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/ADT/StringSet.h" 23 #include "llvm/Support/Debug.h" 24 25 #define DEBUG_TYPE "format-indenter" 26 27 namespace clang { 28 namespace format { 29 30 // Returns true if a TT_SelectorName should be indented when wrapped, 31 // false otherwise. 32 static bool shouldIndentWrappedSelectorName(const FormatStyle &Style, 33 LineType LineType) { 34 return Style.IndentWrappedFunctionNames || LineType == LT_ObjCMethodDecl; 35 } 36 37 // Returns the length of everything up to the first possible line break after 38 // the ), ], } or > matching \c Tok. 39 static unsigned getLengthToMatchingParen(const FormatToken &Tok, 40 const SmallVector<ParenState> &Stack) { 41 // Normally whether or not a break before T is possible is calculated and 42 // stored in T.CanBreakBefore. Braces, array initializers and text proto 43 // messages like `key: < ... >` are an exception: a break is possible 44 // before a closing brace R if a break was inserted after the corresponding 45 // opening brace. The information about whether or not a break is needed 46 // before a closing brace R is stored in the ParenState field 47 // S.BreakBeforeClosingBrace where S is the state that R closes. 48 // 49 // In order to decide whether there can be a break before encountered right 50 // braces, this implementation iterates over the sequence of tokens and over 51 // the paren stack in lockstep, keeping track of the stack level which visited 52 // right braces correspond to in MatchingStackIndex. 53 // 54 // For example, consider: 55 // L. <- line number 56 // 1. { 57 // 2. {1}, 58 // 3. {2}, 59 // 4. {{3}}} 60 // ^ where we call this method with this token. 61 // The paren stack at this point contains 3 brace levels: 62 // 0. { at line 1, BreakBeforeClosingBrace: true 63 // 1. first { at line 4, BreakBeforeClosingBrace: false 64 // 2. second { at line 4, BreakBeforeClosingBrace: false, 65 // where there might be fake parens levels in-between these levels. 66 // The algorithm will start at the first } on line 4, which is the matching 67 // brace of the initial left brace and at level 2 of the stack. Then, 68 // examining BreakBeforeClosingBrace: false at level 2, it will continue to 69 // the second } on line 4, and will traverse the stack downwards until it 70 // finds the matching { on level 1. Then, examining BreakBeforeClosingBrace: 71 // false at level 1, it will continue to the third } on line 4 and will 72 // traverse the stack downwards until it finds the matching { on level 0. 73 // Then, examining BreakBeforeClosingBrace: true at level 0, the algorithm 74 // will stop and will use the second } on line 4 to determine the length to 75 // return, as in this example the range will include the tokens: {3}} 76 // 77 // The algorithm will only traverse the stack if it encounters braces, array 78 // initializer squares or text proto angle brackets. 79 if (!Tok.MatchingParen) 80 return 0; 81 FormatToken *End = Tok.MatchingParen; 82 // Maintains a stack level corresponding to the current End token. 83 int MatchingStackIndex = Stack.size() - 1; 84 // Traverses the stack downwards, looking for the level to which LBrace 85 // corresponds. Returns either a pointer to the matching level or nullptr if 86 // LParen is not found in the initial portion of the stack up to 87 // MatchingStackIndex. 88 auto FindParenState = [&](const FormatToken *LBrace) -> const ParenState * { 89 while (MatchingStackIndex >= 0 && Stack[MatchingStackIndex].Tok != LBrace) 90 --MatchingStackIndex; 91 return MatchingStackIndex >= 0 ? &Stack[MatchingStackIndex] : nullptr; 92 }; 93 for (; End->Next; End = End->Next) { 94 if (End->Next->CanBreakBefore) 95 break; 96 if (!End->Next->closesScope()) 97 continue; 98 if (End->Next->MatchingParen && 99 End->Next->MatchingParen->isOneOf( 100 tok::l_brace, TT_ArrayInitializerLSquare, tok::less)) { 101 const ParenState *State = FindParenState(End->Next->MatchingParen); 102 if (State && State->BreakBeforeClosingBrace) 103 break; 104 } 105 } 106 return End->TotalLength - Tok.TotalLength + 1; 107 } 108 109 static unsigned getLengthToNextOperator(const FormatToken &Tok) { 110 if (!Tok.NextOperator) 111 return 0; 112 return Tok.NextOperator->TotalLength - Tok.TotalLength; 113 } 114 115 // Returns \c true if \c Tok is the "." or "->" of a call and starts the next 116 // segment of a builder type call. 117 static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) { 118 return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope(); 119 } 120 121 // Returns \c true if \c Current starts a new parameter. 122 static bool startsNextParameter(const FormatToken &Current, 123 const FormatStyle &Style) { 124 const FormatToken &Previous = *Current.Previous; 125 if (Current.is(TT_CtorInitializerComma) && 126 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) { 127 return true; 128 } 129 if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName)) 130 return true; 131 return Previous.is(tok::comma) && !Current.isTrailingComment() && 132 ((Previous.isNot(TT_CtorInitializerComma) || 133 Style.BreakConstructorInitializers != 134 FormatStyle::BCIS_BeforeComma) && 135 (Previous.isNot(TT_InheritanceComma) || 136 Style.BreakInheritanceList != FormatStyle::BILS_BeforeComma)); 137 } 138 139 static bool opensProtoMessageField(const FormatToken &LessTok, 140 const FormatStyle &Style) { 141 if (LessTok.isNot(tok::less)) 142 return false; 143 return Style.Language == FormatStyle::LK_TextProto || 144 (Style.Language == FormatStyle::LK_Proto && 145 (LessTok.NestingLevel > 0 || 146 (LessTok.Previous && LessTok.Previous->is(tok::equal)))); 147 } 148 149 // Returns the delimiter of a raw string literal, or None if TokenText is not 150 // the text of a raw string literal. The delimiter could be the empty string. 151 // For example, the delimiter of R"deli(cont)deli" is deli. 152 static llvm::Optional<StringRef> getRawStringDelimiter(StringRef TokenText) { 153 if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'. 154 || !TokenText.startswith("R\"") || !TokenText.endswith("\"")) { 155 return None; 156 } 157 158 // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has 159 // size at most 16 by the standard, so the first '(' must be among the first 160 // 19 bytes. 161 size_t LParenPos = TokenText.substr(0, 19).find_first_of('('); 162 if (LParenPos == StringRef::npos) 163 return None; 164 StringRef Delimiter = TokenText.substr(2, LParenPos - 2); 165 166 // Check that the string ends in ')Delimiter"'. 167 size_t RParenPos = TokenText.size() - Delimiter.size() - 2; 168 if (TokenText[RParenPos] != ')') 169 return None; 170 if (!TokenText.substr(RParenPos + 1).startswith(Delimiter)) 171 return None; 172 return Delimiter; 173 } 174 175 // Returns the canonical delimiter for \p Language, or the empty string if no 176 // canonical delimiter is specified. 177 static StringRef 178 getCanonicalRawStringDelimiter(const FormatStyle &Style, 179 FormatStyle::LanguageKind Language) { 180 for (const auto &Format : Style.RawStringFormats) 181 if (Format.Language == Language) 182 return StringRef(Format.CanonicalDelimiter); 183 return ""; 184 } 185 186 RawStringFormatStyleManager::RawStringFormatStyleManager( 187 const FormatStyle &CodeStyle) { 188 for (const auto &RawStringFormat : CodeStyle.RawStringFormats) { 189 llvm::Optional<FormatStyle> LanguageStyle = 190 CodeStyle.GetLanguageStyle(RawStringFormat.Language); 191 if (!LanguageStyle) { 192 FormatStyle PredefinedStyle; 193 if (!getPredefinedStyle(RawStringFormat.BasedOnStyle, 194 RawStringFormat.Language, &PredefinedStyle)) { 195 PredefinedStyle = getLLVMStyle(); 196 PredefinedStyle.Language = RawStringFormat.Language; 197 } 198 LanguageStyle = PredefinedStyle; 199 } 200 LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit; 201 for (StringRef Delimiter : RawStringFormat.Delimiters) 202 DelimiterStyle.insert({Delimiter, *LanguageStyle}); 203 for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions) 204 EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle}); 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 } 254 State.Line = Line; 255 State.NextToken = Line->First; 256 State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent, 257 /*AvoidBinPacking=*/false, 258 /*NoLineBreak=*/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 auto &CurrentState = State.Stack.back(); 269 CurrentState.AvoidBinPacking = true; 270 CurrentState.BreakBeforeParameter = true; 271 CurrentState.AlignColons = false; 272 } 273 274 // The first token has already been indented and thus consumed. 275 moveStateToNextToken(State, DryRun, /*Newline=*/false); 276 return State; 277 } 278 279 bool ContinuationIndenter::canBreak(const LineState &State) { 280 const FormatToken &Current = *State.NextToken; 281 const FormatToken &Previous = *Current.Previous; 282 const auto &CurrentState = State.Stack.back(); 283 assert(&Previous == Current.Previous); 284 if (!Current.CanBreakBefore && !(CurrentState.BreakBeforeClosingBrace && 285 Current.closesBlockOrBlockTypeList(Style))) { 286 return false; 287 } 288 // The opening "{" of a braced list has to be on the same line as the first 289 // element if it is nested in another braced init list or function call. 290 if (!Current.MustBreakBefore && Previous.is(tok::l_brace) && 291 Previous.isNot(TT_DictLiteral) && Previous.is(BK_BracedInit) && 292 Previous.Previous && 293 Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma)) { 294 return false; 295 } 296 // This prevents breaks like: 297 // ... 298 // SomeParameter, OtherParameter).DoSomething( 299 // ... 300 // As they hide "DoSomething" and are generally bad for readability. 301 if (Previous.opensScope() && Previous.isNot(tok::l_brace) && 302 State.LowestLevelOnLine < State.StartOfLineLevel && 303 State.LowestLevelOnLine < Current.NestingLevel) { 304 return false; 305 } 306 if (Current.isMemberAccess() && CurrentState.ContainsUnwrappedBuilder) 307 return false; 308 309 // Don't create a 'hanging' indent if there are multiple blocks in a single 310 // statement. 311 if (Previous.is(tok::l_brace) && State.Stack.size() > 1 && 312 State.Stack[State.Stack.size() - 2].NestedBlockInlined && 313 State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks) { 314 return false; 315 } 316 317 // Don't break after very short return types (e.g. "void") as that is often 318 // unexpected. 319 if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) { 320 if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None) 321 return false; 322 } 323 324 // If binary operators are moved to the next line (including commas for some 325 // styles of constructor initializers), that's always ok. 326 if (!Current.isOneOf(TT_BinaryOperator, tok::comma) && 327 CurrentState.NoLineBreakInOperand) { 328 return false; 329 } 330 331 if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr)) 332 return false; 333 334 return !CurrentState.NoLineBreak; 335 } 336 337 bool ContinuationIndenter::mustBreak(const LineState &State) { 338 const FormatToken &Current = *State.NextToken; 339 const FormatToken &Previous = *Current.Previous; 340 const auto &CurrentState = State.Stack.back(); 341 if (Style.BraceWrapping.BeforeLambdaBody && Current.CanBreakBefore && 342 Current.is(TT_LambdaLBrace) && Previous.isNot(TT_LineComment)) { 343 auto LambdaBodyLength = getLengthToMatchingParen(Current, State.Stack); 344 return LambdaBodyLength > getColumnLimit(State); 345 } 346 if (Current.MustBreakBefore || Current.is(TT_InlineASMColon)) 347 return true; 348 if (CurrentState.BreakBeforeClosingBrace && 349 Current.closesBlockOrBlockTypeList(Style)) { 350 return true; 351 } 352 if (CurrentState.BreakBeforeClosingParen && Current.is(tok::r_paren)) 353 return true; 354 if (Style.Language == FormatStyle::LK_ObjC && 355 Style.ObjCBreakBeforeNestedBlockParam && 356 Current.ObjCSelectorNameParts > 1 && 357 Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) { 358 return true; 359 } 360 // Avoid producing inconsistent states by requiring breaks where they are not 361 // permitted for C# generic type constraints. 362 if (CurrentState.IsCSharpGenericTypeConstraint && 363 Previous.isNot(TT_CSharpGenericTypeConstraintComma)) { 364 return false; 365 } 366 if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) || 367 (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) && 368 Style.isCpp() && 369 // FIXME: This is a temporary workaround for the case where clang-format 370 // sets BreakBeforeParameter to avoid bin packing and this creates a 371 // completely unnecessary line break after a template type that isn't 372 // line-wrapped. 373 (Previous.NestingLevel == 1 || Style.BinPackParameters)) || 374 (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) && 375 Previous.isNot(tok::question)) || 376 (!Style.BreakBeforeTernaryOperators && 377 Previous.is(TT_ConditionalExpr))) && 378 CurrentState.BreakBeforeParameter && !Current.isTrailingComment() && 379 !Current.isOneOf(tok::r_paren, tok::r_brace)) { 380 return true; 381 } 382 if (CurrentState.IsChainedConditional && 383 ((Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) && 384 Current.is(tok::colon)) || 385 (!Style.BreakBeforeTernaryOperators && Previous.is(TT_ConditionalExpr) && 386 Previous.is(tok::colon)))) { 387 return true; 388 } 389 if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) || 390 (Previous.is(TT_ArrayInitializerLSquare) && 391 Previous.ParameterCount > 1) || 392 opensProtoMessageField(Previous, Style)) && 393 Style.ColumnLimit > 0 && 394 getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 > 395 getColumnLimit(State)) { 396 return true; 397 } 398 399 const FormatToken &BreakConstructorInitializersToken = 400 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon 401 ? Previous 402 : Current; 403 if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) && 404 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength > 405 getColumnLimit(State) || 406 CurrentState.BreakBeforeParameter) && 407 (!Current.isTrailingComment() || Current.NewlinesBefore > 0) && 408 (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All || 409 Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon || 410 Style.ColumnLimit != 0)) { 411 return true; 412 } 413 414 if (Current.is(TT_ObjCMethodExpr) && !Previous.is(TT_SelectorName) && 415 State.Line->startsWith(TT_ObjCMethodSpecifier)) { 416 return true; 417 } 418 if (Current.is(TT_SelectorName) && !Previous.is(tok::at) && 419 CurrentState.ObjCSelectorNameFound && CurrentState.BreakBeforeParameter && 420 (Style.ObjCBreakBeforeNestedBlockParam || 421 !Current.startsSequence(TT_SelectorName, tok::colon, tok::caret))) { 422 return true; 423 } 424 425 unsigned NewLineColumn = getNewLineColumn(State); 426 if (Current.isMemberAccess() && Style.ColumnLimit != 0 && 427 State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit && 428 (State.Column > NewLineColumn || 429 Current.NestingLevel < State.StartOfLineLevel)) { 430 return true; 431 } 432 433 if (startsSegmentOfBuilderTypeCall(Current) && 434 (CurrentState.CallContinuation != 0 || 435 CurrentState.BreakBeforeParameter) && 436 // JavaScript is treated different here as there is a frequent pattern: 437 // SomeFunction(function() { 438 // ... 439 // }.bind(...)); 440 // FIXME: We should find a more generic solution to this problem. 441 !(State.Column <= NewLineColumn && Style.isJavaScript()) && 442 !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) { 443 return true; 444 } 445 446 // If the template declaration spans multiple lines, force wrap before the 447 // function/class declaration 448 if (Previous.ClosesTemplateDeclaration && CurrentState.BreakBeforeParameter && 449 Current.CanBreakBefore) { 450 return true; 451 } 452 453 if (!State.Line->First->is(tok::kw_enum) && State.Column <= NewLineColumn) 454 return false; 455 456 if (Style.AlwaysBreakBeforeMultilineStrings && 457 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth || 458 Previous.is(tok::comma) || Current.NestingLevel < 2) && 459 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at, 460 Keywords.kw_dollar) && 461 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) && 462 nextIsMultilineString(State)) { 463 return true; 464 } 465 466 // Using CanBreakBefore here and below takes care of the decision whether the 467 // current style uses wrapping before or after operators for the given 468 // operator. 469 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) { 470 const auto PreviousPrecedence = Previous.getPrecedence(); 471 if (PreviousPrecedence != prec::Assignment && 472 CurrentState.BreakBeforeParameter && !Current.isTrailingComment()) { 473 const bool LHSIsBinaryExpr = 474 Previous.Previous && Previous.Previous->EndsBinaryExpression; 475 if (LHSIsBinaryExpr) 476 return true; 477 // If we need to break somewhere inside the LHS of a binary expression, we 478 // should also break after the operator. Otherwise, the formatting would 479 // hide the operator precedence, e.g. in: 480 // if (aaaaaaaaaaaaaa == 481 // bbbbbbbbbbbbbb && c) {.. 482 // For comparisons, we only apply this rule, if the LHS is a binary 483 // expression itself as otherwise, the line breaks seem superfluous. 484 // We need special cases for ">>" which we have split into two ">" while 485 // lexing in order to make template parsing easier. 486 const bool IsComparison = 487 (PreviousPrecedence == prec::Relational || 488 PreviousPrecedence == prec::Equality || 489 PreviousPrecedence == prec::Spaceship) && 490 Previous.Previous && 491 Previous.Previous->isNot(TT_BinaryOperator); // For >>. 492 if (!IsComparison) 493 return true; 494 } 495 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore && 496 CurrentState.BreakBeforeParameter) { 497 return true; 498 } 499 500 // Same as above, but for the first "<<" operator. 501 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) && 502 CurrentState.BreakBeforeParameter && CurrentState.FirstLessLess == 0) { 503 return true; 504 } 505 506 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) { 507 // Always break after "template <...>"(*) and leading annotations. This is 508 // only for cases where the entire line does not fit on a single line as a 509 // different LineFormatter would be used otherwise. 510 // *: Except when another option interferes with that, like concepts. 511 if (Previous.ClosesTemplateDeclaration) { 512 if (Current.is(tok::kw_concept)) { 513 switch (Style.BreakBeforeConceptDeclarations) { 514 case FormatStyle::BBCDS_Allowed: 515 break; 516 case FormatStyle::BBCDS_Always: 517 return true; 518 case FormatStyle::BBCDS_Never: 519 return false; 520 } 521 } 522 if (Current.is(TT_RequiresClause)) { 523 switch (Style.RequiresClausePosition) { 524 case FormatStyle::RCPS_SingleLine: 525 case FormatStyle::RCPS_WithPreceding: 526 return false; 527 default: 528 return true; 529 } 530 } 531 return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No; 532 } 533 if (Previous.is(TT_FunctionAnnotationRParen) && 534 State.Line->Type != LT_PreprocessorDirective) { 535 return true; 536 } 537 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) && 538 Current.isNot(TT_LeadingJavaAnnotation)) { 539 return true; 540 } 541 } 542 543 if (Style.isJavaScript() && Previous.is(tok::r_paren) && 544 Previous.is(TT_JavaAnnotation)) { 545 // Break after the closing parenthesis of TypeScript decorators before 546 // functions, getters and setters. 547 static const llvm::StringSet<> BreakBeforeDecoratedTokens = {"get", "set", 548 "function"}; 549 if (BreakBeforeDecoratedTokens.contains(Current.TokenText)) 550 return true; 551 } 552 553 // If the return type spans multiple lines, wrap before the function name. 554 if (((Current.is(TT_FunctionDeclarationName) && 555 // Don't break before a C# function when no break after return type 556 (!Style.isCSharp() || 557 Style.AlwaysBreakAfterReturnType != FormatStyle::RTBS_None) && 558 // Don't always break between a JavaScript `function` and the function 559 // name. 560 !Style.isJavaScript()) || 561 (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) && 562 !Previous.is(tok::kw_template) && CurrentState.BreakBeforeParameter) { 563 return true; 564 } 565 566 // The following could be precomputed as they do not depend on the state. 567 // However, as they should take effect only if the UnwrappedLine does not fit 568 // into the ColumnLimit, they are checked here in the ContinuationIndenter. 569 if (Style.ColumnLimit != 0 && Previous.is(BK_Block) && 570 Previous.is(tok::l_brace) && 571 !Current.isOneOf(tok::r_brace, tok::comment)) { 572 return true; 573 } 574 575 if (Current.is(tok::lessless) && 576 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") || 577 (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") || 578 Previous.TokenText == "\'\\n\'")))) { 579 return true; 580 } 581 582 if (Previous.is(TT_BlockComment) && Previous.IsMultiline) 583 return true; 584 585 if (State.NoContinuation) 586 return true; 587 588 return false; 589 } 590 591 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline, 592 bool DryRun, 593 unsigned ExtraSpaces) { 594 const FormatToken &Current = *State.NextToken; 595 assert(State.NextToken->Previous); 596 const FormatToken &Previous = *State.NextToken->Previous; 597 598 assert(!State.Stack.empty()); 599 State.NoContinuation = false; 600 601 if ((Current.is(TT_ImplicitStringLiteral) && 602 (Previous.Tok.getIdentifierInfo() == nullptr || 603 Previous.Tok.getIdentifierInfo()->getPPKeywordID() == 604 tok::pp_not_keyword))) { 605 unsigned EndColumn = 606 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd()); 607 if (Current.LastNewlineOffset != 0) { 608 // If there is a newline within this token, the final column will solely 609 // determined by the current end column. 610 State.Column = EndColumn; 611 } else { 612 unsigned StartColumn = 613 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin()); 614 assert(EndColumn >= StartColumn); 615 State.Column += EndColumn - StartColumn; 616 } 617 moveStateToNextToken(State, DryRun, /*Newline=*/false); 618 return 0; 619 } 620 621 unsigned Penalty = 0; 622 if (Newline) 623 Penalty = addTokenOnNewLine(State, DryRun); 624 else 625 addTokenOnCurrentLine(State, DryRun, ExtraSpaces); 626 627 return moveStateToNextToken(State, DryRun, Newline) + Penalty; 628 } 629 630 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun, 631 unsigned ExtraSpaces) { 632 FormatToken &Current = *State.NextToken; 633 assert(State.NextToken->Previous); 634 const FormatToken &Previous = *State.NextToken->Previous; 635 auto &CurrentState = State.Stack.back(); 636 637 if (Current.is(tok::equal) && 638 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) && 639 CurrentState.VariablePos == 0) { 640 CurrentState.VariablePos = State.Column; 641 // Move over * and & if they are bound to the variable name. 642 const FormatToken *Tok = &Previous; 643 while (Tok && CurrentState.VariablePos >= Tok->ColumnWidth) { 644 CurrentState.VariablePos -= Tok->ColumnWidth; 645 if (Tok->SpacesRequiredBefore != 0) 646 break; 647 Tok = Tok->Previous; 648 } 649 if (Previous.PartOfMultiVariableDeclStmt) 650 CurrentState.LastSpace = CurrentState.VariablePos; 651 } 652 653 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces; 654 655 // Indent preprocessor directives after the hash if required. 656 int PPColumnCorrection = 0; 657 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 658 Previous.is(tok::hash) && State.FirstIndent > 0 && 659 (State.Line->Type == LT_PreprocessorDirective || 660 State.Line->Type == LT_ImportStatement)) { 661 Spaces += State.FirstIndent; 662 663 // For preprocessor indent with tabs, State.Column will be 1 because of the 664 // hash. This causes second-level indents onward to have an extra space 665 // after the tabs. We avoid this misalignment by subtracting 1 from the 666 // column value passed to replaceWhitespace(). 667 if (Style.UseTab != FormatStyle::UT_Never) 668 PPColumnCorrection = -1; 669 } 670 671 if (!DryRun) { 672 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces, 673 State.Column + Spaces + PPColumnCorrection); 674 } 675 676 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance 677 // declaration unless there is multiple inheritance. 678 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 679 Current.is(TT_InheritanceColon)) { 680 CurrentState.NoLineBreak = true; 681 } 682 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon && 683 Previous.is(TT_InheritanceColon)) { 684 CurrentState.NoLineBreak = true; 685 } 686 687 if (Current.is(TT_SelectorName) && !CurrentState.ObjCSelectorNameFound) { 688 unsigned MinIndent = std::max( 689 State.FirstIndent + Style.ContinuationIndentWidth, CurrentState.Indent); 690 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth; 691 if (Current.LongestObjCSelectorName == 0) 692 CurrentState.AlignColons = false; 693 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos) 694 CurrentState.ColonPos = MinIndent + Current.LongestObjCSelectorName; 695 else 696 CurrentState.ColonPos = FirstColonPos; 697 } 698 699 // In "AlwaysBreak" or "BlockIndent" mode, enforce wrapping directly after the 700 // parenthesis by disallowing any further line breaks if there is no line 701 // break after the opening parenthesis. Don't break if it doesn't conserve 702 // columns. 703 if ((Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak || 704 Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent) && 705 (Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) || 706 (Previous.is(tok::l_brace) && Previous.isNot(BK_Block) && 707 Style.Cpp11BracedListStyle)) && 708 State.Column > getNewLineColumn(State) && 709 (!Previous.Previous || !Previous.Previous->isOneOf( 710 tok::kw_for, tok::kw_while, tok::kw_switch)) && 711 // Don't do this for simple (no expressions) one-argument function calls 712 // as that feels like needlessly wasting whitespace, e.g.: 713 // 714 // caaaaaaaaaaaall( 715 // caaaaaaaaaaaall( 716 // caaaaaaaaaaaall( 717 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa)))); 718 Current.FakeLParens.size() > 0 && 719 Current.FakeLParens.back() > prec::Unknown) { 720 CurrentState.NoLineBreak = true; 721 } 722 if (Previous.is(TT_TemplateString) && Previous.opensScope()) 723 CurrentState.NoLineBreak = true; 724 725 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign && 726 !CurrentState.IsCSharpGenericTypeConstraint && Previous.opensScope() && 727 Previous.isNot(TT_ObjCMethodExpr) && Previous.isNot(TT_RequiresClause) && 728 (Current.isNot(TT_LineComment) || Previous.is(BK_BracedInit))) { 729 CurrentState.Indent = State.Column + Spaces; 730 CurrentState.IsAligned = true; 731 } 732 if (CurrentState.AvoidBinPacking && startsNextParameter(Current, Style)) 733 CurrentState.NoLineBreak = true; 734 if (startsSegmentOfBuilderTypeCall(Current) && 735 State.Column > getNewLineColumn(State)) { 736 CurrentState.ContainsUnwrappedBuilder = true; 737 } 738 739 if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java) 740 CurrentState.NoLineBreak = true; 741 if (Current.isMemberAccess() && Previous.is(tok::r_paren) && 742 (Previous.MatchingParen && 743 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) { 744 // If there is a function call with long parameters, break before trailing 745 // calls. This prevents things like: 746 // EXPECT_CALL(SomeLongParameter).Times( 747 // 2); 748 // We don't want to do this for short parameters as they can just be 749 // indexes. 750 CurrentState.NoLineBreak = true; 751 } 752 753 // Don't allow the RHS of an operator to be split over multiple lines unless 754 // there is a line-break right after the operator. 755 // Exclude relational operators, as there, it is always more desirable to 756 // have the LHS 'left' of the RHS. 757 const FormatToken *P = Current.getPreviousNonComment(); 758 if (!Current.is(tok::comment) && P && 759 (P->isOneOf(TT_BinaryOperator, tok::comma) || 760 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) && 761 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) && 762 P->getPrecedence() != prec::Assignment && 763 P->getPrecedence() != prec::Relational && 764 P->getPrecedence() != prec::Spaceship) { 765 bool BreakBeforeOperator = 766 P->MustBreakBefore || P->is(tok::lessless) || 767 (P->is(TT_BinaryOperator) && 768 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) || 769 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators); 770 // Don't do this if there are only two operands. In these cases, there is 771 // always a nice vertical separation between them and the extra line break 772 // does not help. 773 bool HasTwoOperands = 774 P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr); 775 if ((!BreakBeforeOperator && 776 !(HasTwoOperands && 777 Style.AlignOperands != FormatStyle::OAS_DontAlign)) || 778 (!CurrentState.LastOperatorWrapped && BreakBeforeOperator)) { 779 CurrentState.NoLineBreakInOperand = true; 780 } 781 } 782 783 State.Column += Spaces; 784 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) && 785 Previous.Previous && 786 (Previous.Previous->is(tok::kw_for) || Previous.Previous->isIf())) { 787 // Treat the condition inside an if as if it was a second function 788 // parameter, i.e. let nested calls have a continuation indent. 789 CurrentState.LastSpace = State.Column; 790 CurrentState.NestedBlockIndent = State.Column; 791 } else if (!Current.isOneOf(tok::comment, tok::caret) && 792 ((Previous.is(tok::comma) && 793 !Previous.is(TT_OverloadedOperator)) || 794 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) { 795 CurrentState.LastSpace = State.Column; 796 } else if (Previous.is(TT_CtorInitializerColon) && 797 (!Current.isTrailingComment() || Current.NewlinesBefore > 0) && 798 Style.BreakConstructorInitializers == 799 FormatStyle::BCIS_AfterColon) { 800 CurrentState.Indent = State.Column; 801 CurrentState.LastSpace = State.Column; 802 } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 803 TT_CtorInitializerColon)) && 804 ((Previous.getPrecedence() != prec::Assignment && 805 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 || 806 Previous.NextOperator)) || 807 Current.StartsBinaryExpression)) { 808 // Indent relative to the RHS of the expression unless this is a simple 809 // assignment without binary expression on the RHS. Also indent relative to 810 // unary operators and the colons of constructor initializers. 811 if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None) 812 CurrentState.LastSpace = State.Column; 813 } else if (Previous.is(TT_InheritanceColon)) { 814 CurrentState.Indent = State.Column; 815 CurrentState.LastSpace = State.Column; 816 } else if (Current.is(TT_CSharpGenericTypeConstraintColon)) { 817 CurrentState.ColonPos = State.Column; 818 } else if (Previous.opensScope()) { 819 // If a function has a trailing call, indent all parameters from the 820 // opening parenthesis. This avoids confusing indents like: 821 // OuterFunction(InnerFunctionCall( // break 822 // ParameterToInnerFunction)) // break 823 // .SecondInnerFunctionCall(); 824 if (Previous.MatchingParen) { 825 const FormatToken *Next = Previous.MatchingParen->getNextNonComment(); 826 if (Next && Next->isMemberAccess() && State.Stack.size() > 1 && 827 State.Stack[State.Stack.size() - 2].CallContinuation == 0) { 828 CurrentState.LastSpace = State.Column; 829 } 830 } 831 } 832 } 833 834 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State, 835 bool DryRun) { 836 FormatToken &Current = *State.NextToken; 837 assert(State.NextToken->Previous); 838 const FormatToken &Previous = *State.NextToken->Previous; 839 auto &CurrentState = State.Stack.back(); 840 841 // Extra penalty that needs to be added because of the way certain line 842 // breaks are chosen. 843 unsigned Penalty = 0; 844 845 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 846 const FormatToken *NextNonComment = Previous.getNextNonComment(); 847 if (!NextNonComment) 848 NextNonComment = &Current; 849 // The first line break on any NestingLevel causes an extra penalty in order 850 // prefer similar line breaks. 851 if (!CurrentState.ContainsLineBreak) 852 Penalty += 15; 853 CurrentState.ContainsLineBreak = true; 854 855 Penalty += State.NextToken->SplitPenalty; 856 857 // Breaking before the first "<<" is generally not desirable if the LHS is 858 // short. Also always add the penalty if the LHS is split over multiple lines 859 // to avoid unnecessary line breaks that just work around this penalty. 860 if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess == 0 && 861 (State.Column <= Style.ColumnLimit / 3 || 862 CurrentState.BreakBeforeParameter)) { 863 Penalty += Style.PenaltyBreakFirstLessLess; 864 } 865 866 State.Column = getNewLineColumn(State); 867 868 // Add Penalty proportional to amount of whitespace away from FirstColumn 869 // This tends to penalize several lines that are far-right indented, 870 // and prefers a line-break prior to such a block, e.g: 871 // 872 // Constructor() : 873 // member(value), looooooooooooooooong_member( 874 // looooooooooong_call(param_1, param_2, param_3)) 875 // would then become 876 // Constructor() : 877 // member(value), 878 // looooooooooooooooong_member( 879 // looooooooooong_call(param_1, param_2, param_3)) 880 if (State.Column > State.FirstIndent) { 881 Penalty += 882 Style.PenaltyIndentedWhitespace * (State.Column - State.FirstIndent); 883 } 884 885 // Indent nested blocks relative to this column, unless in a very specific 886 // JavaScript special case where: 887 // 888 // var loooooong_name = 889 // function() { 890 // // code 891 // } 892 // 893 // is common and should be formatted like a free-standing function. The same 894 // goes for wrapping before the lambda return type arrow. 895 if (!Current.is(TT_LambdaArrow) && 896 (!Style.isJavaScript() || Current.NestingLevel != 0 || 897 !PreviousNonComment || !PreviousNonComment->is(tok::equal) || 898 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) { 899 CurrentState.NestedBlockIndent = State.Column; 900 } 901 902 if (NextNonComment->isMemberAccess()) { 903 if (CurrentState.CallContinuation == 0) 904 CurrentState.CallContinuation = State.Column; 905 } else if (NextNonComment->is(TT_SelectorName)) { 906 if (!CurrentState.ObjCSelectorNameFound) { 907 if (NextNonComment->LongestObjCSelectorName == 0) { 908 CurrentState.AlignColons = false; 909 } else { 910 CurrentState.ColonPos = 911 (shouldIndentWrappedSelectorName(Style, State.Line->Type) 912 ? std::max(CurrentState.Indent, 913 State.FirstIndent + Style.ContinuationIndentWidth) 914 : CurrentState.Indent) + 915 std::max(NextNonComment->LongestObjCSelectorName, 916 NextNonComment->ColumnWidth); 917 } 918 } else if (CurrentState.AlignColons && 919 CurrentState.ColonPos <= NextNonComment->ColumnWidth) { 920 CurrentState.ColonPos = State.Column + NextNonComment->ColumnWidth; 921 } 922 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 923 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) { 924 // FIXME: This is hacky, find a better way. The problem is that in an ObjC 925 // method expression, the block should be aligned to the line starting it, 926 // e.g.: 927 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason 928 // ^(int *i) { 929 // // ... 930 // }]; 931 // Thus, we set LastSpace of the next higher NestingLevel, to which we move 932 // when we consume all of the "}"'s FakeRParens at the "{". 933 if (State.Stack.size() > 1) { 934 State.Stack[State.Stack.size() - 2].LastSpace = 935 std::max(CurrentState.LastSpace, CurrentState.Indent) + 936 Style.ContinuationIndentWidth; 937 } 938 } 939 940 if ((PreviousNonComment && 941 PreviousNonComment->isOneOf(tok::comma, tok::semi) && 942 !CurrentState.AvoidBinPacking) || 943 Previous.is(TT_BinaryOperator)) { 944 CurrentState.BreakBeforeParameter = false; 945 } 946 if (PreviousNonComment && 947 (PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) || 948 PreviousNonComment->ClosesRequiresClause) && 949 Current.NestingLevel == 0) { 950 CurrentState.BreakBeforeParameter = false; 951 } 952 if (NextNonComment->is(tok::question) || 953 (PreviousNonComment && PreviousNonComment->is(tok::question))) { 954 CurrentState.BreakBeforeParameter = true; 955 } 956 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore) 957 CurrentState.BreakBeforeParameter = false; 958 959 if (!DryRun) { 960 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1; 961 if (Current.is(tok::r_brace) && Current.MatchingParen && 962 // Only strip trailing empty lines for l_braces that have children, i.e. 963 // for function expressions (lambdas, arrows, etc). 964 !Current.MatchingParen->Children.empty()) { 965 // lambdas and arrow functions are expressions, thus their r_brace is not 966 // on its own line, and thus not covered by UnwrappedLineFormatter's logic 967 // about removing empty lines on closing blocks. Special case them here. 968 MaxEmptyLinesToKeep = 1; 969 } 970 unsigned Newlines = 971 std::max(1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep)); 972 bool ContinuePPDirective = 973 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement; 974 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column, 975 CurrentState.IsAligned, ContinuePPDirective); 976 } 977 978 if (!Current.isTrailingComment()) 979 CurrentState.LastSpace = State.Column; 980 if (Current.is(tok::lessless)) { 981 // If we are breaking before a "<<", we always want to indent relative to 982 // RHS. This is necessary only for "<<", as we special-case it and don't 983 // always indent relative to the RHS. 984 CurrentState.LastSpace += 3; // 3 -> width of "<< ". 985 } 986 987 State.StartOfLineLevel = Current.NestingLevel; 988 State.LowestLevelOnLine = Current.NestingLevel; 989 990 // Any break on this level means that the parent level has been broken 991 // and we need to avoid bin packing there. 992 bool NestedBlockSpecialCase = 993 (!Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 && 994 State.Stack[State.Stack.size() - 2].NestedBlockInlined) || 995 (Style.Language == FormatStyle::LK_ObjC && Current.is(tok::r_brace) && 996 State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam); 997 // Do not force parameter break for statements with requires expressions. 998 NestedBlockSpecialCase = 999 NestedBlockSpecialCase || 1000 (Current.MatchingParen && 1001 Current.MatchingParen->is(TT_RequiresExpressionLBrace)); 1002 if (!NestedBlockSpecialCase) 1003 for (ParenState &PState : llvm::drop_end(State.Stack)) 1004 PState.BreakBeforeParameter = true; 1005 1006 if (PreviousNonComment && 1007 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) && 1008 ((PreviousNonComment->isNot(TT_TemplateCloser) && 1009 !PreviousNonComment->ClosesRequiresClause) || 1010 Current.NestingLevel != 0) && 1011 !PreviousNonComment->isOneOf( 1012 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation, 1013 TT_LeadingJavaAnnotation) && 1014 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope()) { 1015 CurrentState.BreakBeforeParameter = true; 1016 } 1017 1018 // If we break after { or the [ of an array initializer, we should also break 1019 // before the corresponding } or ]. 1020 if (PreviousNonComment && 1021 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 1022 opensProtoMessageField(*PreviousNonComment, Style))) { 1023 CurrentState.BreakBeforeClosingBrace = true; 1024 } 1025 1026 if (PreviousNonComment && PreviousNonComment->is(tok::l_paren)) { 1027 CurrentState.BreakBeforeClosingParen = 1028 Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent; 1029 } 1030 1031 if (CurrentState.AvoidBinPacking) { 1032 // If we are breaking after '(', '{', '<', or this is the break after a ':' 1033 // to start a member initializater list in a constructor, this should not 1034 // be considered bin packing unless the relevant AllowAll option is false or 1035 // this is a dict/object literal. 1036 bool PreviousIsBreakingCtorInitializerColon = 1037 PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && 1038 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 1039 if (!(Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) || 1040 PreviousIsBreakingCtorInitializerColon) || 1041 (!Style.AllowAllParametersOfDeclarationOnNextLine && 1042 State.Line->MustBeDeclaration) || 1043 (!Style.AllowAllArgumentsOnNextLine && 1044 !State.Line->MustBeDeclaration) || 1045 (Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine && 1046 PreviousIsBreakingCtorInitializerColon) || 1047 Previous.is(TT_DictLiteral)) { 1048 CurrentState.BreakBeforeParameter = true; 1049 } 1050 1051 // If we are breaking after a ':' to start a member initializer list, 1052 // and we allow all arguments on the next line, we should not break 1053 // before the next parameter. 1054 if (PreviousIsBreakingCtorInitializerColon && 1055 Style.PackConstructorInitializers == FormatStyle::PCIS_NextLine) { 1056 CurrentState.BreakBeforeParameter = false; 1057 } 1058 } 1059 1060 return Penalty; 1061 } 1062 1063 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) { 1064 if (!State.NextToken || !State.NextToken->Previous) 1065 return 0; 1066 1067 FormatToken &Current = *State.NextToken; 1068 const auto &CurrentState = State.Stack.back(); 1069 1070 if (CurrentState.IsCSharpGenericTypeConstraint && 1071 Current.isNot(TT_CSharpGenericTypeConstraint)) { 1072 return CurrentState.ColonPos + 2; 1073 } 1074 1075 const FormatToken &Previous = *Current.Previous; 1076 // If we are continuing an expression, we want to use the continuation indent. 1077 unsigned ContinuationIndent = 1078 std::max(CurrentState.LastSpace, CurrentState.Indent) + 1079 Style.ContinuationIndentWidth; 1080 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 1081 const FormatToken *NextNonComment = Previous.getNextNonComment(); 1082 if (!NextNonComment) 1083 NextNonComment = &Current; 1084 1085 // Java specific bits. 1086 if (Style.Language == FormatStyle::LK_Java && 1087 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) { 1088 return std::max(CurrentState.LastSpace, 1089 CurrentState.Indent + Style.ContinuationIndentWidth); 1090 } 1091 1092 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths && 1093 State.Line->First->is(tok::kw_enum)) { 1094 return (Style.IndentWidth * State.Line->First->IndentLevel) + 1095 Style.IndentWidth; 1096 } 1097 1098 if (NextNonComment->is(tok::l_brace) && NextNonComment->is(BK_Block)) 1099 return Current.NestingLevel == 0 ? State.FirstIndent : CurrentState.Indent; 1100 if ((Current.isOneOf(tok::r_brace, tok::r_square) || 1101 (Current.is(tok::greater) && 1102 (Style.Language == FormatStyle::LK_Proto || 1103 Style.Language == FormatStyle::LK_TextProto))) && 1104 State.Stack.size() > 1) { 1105 if (Current.closesBlockOrBlockTypeList(Style)) 1106 return State.Stack[State.Stack.size() - 2].NestedBlockIndent; 1107 if (Current.MatchingParen && Current.MatchingParen->is(BK_BracedInit)) 1108 return State.Stack[State.Stack.size() - 2].LastSpace; 1109 return State.FirstIndent; 1110 } 1111 // Indent a closing parenthesis at the previous level if followed by a semi, 1112 // const, or opening brace. This allows indentations such as: 1113 // foo( 1114 // a, 1115 // ); 1116 // int Foo::getter( 1117 // // 1118 // ) const { 1119 // return foo; 1120 // } 1121 // function foo( 1122 // a, 1123 // ) { 1124 // code(); // 1125 // } 1126 if (Current.is(tok::r_paren) && State.Stack.size() > 1 && 1127 (!Current.Next || 1128 Current.Next->isOneOf(tok::semi, tok::kw_const, tok::l_brace))) { 1129 return State.Stack[State.Stack.size() - 2].LastSpace; 1130 } 1131 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent && 1132 Current.is(tok::r_paren) && State.Stack.size() > 1) { 1133 return State.Stack[State.Stack.size() - 2].LastSpace; 1134 } 1135 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope()) 1136 return State.Stack[State.Stack.size() - 2].LastSpace; 1137 if (Current.is(tok::identifier) && Current.Next && 1138 (Current.Next->is(TT_DictLiteral) || 1139 ((Style.Language == FormatStyle::LK_Proto || 1140 Style.Language == FormatStyle::LK_TextProto) && 1141 Current.Next->isOneOf(tok::less, tok::l_brace)))) { 1142 return CurrentState.Indent; 1143 } 1144 if (NextNonComment->is(TT_ObjCStringLiteral) && 1145 State.StartOfStringLiteral != 0) { 1146 return State.StartOfStringLiteral - 1; 1147 } 1148 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0) 1149 return State.StartOfStringLiteral; 1150 if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess != 0) 1151 return CurrentState.FirstLessLess; 1152 if (NextNonComment->isMemberAccess()) { 1153 if (CurrentState.CallContinuation == 0) 1154 return ContinuationIndent; 1155 return CurrentState.CallContinuation; 1156 } 1157 if (CurrentState.QuestionColumn != 0 && 1158 ((NextNonComment->is(tok::colon) && 1159 NextNonComment->is(TT_ConditionalExpr)) || 1160 Previous.is(TT_ConditionalExpr))) { 1161 if (((NextNonComment->is(tok::colon) && NextNonComment->Next && 1162 !NextNonComment->Next->FakeLParens.empty() && 1163 NextNonComment->Next->FakeLParens.back() == prec::Conditional) || 1164 (Previous.is(tok::colon) && !Current.FakeLParens.empty() && 1165 Current.FakeLParens.back() == prec::Conditional)) && 1166 !CurrentState.IsWrappedConditional) { 1167 // NOTE: we may tweak this slightly: 1168 // * not remove the 'lead' ContinuationIndentWidth 1169 // * always un-indent by the operator when 1170 // BreakBeforeTernaryOperators=true 1171 unsigned Indent = CurrentState.Indent; 1172 if (Style.AlignOperands != FormatStyle::OAS_DontAlign) 1173 Indent -= Style.ContinuationIndentWidth; 1174 if (Style.BreakBeforeTernaryOperators && CurrentState.UnindentOperator) 1175 Indent -= 2; 1176 return Indent; 1177 } 1178 return CurrentState.QuestionColumn; 1179 } 1180 if (Previous.is(tok::comma) && CurrentState.VariablePos != 0) 1181 return CurrentState.VariablePos; 1182 if (Current.is(TT_RequiresClause)) { 1183 if (Style.IndentRequiresClause) 1184 return CurrentState.Indent + Style.IndentWidth; 1185 switch (Style.RequiresClausePosition) { 1186 case FormatStyle::RCPS_OwnLine: 1187 case FormatStyle::RCPS_WithFollowing: 1188 return CurrentState.Indent; 1189 default: 1190 break; 1191 } 1192 } 1193 if ((PreviousNonComment && 1194 (PreviousNonComment->ClosesTemplateDeclaration || 1195 PreviousNonComment->ClosesRequiresClause || 1196 PreviousNonComment->isOneOf( 1197 TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen, 1198 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) || 1199 (!Style.IndentWrappedFunctionNames && 1200 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) { 1201 return std::max(CurrentState.LastSpace, CurrentState.Indent); 1202 } 1203 if (NextNonComment->is(TT_SelectorName)) { 1204 if (!CurrentState.ObjCSelectorNameFound) { 1205 unsigned MinIndent = CurrentState.Indent; 1206 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) { 1207 MinIndent = std::max(MinIndent, 1208 State.FirstIndent + Style.ContinuationIndentWidth); 1209 } 1210 // If LongestObjCSelectorName is 0, we are indenting the first 1211 // part of an ObjC selector (or a selector component which is 1212 // not colon-aligned due to block formatting). 1213 // 1214 // Otherwise, we are indenting a subsequent part of an ObjC 1215 // selector which should be colon-aligned to the longest 1216 // component of the ObjC selector. 1217 // 1218 // In either case, we want to respect Style.IndentWrappedFunctionNames. 1219 return MinIndent + 1220 std::max(NextNonComment->LongestObjCSelectorName, 1221 NextNonComment->ColumnWidth) - 1222 NextNonComment->ColumnWidth; 1223 } 1224 if (!CurrentState.AlignColons) 1225 return CurrentState.Indent; 1226 if (CurrentState.ColonPos > NextNonComment->ColumnWidth) 1227 return CurrentState.ColonPos - NextNonComment->ColumnWidth; 1228 return CurrentState.Indent; 1229 } 1230 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr)) 1231 return CurrentState.ColonPos; 1232 if (NextNonComment->is(TT_ArraySubscriptLSquare)) { 1233 if (CurrentState.StartOfArraySubscripts != 0) { 1234 return CurrentState.StartOfArraySubscripts; 1235 } else if (Style.isCSharp()) { // C# allows `["key"] = value` inside object 1236 // initializers. 1237 return CurrentState.Indent; 1238 } 1239 return ContinuationIndent; 1240 } 1241 1242 // This ensure that we correctly format ObjC methods calls without inputs, 1243 // i.e. where the last element isn't selector like: [callee method]; 1244 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 && 1245 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) { 1246 return CurrentState.Indent; 1247 } 1248 1249 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) || 1250 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) { 1251 return ContinuationIndent; 1252 } 1253 if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 1254 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) { 1255 return ContinuationIndent; 1256 } 1257 if (NextNonComment->is(TT_CtorInitializerComma)) 1258 return CurrentState.Indent; 1259 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && 1260 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { 1261 return CurrentState.Indent; 1262 } 1263 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) && 1264 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) { 1265 return CurrentState.Indent; 1266 } 1267 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon, 1268 TT_InheritanceComma)) { 1269 return State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1270 } 1271 if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() && 1272 !Current.isOneOf(tok::colon, tok::comment)) { 1273 return ContinuationIndent; 1274 } 1275 if (Current.is(TT_ProtoExtensionLSquare)) 1276 return CurrentState.Indent; 1277 if (Current.isBinaryOperator() && CurrentState.UnindentOperator) { 1278 return CurrentState.Indent - Current.Tok.getLength() - 1279 Current.SpacesRequiredBefore; 1280 } 1281 if (Current.isOneOf(tok::comment, TT_BlockComment, TT_LineComment) && 1282 NextNonComment->isBinaryOperator() && CurrentState.UnindentOperator) { 1283 return CurrentState.Indent - NextNonComment->Tok.getLength() - 1284 NextNonComment->SpacesRequiredBefore; 1285 } 1286 if (CurrentState.Indent == State.FirstIndent && PreviousNonComment && 1287 !PreviousNonComment->isOneOf(tok::r_brace, TT_CtorInitializerComma)) { 1288 // Ensure that we fall back to the continuation indent width instead of 1289 // just flushing continuations left. 1290 return CurrentState.Indent + Style.ContinuationIndentWidth; 1291 } 1292 return CurrentState.Indent; 1293 } 1294 1295 static bool hasNestedBlockInlined(const FormatToken *Previous, 1296 const FormatToken &Current, 1297 const FormatStyle &Style) { 1298 if (Previous->isNot(tok::l_paren)) 1299 return true; 1300 if (Previous->ParameterCount > 1) 1301 return true; 1302 1303 // Also a nested block if contains a lambda inside function with 1 parameter 1304 return Style.BraceWrapping.BeforeLambdaBody && Current.is(TT_LambdaLSquare); 1305 } 1306 1307 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State, 1308 bool DryRun, bool Newline) { 1309 assert(State.Stack.size()); 1310 const FormatToken &Current = *State.NextToken; 1311 auto &CurrentState = State.Stack.back(); 1312 1313 if (Current.is(TT_CSharpGenericTypeConstraint)) 1314 CurrentState.IsCSharpGenericTypeConstraint = true; 1315 if (Current.isOneOf(tok::comma, TT_BinaryOperator)) 1316 CurrentState.NoLineBreakInOperand = false; 1317 if (Current.isOneOf(TT_InheritanceColon, TT_CSharpGenericTypeConstraintColon)) 1318 CurrentState.AvoidBinPacking = true; 1319 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) { 1320 if (CurrentState.FirstLessLess == 0) 1321 CurrentState.FirstLessLess = State.Column; 1322 else 1323 CurrentState.LastOperatorWrapped = Newline; 1324 } 1325 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless)) 1326 CurrentState.LastOperatorWrapped = Newline; 1327 if (Current.is(TT_ConditionalExpr) && Current.Previous && 1328 !Current.Previous->is(TT_ConditionalExpr)) { 1329 CurrentState.LastOperatorWrapped = Newline; 1330 } 1331 if (Current.is(TT_ArraySubscriptLSquare) && 1332 CurrentState.StartOfArraySubscripts == 0) { 1333 CurrentState.StartOfArraySubscripts = State.Column; 1334 } 1335 1336 auto IsWrappedConditional = [](const FormatToken &Tok) { 1337 if (!(Tok.is(TT_ConditionalExpr) && Tok.is(tok::question))) 1338 return false; 1339 if (Tok.MustBreakBefore) 1340 return true; 1341 1342 const FormatToken *Next = Tok.getNextNonComment(); 1343 return Next && Next->MustBreakBefore; 1344 }; 1345 if (IsWrappedConditional(Current)) 1346 CurrentState.IsWrappedConditional = true; 1347 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question)) 1348 CurrentState.QuestionColumn = State.Column; 1349 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) { 1350 const FormatToken *Previous = Current.Previous; 1351 while (Previous && Previous->isTrailingComment()) 1352 Previous = Previous->Previous; 1353 if (Previous && Previous->is(tok::question)) 1354 CurrentState.QuestionColumn = State.Column; 1355 } 1356 if (!Current.opensScope() && !Current.closesScope() && 1357 !Current.is(TT_PointerOrReference)) { 1358 State.LowestLevelOnLine = 1359 std::min(State.LowestLevelOnLine, Current.NestingLevel); 1360 } 1361 if (Current.isMemberAccess()) 1362 CurrentState.StartOfFunctionCall = !Current.NextOperator ? 0 : State.Column; 1363 if (Current.is(TT_SelectorName)) 1364 CurrentState.ObjCSelectorNameFound = true; 1365 if (Current.is(TT_CtorInitializerColon) && 1366 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) { 1367 // Indent 2 from the column, so: 1368 // SomeClass::SomeClass() 1369 // : First(...), ... 1370 // Next(...) 1371 // ^ line up here. 1372 CurrentState.Indent = State.Column + (Style.BreakConstructorInitializers == 1373 FormatStyle::BCIS_BeforeComma 1374 ? 0 1375 : 2); 1376 CurrentState.NestedBlockIndent = CurrentState.Indent; 1377 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) { 1378 CurrentState.AvoidBinPacking = true; 1379 CurrentState.BreakBeforeParameter = 1380 Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine; 1381 } else { 1382 CurrentState.BreakBeforeParameter = false; 1383 } 1384 } 1385 if (Current.is(TT_CtorInitializerColon) && 1386 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { 1387 CurrentState.Indent = 1388 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1389 CurrentState.NestedBlockIndent = CurrentState.Indent; 1390 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) 1391 CurrentState.AvoidBinPacking = true; 1392 } 1393 if (Current.is(TT_InheritanceColon)) { 1394 CurrentState.Indent = 1395 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1396 } 1397 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline) 1398 CurrentState.NestedBlockIndent = State.Column + Current.ColumnWidth + 1; 1399 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow)) 1400 CurrentState.LastSpace = State.Column; 1401 if (Current.is(TT_RequiresExpression)) 1402 CurrentState.NestedBlockIndent = State.Column; 1403 1404 // Insert scopes created by fake parenthesis. 1405 const FormatToken *Previous = Current.getPreviousNonComment(); 1406 1407 // Add special behavior to support a format commonly used for JavaScript 1408 // closures: 1409 // SomeFunction(function() { 1410 // foo(); 1411 // bar(); 1412 // }, a, b, c); 1413 if (Current.isNot(tok::comment) && !Current.ClosesRequiresClause && 1414 Previous && Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) && 1415 !Previous->is(TT_DictLiteral) && State.Stack.size() > 1 && 1416 !CurrentState.HasMultipleNestedBlocks) { 1417 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline) 1418 for (ParenState &PState : llvm::drop_end(State.Stack)) 1419 PState.NoLineBreak = true; 1420 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false; 1421 } 1422 if (Previous && (Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr) || 1423 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) && 1424 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)))) { 1425 CurrentState.NestedBlockInlined = 1426 !Newline && hasNestedBlockInlined(Previous, Current, Style); 1427 } 1428 1429 moveStatePastFakeLParens(State, Newline); 1430 moveStatePastScopeCloser(State); 1431 // Do not use CurrentState here, since the two functions before may change the 1432 // Stack. 1433 bool AllowBreak = !State.Stack.back().NoLineBreak && 1434 !State.Stack.back().NoLineBreakInOperand; 1435 moveStatePastScopeOpener(State, Newline); 1436 moveStatePastFakeRParens(State); 1437 1438 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0) 1439 State.StartOfStringLiteral = State.Column + 1; 1440 if (Current.is(TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) { 1441 State.StartOfStringLiteral = State.Column + 1; 1442 } else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) { 1443 State.StartOfStringLiteral = State.Column; 1444 } else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) && 1445 !Current.isStringLiteral()) { 1446 State.StartOfStringLiteral = 0; 1447 } 1448 1449 State.Column += Current.ColumnWidth; 1450 State.NextToken = State.NextToken->Next; 1451 1452 unsigned Penalty = 1453 handleEndOfLine(Current, State, DryRun, AllowBreak, Newline); 1454 1455 if (Current.Role) 1456 Current.Role->formatFromToken(State, this, DryRun); 1457 // If the previous has a special role, let it consume tokens as appropriate. 1458 // It is necessary to start at the previous token for the only implemented 1459 // role (comma separated list). That way, the decision whether or not to break 1460 // after the "{" is already done and both options are tried and evaluated. 1461 // FIXME: This is ugly, find a better way. 1462 if (Previous && Previous->Role) 1463 Penalty += Previous->Role->formatAfterToken(State, this, DryRun); 1464 1465 return Penalty; 1466 } 1467 1468 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State, 1469 bool Newline) { 1470 const FormatToken &Current = *State.NextToken; 1471 if (Current.FakeLParens.empty()) 1472 return; 1473 1474 const FormatToken *Previous = Current.getPreviousNonComment(); 1475 1476 // Don't add extra indentation for the first fake parenthesis after 1477 // 'return', assignments, opening <({[, or requires clauses. The indentation 1478 // for these cases is special cased. 1479 bool SkipFirstExtraIndent = 1480 Previous && 1481 (Previous->opensScope() || 1482 Previous->isOneOf(tok::semi, tok::kw_return, TT_RequiresClause) || 1483 (Previous->getPrecedence() == prec::Assignment && 1484 Style.AlignOperands != FormatStyle::OAS_DontAlign) || 1485 Previous->is(TT_ObjCMethodExpr)); 1486 for (const auto &PrecedenceLevel : llvm::reverse(Current.FakeLParens)) { 1487 const auto &CurrentState = State.Stack.back(); 1488 ParenState NewParenState = CurrentState; 1489 NewParenState.Tok = nullptr; 1490 NewParenState.ContainsLineBreak = false; 1491 NewParenState.LastOperatorWrapped = true; 1492 NewParenState.IsChainedConditional = false; 1493 NewParenState.IsWrappedConditional = false; 1494 NewParenState.UnindentOperator = false; 1495 NewParenState.NoLineBreak = 1496 NewParenState.NoLineBreak || CurrentState.NoLineBreakInOperand; 1497 1498 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists. 1499 if (PrecedenceLevel > prec::Comma) 1500 NewParenState.AvoidBinPacking = false; 1501 1502 // Indent from 'LastSpace' unless these are fake parentheses encapsulating 1503 // a builder type call after 'return' or, if the alignment after opening 1504 // brackets is disabled. 1505 if (!Current.isTrailingComment() && 1506 (Style.AlignOperands != FormatStyle::OAS_DontAlign || 1507 PrecedenceLevel < prec::Assignment) && 1508 (!Previous || Previous->isNot(tok::kw_return) || 1509 (Style.Language != FormatStyle::LK_Java && PrecedenceLevel > 0)) && 1510 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign || 1511 PrecedenceLevel != prec::Comma || Current.NestingLevel == 0)) { 1512 NewParenState.Indent = std::max( 1513 std::max(State.Column, NewParenState.Indent), CurrentState.LastSpace); 1514 } 1515 1516 if (Previous && 1517 (Previous->getPrecedence() == prec::Assignment || 1518 Previous->isOneOf(tok::kw_return, TT_RequiresClause) || 1519 (PrecedenceLevel == prec::Conditional && Previous->is(tok::question) && 1520 Previous->is(TT_ConditionalExpr))) && 1521 !Newline) { 1522 // If BreakBeforeBinaryOperators is set, un-indent a bit to account for 1523 // the operator and keep the operands aligned 1524 if (Style.AlignOperands == FormatStyle::OAS_AlignAfterOperator) 1525 NewParenState.UnindentOperator = true; 1526 // Mark indentation as alignment if the expression is aligned. 1527 if (Style.AlignOperands != FormatStyle::OAS_DontAlign) 1528 NewParenState.IsAligned = true; 1529 } 1530 1531 // Do not indent relative to the fake parentheses inserted for "." or "->". 1532 // This is a special case to make the following to statements consistent: 1533 // OuterFunction(InnerFunctionCall( // break 1534 // ParameterToInnerFunction)); 1535 // OuterFunction(SomeObject.InnerFunctionCall( // break 1536 // ParameterToInnerFunction)); 1537 if (PrecedenceLevel > prec::Unknown) 1538 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column); 1539 if (PrecedenceLevel != prec::Conditional && !Current.is(TT_UnaryOperator) && 1540 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) { 1541 NewParenState.StartOfFunctionCall = State.Column; 1542 } 1543 1544 // Indent conditional expressions, unless they are chained "else-if" 1545 // conditionals. Never indent expression where the 'operator' is ',', ';' or 1546 // an assignment (i.e. *I <= prec::Assignment) as those have different 1547 // indentation rules. Indent other expression, unless the indentation needs 1548 // to be skipped. 1549 if (PrecedenceLevel == prec::Conditional && Previous && 1550 Previous->is(tok::colon) && Previous->is(TT_ConditionalExpr) && 1551 &PrecedenceLevel == &Current.FakeLParens.back() && 1552 !CurrentState.IsWrappedConditional) { 1553 NewParenState.IsChainedConditional = true; 1554 NewParenState.UnindentOperator = State.Stack.back().UnindentOperator; 1555 } else if (PrecedenceLevel == prec::Conditional || 1556 (!SkipFirstExtraIndent && PrecedenceLevel > prec::Assignment && 1557 !Current.isTrailingComment())) { 1558 NewParenState.Indent += Style.ContinuationIndentWidth; 1559 } 1560 if ((Previous && !Previous->opensScope()) || PrecedenceLevel != prec::Comma) 1561 NewParenState.BreakBeforeParameter = false; 1562 State.Stack.push_back(NewParenState); 1563 SkipFirstExtraIndent = false; 1564 } 1565 } 1566 1567 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) { 1568 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) { 1569 unsigned VariablePos = State.Stack.back().VariablePos; 1570 if (State.Stack.size() == 1) { 1571 // Do not pop the last element. 1572 break; 1573 } 1574 State.Stack.pop_back(); 1575 State.Stack.back().VariablePos = VariablePos; 1576 } 1577 1578 if (State.NextToken->ClosesRequiresClause && Style.IndentRequiresClause) { 1579 // Remove the indentation of the requires clauses (which is not in Indent, 1580 // but in LastSpace). 1581 State.Stack.back().LastSpace -= Style.IndentWidth; 1582 } 1583 } 1584 1585 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State, 1586 bool Newline) { 1587 const FormatToken &Current = *State.NextToken; 1588 if (!Current.opensScope()) 1589 return; 1590 1591 const auto &CurrentState = State.Stack.back(); 1592 1593 // Don't allow '<' or '(' in C# generic type constraints to start new scopes. 1594 if (Current.isOneOf(tok::less, tok::l_paren) && 1595 CurrentState.IsCSharpGenericTypeConstraint) { 1596 return; 1597 } 1598 1599 if (Current.MatchingParen && Current.is(BK_Block)) { 1600 moveStateToNewBlock(State); 1601 return; 1602 } 1603 1604 unsigned NewIndent; 1605 unsigned LastSpace = CurrentState.LastSpace; 1606 bool AvoidBinPacking; 1607 bool BreakBeforeParameter = false; 1608 unsigned NestedBlockIndent = std::max(CurrentState.StartOfFunctionCall, 1609 CurrentState.NestedBlockIndent); 1610 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 1611 opensProtoMessageField(Current, Style)) { 1612 if (Current.opensBlockOrBlockTypeList(Style)) { 1613 NewIndent = Style.IndentWidth + 1614 std::min(State.Column, CurrentState.NestedBlockIndent); 1615 } else { 1616 NewIndent = CurrentState.LastSpace + Style.ContinuationIndentWidth; 1617 } 1618 const FormatToken *NextNoComment = Current.getNextNonComment(); 1619 bool EndsInComma = Current.MatchingParen && 1620 Current.MatchingParen->Previous && 1621 Current.MatchingParen->Previous->is(tok::comma); 1622 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) || 1623 Style.Language == FormatStyle::LK_Proto || 1624 Style.Language == FormatStyle::LK_TextProto || 1625 !Style.BinPackArguments || 1626 (NextNoComment && 1627 NextNoComment->isOneOf(TT_DesignatedInitializerPeriod, 1628 TT_DesignatedInitializerLSquare)); 1629 BreakBeforeParameter = EndsInComma; 1630 if (Current.ParameterCount > 1) 1631 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1); 1632 } else { 1633 NewIndent = 1634 Style.ContinuationIndentWidth + 1635 std::max(CurrentState.LastSpace, CurrentState.StartOfFunctionCall); 1636 1637 // Ensure that different different brackets force relative alignment, e.g.: 1638 // void SomeFunction(vector< // break 1639 // int> v); 1640 // FIXME: We likely want to do this for more combinations of brackets. 1641 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) { 1642 NewIndent = std::max(NewIndent, CurrentState.Indent); 1643 LastSpace = std::max(LastSpace, CurrentState.Indent); 1644 } 1645 1646 bool EndsInComma = 1647 Current.MatchingParen && 1648 Current.MatchingParen->getPreviousNonComment() && 1649 Current.MatchingParen->getPreviousNonComment()->is(tok::comma); 1650 1651 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters 1652 // for backwards compatibility. 1653 bool ObjCBinPackProtocolList = 1654 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto && 1655 Style.BinPackParameters) || 1656 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always; 1657 1658 bool BinPackDeclaration = 1659 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) || 1660 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList); 1661 1662 AvoidBinPacking = 1663 (CurrentState.IsCSharpGenericTypeConstraint) || 1664 (Style.isJavaScript() && EndsInComma) || 1665 (State.Line->MustBeDeclaration && !BinPackDeclaration) || 1666 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) || 1667 (Style.ExperimentalAutoDetectBinPacking && 1668 (Current.is(PPK_OnePerLine) || 1669 (!BinPackInconclusiveFunctions && Current.is(PPK_Inconclusive)))); 1670 1671 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen && 1672 Style.ObjCBreakBeforeNestedBlockParam) { 1673 if (Style.ColumnLimit) { 1674 // If this '[' opens an ObjC call, determine whether all parameters fit 1675 // into one line and put one per line if they don't. 1676 if (getLengthToMatchingParen(Current, State.Stack) + State.Column > 1677 getColumnLimit(State)) { 1678 BreakBeforeParameter = true; 1679 } 1680 } else { 1681 // For ColumnLimit = 0, we have to figure out whether there is or has to 1682 // be a line break within this call. 1683 for (const FormatToken *Tok = &Current; 1684 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) { 1685 if (Tok->MustBreakBefore || 1686 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) { 1687 BreakBeforeParameter = true; 1688 break; 1689 } 1690 } 1691 } 1692 } 1693 1694 if (Style.isJavaScript() && EndsInComma) 1695 BreakBeforeParameter = true; 1696 } 1697 // Generally inherit NoLineBreak from the current scope to nested scope. 1698 // However, don't do this for non-empty nested blocks, dict literals and 1699 // array literals as these follow different indentation rules. 1700 bool NoLineBreak = 1701 Current.Children.empty() && 1702 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) && 1703 (CurrentState.NoLineBreak || CurrentState.NoLineBreakInOperand || 1704 (Current.is(TT_TemplateOpener) && 1705 CurrentState.ContainsUnwrappedBuilder)); 1706 State.Stack.push_back( 1707 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak)); 1708 auto &NewState = State.Stack.back(); 1709 NewState.NestedBlockIndent = NestedBlockIndent; 1710 NewState.BreakBeforeParameter = BreakBeforeParameter; 1711 NewState.HasMultipleNestedBlocks = (Current.BlockParameterCount > 1); 1712 1713 if (Style.BraceWrapping.BeforeLambdaBody && Current.Next != nullptr && 1714 Current.is(tok::l_paren)) { 1715 // Search for any parameter that is a lambda 1716 FormatToken const *next = Current.Next; 1717 while (next != nullptr) { 1718 if (next->is(TT_LambdaLSquare)) { 1719 NewState.HasMultipleNestedBlocks = true; 1720 break; 1721 } 1722 next = next->Next; 1723 } 1724 } 1725 1726 NewState.IsInsideObjCArrayLiteral = Current.is(TT_ArrayInitializerLSquare) && 1727 Current.Previous && 1728 Current.Previous->is(tok::at); 1729 } 1730 1731 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) { 1732 const FormatToken &Current = *State.NextToken; 1733 if (!Current.closesScope()) 1734 return; 1735 1736 // If we encounter a closing ), ], } or >, we can remove a level from our 1737 // stacks. 1738 if (State.Stack.size() > 1 && 1739 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) || 1740 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) || 1741 State.NextToken->is(TT_TemplateCloser) || 1742 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) { 1743 State.Stack.pop_back(); 1744 } 1745 1746 auto &CurrentState = State.Stack.back(); 1747 1748 // Reevaluate whether ObjC message arguments fit into one line. 1749 // If a receiver spans multiple lines, e.g.: 1750 // [[object block:^{ 1751 // return 42; 1752 // }] a:42 b:42]; 1753 // BreakBeforeParameter is calculated based on an incorrect assumption 1754 // (it is checked whether the whole expression fits into one line without 1755 // considering a line break inside a message receiver). 1756 // We check whether arguments fit after receiver scope closer (into the same 1757 // line). 1758 if (CurrentState.BreakBeforeParameter && Current.MatchingParen && 1759 Current.MatchingParen->Previous) { 1760 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous; 1761 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) && 1762 CurrentScopeOpener.MatchingParen) { 1763 int NecessarySpaceInLine = 1764 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) + 1765 CurrentScopeOpener.TotalLength - Current.TotalLength - 1; 1766 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <= 1767 Style.ColumnLimit) { 1768 CurrentState.BreakBeforeParameter = false; 1769 } 1770 } 1771 } 1772 1773 if (Current.is(tok::r_square)) { 1774 // If this ends the array subscript expr, reset the corresponding value. 1775 const FormatToken *NextNonComment = Current.getNextNonComment(); 1776 if (NextNonComment && NextNonComment->isNot(tok::l_square)) 1777 CurrentState.StartOfArraySubscripts = 0; 1778 } 1779 } 1780 1781 void ContinuationIndenter::moveStateToNewBlock(LineState &State) { 1782 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent; 1783 // ObjC block sometimes follow special indentation rules. 1784 unsigned NewIndent = 1785 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace) 1786 ? Style.ObjCBlockIndentWidth 1787 : Style.IndentWidth); 1788 State.Stack.push_back(ParenState(State.NextToken, NewIndent, 1789 State.Stack.back().LastSpace, 1790 /*AvoidBinPacking=*/true, 1791 /*NoLineBreak=*/false)); 1792 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1793 State.Stack.back().BreakBeforeParameter = true; 1794 } 1795 1796 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn, 1797 unsigned TabWidth, 1798 encoding::Encoding Encoding) { 1799 size_t LastNewlinePos = Text.find_last_of("\n"); 1800 if (LastNewlinePos == StringRef::npos) { 1801 return StartColumn + 1802 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding); 1803 } else { 1804 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos), 1805 /*StartColumn=*/0, TabWidth, Encoding); 1806 } 1807 } 1808 1809 unsigned ContinuationIndenter::reformatRawStringLiteral( 1810 const FormatToken &Current, LineState &State, 1811 const FormatStyle &RawStringStyle, bool DryRun, bool Newline) { 1812 unsigned StartColumn = State.Column - Current.ColumnWidth; 1813 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText); 1814 StringRef NewDelimiter = 1815 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language); 1816 if (NewDelimiter.empty()) 1817 NewDelimiter = OldDelimiter; 1818 // The text of a raw string is between the leading 'R"delimiter(' and the 1819 // trailing 'delimiter)"'. 1820 unsigned OldPrefixSize = 3 + OldDelimiter.size(); 1821 unsigned OldSuffixSize = 2 + OldDelimiter.size(); 1822 // We create a virtual text environment which expects a null-terminated 1823 // string, so we cannot use StringRef. 1824 std::string RawText = std::string( 1825 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize)); 1826 if (NewDelimiter != OldDelimiter) { 1827 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the 1828 // raw string. 1829 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str(); 1830 if (StringRef(RawText).contains(CanonicalDelimiterSuffix)) 1831 NewDelimiter = OldDelimiter; 1832 } 1833 1834 unsigned NewPrefixSize = 3 + NewDelimiter.size(); 1835 unsigned NewSuffixSize = 2 + NewDelimiter.size(); 1836 1837 // The first start column is the column the raw text starts after formatting. 1838 unsigned FirstStartColumn = StartColumn + NewPrefixSize; 1839 1840 // The next start column is the intended indentation a line break inside 1841 // the raw string at level 0. It is determined by the following rules: 1842 // - if the content starts on newline, it is one level more than the current 1843 // indent, and 1844 // - if the content does not start on a newline, it is the first start 1845 // column. 1846 // These rules have the advantage that the formatted content both does not 1847 // violate the rectangle rule and visually flows within the surrounding 1848 // source. 1849 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n'; 1850 // If this token is the last parameter (checked by looking if it's followed by 1851 // `)` and is not on a newline, the base the indent off the line's nested 1852 // block indent. Otherwise, base the indent off the arguments indent, so we 1853 // can achieve: 1854 // 1855 // fffffffffff(1, 2, 3, R"pb( 1856 // key1: 1 # 1857 // key2: 2)pb"); 1858 // 1859 // fffffffffff(1, 2, 3, 1860 // R"pb( 1861 // key1: 1 # 1862 // key2: 2 1863 // )pb"); 1864 // 1865 // fffffffffff(1, 2, 3, 1866 // R"pb( 1867 // key1: 1 # 1868 // key2: 2 1869 // )pb", 1870 // 5); 1871 unsigned CurrentIndent = 1872 (!Newline && Current.Next && Current.Next->is(tok::r_paren)) 1873 ? State.Stack.back().NestedBlockIndent 1874 : State.Stack.back().Indent; 1875 unsigned NextStartColumn = ContentStartsOnNewline 1876 ? CurrentIndent + Style.IndentWidth 1877 : FirstStartColumn; 1878 1879 // The last start column is the column the raw string suffix starts if it is 1880 // put on a newline. 1881 // The last start column is the intended indentation of the raw string postfix 1882 // if it is put on a newline. It is determined by the following rules: 1883 // - if the raw string prefix starts on a newline, it is the column where 1884 // that raw string prefix starts, and 1885 // - if the raw string prefix does not start on a newline, it is the current 1886 // indent. 1887 unsigned LastStartColumn = 1888 Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent; 1889 1890 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat( 1891 RawStringStyle, RawText, {tooling::Range(0, RawText.size())}, 1892 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>", 1893 /*Status=*/nullptr); 1894 1895 auto NewCode = applyAllReplacements(RawText, Fixes.first); 1896 tooling::Replacements NoFixes; 1897 if (!NewCode) 1898 return addMultilineToken(Current, State); 1899 if (!DryRun) { 1900 if (NewDelimiter != OldDelimiter) { 1901 // In 'R"delimiter(...', the delimiter starts 2 characters after the start 1902 // of the token. 1903 SourceLocation PrefixDelimiterStart = 1904 Current.Tok.getLocation().getLocWithOffset(2); 1905 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement( 1906 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1907 if (PrefixErr) { 1908 llvm::errs() 1909 << "Failed to update the prefix delimiter of a raw string: " 1910 << llvm::toString(std::move(PrefixErr)) << "\n"; 1911 } 1912 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at 1913 // position length - 1 - |delimiter|. 1914 SourceLocation SuffixDelimiterStart = 1915 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() - 1916 1 - OldDelimiter.size()); 1917 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement( 1918 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1919 if (SuffixErr) { 1920 llvm::errs() 1921 << "Failed to update the suffix delimiter of a raw string: " 1922 << llvm::toString(std::move(SuffixErr)) << "\n"; 1923 } 1924 } 1925 SourceLocation OriginLoc = 1926 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize); 1927 for (const tooling::Replacement &Fix : Fixes.first) { 1928 auto Err = Whitespaces.addReplacement(tooling::Replacement( 1929 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()), 1930 Fix.getLength(), Fix.getReplacementText())); 1931 if (Err) { 1932 llvm::errs() << "Failed to reformat raw string: " 1933 << llvm::toString(std::move(Err)) << "\n"; 1934 } 1935 } 1936 } 1937 unsigned RawLastLineEndColumn = getLastLineEndColumn( 1938 *NewCode, FirstStartColumn, Style.TabWidth, Encoding); 1939 State.Column = RawLastLineEndColumn + NewSuffixSize; 1940 // Since we're updating the column to after the raw string literal here, we 1941 // have to manually add the penalty for the prefix R"delim( over the column 1942 // limit. 1943 unsigned PrefixExcessCharacters = 1944 StartColumn + NewPrefixSize > Style.ColumnLimit 1945 ? StartColumn + NewPrefixSize - Style.ColumnLimit 1946 : 0; 1947 bool IsMultiline = 1948 ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos); 1949 if (IsMultiline) { 1950 // Break before further function parameters on all levels. 1951 for (ParenState &Paren : State.Stack) 1952 Paren.BreakBeforeParameter = true; 1953 } 1954 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter; 1955 } 1956 1957 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current, 1958 LineState &State) { 1959 // Break before further function parameters on all levels. 1960 for (ParenState &Paren : State.Stack) 1961 Paren.BreakBeforeParameter = true; 1962 1963 unsigned ColumnsUsed = State.Column; 1964 // We can only affect layout of the first and the last line, so the penalty 1965 // for all other lines is constant, and we ignore it. 1966 State.Column = Current.LastLineColumnWidth; 1967 1968 if (ColumnsUsed > getColumnLimit(State)) 1969 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State)); 1970 return 0; 1971 } 1972 1973 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current, 1974 LineState &State, bool DryRun, 1975 bool AllowBreak, bool Newline) { 1976 unsigned Penalty = 0; 1977 // Compute the raw string style to use in case this is a raw string literal 1978 // that can be reformatted. 1979 auto RawStringStyle = getRawStringStyle(Current, State); 1980 if (RawStringStyle && !Current.Finalized) { 1981 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun, 1982 Newline); 1983 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) { 1984 // Don't break multi-line tokens other than block comments and raw string 1985 // literals. Instead, just update the state. 1986 Penalty = addMultilineToken(Current, State); 1987 } else if (State.Line->Type != LT_ImportStatement) { 1988 // We generally don't break import statements. 1989 LineState OriginalState = State; 1990 1991 // Whether we force the reflowing algorithm to stay strictly within the 1992 // column limit. 1993 bool Strict = false; 1994 // Whether the first non-strict attempt at reflowing did intentionally 1995 // exceed the column limit. 1996 bool Exceeded = false; 1997 std::tie(Penalty, Exceeded) = breakProtrudingToken( 1998 Current, State, AllowBreak, /*DryRun=*/true, Strict); 1999 if (Exceeded) { 2000 // If non-strict reflowing exceeds the column limit, try whether strict 2001 // reflowing leads to an overall lower penalty. 2002 LineState StrictState = OriginalState; 2003 unsigned StrictPenalty = 2004 breakProtrudingToken(Current, StrictState, AllowBreak, 2005 /*DryRun=*/true, /*Strict=*/true) 2006 .first; 2007 Strict = StrictPenalty <= Penalty; 2008 if (Strict) { 2009 Penalty = StrictPenalty; 2010 State = StrictState; 2011 } 2012 } 2013 if (!DryRun) { 2014 // If we're not in dry-run mode, apply the changes with the decision on 2015 // strictness made above. 2016 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false, 2017 Strict); 2018 } 2019 } 2020 if (State.Column > getColumnLimit(State)) { 2021 unsigned ExcessCharacters = State.Column - getColumnLimit(State); 2022 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters; 2023 } 2024 return Penalty; 2025 } 2026 2027 // Returns the enclosing function name of a token, or the empty string if not 2028 // found. 2029 static StringRef getEnclosingFunctionName(const FormatToken &Current) { 2030 // Look for: 'function(' or 'function<templates>(' before Current. 2031 auto Tok = Current.getPreviousNonComment(); 2032 if (!Tok || !Tok->is(tok::l_paren)) 2033 return ""; 2034 Tok = Tok->getPreviousNonComment(); 2035 if (!Tok) 2036 return ""; 2037 if (Tok->is(TT_TemplateCloser)) { 2038 Tok = Tok->MatchingParen; 2039 if (Tok) 2040 Tok = Tok->getPreviousNonComment(); 2041 } 2042 if (!Tok || !Tok->is(tok::identifier)) 2043 return ""; 2044 return Tok->TokenText; 2045 } 2046 2047 llvm::Optional<FormatStyle> 2048 ContinuationIndenter::getRawStringStyle(const FormatToken &Current, 2049 const LineState &State) { 2050 if (!Current.isStringLiteral()) 2051 return None; 2052 auto Delimiter = getRawStringDelimiter(Current.TokenText); 2053 if (!Delimiter) 2054 return None; 2055 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter); 2056 if (!RawStringStyle && Delimiter->empty()) { 2057 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle( 2058 getEnclosingFunctionName(Current)); 2059 } 2060 if (!RawStringStyle) 2061 return None; 2062 RawStringStyle->ColumnLimit = getColumnLimit(State); 2063 return RawStringStyle; 2064 } 2065 2066 std::unique_ptr<BreakableToken> 2067 ContinuationIndenter::createBreakableToken(const FormatToken &Current, 2068 LineState &State, bool AllowBreak) { 2069 unsigned StartColumn = State.Column - Current.ColumnWidth; 2070 if (Current.isStringLiteral()) { 2071 // FIXME: String literal breaking is currently disabled for C#, Java, Json 2072 // and JavaScript, as it requires strings to be merged using "+" which we 2073 // don't support. 2074 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 2075 Style.isCSharp() || Style.isJson() || !Style.BreakStringLiterals || 2076 !AllowBreak) { 2077 return nullptr; 2078 } 2079 2080 // Don't break string literals inside preprocessor directives (except for 2081 // #define directives, as their contents are stored in separate lines and 2082 // are not affected by this check). 2083 // This way we avoid breaking code with line directives and unknown 2084 // preprocessor directives that contain long string literals. 2085 if (State.Line->Type == LT_PreprocessorDirective) 2086 return nullptr; 2087 // Exempts unterminated string literals from line breaking. The user will 2088 // likely want to terminate the string before any line breaking is done. 2089 if (Current.IsUnterminatedLiteral) 2090 return nullptr; 2091 // Don't break string literals inside Objective-C array literals (doing so 2092 // raises the warning -Wobjc-string-concatenation). 2093 if (State.Stack.back().IsInsideObjCArrayLiteral) 2094 return nullptr; 2095 2096 StringRef Text = Current.TokenText; 2097 StringRef Prefix; 2098 StringRef Postfix; 2099 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'. 2100 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to 2101 // reduce the overhead) for each FormatToken, which is a string, so that we 2102 // don't run multiple checks here on the hot path. 2103 if ((Text.endswith(Postfix = "\"") && 2104 (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") || 2105 Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") || 2106 Text.startswith(Prefix = "u8\"") || 2107 Text.startswith(Prefix = "L\""))) || 2108 (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) { 2109 // We need this to address the case where there is an unbreakable tail 2110 // only if certain other formatting decisions have been taken. The 2111 // UnbreakableTailLength of Current is an overapproximation is that case 2112 // and we need to be correct here. 2113 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State)) 2114 ? 0 2115 : Current.UnbreakableTailLength; 2116 return std::make_unique<BreakableStringLiteral>( 2117 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength, 2118 State.Line->InPPDirective, Encoding, Style); 2119 } 2120 } else if (Current.is(TT_BlockComment)) { 2121 if (!Style.ReflowComments || 2122 // If a comment token switches formatting, like 2123 // /* clang-format on */, we don't want to break it further, 2124 // but we may still want to adjust its indentation. 2125 switchesFormatting(Current)) { 2126 return nullptr; 2127 } 2128 return std::make_unique<BreakableBlockComment>( 2129 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 2130 State.Line->InPPDirective, Encoding, Style, Whitespaces.useCRLF()); 2131 } else if (Current.is(TT_LineComment) && 2132 (Current.Previous == nullptr || 2133 Current.Previous->isNot(TT_ImplicitStringLiteral))) { 2134 bool RegularComments = [&]() { 2135 for (const FormatToken *T = &Current; T && T->is(TT_LineComment); 2136 T = T->Next) { 2137 if (!(T->TokenText.startswith("//") || T->TokenText.startswith("#"))) 2138 return false; 2139 } 2140 return true; 2141 }(); 2142 if (!Style.ReflowComments || 2143 CommentPragmasRegex.match(Current.TokenText.substr(2)) || 2144 switchesFormatting(Current) || !RegularComments) { 2145 return nullptr; 2146 } 2147 return std::make_unique<BreakableLineCommentSection>( 2148 Current, StartColumn, /*InPPDirective=*/false, Encoding, Style); 2149 } 2150 return nullptr; 2151 } 2152 2153 std::pair<unsigned, bool> 2154 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current, 2155 LineState &State, bool AllowBreak, 2156 bool DryRun, bool Strict) { 2157 std::unique_ptr<const BreakableToken> Token = 2158 createBreakableToken(Current, State, AllowBreak); 2159 if (!Token) 2160 return {0, false}; 2161 assert(Token->getLineCount() > 0); 2162 unsigned ColumnLimit = getColumnLimit(State); 2163 if (Current.is(TT_LineComment)) { 2164 // We don't insert backslashes when breaking line comments. 2165 ColumnLimit = Style.ColumnLimit; 2166 } 2167 if (ColumnLimit == 0) { 2168 // To make the rest of the function easier set the column limit to the 2169 // maximum, if there should be no limit. 2170 ColumnLimit = std::numeric_limits<decltype(ColumnLimit)>::max(); 2171 } 2172 if (Current.UnbreakableTailLength >= ColumnLimit) 2173 return {0, false}; 2174 // ColumnWidth was already accounted into State.Column before calling 2175 // breakProtrudingToken. 2176 unsigned StartColumn = State.Column - Current.ColumnWidth; 2177 unsigned NewBreakPenalty = Current.isStringLiteral() 2178 ? Style.PenaltyBreakString 2179 : Style.PenaltyBreakComment; 2180 // Stores whether we intentionally decide to let a line exceed the column 2181 // limit. 2182 bool Exceeded = false; 2183 // Stores whether we introduce a break anywhere in the token. 2184 bool BreakInserted = Token->introducesBreakBeforeToken(); 2185 // Store whether we inserted a new line break at the end of the previous 2186 // logical line. 2187 bool NewBreakBefore = false; 2188 // We use a conservative reflowing strategy. Reflow starts after a line is 2189 // broken or the corresponding whitespace compressed. Reflow ends as soon as a 2190 // line that doesn't get reflown with the previous line is reached. 2191 bool Reflow = false; 2192 // Keep track of where we are in the token: 2193 // Where we are in the content of the current logical line. 2194 unsigned TailOffset = 0; 2195 // The column number we're currently at. 2196 unsigned ContentStartColumn = 2197 Token->getContentStartColumn(0, /*Break=*/false); 2198 // The number of columns left in the current logical line after TailOffset. 2199 unsigned RemainingTokenColumns = 2200 Token->getRemainingLength(0, TailOffset, ContentStartColumn); 2201 // Adapt the start of the token, for example indent. 2202 if (!DryRun) 2203 Token->adaptStartOfLine(0, Whitespaces); 2204 2205 unsigned ContentIndent = 0; 2206 unsigned Penalty = 0; 2207 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column " 2208 << StartColumn << ".\n"); 2209 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount(); 2210 LineIndex != EndIndex; ++LineIndex) { 2211 LLVM_DEBUG(llvm::dbgs() 2212 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n"); 2213 NewBreakBefore = false; 2214 // If we did reflow the previous line, we'll try reflowing again. Otherwise 2215 // we'll start reflowing if the current line is broken or whitespace is 2216 // compressed. 2217 bool TryReflow = Reflow; 2218 // Break the current token until we can fit the rest of the line. 2219 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2220 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: " 2221 << (ContentStartColumn + RemainingTokenColumns) 2222 << ", space: " << ColumnLimit 2223 << ", reflown prefix: " << ContentStartColumn 2224 << ", offset in line: " << TailOffset << "\n"); 2225 // If the current token doesn't fit, find the latest possible split in the 2226 // current line so that breaking at it will be under the column limit. 2227 // FIXME: Use the earliest possible split while reflowing to correctly 2228 // compress whitespace within a line. 2229 BreakableToken::Split Split = 2230 Token->getSplit(LineIndex, TailOffset, ColumnLimit, 2231 ContentStartColumn, CommentPragmasRegex); 2232 if (Split.first == StringRef::npos) { 2233 // No break opportunity - update the penalty and continue with the next 2234 // logical line. 2235 if (LineIndex < EndIndex - 1) { 2236 // The last line's penalty is handled in addNextStateToQueue() or when 2237 // calling replaceWhitespaceAfterLastLine below. 2238 Penalty += Style.PenaltyExcessCharacter * 2239 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2240 } 2241 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n"); 2242 break; 2243 } 2244 assert(Split.first != 0); 2245 2246 if (Token->supportsReflow()) { 2247 // Check whether the next natural split point after the current one can 2248 // still fit the line, either because we can compress away whitespace, 2249 // or because the penalty the excess characters introduce is lower than 2250 // the break penalty. 2251 // We only do this for tokens that support reflowing, and thus allow us 2252 // to change the whitespace arbitrarily (e.g. comments). 2253 // Other tokens, like string literals, can be broken on arbitrary 2254 // positions. 2255 2256 // First, compute the columns from TailOffset to the next possible split 2257 // position. 2258 // For example: 2259 // ColumnLimit: | 2260 // // Some text that breaks 2261 // ^ tail offset 2262 // ^-- split 2263 // ^-------- to split columns 2264 // ^--- next split 2265 // ^--------------- to next split columns 2266 unsigned ToSplitColumns = Token->getRangeLength( 2267 LineIndex, TailOffset, Split.first, ContentStartColumn); 2268 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n"); 2269 2270 BreakableToken::Split NextSplit = Token->getSplit( 2271 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit, 2272 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex); 2273 // Compute the columns necessary to fit the next non-breakable sequence 2274 // into the current line. 2275 unsigned ToNextSplitColumns = 0; 2276 if (NextSplit.first == StringRef::npos) { 2277 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset, 2278 ContentStartColumn); 2279 } else { 2280 ToNextSplitColumns = Token->getRangeLength( 2281 LineIndex, TailOffset, 2282 Split.first + Split.second + NextSplit.first, ContentStartColumn); 2283 } 2284 // Compress the whitespace between the break and the start of the next 2285 // unbreakable sequence. 2286 ToNextSplitColumns = 2287 Token->getLengthAfterCompression(ToNextSplitColumns, Split); 2288 LLVM_DEBUG(llvm::dbgs() 2289 << " ContentStartColumn: " << ContentStartColumn << "\n"); 2290 LLVM_DEBUG(llvm::dbgs() 2291 << " ToNextSplit: " << ToNextSplitColumns << "\n"); 2292 // If the whitespace compression makes us fit, continue on the current 2293 // line. 2294 bool ContinueOnLine = 2295 ContentStartColumn + ToNextSplitColumns <= ColumnLimit; 2296 unsigned ExcessCharactersPenalty = 0; 2297 if (!ContinueOnLine && !Strict) { 2298 // Similarly, if the excess characters' penalty is lower than the 2299 // penalty of introducing a new break, continue on the current line. 2300 ExcessCharactersPenalty = 2301 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) * 2302 Style.PenaltyExcessCharacter; 2303 LLVM_DEBUG(llvm::dbgs() 2304 << " Penalty excess: " << ExcessCharactersPenalty 2305 << "\n break : " << NewBreakPenalty << "\n"); 2306 if (ExcessCharactersPenalty < NewBreakPenalty) { 2307 Exceeded = true; 2308 ContinueOnLine = true; 2309 } 2310 } 2311 if (ContinueOnLine) { 2312 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n"); 2313 // The current line fits after compressing the whitespace - reflow 2314 // the next line into it if possible. 2315 TryReflow = true; 2316 if (!DryRun) { 2317 Token->compressWhitespace(LineIndex, TailOffset, Split, 2318 Whitespaces); 2319 } 2320 // When we continue on the same line, leave one space between content. 2321 ContentStartColumn += ToSplitColumns + 1; 2322 Penalty += ExcessCharactersPenalty; 2323 TailOffset += Split.first + Split.second; 2324 RemainingTokenColumns = Token->getRemainingLength( 2325 LineIndex, TailOffset, ContentStartColumn); 2326 continue; 2327 } 2328 } 2329 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n"); 2330 // Update the ContentIndent only if the current line was not reflown with 2331 // the previous line, since in that case the previous line should still 2332 // determine the ContentIndent. Also never intent the last line. 2333 if (!Reflow) 2334 ContentIndent = Token->getContentIndent(LineIndex); 2335 LLVM_DEBUG(llvm::dbgs() 2336 << " ContentIndent: " << ContentIndent << "\n"); 2337 ContentStartColumn = ContentIndent + Token->getContentStartColumn( 2338 LineIndex, /*Break=*/true); 2339 2340 unsigned NewRemainingTokenColumns = Token->getRemainingLength( 2341 LineIndex, TailOffset + Split.first + Split.second, 2342 ContentStartColumn); 2343 if (NewRemainingTokenColumns == 0) { 2344 // No content to indent. 2345 ContentIndent = 0; 2346 ContentStartColumn = 2347 Token->getContentStartColumn(LineIndex, /*Break=*/true); 2348 NewRemainingTokenColumns = Token->getRemainingLength( 2349 LineIndex, TailOffset + Split.first + Split.second, 2350 ContentStartColumn); 2351 } 2352 2353 // When breaking before a tab character, it may be moved by a few columns, 2354 // but will still be expanded to the next tab stop, so we don't save any 2355 // columns. 2356 if (NewRemainingTokenColumns >= RemainingTokenColumns) { 2357 // FIXME: Do we need to adjust the penalty? 2358 break; 2359 } 2360 2361 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first 2362 << ", " << Split.second << "\n"); 2363 if (!DryRun) { 2364 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent, 2365 Whitespaces); 2366 } 2367 2368 Penalty += NewBreakPenalty; 2369 TailOffset += Split.first + Split.second; 2370 RemainingTokenColumns = NewRemainingTokenColumns; 2371 BreakInserted = true; 2372 NewBreakBefore = true; 2373 } 2374 // In case there's another line, prepare the state for the start of the next 2375 // line. 2376 if (LineIndex + 1 != EndIndex) { 2377 unsigned NextLineIndex = LineIndex + 1; 2378 if (NewBreakBefore) { 2379 // After breaking a line, try to reflow the next line into the current 2380 // one once RemainingTokenColumns fits. 2381 TryReflow = true; 2382 } 2383 if (TryReflow) { 2384 // We decided that we want to try reflowing the next line into the 2385 // current one. 2386 // We will now adjust the state as if the reflow is successful (in 2387 // preparation for the next line), and see whether that works. If we 2388 // decide that we cannot reflow, we will later reset the state to the 2389 // start of the next line. 2390 Reflow = false; 2391 // As we did not continue breaking the line, RemainingTokenColumns is 2392 // known to fit after ContentStartColumn. Adapt ContentStartColumn to 2393 // the position at which we want to format the next line if we do 2394 // actually reflow. 2395 // When we reflow, we need to add a space between the end of the current 2396 // line and the next line's start column. 2397 ContentStartColumn += RemainingTokenColumns + 1; 2398 // Get the split that we need to reflow next logical line into the end 2399 // of the current one; the split will include any leading whitespace of 2400 // the next logical line. 2401 BreakableToken::Split SplitBeforeNext = 2402 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex); 2403 LLVM_DEBUG(llvm::dbgs() 2404 << " Size of reflown text: " << ContentStartColumn 2405 << "\n Potential reflow split: "); 2406 if (SplitBeforeNext.first != StringRef::npos) { 2407 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", " 2408 << SplitBeforeNext.second << "\n"); 2409 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second; 2410 // If the rest of the next line fits into the current line below the 2411 // column limit, we can safely reflow. 2412 RemainingTokenColumns = Token->getRemainingLength( 2413 NextLineIndex, TailOffset, ContentStartColumn); 2414 Reflow = true; 2415 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2416 LLVM_DEBUG(llvm::dbgs() 2417 << " Over limit after reflow, need: " 2418 << (ContentStartColumn + RemainingTokenColumns) 2419 << ", space: " << ColumnLimit 2420 << ", reflown prefix: " << ContentStartColumn 2421 << ", offset in line: " << TailOffset << "\n"); 2422 // If the whole next line does not fit, try to find a point in 2423 // the next line at which we can break so that attaching the part 2424 // of the next line to that break point onto the current line is 2425 // below the column limit. 2426 BreakableToken::Split Split = 2427 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit, 2428 ContentStartColumn, CommentPragmasRegex); 2429 if (Split.first == StringRef::npos) { 2430 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n"); 2431 Reflow = false; 2432 } else { 2433 // Check whether the first split point gets us below the column 2434 // limit. Note that we will execute this split below as part of 2435 // the normal token breaking and reflow logic within the line. 2436 unsigned ToSplitColumns = Token->getRangeLength( 2437 NextLineIndex, TailOffset, Split.first, ContentStartColumn); 2438 if (ContentStartColumn + ToSplitColumns > ColumnLimit) { 2439 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: " 2440 << (ContentStartColumn + ToSplitColumns) 2441 << ", space: " << ColumnLimit); 2442 unsigned ExcessCharactersPenalty = 2443 (ContentStartColumn + ToSplitColumns - ColumnLimit) * 2444 Style.PenaltyExcessCharacter; 2445 if (NewBreakPenalty < ExcessCharactersPenalty) 2446 Reflow = false; 2447 } 2448 } 2449 } 2450 } else { 2451 LLVM_DEBUG(llvm::dbgs() << "not found.\n"); 2452 } 2453 } 2454 if (!Reflow) { 2455 // If we didn't reflow into the next line, the only space to consider is 2456 // the next logical line. Reset our state to match the start of the next 2457 // line. 2458 TailOffset = 0; 2459 ContentStartColumn = 2460 Token->getContentStartColumn(NextLineIndex, /*Break=*/false); 2461 RemainingTokenColumns = Token->getRemainingLength( 2462 NextLineIndex, TailOffset, ContentStartColumn); 2463 // Adapt the start of the token, for example indent. 2464 if (!DryRun) 2465 Token->adaptStartOfLine(NextLineIndex, Whitespaces); 2466 } else { 2467 // If we found a reflow split and have added a new break before the next 2468 // line, we are going to remove the line break at the start of the next 2469 // logical line. For example, here we'll add a new line break after 2470 // 'text', and subsequently delete the line break between 'that' and 2471 // 'reflows'. 2472 // // some text that 2473 // // reflows 2474 // -> 2475 // // some text 2476 // // that reflows 2477 // When adding the line break, we also added the penalty for it, so we 2478 // need to subtract that penalty again when we remove the line break due 2479 // to reflowing. 2480 if (NewBreakBefore) { 2481 assert(Penalty >= NewBreakPenalty); 2482 Penalty -= NewBreakPenalty; 2483 } 2484 if (!DryRun) 2485 Token->reflow(NextLineIndex, Whitespaces); 2486 } 2487 } 2488 } 2489 2490 BreakableToken::Split SplitAfterLastLine = 2491 Token->getSplitAfterLastLine(TailOffset); 2492 if (SplitAfterLastLine.first != StringRef::npos) { 2493 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n"); 2494 2495 // We add the last line's penalty here, since that line is going to be split 2496 // now. 2497 Penalty += Style.PenaltyExcessCharacter * 2498 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2499 2500 if (!DryRun) { 2501 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine, 2502 Whitespaces); 2503 } 2504 ContentStartColumn = 2505 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true); 2506 RemainingTokenColumns = Token->getRemainingLength( 2507 Token->getLineCount() - 1, 2508 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second, 2509 ContentStartColumn); 2510 } 2511 2512 State.Column = ContentStartColumn + RemainingTokenColumns - 2513 Current.UnbreakableTailLength; 2514 2515 if (BreakInserted) { 2516 // If we break the token inside a parameter list, we need to break before 2517 // the next parameter on all levels, so that the next parameter is clearly 2518 // visible. Line comments already introduce a break. 2519 if (Current.isNot(TT_LineComment)) 2520 for (ParenState &Paren : State.Stack) 2521 Paren.BreakBeforeParameter = true; 2522 2523 if (Current.is(TT_BlockComment)) 2524 State.NoContinuation = true; 2525 2526 State.Stack.back().LastSpace = StartColumn; 2527 } 2528 2529 Token->updateNextToken(State); 2530 2531 return {Penalty, Exceeded}; 2532 } 2533 2534 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const { 2535 // In preprocessor directives reserve two chars for trailing " \" 2536 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0); 2537 } 2538 2539 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) { 2540 const FormatToken &Current = *State.NextToken; 2541 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral)) 2542 return false; 2543 // We never consider raw string literals "multiline" for the purpose of 2544 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased 2545 // (see TokenAnnotator::mustBreakBefore(). 2546 if (Current.TokenText.startswith("R\"")) 2547 return false; 2548 if (Current.IsMultiline) 2549 return true; 2550 if (Current.getNextNonComment() && 2551 Current.getNextNonComment()->isStringLiteral()) { 2552 return true; // Implicit concatenation. 2553 } 2554 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals && 2555 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength > 2556 Style.ColumnLimit) { 2557 return true; // String will be split. 2558 } 2559 return false; 2560 } 2561 2562 } // namespace format 2563 } // namespace clang 2564