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 if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName)) 129 return true; 130 return Previous.is(tok::comma) && !Current.isTrailingComment() && 131 ((Previous.isNot(TT_CtorInitializerComma) || 132 Style.BreakConstructorInitializers != 133 FormatStyle::BCIS_BeforeComma) && 134 (Previous.isNot(TT_InheritanceComma) || 135 Style.BreakInheritanceList != FormatStyle::BILS_BeforeComma)); 136 } 137 138 static bool opensProtoMessageField(const FormatToken &LessTok, 139 const FormatStyle &Style) { 140 if (LessTok.isNot(tok::less)) 141 return false; 142 return Style.Language == FormatStyle::LK_TextProto || 143 (Style.Language == FormatStyle::LK_Proto && 144 (LessTok.NestingLevel > 0 || 145 (LessTok.Previous && LessTok.Previous->is(tok::equal)))); 146 } 147 148 // Returns the delimiter of a raw string literal, or None if TokenText is not 149 // the text of a raw string literal. The delimiter could be the empty string. 150 // For example, the delimiter of R"deli(cont)deli" is deli. 151 static llvm::Optional<StringRef> getRawStringDelimiter(StringRef TokenText) { 152 if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'. 153 || !TokenText.startswith("R\"") || !TokenText.endswith("\"")) 154 return None; 155 156 // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has 157 // size at most 16 by the standard, so the first '(' must be among the first 158 // 19 bytes. 159 size_t LParenPos = TokenText.substr(0, 19).find_first_of('('); 160 if (LParenPos == StringRef::npos) 161 return None; 162 StringRef Delimiter = TokenText.substr(2, LParenPos - 2); 163 164 // Check that the string ends in ')Delimiter"'. 165 size_t RParenPos = TokenText.size() - Delimiter.size() - 2; 166 if (TokenText[RParenPos] != ')') 167 return None; 168 if (!TokenText.substr(RParenPos + 1).startswith(Delimiter)) 169 return None; 170 return Delimiter; 171 } 172 173 // Returns the canonical delimiter for \p Language, or the empty string if no 174 // canonical delimiter is specified. 175 static StringRef 176 getCanonicalRawStringDelimiter(const FormatStyle &Style, 177 FormatStyle::LanguageKind Language) { 178 for (const auto &Format : Style.RawStringFormats) 179 if (Format.Language == Language) 180 return StringRef(Format.CanonicalDelimiter); 181 return ""; 182 } 183 184 RawStringFormatStyleManager::RawStringFormatStyleManager( 185 const FormatStyle &CodeStyle) { 186 for (const auto &RawStringFormat : CodeStyle.RawStringFormats) { 187 llvm::Optional<FormatStyle> LanguageStyle = 188 CodeStyle.GetLanguageStyle(RawStringFormat.Language); 189 if (!LanguageStyle) { 190 FormatStyle PredefinedStyle; 191 if (!getPredefinedStyle(RawStringFormat.BasedOnStyle, 192 RawStringFormat.Language, &PredefinedStyle)) { 193 PredefinedStyle = getLLVMStyle(); 194 PredefinedStyle.Language = RawStringFormat.Language; 195 } 196 LanguageStyle = PredefinedStyle; 197 } 198 LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit; 199 for (StringRef Delimiter : RawStringFormat.Delimiters) 200 DelimiterStyle.insert({Delimiter, *LanguageStyle}); 201 for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions) 202 EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle}); 203 } 204 } 205 206 llvm::Optional<FormatStyle> 207 RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const { 208 auto It = DelimiterStyle.find(Delimiter); 209 if (It == DelimiterStyle.end()) 210 return None; 211 return It->second; 212 } 213 214 llvm::Optional<FormatStyle> 215 RawStringFormatStyleManager::getEnclosingFunctionStyle( 216 StringRef EnclosingFunction) const { 217 auto It = EnclosingFunctionStyle.find(EnclosingFunction); 218 if (It == EnclosingFunctionStyle.end()) 219 return None; 220 return It->second; 221 } 222 223 ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style, 224 const AdditionalKeywords &Keywords, 225 const SourceManager &SourceMgr, 226 WhitespaceManager &Whitespaces, 227 encoding::Encoding Encoding, 228 bool BinPackInconclusiveFunctions) 229 : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr), 230 Whitespaces(Whitespaces), Encoding(Encoding), 231 BinPackInconclusiveFunctions(BinPackInconclusiveFunctions), 232 CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {} 233 234 LineState ContinuationIndenter::getInitialState(unsigned FirstIndent, 235 unsigned FirstStartColumn, 236 const AnnotatedLine *Line, 237 bool DryRun) { 238 LineState State; 239 State.FirstIndent = FirstIndent; 240 if (FirstStartColumn && Line->First->NewlinesBefore == 0) 241 State.Column = FirstStartColumn; 242 else 243 State.Column = FirstIndent; 244 // With preprocessor directive indentation, the line starts on column 0 245 // since it's indented after the hash, but FirstIndent is set to the 246 // preprocessor indent. 247 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 248 (Line->Type == LT_PreprocessorDirective || 249 Line->Type == LT_ImportStatement)) 250 State.Column = 0; 251 State.Line = Line; 252 State.NextToken = Line->First; 253 State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent, 254 /*AvoidBinPacking=*/false, 255 /*NoLineBreak=*/false)); 256 State.NoContinuation = false; 257 State.StartOfStringLiteral = 0; 258 State.StartOfLineLevel = 0; 259 State.LowestLevelOnLine = 0; 260 State.IgnoreStackForComparison = false; 261 262 if (Style.Language == FormatStyle::LK_TextProto) { 263 // We need this in order to deal with the bin packing of text fields at 264 // global scope. 265 auto &CurrentState = State.Stack.back(); 266 CurrentState.AvoidBinPacking = true; 267 CurrentState.BreakBeforeParameter = true; 268 CurrentState.AlignColons = false; 269 } 270 271 // The first token has already been indented and thus consumed. 272 moveStateToNextToken(State, DryRun, /*Newline=*/false); 273 return State; 274 } 275 276 bool ContinuationIndenter::canBreak(const LineState &State) { 277 const FormatToken &Current = *State.NextToken; 278 const FormatToken &Previous = *Current.Previous; 279 const auto &CurrentState = State.Stack.back(); 280 assert(&Previous == Current.Previous); 281 if (!Current.CanBreakBefore && !(CurrentState.BreakBeforeClosingBrace && 282 Current.closesBlockOrBlockTypeList(Style))) 283 return false; 284 // The opening "{" of a braced list has to be on the same line as the first 285 // element if it is nested in another braced init list or function call. 286 if (!Current.MustBreakBefore && Previous.is(tok::l_brace) && 287 Previous.isNot(TT_DictLiteral) && Previous.is(BK_BracedInit) && 288 Previous.Previous && 289 Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma)) 290 return false; 291 // This prevents breaks like: 292 // ... 293 // SomeParameter, OtherParameter).DoSomething( 294 // ... 295 // As they hide "DoSomething" and are generally bad for readability. 296 if (Previous.opensScope() && Previous.isNot(tok::l_brace) && 297 State.LowestLevelOnLine < State.StartOfLineLevel && 298 State.LowestLevelOnLine < Current.NestingLevel) 299 return false; 300 if (Current.isMemberAccess() && CurrentState.ContainsUnwrappedBuilder) 301 return false; 302 303 // Don't create a 'hanging' indent if there are multiple blocks in a single 304 // statement. 305 if (Previous.is(tok::l_brace) && State.Stack.size() > 1 && 306 State.Stack[State.Stack.size() - 2].NestedBlockInlined && 307 State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks) 308 return false; 309 310 // Don't break after very short return types (e.g. "void") as that is often 311 // unexpected. 312 if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) { 313 if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None) 314 return false; 315 } 316 317 // If binary operators are moved to the next line (including commas for some 318 // styles of constructor initializers), that's always ok. 319 if (!Current.isOneOf(TT_BinaryOperator, tok::comma) && 320 CurrentState.NoLineBreakInOperand) 321 return false; 322 323 if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr)) 324 return false; 325 326 return !CurrentState.NoLineBreak; 327 } 328 329 bool ContinuationIndenter::mustBreak(const LineState &State) { 330 const FormatToken &Current = *State.NextToken; 331 const FormatToken &Previous = *Current.Previous; 332 const auto &CurrentState = State.Stack.back(); 333 if (Style.BraceWrapping.BeforeLambdaBody && Current.CanBreakBefore && 334 Current.is(TT_LambdaLBrace) && Previous.isNot(TT_LineComment)) { 335 auto LambdaBodyLength = getLengthToMatchingParen(Current, State.Stack); 336 return LambdaBodyLength > getColumnLimit(State); 337 } 338 if (Current.MustBreakBefore || Current.is(TT_InlineASMColon)) 339 return true; 340 if (CurrentState.BreakBeforeClosingBrace && 341 Current.closesBlockOrBlockTypeList(Style)) 342 return true; 343 if (CurrentState.BreakBeforeClosingParen && Current.is(tok::r_paren)) 344 return true; 345 if (Style.Language == FormatStyle::LK_ObjC && 346 Style.ObjCBreakBeforeNestedBlockParam && 347 Current.ObjCSelectorNameParts > 1 && 348 Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) 349 return true; 350 // Avoid producing inconsistent states by requiring breaks where they are not 351 // permitted for C# generic type constraints. 352 if (CurrentState.IsCSharpGenericTypeConstraint && 353 Previous.isNot(TT_CSharpGenericTypeConstraintComma)) 354 return false; 355 if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) || 356 (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) && 357 Style.isCpp() && 358 // FIXME: This is a temporary workaround for the case where clang-format 359 // sets BreakBeforeParameter to avoid bin packing and this creates a 360 // completely unnecessary line break after a template type that isn't 361 // line-wrapped. 362 (Previous.NestingLevel == 1 || Style.BinPackParameters)) || 363 (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) && 364 Previous.isNot(tok::question)) || 365 (!Style.BreakBeforeTernaryOperators && 366 Previous.is(TT_ConditionalExpr))) && 367 CurrentState.BreakBeforeParameter && !Current.isTrailingComment() && 368 !Current.isOneOf(tok::r_paren, tok::r_brace)) 369 return true; 370 if (CurrentState.IsChainedConditional && 371 ((Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) && 372 Current.is(tok::colon)) || 373 (!Style.BreakBeforeTernaryOperators && Previous.is(TT_ConditionalExpr) && 374 Previous.is(tok::colon)))) 375 return true; 376 if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) || 377 (Previous.is(TT_ArrayInitializerLSquare) && 378 Previous.ParameterCount > 1) || 379 opensProtoMessageField(Previous, Style)) && 380 Style.ColumnLimit > 0 && 381 getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 > 382 getColumnLimit(State)) 383 return true; 384 385 const FormatToken &BreakConstructorInitializersToken = 386 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon 387 ? Previous 388 : Current; 389 if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) && 390 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength > 391 getColumnLimit(State) || 392 CurrentState.BreakBeforeParameter) && 393 (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All || 394 Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon || 395 Style.ColumnLimit != 0)) 396 return true; 397 398 if (Current.is(TT_ObjCMethodExpr) && !Previous.is(TT_SelectorName) && 399 State.Line->startsWith(TT_ObjCMethodSpecifier)) 400 return true; 401 if (Current.is(TT_SelectorName) && !Previous.is(tok::at) && 402 CurrentState.ObjCSelectorNameFound && CurrentState.BreakBeforeParameter && 403 (Style.ObjCBreakBeforeNestedBlockParam || 404 !Current.startsSequence(TT_SelectorName, tok::colon, tok::caret))) 405 return true; 406 407 unsigned NewLineColumn = getNewLineColumn(State); 408 if (Current.isMemberAccess() && Style.ColumnLimit != 0 && 409 State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit && 410 (State.Column > NewLineColumn || 411 Current.NestingLevel < State.StartOfLineLevel)) 412 return true; 413 414 if (startsSegmentOfBuilderTypeCall(Current) && 415 (CurrentState.CallContinuation != 0 || 416 CurrentState.BreakBeforeParameter) && 417 // JavaScript is treated different here as there is a frequent pattern: 418 // SomeFunction(function() { 419 // ... 420 // }.bind(...)); 421 // FIXME: We should find a more generic solution to this problem. 422 !(State.Column <= NewLineColumn && Style.isJavaScript()) && 423 !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) 424 return true; 425 426 // If the template declaration spans multiple lines, force wrap before the 427 // function/class declaration 428 if (Previous.ClosesTemplateDeclaration && CurrentState.BreakBeforeParameter && 429 Current.CanBreakBefore) 430 return true; 431 432 if (!State.Line->First->is(tok::kw_enum) && State.Column <= NewLineColumn) 433 return false; 434 435 if (Style.AlwaysBreakBeforeMultilineStrings && 436 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth || 437 Previous.is(tok::comma) || Current.NestingLevel < 2) && 438 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at, 439 Keywords.kw_dollar) && 440 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) && 441 nextIsMultilineString(State)) 442 return true; 443 444 // Using CanBreakBefore here and below takes care of the decision whether the 445 // current style uses wrapping before or after operators for the given 446 // operator. 447 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) { 448 const auto PreviousPrecedence = Previous.getPrecedence(); 449 if (PreviousPrecedence != prec::Assignment && 450 CurrentState.BreakBeforeParameter && !Current.isTrailingComment()) { 451 const bool LHSIsBinaryExpr = 452 Previous.Previous && Previous.Previous->EndsBinaryExpression; 453 if (LHSIsBinaryExpr) 454 return true; 455 // If we need to break somewhere inside the LHS of a binary expression, we 456 // should also break after the operator. Otherwise, the formatting would 457 // hide the operator precedence, e.g. in: 458 // if (aaaaaaaaaaaaaa == 459 // bbbbbbbbbbbbbb && c) {.. 460 // For comparisons, we only apply this rule, if the LHS is a binary 461 // expression itself as otherwise, the line breaks seem superfluous. 462 // We need special cases for ">>" which we have split into two ">" while 463 // lexing in order to make template parsing easier. 464 const bool IsComparison = 465 (PreviousPrecedence == prec::Relational || 466 PreviousPrecedence == prec::Equality || 467 PreviousPrecedence == prec::Spaceship) && 468 Previous.Previous && 469 Previous.Previous->isNot(TT_BinaryOperator); // For >>. 470 if (!IsComparison) 471 return true; 472 } 473 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore && 474 CurrentState.BreakBeforeParameter) { 475 return true; 476 } 477 478 // Same as above, but for the first "<<" operator. 479 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) && 480 CurrentState.BreakBeforeParameter && CurrentState.FirstLessLess == 0) 481 return true; 482 483 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) { 484 // Always break after "template <...>"(*) and leading annotations. This is 485 // only for cases where the entire line does not fit on a single line as a 486 // different LineFormatter would be used otherwise. 487 // *: Except when another option interferes with that, like concepts. 488 if (Previous.ClosesTemplateDeclaration) { 489 if (Current.is(tok::kw_concept)) { 490 switch (Style.BreakBeforeConceptDeclarations) { 491 case FormatStyle::BBCDS_Allowed: 492 break; 493 case FormatStyle::BBCDS_Always: 494 return true; 495 case FormatStyle::BBCDS_Never: 496 return false; 497 } 498 } 499 if (Current.is(TT_RequiresClause)) { 500 switch (Style.RequiresClausePosition) { 501 case FormatStyle::RCPS_SingleLine: 502 case FormatStyle::RCPS_WithPreceding: 503 return false; 504 default: 505 return true; 506 } 507 } 508 return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No; 509 } 510 if (Previous.is(TT_FunctionAnnotationRParen) && 511 State.Line->Type != LT_PreprocessorDirective) 512 return true; 513 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) && 514 Current.isNot(TT_LeadingJavaAnnotation)) 515 return true; 516 } 517 518 if (Style.isJavaScript() && Previous.is(tok::r_paren) && 519 Previous.is(TT_JavaAnnotation)) { 520 // Break after the closing parenthesis of TypeScript decorators before 521 // functions, getters and setters. 522 static const llvm::StringSet<> BreakBeforeDecoratedTokens = {"get", "set", 523 "function"}; 524 if (BreakBeforeDecoratedTokens.contains(Current.TokenText)) 525 return true; 526 } 527 528 // If the return type spans multiple lines, wrap before the function name. 529 if (((Current.is(TT_FunctionDeclarationName) && 530 // Don't break before a C# function when no break after return type 531 (!Style.isCSharp() || 532 Style.AlwaysBreakAfterReturnType != FormatStyle::RTBS_None) && 533 // Don't always break between a JavaScript `function` and the function 534 // name. 535 !Style.isJavaScript()) || 536 (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) && 537 !Previous.is(tok::kw_template) && CurrentState.BreakBeforeParameter) 538 return true; 539 540 // The following could be precomputed as they do not depend on the state. 541 // However, as they should take effect only if the UnwrappedLine does not fit 542 // into the ColumnLimit, they are checked here in the ContinuationIndenter. 543 if (Style.ColumnLimit != 0 && Previous.is(BK_Block) && 544 Previous.is(tok::l_brace) && !Current.isOneOf(tok::r_brace, tok::comment)) 545 return true; 546 547 if (Current.is(tok::lessless) && 548 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") || 549 (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") || 550 Previous.TokenText == "\'\\n\'")))) 551 return true; 552 553 if (Previous.is(TT_BlockComment) && Previous.IsMultiline) 554 return true; 555 556 if (State.NoContinuation) 557 return true; 558 559 return false; 560 } 561 562 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline, 563 bool DryRun, 564 unsigned ExtraSpaces) { 565 const FormatToken &Current = *State.NextToken; 566 assert(State.NextToken->Previous); 567 const FormatToken &Previous = *State.NextToken->Previous; 568 569 assert(!State.Stack.empty()); 570 State.NoContinuation = false; 571 572 if ((Current.is(TT_ImplicitStringLiteral) && 573 (Previous.Tok.getIdentifierInfo() == nullptr || 574 Previous.Tok.getIdentifierInfo()->getPPKeywordID() == 575 tok::pp_not_keyword))) { 576 unsigned EndColumn = 577 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd()); 578 if (Current.LastNewlineOffset != 0) { 579 // If there is a newline within this token, the final column will solely 580 // determined by the current end column. 581 State.Column = EndColumn; 582 } else { 583 unsigned StartColumn = 584 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin()); 585 assert(EndColumn >= StartColumn); 586 State.Column += EndColumn - StartColumn; 587 } 588 moveStateToNextToken(State, DryRun, /*Newline=*/false); 589 return 0; 590 } 591 592 unsigned Penalty = 0; 593 if (Newline) 594 Penalty = addTokenOnNewLine(State, DryRun); 595 else 596 addTokenOnCurrentLine(State, DryRun, ExtraSpaces); 597 598 return moveStateToNextToken(State, DryRun, Newline) + Penalty; 599 } 600 601 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun, 602 unsigned ExtraSpaces) { 603 FormatToken &Current = *State.NextToken; 604 assert(State.NextToken->Previous); 605 const FormatToken &Previous = *State.NextToken->Previous; 606 auto &CurrentState = State.Stack.back(); 607 608 if (Current.is(tok::equal) && 609 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) && 610 CurrentState.VariablePos == 0) { 611 CurrentState.VariablePos = State.Column; 612 // Move over * and & if they are bound to the variable name. 613 const FormatToken *Tok = &Previous; 614 while (Tok && CurrentState.VariablePos >= Tok->ColumnWidth) { 615 CurrentState.VariablePos -= Tok->ColumnWidth; 616 if (Tok->SpacesRequiredBefore != 0) 617 break; 618 Tok = Tok->Previous; 619 } 620 if (Previous.PartOfMultiVariableDeclStmt) 621 CurrentState.LastSpace = CurrentState.VariablePos; 622 } 623 624 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces; 625 626 // Indent preprocessor directives after the hash if required. 627 int PPColumnCorrection = 0; 628 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 629 Previous.is(tok::hash) && State.FirstIndent > 0 && 630 (State.Line->Type == LT_PreprocessorDirective || 631 State.Line->Type == LT_ImportStatement)) { 632 Spaces += State.FirstIndent; 633 634 // For preprocessor indent with tabs, State.Column will be 1 because of the 635 // hash. This causes second-level indents onward to have an extra space 636 // after the tabs. We avoid this misalignment by subtracting 1 from the 637 // column value passed to replaceWhitespace(). 638 if (Style.UseTab != FormatStyle::UT_Never) 639 PPColumnCorrection = -1; 640 } 641 642 if (!DryRun) 643 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces, 644 State.Column + Spaces + PPColumnCorrection); 645 646 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance 647 // declaration unless there is multiple inheritance. 648 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 649 Current.is(TT_InheritanceColon)) 650 CurrentState.NoLineBreak = true; 651 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon && 652 Previous.is(TT_InheritanceColon)) 653 CurrentState.NoLineBreak = true; 654 655 if (Current.is(TT_SelectorName) && !CurrentState.ObjCSelectorNameFound) { 656 unsigned MinIndent = std::max( 657 State.FirstIndent + Style.ContinuationIndentWidth, CurrentState.Indent); 658 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth; 659 if (Current.LongestObjCSelectorName == 0) 660 CurrentState.AlignColons = false; 661 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos) 662 CurrentState.ColonPos = MinIndent + Current.LongestObjCSelectorName; 663 else 664 CurrentState.ColonPos = FirstColonPos; 665 } 666 667 // In "AlwaysBreak" or "BlockIndent" mode, enforce wrapping directly after the 668 // parenthesis by disallowing any further line breaks if there is no line 669 // break after the opening parenthesis. Don't break if it doesn't conserve 670 // columns. 671 if ((Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak || 672 Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent) && 673 (Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) || 674 (Previous.is(tok::l_brace) && Previous.isNot(BK_Block) && 675 Style.Cpp11BracedListStyle)) && 676 State.Column > getNewLineColumn(State) && 677 (!Previous.Previous || !Previous.Previous->isOneOf( 678 tok::kw_for, tok::kw_while, tok::kw_switch)) && 679 // Don't do this for simple (no expressions) one-argument function calls 680 // as that feels like needlessly wasting whitespace, e.g.: 681 // 682 // caaaaaaaaaaaall( 683 // caaaaaaaaaaaall( 684 // caaaaaaaaaaaall( 685 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa)))); 686 Current.FakeLParens.size() > 0 && 687 Current.FakeLParens.back() > prec::Unknown) 688 CurrentState.NoLineBreak = true; 689 if (Previous.is(TT_TemplateString) && Previous.opensScope()) 690 CurrentState.NoLineBreak = true; 691 692 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign && 693 !CurrentState.IsCSharpGenericTypeConstraint && Previous.opensScope() && 694 Previous.isNot(TT_ObjCMethodExpr) && Previous.isNot(TT_RequiresClause) && 695 (Current.isNot(TT_LineComment) || Previous.is(BK_BracedInit))) { 696 CurrentState.Indent = State.Column + Spaces; 697 CurrentState.IsAligned = true; 698 } 699 if (CurrentState.AvoidBinPacking && startsNextParameter(Current, Style)) 700 CurrentState.NoLineBreak = true; 701 if (startsSegmentOfBuilderTypeCall(Current) && 702 State.Column > getNewLineColumn(State)) 703 CurrentState.ContainsUnwrappedBuilder = true; 704 705 if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java) 706 CurrentState.NoLineBreak = true; 707 if (Current.isMemberAccess() && Previous.is(tok::r_paren) && 708 (Previous.MatchingParen && 709 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) 710 // If there is a function call with long parameters, break before trailing 711 // calls. This prevents things like: 712 // EXPECT_CALL(SomeLongParameter).Times( 713 // 2); 714 // We don't want to do this for short parameters as they can just be 715 // indexes. 716 CurrentState.NoLineBreak = true; 717 718 // Don't allow the RHS of an operator to be split over multiple lines unless 719 // there is a line-break right after the operator. 720 // Exclude relational operators, as there, it is always more desirable to 721 // have the LHS 'left' of the RHS. 722 const FormatToken *P = Current.getPreviousNonComment(); 723 if (!Current.is(tok::comment) && P && 724 (P->isOneOf(TT_BinaryOperator, tok::comma) || 725 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) && 726 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) && 727 P->getPrecedence() != prec::Assignment && 728 P->getPrecedence() != prec::Relational && 729 P->getPrecedence() != prec::Spaceship) { 730 bool BreakBeforeOperator = 731 P->MustBreakBefore || P->is(tok::lessless) || 732 (P->is(TT_BinaryOperator) && 733 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) || 734 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators); 735 // Don't do this if there are only two operands. In these cases, there is 736 // always a nice vertical separation between them and the extra line break 737 // does not help. 738 bool HasTwoOperands = 739 P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr); 740 if ((!BreakBeforeOperator && 741 !(HasTwoOperands && 742 Style.AlignOperands != FormatStyle::OAS_DontAlign)) || 743 (!CurrentState.LastOperatorWrapped && BreakBeforeOperator)) 744 CurrentState.NoLineBreakInOperand = true; 745 } 746 747 State.Column += Spaces; 748 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) && 749 Previous.Previous && 750 (Previous.Previous->is(tok::kw_for) || Previous.Previous->isIf())) { 751 // Treat the condition inside an if as if it was a second function 752 // parameter, i.e. let nested calls have a continuation indent. 753 CurrentState.LastSpace = State.Column; 754 CurrentState.NestedBlockIndent = State.Column; 755 } else if (!Current.isOneOf(tok::comment, tok::caret) && 756 ((Previous.is(tok::comma) && 757 !Previous.is(TT_OverloadedOperator)) || 758 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) { 759 CurrentState.LastSpace = State.Column; 760 } else if (Previous.is(TT_CtorInitializerColon) && 761 Style.BreakConstructorInitializers == 762 FormatStyle::BCIS_AfterColon) { 763 CurrentState.Indent = State.Column; 764 CurrentState.LastSpace = State.Column; 765 } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 766 TT_CtorInitializerColon)) && 767 ((Previous.getPrecedence() != prec::Assignment && 768 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 || 769 Previous.NextOperator)) || 770 Current.StartsBinaryExpression)) { 771 // Indent relative to the RHS of the expression unless this is a simple 772 // assignment without binary expression on the RHS. Also indent relative to 773 // unary operators and the colons of constructor initializers. 774 if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None) 775 CurrentState.LastSpace = State.Column; 776 } else if (Previous.is(TT_InheritanceColon)) { 777 CurrentState.Indent = State.Column; 778 CurrentState.LastSpace = State.Column; 779 } else if (Current.is(TT_CSharpGenericTypeConstraintColon)) { 780 CurrentState.ColonPos = State.Column; 781 } else if (Previous.opensScope()) { 782 // If a function has a trailing call, indent all parameters from the 783 // opening parenthesis. This avoids confusing indents like: 784 // OuterFunction(InnerFunctionCall( // break 785 // ParameterToInnerFunction)) // break 786 // .SecondInnerFunctionCall(); 787 if (Previous.MatchingParen) { 788 const FormatToken *Next = Previous.MatchingParen->getNextNonComment(); 789 if (Next && Next->isMemberAccess() && State.Stack.size() > 1 && 790 State.Stack[State.Stack.size() - 2].CallContinuation == 0) 791 CurrentState.LastSpace = State.Column; 792 } 793 } 794 } 795 796 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State, 797 bool DryRun) { 798 FormatToken &Current = *State.NextToken; 799 assert(State.NextToken->Previous); 800 const FormatToken &Previous = *State.NextToken->Previous; 801 auto &CurrentState = State.Stack.back(); 802 803 // Extra penalty that needs to be added because of the way certain line 804 // breaks are chosen. 805 unsigned Penalty = 0; 806 807 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 808 const FormatToken *NextNonComment = Previous.getNextNonComment(); 809 if (!NextNonComment) 810 NextNonComment = &Current; 811 // The first line break on any NestingLevel causes an extra penalty in order 812 // prefer similar line breaks. 813 if (!CurrentState.ContainsLineBreak) 814 Penalty += 15; 815 CurrentState.ContainsLineBreak = true; 816 817 Penalty += State.NextToken->SplitPenalty; 818 819 // Breaking before the first "<<" is generally not desirable if the LHS is 820 // short. Also always add the penalty if the LHS is split over multiple lines 821 // to avoid unnecessary line breaks that just work around this penalty. 822 if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess == 0 && 823 (State.Column <= Style.ColumnLimit / 3 || 824 CurrentState.BreakBeforeParameter)) 825 Penalty += Style.PenaltyBreakFirstLessLess; 826 827 State.Column = getNewLineColumn(State); 828 829 // Add Penalty proportional to amount of whitespace away from FirstColumn 830 // This tends to penalize several lines that are far-right indented, 831 // and prefers a line-break prior to such a block, e.g: 832 // 833 // Constructor() : 834 // member(value), looooooooooooooooong_member( 835 // looooooooooong_call(param_1, param_2, param_3)) 836 // would then become 837 // Constructor() : 838 // member(value), 839 // looooooooooooooooong_member( 840 // looooooooooong_call(param_1, param_2, param_3)) 841 if (State.Column > State.FirstIndent) 842 Penalty += 843 Style.PenaltyIndentedWhitespace * (State.Column - State.FirstIndent); 844 845 // Indent nested blocks relative to this column, unless in a very specific 846 // JavaScript special case where: 847 // 848 // var loooooong_name = 849 // function() { 850 // // code 851 // } 852 // 853 // is common and should be formatted like a free-standing function. The same 854 // goes for wrapping before the lambda return type arrow. 855 if (!Current.is(TT_LambdaArrow) && 856 (!Style.isJavaScript() || Current.NestingLevel != 0 || 857 !PreviousNonComment || !PreviousNonComment->is(tok::equal) || 858 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) 859 CurrentState.NestedBlockIndent = State.Column; 860 861 if (NextNonComment->isMemberAccess()) { 862 if (CurrentState.CallContinuation == 0) 863 CurrentState.CallContinuation = State.Column; 864 } else if (NextNonComment->is(TT_SelectorName)) { 865 if (!CurrentState.ObjCSelectorNameFound) { 866 if (NextNonComment->LongestObjCSelectorName == 0) { 867 CurrentState.AlignColons = false; 868 } else { 869 CurrentState.ColonPos = 870 (shouldIndentWrappedSelectorName(Style, State.Line->Type) 871 ? std::max(CurrentState.Indent, 872 State.FirstIndent + Style.ContinuationIndentWidth) 873 : CurrentState.Indent) + 874 std::max(NextNonComment->LongestObjCSelectorName, 875 NextNonComment->ColumnWidth); 876 } 877 } else if (CurrentState.AlignColons && 878 CurrentState.ColonPos <= NextNonComment->ColumnWidth) { 879 CurrentState.ColonPos = State.Column + NextNonComment->ColumnWidth; 880 } 881 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 882 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) { 883 // FIXME: This is hacky, find a better way. The problem is that in an ObjC 884 // method expression, the block should be aligned to the line starting it, 885 // e.g.: 886 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason 887 // ^(int *i) { 888 // // ... 889 // }]; 890 // Thus, we set LastSpace of the next higher NestingLevel, to which we move 891 // when we consume all of the "}"'s FakeRParens at the "{". 892 if (State.Stack.size() > 1) 893 State.Stack[State.Stack.size() - 2].LastSpace = 894 std::max(CurrentState.LastSpace, CurrentState.Indent) + 895 Style.ContinuationIndentWidth; 896 } 897 898 if ((PreviousNonComment && 899 PreviousNonComment->isOneOf(tok::comma, tok::semi) && 900 !CurrentState.AvoidBinPacking) || 901 Previous.is(TT_BinaryOperator)) 902 CurrentState.BreakBeforeParameter = false; 903 if (PreviousNonComment && 904 (PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) || 905 PreviousNonComment->ClosesRequiresClause) && 906 Current.NestingLevel == 0) 907 CurrentState.BreakBeforeParameter = false; 908 if (NextNonComment->is(tok::question) || 909 (PreviousNonComment && PreviousNonComment->is(tok::question))) 910 CurrentState.BreakBeforeParameter = true; 911 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore) 912 CurrentState.BreakBeforeParameter = false; 913 914 if (!DryRun) { 915 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1; 916 if (Current.is(tok::r_brace) && Current.MatchingParen && 917 // Only strip trailing empty lines for l_braces that have children, i.e. 918 // for function expressions (lambdas, arrows, etc). 919 !Current.MatchingParen->Children.empty()) { 920 // lambdas and arrow functions are expressions, thus their r_brace is not 921 // on its own line, and thus not covered by UnwrappedLineFormatter's logic 922 // about removing empty lines on closing blocks. Special case them here. 923 MaxEmptyLinesToKeep = 1; 924 } 925 unsigned Newlines = 926 std::max(1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep)); 927 bool ContinuePPDirective = 928 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement; 929 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column, 930 CurrentState.IsAligned, ContinuePPDirective); 931 } 932 933 if (!Current.isTrailingComment()) 934 CurrentState.LastSpace = State.Column; 935 if (Current.is(tok::lessless)) 936 // If we are breaking before a "<<", we always want to indent relative to 937 // RHS. This is necessary only for "<<", as we special-case it and don't 938 // always indent relative to the RHS. 939 CurrentState.LastSpace += 3; // 3 -> width of "<< ". 940 941 State.StartOfLineLevel = Current.NestingLevel; 942 State.LowestLevelOnLine = Current.NestingLevel; 943 944 // Any break on this level means that the parent level has been broken 945 // and we need to avoid bin packing there. 946 bool NestedBlockSpecialCase = 947 (!Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 && 948 State.Stack[State.Stack.size() - 2].NestedBlockInlined) || 949 (Style.Language == FormatStyle::LK_ObjC && Current.is(tok::r_brace) && 950 State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam); 951 // Do not force parameter break for statements with requires expressions. 952 NestedBlockSpecialCase = 953 NestedBlockSpecialCase || 954 (Current.MatchingParen && 955 Current.MatchingParen->is(TT_RequiresExpressionLBrace)); 956 if (!NestedBlockSpecialCase) 957 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 958 State.Stack[i].BreakBeforeParameter = true; 959 960 if (PreviousNonComment && 961 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) && 962 ((PreviousNonComment->isNot(TT_TemplateCloser) && 963 !PreviousNonComment->ClosesRequiresClause) || 964 Current.NestingLevel != 0) && 965 !PreviousNonComment->isOneOf( 966 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation, 967 TT_LeadingJavaAnnotation) && 968 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope()) 969 CurrentState.BreakBeforeParameter = true; 970 971 // If we break after { or the [ of an array initializer, we should also break 972 // before the corresponding } or ]. 973 if (PreviousNonComment && 974 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 975 opensProtoMessageField(*PreviousNonComment, Style))) 976 CurrentState.BreakBeforeClosingBrace = true; 977 978 if (PreviousNonComment && PreviousNonComment->is(tok::l_paren)) 979 CurrentState.BreakBeforeClosingParen = 980 Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent; 981 982 if (CurrentState.AvoidBinPacking) { 983 // If we are breaking after '(', '{', '<', or this is the break after a ':' 984 // to start a member initializater list in a constructor, this should not 985 // be considered bin packing unless the relevant AllowAll option is false or 986 // this is a dict/object literal. 987 bool PreviousIsBreakingCtorInitializerColon = 988 Previous.is(TT_CtorInitializerColon) && 989 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 990 if (!(Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) || 991 PreviousIsBreakingCtorInitializerColon) || 992 (!Style.AllowAllParametersOfDeclarationOnNextLine && 993 State.Line->MustBeDeclaration) || 994 (!Style.AllowAllArgumentsOnNextLine && 995 !State.Line->MustBeDeclaration) || 996 (Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine && 997 PreviousIsBreakingCtorInitializerColon) || 998 Previous.is(TT_DictLiteral)) 999 CurrentState.BreakBeforeParameter = true; 1000 1001 // If we are breaking after a ':' to start a member initializer list, 1002 // and we allow all arguments on the next line, we should not break 1003 // before the next parameter. 1004 if (PreviousIsBreakingCtorInitializerColon && 1005 Style.PackConstructorInitializers == FormatStyle::PCIS_NextLine) 1006 CurrentState.BreakBeforeParameter = false; 1007 } 1008 1009 return Penalty; 1010 } 1011 1012 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) { 1013 if (!State.NextToken || !State.NextToken->Previous) 1014 return 0; 1015 1016 FormatToken &Current = *State.NextToken; 1017 const auto &CurrentState = State.Stack.back(); 1018 1019 if (CurrentState.IsCSharpGenericTypeConstraint && 1020 Current.isNot(TT_CSharpGenericTypeConstraint)) 1021 return CurrentState.ColonPos + 2; 1022 1023 const FormatToken &Previous = *Current.Previous; 1024 // If we are continuing an expression, we want to use the continuation indent. 1025 unsigned ContinuationIndent = 1026 std::max(CurrentState.LastSpace, CurrentState.Indent) + 1027 Style.ContinuationIndentWidth; 1028 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 1029 const FormatToken *NextNonComment = Previous.getNextNonComment(); 1030 if (!NextNonComment) 1031 NextNonComment = &Current; 1032 1033 // Java specific bits. 1034 if (Style.Language == FormatStyle::LK_Java && 1035 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) 1036 return std::max(CurrentState.LastSpace, 1037 CurrentState.Indent + Style.ContinuationIndentWidth); 1038 1039 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths && 1040 State.Line->First->is(tok::kw_enum)) 1041 return (Style.IndentWidth * State.Line->First->IndentLevel) + 1042 Style.IndentWidth; 1043 1044 if (NextNonComment->is(tok::l_brace) && NextNonComment->is(BK_Block)) 1045 return Current.NestingLevel == 0 ? State.FirstIndent : CurrentState.Indent; 1046 if ((Current.isOneOf(tok::r_brace, tok::r_square) || 1047 (Current.is(tok::greater) && 1048 (Style.Language == FormatStyle::LK_Proto || 1049 Style.Language == FormatStyle::LK_TextProto))) && 1050 State.Stack.size() > 1) { 1051 if (Current.closesBlockOrBlockTypeList(Style)) 1052 return State.Stack[State.Stack.size() - 2].NestedBlockIndent; 1053 if (Current.MatchingParen && Current.MatchingParen->is(BK_BracedInit)) 1054 return State.Stack[State.Stack.size() - 2].LastSpace; 1055 return State.FirstIndent; 1056 } 1057 // Indent a closing parenthesis at the previous level if followed by a semi, 1058 // const, or opening brace. This allows indentations such as: 1059 // foo( 1060 // a, 1061 // ); 1062 // int Foo::getter( 1063 // // 1064 // ) const { 1065 // return foo; 1066 // } 1067 // function foo( 1068 // a, 1069 // ) { 1070 // code(); // 1071 // } 1072 if (Current.is(tok::r_paren) && State.Stack.size() > 1 && 1073 (!Current.Next || 1074 Current.Next->isOneOf(tok::semi, tok::kw_const, tok::l_brace))) 1075 return State.Stack[State.Stack.size() - 2].LastSpace; 1076 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent && 1077 Current.is(tok::r_paren) && State.Stack.size() > 1) 1078 return State.Stack[State.Stack.size() - 2].LastSpace; 1079 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope()) 1080 return State.Stack[State.Stack.size() - 2].LastSpace; 1081 if (Current.is(tok::identifier) && Current.Next && 1082 (Current.Next->is(TT_DictLiteral) || 1083 ((Style.Language == FormatStyle::LK_Proto || 1084 Style.Language == FormatStyle::LK_TextProto) && 1085 Current.Next->isOneOf(tok::less, tok::l_brace)))) 1086 return CurrentState.Indent; 1087 if (NextNonComment->is(TT_ObjCStringLiteral) && 1088 State.StartOfStringLiteral != 0) 1089 return State.StartOfStringLiteral - 1; 1090 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0) 1091 return State.StartOfStringLiteral; 1092 if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess != 0) 1093 return CurrentState.FirstLessLess; 1094 if (NextNonComment->isMemberAccess()) { 1095 if (CurrentState.CallContinuation == 0) 1096 return ContinuationIndent; 1097 return CurrentState.CallContinuation; 1098 } 1099 if (CurrentState.QuestionColumn != 0 && 1100 ((NextNonComment->is(tok::colon) && 1101 NextNonComment->is(TT_ConditionalExpr)) || 1102 Previous.is(TT_ConditionalExpr))) { 1103 if (((NextNonComment->is(tok::colon) && NextNonComment->Next && 1104 !NextNonComment->Next->FakeLParens.empty() && 1105 NextNonComment->Next->FakeLParens.back() == prec::Conditional) || 1106 (Previous.is(tok::colon) && !Current.FakeLParens.empty() && 1107 Current.FakeLParens.back() == prec::Conditional)) && 1108 !CurrentState.IsWrappedConditional) { 1109 // NOTE: we may tweak this slightly: 1110 // * not remove the 'lead' ContinuationIndentWidth 1111 // * always un-indent by the operator when 1112 // BreakBeforeTernaryOperators=true 1113 unsigned Indent = CurrentState.Indent; 1114 if (Style.AlignOperands != FormatStyle::OAS_DontAlign) 1115 Indent -= Style.ContinuationIndentWidth; 1116 if (Style.BreakBeforeTernaryOperators && CurrentState.UnindentOperator) 1117 Indent -= 2; 1118 return Indent; 1119 } 1120 return CurrentState.QuestionColumn; 1121 } 1122 if (Previous.is(tok::comma) && CurrentState.VariablePos != 0) 1123 return CurrentState.VariablePos; 1124 if (Current.is(TT_RequiresClause)) { 1125 if (Style.IndentRequiresClause) 1126 return CurrentState.Indent + Style.IndentWidth; 1127 switch (Style.RequiresClausePosition) { 1128 case FormatStyle::RCPS_OwnLine: 1129 case FormatStyle::RCPS_WithFollowing: 1130 return CurrentState.Indent; 1131 default: 1132 break; 1133 } 1134 } 1135 if ((PreviousNonComment && 1136 (PreviousNonComment->ClosesTemplateDeclaration || 1137 PreviousNonComment->ClosesRequiresClause || 1138 PreviousNonComment->isOneOf( 1139 TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen, 1140 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) || 1141 (!Style.IndentWrappedFunctionNames && 1142 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) 1143 return std::max(CurrentState.LastSpace, CurrentState.Indent); 1144 if (NextNonComment->is(TT_SelectorName)) { 1145 if (!CurrentState.ObjCSelectorNameFound) { 1146 unsigned MinIndent = CurrentState.Indent; 1147 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) 1148 MinIndent = std::max(MinIndent, 1149 State.FirstIndent + Style.ContinuationIndentWidth); 1150 // If LongestObjCSelectorName is 0, we are indenting the first 1151 // part of an ObjC selector (or a selector component which is 1152 // not colon-aligned due to block formatting). 1153 // 1154 // Otherwise, we are indenting a subsequent part of an ObjC 1155 // selector which should be colon-aligned to the longest 1156 // component of the ObjC selector. 1157 // 1158 // In either case, we want to respect Style.IndentWrappedFunctionNames. 1159 return MinIndent + 1160 std::max(NextNonComment->LongestObjCSelectorName, 1161 NextNonComment->ColumnWidth) - 1162 NextNonComment->ColumnWidth; 1163 } 1164 if (!CurrentState.AlignColons) 1165 return CurrentState.Indent; 1166 if (CurrentState.ColonPos > NextNonComment->ColumnWidth) 1167 return CurrentState.ColonPos - NextNonComment->ColumnWidth; 1168 return CurrentState.Indent; 1169 } 1170 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr)) 1171 return CurrentState.ColonPos; 1172 if (NextNonComment->is(TT_ArraySubscriptLSquare)) { 1173 if (CurrentState.StartOfArraySubscripts != 0) 1174 return CurrentState.StartOfArraySubscripts; 1175 else if (Style.isCSharp()) // C# allows `["key"] = value` inside object 1176 // initializers. 1177 return CurrentState.Indent; 1178 return ContinuationIndent; 1179 } 1180 1181 // This ensure that we correctly format ObjC methods calls without inputs, 1182 // i.e. where the last element isn't selector like: [callee method]; 1183 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 && 1184 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) 1185 return CurrentState.Indent; 1186 1187 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) || 1188 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) 1189 return ContinuationIndent; 1190 if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 1191 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) 1192 return ContinuationIndent; 1193 if (NextNonComment->is(TT_CtorInitializerComma)) 1194 return CurrentState.Indent; 1195 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && 1196 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) 1197 return CurrentState.Indent; 1198 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) && 1199 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) 1200 return CurrentState.Indent; 1201 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon, 1202 TT_InheritanceComma)) 1203 return State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1204 if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() && 1205 !Current.isOneOf(tok::colon, tok::comment)) 1206 return ContinuationIndent; 1207 if (Current.is(TT_ProtoExtensionLSquare)) 1208 return CurrentState.Indent; 1209 if (Current.isBinaryOperator() && CurrentState.UnindentOperator) 1210 return CurrentState.Indent - Current.Tok.getLength() - 1211 Current.SpacesRequiredBefore; 1212 if (Current.isOneOf(tok::comment, TT_BlockComment, TT_LineComment) && 1213 NextNonComment->isBinaryOperator() && CurrentState.UnindentOperator) 1214 return CurrentState.Indent - NextNonComment->Tok.getLength() - 1215 NextNonComment->SpacesRequiredBefore; 1216 if (CurrentState.Indent == State.FirstIndent && PreviousNonComment && 1217 !PreviousNonComment->isOneOf(tok::r_brace, TT_CtorInitializerComma)) 1218 // Ensure that we fall back to the continuation indent width instead of 1219 // just flushing continuations left. 1220 return CurrentState.Indent + Style.ContinuationIndentWidth; 1221 return CurrentState.Indent; 1222 } 1223 1224 static bool hasNestedBlockInlined(const FormatToken *Previous, 1225 const FormatToken &Current, 1226 const FormatStyle &Style) { 1227 if (Previous->isNot(tok::l_paren)) 1228 return true; 1229 if (Previous->ParameterCount > 1) 1230 return true; 1231 1232 // Also a nested block if contains a lambda inside function with 1 parameter 1233 return Style.BraceWrapping.BeforeLambdaBody && Current.is(TT_LambdaLSquare); 1234 } 1235 1236 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State, 1237 bool DryRun, bool Newline) { 1238 assert(State.Stack.size()); 1239 const FormatToken &Current = *State.NextToken; 1240 auto &CurrentState = State.Stack.back(); 1241 1242 if (Current.is(TT_CSharpGenericTypeConstraint)) 1243 CurrentState.IsCSharpGenericTypeConstraint = true; 1244 if (Current.isOneOf(tok::comma, TT_BinaryOperator)) 1245 CurrentState.NoLineBreakInOperand = false; 1246 if (Current.isOneOf(TT_InheritanceColon, TT_CSharpGenericTypeConstraintColon)) 1247 CurrentState.AvoidBinPacking = true; 1248 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) { 1249 if (CurrentState.FirstLessLess == 0) 1250 CurrentState.FirstLessLess = State.Column; 1251 else 1252 CurrentState.LastOperatorWrapped = Newline; 1253 } 1254 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless)) 1255 CurrentState.LastOperatorWrapped = Newline; 1256 if (Current.is(TT_ConditionalExpr) && Current.Previous && 1257 !Current.Previous->is(TT_ConditionalExpr)) 1258 CurrentState.LastOperatorWrapped = Newline; 1259 if (Current.is(TT_ArraySubscriptLSquare) && 1260 CurrentState.StartOfArraySubscripts == 0) 1261 CurrentState.StartOfArraySubscripts = State.Column; 1262 1263 auto IsWrappedConditional = [](const FormatToken &Tok) { 1264 if (!(Tok.is(TT_ConditionalExpr) && Tok.is(tok::question))) 1265 return false; 1266 if (Tok.MustBreakBefore) 1267 return true; 1268 1269 const FormatToken *Next = Tok.getNextNonComment(); 1270 return Next && Next->MustBreakBefore; 1271 }; 1272 if (IsWrappedConditional(Current)) 1273 CurrentState.IsWrappedConditional = true; 1274 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question)) 1275 CurrentState.QuestionColumn = State.Column; 1276 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) { 1277 const FormatToken *Previous = Current.Previous; 1278 while (Previous && Previous->isTrailingComment()) 1279 Previous = Previous->Previous; 1280 if (Previous && Previous->is(tok::question)) 1281 CurrentState.QuestionColumn = State.Column; 1282 } 1283 if (!Current.opensScope() && !Current.closesScope() && 1284 !Current.is(TT_PointerOrReference)) 1285 State.LowestLevelOnLine = 1286 std::min(State.LowestLevelOnLine, Current.NestingLevel); 1287 if (Current.isMemberAccess()) 1288 CurrentState.StartOfFunctionCall = !Current.NextOperator ? 0 : State.Column; 1289 if (Current.is(TT_SelectorName)) 1290 CurrentState.ObjCSelectorNameFound = true; 1291 if (Current.is(TT_CtorInitializerColon) && 1292 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) { 1293 // Indent 2 from the column, so: 1294 // SomeClass::SomeClass() 1295 // : First(...), ... 1296 // Next(...) 1297 // ^ line up here. 1298 CurrentState.Indent = State.Column + (Style.BreakConstructorInitializers == 1299 FormatStyle::BCIS_BeforeComma 1300 ? 0 1301 : 2); 1302 CurrentState.NestedBlockIndent = CurrentState.Indent; 1303 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) { 1304 CurrentState.AvoidBinPacking = true; 1305 CurrentState.BreakBeforeParameter = 1306 Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine; 1307 } else { 1308 CurrentState.BreakBeforeParameter = false; 1309 } 1310 } 1311 if (Current.is(TT_CtorInitializerColon) && 1312 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { 1313 CurrentState.Indent = 1314 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1315 CurrentState.NestedBlockIndent = CurrentState.Indent; 1316 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) 1317 CurrentState.AvoidBinPacking = true; 1318 } 1319 if (Current.is(TT_InheritanceColon)) 1320 CurrentState.Indent = 1321 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1322 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline) 1323 CurrentState.NestedBlockIndent = State.Column + Current.ColumnWidth + 1; 1324 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow)) 1325 CurrentState.LastSpace = State.Column; 1326 if (Current.is(TT_RequiresExpression)) 1327 CurrentState.NestedBlockIndent = State.Column; 1328 1329 // Insert scopes created by fake parenthesis. 1330 const FormatToken *Previous = Current.getPreviousNonComment(); 1331 1332 // Add special behavior to support a format commonly used for JavaScript 1333 // closures: 1334 // SomeFunction(function() { 1335 // foo(); 1336 // bar(); 1337 // }, a, b, c); 1338 if (Current.isNot(tok::comment) && !Current.ClosesRequiresClause && 1339 Previous && Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) && 1340 !Previous->is(TT_DictLiteral) && State.Stack.size() > 1 && 1341 !CurrentState.HasMultipleNestedBlocks) { 1342 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline) 1343 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 1344 State.Stack[i].NoLineBreak = true; 1345 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false; 1346 } 1347 if (Previous && (Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr) || 1348 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) && 1349 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)))) { 1350 CurrentState.NestedBlockInlined = 1351 !Newline && hasNestedBlockInlined(Previous, Current, Style); 1352 } 1353 1354 moveStatePastFakeLParens(State, Newline); 1355 moveStatePastScopeCloser(State); 1356 // Do not use CurrentState here, since the two functions before may change the 1357 // Stack. 1358 bool AllowBreak = !State.Stack.back().NoLineBreak && 1359 !State.Stack.back().NoLineBreakInOperand; 1360 moveStatePastScopeOpener(State, Newline); 1361 moveStatePastFakeRParens(State); 1362 1363 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0) 1364 State.StartOfStringLiteral = State.Column + 1; 1365 if (Current.is(TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) 1366 State.StartOfStringLiteral = State.Column + 1; 1367 else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) 1368 State.StartOfStringLiteral = State.Column; 1369 else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) && 1370 !Current.isStringLiteral()) 1371 State.StartOfStringLiteral = 0; 1372 1373 State.Column += Current.ColumnWidth; 1374 State.NextToken = State.NextToken->Next; 1375 1376 unsigned Penalty = 1377 handleEndOfLine(Current, State, DryRun, AllowBreak, Newline); 1378 1379 if (Current.Role) 1380 Current.Role->formatFromToken(State, this, DryRun); 1381 // If the previous has a special role, let it consume tokens as appropriate. 1382 // It is necessary to start at the previous token for the only implemented 1383 // role (comma separated list). That way, the decision whether or not to break 1384 // after the "{" is already done and both options are tried and evaluated. 1385 // FIXME: This is ugly, find a better way. 1386 if (Previous && Previous->Role) 1387 Penalty += Previous->Role->formatAfterToken(State, this, DryRun); 1388 1389 return Penalty; 1390 } 1391 1392 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State, 1393 bool Newline) { 1394 const FormatToken &Current = *State.NextToken; 1395 if (Current.FakeLParens.empty()) 1396 return; 1397 1398 const FormatToken *Previous = Current.getPreviousNonComment(); 1399 1400 // Don't add extra indentation for the first fake parenthesis after 1401 // 'return', assignments, opening <({[, or requires clauses. The indentation 1402 // for these cases is special cased. 1403 bool SkipFirstExtraIndent = 1404 Previous && 1405 (Previous->opensScope() || 1406 Previous->isOneOf(tok::semi, tok::kw_return, TT_RequiresClause) || 1407 (Previous->getPrecedence() == prec::Assignment && 1408 Style.AlignOperands != FormatStyle::OAS_DontAlign) || 1409 Previous->is(TT_ObjCMethodExpr)); 1410 for (const auto &PrecedenceLevel : llvm::reverse(Current.FakeLParens)) { 1411 const auto &CurrentState = State.Stack.back(); 1412 ParenState NewParenState = CurrentState; 1413 NewParenState.Tok = nullptr; 1414 NewParenState.ContainsLineBreak = false; 1415 NewParenState.LastOperatorWrapped = true; 1416 NewParenState.IsChainedConditional = false; 1417 NewParenState.IsWrappedConditional = false; 1418 NewParenState.UnindentOperator = false; 1419 NewParenState.NoLineBreak = 1420 NewParenState.NoLineBreak || CurrentState.NoLineBreakInOperand; 1421 1422 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists. 1423 if (PrecedenceLevel > prec::Comma) 1424 NewParenState.AvoidBinPacking = false; 1425 1426 // Indent from 'LastSpace' unless these are fake parentheses encapsulating 1427 // a builder type call after 'return' or, if the alignment after opening 1428 // brackets is disabled. 1429 if (!Current.isTrailingComment() && 1430 (Style.AlignOperands != FormatStyle::OAS_DontAlign || 1431 PrecedenceLevel < prec::Assignment) && 1432 (!Previous || Previous->isNot(tok::kw_return) || 1433 (Style.Language != FormatStyle::LK_Java && PrecedenceLevel > 0)) && 1434 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign || 1435 PrecedenceLevel != prec::Comma || Current.NestingLevel == 0)) { 1436 NewParenState.Indent = std::max( 1437 std::max(State.Column, NewParenState.Indent), CurrentState.LastSpace); 1438 } 1439 1440 if (Previous && 1441 (Previous->getPrecedence() == prec::Assignment || 1442 Previous->isOneOf(tok::kw_return, TT_RequiresClause) || 1443 (PrecedenceLevel == prec::Conditional && Previous->is(tok::question) && 1444 Previous->is(TT_ConditionalExpr))) && 1445 !Newline) { 1446 // If BreakBeforeBinaryOperators is set, un-indent a bit to account for 1447 // the operator and keep the operands aligned 1448 if (Style.AlignOperands == FormatStyle::OAS_AlignAfterOperator) 1449 NewParenState.UnindentOperator = true; 1450 // Mark indentation as alignment if the expression is aligned. 1451 if (Style.AlignOperands != FormatStyle::OAS_DontAlign) 1452 NewParenState.IsAligned = true; 1453 } 1454 1455 // Do not indent relative to the fake parentheses inserted for "." or "->". 1456 // This is a special case to make the following to statements consistent: 1457 // OuterFunction(InnerFunctionCall( // break 1458 // ParameterToInnerFunction)); 1459 // OuterFunction(SomeObject.InnerFunctionCall( // break 1460 // ParameterToInnerFunction)); 1461 if (PrecedenceLevel > prec::Unknown) 1462 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column); 1463 if (PrecedenceLevel != prec::Conditional && !Current.is(TT_UnaryOperator) && 1464 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 1465 NewParenState.StartOfFunctionCall = State.Column; 1466 1467 // Indent conditional expressions, unless they are chained "else-if" 1468 // conditionals. Never indent expression where the 'operator' is ',', ';' or 1469 // an assignment (i.e. *I <= prec::Assignment) as those have different 1470 // indentation rules. Indent other expression, unless the indentation needs 1471 // to be skipped. 1472 if (PrecedenceLevel == prec::Conditional && Previous && 1473 Previous->is(tok::colon) && Previous->is(TT_ConditionalExpr) && 1474 &PrecedenceLevel == &Current.FakeLParens.back() && 1475 !CurrentState.IsWrappedConditional) { 1476 NewParenState.IsChainedConditional = true; 1477 NewParenState.UnindentOperator = State.Stack.back().UnindentOperator; 1478 } else if (PrecedenceLevel == prec::Conditional || 1479 (!SkipFirstExtraIndent && PrecedenceLevel > prec::Assignment && 1480 !Current.isTrailingComment())) { 1481 NewParenState.Indent += Style.ContinuationIndentWidth; 1482 } 1483 if ((Previous && !Previous->opensScope()) || PrecedenceLevel != prec::Comma) 1484 NewParenState.BreakBeforeParameter = false; 1485 State.Stack.push_back(NewParenState); 1486 SkipFirstExtraIndent = false; 1487 } 1488 } 1489 1490 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) { 1491 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) { 1492 unsigned VariablePos = State.Stack.back().VariablePos; 1493 if (State.Stack.size() == 1) { 1494 // Do not pop the last element. 1495 break; 1496 } 1497 State.Stack.pop_back(); 1498 State.Stack.back().VariablePos = VariablePos; 1499 } 1500 1501 if (State.NextToken->ClosesRequiresClause && Style.IndentRequiresClause) { 1502 // Remove the indentation of the requires clauses (which is not in Indent, 1503 // but in LastSpace). 1504 State.Stack.back().LastSpace -= Style.IndentWidth; 1505 } 1506 } 1507 1508 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State, 1509 bool Newline) { 1510 const FormatToken &Current = *State.NextToken; 1511 if (!Current.opensScope()) 1512 return; 1513 1514 const auto &CurrentState = State.Stack.back(); 1515 1516 // Don't allow '<' or '(' in C# generic type constraints to start new scopes. 1517 if (Current.isOneOf(tok::less, tok::l_paren) && 1518 CurrentState.IsCSharpGenericTypeConstraint) 1519 return; 1520 1521 if (Current.MatchingParen && Current.is(BK_Block)) { 1522 moveStateToNewBlock(State); 1523 return; 1524 } 1525 1526 unsigned NewIndent; 1527 unsigned LastSpace = CurrentState.LastSpace; 1528 bool AvoidBinPacking; 1529 bool BreakBeforeParameter = false; 1530 unsigned NestedBlockIndent = std::max(CurrentState.StartOfFunctionCall, 1531 CurrentState.NestedBlockIndent); 1532 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 1533 opensProtoMessageField(Current, Style)) { 1534 if (Current.opensBlockOrBlockTypeList(Style)) { 1535 NewIndent = Style.IndentWidth + 1536 std::min(State.Column, CurrentState.NestedBlockIndent); 1537 } else { 1538 NewIndent = CurrentState.LastSpace + Style.ContinuationIndentWidth; 1539 } 1540 const FormatToken *NextNoComment = Current.getNextNonComment(); 1541 bool EndsInComma = Current.MatchingParen && 1542 Current.MatchingParen->Previous && 1543 Current.MatchingParen->Previous->is(tok::comma); 1544 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) || 1545 Style.Language == FormatStyle::LK_Proto || 1546 Style.Language == FormatStyle::LK_TextProto || 1547 !Style.BinPackArguments || 1548 (NextNoComment && 1549 NextNoComment->isOneOf(TT_DesignatedInitializerPeriod, 1550 TT_DesignatedInitializerLSquare)); 1551 BreakBeforeParameter = EndsInComma; 1552 if (Current.ParameterCount > 1) 1553 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1); 1554 } else { 1555 NewIndent = 1556 Style.ContinuationIndentWidth + 1557 std::max(CurrentState.LastSpace, CurrentState.StartOfFunctionCall); 1558 1559 // Ensure that different different brackets force relative alignment, e.g.: 1560 // void SomeFunction(vector< // break 1561 // int> v); 1562 // FIXME: We likely want to do this for more combinations of brackets. 1563 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) { 1564 NewIndent = std::max(NewIndent, CurrentState.Indent); 1565 LastSpace = std::max(LastSpace, CurrentState.Indent); 1566 } 1567 1568 bool EndsInComma = 1569 Current.MatchingParen && 1570 Current.MatchingParen->getPreviousNonComment() && 1571 Current.MatchingParen->getPreviousNonComment()->is(tok::comma); 1572 1573 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters 1574 // for backwards compatibility. 1575 bool ObjCBinPackProtocolList = 1576 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto && 1577 Style.BinPackParameters) || 1578 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always; 1579 1580 bool BinPackDeclaration = 1581 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) || 1582 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList); 1583 1584 AvoidBinPacking = 1585 (CurrentState.IsCSharpGenericTypeConstraint) || 1586 (Style.isJavaScript() && EndsInComma) || 1587 (State.Line->MustBeDeclaration && !BinPackDeclaration) || 1588 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) || 1589 (Style.ExperimentalAutoDetectBinPacking && 1590 (Current.is(PPK_OnePerLine) || 1591 (!BinPackInconclusiveFunctions && Current.is(PPK_Inconclusive)))); 1592 1593 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen && 1594 Style.ObjCBreakBeforeNestedBlockParam) { 1595 if (Style.ColumnLimit) { 1596 // If this '[' opens an ObjC call, determine whether all parameters fit 1597 // into one line and put one per line if they don't. 1598 if (getLengthToMatchingParen(Current, State.Stack) + State.Column > 1599 getColumnLimit(State)) 1600 BreakBeforeParameter = true; 1601 } else { 1602 // For ColumnLimit = 0, we have to figure out whether there is or has to 1603 // be a line break within this call. 1604 for (const FormatToken *Tok = &Current; 1605 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) { 1606 if (Tok->MustBreakBefore || 1607 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) { 1608 BreakBeforeParameter = true; 1609 break; 1610 } 1611 } 1612 } 1613 } 1614 1615 if (Style.isJavaScript() && EndsInComma) 1616 BreakBeforeParameter = true; 1617 } 1618 // Generally inherit NoLineBreak from the current scope to nested scope. 1619 // However, don't do this for non-empty nested blocks, dict literals and 1620 // array literals as these follow different indentation rules. 1621 bool NoLineBreak = 1622 Current.Children.empty() && 1623 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) && 1624 (CurrentState.NoLineBreak || CurrentState.NoLineBreakInOperand || 1625 (Current.is(TT_TemplateOpener) && 1626 CurrentState.ContainsUnwrappedBuilder)); 1627 State.Stack.push_back( 1628 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak)); 1629 auto &NewState = State.Stack.back(); 1630 NewState.NestedBlockIndent = NestedBlockIndent; 1631 NewState.BreakBeforeParameter = BreakBeforeParameter; 1632 NewState.HasMultipleNestedBlocks = (Current.BlockParameterCount > 1); 1633 1634 if (Style.BraceWrapping.BeforeLambdaBody && Current.Next != nullptr && 1635 Current.is(tok::l_paren)) { 1636 // Search for any parameter that is a lambda 1637 FormatToken const *next = Current.Next; 1638 while (next != nullptr) { 1639 if (next->is(TT_LambdaLSquare)) { 1640 NewState.HasMultipleNestedBlocks = true; 1641 break; 1642 } 1643 next = next->Next; 1644 } 1645 } 1646 1647 NewState.IsInsideObjCArrayLiteral = Current.is(TT_ArrayInitializerLSquare) && 1648 Current.Previous && 1649 Current.Previous->is(tok::at); 1650 } 1651 1652 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) { 1653 const FormatToken &Current = *State.NextToken; 1654 if (!Current.closesScope()) 1655 return; 1656 1657 // If we encounter a closing ), ], } or >, we can remove a level from our 1658 // stacks. 1659 if (State.Stack.size() > 1 && 1660 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) || 1661 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) || 1662 State.NextToken->is(TT_TemplateCloser) || 1663 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) 1664 State.Stack.pop_back(); 1665 1666 auto &CurrentState = State.Stack.back(); 1667 1668 // Reevaluate whether ObjC message arguments fit into one line. 1669 // If a receiver spans multiple lines, e.g.: 1670 // [[object block:^{ 1671 // return 42; 1672 // }] a:42 b:42]; 1673 // BreakBeforeParameter is calculated based on an incorrect assumption 1674 // (it is checked whether the whole expression fits into one line without 1675 // considering a line break inside a message receiver). 1676 // We check whether arguments fit after receiver scope closer (into the same 1677 // line). 1678 if (CurrentState.BreakBeforeParameter && Current.MatchingParen && 1679 Current.MatchingParen->Previous) { 1680 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous; 1681 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) && 1682 CurrentScopeOpener.MatchingParen) { 1683 int NecessarySpaceInLine = 1684 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) + 1685 CurrentScopeOpener.TotalLength - Current.TotalLength - 1; 1686 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <= 1687 Style.ColumnLimit) 1688 CurrentState.BreakBeforeParameter = false; 1689 } 1690 } 1691 1692 if (Current.is(tok::r_square)) { 1693 // If this ends the array subscript expr, reset the corresponding value. 1694 const FormatToken *NextNonComment = Current.getNextNonComment(); 1695 if (NextNonComment && NextNonComment->isNot(tok::l_square)) 1696 CurrentState.StartOfArraySubscripts = 0; 1697 } 1698 } 1699 1700 void ContinuationIndenter::moveStateToNewBlock(LineState &State) { 1701 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent; 1702 // ObjC block sometimes follow special indentation rules. 1703 unsigned NewIndent = 1704 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace) 1705 ? Style.ObjCBlockIndentWidth 1706 : Style.IndentWidth); 1707 State.Stack.push_back(ParenState(State.NextToken, NewIndent, 1708 State.Stack.back().LastSpace, 1709 /*AvoidBinPacking=*/true, 1710 /*NoLineBreak=*/false)); 1711 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1712 State.Stack.back().BreakBeforeParameter = true; 1713 } 1714 1715 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn, 1716 unsigned TabWidth, 1717 encoding::Encoding Encoding) { 1718 size_t LastNewlinePos = Text.find_last_of("\n"); 1719 if (LastNewlinePos == StringRef::npos) { 1720 return StartColumn + 1721 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding); 1722 } else { 1723 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos), 1724 /*StartColumn=*/0, TabWidth, Encoding); 1725 } 1726 } 1727 1728 unsigned ContinuationIndenter::reformatRawStringLiteral( 1729 const FormatToken &Current, LineState &State, 1730 const FormatStyle &RawStringStyle, bool DryRun, bool Newline) { 1731 unsigned StartColumn = State.Column - Current.ColumnWidth; 1732 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText); 1733 StringRef NewDelimiter = 1734 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language); 1735 if (NewDelimiter.empty()) 1736 NewDelimiter = OldDelimiter; 1737 // The text of a raw string is between the leading 'R"delimiter(' and the 1738 // trailing 'delimiter)"'. 1739 unsigned OldPrefixSize = 3 + OldDelimiter.size(); 1740 unsigned OldSuffixSize = 2 + OldDelimiter.size(); 1741 // We create a virtual text environment which expects a null-terminated 1742 // string, so we cannot use StringRef. 1743 std::string RawText = std::string( 1744 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize)); 1745 if (NewDelimiter != OldDelimiter) { 1746 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the 1747 // raw string. 1748 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str(); 1749 if (StringRef(RawText).contains(CanonicalDelimiterSuffix)) 1750 NewDelimiter = OldDelimiter; 1751 } 1752 1753 unsigned NewPrefixSize = 3 + NewDelimiter.size(); 1754 unsigned NewSuffixSize = 2 + NewDelimiter.size(); 1755 1756 // The first start column is the column the raw text starts after formatting. 1757 unsigned FirstStartColumn = StartColumn + NewPrefixSize; 1758 1759 // The next start column is the intended indentation a line break inside 1760 // the raw string at level 0. It is determined by the following rules: 1761 // - if the content starts on newline, it is one level more than the current 1762 // indent, and 1763 // - if the content does not start on a newline, it is the first start 1764 // column. 1765 // These rules have the advantage that the formatted content both does not 1766 // violate the rectangle rule and visually flows within the surrounding 1767 // source. 1768 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n'; 1769 // If this token is the last parameter (checked by looking if it's followed by 1770 // `)` and is not on a newline, the base the indent off the line's nested 1771 // block indent. Otherwise, base the indent off the arguments indent, so we 1772 // can achieve: 1773 // 1774 // fffffffffff(1, 2, 3, R"pb( 1775 // key1: 1 # 1776 // key2: 2)pb"); 1777 // 1778 // fffffffffff(1, 2, 3, 1779 // R"pb( 1780 // key1: 1 # 1781 // key2: 2 1782 // )pb"); 1783 // 1784 // fffffffffff(1, 2, 3, 1785 // R"pb( 1786 // key1: 1 # 1787 // key2: 2 1788 // )pb", 1789 // 5); 1790 unsigned CurrentIndent = 1791 (!Newline && Current.Next && Current.Next->is(tok::r_paren)) 1792 ? State.Stack.back().NestedBlockIndent 1793 : State.Stack.back().Indent; 1794 unsigned NextStartColumn = ContentStartsOnNewline 1795 ? CurrentIndent + Style.IndentWidth 1796 : FirstStartColumn; 1797 1798 // The last start column is the column the raw string suffix starts if it is 1799 // put on a newline. 1800 // The last start column is the intended indentation of the raw string postfix 1801 // if it is put on a newline. It is determined by the following rules: 1802 // - if the raw string prefix starts on a newline, it is the column where 1803 // that raw string prefix starts, and 1804 // - if the raw string prefix does not start on a newline, it is the current 1805 // indent. 1806 unsigned LastStartColumn = 1807 Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent; 1808 1809 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat( 1810 RawStringStyle, RawText, {tooling::Range(0, RawText.size())}, 1811 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>", 1812 /*Status=*/nullptr); 1813 1814 auto NewCode = applyAllReplacements(RawText, Fixes.first); 1815 tooling::Replacements NoFixes; 1816 if (!NewCode) 1817 return addMultilineToken(Current, State); 1818 if (!DryRun) { 1819 if (NewDelimiter != OldDelimiter) { 1820 // In 'R"delimiter(...', the delimiter starts 2 characters after the start 1821 // of the token. 1822 SourceLocation PrefixDelimiterStart = 1823 Current.Tok.getLocation().getLocWithOffset(2); 1824 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement( 1825 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1826 if (PrefixErr) { 1827 llvm::errs() 1828 << "Failed to update the prefix delimiter of a raw string: " 1829 << llvm::toString(std::move(PrefixErr)) << "\n"; 1830 } 1831 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at 1832 // position length - 1 - |delimiter|. 1833 SourceLocation SuffixDelimiterStart = 1834 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() - 1835 1 - OldDelimiter.size()); 1836 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement( 1837 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1838 if (SuffixErr) { 1839 llvm::errs() 1840 << "Failed to update the suffix delimiter of a raw string: " 1841 << llvm::toString(std::move(SuffixErr)) << "\n"; 1842 } 1843 } 1844 SourceLocation OriginLoc = 1845 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize); 1846 for (const tooling::Replacement &Fix : Fixes.first) { 1847 auto Err = Whitespaces.addReplacement(tooling::Replacement( 1848 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()), 1849 Fix.getLength(), Fix.getReplacementText())); 1850 if (Err) { 1851 llvm::errs() << "Failed to reformat raw string: " 1852 << llvm::toString(std::move(Err)) << "\n"; 1853 } 1854 } 1855 } 1856 unsigned RawLastLineEndColumn = getLastLineEndColumn( 1857 *NewCode, FirstStartColumn, Style.TabWidth, Encoding); 1858 State.Column = RawLastLineEndColumn + NewSuffixSize; 1859 // Since we're updating the column to after the raw string literal here, we 1860 // have to manually add the penalty for the prefix R"delim( over the column 1861 // limit. 1862 unsigned PrefixExcessCharacters = 1863 StartColumn + NewPrefixSize > Style.ColumnLimit 1864 ? StartColumn + NewPrefixSize - Style.ColumnLimit 1865 : 0; 1866 bool IsMultiline = 1867 ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos); 1868 if (IsMultiline) { 1869 // Break before further function parameters on all levels. 1870 for (ParenState &Paren : State.Stack) 1871 Paren.BreakBeforeParameter = true; 1872 } 1873 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter; 1874 } 1875 1876 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current, 1877 LineState &State) { 1878 // Break before further function parameters on all levels. 1879 for (ParenState &Paren : State.Stack) 1880 Paren.BreakBeforeParameter = true; 1881 1882 unsigned ColumnsUsed = State.Column; 1883 // We can only affect layout of the first and the last line, so the penalty 1884 // for all other lines is constant, and we ignore it. 1885 State.Column = Current.LastLineColumnWidth; 1886 1887 if (ColumnsUsed > getColumnLimit(State)) 1888 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State)); 1889 return 0; 1890 } 1891 1892 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current, 1893 LineState &State, bool DryRun, 1894 bool AllowBreak, bool Newline) { 1895 unsigned Penalty = 0; 1896 // Compute the raw string style to use in case this is a raw string literal 1897 // that can be reformatted. 1898 auto RawStringStyle = getRawStringStyle(Current, State); 1899 if (RawStringStyle && !Current.Finalized) { 1900 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun, 1901 Newline); 1902 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) { 1903 // Don't break multi-line tokens other than block comments and raw string 1904 // literals. Instead, just update the state. 1905 Penalty = addMultilineToken(Current, State); 1906 } else if (State.Line->Type != LT_ImportStatement) { 1907 // We generally don't break import statements. 1908 LineState OriginalState = State; 1909 1910 // Whether we force the reflowing algorithm to stay strictly within the 1911 // column limit. 1912 bool Strict = false; 1913 // Whether the first non-strict attempt at reflowing did intentionally 1914 // exceed the column limit. 1915 bool Exceeded = false; 1916 std::tie(Penalty, Exceeded) = breakProtrudingToken( 1917 Current, State, AllowBreak, /*DryRun=*/true, Strict); 1918 if (Exceeded) { 1919 // If non-strict reflowing exceeds the column limit, try whether strict 1920 // reflowing leads to an overall lower penalty. 1921 LineState StrictState = OriginalState; 1922 unsigned StrictPenalty = 1923 breakProtrudingToken(Current, StrictState, AllowBreak, 1924 /*DryRun=*/true, /*Strict=*/true) 1925 .first; 1926 Strict = StrictPenalty <= Penalty; 1927 if (Strict) { 1928 Penalty = StrictPenalty; 1929 State = StrictState; 1930 } 1931 } 1932 if (!DryRun) { 1933 // If we're not in dry-run mode, apply the changes with the decision on 1934 // strictness made above. 1935 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false, 1936 Strict); 1937 } 1938 } 1939 if (State.Column > getColumnLimit(State)) { 1940 unsigned ExcessCharacters = State.Column - getColumnLimit(State); 1941 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters; 1942 } 1943 return Penalty; 1944 } 1945 1946 // Returns the enclosing function name of a token, or the empty string if not 1947 // found. 1948 static StringRef getEnclosingFunctionName(const FormatToken &Current) { 1949 // Look for: 'function(' or 'function<templates>(' before Current. 1950 auto Tok = Current.getPreviousNonComment(); 1951 if (!Tok || !Tok->is(tok::l_paren)) 1952 return ""; 1953 Tok = Tok->getPreviousNonComment(); 1954 if (!Tok) 1955 return ""; 1956 if (Tok->is(TT_TemplateCloser)) { 1957 Tok = Tok->MatchingParen; 1958 if (Tok) 1959 Tok = Tok->getPreviousNonComment(); 1960 } 1961 if (!Tok || !Tok->is(tok::identifier)) 1962 return ""; 1963 return Tok->TokenText; 1964 } 1965 1966 llvm::Optional<FormatStyle> 1967 ContinuationIndenter::getRawStringStyle(const FormatToken &Current, 1968 const LineState &State) { 1969 if (!Current.isStringLiteral()) 1970 return None; 1971 auto Delimiter = getRawStringDelimiter(Current.TokenText); 1972 if (!Delimiter) 1973 return None; 1974 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter); 1975 if (!RawStringStyle && Delimiter->empty()) 1976 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle( 1977 getEnclosingFunctionName(Current)); 1978 if (!RawStringStyle) 1979 return None; 1980 RawStringStyle->ColumnLimit = getColumnLimit(State); 1981 return RawStringStyle; 1982 } 1983 1984 std::unique_ptr<BreakableToken> 1985 ContinuationIndenter::createBreakableToken(const FormatToken &Current, 1986 LineState &State, bool AllowBreak) { 1987 unsigned StartColumn = State.Column - Current.ColumnWidth; 1988 if (Current.isStringLiteral()) { 1989 // FIXME: String literal breaking is currently disabled for C#, Java, Json 1990 // and JavaScript, as it requires strings to be merged using "+" which we 1991 // don't support. 1992 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 1993 Style.isCSharp() || Style.isJson() || !Style.BreakStringLiterals || 1994 !AllowBreak) 1995 return nullptr; 1996 1997 // Don't break string literals inside preprocessor directives (except for 1998 // #define directives, as their contents are stored in separate lines and 1999 // are not affected by this check). 2000 // This way we avoid breaking code with line directives and unknown 2001 // preprocessor directives that contain long string literals. 2002 if (State.Line->Type == LT_PreprocessorDirective) 2003 return nullptr; 2004 // Exempts unterminated string literals from line breaking. The user will 2005 // likely want to terminate the string before any line breaking is done. 2006 if (Current.IsUnterminatedLiteral) 2007 return nullptr; 2008 // Don't break string literals inside Objective-C array literals (doing so 2009 // raises the warning -Wobjc-string-concatenation). 2010 if (State.Stack.back().IsInsideObjCArrayLiteral) 2011 return nullptr; 2012 2013 StringRef Text = Current.TokenText; 2014 StringRef Prefix; 2015 StringRef Postfix; 2016 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'. 2017 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to 2018 // reduce the overhead) for each FormatToken, which is a string, so that we 2019 // don't run multiple checks here on the hot path. 2020 if ((Text.endswith(Postfix = "\"") && 2021 (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") || 2022 Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") || 2023 Text.startswith(Prefix = "u8\"") || 2024 Text.startswith(Prefix = "L\""))) || 2025 (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) { 2026 // We need this to address the case where there is an unbreakable tail 2027 // only if certain other formatting decisions have been taken. The 2028 // UnbreakableTailLength of Current is an overapproximation is that case 2029 // and we need to be correct here. 2030 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State)) 2031 ? 0 2032 : Current.UnbreakableTailLength; 2033 return std::make_unique<BreakableStringLiteral>( 2034 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength, 2035 State.Line->InPPDirective, Encoding, Style); 2036 } 2037 } else if (Current.is(TT_BlockComment)) { 2038 if (!Style.ReflowComments || 2039 // If a comment token switches formatting, like 2040 // /* clang-format on */, we don't want to break it further, 2041 // but we may still want to adjust its indentation. 2042 switchesFormatting(Current)) 2043 return nullptr; 2044 return std::make_unique<BreakableBlockComment>( 2045 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 2046 State.Line->InPPDirective, Encoding, Style, Whitespaces.useCRLF()); 2047 } else if (Current.is(TT_LineComment) && 2048 (Current.Previous == nullptr || 2049 Current.Previous->isNot(TT_ImplicitStringLiteral))) { 2050 bool RegularComments = [&]() { 2051 for (const FormatToken *T = &Current; T && T->is(TT_LineComment); 2052 T = T->Next) 2053 if (!(T->TokenText.startswith("//") || T->TokenText.startswith("#"))) 2054 return false; 2055 return true; 2056 }(); 2057 if (!Style.ReflowComments || 2058 CommentPragmasRegex.match(Current.TokenText.substr(2)) || 2059 switchesFormatting(Current) || !RegularComments) 2060 return nullptr; 2061 return std::make_unique<BreakableLineCommentSection>( 2062 Current, StartColumn, /*InPPDirective=*/false, Encoding, Style); 2063 } 2064 return nullptr; 2065 } 2066 2067 std::pair<unsigned, bool> 2068 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current, 2069 LineState &State, bool AllowBreak, 2070 bool DryRun, bool Strict) { 2071 std::unique_ptr<const BreakableToken> Token = 2072 createBreakableToken(Current, State, AllowBreak); 2073 if (!Token) 2074 return {0, false}; 2075 assert(Token->getLineCount() > 0); 2076 unsigned ColumnLimit = getColumnLimit(State); 2077 if (Current.is(TT_LineComment)) { 2078 // We don't insert backslashes when breaking line comments. 2079 ColumnLimit = Style.ColumnLimit; 2080 } 2081 if (ColumnLimit == 0) { 2082 // To make the rest of the function easier set the column limit to the 2083 // maximum, if there should be no limit. 2084 ColumnLimit = std::numeric_limits<decltype(ColumnLimit)>::max(); 2085 } 2086 if (Current.UnbreakableTailLength >= ColumnLimit) 2087 return {0, false}; 2088 // ColumnWidth was already accounted into State.Column before calling 2089 // breakProtrudingToken. 2090 unsigned StartColumn = State.Column - Current.ColumnWidth; 2091 unsigned NewBreakPenalty = Current.isStringLiteral() 2092 ? Style.PenaltyBreakString 2093 : Style.PenaltyBreakComment; 2094 // Stores whether we intentionally decide to let a line exceed the column 2095 // limit. 2096 bool Exceeded = false; 2097 // Stores whether we introduce a break anywhere in the token. 2098 bool BreakInserted = Token->introducesBreakBeforeToken(); 2099 // Store whether we inserted a new line break at the end of the previous 2100 // logical line. 2101 bool NewBreakBefore = false; 2102 // We use a conservative reflowing strategy. Reflow starts after a line is 2103 // broken or the corresponding whitespace compressed. Reflow ends as soon as a 2104 // line that doesn't get reflown with the previous line is reached. 2105 bool Reflow = false; 2106 // Keep track of where we are in the token: 2107 // Where we are in the content of the current logical line. 2108 unsigned TailOffset = 0; 2109 // The column number we're currently at. 2110 unsigned ContentStartColumn = 2111 Token->getContentStartColumn(0, /*Break=*/false); 2112 // The number of columns left in the current logical line after TailOffset. 2113 unsigned RemainingTokenColumns = 2114 Token->getRemainingLength(0, TailOffset, ContentStartColumn); 2115 // Adapt the start of the token, for example indent. 2116 if (!DryRun) 2117 Token->adaptStartOfLine(0, Whitespaces); 2118 2119 unsigned ContentIndent = 0; 2120 unsigned Penalty = 0; 2121 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column " 2122 << StartColumn << ".\n"); 2123 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount(); 2124 LineIndex != EndIndex; ++LineIndex) { 2125 LLVM_DEBUG(llvm::dbgs() 2126 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n"); 2127 NewBreakBefore = false; 2128 // If we did reflow the previous line, we'll try reflowing again. Otherwise 2129 // we'll start reflowing if the current line is broken or whitespace is 2130 // compressed. 2131 bool TryReflow = Reflow; 2132 // Break the current token until we can fit the rest of the line. 2133 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2134 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: " 2135 << (ContentStartColumn + RemainingTokenColumns) 2136 << ", space: " << ColumnLimit 2137 << ", reflown prefix: " << ContentStartColumn 2138 << ", offset in line: " << TailOffset << "\n"); 2139 // If the current token doesn't fit, find the latest possible split in the 2140 // current line so that breaking at it will be under the column limit. 2141 // FIXME: Use the earliest possible split while reflowing to correctly 2142 // compress whitespace within a line. 2143 BreakableToken::Split Split = 2144 Token->getSplit(LineIndex, TailOffset, ColumnLimit, 2145 ContentStartColumn, CommentPragmasRegex); 2146 if (Split.first == StringRef::npos) { 2147 // No break opportunity - update the penalty and continue with the next 2148 // logical line. 2149 if (LineIndex < EndIndex - 1) 2150 // The last line's penalty is handled in addNextStateToQueue() or when 2151 // calling replaceWhitespaceAfterLastLine below. 2152 Penalty += Style.PenaltyExcessCharacter * 2153 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2154 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n"); 2155 break; 2156 } 2157 assert(Split.first != 0); 2158 2159 if (Token->supportsReflow()) { 2160 // Check whether the next natural split point after the current one can 2161 // still fit the line, either because we can compress away whitespace, 2162 // or because the penalty the excess characters introduce is lower than 2163 // the break penalty. 2164 // We only do this for tokens that support reflowing, and thus allow us 2165 // to change the whitespace arbitrarily (e.g. comments). 2166 // Other tokens, like string literals, can be broken on arbitrary 2167 // positions. 2168 2169 // First, compute the columns from TailOffset to the next possible split 2170 // position. 2171 // For example: 2172 // ColumnLimit: | 2173 // // Some text that breaks 2174 // ^ tail offset 2175 // ^-- split 2176 // ^-------- to split columns 2177 // ^--- next split 2178 // ^--------------- to next split columns 2179 unsigned ToSplitColumns = Token->getRangeLength( 2180 LineIndex, TailOffset, Split.first, ContentStartColumn); 2181 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n"); 2182 2183 BreakableToken::Split NextSplit = Token->getSplit( 2184 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit, 2185 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex); 2186 // Compute the columns necessary to fit the next non-breakable sequence 2187 // into the current line. 2188 unsigned ToNextSplitColumns = 0; 2189 if (NextSplit.first == StringRef::npos) { 2190 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset, 2191 ContentStartColumn); 2192 } else { 2193 ToNextSplitColumns = Token->getRangeLength( 2194 LineIndex, TailOffset, 2195 Split.first + Split.second + NextSplit.first, ContentStartColumn); 2196 } 2197 // Compress the whitespace between the break and the start of the next 2198 // unbreakable sequence. 2199 ToNextSplitColumns = 2200 Token->getLengthAfterCompression(ToNextSplitColumns, Split); 2201 LLVM_DEBUG(llvm::dbgs() 2202 << " ContentStartColumn: " << ContentStartColumn << "\n"); 2203 LLVM_DEBUG(llvm::dbgs() 2204 << " ToNextSplit: " << ToNextSplitColumns << "\n"); 2205 // If the whitespace compression makes us fit, continue on the current 2206 // line. 2207 bool ContinueOnLine = 2208 ContentStartColumn + ToNextSplitColumns <= ColumnLimit; 2209 unsigned ExcessCharactersPenalty = 0; 2210 if (!ContinueOnLine && !Strict) { 2211 // Similarly, if the excess characters' penalty is lower than the 2212 // penalty of introducing a new break, continue on the current line. 2213 ExcessCharactersPenalty = 2214 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) * 2215 Style.PenaltyExcessCharacter; 2216 LLVM_DEBUG(llvm::dbgs() 2217 << " Penalty excess: " << ExcessCharactersPenalty 2218 << "\n break : " << NewBreakPenalty << "\n"); 2219 if (ExcessCharactersPenalty < NewBreakPenalty) { 2220 Exceeded = true; 2221 ContinueOnLine = true; 2222 } 2223 } 2224 if (ContinueOnLine) { 2225 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n"); 2226 // The current line fits after compressing the whitespace - reflow 2227 // the next line into it if possible. 2228 TryReflow = true; 2229 if (!DryRun) 2230 Token->compressWhitespace(LineIndex, TailOffset, Split, 2231 Whitespaces); 2232 // When we continue on the same line, leave one space between content. 2233 ContentStartColumn += ToSplitColumns + 1; 2234 Penalty += ExcessCharactersPenalty; 2235 TailOffset += Split.first + Split.second; 2236 RemainingTokenColumns = Token->getRemainingLength( 2237 LineIndex, TailOffset, ContentStartColumn); 2238 continue; 2239 } 2240 } 2241 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n"); 2242 // Update the ContentIndent only if the current line was not reflown with 2243 // the previous line, since in that case the previous line should still 2244 // determine the ContentIndent. Also never intent the last line. 2245 if (!Reflow) 2246 ContentIndent = Token->getContentIndent(LineIndex); 2247 LLVM_DEBUG(llvm::dbgs() 2248 << " ContentIndent: " << ContentIndent << "\n"); 2249 ContentStartColumn = ContentIndent + Token->getContentStartColumn( 2250 LineIndex, /*Break=*/true); 2251 2252 unsigned NewRemainingTokenColumns = Token->getRemainingLength( 2253 LineIndex, TailOffset + Split.first + Split.second, 2254 ContentStartColumn); 2255 if (NewRemainingTokenColumns == 0) { 2256 // No content to indent. 2257 ContentIndent = 0; 2258 ContentStartColumn = 2259 Token->getContentStartColumn(LineIndex, /*Break=*/true); 2260 NewRemainingTokenColumns = Token->getRemainingLength( 2261 LineIndex, TailOffset + Split.first + Split.second, 2262 ContentStartColumn); 2263 } 2264 2265 // When breaking before a tab character, it may be moved by a few columns, 2266 // but will still be expanded to the next tab stop, so we don't save any 2267 // columns. 2268 if (NewRemainingTokenColumns >= RemainingTokenColumns) { 2269 // FIXME: Do we need to adjust the penalty? 2270 break; 2271 } 2272 2273 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first 2274 << ", " << Split.second << "\n"); 2275 if (!DryRun) 2276 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent, 2277 Whitespaces); 2278 2279 Penalty += NewBreakPenalty; 2280 TailOffset += Split.first + Split.second; 2281 RemainingTokenColumns = NewRemainingTokenColumns; 2282 BreakInserted = true; 2283 NewBreakBefore = true; 2284 } 2285 // In case there's another line, prepare the state for the start of the next 2286 // line. 2287 if (LineIndex + 1 != EndIndex) { 2288 unsigned NextLineIndex = LineIndex + 1; 2289 if (NewBreakBefore) 2290 // After breaking a line, try to reflow the next line into the current 2291 // one once RemainingTokenColumns fits. 2292 TryReflow = true; 2293 if (TryReflow) { 2294 // We decided that we want to try reflowing the next line into the 2295 // current one. 2296 // We will now adjust the state as if the reflow is successful (in 2297 // preparation for the next line), and see whether that works. If we 2298 // decide that we cannot reflow, we will later reset the state to the 2299 // start of the next line. 2300 Reflow = false; 2301 // As we did not continue breaking the line, RemainingTokenColumns is 2302 // known to fit after ContentStartColumn. Adapt ContentStartColumn to 2303 // the position at which we want to format the next line if we do 2304 // actually reflow. 2305 // When we reflow, we need to add a space between the end of the current 2306 // line and the next line's start column. 2307 ContentStartColumn += RemainingTokenColumns + 1; 2308 // Get the split that we need to reflow next logical line into the end 2309 // of the current one; the split will include any leading whitespace of 2310 // the next logical line. 2311 BreakableToken::Split SplitBeforeNext = 2312 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex); 2313 LLVM_DEBUG(llvm::dbgs() 2314 << " Size of reflown text: " << ContentStartColumn 2315 << "\n Potential reflow split: "); 2316 if (SplitBeforeNext.first != StringRef::npos) { 2317 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", " 2318 << SplitBeforeNext.second << "\n"); 2319 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second; 2320 // If the rest of the next line fits into the current line below the 2321 // column limit, we can safely reflow. 2322 RemainingTokenColumns = Token->getRemainingLength( 2323 NextLineIndex, TailOffset, ContentStartColumn); 2324 Reflow = true; 2325 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2326 LLVM_DEBUG(llvm::dbgs() 2327 << " Over limit after reflow, need: " 2328 << (ContentStartColumn + RemainingTokenColumns) 2329 << ", space: " << ColumnLimit 2330 << ", reflown prefix: " << ContentStartColumn 2331 << ", offset in line: " << TailOffset << "\n"); 2332 // If the whole next line does not fit, try to find a point in 2333 // the next line at which we can break so that attaching the part 2334 // of the next line to that break point onto the current line is 2335 // below the column limit. 2336 BreakableToken::Split Split = 2337 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit, 2338 ContentStartColumn, CommentPragmasRegex); 2339 if (Split.first == StringRef::npos) { 2340 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n"); 2341 Reflow = false; 2342 } else { 2343 // Check whether the first split point gets us below the column 2344 // limit. Note that we will execute this split below as part of 2345 // the normal token breaking and reflow logic within the line. 2346 unsigned ToSplitColumns = Token->getRangeLength( 2347 NextLineIndex, TailOffset, Split.first, ContentStartColumn); 2348 if (ContentStartColumn + ToSplitColumns > ColumnLimit) { 2349 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: " 2350 << (ContentStartColumn + ToSplitColumns) 2351 << ", space: " << ColumnLimit); 2352 unsigned ExcessCharactersPenalty = 2353 (ContentStartColumn + ToSplitColumns - ColumnLimit) * 2354 Style.PenaltyExcessCharacter; 2355 if (NewBreakPenalty < ExcessCharactersPenalty) 2356 Reflow = false; 2357 } 2358 } 2359 } 2360 } else { 2361 LLVM_DEBUG(llvm::dbgs() << "not found.\n"); 2362 } 2363 } 2364 if (!Reflow) { 2365 // If we didn't reflow into the next line, the only space to consider is 2366 // the next logical line. Reset our state to match the start of the next 2367 // line. 2368 TailOffset = 0; 2369 ContentStartColumn = 2370 Token->getContentStartColumn(NextLineIndex, /*Break=*/false); 2371 RemainingTokenColumns = Token->getRemainingLength( 2372 NextLineIndex, TailOffset, ContentStartColumn); 2373 // Adapt the start of the token, for example indent. 2374 if (!DryRun) 2375 Token->adaptStartOfLine(NextLineIndex, Whitespaces); 2376 } else { 2377 // If we found a reflow split and have added a new break before the next 2378 // line, we are going to remove the line break at the start of the next 2379 // logical line. For example, here we'll add a new line break after 2380 // 'text', and subsequently delete the line break between 'that' and 2381 // 'reflows'. 2382 // // some text that 2383 // // reflows 2384 // -> 2385 // // some text 2386 // // that reflows 2387 // When adding the line break, we also added the penalty for it, so we 2388 // need to subtract that penalty again when we remove the line break due 2389 // to reflowing. 2390 if (NewBreakBefore) { 2391 assert(Penalty >= NewBreakPenalty); 2392 Penalty -= NewBreakPenalty; 2393 } 2394 if (!DryRun) 2395 Token->reflow(NextLineIndex, Whitespaces); 2396 } 2397 } 2398 } 2399 2400 BreakableToken::Split SplitAfterLastLine = 2401 Token->getSplitAfterLastLine(TailOffset); 2402 if (SplitAfterLastLine.first != StringRef::npos) { 2403 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n"); 2404 2405 // We add the last line's penalty here, since that line is going to be split 2406 // now. 2407 Penalty += Style.PenaltyExcessCharacter * 2408 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2409 2410 if (!DryRun) 2411 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine, 2412 Whitespaces); 2413 ContentStartColumn = 2414 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true); 2415 RemainingTokenColumns = Token->getRemainingLength( 2416 Token->getLineCount() - 1, 2417 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second, 2418 ContentStartColumn); 2419 } 2420 2421 State.Column = ContentStartColumn + RemainingTokenColumns - 2422 Current.UnbreakableTailLength; 2423 2424 if (BreakInserted) { 2425 // If we break the token inside a parameter list, we need to break before 2426 // the next parameter on all levels, so that the next parameter is clearly 2427 // visible. Line comments already introduce a break. 2428 if (Current.isNot(TT_LineComment)) 2429 for (ParenState &Paren : State.Stack) 2430 Paren.BreakBeforeParameter = true; 2431 2432 if (Current.is(TT_BlockComment)) 2433 State.NoContinuation = true; 2434 2435 State.Stack.back().LastSpace = StartColumn; 2436 } 2437 2438 Token->updateNextToken(State); 2439 2440 return {Penalty, Exceeded}; 2441 } 2442 2443 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const { 2444 // In preprocessor directives reserve two chars for trailing " \" 2445 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0); 2446 } 2447 2448 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) { 2449 const FormatToken &Current = *State.NextToken; 2450 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral)) 2451 return false; 2452 // We never consider raw string literals "multiline" for the purpose of 2453 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased 2454 // (see TokenAnnotator::mustBreakBefore(). 2455 if (Current.TokenText.startswith("R\"")) 2456 return false; 2457 if (Current.IsMultiline) 2458 return true; 2459 if (Current.getNextNonComment() && 2460 Current.getNextNonComment()->isStringLiteral()) 2461 return true; // Implicit concatenation. 2462 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals && 2463 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength > 2464 Style.ColumnLimit) 2465 return true; // String will be split. 2466 return false; 2467 } 2468 2469 } // namespace format 2470 } // namespace clang 2471