirstStartColumn - NewPrefixSize 1517 : State.Stack.back().NestedBlockIndent; 1518 1519 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat( 1520 RawStringStyle, RawText, {tooling::Range(0, RawText.size())}, 1521 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>", 1522 /*Status=*/nullptr); 1523 1524 auto NewCode = applyAllReplacements(RawText, Fixes.first); 1525 tooling::Replacements NoFixes; 1526 if (!NewCode) { 1527 State.Column += Current.ColumnWidth; 1528 return 0; 1529 } 1530 if (!DryRun) { 1531 if (NewDelimiter != OldDelimiter) { 1532 // In 'R"delimiter(...', the delimiter starts 2 characters after the start 1533 // of the token. 1534 SourceLocation PrefixDelimiterStart = 1535 Current.Tok.getLocation().getLocWithOffset(2); 1536 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement( 1537 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1538 if (PrefixErr) { 1539 llvm::errs() 1540 << "Failed to update the prefix delimiter of a raw string: " 1541 << llvm::toString(std::move(PrefixErr)) << "\n"; 1542 } 1543 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at 1544 // position length - 1 - |delimiter|. 1545 SourceLocation SuffixDelimiterStart = 1546 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() - 1547 1 - OldDelimiter.size()); 1548 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement( 1549 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1550 if (SuffixErr) { 1551 llvm::errs() 1552 << "Failed to update the suffix delimiter of a raw string: " 1553 << llvm::toString(std::move(SuffixErr)) << "\n"; 1554 } 1555 } 1556 SourceLocation OriginLoc = 1557 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize); 1558 for (const tooling::Replacement &Fix : Fixes.first) { 1559 auto Err = Whitespaces.addReplacement(tooling::Replacement( 1560 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()), 1561 Fix.getLength(), Fix.getReplacementText())); 1562 if (Err) { 1563 llvm::errs() << "Failed to reformat raw string: " 1564 << llvm::toString(std::move(Err)) << "\n"; 1565 } 1566 } 1567 } 1568 unsigned RawLastLineEndColumn = getLastLineEndColumn( 1569 *NewCode, FirstStartColumn, Style.TabWidth, Encoding); 1570 State.Column = RawLastLineEndColumn + NewSuffixSize; 1571 // Since we're updating the column to after the raw string literal here, we 1572 // have to manually add the penalty for the prefix R"delim( over the column 1573 // limit. 1574 unsigned PrefixExcessCharacters = 1575 StartColumn + NewPrefixSize > Style.ColumnLimit ? 1576 StartColumn + NewPrefixSize - Style.ColumnLimit : 0; 1577 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter; 1578 } 1579 1580 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current, 1581 LineState &State) { 1582 // Break before further function parameters on all levels. 1583 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 1584 State.Stack[i].BreakBeforeParameter = true; 1585 1586 unsigned ColumnsUsed = State.Column; 1587 // We can only affect layout of the first and the last line, so the penalty 1588 // for all other lines is constant, and we ignore it. 1589 State.Column = Current.LastLineColumnWidth; 1590 1591 if (ColumnsUsed > getColumnLimit(State)) 1592 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State)); 1593 return 0; 1594 } 1595 1596 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current, 1597 LineState &State, bool DryRun, 1598 bool AllowBreak) { 1599 unsigned Penalty = 0; 1600 // Compute the raw string style to use in case this is a raw string literal 1601 // that can be reformatted. 1602 auto RawStringStyle = getRawStringStyle(Current, State); 1603 if (RawStringStyle && !Current.Finalized) { 1604 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun); 1605 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) { 1606 // Don't break multi-line tokens other than block comments and raw string 1607 // literals. Instead, just update the state. 1608 Penalty = addMultilineToken(Current, State); 1609 } else if (State.Line->Type != LT_ImportStatement) { 1610 // We generally don't break import statements. 1611 LineState OriginalState = State; 1612 1613 // Whether we force the reflowing algorithm to stay strictly within the 1614 // column limit. 1615 bool Strict = false; 1616 // Whether the first non-strict attempt at reflowing did intentionally 1617 // exceed the column limit. 1618 bool Exceeded = false; 1619 std::tie(Penalty, Exceeded) = breakProtrudingToken( 1620 Current, State, AllowBreak, /*DryRun=*/true, Strict); 1621 if (Exceeded) { 1622 // If non-strict reflowing exceeds the column limit, try whether strict 1623 // reflowing leads to an overall lower penalty. 1624 LineState StrictState = OriginalState; 1625 unsigned StrictPenalty = 1626 breakProtrudingToken(Current, StrictState, AllowBreak, 1627 /*DryRun=*/true, /*Strict=*/true) 1628 .first; 1629 Strict = StrictPenalty <= Penalty; 1630 if (Strict) { 1631 Penalty = StrictPenalty; 1632 State = StrictState; 1633 } 1634 } 1635 if (!DryRun) { 1636 // If we're not in dry-run mode, apply the changes with the decision on 1637 // strictness made above. 1638 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false, 1639 Strict); 1640 } 1641 } 1642 if (State.Column > getColumnLimit(State)) { 1643 unsigned ExcessCharacters = State.Column - getColumnLimit(State); 1644 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters; 1645 } 1646 return Penalty; 1647 } 1648 1649 // Returns the enclosing function name of a token, or the empty string if not 1650 // found. 1651 static StringRef getEnclosingFunctionName(const FormatToken &Current) { 1652 // Look for: 'function(' or 'function<templates>(' before Current. 1653 auto Tok = Current.getPreviousNonComment(); 1654 if (!Tok || !Tok->is(tok::l_paren)) 1655 return ""; 1656 Tok = Tok->getPreviousNonComment(); 1657 if (!Tok) 1658 return ""; 1659 if (Tok->is(TT_TemplateCloser)) { 1660 Tok = Tok->MatchingParen; 1661 if (Tok) 1662 Tok = Tok->getPreviousNonComment(); 1663 } 1664 if (!Tok || !Tok->is(tok::identifier)) 1665 return ""; 1666 return Tok->TokenText; 1667 } 1668 1669 llvm::Optional<FormatStyle> 1670 ContinuationIndenter::getRawStringStyle(const FormatToken &Current, 1671 const LineState &State) { 1672 if (!Current.isStringLiteral()) 1673 return None; 1674 auto Delimiter = getRawStringDelimiter(Current.TokenText); 1675 if (!Delimiter) 1676 return None; 1677 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter); 1678 if (!RawStringStyle && Delimiter->empty()) 1679 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle( 1680 getEnclosingFunctionName(Current)); 1681 if (!RawStringStyle) 1682 return None; 1683 RawStringStyle->ColumnLimit = getColumnLimit(State); 1684 return RawStringStyle; 1685 } 1686 1687 std::unique_ptr<BreakableToken> ContinuationIndenter::createBreakableToken( 1688 const FormatToken &Current, LineState &State, bool AllowBreak) { 1689 unsigned StartColumn = State.Column - Current.ColumnWidth; 1690 if (Current.isStringLiteral()) { 1691 // FIXME: String literal breaking is currently disabled for Java and JS, as 1692 // it requires strings to be merged using "+" which we don't support. 1693 if (Style.Language == FormatStyle::LK_Java || 1694 Style.Language == FormatStyle::LK_JavaScript || 1695 !Style.BreakStringLiterals || 1696 !AllowBreak) 1697 return nullptr; 1698 1699 // Don't break string literals inside preprocessor directives (except for 1700 // #define directives, as their contents are stored in separate lines and 1701 // are not affected by this check). 1702 // This way we avoid breaking code with line directives and unknown 1703 // preprocessor directives that contain long string literals. 1704 if (State.Line->Type == LT_PreprocessorDirective) 1705 return nullptr; 1706 // Exempts unterminated string literals from line breaking. The user will 1707 // likely want to terminate the string before any line breaking is done. 1708 if (Current.IsUnterminatedLiteral) 1709 return nullptr; 1710 // Don't break string literals inside Objective-C array literals (doing so 1711 // raises the warning -Wobjc-string-concatenation). 1712 if (State.Stack.back().IsInsideObjCArrayLiteral) { 1713 return nullptr; 1714 } 1715 1716 StringRef Text = Current.TokenText; 1717 StringRef Prefix; 1718 StringRef Postfix; 1719 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'. 1720 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to 1721 // reduce the overhead) for each FormatToken, which is a string, so that we 1722 // don't run multiple checks here on the hot path. 1723 if ((Text.endswith(Postfix = "\"") && 1724 (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") || 1725 Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") || 1726 Text.startswith(Prefix = "u8\"") || 1727 Text.startswith(Prefix = "L\""))) || 1728 (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) { 1729 // We need this to address the case where there is an unbreakable tail 1730 // only if certain other formatting decisions have been taken. The 1731 // UnbreakableTailLength of Current is an overapproximation is that case 1732 // and we need to be correct here. 1733 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State)) 1734 ? 0 1735 : Current.UnbreakableTailLength; 1736 return llvm::make_unique<BreakableStringLiteral>( 1737 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength, 1738 State.Line->InPPDirective, Encoding, Style); 1739 } 1740 } else if (Current.is(TT_BlockComment)) { 1741 if (!Style.ReflowComments || 1742 // If a comment token switches formatting, like 1743 // /* clang-format on */, we don't want to break it further, 1744 // but we may still want to adjust its indentation. 1745 switchesFormatting(Current)) { 1746 return nullptr; 1747 } 1748 return llvm::make_unique<BreakableBlockComment>( 1749 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1750 State.Line->InPPDirective, Encoding, Style); 1751 } else if (Current.is(TT_LineComment) && 1752 (Current.Previous == nullptr || 1753 Current.Previous->isNot(TT_ImplicitStringLiteral))) { 1754 if (!Style.ReflowComments || 1755 CommentPragmasRegex.match(Current.TokenText.substr(2)) || 1756 switchesFormatting(Current)) 1757 return nullptr; 1758 return llvm::make_unique<BreakableLineCommentSection>( 1759 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1760 /*InPPDirective=*/false, Encoding, Style); 1761 } 1762 return nullptr; 1763 } 1764 1765 std::pair<unsigned, bool> 1766 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current, 1767 LineState &State, bool AllowBreak, 1768 bool DryRun, bool Strict) { 1769 std::unique_ptr<const BreakableToken> Token = 1770 createBreakableToken(Current, State, AllowBreak); 1771 if (!Token) 1772 return {0, false}; 1773 assert(Token->getLineCount() > 0); 1774 unsigned ColumnLimit = getColumnLimit(State); 1775 if (Current.is(TT_LineComment)) { 1776 // We don't insert backslashes when breaking line comments. 1777 ColumnLimit = Style.ColumnLimit; 1778 } 1779 if (Current.UnbreakableTailLength >= ColumnLimit) 1780 return {0, false}; 1781 // ColumnWidth was already accounted into State.Column before calling 1782 // breakProtrudingToken. 1783 unsigned StartColumn = State.Column - Current.ColumnWidth; 1784 unsigned NewBreakPenalty = Current.isStringLiteral() 1785 ? Style.PenaltyBreakString 1786 : Style.PenaltyBreakComment; 1787 // Stores whether we intentionally decide to let a line exceed the column 1788 // limit. 1789 bool Exceeded = false; 1790 // Stores whether we introduce a break anywhere in the token. 1791 bool BreakInserted = Token->introducesBreakBeforeToken(); 1792 // Store whether we inserted a new line break at the end of the previous 1793 // logical line. 1794 bool NewBreakBefore = false; 1795 // We use a conservative reflowing strategy. Reflow starts after a line is 1796 // broken or the corresponding whitespace compressed. Reflow ends as soon as a 1797 // line that doesn't get reflown with the previous line is reached. 1798 bool Reflow = false; 1799 // Keep track of where we are in the token: 1800 // Where we are in the content of the current logical line. 1801 unsigned TailOffset = 0; 1802 // The column number we're currently at. 1803 unsigned ContentStartColumn = 1804 Token->getContentStartColumn(0, /*Break=*/false); 1805 // The number of columns left in the current logical line after TailOffset. 1806 unsigned RemainingTokenColumns = 1807 Token->getRemainingLength(0, TailOffset, ContentStartColumn); 1808 // Adapt the start of the token, for example indent. 1809 if (!DryRun) 1810 Token->adaptStartOfLine(0, Whitespaces); 1811 1812 unsigned ContentIndent = 0; 1813 unsigned Penalty = 0; 1814 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column " 1815 << StartColumn << ".\n"); 1816 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount(); 1817 LineIndex != EndIndex; ++LineIndex) { 1818 LLVM_DEBUG(llvm::dbgs() 1819 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n"); 1820 NewBreakBefore = false; 1821 // If we did reflow the previous line, we'll try reflowing again. Otherwise 1822 // we'll start reflowing if the current line is broken or whitespace is 1823 // compressed. 1824 bool TryReflow = Reflow; 1825 // Break the current token until we can fit the rest of the line. 1826 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 1827 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: " 1828 << (ContentStartColumn + RemainingTokenColumns) 1829 << ", space: " << ColumnLimit 1830 << ", reflown prefix: " << ContentStartColumn 1831 << ", offset in line: " << TailOffset << "\n"); 1832 // If the current token doesn't fit, find the latest possible split in the 1833 // current line so that breaking at it will be under the column limit. 1834 // FIXME: Use the earliest possible split while reflowing to correctly 1835 // compress whitespace within a line. 1836 BreakableToken::Split Split = 1837 Token->getSplit(LineIndex, TailOffset, ColumnLimit, 1838 ContentStartColumn, CommentPragmasRegex); 1839 if (Split.first == StringRef::npos) { 1840 // No break opportunity - update the penalty and continue with the next 1841 // logical line. 1842 if (LineIndex < EndIndex - 1) 1843 // The last line's penalty is handled in addNextStateToQueue() or when 1844 // calling replaceWhitespaceAfterLastLine below. 1845 Penalty += Style.PenaltyExcessCharacter * 1846 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 1847 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n"); 1848 break; 1849 } 1850 assert(Split.first != 0); 1851 1852 if (Token->supportsReflow()) { 1853 // Check whether the next natural split point after the current one can 1854 // still fit the line, either because we can compress away whitespace, 1855 // or because the penalty the excess characters introduce is lower than 1856 // the break penalty. 1857 // We only do this for tokens that support reflowing, and thus allow us 1858 // to change the whitespace arbitrarily (e.g. comments). 1859 // Other tokens, like string literals, can be broken on arbitrary 1860 // positions. 1861 1862 // First, compute the columns from TailOffset to the next possible split 1863 // position. 1864 // For example: 1865 // ColumnLimit: | 1866 // // Some text that breaks 1867 // ^ tail offset 1868 // ^-- split 1869 // ^-------- to split columns 1870 // ^--- next split 1871 // ^--------------- to next split columns 1872 unsigned ToSplitColumns = Token->getRangeLength( 1873 LineIndex, TailOffset, Split.first, ContentStartColumn); 1874 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n"); 1875 1876 BreakableToken::Split NextSplit = Token->getSplit( 1877 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit, 1878 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex); 1879 // Compute the columns necessary to fit the next non-breakable sequence 1880 // into the current line. 1881 unsigned ToNextSplitColumns = 0; 1882 if (NextSplit.first == StringRef::npos) { 1883 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset, 1884 ContentStartColumn); 1885 } else { 1886 ToNextSplitColumns = Token->getRangeLength( 1887 LineIndex, TailOffset, 1888 Split.first + Split.second + NextSplit.first, ContentStartColumn); 1889 } 1890 // Compress the whitespace between the break and the start of the next 1891 // unbreakable sequence. 1892 ToNextSplitColumns = 1893 Token->getLengthAfterCompression(ToNextSplitColumns, Split); 1894 LLVM_DEBUG(llvm::dbgs() 1895 << " ContentStartColumn: " << ContentStartColumn << "\n"); 1896 LLVM_DEBUG(llvm::dbgs() 1897 << " ToNextSplit: " << ToNextSplitColumns << "\n"); 1898 // If the whitespace compression makes us fit, continue on the current 1899 // line. 1900 bool ContinueOnLine = 1901 ContentStartColumn + ToNextSplitColumns <= ColumnLimit; 1902 unsigned ExcessCharactersPenalty = 0; 1903 if (!ContinueOnLine && !Strict) { 1904 // Similarly, if the excess characters' penalty is lower than the 1905 // penalty of introducing a new break, continue on the current line. 1906 ExcessCharactersPenalty = 1907 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) * 1908 Style.PenaltyExcessCharacter; 1909 LLVM_DEBUG(llvm::dbgs() 1910 << " Penalty excess: " << ExcessCharactersPenalty 1911 << "\n break : " << NewBreakPenalty << "\n"); 1912 if (ExcessCharactersPenalty < NewBreakPenalty) { 1913 Exceeded = true; 1914 ContinueOnLine = true; 1915 } 1916 } 1917 if (ContinueOnLine) { 1918 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n"); 1919 // The current line fits after compressing the whitespace - reflow 1920 // the next line into it if possible. 1921 TryReflow = true; 1922 if (!DryRun) 1923 Token->compressWhitespace(LineIndex, TailOffset, Split, 1924 Whitespaces); 1925 // When we continue on the same line, leave one space between content. 1926 ContentStartColumn += ToSplitColumns + 1; 1927 Penalty += ExcessCharactersPenalty; 1928 TailOffset += Split.first + Split.second; 1929 RemainingTokenColumns = Token->getRemainingLength( 1930 LineIndex, TailOffset, ContentStartColumn); 1931 continue; 1932 } 1933 } 1934 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n"); 1935 // Update the ContentIndent only if the current line was not reflown with 1936 // the previous line, since in that case the previous line should still 1937 // determine the ContentIndent. Also never intent the last line. 1938 if (!Reflow) 1939 ContentIndent = Token->getContentIndent(LineIndex); 1940 LLVM_DEBUG(llvm::dbgs() 1941 << " ContentIndent: " << ContentIndent << "\n"); 1942 ContentStartColumn = ContentIndent + Token->getContentStartColumn( 1943 LineIndex, /*Break=*/true); 1944 1945 unsigned NewRemainingTokenColumns = Token->getRemainingLength( 1946 LineIndex, TailOffset + Split.first + Split.second, 1947 ContentStartColumn); 1948 if (NewRemainingTokenColumns == 0) { 1949 // No content to indent. 1950 ContentIndent = 0; 1951 ContentStartColumn = 1952 Token->getContentStartColumn(LineIndex, /*Break=*/true); 1953 NewRemainingTokenColumns = Token->getRemainingLength( 1954 LineIndex, TailOffset + Split.first + Split.second, 1955 ContentStartColumn); 1956 } 1957 1958 // When breaking before a tab character, it may be moved by a few columns, 1959 // but will still be expanded to the next tab stop, so we don't save any 1960 // columns. 1961 if (NewRemainingTokenColumns == RemainingTokenColumns) { 1962 // FIXME: Do we need to adjust the penalty? 1963 break; 1964 } 1965 assert(NewRemainingTokenColumns < RemainingTokenColumns); 1966 1967 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first 1968 << ", " << Split.second << "\n"); 1969 if (!DryRun) 1970 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent, 1971 Whitespaces); 1972 1973 Penalty += NewBreakPenalty; 1974 TailOffset += Split.first + Split.second; 1975 RemainingTokenColumns = NewRemainingTokenColumns; 1976 BreakInserted = true; 1977 NewBreakBefore = true; 1978 } 1979 // In case there's another line, prepare the state for the start of the next 1980 // line. 1981 if (LineIndex + 1 != EndIndex) { 1982 unsigned NextLineIndex = LineIndex + 1; 1983 if (NewBreakBefore) 1984 // After breaking a line, try to reflow the next line into the current 1985 // one once RemainingTokenColumns fits. 1986 TryReflow = true; 1987 if (TryReflow) { 1988 // We decided that we want to try reflowing the next line into the 1989 // current one. 1990 // We will now adjust the state as if the reflow is successful (in 1991 // preparation for the next line), and see whether that works. If we 1992 // decide that we cannot reflow, we will later reset the state to the 1993 // start of the next line. 1994 Reflow = false; 1995 // As we did not continue breaking the line, RemainingTokenColumns is 1996 // known to fit after ContentStartColumn. Adapt ContentStartColumn to 1997 // the position at which we want to format the next line if we do 1998 // actually reflow. 1999 // When we reflow, we need to add a space between the end of the current 2000 // line and the next line's start column. 2001 ContentStartColumn += RemainingTokenColumns + 1; 2002 // Get the split that we need to reflow next logical line into the end 2003 // of the current one; the split will include any leading whitespace of 2004 // the next logical line. 2005 BreakableToken::Split SplitBeforeNext = 2006 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex); 2007 LLVM_DEBUG(llvm::dbgs() 2008 << " Size of reflown text: " << ContentStartColumn 2009 << "\n Potential reflow split: "); 2010 if (SplitBeforeNext.first != StringRef::npos) { 2011 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", " 2012 << SplitBeforeNext.second << "\n"); 2013 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second; 2014 // If the rest of the next line fits into the current line below the 2015 // column limit, we can safely reflow. 2016 RemainingTokenColumns = Token->getRemainingLength( 2017 NextLineIndex, TailOffset, ContentStartColumn); 2018 Reflow = true; 2019 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2020 LLVM_DEBUG(llvm::dbgs() 2021 << " Over limit after reflow, need: " 2022 << (ContentStartColumn + RemainingTokenColumns) 2023 << ", space: " << ColumnLimit 2024 << ", reflown prefix: " << ContentStartColumn 2025 << ", offset in line: " << TailOffset << "\n"); 2026 // If the whole next line does not fit, try to find a point in 2027 // the next line at which we can break so that attaching the part 2028 // of the next line to that break point onto the current line is 2029 // below the column limit. 2030 BreakableToken::Split Split = 2031 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit, 2032 ContentStartColumn, CommentPragmasRegex); 2033 if (Split.first == StringRef::npos) { 2034 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n"); 2035 Reflow = false; 2036 } else { 2037 // Check whether the first split point gets us below the column 2038 // limit. Note that we will execute this split below as part of 2039 // the normal token breaking and reflow logic within the line. 2040 unsigned ToSplitColumns = Token->getRangeLength( 2041 NextLineIndex, TailOffset, Split.first, ContentStartColumn); 2042 if (ContentStartColumn + ToSplitColumns > ColumnLimit) { 2043 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: " 2044 << (ContentStartColumn + ToSplitColumns) 2045 << ", space: " << ColumnLimit); 2046 unsigned ExcessCharactersPenalty = 2047 (ContentStartColumn + ToSplitColumns - ColumnLimit) * 2048 Style.PenaltyExcessCharacter; 2049 if (NewBreakPenalty < ExcessCharactersPenalty) { 2050 Reflow = false; 2051 } 2052 } 2053 } 2054 } 2055 } else { 2056 LLVM_DEBUG(llvm::dbgs() << "not found.\n"); 2057 } 2058 } 2059 if (!Reflow) { 2060 // If we didn't reflow into the next line, the only space to consider is 2061 // the next logical line. Reset our state to match the start of the next 2062 // line. 2063 TailOffset = 0; 2064 ContentStartColumn = 2065 Token->getContentStartColumn(NextLineIndex, /*Break=*/false); 2066 RemainingTokenColumns = Token->getRemainingLength( 2067 NextLineIndex, TailOffset, ContentStartColumn); 2068 // Adapt the start of the token, for example indent. 2069 if (!DryRun) 2070 Token->adaptStartOfLine(NextLineIndex, Whitespaces); 2071 } else { 2072 // If we found a reflow split and have added a new break before the next 2073 // line, we are going to remove the line break at the start of the next 2074 // logical line. For example, here we'll add a new line break after 2075 // 'text', and subsequently delete the line break between 'that' and 2076 // 'reflows'. 2077 // // some text that 2078 // // reflows 2079 // -> 2080 // // some text 2081 // // that reflows 2082 // When adding the line break, we also added the penalty for it, so we 2083 // need to subtract that penalty again when we remove the line break due 2084 // to reflowing. 2085 if (NewBreakBefore) { 2086 assert(Penalty >= NewBreakPenalty); 2087 Penalty -= NewBreakPenalty; 2088 } 2089 if (!DryRun) 2090 Token->reflow(NextLineIndex, Whitespaces); 2091 } 2092 } 2093 } 2094 2095 BreakableToken::Split SplitAfterLastLine = 2096 Token->getSplitAfterLastLine(TailOffset); 2097 if (SplitAfterLastLine.first != StringRef::npos) { 2098 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n"); 2099 2100 // We add the last line's penalty here, since that line is going to be split 2101 // now. 2102 Penalty += Style.PenaltyExcessCharacter * 2103 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2104 2105 if (!DryRun) 2106 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine, 2107 Whitespaces); 2108 ContentStartColumn = 2109 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true); 2110 RemainingTokenColumns = Token->getRemainingLength( 2111 Token->getLineCount() - 1, 2112 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second, 2113 ContentStartColumn); 2114 } 2115 2116 State.Column = ContentStartColumn + RemainingTokenColumns - 2117 Current.UnbreakableTailLength; 2118 2119 if (BreakInserted) { 2120 // If we break the token inside a parameter list, we need to break before 2121 // the next parameter on all levels, so that the next parameter is clearly 2122 // visible. Line comments already introduce a break. 2123 if (Current.isNot(TT_LineComment)) { 2124 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 2125 State.Stack[i].BreakBeforeParameter = true; 2126 } 2127 2128 if (Current.is(TT_BlockComment)) 2129 State.NoContinuation = true; 2130 2131 State.Stack.back().LastSpace = StartColumn; 2132 } 2133 2134 Token->updateNextToken(State); 2135 2136 return {Penalty, Exceeded}; 2137 } 2138 2139 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const { 2140 // In preprocessor directives reserve two chars for trailing " \" 2141 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0); 2142 } 2143 2144 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) { 2145 const FormatToken &Current = *State.NextToken; 2146 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral)) 2147 return false; 2148 // We never consider raw string literals "multiline" for the purpose of 2149 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased 2150 // (see TokenAnnotator::mustBreakBefore(). 2151 if (Current.TokenText.startswith("R\"")) 2152 return false; 2153 if (Current.IsMultiline) 2154 return true; 2155 if (Current.getNextNonComment() && 2156 Current.getNextNonComment()->isStringLiteral()) 2157 return true; // Implicit concatenation. 2158 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals && 2159 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength > 2160 Style.ColumnLimit) 2161 return true; // String will be split. 2162 return false; 2163 } 2164 2165 } // namespace format 2166 } // namespace clang 2167