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 "FormatToken.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 true if a TT_SelectorName should be indented when wrapped, 30 // false otherwise. 31 static bool shouldIndentWrappedSelectorName(const FormatStyle &Style, 32 LineType LineType) { 33 return Style.IndentWrappedFunctionNames || LineType == LT_ObjCMethodDecl; 34 } 35 36 // Returns the length of everything up to the first possible line break after 37 // the ), ], } or > matching \c Tok. 38 static unsigned getLengthToMatchingParen(const FormatToken &Tok, 39 const std::vector<ParenState> &Stack) { 40 // Normally whether or not a break before T is possible is calculated and 41 // stored in T.CanBreakBefore. Braces, array initializers and text proto 42 // messages like `key: < ... >` are an exception: a break is possible 43 // before a closing brace R if a break was inserted after the corresponding 44 // opening brace. The information about whether or not a break is needed 45 // before a closing brace R is stored in the ParenState field 46 // S.BreakBeforeClosingBrace where S is the state that R closes. 47 // 48 // In order to decide whether there can be a break before encountered right 49 // braces, this implementation iterates over the sequence of tokens and over 50 // the paren stack in lockstep, keeping track of the stack level which visited 51 // right braces correspond to in MatchingStackIndex. 52 // 53 // For example, consider: 54 // L. <- line number 55 // 1. { 56 // 2. {1}, 57 // 3. {2}, 58 // 4. {{3}}} 59 // ^ where we call this method with this token. 60 // The paren stack at this point contains 3 brace levels: 61 // 0. { at line 1, BreakBeforeClosingBrace: true 62 // 1. first { at line 4, BreakBeforeClosingBrace: false 63 // 2. second { at line 4, BreakBeforeClosingBrace: false, 64 // where there might be fake parens levels in-between these levels. 65 // The algorithm will start at the first } on line 4, which is the matching 66 // brace of the initial left brace and at level 2 of the stack. Then, 67 // examining BreakBeforeClosingBrace: false at level 2, it will continue to 68 // the second } on line 4, and will traverse the stack downwards until it 69 // finds the matching { on level 1. Then, examining BreakBeforeClosingBrace: 70 // false at level 1, it will continue to the third } on line 4 and will 71 // traverse the stack downwards until it finds the matching { on level 0. 72 // Then, examining BreakBeforeClosingBrace: true at level 0, the algorithm 73 // will stop and will use the second } on line 4 to determine the length to 74 // return, as in this example the range will include the tokens: {3}} 75 // 76 // The algorithm will only traverse the stack if it encounters braces, array 77 // initializer squares or text proto angle brackets. 78 if (!Tok.MatchingParen) 79 return 0; 80 FormatToken *End = Tok.MatchingParen; 81 // Maintains a stack level corresponding to the current End token. 82 int MatchingStackIndex = Stack.size() - 1; 83 // Traverses the stack downwards, looking for the level to which LBrace 84 // corresponds. Returns either a pointer to the matching level or nullptr if 85 // LParen is not found in the initial portion of the stack up to 86 // MatchingStackIndex. 87 auto FindParenState = [&](const FormatToken *LBrace) -> const ParenState * { 88 while (MatchingStackIndex >= 0 && Stack[MatchingStackIndex].Tok != LBrace) 89 --MatchingStackIndex; 90 return MatchingStackIndex >= 0 ? &Stack[MatchingStackIndex] : nullptr; 91 }; 92 for (; End->Next; End = End->Next) { 93 if (End->Next->CanBreakBefore) 94 break; 95 if (!End->Next->closesScope()) 96 continue; 97 if (End->Next->MatchingParen && 98 End->Next->MatchingParen->isOneOf( 99 tok::l_brace, TT_ArrayInitializerLSquare, tok::less)) { 100 const ParenState *State = FindParenState(End->Next->MatchingParen); 101 if (State && State->BreakBeforeClosingBrace) 102 break; 103 } 104 } 105 return End->TotalLength - Tok.TotalLength + 1; 106 } 107 108 static unsigned getLengthToNextOperator(const FormatToken &Tok) { 109 if (!Tok.NextOperator) 110 return 0; 111 return Tok.NextOperator->TotalLength - Tok.TotalLength; 112 } 113 114 // Returns \c true if \c Tok is the "." or "->" of a call and starts the next 115 // segment of a builder type call. 116 static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) { 117 return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope(); 118 } 119 120 // Returns \c true if \c Current starts a new parameter. 121 static bool startsNextParameter(const FormatToken &Current, 122 const FormatStyle &Style) { 123 const FormatToken &Previous = *Current.Previous; 124 if (Current.is(TT_CtorInitializerComma) && 125 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) 126 return true; 127 if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName)) 128 return true; 129 return Previous.is(tok::comma) && !Current.isTrailingComment() && 130 ((Previous.isNot(TT_CtorInitializerComma) || 131 Style.BreakConstructorInitializers != 132 FormatStyle::BCIS_BeforeComma) && 133 (Previous.isNot(TT_InheritanceComma) || 134 Style.BreakInheritanceList != FormatStyle::BILS_BeforeComma)); 135 } 136 137 static bool opensProtoMessageField(const FormatToken &LessTok, 138 const FormatStyle &Style) { 139 if (LessTok.isNot(tok::less)) 140 return false; 141 return Style.Language == FormatStyle::LK_TextProto || 142 (Style.Language == FormatStyle::LK_Proto && 143 (LessTok.NestingLevel > 0 || 144 (LessTok.Previous && LessTok.Previous->is(tok::equal)))); 145 } 146 147 // Returns the delimiter of a raw string literal, or None if TokenText is not 148 // the text of a raw string literal. The delimiter could be the empty string. 149 // For example, the delimiter of R"deli(cont)deli" is deli. 150 static llvm::Optional<StringRef> getRawStringDelimiter(StringRef TokenText) { 151 if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'. 152 || !TokenText.startswith("R\"") || !TokenText.endswith("\"")) 153 return None; 154 155 // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has 156 // size at most 16 by the standard, so the first '(' must be among the first 157 // 19 bytes. 158 size_t LParenPos = TokenText.substr(0, 19).find_first_of('('); 159 if (LParenPos == StringRef::npos) 160 return None; 161 StringRef Delimiter = TokenText.substr(2, LParenPos - 2); 162 163 // Check that the string ends in ')Delimiter"'. 164 size_t RParenPos = TokenText.size() - Delimiter.size() - 2; 165 if (TokenText[RParenPos] != ')') 166 return None; 167 if (!TokenText.substr(RParenPos + 1).startswith(Delimiter)) 168 return None; 169 return Delimiter; 170 } 171 172 // Returns the canonical delimiter for \p Language, or the empty string if no 173 // canonical delimiter is specified. 174 static StringRef 175 getCanonicalRawStringDelimiter(const FormatStyle &Style, 176 FormatStyle::LanguageKind Language) { 177 for (const auto &Format : Style.RawStringFormats) { 178 if (Format.Language == Language) 179 return StringRef(Format.CanonicalDelimiter); 180 } 181 return ""; 182 } 183 184 RawStringFormatStyleManager::RawStringFormatStyleManager( 185 const FormatStyle &CodeStyle) { 186 for (const auto &RawStringFormat : CodeStyle.RawStringFormats) { 187 llvm::Optional<FormatStyle> LanguageStyle = 188 CodeStyle.GetLanguageStyle(RawStringFormat.Language); 189 if (!LanguageStyle) { 190 FormatStyle PredefinedStyle; 191 if (!getPredefinedStyle(RawStringFormat.BasedOnStyle, 192 RawStringFormat.Language, &PredefinedStyle)) { 193 PredefinedStyle = getLLVMStyle(); 194 PredefinedStyle.Language = RawStringFormat.Language; 195 } 196 LanguageStyle = PredefinedStyle; 197 } 198 LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit; 199 for (StringRef Delimiter : RawStringFormat.Delimiters) { 200 DelimiterStyle.insert({Delimiter, *LanguageStyle}); 201 } 202 for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions) { 203 EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle}); 204 } 205 } 206 } 207 208 llvm::Optional<FormatStyle> 209 RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const { 210 auto It = DelimiterStyle.find(Delimiter); 211 if (It == DelimiterStyle.end()) 212 return None; 213 return It->second; 214 } 215 216 llvm::Optional<FormatStyle> 217 RawStringFormatStyleManager::getEnclosingFunctionStyle( 218 StringRef EnclosingFunction) const { 219 auto It = EnclosingFunctionStyle.find(EnclosingFunction); 220 if (It == EnclosingFunctionStyle.end()) 221 return None; 222 return It->second; 223 } 224 225 ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style, 226 const AdditionalKeywords &Keywords, 227 const SourceManager &SourceMgr, 228 WhitespaceManager &Whitespaces, 229 encoding::Encoding Encoding, 230 bool BinPackInconclusiveFunctions) 231 : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr), 232 Whitespaces(Whitespaces), Encoding(Encoding), 233 BinPackInconclusiveFunctions(BinPackInconclusiveFunctions), 234 CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {} 235 236 LineState ContinuationIndenter::getInitialState(unsigned FirstIndent, 237 unsigned FirstStartColumn, 238 const AnnotatedLine *Line, 239 bool DryRun) { 240 LineState State; 241 State.FirstIndent = FirstIndent; 242 if (FirstStartColumn && Line->First->NewlinesBefore == 0) 243 State.Column = FirstStartColumn; 244 else 245 State.Column = FirstIndent; 246 // With preprocessor directive indentation, the line starts on column 0 247 // since it's indented after the hash, but FirstIndent is set to the 248 // preprocessor indent. 249 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 250 (Line->Type == LT_PreprocessorDirective || 251 Line->Type == LT_ImportStatement)) 252 State.Column = 0; 253 State.Line = Line; 254 State.NextToken = Line->First; 255 State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent, 256 /*AvoidBinPacking=*/false, 257 /*NoLineBreak=*/false)); 258 State.LineContainsContinuedForLoopSection = false; 259 State.NoContinuation = false; 260 State.StartOfStringLiteral = 0; 261 State.StartOfLineLevel = 0; 262 State.LowestLevelOnLine = 0; 263 State.IgnoreStackForComparison = false; 264 265 if (Style.Language == FormatStyle::LK_TextProto) { 266 // We need this in order to deal with the bin packing of text fields at 267 // global scope. 268 State.Stack.back().AvoidBinPacking = true; 269 State.Stack.back().BreakBeforeParameter = true; 270 State.Stack.back().AlignColons = false; 271 } 272 273 // The first token has already been indented and thus consumed. 274 moveStateToNextToken(State, DryRun, /*Newline=*/false); 275 return State; 276 } 277 278 bool ContinuationIndenter::canBreak(const LineState &State) { 279 const FormatToken &Current = *State.NextToken; 280 const FormatToken &Previous = *Current.Previous; 281 assert(&Previous == Current.Previous); 282 if (!Current.CanBreakBefore && !(State.Stack.back().BreakBeforeClosingBrace && 283 Current.closesBlockOrBlockTypeList(Style))) 284 return false; 285 // The opening "{" of a braced list has to be on the same line as the first 286 // element if it is nested in another braced init list or function call. 287 if (!Current.MustBreakBefore && Previous.is(tok::l_brace) && 288 Previous.isNot(TT_DictLiteral) && Previous.is(BK_BracedInit) && 289 Previous.Previous && 290 Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma)) 291 return false; 292 // This prevents breaks like: 293 // ... 294 // SomeParameter, OtherParameter).DoSomething( 295 // ... 296 // As they hide "DoSomething" and are generally bad for readability. 297 if (Previous.opensScope() && Previous.isNot(tok::l_brace) && 298 State.LowestLevelOnLine < State.StartOfLineLevel && 299 State.LowestLevelOnLine < Current.NestingLevel) 300 return false; 301 if (Current.isMemberAccess() && State.Stack.back().ContainsUnwrappedBuilder) 302 return false; 303 304 // Don't create a 'hanging' indent if there are multiple blocks in a single 305 // statement. 306 if (Previous.is(tok::l_brace) && State.Stack.size() > 1 && 307 State.Stack[State.Stack.size() - 2].NestedBlockInlined && 308 State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks) 309 return false; 310 311 // Don't break after very short return types (e.g. "void") as that is often 312 // unexpected. 313 if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) { 314 if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None) 315 return false; 316 } 317 318 // If binary operators are moved to the next line (including commas for some 319 // styles of constructor initializers), that's always ok. 320 if (!Current.isOneOf(TT_BinaryOperator, tok::comma) && 321 State.Stack.back().NoLineBreakInOperand) 322 return false; 323 324 if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr)) 325 return false; 326 327 return !State.Stack.back().NoLineBreak; 328 } 329 330 bool ContinuationIndenter::mustBreak(const LineState &State) { 331 const FormatToken &Current = *State.NextToken; 332 const FormatToken &Previous = *Current.Previous; 333 if (Style.BraceWrapping.BeforeLambdaBody && Current.CanBreakBefore && 334 Current.is(TT_LambdaLBrace) && Previous.isNot(TT_LineComment)) { 335 auto LambdaBodyLength = getLengthToMatchingParen(Current, State.Stack); 336 return (LambdaBodyLength > getColumnLimit(State)); 337 } 338 if (Current.MustBreakBefore || Current.is(TT_InlineASMColon)) 339 return true; 340 if (State.Stack.back().BreakBeforeClosingBrace && 341 Current.closesBlockOrBlockTypeList(Style)) 342 return true; 343 if (Previous.is(tok::semi) && State.LineContainsContinuedForLoopSection) 344 return true; 345 if (Style.Language == FormatStyle::LK_ObjC && 346 Style.ObjCBreakBeforeNestedBlockParam && 347 Current.ObjCSelectorNameParts > 1 && 348 Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) { 349 return true; 350 } 351 // Avoid producing inconsistent states by requiring breaks where they are not 352 // permitted for C# generic type constraints. 353 if (State.Stack.back().IsCSharpGenericTypeConstraint && 354 Previous.isNot(TT_CSharpGenericTypeConstraintComma)) 355 return false; 356 if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) || 357 (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) && 358 Style.isCpp() && 359 // FIXME: This is a temporary workaround for the case where clang-format 360 // sets BreakBeforeParameter to avoid bin packing and this creates a 361 // completely unnecessary line break after a template type that isn't 362 // line-wrapped. 363 (Previous.NestingLevel == 1 || Style.BinPackParameters)) || 364 (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) && 365 Previous.isNot(tok::question)) || 366 (!Style.BreakBeforeTernaryOperators && 367 Previous.is(TT_ConditionalExpr))) && 368 State.Stack.back().BreakBeforeParameter && !Current.isTrailingComment() && 369 !Current.isOneOf(tok::r_paren, tok::r_brace)) 370 return true; 371 if (State.Stack.back().IsChainedConditional && 372 ((Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) && 373 Current.is(tok::colon)) || 374 (!Style.BreakBeforeTernaryOperators && Previous.is(TT_ConditionalExpr) && 375 Previous.is(tok::colon)))) 376 return true; 377 if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) || 378 (Previous.is(TT_ArrayInitializerLSquare) && 379 Previous.ParameterCount > 1) || 380 opensProtoMessageField(Previous, Style)) && 381 Style.ColumnLimit > 0 && 382 getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 > 383 getColumnLimit(State)) 384 return true; 385 386 const FormatToken &BreakConstructorInitializersToken = 387 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon 388 ? Previous 389 : Current; 390 if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) && 391 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength > 392 getColumnLimit(State) || 393 State.Stack.back().BreakBeforeParameter) && 394 (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All || 395 Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon || 396 Style.ColumnLimit != 0)) 397 return true; 398 399 if (Current.is(TT_ObjCMethodExpr) && !Previous.is(TT_SelectorName) && 400 State.Line->startsWith(TT_ObjCMethodSpecifier)) 401 return true; 402 if (Current.is(TT_SelectorName) && !Previous.is(tok::at) && 403 State.Stack.back().ObjCSelectorNameFound && 404 State.Stack.back().BreakBeforeParameter && 405 (Style.ObjCBreakBeforeNestedBlockParam || 406 !Current.startsSequence(TT_SelectorName, tok::colon, tok::caret))) 407 return true; 408 409 unsigned NewLineColumn = getNewLineColumn(State); 410 if (Current.isMemberAccess() && Style.ColumnLimit != 0 && 411 State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit && 412 (State.Column > NewLineColumn || 413 Current.NestingLevel < State.StartOfLineLevel)) 414 return true; 415 416 if (startsSegmentOfBuilderTypeCall(Current) && 417 (State.Stack.back().CallContinuation != 0 || 418 State.Stack.back().BreakBeforeParameter) && 419 // JavaScript is treated different here as there is a frequent pattern: 420 // SomeFunction(function() { 421 // ... 422 // }.bind(...)); 423 // FIXME: We should find a more generic solution to this problem. 424 !(State.Column <= NewLineColumn && 425 Style.Language == FormatStyle::LK_JavaScript) && 426 !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) 427 return true; 428 429 // If the template declaration spans multiple lines, force wrap before the 430 // function/class declaration 431 if (Previous.ClosesTemplateDeclaration && 432 State.Stack.back().BreakBeforeParameter && Current.CanBreakBefore) 433 return true; 434 435 if (!State.Line->First->is(tok::kw_enum) && State.Column <= NewLineColumn) 436 return false; 437 438 if (Style.AlwaysBreakBeforeMultilineStrings && 439 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth || 440 Previous.is(tok::comma) || Current.NestingLevel < 2) && 441 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at, 442 Keywords.kw_dollar) && 443 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) && 444 nextIsMultilineString(State)) 445 return true; 446 447 // Using CanBreakBefore here and below takes care of the decision whether the 448 // current style uses wrapping before or after operators for the given 449 // operator. 450 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) { 451 // If we need to break somewhere inside the LHS of a binary expression, we 452 // should also break after the operator. Otherwise, the formatting would 453 // hide the operator precedence, e.g. in: 454 // if (aaaaaaaaaaaaaa == 455 // bbbbbbbbbbbbbb && c) {.. 456 // For comparisons, we only apply this rule, if the LHS is a binary 457 // expression itself as otherwise, the line breaks seem superfluous. 458 // We need special cases for ">>" which we have split into two ">" while 459 // lexing in order to make template parsing easier. 460 bool IsComparison = (Previous.getPrecedence() == prec::Relational || 461 Previous.getPrecedence() == prec::Equality || 462 Previous.getPrecedence() == prec::Spaceship) && 463 Previous.Previous && 464 Previous.Previous->isNot(TT_BinaryOperator); // For >>. 465 bool LHSIsBinaryExpr = 466 Previous.Previous && Previous.Previous->EndsBinaryExpression; 467 if ((!IsComparison || LHSIsBinaryExpr) && !Current.isTrailingComment() && 468 Previous.getPrecedence() != prec::Assignment && 469 State.Stack.back().BreakBeforeParameter) 470 return true; 471 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore && 472 State.Stack.back().BreakBeforeParameter) { 473 return true; 474 } 475 476 // Same as above, but for the first "<<" operator. 477 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) && 478 State.Stack.back().BreakBeforeParameter && 479 State.Stack.back().FirstLessLess == 0) 480 return true; 481 482 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) { 483 // Always break after "template <...>" and leading annotations. This is only 484 // for cases where the entire line does not fit on a single line as a 485 // different LineFormatter would be used otherwise. 486 if (Previous.ClosesTemplateDeclaration) 487 return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No; 488 if (Previous.is(TT_FunctionAnnotationRParen)) 489 return true; 490 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) && 491 Current.isNot(TT_LeadingJavaAnnotation)) 492 return true; 493 } 494 495 // Break after the closing parenthesis of TypeScript decorators. 496 if (Style.Language == FormatStyle::LK_JavaScript && 497 Previous.is(tok::r_paren) && Previous.is(TT_JavaAnnotation)) { 498 return true; 499 } 500 501 // If the return type spans multiple lines, wrap before the function name. 502 if (((Current.is(TT_FunctionDeclarationName) && 503 // Don't break before a C# function when no break after return type 504 (!Style.isCSharp() || 505 Style.AlwaysBreakAfterReturnType != FormatStyle::RTBS_None) && 506 // Don't always break between a JavaScript `function` and the function 507 // name. 508 Style.Language != FormatStyle::LK_JavaScript) || 509 (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) && 510 !Previous.is(tok::kw_template) && State.Stack.back().BreakBeforeParameter) 511 return true; 512 513 // The following could be precomputed as they do not depend on the state. 514 // However, as they should take effect only if the UnwrappedLine does not fit 515 // into the ColumnLimit, they are checked here in the ContinuationIndenter. 516 if (Style.ColumnLimit != 0 && Previous.is(BK_Block) && 517 Previous.is(tok::l_brace) && !Current.isOneOf(tok::r_brace, tok::comment)) 518 return true; 519 520 if (Current.is(tok::lessless) && 521 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") || 522 (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") || 523 Previous.TokenText == "\'\\n\'")))) 524 return true; 525 526 if (Previous.is(TT_BlockComment) && Previous.IsMultiline) 527 return true; 528 529 if (State.NoContinuation) 530 return true; 531 532 return false; 533 } 534 535 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline, 536 bool DryRun, 537 unsigned ExtraSpaces) { 538 const FormatToken &Current = *State.NextToken; 539 540 assert(!State.Stack.empty()); 541 State.NoContinuation = false; 542 543 if ((Current.is(TT_ImplicitStringLiteral) && 544 (Current.Previous->Tok.getIdentifierInfo() == nullptr || 545 Current.Previous->Tok.getIdentifierInfo()->getPPKeywordID() == 546 tok::pp_not_keyword))) { 547 unsigned EndColumn = 548 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd()); 549 if (Current.LastNewlineOffset != 0) { 550 // If there is a newline within this token, the final column will solely 551 // determined by the current end column. 552 State.Column = EndColumn; 553 } else { 554 unsigned StartColumn = 555 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin()); 556 assert(EndColumn >= StartColumn); 557 State.Column += EndColumn - StartColumn; 558 } 559 moveStateToNextToken(State, DryRun, /*Newline=*/false); 560 return 0; 561 } 562 563 unsigned Penalty = 0; 564 if (Newline) 565 Penalty = addTokenOnNewLine(State, DryRun); 566 else 567 addTokenOnCurrentLine(State, DryRun, ExtraSpaces); 568 569 return moveStateToNextToken(State, DryRun, Newline) + Penalty; 570 } 571 572 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun, 573 unsigned ExtraSpaces) { 574 FormatToken &Current = *State.NextToken; 575 const FormatToken &Previous = *State.NextToken->Previous; 576 if (Current.is(tok::equal) && 577 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) && 578 State.Stack.back().VariablePos == 0) { 579 State.Stack.back().VariablePos = State.Column; 580 // Move over * and & if they are bound to the variable name. 581 const FormatToken *Tok = &Previous; 582 while (Tok && State.Stack.back().VariablePos >= Tok->ColumnWidth) { 583 State.Stack.back().VariablePos -= Tok->ColumnWidth; 584 if (Tok->SpacesRequiredBefore != 0) 585 break; 586 Tok = Tok->Previous; 587 } 588 if (Previous.PartOfMultiVariableDeclStmt) 589 State.Stack.back().LastSpace = State.Stack.back().VariablePos; 590 } 591 592 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces; 593 594 // Indent preprocessor directives after the hash if required. 595 int PPColumnCorrection = 0; 596 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 597 Previous.is(tok::hash) && State.FirstIndent > 0 && 598 (State.Line->Type == LT_PreprocessorDirective || 599 State.Line->Type == LT_ImportStatement)) { 600 Spaces += State.FirstIndent; 601 602 // For preprocessor indent with tabs, State.Column will be 1 because of the 603 // hash. This causes second-level indents onward to have an extra space 604 // after the tabs. We avoid this misalignment by subtracting 1 from the 605 // column value passed to replaceWhitespace(). 606 if (Style.UseTab != FormatStyle::UT_Never) 607 PPColumnCorrection = -1; 608 } 609 610 if (!DryRun) 611 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces, 612 State.Column + Spaces + PPColumnCorrection); 613 614 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance 615 // declaration unless there is multiple inheritance. 616 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 617 Current.is(TT_InheritanceColon)) 618 State.Stack.back().NoLineBreak = true; 619 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon && 620 Previous.is(TT_InheritanceColon)) 621 State.Stack.back().NoLineBreak = true; 622 623 if (Current.is(TT_SelectorName) && 624 !State.Stack.back().ObjCSelectorNameFound) { 625 unsigned MinIndent = 626 std::max(State.FirstIndent + Style.ContinuationIndentWidth, 627 State.Stack.back().Indent); 628 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth; 629 if (Current.LongestObjCSelectorName == 0) 630 State.Stack.back().AlignColons = false; 631 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos) 632 State.Stack.back().ColonPos = MinIndent + Current.LongestObjCSelectorName; 633 else 634 State.Stack.back().ColonPos = FirstColonPos; 635 } 636 637 // In "AlwaysBreak" mode, enforce wrapping directly after the parenthesis by 638 // disallowing any further line breaks if there is no line break after the 639 // opening parenthesis. Don't break if it doesn't conserve columns. 640 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak && 641 (Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) || 642 (Previous.is(tok::l_brace) && Previous.isNot(BK_Block) && 643 Style.Cpp11BracedListStyle)) && 644 State.Column > getNewLineColumn(State) && 645 (!Previous.Previous || !Previous.Previous->isOneOf( 646 tok::kw_for, tok::kw_while, tok::kw_switch)) && 647 // Don't do this for simple (no expressions) one-argument function calls 648 // as that feels like needlessly wasting whitespace, e.g.: 649 // 650 // caaaaaaaaaaaall( 651 // caaaaaaaaaaaall( 652 // caaaaaaaaaaaall( 653 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa)))); 654 Current.FakeLParens.size() > 0 && 655 Current.FakeLParens.back() > prec::Unknown) 656 State.Stack.back().NoLineBreak = true; 657 if (Previous.is(TT_TemplateString) && Previous.opensScope()) 658 State.Stack.back().NoLineBreak = true; 659 660 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign && 661 !State.Stack.back().IsCSharpGenericTypeConstraint && 662 Previous.opensScope() && Previous.isNot(TT_ObjCMethodExpr) && 663 (Current.isNot(TT_LineComment) || Previous.is(BK_BracedInit))) { 664 State.Stack.back().Indent = State.Column + Spaces; 665 State.Stack.back().IsAligned = true; 666 } 667 if (State.Stack.back().AvoidBinPacking && startsNextParameter(Current, Style)) 668 State.Stack.back().NoLineBreak = true; 669 if (startsSegmentOfBuilderTypeCall(Current) && 670 State.Column > getNewLineColumn(State)) 671 State.Stack.back().ContainsUnwrappedBuilder = true; 672 673 if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java) 674 State.Stack.back().NoLineBreak = true; 675 if (Current.isMemberAccess() && Previous.is(tok::r_paren) && 676 (Previous.MatchingParen && 677 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) 678 // If there is a function call with long parameters, break before trailing 679 // calls. This prevents things like: 680 // EXPECT_CALL(SomeLongParameter).Times( 681 // 2); 682 // We don't want to do this for short parameters as they can just be 683 // indexes. 684 State.Stack.back().NoLineBreak = true; 685 686 // Don't allow the RHS of an operator to be split over multiple lines unless 687 // there is a line-break right after the operator. 688 // Exclude relational operators, as there, it is always more desirable to 689 // have the LHS 'left' of the RHS. 690 const FormatToken *P = Current.getPreviousNonComment(); 691 if (!Current.is(tok::comment) && P && 692 (P->isOneOf(TT_BinaryOperator, tok::comma) || 693 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) && 694 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) && 695 P->getPrecedence() != prec::Assignment && 696 P->getPrecedence() != prec::Relational && 697 P->getPrecedence() != prec::Spaceship) { 698 bool BreakBeforeOperator = 699 P->MustBreakBefore || P->is(tok::lessless) || 700 (P->is(TT_BinaryOperator) && 701 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) || 702 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators); 703 // Don't do this if there are only two operands. In these cases, there is 704 // always a nice vertical separation between them and the extra line break 705 // does not help. 706 bool HasTwoOperands = 707 P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr); 708 if ((!BreakBeforeOperator && 709 !(HasTwoOperands && 710 Style.AlignOperands != FormatStyle::OAS_DontAlign)) || 711 (!State.Stack.back().LastOperatorWrapped && BreakBeforeOperator)) 712 State.Stack.back().NoLineBreakInOperand = true; 713 } 714 715 State.Column += Spaces; 716 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) && 717 Previous.Previous && 718 (Previous.Previous->is(tok::kw_for) || Previous.Previous->isIf())) { 719 // Treat the condition inside an if as if it was a second function 720 // parameter, i.e. let nested calls have a continuation indent. 721 State.Stack.back().LastSpace = State.Column; 722 State.Stack.back().NestedBlockIndent = State.Column; 723 } else if (!Current.isOneOf(tok::comment, tok::caret) && 724 ((Previous.is(tok::comma) && 725 !Previous.is(TT_OverloadedOperator)) || 726 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) { 727 State.Stack.back().LastSpace = State.Column; 728 } else if (Previous.is(TT_CtorInitializerColon) && 729 Style.BreakConstructorInitializers == 730 FormatStyle::BCIS_AfterColon) { 731 State.Stack.back().Indent = State.Column; 732 State.Stack.back().LastSpace = State.Column; 733 } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 734 TT_CtorInitializerColon)) && 735 ((Previous.getPrecedence() != prec::Assignment && 736 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 || 737 Previous.NextOperator)) || 738 Current.StartsBinaryExpression)) { 739 // Indent relative to the RHS of the expression unless this is a simple 740 // assignment without binary expression on the RHS. Also indent relative to 741 // unary operators and the colons of constructor initializers. 742 if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None) 743 State.Stack.back().LastSpace = State.Column; 744 } else if (Previous.is(TT_InheritanceColon)) { 745 State.Stack.back().Indent = State.Column; 746 State.Stack.back().LastSpace = State.Column; 747 } else if (Current.is(TT_CSharpGenericTypeConstraintColon)) { 748 State.Stack.back().ColonPos = State.Column; 749 } else if (Previous.opensScope()) { 750 // If a function has a trailing call, indent all parameters from the 751 // opening parenthesis. This avoids confusing indents like: 752 // OuterFunction(InnerFunctionCall( // break 753 // ParameterToInnerFunction)) // break 754 // .SecondInnerFunctionCall(); 755 bool HasTrailingCall = false; 756 if (Previous.MatchingParen) { 757 const FormatToken *Next = Previous.MatchingParen->getNextNonComment(); 758 HasTrailingCall = Next && Next->isMemberAccess(); 759 } 760 if (HasTrailingCall && State.Stack.size() > 1 && 761 State.Stack[State.Stack.size() - 2].CallContinuation == 0) 762 State.Stack.back().LastSpace = State.Column; 763 } 764 } 765 766 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State, 767 bool DryRun) { 768 FormatToken &Current = *State.NextToken; 769 const FormatToken &Previous = *State.NextToken->Previous; 770 771 // Extra penalty that needs to be added because of the way certain line 772 // breaks are chosen. 773 unsigned Penalty = 0; 774 775 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 776 const FormatToken *NextNonComment = Previous.getNextNonComment(); 777 if (!NextNonComment) 778 NextNonComment = &Current; 779 // The first line break on any NestingLevel causes an extra penalty in order 780 // prefer similar line breaks. 781 if (!State.Stack.back().ContainsLineBreak) 782 Penalty += 15; 783 State.Stack.back().ContainsLineBreak = true; 784 785 Penalty += State.NextToken->SplitPenalty; 786 787 // Breaking before the first "<<" is generally not desirable if the LHS is 788 // short. Also always add the penalty if the LHS is split over multiple lines 789 // to avoid unnecessary line breaks that just work around this penalty. 790 if (NextNonComment->is(tok::lessless) && 791 State.Stack.back().FirstLessLess == 0 && 792 (State.Column <= Style.ColumnLimit / 3 || 793 State.Stack.back().BreakBeforeParameter)) 794 Penalty += Style.PenaltyBreakFirstLessLess; 795 796 State.Column = getNewLineColumn(State); 797 798 // Add Penalty proportional to amount of whitespace away from FirstColumn 799 // This tends to penalize several lines that are far-right indented, 800 // and prefers a line-break prior to such a block, e.g: 801 // 802 // Constructor() : 803 // member(value), looooooooooooooooong_member( 804 // looooooooooong_call(param_1, param_2, param_3)) 805 // would then become 806 // Constructor() : 807 // member(value), 808 // looooooooooooooooong_member( 809 // looooooooooong_call(param_1, param_2, param_3)) 810 if (State.Column > State.FirstIndent) 811 Penalty += 812 Style.PenaltyIndentedWhitespace * (State.Column - State.FirstIndent); 813 814 // Indent nested blocks relative to this column, unless in a very specific 815 // JavaScript special case where: 816 // 817 // var loooooong_name = 818 // function() { 819 // // code 820 // } 821 // 822 // is common and should be formatted like a free-standing function. The same 823 // goes for wrapping before the lambda return type arrow. 824 if (!Current.is(TT_LambdaArrow) && 825 (Style.Language != FormatStyle::LK_JavaScript || 826 Current.NestingLevel != 0 || !PreviousNonComment || 827 !PreviousNonComment->is(tok::equal) || 828 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) 829 State.Stack.back().NestedBlockIndent = State.Column; 830 831 if (NextNonComment->isMemberAccess()) { 832 if (State.Stack.back().CallContinuation == 0) 833 State.Stack.back().CallContinuation = State.Column; 834 } else if (NextNonComment->is(TT_SelectorName)) { 835 if (!State.Stack.back().ObjCSelectorNameFound) { 836 if (NextNonComment->LongestObjCSelectorName == 0) { 837 State.Stack.back().AlignColons = false; 838 } else { 839 State.Stack.back().ColonPos = 840 (shouldIndentWrappedSelectorName(Style, State.Line->Type) 841 ? std::max(State.Stack.back().Indent, 842 State.FirstIndent + Style.ContinuationIndentWidth) 843 : State.Stack.back().Indent) + 844 std::max(NextNonComment->LongestObjCSelectorName, 845 NextNonComment->ColumnWidth); 846 } 847 } else if (State.Stack.back().AlignColons && 848 State.Stack.back().ColonPos <= NextNonComment->ColumnWidth) { 849 State.Stack.back().ColonPos = State.Column + NextNonComment->ColumnWidth; 850 } 851 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 852 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) { 853 // FIXME: This is hacky, find a better way. The problem is that in an ObjC 854 // method expression, the block should be aligned to the line starting it, 855 // e.g.: 856 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason 857 // ^(int *i) { 858 // // ... 859 // }]; 860 // Thus, we set LastSpace of the next higher NestingLevel, to which we move 861 // when we consume all of the "}"'s FakeRParens at the "{". 862 if (State.Stack.size() > 1) 863 State.Stack[State.Stack.size() - 2].LastSpace = 864 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 865 Style.ContinuationIndentWidth; 866 } 867 868 if ((PreviousNonComment && 869 PreviousNonComment->isOneOf(tok::comma, tok::semi) && 870 !State.Stack.back().AvoidBinPacking) || 871 Previous.is(TT_BinaryOperator)) 872 State.Stack.back().BreakBeforeParameter = false; 873 if (PreviousNonComment && 874 PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) && 875 Current.NestingLevel == 0) 876 State.Stack.back().BreakBeforeParameter = false; 877 if (NextNonComment->is(tok::question) || 878 (PreviousNonComment && PreviousNonComment->is(tok::question))) 879 State.Stack.back().BreakBeforeParameter = true; 880 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore) 881 State.Stack.back().BreakBeforeParameter = false; 882 883 if (!DryRun) { 884 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1; 885 if (Current.is(tok::r_brace) && Current.MatchingParen && 886 // Only strip trailing empty lines for l_braces that have children, i.e. 887 // for function expressions (lambdas, arrows, etc). 888 !Current.MatchingParen->Children.empty()) { 889 // lambdas and arrow functions are expressions, thus their r_brace is not 890 // on its own line, and thus not covered by UnwrappedLineFormatter's logic 891 // about removing empty lines on closing blocks. Special case them here. 892 MaxEmptyLinesToKeep = 1; 893 } 894 unsigned Newlines = 895 std::max(1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep)); 896 bool ContinuePPDirective = 897 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement; 898 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column, 899 State.Stack.back().IsAligned, 900 ContinuePPDirective); 901 } 902 903 if (!Current.isTrailingComment()) 904 State.Stack.back().LastSpace = State.Column; 905 if (Current.is(tok::lessless)) 906 // If we are breaking before a "<<", we always want to indent relative to 907 // RHS. This is necessary only for "<<", as we special-case it and don't 908 // always indent relative to the RHS. 909 State.Stack.back().LastSpace += 3; // 3 -> width of "<< ". 910 911 State.StartOfLineLevel = Current.NestingLevel; 912 State.LowestLevelOnLine = Current.NestingLevel; 913 914 // Any break on this level means that the parent level has been broken 915 // and we need to avoid bin packing there. 916 bool NestedBlockSpecialCase = 917 (!Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 && 918 State.Stack[State.Stack.size() - 2].NestedBlockInlined) || 919 (Style.Language == FormatStyle::LK_ObjC && Current.is(tok::r_brace) && 920 State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam); 921 if (!NestedBlockSpecialCase) 922 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 923 State.Stack[i].BreakBeforeParameter = true; 924 925 if (PreviousNonComment && 926 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) && 927 (PreviousNonComment->isNot(TT_TemplateCloser) || 928 Current.NestingLevel != 0) && 929 !PreviousNonComment->isOneOf( 930 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation, 931 TT_LeadingJavaAnnotation) && 932 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope()) 933 State.Stack.back().BreakBeforeParameter = true; 934 935 // If we break after { or the [ of an array initializer, we should also break 936 // before the corresponding } or ]. 937 if (PreviousNonComment && 938 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 939 opensProtoMessageField(*PreviousNonComment, Style))) 940 State.Stack.back().BreakBeforeClosingBrace = true; 941 942 if (State.Stack.back().AvoidBinPacking) { 943 // If we are breaking after '(', '{', '<', or this is the break after a ':' 944 // to start a member initializater list in a constructor, this should not 945 // be considered bin packing unless the relevant AllowAll option is false or 946 // this is a dict/object literal. 947 bool PreviousIsBreakingCtorInitializerColon = 948 Previous.is(TT_CtorInitializerColon) && 949 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 950 if (!(Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) || 951 PreviousIsBreakingCtorInitializerColon) || 952 (!Style.AllowAllParametersOfDeclarationOnNextLine && 953 State.Line->MustBeDeclaration) || 954 (!Style.AllowAllArgumentsOnNextLine && 955 !State.Line->MustBeDeclaration) || 956 (!Style.AllowAllConstructorInitializersOnNextLine && 957 PreviousIsBreakingCtorInitializerColon) || 958 Previous.is(TT_DictLiteral)) 959 State.Stack.back().BreakBeforeParameter = true; 960 961 // If we are breaking after a ':' to start a member initializer list, 962 // and we allow all arguments on the next line, we should not break 963 // before the next parameter. 964 if (PreviousIsBreakingCtorInitializerColon && 965 Style.AllowAllConstructorInitializersOnNextLine) 966 State.Stack.back().BreakBeforeParameter = false; 967 } 968 969 return Penalty; 970 } 971 972 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) { 973 if (!State.NextToken || !State.NextToken->Previous) 974 return 0; 975 976 FormatToken &Current = *State.NextToken; 977 978 if (State.Stack.back().IsCSharpGenericTypeConstraint && 979 Current.isNot(TT_CSharpGenericTypeConstraint)) 980 return State.Stack.back().ColonPos + 2; 981 982 const FormatToken &Previous = *Current.Previous; 983 // If we are continuing an expression, we want to use the continuation indent. 984 unsigned ContinuationIndent = 985 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 986 Style.ContinuationIndentWidth; 987 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 988 const FormatToken *NextNonComment = Previous.getNextNonComment(); 989 if (!NextNonComment) 990 NextNonComment = &Current; 991 992 // Java specific bits. 993 if (Style.Language == FormatStyle::LK_Java && 994 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) 995 return std::max(State.Stack.back().LastSpace, 996 State.Stack.back().Indent + Style.ContinuationIndentWidth); 997 998 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths && 999 State.Line->First->is(tok::kw_enum)) 1000 return (Style.IndentWidth * State.Line->First->IndentLevel) + 1001 Style.IndentWidth; 1002 1003 if (NextNonComment->is(tok::l_brace) && NextNonComment->is(BK_Block)) 1004 return Current.NestingLevel == 0 ? State.FirstIndent 1005 : State.Stack.back().Indent; 1006 if ((Current.isOneOf(tok::r_brace, tok::r_square) || 1007 (Current.is(tok::greater) && 1008 (Style.Language == FormatStyle::LK_Proto || 1009 Style.Language == FormatStyle::LK_TextProto))) && 1010 State.Stack.size() > 1) { 1011 if (Current.closesBlockOrBlockTypeList(Style)) 1012 return State.Stack[State.Stack.size() - 2].NestedBlockIndent; 1013 if (Current.MatchingParen && Current.MatchingParen->is(BK_BracedInit)) 1014 return State.Stack[State.Stack.size() - 2].LastSpace; 1015 return State.FirstIndent; 1016 } 1017 // Indent a closing parenthesis at the previous level if followed by a semi, 1018 // const, or opening brace. This allows indentations such as: 1019 // foo( 1020 // a, 1021 // ); 1022 // int Foo::getter( 1023 // // 1024 // ) const { 1025 // return foo; 1026 // } 1027 // function foo( 1028 // a, 1029 // ) { 1030 // code(); // 1031 // } 1032 if (Current.is(tok::r_paren) && State.Stack.size() > 1 && 1033 (!Current.Next || 1034 Current.Next->isOneOf(tok::semi, tok::kw_const, tok::l_brace))) 1035 return State.Stack[State.Stack.size() - 2].LastSpace; 1036 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope()) 1037 return State.Stack[State.Stack.size() - 2].LastSpace; 1038 if (Current.is(tok::identifier) && Current.Next && 1039 (Current.Next->is(TT_DictLiteral) || 1040 ((Style.Language == FormatStyle::LK_Proto || 1041 Style.Language == FormatStyle::LK_TextProto) && 1042 Current.Next->isOneOf(tok::less, tok::l_brace)))) 1043 return State.Stack.back().Indent; 1044 if (NextNonComment->is(TT_ObjCStringLiteral) && 1045 State.StartOfStringLiteral != 0) 1046 return State.StartOfStringLiteral - 1; 1047 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0) 1048 return State.StartOfStringLiteral; 1049 if (NextNonComment->is(tok::lessless) && 1050 State.Stack.back().FirstLessLess != 0) 1051 return State.Stack.back().FirstLessLess; 1052 if (NextNonComment->isMemberAccess()) { 1053 if (State.Stack.back().CallContinuation == 0) 1054 return ContinuationIndent; 1055 return State.Stack.back().CallContinuation; 1056 } 1057 if (State.Stack.back().QuestionColumn != 0 && 1058 ((NextNonComment->is(tok::colon) && 1059 NextNonComment->is(TT_ConditionalExpr)) || 1060 Previous.is(TT_ConditionalExpr))) { 1061 if (((NextNonComment->is(tok::colon) && NextNonComment->Next && 1062 !NextNonComment->Next->FakeLParens.empty() && 1063 NextNonComment->Next->FakeLParens.back() == prec::Conditional) || 1064 (Previous.is(tok::colon) && !Current.FakeLParens.empty() && 1065 Current.FakeLParens.back() == prec::Conditional)) && 1066 !State.Stack.back().IsWrappedConditional) { 1067 // NOTE: we may tweak this slightly: 1068 // * not remove the 'lead' ContinuationIndentWidth 1069 // * always un-indent by the operator when 1070 // BreakBeforeTernaryOperators=true 1071 unsigned Indent = State.Stack.back().Indent; 1072 if (Style.AlignOperands != FormatStyle::OAS_DontAlign) { 1073 Indent -= Style.ContinuationIndentWidth; 1074 } 1075 if (Style.BreakBeforeTernaryOperators && 1076 State.Stack.back().UnindentOperator) 1077 Indent -= 2; 1078 return Indent; 1079 } 1080 return State.Stack.back().QuestionColumn; 1081 } 1082 if (Previous.is(tok::comma) && State.Stack.back().VariablePos != 0) 1083 return State.Stack.back().VariablePos; 1084 if ((PreviousNonComment && 1085 (PreviousNonComment->ClosesTemplateDeclaration || 1086 PreviousNonComment->isOneOf( 1087 TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen, 1088 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) || 1089 (!Style.IndentWrappedFunctionNames && 1090 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) 1091 return std::max(State.Stack.back().LastSpace, State.Stack.back().Indent); 1092 if (NextNonComment->is(TT_SelectorName)) { 1093 if (!State.Stack.back().ObjCSelectorNameFound) { 1094 unsigned MinIndent = State.Stack.back().Indent; 1095 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) 1096 MinIndent = std::max(MinIndent, 1097 State.FirstIndent + Style.ContinuationIndentWidth); 1098 // If LongestObjCSelectorName is 0, we are indenting the first 1099 // part of an ObjC selector (or a selector component which is 1100 // not colon-aligned due to block formatting). 1101 // 1102 // Otherwise, we are indenting a subsequent part of an ObjC 1103 // selector which should be colon-aligned to the longest 1104 // component of the ObjC selector. 1105 // 1106 // In either case, we want to respect Style.IndentWrappedFunctionNames. 1107 return MinIndent + 1108 std::max(NextNonComment->LongestObjCSelectorName, 1109 NextNonComment->ColumnWidth) - 1110 NextNonComment->ColumnWidth; 1111 } 1112 if (!State.Stack.back().AlignColons) 1113 return State.Stack.back().Indent; 1114 if (State.Stack.back().ColonPos > NextNonComment->ColumnWidth) 1115 return State.Stack.back().ColonPos - NextNonComment->ColumnWidth; 1116 return State.Stack.back().Indent; 1117 } 1118 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr)) 1119 return State.Stack.back().ColonPos; 1120 if (NextNonComment->is(TT_ArraySubscriptLSquare)) { 1121 if (State.Stack.back().StartOfArraySubscripts != 0) 1122 return State.Stack.back().StartOfArraySubscripts; 1123 else if (Style.isCSharp()) // C# allows `["key"] = value` inside object 1124 // initializers. 1125 return State.Stack.back().Indent; 1126 return ContinuationIndent; 1127 } 1128 1129 // This ensure that we correctly format ObjC methods calls without inputs, 1130 // i.e. where the last element isn't selector like: [callee method]; 1131 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 && 1132 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) 1133 return State.Stack.back().Indent; 1134 1135 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) || 1136 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) 1137 return ContinuationIndent; 1138 if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 1139 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) 1140 return ContinuationIndent; 1141 if (NextNonComment->is(TT_CtorInitializerComma)) 1142 return State.Stack.back().Indent; 1143 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && 1144 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) 1145 return State.Stack.back().Indent; 1146 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) && 1147 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) 1148 return State.Stack.back().Indent; 1149 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon, 1150 TT_InheritanceComma)) 1151 return State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1152 if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() && 1153 !Current.isOneOf(tok::colon, tok::comment)) 1154 return ContinuationIndent; 1155 if (Current.is(TT_ProtoExtensionLSquare)) 1156 return State.Stack.back().Indent; 1157 if (Current.isBinaryOperator() && State.Stack.back().UnindentOperator) 1158 return State.Stack.back().Indent - Current.Tok.getLength() - 1159 Current.SpacesRequiredBefore; 1160 if (Current.isOneOf(tok::comment, TT_BlockComment, TT_LineComment) && 1161 NextNonComment->isBinaryOperator() && State.Stack.back().UnindentOperator) 1162 return State.Stack.back().Indent - NextNonComment->Tok.getLength() - 1163 NextNonComment->SpacesRequiredBefore; 1164 if (State.Stack.back().Indent == State.FirstIndent && PreviousNonComment && 1165 !PreviousNonComment->isOneOf(tok::r_brace, TT_CtorInitializerComma)) 1166 // Ensure that we fall back to the continuation indent width instead of 1167 // just flushing continuations left. 1168 return State.Stack.back().Indent + Style.ContinuationIndentWidth; 1169 return State.Stack.back().Indent; 1170 } 1171 1172 static bool hasNestedBlockInlined(const FormatToken *Previous, 1173 const FormatToken &Current, 1174 const FormatStyle &Style) { 1175 if (Previous->isNot(tok::l_paren)) 1176 return true; 1177 if (Previous->ParameterCount > 1) 1178 return true; 1179 1180 // Also a nested block if contains a lambda inside function with 1 parameter 1181 return (Style.BraceWrapping.BeforeLambdaBody && Current.is(TT_LambdaLSquare)); 1182 } 1183 1184 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State, 1185 bool DryRun, bool Newline) { 1186 assert(State.Stack.size()); 1187 const FormatToken &Current = *State.NextToken; 1188 1189 if (Current.is(TT_CSharpGenericTypeConstraint)) 1190 State.Stack.back().IsCSharpGenericTypeConstraint = true; 1191 if (Current.isOneOf(tok::comma, TT_BinaryOperator)) 1192 State.Stack.back().NoLineBreakInOperand = false; 1193 if (Current.isOneOf(TT_InheritanceColon, TT_CSharpGenericTypeConstraintColon)) 1194 State.Stack.back().AvoidBinPacking = true; 1195 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) { 1196 if (State.Stack.back().FirstLessLess == 0) 1197 State.Stack.back().FirstLessLess = State.Column; 1198 else 1199 State.Stack.back().LastOperatorWrapped = Newline; 1200 } 1201 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless)) 1202 State.Stack.back().LastOperatorWrapped = Newline; 1203 if (Current.is(TT_ConditionalExpr) && Current.Previous && 1204 !Current.Previous->is(TT_ConditionalExpr)) 1205 State.Stack.back().LastOperatorWrapped = Newline; 1206 if (Current.is(TT_ArraySubscriptLSquare) && 1207 State.Stack.back().StartOfArraySubscripts == 0) 1208 State.Stack.back().StartOfArraySubscripts = State.Column; 1209 if (Current.is(TT_ConditionalExpr) && Current.is(tok::question) && 1210 ((Current.MustBreakBefore) || 1211 (Current.getNextNonComment() && 1212 Current.getNextNonComment()->MustBreakBefore))) 1213 State.Stack.back().IsWrappedConditional = true; 1214 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question)) 1215 State.Stack.back().QuestionColumn = State.Column; 1216 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) { 1217 const FormatToken *Previous = Current.Previous; 1218 while (Previous && Previous->isTrailingComment()) 1219 Previous = Previous->Previous; 1220 if (Previous && Previous->is(tok::question)) 1221 State.Stack.back().QuestionColumn = State.Column; 1222 } 1223 if (!Current.opensScope() && !Current.closesScope() && 1224 !Current.is(TT_PointerOrReference)) 1225 State.LowestLevelOnLine = 1226 std::min(State.LowestLevelOnLine, Current.NestingLevel); 1227 if (Current.isMemberAccess()) 1228 State.Stack.back().StartOfFunctionCall = 1229 !Current.NextOperator ? 0 : State.Column; 1230 if (Current.is(TT_SelectorName)) 1231 State.Stack.back().ObjCSelectorNameFound = true; 1232 if (Current.is(TT_CtorInitializerColon) && 1233 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) { 1234 // Indent 2 from the column, so: 1235 // SomeClass::SomeClass() 1236 // : First(...), ... 1237 // Next(...) 1238 // ^ line up here. 1239 State.Stack.back().Indent = 1240 State.Column + 1241 (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma 1242 ? 0 1243 : 2); 1244 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1245 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) { 1246 State.Stack.back().AvoidBinPacking = true; 1247 State.Stack.back().BreakBeforeParameter = 1248 !Style.AllowAllConstructorInitializersOnNextLine; 1249 } else { 1250 State.Stack.back().BreakBeforeParameter = false; 1251 } 1252 } 1253 if (Current.is(TT_CtorInitializerColon) && 1254 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { 1255 State.Stack.back().Indent = 1256 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1257 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1258 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) 1259 State.Stack.back().AvoidBinPacking = true; 1260 } 1261 if (Current.is(TT_InheritanceColon)) 1262 State.Stack.back().Indent = 1263 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1264 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline) 1265 State.Stack.back().NestedBlockIndent = 1266 State.Column + Current.ColumnWidth + 1; 1267 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow)) 1268 State.Stack.back().LastSpace = State.Column; 1269 1270 // Insert scopes created by fake parenthesis. 1271 const FormatToken *Previous = Current.getPreviousNonComment(); 1272 1273 // Add special behavior to support a format commonly used for JavaScript 1274 // closures: 1275 // SomeFunction(function() { 1276 // foo(); 1277 // bar(); 1278 // }, a, b, c); 1279 if (Current.isNot(tok::comment) && Previous && 1280 Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) && 1281 !Previous->is(TT_DictLiteral) && State.Stack.size() > 1 && 1282 !State.Stack.back().HasMultipleNestedBlocks) { 1283 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline) 1284 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 1285 State.Stack[i].NoLineBreak = true; 1286 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false; 1287 } 1288 if (Previous && 1289 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) || 1290 Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr)) && 1291 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) { 1292 State.Stack.back().NestedBlockInlined = 1293 !Newline && hasNestedBlockInlined(Previous, Current, Style); 1294 } 1295 1296 moveStatePastFakeLParens(State, Newline); 1297 moveStatePastScopeCloser(State); 1298 bool AllowBreak = !State.Stack.back().NoLineBreak && 1299 !State.Stack.back().NoLineBreakInOperand; 1300 moveStatePastScopeOpener(State, Newline); 1301 moveStatePastFakeRParens(State); 1302 1303 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0) 1304 State.StartOfStringLiteral = State.Column + 1; 1305 if (Current.is(TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) 1306 State.StartOfStringLiteral = State.Column + 1; 1307 else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) 1308 State.StartOfStringLiteral = State.Column; 1309 else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) && 1310 !Current.isStringLiteral()) 1311 State.StartOfStringLiteral = 0; 1312 1313 State.Column += Current.ColumnWidth; 1314 State.NextToken = State.NextToken->Next; 1315 1316 unsigned Penalty = 1317 handleEndOfLine(Current, State, DryRun, AllowBreak, Newline); 1318 1319 if (Current.Role) 1320 Current.Role->formatFromToken(State, this, DryRun); 1321 // If the previous has a special role, let it consume tokens as appropriate. 1322 // It is necessary to start at the previous token for the only implemented 1323 // role (comma separated list). That way, the decision whether or not to break 1324 // after the "{" is already done and both options are tried and evaluated. 1325 // FIXME: This is ugly, find a better way. 1326 if (Previous && Previous->Role) 1327 Penalty += Previous->Role->formatAfterToken(State, this, DryRun); 1328 1329 return Penalty; 1330 } 1331 1332 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State, 1333 bool Newline) { 1334 const FormatToken &Current = *State.NextToken; 1335 const FormatToken *Previous = Current.getPreviousNonComment(); 1336 1337 // Don't add extra indentation for the first fake parenthesis after 1338 // 'return', assignments or opening <({[. The indentation for these cases 1339 // is special cased. 1340 bool SkipFirstExtraIndent = 1341 (Previous && (Previous->opensScope() || 1342 Previous->isOneOf(tok::semi, tok::kw_return) || 1343 (Previous->getPrecedence() == prec::Assignment && 1344 Style.AlignOperands != FormatStyle::OAS_DontAlign) || 1345 Previous->is(TT_ObjCMethodExpr))); 1346 for (SmallVectorImpl<prec::Level>::const_reverse_iterator 1347 I = Current.FakeLParens.rbegin(), 1348 E = Current.FakeLParens.rend(); 1349 I != E; ++I) { 1350 ParenState NewParenState = State.Stack.back(); 1351 NewParenState.Tok = nullptr; 1352 NewParenState.ContainsLineBreak = false; 1353 NewParenState.LastOperatorWrapped = true; 1354 NewParenState.IsChainedConditional = false; 1355 NewParenState.IsWrappedConditional = false; 1356 NewParenState.UnindentOperator = false; 1357 NewParenState.NoLineBreak = 1358 NewParenState.NoLineBreak || State.Stack.back().NoLineBreakInOperand; 1359 1360 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists. 1361 if (*I > prec::Comma) 1362 NewParenState.AvoidBinPacking = false; 1363 1364 // Indent from 'LastSpace' unless these are fake parentheses encapsulating 1365 // a builder type call after 'return' or, if the alignment after opening 1366 // brackets is disabled. 1367 if (!Current.isTrailingComment() && 1368 (Style.AlignOperands != FormatStyle::OAS_DontAlign || 1369 *I < prec::Assignment) && 1370 (!Previous || Previous->isNot(tok::kw_return) || 1371 (Style.Language != FormatStyle::LK_Java && *I > 0)) && 1372 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign || 1373 *I != prec::Comma || Current.NestingLevel == 0)) { 1374 NewParenState.Indent = 1375 std::max(std::max(State.Column, NewParenState.Indent), 1376 State.Stack.back().LastSpace); 1377 } 1378 1379 if (Previous && 1380 (Previous->getPrecedence() == prec::Assignment || 1381 Previous->is(tok::kw_return) || 1382 (*I == prec::Conditional && Previous->is(tok::question) && 1383 Previous->is(TT_ConditionalExpr))) && 1384 !Newline) { 1385 // If BreakBeforeBinaryOperators is set, un-indent a bit to account for 1386 // the operator and keep the operands aligned 1387 if (Style.AlignOperands == FormatStyle::OAS_AlignAfterOperator) 1388 NewParenState.UnindentOperator = true; 1389 // Mark indentation as alignment if the expression is aligned. 1390 if (Style.AlignOperands != FormatStyle::OAS_DontAlign) 1391 NewParenState.IsAligned = true; 1392 } 1393 1394 // Do not indent relative to the fake parentheses inserted for "." or "->". 1395 // This is a special case to make the following to statements consistent: 1396 // OuterFunction(InnerFunctionCall( // break 1397 // ParameterToInnerFunction)); 1398 // OuterFunction(SomeObject.InnerFunctionCall( // break 1399 // ParameterToInnerFunction)); 1400 if (*I > prec::Unknown) 1401 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column); 1402 if (*I != prec::Conditional && !Current.is(TT_UnaryOperator) && 1403 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 1404 NewParenState.StartOfFunctionCall = State.Column; 1405 1406 // Indent conditional expressions, unless they are chained "else-if" 1407 // conditionals. Never indent expression where the 'operator' is ',', ';' or 1408 // an assignment (i.e. *I <= prec::Assignment) as those have different 1409 // indentation rules. Indent other expression, unless the indentation needs 1410 // to be skipped. 1411 if (*I == prec::Conditional && Previous && Previous->is(tok::colon) && 1412 Previous->is(TT_ConditionalExpr) && I == Current.FakeLParens.rbegin() && 1413 !State.Stack.back().IsWrappedConditional) { 1414 NewParenState.IsChainedConditional = true; 1415 NewParenState.UnindentOperator = State.Stack.back().UnindentOperator; 1416 } else if (*I == prec::Conditional || 1417 (!SkipFirstExtraIndent && *I > prec::Assignment && 1418 !Current.isTrailingComment())) { 1419 NewParenState.Indent += Style.ContinuationIndentWidth; 1420 } 1421 if ((Previous && !Previous->opensScope()) || *I != prec::Comma) 1422 NewParenState.BreakBeforeParameter = false; 1423 State.Stack.push_back(NewParenState); 1424 SkipFirstExtraIndent = false; 1425 } 1426 } 1427 1428 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) { 1429 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) { 1430 unsigned VariablePos = State.Stack.back().VariablePos; 1431 if (State.Stack.size() == 1) { 1432 // Do not pop the last element. 1433 break; 1434 } 1435 State.Stack.pop_back(); 1436 State.Stack.back().VariablePos = VariablePos; 1437 } 1438 } 1439 1440 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State, 1441 bool Newline) { 1442 const FormatToken &Current = *State.NextToken; 1443 if (!Current.opensScope()) 1444 return; 1445 1446 // Don't allow '<' or '(' in C# generic type constraints to start new scopes. 1447 if (Current.isOneOf(tok::less, tok::l_paren) && 1448 State.Stack.back().IsCSharpGenericTypeConstraint) 1449 return; 1450 1451 if (Current.MatchingParen && Current.is(BK_Block)) { 1452 moveStateToNewBlock(State); 1453 return; 1454 } 1455 1456 unsigned NewIndent; 1457 unsigned LastSpace = State.Stack.back().LastSpace; 1458 bool AvoidBinPacking; 1459 bool BreakBeforeParameter = false; 1460 unsigned NestedBlockIndent = std::max(State.Stack.back().StartOfFunctionCall, 1461 State.Stack.back().NestedBlockIndent); 1462 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 1463 opensProtoMessageField(Current, Style)) { 1464 if (Current.opensBlockOrBlockTypeList(Style)) { 1465 NewIndent = Style.IndentWidth + 1466 std::min(State.Column, State.Stack.back().NestedBlockIndent); 1467 } else { 1468 NewIndent = State.Stack.back().LastSpace + Style.ContinuationIndentWidth; 1469 } 1470 const FormatToken *NextNoComment = Current.getNextNonComment(); 1471 bool EndsInComma = Current.MatchingParen && 1472 Current.MatchingParen->Previous && 1473 Current.MatchingParen->Previous->is(tok::comma); 1474 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) || 1475 Style.Language == FormatStyle::LK_Proto || 1476 Style.Language == FormatStyle::LK_TextProto || 1477 !Style.BinPackArguments || 1478 (NextNoComment && 1479 NextNoComment->isOneOf(TT_DesignatedInitializerPeriod, 1480 TT_DesignatedInitializerLSquare)); 1481 BreakBeforeParameter = EndsInComma; 1482 if (Current.ParameterCount > 1) 1483 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1); 1484 } else { 1485 NewIndent = Style.ContinuationIndentWidth + 1486 std::max(State.Stack.back().LastSpace, 1487 State.Stack.back().StartOfFunctionCall); 1488 1489 // Ensure that different different brackets force relative alignment, e.g.: 1490 // void SomeFunction(vector< // break 1491 // int> v); 1492 // FIXME: We likely want to do this for more combinations of brackets. 1493 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) { 1494 NewIndent = std::max(NewIndent, State.Stack.back().Indent); 1495 LastSpace = std::max(LastSpace, State.Stack.back().Indent); 1496 } 1497 1498 bool EndsInComma = 1499 Current.MatchingParen && 1500 Current.MatchingParen->getPreviousNonComment() && 1501 Current.MatchingParen->getPreviousNonComment()->is(tok::comma); 1502 1503 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters 1504 // for backwards compatibility. 1505 bool ObjCBinPackProtocolList = 1506 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto && 1507 Style.BinPackParameters) || 1508 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always; 1509 1510 bool BinPackDeclaration = 1511 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) || 1512 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList); 1513 1514 AvoidBinPacking = 1515 (State.Stack.back().IsCSharpGenericTypeConstraint) || 1516 (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) || 1517 (State.Line->MustBeDeclaration && !BinPackDeclaration) || 1518 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) || 1519 (Style.ExperimentalAutoDetectBinPacking && 1520 (Current.is(PPK_OnePerLine) || 1521 (!BinPackInconclusiveFunctions && Current.is(PPK_Inconclusive)))); 1522 1523 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen && 1524 Style.ObjCBreakBeforeNestedBlockParam) { 1525 if (Style.ColumnLimit) { 1526 // If this '[' opens an ObjC call, determine whether all parameters fit 1527 // into one line and put one per line if they don't. 1528 if (getLengthToMatchingParen(Current, State.Stack) + State.Column > 1529 getColumnLimit(State)) 1530 BreakBeforeParameter = true; 1531 } else { 1532 // For ColumnLimit = 0, we have to figure out whether there is or has to 1533 // be a line break within this call. 1534 for (const FormatToken *Tok = &Current; 1535 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) { 1536 if (Tok->MustBreakBefore || 1537 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) { 1538 BreakBeforeParameter = true; 1539 break; 1540 } 1541 } 1542 } 1543 } 1544 1545 if (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) 1546 BreakBeforeParameter = true; 1547 } 1548 // Generally inherit NoLineBreak from the current scope to nested scope. 1549 // However, don't do this for non-empty nested blocks, dict literals and 1550 // array literals as these follow different indentation rules. 1551 bool NoLineBreak = 1552 Current.Children.empty() && 1553 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) && 1554 (State.Stack.back().NoLineBreak || 1555 State.Stack.back().NoLineBreakInOperand || 1556 (Current.is(TT_TemplateOpener) && 1557 State.Stack.back().ContainsUnwrappedBuilder)); 1558 State.Stack.push_back( 1559 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak)); 1560 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1561 State.Stack.back().BreakBeforeParameter = BreakBeforeParameter; 1562 State.Stack.back().HasMultipleNestedBlocks = 1563 (Current.BlockParameterCount > 1); 1564 1565 if (Style.BraceWrapping.BeforeLambdaBody && Current.Next != nullptr && 1566 Current.Tok.is(tok::l_paren)) { 1567 // Search for any parameter that is a lambda 1568 FormatToken const *next = Current.Next; 1569 while (next != nullptr) { 1570 if (next->is(TT_LambdaLSquare)) { 1571 State.Stack.back().HasMultipleNestedBlocks = true; 1572 break; 1573 } 1574 next = next->Next; 1575 } 1576 } 1577 1578 State.Stack.back().IsInsideObjCArrayLiteral = 1579 Current.is(TT_ArrayInitializerLSquare) && Current.Previous && 1580 Current.Previous->is(tok::at); 1581 } 1582 1583 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) { 1584 const FormatToken &Current = *State.NextToken; 1585 if (!Current.closesScope()) 1586 return; 1587 1588 // If we encounter a closing ), ], } or >, we can remove a level from our 1589 // stacks. 1590 if (State.Stack.size() > 1 && 1591 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) || 1592 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) || 1593 State.NextToken->is(TT_TemplateCloser) || 1594 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) 1595 State.Stack.pop_back(); 1596 1597 // Reevaluate whether ObjC message arguments fit into one line. 1598 // If a receiver spans multiple lines, e.g.: 1599 // [[object block:^{ 1600 // return 42; 1601 // }] a:42 b:42]; 1602 // BreakBeforeParameter is calculated based on an incorrect assumption 1603 // (it is checked whether the whole expression fits into one line without 1604 // considering a line break inside a message receiver). 1605 // We check whether arguements fit after receiver scope closer (into the same 1606 // line). 1607 if (State.Stack.back().BreakBeforeParameter && Current.MatchingParen && 1608 Current.MatchingParen->Previous) { 1609 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous; 1610 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) && 1611 CurrentScopeOpener.MatchingParen) { 1612 int NecessarySpaceInLine = 1613 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) + 1614 CurrentScopeOpener.TotalLength - Current.TotalLength - 1; 1615 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <= 1616 Style.ColumnLimit) 1617 State.Stack.back().BreakBeforeParameter = false; 1618 } 1619 } 1620 1621 if (Current.is(tok::r_square)) { 1622 // If this ends the array subscript expr, reset the corresponding value. 1623 const FormatToken *NextNonComment = Current.getNextNonComment(); 1624 if (NextNonComment && NextNonComment->isNot(tok::l_square)) 1625 State.Stack.back().StartOfArraySubscripts = 0; 1626 } 1627 } 1628 1629 void ContinuationIndenter::moveStateToNewBlock(LineState &State) { 1630 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent; 1631 // ObjC block sometimes follow special indentation rules. 1632 unsigned NewIndent = 1633 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace) 1634 ? Style.ObjCBlockIndentWidth 1635 : Style.IndentWidth); 1636 State.Stack.push_back(ParenState(State.NextToken, NewIndent, 1637 State.Stack.back().LastSpace, 1638 /*AvoidBinPacking=*/true, 1639 /*NoLineBreak=*/false)); 1640 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1641 State.Stack.back().BreakBeforeParameter = true; 1642 } 1643 1644 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn, 1645 unsigned TabWidth, 1646 encoding::Encoding Encoding) { 1647 size_t LastNewlinePos = Text.find_last_of("\n"); 1648 if (LastNewlinePos == StringRef::npos) { 1649 return StartColumn + 1650 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding); 1651 } else { 1652 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos), 1653 /*StartColumn=*/0, TabWidth, Encoding); 1654 } 1655 } 1656 1657 unsigned ContinuationIndenter::reformatRawStringLiteral( 1658 const FormatToken &Current, LineState &State, 1659 const FormatStyle &RawStringStyle, bool DryRun, bool Newline) { 1660 unsigned StartColumn = State.Column - Current.ColumnWidth; 1661 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText); 1662 StringRef NewDelimiter = 1663 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language); 1664 if (NewDelimiter.empty()) 1665 NewDelimiter = OldDelimiter; 1666 // The text of a raw string is between the leading 'R"delimiter(' and the 1667 // trailing 'delimiter)"'. 1668 unsigned OldPrefixSize = 3 + OldDelimiter.size(); 1669 unsigned OldSuffixSize = 2 + OldDelimiter.size(); 1670 // We create a virtual text environment which expects a null-terminated 1671 // string, so we cannot use StringRef. 1672 std::string RawText = std::string( 1673 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize)); 1674 if (NewDelimiter != OldDelimiter) { 1675 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the 1676 // raw string. 1677 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str(); 1678 if (StringRef(RawText).contains(CanonicalDelimiterSuffix)) 1679 NewDelimiter = OldDelimiter; 1680 } 1681 1682 unsigned NewPrefixSize = 3 + NewDelimiter.size(); 1683 unsigned NewSuffixSize = 2 + NewDelimiter.size(); 1684 1685 // The first start column is the column the raw text starts after formatting. 1686 unsigned FirstStartColumn = StartColumn + NewPrefixSize; 1687 1688 // The next start column is the intended indentation a line break inside 1689 // the raw string at level 0. It is determined by the following rules: 1690 // - if the content starts on newline, it is one level more than the current 1691 // indent, and 1692 // - if the content does not start on a newline, it is the first start 1693 // column. 1694 // These rules have the advantage that the formatted content both does not 1695 // violate the rectangle rule and visually flows within the surrounding 1696 // source. 1697 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n'; 1698 // If this token is the last parameter (checked by looking if it's followed by 1699 // `)` and is not on a newline, the base the indent off the line's nested 1700 // block indent. Otherwise, base the indent off the arguments indent, so we 1701 // can achieve: 1702 // 1703 // fffffffffff(1, 2, 3, R"pb( 1704 // key1: 1 # 1705 // key2: 2)pb"); 1706 // 1707 // fffffffffff(1, 2, 3, 1708 // R"pb( 1709 // key1: 1 # 1710 // key2: 2 1711 // )pb"); 1712 // 1713 // fffffffffff(1, 2, 3, 1714 // R"pb( 1715 // key1: 1 # 1716 // key2: 2 1717 // )pb", 1718 // 5); 1719 unsigned CurrentIndent = 1720 (!Newline && Current.Next && Current.Next->is(tok::r_paren)) 1721 ? State.Stack.back().NestedBlockIndent 1722 : State.Stack.back().Indent; 1723 unsigned NextStartColumn = ContentStartsOnNewline 1724 ? CurrentIndent + Style.IndentWidth 1725 : FirstStartColumn; 1726 1727 // The last start column is the column the raw string suffix starts if it is 1728 // put on a newline. 1729 // The last start column is the intended indentation of the raw string postfix 1730 // if it is put on a newline. It is determined by the following rules: 1731 // - if the raw string prefix starts on a newline, it is the column where 1732 // that raw string prefix starts, and 1733 // - if the raw string prefix does not start on a newline, it is the current 1734 // indent. 1735 unsigned LastStartColumn = 1736 Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent; 1737 1738 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat( 1739 RawStringStyle, RawText, {tooling::Range(0, RawText.size())}, 1740 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>", 1741 /*Status=*/nullptr); 1742 1743 auto NewCode = applyAllReplacements(RawText, Fixes.first); 1744 tooling::Replacements NoFixes; 1745 if (!NewCode) { 1746 return addMultilineToken(Current, State); 1747 } 1748 if (!DryRun) { 1749 if (NewDelimiter != OldDelimiter) { 1750 // In 'R"delimiter(...', the delimiter starts 2 characters after the start 1751 // of the token. 1752 SourceLocation PrefixDelimiterStart = 1753 Current.Tok.getLocation().getLocWithOffset(2); 1754 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement( 1755 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1756 if (PrefixErr) { 1757 llvm::errs() 1758 << "Failed to update the prefix delimiter of a raw string: " 1759 << llvm::toString(std::move(PrefixErr)) << "\n"; 1760 } 1761 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at 1762 // position length - 1 - |delimiter|. 1763 SourceLocation SuffixDelimiterStart = 1764 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() - 1765 1 - OldDelimiter.size()); 1766 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement( 1767 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1768 if (SuffixErr) { 1769 llvm::errs() 1770 << "Failed to update the suffix delimiter of a raw string: " 1771 << llvm::toString(std::move(SuffixErr)) << "\n"; 1772 } 1773 } 1774 SourceLocation OriginLoc = 1775 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize); 1776 for (const tooling::Replacement &Fix : Fixes.first) { 1777 auto Err = Whitespaces.addReplacement(tooling::Replacement( 1778 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()), 1779 Fix.getLength(), Fix.getReplacementText())); 1780 if (Err) { 1781 llvm::errs() << "Failed to reformat raw string: " 1782 << llvm::toString(std::move(Err)) << "\n"; 1783 } 1784 } 1785 } 1786 unsigned RawLastLineEndColumn = getLastLineEndColumn( 1787 *NewCode, FirstStartColumn, Style.TabWidth, Encoding); 1788 State.Column = RawLastLineEndColumn + NewSuffixSize; 1789 // Since we're updating the column to after the raw string literal here, we 1790 // have to manually add the penalty for the prefix R"delim( over the column 1791 // limit. 1792 unsigned PrefixExcessCharacters = 1793 StartColumn + NewPrefixSize > Style.ColumnLimit 1794 ? StartColumn + NewPrefixSize - Style.ColumnLimit 1795 : 0; 1796 bool IsMultiline = 1797 ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos); 1798 if (IsMultiline) { 1799 // Break before further function parameters on all levels. 1800 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 1801 State.Stack[i].BreakBeforeParameter = true; 1802 } 1803 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter; 1804 } 1805 1806 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current, 1807 LineState &State) { 1808 // Break before further function parameters on all levels. 1809 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 1810 State.Stack[i].BreakBeforeParameter = true; 1811 1812 unsigned ColumnsUsed = State.Column; 1813 // We can only affect layout of the first and the last line, so the penalty 1814 // for all other lines is constant, and we ignore it. 1815 State.Column = Current.LastLineColumnWidth; 1816 1817 if (ColumnsUsed > getColumnLimit(State)) 1818 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State)); 1819 return 0; 1820 } 1821 1822 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current, 1823 LineState &State, bool DryRun, 1824 bool AllowBreak, bool Newline) { 1825 unsigned Penalty = 0; 1826 // Compute the raw string style to use in case this is a raw string literal 1827 // that can be reformatted. 1828 auto RawStringStyle = getRawStringStyle(Current, State); 1829 if (RawStringStyle && !Current.Finalized) { 1830 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun, 1831 Newline); 1832 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) { 1833 // Don't break multi-line tokens other than block comments and raw string 1834 // literals. Instead, just update the state. 1835 Penalty = addMultilineToken(Current, State); 1836 } else if (State.Line->Type != LT_ImportStatement) { 1837 // We generally don't break import statements. 1838 LineState OriginalState = State; 1839 1840 // Whether we force the reflowing algorithm to stay strictly within the 1841 // column limit. 1842 bool Strict = false; 1843 // Whether the first non-strict attempt at reflowing did intentionally 1844 // exceed the column limit. 1845 bool Exceeded = false; 1846 std::tie(Penalty, Exceeded) = breakProtrudingToken( 1847 Current, State, AllowBreak, /*DryRun=*/true, Strict); 1848 if (Exceeded) { 1849 // If non-strict reflowing exceeds the column limit, try whether strict 1850 // reflowing leads to an overall lower penalty. 1851 LineState StrictState = OriginalState; 1852 unsigned StrictPenalty = 1853 breakProtrudingToken(Current, StrictState, AllowBreak, 1854 /*DryRun=*/true, /*Strict=*/true) 1855 .first; 1856 Strict = StrictPenalty <= Penalty; 1857 if (Strict) { 1858 Penalty = StrictPenalty; 1859 State = StrictState; 1860 } 1861 } 1862 if (!DryRun) { 1863 // If we're not in dry-run mode, apply the changes with the decision on 1864 // strictness made above. 1865 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false, 1866 Strict); 1867 } 1868 } 1869 if (State.Column > getColumnLimit(State)) { 1870 unsigned ExcessCharacters = State.Column - getColumnLimit(State); 1871 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters; 1872 } 1873 return Penalty; 1874 } 1875 1876 // Returns the enclosing function name of a token, or the empty string if not 1877 // found. 1878 static StringRef getEnclosingFunctionName(const FormatToken &Current) { 1879 // Look for: 'function(' or 'function<templates>(' before Current. 1880 auto Tok = Current.getPreviousNonComment(); 1881 if (!Tok || !Tok->is(tok::l_paren)) 1882 return ""; 1883 Tok = Tok->getPreviousNonComment(); 1884 if (!Tok) 1885 return ""; 1886 if (Tok->is(TT_TemplateCloser)) { 1887 Tok = Tok->MatchingParen; 1888 if (Tok) 1889 Tok = Tok->getPreviousNonComment(); 1890 } 1891 if (!Tok || !Tok->is(tok::identifier)) 1892 return ""; 1893 return Tok->TokenText; 1894 } 1895 1896 llvm::Optional<FormatStyle> 1897 ContinuationIndenter::getRawStringStyle(const FormatToken &Current, 1898 const LineState &State) { 1899 if (!Current.isStringLiteral()) 1900 return None; 1901 auto Delimiter = getRawStringDelimiter(Current.TokenText); 1902 if (!Delimiter) 1903 return None; 1904 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter); 1905 if (!RawStringStyle && Delimiter->empty()) 1906 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle( 1907 getEnclosingFunctionName(Current)); 1908 if (!RawStringStyle) 1909 return None; 1910 RawStringStyle->ColumnLimit = getColumnLimit(State); 1911 return RawStringStyle; 1912 } 1913 1914 std::unique_ptr<BreakableToken> 1915 ContinuationIndenter::createBreakableToken(const FormatToken &Current, 1916 LineState &State, bool AllowBreak) { 1917 unsigned StartColumn = State.Column - Current.ColumnWidth; 1918 if (Current.isStringLiteral()) { 1919 // FIXME: String literal breaking is currently disabled for C#, Java, Json 1920 // and JavaScript, as it requires strings to be merged using "+" which we 1921 // don't support. 1922 if (Style.Language == FormatStyle::LK_Java || 1923 Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp() || 1924 Style.isJson() || !Style.BreakStringLiterals || !AllowBreak) 1925 return nullptr; 1926 1927 // Don't break string literals inside preprocessor directives (except for 1928 // #define directives, as their contents are stored in separate lines and 1929 // are not affected by this check). 1930 // This way we avoid breaking code with line directives and unknown 1931 // preprocessor directives that contain long string literals. 1932 if (State.Line->Type == LT_PreprocessorDirective) 1933 return nullptr; 1934 // Exempts unterminated string literals from line breaking. The user will 1935 // likely want to terminate the string before any line breaking is done. 1936 if (Current.IsUnterminatedLiteral) 1937 return nullptr; 1938 // Don't break string literals inside Objective-C array literals (doing so 1939 // raises the warning -Wobjc-string-concatenation). 1940 if (State.Stack.back().IsInsideObjCArrayLiteral) { 1941 return nullptr; 1942 } 1943 1944 StringRef Text = Current.TokenText; 1945 StringRef Prefix; 1946 StringRef Postfix; 1947 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'. 1948 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to 1949 // reduce the overhead) for each FormatToken, which is a string, so that we 1950 // don't run multiple checks here on the hot path. 1951 if ((Text.endswith(Postfix = "\"") && 1952 (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") || 1953 Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") || 1954 Text.startswith(Prefix = "u8\"") || 1955 Text.startswith(Prefix = "L\""))) || 1956 (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) { 1957 // We need this to address the case where there is an unbreakable tail 1958 // only if certain other formatting decisions have been taken. The 1959 // UnbreakableTailLength of Current is an overapproximation is that case 1960 // and we need to be correct here. 1961 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State)) 1962 ? 0 1963 : Current.UnbreakableTailLength; 1964 return std::make_unique<BreakableStringLiteral>( 1965 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength, 1966 State.Line->InPPDirective, Encoding, Style); 1967 } 1968 } else if (Current.is(TT_BlockComment)) { 1969 if (!Style.ReflowComments || 1970 // If a comment token switches formatting, like 1971 // /* clang-format on */, we don't want to break it further, 1972 // but we may still want to adjust its indentation. 1973 switchesFormatting(Current)) { 1974 return nullptr; 1975 } 1976 return std::make_unique<BreakableBlockComment>( 1977 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 1978 State.Line->InPPDirective, Encoding, Style, Whitespaces.useCRLF()); 1979 } else if (Current.is(TT_LineComment) && 1980 (Current.Previous == nullptr || 1981 Current.Previous->isNot(TT_ImplicitStringLiteral))) { 1982 if (!Style.ReflowComments || 1983 CommentPragmasRegex.match(Current.TokenText.substr(2)) || 1984 switchesFormatting(Current)) 1985 return nullptr; 1986 return std::make_unique<BreakableLineCommentSection>( 1987 Current, StartColumn, /*InPPDirective=*/false, Encoding, Style); 1988 } 1989 return nullptr; 1990 } 1991 1992 std::pair<unsigned, bool> 1993 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current, 1994 LineState &State, bool AllowBreak, 1995 bool DryRun, bool Strict) { 1996 std::unique_ptr<const BreakableToken> Token = 1997 createBreakableToken(Current, State, AllowBreak); 1998 if (!Token) 1999 return {0, false}; 2000 assert(Token->getLineCount() > 0); 2001 unsigned ColumnLimit = getColumnLimit(State); 2002 if (Current.is(TT_LineComment)) { 2003 // We don't insert backslashes when breaking line comments. 2004 ColumnLimit = Style.ColumnLimit; 2005 } 2006 if (ColumnLimit == 0) { 2007 // To make the rest of the function easier set the column limit to the 2008 // maximum, if there should be no limit. 2009 ColumnLimit = std::numeric_limits<decltype(ColumnLimit)>::max(); 2010 } 2011 if (Current.UnbreakableTailLength >= ColumnLimit) 2012 return {0, false}; 2013 // ColumnWidth was already accounted into State.Column before calling 2014 // breakProtrudingToken. 2015 unsigned StartColumn = State.Column - Current.ColumnWidth; 2016 unsigned NewBreakPenalty = Current.isStringLiteral() 2017 ? Style.PenaltyBreakString 2018 : Style.PenaltyBreakComment; 2019 // Stores whether we intentionally decide to let a line exceed the column 2020 // limit. 2021 bool Exceeded = false; 2022 // Stores whether we introduce a break anywhere in the token. 2023 bool BreakInserted = Token->introducesBreakBeforeToken(); 2024 // Store whether we inserted a new line break at the end of the previous 2025 // logical line. 2026 bool NewBreakBefore = false; 2027 // We use a conservative reflowing strategy. Reflow starts after a line is 2028 // broken or the corresponding whitespace compressed. Reflow ends as soon as a 2029 // line that doesn't get reflown with the previous line is reached. 2030 bool Reflow = false; 2031 // Keep track of where we are in the token: 2032 // Where we are in the content of the current logical line. 2033 unsigned TailOffset = 0; 2034 // The column number we're currently at. 2035 unsigned ContentStartColumn = 2036 Token->getContentStartColumn(0, /*Break=*/false); 2037 // The number of columns left in the current logical line after TailOffset. 2038 unsigned RemainingTokenColumns = 2039 Token->getRemainingLength(0, TailOffset, ContentStartColumn); 2040 // Adapt the start of the token, for example indent. 2041 if (!DryRun) 2042 Token->adaptStartOfLine(0, Whitespaces); 2043 2044 unsigned ContentIndent = 0; 2045 unsigned Penalty = 0; 2046 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column " 2047 << StartColumn << ".\n"); 2048 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount(); 2049 LineIndex != EndIndex; ++LineIndex) { 2050 LLVM_DEBUG(llvm::dbgs() 2051 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n"); 2052 NewBreakBefore = false; 2053 // If we did reflow the previous line, we'll try reflowing again. Otherwise 2054 // we'll start reflowing if the current line is broken or whitespace is 2055 // compressed. 2056 bool TryReflow = Reflow; 2057 // Break the current token until we can fit the rest of the line. 2058 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2059 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: " 2060 << (ContentStartColumn + RemainingTokenColumns) 2061 << ", space: " << ColumnLimit 2062 << ", reflown prefix: " << ContentStartColumn 2063 << ", offset in line: " << TailOffset << "\n"); 2064 // If the current token doesn't fit, find the latest possible split in the 2065 // current line so that breaking at it will be under the column limit. 2066 // FIXME: Use the earliest possible split while reflowing to correctly 2067 // compress whitespace within a line. 2068 BreakableToken::Split Split = 2069 Token->getSplit(LineIndex, TailOffset, ColumnLimit, 2070 ContentStartColumn, CommentPragmasRegex); 2071 if (Split.first == StringRef::npos) { 2072 // No break opportunity - update the penalty and continue with the next 2073 // logical line. 2074 if (LineIndex < EndIndex - 1) 2075 // The last line's penalty is handled in addNextStateToQueue() or when 2076 // calling replaceWhitespaceAfterLastLine below. 2077 Penalty += Style.PenaltyExcessCharacter * 2078 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2079 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n"); 2080 break; 2081 } 2082 assert(Split.first != 0); 2083 2084 if (Token->supportsReflow()) { 2085 // Check whether the next natural split point after the current one can 2086 // still fit the line, either because we can compress away whitespace, 2087 // or because the penalty the excess characters introduce is lower than 2088 // the break penalty. 2089 // We only do this for tokens that support reflowing, and thus allow us 2090 // to change the whitespace arbitrarily (e.g. comments). 2091 // Other tokens, like string literals, can be broken on arbitrary 2092 // positions. 2093 2094 // First, compute the columns from TailOffset to the next possible split 2095 // position. 2096 // For example: 2097 // ColumnLimit: | 2098 // // Some text that breaks 2099 // ^ tail offset 2100 // ^-- split 2101 // ^-------- to split columns 2102 // ^--- next split 2103 // ^--------------- to next split columns 2104 unsigned ToSplitColumns = Token->getRangeLength( 2105 LineIndex, TailOffset, Split.first, ContentStartColumn); 2106 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n"); 2107 2108 BreakableToken::Split NextSplit = Token->getSplit( 2109 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit, 2110 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex); 2111 // Compute the columns necessary to fit the next non-breakable sequence 2112 // into the current line. 2113 unsigned ToNextSplitColumns = 0; 2114 if (NextSplit.first == StringRef::npos) { 2115 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset, 2116 ContentStartColumn); 2117 } else { 2118 ToNextSplitColumns = Token->getRangeLength( 2119 LineIndex, TailOffset, 2120 Split.first + Split.second + NextSplit.first, ContentStartColumn); 2121 } 2122 // Compress the whitespace between the break and the start of the next 2123 // unbreakable sequence. 2124 ToNextSplitColumns = 2125 Token->getLengthAfterCompression(ToNextSplitColumns, Split); 2126 LLVM_DEBUG(llvm::dbgs() 2127 << " ContentStartColumn: " << ContentStartColumn << "\n"); 2128 LLVM_DEBUG(llvm::dbgs() 2129 << " ToNextSplit: " << ToNextSplitColumns << "\n"); 2130 // If the whitespace compression makes us fit, continue on the current 2131 // line. 2132 bool ContinueOnLine = 2133 ContentStartColumn + ToNextSplitColumns <= ColumnLimit; 2134 unsigned ExcessCharactersPenalty = 0; 2135 if (!ContinueOnLine && !Strict) { 2136 // Similarly, if the excess characters' penalty is lower than the 2137 // penalty of introducing a new break, continue on the current line. 2138 ExcessCharactersPenalty = 2139 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) * 2140 Style.PenaltyExcessCharacter; 2141 LLVM_DEBUG(llvm::dbgs() 2142 << " Penalty excess: " << ExcessCharactersPenalty 2143 << "\n break : " << NewBreakPenalty << "\n"); 2144 if (ExcessCharactersPenalty < NewBreakPenalty) { 2145 Exceeded = true; 2146 ContinueOnLine = true; 2147 } 2148 } 2149 if (ContinueOnLine) { 2150 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n"); 2151 // The current line fits after compressing the whitespace - reflow 2152 // the next line into it if possible. 2153 TryReflow = true; 2154 if (!DryRun) 2155 Token->compressWhitespace(LineIndex, TailOffset, Split, 2156 Whitespaces); 2157 // When we continue on the same line, leave one space between content. 2158 ContentStartColumn += ToSplitColumns + 1; 2159 Penalty += ExcessCharactersPenalty; 2160 TailOffset += Split.first + Split.second; 2161 RemainingTokenColumns = Token->getRemainingLength( 2162 LineIndex, TailOffset, ContentStartColumn); 2163 continue; 2164 } 2165 } 2166 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n"); 2167 // Update the ContentIndent only if the current line was not reflown with 2168 // the previous line, since in that case the previous line should still 2169 // determine the ContentIndent. Also never intent the last line. 2170 if (!Reflow) 2171 ContentIndent = Token->getContentIndent(LineIndex); 2172 LLVM_DEBUG(llvm::dbgs() 2173 << " ContentIndent: " << ContentIndent << "\n"); 2174 ContentStartColumn = ContentIndent + Token->getContentStartColumn( 2175 LineIndex, /*Break=*/true); 2176 2177 unsigned NewRemainingTokenColumns = Token->getRemainingLength( 2178 LineIndex, TailOffset + Split.first + Split.second, 2179 ContentStartColumn); 2180 if (NewRemainingTokenColumns == 0) { 2181 // No content to indent. 2182 ContentIndent = 0; 2183 ContentStartColumn = 2184 Token->getContentStartColumn(LineIndex, /*Break=*/true); 2185 NewRemainingTokenColumns = Token->getRemainingLength( 2186 LineIndex, TailOffset + Split.first + Split.second, 2187 ContentStartColumn); 2188 } 2189 2190 // When breaking before a tab character, it may be moved by a few columns, 2191 // but will still be expanded to the next tab stop, so we don't save any 2192 // columns. 2193 if (NewRemainingTokenColumns == RemainingTokenColumns) { 2194 // FIXME: Do we need to adjust the penalty? 2195 break; 2196 } 2197 assert(NewRemainingTokenColumns < RemainingTokenColumns); 2198 2199 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first 2200 << ", " << Split.second << "\n"); 2201 if (!DryRun) 2202 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent, 2203 Whitespaces); 2204 2205 Penalty += NewBreakPenalty; 2206 TailOffset += Split.first + Split.second; 2207 RemainingTokenColumns = NewRemainingTokenColumns; 2208 BreakInserted = true; 2209 NewBreakBefore = true; 2210 } 2211 // In case there's another line, prepare the state for the start of the next 2212 // line. 2213 if (LineIndex + 1 != EndIndex) { 2214 unsigned NextLineIndex = LineIndex + 1; 2215 if (NewBreakBefore) 2216 // After breaking a line, try to reflow the next line into the current 2217 // one once RemainingTokenColumns fits. 2218 TryReflow = true; 2219 if (TryReflow) { 2220 // We decided that we want to try reflowing the next line into the 2221 // current one. 2222 // We will now adjust the state as if the reflow is successful (in 2223 // preparation for the next line), and see whether that works. If we 2224 // decide that we cannot reflow, we will later reset the state to the 2225 // start of the next line. 2226 Reflow = false; 2227 // As we did not continue breaking the line, RemainingTokenColumns is 2228 // known to fit after ContentStartColumn. Adapt ContentStartColumn to 2229 // the position at which we want to format the next line if we do 2230 // actually reflow. 2231 // When we reflow, we need to add a space between the end of the current 2232 // line and the next line's start column. 2233 ContentStartColumn += RemainingTokenColumns + 1; 2234 // Get the split that we need to reflow next logical line into the end 2235 // of the current one; the split will include any leading whitespace of 2236 // the next logical line. 2237 BreakableToken::Split SplitBeforeNext = 2238 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex); 2239 LLVM_DEBUG(llvm::dbgs() 2240 << " Size of reflown text: " << ContentStartColumn 2241 << "\n Potential reflow split: "); 2242 if (SplitBeforeNext.first != StringRef::npos) { 2243 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", " 2244 << SplitBeforeNext.second << "\n"); 2245 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second; 2246 // If the rest of the next line fits into the current line below the 2247 // column limit, we can safely reflow. 2248 RemainingTokenColumns = Token->getRemainingLength( 2249 NextLineIndex, TailOffset, ContentStartColumn); 2250 Reflow = true; 2251 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2252 LLVM_DEBUG(llvm::dbgs() 2253 << " Over limit after reflow, need: " 2254 << (ContentStartColumn + RemainingTokenColumns) 2255 << ", space: " << ColumnLimit 2256 << ", reflown prefix: " << ContentStartColumn 2257 << ", offset in line: " << TailOffset << "\n"); 2258 // If the whole next line does not fit, try to find a point in 2259 // the next line at which we can break so that attaching the part 2260 // of the next line to that break point onto the current line is 2261 // below the column limit. 2262 BreakableToken::Split Split = 2263 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit, 2264 ContentStartColumn, CommentPragmasRegex); 2265 if (Split.first == StringRef::npos) { 2266 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n"); 2267 Reflow = false; 2268 } else { 2269 // Check whether the first split point gets us below the column 2270 // limit. Note that we will execute this split below as part of 2271 // the normal token breaking and reflow logic within the line. 2272 unsigned ToSplitColumns = Token->getRangeLength( 2273 NextLineIndex, TailOffset, Split.first, ContentStartColumn); 2274 if (ContentStartColumn + ToSplitColumns > ColumnLimit) { 2275 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: " 2276 << (ContentStartColumn + ToSplitColumns) 2277 << ", space: " << ColumnLimit); 2278 unsigned ExcessCharactersPenalty = 2279 (ContentStartColumn + ToSplitColumns - ColumnLimit) * 2280 Style.PenaltyExcessCharacter; 2281 if (NewBreakPenalty < ExcessCharactersPenalty) { 2282 Reflow = false; 2283 } 2284 } 2285 } 2286 } 2287 } else { 2288 LLVM_DEBUG(llvm::dbgs() << "not found.\n"); 2289 } 2290 } 2291 if (!Reflow) { 2292 // If we didn't reflow into the next line, the only space to consider is 2293 // the next logical line. Reset our state to match the start of the next 2294 // line. 2295 TailOffset = 0; 2296 ContentStartColumn = 2297 Token->getContentStartColumn(NextLineIndex, /*Break=*/false); 2298 RemainingTokenColumns = Token->getRemainingLength( 2299 NextLineIndex, TailOffset, ContentStartColumn); 2300 // Adapt the start of the token, for example indent. 2301 if (!DryRun) 2302 Token->adaptStartOfLine(NextLineIndex, Whitespaces); 2303 } else { 2304 // If we found a reflow split and have added a new break before the next 2305 // line, we are going to remove the line break at the start of the next 2306 // logical line. For example, here we'll add a new line break after 2307 // 'text', and subsequently delete the line break between 'that' and 2308 // 'reflows'. 2309 // // some text that 2310 // // reflows 2311 // -> 2312 // // some text 2313 // // that reflows 2314 // When adding the line break, we also added the penalty for it, so we 2315 // need to subtract that penalty again when we remove the line break due 2316 // to reflowing. 2317 if (NewBreakBefore) { 2318 assert(Penalty >= NewBreakPenalty); 2319 Penalty -= NewBreakPenalty; 2320 } 2321 if (!DryRun) 2322 Token->reflow(NextLineIndex, Whitespaces); 2323 } 2324 } 2325 } 2326 2327 BreakableToken::Split SplitAfterLastLine = 2328 Token->getSplitAfterLastLine(TailOffset); 2329 if (SplitAfterLastLine.first != StringRef::npos) { 2330 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n"); 2331 2332 // We add the last line's penalty here, since that line is going to be split 2333 // now. 2334 Penalty += Style.PenaltyExcessCharacter * 2335 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2336 2337 if (!DryRun) 2338 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine, 2339 Whitespaces); 2340 ContentStartColumn = 2341 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true); 2342 RemainingTokenColumns = Token->getRemainingLength( 2343 Token->getLineCount() - 1, 2344 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second, 2345 ContentStartColumn); 2346 } 2347 2348 State.Column = ContentStartColumn + RemainingTokenColumns - 2349 Current.UnbreakableTailLength; 2350 2351 if (BreakInserted) { 2352 // If we break the token inside a parameter list, we need to break before 2353 // the next parameter on all levels, so that the next parameter is clearly 2354 // visible. Line comments already introduce a break. 2355 if (Current.isNot(TT_LineComment)) { 2356 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i) 2357 State.Stack[i].BreakBeforeParameter = true; 2358 } 2359 2360 if (Current.is(TT_BlockComment)) 2361 State.NoContinuation = true; 2362 2363 State.Stack.back().LastSpace = StartColumn; 2364 } 2365 2366 Token->updateNextToken(State); 2367 2368 return {Penalty, Exceeded}; 2369 } 2370 2371 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const { 2372 // In preprocessor directives reserve two chars for trailing " \" 2373 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0); 2374 } 2375 2376 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) { 2377 const FormatToken &Current = *State.NextToken; 2378 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral)) 2379 return false; 2380 // We never consider raw string literals "multiline" for the purpose of 2381 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased 2382 // (see TokenAnnotator::mustBreakBefore(). 2383 if (Current.TokenText.startswith("R\"")) 2384 return false; 2385 if (Current.IsMultiline) 2386 return true; 2387 if (Current.getNextNonComment() && 2388 Current.getNextNonComment()->isStringLiteral()) 2389 return true; // Implicit concatenation. 2390 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals && 2391 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength > 2392 Style.ColumnLimit) 2393 return true; // String will be split. 2394 return false; 2395 } 2396 2397 } // namespace format 2398 } // namespace clang 2399