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