1 //===--- ContinuationIndenter.cpp - Format C++ code -----------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// 9 /// \file 10 /// This file implements the continuation indenter. 11 /// 12 //===----------------------------------------------------------------------===// 13 14 #include "ContinuationIndenter.h" 15 #include "BreakableToken.h" 16 #include "FormatInternal.h" 17 #include "WhitespaceManager.h" 18 #include "clang/Basic/OperatorPrecedence.h" 19 #include "clang/Basic/SourceManager.h" 20 #include "clang/Format/Format.h" 21 #include "llvm/Support/Debug.h" 22 23 #define DEBUG_TYPE "format-indenter" 24 25 namespace clang { 26 namespace format { 27 28 // Returns true if a TT_SelectorName should be indented when wrapped, 29 // false otherwise. 30 static bool shouldIndentWrappedSelectorName(const FormatStyle &Style, 31 LineType LineType) { 32 return Style.IndentWrappedFunctionNames || LineType == LT_ObjCMethodDecl; 33 } 34 35 // Returns the length of everything up to the first possible line break after 36 // the ), ], } or > matching \c Tok. 37 static unsigned getLengthToMatchingParen(const FormatToken &Tok, 38 const std::vector<ParenState> &Stack) { 39 // Normally whether or not a break before T is possible is calculated and 40 // stored in T.CanBreakBefore. Braces, array initializers and text proto 41 // messages like `key: < ... >` are an exception: a break is possible 42 // before a closing brace R if a break was inserted after the corresponding 43 // opening brace. The information about whether or not a break is needed 44 // before a closing brace R is stored in the ParenState field 45 // S.BreakBeforeClosingBrace where S is the state that R closes. 46 // 47 // In order to decide whether there can be a break before encountered right 48 // braces, this implementation iterates over the sequence of tokens and over 49 // the paren stack in lockstep, keeping track of the stack level which visited 50 // right braces correspond to in MatchingStackIndex. 51 // 52 // For example, consider: 53 // L. <- line number 54 // 1. { 55 // 2. {1}, 56 // 3. {2}, 57 // 4. {{3}}} 58 // ^ where we call this method with this token. 59 // The paren stack at this point contains 3 brace levels: 60 // 0. { at line 1, BreakBeforeClosingBrace: true 61 // 1. first { at line 4, BreakBeforeClosingBrace: false 62 // 2. second { at line 4, BreakBeforeClosingBrace: false, 63 // where there might be fake parens levels in-between these levels. 64 // The algorithm will start at the first } on line 4, which is the matching 65 // brace of the initial left brace and at level 2 of the stack. Then, 66 // examining BreakBeforeClosingBrace: false at level 2, it will continue to 67 // the second } on line 4, and will traverse the stack downwards until it 68 // finds the matching { on level 1. Then, examining BreakBeforeClosingBrace: 69 // false at level 1, it will continue to the third } on line 4 and will 70 // traverse the stack downwards until it finds the matching { on level 0. 71 // Then, examining BreakBeforeClosingBrace: true at level 0, the algorithm 72 // will stop and will use the second } on line 4 to determine the length to 73 // return, as in this example the range will include the tokens: {3}} 74 // 75 // The algorithm will only traverse the stack if it encounters braces, array 76 // initializer squares or text proto angle brackets. 77 if (!Tok.MatchingParen) 78 return 0; 79 FormatToken *End = Tok.MatchingParen; 80 // Maintains a stack level corresponding to the current End token. 81 int MatchingStackIndex = Stack.size() - 1; 82 // Traverses the stack downwards, looking for the level to which LBrace 83 // corresponds. Returns either a pointer to the matching level or nullptr if 84 // LParen is not found in the initial portion of the stack up to 85 // MatchingStackIndex. 86 auto FindParenState = [&](const FormatToken *LBrace) -> const ParenState * { 87 while (MatchingStackIndex >= 0 && Stack[MatchingStackIndex].Tok != LBrace) 88 --MatchingStackIndex; 89 return MatchingStackIndex >= 0 ? &Stack[MatchingStackIndex] : nullptr; 90 }; 91 for (; End->Next; End = End->Next) { 92 if (End->Next->CanBreakBefore) 93 break; 94 if (!End->Next->closesScope()) 95 continue; 96 if (End->Next->MatchingParen && 97 End->Next->MatchingParen->isOneOf( 98 tok::l_brace, TT_ArrayInitializerLSquare, tok::less)) { 99 const ParenState *State = FindParenState(End->Next->MatchingParen); 100 if (State && State->BreakBeforeClosingBrace) 101 break; 102 } 103 } 104 return End->TotalLength - Tok.TotalLength + 1; 105 } 106 107 static unsigned getLengthToNextOperator(const FormatToken &Tok) { 108 if (!Tok.NextOperator) 109 return 0; 110 return Tok.NextOperator->TotalLength - Tok.TotalLength; 111 } 112 113 // Returns \c true if \c Tok is the "." or "->" of a call and starts the next 114 // segment of a builder type call. 115 static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) { 116 return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope(); 117 } 118 119 // Returns \c true if \c Current starts a new parameter. 120 static bool startsNextParameter(const FormatToken &Current, 121 const FormatStyle &Style) { 122 const FormatToken &Previous = *Current.Previous; 123 if (Current.is(TT_CtorInitializerComma) && 124 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 125 return true; 126 if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName)) 127 return true; 128 return Previous.is(tok::comma) && !Current.isTrailingComment() && 129 ((Previous.isNot(TT_CtorInitializerComma) || 130 Style.BreakConstructorInitializers != 131 FormatStyle::BCIS_BeforeComma) && 132 (Previous.isNot(TT_InheritanceComma) || 133 Style.BreakInheritanceList != FormatStyle::BILS_BeforeComma)); 134 } 135 136 static bool opensProtoMessageField(const FormatToken &LessTok, 137 const FormatStyle &Style) { 138 if (LessTok.isNot(tok::less)) 139 return false; 140 return Style.Language == FormatStyle::LK_TextProto || 141 (Style.Language == FormatStyle::LK_Proto && 142 (LessTok.NestingLevel > 0 || 143 (LessTok.Previous && LessTok.Previous->is(tok::equal)))); 144 } 145 146 // Returns the delimiter of a raw string literal, or None if TokenText is not 147 // the text of a raw string literal. The delimiter could be the empty string. 148 // For example, the delimiter of R"deli(cont)deli" is deli. 149 static llvm::Optional<StringRef> getRawStringDelimiter(StringRef TokenText) { 150 if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'. 151 || !TokenText.startswith("R\"") || !TokenText.endswith("\"")) 152 return None; 153 154 // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has 155 // size at most 16 by the standard, so the first '(' must be among the first 156 // 19 bytes. 157 size_t LParenPos = TokenText.substr(0, 19).find_first_of('('); 158 if (LParenPos == StringRef::npos) 159 return None; 160 StringRef Delimiter = TokenText.substr(2, LParenPos - 2); 161 162 // Check that the string ends in ')Delimiter"'. 163 size_t RParenPos = TokenText.size() - Delimiter.size() - 2; 164 if (TokenText[RParenPos] != ')') 165 return None; 166 if (!TokenText.substr(RParenPos + 1).startswith(Delimiter)) 167 return None; 168 return Delimiter; 169 } 170 171 // Returns the canonical delimiter for \p Language, or the empty string if no 172 // canonical delimiter is specified. 173 static StringRef 174 getCanonicalRawStringDelimiter(const FormatStyle &Style, 175 FormatStyle::LanguageKind Language) { 176 for (const auto &Format : Style.RawStringFormats) { 177 if (Format.Language == Language) 178 return StringRef(Format.CanonicalDelimiter); 179 } 180 return ""; 181 } 182 183 RawStringFormatStyleManager::RawStringFormatStyleManager( 184 const FormatStyle &CodeStyle) { 185 for (const auto &RawStringFormat : CodeStyle.RawStringFormats) { 186 llvm::Optional<FormatStyle> LanguageStyle = 187 CodeStyle.GetLanguageStyle(RawStringFormat.Language); 188 if (!LanguageStyle) { 189 FormatStyle PredefinedStyle; 190 if (!getPredefinedStyle(RawStringFormat.BasedOnStyle, 191 RawStringFormat.Language, &PredefinedStyle)) { 192 PredefinedStyle = getLLVMStyle(); 193 PredefinedStyle.Language = RawStringFormat.Language; 194 } 195 LanguageStyle = PredefinedStyle; 196 } 197 LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit; 198 for (StringRef Delimiter : RawStringFormat.Delimiters) { 199 DelimiterStyle.insert({Delimiter, *LanguageStyle}); 200 } 201 for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions) { 202 EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle}); 203 } 204 } 205 } 206 207 llvm::Optional<FormatStyle> 208 RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const { 209 auto It = DelimiterStyle.find(Delimiter); 210 if (It == DelimiterStyle.end()) 211 return None; 212 return It->second; 213 } 214 215 llvm::Optional<FormatStyle> 216 RawStringFormatStyleManager::getEnclosingFunctionStyle( 217 StringRef EnclosingFunction) const { 218 auto It = EnclosingFunctionStyle.find(EnclosingFunction); 219 if (It == EnclosingFunctionStyle.end()) 220 return None; 221 return It->second; 222 } 223 224 ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style, 225 const AdditionalKeywords &Keywords, 226 const SourceManager &SourceMgr, 227 WhitespaceManager &Whitespaces, 228 encoding::Encoding Encoding, 229 bool BinPackInconclusiveFunctions) 230 : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr), 231 Whitespaces(Whitespaces), Encoding(Encoding), 232 BinPackInconclusiveFunctions(BinPackInconclusiveFunctions), 233 CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {} 234 235 LineState ContinuationIndenter::getInitialState(unsigned FirstIndent, 236 unsigned FirstStartColumn, 237 const AnnotatedLine *Line, 238 bool DryRun) { 239 LineState State; 240 State.FirstIndent = FirstIndent; 241 if (FirstStartColumn && Line->First->NewlinesBefore == 0) 242 State.Column = FirstStartColumn; 243 else 244 State.Column = FirstIndent; 245 // With preprocessor directive indentation, the line starts on column 0 246 // since it's indented after the hash, but FirstIndent is set to the 247 // preprocessor indent. 248 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 249 (Line->Type == LT_PreprocessorDirective || 250 Line->Type == LT_ImportStatement)) 251 State.Column = 0; 252 State.Line = Line; 253 State.NextToken = Line->First; 254 State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent, 255 /*AvoidBinPacking=*/false, 256 /*NoLineBreak=*/false)); 257 State.LineContainsContinuedForLoopSection = false; 258 State.NoContinuation = false; 259 State.StartOfStringLiteral = 0; 260 State.StartOfLineLevel = 0; 261 State.LowestLevelOnLine = 0; 262 State.IgnoreStackForComparison = false; 263 264 if (Style.Language == FormatStyle::LK_TextProto) { 265 // We need this in order to deal with the bin packing of text fields at 266 // global scope. 267 State.Stack.back().AvoidBinPacking = true; 268 State.Stack.back().BreakBeforeParameter = true; 269 State.Stack.back().AlignColons = false; 270 } 271 272 // The first token has already been indented and thus consumed. 273 moveStateToNextToken(State, DryRun, /*Newline=*/false); 274 return State; 275 } 276 277 bool ContinuationIndenter::canBreak(const LineState &State) { 278 const FormatToken &Current = *State.NextToken; 279 const FormatToken &Previous = *Current.Previous; 280 assert(&Previous == Current.Previous); 281 if (!Current.CanBreakBefore && !(State.Stack.back().BreakBeforeClosingBrace && 282 Current.closesBlockOrBlockTypeList(Style))) 283 return false; 284 // The opening "{" of a braced list has to be on the same line as the first 285 // element if it is nested in another braced init list or function call. 286 if (!Current.MustBreakBefore && Previous.is(tok::l_brace) && 287 Previous.isNot(TT_DictLiteral) && Previous.BlockKind == BK_BracedInit && 288 Previous.Previous && 289 Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma)) 290 return false; 291 // This prevents breaks like: 292 // ... 293 // SomeParameter, OtherParameter).DoSomething( 294 // ... 295 // As they hide "DoSomething" and are generally bad for readability. 296 if (Previous.opensScope() && Previous.isNot(tok::l_brace) && 297 State.LowestLevelOnLine < State.StartOfLineLevel && 298 State.LowestLevelOnLine < Current.NestingLevel) 299 return false; 300 if (Current.isMemberAccess() && State.Stack.back().ContainsUnwrappedBuilder) 301 return false; 302 303 // Don't create a 'hanging' indent if there are multiple blocks in a single 304 // statement. 305 if (Previous.is(tok::l_brace) && State.Stack.size() > 1 && 306 State.Stack[State.Stack.size() - 2].NestedBlockInlined && 307 State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks) 308 return false; 309 310 // Don't break after very short return types (e.g. "void") as that is often 311 // unexpected. 312 if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) { 313 if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None) 314 return false; 315 } 316 317 // If binary operators are moved to the next line (including commas for some 318 // styles of constructor initializers), that's always ok. 319 if (!Current.isOneOf(TT_BinaryOperator, tok::comma) && 320 State.Stack.back().NoLineBreakInOperand) 321 return false; 322 323 if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr)) 324 return false; 325 326 return !State.Stack.back().NoLineBreak; 327 } 328 329 bool ContinuationIndenter::mustBreak(const LineState &State) { 330 const FormatToken &Current = *State.NextToken; 331 const FormatToken &Previous = *Current.Previous; 332 if (Style.BraceWrapping.BeforeLambdaBody && Current.CanBreakBefore && 333 Current.is(TT_LambdaLBrace)) { 334 auto LambdaBodyLength = getLengthToMatchingParen(Current, State.Stack); 335 return (LambdaBodyLength > getColumnLimit(State)); 336 } 337 if (Current.MustBreakBefore || Current.is(TT_InlineASMColon)) 338 return true; 339 if (State.Stack.back().BreakBeforeClosingBrace && 340 Current.closesBlockOrBlockTypeList(Style)) 341 return true; 342 if (Previous.is(tok::semi) && State.LineContainsContinuedForLoopSection) 343 return true; 344 if (Style.Language == FormatStyle::LK_ObjC && 345 Current.ObjCSelectorNameParts > 1 && 346 Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) { 347 return true; 348 } 349 if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) || 350 (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) && 351 Style.isCpp() && 352 // FIXME: This is a temporary workaround for the case where clang-format 353 // sets BreakBeforeParameter to avoid bin packing and this creates a 354 // completely unnecessary line break after a template type that isn't 355 // line-wrapped. 356 (Previous.NestingLevel == 1 || Style.BinPackParameters)) || 357 (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) && 358 Previous.isNot(tok::question)) || 359 (!Style.BreakBeforeTernaryOperators && 360 Previous.is(TT_ConditionalExpr))) && 361 State.Stack.back().BreakBeforeParameter && !Current.isTrailingComment() && 362 !Current.isOneOf(tok::r_paren, tok::r_brace)) 363 return true; 364 if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) || 365 (Previous.is(TT_ArrayInitializerLSquare) && 366 Previous.ParameterCount > 1) || 367 opensProtoMessageField(Previous, Style)) && 368 Style.ColumnLimit > 0 && 369 getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 > 370 getColumnLimit(State)) 371 return true; 372 373 const FormatToken &BreakConstructorInitializersToken = 374 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon 375 ? Previous 376 : Current; 377 if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) && 378 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength > 379 getColumnLimit(State) || 380 State.Stack.back().BreakBeforeParameter) && 381 (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All || 382 Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon || 383 Style.ColumnLimit != 0)) 384 return true; 385 386 if (Current.is(TT_ObjCMethodExpr) && !Previous.is(TT_SelectorName) && 387 State.Line->startsWith(TT_ObjCMethodSpecifier)) 388 return true; 389 if (Current.is(TT_SelectorName) && !Previous.is(tok::at) && 390 State.Stack.back().ObjCSelectorNameFound && 391 State.Stack.back().BreakBeforeParameter) 392 return true; 393 394 unsigned NewLineColumn = getNewLineColumn(State); 395 if (Current.isMemberAccess() && Style.ColumnLimit != 0 && 396 State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit && 397 (State.Column > NewLineColumn || 398 Current.NestingLevel < State.StartOfLineLevel)) 399 return true; 400 401 if (startsSegmentOfBuilderTypeCall(Current) && 402 (State.Stack.back().CallContinuation != 0 || 403 State.Stack.back().BreakBeforeParameter) && 404 // JavaScript is treated different here as there is a frequent pattern: 405 // SomeFunction(function() { 406 // ... 407 // }.bind(...)); 408 // FIXME: We should find a more generic solution to this problem. 409 !(State.Column <= NewLineColumn && 410 Style.Language == FormatStyle::LK_JavaScript) && 411 !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) 412 return true; 413 414 // If the template declaration spans multiple lines, force wrap before the 415 // function/class declaration 416 if (Previous.ClosesTemplateDeclaration && 417 State.Stack.back().BreakBeforeParameter && Current.CanBreakBefore) 418 return true; 419 420 if (State.Column <= NewLineColumn) 421 return false; 422 423 if (Style.AlwaysBreakBeforeMultilineStrings && 424 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth || 425 Previous.is(tok::comma) || Current.NestingLevel < 2) && 426 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at, 427 Keywords.kw_dollar) && 428 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) && 429 nextIsMultilineString(State)) 430 return true; 431 432 // Using CanBreakBefore here and below takes care of the decision whether the 433 // current style uses wrapping before or after operators for the given 434 // operator. 435 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) { 436 // If we need to break somewhere inside the LHS of a binary expression, we 437 // should also break after the operator. Otherwise, the formatting would 438 // hide the operator precedence, e.g. in: 439 // if (aaaaaaaaaaaaaa == 440 // bbbbbbbbbbbbbb && c) {.. 441 // For comparisons, we only apply this rule, if the LHS is a binary 442 // expression itself as otherwise, the line breaks seem superfluous. 443 // We need special cases for ">>" which we have split into two ">" while 444 // lexing in order to make template parsing easier. 445 bool IsComparison = (Previous.getPrecedence() == prec::Relational || 446 Previous.getPrecedence() == prec::Equality || 447 Previous.getPrecedence() == prec::Spaceship) && 448 Previous.Previous && 449 Previous.Previous->isNot(TT_BinaryOperator); // For >>. 450 bool LHSIsBinaryExpr = 451 Previous.Previous && Previous.Previous->EndsBinaryExpression; 452 if ((!IsComparison || LHSIsBinaryExpr) && !Current.isTrailingComment() && 453 Previous.getPrecedence() != prec::Assignment && 454 State.Stack.back().BreakBeforeParameter) 455 return true; 456 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore && 457 State.Stack.back().BreakBeforeParameter) { 458 return true; 459 } 460 461 // Same as above, but for the first "<<" operator. 462 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) && 463 State.Stack.back().BreakBeforeParameter && 464 State.Stack.back().FirstLessLess == 0) 465 return true; 466 467 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) { 468 // Always break after "template <...>" and leading annotations. This is only 469 // for cases where the entire line does not fit on a single line as a 470 // different LineFormatter would be used otherwise. 471 if (Previous.ClosesTemplateDeclaration) 472 return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No; 473 if (Previous.is(TT_FunctionAnnotationRParen)) 474 return true; 475 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) && 476 Current.isNot(TT_LeadingJavaAnnotation)) 477 return true; 478 } 479 480 // If the return type spans multiple lines, wrap before the function name. 481 if (((Current.is(TT_FunctionDeclarationName) && 482 // Don't break before a C# function when no break after return type 483 (!Style.isCSharp() || 484 Style.AlwaysBreakAfterReturnType != FormatStyle::RTBS_None)) || 485 (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) && 486 !Previous.is(tok::kw_template) && State.Stack.back().BreakBeforeParameter) 487 return true; 488 489 // The following could be precomputed as they do not depend on the state. 490 // However, as they should take effect only if the UnwrappedLine does not fit 491 // into the ColumnLimit, they are checked here in the ContinuationIndenter. 492 if (Style.ColumnLimit != 0 && Previous.BlockKind == BK_Block && 493 Previous.is(tok::l_brace) && !Current.isOneOf(tok::r_brace, tok::comment)) 494 return true; 495 496 if (Current.is(tok::lessless) && 497 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") || 498 (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") || 499 Previous.TokenText == "\'\\n\'")))) 500 return true; 501 502 if (Previous.is(TT_BlockComment) && Previous.IsMultiline) 503 return true; 504 505 if (State.NoContinuation) 506 return true; 507 508 return false; 509 } 510 511 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline, 512 bool DryRun, 513 unsigned ExtraSpaces) { 514 const FormatToken &Current = *State.NextToken; 515 516 assert(!State.Stack.empty()); 517 State.NoContinuation = false; 518 519 if ((Current.is(TT_ImplicitStringLiteral) && 520 (Current.Previous->Tok.getIdentifierInfo() == nullptr || 521 Current.Previous->Tok.getIdentifierInfo()->getPPKeywordID() == 522 tok::pp_not_keyword))) { 523 unsigned EndColumn = 524 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd()); 525 if (Current.LastNewlineOffset != 0) { 526 // If there is a newline within this token, the final column will solely 527 // determined by the current end column. 528 State.Column = EndColumn; 529 } else { 530 unsigned StartColumn = 531 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin()); 532 assert(EndColumn >= StartColumn); 533 State.Column += EndColumn - StartColumn; 534 } 535 moveStateToNextToken(State, DryRun, /*Newline=*/false); 536 return 0; 537 } 538 539 unsigned Penalty = 0; 540 if (Newline) 541 Penalty = addTokenOnNewLine(State, DryRun); 542 else 543 addTokenOnCurrentLine(State, DryRun, ExtraSpaces); 544 545 return moveStateToNextToken(State, DryRun, Newline) + Penalty; 546 } 547 548 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun, 549 unsigned ExtraSpaces) { 550 FormatToken &Current = *State.NextToken; 551 const FormatToken &Previous = *State.NextToken->Previous; 552 if (Current.is(tok::equal) && 553 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) && 554 State.Stack.back().VariablePos == 0) { 555 State.Stack.back().VariablePos = State.Column; 556 // Move over * and & if they are bound to the variable name. 557 const FormatToken *Tok = &Previous; 558 while (Tok && State.Stack.back().VariablePos >= Tok->ColumnWidth) { 559 State.Stack.back().VariablePos -= Tok->ColumnWidth; 560 if (Tok->SpacesRequiredBefore != 0) 561 break; 562 Tok = Tok->Previous; 563 } 564 if (Previous.PartOfMultiVariableDeclStmt) 565 State.Stack.back().LastSpace = State.Stack.back().VariablePos; 566 } 567 568 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces; 569 570 // Indent preprocessor directives after the hash if required. 571 int PPColumnCorrection = 0; 572 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 573 Previous.is(tok::hash) && State.FirstIndent > 0 && 574 (State.Line->Type == LT_PreprocessorDirective || 575 State.Line->Type == LT_ImportStatement)) { 576 Spaces += State.FirstIndent; 577 578 // For preprocessor indent with tabs, State.Column will be 1 because of the 579 // hash. This causes second-level indents onward to have an extra space 580 // after the tabs. We avoid this misalignment by subtracting 1 from the 581 // column value passed to replaceWhitespace(). 582 if (Style.UseTab != FormatStyle::UT_Never) 583 PPColumnCorrection = -1; 584 } 585 586 if (!DryRun) 587 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces, 588 State.Column + Spaces + PPColumnCorrection); 589 590 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance 591 // declaration unless there is multiple inheritance. 592 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 593 Current.is(TT_InheritanceColon)) 594 State.Stack.back().NoLineBreak = true; 595 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon && 596 Previous.is(TT_InheritanceColon)) 597 State.Stack.back().NoLineBreak = true; 598 599 if (Current.is(TT_SelectorName) && 600 !State.Stack.back().ObjCSelectorNameFound) { 601 unsigned MinIndent = 602 std::max(State.FirstIndent + Style.ContinuationIndentWidth, 603 State.Stack.back().Indent); 604 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth; 605 if (Current.LongestObjCSelectorName == 0) 606 State.Stack.back().AlignColons = false; 607 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos) 608 State.Stack.back().ColonPos = MinIndent + Current.LongestObjCSelectorName; 609 else 610 State.Stack.back().ColonPos = FirstColonPos; 611 } 612 613 // In "AlwaysBreak" mode, enforce wrapping directly after the parenthesis by 614 // disallowing any further line breaks if there is no line break after the 615 // opening parenthesis. Don't break if it doesn't conserve columns. 616 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak && 617 (Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) || 618 (Previous.is(tok::l_brace) && Previous.BlockKind != BK_Block && 619 Style.Cpp11BracedListStyle)) && 620 State.Column > getNewLineColumn(State) && 621 (!Previous.Previous || !Previous.Previous->isOneOf( 622 tok::kw_for, tok::kw_while, tok::kw_switch)) && 623 // Don't do this for simple (no expressions) one-argument function calls 624 // as that feels like needlessly wasting whitespace, e.g.: 625 // 626 // caaaaaaaaaaaall( 627 // caaaaaaaaaaaall( 628 // caaaaaaaaaaaall( 629 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa)))); 630 Current.FakeLParens.size() > 0 && 631 Current.FakeLParens.back() > prec::Unknown) 632 State.Stack.back().NoLineBreak = true; 633 if (Previous.is(TT_TemplateString) && Previous.opensScope()) 634 State.Stack.back().NoLineBreak = true; 635 636 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign && 637 !State.Stack.back().IsCSharpGenericTypeConstraint && 638 Previous.opensScope() && Previous.isNot(TT_ObjCMethodExpr) && 639 (Current.isNot(TT_LineComment) || Previous.BlockKind == BK_BracedInit)) 640 State.Stack.back().Indent = State.Column + Spaces; 641 if (State.Stack.back().AvoidBinPacking && startsNextParameter(Current, Style)) 642 State.Stack.back().NoLineBreak = true; 643 if (startsSegmentOfBuilderTypeCall(Current) && 644 State.Column > getNewLineColumn(State)) 645 State.Stack.back().ContainsUnwrappedBuilder = true; 646 647 if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java) 648 State.Stack.back().NoLineBreak = true; 649 if (Current.isMemberAccess() && Previous.is(tok::r_paren) && 650 (Previous.MatchingParen && 651 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) 652 // If there is a function call with long parameters, break before trailing 653 // calls. This prevents things like: 654 // EXPECT_CALL(SomeLongParameter).Times( 655 // 2); 656 // We don't want to do this for short parameters as they can just be 657 // indexes. 658 State.Stack.back().NoLineBreak = true; 659 660 // Don't allow the RHS of an operator to be split over multiple lines unless 661 // there is a line-break right after the operator. 662 // Exclude relational operators, as there, it is always more desirable to 663 // have the LHS 'left' of the RHS. 664 const FormatToken *P = Current.getPreviousNonComment(); 665 if (!Current.is(tok::comment) && P && 666 (P->isOneOf(TT_BinaryOperator, tok::comma) || 667 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) && 668 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) && 669 P->getPrecedence() != prec::Assignment && 670 P->getPrecedence() != prec::Relational && 671 P->getPrecedence() != prec::Spaceship) { 672 bool BreakBeforeOperator = 673 P->MustBreakBefore || P->is(tok::lessless) || 674 (P->is(TT_BinaryOperator) && 675 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) || 676 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators); 677 // Don't do this if there are only two operands. In these cases, there is 678 // always a nice vertical separation between them and the extra line break 679 // does not help. 680 bool HasTwoOperands = 681 P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr); 682 if ((!BreakBeforeOperator && !(HasTwoOperands && Style.AlignOperands)) || 683 (!State.Stack.back().LastOperatorWrapped && BreakBeforeOperator)) 684 State.Stack.back().NoLineBreakInOperand = true; 685 } 686 687 State.Column += Spaces; 688 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) && 689 Previous.Previous && 690 (Previous.Previous->is(tok::kw_for) || Previous.Previous->isIf())) { 691 // Treat the condition inside an if as if it was a second function 692 // parameter, i.e. let nested calls have a continuation indent. 693 State.Stack.back().LastSpace = State.Column; 694 State.Stack.back().NestedBlockIndent = State.Column; 695 } else if (!Current.isOneOf(tok::comment, tok::caret) && 696 ((Previous.is(tok::comma) && 697 !Previous.is(TT_OverloadedOperator)) || 698 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) { 699 State.Stack.back().LastSpace = State.Column; 700 } else if (Previous.is(TT_CtorInitializerColon) && 701 Style.BreakConstructorInitializers == 702 FormatStyle::BCIS_AfterColon) { 703 State.Stack.back().Indent = State.Column; 704 State.Stack.back().LastSpace = State.Column; 705 } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 706 TT_CtorInitializerColon)) && 707 ((Previous.getPrecedence() != prec::Assignment && 708 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 || 709 Previous.NextOperator)) || 710 Current.StartsBinaryExpression)) { 711 // Indent relative to the RHS of the expression unless this is a simple 712 // assignment without binary expression on the RHS. Also indent relative to 713 // unary operators and the colons of constructor initializers. 714 if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None) 715 State.Stack.back().LastSpace = State.Column; 716 } else if (Previous.is(TT_InheritanceColon)) { 717 State.Stack.back().Indent = State.Column; 718 State.Stack.back().LastSpace = State.Column; 719 } else if (Current.is(TT_CSharpGenericTypeConstraintColon)) { 720 State.Stack.back().ColonPos = State.Column; 721 } else if (Previous.opensScope()) { 722 // If a function has a trailing call, indent all parameters from the 723 // opening parenthesis. This avoids confusing indents like: 724 // OuterFunction(InnerFunctionCall( // break 725 // ParameterToInnerFunction)) // break 726 // .SecondInnerFunctionCall(); 727 bool HasTrailingCall = false; 728 if (Previous.MatchingParen) { 729 const FormatToken *Next = Previous.MatchingParen->getNextNonComment(); 730 HasTrailingCall = Next && Next->isMemberAccess(); 731 } 732 if (HasTrailingCall && State.Stack.size() > 1 && 733 State.Stack[State.Stack.size() - 2].CallContinuation == 0) 734 State.Stack.back().LastSpace = State.Column; 735 } 736 } 737 738 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State, 739 bool DryRun) { 740 FormatToken &Current = *State.NextToken; 741 const FormatToken &Previous = *State.NextToken->Previous; 742 743 // Extra penalty that needs to be added because of the way certain line 744 // breaks are chosen. 745 unsigned Penalty = 0; 746 747 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 748 const FormatToken *NextNonComment = Previous.getNextNonComment(); 749 if (!NextNonComment) 750 NextNonComment = &Current; 751 // The first line break on any NestingLevel causes an extra penalty in order 752 // prefer similar line breaks. 753 if (!State.Stack.back().ContainsLineBreak) 754 Penalty += 15; 755 State.Stack.back().ContainsLineBreak = true; 756 757 Penalty += State.NextToken->SplitPenalty; 758 759 // Breaking before the first "<<" is generally not desirable if the LHS is 760 // short. Also always add the penalty if the LHS is split over multiple lines 761 // to avoid unnecessary line breaks that just work around this penalty. 762 if (NextNonComment->is(tok::lessless) && 763 State.Stack.back().FirstLessLess == 0 && 764 (State.Column <= Style.ColumnLimit / 3 || 765 State.Stack.back().BreakBeforeParameter)) 766 Penalty += Style.PenaltyBreakFirstLessLess; 767 768 State.Column = getNewLineColumn(State); 769 770 // Indent nested blocks relative to this column, unless in a very specific 771 // JavaScript special case where: 772 // 773 // var loooooong_name = 774 // function() { 775 // // code 776 // } 777 // 778 // is common and should be formatted like a free-standing function. The same 779 // goes for wrapping before the lambda return type arrow. 780 if (!Current.is(TT_LambdaArrow) && 781 (Style.Language != FormatStyle::LK_JavaScript || 782 Current.NestingLevel != 0 || !PreviousNonComment || 783 !PreviousNonComment->is(tok::equal) || 784 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) 785 State.Stack.back().NestedBlockIndent = State.Column; 786 787 if (NextNonComment->isMemberAccess()) { 788 if (State.Stack.back().CallContinuation == 0) 789 State.Stack.back().CallContinuation = State.Column; 790 } else if (NextNonComment->is(TT_SelectorName)) { 791 if (!State.Stack.back().ObjCSelectorNameFound) { 792 if (NextNonComment->LongestObjCSelectorName == 0) { 793 State.Stack.back().AlignColons = false; 794 } else { 795 State.Stack.back().ColonPos = 796 (shouldIndentWrappedSelectorName(Style, State.Line->Type) 797 ? std::max(State.Stack.back().Indent, 798 State.FirstIndent + Style.ContinuationIndentWidth) 799 : State.Stack.back().Indent) + 800 std::max(NextNonComment->LongestObjCSelectorName, 801 NextNonComment->ColumnWidth); 802 } 803 } else if (State.Stack.back().AlignColons && 804 State.Stack.back().ColonPos <= NextNonComment->ColumnWidth) { 805 State.Stack.back().ColonPos = State.Column + NextNonComment->ColumnWidth; 806 } 807 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 808 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) { 809 // FIXME: This is hacky, find a better way. The problem is that in an ObjC 810 // method expression, the block should be aligned to the line starting it, 811 // e.g.: 812 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason 813 // ^(int *i) { 814 // // ... 815 // }]; 816 // Thus, we set LastSpace of the next higher NestingLevel, to which we move 817 // when we consume all of the "}"'s FakeRParens at the "{". 818 if (State.Stack.size() > 1) 819 State.Stack[State.Stack.size() - 2].LastSpace = 820 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 821 Style.ContinuationIndentWidth; 822 } 823 824 if ((PreviousNonComment && 825 PreviousNonComment->isOneOf(tok::comma, tok::semi) && 826 !State.Stack.back().AvoidBinPacking) || 827 Previous.is(TT_BinaryOperator)) 828 State.Stack.back().BreakBeforeParameter = false; 829 if (PreviousNonComment && 830 PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) && 831 Current.NestingLevel == 0) 832 State.Stack.back().BreakBeforeParameter = false; 833 if (NextNonComment->is(tok::question) || 834 (PreviousNonComment && PreviousNonComment->is(tok::question))) 835 State.Stack.back().BreakBeforeParameter = true; 836 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore) 837 State.Stack.back().BreakBeforeParameter = false; 838 839 if (!DryRun) { 840 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1; 841 if (Current.is(tok::r_brace) && Current.MatchingParen && 842 // Only strip trailing empty lines for l_braces that have children, i.e. 843 // for function expressions (lambdas, arrows, etc). 844 !Current.MatchingParen->Children.empty()) { 845 // lambdas and arrow functions are expressions, thus their r_brace is not 846 // on its own line, and thus not covered by UnwrappedLineFormatter's logic 847 // about removing empty lines on closing blocks. Special case them here. 848 MaxEmptyLinesToKeep = 1; 849 } 850 unsigned Newlines = 851 std::max(1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep)); 852 bool ContinuePPDirective = 853 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement; 854 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column, 855 ContinuePPDirective); 856 } 857 858 if (!Current.isTrailingComment()) 859 State.Stack.back().LastSpace = State.Column; 860 if (Current.is(tok::lessless)) 861 // If we are breaking before a "<<", we always want to indent relative to 862 // RHS. This is necessary only for "<<", as we special-case it and don't 863 // always indent relative to the RHS. 864 State.Stack.back().LastSpace += 3; // 3 -> width of "<< ". 865 866 State.StartOfLineLevel = Current.NestingLevel; 867 State.LowestLevelOnLine = Current.NestingLevel; 868 869 // Any break on this level means that the parent level has been broken 870 // and we need to avoid bin packing there. 871 bool NestedBlockSpecialCase = 872 (!Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 && 873 State.Stack[State.Stack.size() - 2].NestedBlockInlined) || 874 (Style.Language == FormatStyle::LK_ObjC && Current.is(tok::r_brace) && 875 State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam); 876 if (!NestedBlockSpecialCase) 877 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 878 State.Stack[i].BreakBeforeParameter = true; 879 880 if (PreviousNonComment && 881 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) && 882 (PreviousNonComment->isNot(TT_TemplateCloser) || 883 Current.NestingLevel != 0) && 884 !PreviousNonComment->isOneOf( 885 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation, 886 TT_LeadingJavaAnnotation) && 887 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope()) 888 State.Stack.back().BreakBeforeParameter = true; 889 890 // If we break after { or the [ of an array initializer, we should also break 891 // before the corresponding } or ]. 892 if (PreviousNonComment && 893 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 894 opensProtoMessageField(*PreviousNonComment, Style))) 895 State.Stack.back().BreakBeforeClosingBrace = true; 896 897 if (State.Stack.back().AvoidBinPacking) { 898 // If we are breaking after '(', '{', '<', or this is the break after a ':' 899 // to start a member initializater list in a constructor, this should not 900 // be considered bin packing unless the relevant AllowAll option is false or 901 // this is a dict/object literal. 902 bool PreviousIsBreakingCtorInitializerColon = 903 Previous.is(TT_CtorInitializerColon) && 904 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 905 if (!(Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) || 906 PreviousIsBreakingCtorInitializerColon) || 907 (!Style.AllowAllParametersOfDeclarationOnNextLine && 908 State.Line->MustBeDeclaration) || 909 (!Style.AllowAllArgumentsOnNextLine && 910 !State.Line->MustBeDeclaration) || 911 (!Style.AllowAllConstructorInitializersOnNextLine && 912 PreviousIsBreakingCtorInitializerColon) || 913 Previous.is(TT_DictLiteral)) 914 State.Stack.back().BreakBeforeParameter = true; 915 916 // If we are breaking after a ':' to start a member initializer list, 917 // and we allow all arguments on the next line, we should not break 918 // before the next parameter. 919 if (PreviousIsBreakingCtorInitializerColon && 920 Style.AllowAllConstructorInitializersOnNextLine) 921 State.Stack.back().BreakBeforeParameter = false; 922 } 923 924 return Penalty; 925 } 926 927 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) { 928 if (!State.NextToken || !State.NextToken->Previous) 929 return 0; 930 931 FormatToken &Current = *State.NextToken; 932 933 if (State.Stack.back().IsCSharpGenericTypeConstraint && 934 Current.isNot(TT_CSharpGenericTypeConstraint)) 935 return State.Stack.back().ColonPos + 2; 936 937 const FormatToken &Previous = *Current.Previous; 938 // If we are continuing an expression, we want to use the continuation indent. 939 unsigned ContinuationIndent = 940 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 941 Style.ContinuationIndentWidth; 942 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 943 const FormatToken *NextNonComment = Previous.getNextNonComment(); 944 if (!NextNonComment) 945 NextNonComment = &Current; 946 947 // Java specific bits. 948 if (Style.Language == FormatStyle::LK_Java && 949 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) 950 return std::max(State.Stack.back().LastSpace, 951 State.Stack.back().Indent + Style.ContinuationIndentWidth); 952 953 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths && 954 State.Line->First->is(tok::kw_enum)) 955 return (Style.IndentWidth * State.Line->First->IndentLevel) + 956 Style.IndentWidth; 957 958 if (NextNonComment->is(tok::l_brace) && NextNonComment->BlockKind == BK_Block) 959 return Current.NestingLevel == 0 ? State.FirstIndent 960 : State.Stack.back().Indent; 961 if ((Current.isOneOf(tok::r_brace, tok::r_square) || 962 (Current.is(tok::greater) && 963 (Style.Language == FormatStyle::LK_Proto || 964 Style.Language == FormatStyle::LK_TextProto))) && 965 State.Stack.size() > 1) { 966 if (Current.closesBlockOrBlockTypeList(Style)) 967 return State.Stack[State.Stack.size() - 2].NestedBlockIndent; 968 if (Current.MatchingParen && 969 Current.MatchingParen->BlockKind == BK_BracedInit) 970 return State.Stack[State.Stack.size() - 2].LastSpace; 971 return State.FirstIndent; 972 } 973 // Indent a closing parenthesis at the previous level if followed by a semi, 974 // const, or opening brace. This allows indentations such as: 975 // foo( 976 // a, 977 // ); 978 // int Foo::getter( 979 // // 980 // ) const { 981 // return foo; 982 // } 983 // function foo( 984 // a, 985 // ) { 986 // code(); // 987 // } 988 if (Current.is(tok::r_paren) && State.Stack.size() > 1 && 989 (!Current.Next || 990 Current.Next->isOneOf(tok::semi, tok::kw_const, tok::l_brace))) 991 return State.Stack[State.Stack.size() - 2].LastSpace; 992 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope()) 993 return State.Stack[State.Stack.size() - 2].LastSpace; 994 if (Current.is(tok::identifier) && Current.Next && 995 (Current.Next->is(TT_DictLiteral) || 996 ((Style.Language == FormatStyle::LK_Proto || 997 Style.Language == FormatStyle::LK_TextProto) && 998 Current.Next->isOneOf(tok::less, tok::l_brace)))) 999 return State.Stack.back().Indent; 1000 if (NextNonComment->is(TT_ObjCStringLiteral) && 1001 State.StartOfStringLiteral != 0) 1002 return State.StartOfStringLiteral - 1; 1003 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0) 1004 return State.StartOfStringLiteral; 1005 if (NextNonComment->is(tok::lessless) && 1006 State.Stack.back().FirstLessLess != 0) 1007 return State.Stack.back().FirstLessLess; 1008 if (NextNonComment->isMemberAccess()) { 1009 if (State.Stack.back().CallContinuation == 0) 1010 return ContinuationIndent; 1011 return State.Stack.back().CallContinuation; 1012 } 1013 if (State.Stack.back().QuestionColumn != 0 && 1014 ((NextNonComment->is(tok::colon) && 1015 NextNonComment->is(TT_ConditionalExpr)) || 1016 Previous.is(TT_ConditionalExpr))) 1017 return State.Stack.back().QuestionColumn; 1018 if (Previous.is(tok::comma) && State.Stack.back().VariablePos != 0) 1019 return State.Stack.back().VariablePos; 1020 if ((PreviousNonComment && 1021 (PreviousNonComment->ClosesTemplateDeclaration || 1022 PreviousNonComment->isOneOf( 1023 TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen, 1024 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) || 1025 (!Style.IndentWrappedFunctionNames && 1026 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) 1027 return std::max(State.Stack.back().LastSpace, State.Stack.back().Indent); 1028 if (NextNonComment->is(TT_SelectorName)) { 1029 if (!State.Stack.back().ObjCSelectorNameFound) { 1030 unsigned MinIndent = State.Stack.back().Indent; 1031 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) 1032 MinIndent = std::max(MinIndent, 1033 State.FirstIndent + Style.ContinuationIndentWidth); 1034 // If LongestObjCSelectorName is 0, we are indenting the first 1035 // part of an ObjC selector (or a selector component which is 1036 // not colon-aligned due to block formatting). 1037 // 1038 // Otherwise, we are indenting a subsequent part of an ObjC 1039 // selector which should be colon-aligned to the longest 1040 // component of the ObjC selector. 1041 // 1042 // In either case, we want to respect Style.IndentWrappedFunctionNames. 1043 return MinIndent + 1044 std::max(NextNonComment->LongestObjCSelectorName, 1045 NextNonComment->ColumnWidth) - 1046 NextNonComment->ColumnWidth; 1047 } 1048 if (!State.Stack.back().AlignColons) 1049 return State.Stack.back().Indent; 1050 if (State.Stack.back().ColonPos > NextNonComment->ColumnWidth) 1051 return State.Stack.back().ColonPos - NextNonComment->ColumnWidth; 1052 return State.Stack.back().Indent; 1053 } 1054 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr)) 1055 return State.Stack.back().ColonPos; 1056 if (NextNonComment->is(TT_ArraySubscriptLSquare)) { 1057 if (State.Stack.back().StartOfArraySubscripts != 0) 1058 return State.Stack.back().StartOfArraySubscripts; 1059 else if (Style.isCSharp()) // C# allows `["key"] = value` inside object 1060 // initializers. 1061 return State.Stack.back().Indent; 1062 return ContinuationIndent; 1063 } 1064 1065 // This ensure that we correctly format ObjC methods calls without inputs, 1066 // i.e. where the last element isn't selector like: [callee method]; 1067 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 && 1068 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) 1069 return State.Stack.back().Indent; 1070 1071 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) || 1072 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) 1073 return ContinuationIndent; 1074 if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 1075 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) 1076 return ContinuationIndent; 1077 if (NextNonComment->is(TT_CtorInitializerComma)) 1078 return State.Stack.back().Indent; 1079 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && 1080 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) 1081 return State.Stack.back().Indent; 1082 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) && 1083 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) 1084 return State.Stack.back().Indent; 1085 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon, 1086 TT_InheritanceComma)) 1087 return State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1088 if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() && 1089 !Current.isOneOf(tok::colon, tok::comment)) 1090 return ContinuationIndent; 1091 if (Current.is(TT_ProtoExtensionLSquare)) 1092 return State.Stack.back().Indent; 1093 if (State.Stack.back().Indent == State.FirstIndent && PreviousNonComment && 1094 !PreviousNonComment->isOneOf(tok::r_brace, TT_CtorInitializerComma)) 1095 // Ensure that we fall back to the continuation indent width instead of 1096 // just flushing continuations left. 1097 return State.Stack.back().Indent + Style.ContinuationIndentWidth; 1098 return State.Stack.back().Indent; 1099 } 1100 1101 static bool hasNestedBlockInlined(const FormatToken *Previous, 1102 const FormatToken &Current, 1103 const FormatStyle &Style) { 1104 if (Previous->isNot(tok::l_paren)) 1105 return true; 1106 if (Previous->ParameterCount > 1) 1107 return true; 1108 1109 // Also a nested block if contains a lambda inside function with 1 parameter 1110 return (Style.BraceWrapping.BeforeLambdaBody && Current.is(TT_LambdaLSquare)); 1111 } 1112 1113 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State, 1114 bool DryRun, bool Newline) { 1115 assert(State.Stack.size()); 1116 const FormatToken &Current = *State.NextToken; 1117 1118 if (Current.is(TT_CSharpGenericTypeConstraint)) 1119 State.Stack.back().IsCSharpGenericTypeConstraint = true; 1120 if (Current.isOneOf(tok::comma, TT_BinaryOperator)) 1121 State.Stack.back().NoLineBreakInOperand = false; 1122 if (Current.isOneOf(TT_InheritanceColon, TT_CSharpGenericTypeConstraintColon)) 1123 State.Stack.back().AvoidBinPacking = true; 1124 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) { 1125 if (State.Stack.back().FirstLessLess == 0) 1126 State.Stack.back().FirstLessLess = State.Column; 1127 else 1128 State.Stack.back().LastOperatorWrapped = Newline; 1129 } 1130 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless)) 1131 State.Stack.back().LastOperatorWrapped = Newline; 1132 if (Current.is(TT_ConditionalExpr) && Current.Previous && 1133 !Current.Previous->is(TT_ConditionalExpr)) 1134 State.Stack.back().LastOperatorWrapped = Newline; 1135 if (Current.is(TT_ArraySubscriptLSquare) && 1136 State.Stack.back().StartOfArraySubscripts == 0) 1137 State.Stack.back().StartOfArraySubscripts = State.Column; 1138 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question)) 1139 State.Stack.back().QuestionColumn = State.Column; 1140 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) { 1141 const FormatToken *Previous = Current.Previous; 1142 while (Previous && Previous->isTrailingComment()) 1143 Previous = Previous->Previous; 1144 if (Previous && Previous->is(tok::question)) 1145 State.Stack.back().QuestionColumn = State.Column; 1146 } 1147 if (!Current.opensScope() && !Current.closesScope() && 1148 !Current.is(TT_PointerOrReference)) 1149 State.LowestLevelOnLine = 1150 std::min(State.LowestLevelOnLine, Current.NestingLevel); 1151 if (Current.isMemberAccess()) 1152 State.Stack.back().StartOfFunctionCall = 1153 !Current.NextOperator ? 0 : State.Column; 1154 if (Current.is(TT_SelectorName)) 1155 State.Stack.back().ObjCSelectorNameFound = true; 1156 if (Current.is(TT_CtorInitializerColon) && 1157 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) { 1158 // Indent 2 from the column, so: 1159 // SomeClass::SomeClass() 1160 // : First(...), ... 1161 // Next(...) 1162 // ^ line up here. 1163 State.Stack.back().Indent = 1164 State.Column + 1165 (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma 1166 ? 0 1167 : 2); 1168 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1169 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) { 1170 State.Stack.back().AvoidBinPacking = true; 1171 State.Stack.back().BreakBeforeParameter = 1172 !Style.AllowAllConstructorInitializersOnNextLine; 1173 } else { 1174 State.Stack.back().BreakBeforeParameter = false; 1175 } 1176 } 1177 if (Current.is(TT_CtorInitializerColon) && 1178 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { 1179 State.Stack.back().Indent = 1180 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1181 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1182 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 1183 State.Stack.back().AvoidBinPacking = true; 1184 } 1185 if (Current.is(TT_InheritanceColon)) 1186 State.Stack.back().Indent = 1187 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1188 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline) 1189 State.Stack.back().NestedBlockIndent = 1190 State.Column + Current.ColumnWidth + 1; 1191 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow)) 1192 State.Stack.back().LastSpace = State.Column; 1193 1194 // Insert scopes created by fake parenthesis. 1195 const FormatToken *Previous = Current.getPreviousNonComment(); 1196 1197 // Add special behavior to support a format commonly used for JavaScript 1198 // closures: 1199 // SomeFunction(function() { 1200 // foo(); 1201 // bar(); 1202 // }, a, b, c); 1203 if (Current.isNot(tok::comment) && Previous && 1204 Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) && 1205 !Previous->is(TT_DictLiteral) && State.Stack.size() > 1 && 1206 !State.Stack.back().HasMultipleNestedBlocks) { 1207 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline) 1208 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 1209 State.Stack[i].NoLineBreak = true; 1210 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false; 1211 } 1212 if (Previous && 1213 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) || 1214 Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr)) && 1215 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 1216 State.Stack.back().NestedBlockInlined = 1217 !Newline && hasNestedBlockInlined(Previous, Current, Style); 1218 } 1219 1220 moveStatePastFakeLParens(State, Newline); 1221 moveStatePastScopeCloser(State); 1222 bool AllowBreak = !State.Stack.back().NoLineBreak && 1223 !State.Stack.back().NoLineBreakInOperand; 1224 moveStatePastScopeOpener(State, Newline); 1225 moveStatePastFakeRParens(State); 1226 1227 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0) 1228 State.StartOfStringLiteral = State.Column + 1; 1229 if (Current.is(TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) 1230 State.StartOfStringLiteral = State.Column + 1; 1231 else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) 1232 State.StartOfStringLiteral = State.Column; 1233 else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) && 1234 !Current.isStringLiteral()) 1235 State.StartOfStringLiteral = 0; 1236 1237 State.Column += Current.ColumnWidth; 1238 State.NextToken = State.NextToken->Next; 1239 1240 unsigned Penalty = 1241 handleEndOfLine(Current, State, DryRun, AllowBreak, Newline); 1242 1243 if (Current.Role) 1244 Current.Role->formatFromToken(State, this, DryRun); 1245 // If the previous has a special role, let it consume tokens as appropriate. 1246 // It is necessary to start at the previous token for the only implemented 1247 // role (comma separated list). That way, the decision whether or not to break 1248 // after the "{" is already done and both options are tried and evaluated. 1249 // FIXME: This is ugly, find a better way. 1250 if (Previous && Previous->Role) 1251 Penalty += Previous->Role->formatAfterToken(State, this, DryRun); 1252 1253 return Penalty; 1254 } 1255 1256 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State, 1257 bool Newline) { 1258 const FormatToken &Current = *State.NextToken; 1259 const FormatToken *Previous = Current.getPreviousNonComment(); 1260 1261 // Don't add extra indentation for the first fake parenthesis after 1262 // 'return', assignments or opening <({[. The indentation for these cases 1263 // is special cased. 1264 bool SkipFirstExtraIndent = 1265 (Previous && (Previous->opensScope() || 1266 Previous->isOneOf(tok::semi, tok::kw_return) || 1267 (Previous->getPrecedence() == prec::Assignment && 1268 Style.AlignOperands) || 1269 Previous->is(TT_ObjCMethodExpr))); 1270 for (SmallVectorImpl<prec::Level>::const_reverse_iterator 1271 I = Current.FakeLParens.rbegin(), 1272 E = Current.FakeLParens.rend(); 1273 I != E; ++I) { 1274 ParenState NewParenState = State.Stack.back(); 1275 NewParenState.Tok = nullptr; 1276 NewParenState.ContainsLineBreak = false; 1277 NewParenState.LastOperatorWrapped = true; 1278 NewParenState.NoLineBreak = 1279 NewParenState.NoLineBreak || State.Stack.back().NoLineBreakInOperand; 1280 1281 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists. 1282 if (*I > prec::Comma) 1283 NewParenState.AvoidBinPacking = false; 1284 1285 // Indent from 'LastSpace' unless these are fake parentheses encapsulating 1286 // a builder type call after 'return' or, if the alignment after opening 1287 // brackets is disabled. 1288 if (!Current.isTrailingComment() && 1289 (Style.AlignOperands || *I < prec::Assignment) && 1290 (!Previous || Previous->isNot(tok::kw_return) || 1291 (Style.Language != FormatStyle::LK_Java && *I > 0)) && 1292 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign || 1293 *I != prec::Comma || Current.NestingLevel == 0)) 1294 NewParenState.Indent = 1295 std::max(std::max(State.Column, NewParenState.Indent), 1296 State.Stack.back().LastSpace); 1297 1298 // Do not indent relative to the fake parentheses inserted for "." or "->". 1299 // This is a special case to make the following to statements consistent: 1300 // OuterFunction(InnerFunctionCall( // break 1301 // ParameterToInnerFunction)); 1302 // OuterFunction(SomeObject.InnerFunctionCall( // break 1303 // ParameterToInnerFunction)); 1304 if (*I > prec::Unknown) 1305 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column); 1306 if (*I != prec::Conditional && !Current.is(TT_UnaryOperator) && 1307 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 1308 NewParenState.StartOfFunctionCall = State.Column; 1309 1310 // Always indent conditional expressions. Never indent expression where 1311 // the 'operator' is ',', ';' or an assignment (i.e. *I <= 1312 // prec::Assignment) as those have different indentation rules. Indent 1313 // other expression, unless the indentation needs to be skipped. 1314 if (*I == prec::Conditional || 1315 (!SkipFirstExtraIndent && *I > prec::Assignment && 1316 !Current.isTrailingComment())) 1317 NewParenState.Indent += Style.ContinuationIndentWidth; 1318 if ((Previous && !Previous->opensScope()) || *I != prec::Comma) 1319 NewParenState.BreakBeforeParameter = false; 1320 State.Stack.push_back(NewParenState); 1321 SkipFirstExtraIndent = false; 1322 } 1323 } 1324 1325 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) { 1326 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) { 1327 unsigned VariablePos = State.Stack.back().VariablePos; 1328 if (State.Stack.size() == 1) { 1329 // Do not pop the last element. 1330 break; 1331 } 1332 State.Stack.pop_back(); 1333 State.Stack.back().VariablePos = VariablePos; 1334 } 1335 } 1336 1337 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State, 1338 bool Newline) { 1339 const FormatToken &Current = *State.NextToken; 1340 if (!Current.opensScope()) 1341 return; 1342 1343 // Don't allow '<' or '(' in C# generic type constraints to start new scopes. 1344 if (Current.isOneOf(tok::less, tok::l_paren) && 1345 State.Stack.back().IsCSharpGenericTypeConstraint) 1346 return; 1347 1348 if (Current.MatchingParen && Current.BlockKind == BK_Block) { 1349 moveStateToNewBlock(State); 1350 return; 1351 } 1352 1353 unsigned NewIndent; 1354 unsigned LastSpace = State.Stack.back().LastSpace; 1355 bool AvoidBinPacking; 1356 bool BreakBeforeParameter = false; 1357 unsigned NestedBlockIndent = std::max(State.Stack.back().StartOfFunctionCall, 1358 State.Stack.back().NestedBlockIndent); 1359 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 1360 opensProtoMessageField(Current, Style)) { 1361 if (Current.opensBlockOrBlockTypeList(Style)) { 1362 NewIndent = Style.IndentWidth + 1363 std::min(State.Column, State.Stack.back().NestedBlockIndent); 1364 } else { 1365 NewIndent = State.Stack.back().LastSpace + Style.ContinuationIndentWidth; 1366 } 1367 const FormatToken *NextNoComment = Current.getNextNonComment(); 1368 bool EndsInComma = Current.MatchingParen && 1369 Current.MatchingParen->Previous && 1370 Current.MatchingParen->Previous->is(tok::comma); 1371 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) || 1372 Style.Language == FormatStyle::LK_Proto || 1373 Style.Language == FormatStyle::LK_TextProto || 1374 !Style.BinPackArguments || 1375 (NextNoComment && 1376 NextNoComment->isOneOf(TT_DesignatedInitializerPeriod, 1377 TT_DesignatedInitializerLSquare)); 1378 BreakBeforeParameter = EndsInComma; 1379 if (Current.ParameterCount > 1) 1380 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1); 1381 } else { 1382 NewIndent = Style.ContinuationIndentWidth + 1383 std::max(State.Stack.back().LastSpace, 1384 State.Stack.back().StartOfFunctionCall); 1385 1386 // Ensure that different different brackets force relative alignment, e.g.: 1387 // void SomeFunction(vector< // break 1388 // int> v); 1389 // FIXME: We likely want to do this for more combinations of brackets. 1390 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) { 1391 NewIndent = std::max(NewIndent, State.Stack.back().Indent); 1392 LastSpace = std::max(LastSpace, State.Stack.back().Indent); 1393 } 1394 1395 bool EndsInComma = 1396 Current.MatchingParen && 1397 Current.MatchingParen->getPreviousNonComment() && 1398 Current.MatchingParen->getPreviousNonComment()->is(tok::comma); 1399 1400 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters 1401 // for backwards compatibility. 1402 bool ObjCBinPackProtocolList = 1403 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto && 1404 Style.BinPackParameters) || 1405 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always; 1406 1407 bool BinPackDeclaration = 1408 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) || 1409 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList); 1410 1411 AvoidBinPacking = 1412 (State.Stack.back().IsCSharpGenericTypeConstraint) || 1413 (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) || 1414 (State.Line->MustBeDeclaration && !BinPackDeclaration) || 1415 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) || 1416 (Style.ExperimentalAutoDetectBinPacking && 1417 (Current.PackingKind == PPK_OnePerLine || 1418 (!BinPackInconclusiveFunctions && 1419 Current.PackingKind == PPK_Inconclusive))); 1420 1421 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen && 1422 Style.ObjCBreakBeforeNestedBlockParam) { 1423 if (Style.ColumnLimit) { 1424 // If this '[' opens an ObjC call, determine whether all parameters fit 1425 // into one line and put one per line if they don't. 1426 if (getLengthToMatchingParen(Current, State.Stack) + State.Column > 1427 getColumnLimit(State)) 1428 BreakBeforeParameter = true; 1429 } else { 1430 // For ColumnLimit = 0, we have to figure out whether there is or has to 1431 // be a line break within this call. 1432 for (const FormatToken *Tok = &Current; 1433 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) { 1434 if (Tok->MustBreakBefore || 1435 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) { 1436 BreakBeforeParameter = true; 1437 break; 1438 } 1439 } 1440 } 1441 } 1442 1443 if (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) 1444 BreakBeforeParameter = true; 1445 } 1446 // Generally inherit NoLineBreak from the current scope to nested scope. 1447 // However, don't do this for non-empty nested blocks, dict literals and 1448 // array literals as these follow different indentation rules. 1449 bool NoLineBreak = 1450 Current.Children.empty() && 1451 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) && 1452 (State.Stack.back().NoLineBreak || 1453 State.Stack.back().NoLineBreakInOperand || 1454 (Current.is(TT_TemplateOpener) && 1455 State.Stack.back().ContainsUnwrappedBuilder)); 1456 State.Stack.push_back( 1457 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak)); 1458 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1459 State.Stack.back().BreakBeforeParameter = BreakBeforeParameter; 1460 State.Stack.back().HasMultipleNestedBlocks = (Current.BlockParameterCount > 1); 1461 1462 if (Style.BraceWrapping.BeforeLambdaBody && 1463 Current.Next != nullptr && Current.Tok.is(tok::l_paren)) { 1464 // Search for any parameter that is a lambda 1465 FormatToken const *next = Current.Next; 1466 while (next != nullptr) { 1467 if (next->is(TT_LambdaLSquare)) { 1468 State.Stack.back().HasMultipleNestedBlocks = true; 1469 break; 1470 } 1471 next = next->Next; 1472 } 1473 } 1474 1475 State.Stack.back().IsInsideObjCArrayLiteral = 1476 Current.is(TT_ArrayInitializerLSquare) && Current.Previous && 1477 Current.Previous->is(tok::at); 1478 } 1479 1480 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) { 1481 const FormatToken &Current = *State.NextToken; 1482 if (!Current.closesScope()) 1483 return; 1484 1485 // If we encounter a closing ), ], } or >, we can remove a level from our 1486 // stacks. 1487 if (State.Stack.size() > 1 && 1488 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) || 1489 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) || 1490 State.NextToken->is(TT_TemplateCloser) || 1491 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) 1492 State.Stack.pop_back(); 1493 1494 // Reevaluate whether ObjC message arguments fit into one line. 1495 // If a receiver spans multiple lines, e.g.: 1496 // [[object block:^{ 1497 // return 42; 1498 // }] a:42 b:42]; 1499 // BreakBeforeParameter is calculated based on an incorrect assumption 1500 // (it is checked whether the whole expression fits into one line without 1501 // considering a line break inside a message receiver). 1502 // We check whether arguements fit after receiver scope closer (into the same 1503 // line). 1504 if (State.Stack.back().BreakBeforeParameter && Current.MatchingParen && 1505 Current.MatchingParen->Previous) { 1506 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous; 1507 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) && 1508 CurrentScopeOpener.MatchingParen) { 1509 int NecessarySpaceInLine = 1510 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) + 1511 CurrentScopeOpener.TotalLength - Current.TotalLength - 1; 1512 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <= 1513 Style.ColumnLimit) 1514 State.Stack.back().BreakBeforeParameter = false; 1515 } 1516 } 1517 1518 if (Current.is(tok::r_square)) { 1519 // If this ends the array subscript expr, reset the corresponding value. 1520 const FormatToken *NextNonComment = Current.getNextNonComment(); 1521 if (NextNonComment && NextNonComment->isNot(tok::l_square)) 1522 State.Stack.back().StartOfArraySubscripts = 0; 1523 } 1524 } 1525 1526 void ContinuationIndenter::moveStateToNewBlock(LineState &State) { 1527 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent; 1528 // ObjC block sometimes follow special indentation rules. 1529 unsigned NewIndent = 1530 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace) 1531 ? Style.ObjCBlockIndentWidth 1532 : Style.IndentWidth); 1533 State.Stack.push_back(ParenState(State.NextToken, NewIndent, 1534 State.Stack.back().LastSpace, 1535 /*AvoidBinPacking=*/true, 1536 /*NoLineBreak=*/false)); 1537 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1538 State.Stack.back().BreakBeforeParameter = true; 1539 } 1540 1541 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn, 1542 unsigned TabWidth, 1543 encoding::Encoding Encoding) { 1544 size_t LastNewlinePos = Text.find_last_of("\n"); 1545 if (LastNewlinePos == StringRef::npos) { 1546 return StartColumn + 1547 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding); 1548 } else { 1549 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos), 1550 /*StartColumn=*/0, TabWidth, Encoding); 1551 } 1552 } 1553 1554 unsigned ContinuationIndenter::reformatRawStringLiteral( 1555 const FormatToken &Current, LineState &State, 1556 const FormatStyle &RawStringStyle, bool DryRun, bool Newline) { 1557 unsigned StartColumn = State.Column - Current.ColumnWidth; 1558 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText); 1559 StringRef NewDelimiter = 1560 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language); 1561 if (NewDelimiter.empty() || OldDelimiter.empty()) 1562 NewDelimiter = OldDelimiter; 1563 // The text of a raw string is between the leading 'R"delimiter(' and the 1564 // trailing 'delimiter)"'. 1565 unsigned OldPrefixSize = 3 + OldDelimiter.size(); 1566 unsigned OldSuffixSize = 2 + OldDelimiter.size(); 1567 // We create a virtual text environment which expects a null-terminated 1568 // string, so we cannot use StringRef. 1569 std::string RawText = std::string( 1570 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize)); 1571 if (NewDelimiter != OldDelimiter) { 1572 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the 1573 // raw string. 1574 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str(); 1575 if (StringRef(RawText).contains(CanonicalDelimiterSuffix)) 1576 NewDelimiter = OldDelimiter; 1577 } 1578 1579 unsigned NewPrefixSize = 3 + NewDelimiter.size(); 1580 unsigned NewSuffixSize = 2 + NewDelimiter.size(); 1581 1582 // The first start column is the column the raw text starts after formatting. 1583 unsigned FirstStartColumn = StartColumn + NewPrefixSize; 1584 1585 // The next start column is the intended indentation a line break inside 1586 // the raw string at level 0. It is determined by the following rules: 1587 // - if the content starts on newline, it is one level more than the current 1588 // indent, and 1589 // - if the content does not start on a newline, it is the first start 1590 // column. 1591 // These rules have the advantage that the formatted content both does not 1592 // violate the rectangle rule and visually flows within the surrounding 1593 // source. 1594 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n'; 1595 // If this token is the last parameter (checked by looking if it's followed by 1596 // `)` and is not on a newline, the base the indent off the line's nested 1597 // block indent. Otherwise, base the indent off the arguments indent, so we 1598 // can achieve: 1599 // 1600 // fffffffffff(1, 2, 3, R"pb( 1601 // key1: 1 # 1602 // key2: 2)pb"); 1603 // 1604 // fffffffffff(1, 2, 3, 1605 // R"pb( 1606 // key1: 1 # 1607 // key2: 2 1608 // )pb"); 1609 // 1610 // fffffffffff(1, 2, 3, 1611 // R"pb( 1612 // key1: 1 # 1613 // key2: 2 1614 // )pb", 1615 // 5); 1616 unsigned CurrentIndent = 1617 (!Newline && Current.Next && Current.Next->is(tok::r_paren)) 1618 ? State.Stack.back().NestedBlockIndent 1619 : State.Stack.back().Indent; 1620 unsigned NextStartColumn = ContentStartsOnNewline 1621 ? CurrentIndent + Style.IndentWidth 1622 : FirstStartColumn; 1623 1624 // The last start column is the column the raw string suffix starts if it is 1625 // put on a newline. 1626 // The last start column is the intended indentation of the raw string postfix 1627 // if it is put on a newline. It is determined by the following rules: 1628 // - if the raw string prefix starts on a newline, it is the column where 1629 // that raw string prefix starts, and 1630 // - if the raw string prefix does not start on a newline, it is the current 1631 // indent. 1632 unsigned LastStartColumn = 1633 Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent; 1634 1635 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat( 1636 RawStringStyle, RawText, {tooling::Range(0, RawText.size())}, 1637 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>", 1638 /*Status=*/nullptr); 1639 1640 auto NewCode = applyAllReplacements(RawText, Fixes.first); 1641 tooling::Replacements NoFixes; 1642 if (!NewCode) { 1643 return addMultilineToken(Current, State); 1644 } 1645 if (!DryRun) { 1646 if (NewDelimiter != OldDelimiter) { 1647 // In 'R"delimiter(...', the delimiter starts 2 characters after the start 1648 // of the token. 1649 SourceLocation PrefixDelimiterStart = 1650 Current.Tok.getLocation().getLocWithOffset(2); 1651 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement( 1652 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1653 if (PrefixErr) { 1654 llvm::errs() 1655 << "Failed to update the prefix delimiter of a raw string: " 1656 << llvm::toString(std::move(PrefixErr)) << "\n"; 1657 } 1658 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at 1659 // position length - 1 - |delimiter|. 1660 SourceLocation SuffixDelimiterStart = 1661 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() - 1662 1 - OldDelimiter.size()); 1663 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement( 1664 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1665 if (SuffixErr) { 1666 llvm::errs() 1667 << "Failed to update the suffix delimiter of a raw string: " 1668 << llvm::toString(std::move(SuffixErr)) << "\n"; 1669 } 1670 } 1671 SourceLocation OriginLoc = 1672 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize); 1673 for (const tooling::Replacement &Fix : Fixes.first) { 1674 auto Err = Whitespaces.addReplacement(tooling::Replacement( 1675 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()), 1676 Fix.getLength(), Fix.getReplacementText())); 1677 if (Err) { 1678 llvm::errs() << "Failed to reformat raw string: " 1679 << llvm::toString(std::move(Err)) << "\n"; 1680 } 1681 } 1682 } 1683 unsigned RawLastLineEndColumn = getLastLineEndColumn( 1684 *NewCode, FirstStartColumn, Style.TabWidth, Encoding); 1685 State.Column = RawLastLineEndColumn + NewSuffixSize; 1686 // Since we're updating the column to after the raw string literal here, we 1687 // have to manually add the penalty for the prefix R"delim( over the column 1688 // limit. 1689 unsigned PrefixExcessCharacters = 1690 StartColumn + NewPrefixSize > Style.ColumnLimit 1691 ? StartColumn + NewPrefixSize - Style.ColumnLimit 1692 : 0; 1693 bool IsMultiline = 1694 ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos); 1695 if (IsMultiline) { 1696 // Break before further function parameters on all levels. 1697 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 1698 State.Stack[i].BreakBeforeParameter = true; 1699 } 1700 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter; 1701 } 1702 1703 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current, 1704 LineState &State) { 1705 // Break before further function parameters on all levels. 1706 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 1707 State.Stack[i].BreakBeforeParameter = true; 1708 1709 unsigned ColumnsUsed = State.Column; 1710 // We can only affect layout of the first and the last line, so the penalty 1711 // for all other lines is constant, and we ignore it. 1712 State.Column = Current.LastLineColumnWidth; 1713 1714 if (ColumnsUsed > getColumnLimit(State)) 1715 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State)); 1716 return 0; 1717 } 1718 1719 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current, 1720 LineState &State, bool DryRun, 1721 bool AllowBreak, bool Newline) { 1722 unsigned Penalty = 0; 1723 // Compute the raw string style to use in case this is a raw string literal 1724 // that can be reformatted. 1725 auto RawStringStyle = getRawStringStyle(Current, State); 1726 if (RawStringStyle && !Current.Finalized) { 1727 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun, 1728 Newline); 1729 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) { 1730 // Don't break multi-line tokens other than block comments and raw string 1731 // literals. Instead, just update the state. 1732 Penalty = addMultilineToken(Current, State); 1733 } else if (State.Line->Type != LT_ImportStatement) { 1734 // We generally don't break import statements. 1735 LineState OriginalState = State; 1736 1737 // Whether we force the reflowing algorithm to stay strictly within the 1738 // column limit. 1739 bool Strict = false; 1740 // Whether the first non-strict attempt at reflowing did intentionally 1741 // exceed the column limit. 1742 bool Exceeded = false; 1743 std::tie(Penalty, Exceeded) = breakProtrudingToken( 1744 Current, State, AllowBreak, /*DryRun=*/true, Strict); 1745 if (Exceeded) { 1746 // If non-strict reflowing exceeds the column limit, try whether strict 1747 // reflowing leads to an overall lower penalty. 1748 LineState StrictState = OriginalState; 1749 unsigned StrictPenalty = 1750 breakProtrudingToken(Current, StrictState, AllowBreak, 1751 /*DryRun=*/true, /*Strict=*/true) 1752 .first; 1753 Strict = StrictPenalty <= Penalty; 1754 if (Strict) { 1755 Penalty = StrictPenalty; 1756 State = StrictState; 1757 } 1758 } 1759 if (!DryRun) { 1760 // If we're not in dry-run mode, apply the changes with the decision on 1761 // strictness made above. 1762 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false, 1763 Strict); 1764 } 1765 } 1766 if (State.Column > getColumnLimit(State)) { 1767 unsigned ExcessCharacters = State.Column - getColumnLimit(State); 1768 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters; 1769 } 1770 return Penalty; 1771 } 1772 1773 // Returns the enclosing function name of a token, or the empty string if not 1774 // found. 1775 static StringRef getEnclosingFunctionName(const FormatToken &Current) { 1776 // Look for: 'function(' or 'function<templates>(' before Current. 1777 auto Tok = Current.getPreviousNonComment(); 1778 if (!Tok || !Tok->is(tok::l_paren)) 1779 return ""; 1780 Tok = Tok->getPreviousNonComment(); 1781 if (!Tok) 1782 return ""; 1783 if (Tok->is(TT_TemplateCloser)) { 1784 Tok = Tok->MatchingParen; 1785 if (Tok) 1786 Tok = Tok->getPreviousNonComment(); 1787 } 1788 if (!Tok || !Tok->is(tok::identifier)) 1789 return ""; 1790 return Tok->TokenText; 1791 } 1792 1793 llvm::Optional<FormatStyle> 1794 ContinuationIndenter::getRawStringStyle(const FormatToken &Current, 1795 const LineState &State) { 1796 if (!Current.isStringLiteral()) 1797 return None; 1798 auto Delimiter = getRawStringDelimiter(Current.TokenText); 1799 if (!Delimiter) 1800 return None; 1801 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter); 1802 if (!RawStringStyle && Delimiter->empty()) 1803 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle( 1804 getEnclosingFunctionName(Current)); 1805 if (!RawStringStyle) 1806 return None; 1807 RawStringStyle->ColumnLimit = getColumnLimit(State); 1808 return RawStringStyle; 1809 } 1810 1811 std::unique_ptr<BreakableToken> 1812 ContinuationIndenter::createBreakableToken(const FormatToken &Current, 1813 LineState &State, bool AllowBreak) { 1814 unsigned StartColumn = State.Column - Current.ColumnWidth; 1815 if (Current.isStringLiteral()) { 1816 // FIXME: String literal breaking is currently disabled for C#, Java and 1817 // JavaScript, as it requires strings to be merged using "+" which we 1818 // don't support. 1819 if (Style.Language == FormatStyle::LK_Java || 1820 Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp() || 1821 !Style.BreakStringLiterals || !AllowBreak) 1822 return nullptr; 1823 1824 // Don't break string literals inside preprocessor directives (except for 1825 // #define directives, as their contents are stored in separate lines and 1826 // are not affected by this check). 1827 // This way we avoid breaking code with line directives and unknown 1828 // preprocessor directives that contain long string literals. 1829 if (State.Line->Type == LT_PreprocessorDirective) 1830 return nullptr; 1831 // Exempts unterminated string literals from line breaking. The user will 1832 // likely want to terminate the string before any line breaking is done. 1833 if (Current.IsUnterminatedLiteral) 1834 return nullptr; 1835 // Don't break string literals inside Objective-C array literals (doing so 1836 // raises the warning -Wobjc-string-concatenation). 1837 if (State.Stack.back().IsInsideObjCArrayLiteral) { 1838 return nullptr; 1839 } 1840 1841 StringRef Text = Current.TokenText; 1842 StringRef Prefix; 1843 StringRef Postfix; 1844 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'. 1845 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to 1846 // reduce the overhead) for each FormatToken, which is a string, so that we 1847 // don't run multiple checks here on the hot path. 1848 if ((Text.endswith(Postfix = "\"") && 1849 (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") || 1850 Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") || 1851 Text.startswith(Prefix = "u8\"") || 1852 Text.startswith(Prefix = "L\""))) || 1853 (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) { 1854 // We need this to address the case where there is an unbreakable tail 1855 // only if certain other formatting decisions have been taken. The 1856 // UnbreakableTailLength of Current is an overapproximation is that case 1857 // and we need to be correct here. 1858 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State)) 1859 ? 0 1860 : Current.UnbreakableTailLength; 1861 return std::make_unique<BreakableStringLiteral>( 1862 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength, 1863 State.Line->InPPDirective, Encoding, Style); 1864 } 1865 } else if (Current.is(TT_BlockComment)) { 1866 if (!Style.ReflowComments || 1867 // If a comment token switches formatting, like 1868 // /* clang-format on */, we don't want to break it further, 1869 // but we may still want to adjust its indentation. 1870 switchesFormatting(Current)) { 1871 return nullptr; 1872 } 1873 return std::make_unique<BreakableBlockComment>( 1874 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1875 State.Line->InPPDirective, Encoding, Style, Whitespaces.useCRLF()); 1876 } else if (Current.is(TT_LineComment) && 1877 (Current.Previous == nullptr || 1878 Current.Previous->isNot(TT_ImplicitStringLiteral))) { 1879 if (!Style.ReflowComments || 1880 CommentPragmasRegex.match(Current.TokenText.substr(2)) || 1881 switchesFormatting(Current)) 1882 return nullptr; 1883 return std::make_unique<BreakableLineCommentSection>( 1884 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1885 /*InPPDirective=*/false, Encoding, Style); 1886 } 1887 return nullptr; 1888 } 1889 1890 std::pair<unsigned, bool> 1891 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current, 1892 LineState &State, bool AllowBreak, 1893 bool DryRun, bool Strict) { 1894 std::unique_ptr<const BreakableToken> Token = 1895 createBreakableToken(Current, State, AllowBreak); 1896 if (!Token) 1897 return {0, false}; 1898 assert(Token->getLineCount() > 0); 1899 unsigned ColumnLimit = getColumnLimit(State); 1900 if (Current.is(TT_LineComment)) { 1901 // We don't insert backslashes when breaking line comments. 1902 ColumnLimit = Style.ColumnLimit; 1903 } 1904 if (Current.UnbreakableTailLength >= ColumnLimit) 1905 return {0, false}; 1906 // ColumnWidth was already accounted into State.Column before calling 1907 // breakProtrudingToken. 1908 unsigned StartColumn = State.Column - Current.ColumnWidth; 1909 unsigned NewBreakPenalty = Current.isStringLiteral() 1910 ? Style.PenaltyBreakString 1911 : Style.PenaltyBreakComment; 1912 // Stores whether we intentionally decide to let a line exceed the column 1913 // limit. 1914 bool Exceeded = false; 1915 // Stores whether we introduce a break anywhere in the token. 1916 bool BreakInserted = Token->introducesBreakBeforeToken(); 1917 // Store whether we inserted a new line break at the end of the previous 1918 // logical line. 1919 bool NewBreakBefore = false; 1920 // We use a conservative reflowing strategy. Reflow starts after a line is 1921 // broken or the corresponding whitespace compressed. Reflow ends as soon as a 1922 // line that doesn't get reflown with the previous line is reached. 1923 bool Reflow = false; 1924 // Keep track of where we are in the token: 1925 // Where we are in the content of the current logical line. 1926 unsigned TailOffset = 0; 1927 // The column number we're currently at. 1928 unsigned ContentStartColumn = 1929 Token->getContentStartColumn(0, /*Break=*/false); 1930 // The number of columns left in the current logical line after TailOffset. 1931 unsigned RemainingTokenColumns = 1932 Token->getRemainingLength(0, TailOffset, ContentStartColumn); 1933 // Adapt the start of the token, for example indent. 1934 if (!DryRun) 1935 Token->adaptStartOfLine(0, Whitespaces); 1936 1937 unsigned ContentIndent = 0; 1938 unsigned Penalty = 0; 1939 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column " 1940 << StartColumn << ".\n"); 1941 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount(); 1942 LineIndex != EndIndex; ++LineIndex) { 1943 LLVM_DEBUG(llvm::dbgs() 1944 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n"); 1945 NewBreakBefore = false; 1946 // If we did reflow the previous line, we'll try reflowing again. Otherwise 1947 // we'll start reflowing if the current line is broken or whitespace is 1948 // compressed. 1949 bool TryReflow = Reflow; 1950 // Break the current token until we can fit the rest of the line. 1951 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 1952 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: " 1953 << (ContentStartColumn + RemainingTokenColumns) 1954 << ", space: " << ColumnLimit 1955 << ", reflown prefix: " << ContentStartColumn 1956 << ", offset in line: " << TailOffset << "\n"); 1957 // If the current token doesn't fit, find the latest possible split in the 1958 // current line so that breaking at it will be under the column limit. 1959 // FIXME: Use the earliest possible split while reflowing to correctly 1960 // compress whitespace within a line. 1961 BreakableToken::Split Split = 1962 Token->getSplit(LineIndex, TailOffset, ColumnLimit, 1963 ContentStartColumn, CommentPragmasRegex); 1964 if (Split.first == StringRef::npos) { 1965 // No break opportunity - update the penalty and continue with the next 1966 // logical line. 1967 if (LineIndex < EndIndex - 1) 1968 // The last line's penalty is handled in addNextStateToQueue() or when 1969 // calling replaceWhitespaceAfterLastLine below. 1970 Penalty += Style.PenaltyExcessCharacter * 1971 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 1972 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n"); 1973 break; 1974 } 1975 assert(Split.first != 0); 1976 1977 if (Token->supportsReflow()) { 1978 // Check whether the next natural split point after the current one can 1979 // still fit the line, either because we can compress away whitespace, 1980 // or because the penalty the excess characters introduce is lower than 1981 // the break penalty. 1982 // We only do this for tokens that support reflowing, and thus allow us 1983 // to change the whitespace arbitrarily (e.g. comments). 1984 // Other tokens, like string literals, can be broken on arbitrary 1985 // positions. 1986 1987 // First, compute the columns from TailOffset to the next possible split 1988 // position. 1989 // For example: 1990 // ColumnLimit: | 1991 // // Some text that breaks 1992 // ^ tail offset 1993 // ^-- split 1994 // ^-------- to split columns 1995 // ^--- next split 1996 // ^--------------- to next split columns 1997 unsigned ToSplitColumns = Token->getRangeLength( 1998 LineIndex, TailOffset, Split.first, ContentStartColumn); 1999 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n"); 2000 2001 BreakableToken::Split NextSplit = Token->getSplit( 2002 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit, 2003 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex); 2004 // Compute the columns necessary to fit the next non-breakable sequence 2005 // into the current line. 2006 unsigned ToNextSplitColumns = 0; 2007 if (NextSplit.first == StringRef::npos) { 2008 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset, 2009 ContentStartColumn); 2010 } else { 2011 ToNextSplitColumns = Token->getRangeLength( 2012 LineIndex, TailOffset, 2013 Split.first + Split.second + NextSplit.first, ContentStartColumn); 2014 } 2015 // Compress the whitespace between the break and the start of the next 2016 // unbreakable sequence. 2017 ToNextSplitColumns = 2018 Token->getLengthAfterCompression(ToNextSplitColumns, Split); 2019 LLVM_DEBUG(llvm::dbgs() 2020 << " ContentStartColumn: " << ContentStartColumn << "\n"); 2021 LLVM_DEBUG(llvm::dbgs() 2022 << " ToNextSplit: " << ToNextSplitColumns << "\n"); 2023 // If the whitespace compression makes us fit, continue on the current 2024 // line. 2025 bool ContinueOnLine = 2026 ContentStartColumn + ToNextSplitColumns <= ColumnLimit; 2027 unsigned ExcessCharactersPenalty = 0; 2028 if (!ContinueOnLine && !Strict) { 2029 // Similarly, if the excess characters' penalty is lower than the 2030 // penalty of introducing a new break, continue on the current line. 2031 ExcessCharactersPenalty = 2032 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) * 2033 Style.PenaltyExcessCharacter; 2034 LLVM_DEBUG(llvm::dbgs() 2035 << " Penalty excess: " << ExcessCharactersPenalty 2036 << "\n break : " << NewBreakPenalty << "\n"); 2037 if (ExcessCharactersPenalty < NewBreakPenalty) { 2038 Exceeded = true; 2039 ContinueOnLine = true; 2040 } 2041 } 2042 if (ContinueOnLine) { 2043 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n"); 2044 // The current line fits after compressing the whitespace - reflow 2045 // the next line into it if possible. 2046 TryReflow = true; 2047 if (!DryRun) 2048 Token->compressWhitespace(LineIndex, TailOffset, Split, 2049 Whitespaces); 2050 // When we continue on the same line, leave one space between content. 2051 ContentStartColumn += ToSplitColumns + 1; 2052 Penalty += ExcessCharactersPenalty; 2053 TailOffset += Split.first + Split.second; 2054 RemainingTokenColumns = Token->getRemainingLength( 2055 LineIndex, TailOffset, ContentStartColumn); 2056 continue; 2057 } 2058 } 2059 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n"); 2060 // Update the ContentIndent only if the current line was not reflown with 2061 // the previous line, since in that case the previous line should still 2062 // determine the ContentIndent. Also never intent the last line. 2063 if (!Reflow) 2064 ContentIndent = Token->getContentIndent(LineIndex); 2065 LLVM_DEBUG(llvm::dbgs() 2066 << " ContentIndent: " << ContentIndent << "\n"); 2067 ContentStartColumn = ContentIndent + Token->getContentStartColumn( 2068 LineIndex, /*Break=*/true); 2069 2070 unsigned NewRemainingTokenColumns = Token->getRemainingLength( 2071 LineIndex, TailOffset + Split.first + Split.second, 2072 ContentStartColumn); 2073 if (NewRemainingTokenColumns == 0) { 2074 // No content to indent. 2075 ContentIndent = 0; 2076 ContentStartColumn = 2077 Token->getContentStartColumn(LineIndex, /*Break=*/true); 2078 NewRemainingTokenColumns = Token->getRemainingLength( 2079 LineIndex, TailOffset + Split.first + Split.second, 2080 ContentStartColumn); 2081 } 2082 2083 // When breaking before a tab character, it may be moved by a few columns, 2084 // but will still be expanded to the next tab stop, so we don't save any 2085 // columns. 2086 if (NewRemainingTokenColumns == RemainingTokenColumns) { 2087 // FIXME: Do we need to adjust the penalty? 2088 break; 2089 } 2090 assert(NewRemainingTokenColumns < RemainingTokenColumns); 2091 2092 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first 2093 << ", " << Split.second << "\n"); 2094 if (!DryRun) 2095 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent, 2096 Whitespaces); 2097 2098 Penalty += NewBreakPenalty; 2099 TailOffset += Split.first + Split.second; 2100 RemainingTokenColumns = NewRemainingTokenColumns; 2101 BreakInserted = true; 2102 NewBreakBefore = true; 2103 } 2104 // In case there's another line, prepare the state for the start of the next 2105 // line. 2106 if (LineIndex + 1 != EndIndex) { 2107 unsigned NextLineIndex = LineIndex + 1; 2108 if (NewBreakBefore) 2109 // After breaking a line, try to reflow the next line into the current 2110 // one once RemainingTokenColumns fits. 2111 TryReflow = true; 2112 if (TryReflow) { 2113 // We decided that we want to try reflowing the next line into the 2114 // current one. 2115 // We will now adjust the state as if the reflow is successful (in 2116 // preparation for the next line), and see whether that works. If we 2117 // decide that we cannot reflow, we will later reset the state to the 2118 // start of the next line. 2119 Reflow = false; 2120 // As we did not continue breaking the line, RemainingTokenColumns is 2121 // known to fit after ContentStartColumn. Adapt ContentStartColumn to 2122 // the position at which we want to format the next line if we do 2123 // actually reflow. 2124 // When we reflow, we need to add a space between the end of the current 2125 // line and the next line's start column. 2126 ContentStartColumn += RemainingTokenColumns + 1; 2127 // Get the split that we need to reflow next logical line into the end 2128 // of the current one; the split will include any leading whitespace of 2129 // the next logical line. 2130 BreakableToken::Split SplitBeforeNext = 2131 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex); 2132 LLVM_DEBUG(llvm::dbgs() 2133 << " Size of reflown text: " << ContentStartColumn 2134 << "\n Potential reflow split: "); 2135 if (SplitBeforeNext.first != StringRef::npos) { 2136 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", " 2137 << SplitBeforeNext.second << "\n"); 2138 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second; 2139 // If the rest of the next line fits into the current line below the 2140 // column limit, we can safely reflow. 2141 RemainingTokenColumns = Token->getRemainingLength( 2142 NextLineIndex, TailOffset, ContentStartColumn); 2143 Reflow = true; 2144 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2145 LLVM_DEBUG(llvm::dbgs() 2146 << " Over limit after reflow, need: " 2147 << (ContentStartColumn + RemainingTokenColumns) 2148 << ", space: " << ColumnLimit 2149 << ", reflown prefix: " << ContentStartColumn 2150 << ", offset in line: " << TailOffset << "\n"); 2151 // If the whole next line does not fit, try to find a point in 2152 // the next line at which we can break so that attaching the part 2153 // of the next line to that break point onto the current line is 2154 // below the column limit. 2155 BreakableToken::Split Split = 2156 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit, 2157 ContentStartColumn, CommentPragmasRegex); 2158 if (Split.first == StringRef::npos) { 2159 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n"); 2160 Reflow = false; 2161 } else { 2162 // Check whether the first split point gets us below the column 2163 // limit. Note that we will execute this split below as part of 2164 // the normal token breaking and reflow logic within the line. 2165 unsigned ToSplitColumns = Token->getRangeLength( 2166 NextLineIndex, TailOffset, Split.first, ContentStartColumn); 2167 if (ContentStartColumn + ToSplitColumns > ColumnLimit) { 2168 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: " 2169 << (ContentStartColumn + ToSplitColumns) 2170 << ", space: " << ColumnLimit); 2171 unsigned ExcessCharactersPenalty = 2172 (ContentStartColumn + ToSplitColumns - ColumnLimit) * 2173 Style.PenaltyExcessCharacter; 2174 if (NewBreakPenalty < ExcessCharactersPenalty) { 2175 Reflow = false; 2176 } 2177 } 2178 } 2179 } 2180 } else { 2181 LLVM_DEBUG(llvm::dbgs() << "not found.\n"); 2182 } 2183 } 2184 if (!Reflow) { 2185 // If we didn't reflow into the next line, the only space to consider is 2186 // the next logical line. Reset our state to match the start of the next 2187 // line. 2188 TailOffset = 0; 2189 ContentStartColumn = 2190 Token->getContentStartColumn(NextLineIndex, /*Break=*/false); 2191 RemainingTokenColumns = Token->getRemainingLength( 2192 NextLineIndex, TailOffset, ContentStartColumn); 2193 // Adapt the start of the token, for example indent. 2194 if (!DryRun) 2195 Token->adaptStartOfLine(NextLineIndex, Whitespaces); 2196 } else { 2197 // If we found a reflow split and have added a new break before the next 2198 // line, we are going to remove the line break at the start of the next 2199 // logical line. For example, here we'll add a new line break after 2200 // 'text', and subsequently delete the line break between 'that' and 2201 // 'reflows'. 2202 // // some text that 2203 // // reflows 2204 // -> 2205 // // some text 2206 // // that reflows 2207 // When adding the line break, we also added the penalty for it, so we 2208 // need to subtract that penalty again when we remove the line break due 2209 // to reflowing. 2210 if (NewBreakBefore) { 2211 assert(Penalty >= NewBreakPenalty); 2212 Penalty -= NewBreakPenalty; 2213 } 2214 if (!DryRun) 2215 Token->reflow(NextLineIndex, Whitespaces); 2216 } 2217 } 2218 } 2219 2220 BreakableToken::Split SplitAfterLastLine = 2221 Token->getSplitAfterLastLine(TailOffset); 2222 if (SplitAfterLastLine.first != StringRef::npos) { 2223 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n"); 2224 2225 // We add the last line's penalty here, since that line is going to be split 2226 // now. 2227 Penalty += Style.PenaltyExcessCharacter * 2228 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2229 2230 if (!DryRun) 2231 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine, 2232 Whitespaces); 2233 ContentStartColumn = 2234 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true); 2235 RemainingTokenColumns = Token->getRemainingLength( 2236 Token->getLineCount() - 1, 2237 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second, 2238 ContentStartColumn); 2239 } 2240 2241 State.Column = ContentStartColumn + RemainingTokenColumns - 2242 Current.UnbreakableTailLength; 2243 2244 if (BreakInserted) { 2245 // If we break the token inside a parameter list, we need to break before 2246 // the next parameter on all levels, so that the next parameter is clearly 2247 // visible. Line comments already introduce a break. 2248 if (Current.isNot(TT_LineComment)) { 2249 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 2250 State.Stack[i].BreakBeforeParameter = true; 2251 } 2252 2253 if (Current.is(TT_BlockComment)) 2254 State.NoContinuation = true; 2255 2256 State.Stack.back().LastSpace = StartColumn; 2257 } 2258 2259 Token->updateNextToken(State); 2260 2261 return {Penalty, Exceeded}; 2262 } 2263 2264 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const { 2265 // In preprocessor directives reserve two chars for trailing " \" 2266 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0); 2267 } 2268 2269 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) { 2270 const FormatToken &Current = *State.NextToken; 2271 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral)) 2272 return false; 2273 // We never consider raw string literals "multiline" for the purpose of 2274 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased 2275 // (see TokenAnnotator::mustBreakBefore(). 2276 if (Current.TokenText.startswith("R\"")) 2277 return false; 2278 if (Current.IsMultiline) 2279 return true; 2280 if (Current.getNextNonComment() && 2281 Current.getNextNonComment()->isStringLiteral()) 2282 return true; // Implicit concatenation. 2283 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals && 2284 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength > 2285 Style.ColumnLimit) 2286 return true; // String will be split. 2287 return false; 2288 } 2289 2290 } // namespace format 2291 } // namespace clang 2292