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.LineContainsContinuedForLoopSection = false; 257 State.NoContinuation = false; 258 State.StartOfStringLiteral = 0; 259 State.StartOfLineLevel = 0; 260 State.LowestLevelOnLine = 0; 261 State.IgnoreStackForComparison = false; 262 263 if (Style.Language == FormatStyle::LK_TextProto) { 264 // We need this in order to deal with the bin packing of text fields at 265 // global scope. 266 State.Stack.back().AvoidBinPacking = true; 267 State.Stack.back().BreakBeforeParameter = true; 268 State.Stack.back().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 assert(&Previous == Current.Previous); 280 if (!Current.CanBreakBefore && !(State.Stack.back().BreakBeforeClosingBrace && 281 Current.closesBlockOrBlockTypeList(Style))) 282 return false; 283 // The opening "{" of a braced list has to be on the same line as the first 284 // element if it is nested in another braced init list or function call. 285 if (!Current.MustBreakBefore && Previous.is(tok::l_brace) && 286 Previous.isNot(TT_DictLiteral) && Previous.is(BK_BracedInit) && 287 Previous.Previous && 288 Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma)) 289 return false; 290 // This prevents breaks like: 291 // ... 292 // SomeParameter, OtherParameter).DoSomething( 293 // ... 294 // As they hide "DoSomething" and are generally bad for readability. 295 if (Previous.opensScope() && Previous.isNot(tok::l_brace) && 296 State.LowestLevelOnLine < State.StartOfLineLevel && 297 State.LowestLevelOnLine < Current.NestingLevel) 298 return false; 299 if (Current.isMemberAccess() && State.Stack.back().ContainsUnwrappedBuilder) 300 return false; 301 302 // Don't create a 'hanging' indent if there are multiple blocks in a single 303 // statement. 304 if (Previous.is(tok::l_brace) && State.Stack.size() > 1 && 305 State.Stack[State.Stack.size() - 2].NestedBlockInlined && 306 State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks) 307 return false; 308 309 // Don't break after very short return types (e.g. "void") as that is often 310 // unexpected. 311 if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) { 312 if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None) 313 return false; 314 } 315 316 // If binary operators are moved to the next line (including commas for some 317 // styles of constructor initializers), that's always ok. 318 if (!Current.isOneOf(TT_BinaryOperator, tok::comma) && 319 State.Stack.back().NoLineBreakInOperand) 320 return false; 321 322 if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr)) 323 return false; 324 325 return !State.Stack.back().NoLineBreak; 326 } 327 328 bool ContinuationIndenter::mustBreak(const LineState &State) { 329 const FormatToken &Current = *State.NextToken; 330 const FormatToken &Previous = *Current.Previous; 331 if (Style.BraceWrapping.BeforeLambdaBody && Current.CanBreakBefore && 332 Current.is(TT_LambdaLBrace) && Previous.isNot(TT_LineComment)) { 333 auto LambdaBodyLength = getLengthToMatchingParen(Current, State.Stack); 334 return (LambdaBodyLength > getColumnLimit(State)); 335 } 336 if (Current.MustBreakBefore || Current.is(TT_InlineASMColon)) 337 return true; 338 if (State.Stack.back().BreakBeforeClosingBrace && 339 Current.closesBlockOrBlockTypeList(Style)) 340 return true; 341 if (State.Stack.back().BreakBeforeClosingParen && Current.is(tok::r_paren)) 342 return true; 343 if (Previous.is(tok::semi) && State.LineContainsContinuedForLoopSection) 344 return true; 345 if (Style.Language == FormatStyle::LK_ObjC && 346 Style.ObjCBreakBeforeNestedBlockParam && 347 Current.ObjCSelectorNameParts > 1 && 348 Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) 349 return true; 350 // Avoid producing inconsistent states by requiring breaks where they are not 351 // permitted for C# generic type constraints. 352 if (State.Stack.back().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 State.Stack.back().BreakBeforeParameter && !Current.isTrailingComment() && 368 !Current.isOneOf(tok::r_paren, tok::r_brace)) 369 return true; 370 if (State.Stack.back().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 State.Stack.back().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 State.Stack.back().ObjCSelectorNameFound && 403 State.Stack.back().BreakBeforeParameter && 404 (Style.ObjCBreakBeforeNestedBlockParam || 405 !Current.startsSequence(TT_SelectorName, tok::colon, tok::caret))) 406 return true; 407 408 unsigned NewLineColumn = getNewLineColumn(State); 409 if (Current.isMemberAccess() && Style.ColumnLimit != 0 && 410 State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit && 411 (State.Column > NewLineColumn || 412 Current.NestingLevel < State.StartOfLineLevel)) 413 return true; 414 415 if (startsSegmentOfBuilderTypeCall(Current) && 416 (State.Stack.back().CallContinuation != 0 || 417 State.Stack.back().BreakBeforeParameter) && 418 // JavaScript is treated different here as there is a frequent pattern: 419 // SomeFunction(function() { 420 // ... 421 // }.bind(...)); 422 // FIXME: We should find a more generic solution to this problem. 423 !(State.Column <= NewLineColumn && Style.isJavaScript()) && 424 !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) 425 return true; 426 427 // If the template declaration spans multiple lines, force wrap before the 428 // function/class declaration 429 if (Previous.ClosesTemplateDeclaration && 430 State.Stack.back().BreakBeforeParameter && Current.CanBreakBefore) 431 return true; 432 433 if (!State.Line->First->is(tok::kw_enum) && State.Column <= NewLineColumn) 434 return false; 435 436 if (Style.AlwaysBreakBeforeMultilineStrings && 437 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth || 438 Previous.is(tok::comma) || Current.NestingLevel < 2) && 439 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at, 440 Keywords.kw_dollar) && 441 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) && 442 nextIsMultilineString(State)) 443 return true; 444 445 // Using CanBreakBefore here and below takes care of the decision whether the 446 // current style uses wrapping before or after operators for the given 447 // operator. 448 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) { 449 // If we need to break somewhere inside the LHS of a binary expression, we 450 // should also break after the operator. Otherwise, the formatting would 451 // hide the operator precedence, e.g. in: 452 // if (aaaaaaaaaaaaaa == 453 // bbbbbbbbbbbbbb && c) {.. 454 // For comparisons, we only apply this rule, if the LHS is a binary 455 // expression itself as otherwise, the line breaks seem superfluous. 456 // We need special cases for ">>" which we have split into two ">" while 457 // lexing in order to make template parsing easier. 458 bool IsComparison = (Previous.getPrecedence() == prec::Relational || 459 Previous.getPrecedence() == prec::Equality || 460 Previous.getPrecedence() == prec::Spaceship) && 461 Previous.Previous && 462 Previous.Previous->isNot(TT_BinaryOperator); // For >>. 463 bool LHSIsBinaryExpr = 464 Previous.Previous && Previous.Previous->EndsBinaryExpression; 465 if ((!IsComparison || LHSIsBinaryExpr) && !Current.isTrailingComment() && 466 Previous.getPrecedence() != prec::Assignment && 467 State.Stack.back().BreakBeforeParameter) 468 return true; 469 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore && 470 State.Stack.back().BreakBeforeParameter) { 471 return true; 472 } 473 474 // Same as above, but for the first "<<" operator. 475 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) && 476 State.Stack.back().BreakBeforeParameter && 477 State.Stack.back().FirstLessLess == 0) 478 return true; 479 480 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) { 481 // Always break after "template <...>"(*) and leading annotations. This is 482 // only for cases where the entire line does not fit on a single line as a 483 // different LineFormatter would be used otherwise. 484 // *: Except when another option interferes with that, like concepts. 485 if (Previous.ClosesTemplateDeclaration) { 486 if (Current.is(tok::kw_concept)) { 487 switch (Style.BreakBeforeConceptDeclarations) { 488 case FormatStyle::BBCDS_Allowed: 489 break; 490 case FormatStyle::BBCDS_Always: 491 return true; 492 case FormatStyle::BBCDS_Never: 493 return false; 494 } 495 } 496 if (Current.is(TT_RequiresClause)) { 497 switch (Style.RequiresClausePosition) { 498 case FormatStyle::RCPS_SingleLine: 499 case FormatStyle::RCPS_WithPreceding: 500 return false; 501 default: 502 return true; 503 } 504 } 505 return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No; 506 } 507 if (Previous.is(TT_FunctionAnnotationRParen) && 508 State.Line->Type != LT_PreprocessorDirective) 509 return true; 510 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) && 511 Current.isNot(TT_LeadingJavaAnnotation)) 512 return true; 513 } 514 515 if (Style.isJavaScript() && Previous.is(tok::r_paren) && 516 Previous.is(TT_JavaAnnotation)) { 517 // Break after the closing parenthesis of TypeScript decorators before 518 // functions, getters and setters. 519 static const llvm::StringSet<> BreakBeforeDecoratedTokens = {"get", "set", 520 "function"}; 521 if (BreakBeforeDecoratedTokens.contains(Current.TokenText)) 522 return true; 523 } 524 525 // If the return type spans multiple lines, wrap before the function name. 526 if (((Current.is(TT_FunctionDeclarationName) && 527 // Don't break before a C# function when no break after return type 528 (!Style.isCSharp() || 529 Style.AlwaysBreakAfterReturnType != FormatStyle::RTBS_None) && 530 // Don't always break between a JavaScript `function` and the function 531 // name. 532 !Style.isJavaScript()) || 533 (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) && 534 !Previous.is(tok::kw_template) && State.Stack.back().BreakBeforeParameter) 535 return true; 536 537 // The following could be precomputed as they do not depend on the state. 538 // However, as they should take effect only if the UnwrappedLine does not fit 539 // into the ColumnLimit, they are checked here in the ContinuationIndenter. 540 if (Style.ColumnLimit != 0 && Previous.is(BK_Block) && 541 Previous.is(tok::l_brace) && !Current.isOneOf(tok::r_brace, tok::comment)) 542 return true; 543 544 if (Current.is(tok::lessless) && 545 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") || 546 (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") || 547 Previous.TokenText == "\'\\n\'")))) 548 return true; 549 550 if (Previous.is(TT_BlockComment) && Previous.IsMultiline) 551 return true; 552 553 if (State.NoContinuation) 554 return true; 555 556 return false; 557 } 558 559 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline, 560 bool DryRun, 561 unsigned ExtraSpaces) { 562 const FormatToken &Current = *State.NextToken; 563 assert(State.NextToken->Previous); 564 const FormatToken &Previous = *State.NextToken->Previous; 565 566 assert(!State.Stack.empty()); 567 State.NoContinuation = false; 568 569 if ((Current.is(TT_ImplicitStringLiteral) && 570 (Previous.Tok.getIdentifierInfo() == nullptr || 571 Previous.Tok.getIdentifierInfo()->getPPKeywordID() == 572 tok::pp_not_keyword))) { 573 unsigned EndColumn = 574 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd()); 575 if (Current.LastNewlineOffset != 0) { 576 // If there is a newline within this token, the final column will solely 577 // determined by the current end column. 578 State.Column = EndColumn; 579 } else { 580 unsigned StartColumn = 581 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin()); 582 assert(EndColumn >= StartColumn); 583 State.Column += EndColumn - StartColumn; 584 } 585 moveStateToNextToken(State, DryRun, /*Newline=*/false); 586 return 0; 587 } 588 589 unsigned Penalty = 0; 590 if (Newline) 591 Penalty = addTokenOnNewLine(State, DryRun); 592 else 593 addTokenOnCurrentLine(State, DryRun, ExtraSpaces); 594 595 return moveStateToNextToken(State, DryRun, Newline) + Penalty; 596 } 597 598 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun, 599 unsigned ExtraSpaces) { 600 FormatToken &Current = *State.NextToken; 601 assert(State.NextToken->Previous); 602 const FormatToken &Previous = *State.NextToken->Previous; 603 604 if (Current.is(tok::equal) && 605 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) && 606 State.Stack.back().VariablePos == 0) { 607 State.Stack.back().VariablePos = State.Column; 608 // Move over * and & if they are bound to the variable name. 609 const FormatToken *Tok = &Previous; 610 while (Tok && State.Stack.back().VariablePos >= Tok->ColumnWidth) { 611 State.Stack.back().VariablePos -= Tok->ColumnWidth; 612 if (Tok->SpacesRequiredBefore != 0) 613 break; 614 Tok = Tok->Previous; 615 } 616 if (Previous.PartOfMultiVariableDeclStmt) 617 State.Stack.back().LastSpace = State.Stack.back().VariablePos; 618 } 619 620 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces; 621 622 // Indent preprocessor directives after the hash if required. 623 int PPColumnCorrection = 0; 624 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && 625 Previous.is(tok::hash) && State.FirstIndent > 0 && 626 (State.Line->Type == LT_PreprocessorDirective || 627 State.Line->Type == LT_ImportStatement)) { 628 Spaces += State.FirstIndent; 629 630 // For preprocessor indent with tabs, State.Column will be 1 because of the 631 // hash. This causes second-level indents onward to have an extra space 632 // after the tabs. We avoid this misalignment by subtracting 1 from the 633 // column value passed to replaceWhitespace(). 634 if (Style.UseTab != FormatStyle::UT_Never) 635 PPColumnCorrection = -1; 636 } 637 638 if (!DryRun) 639 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces, 640 State.Column + Spaces + PPColumnCorrection); 641 642 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance 643 // declaration unless there is multiple inheritance. 644 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && 645 Current.is(TT_InheritanceColon)) 646 State.Stack.back().NoLineBreak = true; 647 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon && 648 Previous.is(TT_InheritanceColon)) 649 State.Stack.back().NoLineBreak = true; 650 651 if (Current.is(TT_SelectorName) && 652 !State.Stack.back().ObjCSelectorNameFound) { 653 unsigned MinIndent = 654 std::max(State.FirstIndent + Style.ContinuationIndentWidth, 655 State.Stack.back().Indent); 656 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth; 657 if (Current.LongestObjCSelectorName == 0) 658 State.Stack.back().AlignColons = false; 659 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos) 660 State.Stack.back().ColonPos = MinIndent + Current.LongestObjCSelectorName; 661 else 662 State.Stack.back().ColonPos = FirstColonPos; 663 } 664 665 // In "AlwaysBreak" or "BlockIndent" mode, enforce wrapping directly after the 666 // parenthesis by disallowing any further line breaks if there is no line 667 // break after the opening parenthesis. Don't break if it doesn't conserve 668 // columns. 669 if ((Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak || 670 Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent) && 671 (Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) || 672 (Previous.is(tok::l_brace) && Previous.isNot(BK_Block) && 673 Style.Cpp11BracedListStyle)) && 674 State.Column > getNewLineColumn(State) && 675 (!Previous.Previous || !Previous.Previous->isOneOf( 676 tok::kw_for, tok::kw_while, tok::kw_switch)) && 677 // Don't do this for simple (no expressions) one-argument function calls 678 // as that feels like needlessly wasting whitespace, e.g.: 679 // 680 // caaaaaaaaaaaall( 681 // caaaaaaaaaaaall( 682 // caaaaaaaaaaaall( 683 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa)))); 684 Current.FakeLParens.size() > 0 && 685 Current.FakeLParens.back() > prec::Unknown) 686 State.Stack.back().NoLineBreak = true; 687 if (Previous.is(TT_TemplateString) && Previous.opensScope()) 688 State.Stack.back().NoLineBreak = true; 689 690 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign && 691 !State.Stack.back().IsCSharpGenericTypeConstraint && 692 Previous.opensScope() && Previous.isNot(TT_ObjCMethodExpr) && 693 Previous.isNot(TT_RequiresClause) && 694 (Current.isNot(TT_LineComment) || Previous.is(BK_BracedInit))) { 695 State.Stack.back().Indent = State.Column + Spaces; 696 State.Stack.back().IsAligned = true; 697 } 698 if (State.Stack.back().AvoidBinPacking && startsNextParameter(Current, Style)) 699 State.Stack.back().NoLineBreak = true; 700 if (startsSegmentOfBuilderTypeCall(Current) && 701 State.Column > getNewLineColumn(State)) 702 State.Stack.back().ContainsUnwrappedBuilder = true; 703 704 if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java) 705 State.Stack.back().NoLineBreak = true; 706 if (Current.isMemberAccess() && Previous.is(tok::r_paren) && 707 (Previous.MatchingParen && 708 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) 709 // If there is a function call with long parameters, break before trailing 710 // calls. This prevents things like: 711 // EXPECT_CALL(SomeLongParameter).Times( 712 // 2); 713 // We don't want to do this for short parameters as they can just be 714 // indexes. 715 State.Stack.back().NoLineBreak = true; 716 717 // Don't allow the RHS of an operator to be split over multiple lines unless 718 // there is a line-break right after the operator. 719 // Exclude relational operators, as there, it is always more desirable to 720 // have the LHS 'left' of the RHS. 721 const FormatToken *P = Current.getPreviousNonComment(); 722 if (!Current.is(tok::comment) && P && 723 (P->isOneOf(TT_BinaryOperator, tok::comma) || 724 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) && 725 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) && 726 P->getPrecedence() != prec::Assignment && 727 P->getPrecedence() != prec::Relational && 728 P->getPrecedence() != prec::Spaceship) { 729 bool BreakBeforeOperator = 730 P->MustBreakBefore || P->is(tok::lessless) || 731 (P->is(TT_BinaryOperator) && 732 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) || 733 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators); 734 // Don't do this if there are only two operands. In these cases, there is 735 // always a nice vertical separation between them and the extra line break 736 // does not help. 737 bool HasTwoOperands = 738 P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr); 739 if ((!BreakBeforeOperator && 740 !(HasTwoOperands && 741 Style.AlignOperands != FormatStyle::OAS_DontAlign)) || 742 (!State.Stack.back().LastOperatorWrapped && BreakBeforeOperator)) 743 State.Stack.back().NoLineBreakInOperand = true; 744 } 745 746 State.Column += Spaces; 747 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) && 748 Previous.Previous && 749 (Previous.Previous->is(tok::kw_for) || Previous.Previous->isIf())) { 750 // Treat the condition inside an if as if it was a second function 751 // parameter, i.e. let nested calls have a continuation indent. 752 State.Stack.back().LastSpace = State.Column; 753 State.Stack.back().NestedBlockIndent = State.Column; 754 } else if (!Current.isOneOf(tok::comment, tok::caret) && 755 ((Previous.is(tok::comma) && 756 !Previous.is(TT_OverloadedOperator)) || 757 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) { 758 State.Stack.back().LastSpace = State.Column; 759 } else if (Previous.is(TT_CtorInitializerColon) && 760 Style.BreakConstructorInitializers == 761 FormatStyle::BCIS_AfterColon) { 762 State.Stack.back().Indent = State.Column; 763 State.Stack.back().LastSpace = State.Column; 764 } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr, 765 TT_CtorInitializerColon)) && 766 ((Previous.getPrecedence() != prec::Assignment && 767 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 || 768 Previous.NextOperator)) || 769 Current.StartsBinaryExpression)) { 770 // Indent relative to the RHS of the expression unless this is a simple 771 // assignment without binary expression on the RHS. Also indent relative to 772 // unary operators and the colons of constructor initializers. 773 if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None) 774 State.Stack.back().LastSpace = State.Column; 775 } else if (Previous.is(TT_InheritanceColon)) { 776 State.Stack.back().Indent = State.Column; 777 State.Stack.back().LastSpace = State.Column; 778 } else if (Current.is(TT_CSharpGenericTypeConstraintColon)) { 779 State.Stack.back().ColonPos = State.Column; 780 } else if (Previous.opensScope()) { 781 // If a function has a trailing call, indent all parameters from the 782 // opening parenthesis. This avoids confusing indents like: 783 // OuterFunction(InnerFunctionCall( // break 784 // ParameterToInnerFunction)) // break 785 // .SecondInnerFunctionCall(); 786 bool HasTrailingCall = false; 787 if (Previous.MatchingParen) { 788 const FormatToken *Next = Previous.MatchingParen->getNextNonComment(); 789 HasTrailingCall = Next && Next->isMemberAccess(); 790 } 791 if (HasTrailingCall && State.Stack.size() > 1 && 792 State.Stack[State.Stack.size() - 2].CallContinuation == 0) 793 State.Stack.back().LastSpace = State.Column; 794 } 795 } 796 797 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State, 798 bool DryRun) { 799 FormatToken &Current = *State.NextToken; 800 assert(State.NextToken->Previous); 801 const FormatToken &Previous = *State.NextToken->Previous; 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 (!State.Stack.back().ContainsLineBreak) 814 Penalty += 15; 815 State.Stack.back().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) && 823 State.Stack.back().FirstLessLess == 0 && 824 (State.Column <= Style.ColumnLimit / 3 || 825 State.Stack.back().BreakBeforeParameter)) 826 Penalty += Style.PenaltyBreakFirstLessLess; 827 828 State.Column = getNewLineColumn(State); 829 830 // Add Penalty proportional to amount of whitespace away from FirstColumn 831 // This tends to penalize several lines that are far-right indented, 832 // and prefers a line-break prior to such a block, e.g: 833 // 834 // Constructor() : 835 // member(value), looooooooooooooooong_member( 836 // looooooooooong_call(param_1, param_2, param_3)) 837 // would then become 838 // Constructor() : 839 // member(value), 840 // looooooooooooooooong_member( 841 // looooooooooong_call(param_1, param_2, param_3)) 842 if (State.Column > State.FirstIndent) 843 Penalty += 844 Style.PenaltyIndentedWhitespace * (State.Column - State.FirstIndent); 845 846 // Indent nested blocks relative to this column, unless in a very specific 847 // JavaScript special case where: 848 // 849 // var loooooong_name = 850 // function() { 851 // // code 852 // } 853 // 854 // is common and should be formatted like a free-standing function. The same 855 // goes for wrapping before the lambda return type arrow. 856 if (!Current.is(TT_LambdaArrow) && 857 (!Style.isJavaScript() || Current.NestingLevel != 0 || 858 !PreviousNonComment || !PreviousNonComment->is(tok::equal) || 859 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) 860 State.Stack.back().NestedBlockIndent = State.Column; 861 862 if (NextNonComment->isMemberAccess()) { 863 if (State.Stack.back().CallContinuation == 0) 864 State.Stack.back().CallContinuation = State.Column; 865 } else if (NextNonComment->is(TT_SelectorName)) { 866 if (!State.Stack.back().ObjCSelectorNameFound) { 867 if (NextNonComment->LongestObjCSelectorName == 0) { 868 State.Stack.back().AlignColons = false; 869 } else { 870 State.Stack.back().ColonPos = 871 (shouldIndentWrappedSelectorName(Style, State.Line->Type) 872 ? std::max(State.Stack.back().Indent, 873 State.FirstIndent + Style.ContinuationIndentWidth) 874 : State.Stack.back().Indent) + 875 std::max(NextNonComment->LongestObjCSelectorName, 876 NextNonComment->ColumnWidth); 877 } 878 } else if (State.Stack.back().AlignColons && 879 State.Stack.back().ColonPos <= NextNonComment->ColumnWidth) { 880 State.Stack.back().ColonPos = State.Column + NextNonComment->ColumnWidth; 881 } 882 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 883 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) { 884 // FIXME: This is hacky, find a better way. The problem is that in an ObjC 885 // method expression, the block should be aligned to the line starting it, 886 // e.g.: 887 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason 888 // ^(int *i) { 889 // // ... 890 // }]; 891 // Thus, we set LastSpace of the next higher NestingLevel, to which we move 892 // when we consume all of the "}"'s FakeRParens at the "{". 893 if (State.Stack.size() > 1) 894 State.Stack[State.Stack.size() - 2].LastSpace = 895 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 896 Style.ContinuationIndentWidth; 897 } 898 899 if ((PreviousNonComment && 900 PreviousNonComment->isOneOf(tok::comma, tok::semi) && 901 !State.Stack.back().AvoidBinPacking) || 902 Previous.is(TT_BinaryOperator)) 903 State.Stack.back().BreakBeforeParameter = false; 904 if (PreviousNonComment && 905 (PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) || 906 PreviousNonComment->ClosesRequiresClause) && 907 Current.NestingLevel == 0) 908 State.Stack.back().BreakBeforeParameter = false; 909 if (NextNonComment->is(tok::question) || 910 (PreviousNonComment && PreviousNonComment->is(tok::question))) 911 State.Stack.back().BreakBeforeParameter = true; 912 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore) 913 State.Stack.back().BreakBeforeParameter = false; 914 915 if (!DryRun) { 916 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1; 917 if (Current.is(tok::r_brace) && Current.MatchingParen && 918 // Only strip trailing empty lines for l_braces that have children, i.e. 919 // for function expressions (lambdas, arrows, etc). 920 !Current.MatchingParen->Children.empty()) { 921 // lambdas and arrow functions are expressions, thus their r_brace is not 922 // on its own line, and thus not covered by UnwrappedLineFormatter's logic 923 // about removing empty lines on closing blocks. Special case them here. 924 MaxEmptyLinesToKeep = 1; 925 } 926 unsigned Newlines = 927 std::max(1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep)); 928 bool ContinuePPDirective = 929 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement; 930 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column, 931 State.Stack.back().IsAligned, 932 ContinuePPDirective); 933 } 934 935 if (!Current.isTrailingComment()) 936 State.Stack.back().LastSpace = State.Column; 937 if (Current.is(tok::lessless)) 938 // If we are breaking before a "<<", we always want to indent relative to 939 // RHS. This is necessary only for "<<", as we special-case it and don't 940 // always indent relative to the RHS. 941 State.Stack.back().LastSpace += 3; // 3 -> width of "<< ". 942 943 State.StartOfLineLevel = Current.NestingLevel; 944 State.LowestLevelOnLine = Current.NestingLevel; 945 946 // Any break on this level means that the parent level has been broken 947 // and we need to avoid bin packing there. 948 bool NestedBlockSpecialCase = 949 (!Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 && 950 State.Stack[State.Stack.size() - 2].NestedBlockInlined) || 951 (Style.Language == FormatStyle::LK_ObjC && Current.is(tok::r_brace) && 952 State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam); 953 // Do not force parameter break for statements with requires expressions. 954 NestedBlockSpecialCase = 955 NestedBlockSpecialCase || 956 (Current.MatchingParen && 957 Current.MatchingParen->is(TT_RequiresExpressionLBrace)); 958 if (!NestedBlockSpecialCase) 959 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 960 State.Stack[i].BreakBeforeParameter = true; 961 962 if (PreviousNonComment && 963 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) && 964 ((PreviousNonComment->isNot(TT_TemplateCloser) && 965 !PreviousNonComment->ClosesRequiresClause) || 966 Current.NestingLevel != 0) && 967 !PreviousNonComment->isOneOf( 968 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation, 969 TT_LeadingJavaAnnotation) && 970 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope()) 971 State.Stack.back().BreakBeforeParameter = true; 972 973 // If we break after { or the [ of an array initializer, we should also break 974 // before the corresponding } or ]. 975 if (PreviousNonComment && 976 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 977 opensProtoMessageField(*PreviousNonComment, Style))) 978 State.Stack.back().BreakBeforeClosingBrace = true; 979 980 if (PreviousNonComment && PreviousNonComment->is(tok::l_paren)) 981 State.Stack.back().BreakBeforeClosingParen = 982 Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent; 983 984 if (State.Stack.back().AvoidBinPacking) { 985 // If we are breaking after '(', '{', '<', or this is the break after a ':' 986 // to start a member initializater list in a constructor, this should not 987 // be considered bin packing unless the relevant AllowAll option is false or 988 // this is a dict/object literal. 989 bool PreviousIsBreakingCtorInitializerColon = 990 Previous.is(TT_CtorInitializerColon) && 991 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; 992 if (!(Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) || 993 PreviousIsBreakingCtorInitializerColon) || 994 (!Style.AllowAllParametersOfDeclarationOnNextLine && 995 State.Line->MustBeDeclaration) || 996 (!Style.AllowAllArgumentsOnNextLine && 997 !State.Line->MustBeDeclaration) || 998 (Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine && 999 PreviousIsBreakingCtorInitializerColon) || 1000 Previous.is(TT_DictLiteral)) 1001 State.Stack.back().BreakBeforeParameter = true; 1002 1003 // If we are breaking after a ':' to start a member initializer list, 1004 // and we allow all arguments on the next line, we should not break 1005 // before the next parameter. 1006 if (PreviousIsBreakingCtorInitializerColon && 1007 Style.PackConstructorInitializers == FormatStyle::PCIS_NextLine) 1008 State.Stack.back().BreakBeforeParameter = false; 1009 } 1010 1011 return Penalty; 1012 } 1013 1014 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) { 1015 if (!State.NextToken || !State.NextToken->Previous) 1016 return 0; 1017 1018 FormatToken &Current = *State.NextToken; 1019 1020 if (State.Stack.back().IsCSharpGenericTypeConstraint && 1021 Current.isNot(TT_CSharpGenericTypeConstraint)) 1022 return State.Stack.back().ColonPos + 2; 1023 1024 const FormatToken &Previous = *Current.Previous; 1025 // If we are continuing an expression, we want to use the continuation indent. 1026 unsigned ContinuationIndent = 1027 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 1028 Style.ContinuationIndentWidth; 1029 const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); 1030 const FormatToken *NextNonComment = Previous.getNextNonComment(); 1031 if (!NextNonComment) 1032 NextNonComment = &Current; 1033 1034 // Java specific bits. 1035 if (Style.Language == FormatStyle::LK_Java && 1036 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) 1037 return std::max(State.Stack.back().LastSpace, 1038 State.Stack.back().Indent + Style.ContinuationIndentWidth); 1039 1040 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths && 1041 State.Line->First->is(tok::kw_enum)) 1042 return (Style.IndentWidth * State.Line->First->IndentLevel) + 1043 Style.IndentWidth; 1044 1045 if (NextNonComment->is(tok::l_brace) && NextNonComment->is(BK_Block)) 1046 return Current.NestingLevel == 0 ? State.FirstIndent 1047 : State.Stack.back().Indent; 1048 if ((Current.isOneOf(tok::r_brace, tok::r_square) || 1049 (Current.is(tok::greater) && 1050 (Style.Language == FormatStyle::LK_Proto || 1051 Style.Language == FormatStyle::LK_TextProto))) && 1052 State.Stack.size() > 1) { 1053 if (Current.closesBlockOrBlockTypeList(Style)) 1054 return State.Stack[State.Stack.size() - 2].NestedBlockIndent; 1055 if (Current.MatchingParen && Current.MatchingParen->is(BK_BracedInit)) 1056 return State.Stack[State.Stack.size() - 2].LastSpace; 1057 return State.FirstIndent; 1058 } 1059 // Indent a closing parenthesis at the previous level if followed by a semi, 1060 // const, or opening brace. This allows indentations such as: 1061 // foo( 1062 // a, 1063 // ); 1064 // int Foo::getter( 1065 // // 1066 // ) const { 1067 // return foo; 1068 // } 1069 // function foo( 1070 // a, 1071 // ) { 1072 // code(); // 1073 // } 1074 if (Current.is(tok::r_paren) && State.Stack.size() > 1 && 1075 (!Current.Next || 1076 Current.Next->isOneOf(tok::semi, tok::kw_const, tok::l_brace))) 1077 return State.Stack[State.Stack.size() - 2].LastSpace; 1078 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent && 1079 Current.is(tok::r_paren) && State.Stack.size() > 1) 1080 return State.Stack[State.Stack.size() - 2].LastSpace; 1081 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope()) 1082 return State.Stack[State.Stack.size() - 2].LastSpace; 1083 if (Current.is(tok::identifier) && Current.Next && 1084 (Current.Next->is(TT_DictLiteral) || 1085 ((Style.Language == FormatStyle::LK_Proto || 1086 Style.Language == FormatStyle::LK_TextProto) && 1087 Current.Next->isOneOf(tok::less, tok::l_brace)))) 1088 return State.Stack.back().Indent; 1089 if (NextNonComment->is(TT_ObjCStringLiteral) && 1090 State.StartOfStringLiteral != 0) 1091 return State.StartOfStringLiteral - 1; 1092 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0) 1093 return State.StartOfStringLiteral; 1094 if (NextNonComment->is(tok::lessless) && 1095 State.Stack.back().FirstLessLess != 0) 1096 return State.Stack.back().FirstLessLess; 1097 if (NextNonComment->isMemberAccess()) { 1098 if (State.Stack.back().CallContinuation == 0) 1099 return ContinuationIndent; 1100 return State.Stack.back().CallContinuation; 1101 } 1102 if (State.Stack.back().QuestionColumn != 0 && 1103 ((NextNonComment->is(tok::colon) && 1104 NextNonComment->is(TT_ConditionalExpr)) || 1105 Previous.is(TT_ConditionalExpr))) { 1106 if (((NextNonComment->is(tok::colon) && NextNonComment->Next && 1107 !NextNonComment->Next->FakeLParens.empty() && 1108 NextNonComment->Next->FakeLParens.back() == prec::Conditional) || 1109 (Previous.is(tok::colon) && !Current.FakeLParens.empty() && 1110 Current.FakeLParens.back() == prec::Conditional)) && 1111 !State.Stack.back().IsWrappedConditional) { 1112 // NOTE: we may tweak this slightly: 1113 // * not remove the 'lead' ContinuationIndentWidth 1114 // * always un-indent by the operator when 1115 // BreakBeforeTernaryOperators=true 1116 unsigned Indent = State.Stack.back().Indent; 1117 if (Style.AlignOperands != FormatStyle::OAS_DontAlign) 1118 Indent -= Style.ContinuationIndentWidth; 1119 if (Style.BreakBeforeTernaryOperators && 1120 State.Stack.back().UnindentOperator) 1121 Indent -= 2; 1122 return Indent; 1123 } 1124 return State.Stack.back().QuestionColumn; 1125 } 1126 if (Previous.is(tok::comma) && State.Stack.back().VariablePos != 0) 1127 return State.Stack.back().VariablePos; 1128 if (Current.is(TT_RequiresClause)) { 1129 if (Style.IndentRequiresClause) 1130 return State.Stack.back().Indent + Style.IndentWidth; 1131 switch (Style.RequiresClausePosition) { 1132 case FormatStyle::RCPS_OwnLine: 1133 case FormatStyle::RCPS_WithFollowing: 1134 return State.Stack.back().Indent; 1135 default: 1136 break; 1137 } 1138 } 1139 if ((PreviousNonComment && 1140 (PreviousNonComment->ClosesTemplateDeclaration || 1141 PreviousNonComment->ClosesRequiresClause || 1142 PreviousNonComment->isOneOf( 1143 TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen, 1144 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) || 1145 (!Style.IndentWrappedFunctionNames && 1146 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) 1147 return std::max(State.Stack.back().LastSpace, State.Stack.back().Indent); 1148 if (NextNonComment->is(TT_SelectorName)) { 1149 if (!State.Stack.back().ObjCSelectorNameFound) { 1150 unsigned MinIndent = State.Stack.back().Indent; 1151 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) 1152 MinIndent = std::max(MinIndent, 1153 State.FirstIndent + Style.ContinuationIndentWidth); 1154 // If LongestObjCSelectorName is 0, we are indenting the first 1155 // part of an ObjC selector (or a selector component which is 1156 // not colon-aligned due to block formatting). 1157 // 1158 // Otherwise, we are indenting a subsequent part of an ObjC 1159 // selector which should be colon-aligned to the longest 1160 // component of the ObjC selector. 1161 // 1162 // In either case, we want to respect Style.IndentWrappedFunctionNames. 1163 return MinIndent + 1164 std::max(NextNonComment->LongestObjCSelectorName, 1165 NextNonComment->ColumnWidth) - 1166 NextNonComment->ColumnWidth; 1167 } 1168 if (!State.Stack.back().AlignColons) 1169 return State.Stack.back().Indent; 1170 if (State.Stack.back().ColonPos > NextNonComment->ColumnWidth) 1171 return State.Stack.back().ColonPos - NextNonComment->ColumnWidth; 1172 return State.Stack.back().Indent; 1173 } 1174 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr)) 1175 return State.Stack.back().ColonPos; 1176 if (NextNonComment->is(TT_ArraySubscriptLSquare)) { 1177 if (State.Stack.back().StartOfArraySubscripts != 0) 1178 return State.Stack.back().StartOfArraySubscripts; 1179 else if (Style.isCSharp()) // C# allows `["key"] = value` inside object 1180 // initializers. 1181 return State.Stack.back().Indent; 1182 return ContinuationIndent; 1183 } 1184 1185 // This ensure that we correctly format ObjC methods calls without inputs, 1186 // i.e. where the last element isn't selector like: [callee method]; 1187 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 && 1188 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) 1189 return State.Stack.back().Indent; 1190 1191 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) || 1192 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) 1193 return ContinuationIndent; 1194 if (PreviousNonComment && PreviousNonComment->is(tok::colon) && 1195 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) 1196 return ContinuationIndent; 1197 if (NextNonComment->is(TT_CtorInitializerComma)) 1198 return State.Stack.back().Indent; 1199 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && 1200 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) 1201 return State.Stack.back().Indent; 1202 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) && 1203 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) 1204 return State.Stack.back().Indent; 1205 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon, 1206 TT_InheritanceComma)) 1207 return State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1208 if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() && 1209 !Current.isOneOf(tok::colon, tok::comment)) 1210 return ContinuationIndent; 1211 if (Current.is(TT_ProtoExtensionLSquare)) 1212 return State.Stack.back().Indent; 1213 if (Current.isBinaryOperator() && State.Stack.back().UnindentOperator) 1214 return State.Stack.back().Indent - Current.Tok.getLength() - 1215 Current.SpacesRequiredBefore; 1216 if (Current.isOneOf(tok::comment, TT_BlockComment, TT_LineComment) && 1217 NextNonComment->isBinaryOperator() && State.Stack.back().UnindentOperator) 1218 return State.Stack.back().Indent - NextNonComment->Tok.getLength() - 1219 NextNonComment->SpacesRequiredBefore; 1220 if (State.Stack.back().Indent == State.FirstIndent && PreviousNonComment && 1221 !PreviousNonComment->isOneOf(tok::r_brace, TT_CtorInitializerComma)) 1222 // Ensure that we fall back to the continuation indent width instead of 1223 // just flushing continuations left. 1224 return State.Stack.back().Indent + Style.ContinuationIndentWidth; 1225 return State.Stack.back().Indent; 1226 } 1227 1228 static bool hasNestedBlockInlined(const FormatToken *Previous, 1229 const FormatToken &Current, 1230 const FormatStyle &Style) { 1231 if (Previous->isNot(tok::l_paren)) 1232 return true; 1233 if (Previous->ParameterCount > 1) 1234 return true; 1235 1236 // Also a nested block if contains a lambda inside function with 1 parameter 1237 return (Style.BraceWrapping.BeforeLambdaBody && Current.is(TT_LambdaLSquare)); 1238 } 1239 1240 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State, 1241 bool DryRun, bool Newline) { 1242 assert(State.Stack.size()); 1243 const FormatToken &Current = *State.NextToken; 1244 1245 if (Current.is(TT_CSharpGenericTypeConstraint)) 1246 State.Stack.back().IsCSharpGenericTypeConstraint = true; 1247 if (Current.isOneOf(tok::comma, TT_BinaryOperator)) 1248 State.Stack.back().NoLineBreakInOperand = false; 1249 if (Current.isOneOf(TT_InheritanceColon, TT_CSharpGenericTypeConstraintColon)) 1250 State.Stack.back().AvoidBinPacking = true; 1251 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) { 1252 if (State.Stack.back().FirstLessLess == 0) 1253 State.Stack.back().FirstLessLess = State.Column; 1254 else 1255 State.Stack.back().LastOperatorWrapped = Newline; 1256 } 1257 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless)) 1258 State.Stack.back().LastOperatorWrapped = Newline; 1259 if (Current.is(TT_ConditionalExpr) && Current.Previous && 1260 !Current.Previous->is(TT_ConditionalExpr)) 1261 State.Stack.back().LastOperatorWrapped = Newline; 1262 if (Current.is(TT_ArraySubscriptLSquare) && 1263 State.Stack.back().StartOfArraySubscripts == 0) 1264 State.Stack.back().StartOfArraySubscripts = State.Column; 1265 1266 auto IsWrappedConditional = [](const FormatToken &Tok) { 1267 if (!(Tok.is(TT_ConditionalExpr) && Tok.is(tok::question))) 1268 return false; 1269 if (Tok.MustBreakBefore) 1270 return true; 1271 1272 const FormatToken *Next = Tok.getNextNonComment(); 1273 return Next && Next->MustBreakBefore; 1274 }; 1275 if (IsWrappedConditional(Current)) 1276 State.Stack.back().IsWrappedConditional = true; 1277 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question)) 1278 State.Stack.back().QuestionColumn = State.Column; 1279 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) { 1280 const FormatToken *Previous = Current.Previous; 1281 while (Previous && Previous->isTrailingComment()) 1282 Previous = Previous->Previous; 1283 if (Previous && Previous->is(tok::question)) 1284 State.Stack.back().QuestionColumn = State.Column; 1285 } 1286 if (!Current.opensScope() && !Current.closesScope() && 1287 !Current.is(TT_PointerOrReference)) 1288 State.LowestLevelOnLine = 1289 std::min(State.LowestLevelOnLine, Current.NestingLevel); 1290 if (Current.isMemberAccess()) 1291 State.Stack.back().StartOfFunctionCall = 1292 !Current.NextOperator ? 0 : State.Column; 1293 if (Current.is(TT_SelectorName)) 1294 State.Stack.back().ObjCSelectorNameFound = true; 1295 if (Current.is(TT_CtorInitializerColon) && 1296 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) { 1297 // Indent 2 from the column, so: 1298 // SomeClass::SomeClass() 1299 // : First(...), ... 1300 // Next(...) 1301 // ^ line up here. 1302 State.Stack.back().Indent = 1303 State.Column + 1304 (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma 1305 ? 0 1306 : 2); 1307 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1308 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) { 1309 State.Stack.back().AvoidBinPacking = true; 1310 State.Stack.back().BreakBeforeParameter = 1311 Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine; 1312 } else { 1313 State.Stack.back().BreakBeforeParameter = false; 1314 } 1315 } 1316 if (Current.is(TT_CtorInitializerColon) && 1317 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { 1318 State.Stack.back().Indent = 1319 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1320 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent; 1321 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) 1322 State.Stack.back().AvoidBinPacking = true; 1323 } 1324 if (Current.is(TT_InheritanceColon)) 1325 State.Stack.back().Indent = 1326 State.FirstIndent + Style.ConstructorInitializerIndentWidth; 1327 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline) 1328 State.Stack.back().NestedBlockIndent = 1329 State.Column + Current.ColumnWidth + 1; 1330 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow)) 1331 State.Stack.back().LastSpace = State.Column; 1332 if (Current.is(TT_RequiresExpression)) 1333 State.Stack.back().NestedBlockIndent = State.Column; 1334 1335 // Insert scopes created by fake parenthesis. 1336 const FormatToken *Previous = Current.getPreviousNonComment(); 1337 1338 // Add special behavior to support a format commonly used for JavaScript 1339 // closures: 1340 // SomeFunction(function() { 1341 // foo(); 1342 // bar(); 1343 // }, a, b, c); 1344 if (Current.isNot(tok::comment) && !Current.ClosesRequiresClause && 1345 Previous && Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) && 1346 !Previous->is(TT_DictLiteral) && State.Stack.size() > 1 && 1347 !State.Stack.back().HasMultipleNestedBlocks) { 1348 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline) 1349 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) 1350 State.Stack[i].NoLineBreak = true; 1351 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false; 1352 } 1353 if (Previous && (Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr) || 1354 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) && 1355 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)))) { 1356 State.Stack.back().NestedBlockInlined = 1357 !Newline && hasNestedBlockInlined(Previous, Current, Style); 1358 } 1359 1360 moveStatePastFakeLParens(State, Newline); 1361 moveStatePastScopeCloser(State); 1362 bool AllowBreak = !State.Stack.back().NoLineBreak && 1363 !State.Stack.back().NoLineBreakInOperand; 1364 moveStatePastScopeOpener(State, Newline); 1365 moveStatePastFakeRParens(State); 1366 1367 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0) 1368 State.StartOfStringLiteral = State.Column + 1; 1369 if (Current.is(TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) 1370 State.StartOfStringLiteral = State.Column + 1; 1371 else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) 1372 State.StartOfStringLiteral = State.Column; 1373 else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) && 1374 !Current.isStringLiteral()) 1375 State.StartOfStringLiteral = 0; 1376 1377 State.Column += Current.ColumnWidth; 1378 State.NextToken = State.NextToken->Next; 1379 1380 unsigned Penalty = 1381 handleEndOfLine(Current, State, DryRun, AllowBreak, Newline); 1382 1383 if (Current.Role) 1384 Current.Role->formatFromToken(State, this, DryRun); 1385 // If the previous has a special role, let it consume tokens as appropriate. 1386 // It is necessary to start at the previous token for the only implemented 1387 // role (comma separated list). That way, the decision whether or not to break 1388 // after the "{" is already done and both options are tried and evaluated. 1389 // FIXME: This is ugly, find a better way. 1390 if (Previous && Previous->Role) 1391 Penalty += Previous->Role->formatAfterToken(State, this, DryRun); 1392 1393 return Penalty; 1394 } 1395 1396 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State, 1397 bool Newline) { 1398 const FormatToken &Current = *State.NextToken; 1399 if (Current.FakeLParens.empty()) 1400 return; 1401 1402 const FormatToken *Previous = Current.getPreviousNonComment(); 1403 1404 // Don't add extra indentation for the first fake parenthesis after 1405 // 'return', assignments, opening <({[, or requires clauses. The indentation 1406 // for these cases is special cased. 1407 bool SkipFirstExtraIndent = 1408 Previous && 1409 (Previous->opensScope() || 1410 Previous->isOneOf(tok::semi, tok::kw_return, TT_RequiresClause) || 1411 (Previous->getPrecedence() == prec::Assignment && 1412 Style.AlignOperands != FormatStyle::OAS_DontAlign) || 1413 Previous->is(TT_ObjCMethodExpr)); 1414 for (const auto &PrecedenceLevel : llvm::reverse(Current.FakeLParens)) { 1415 ParenState NewParenState = State.Stack.back(); 1416 NewParenState.Tok = nullptr; 1417 NewParenState.ContainsLineBreak = false; 1418 NewParenState.LastOperatorWrapped = true; 1419 NewParenState.IsChainedConditional = false; 1420 NewParenState.IsWrappedConditional = false; 1421 NewParenState.UnindentOperator = false; 1422 NewParenState.NoLineBreak = 1423 NewParenState.NoLineBreak || State.Stack.back().NoLineBreakInOperand; 1424 1425 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists. 1426 if (PrecedenceLevel > prec::Comma) 1427 NewParenState.AvoidBinPacking = false; 1428 1429 // Indent from 'LastSpace' unless these are fake parentheses encapsulating 1430 // a builder type call after 'return' or, if the alignment after opening 1431 // brackets is disabled. 1432 if (!Current.isTrailingComment() && 1433 (Style.AlignOperands != FormatStyle::OAS_DontAlign || 1434 PrecedenceLevel < prec::Assignment) && 1435 (!Previous || Previous->isNot(tok::kw_return) || 1436 (Style.Language != FormatStyle::LK_Java && PrecedenceLevel > 0)) && 1437 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign || 1438 PrecedenceLevel != prec::Comma || Current.NestingLevel == 0)) { 1439 NewParenState.Indent = 1440 std::max(std::max(State.Column, NewParenState.Indent), 1441 State.Stack.back().LastSpace); 1442 } 1443 1444 if (Previous && 1445 (Previous->getPrecedence() == prec::Assignment || 1446 Previous->isOneOf(tok::kw_return, TT_RequiresClause) || 1447 (PrecedenceLevel == prec::Conditional && Previous->is(tok::question) && 1448 Previous->is(TT_ConditionalExpr))) && 1449 !Newline) { 1450 // If BreakBeforeBinaryOperators is set, un-indent a bit to account for 1451 // the operator and keep the operands aligned 1452 if (Style.AlignOperands == FormatStyle::OAS_AlignAfterOperator) 1453 NewParenState.UnindentOperator = true; 1454 // Mark indentation as alignment if the expression is aligned. 1455 if (Style.AlignOperands != FormatStyle::OAS_DontAlign) 1456 NewParenState.IsAligned = true; 1457 } 1458 1459 // Do not indent relative to the fake parentheses inserted for "." or "->". 1460 // This is a special case to make the following to statements consistent: 1461 // OuterFunction(InnerFunctionCall( // break 1462 // ParameterToInnerFunction)); 1463 // OuterFunction(SomeObject.InnerFunctionCall( // break 1464 // ParameterToInnerFunction)); 1465 if (PrecedenceLevel > prec::Unknown) 1466 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column); 1467 if (PrecedenceLevel != prec::Conditional && !Current.is(TT_UnaryOperator) && 1468 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) 1469 NewParenState.StartOfFunctionCall = State.Column; 1470 1471 // Indent conditional expressions, unless they are chained "else-if" 1472 // conditionals. Never indent expression where the 'operator' is ',', ';' or 1473 // an assignment (i.e. *I <= prec::Assignment) as those have different 1474 // indentation rules. Indent other expression, unless the indentation needs 1475 // to be skipped. 1476 if (PrecedenceLevel == prec::Conditional && Previous && 1477 Previous->is(tok::colon) && Previous->is(TT_ConditionalExpr) && 1478 &PrecedenceLevel == &Current.FakeLParens.back() && 1479 !State.Stack.back().IsWrappedConditional) { 1480 NewParenState.IsChainedConditional = true; 1481 NewParenState.UnindentOperator = State.Stack.back().UnindentOperator; 1482 } else if (PrecedenceLevel == prec::Conditional || 1483 (!SkipFirstExtraIndent && PrecedenceLevel > prec::Assignment && 1484 !Current.isTrailingComment())) { 1485 NewParenState.Indent += Style.ContinuationIndentWidth; 1486 } 1487 if ((Previous && !Previous->opensScope()) || PrecedenceLevel != prec::Comma) 1488 NewParenState.BreakBeforeParameter = false; 1489 State.Stack.push_back(NewParenState); 1490 SkipFirstExtraIndent = false; 1491 } 1492 } 1493 1494 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) { 1495 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) { 1496 unsigned VariablePos = State.Stack.back().VariablePos; 1497 if (State.Stack.size() == 1) { 1498 // Do not pop the last element. 1499 break; 1500 } 1501 State.Stack.pop_back(); 1502 State.Stack.back().VariablePos = VariablePos; 1503 } 1504 1505 if (State.NextToken->ClosesRequiresClause && Style.IndentRequiresClause) { 1506 // Remove the indentation of the requires clauses (which is not in Indent, 1507 // but in LastSpace). 1508 State.Stack.back().LastSpace -= Style.IndentWidth; 1509 } 1510 } 1511 1512 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State, 1513 bool Newline) { 1514 const FormatToken &Current = *State.NextToken; 1515 if (!Current.opensScope()) 1516 return; 1517 1518 // Don't allow '<' or '(' in C# generic type constraints to start new scopes. 1519 if (Current.isOneOf(tok::less, tok::l_paren) && 1520 State.Stack.back().IsCSharpGenericTypeConstraint) 1521 return; 1522 1523 if (Current.MatchingParen && Current.is(BK_Block)) { 1524 moveStateToNewBlock(State); 1525 return; 1526 } 1527 1528 unsigned NewIndent; 1529 unsigned LastSpace = State.Stack.back().LastSpace; 1530 bool AvoidBinPacking; 1531 bool BreakBeforeParameter = false; 1532 unsigned NestedBlockIndent = std::max(State.Stack.back().StartOfFunctionCall, 1533 State.Stack.back().NestedBlockIndent); 1534 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || 1535 opensProtoMessageField(Current, Style)) { 1536 if (Current.opensBlockOrBlockTypeList(Style)) { 1537 NewIndent = Style.IndentWidth + 1538 std::min(State.Column, State.Stack.back().NestedBlockIndent); 1539 } else { 1540 NewIndent = State.Stack.back().LastSpace + Style.ContinuationIndentWidth; 1541 } 1542 const FormatToken *NextNoComment = Current.getNextNonComment(); 1543 bool EndsInComma = Current.MatchingParen && 1544 Current.MatchingParen->Previous && 1545 Current.MatchingParen->Previous->is(tok::comma); 1546 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) || 1547 Style.Language == FormatStyle::LK_Proto || 1548 Style.Language == FormatStyle::LK_TextProto || 1549 !Style.BinPackArguments || 1550 (NextNoComment && 1551 NextNoComment->isOneOf(TT_DesignatedInitializerPeriod, 1552 TT_DesignatedInitializerLSquare)); 1553 BreakBeforeParameter = EndsInComma; 1554 if (Current.ParameterCount > 1) 1555 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1); 1556 } else { 1557 NewIndent = Style.ContinuationIndentWidth + 1558 std::max(State.Stack.back().LastSpace, 1559 State.Stack.back().StartOfFunctionCall); 1560 1561 // Ensure that different different brackets force relative alignment, e.g.: 1562 // void SomeFunction(vector< // break 1563 // int> v); 1564 // FIXME: We likely want to do this for more combinations of brackets. 1565 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) { 1566 NewIndent = std::max(NewIndent, State.Stack.back().Indent); 1567 LastSpace = std::max(LastSpace, State.Stack.back().Indent); 1568 } 1569 1570 bool EndsInComma = 1571 Current.MatchingParen && 1572 Current.MatchingParen->getPreviousNonComment() && 1573 Current.MatchingParen->getPreviousNonComment()->is(tok::comma); 1574 1575 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters 1576 // for backwards compatibility. 1577 bool ObjCBinPackProtocolList = 1578 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto && 1579 Style.BinPackParameters) || 1580 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always; 1581 1582 bool BinPackDeclaration = 1583 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) || 1584 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList); 1585 1586 AvoidBinPacking = 1587 (State.Stack.back().IsCSharpGenericTypeConstraint) || 1588 (Style.isJavaScript() && EndsInComma) || 1589 (State.Line->MustBeDeclaration && !BinPackDeclaration) || 1590 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) || 1591 (Style.ExperimentalAutoDetectBinPacking && 1592 (Current.is(PPK_OnePerLine) || 1593 (!BinPackInconclusiveFunctions && Current.is(PPK_Inconclusive)))); 1594 1595 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen && 1596 Style.ObjCBreakBeforeNestedBlockParam) { 1597 if (Style.ColumnLimit) { 1598 // If this '[' opens an ObjC call, determine whether all parameters fit 1599 // into one line and put one per line if they don't. 1600 if (getLengthToMatchingParen(Current, State.Stack) + State.Column > 1601 getColumnLimit(State)) 1602 BreakBeforeParameter = true; 1603 } else { 1604 // For ColumnLimit = 0, we have to figure out whether there is or has to 1605 // be a line break within this call. 1606 for (const FormatToken *Tok = &Current; 1607 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) { 1608 if (Tok->MustBreakBefore || 1609 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) { 1610 BreakBeforeParameter = true; 1611 break; 1612 } 1613 } 1614 } 1615 } 1616 1617 if (Style.isJavaScript() && EndsInComma) 1618 BreakBeforeParameter = true; 1619 } 1620 // Generally inherit NoLineBreak from the current scope to nested scope. 1621 // However, don't do this for non-empty nested blocks, dict literals and 1622 // array literals as these follow different indentation rules. 1623 bool NoLineBreak = 1624 Current.Children.empty() && 1625 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) && 1626 (State.Stack.back().NoLineBreak || 1627 State.Stack.back().NoLineBreakInOperand || 1628 (Current.is(TT_TemplateOpener) && 1629 State.Stack.back().ContainsUnwrappedBuilder)); 1630 State.Stack.push_back( 1631 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak)); 1632 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1633 State.Stack.back().BreakBeforeParameter = BreakBeforeParameter; 1634 State.Stack.back().HasMultipleNestedBlocks = 1635 (Current.BlockParameterCount > 1); 1636 1637 if (Style.BraceWrapping.BeforeLambdaBody && Current.Next != nullptr && 1638 Current.Tok.is(tok::l_paren)) { 1639 // Search for any parameter that is a lambda 1640 FormatToken const *next = Current.Next; 1641 while (next != nullptr) { 1642 if (next->is(TT_LambdaLSquare)) { 1643 State.Stack.back().HasMultipleNestedBlocks = true; 1644 break; 1645 } 1646 next = next->Next; 1647 } 1648 } 1649 1650 State.Stack.back().IsInsideObjCArrayLiteral = 1651 Current.is(TT_ArrayInitializerLSquare) && Current.Previous && 1652 Current.Previous->is(tok::at); 1653 } 1654 1655 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) { 1656 const FormatToken &Current = *State.NextToken; 1657 if (!Current.closesScope()) 1658 return; 1659 1660 // If we encounter a closing ), ], } or >, we can remove a level from our 1661 // stacks. 1662 if (State.Stack.size() > 1 && 1663 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) || 1664 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) || 1665 State.NextToken->is(TT_TemplateCloser) || 1666 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) 1667 State.Stack.pop_back(); 1668 1669 // Reevaluate whether ObjC message arguments fit into one line. 1670 // If a receiver spans multiple lines, e.g.: 1671 // [[object block:^{ 1672 // return 42; 1673 // }] a:42 b:42]; 1674 // BreakBeforeParameter is calculated based on an incorrect assumption 1675 // (it is checked whether the whole expression fits into one line without 1676 // considering a line break inside a message receiver). 1677 // We check whether arguments fit after receiver scope closer (into the same 1678 // line). 1679 if (State.Stack.back().BreakBeforeParameter && Current.MatchingParen && 1680 Current.MatchingParen->Previous) { 1681 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous; 1682 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) && 1683 CurrentScopeOpener.MatchingParen) { 1684 int NecessarySpaceInLine = 1685 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) + 1686 CurrentScopeOpener.TotalLength - Current.TotalLength - 1; 1687 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <= 1688 Style.ColumnLimit) 1689 State.Stack.back().BreakBeforeParameter = false; 1690 } 1691 } 1692 1693 if (Current.is(tok::r_square)) { 1694 // If this ends the array subscript expr, reset the corresponding value. 1695 const FormatToken *NextNonComment = Current.getNextNonComment(); 1696 if (NextNonComment && NextNonComment->isNot(tok::l_square)) 1697 State.Stack.back().StartOfArraySubscripts = 0; 1698 } 1699 } 1700 1701 void ContinuationIndenter::moveStateToNewBlock(LineState &State) { 1702 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent; 1703 // ObjC block sometimes follow special indentation rules. 1704 unsigned NewIndent = 1705 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace) 1706 ? Style.ObjCBlockIndentWidth 1707 : Style.IndentWidth); 1708 State.Stack.push_back(ParenState(State.NextToken, NewIndent, 1709 State.Stack.back().LastSpace, 1710 /*AvoidBinPacking=*/true, 1711 /*NoLineBreak=*/false)); 1712 State.Stack.back().NestedBlockIndent = NestedBlockIndent; 1713 State.Stack.back().BreakBeforeParameter = true; 1714 } 1715 1716 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn, 1717 unsigned TabWidth, 1718 encoding::Encoding Encoding) { 1719 size_t LastNewlinePos = Text.find_last_of("\n"); 1720 if (LastNewlinePos == StringRef::npos) { 1721 return StartColumn + 1722 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding); 1723 } else { 1724 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos), 1725 /*StartColumn=*/0, TabWidth, Encoding); 1726 } 1727 } 1728 1729 unsigned ContinuationIndenter::reformatRawStringLiteral( 1730 const FormatToken &Current, LineState &State, 1731 const FormatStyle &RawStringStyle, bool DryRun, bool Newline) { 1732 unsigned StartColumn = State.Column - Current.ColumnWidth; 1733 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText); 1734 StringRef NewDelimiter = 1735 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language); 1736 if (NewDelimiter.empty()) 1737 NewDelimiter = OldDelimiter; 1738 // The text of a raw string is between the leading 'R"delimiter(' and the 1739 // trailing 'delimiter)"'. 1740 unsigned OldPrefixSize = 3 + OldDelimiter.size(); 1741 unsigned OldSuffixSize = 2 + OldDelimiter.size(); 1742 // We create a virtual text environment which expects a null-terminated 1743 // string, so we cannot use StringRef. 1744 std::string RawText = std::string( 1745 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize)); 1746 if (NewDelimiter != OldDelimiter) { 1747 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the 1748 // raw string. 1749 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str(); 1750 if (StringRef(RawText).contains(CanonicalDelimiterSuffix)) 1751 NewDelimiter = OldDelimiter; 1752 } 1753 1754 unsigned NewPrefixSize = 3 + NewDelimiter.size(); 1755 unsigned NewSuffixSize = 2 + NewDelimiter.size(); 1756 1757 // The first start column is the column the raw text starts after formatting. 1758 unsigned FirstStartColumn = StartColumn + NewPrefixSize; 1759 1760 // The next start column is the intended indentation a line break inside 1761 // the raw string at level 0. It is determined by the following rules: 1762 // - if the content starts on newline, it is one level more than the current 1763 // indent, and 1764 // - if the content does not start on a newline, it is the first start 1765 // column. 1766 // These rules have the advantage that the formatted content both does not 1767 // violate the rectangle rule and visually flows within the surrounding 1768 // source. 1769 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n'; 1770 // If this token is the last parameter (checked by looking if it's followed by 1771 // `)` and is not on a newline, the base the indent off the line's nested 1772 // block indent. Otherwise, base the indent off the arguments indent, so we 1773 // can achieve: 1774 // 1775 // fffffffffff(1, 2, 3, R"pb( 1776 // key1: 1 # 1777 // key2: 2)pb"); 1778 // 1779 // fffffffffff(1, 2, 3, 1780 // R"pb( 1781 // key1: 1 # 1782 // key2: 2 1783 // )pb"); 1784 // 1785 // fffffffffff(1, 2, 3, 1786 // R"pb( 1787 // key1: 1 # 1788 // key2: 2 1789 // )pb", 1790 // 5); 1791 unsigned CurrentIndent = 1792 (!Newline && Current.Next && Current.Next->is(tok::r_paren)) 1793 ? State.Stack.back().NestedBlockIndent 1794 : State.Stack.back().Indent; 1795 unsigned NextStartColumn = ContentStartsOnNewline 1796 ? CurrentIndent + Style.IndentWidth 1797 : FirstStartColumn; 1798 1799 // The last start column is the column the raw string suffix starts if it is 1800 // put on a newline. 1801 // The last start column is the intended indentation of the raw string postfix 1802 // if it is put on a newline. It is determined by the following rules: 1803 // - if the raw string prefix starts on a newline, it is the column where 1804 // that raw string prefix starts, and 1805 // - if the raw string prefix does not start on a newline, it is the current 1806 // indent. 1807 unsigned LastStartColumn = 1808 Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent; 1809 1810 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat( 1811 RawStringStyle, RawText, {tooling::Range(0, RawText.size())}, 1812 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>", 1813 /*Status=*/nullptr); 1814 1815 auto NewCode = applyAllReplacements(RawText, Fixes.first); 1816 tooling::Replacements NoFixes; 1817 if (!NewCode) 1818 return addMultilineToken(Current, State); 1819 if (!DryRun) { 1820 if (NewDelimiter != OldDelimiter) { 1821 // In 'R"delimiter(...', the delimiter starts 2 characters after the start 1822 // of the token. 1823 SourceLocation PrefixDelimiterStart = 1824 Current.Tok.getLocation().getLocWithOffset(2); 1825 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement( 1826 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1827 if (PrefixErr) { 1828 llvm::errs() 1829 << "Failed to update the prefix delimiter of a raw string: " 1830 << llvm::toString(std::move(PrefixErr)) << "\n"; 1831 } 1832 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at 1833 // position length - 1 - |delimiter|. 1834 SourceLocation SuffixDelimiterStart = 1835 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() - 1836 1 - OldDelimiter.size()); 1837 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement( 1838 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter)); 1839 if (SuffixErr) { 1840 llvm::errs() 1841 << "Failed to update the suffix delimiter of a raw string: " 1842 << llvm::toString(std::move(SuffixErr)) << "\n"; 1843 } 1844 } 1845 SourceLocation OriginLoc = 1846 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize); 1847 for (const tooling::Replacement &Fix : Fixes.first) { 1848 auto Err = Whitespaces.addReplacement(tooling::Replacement( 1849 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()), 1850 Fix.getLength(), Fix.getReplacementText())); 1851 if (Err) { 1852 llvm::errs() << "Failed to reformat raw string: " 1853 << llvm::toString(std::move(Err)) << "\n"; 1854 } 1855 } 1856 } 1857 unsigned RawLastLineEndColumn = getLastLineEndColumn( 1858 *NewCode, FirstStartColumn, Style.TabWidth, Encoding); 1859 State.Column = RawLastLineEndColumn + NewSuffixSize; 1860 // Since we're updating the column to after the raw string literal here, we 1861 // have to manually add the penalty for the prefix R"delim( over the column 1862 // limit. 1863 unsigned PrefixExcessCharacters = 1864 StartColumn + NewPrefixSize > Style.ColumnLimit 1865 ? StartColumn + NewPrefixSize - Style.ColumnLimit 1866 : 0; 1867 bool IsMultiline = 1868 ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos); 1869 if (IsMultiline) { 1870 // Break before further function parameters on all levels. 1871 for (ParenState &Paren : State.Stack) 1872 Paren.BreakBeforeParameter = true; 1873 } 1874 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter; 1875 } 1876 1877 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current, 1878 LineState &State) { 1879 // Break before further function parameters on all levels. 1880 for (ParenState &Paren : State.Stack) 1881 Paren.BreakBeforeParameter = true; 1882 1883 unsigned ColumnsUsed = State.Column; 1884 // We can only affect layout of the first and the last line, so the penalty 1885 // for all other lines is constant, and we ignore it. 1886 State.Column = Current.LastLineColumnWidth; 1887 1888 if (ColumnsUsed > getColumnLimit(State)) 1889 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State)); 1890 return 0; 1891 } 1892 1893 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current, 1894 LineState &State, bool DryRun, 1895 bool AllowBreak, bool Newline) { 1896 unsigned Penalty = 0; 1897 // Compute the raw string style to use in case this is a raw string literal 1898 // that can be reformatted. 1899 auto RawStringStyle = getRawStringStyle(Current, State); 1900 if (RawStringStyle && !Current.Finalized) { 1901 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun, 1902 Newline); 1903 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) { 1904 // Don't break multi-line tokens other than block comments and raw string 1905 // literals. Instead, just update the state. 1906 Penalty = addMultilineToken(Current, State); 1907 } else if (State.Line->Type != LT_ImportStatement) { 1908 // We generally don't break import statements. 1909 LineState OriginalState = State; 1910 1911 // Whether we force the reflowing algorithm to stay strictly within the 1912 // column limit. 1913 bool Strict = false; 1914 // Whether the first non-strict attempt at reflowing did intentionally 1915 // exceed the column limit. 1916 bool Exceeded = false; 1917 std::tie(Penalty, Exceeded) = breakProtrudingToken( 1918 Current, State, AllowBreak, /*DryRun=*/true, Strict); 1919 if (Exceeded) { 1920 // If non-strict reflowing exceeds the column limit, try whether strict 1921 // reflowing leads to an overall lower penalty. 1922 LineState StrictState = OriginalState; 1923 unsigned StrictPenalty = 1924 breakProtrudingToken(Current, StrictState, AllowBreak, 1925 /*DryRun=*/true, /*Strict=*/true) 1926 .first; 1927 Strict = StrictPenalty <= Penalty; 1928 if (Strict) { 1929 Penalty = StrictPenalty; 1930 State = StrictState; 1931 } 1932 } 1933 if (!DryRun) { 1934 // If we're not in dry-run mode, apply the changes with the decision on 1935 // strictness made above. 1936 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false, 1937 Strict); 1938 } 1939 } 1940 if (State.Column > getColumnLimit(State)) { 1941 unsigned ExcessCharacters = State.Column - getColumnLimit(State); 1942 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters; 1943 } 1944 return Penalty; 1945 } 1946 1947 // Returns the enclosing function name of a token, or the empty string if not 1948 // found. 1949 static StringRef getEnclosingFunctionName(const FormatToken &Current) { 1950 // Look for: 'function(' or 'function<templates>(' before Current. 1951 auto Tok = Current.getPreviousNonComment(); 1952 if (!Tok || !Tok->is(tok::l_paren)) 1953 return ""; 1954 Tok = Tok->getPreviousNonComment(); 1955 if (!Tok) 1956 return ""; 1957 if (Tok->is(TT_TemplateCloser)) { 1958 Tok = Tok->MatchingParen; 1959 if (Tok) 1960 Tok = Tok->getPreviousNonComment(); 1961 } 1962 if (!Tok || !Tok->is(tok::identifier)) 1963 return ""; 1964 return Tok->TokenText; 1965 } 1966 1967 llvm::Optional<FormatStyle> 1968 ContinuationIndenter::getRawStringStyle(const FormatToken &Current, 1969 const LineState &State) { 1970 if (!Current.isStringLiteral()) 1971 return None; 1972 auto Delimiter = getRawStringDelimiter(Current.TokenText); 1973 if (!Delimiter) 1974 return None; 1975 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter); 1976 if (!RawStringStyle && Delimiter->empty()) 1977 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle( 1978 getEnclosingFunctionName(Current)); 1979 if (!RawStringStyle) 1980 return None; 1981 RawStringStyle->ColumnLimit = getColumnLimit(State); 1982 return RawStringStyle; 1983 } 1984 1985 std::unique_ptr<BreakableToken> 1986 ContinuationIndenter::createBreakableToken(const FormatToken &Current, 1987 LineState &State, bool AllowBreak) { 1988 unsigned StartColumn = State.Column - Current.ColumnWidth; 1989 if (Current.isStringLiteral()) { 1990 // FIXME: String literal breaking is currently disabled for C#, Java, Json 1991 // and JavaScript, as it requires strings to be merged using "+" which we 1992 // don't support. 1993 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 1994 Style.isCSharp() || Style.isJson() || !Style.BreakStringLiterals || 1995 !AllowBreak) 1996 return nullptr; 1997 1998 // Don't break string literals inside preprocessor directives (except for 1999 // #define directives, as their contents are stored in separate lines and 2000 // are not affected by this check). 2001 // This way we avoid breaking code with line directives and unknown 2002 // preprocessor directives that contain long string literals. 2003 if (State.Line->Type == LT_PreprocessorDirective) 2004 return nullptr; 2005 // Exempts unterminated string literals from line breaking. The user will 2006 // likely want to terminate the string before any line breaking is done. 2007 if (Current.IsUnterminatedLiteral) 2008 return nullptr; 2009 // Don't break string literals inside Objective-C array literals (doing so 2010 // raises the warning -Wobjc-string-concatenation). 2011 if (State.Stack.back().IsInsideObjCArrayLiteral) 2012 return nullptr; 2013 2014 StringRef Text = Current.TokenText; 2015 StringRef Prefix; 2016 StringRef Postfix; 2017 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'. 2018 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to 2019 // reduce the overhead) for each FormatToken, which is a string, so that we 2020 // don't run multiple checks here on the hot path. 2021 if ((Text.endswith(Postfix = "\"") && 2022 (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") || 2023 Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") || 2024 Text.startswith(Prefix = "u8\"") || 2025 Text.startswith(Prefix = "L\""))) || 2026 (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) { 2027 // We need this to address the case where there is an unbreakable tail 2028 // only if certain other formatting decisions have been taken. The 2029 // UnbreakableTailLength of Current is an overapproximation is that case 2030 // and we need to be correct here. 2031 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State)) 2032 ? 0 2033 : Current.UnbreakableTailLength; 2034 return std::make_unique<BreakableStringLiteral>( 2035 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength, 2036 State.Line->InPPDirective, Encoding, Style); 2037 } 2038 } else if (Current.is(TT_BlockComment)) { 2039 if (!Style.ReflowComments || 2040 // If a comment token switches formatting, like 2041 // /* clang-format on */, we don't want to break it further, 2042 // but we may still want to adjust its indentation. 2043 switchesFormatting(Current)) 2044 return nullptr; 2045 return std::make_unique<BreakableBlockComment>( 2046 Current, StartColumn, Current.OriginalColumn, !Current.Previous, 2047 State.Line->InPPDirective, Encoding, Style, Whitespaces.useCRLF()); 2048 } else if (Current.is(TT_LineComment) && 2049 (Current.Previous == nullptr || 2050 Current.Previous->isNot(TT_ImplicitStringLiteral))) { 2051 bool RegularComments = [&]() { 2052 for (const FormatToken *T = &Current; T && T->is(TT_LineComment); 2053 T = T->Next) 2054 if (!(T->TokenText.startswith("//") || T->TokenText.startswith("#"))) 2055 return false; 2056 return true; 2057 }(); 2058 if (!Style.ReflowComments || 2059 CommentPragmasRegex.match(Current.TokenText.substr(2)) || 2060 switchesFormatting(Current) || !RegularComments) 2061 return nullptr; 2062 return std::make_unique<BreakableLineCommentSection>( 2063 Current, StartColumn, /*InPPDirective=*/false, Encoding, Style); 2064 } 2065 return nullptr; 2066 } 2067 2068 std::pair<unsigned, bool> 2069 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current, 2070 LineState &State, bool AllowBreak, 2071 bool DryRun, bool Strict) { 2072 std::unique_ptr<const BreakableToken> Token = 2073 createBreakableToken(Current, State, AllowBreak); 2074 if (!Token) 2075 return {0, false}; 2076 assert(Token->getLineCount() > 0); 2077 unsigned ColumnLimit = getColumnLimit(State); 2078 if (Current.is(TT_LineComment)) { 2079 // We don't insert backslashes when breaking line comments. 2080 ColumnLimit = Style.ColumnLimit; 2081 } 2082 if (ColumnLimit == 0) { 2083 // To make the rest of the function easier set the column limit to the 2084 // maximum, if there should be no limit. 2085 ColumnLimit = std::numeric_limits<decltype(ColumnLimit)>::max(); 2086 } 2087 if (Current.UnbreakableTailLength >= ColumnLimit) 2088 return {0, false}; 2089 // ColumnWidth was already accounted into State.Column before calling 2090 // breakProtrudingToken. 2091 unsigned StartColumn = State.Column - Current.ColumnWidth; 2092 unsigned NewBreakPenalty = Current.isStringLiteral() 2093 ? Style.PenaltyBreakString 2094 : Style.PenaltyBreakComment; 2095 // Stores whether we intentionally decide to let a line exceed the column 2096 // limit. 2097 bool Exceeded = false; 2098 // Stores whether we introduce a break anywhere in the token. 2099 bool BreakInserted = Token->introducesBreakBeforeToken(); 2100 // Store whether we inserted a new line break at the end of the previous 2101 // logical line. 2102 bool NewBreakBefore = false; 2103 // We use a conservative reflowing strategy. Reflow starts after a line is 2104 // broken or the corresponding whitespace compressed. Reflow ends as soon as a 2105 // line that doesn't get reflown with the previous line is reached. 2106 bool Reflow = false; 2107 // Keep track of where we are in the token: 2108 // Where we are in the content of the current logical line. 2109 unsigned TailOffset = 0; 2110 // The column number we're currently at. 2111 unsigned ContentStartColumn = 2112 Token->getContentStartColumn(0, /*Break=*/false); 2113 // The number of columns left in the current logical line after TailOffset. 2114 unsigned RemainingTokenColumns = 2115 Token->getRemainingLength(0, TailOffset, ContentStartColumn); 2116 // Adapt the start of the token, for example indent. 2117 if (!DryRun) 2118 Token->adaptStartOfLine(0, Whitespaces); 2119 2120 unsigned ContentIndent = 0; 2121 unsigned Penalty = 0; 2122 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column " 2123 << StartColumn << ".\n"); 2124 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount(); 2125 LineIndex != EndIndex; ++LineIndex) { 2126 LLVM_DEBUG(llvm::dbgs() 2127 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n"); 2128 NewBreakBefore = false; 2129 // If we did reflow the previous line, we'll try reflowing again. Otherwise 2130 // we'll start reflowing if the current line is broken or whitespace is 2131 // compressed. 2132 bool TryReflow = Reflow; 2133 // Break the current token until we can fit the rest of the line. 2134 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2135 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: " 2136 << (ContentStartColumn + RemainingTokenColumns) 2137 << ", space: " << ColumnLimit 2138 << ", reflown prefix: " << ContentStartColumn 2139 << ", offset in line: " << TailOffset << "\n"); 2140 // If the current token doesn't fit, find the latest possible split in the 2141 // current line so that breaking at it will be under the column limit. 2142 // FIXME: Use the earliest possible split while reflowing to correctly 2143 // compress whitespace within a line. 2144 BreakableToken::Split Split = 2145 Token->getSplit(LineIndex, TailOffset, ColumnLimit, 2146 ContentStartColumn, CommentPragmasRegex); 2147 if (Split.first == StringRef::npos) { 2148 // No break opportunity - update the penalty and continue with the next 2149 // logical line. 2150 if (LineIndex < EndIndex - 1) 2151 // The last line's penalty is handled in addNextStateToQueue() or when 2152 // calling replaceWhitespaceAfterLastLine below. 2153 Penalty += Style.PenaltyExcessCharacter * 2154 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2155 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n"); 2156 break; 2157 } 2158 assert(Split.first != 0); 2159 2160 if (Token->supportsReflow()) { 2161 // Check whether the next natural split point after the current one can 2162 // still fit the line, either because we can compress away whitespace, 2163 // or because the penalty the excess characters introduce is lower than 2164 // the break penalty. 2165 // We only do this for tokens that support reflowing, and thus allow us 2166 // to change the whitespace arbitrarily (e.g. comments). 2167 // Other tokens, like string literals, can be broken on arbitrary 2168 // positions. 2169 2170 // First, compute the columns from TailOffset to the next possible split 2171 // position. 2172 // For example: 2173 // ColumnLimit: | 2174 // // Some text that breaks 2175 // ^ tail offset 2176 // ^-- split 2177 // ^-------- to split columns 2178 // ^--- next split 2179 // ^--------------- to next split columns 2180 unsigned ToSplitColumns = Token->getRangeLength( 2181 LineIndex, TailOffset, Split.first, ContentStartColumn); 2182 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n"); 2183 2184 BreakableToken::Split NextSplit = Token->getSplit( 2185 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit, 2186 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex); 2187 // Compute the columns necessary to fit the next non-breakable sequence 2188 // into the current line. 2189 unsigned ToNextSplitColumns = 0; 2190 if (NextSplit.first == StringRef::npos) { 2191 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset, 2192 ContentStartColumn); 2193 } else { 2194 ToNextSplitColumns = Token->getRangeLength( 2195 LineIndex, TailOffset, 2196 Split.first + Split.second + NextSplit.first, ContentStartColumn); 2197 } 2198 // Compress the whitespace between the break and the start of the next 2199 // unbreakable sequence. 2200 ToNextSplitColumns = 2201 Token->getLengthAfterCompression(ToNextSplitColumns, Split); 2202 LLVM_DEBUG(llvm::dbgs() 2203 << " ContentStartColumn: " << ContentStartColumn << "\n"); 2204 LLVM_DEBUG(llvm::dbgs() 2205 << " ToNextSplit: " << ToNextSplitColumns << "\n"); 2206 // If the whitespace compression makes us fit, continue on the current 2207 // line. 2208 bool ContinueOnLine = 2209 ContentStartColumn + ToNextSplitColumns <= ColumnLimit; 2210 unsigned ExcessCharactersPenalty = 0; 2211 if (!ContinueOnLine && !Strict) { 2212 // Similarly, if the excess characters' penalty is lower than the 2213 // penalty of introducing a new break, continue on the current line. 2214 ExcessCharactersPenalty = 2215 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) * 2216 Style.PenaltyExcessCharacter; 2217 LLVM_DEBUG(llvm::dbgs() 2218 << " Penalty excess: " << ExcessCharactersPenalty 2219 << "\n break : " << NewBreakPenalty << "\n"); 2220 if (ExcessCharactersPenalty < NewBreakPenalty) { 2221 Exceeded = true; 2222 ContinueOnLine = true; 2223 } 2224 } 2225 if (ContinueOnLine) { 2226 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n"); 2227 // The current line fits after compressing the whitespace - reflow 2228 // the next line into it if possible. 2229 TryReflow = true; 2230 if (!DryRun) 2231 Token->compressWhitespace(LineIndex, TailOffset, Split, 2232 Whitespaces); 2233 // When we continue on the same line, leave one space between content. 2234 ContentStartColumn += ToSplitColumns + 1; 2235 Penalty += ExcessCharactersPenalty; 2236 TailOffset += Split.first + Split.second; 2237 RemainingTokenColumns = Token->getRemainingLength( 2238 LineIndex, TailOffset, ContentStartColumn); 2239 continue; 2240 } 2241 } 2242 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n"); 2243 // Update the ContentIndent only if the current line was not reflown with 2244 // the previous line, since in that case the previous line should still 2245 // determine the ContentIndent. Also never intent the last line. 2246 if (!Reflow) 2247 ContentIndent = Token->getContentIndent(LineIndex); 2248 LLVM_DEBUG(llvm::dbgs() 2249 << " ContentIndent: " << ContentIndent << "\n"); 2250 ContentStartColumn = ContentIndent + Token->getContentStartColumn( 2251 LineIndex, /*Break=*/true); 2252 2253 unsigned NewRemainingTokenColumns = Token->getRemainingLength( 2254 LineIndex, TailOffset + Split.first + Split.second, 2255 ContentStartColumn); 2256 if (NewRemainingTokenColumns == 0) { 2257 // No content to indent. 2258 ContentIndent = 0; 2259 ContentStartColumn = 2260 Token->getContentStartColumn(LineIndex, /*Break=*/true); 2261 NewRemainingTokenColumns = Token->getRemainingLength( 2262 LineIndex, TailOffset + Split.first + Split.second, 2263 ContentStartColumn); 2264 } 2265 2266 // When breaking before a tab character, it may be moved by a few columns, 2267 // but will still be expanded to the next tab stop, so we don't save any 2268 // columns. 2269 if (NewRemainingTokenColumns >= RemainingTokenColumns) { 2270 // FIXME: Do we need to adjust the penalty? 2271 break; 2272 } 2273 2274 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first 2275 << ", " << Split.second << "\n"); 2276 if (!DryRun) 2277 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent, 2278 Whitespaces); 2279 2280 Penalty += NewBreakPenalty; 2281 TailOffset += Split.first + Split.second; 2282 RemainingTokenColumns = NewRemainingTokenColumns; 2283 BreakInserted = true; 2284 NewBreakBefore = true; 2285 } 2286 // In case there's another line, prepare the state for the start of the next 2287 // line. 2288 if (LineIndex + 1 != EndIndex) { 2289 unsigned NextLineIndex = LineIndex + 1; 2290 if (NewBreakBefore) 2291 // After breaking a line, try to reflow the next line into the current 2292 // one once RemainingTokenColumns fits. 2293 TryReflow = true; 2294 if (TryReflow) { 2295 // We decided that we want to try reflowing the next line into the 2296 // current one. 2297 // We will now adjust the state as if the reflow is successful (in 2298 // preparation for the next line), and see whether that works. If we 2299 // decide that we cannot reflow, we will later reset the state to the 2300 // start of the next line. 2301 Reflow = false; 2302 // As we did not continue breaking the line, RemainingTokenColumns is 2303 // known to fit after ContentStartColumn. Adapt ContentStartColumn to 2304 // the position at which we want to format the next line if we do 2305 // actually reflow. 2306 // When we reflow, we need to add a space between the end of the current 2307 // line and the next line's start column. 2308 ContentStartColumn += RemainingTokenColumns + 1; 2309 // Get the split that we need to reflow next logical line into the end 2310 // of the current one; the split will include any leading whitespace of 2311 // the next logical line. 2312 BreakableToken::Split SplitBeforeNext = 2313 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex); 2314 LLVM_DEBUG(llvm::dbgs() 2315 << " Size of reflown text: " << ContentStartColumn 2316 << "\n Potential reflow split: "); 2317 if (SplitBeforeNext.first != StringRef::npos) { 2318 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", " 2319 << SplitBeforeNext.second << "\n"); 2320 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second; 2321 // If the rest of the next line fits into the current line below the 2322 // column limit, we can safely reflow. 2323 RemainingTokenColumns = Token->getRemainingLength( 2324 NextLineIndex, TailOffset, ContentStartColumn); 2325 Reflow = true; 2326 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { 2327 LLVM_DEBUG(llvm::dbgs() 2328 << " Over limit after reflow, need: " 2329 << (ContentStartColumn + RemainingTokenColumns) 2330 << ", space: " << ColumnLimit 2331 << ", reflown prefix: " << ContentStartColumn 2332 << ", offset in line: " << TailOffset << "\n"); 2333 // If the whole next line does not fit, try to find a point in 2334 // the next line at which we can break so that attaching the part 2335 // of the next line to that break point onto the current line is 2336 // below the column limit. 2337 BreakableToken::Split Split = 2338 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit, 2339 ContentStartColumn, CommentPragmasRegex); 2340 if (Split.first == StringRef::npos) { 2341 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n"); 2342 Reflow = false; 2343 } else { 2344 // Check whether the first split point gets us below the column 2345 // limit. Note that we will execute this split below as part of 2346 // the normal token breaking and reflow logic within the line. 2347 unsigned ToSplitColumns = Token->getRangeLength( 2348 NextLineIndex, TailOffset, Split.first, ContentStartColumn); 2349 if (ContentStartColumn + ToSplitColumns > ColumnLimit) { 2350 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: " 2351 << (ContentStartColumn + ToSplitColumns) 2352 << ", space: " << ColumnLimit); 2353 unsigned ExcessCharactersPenalty = 2354 (ContentStartColumn + ToSplitColumns - ColumnLimit) * 2355 Style.PenaltyExcessCharacter; 2356 if (NewBreakPenalty < ExcessCharactersPenalty) 2357 Reflow = false; 2358 } 2359 } 2360 } 2361 } else { 2362 LLVM_DEBUG(llvm::dbgs() << "not found.\n"); 2363 } 2364 } 2365 if (!Reflow) { 2366 // If we didn't reflow into the next line, the only space to consider is 2367 // the next logical line. Reset our state to match the start of the next 2368 // line. 2369 TailOffset = 0; 2370 ContentStartColumn = 2371 Token->getContentStartColumn(NextLineIndex, /*Break=*/false); 2372 RemainingTokenColumns = Token->getRemainingLength( 2373 NextLineIndex, TailOffset, ContentStartColumn); 2374 // Adapt the start of the token, for example indent. 2375 if (!DryRun) 2376 Token->adaptStartOfLine(NextLineIndex, Whitespaces); 2377 } else { 2378 // If we found a reflow split and have added a new break before the next 2379 // line, we are going to remove the line break at the start of the next 2380 // logical line. For example, here we'll add a new line break after 2381 // 'text', and subsequently delete the line break between 'that' and 2382 // 'reflows'. 2383 // // some text that 2384 // // reflows 2385 // -> 2386 // // some text 2387 // // that reflows 2388 // When adding the line break, we also added the penalty for it, so we 2389 // need to subtract that penalty again when we remove the line break due 2390 // to reflowing. 2391 if (NewBreakBefore) { 2392 assert(Penalty >= NewBreakPenalty); 2393 Penalty -= NewBreakPenalty; 2394 } 2395 if (!DryRun) 2396 Token->reflow(NextLineIndex, Whitespaces); 2397 } 2398 } 2399 } 2400 2401 BreakableToken::Split SplitAfterLastLine = 2402 Token->getSplitAfterLastLine(TailOffset); 2403 if (SplitAfterLastLine.first != StringRef::npos) { 2404 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n"); 2405 2406 // We add the last line's penalty here, since that line is going to be split 2407 // now. 2408 Penalty += Style.PenaltyExcessCharacter * 2409 (ContentStartColumn + RemainingTokenColumns - ColumnLimit); 2410 2411 if (!DryRun) 2412 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine, 2413 Whitespaces); 2414 ContentStartColumn = 2415 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true); 2416 RemainingTokenColumns = Token->getRemainingLength( 2417 Token->getLineCount() - 1, 2418 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second, 2419 ContentStartColumn); 2420 } 2421 2422 State.Column = ContentStartColumn + RemainingTokenColumns - 2423 Current.UnbreakableTailLength; 2424 2425 if (BreakInserted) { 2426 // If we break the token inside a parameter list, we need to break before 2427 // the next parameter on all levels, so that the next parameter is clearly 2428 // visible. Line comments already introduce a break. 2429 if (Current.isNot(TT_LineComment)) 2430 for (ParenState &Paren : State.Stack) 2431 Paren.BreakBeforeParameter = true; 2432 2433 if (Current.is(TT_BlockComment)) 2434 State.NoContinuation = true; 2435 2436 State.Stack.back().LastSpace = StartColumn; 2437 } 2438 2439 Token->updateNextToken(State); 2440 2441 return {Penalty, Exceeded}; 2442 } 2443 2444 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const { 2445 // In preprocessor directives reserve two chars for trailing " \" 2446 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0); 2447 } 2448 2449 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) { 2450 const FormatToken &Current = *State.NextToken; 2451 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral)) 2452 return false; 2453 // We never consider raw string literals "multiline" for the purpose of 2454 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased 2455 // (see TokenAnnotator::mustBreakBefore(). 2456 if (Current.TokenText.startswith("R\"")) 2457 return false; 2458 if (Current.IsMultiline) 2459 return true; 2460 if (Current.getNextNonComment() && 2461 Current.getNextNonComment()->isStringLiteral()) 2462 return true; // Implicit concatenation. 2463 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals && 2464 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength > 2465 Style.ColumnLimit) 2466 return true; // String will be split. 2467 return false; 2468 } 2469 2470 } // namespace format 2471 } // namespace clang 2472