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() && 407 State.Column <= NewLineColumn)) 408 return true; 409 410 // If the template declaration spans multiple lines, force wrap before the 411 // function/class declaration 412 if (Previous.ClosesTemplateDeclaration && 413 State.Stack.back().BreakBeforeParameter && Current.CanBreakBefore) 414 return true; 415 416 if (State.Column <= NewLineColumn) 417 return false; 418 419 if (Style.AlwaysBreakBeforeMultilineStrings && 420 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth || 421 Previous.is(tok::comma) || Current.NestingLevel < 2) && 422 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at) && 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 = std::max( 839 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 '(', '{', '<', this is not bin packing 885 // unless AllowAllParametersOfDeclarationOnNextLine is false or this is a 886 // dict/object literal. 887 if (!Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) || 888 (!Style.AllowAllParametersOfDeclarationOnNextLine && 889 State.Line->MustBeDeclaration) || 890 Previous.is(TT_DictLiteral)) 891 State.Stack.back().BreakBeforeParameter = true; 892 } 893 894 return Penalty; 895 } 896 897 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) { 898 if (!State.NextToken || !State.NextToken->Previous) 899 return 0; 900 FormatToken &Current = *State.NextToken; 901 const FormatToken &Previous = *Current.Previous; 902 // If we are continuing an expression, we want to use the continuation indent. 903 unsigned ContinuationIndent = 904 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 905 Style.ContinuationIndentWidth; 906 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 907 const FormatToken *NextNonComment = Previous.getNextNonComment(); 908 if (!NextNonComment) 909 NextNonComment = &Current; 910 911 // Java specific bits. 912 if (Style.Language == FormatStyle::LK_Java && 913 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) 914 return std::max(State.Stack.back().LastSpace, 915 State.Stack.back().Indent + Style.ContinuationIndentWidth); 916 917 if (NextNonComment->is(tok::l_brace) && NextNonComment->BlockKind == BK_Block) 918 return Current.NestingLevel == 0 ? State.FirstIndent 919 : State.Stack.back().Indent; 920 if ((Current.isOneOf(tok::r_brace, tok::r_square) || 921 (Current.is(tok::greater) && 922 (Style.Language == FormatStyle::LK_Proto || 923 Style.Language == FormatStyle::LK_TextProto))) && 924 State.Stack.size() > 1) { 925 if (Current.closesBlockOrBlockTypeList(Style)) 926 return State.Stack[State.Stack.size() - 2].NestedBlockIndent; 927 if (Current.MatchingParen && 928 Current.MatchingParen->BlockKind == BK_BracedInit) 929 return State.Stack[State.Stack.size() - 2].LastSpace; 930 return State.FirstIndent; 931 } 932 // Indent a closing parenthesis at the previous level if followed by a semi or 933 // opening brace. This allows indentations such as: 934 // foo( 935 // a, 936 // ); 937 // function foo( 938 // a, 939 // ) { 940 // code(); // 941 // } 942 if (Current.is(tok::r_paren) && State.Stack.size() > 1 && 943 (!Current.Next || Current.Next->isOneOf(tok::semi, tok::l_brace))) 944 return State.Stack[State.Stack.size() - 2].LastSpace; 945 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope()) 946 return State.Stack[State.Stack.size() - 2].LastSpace; 947 if (Current.is(tok::identifier) && Current.Next && 948 (Current.Next->is(TT_DictLiteral) || 949 ((Style.Language == FormatStyle::LK_Proto || 950 Style.Language == FormatStyle::LK_TextProto) && 951 Current.Next->isOneOf(tok::less, tok::l_brace)))) 952 return State.Stack.back().Indent; 953 if (NextNonComment->is(TT_ObjCStringLiteral) && 954 State.StartOfStringLiteral != 0) 955 return State.StartOfStringLiteral - 1; 956 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0) 957 return State.StartOfStringLiteral; 958 if (NextNonComment->is(tok::lessless) && 959 State.Stack.back().FirstLessLess != 0) 960 return State.Stack.back().FirstLessLess; 961 if (NextNonComment->isMemberAccess()) { 962 if (State.Stack.back().CallContinuation == 0) 963 return ContinuationIndent; 964 return State.Stack.back().CallContinuation; 965 } 966 if (State.Stack.back().QuestionColumn != 0 && 967 ((NextNonComment->is(tok::colon) && 968 NextNonComment->is(TT_ConditionalExpr)) || 969 Previous.is(TT_ConditionalExpr))) 970 return State.Stack.back().QuestionColumn; 971 if (Previous.is(tok::comma) && State.Stack.back().VariablePos != 0) 972 return State.Stack.back().VariablePos; 973 if ((PreviousNonComment && 974 (PreviousNonComment->ClosesTemplateDeclaration || 975 PreviousNonComment->isOneOf( 976 TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen, 977 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) || 978 (!Style.IndentWrappedFunctionNames && 979 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) 980 return std::max(State.Stack.back().LastSpace, State.Stack.back().Indent); 981 if (NextNonComment->is(TT_SelectorName)) { 982 if (!State.Stack.back().ObjCSelectorNameFound) { 983 unsigned MinIndent = State.Stack.back().Indent; 984 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) 985 MinIndent = std::max(MinIndent, 986 State.FirstIndent + Style.ContinuationIndentWidth); 987 // If LongestObjCSelectorName is 0, we are indenting the first 988 // part of an ObjC selector (or a selector component which is 989 // not colon-aligned due to block formatting). 990 // 991 // Otherwise, we are indenting a subsequent part of an ObjC 992 // selector which should be colon-aligned to the longest 993 // component of the ObjC selector. 994 // 995 // In either case, we want to respect Style.IndentWrappedFunctionNames. 996 return MinIndent + 997 std::max(NextNonComment->LongestObjCSelectorName, 998 NextNonComment->ColumnWidth) - 999 NextNonComment->ColumnWidth; 1000 } 1001 if (!State.Stack.back().AlignColons) 1002 return State.Stack.back().Indent; 1003 if (State.Stack.back().ColonPos > NextNonComment->ColumnWidth) 1004 return State.Stack.back().ColonPos - NextNonComment->ColumnWidth; 1005 return State.Stack.back().Indent; 1006 } 1007 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr)) 1008 return State.Stack.back().ColonPos; 1009 if (NextNonComment->is(TT_ArraySubscriptLSquare)) { 1010 if (State.Stack.back().StartOfArraySubscripts != 0) 1011 return State.Stack.back().StartOfArraySubscripts; 1012 return ContinuationIndent; 1013 } 1014 1015 // This ensure that we correctly format ObjC methods calls without inputs, 1016 // i.e. where the last element isn't selector like: [callee method]; 1017 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 && 1018 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) 1019 return State.Stack.back().Indent; 1020 1021 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) || 1022 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) 1023 return ContinuationIndent; 1024 if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 1025 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) 1026 return ContinuationIndent; 1027 if (NextNonComment->is(TT_CtorInitializerComma)) 1028 return State.Stack.back().Indent; 1029 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && 1030 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) 1031 return State.Stack.back().Indent; 1032 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) && 1033 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) 1034 return State.Stack.back().Indent; 1035 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon, 1036 TT_InheritanceComma)) 1037 return State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1038 if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() && 1039 !Current.isOneOf(tok::colon, tok::comment)) 1040 return ContinuationIndent; 1041 if (Current.is(TT_ProtoExtensionLSquare)) 1042 return State.Stack.back().Indent; 1043 if (State.Stack.back().Indent == State.FirstIndent && PreviousNonComment && 1044 PreviousNonComment->isNot(tok::r_brace)) 1045 // Ensure that we fall back to the continuation indent width instead of 1046 // just flushing continuations left. 1047 return State.Stack.back().Indent + Style.ContinuationIndentWidth; 1048 return State.Stack.back().Indent; 1049 } 1050 1051 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State, 1052 bool DryRun, bool Newline) { 1053 assert(State.Stack.size()); 1054 const FormatToken &Current = *State.NextToken; 1055 1056 if (Current.isOneOf(tok::comma, TT_BinaryOperator)) 1057 State.Stack.back().NoLineBreakInOperand = false; 1058 if (Current.is(TT_InheritanceColon)) 1059 State.Stack.back().AvoidBinPacking = true; 1060 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) { 1061 if (State.Stack.back().FirstLessLess == 0) 1062 State.Stack.back().FirstLessLess = State.Column; 1063 else 1064 State.Stack.back().LastOperatorWrapped = Newline; 1065 } 1066 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless)) 1067 State.Stack.back().LastOperatorWrapped = Newline; 1068 if (Current.is(TT_ConditionalExpr) && Current.Previous && 1069 !Current.Previous->is(TT_ConditionalExpr)) 1070 State.Stack.back().LastOperatorWrapped = Newline; 1071 if (Current.is(TT_ArraySubscriptLSquare) && 1072 State.Stack.back().StartOfArraySubscripts == 0) 1073 State.Stack.back().StartOfArraySubscripts = State.Column; 1074 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question)) 1075 State.Stack.back().QuestionColumn = State.Column; 1076 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) { 1077 const FormatToken *Previous = Current.Previous; 1078 while (Previous && Previous->isTrailingComment()) 1079 Previous = Previous->Previous; 1080 if (Previous && Previous->is(tok::question)) 1081 State.Stack.back().QuestionColumn = State.Column; 1082 } 1083 if (!Current.opensScope() && !Current.closesScope() && 1084 !Current.is(TT_PointerOrReference)) 1085 State.LowestLevelOnLine = 1086 std::min(State.LowestLevelOnLine, Current.NestingLevel); 1087 if (Current.isMemberAccess()) 1088 State.Stack.back().StartOfFunctionCall = 1089 !Current.NextOperator ? 0 : State.Column; 1090 if (Current.is(TT_SelectorName)) 1091 State.Stack.back().ObjCSelectorNameFound = true; 1092 if (Current.is(TT_CtorInitializerColon) && 1093 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) { 1094 // Indent 2 from the column, so: 1095 // SomeClass::SomeClass() 1096 // : First(...), ... 1097 // Next(...) 1098 // ^ line up here. 1099 State.Stack.back().Indent = 1100 State.Column + 1101 (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma 1102 ? 0 1103 : 2); 1104 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1105 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 1106 State.Stack.back().AvoidBinPacking = true; 1107 State.Stack.back().BreakBeforeParameter = false; 1108 } 1109 if (Current.is(TT_CtorInitializerColon) && 1110 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { 1111 State.Stack.back().Indent = 1112 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1113 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1114 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 1115 State.Stack.back().AvoidBinPacking = true; 1116 } 1117 if (Current.is(TT_InheritanceColon)) 1118 State.Stack.back().Indent = 1119 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1120 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline) 1121 State.Stack.back().NestedBlockIndent = 1122 State.Column + Current.ColumnWidth + 1; 1123 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow)) 1124 State.Stack.back().LastSpace = State.Column; 1125 1126 // Insert scopes created by fake parenthesis. 1127 const FormatToken *Previous = Current.getPreviousNonComment(); 1128 1129 // Add special behavior to support a format commonly used for JavaScript 1130 // closures: 1131 // SomeFunction(function() { 1132 // foo(); 1133 // bar(); 1134 // }, a, b, c); 1135 if (Current.isNot(tok::comment) && Previous && 1136 Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) && 1137 !Previous->is(TT_DictLiteral) && State.Stack.size() > 1 && 1138 !State.Stack.back().HasMultipleNestedBlocks) { 1139 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline) 1140 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 1141 State.Stack[i].NoLineBreak = true; 1142 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false; 1143 } 1144 if (Previous && 1145 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) || 1146 Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr)) && 1147 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 1148 State.Stack.back().NestedBlockInlined = 1149 !Newline && 1150 (Previous->isNot(tok::l_paren) || Previous->ParameterCount > 1); 1151 } 1152 1153 moveStatePastFakeLParens(State, Newline); 1154 moveStatePastScopeCloser(State); 1155 bool AllowBreak = !State.Stack.back().NoLineBreak && 1156 !State.Stack.back().NoLineBreakInOperand; 1157 moveStatePastScopeOpener(State, Newline); 1158 moveStatePastFakeRParens(State); 1159 1160 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0) 1161 State.StartOfStringLiteral = State.Column + 1; 1162 else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) 1163 State.StartOfStringLiteral = State.Column; 1164 else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) && 1165 !Current.isStringLiteral()) 1166 State.StartOfStringLiteral = 0; 1167 1168 State.Column += Current.ColumnWidth; 1169 State.NextToken = State.NextToken->Next; 1170 1171 unsigned Penalty = 1172 handleEndOfLine(Current, State, DryRun, AllowBreak); 1173 1174 if (Current.Role) 1175 Current.Role->formatFromToken(State, this, DryRun); 1176 // If the previous has a special role, let it consume tokens as appropriate. 1177 // It is necessary to start at the previous token for the only implemented 1178 // role (comma separated list). That way, the decision whether or not to break 1179 // after the "{" is already done and both options are tried and evaluated. 1180 // FIXME: This is ugly, find a better way. 1181 if (Previous && Previous->Role) 1182 Penalty += Previous->Role->formatAfterToken(State, this, DryRun); 1183 1184 return Penalty; 1185 } 1186 1187 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State, 1188 bool Newline) { 1189 const FormatToken &Current = *State.NextToken; 1190 const FormatToken *Previous = Current.getPreviousNonComment(); 1191 1192 // Don't add extra indentation for the first fake parenthesis after 1193 // 'return', assignments or opening <({[. The indentation for these cases 1194 // is special cased. 1195 bool SkipFirstExtraIndent = 1196 (Previous && (Previous->opensScope() || 1197 Previous->isOneOf(tok::semi, tok::kw_return) || 1198 (Previous->getPrecedence() == prec::Assignment && 1199 Style.AlignOperands) || 1200 Previous->is(TT_ObjCMethodExpr))); 1201 for (SmallVectorImpl<prec::Level>::const_reverse_iterator 1202 I = Current.FakeLParens.rbegin(), 1203 E = Current.FakeLParens.rend(); 1204 I != E; ++I) { 1205 ParenState NewParenState = State.Stack.back(); 1206 NewParenState.Tok = nullptr; 1207 NewParenState.ContainsLineBreak = false; 1208 NewParenState.LastOperatorWrapped = true; 1209 NewParenState.NoLineBreak = 1210 NewParenState.NoLineBreak || State.Stack.back().NoLineBreakInOperand; 1211 1212 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists. 1213 if (*I > prec::Comma) 1214 NewParenState.AvoidBinPacking = false; 1215 1216 // Indent from 'LastSpace' unless these are fake parentheses encapsulating 1217 // a builder type call after 'return' or, if the alignment after opening 1218 // brackets is disabled. 1219 if (!Current.isTrailingComment() && 1220 (Style.AlignOperands || *I < prec::Assignment) && 1221 (!Previous || Previous->isNot(tok::kw_return) || 1222 (Style.Language != FormatStyle::LK_Java && *I > 0)) && 1223 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign || 1224 *I != prec::Comma || Current.NestingLevel == 0)) 1225 NewParenState.Indent = 1226 std::max(std::max(State.Column, NewParenState.Indent), 1227 State.Stack.back().LastSpace); 1228 1229 // Do not indent relative to the fake parentheses inserted for "." or "->". 1230 // This is a special case to make the following to statements consistent: 1231 // OuterFunction(InnerFunctionCall( // break 1232 // ParameterToInnerFunction)); 1233 // OuterFunction(SomeObject.InnerFunctionCall( // break 1234 // ParameterToInnerFunction)); 1235 if (*I > prec::Unknown) 1236 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column); 1237 if (*I != prec::Conditional && !Current.is(TT_UnaryOperator) && 1238 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 1239 NewParenState.StartOfFunctionCall = State.Column; 1240 1241 // Always indent conditional expressions. Never indent expression where 1242 // the 'operator' is ',', ';' or an assignment (i.e. *I <= 1243 // prec::Assignment) as those have different indentation rules. Indent 1244 // other expression, unless the indentation needs to be skipped. 1245 if (*I == prec::Conditional || 1246 (!SkipFirstExtraIndent && *I > prec::Assignment && 1247 !Current.isTrailingComment())) 1248 NewParenState.Indent += Style.ContinuationIndentWidth; 1249 if ((Previous && !Previous->opensScope()) || *I != prec::Comma) 1250 NewParenState.BreakBeforeParameter = false; 1251 State.Stack.push_back(NewParenState); 1252 SkipFirstExtraIndent = false; 1253 } 1254 } 1255 1256 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) { 1257 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) { 1258 unsigned VariablePos = State.Stack.back().VariablePos; 1259 if (State.Stack.size() == 1) { 1260 // Do not pop the last element. 1261 break; 1262 } 1263 State.Stack.pop_back(); 1264 State.Stack.back().VariablePos = VariablePos; 1265 } 1266 } 1267 1268 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State, 1269 bool Newline) { 1270 const FormatToken &Current = *State.NextToken; 1271 if (!Current.opensScope()) 1272 return; 1273 1274 if (Current.MatchingParen && Current.BlockKind == BK_Block) { 1275 moveStateToNewBlock(State); 1276 return; 1277 } 1278 1279 unsigned NewIndent; 1280 unsigned LastSpace = State.Stack.back().LastSpace; 1281 bool AvoidBinPacking; 1282 bool BreakBeforeParameter = false; 1283 unsigned NestedBlockIndent = std::max(State.Stack.back().StartOfFunctionCall, 1284 State.Stack.back().NestedBlockIndent); 1285 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 1286 opensProtoMessageField(Current, Style)) { 1287 if (Current.opensBlockOrBlockTypeList(Style)) { 1288 NewIndent = Style.IndentWidth + 1289 std::min(State.Column, State.Stack.back().NestedBlockIndent); 1290 } else { 1291 NewIndent = State.Stack.back().LastSpace + Style.ContinuationIndentWidth; 1292 } 1293 const FormatToken *NextNoComment = Current.getNextNonComment(); 1294 bool EndsInComma = Current.MatchingParen && 1295 Current.MatchingParen->Previous && 1296 Current.MatchingParen->Previous->is(tok::comma); 1297 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) || 1298 Style.Language == FormatStyle::LK_Proto || 1299 Style.Language == FormatStyle::LK_TextProto || 1300 !Style.BinPackArguments || 1301 (NextNoComment && 1302 NextNoComment->isOneOf(TT_DesignatedInitializerPeriod, 1303 TT_DesignatedInitializerLSquare)); 1304 BreakBeforeParameter = EndsInComma; 1305 if (Current.ParameterCount > 1) 1306 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1); 1307 } else { 1308 NewIndent = Style.ContinuationIndentWidth + 1309 std::max(State.Stack.back().LastSpace, 1310 State.Stack.back().StartOfFunctionCall); 1311 1312 // Ensure that different different brackets force relative alignment, e.g.: 1313 // void SomeFunction(vector< // break 1314 // int> v); 1315 // FIXME: We likely want to do this for more combinations of brackets. 1316 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) { 1317 NewIndent = std::max(NewIndent, State.Stack.back().Indent); 1318 LastSpace = std::max(LastSpace, State.Stack.back().Indent); 1319 } 1320 1321 bool EndsInComma = 1322 Current.MatchingParen && 1323 Current.MatchingParen->getPreviousNonComment() && 1324 Current.MatchingParen->getPreviousNonComment()->is(tok::comma); 1325 1326 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters 1327 // for backwards compatibility. 1328 bool ObjCBinPackProtocolList = 1329 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto && 1330 Style.BinPackParameters) || 1331 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always; 1332 1333 bool BinPackDeclaration = 1334 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) || 1335 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList); 1336 1337 AvoidBinPacking = 1338 (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) || 1339 (State.Line->MustBeDeclaration && !BinPackDeclaration) || 1340 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) || 1341 (Style.ExperimentalAutoDetectBinPacking && 1342 (Current.PackingKind == PPK_OnePerLine || 1343 (!BinPackInconclusiveFunctions && 1344 Current.PackingKind == PPK_Inconclusive))); 1345 1346 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen) { 1347 if (Style.ColumnLimit) { 1348 // If this '[' opens an ObjC call, determine whether all parameters fit 1349 // into one line and put one per line if they don't. 1350 if (getLengthToMatchingParen(Current, State.Stack) + State.Column > 1351 getColumnLimit(State)) 1352 BreakBeforeParameter = true; 1353 } else { 1354 // For ColumnLimit = 0, we have to figure out whether there is or has to 1355 // be a line break within this call. 1356 for (const FormatToken *Tok = &Current; 1357 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) { 1358 if (Tok->MustBreakBefore || 1359 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) { 1360 BreakBeforeParameter = true; 1361 break; 1362 } 1363 } 1364 } 1365 } 1366 1367 if (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) 1368 BreakBeforeParameter = true; 1369 } 1370 // Generally inherit NoLineBreak from the current scope to nested scope. 1371 // However, don't do this for non-empty nested blocks, dict literals and 1372 // array literals as these follow different indentation rules. 1373 bool NoLineBreak = 1374 Current.Children.empty() && 1375 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) && 1376 (State.Stack.back().NoLineBreak || 1377 State.Stack.back().NoLineBreakInOperand || 1378 (Current.is(TT_TemplateOpener) && 1379 State.Stack.back().ContainsUnwrappedBuilder)); 1380 State.Stack.push_back( 1381 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak)); 1382 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1383 State.Stack.back().BreakBeforeParameter = BreakBeforeParameter; 1384 State.Stack.back().HasMultipleNestedBlocks = Current.BlockParameterCount > 1; 1385 State.Stack.back().IsInsideObjCArrayLiteral = 1386 Current.is(TT_ArrayInitializerLSquare) && Current.Previous && 1387 Current.Previous->is(tok::at); 1388 } 1389 1390 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) { 1391 const FormatToken &Current = *State.NextToken; 1392 if (!Current.closesScope()) 1393 return; 1394 1395 // If we encounter a closing ), ], } or >, we can remove a level from our 1396 // stacks. 1397 if (State.Stack.size() > 1 && 1398 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) || 1399 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) || 1400 State.NextToken->is(TT_TemplateCloser) || 1401 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) 1402 State.Stack.pop_back(); 1403 1404 // Reevaluate whether ObjC message arguments fit into one line. 1405 // If a receiver spans multiple lines, e.g.: 1406 // [[object block:^{ 1407 // return 42; 1408 // }] a:42 b:42]; 1409 // BreakBeforeParameter is calculated based on an incorrect assumption 1410 // (it is checked whether the whole expression fits into one line without 1411 // considering a line break inside a message receiver). 1412 // We check whether arguements fit after receiver scope closer (into the same 1413 // line). 1414 if (State.Stack.back().BreakBeforeParameter && Current.MatchingParen && 1415 Current.MatchingParen->Previous) { 1416 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous; 1417 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) && 1418 CurrentScopeOpener.MatchingParen) { 1419 int NecessarySpaceInLine = 1420 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) + 1421 CurrentScopeOpener.TotalLength - Current.TotalLength - 1; 1422 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <= 1423 Style.ColumnLimit) 1424 State.Stack.back().BreakBeforeParameter = false; 1425 } 1426 } 1427 1428 if (Current.is(tok::r_square)) { 1429 // If this ends the array subscript expr, reset the corresponding value. 1430 const FormatToken *NextNonComment = Current.getNextNonComment(); 1431 if (NextNonComment && NextNonComment->isNot(tok::l_square)) 1432 State.Stack.back().StartOfArraySubscripts = 0; 1433 } 1434 } 1435 1436 void ContinuationIndenter::moveStateToNewBlock(LineState &State) { 1437 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent; 1438 // ObjC block sometimes follow special indentation rules. 1439 unsigned NewIndent = 1440 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace) 1441 ? Style.ObjCBlockIndentWidth 1442 : Style.IndentWidth); 1443 State.Stack.push_back(ParenState(State.NextToken, NewIndent, 1444 State.Stack.back().LastSpace, 1445 /*AvoidBinPacking=*/true, 1446 /*NoLineBreak=*/false)); 1447 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1448 State.Stack.back().BreakBeforeParameter = true; 1449 } 1450 1451 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn, 1452 unsigned TabWidth, 1453 encoding::Encoding Encoding) { 1454 size_t LastNewlinePos = Text.find_last_of("\n"); 1455 if (LastNewlinePos == StringRef::npos) { 1456 return StartColumn + 1457 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding); 1458 } else { 1459 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos), 1460 /*StartColumn=*/0, TabWidth, Encoding); 1461 } 1462 } 1463 1464 unsigned ContinuationIndenter::reformatRawStringLiteral( 1465 const FormatToken &Current, LineState &State, 1466 const FormatStyle &RawStringStyle, bool DryRun) { 1467 unsigned StartColumn = State.Column - Current.ColumnWidth; 1468 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText); 1469 StringRef NewDelimiter = 1470 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language); 1471 if (NewDelimiter.empty() || OldDelimiter.empty()) 1472 NewDelimiter = OldDelimiter; 1473 // The text of a raw string is between the leading 'R"delimiter(' and the 1474 // trailing 'delimiter)"'. 1475 unsigned OldPrefixSize = 3 + OldDelimiter.size(); 1476 unsigned OldSuffixSize = 2 + OldDelimiter.size(); 1477 // We create a virtual text environment which expects a null-terminated 1478 // string, so we cannot use StringRef. 1479 std::string RawText = 1480 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize); 1481 if (NewDelimiter != OldDelimiter) { 1482 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the 1483 // raw string. 1484 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str(); 1485 if (StringRef(RawText).contains(CanonicalDelimiterSuffix)) 1486 NewDelimiter = OldDelimiter; 1487 } 1488 1489 unsigned NewPrefixSize = 3 + NewDelimiter.size(); 1490 unsigned NewSuffixSize = 2 + NewDelimiter.size(); 1491 1492 // The first start column is the column the raw text starts after formatting. 1493 unsigned FirstStartColumn = StartColumn + NewPrefixSize; 1494 1495 // The next start column is the intended indentation a line break inside 1496 // the raw string at level 0. It is determined by the following rules: 1497 // - if the content starts on newline, it is one level more than the current 1498 // indent, and 1499 // - if the content does not start on a newline, it is the first start 1500 // column. 1501 // These rules have the advantage that the formatted content both does not 1502 // violate the rectangle rule and visually flows within the surrounding 1503 // source. 1504 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n'; 1505 // If this token is the last parameter (checked by looking if it's followed by 1506 // `)`, the base the indent off the line's nested block indent. Otherwise, 1507 // base the indent off the arguments indent, so we can achieve: 1508 // fffffffffff(1, 2, 3, R"pb( 1509 // key1: 1 # 1510 // key2: 2)pb"); 1511 // 1512 // fffffffffff(1, 2, 3, 1513 // R"pb( 1514 // key1: 1 # 1515 // key2: 2 1516 // )pb", 1517 // 5); 1518 unsigned CurrentIndent = (Current.Next && Current.Next->is(tok::r_paren)) 1519 ? State.Stack.back().NestedBlockIndent 1520 : State.Stack.back().Indent; 1521 unsigned NextStartColumn = ContentStartsOnNewline 1522 ? CurrentIndent + Style.IndentWidth 1523 : FirstStartColumn; 1524 1525 // The last start column is the column the raw string suffix starts if it is 1526 // put on a newline. 1527 // The last start column is the intended indentation of the raw string postfix 1528 // if it is put on a newline. It is determined by the following rules: 1529 // - if the raw string prefix starts on a newline, it is the column where 1530 // that raw string prefix starts, and 1531 // - if the raw string prefix does not start on a newline, it is the current 1532 // indent. 1533 unsigned LastStartColumn = Current.NewlinesBefore 1534 ? FirstStartColumn - NewPrefixSize 1535 : CurrentIndent; 1536 1537 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat( 1538 RawStringStyle, RawText, {tooling::Range(0, RawText.size())}, 1539 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>", 1540 /*Status=*/nullptr); 1541 1542 auto NewCode = applyAllReplacements(RawText, Fixes.first); 1543 tooling::Replacements NoFixes; 1544 if (!NewCode) { 1545 return addMultilineToken(Current, State); 1546 } 1547 if (!DryRun) { 1548 if (NewDelimiter != OldDelimiter) { 1549 // In 'R"delimiter(...', the delimiter starts 2 characters after the start 1550 // of the token. 1551 SourceLocation PrefixDelimiterStart = 1552 Current.Tok.getLocation().getLocWithOffset(2); 1553 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement( 1554 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1555 if (PrefixErr) { 1556 llvm::errs() 1557 << "Failed to update the prefix delimiter of a raw string: " 1558 << llvm::toString(std::move(PrefixErr)) << "\n"; 1559 } 1560 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at 1561 // position length - 1 - |delimiter|. 1562 SourceLocation SuffixDelimiterStart = 1563 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() - 1564 1 - OldDelimiter.size()); 1565 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement( 1566 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1567 if (SuffixErr) { 1568 llvm::errs() 1569 << "Failed to update the suffix delimiter of a raw string: " 1570 << llvm::toString(std::move(SuffixErr)) << "\n"; 1571 } 1572 } 1573 SourceLocation OriginLoc = 1574 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize); 1575 for (const tooling::Replacement &Fix : Fixes.first) { 1576 auto Err = Whitespaces.addReplacement(tooling::Replacement( 1577 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()), 1578 Fix.getLength(), Fix.getReplacementText())); 1579 if (Err) { 1580 llvm::errs() << "Failed to reformat raw string: " 1581 << llvm::toString(std::move(Err)) << "\n"; 1582 } 1583 } 1584 } 1585 unsigned RawLastLineEndColumn = getLastLineEndColumn( 1586 *NewCode, FirstStartColumn, Style.TabWidth, Encoding); 1587 State.Column = RawLastLineEndColumn + NewSuffixSize; 1588 // Since we're updating the column to after the raw string literal here, we 1589 // have to manually add the penalty for the prefix R"delim( over the column 1590 // limit. 1591 unsigned PrefixExcessCharacters = 1592 StartColumn + NewPrefixSize > Style.ColumnLimit ? 1593 StartColumn + NewPrefixSize - Style.ColumnLimit : 0; 1594 bool IsMultiline = 1595 ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos); 1596 if (IsMultiline) { 1597 // Break before further function parameters on all levels. 1598 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 1599 State.Stack[i].BreakBeforeParameter = true; 1600 } 1601 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter; 1602 } 1603 1604 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current, 1605 LineState &State) { 1606 // Break before further function parameters on all levels. 1607 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 1608 State.Stack[i].BreakBeforeParameter = true; 1609 1610 unsigned ColumnsUsed = State.Column; 1611 // We can only affect layout of the first and the last line, so the penalty 1612 // for all other lines is constant, and we ignore it. 1613 State.Column = Current.LastLineColumnWidth; 1614 1615 if (ColumnsUsed > getColumnLimit(State)) 1616 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State)); 1617 return 0; 1618 } 1619 1620 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current, 1621 LineState &State, bool DryRun, 1622 bool AllowBreak) { 1623 unsigned Penalty = 0; 1624 // Compute the raw string style to use in case this is a raw string literal 1625 // that can be reformatted. 1626 auto RawStringStyle = getRawStringStyle(Current, State); 1627 if (RawStringStyle && !Current.Finalized) { 1628 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun); 1629 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) { 1630 // Don't break multi-line tokens other than block comments and raw string 1631 // literals. Instead, just update the state. 1632 Penalty = addMultilineToken(Current, State); 1633 } else if (State.Line->Type != LT_ImportStatement) { 1634 // We generally don't break import statements. 1635 LineState OriginalState = State; 1636 1637 // Whether we force the reflowing algorithm to stay strictly within the 1638 // column limit. 1639 bool Strict = false; 1640 // Whether the first non-strict attempt at reflowing did intentionally 1641 // exceed the column limit. 1642 bool Exceeded = false; 1643 std::tie(Penalty, Exceeded) = breakProtrudingToken( 1644 Current, State, AllowBreak, /*DryRun=*/true, Strict); 1645 if (Exceeded) { 1646 // If non-strict reflowing exceeds the column limit, try whether strict 1647 // reflowing leads to an overall lower penalty. 1648 LineState StrictState = OriginalState; 1649 unsigned StrictPenalty = 1650 breakProtrudingToken(Current, StrictState, AllowBreak, 1651 /*DryRun=*/true, /*Strict=*/true) 1652 .first; 1653 Strict = StrictPenalty <= Penalty; 1654 if (Strict) { 1655 Penalty = StrictPenalty; 1656 State = StrictState; 1657 } 1658 } 1659 if (!DryRun) { 1660 // If we're not in dry-run mode, apply the changes with the decision on 1661 // strictness made above. 1662 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false, 1663 Strict); 1664 } 1665 } 1666 if (State.Column > getColumnLimit(State)) { 1667 unsigned ExcessCharacters = State.Column - getColumnLimit(State); 1668 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters; 1669 } 1670 return Penalty; 1671 } 1672 1673 // Returns the enclosing function name of a token, or the empty string if not 1674 // found. 1675 static StringRef getEnclosingFunctionName(const FormatToken &Current) { 1676 // Look for: 'function(' or 'function<templates>(' before Current. 1677 auto Tok = Current.getPreviousNonComment(); 1678 if (!Tok || !Tok->is(tok::l_paren)) 1679 return ""; 1680 Tok = Tok->getPreviousNonComment(); 1681 if (!Tok) 1682 return ""; 1683 if (Tok->is(TT_TemplateCloser)) { 1684 Tok = Tok->MatchingParen; 1685 if (Tok) 1686 Tok = Tok->getPreviousNonComment(); 1687 } 1688 if (!Tok || !Tok->is(tok::identifier)) 1689 return ""; 1690 return Tok->TokenText; 1691 } 1692 1693 llvm::Optional<FormatStyle> 1694 ContinuationIndenter::getRawStringStyle(const FormatToken &Current, 1695 const LineState &State) { 1696 if (!Current.isStringLiteral()) 1697 return None; 1698 auto Delimiter = getRawStringDelimiter(Current.TokenText); 1699 if (!Delimiter) 1700 return None; 1701 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter); 1702 if (!RawStringStyle && Delimiter->empty()) 1703 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle( 1704 getEnclosingFunctionName(Current)); 1705 if (!RawStringStyle) 1706 return None; 1707 RawStringStyle->ColumnLimit = getColumnLimit(State); 1708 return RawStringStyle; 1709 } 1710 1711 std::unique_ptr<BreakableToken> ContinuationIndenter::createBreakableToken( 1712 const FormatToken &Current, LineState &State, bool AllowBreak) { 1713 unsigned StartColumn = State.Column - Current.ColumnWidth; 1714 if (Current.isStringLiteral()) { 1715 // FIXME: String literal breaking is currently disabled for Java and JS, as 1716 // it requires strings to be merged using "+" which we don't support. 1717 if (Style.Language == FormatStyle::LK_Java || 1718 Style.Language == FormatStyle::LK_JavaScript || 1719 !Style.BreakStringLiterals || 1720 !AllowBreak) 1721 return nullptr; 1722 1723 // Don't break string literals inside preprocessor directives (except for 1724 // #define directives, as their contents are stored in separate lines and 1725 // are not affected by this check). 1726 // This way we avoid breaking code with line directives and unknown 1727 // preprocessor directives that contain long string literals. 1728 if (State.Line->Type == LT_PreprocessorDirective) 1729 return nullptr; 1730 // Exempts unterminated string literals from line breaking. The user will 1731 // likely want to terminate the string before any line breaking is done. 1732 if (Current.IsUnterminatedLiteral) 1733 return nullptr; 1734 // Don't break string literals inside Objective-C array literals (doing so 1735 // raises the warning -Wobjc-string-concatenation). 1736 if (State.Stack.back().IsInsideObjCArrayLiteral) { 1737 return nullptr; 1738 } 1739 1740 StringRef Text = Current.TokenText; 1741 StringRef Prefix; 1742 StringRef Postfix; 1743 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'. 1744 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to 1745 // reduce the overhead) for each FormatToken, which is a string, so that we 1746 // don't run multiple checks here on the hot path. 1747 if ((Text.endswith(Postfix = "\"") && 1748 (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") || 1749 Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") || 1750 Text.startswith(Prefix = "u8\"") || 1751 Text.startswith(Prefix = "L\""))) || 1752 (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) { 1753 // We need this to address the case where there is an unbreakable tail 1754 // only if certain other formatting decisions have been taken. The 1755 // UnbreakableTailLength of Current is an overapproximation is that case 1756 // and we need to be correct here. 1757 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State)) 1758 ? 0 1759 : Current.UnbreakableTailLength; 1760 return llvm::make_unique<BreakableStringLiteral>( 1761 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength, 1762 State.Line->InPPDirective, Encoding, Style); 1763 } 1764 } else if (Current.is(TT_BlockComment)) { 1765 if (!Style.ReflowComments || 1766 // If a comment token switches formatting, like 1767 // /* clang-format on */, we don't want to break it further, 1768 // but we may still want to adjust its indentation. 1769 switchesFormatting(Current)) { 1770 return nullptr; 1771 } 1772 return llvm::make_unique<BreakableBlockComment>( 1773 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1774 State.Line->InPPDirective, Encoding, Style); 1775 } else if (Current.is(TT_LineComment) && 1776 (Current.Previous == nullptr || 1777 Current.Previous->isNot(TT_ImplicitStringLiteral))) { 1778 if (!Style.ReflowComments || 1779 CommentPragmasRegex.match(Current.TokenText.substr(2)) || 1780 switchesFormatting(Current)) 1781 return nullptr; 1782 return llvm::make_unique<BreakableLineCommentSection>( 1783 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1784 /*InPPDirective=*/false, Encoding, Style); 1785 } 1786 return nullptr; 1787 } 1788 1789 std::pair<unsigned, bool> 1790 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current, 1791 LineState &State, bool AllowBreak, 1792 bool DryRun, bool Strict) { 1793 std::unique_ptr<const BreakableToken> Token = 1794 createBreakableToken(Current, State, AllowBreak); 1795 if (!Token) 1796 return {0, false}; 1797 assert(Token->getLineCount() > 0); 1798 unsigned ColumnLimit = getColumnLimit(State); 1799 if (Current.is(TT_LineComment)) { 1800 // We don't insert backslashes when breaking line comments. 1801 ColumnLimit = Style.ColumnLimit; 1802 } 1803 if (Current.UnbreakableTailLength >= ColumnLimit) 1804 return {0, false}; 1805 // ColumnWidth was already accounted into State.Column before calling 1806 // breakProtrudingToken. 1807 unsigned StartColumn = State.Column - Current.ColumnWidth; 1808 unsigned NewBreakPenalty = Current.isStringLiteral() 1809 ? Style.PenaltyBreakString 1810 : Style.PenaltyBreakComment; 1811 // Stores whether we intentionally decide to let a line exceed the column 1812 // limit. 1813 bool Exceeded = false; 1814 // Stores whether we introduce a break anywhere in the token. 1815 bool BreakInserted = Token->introducesBreakBeforeToken(); 1816 // Store whether we inserted a new line break at the end of the previous 1817 // logical line. 1818 bool NewBreakBefore = false; 1819 // We use a conservative reflowing strategy. Reflow starts after a line is 1820 // broken or the corresponding whitespace compressed. Reflow ends as soon as a 1821 // line that doesn't get reflown with the previous line is reached. 1822 bool Reflow = false; 1823 // Keep track of where we are in the token: 1824 // Where we are in the content of the current logical line. 1825 unsigned TailOffset = 0; 1826 // The column number we're currently at. 1827 unsigned ContentStartColumn = 1828 Token->getContentStartColumn(0, /*Break=*/false); 1829 // The number of columns left in the current logical line after TailOffset. 1830 unsigned RemainingTokenColumns = 1831 Token->getRemainingLength(0, TailOffset, ContentStartColumn); 1832 // Adapt the start of the token, for example indent. 1833 if (!DryRun) 1834 Token->adaptStartOfLine(0, Whitespaces); 1835 1836 unsigned ContentIndent = 0; 1837 unsigned Penalty = 0; 1838 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column " 1839 << StartColumn << ".\n"); 1840 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount(); 1841 LineIndex != EndIndex; ++LineIndex) { 1842 LLVM_DEBUG(llvm::dbgs() 1843 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n"); 1844 NewBreakBefore = false; 1845 // If we did reflow the previous line, we'll try reflowing again. Otherwise 1846 // we'll start reflowing if the current line is broken or whitespace is 1847 // compressed. 1848 bool TryReflow = Reflow; 1849 // Break the current token until we can fit the rest of the line. 1850 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 1851 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: " 1852 << (ContentStartColumn + RemainingTokenColumns) 1853 << ", space: " << ColumnLimit 1854 << ", reflown prefix: " << ContentStartColumn 1855 << ", offset in line: " << TailOffset << "\n"); 1856 // If the current token doesn't fit, find the latest possible split in the 1857 // current line so that breaking at it will be under the column limit. 1858 // FIXME: Use the earliest possible split while reflowing to correctly 1859 // compress whitespace within a line. 1860 BreakableToken::Split Split = 1861 Token->getSplit(LineIndex, TailOffset, ColumnLimit, 1862 ContentStartColumn, CommentPragmasRegex); 1863 if (Split.first == StringRef::npos) { 1864 // No break opportunity - update the penalty and continue with the next 1865 // logical line. 1866 if (LineIndex < EndIndex - 1) 1867 // The last line's penalty is handled in addNextStateToQueue() or when 1868 // calling replaceWhitespaceAfterLastLine below. 1869 Penalty += Style.PenaltyExcessCharacter * 1870 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 1871 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n"); 1872 break; 1873 } 1874 assert(Split.first != 0); 1875 1876 if (Token->supportsReflow()) { 1877 // Check whether the next natural split point after the current one can 1878 // still fit the line, either because we can compress away whitespace, 1879 // or because the penalty the excess characters introduce is lower than 1880 // the break penalty. 1881 // We only do this for tokens that support reflowing, and thus allow us 1882 // to change the whitespace arbitrarily (e.g. comments). 1883 // Other tokens, like string literals, can be broken on arbitrary 1884 // positions. 1885 1886 // First, compute the columns from TailOffset to the next possible split 1887 // position. 1888 // For example: 1889 // ColumnLimit: | 1890 // // Some text that breaks 1891 // ^ tail offset 1892 // ^-- split 1893 // ^-------- to split columns 1894 // ^--- next split 1895 // ^--------------- to next split columns 1896 unsigned ToSplitColumns = Token->getRangeLength( 1897 LineIndex, TailOffset, Split.first, ContentStartColumn); 1898 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n"); 1899 1900 BreakableToken::Split NextSplit = Token->getSplit( 1901 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit, 1902 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex); 1903 // Compute the columns necessary to fit the next non-breakable sequence 1904 // into the current line. 1905 unsigned ToNextSplitColumns = 0; 1906 if (NextSplit.first == StringRef::npos) { 1907 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset, 1908 ContentStartColumn); 1909 } else { 1910 ToNextSplitColumns = Token->getRangeLength( 1911 LineIndex, TailOffset, 1912 Split.first + Split.second + NextSplit.first, ContentStartColumn); 1913 } 1914 // Compress the whitespace between the break and the start of the next 1915 // unbreakable sequence. 1916 ToNextSplitColumns = 1917 Token->getLengthAfterCompression(ToNextSplitColumns, Split); 1918 LLVM_DEBUG(llvm::dbgs() 1919 << " ContentStartColumn: " << ContentStartColumn << "\n"); 1920 LLVM_DEBUG(llvm::dbgs() 1921 << " ToNextSplit: " << ToNextSplitColumns << "\n"); 1922 // If the whitespace compression makes us fit, continue on the current 1923 // line. 1924 bool ContinueOnLine = 1925 ContentStartColumn + ToNextSplitColumns <= ColumnLimit; 1926 unsigned ExcessCharactersPenalty = 0; 1927 if (!ContinueOnLine && !Strict) { 1928 // Similarly, if the excess characters' penalty is lower than the 1929 // penalty of introducing a new break, continue on the current line. 1930 ExcessCharactersPenalty = 1931 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) * 1932 Style.PenaltyExcessCharacter; 1933 LLVM_DEBUG(llvm::dbgs() 1934 << " Penalty excess: " << ExcessCharactersPenalty 1935 << "\n break : " << NewBreakPenalty << "\n"); 1936 if (ExcessCharactersPenalty < NewBreakPenalty) { 1937 Exceeded = true; 1938 ContinueOnLine = true; 1939 } 1940 } 1941 if (ContinueOnLine) { 1942 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n"); 1943 // The current line fits after compressing the whitespace - reflow 1944 // the next line into it if possible. 1945 TryReflow = true; 1946 if (!DryRun) 1947 Token->compressWhitespace(LineIndex, TailOffset, Split, 1948 Whitespaces); 1949 // When we continue on the same line, leave one space between content. 1950 ContentStartColumn += ToSplitColumns + 1; 1951 Penalty += ExcessCharactersPenalty; 1952 TailOffset += Split.first + Split.second; 1953 RemainingTokenColumns = Token->getRemainingLength( 1954 LineIndex, TailOffset, ContentStartColumn); 1955 continue; 1956 } 1957 } 1958 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n"); 1959 // Update the ContentIndent only if the current line was not reflown with 1960 // the previous line, since in that case the previous line should still 1961 // determine the ContentIndent. Also never intent the last line. 1962 if (!Reflow) 1963 ContentIndent = Token->getContentIndent(LineIndex); 1964 LLVM_DEBUG(llvm::dbgs() 1965 << " ContentIndent: " << ContentIndent << "\n"); 1966 ContentStartColumn = ContentIndent + Token->getContentStartColumn( 1967 LineIndex, /*Break=*/true); 1968 1969 unsigned NewRemainingTokenColumns = Token->getRemainingLength( 1970 LineIndex, TailOffset + Split.first + Split.second, 1971 ContentStartColumn); 1972 if (NewRemainingTokenColumns == 0) { 1973 // No content to indent. 1974 ContentIndent = 0; 1975 ContentStartColumn = 1976 Token->getContentStartColumn(LineIndex, /*Break=*/true); 1977 NewRemainingTokenColumns = Token->getRemainingLength( 1978 LineIndex, TailOffset + Split.first + Split.second, 1979 ContentStartColumn); 1980 } 1981 1982 // When breaking before a tab character, it may be moved by a few columns, 1983 // but will still be expanded to the next tab stop, so we don't save any 1984 // columns. 1985 if (NewRemainingTokenColumns == RemainingTokenColumns) { 1986 // FIXME: Do we need to adjust the penalty? 1987 break; 1988 } 1989 assert(NewRemainingTokenColumns < RemainingTokenColumns); 1990 1991 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first 1992 << ", " << Split.second << "\n"); 1993 if (!DryRun) 1994 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent, 1995 Whitespaces); 1996 1997 Penalty += NewBreakPenalty; 1998 TailOffset += Split.first + Split.second; 1999 RemainingTokenColumns = NewRemainingTokenColumns; 2000 BreakInserted = true; 2001 NewBreakBefore = true; 2002 } 2003 // In case there's another line, prepare the state for the start of the next 2004 // line. 2005 if (LineIndex + 1 != EndIndex) { 2006 unsigned NextLineIndex = LineIndex + 1; 2007 if (NewBreakBefore) 2008 // After breaking a line, try to reflow the next line into the current 2009 // one once RemainingTokenColumns fits. 2010 TryReflow = true; 2011 if (TryReflow) { 2012 // We decided that we want to try reflowing the next line into the 2013 // current one. 2014 // We will now adjust the state as if the reflow is successful (in 2015 // preparation for the next line), and see whether that works. If we 2016 // decide that we cannot reflow, we will later reset the state to the 2017 // start of the next line. 2018 Reflow = false; 2019 // As we did not continue breaking the line, RemainingTokenColumns is 2020 // known to fit after ContentStartColumn. Adapt ContentStartColumn to 2021 // the position at which we want to format the next line if we do 2022 // actually reflow. 2023 // When we reflow, we need to add a space between the end of the current 2024 // line and the next line's start column. 2025 ContentStartColumn += RemainingTokenColumns + 1; 2026 // Get the split that we need to reflow next logical line into the end 2027 // of the current one; the split will include any leading whitespace of 2028 // the next logical line. 2029 BreakableToken::Split SplitBeforeNext = 2030 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex); 2031 LLVM_DEBUG(llvm::dbgs() 2032 << " Size of reflown text: " << ContentStartColumn 2033 << "\n Potential reflow split: "); 2034 if (SplitBeforeNext.first != StringRef::npos) { 2035 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", " 2036 << SplitBeforeNext.second << "\n"); 2037 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second; 2038 // If the rest of the next line fits into the current line below the 2039 // column limit, we can safely reflow. 2040 RemainingTokenColumns = Token->getRemainingLength( 2041 NextLineIndex, TailOffset, ContentStartColumn); 2042 Reflow = true; 2043 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2044 LLVM_DEBUG(llvm::dbgs() 2045 << " Over limit after reflow, need: " 2046 << (ContentStartColumn + RemainingTokenColumns) 2047 << ", space: " << ColumnLimit 2048 << ", reflown prefix: " << ContentStartColumn 2049 << ", offset in line: " << TailOffset << "\n"); 2050 // If the whole next line does not fit, try to find a point in 2051 // the next line at which we can break so that attaching the part 2052 // of the next line to that break point onto the current line is 2053 // below the column limit. 2054 BreakableToken::Split Split = 2055 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit, 2056 ContentStartColumn, CommentPragmasRegex); 2057 if (Split.first == StringRef::npos) { 2058 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n"); 2059 Reflow = false; 2060 } else { 2061 // Check whether the first split point gets us below the column 2062 // limit. Note that we will execute this split below as part of 2063 // the normal token breaking and reflow logic within the line. 2064 unsigned ToSplitColumns = Token->getRangeLength( 2065 NextLineIndex, TailOffset, Split.first, ContentStartColumn); 2066 if (ContentStartColumn + ToSplitColumns > ColumnLimit) { 2067 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: " 2068 << (ContentStartColumn + ToSplitColumns) 2069 << ", space: " << ColumnLimit); 2070 unsigned ExcessCharactersPenalty = 2071 (ContentStartColumn + ToSplitColumns - ColumnLimit) * 2072 Style.PenaltyExcessCharacter; 2073 if (NewBreakPenalty < ExcessCharactersPenalty) { 2074 Reflow = false; 2075 } 2076 } 2077 } 2078 } 2079 } else { 2080 LLVM_DEBUG(llvm::dbgs() << "not found.\n"); 2081 } 2082 } 2083 if (!Reflow) { 2084 // If we didn't reflow into the next line, the only space to consider is 2085 // the next logical line. Reset our state to match the start of the next 2086 // line. 2087 TailOffset = 0; 2088 ContentStartColumn = 2089 Token->getContentStartColumn(NextLineIndex, /*Break=*/false); 2090 RemainingTokenColumns = Token->getRemainingLength( 2091 NextLineIndex, TailOffset, ContentStartColumn); 2092 // Adapt the start of the token, for example indent. 2093 if (!DryRun) 2094 Token->adaptStartOfLine(NextLineIndex, Whitespaces); 2095 } else { 2096 // If we found a reflow split and have added a new break before the next 2097 // line, we are going to remove the line break at the start of the next 2098 // logical line. For example, here we'll add a new line break after 2099 // 'text', and subsequently delete the line break between 'that' and 2100 // 'reflows'. 2101 // // some text that 2102 // // reflows 2103 // -> 2104 // // some text 2105 // // that reflows 2106 // When adding the line break, we also added the penalty for it, so we 2107 // need to subtract that penalty again when we remove the line break due 2108 // to reflowing. 2109 if (NewBreakBefore) { 2110 assert(Penalty >= NewBreakPenalty); 2111 Penalty -= NewBreakPenalty; 2112 } 2113 if (!DryRun) 2114 Token->reflow(NextLineIndex, Whitespaces); 2115 } 2116 } 2117 } 2118 2119 BreakableToken::Split SplitAfterLastLine = 2120 Token->getSplitAfterLastLine(TailOffset); 2121 if (SplitAfterLastLine.first != StringRef::npos) { 2122 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n"); 2123 2124 // We add the last line's penalty here, since that line is going to be split 2125 // now. 2126 Penalty += Style.PenaltyExcessCharacter * 2127 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2128 2129 if (!DryRun) 2130 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine, 2131 Whitespaces); 2132 ContentStartColumn = 2133 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true); 2134 RemainingTokenColumns = Token->getRemainingLength( 2135 Token->getLineCount() - 1, 2136 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second, 2137 ContentStartColumn); 2138 } 2139 2140 State.Column = ContentStartColumn + RemainingTokenColumns - 2141 Current.UnbreakableTailLength; 2142 2143 if (BreakInserted) { 2144 // If we break the token inside a parameter list, we need to break before 2145 // the next parameter on all levels, so that the next parameter is clearly 2146 // visible. Line comments already introduce a break. 2147 if (Current.isNot(TT_LineComment)) { 2148 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 2149 State.Stack[i].BreakBeforeParameter = true; 2150 } 2151 2152 if (Current.is(TT_BlockComment)) 2153 State.NoContinuation = true; 2154 2155 State.Stack.back().LastSpace = StartColumn; 2156 } 2157 2158 Token->updateNextToken(State); 2159 2160 return {Penalty, Exceeded}; 2161 } 2162 2163 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const { 2164 // In preprocessor directives reserve two chars for trailing " \" 2165 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0); 2166 } 2167 2168 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) { 2169 const FormatToken &Current = *State.NextToken; 2170 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral)) 2171 return false; 2172 // We never consider raw string literals "multiline" for the purpose of 2173 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased 2174 // (see TokenAnnotator::mustBreakBefore(). 2175 if (Current.TokenText.startswith("R\"")) 2176 return false; 2177 if (Current.IsMultiline) 2178 return true; 2179 if (Current.getNextNonComment() && 2180 Current.getNextNonComment()->isStringLiteral()) 2181 return true; // Implicit concatenation. 2182 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals && 2183 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength > 2184 Style.ColumnLimit) 2185 return true; // String will be split. 2186 return false; 2187 } 2188 2189 } // namespace format 2190 } // namespace clang 2191