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