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 !(Previous.closesScopeAfterBlock() && 408 State.Column <= NewLineColumn)) 409 return true; 410 411 // If the template declaration spans multiple lines, force wrap before the 412 // function/class declaration 413 if (Previous.ClosesTemplateDeclaration && 414 State.Stack.back().BreakBeforeParameter && Current.CanBreakBefore) 415 return true; 416 417 if (State.Column <= NewLineColumn) 418 return false; 419 420 if (Style.AlwaysBreakBeforeMultilineStrings && 421 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth || 422 Previous.is(tok::comma) || Current.NestingLevel < 2) && 423 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at) && 424 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) && 425 nextIsMultilineString(State)) 426 return true; 427 428 // Using CanBreakBefore here and below takes care of the decision whether the 429 // current style uses wrapping before or after operators for the given 430 // operator. 431 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) { 432 // If we need to break somewhere inside the LHS of a binary expression, we 433 // should also break after the operator. Otherwise, the formatting would 434 // hide the operator precedence, e.g. in: 435 // if (aaaaaaaaaaaaaa == 436 // bbbbbbbbbbbbbb && c) {.. 437 // For comparisons, we only apply this rule, if the LHS is a binary 438 // expression itself as otherwise, the line breaks seem superfluous. 439 // We need special cases for ">>" which we have split into two ">" while 440 // lexing in order to make template parsing easier. 441 bool IsComparison = (Previous.getPrecedence() == prec::Relational || 442 Previous.getPrecedence() == prec::Equality || 443 Previous.getPrecedence() == prec::Spaceship) && 444 Previous.Previous && 445 Previous.Previous->isNot(TT_BinaryOperator); // For >>. 446 bool LHSIsBinaryExpr = 447 Previous.Previous && Previous.Previous->EndsBinaryExpression; 448 if ((!IsComparison || LHSIsBinaryExpr) && !Current.isTrailingComment() && 449 Previous.getPrecedence() != prec::Assignment && 450 State.Stack.back().BreakBeforeParameter) 451 return true; 452 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore && 453 State.Stack.back().BreakBeforeParameter) { 454 return true; 455 } 456 457 // Same as above, but for the first "<<" operator. 458 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) && 459 State.Stack.back().BreakBeforeParameter && 460 State.Stack.back().FirstLessLess == 0) 461 return true; 462 463 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) { 464 // Always break after "template <...>" and leading annotations. This is only 465 // for cases where the entire line does not fit on a single line as a 466 // different LineFormatter would be used otherwise. 467 if (Previous.ClosesTemplateDeclaration) 468 return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No; 469 if (Previous.is(TT_FunctionAnnotationRParen)) 470 return true; 471 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) && 472 Current.isNot(TT_LeadingJavaAnnotation)) 473 return true; 474 } 475 476 // If the return type spans multiple lines, wrap before the function name. 477 if ((Current.is(TT_FunctionDeclarationName) || 478 (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) && 479 !Previous.is(tok::kw_template) && State.Stack.back().BreakBeforeParameter) 480 return true; 481 482 // The following could be precomputed as they do not depend on the state. 483 // However, as they should take effect only if the UnwrappedLine does not fit 484 // into the ColumnLimit, they are checked here in the ContinuationIndenter. 485 if (Style.ColumnLimit != 0 && Previous.BlockKind == BK_Block && 486 Previous.is(tok::l_brace) && !Current.isOneOf(tok::r_brace, tok::comment)) 487 return true; 488 489 if (Current.is(tok::lessless) && 490 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") || 491 (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") || 492 Previous.TokenText == "\'\\n\'")))) 493 return true; 494 495 if (Previous.is(TT_BlockComment) && Previous.IsMultiline) 496 return true; 497 498 if (State.NoContinuation) 499 return true; 500 501 return false; 502 } 503 504 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline, 505 bool DryRun, 506 unsigned ExtraSpaces) { 507 const FormatToken &Current = *State.NextToken; 508 509 assert(!State.Stack.empty()); 510 State.NoContinuation = false; 511 512 if ((Current.is(TT_ImplicitStringLiteral) && 513 (Current.Previous->Tok.getIdentifierInfo() == nullptr || 514 Current.Previous->Tok.getIdentifierInfo()->getPPKeywordID() == 515 tok::pp_not_keyword))) { 516 unsigned EndColumn = 517 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd()); 518 if (Current.LastNewlineOffset != 0) { 519 // If there is a newline within this token, the final column will solely 520 // determined by the current end column. 521 State.Column = EndColumn; 522 } else { 523 unsigned StartColumn = 524 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin()); 525 assert(EndColumn >= StartColumn); 526 State.Column += EndColumn - StartColumn; 527 } 528 moveStateToNextToken(State, DryRun, /*Newline=*/false); 529 return 0; 530 } 531 532 unsigned Penalty = 0; 533 if (Newline) 534 Penalty = addTokenOnNewLine(State, DryRun); 535 else 536 addTokenOnCurrentLine(State, DryRun, ExtraSpaces); 537 538 return moveStateToNextToken(State, DryRun, Newline) + Penalty; 539 } 540 541 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun, 542 unsigned ExtraSpaces) { 543 FormatToken &Current = *State.NextToken; 544 const FormatToken &Previous = *State.NextToken->Previous; 545 if (Current.is(tok::equal) && 546 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) && 547 State.Stack.back().VariablePos == 0) { 548 State.Stack.back().VariablePos = State.Column; 549 // Move over * and & if they are bound to the variable name. 550 const FormatToken *Tok = &Previous; 551 while (Tok && State.Stack.back().VariablePos >= Tok->ColumnWidth) { 552 State.Stack.back().VariablePos -= Tok->ColumnWidth; 553 if (Tok->SpacesRequiredBefore != 0) 554 break; 555 Tok = Tok->Previous; 556 } 557 if (Previous.PartOfMultiVariableDeclStmt) 558 State.Stack.back().LastSpace = State.Stack.back().VariablePos; 559 } 560 561 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces; 562 563 // Indent preprocessor directives after the hash if required. 564 int PPColumnCorrection = 0; 565 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 566 Previous.is(tok::hash) && State.FirstIndent > 0 && 567 (State.Line->Type == LT_PreprocessorDirective || 568 State.Line->Type == LT_ImportStatement)) { 569 Spaces += State.FirstIndent; 570 571 // For preprocessor indent with tabs, State.Column will be 1 because of the 572 // hash. This causes second-level indents onward to have an extra space 573 // after the tabs. We avoid this misalignment by subtracting 1 from the 574 // column value passed to replaceWhitespace(). 575 if (Style.UseTab != FormatStyle::UT_Never) 576 PPColumnCorrection = -1; 577 } 578 579 if (!DryRun) 580 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces, 581 State.Column + Spaces + PPColumnCorrection); 582 583 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance 584 // declaration unless there is multiple inheritance. 585 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 586 Current.is(TT_InheritanceColon)) 587 State.Stack.back().NoLineBreak = true; 588 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon && 589 Previous.is(TT_InheritanceColon)) 590 State.Stack.back().NoLineBreak = true; 591 592 if (Current.is(TT_SelectorName) && 593 !State.Stack.back().ObjCSelectorNameFound) { 594 unsigned MinIndent = 595 std::max(State.FirstIndent + Style.ContinuationIndentWidth, 596 State.Stack.back().Indent); 597 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth; 598 if (Current.LongestObjCSelectorName == 0) 599 State.Stack.back().AlignColons = false; 600 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos) 601 State.Stack.back().ColonPos = MinIndent + Current.LongestObjCSelectorName; 602 else 603 State.Stack.back().ColonPos = FirstColonPos; 604 } 605 606 // In "AlwaysBreak" mode, enforce wrapping directly after the parenthesis by 607 // disallowing any further line breaks if there is no line break after the 608 // opening parenthesis. Don't break if it doesn't conserve columns. 609 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak && 610 Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) && 611 State.Column > getNewLineColumn(State) && 612 (!Previous.Previous || !Previous.Previous->isOneOf( 613 tok::kw_for, tok::kw_while, tok::kw_switch)) && 614 // Don't do this for simple (no expressions) one-argument function calls 615 // as that feels like needlessly wasting whitespace, e.g.: 616 // 617 // caaaaaaaaaaaall( 618 // caaaaaaaaaaaall( 619 // caaaaaaaaaaaall( 620 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa)))); 621 Current.FakeLParens.size() > 0 && 622 Current.FakeLParens.back() > prec::Unknown) 623 State.Stack.back().NoLineBreak = true; 624 if (Previous.is(TT_TemplateString) && Previous.opensScope()) 625 State.Stack.back().NoLineBreak = true; 626 627 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign && 628 Previous.opensScope() && Previous.isNot(TT_ObjCMethodExpr) && 629 (Current.isNot(TT_LineComment) || Previous.BlockKind == BK_BracedInit)) 630 State.Stack.back().Indent = State.Column + Spaces; 631 if (State.Stack.back().AvoidBinPacking && startsNextParameter(Current, Style)) 632 State.Stack.back().NoLineBreak = true; 633 if (startsSegmentOfBuilderTypeCall(Current) && 634 State.Column > getNewLineColumn(State)) 635 State.Stack.back().ContainsUnwrappedBuilder = true; 636 637 if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java) 638 State.Stack.back().NoLineBreak = true; 639 if (Current.isMemberAccess() && Previous.is(tok::r_paren) && 640 (Previous.MatchingParen && 641 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) 642 // If there is a function call with long parameters, break before trailing 643 // calls. This prevents things like: 644 // EXPECT_CALL(SomeLongParameter).Times( 645 // 2); 646 // We don't want to do this for short parameters as they can just be 647 // indexes. 648 State.Stack.back().NoLineBreak = true; 649 650 // Don't allow the RHS of an operator to be split over multiple lines unless 651 // there is a line-break right after the operator. 652 // Exclude relational operators, as there, it is always more desirable to 653 // have the LHS 'left' of the RHS. 654 const FormatToken *P = Current.getPreviousNonComment(); 655 if (!Current.is(tok::comment) && P && 656 (P->isOneOf(TT_BinaryOperator, tok::comma) || 657 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) && 658 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) && 659 P->getPrecedence() != prec::Assignment && 660 P->getPrecedence() != prec::Relational && 661 P->getPrecedence() != prec::Spaceship) { 662 bool BreakBeforeOperator = 663 P->MustBreakBefore || P->is(tok::lessless) || 664 (P->is(TT_BinaryOperator) && 665 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) || 666 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators); 667 // Don't do this if there are only two operands. In these cases, there is 668 // always a nice vertical separation between them and the extra line break 669 // does not help. 670 bool HasTwoOperands = 671 P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr); 672 if ((!BreakBeforeOperator && !(HasTwoOperands && Style.AlignOperands)) || 673 (!State.Stack.back().LastOperatorWrapped && BreakBeforeOperator)) 674 State.Stack.back().NoLineBreakInOperand = true; 675 } 676 677 State.Column += Spaces; 678 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) && 679 Previous.Previous && 680 (Previous.Previous->isOneOf(tok::kw_if, tok::kw_for) || 681 Previous.Previous->endsSequence(tok::kw_constexpr, tok::kw_if))) { 682 // Treat the condition inside an if as if it was a second function 683 // parameter, i.e. let nested calls have a continuation indent. 684 State.Stack.back().LastSpace = State.Column; 685 State.Stack.back().NestedBlockIndent = State.Column; 686 } else if (!Current.isOneOf(tok::comment, tok::caret) && 687 ((Previous.is(tok::comma) && 688 !Previous.is(TT_OverloadedOperator)) || 689 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) { 690 State.Stack.back().LastSpace = State.Column; 691 } else if (Previous.is(TT_CtorInitializerColon) && 692 Style.BreakConstructorInitializers == 693 FormatStyle::BCIS_AfterColon) { 694 State.Stack.back().Indent = State.Column; 695 State.Stack.back().LastSpace = State.Column; 696 } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 697 TT_CtorInitializerColon)) && 698 ((Previous.getPrecedence() != prec::Assignment && 699 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 || 700 Previous.NextOperator)) || 701 Current.StartsBinaryExpression)) { 702 // Indent relative to the RHS of the expression unless this is a simple 703 // assignment without binary expression on the RHS. Also indent relative to 704 // unary operators and the colons of constructor initializers. 705 if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None) 706 State.Stack.back().LastSpace = State.Column; 707 } else if (Previous.is(TT_InheritanceColon)) { 708 State.Stack.back().Indent = State.Column; 709 State.Stack.back().LastSpace = State.Column; 710 } else if (Previous.opensScope()) { 711 // If a function has a trailing call, indent all parameters from the 712 // opening parenthesis. This avoids confusing indents like: 713 // OuterFunction(InnerFunctionCall( // break 714 // ParameterToInnerFunction)) // break 715 // .SecondInnerFunctionCall(); 716 bool HasTrailingCall = false; 717 if (Previous.MatchingParen) { 718 const FormatToken *Next = Previous.MatchingParen->getNextNonComment(); 719 HasTrailingCall = Next && Next->isMemberAccess(); 720 } 721 if (HasTrailingCall && State.Stack.size() > 1 && 722 State.Stack[State.Stack.size() - 2].CallContinuation == 0) 723 State.Stack.back().LastSpace = State.Column; 724 } 725 } 726 727 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State, 728 bool DryRun) { 729 FormatToken &Current = *State.NextToken; 730 const FormatToken &Previous = *State.NextToken->Previous; 731 732 // Extra penalty that needs to be added because of the way certain line 733 // breaks are chosen. 734 unsigned Penalty = 0; 735 736 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 737 const FormatToken *NextNonComment = Previous.getNextNonComment(); 738 if (!NextNonComment) 739 NextNonComment = &Current; 740 // The first line break on any NestingLevel causes an extra penalty in order 741 // prefer similar line breaks. 742 if (!State.Stack.back().ContainsLineBreak) 743 Penalty += 15; 744 State.Stack.back().ContainsLineBreak = true; 745 746 Penalty += State.NextToken->SplitPenalty; 747 748 // Breaking before the first "<<" is generally not desirable if the LHS is 749 // short. Also always add the penalty if the LHS is split over multiple lines 750 // to avoid unnecessary line breaks that just work around this penalty. 751 if (NextNonComment->is(tok::lessless) && 752 State.Stack.back().FirstLessLess == 0 && 753 (State.Column <= Style.ColumnLimit / 3 || 754 State.Stack.back().BreakBeforeParameter)) 755 Penalty += Style.PenaltyBreakFirstLessLess; 756 757 State.Column = getNewLineColumn(State); 758 759 // Indent nested blocks relative to this column, unless in a very specific 760 // JavaScript special case where: 761 // 762 // var loooooong_name = 763 // function() { 764 // // code 765 // } 766 // 767 // is common and should be formatted like a free-standing function. The same 768 // goes for wrapping before the lambda return type arrow. 769 if (!Current.is(TT_LambdaArrow) && 770 (Style.Language != FormatStyle::LK_JavaScript || 771 Current.NestingLevel != 0 || !PreviousNonComment || 772 !PreviousNonComment->is(tok::equal) || 773 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) 774 State.Stack.back().NestedBlockIndent = State.Column; 775 776 if (NextNonComment->isMemberAccess()) { 777 if (State.Stack.back().CallContinuation == 0) 778 State.Stack.back().CallContinuation = State.Column; 779 } else if (NextNonComment->is(TT_SelectorName)) { 780 if (!State.Stack.back().ObjCSelectorNameFound) { 781 if (NextNonComment->LongestObjCSelectorName == 0) { 782 State.Stack.back().AlignColons = false; 783 } else { 784 State.Stack.back().ColonPos = 785 (shouldIndentWrappedSelectorName(Style, State.Line->Type) 786 ? std::max(State.Stack.back().Indent, 787 State.FirstIndent + Style.ContinuationIndentWidth) 788 : State.Stack.back().Indent) + 789 std::max(NextNonComment->LongestObjCSelectorName, 790 NextNonComment->ColumnWidth); 791 } 792 } else if (State.Stack.back().AlignColons && 793 State.Stack.back().ColonPos <= NextNonComment->ColumnWidth) { 794 State.Stack.back().ColonPos = State.Column + NextNonComment->ColumnWidth; 795 } 796 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 797 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) { 798 // FIXME: This is hacky, find a better way. The problem is that in an ObjC 799 // method expression, the block should be aligned to the line starting it, 800 // e.g.: 801 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason 802 // ^(int *i) { 803 // // ... 804 // }]; 805 // Thus, we set LastSpace of the next higher NestingLevel, to which we move 806 // when we consume all of the "}"'s FakeRParens at the "{". 807 if (State.Stack.size() > 1) 808 State.Stack[State.Stack.size() - 2].LastSpace = 809 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 810 Style.ContinuationIndentWidth; 811 } 812 813 if ((PreviousNonComment && 814 PreviousNonComment->isOneOf(tok::comma, tok::semi) && 815 !State.Stack.back().AvoidBinPacking) || 816 Previous.is(TT_BinaryOperator)) 817 State.Stack.back().BreakBeforeParameter = false; 818 if (PreviousNonComment && 819 PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) && 820 Current.NestingLevel == 0) 821 State.Stack.back().BreakBeforeParameter = false; 822 if (NextNonComment->is(tok::question) || 823 (PreviousNonComment && PreviousNonComment->is(tok::question))) 824 State.Stack.back().BreakBeforeParameter = true; 825 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore) 826 State.Stack.back().BreakBeforeParameter = false; 827 828 if (!DryRun) { 829 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1; 830 if (Current.is(tok::r_brace) && Current.MatchingParen && 831 // Only strip trailing empty lines for l_braces that have children, i.e. 832 // for function expressions (lambdas, arrows, etc). 833 !Current.MatchingParen->Children.empty()) { 834 // lambdas and arrow functions are expressions, thus their r_brace is not 835 // on its own line, and thus not covered by UnwrappedLineFormatter's logic 836 // about removing empty lines on closing blocks. Special case them here. 837 MaxEmptyLinesToKeep = 1; 838 } 839 unsigned Newlines = std::max( 840 1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep)); 841 bool ContinuePPDirective = 842 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement; 843 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column, 844 ContinuePPDirective); 845 } 846 847 if (!Current.isTrailingComment()) 848 State.Stack.back().LastSpace = State.Column; 849 if (Current.is(tok::lessless)) 850 // If we are breaking before a "<<", we always want to indent relative to 851 // RHS. This is necessary only for "<<", as we special-case it and don't 852 // always indent relative to the RHS. 853 State.Stack.back().LastSpace += 3; // 3 -> width of "<< ". 854 855 State.StartOfLineLevel = Current.NestingLevel; 856 State.LowestLevelOnLine = Current.NestingLevel; 857 858 // Any break on this level means that the parent level has been broken 859 // and we need to avoid bin packing there. 860 bool NestedBlockSpecialCase = 861 !Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 && 862 State.Stack[State.Stack.size() - 2].NestedBlockInlined; 863 if (!NestedBlockSpecialCase) 864 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 865 State.Stack[i].BreakBeforeParameter = true; 866 867 if (PreviousNonComment && 868 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) && 869 (PreviousNonComment->isNot(TT_TemplateCloser) || 870 Current.NestingLevel != 0) && 871 !PreviousNonComment->isOneOf( 872 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation, 873 TT_LeadingJavaAnnotation) && 874 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope()) 875 State.Stack.back().BreakBeforeParameter = true; 876 877 // If we break after { or the [ of an array initializer, we should also break 878 // before the corresponding } or ]. 879 if (PreviousNonComment && 880 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 881 opensProtoMessageField(*PreviousNonComment, Style))) 882 State.Stack.back().BreakBeforeClosingBrace = true; 883 884 if (State.Stack.back().AvoidBinPacking) { 885 // If we are breaking after '(', '{', '<', this is not bin packing 886 // unless AllowAllParametersOfDeclarationOnNextLine is false or this is a 887 // dict/object literal. 888 if (!Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) || 889 (!Style.AllowAllParametersOfDeclarationOnNextLine && 890 State.Line->MustBeDeclaration) || 891 Previous.is(TT_DictLiteral)) 892 State.Stack.back().BreakBeforeParameter = true; 893 } 894 895 return Penalty; 896 } 897 898 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) { 899 if (!State.NextToken || !State.NextToken->Previous) 900 return 0; 901 FormatToken &Current = *State.NextToken; 902 const FormatToken &Previous = *Current.Previous; 903 // If we are continuing an expression, we want to use the continuation indent. 904 unsigned ContinuationIndent = 905 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 906 Style.ContinuationIndentWidth; 907 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 908 const FormatToken *NextNonComment = Previous.getNextNonComment(); 909 if (!NextNonComment) 910 NextNonComment = &Current; 911 912 // Java specific bits. 913 if (Style.Language == FormatStyle::LK_Java && 914 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) 915 return std::max(State.Stack.back().LastSpace, 916 State.Stack.back().Indent + Style.ContinuationIndentWidth); 917 918 if (NextNonComment->is(tok::l_brace) && NextNonComment->BlockKind == BK_Block) 919 return Current.NestingLevel == 0 ? State.FirstIndent 920 : State.Stack.back().Indent; 921 if ((Current.isOneOf(tok::r_brace, tok::r_square) || 922 (Current.is(tok::greater) && 923 (Style.Language == FormatStyle::LK_Proto || 924 Style.Language == FormatStyle::LK_TextProto))) && 925 State.Stack.size() > 1) { 926 if (Current.closesBlockOrBlockTypeList(Style)) 927 return State.Stack[State.Stack.size() - 2].NestedBlockIndent; 928 if (Current.MatchingParen && 929 Current.MatchingParen->BlockKind == BK_BracedInit) 930 return State.Stack[State.Stack.size() - 2].LastSpace; 931 return State.FirstIndent; 932 } 933 // Indent a closing parenthesis at the previous level if followed by a semi or 934 // opening brace. This allows indentations such as: 935 // foo( 936 // a, 937 // ); 938 // function foo( 939 // a, 940 // ) { 941 // code(); // 942 // } 943 if (Current.is(tok::r_paren) && State.Stack.size() > 1 && 944 (!Current.Next || Current.Next->isOneOf(tok::semi, tok::l_brace))) 945 return State.Stack[State.Stack.size() - 2].LastSpace; 946 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope()) 947 return State.Stack[State.Stack.size() - 2].LastSpace; 948 if (Current.is(tok::identifier) && Current.Next && 949 (Current.Next->is(TT_DictLiteral) || 950 ((Style.Language == FormatStyle::LK_Proto || 951 Style.Language == FormatStyle::LK_TextProto) && 952 Current.Next->isOneOf(tok::less, tok::l_brace)))) 953 return State.Stack.back().Indent; 954 if (NextNonComment->is(TT_ObjCStringLiteral) && 955 State.StartOfStringLiteral != 0) 956 return State.StartOfStringLiteral - 1; 957 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0) 958 return State.StartOfStringLiteral; 959 if (NextNonComment->is(tok::lessless) && 960 State.Stack.back().FirstLessLess != 0) 961 return State.Stack.back().FirstLessLess; 962 if (NextNonComment->isMemberAccess()) { 963 if (State.Stack.back().CallContinuation == 0) 964 return ContinuationIndent; 965 return State.Stack.back().CallContinuation; 966 } 967 if (State.Stack.back().QuestionColumn != 0 && 968 ((NextNonComment->is(tok::colon) && 969 NextNonComment->is(TT_ConditionalExpr)) || 970 Previous.is(TT_ConditionalExpr))) 971 return State.Stack.back().QuestionColumn; 972 if (Previous.is(tok::comma) && State.Stack.back().VariablePos != 0) 973 return State.Stack.back().VariablePos; 974 if ((PreviousNonComment && 975 (PreviousNonComment->ClosesTemplateDeclaration || 976 PreviousNonComment->isOneOf( 977 TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen, 978 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) || 979 (!Style.IndentWrappedFunctionNames && 980 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) 981 return std::max(State.Stack.back().LastSpace, State.Stack.back().Indent); 982 if (NextNonComment->is(TT_SelectorName)) { 983 if (!State.Stack.back().ObjCSelectorNameFound) { 984 unsigned MinIndent = State.Stack.back().Indent; 985 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) 986 MinIndent = std::max(MinIndent, 987 State.FirstIndent + Style.ContinuationIndentWidth); 988 // If LongestObjCSelectorName is 0, we are indenting the first 989 // part of an ObjC selector (or a selector component which is 990 // not colon-aligned due to block formatting). 991 // 992 // Otherwise, we are indenting a subsequent part of an ObjC 993 // selector which should be colon-aligned to the longest 994 // component of the ObjC selector. 995 // 996 // In either case, we want to respect Style.IndentWrappedFunctionNames. 997 return MinIndent + 998 std::max(NextNonComment->LongestObjCSelectorName, 999 NextNonComment->ColumnWidth) - 1000 NextNonComment->ColumnWidth; 1001 } 1002 if (!State.Stack.back().AlignColons) 1003 return State.Stack.back().Indent; 1004 if (State.Stack.back().ColonPos > NextNonComment->ColumnWidth) 1005 return State.Stack.back().ColonPos - NextNonComment->ColumnWidth; 1006 return State.Stack.back().Indent; 1007 } 1008 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr)) 1009 return State.Stack.back().ColonPos; 1010 if (NextNonComment->is(TT_ArraySubscriptLSquare)) { 1011 if (State.Stack.back().StartOfArraySubscripts != 0) 1012 return State.Stack.back().StartOfArraySubscripts; 1013 return ContinuationIndent; 1014 } 1015 1016 // This ensure that we correctly format ObjC methods calls without inputs, 1017 // i.e. where the last element isn't selector like: [callee method]; 1018 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 && 1019 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) 1020 return State.Stack.back().Indent; 1021 1022 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) || 1023 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) 1024 return ContinuationIndent; 1025 if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 1026 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) 1027 return ContinuationIndent; 1028 if (NextNonComment->is(TT_CtorInitializerComma)) 1029 return State.Stack.back().Indent; 1030 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && 1031 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) 1032 return State.Stack.back().Indent; 1033 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) && 1034 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) 1035 return State.Stack.back().Indent; 1036 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon, 1037 TT_InheritanceComma)) 1038 return State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1039 if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() && 1040 !Current.isOneOf(tok::colon, tok::comment)) 1041 return ContinuationIndent; 1042 if (Current.is(TT_ProtoExtensionLSquare)) 1043 return State.Stack.back().Indent; 1044 if (State.Stack.back().Indent == State.FirstIndent && PreviousNonComment && 1045 PreviousNonComment->isNot(tok::r_brace)) 1046 // Ensure that we fall back to the continuation indent width instead of 1047 // just flushing continuations left. 1048 return State.Stack.back().Indent + Style.ContinuationIndentWidth; 1049 return State.Stack.back().Indent; 1050 } 1051 1052 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State, 1053 bool DryRun, bool Newline) { 1054 assert(State.Stack.size()); 1055 const FormatToken &Current = *State.NextToken; 1056 1057 if (Current.isOneOf(tok::comma, TT_BinaryOperator)) 1058 State.Stack.back().NoLineBreakInOperand = false; 1059 if (Current.is(TT_InheritanceColon)) 1060 State.Stack.back().AvoidBinPacking = true; 1061 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) { 1062 if (State.Stack.back().FirstLessLess == 0) 1063 State.Stack.back().FirstLessLess = State.Column; 1064 else 1065 State.Stack.back().LastOperatorWrapped = Newline; 1066 } 1067 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless)) 1068 State.Stack.back().LastOperatorWrapped = Newline; 1069 if (Current.is(TT_ConditionalExpr) && Current.Previous && 1070 !Current.Previous->is(TT_ConditionalExpr)) 1071 State.Stack.back().LastOperatorWrapped = Newline; 1072 if (Current.is(TT_ArraySubscriptLSquare) && 1073 State.Stack.back().StartOfArraySubscripts == 0) 1074 State.Stack.back().StartOfArraySubscripts = State.Column; 1075 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question)) 1076 State.Stack.back().QuestionColumn = State.Column; 1077 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) { 1078 const FormatToken *Previous = Current.Previous; 1079 while (Previous && Previous->isTrailingComment()) 1080 Previous = Previous->Previous; 1081 if (Previous && Previous->is(tok::question)) 1082 State.Stack.back().QuestionColumn = State.Column; 1083 } 1084 if (!Current.opensScope() && !Current.closesScope() && 1085 !Current.is(TT_PointerOrReference)) 1086 State.LowestLevelOnLine = 1087 std::min(State.LowestLevelOnLine, Current.NestingLevel); 1088 if (Current.isMemberAccess()) 1089 State.Stack.back().StartOfFunctionCall = 1090 !Current.NextOperator ? 0 : State.Column; 1091 if (Current.is(TT_SelectorName)) 1092 State.Stack.back().ObjCSelectorNameFound = true; 1093 if (Current.is(TT_CtorInitializerColon) && 1094 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) { 1095 // Indent 2 from the column, so: 1096 // SomeClass::SomeClass() 1097 // : First(...), ... 1098 // Next(...) 1099 // ^ line up here. 1100 State.Stack.back().Indent = 1101 State.Column + 1102 (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma 1103 ? 0 1104 : 2); 1105 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1106 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 1107 State.Stack.back().AvoidBinPacking = true; 1108 State.Stack.back().BreakBeforeParameter = false; 1109 } 1110 if (Current.is(TT_CtorInitializerColon) && 1111 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { 1112 State.Stack.back().Indent = 1113 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1114 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1115 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 1116 State.Stack.back().AvoidBinPacking = true; 1117 } 1118 if (Current.is(TT_InheritanceColon)) 1119 State.Stack.back().Indent = 1120 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1121 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline) 1122 State.Stack.back().NestedBlockIndent = 1123 State.Column + Current.ColumnWidth + 1; 1124 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow)) 1125 State.Stack.back().LastSpace = State.Column; 1126 1127 // Insert scopes created by fake parenthesis. 1128 const FormatToken *Previous = Current.getPreviousNonComment(); 1129 1130 // Add special behavior to support a format commonly used for JavaScript 1131 // closures: 1132 // SomeFunction(function() { 1133 // foo(); 1134 // bar(); 1135 // }, a, b, c); 1136 if (Current.isNot(tok::comment) && Previous && 1137 Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) && 1138 !Previous->is(TT_DictLiteral) && State.Stack.size() > 1) { 1139 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline) 1140 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 1141 State.Stack[i].NoLineBreak = true; 1142 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false; 1143 } 1144 if (Previous && 1145 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) || 1146 Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr)) && 1147 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 1148 State.Stack.back().NestedBlockInlined = 1149 !Newline && 1150 (Previous->isNot(tok::l_paren) || Previous->ParameterCount > 1); 1151 } 1152 1153 moveStatePastFakeLParens(State, Newline); 1154 moveStatePastScopeCloser(State); 1155 bool AllowBreak = !State.Stack.back().NoLineBreak && 1156 !State.Stack.back().NoLineBreakInOperand; 1157 moveStatePastScopeOpener(State, Newline); 1158 moveStatePastFakeRParens(State); 1159 1160 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0) 1161 State.StartOfStringLiteral = State.Column + 1; 1162 else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) 1163 State.StartOfStringLiteral = State.Column; 1164 else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) && 1165 !Current.isStringLiteral()) 1166 State.StartOfStringLiteral = 0; 1167 1168 State.Column += Current.ColumnWidth; 1169 State.NextToken = State.NextToken->Next; 1170 1171 unsigned Penalty = 1172 handleEndOfLine(Current, State, DryRun, AllowBreak); 1173 1174 if (Current.Role) 1175 Current.Role->formatFromToken(State, this, DryRun); 1176 // If the previous has a special role, let it consume tokens as appropriate. 1177 // It is necessary to start at the previous token for the only implemented 1178 // role (comma separated list). That way, the decision whether or not to break 1179 // after the "{" is already done and both options are tried and evaluated. 1180 // FIXME: This is ugly, find a better way. 1181 if (Previous && Previous->Role) 1182 Penalty += Previous->Role->formatAfterToken(State, this, DryRun); 1183 1184 return Penalty; 1185 } 1186 1187 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State, 1188 bool Newline) { 1189 const FormatToken &Current = *State.NextToken; 1190 const FormatToken *Previous = Current.getPreviousNonComment(); 1191 1192 // Don't add extra indentation for the first fake parenthesis after 1193 // 'return', assignments or opening <({[. The indentation for these cases 1194 // is special cased. 1195 bool SkipFirstExtraIndent = 1196 (Previous && (Previous->opensScope() || 1197 Previous->isOneOf(tok::semi, tok::kw_return) || 1198 (Previous->getPrecedence() == prec::Assignment && 1199 Style.AlignOperands) || 1200 Previous->is(TT_ObjCMethodExpr))); 1201 for (SmallVectorImpl<prec::Level>::const_reverse_iterator 1202 I = Current.FakeLParens.rbegin(), 1203 E = Current.FakeLParens.rend(); 1204 I != E; ++I) { 1205 ParenState NewParenState = State.Stack.back(); 1206 NewParenState.Tok = nullptr; 1207 NewParenState.ContainsLineBreak = false; 1208 NewParenState.LastOperatorWrapped = true; 1209 NewParenState.NoLineBreak = 1210 NewParenState.NoLineBreak || State.Stack.back().NoLineBreakInOperand; 1211 1212 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists. 1213 if (*I > prec::Comma) 1214 NewParenState.AvoidBinPacking = false; 1215 1216 // Indent from 'LastSpace' unless these are fake parentheses encapsulating 1217 // a builder type call after 'return' or, if the alignment after opening 1218 // brackets is disabled. 1219 if (!Current.isTrailingComment() && 1220 (Style.AlignOperands || *I < prec::Assignment) && 1221 (!Previous || Previous->isNot(tok::kw_return) || 1222 (Style.Language != FormatStyle::LK_Java && *I > 0)) && 1223 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign || 1224 *I != prec::Comma || Current.NestingLevel == 0)) 1225 NewParenState.Indent = 1226 std::max(std::max(State.Column, NewParenState.Indent), 1227 State.Stack.back().LastSpace); 1228 1229 // Do not indent relative to the fake parentheses inserted for "." or "->". 1230 // This is a special case to make the following to statements consistent: 1231 // OuterFunction(InnerFunctionCall( // break 1232 // ParameterToInnerFunction)); 1233 // OuterFunction(SomeObject.InnerFunctionCall( // break 1234 // ParameterToInnerFunction)); 1235 if (*I > prec::Unknown) 1236 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column); 1237 if (*I != prec::Conditional && !Current.is(TT_UnaryOperator) && 1238 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 1239 NewParenState.StartOfFunctionCall = State.Column; 1240 1241 // Always indent conditional expressions. Never indent expression where 1242 // the 'operator' is ',', ';' or an assignment (i.e. *I <= 1243 // prec::Assignment) as those have different indentation rules. Indent 1244 // other expression, unless the indentation needs to be skipped. 1245 if (*I == prec::Conditional || 1246 (!SkipFirstExtraIndent && *I > prec::Assignment && 1247 !Current.isTrailingComment())) 1248 NewParenState.Indent += Style.ContinuationIndentWidth; 1249 if ((Previous && !Previous->opensScope()) || *I != prec::Comma) 1250 NewParenState.BreakBeforeParameter = false; 1251 State.Stack.push_back(NewParenState); 1252 SkipFirstExtraIndent = false; 1253 } 1254 } 1255 1256 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) { 1257 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) { 1258 unsigned VariablePos = State.Stack.back().VariablePos; 1259 if (State.Stack.size() == 1) { 1260 // Do not pop the last element. 1261 break; 1262 } 1263 State.Stack.pop_back(); 1264 State.Stack.back().VariablePos = VariablePos; 1265 } 1266 } 1267 1268 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State, 1269 bool Newline) { 1270 const FormatToken &Current = *State.NextToken; 1271 if (!Current.opensScope()) 1272 return; 1273 1274 if (Current.MatchingParen && Current.BlockKind == BK_Block) { 1275 moveStateToNewBlock(State); 1276 return; 1277 } 1278 1279 unsigned NewIndent; 1280 unsigned LastSpace = State.Stack.back().LastSpace; 1281 bool AvoidBinPacking; 1282 bool BreakBeforeParameter = false; 1283 unsigned NestedBlockIndent = std::max(State.Stack.back().StartOfFunctionCall, 1284 State.Stack.back().NestedBlockIndent); 1285 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 1286 opensProtoMessageField(Current, Style)) { 1287 if (Current.opensBlockOrBlockTypeList(Style)) { 1288 NewIndent = Style.IndentWidth + 1289 std::min(State.Column, State.Stack.back().NestedBlockIndent); 1290 } else { 1291 NewIndent = State.Stack.back().LastSpace + Style.ContinuationIndentWidth; 1292 } 1293 const FormatToken *NextNoComment = Current.getNextNonComment(); 1294 bool EndsInComma = Current.MatchingParen && 1295 Current.MatchingParen->Previous && 1296 Current.MatchingParen->Previous->is(tok::comma); 1297 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) || 1298 Style.Language == FormatStyle::LK_Proto || 1299 Style.Language == FormatStyle::LK_TextProto || 1300 !Style.BinPackArguments || 1301 (NextNoComment && 1302 NextNoComment->isOneOf(TT_DesignatedInitializerPeriod, 1303 TT_DesignatedInitializerLSquare)); 1304 BreakBeforeParameter = EndsInComma; 1305 if (Current.ParameterCount > 1) 1306 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1); 1307 } else { 1308 NewIndent = Style.ContinuationIndentWidth + 1309 std::max(State.Stack.back().LastSpace, 1310 State.Stack.back().StartOfFunctionCall); 1311 1312 // Ensure that different different brackets force relative alignment, e.g.: 1313 // void SomeFunction(vector< // break 1314 // int> v); 1315 // FIXME: We likely want to do this for more combinations of brackets. 1316 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) { 1317 NewIndent = std::max(NewIndent, State.Stack.back().Indent); 1318 LastSpace = std::max(LastSpace, State.Stack.back().Indent); 1319 } 1320 1321 bool EndsInComma = 1322 Current.MatchingParen && 1323 Current.MatchingParen->getPreviousNonComment() && 1324 Current.MatchingParen->getPreviousNonComment()->is(tok::comma); 1325 1326 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters 1327 // for backwards compatibility. 1328 bool ObjCBinPackProtocolList = 1329 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto && 1330 Style.BinPackParameters) || 1331 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always; 1332 1333 bool BinPackDeclaration = 1334 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) || 1335 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList); 1336 1337 AvoidBinPacking = 1338 (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) || 1339 (State.Line->MustBeDeclaration && !BinPackDeclaration) || 1340 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) || 1341 (Style.ExperimentalAutoDetectBinPacking && 1342 (Current.PackingKind == PPK_OnePerLine || 1343 (!BinPackInconclusiveFunctions && 1344 Current.PackingKind == PPK_Inconclusive))); 1345 1346 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen) { 1347 if (Style.ColumnLimit) { 1348 // If this '[' opens an ObjC call, determine whether all parameters fit 1349 // into one line and put one per line if they don't. 1350 if (getLengthToMatchingParen(Current, State.Stack) + State.Column > 1351 getColumnLimit(State)) 1352 BreakBeforeParameter = true; 1353 } else { 1354 // For ColumnLimit = 0, we have to figure out whether there is or has to 1355 // be a line break within this call. 1356 for (const FormatToken *Tok = &Current; 1357 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) { 1358 if (Tok->MustBreakBefore || 1359 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) { 1360 BreakBeforeParameter = true; 1361 break; 1362 } 1363 } 1364 } 1365 } 1366 1367 if (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) 1368 BreakBeforeParameter = true; 1369 } 1370 // Generally inherit NoLineBreak from the current scope to nested scope. 1371 // However, don't do this for non-empty nested blocks, dict literals and 1372 // array literals as these follow different indentation rules. 1373 bool NoLineBreak = 1374 Current.Children.empty() && 1375 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) && 1376 (State.Stack.back().NoLineBreak || 1377 State.Stack.back().NoLineBreakInOperand || 1378 (Current.is(TT_TemplateOpener) && 1379 State.Stack.back().ContainsUnwrappedBuilder)); 1380 State.Stack.push_back( 1381 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak)); 1382 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1383 State.Stack.back().BreakBeforeParameter = BreakBeforeParameter; 1384 State.Stack.back().HasMultipleNestedBlocks = Current.BlockParameterCount > 1; 1385 State.Stack.back().IsInsideObjCArrayLiteral = 1386 Current.is(TT_ArrayInitializerLSquare) && Current.Previous && 1387 Current.Previous->is(tok::at); 1388 } 1389 1390 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) { 1391 const FormatToken &Current = *State.NextToken; 1392 if (!Current.closesScope()) 1393 return; 1394 1395 // If we encounter a closing ), ], } or >, we can remove a level from our 1396 // stacks. 1397 if (State.Stack.size() > 1 && 1398 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) || 1399 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) || 1400 State.NextToken->is(TT_TemplateCloser) || 1401 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) 1402 State.Stack.pop_back(); 1403 1404 // Reevaluate whether ObjC message arguments fit into one line. 1405 // If a receiver spans multiple lines, e.g.: 1406 // [[object block:^{ 1407 // return 42; 1408 // }] a:42 b:42]; 1409 // BreakBeforeParameter is calculated based on an incorrect assumption 1410 // (it is checked whether the whole expression fits into one line without 1411 // considering a line break inside a message receiver). 1412 // We check whether arguements fit after receiver scope closer (into the same 1413 // line). 1414 if (State.Stack.back().BreakBeforeParameter && Current.MatchingParen && 1415 Current.MatchingParen->Previous) { 1416 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous; 1417 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) && 1418 CurrentScopeOpener.MatchingParen) { 1419 int NecessarySpaceInLine = 1420 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) + 1421 CurrentScopeOpener.TotalLength - Current.TotalLength - 1; 1422 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <= 1423 Style.ColumnLimit) 1424 State.Stack.back().BreakBeforeParameter = false; 1425 } 1426 } 1427 1428 if (Current.is(tok::r_square)) { 1429 // If this ends the array subscript expr, reset the corresponding value. 1430 const FormatToken *NextNonComment = Current.getNextNonComment(); 1431 if (NextNonComment && NextNonComment->isNot(tok::l_square)) 1432 State.Stack.back().StartOfArraySubscripts = 0; 1433 } 1434 } 1435 1436 void ContinuationIndenter::moveStateToNewBlock(LineState &State) { 1437 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent; 1438 // ObjC block sometimes follow special indentation rules. 1439 unsigned NewIndent = 1440 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace) 1441 ? Style.ObjCBlockIndentWidth 1442 : Style.IndentWidth); 1443 State.Stack.push_back(ParenState(State.NextToken, NewIndent, 1444 State.Stack.back().LastSpace, 1445 /*AvoidBinPacking=*/true, 1446 /*NoLineBreak=*/false)); 1447 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1448 State.Stack.back().BreakBeforeParameter = true; 1449 } 1450 1451 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn, 1452 unsigned TabWidth, 1453 encoding::Encoding Encoding) { 1454 size_t LastNewlinePos = Text.find_last_of("\n"); 1455 if (LastNewlinePos == StringRef::npos) { 1456 return StartColumn + 1457 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding); 1458 } else { 1459 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos), 1460 /*StartColumn=*/0, TabWidth, Encoding); 1461 } 1462 } 1463 1464 unsigned ContinuationIndenter::reformatRawStringLiteral( 1465 const FormatToken &Current, LineState &State, 1466 const FormatStyle &RawStringStyle, bool DryRun) { 1467 unsigned StartColumn = State.Column - Current.ColumnWidth; 1468 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText); 1469 StringRef NewDelimiter = 1470 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language); 1471 if (NewDelimiter.empty() || OldDelimiter.empty()) 1472 NewDelimiter = OldDelimiter; 1473 // The text of a raw string is between the leading 'R"delimiter(' and the 1474 // trailing 'delimiter)"'. 1475 unsigned OldPrefixSize = 3 + OldDelimiter.size(); 1476 unsigned OldSuffixSize = 2 + OldDelimiter.size(); 1477 // We create a virtual text environment which expects a null-terminated 1478 // string, so we cannot use StringRef. 1479 std::string RawText = 1480 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize); 1481 if (NewDelimiter != OldDelimiter) { 1482 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the 1483 // raw string. 1484 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str(); 1485 if (StringRef(RawText).contains(CanonicalDelimiterSuffix)) 1486 NewDelimiter = OldDelimiter; 1487 } 1488 1489 unsigned NewPrefixSize = 3 + NewDelimiter.size(); 1490 unsigned NewSuffixSize = 2 + NewDelimiter.size(); 1491 1492 // The first start column is the column the raw text starts after formatting. 1493 unsigned FirstStartColumn = StartColumn + NewPrefixSize; 1494 1495 // The next start column is the intended indentation a line break inside 1496 // the raw string at level 0. It is determined by the following rules: 1497 // - if the content starts on newline, it is one level more than the current 1498 // indent, and 1499 // - if the content does not start on a newline, it is the first start 1500 // column. 1501 // These rules have the advantage that the formatted content both does not 1502 // violate the rectangle rule and visually flows within the surrounding 1503 // source. 1504 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n'; 1505 unsigned NextStartColumn = 1506 ContentStartsOnNewline 1507 ? State.Stack.back().NestedBlockIndent + Style.IndentWidth 1508 : FirstStartColumn; 1509 1510 // The last start column is the column the raw string suffix starts if it is 1511 // put on a newline. 1512 // The last start column is the intended indentation of the raw string postfix 1513 // if it is put on a newline. It is determined by the following rules: 1514 // - if the raw string prefix starts on a newline, it is the column where 1515 // that raw string prefix starts, and 1516 // - if the raw string prefix does not start on a newline, it is the current 1517 // indent. 1518 unsigned LastStartColumn = Current.NewlinesBefore 1519 ? FirstStartColumn - NewPrefixSize 1520 : State.Stack.back().NestedBlockIndent; 1521 1522 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat( 1523 RawStringStyle, RawText, {tooling::Range(0, RawText.size())}, 1524 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>", 1525 /*Status=*/nullptr); 1526 1527 auto NewCode = applyAllReplacements(RawText, Fixes.first); 1528 tooling::Replacements NoFixes; 1529 if (!NewCode) { 1530 State.Column += Current.ColumnWidth; 1531 return 0; 1532 } 1533 if (!DryRun) { 1534 if (NewDelimiter != OldDelimiter) { 1535 // In 'R"delimiter(...', the delimiter starts 2 characters after the start 1536 // of the token. 1537 SourceLocation PrefixDelimiterStart = 1538 Current.Tok.getLocation().getLocWithOffset(2); 1539 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement( 1540 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1541 if (PrefixErr) { 1542 llvm::errs() 1543 << "Failed to update the prefix delimiter of a raw string: " 1544 << llvm::toString(std::move(PrefixErr)) << "\n"; 1545 } 1546 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at 1547 // position length - 1 - |delimiter|. 1548 SourceLocation SuffixDelimiterStart = 1549 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() - 1550 1 - OldDelimiter.size()); 1551 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement( 1552 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1553 if (SuffixErr) { 1554 llvm::errs() 1555 << "Failed to update the suffix delimiter of a raw string: " 1556 << llvm::toString(std::move(SuffixErr)) << "\n"; 1557 } 1558 } 1559 SourceLocation OriginLoc = 1560 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize); 1561 for (const tooling::Replacement &Fix : Fixes.first) { 1562 auto Err = Whitespaces.addReplacement(tooling::Replacement( 1563 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()), 1564 Fix.getLength(), Fix.getReplacementText())); 1565 if (Err) { 1566 llvm::errs() << "Failed to reformat raw string: " 1567 << llvm::toString(std::move(Err)) << "\n"; 1568 } 1569 } 1570 } 1571 unsigned RawLastLineEndColumn = getLastLineEndColumn( 1572 *NewCode, FirstStartColumn, Style.TabWidth, Encoding); 1573 State.Column = RawLastLineEndColumn + NewSuffixSize; 1574 // Since we're updating the column to after the raw string literal here, we 1575 // have to manually add the penalty for the prefix R"delim( over the column 1576 // limit. 1577 unsigned PrefixExcessCharacters = 1578 StartColumn + NewPrefixSize > Style.ColumnLimit ? 1579 StartColumn + NewPrefixSize - Style.ColumnLimit : 0; 1580 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter; 1581 } 1582 1583 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current, 1584 LineState &State) { 1585 // Break before further function parameters on all levels. 1586 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 1587 State.Stack[i].BreakBeforeParameter = true; 1588 1589 unsigned ColumnsUsed = State.Column; 1590 // We can only affect layout of the first and the last line, so the penalty 1591 // for all other lines is constant, and we ignore it. 1592 State.Column = Current.LastLineColumnWidth; 1593 1594 if (ColumnsUsed > getColumnLimit(State)) 1595 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State)); 1596 return 0; 1597 } 1598 1599 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current, 1600 LineState &State, bool DryRun, 1601 bool AllowBreak) { 1602 unsigned Penalty = 0; 1603 // Compute the raw string style to use in case this is a raw string literal 1604 // that can be reformatted. 1605 auto RawStringStyle = getRawStringStyle(Current, State); 1606 if (RawStringStyle && !Current.Finalized) { 1607 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun); 1608 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) { 1609 // Don't break multi-line tokens other than block comments and raw string 1610 // literals. Instead, just update the state. 1611 Penalty = addMultilineToken(Current, State); 1612 } else if (State.Line->Type != LT_ImportStatement) { 1613 // We generally don't break import statements. 1614 LineState OriginalState = State; 1615 1616 // Whether we force the reflowing algorithm to stay strictly within the 1617 // column limit. 1618 bool Strict = false; 1619 // Whether the first non-strict attempt at reflowing did intentionally 1620 // exceed the column limit. 1621 bool Exceeded = false; 1622 std::tie(Penalty, Exceeded) = breakProtrudingToken( 1623 Current, State, AllowBreak, /*DryRun=*/true, Strict); 1624 if (Exceeded) { 1625 // If non-strict reflowing exceeds the column limit, try whether strict 1626 // reflowing leads to an overall lower penalty. 1627 LineState StrictState = OriginalState; 1628 unsigned StrictPenalty = 1629 breakProtrudingToken(Current, StrictState, AllowBreak, 1630 /*DryRun=*/true, /*Strict=*/true) 1631 .first; 1632 Strict = StrictPenalty <= Penalty; 1633 if (Strict) { 1634 Penalty = StrictPenalty; 1635 State = StrictState; 1636 } 1637 } 1638 if (!DryRun) { 1639 // If we're not in dry-run mode, apply the changes with the decision on 1640 // strictness made above. 1641 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false, 1642 Strict); 1643 } 1644 } 1645 if (State.Column > getColumnLimit(State)) { 1646 unsigned ExcessCharacters = State.Column - getColumnLimit(State); 1647 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters; 1648 } 1649 return Penalty; 1650 } 1651 1652 // Returns the enclosing function name of a token, or the empty string if not 1653 // found. 1654 static StringRef getEnclosingFunctionName(const FormatToken &Current) { 1655 // Look for: 'function(' or 'function<templates>(' before Current. 1656 auto Tok = Current.getPreviousNonComment(); 1657 if (!Tok || !Tok->is(tok::l_paren)) 1658 return ""; 1659 Tok = Tok->getPreviousNonComment(); 1660 if (!Tok) 1661 return ""; 1662 if (Tok->is(TT_TemplateCloser)) { 1663 Tok = Tok->MatchingParen; 1664 if (Tok) 1665 Tok = Tok->getPreviousNonComment(); 1666 } 1667 if (!Tok || !Tok->is(tok::identifier)) 1668 return ""; 1669 return Tok->TokenText; 1670 } 1671 1672 llvm::Optional<FormatStyle> 1673 ContinuationIndenter::getRawStringStyle(const FormatToken &Current, 1674 const LineState &State) { 1675 if (!Current.isStringLiteral()) 1676 return None; 1677 auto Delimiter = getRawStringDelimiter(Current.TokenText); 1678 if (!Delimiter) 1679 return None; 1680 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter); 1681 if (!RawStringStyle && Delimiter->empty()) 1682 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle( 1683 getEnclosingFunctionName(Current)); 1684 if (!RawStringStyle) 1685 return None; 1686 RawStringStyle->ColumnLimit = getColumnLimit(State); 1687 return RawStringStyle; 1688 } 1689 1690 std::unique_ptr<BreakableToken> ContinuationIndenter::createBreakableToken( 1691 const FormatToken &Current, LineState &State, bool AllowBreak) { 1692 unsigned StartColumn = State.Column - Current.ColumnWidth; 1693 if (Current.isStringLiteral()) { 1694 // FIXME: String literal breaking is currently disabled for Java and JS, as 1695 // it requires strings to be merged using "+" which we don't support. 1696 if (Style.Language == FormatStyle::LK_Java || 1697 Style.Language == FormatStyle::LK_JavaScript || 1698 !Style.BreakStringLiterals || 1699 !AllowBreak) 1700 return nullptr; 1701 1702 // Don't break string literals inside preprocessor directives (except for 1703 // #define directives, as their contents are stored in separate lines and 1704 // are not affected by this check). 1705 // This way we avoid breaking code with line directives and unknown 1706 // preprocessor directives that contain long string literals. 1707 if (State.Line->Type == LT_PreprocessorDirective) 1708 return nullptr; 1709 // Exempts unterminated string literals from line breaking. The user will 1710 // likely want to terminate the string before any line breaking is done. 1711 if (Current.IsUnterminatedLiteral) 1712 return nullptr; 1713 // Don't break string literals inside Objective-C array literals (doing so 1714 // raises the warning -Wobjc-string-concatenation). 1715 if (State.Stack.back().IsInsideObjCArrayLiteral) { 1716 return nullptr; 1717 } 1718 1719 StringRef Text = Current.TokenText; 1720 StringRef Prefix; 1721 StringRef Postfix; 1722 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'. 1723 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to 1724 // reduce the overhead) for each FormatToken, which is a string, so that we 1725 // don't run multiple checks here on the hot path. 1726 if ((Text.endswith(Postfix = "\"") && 1727 (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") || 1728 Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") || 1729 Text.startswith(Prefix = "u8\"") || 1730 Text.startswith(Prefix = "L\""))) || 1731 (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) { 1732 // We need this to address the case where there is an unbreakable tail 1733 // only if certain other formatting decisions have been taken. The 1734 // UnbreakableTailLength of Current is an overapproximation is that case 1735 // and we need to be correct here. 1736 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State)) 1737 ? 0 1738 : Current.UnbreakableTailLength; 1739 return llvm::make_unique<BreakableStringLiteral>( 1740 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength, 1741 State.Line->InPPDirective, Encoding, Style); 1742 } 1743 } else if (Current.is(TT_BlockComment)) { 1744 if (!Style.ReflowComments || 1745 // If a comment token switches formatting, like 1746 // /* clang-format on */, we don't want to break it further, 1747 // but we may still want to adjust its indentation. 1748 switchesFormatting(Current)) { 1749 return nullptr; 1750 } 1751 return llvm::make_unique<BreakableBlockComment>( 1752 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1753 State.Line->InPPDirective, Encoding, Style); 1754 } else if (Current.is(TT_LineComment) && 1755 (Current.Previous == nullptr || 1756 Current.Previous->isNot(TT_ImplicitStringLiteral))) { 1757 if (!Style.ReflowComments || 1758 CommentPragmasRegex.match(Current.TokenText.substr(2)) || 1759 switchesFormatting(Current)) 1760 return nullptr; 1761 return llvm::make_unique<BreakableLineCommentSection>( 1762 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1763 /*InPPDirective=*/false, Encoding, Style); 1764 } 1765 return nullptr; 1766 } 1767 1768 std::pair<unsigned, bool> 1769 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current, 1770 LineState &State, bool AllowBreak, 1771 bool DryRun, bool Strict) { 1772 std::unique_ptr<const BreakableToken> Token = 1773 createBreakableToken(Current, State, AllowBreak); 1774 if (!Token) 1775 return {0, false}; 1776 assert(Token->getLineCount() > 0); 1777 unsigned ColumnLimit = getColumnLimit(State); 1778 if (Current.is(TT_LineComment)) { 1779 // We don't insert backslashes when breaking line comments. 1780 ColumnLimit = Style.ColumnLimit; 1781 } 1782 if (Current.UnbreakableTailLength >= ColumnLimit) 1783 return {0, false}; 1784 // ColumnWidth was already accounted into State.Column before calling 1785 // breakProtrudingToken. 1786 unsigned StartColumn = State.Column - Current.ColumnWidth; 1787 unsigned NewBreakPenalty = Current.isStringLiteral() 1788 ? Style.PenaltyBreakString 1789 : Style.PenaltyBreakComment; 1790 // Stores whether we intentionally decide to let a line exceed the column 1791 // limit. 1792 bool Exceeded = false; 1793 // Stores whether we introduce a break anywhere in the token. 1794 bool BreakInserted = Token->introducesBreakBeforeToken(); 1795 // Store whether we inserted a new line break at the end of the previous 1796 // logical line. 1797 bool NewBreakBefore = false; 1798 // We use a conservative reflowing strategy. Reflow starts after a line is 1799 // broken or the corresponding whitespace compressed. Reflow ends as soon as a 1800 // line that doesn't get reflown with the previous line is reached. 1801 bool Reflow = false; 1802 // Keep track of where we are in the token: 1803 // Where we are in the content of the current logical line. 1804 unsigned TailOffset = 0; 1805 // The column number we're currently at. 1806 unsigned ContentStartColumn = 1807 Token->getContentStartColumn(0, /*Break=*/false); 1808 // The number of columns left in the current logical line after TailOffset. 1809 unsigned RemainingTokenColumns = 1810 Token->getRemainingLength(0, TailOffset, ContentStartColumn); 1811 // Adapt the start of the token, for example indent. 1812 if (!DryRun) 1813 Token->adaptStartOfLine(0, Whitespaces); 1814 1815 unsigned ContentIndent = 0; 1816 unsigned Penalty = 0; 1817 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column " 1818 << StartColumn << ".\n"); 1819 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount(); 1820 LineIndex != EndIndex; ++LineIndex) { 1821 LLVM_DEBUG(llvm::dbgs() 1822 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n"); 1823 NewBreakBefore = false; 1824 // If we did reflow the previous line, we'll try reflowing again. Otherwise 1825 // we'll start reflowing if the current line is broken or whitespace is 1826 // compressed. 1827 bool TryReflow = Reflow; 1828 // Break the current token until we can fit the rest of the line. 1829 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 1830 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: " 1831 << (ContentStartColumn + RemainingTokenColumns) 1832 << ", space: " << ColumnLimit 1833 << ", reflown prefix: " << ContentStartColumn 1834 << ", offset in line: " << TailOffset << "\n"); 1835 // If the current token doesn't fit, find the latest possible split in the 1836 // current line so that breaking at it will be under the column limit. 1837 // FIXME: Use the earliest possible split while reflowing to correctly 1838 // compress whitespace within a line. 1839 BreakableToken::Split Split = 1840 Token->getSplit(LineIndex, TailOffset, ColumnLimit, 1841 ContentStartColumn, CommentPragmasRegex); 1842 if (Split.first == StringRef::npos) { 1843 // No break opportunity - update the penalty and continue with the next 1844 // logical line. 1845 if (LineIndex < EndIndex - 1) 1846 // The last line's penalty is handled in addNextStateToQueue() or when 1847 // calling replaceWhitespaceAfterLastLine below. 1848 Penalty += Style.PenaltyExcessCharacter * 1849 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 1850 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n"); 1851 break; 1852 } 1853 assert(Split.first != 0); 1854 1855 if (Token->supportsReflow()) { 1856 // Check whether the next natural split point after the current one can 1857 // still fit the line, either because we can compress away whitespace, 1858 // or because the penalty the excess characters introduce is lower than 1859 // the break penalty. 1860 // We only do this for tokens that support reflowing, and thus allow us 1861 // to change the whitespace arbitrarily (e.g. comments). 1862 // Other tokens, like string literals, can be broken on arbitrary 1863 // positions. 1864 1865 // First, compute the columns from TailOffset to the next possible split 1866 // position. 1867 // For example: 1868 // ColumnLimit: | 1869 // // Some text that breaks 1870 // ^ tail offset 1871 // ^-- split 1872 // ^-------- to split columns 1873 // ^--- next split 1874 // ^--------------- to next split columns 1875 unsigned ToSplitColumns = Token->getRangeLength( 1876 LineIndex, TailOffset, Split.first, ContentStartColumn); 1877 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n"); 1878 1879 BreakableToken::Split NextSplit = Token->getSplit( 1880 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit, 1881 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex); 1882 // Compute the columns necessary to fit the next non-breakable sequence 1883 // into the current line. 1884 unsigned ToNextSplitColumns = 0; 1885 if (NextSplit.first == StringRef::npos) { 1886 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset, 1887 ContentStartColumn); 1888 } else { 1889 ToNextSplitColumns = Token->getRangeLength( 1890 LineIndex, TailOffset, 1891 Split.first + Split.second + NextSplit.first, ContentStartColumn); 1892 } 1893 // Compress the whitespace between the break and the start of the next 1894 // unbreakable sequence. 1895 ToNextSplitColumns = 1896 Token->getLengthAfterCompression(ToNextSplitColumns, Split); 1897 LLVM_DEBUG(llvm::dbgs() 1898 << " ContentStartColumn: " << ContentStartColumn << "\n"); 1899 LLVM_DEBUG(llvm::dbgs() 1900 << " ToNextSplit: " << ToNextSplitColumns << "\n"); 1901 // If the whitespace compression makes us fit, continue on the current 1902 // line. 1903 bool ContinueOnLine = 1904 ContentStartColumn + ToNextSplitColumns <= ColumnLimit; 1905 unsigned ExcessCharactersPenalty = 0; 1906 if (!ContinueOnLine && !Strict) { 1907 // Similarly, if the excess characters' penalty is lower than the 1908 // penalty of introducing a new break, continue on the current line. 1909 ExcessCharactersPenalty = 1910 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) * 1911 Style.PenaltyExcessCharacter; 1912 LLVM_DEBUG(llvm::dbgs() 1913 << " Penalty excess: " << ExcessCharactersPenalty 1914 << "\n break : " << NewBreakPenalty << "\n"); 1915 if (ExcessCharactersPenalty < NewBreakPenalty) { 1916 Exceeded = true; 1917 ContinueOnLine = true; 1918 } 1919 } 1920 if (ContinueOnLine) { 1921 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n"); 1922 // The current line fits after compressing the whitespace - reflow 1923 // the next line into it if possible. 1924 TryReflow = true; 1925 if (!DryRun) 1926 Token->compressWhitespace(LineIndex, TailOffset, Split, 1927 Whitespaces); 1928 // When we continue on the same line, leave one space between content. 1929 ContentStartColumn += ToSplitColumns + 1; 1930 Penalty += ExcessCharactersPenalty; 1931 TailOffset += Split.first + Split.second; 1932 RemainingTokenColumns = Token->getRemainingLength( 1933 LineIndex, TailOffset, ContentStartColumn); 1934 continue; 1935 } 1936 } 1937 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n"); 1938 // Update the ContentIndent only if the current line was not reflown with 1939 // the previous line, since in that case the previous line should still 1940 // determine the ContentIndent. Also never intent the last line. 1941 if (!Reflow) 1942 ContentIndent = Token->getContentIndent(LineIndex); 1943 LLVM_DEBUG(llvm::dbgs() 1944 << " ContentIndent: " << ContentIndent << "\n"); 1945 ContentStartColumn = ContentIndent + Token->getContentStartColumn( 1946 LineIndex, /*Break=*/true); 1947 1948 unsigned NewRemainingTokenColumns = Token->getRemainingLength( 1949 LineIndex, TailOffset + Split.first + Split.second, 1950 ContentStartColumn); 1951 if (NewRemainingTokenColumns == 0) { 1952 // No content to indent. 1953 ContentIndent = 0; 1954 ContentStartColumn = 1955 Token->getContentStartColumn(LineIndex, /*Break=*/true); 1956 NewRemainingTokenColumns = Token->getRemainingLength( 1957 LineIndex, TailOffset + Split.first + Split.second, 1958 ContentStartColumn); 1959 } 1960 1961 // When breaking before a tab character, it may be moved by a few columns, 1962 // but will still be expanded to the next tab stop, so we don't save any 1963 // columns. 1964 if (NewRemainingTokenColumns == RemainingTokenColumns) { 1965 // FIXME: Do we need to adjust the penalty? 1966 break; 1967 } 1968 assert(NewRemainingTokenColumns < RemainingTokenColumns); 1969 1970 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first 1971 << ", " << Split.second << "\n"); 1972 if (!DryRun) 1973 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent, 1974 Whitespaces); 1975 1976 Penalty += NewBreakPenalty; 1977 TailOffset += Split.first + Split.second; 1978 RemainingTokenColumns = NewRemainingTokenColumns; 1979 BreakInserted = true; 1980 NewBreakBefore = true; 1981 } 1982 // In case there's another line, prepare the state for the start of the next 1983 // line. 1984 if (LineIndex + 1 != EndIndex) { 1985 unsigned NextLineIndex = LineIndex + 1; 1986 if (NewBreakBefore) 1987 // After breaking a line, try to reflow the next line into the current 1988 // one once RemainingTokenColumns fits. 1989 TryReflow = true; 1990 if (TryReflow) { 1991 // We decided that we want to try reflowing the next line into the 1992 // current one. 1993 // We will now adjust the state as if the reflow is successful (in 1994 // preparation for the next line), and see whether that works. If we 1995 // decide that we cannot reflow, we will later reset the state to the 1996 // start of the next line. 1997 Reflow = false; 1998 // As we did not continue breaking the line, RemainingTokenColumns is 1999 // known to fit after ContentStartColumn. Adapt ContentStartColumn to 2000 // the position at which we want to format the next line if we do 2001 // actually reflow. 2002 // When we reflow, we need to add a space between the end of the current 2003 // line and the next line's start column. 2004 ContentStartColumn += RemainingTokenColumns + 1; 2005 // Get the split that we need to reflow next logical line into the end 2006 // of the current one; the split will include any leading whitespace of 2007 // the next logical line. 2008 BreakableToken::Split SplitBeforeNext = 2009 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex); 2010 LLVM_DEBUG(llvm::dbgs() 2011 << " Size of reflown text: " << ContentStartColumn 2012 << "\n Potential reflow split: "); 2013 if (SplitBeforeNext.first != StringRef::npos) { 2014 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", " 2015 << SplitBeforeNext.second << "\n"); 2016 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second; 2017 // If the rest of the next line fits into the current line below the 2018 // column limit, we can safely reflow. 2019 RemainingTokenColumns = Token->getRemainingLength( 2020 NextLineIndex, TailOffset, ContentStartColumn); 2021 Reflow = true; 2022 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2023 LLVM_DEBUG(llvm::dbgs() 2024 << " Over limit after reflow, need: " 2025 << (ContentStartColumn + RemainingTokenColumns) 2026 << ", space: " << ColumnLimit 2027 << ", reflown prefix: " << ContentStartColumn 2028 << ", offset in line: " << TailOffset << "\n"); 2029 // If the whole next line does not fit, try to find a point in 2030 // the next line at which we can break so that attaching the part 2031 // of the next line to that break point onto the current line is 2032 // below the column limit. 2033 BreakableToken::Split Split = 2034 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit, 2035 ContentStartColumn, CommentPragmasRegex); 2036 if (Split.first == StringRef::npos) { 2037 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n"); 2038 Reflow = false; 2039 } else { 2040 // Check whether the first split point gets us below the column 2041 // limit. Note that we will execute this split below as part of 2042 // the normal token breaking and reflow logic within the line. 2043 unsigned ToSplitColumns = Token->getRangeLength( 2044 NextLineIndex, TailOffset, Split.first, ContentStartColumn); 2045 if (ContentStartColumn + ToSplitColumns > ColumnLimit) { 2046 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: " 2047 << (ContentStartColumn + ToSplitColumns) 2048 << ", space: " << ColumnLimit); 2049 unsigned ExcessCharactersPenalty = 2050 (ContentStartColumn + ToSplitColumns - ColumnLimit) * 2051 Style.PenaltyExcessCharacter; 2052 if (NewBreakPenalty < ExcessCharactersPenalty) { 2053 Reflow = false; 2054 } 2055 } 2056 } 2057 } 2058 } else { 2059 LLVM_DEBUG(llvm::dbgs() << "not found.\n"); 2060 } 2061 } 2062 if (!Reflow) { 2063 // If we didn't reflow into the next line, the only space to consider is 2064 // the next logical line. Reset our state to match the start of the next 2065 // line. 2066 TailOffset = 0; 2067 ContentStartColumn = 2068 Token->getContentStartColumn(NextLineIndex, /*Break=*/false); 2069 RemainingTokenColumns = Token->getRemainingLength( 2070 NextLineIndex, TailOffset, ContentStartColumn); 2071 // Adapt the start of the token, for example indent. 2072 if (!DryRun) 2073 Token->adaptStartOfLine(NextLineIndex, Whitespaces); 2074 } else { 2075 // If we found a reflow split and have added a new break before the next 2076 // line, we are going to remove the line break at the start of the next 2077 // logical line. For example, here we'll add a new line break after 2078 // 'text', and subsequently delete the line break between 'that' and 2079 // 'reflows'. 2080 // // some text that 2081 // // reflows 2082 // -> 2083 // // some text 2084 // // that reflows 2085 // When adding the line break, we also added the penalty for it, so we 2086 // need to subtract that penalty again when we remove the line break due 2087 // to reflowing. 2088 if (NewBreakBefore) { 2089 assert(Penalty >= NewBreakPenalty); 2090 Penalty -= NewBreakPenalty; 2091 } 2092 if (!DryRun) 2093 Token->reflow(NextLineIndex, Whitespaces); 2094 } 2095 } 2096 } 2097 2098 BreakableToken::Split SplitAfterLastLine = 2099 Token->getSplitAfterLastLine(TailOffset); 2100 if (SplitAfterLastLine.first != StringRef::npos) { 2101 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n"); 2102 2103 // We add the last line's penalty here, since that line is going to be split 2104 // now. 2105 Penalty += Style.PenaltyExcessCharacter * 2106 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2107 2108 if (!DryRun) 2109 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine, 2110 Whitespaces); 2111 ContentStartColumn = 2112 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true); 2113 RemainingTokenColumns = Token->getRemainingLength( 2114 Token->getLineCount() - 1, 2115 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second, 2116 ContentStartColumn); 2117 } 2118 2119 State.Column = ContentStartColumn + RemainingTokenColumns - 2120 Current.UnbreakableTailLength; 2121 2122 if (BreakInserted) { 2123 // If we break the token inside a parameter list, we need to break before 2124 // the next parameter on all levels, so that the next parameter is clearly 2125 // visible. Line comments already introduce a break. 2126 if (Current.isNot(TT_LineComment)) { 2127 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 2128 State.Stack[i].BreakBeforeParameter = true; 2129 } 2130 2131 if (Current.is(TT_BlockComment)) 2132 State.NoContinuation = true; 2133 2134 State.Stack.back().LastSpace = StartColumn; 2135 } 2136 2137 Token->updateNextToken(State); 2138 2139 return {Penalty, Exceeded}; 2140 } 2141 2142 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const { 2143 // In preprocessor directives reserve two chars for trailing " \" 2144 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0); 2145 } 2146 2147 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) { 2148 const FormatToken &Current = *State.NextToken; 2149 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral)) 2150 return false; 2151 // We never consider raw string literals "multiline" for the purpose of 2152 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased 2153 // (see TokenAnnotator::mustBreakBefore(). 2154 if (Current.TokenText.startswith("R\"")) 2155 return false; 2156 if (Current.IsMultiline) 2157 return true; 2158 if (Current.getNextNonComment() && 2159 Current.getNextNonComment()->isStringLiteral()) 2160 return true; // Implicit concatenation. 2161 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals && 2162 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength > 2163 Style.ColumnLimit) 2164 return true; // String will be split. 2165 return false; 2166 } 2167 2168 } // namespace format 2169 } // namespace clang 2170