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