1 //===--- ContinuationIndenter.cpp - Format C++ code -----------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// 9 /// \file 10 /// This file implements the continuation indenter. 11 /// 12 //===----------------------------------------------------------------------===// 13 14 #include "ContinuationIndenter.h" 15 #include "BreakableToken.h" 16 #include "FormatInternal.h" 17 #include "WhitespaceManager.h" 18 #include "clang/Basic/OperatorPrecedence.h" 19 #include "clang/Basic/SourceManager.h" 20 #include "clang/Format/Format.h" 21 #include "llvm/Support/Debug.h" 22 23 #define DEBUG_TYPE "format-indenter" 24 25 namespace clang { 26 namespace format { 27 28 // Returns true if a TT_SelectorName should be indented when wrapped, 29 // false otherwise. 30 static bool shouldIndentWrappedSelectorName(const FormatStyle &Style, 31 LineType LineType) { 32 return Style.IndentWrappedFunctionNames || LineType == LT_ObjCMethodDecl; 33 } 34 35 // Returns the length of everything up to the first possible line break after 36 // the ), ], } or > matching \c Tok. 37 static unsigned getLengthToMatchingParen(const FormatToken &Tok, 38 const std::vector<ParenState> &Stack) { 39 // Normally whether or not a break before T is possible is calculated and 40 // stored in T.CanBreakBefore. Braces, array initializers and text proto 41 // messages like `key: < ... >` are an exception: a break is possible 42 // before a closing brace R if a break was inserted after the corresponding 43 // opening brace. The information about whether or not a break is needed 44 // before a closing brace R is stored in the ParenState field 45 // S.BreakBeforeClosingBrace where S is the state that R closes. 46 // 47 // In order to decide whether there can be a break before encountered right 48 // braces, this implementation iterates over the sequence of tokens and over 49 // the paren stack in lockstep, keeping track of the stack level which visited 50 // right braces correspond to in MatchingStackIndex. 51 // 52 // For example, consider: 53 // L. <- line number 54 // 1. { 55 // 2. {1}, 56 // 3. {2}, 57 // 4. {{3}}} 58 // ^ where we call this method with this token. 59 // The paren stack at this point contains 3 brace levels: 60 // 0. { at line 1, BreakBeforeClosingBrace: true 61 // 1. first { at line 4, BreakBeforeClosingBrace: false 62 // 2. second { at line 4, BreakBeforeClosingBrace: false, 63 // where there might be fake parens levels in-between these levels. 64 // The algorithm will start at the first } on line 4, which is the matching 65 // brace of the initial left brace and at level 2 of the stack. Then, 66 // examining BreakBeforeClosingBrace: false at level 2, it will continue to 67 // the second } on line 4, and will traverse the stack downwards until it 68 // finds the matching { on level 1. Then, examining BreakBeforeClosingBrace: 69 // false at level 1, it will continue to the third } on line 4 and will 70 // traverse the stack downwards until it finds the matching { on level 0. 71 // Then, examining BreakBeforeClosingBrace: true at level 0, the algorithm 72 // will stop and will use the second } on line 4 to determine the length to 73 // return, as in this example the range will include the tokens: {3}} 74 // 75 // The algorithm will only traverse the stack if it encounters braces, array 76 // initializer squares or text proto angle brackets. 77 if (!Tok.MatchingParen) 78 return 0; 79 FormatToken *End = Tok.MatchingParen; 80 // Maintains a stack level corresponding to the current End token. 81 int MatchingStackIndex = Stack.size() - 1; 82 // Traverses the stack downwards, looking for the level to which LBrace 83 // corresponds. Returns either a pointer to the matching level or nullptr if 84 // LParen is not found in the initial portion of the stack up to 85 // MatchingStackIndex. 86 auto FindParenState = [&](const FormatToken *LBrace) -> const ParenState * { 87 while (MatchingStackIndex >= 0 && Stack[MatchingStackIndex].Tok != LBrace) 88 --MatchingStackIndex; 89 return MatchingStackIndex >= 0 ? &Stack[MatchingStackIndex] : nullptr; 90 }; 91 for (; End->Next; End = End->Next) { 92 if (End->Next->CanBreakBefore) 93 break; 94 if (!End->Next->closesScope()) 95 continue; 96 if (End->Next->MatchingParen && 97 End->Next->MatchingParen->isOneOf( 98 tok::l_brace, TT_ArrayInitializerLSquare, tok::less)) { 99 const ParenState *State = FindParenState(End->Next->MatchingParen); 100 if (State && State->BreakBeforeClosingBrace) 101 break; 102 } 103 } 104 return End->TotalLength - Tok.TotalLength + 1; 105 } 106 107 static unsigned getLengthToNextOperator(const FormatToken &Tok) { 108 if (!Tok.NextOperator) 109 return 0; 110 return Tok.NextOperator->TotalLength - Tok.TotalLength; 111 } 112 113 // Returns \c true if \c Tok is the "." or "->" of a call and starts the next 114 // segment of a builder type call. 115 static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) { 116 return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope(); 117 } 118 119 // Returns \c true if \c Current starts a new parameter. 120 static bool startsNextParameter(const FormatToken &Current, 121 const FormatStyle &Style) { 122 const FormatToken &Previous = *Current.Previous; 123 if (Current.is(TT_CtorInitializerComma) && 124 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 125 return true; 126 if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName)) 127 return true; 128 return Previous.is(tok::comma) && !Current.isTrailingComment() && 129 ((Previous.isNot(TT_CtorInitializerComma) || 130 Style.BreakConstructorInitializers != 131 FormatStyle::BCIS_BeforeComma) && 132 (Previous.isNot(TT_InheritanceComma) || 133 Style.BreakInheritanceList != FormatStyle::BILS_BeforeComma)); 134 } 135 136 static bool opensProtoMessageField(const FormatToken &LessTok, 137 const FormatStyle &Style) { 138 if (LessTok.isNot(tok::less)) 139 return false; 140 return Style.Language == FormatStyle::LK_TextProto || 141 (Style.Language == FormatStyle::LK_Proto && 142 (LessTok.NestingLevel > 0 || 143 (LessTok.Previous && LessTok.Previous->is(tok::equal)))); 144 } 145 146 // Returns the delimiter of a raw string literal, or None if TokenText is not 147 // the text of a raw string literal. The delimiter could be the empty string. 148 // For example, the delimiter of R"deli(cont)deli" is deli. 149 static llvm::Optional<StringRef> getRawStringDelimiter(StringRef TokenText) { 150 if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'. 151 || !TokenText.startswith("R\"") || !TokenText.endswith("\"")) 152 return None; 153 154 // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has 155 // size at most 16 by the standard, so the first '(' must be among the first 156 // 19 bytes. 157 size_t LParenPos = TokenText.substr(0, 19).find_first_of('('); 158 if (LParenPos == StringRef::npos) 159 return None; 160 StringRef Delimiter = TokenText.substr(2, LParenPos - 2); 161 162 // Check that the string ends in ')Delimiter"'. 163 size_t RParenPos = TokenText.size() - Delimiter.size() - 2; 164 if (TokenText[RParenPos] != ')') 165 return None; 166 if (!TokenText.substr(RParenPos + 1).startswith(Delimiter)) 167 return None; 168 return Delimiter; 169 } 170 171 // Returns the canonical delimiter for \p Language, or the empty string if no 172 // canonical delimiter is specified. 173 static StringRef 174 getCanonicalRawStringDelimiter(const FormatStyle &Style, 175 FormatStyle::LanguageKind Language) { 176 for (const auto &Format : Style.RawStringFormats) { 177 if (Format.Language == Language) 178 return StringRef(Format.CanonicalDelimiter); 179 } 180 return ""; 181 } 182 183 RawStringFormatStyleManager::RawStringFormatStyleManager( 184 const FormatStyle &CodeStyle) { 185 for (const auto &RawStringFormat : CodeStyle.RawStringFormats) { 186 llvm::Optional<FormatStyle> LanguageStyle = 187 CodeStyle.GetLanguageStyle(RawStringFormat.Language); 188 if (!LanguageStyle) { 189 FormatStyle PredefinedStyle; 190 if (!getPredefinedStyle(RawStringFormat.BasedOnStyle, 191 RawStringFormat.Language, &PredefinedStyle)) { 192 PredefinedStyle = getLLVMStyle(); 193 PredefinedStyle.Language = RawStringFormat.Language; 194 } 195 LanguageStyle = PredefinedStyle; 196 } 197 LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit; 198 for (StringRef Delimiter : RawStringFormat.Delimiters) { 199 DelimiterStyle.insert({Delimiter, *LanguageStyle}); 200 } 201 for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions) { 202 EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle}); 203 } 204 } 205 } 206 207 llvm::Optional<FormatStyle> 208 RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const { 209 auto It = DelimiterStyle.find(Delimiter); 210 if (It == DelimiterStyle.end()) 211 return None; 212 return It->second; 213 } 214 215 llvm::Optional<FormatStyle> 216 RawStringFormatStyleManager::getEnclosingFunctionStyle( 217 StringRef EnclosingFunction) const { 218 auto It = EnclosingFunctionStyle.find(EnclosingFunction); 219 if (It == EnclosingFunctionStyle.end()) 220 return None; 221 return It->second; 222 } 223 224 ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style, 225 const AdditionalKeywords &Keywords, 226 const SourceManager &SourceMgr, 227 WhitespaceManager &Whitespaces, 228 encoding::Encoding Encoding, 229 bool BinPackInconclusiveFunctions) 230 : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr), 231 Whitespaces(Whitespaces), Encoding(Encoding), 232 BinPackInconclusiveFunctions(BinPackInconclusiveFunctions), 233 CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {} 234 235 LineState ContinuationIndenter::getInitialState(unsigned FirstIndent, 236 unsigned FirstStartColumn, 237 const AnnotatedLine *Line, 238 bool DryRun) { 239 LineState State; 240 State.FirstIndent = FirstIndent; 241 if (FirstStartColumn && Line->First->NewlinesBefore == 0) 242 State.Column = FirstStartColumn; 243 else 244 State.Column = FirstIndent; 245 // With preprocessor directive indentation, the line starts on column 0 246 // since it's indented after the hash, but FirstIndent is set to the 247 // preprocessor indent. 248 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 249 (Line->Type == LT_PreprocessorDirective || 250 Line->Type == LT_ImportStatement)) 251 State.Column = 0; 252 State.Line = Line; 253 State.NextToken = Line->First; 254 State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent, 255 /*AvoidBinPacking=*/false, 256 /*NoLineBreak=*/false)); 257 State.LineContainsContinuedForLoopSection = false; 258 State.NoContinuation = false; 259 State.StartOfStringLiteral = 0; 260 State.StartOfLineLevel = 0; 261 State.LowestLevelOnLine = 0; 262 State.IgnoreStackForComparison = false; 263 264 if (Style.Language == FormatStyle::LK_TextProto) { 265 // We need this in order to deal with the bin packing of text fields at 266 // global scope. 267 State.Stack.back().AvoidBinPacking = true; 268 State.Stack.back().BreakBeforeParameter = true; 269 State.Stack.back().AlignColons = false; 270 } 271 272 // The first token has already been indented and thus consumed. 273 moveStateToNextToken(State, DryRun, /*Newline=*/false); 274 return State; 275 } 276 277 bool ContinuationIndenter::canBreak(const LineState &State) { 278 const FormatToken &Current = *State.NextToken; 279 const FormatToken &Previous = *Current.Previous; 280 assert(&Previous == Current.Previous); 281 if (!Current.CanBreakBefore && !(State.Stack.back().BreakBeforeClosingBrace && 282 Current.closesBlockOrBlockTypeList(Style))) 283 return false; 284 // The opening "{" of a braced list has to be on the same line as the first 285 // element if it is nested in another braced init list or function call. 286 if (!Current.MustBreakBefore && Previous.is(tok::l_brace) && 287 Previous.isNot(TT_DictLiteral) && Previous.BlockKind == BK_BracedInit && 288 Previous.Previous && 289 Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma)) 290 return false; 291 // This prevents breaks like: 292 // ... 293 // SomeParameter, OtherParameter).DoSomething( 294 // ... 295 // As they hide "DoSomething" and are generally bad for readability. 296 if (Previous.opensScope() && Previous.isNot(tok::l_brace) && 297 State.LowestLevelOnLine < State.StartOfLineLevel && 298 State.LowestLevelOnLine < Current.NestingLevel) 299 return false; 300 if (Current.isMemberAccess() && State.Stack.back().ContainsUnwrappedBuilder) 301 return false; 302 303 // Don't create a 'hanging' indent if there are multiple blocks in a single 304 // statement. 305 if (Previous.is(tok::l_brace) && State.Stack.size() > 1 && 306 State.Stack[State.Stack.size() - 2].NestedBlockInlined && 307 State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks) 308 return false; 309 310 // Don't break after very short return types (e.g. "void") as that is often 311 // unexpected. 312 if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) { 313 if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None) 314 return false; 315 } 316 317 // If binary operators are moved to the next line (including commas for some 318 // styles of constructor initializers), that's always ok. 319 if (!Current.isOneOf(TT_BinaryOperator, tok::comma) && 320 State.Stack.back().NoLineBreakInOperand) 321 return false; 322 323 if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr)) 324 return false; 325 326 return !State.Stack.back().NoLineBreak; 327 } 328 329 bool ContinuationIndenter::mustBreak(const LineState &State) { 330 const FormatToken &Current = *State.NextToken; 331 const FormatToken &Previous = *Current.Previous; 332 if (Style.BraceWrapping.BeforeLambdaBody && Current.CanBreakBefore && 333 Current.is(TT_LambdaLBrace)) { 334 auto LambdaBodyLength = getLengthToMatchingParen(Current, State.Stack); 335 return (LambdaBodyLength > getColumnLimit(State)); 336 } 337 if (Current.MustBreakBefore || Current.is(TT_InlineASMColon)) 338 return true; 339 if (State.Stack.back().BreakBeforeClosingBrace && 340 Current.closesBlockOrBlockTypeList(Style)) 341 return true; 342 if (Previous.is(tok::semi) && State.LineContainsContinuedForLoopSection) 343 return true; 344 if (Style.Language == FormatStyle::LK_ObjC && 345 Current.ObjCSelectorNameParts > 1 && 346 Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) { 347 return true; 348 } 349 if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) || 350 (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) && 351 Style.isCpp() && 352 // FIXME: This is a temporary workaround for the case where clang-format 353 // sets BreakBeforeParameter to avoid bin packing and this creates a 354 // completely unnecessary line break after a template type that isn't 355 // line-wrapped. 356 (Previous.NestingLevel == 1 || Style.BinPackParameters)) || 357 (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) && 358 Previous.isNot(tok::question)) || 359 (!Style.BreakBeforeTernaryOperators && 360 Previous.is(TT_ConditionalExpr))) && 361 State.Stack.back().BreakBeforeParameter && !Current.isTrailingComment() && 362 !Current.isOneOf(tok::r_paren, tok::r_brace)) 363 return true; 364 if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) || 365 (Previous.is(TT_ArrayInitializerLSquare) && 366 Previous.ParameterCount > 1) || 367 opensProtoMessageField(Previous, Style)) && 368 Style.ColumnLimit > 0 && 369 getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 > 370 getColumnLimit(State)) 371 return true; 372 373 const FormatToken &BreakConstructorInitializersToken = 374 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon 375 ? Previous 376 : Current; 377 if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) && 378 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength > 379 getColumnLimit(State) || 380 State.Stack.back().BreakBeforeParameter) && 381 (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All || 382 Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon || 383 Style.ColumnLimit != 0)) 384 return true; 385 386 if (Current.is(TT_ObjCMethodExpr) && !Previous.is(TT_SelectorName) && 387 State.Line->startsWith(TT_ObjCMethodSpecifier)) 388 return true; 389 if (Current.is(TT_SelectorName) && !Previous.is(tok::at) && 390 State.Stack.back().ObjCSelectorNameFound && 391 State.Stack.back().BreakBeforeParameter) 392 return true; 393 394 unsigned NewLineColumn = getNewLineColumn(State); 395 if (Current.isMemberAccess() && Style.ColumnLimit != 0 && 396 State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit && 397 (State.Column > NewLineColumn || 398 Current.NestingLevel < State.StartOfLineLevel)) 399 return true; 400 401 if (startsSegmentOfBuilderTypeCall(Current) && 402 (State.Stack.back().CallContinuation != 0 || 403 State.Stack.back().BreakBeforeParameter) && 404 // JavaScript is treated different here as there is a frequent pattern: 405 // SomeFunction(function() { 406 // ... 407 // }.bind(...)); 408 // FIXME: We should find a more generic solution to this problem. 409 !(State.Column <= NewLineColumn && 410 Style.Language == FormatStyle::LK_JavaScript) && 411 !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) 412 return true; 413 414 // If the template declaration spans multiple lines, force wrap before the 415 // function/class declaration 416 if (Previous.ClosesTemplateDeclaration && 417 State.Stack.back().BreakBeforeParameter && Current.CanBreakBefore) 418 return true; 419 420 if (State.Column <= NewLineColumn) 421 return false; 422 423 if (Style.AlwaysBreakBeforeMultilineStrings && 424 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth || 425 Previous.is(tok::comma) || Current.NestingLevel < 2) && 426 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at, 427 Keywords.kw_dollar) && 428 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) && 429 nextIsMultilineString(State)) 430 return true; 431 432 // Using CanBreakBefore here and below takes care of the decision whether the 433 // current style uses wrapping before or after operators for the given 434 // operator. 435 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) { 436 // If we need to break somewhere inside the LHS of a binary expression, we 437 // should also break after the operator. Otherwise, the formatting would 438 // hide the operator precedence, e.g. in: 439 // if (aaaaaaaaaaaaaa == 440 // bbbbbbbbbbbbbb && c) {.. 441 // For comparisons, we only apply this rule, if the LHS is a binary 442 // expression itself as otherwise, the line breaks seem superfluous. 443 // We need special cases for ">>" which we have split into two ">" while 444 // lexing in order to make template parsing easier. 445 bool IsComparison = (Previous.getPrecedence() == prec::Relational || 446 Previous.getPrecedence() == prec::Equality || 447 Previous.getPrecedence() == prec::Spaceship) && 448 Previous.Previous && 449 Previous.Previous->isNot(TT_BinaryOperator); // For >>. 450 bool LHSIsBinaryExpr = 451 Previous.Previous && Previous.Previous->EndsBinaryExpression; 452 if ((!IsComparison || LHSIsBinaryExpr) && !Current.isTrailingComment() && 453 Previous.getPrecedence() != prec::Assignment && 454 State.Stack.back().BreakBeforeParameter) 455 return true; 456 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore && 457 State.Stack.back().BreakBeforeParameter) { 458 return true; 459 } 460 461 // Same as above, but for the first "<<" operator. 462 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) && 463 State.Stack.back().BreakBeforeParameter && 464 State.Stack.back().FirstLessLess == 0) 465 return true; 466 467 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) { 468 // Always break after "template <...>" and leading annotations. This is only 469 // for cases where the entire line does not fit on a single line as a 470 // different LineFormatter would be used otherwise. 471 if (Previous.ClosesTemplateDeclaration) 472 return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No; 473 if (Previous.is(TT_FunctionAnnotationRParen)) 474 return true; 475 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) && 476 Current.isNot(TT_LeadingJavaAnnotation)) 477 return true; 478 } 479 480 // If the return type spans multiple lines, wrap before the function name. 481 if (((Current.is(TT_FunctionDeclarationName) && 482 // Don't break before a C# function when no break after return type 483 (!Style.isCSharp() || 484 Style.AlwaysBreakAfterReturnType != FormatStyle::RTBS_None)) || 485 (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) && 486 !Previous.is(tok::kw_template) && State.Stack.back().BreakBeforeParameter) 487 return true; 488 489 // The following could be precomputed as they do not depend on the state. 490 // However, as they should take effect only if the UnwrappedLine does not fit 491 // into the ColumnLimit, they are checked here in the ContinuationIndenter. 492 if (Style.ColumnLimit != 0 && Previous.BlockKind == BK_Block && 493 Previous.is(tok::l_brace) && !Current.isOneOf(tok::r_brace, tok::comment)) 494 return true; 495 496 if (Current.is(tok::lessless) && 497 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") || 498 (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") || 499 Previous.TokenText == "\'\\n\'")))) 500 return true; 501 502 if (Previous.is(TT_BlockComment) && Previous.IsMultiline) 503 return true; 504 505 if (State.NoContinuation) 506 return true; 507 508 return false; 509 } 510 511 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline, 512 bool DryRun, 513 unsigned ExtraSpaces) { 514 const FormatToken &Current = *State.NextToken; 515 516 assert(!State.Stack.empty()); 517 State.NoContinuation = false; 518 519 if ((Current.is(TT_ImplicitStringLiteral) && 520 (Current.Previous->Tok.getIdentifierInfo() == nullptr || 521 Current.Previous->Tok.getIdentifierInfo()->getPPKeywordID() == 522 tok::pp_not_keyword))) { 523 unsigned EndColumn = 524 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd()); 525 if (Current.LastNewlineOffset != 0) { 526 // If there is a newline within this token, the final column will solely 527 // determined by the current end column. 528 State.Column = EndColumn; 529 } else { 530 unsigned StartColumn = 531 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin()); 532 assert(EndColumn >= StartColumn); 533 State.Column += EndColumn - StartColumn; 534 } 535 moveStateToNextToken(State, DryRun, /*Newline=*/false); 536 return 0; 537 } 538 539 unsigned Penalty = 0; 540 if (Newline) 541 Penalty = addTokenOnNewLine(State, DryRun); 542 else 543 addTokenOnCurrentLine(State, DryRun, ExtraSpaces); 544 545 return moveStateToNextToken(State, DryRun, Newline) + Penalty; 546 } 547 548 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun, 549 unsigned ExtraSpaces) { 550 FormatToken &Current = *State.NextToken; 551 const FormatToken &Previous = *State.NextToken->Previous; 552 if (Current.is(tok::equal) && 553 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) && 554 State.Stack.back().VariablePos == 0) { 555 State.Stack.back().VariablePos = State.Column; 556 // Move over * and & if they are bound to the variable name. 557 const FormatToken *Tok = &Previous; 558 while (Tok && State.Stack.back().VariablePos >= Tok->ColumnWidth) { 559 State.Stack.back().VariablePos -= Tok->ColumnWidth; 560 if (Tok->SpacesRequiredBefore != 0) 561 break; 562 Tok = Tok->Previous; 563 } 564 if (Previous.PartOfMultiVariableDeclStmt) 565 State.Stack.back().LastSpace = State.Stack.back().VariablePos; 566 } 567 568 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces; 569 570 // Indent preprocessor directives after the hash if required. 571 int PPColumnCorrection = 0; 572 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 573 Previous.is(tok::hash) && State.FirstIndent > 0 && 574 (State.Line->Type == LT_PreprocessorDirective || 575 State.Line->Type == LT_ImportStatement)) { 576 Spaces += State.FirstIndent; 577 578 // For preprocessor indent with tabs, State.Column will be 1 because of the 579 // hash. This causes second-level indents onward to have an extra space 580 // after the tabs. We avoid this misalignment by subtracting 1 from the 581 // column value passed to replaceWhitespace(). 582 if (Style.UseTab != FormatStyle::UT_Never) 583 PPColumnCorrection = -1; 584 } 585 586 if (!DryRun) 587 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces, 588 State.Column + Spaces + PPColumnCorrection); 589 590 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance 591 // declaration unless there is multiple inheritance. 592 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 593 Current.is(TT_InheritanceColon)) 594 State.Stack.back().NoLineBreak = true; 595 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon && 596 Previous.is(TT_InheritanceColon)) 597 State.Stack.back().NoLineBreak = true; 598 599 if (Current.is(TT_SelectorName) && 600 !State.Stack.back().ObjCSelectorNameFound) { 601 unsigned MinIndent = 602 std::max(State.FirstIndent + Style.ContinuationIndentWidth, 603 State.Stack.back().Indent); 604 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth; 605 if (Current.LongestObjCSelectorName == 0) 606 State.Stack.back().AlignColons = false; 607 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos) 608 State.Stack.back().ColonPos = MinIndent + Current.LongestObjCSelectorName; 609 else 610 State.Stack.back().ColonPos = FirstColonPos; 611 } 612 613 // In "AlwaysBreak" mode, enforce wrapping directly after the parenthesis by 614 // disallowing any further line breaks if there is no line break after the 615 // opening parenthesis. Don't break if it doesn't conserve columns. 616 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak && 617 (Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) || 618 (Previous.is(tok::l_brace) && Previous.BlockKind != BK_Block && 619 Style.Cpp11BracedListStyle)) && 620 State.Column > getNewLineColumn(State) && 621 (!Previous.Previous || !Previous.Previous->isOneOf( 622 tok::kw_for, tok::kw_while, tok::kw_switch)) && 623 // Don't do this for simple (no expressions) one-argument function calls 624 // as that feels like needlessly wasting whitespace, e.g.: 625 // 626 // caaaaaaaaaaaall( 627 // caaaaaaaaaaaall( 628 // caaaaaaaaaaaall( 629 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa)))); 630 Current.FakeLParens.size() > 0 && 631 Current.FakeLParens.back() > prec::Unknown) 632 State.Stack.back().NoLineBreak = true; 633 if (Previous.is(TT_TemplateString) && Previous.opensScope()) 634 State.Stack.back().NoLineBreak = true; 635 636 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign && 637 Previous.opensScope() && Previous.isNot(TT_ObjCMethodExpr) && 638 (Current.isNot(TT_LineComment) || Previous.BlockKind == BK_BracedInit)) 639 State.Stack.back().Indent = State.Column + Spaces; 640 if (State.Stack.back().AvoidBinPacking && startsNextParameter(Current, Style)) 641 State.Stack.back().NoLineBreak = true; 642 if (startsSegmentOfBuilderTypeCall(Current) && 643 State.Column > getNewLineColumn(State)) 644 State.Stack.back().ContainsUnwrappedBuilder = true; 645 646 if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java) 647 State.Stack.back().NoLineBreak = true; 648 if (Current.isMemberAccess() && Previous.is(tok::r_paren) && 649 (Previous.MatchingParen && 650 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) 651 // If there is a function call with long parameters, break before trailing 652 // calls. This prevents things like: 653 // EXPECT_CALL(SomeLongParameter).Times( 654 // 2); 655 // We don't want to do this for short parameters as they can just be 656 // indexes. 657 State.Stack.back().NoLineBreak = true; 658 659 // Don't allow the RHS of an operator to be split over multiple lines unless 660 // there is a line-break right after the operator. 661 // Exclude relational operators, as there, it is always more desirable to 662 // have the LHS 'left' of the RHS. 663 const FormatToken *P = Current.getPreviousNonComment(); 664 if (!Current.is(tok::comment) && P && 665 (P->isOneOf(TT_BinaryOperator, tok::comma) || 666 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) && 667 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) && 668 P->getPrecedence() != prec::Assignment && 669 P->getPrecedence() != prec::Relational && 670 P->getPrecedence() != prec::Spaceship) { 671 bool BreakBeforeOperator = 672 P->MustBreakBefore || P->is(tok::lessless) || 673 (P->is(TT_BinaryOperator) && 674 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) || 675 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators); 676 // Don't do this if there are only two operands. In these cases, there is 677 // always a nice vertical separation between them and the extra line break 678 // does not help. 679 bool HasTwoOperands = 680 P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr); 681 if ((!BreakBeforeOperator && !(HasTwoOperands && Style.AlignOperands)) || 682 (!State.Stack.back().LastOperatorWrapped && BreakBeforeOperator)) 683 State.Stack.back().NoLineBreakInOperand = true; 684 } 685 686 State.Column += Spaces; 687 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) && 688 Previous.Previous && 689 (Previous.Previous->is(tok::kw_for) || Previous.Previous->isIf())) { 690 // Treat the condition inside an if as if it was a second function 691 // parameter, i.e. let nested calls have a continuation indent. 692 State.Stack.back().LastSpace = State.Column; 693 State.Stack.back().NestedBlockIndent = State.Column; 694 } else if (!Current.isOneOf(tok::comment, tok::caret) && 695 ((Previous.is(tok::comma) && 696 !Previous.is(TT_OverloadedOperator)) || 697 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) { 698 State.Stack.back().LastSpace = State.Column; 699 } else if (Previous.is(TT_CtorInitializerColon) && 700 Style.BreakConstructorInitializers == 701 FormatStyle::BCIS_AfterColon) { 702 State.Stack.back().Indent = State.Column; 703 State.Stack.back().LastSpace = State.Column; 704 } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 705 TT_CtorInitializerColon)) && 706 ((Previous.getPrecedence() != prec::Assignment && 707 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 || 708 Previous.NextOperator)) || 709 Current.StartsBinaryExpression)) { 710 // Indent relative to the RHS of the expression unless this is a simple 711 // assignment without binary expression on the RHS. Also indent relative to 712 // unary operators and the colons of constructor initializers. 713 if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None) 714 State.Stack.back().LastSpace = State.Column; 715 } else if (Previous.is(TT_InheritanceColon)) { 716 State.Stack.back().Indent = State.Column; 717 State.Stack.back().LastSpace = State.Column; 718 } else if (Previous.opensScope()) { 719 // If a function has a trailing call, indent all parameters from the 720 // opening parenthesis. This avoids confusing indents like: 721 // OuterFunction(InnerFunctionCall( // break 722 // ParameterToInnerFunction)) // break 723 // .SecondInnerFunctionCall(); 724 bool HasTrailingCall = false; 725 if (Previous.MatchingParen) { 726 const FormatToken *Next = Previous.MatchingParen->getNextNonComment(); 727 HasTrailingCall = Next && Next->isMemberAccess(); 728 } 729 if (HasTrailingCall && State.Stack.size() > 1 && 730 State.Stack[State.Stack.size() - 2].CallContinuation == 0) 731 State.Stack.back().LastSpace = State.Column; 732 } 733 } 734 735 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State, 736 bool DryRun) { 737 FormatToken &Current = *State.NextToken; 738 const FormatToken &Previous = *State.NextToken->Previous; 739 740 // Extra penalty that needs to be added because of the way certain line 741 // breaks are chosen. 742 unsigned Penalty = 0; 743 744 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 745 const FormatToken *NextNonComment = Previous.getNextNonComment(); 746 if (!NextNonComment) 747 NextNonComment = &Current; 748 // The first line break on any NestingLevel causes an extra penalty in order 749 // prefer similar line breaks. 750 if (!State.Stack.back().ContainsLineBreak) 751 Penalty += 15; 752 State.Stack.back().ContainsLineBreak = true; 753 754 Penalty += State.NextToken->SplitPenalty; 755 756 // Breaking before the first "<<" is generally not desirable if the LHS is 757 // short. Also always add the penalty if the LHS is split over multiple lines 758 // to avoid unnecessary line breaks that just work around this penalty. 759 if (NextNonComment->is(tok::lessless) && 760 State.Stack.back().FirstLessLess == 0 && 761 (State.Column <= Style.ColumnLimit / 3 || 762 State.Stack.back().BreakBeforeParameter)) 763 Penalty += Style.PenaltyBreakFirstLessLess; 764 765 State.Column = getNewLineColumn(State); 766 767 // Indent nested blocks relative to this column, unless in a very specific 768 // JavaScript special case where: 769 // 770 // var loooooong_name = 771 // function() { 772 // // code 773 // } 774 // 775 // is common and should be formatted like a free-standing function. The same 776 // goes for wrapping before the lambda return type arrow. 777 if (!Current.is(TT_LambdaArrow) && 778 (Style.Language != FormatStyle::LK_JavaScript || 779 Current.NestingLevel != 0 || !PreviousNonComment || 780 !PreviousNonComment->is(tok::equal) || 781 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) 782 State.Stack.back().NestedBlockIndent = State.Column; 783 784 if (NextNonComment->isMemberAccess()) { 785 if (State.Stack.back().CallContinuation == 0) 786 State.Stack.back().CallContinuation = State.Column; 787 } else if (NextNonComment->is(TT_SelectorName)) { 788 if (!State.Stack.back().ObjCSelectorNameFound) { 789 if (NextNonComment->LongestObjCSelectorName == 0) { 790 State.Stack.back().AlignColons = false; 791 } else { 792 State.Stack.back().ColonPos = 793 (shouldIndentWrappedSelectorName(Style, State.Line->Type) 794 ? std::max(State.Stack.back().Indent, 795 State.FirstIndent + Style.ContinuationIndentWidth) 796 : State.Stack.back().Indent) + 797 std::max(NextNonComment->LongestObjCSelectorName, 798 NextNonComment->ColumnWidth); 799 } 800 } else if (State.Stack.back().AlignColons && 801 State.Stack.back().ColonPos <= NextNonComment->ColumnWidth) { 802 State.Stack.back().ColonPos = State.Column + NextNonComment->ColumnWidth; 803 } 804 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 805 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) { 806 // FIXME: This is hacky, find a better way. The problem is that in an ObjC 807 // method expression, the block should be aligned to the line starting it, 808 // e.g.: 809 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason 810 // ^(int *i) { 811 // // ... 812 // }]; 813 // Thus, we set LastSpace of the next higher NestingLevel, to which we move 814 // when we consume all of the "}"'s FakeRParens at the "{". 815 if (State.Stack.size() > 1) 816 State.Stack[State.Stack.size() - 2].LastSpace = 817 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 818 Style.ContinuationIndentWidth; 819 } 820 821 if ((PreviousNonComment && 822 PreviousNonComment->isOneOf(tok::comma, tok::semi) && 823 !State.Stack.back().AvoidBinPacking) || 824 Previous.is(TT_BinaryOperator)) 825 State.Stack.back().BreakBeforeParameter = false; 826 if (PreviousNonComment && 827 PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) && 828 Current.NestingLevel == 0) 829 State.Stack.back().BreakBeforeParameter = false; 830 if (NextNonComment->is(tok::question) || 831 (PreviousNonComment && PreviousNonComment->is(tok::question))) 832 State.Stack.back().BreakBeforeParameter = true; 833 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore) 834 State.Stack.back().BreakBeforeParameter = false; 835 836 if (!DryRun) { 837 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1; 838 if (Current.is(tok::r_brace) && Current.MatchingParen && 839 // Only strip trailing empty lines for l_braces that have children, i.e. 840 // for function expressions (lambdas, arrows, etc). 841 !Current.MatchingParen->Children.empty()) { 842 // lambdas and arrow functions are expressions, thus their r_brace is not 843 // on its own line, and thus not covered by UnwrappedLineFormatter's logic 844 // about removing empty lines on closing blocks. Special case them here. 845 MaxEmptyLinesToKeep = 1; 846 } 847 unsigned Newlines = 848 std::max(1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep)); 849 bool ContinuePPDirective = 850 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement; 851 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column, 852 ContinuePPDirective); 853 } 854 855 if (!Current.isTrailingComment()) 856 State.Stack.back().LastSpace = State.Column; 857 if (Current.is(tok::lessless)) 858 // If we are breaking before a "<<", we always want to indent relative to 859 // RHS. This is necessary only for "<<", as we special-case it and don't 860 // always indent relative to the RHS. 861 State.Stack.back().LastSpace += 3; // 3 -> width of "<< ". 862 863 State.StartOfLineLevel = Current.NestingLevel; 864 State.LowestLevelOnLine = Current.NestingLevel; 865 866 // Any break on this level means that the parent level has been broken 867 // and we need to avoid bin packing there. 868 bool NestedBlockSpecialCase = 869 (!Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 && 870 State.Stack[State.Stack.size() - 2].NestedBlockInlined) || 871 (Style.Language == FormatStyle::LK_ObjC && Current.is(tok::r_brace) && 872 State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam); 873 if (!NestedBlockSpecialCase) 874 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 875 State.Stack[i].BreakBeforeParameter = true; 876 877 if (PreviousNonComment && 878 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) && 879 (PreviousNonComment->isNot(TT_TemplateCloser) || 880 Current.NestingLevel != 0) && 881 !PreviousNonComment->isOneOf( 882 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation, 883 TT_LeadingJavaAnnotation) && 884 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope()) 885 State.Stack.back().BreakBeforeParameter = true; 886 887 // If we break after { or the [ of an array initializer, we should also break 888 // before the corresponding } or ]. 889 if (PreviousNonComment && 890 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 891 opensProtoMessageField(*PreviousNonComment, Style))) 892 State.Stack.back().BreakBeforeClosingBrace = true; 893 894 if (State.Stack.back().AvoidBinPacking) { 895 // If we are breaking after '(', '{', '<', or this is the break after a ':' 896 // to start a member initializater list in a constructor, this should not 897 // be considered bin packing unless the relevant AllowAll option is false or 898 // this is a dict/object literal. 899 bool PreviousIsBreakingCtorInitializerColon = 900 Previous.is(TT_CtorInitializerColon) && 901 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 902 if (!(Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) || 903 PreviousIsBreakingCtorInitializerColon) || 904 (!Style.AllowAllParametersOfDeclarationOnNextLine && 905 State.Line->MustBeDeclaration) || 906 (!Style.AllowAllArgumentsOnNextLine && 907 !State.Line->MustBeDeclaration) || 908 (!Style.AllowAllConstructorInitializersOnNextLine && 909 PreviousIsBreakingCtorInitializerColon) || 910 Previous.is(TT_DictLiteral)) 911 State.Stack.back().BreakBeforeParameter = true; 912 913 // If we are breaking after a ':' to start a member initializer list, 914 // and we allow all arguments on the next line, we should not break 915 // before the next parameter. 916 if (PreviousIsBreakingCtorInitializerColon && 917 Style.AllowAllConstructorInitializersOnNextLine) 918 State.Stack.back().BreakBeforeParameter = false; 919 } 920 921 return Penalty; 922 } 923 924 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) { 925 if (!State.NextToken || !State.NextToken->Previous) 926 return 0; 927 FormatToken &Current = *State.NextToken; 928 const FormatToken &Previous = *Current.Previous; 929 // If we are continuing an expression, we want to use the continuation indent. 930 unsigned ContinuationIndent = 931 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 932 Style.ContinuationIndentWidth; 933 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 934 const FormatToken *NextNonComment = Previous.getNextNonComment(); 935 if (!NextNonComment) 936 NextNonComment = &Current; 937 938 // Java specific bits. 939 if (Style.Language == FormatStyle::LK_Java && 940 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) 941 return std::max(State.Stack.back().LastSpace, 942 State.Stack.back().Indent + Style.ContinuationIndentWidth); 943 944 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths && 945 State.Line->First->is(tok::kw_enum)) 946 return (Style.IndentWidth * State.Line->First->IndentLevel) + 947 Style.IndentWidth; 948 949 if (NextNonComment->is(tok::l_brace) && NextNonComment->BlockKind == BK_Block) 950 return Current.NestingLevel == 0 ? State.FirstIndent 951 : State.Stack.back().Indent; 952 if ((Current.isOneOf(tok::r_brace, tok::r_square) || 953 (Current.is(tok::greater) && 954 (Style.Language == FormatStyle::LK_Proto || 955 Style.Language == FormatStyle::LK_TextProto))) && 956 State.Stack.size() > 1) { 957 if (Current.closesBlockOrBlockTypeList(Style)) 958 return State.Stack[State.Stack.size() - 2].NestedBlockIndent; 959 if (Current.MatchingParen && 960 Current.MatchingParen->BlockKind == BK_BracedInit) 961 return State.Stack[State.Stack.size() - 2].LastSpace; 962 return State.FirstIndent; 963 } 964 // Indent a closing parenthesis at the previous level if followed by a semi, 965 // const, or opening brace. This allows indentations such as: 966 // foo( 967 // a, 968 // ); 969 // int Foo::getter( 970 // // 971 // ) const { 972 // return foo; 973 // } 974 // function foo( 975 // a, 976 // ) { 977 // code(); // 978 // } 979 if (Current.is(tok::r_paren) && State.Stack.size() > 1 && 980 (!Current.Next || 981 Current.Next->isOneOf(tok::semi, tok::kw_const, tok::l_brace))) 982 return State.Stack[State.Stack.size() - 2].LastSpace; 983 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope()) 984 return State.Stack[State.Stack.size() - 2].LastSpace; 985 if (Current.is(tok::identifier) && Current.Next && 986 (Current.Next->is(TT_DictLiteral) || 987 ((Style.Language == FormatStyle::LK_Proto || 988 Style.Language == FormatStyle::LK_TextProto) && 989 Current.Next->isOneOf(tok::less, tok::l_brace)))) 990 return State.Stack.back().Indent; 991 if (NextNonComment->is(TT_ObjCStringLiteral) && 992 State.StartOfStringLiteral != 0) 993 return State.StartOfStringLiteral - 1; 994 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0) 995 return State.StartOfStringLiteral; 996 if (NextNonComment->is(tok::lessless) && 997 State.Stack.back().FirstLessLess != 0) 998 return State.Stack.back().FirstLessLess; 999 if (NextNonComment->isMemberAccess()) { 1000 if (State.Stack.back().CallContinuation == 0) 1001 return ContinuationIndent; 1002 return State.Stack.back().CallContinuation; 1003 } 1004 if (State.Stack.back().QuestionColumn != 0 && 1005 ((NextNonComment->is(tok::colon) && 1006 NextNonComment->is(TT_ConditionalExpr)) || 1007 Previous.is(TT_ConditionalExpr))) 1008 return State.Stack.back().QuestionColumn; 1009 if (Previous.is(tok::comma) && State.Stack.back().VariablePos != 0) 1010 return State.Stack.back().VariablePos; 1011 if ((PreviousNonComment && 1012 (PreviousNonComment->ClosesTemplateDeclaration || 1013 PreviousNonComment->isOneOf( 1014 TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen, 1015 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) || 1016 (!Style.IndentWrappedFunctionNames && 1017 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) 1018 return std::max(State.Stack.back().LastSpace, State.Stack.back().Indent); 1019 if (NextNonComment->is(TT_SelectorName)) { 1020 if (!State.Stack.back().ObjCSelectorNameFound) { 1021 unsigned MinIndent = State.Stack.back().Indent; 1022 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) 1023 MinIndent = std::max(MinIndent, 1024 State.FirstIndent + Style.ContinuationIndentWidth); 1025 // If LongestObjCSelectorName is 0, we are indenting the first 1026 // part of an ObjC selector (or a selector component which is 1027 // not colon-aligned due to block formatting). 1028 // 1029 // Otherwise, we are indenting a subsequent part of an ObjC 1030 // selector which should be colon-aligned to the longest 1031 // component of the ObjC selector. 1032 // 1033 // In either case, we want to respect Style.IndentWrappedFunctionNames. 1034 return MinIndent + 1035 std::max(NextNonComment->LongestObjCSelectorName, 1036 NextNonComment->ColumnWidth) - 1037 NextNonComment->ColumnWidth; 1038 } 1039 if (!State.Stack.back().AlignColons) 1040 return State.Stack.back().Indent; 1041 if (State.Stack.back().ColonPos > NextNonComment->ColumnWidth) 1042 return State.Stack.back().ColonPos - NextNonComment->ColumnWidth; 1043 return State.Stack.back().Indent; 1044 } 1045 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr)) 1046 return State.Stack.back().ColonPos; 1047 if (NextNonComment->is(TT_ArraySubscriptLSquare)) { 1048 if (State.Stack.back().StartOfArraySubscripts != 0) 1049 return State.Stack.back().StartOfArraySubscripts; 1050 return ContinuationIndent; 1051 } 1052 1053 // This ensure that we correctly format ObjC methods calls without inputs, 1054 // i.e. where the last element isn't selector like: [callee method]; 1055 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 && 1056 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) 1057 return State.Stack.back().Indent; 1058 1059 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) || 1060 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) 1061 return ContinuationIndent; 1062 if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 1063 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) 1064 return ContinuationIndent; 1065 if (NextNonComment->is(TT_CtorInitializerComma)) 1066 return State.Stack.back().Indent; 1067 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && 1068 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) 1069 return State.Stack.back().Indent; 1070 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) && 1071 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) 1072 return State.Stack.back().Indent; 1073 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon, 1074 TT_InheritanceComma)) 1075 return State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1076 if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() && 1077 !Current.isOneOf(tok::colon, tok::comment)) 1078 return ContinuationIndent; 1079 if (Current.is(TT_ProtoExtensionLSquare)) 1080 return State.Stack.back().Indent; 1081 if (State.Stack.back().Indent == State.FirstIndent && PreviousNonComment && 1082 !PreviousNonComment->isOneOf(tok::r_brace, TT_CtorInitializerComma)) 1083 // Ensure that we fall back to the continuation indent width instead of 1084 // just flushing continuations left. 1085 return State.Stack.back().Indent + Style.ContinuationIndentWidth; 1086 return State.Stack.back().Indent; 1087 } 1088 1089 static bool hasNestedBlockInlined(const FormatToken *Previous, 1090 const FormatToken &Current, 1091 const FormatStyle &Style) { 1092 if (Previous->isNot(tok::l_paren)) 1093 return true; 1094 if (Previous->ParameterCount > 1) 1095 return true; 1096 1097 // Also a nested block if contains a lambda inside function with 1 parameter 1098 return (Style.BraceWrapping.BeforeLambdaBody && Current.is(TT_LambdaLSquare)); 1099 } 1100 1101 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State, 1102 bool DryRun, bool Newline) { 1103 assert(State.Stack.size()); 1104 const FormatToken &Current = *State.NextToken; 1105 1106 if (Current.isOneOf(tok::comma, TT_BinaryOperator)) 1107 State.Stack.back().NoLineBreakInOperand = false; 1108 if (Current.is(TT_InheritanceColon)) 1109 State.Stack.back().AvoidBinPacking = true; 1110 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) { 1111 if (State.Stack.back().FirstLessLess == 0) 1112 State.Stack.back().FirstLessLess = State.Column; 1113 else 1114 State.Stack.back().LastOperatorWrapped = Newline; 1115 } 1116 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless)) 1117 State.Stack.back().LastOperatorWrapped = Newline; 1118 if (Current.is(TT_ConditionalExpr) && Current.Previous && 1119 !Current.Previous->is(TT_ConditionalExpr)) 1120 State.Stack.back().LastOperatorWrapped = Newline; 1121 if (Current.is(TT_ArraySubscriptLSquare) && 1122 State.Stack.back().StartOfArraySubscripts == 0) 1123 State.Stack.back().StartOfArraySubscripts = State.Column; 1124 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question)) 1125 State.Stack.back().QuestionColumn = State.Column; 1126 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) { 1127 const FormatToken *Previous = Current.Previous; 1128 while (Previous && Previous->isTrailingComment()) 1129 Previous = Previous->Previous; 1130 if (Previous && Previous->is(tok::question)) 1131 State.Stack.back().QuestionColumn = State.Column; 1132 } 1133 if (!Current.opensScope() && !Current.closesScope() && 1134 !Current.is(TT_PointerOrReference)) 1135 State.LowestLevelOnLine = 1136 std::min(State.LowestLevelOnLine, Current.NestingLevel); 1137 if (Current.isMemberAccess()) 1138 State.Stack.back().StartOfFunctionCall = 1139 !Current.NextOperator ? 0 : State.Column; 1140 if (Current.is(TT_SelectorName)) 1141 State.Stack.back().ObjCSelectorNameFound = true; 1142 if (Current.is(TT_CtorInitializerColon) && 1143 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) { 1144 // Indent 2 from the column, so: 1145 // SomeClass::SomeClass() 1146 // : First(...), ... 1147 // Next(...) 1148 // ^ line up here. 1149 State.Stack.back().Indent = 1150 State.Column + 1151 (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma 1152 ? 0 1153 : 2); 1154 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1155 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) { 1156 State.Stack.back().AvoidBinPacking = true; 1157 State.Stack.back().BreakBeforeParameter = 1158 !Style.AllowAllConstructorInitializersOnNextLine; 1159 } else { 1160 State.Stack.back().BreakBeforeParameter = false; 1161 } 1162 } 1163 if (Current.is(TT_CtorInitializerColon) && 1164 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { 1165 State.Stack.back().Indent = 1166 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1167 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1168 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 1169 State.Stack.back().AvoidBinPacking = true; 1170 } 1171 if (Current.is(TT_InheritanceColon)) 1172 State.Stack.back().Indent = 1173 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1174 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline) 1175 State.Stack.back().NestedBlockIndent = 1176 State.Column + Current.ColumnWidth + 1; 1177 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow)) 1178 State.Stack.back().LastSpace = State.Column; 1179 1180 // Insert scopes created by fake parenthesis. 1181 const FormatToken *Previous = Current.getPreviousNonComment(); 1182 1183 // Add special behavior to support a format commonly used for JavaScript 1184 // closures: 1185 // SomeFunction(function() { 1186 // foo(); 1187 // bar(); 1188 // }, a, b, c); 1189 if (Current.isNot(tok::comment) && Previous && 1190 Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) && 1191 !Previous->is(TT_DictLiteral) && State.Stack.size() > 1 && 1192 !State.Stack.back().HasMultipleNestedBlocks) { 1193 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline) 1194 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 1195 State.Stack[i].NoLineBreak = true; 1196 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false; 1197 } 1198 if (Previous && 1199 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) || 1200 Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr)) && 1201 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 1202 State.Stack.back().NestedBlockInlined = 1203 !Newline && hasNestedBlockInlined(Previous, Current, Style); 1204 } 1205 1206 moveStatePastFakeLParens(State, Newline); 1207 moveStatePastScopeCloser(State); 1208 bool AllowBreak = !State.Stack.back().NoLineBreak && 1209 !State.Stack.back().NoLineBreakInOperand; 1210 moveStatePastScopeOpener(State, Newline); 1211 moveStatePastFakeRParens(State); 1212 1213 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0) 1214 State.StartOfStringLiteral = State.Column + 1; 1215 if (Current.is(TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) 1216 State.StartOfStringLiteral = State.Column + 1; 1217 else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) 1218 State.StartOfStringLiteral = State.Column; 1219 else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) && 1220 !Current.isStringLiteral()) 1221 State.StartOfStringLiteral = 0; 1222 1223 State.Column += Current.ColumnWidth; 1224 State.NextToken = State.NextToken->Next; 1225 1226 unsigned Penalty = 1227 handleEndOfLine(Current, State, DryRun, AllowBreak, Newline); 1228 1229 if (Current.Role) 1230 Current.Role->formatFromToken(State, this, DryRun); 1231 // If the previous has a special role, let it consume tokens as appropriate. 1232 // It is necessary to start at the previous token for the only implemented 1233 // role (comma separated list). That way, the decision whether or not to break 1234 // after the "{" is already done and both options are tried and evaluated. 1235 // FIXME: This is ugly, find a better way. 1236 if (Previous && Previous->Role) 1237 Penalty += Previous->Role->formatAfterToken(State, this, DryRun); 1238 1239 return Penalty; 1240 } 1241 1242 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State, 1243 bool Newline) { 1244 const FormatToken &Current = *State.NextToken; 1245 const FormatToken *Previous = Current.getPreviousNonComment(); 1246 1247 // Don't add extra indentation for the first fake parenthesis after 1248 // 'return', assignments or opening <({[. The indentation for these cases 1249 // is special cased. 1250 bool SkipFirstExtraIndent = 1251 (Previous && (Previous->opensScope() || 1252 Previous->isOneOf(tok::semi, tok::kw_return) || 1253 (Previous->getPrecedence() == prec::Assignment && 1254 Style.AlignOperands) || 1255 Previous->is(TT_ObjCMethodExpr))); 1256 for (SmallVectorImpl<prec::Level>::const_reverse_iterator 1257 I = Current.FakeLParens.rbegin(), 1258 E = Current.FakeLParens.rend(); 1259 I != E; ++I) { 1260 ParenState NewParenState = State.Stack.back(); 1261 NewParenState.Tok = nullptr; 1262 NewParenState.ContainsLineBreak = false; 1263 NewParenState.LastOperatorWrapped = true; 1264 NewParenState.NoLineBreak = 1265 NewParenState.NoLineBreak || State.Stack.back().NoLineBreakInOperand; 1266 1267 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists. 1268 if (*I > prec::Comma) 1269 NewParenState.AvoidBinPacking = false; 1270 1271 // Indent from 'LastSpace' unless these are fake parentheses encapsulating 1272 // a builder type call after 'return' or, if the alignment after opening 1273 // brackets is disabled. 1274 if (!Current.isTrailingComment() && 1275 (Style.AlignOperands || *I < prec::Assignment) && 1276 (!Previous || Previous->isNot(tok::kw_return) || 1277 (Style.Language != FormatStyle::LK_Java && *I > 0)) && 1278 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign || 1279 *I != prec::Comma || Current.NestingLevel == 0)) 1280 NewParenState.Indent = 1281 std::max(std::max(State.Column, NewParenState.Indent), 1282 State.Stack.back().LastSpace); 1283 1284 // Do not indent relative to the fake parentheses inserted for "." or "->". 1285 // This is a special case to make the following to statements consistent: 1286 // OuterFunction(InnerFunctionCall( // break 1287 // ParameterToInnerFunction)); 1288 // OuterFunction(SomeObject.InnerFunctionCall( // break 1289 // ParameterToInnerFunction)); 1290 if (*I > prec::Unknown) 1291 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column); 1292 if (*I != prec::Conditional && !Current.is(TT_UnaryOperator) && 1293 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 1294 NewParenState.StartOfFunctionCall = State.Column; 1295 1296 // Always indent conditional expressions. Never indent expression where 1297 // the 'operator' is ',', ';' or an assignment (i.e. *I <= 1298 // prec::Assignment) as those have different indentation rules. Indent 1299 // other expression, unless the indentation needs to be skipped. 1300 if (*I == prec::Conditional || 1301 (!SkipFirstExtraIndent && *I > prec::Assignment && 1302 !Current.isTrailingComment())) 1303 NewParenState.Indent += Style.ContinuationIndentWidth; 1304 if ((Previous && !Previous->opensScope()) || *I != prec::Comma) 1305 NewParenState.BreakBeforeParameter = false; 1306 State.Stack.push_back(NewParenState); 1307 SkipFirstExtraIndent = false; 1308 } 1309 } 1310 1311 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) { 1312 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) { 1313 unsigned VariablePos = State.Stack.back().VariablePos; 1314 if (State.Stack.size() == 1) { 1315 // Do not pop the last element. 1316 break; 1317 } 1318 State.Stack.pop_back(); 1319 State.Stack.back().VariablePos = VariablePos; 1320 } 1321 } 1322 1323 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State, 1324 bool Newline) { 1325 const FormatToken &Current = *State.NextToken; 1326 if (!Current.opensScope()) 1327 return; 1328 1329 if (Current.MatchingParen && Current.BlockKind == BK_Block) { 1330 moveStateToNewBlock(State); 1331 return; 1332 } 1333 1334 unsigned NewIndent; 1335 unsigned LastSpace = State.Stack.back().LastSpace; 1336 bool AvoidBinPacking; 1337 bool BreakBeforeParameter = false; 1338 unsigned NestedBlockIndent = std::max(State.Stack.back().StartOfFunctionCall, 1339 State.Stack.back().NestedBlockIndent); 1340 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 1341 opensProtoMessageField(Current, Style)) { 1342 if (Current.opensBlockOrBlockTypeList(Style)) { 1343 NewIndent = Style.IndentWidth + 1344 std::min(State.Column, State.Stack.back().NestedBlockIndent); 1345 } else { 1346 NewIndent = State.Stack.back().LastSpace + Style.ContinuationIndentWidth; 1347 } 1348 const FormatToken *NextNoComment = Current.getNextNonComment(); 1349 bool EndsInComma = Current.MatchingParen && 1350 Current.MatchingParen->Previous && 1351 Current.MatchingParen->Previous->is(tok::comma); 1352 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) || 1353 Style.Language == FormatStyle::LK_Proto || 1354 Style.Language == FormatStyle::LK_TextProto || 1355 !Style.BinPackArguments || 1356 (NextNoComment && 1357 NextNoComment->isOneOf(TT_DesignatedInitializerPeriod, 1358 TT_DesignatedInitializerLSquare)); 1359 BreakBeforeParameter = EndsInComma; 1360 if (Current.ParameterCount > 1) 1361 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1); 1362 } else { 1363 NewIndent = Style.ContinuationIndentWidth + 1364 std::max(State.Stack.back().LastSpace, 1365 State.Stack.back().StartOfFunctionCall); 1366 1367 // Ensure that different different brackets force relative alignment, e.g.: 1368 // void SomeFunction(vector< // break 1369 // int> v); 1370 // FIXME: We likely want to do this for more combinations of brackets. 1371 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) { 1372 NewIndent = std::max(NewIndent, State.Stack.back().Indent); 1373 LastSpace = std::max(LastSpace, State.Stack.back().Indent); 1374 } 1375 1376 bool EndsInComma = 1377 Current.MatchingParen && 1378 Current.MatchingParen->getPreviousNonComment() && 1379 Current.MatchingParen->getPreviousNonComment()->is(tok::comma); 1380 1381 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters 1382 // for backwards compatibility. 1383 bool ObjCBinPackProtocolList = 1384 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto && 1385 Style.BinPackParameters) || 1386 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always; 1387 1388 bool BinPackDeclaration = 1389 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) || 1390 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList); 1391 1392 AvoidBinPacking = 1393 (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) || 1394 (State.Line->MustBeDeclaration && !BinPackDeclaration) || 1395 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) || 1396 (Style.ExperimentalAutoDetectBinPacking && 1397 (Current.PackingKind == PPK_OnePerLine || 1398 (!BinPackInconclusiveFunctions && 1399 Current.PackingKind == PPK_Inconclusive))); 1400 1401 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen && 1402 Style.ObjCBreakBeforeNestedBlockParam) { 1403 if (Style.ColumnLimit) { 1404 // If this '[' opens an ObjC call, determine whether all parameters fit 1405 // into one line and put one per line if they don't. 1406 if (getLengthToMatchingParen(Current, State.Stack) + State.Column > 1407 getColumnLimit(State)) 1408 BreakBeforeParameter = true; 1409 } else { 1410 // For ColumnLimit = 0, we have to figure out whether there is or has to 1411 // be a line break within this call. 1412 for (const FormatToken *Tok = &Current; 1413 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) { 1414 if (Tok->MustBreakBefore || 1415 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) { 1416 BreakBeforeParameter = true; 1417 break; 1418 } 1419 } 1420 } 1421 } 1422 1423 if (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) 1424 BreakBeforeParameter = true; 1425 } 1426 // Generally inherit NoLineBreak from the current scope to nested scope. 1427 // However, don't do this for non-empty nested blocks, dict literals and 1428 // array literals as these follow different indentation rules. 1429 bool NoLineBreak = 1430 Current.Children.empty() && 1431 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) && 1432 (State.Stack.back().NoLineBreak || 1433 State.Stack.back().NoLineBreakInOperand || 1434 (Current.is(TT_TemplateOpener) && 1435 State.Stack.back().ContainsUnwrappedBuilder)); 1436 State.Stack.push_back( 1437 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak)); 1438 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1439 State.Stack.back().BreakBeforeParameter = BreakBeforeParameter; 1440 State.Stack.back().HasMultipleNestedBlocks = (Current.BlockParameterCount > 1); 1441 1442 if (Style.BraceWrapping.BeforeLambdaBody && 1443 Current.Next != nullptr && Current.Tok.is(tok::l_paren)) { 1444 // Search for any parameter that is a lambda 1445 FormatToken const *next = Current.Next; 1446 while (next != nullptr) { 1447 if (next->is(TT_LambdaLSquare)) { 1448 State.Stack.back().HasMultipleNestedBlocks = true; 1449 break; 1450 } 1451 next = next->Next; 1452 } 1453 } 1454 1455 State.Stack.back().IsInsideObjCArrayLiteral = 1456 Current.is(TT_ArrayInitializerLSquare) && Current.Previous && 1457 Current.Previous->is(tok::at); 1458 } 1459 1460 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) { 1461 const FormatToken &Current = *State.NextToken; 1462 if (!Current.closesScope()) 1463 return; 1464 1465 // If we encounter a closing ), ], } or >, we can remove a level from our 1466 // stacks. 1467 if (State.Stack.size() > 1 && 1468 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) || 1469 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) || 1470 State.NextToken->is(TT_TemplateCloser) || 1471 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) 1472 State.Stack.pop_back(); 1473 1474 // Reevaluate whether ObjC message arguments fit into one line. 1475 // If a receiver spans multiple lines, e.g.: 1476 // [[object block:^{ 1477 // return 42; 1478 // }] a:42 b:42]; 1479 // BreakBeforeParameter is calculated based on an incorrect assumption 1480 // (it is checked whether the whole expression fits into one line without 1481 // considering a line break inside a message receiver). 1482 // We check whether arguements fit after receiver scope closer (into the same 1483 // line). 1484 if (State.Stack.back().BreakBeforeParameter && Current.MatchingParen && 1485 Current.MatchingParen->Previous) { 1486 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous; 1487 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) && 1488 CurrentScopeOpener.MatchingParen) { 1489 int NecessarySpaceInLine = 1490 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) + 1491 CurrentScopeOpener.TotalLength - Current.TotalLength - 1; 1492 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <= 1493 Style.ColumnLimit) 1494 State.Stack.back().BreakBeforeParameter = false; 1495 } 1496 } 1497 1498 if (Current.is(tok::r_square)) { 1499 // If this ends the array subscript expr, reset the corresponding value. 1500 const FormatToken *NextNonComment = Current.getNextNonComment(); 1501 if (NextNonComment && NextNonComment->isNot(tok::l_square)) 1502 State.Stack.back().StartOfArraySubscripts = 0; 1503 } 1504 } 1505 1506 void ContinuationIndenter::moveStateToNewBlock(LineState &State) { 1507 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent; 1508 // ObjC block sometimes follow special indentation rules. 1509 unsigned NewIndent = 1510 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace) 1511 ? Style.ObjCBlockIndentWidth 1512 : Style.IndentWidth); 1513 State.Stack.push_back(ParenState(State.NextToken, NewIndent, 1514 State.Stack.back().LastSpace, 1515 /*AvoidBinPacking=*/true, 1516 /*NoLineBreak=*/false)); 1517 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1518 State.Stack.back().BreakBeforeParameter = true; 1519 } 1520 1521 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn, 1522 unsigned TabWidth, 1523 encoding::Encoding Encoding) { 1524 size_t LastNewlinePos = Text.find_last_of("\n"); 1525 if (LastNewlinePos == StringRef::npos) { 1526 return StartColumn + 1527 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding); 1528 } else { 1529 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos), 1530 /*StartColumn=*/0, TabWidth, Encoding); 1531 } 1532 } 1533 1534 unsigned ContinuationIndenter::reformatRawStringLiteral( 1535 const FormatToken &Current, LineState &State, 1536 const FormatStyle &RawStringStyle, bool DryRun, bool Newline) { 1537 unsigned StartColumn = State.Column - Current.ColumnWidth; 1538 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText); 1539 StringRef NewDelimiter = 1540 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language); 1541 if (NewDelimiter.empty() || OldDelimiter.empty()) 1542 NewDelimiter = OldDelimiter; 1543 // The text of a raw string is between the leading 'R"delimiter(' and the 1544 // trailing 'delimiter)"'. 1545 unsigned OldPrefixSize = 3 + OldDelimiter.size(); 1546 unsigned OldSuffixSize = 2 + OldDelimiter.size(); 1547 // We create a virtual text environment which expects a null-terminated 1548 // string, so we cannot use StringRef. 1549 std::string RawText = std::string( 1550 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize)); 1551 if (NewDelimiter != OldDelimiter) { 1552 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the 1553 // raw string. 1554 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str(); 1555 if (StringRef(RawText).contains(CanonicalDelimiterSuffix)) 1556 NewDelimiter = OldDelimiter; 1557 } 1558 1559 unsigned NewPrefixSize = 3 + NewDelimiter.size(); 1560 unsigned NewSuffixSize = 2 + NewDelimiter.size(); 1561 1562 // The first start column is the column the raw text starts after formatting. 1563 unsigned FirstStartColumn = StartColumn + NewPrefixSize; 1564 1565 // The next start column is the intended indentation a line break inside 1566 // the raw string at level 0. It is determined by the following rules: 1567 // - if the content starts on newline, it is one level more than the current 1568 // indent, and 1569 // - if the content does not start on a newline, it is the first start 1570 // column. 1571 // These rules have the advantage that the formatted content both does not 1572 // violate the rectangle rule and visually flows within the surrounding 1573 // source. 1574 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n'; 1575 // If this token is the last parameter (checked by looking if it's followed by 1576 // `)` and is not on a newline, the base the indent off the line's nested 1577 // block indent. Otherwise, base the indent off the arguments indent, so we 1578 // can achieve: 1579 // 1580 // fffffffffff(1, 2, 3, R"pb( 1581 // key1: 1 # 1582 // key2: 2)pb"); 1583 // 1584 // fffffffffff(1, 2, 3, 1585 // R"pb( 1586 // key1: 1 # 1587 // key2: 2 1588 // )pb"); 1589 // 1590 // fffffffffff(1, 2, 3, 1591 // R"pb( 1592 // key1: 1 # 1593 // key2: 2 1594 // )pb", 1595 // 5); 1596 unsigned CurrentIndent = 1597 (!Newline && Current.Next && Current.Next->is(tok::r_paren)) 1598 ? State.Stack.back().NestedBlockIndent 1599 : State.Stack.back().Indent; 1600 unsigned NextStartColumn = ContentStartsOnNewline 1601 ? CurrentIndent + Style.IndentWidth 1602 : FirstStartColumn; 1603 1604 // The last start column is the column the raw string suffix starts if it is 1605 // put on a newline. 1606 // The last start column is the intended indentation of the raw string postfix 1607 // if it is put on a newline. It is determined by the following rules: 1608 // - if the raw string prefix starts on a newline, it is the column where 1609 // that raw string prefix starts, and 1610 // - if the raw string prefix does not start on a newline, it is the current 1611 // indent. 1612 unsigned LastStartColumn = 1613 Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent; 1614 1615 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat( 1616 RawStringStyle, RawText, {tooling::Range(0, RawText.size())}, 1617 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>", 1618 /*Status=*/nullptr); 1619 1620 auto NewCode = applyAllReplacements(RawText, Fixes.first); 1621 tooling::Replacements NoFixes; 1622 if (!NewCode) { 1623 return addMultilineToken(Current, State); 1624 } 1625 if (!DryRun) { 1626 if (NewDelimiter != OldDelimiter) { 1627 // In 'R"delimiter(...', the delimiter starts 2 characters after the start 1628 // of the token. 1629 SourceLocation PrefixDelimiterStart = 1630 Current.Tok.getLocation().getLocWithOffset(2); 1631 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement( 1632 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1633 if (PrefixErr) { 1634 llvm::errs() 1635 << "Failed to update the prefix delimiter of a raw string: " 1636 << llvm::toString(std::move(PrefixErr)) << "\n"; 1637 } 1638 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at 1639 // position length - 1 - |delimiter|. 1640 SourceLocation SuffixDelimiterStart = 1641 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() - 1642 1 - OldDelimiter.size()); 1643 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement( 1644 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1645 if (SuffixErr) { 1646 llvm::errs() 1647 << "Failed to update the suffix delimiter of a raw string: " 1648 << llvm::toString(std::move(SuffixErr)) << "\n"; 1649 } 1650 } 1651 SourceLocation OriginLoc = 1652 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize); 1653 for (const tooling::Replacement &Fix : Fixes.first) { 1654 auto Err = Whitespaces.addReplacement(tooling::Replacement( 1655 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()), 1656 Fix.getLength(), Fix.getReplacementText())); 1657 if (Err) { 1658 llvm::errs() << "Failed to reformat raw string: " 1659 << llvm::toString(std::move(Err)) << "\n"; 1660 } 1661 } 1662 } 1663 unsigned RawLastLineEndColumn = getLastLineEndColumn( 1664 *NewCode, FirstStartColumn, Style.TabWidth, Encoding); 1665 State.Column = RawLastLineEndColumn + NewSuffixSize; 1666 // Since we're updating the column to after the raw string literal here, we 1667 // have to manually add the penalty for the prefix R"delim( over the column 1668 // limit. 1669 unsigned PrefixExcessCharacters = 1670 StartColumn + NewPrefixSize > Style.ColumnLimit 1671 ? StartColumn + NewPrefixSize - Style.ColumnLimit 1672 : 0; 1673 bool IsMultiline = 1674 ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos); 1675 if (IsMultiline) { 1676 // Break before further function parameters on all levels. 1677 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 1678 State.Stack[i].BreakBeforeParameter = true; 1679 } 1680 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter; 1681 } 1682 1683 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current, 1684 LineState &State) { 1685 // Break before further function parameters on all levels. 1686 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 1687 State.Stack[i].BreakBeforeParameter = true; 1688 1689 unsigned ColumnsUsed = State.Column; 1690 // We can only affect layout of the first and the last line, so the penalty 1691 // for all other lines is constant, and we ignore it. 1692 State.Column = Current.LastLineColumnWidth; 1693 1694 if (ColumnsUsed > getColumnLimit(State)) 1695 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State)); 1696 return 0; 1697 } 1698 1699 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current, 1700 LineState &State, bool DryRun, 1701 bool AllowBreak, bool Newline) { 1702 unsigned Penalty = 0; 1703 // Compute the raw string style to use in case this is a raw string literal 1704 // that can be reformatted. 1705 auto RawStringStyle = getRawStringStyle(Current, State); 1706 if (RawStringStyle && !Current.Finalized) { 1707 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun, 1708 Newline); 1709 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) { 1710 // Don't break multi-line tokens other than block comments and raw string 1711 // literals. Instead, just update the state. 1712 Penalty = addMultilineToken(Current, State); 1713 } else if (State.Line->Type != LT_ImportStatement) { 1714 // We generally don't break import statements. 1715 LineState OriginalState = State; 1716 1717 // Whether we force the reflowing algorithm to stay strictly within the 1718 // column limit. 1719 bool Strict = false; 1720 // Whether the first non-strict attempt at reflowing did intentionally 1721 // exceed the column limit. 1722 bool Exceeded = false; 1723 std::tie(Penalty, Exceeded) = breakProtrudingToken( 1724 Current, State, AllowBreak, /*DryRun=*/true, Strict); 1725 if (Exceeded) { 1726 // If non-strict reflowing exceeds the column limit, try whether strict 1727 // reflowing leads to an overall lower penalty. 1728 LineState StrictState = OriginalState; 1729 unsigned StrictPenalty = 1730 breakProtrudingToken(Current, StrictState, AllowBreak, 1731 /*DryRun=*/true, /*Strict=*/true) 1732 .first; 1733 Strict = StrictPenalty <= Penalty; 1734 if (Strict) { 1735 Penalty = StrictPenalty; 1736 State = StrictState; 1737 } 1738 } 1739 if (!DryRun) { 1740 // If we're not in dry-run mode, apply the changes with the decision on 1741 // strictness made above. 1742 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false, 1743 Strict); 1744 } 1745 } 1746 if (State.Column > getColumnLimit(State)) { 1747 unsigned ExcessCharacters = State.Column - getColumnLimit(State); 1748 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters; 1749 } 1750 return Penalty; 1751 } 1752 1753 // Returns the enclosing function name of a token, or the empty string if not 1754 // found. 1755 static StringRef getEnclosingFunctionName(const FormatToken &Current) { 1756 // Look for: 'function(' or 'function<templates>(' before Current. 1757 auto Tok = Current.getPreviousNonComment(); 1758 if (!Tok || !Tok->is(tok::l_paren)) 1759 return ""; 1760 Tok = Tok->getPreviousNonComment(); 1761 if (!Tok) 1762 return ""; 1763 if (Tok->is(TT_TemplateCloser)) { 1764 Tok = Tok->MatchingParen; 1765 if (Tok) 1766 Tok = Tok->getPreviousNonComment(); 1767 } 1768 if (!Tok || !Tok->is(tok::identifier)) 1769 return ""; 1770 return Tok->TokenText; 1771 } 1772 1773 llvm::Optional<FormatStyle> 1774 ContinuationIndenter::getRawStringStyle(const FormatToken &Current, 1775 const LineState &State) { 1776 if (!Current.isStringLiteral()) 1777 return None; 1778 auto Delimiter = getRawStringDelimiter(Current.TokenText); 1779 if (!Delimiter) 1780 return None; 1781 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter); 1782 if (!RawStringStyle && Delimiter->empty()) 1783 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle( 1784 getEnclosingFunctionName(Current)); 1785 if (!RawStringStyle) 1786 return None; 1787 RawStringStyle->ColumnLimit = getColumnLimit(State); 1788 return RawStringStyle; 1789 } 1790 1791 std::unique_ptr<BreakableToken> 1792 ContinuationIndenter::createBreakableToken(const FormatToken &Current, 1793 LineState &State, bool AllowBreak) { 1794 unsigned StartColumn = State.Column - Current.ColumnWidth; 1795 if (Current.isStringLiteral()) { 1796 // FIXME: String literal breaking is currently disabled for C#, Java and 1797 // JavaScript, as it requires strings to be merged using "+" which we 1798 // don't support. 1799 if (Style.Language == FormatStyle::LK_Java || 1800 Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp() || 1801 !Style.BreakStringLiterals || !AllowBreak) 1802 return nullptr; 1803 1804 // Don't break string literals inside preprocessor directives (except for 1805 // #define directives, as their contents are stored in separate lines and 1806 // are not affected by this check). 1807 // This way we avoid breaking code with line directives and unknown 1808 // preprocessor directives that contain long string literals. 1809 if (State.Line->Type == LT_PreprocessorDirective) 1810 return nullptr; 1811 // Exempts unterminated string literals from line breaking. The user will 1812 // likely want to terminate the string before any line breaking is done. 1813 if (Current.IsUnterminatedLiteral) 1814 return nullptr; 1815 // Don't break string literals inside Objective-C array literals (doing so 1816 // raises the warning -Wobjc-string-concatenation). 1817 if (State.Stack.back().IsInsideObjCArrayLiteral) { 1818 return nullptr; 1819 } 1820 1821 StringRef Text = Current.TokenText; 1822 StringRef Prefix; 1823 StringRef Postfix; 1824 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'. 1825 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to 1826 // reduce the overhead) for each FormatToken, which is a string, so that we 1827 // don't run multiple checks here on the hot path. 1828 if ((Text.endswith(Postfix = "\"") && 1829 (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") || 1830 Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") || 1831 Text.startswith(Prefix = "u8\"") || 1832 Text.startswith(Prefix = "L\""))) || 1833 (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) { 1834 // We need this to address the case where there is an unbreakable tail 1835 // only if certain other formatting decisions have been taken. The 1836 // UnbreakableTailLength of Current is an overapproximation is that case 1837 // and we need to be correct here. 1838 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State)) 1839 ? 0 1840 : Current.UnbreakableTailLength; 1841 return std::make_unique<BreakableStringLiteral>( 1842 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength, 1843 State.Line->InPPDirective, Encoding, Style); 1844 } 1845 } else if (Current.is(TT_BlockComment)) { 1846 if (!Style.ReflowComments || 1847 // If a comment token switches formatting, like 1848 // /* clang-format on */, we don't want to break it further, 1849 // but we may still want to adjust its indentation. 1850 switchesFormatting(Current)) { 1851 return nullptr; 1852 } 1853 return std::make_unique<BreakableBlockComment>( 1854 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1855 State.Line->InPPDirective, Encoding, Style, Whitespaces.useCRLF()); 1856 } else if (Current.is(TT_LineComment) && 1857 (Current.Previous == nullptr || 1858 Current.Previous->isNot(TT_ImplicitStringLiteral))) { 1859 if (!Style.ReflowComments || 1860 CommentPragmasRegex.match(Current.TokenText.substr(2)) || 1861 switchesFormatting(Current)) 1862 return nullptr; 1863 return std::make_unique<BreakableLineCommentSection>( 1864 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1865 /*InPPDirective=*/false, Encoding, Style); 1866 } 1867 return nullptr; 1868 } 1869 1870 std::pair<unsigned, bool> 1871 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current, 1872 LineState &State, bool AllowBreak, 1873 bool DryRun, bool Strict) { 1874 std::unique_ptr<const BreakableToken> Token = 1875 createBreakableToken(Current, State, AllowBreak); 1876 if (!Token) 1877 return {0, false}; 1878 assert(Token->getLineCount() > 0); 1879 unsigned ColumnLimit = getColumnLimit(State); 1880 if (Current.is(TT_LineComment)) { 1881 // We don't insert backslashes when breaking line comments. 1882 ColumnLimit = Style.ColumnLimit; 1883 } 1884 if (Current.UnbreakableTailLength >= ColumnLimit) 1885 return {0, false}; 1886 // ColumnWidth was already accounted into State.Column before calling 1887 // breakProtrudingToken. 1888 unsigned StartColumn = State.Column - Current.ColumnWidth; 1889 unsigned NewBreakPenalty = Current.isStringLiteral() 1890 ? Style.PenaltyBreakString 1891 : Style.PenaltyBreakComment; 1892 // Stores whether we intentionally decide to let a line exceed the column 1893 // limit. 1894 bool Exceeded = false; 1895 // Stores whether we introduce a break anywhere in the token. 1896 bool BreakInserted = Token->introducesBreakBeforeToken(); 1897 // Store whether we inserted a new line break at the end of the previous 1898 // logical line. 1899 bool NewBreakBefore = false; 1900 // We use a conservative reflowing strategy. Reflow starts after a line is 1901 // broken or the corresponding whitespace compressed. Reflow ends as soon as a 1902 // line that doesn't get reflown with the previous line is reached. 1903 bool Reflow = false; 1904 // Keep track of where we are in the token: 1905 // Where we are in the content of the current logical line. 1906 unsigned TailOffset = 0; 1907 // The column number we're currently at. 1908 unsigned ContentStartColumn = 1909 Token->getContentStartColumn(0, /*Break=*/false); 1910 // The number of columns left in the current logical line after TailOffset. 1911 unsigned RemainingTokenColumns = 1912 Token->getRemainingLength(0, TailOffset, ContentStartColumn); 1913 // Adapt the start of the token, for example indent. 1914 if (!DryRun) 1915 Token->adaptStartOfLine(0, Whitespaces); 1916 1917 unsigned ContentIndent = 0; 1918 unsigned Penalty = 0; 1919 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column " 1920 << StartColumn << ".\n"); 1921 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount(); 1922 LineIndex != EndIndex; ++LineIndex) { 1923 LLVM_DEBUG(llvm::dbgs() 1924 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n"); 1925 NewBreakBefore = false; 1926 // If we did reflow the previous line, we'll try reflowing again. Otherwise 1927 // we'll start reflowing if the current line is broken or whitespace is 1928 // compressed. 1929 bool TryReflow = Reflow; 1930 // Break the current token until we can fit the rest of the line. 1931 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 1932 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: " 1933 << (ContentStartColumn + RemainingTokenColumns) 1934 << ", space: " << ColumnLimit 1935 << ", reflown prefix: " << ContentStartColumn 1936 << ", offset in line: " << TailOffset << "\n"); 1937 // If the current token doesn't fit, find the latest possible split in the 1938 // current line so that breaking at it will be under the column limit. 1939 // FIXME: Use the earliest possible split while reflowing to correctly 1940 // compress whitespace within a line. 1941 BreakableToken::Split Split = 1942 Token->getSplit(LineIndex, TailOffset, ColumnLimit, 1943 ContentStartColumn, CommentPragmasRegex); 1944 if (Split.first == StringRef::npos) { 1945 // No break opportunity - update the penalty and continue with the next 1946 // logical line. 1947 if (LineIndex < EndIndex - 1) 1948 // The last line's penalty is handled in addNextStateToQueue() or when 1949 // calling replaceWhitespaceAfterLastLine below. 1950 Penalty += Style.PenaltyExcessCharacter * 1951 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 1952 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n"); 1953 break; 1954 } 1955 assert(Split.first != 0); 1956 1957 if (Token->supportsReflow()) { 1958 // Check whether the next natural split point after the current one can 1959 // still fit the line, either because we can compress away whitespace, 1960 // or because the penalty the excess characters introduce is lower than 1961 // the break penalty. 1962 // We only do this for tokens that support reflowing, and thus allow us 1963 // to change the whitespace arbitrarily (e.g. comments). 1964 // Other tokens, like string literals, can be broken on arbitrary 1965 // positions. 1966 1967 // First, compute the columns from TailOffset to the next possible split 1968 // position. 1969 // For example: 1970 // ColumnLimit: | 1971 // // Some text that breaks 1972 // ^ tail offset 1973 // ^-- split 1974 // ^-------- to split columns 1975 // ^--- next split 1976 // ^--------------- to next split columns 1977 unsigned ToSplitColumns = Token->getRangeLength( 1978 LineIndex, TailOffset, Split.first, ContentStartColumn); 1979 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n"); 1980 1981 BreakableToken::Split NextSplit = Token->getSplit( 1982 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit, 1983 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex); 1984 // Compute the columns necessary to fit the next non-breakable sequence 1985 // into the current line. 1986 unsigned ToNextSplitColumns = 0; 1987 if (NextSplit.first == StringRef::npos) { 1988 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset, 1989 ContentStartColumn); 1990 } else { 1991 ToNextSplitColumns = Token->getRangeLength( 1992 LineIndex, TailOffset, 1993 Split.first + Split.second + NextSplit.first, ContentStartColumn); 1994 } 1995 // Compress the whitespace between the break and the start of the next 1996 // unbreakable sequence. 1997 ToNextSplitColumns = 1998 Token->getLengthAfterCompression(ToNextSplitColumns, Split); 1999 LLVM_DEBUG(llvm::dbgs() 2000 << " ContentStartColumn: " << ContentStartColumn << "\n"); 2001 LLVM_DEBUG(llvm::dbgs() 2002 << " ToNextSplit: " << ToNextSplitColumns << "\n"); 2003 // If the whitespace compression makes us fit, continue on the current 2004 // line. 2005 bool ContinueOnLine = 2006 ContentStartColumn + ToNextSplitColumns <= ColumnLimit; 2007 unsigned ExcessCharactersPenalty = 0; 2008 if (!ContinueOnLine && !Strict) { 2009 // Similarly, if the excess characters' penalty is lower than the 2010 // penalty of introducing a new break, continue on the current line. 2011 ExcessCharactersPenalty = 2012 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) * 2013 Style.PenaltyExcessCharacter; 2014 LLVM_DEBUG(llvm::dbgs() 2015 << " Penalty excess: " << ExcessCharactersPenalty 2016 << "\n break : " << NewBreakPenalty << "\n"); 2017 if (ExcessCharactersPenalty < NewBreakPenalty) { 2018 Exceeded = true; 2019 ContinueOnLine = true; 2020 } 2021 } 2022 if (ContinueOnLine) { 2023 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n"); 2024 // The current line fits after compressing the whitespace - reflow 2025 // the next line into it if possible. 2026 TryReflow = true; 2027 if (!DryRun) 2028 Token->compressWhitespace(LineIndex, TailOffset, Split, 2029 Whitespaces); 2030 // When we continue on the same line, leave one space between content. 2031 ContentStartColumn += ToSplitColumns + 1; 2032 Penalty += ExcessCharactersPenalty; 2033 TailOffset += Split.first + Split.second; 2034 RemainingTokenColumns = Token->getRemainingLength( 2035 LineIndex, TailOffset, ContentStartColumn); 2036 continue; 2037 } 2038 } 2039 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n"); 2040 // Update the ContentIndent only if the current line was not reflown with 2041 // the previous line, since in that case the previous line should still 2042 // determine the ContentIndent. Also never intent the last line. 2043 if (!Reflow) 2044 ContentIndent = Token->getContentIndent(LineIndex); 2045 LLVM_DEBUG(llvm::dbgs() 2046 << " ContentIndent: " << ContentIndent << "\n"); 2047 ContentStartColumn = ContentIndent + Token->getContentStartColumn( 2048 LineIndex, /*Break=*/true); 2049 2050 unsigned NewRemainingTokenColumns = Token->getRemainingLength( 2051 LineIndex, TailOffset + Split.first + Split.second, 2052 ContentStartColumn); 2053 if (NewRemainingTokenColumns == 0) { 2054 // No content to indent. 2055 ContentIndent = 0; 2056 ContentStartColumn = 2057 Token->getContentStartColumn(LineIndex, /*Break=*/true); 2058 NewRemainingTokenColumns = Token->getRemainingLength( 2059 LineIndex, TailOffset + Split.first + Split.second, 2060 ContentStartColumn); 2061 } 2062 2063 // When breaking before a tab character, it may be moved by a few columns, 2064 // but will still be expanded to the next tab stop, so we don't save any 2065 // columns. 2066 if (NewRemainingTokenColumns == RemainingTokenColumns) { 2067 // FIXME: Do we need to adjust the penalty? 2068 break; 2069 } 2070 assert(NewRemainingTokenColumns < RemainingTokenColumns); 2071 2072 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first 2073 << ", " << Split.second << "\n"); 2074 if (!DryRun) 2075 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent, 2076 Whitespaces); 2077 2078 Penalty += NewBreakPenalty; 2079 TailOffset += Split.first + Split.second; 2080 RemainingTokenColumns = NewRemainingTokenColumns; 2081 BreakInserted = true; 2082 NewBreakBefore = true; 2083 } 2084 // In case there's another line, prepare the state for the start of the next 2085 // line. 2086 if (LineIndex + 1 != EndIndex) { 2087 unsigned NextLineIndex = LineIndex + 1; 2088 if (NewBreakBefore) 2089 // After breaking a line, try to reflow the next line into the current 2090 // one once RemainingTokenColumns fits. 2091 TryReflow = true; 2092 if (TryReflow) { 2093 // We decided that we want to try reflowing the next line into the 2094 // current one. 2095 // We will now adjust the state as if the reflow is successful (in 2096 // preparation for the next line), and see whether that works. If we 2097 // decide that we cannot reflow, we will later reset the state to the 2098 // start of the next line. 2099 Reflow = false; 2100 // As we did not continue breaking the line, RemainingTokenColumns is 2101 // known to fit after ContentStartColumn. Adapt ContentStartColumn to 2102 // the position at which we want to format the next line if we do 2103 // actually reflow. 2104 // When we reflow, we need to add a space between the end of the current 2105 // line and the next line's start column. 2106 ContentStartColumn += RemainingTokenColumns + 1; 2107 // Get the split that we need to reflow next logical line into the end 2108 // of the current one; the split will include any leading whitespace of 2109 // the next logical line. 2110 BreakableToken::Split SplitBeforeNext = 2111 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex); 2112 LLVM_DEBUG(llvm::dbgs() 2113 << " Size of reflown text: " << ContentStartColumn 2114 << "\n Potential reflow split: "); 2115 if (SplitBeforeNext.first != StringRef::npos) { 2116 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", " 2117 << SplitBeforeNext.second << "\n"); 2118 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second; 2119 // If the rest of the next line fits into the current line below the 2120 // column limit, we can safely reflow. 2121 RemainingTokenColumns = Token->getRemainingLength( 2122 NextLineIndex, TailOffset, ContentStartColumn); 2123 Reflow = true; 2124 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2125 LLVM_DEBUG(llvm::dbgs() 2126 << " Over limit after reflow, need: " 2127 << (ContentStartColumn + RemainingTokenColumns) 2128 << ", space: " << ColumnLimit 2129 << ", reflown prefix: " << ContentStartColumn 2130 << ", offset in line: " << TailOffset << "\n"); 2131 // If the whole next line does not fit, try to find a point in 2132 // the next line at which we can break so that attaching the part 2133 // of the next line to that break point onto the current line is 2134 // below the column limit. 2135 BreakableToken::Split Split = 2136 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit, 2137 ContentStartColumn, CommentPragmasRegex); 2138 if (Split.first == StringRef::npos) { 2139 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n"); 2140 Reflow = false; 2141 } else { 2142 // Check whether the first split point gets us below the column 2143 // limit. Note that we will execute this split below as part of 2144 // the normal token breaking and reflow logic within the line. 2145 unsigned ToSplitColumns = Token->getRangeLength( 2146 NextLineIndex, TailOffset, Split.first, ContentStartColumn); 2147 if (ContentStartColumn + ToSplitColumns > ColumnLimit) { 2148 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: " 2149 << (ContentStartColumn + ToSplitColumns) 2150 << ", space: " << ColumnLimit); 2151 unsigned ExcessCharactersPenalty = 2152 (ContentStartColumn + ToSplitColumns - ColumnLimit) * 2153 Style.PenaltyExcessCharacter; 2154 if (NewBreakPenalty < ExcessCharactersPenalty) { 2155 Reflow = false; 2156 } 2157 } 2158 } 2159 } 2160 } else { 2161 LLVM_DEBUG(llvm::dbgs() << "not found.\n"); 2162 } 2163 } 2164 if (!Reflow) { 2165 // If we didn't reflow into the next line, the only space to consider is 2166 // the next logical line. Reset our state to match the start of the next 2167 // line. 2168 TailOffset = 0; 2169 ContentStartColumn = 2170 Token->getContentStartColumn(NextLineIndex, /*Break=*/false); 2171 RemainingTokenColumns = Token->getRemainingLength( 2172 NextLineIndex, TailOffset, ContentStartColumn); 2173 // Adapt the start of the token, for example indent. 2174 if (!DryRun) 2175 Token->adaptStartOfLine(NextLineIndex, Whitespaces); 2176 } else { 2177 // If we found a reflow split and have added a new break before the next 2178 // line, we are going to remove the line break at the start of the next 2179 // logical line. For example, here we'll add a new line break after 2180 // 'text', and subsequently delete the line break between 'that' and 2181 // 'reflows'. 2182 // // some text that 2183 // // reflows 2184 // -> 2185 // // some text 2186 // // that reflows 2187 // When adding the line break, we also added the penalty for it, so we 2188 // need to subtract that penalty again when we remove the line break due 2189 // to reflowing. 2190 if (NewBreakBefore) { 2191 assert(Penalty >= NewBreakPenalty); 2192 Penalty -= NewBreakPenalty; 2193 } 2194 if (!DryRun) 2195 Token->reflow(NextLineIndex, Whitespaces); 2196 } 2197 } 2198 } 2199 2200 BreakableToken::Split SplitAfterLastLine = 2201 Token->getSplitAfterLastLine(TailOffset); 2202 if (SplitAfterLastLine.first != StringRef::npos) { 2203 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n"); 2204 2205 // We add the last line's penalty here, since that line is going to be split 2206 // now. 2207 Penalty += Style.PenaltyExcessCharacter * 2208 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2209 2210 if (!DryRun) 2211 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine, 2212 Whitespaces); 2213 ContentStartColumn = 2214 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true); 2215 RemainingTokenColumns = Token->getRemainingLength( 2216 Token->getLineCount() - 1, 2217 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second, 2218 ContentStartColumn); 2219 } 2220 2221 State.Column = ContentStartColumn + RemainingTokenColumns - 2222 Current.UnbreakableTailLength; 2223 2224 if (BreakInserted) { 2225 // If we break the token inside a parameter list, we need to break before 2226 // the next parameter on all levels, so that the next parameter is clearly 2227 // visible. Line comments already introduce a break. 2228 if (Current.isNot(TT_LineComment)) { 2229 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 2230 State.Stack[i].BreakBeforeParameter = true; 2231 } 2232 2233 if (Current.is(TT_BlockComment)) 2234 State.NoContinuation = true; 2235 2236 State.Stack.back().LastSpace = StartColumn; 2237 } 2238 2239 Token->updateNextToken(State); 2240 2241 return {Penalty, Exceeded}; 2242 } 2243 2244 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const { 2245 // In preprocessor directives reserve two chars for trailing " \" 2246 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0); 2247 } 2248 2249 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) { 2250 const FormatToken &Current = *State.NextToken; 2251 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral)) 2252 return false; 2253 // We never consider raw string literals "multiline" for the purpose of 2254 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased 2255 // (see TokenAnnotator::mustBreakBefore(). 2256 if (Current.TokenText.startswith("R\"")) 2257 return false; 2258 if (Current.IsMultiline) 2259 return true; 2260 if (Current.getNextNonComment() && 2261 Current.getNextNonComment()->isStringLiteral()) 2262 return true; // Implicit concatenation. 2263 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals && 2264 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength > 2265 Style.ColumnLimit) 2266 return true; // String will be split. 2267 return false; 2268 } 2269 2270 } // namespace format 2271 } // namespace clang 2272