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