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