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