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