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