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