1 //===--- ContinuationIndenter.cpp - Format C++ code -----------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 ///
10 /// \file
11 /// \brief This file implements the continuation indenter.
12 ///
13 //===----------------------------------------------------------------------===//
14 
15 #include "ContinuationIndenter.h"
16 #include "BreakableToken.h"
17 #include "FormatInternal.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/Support/Debug.h"
23 
24 #define DEBUG_TYPE "format-indenter"
25 
26 namespace clang {
27 namespace format {
28 
29 // Returns the length of everything up to the first possible line break after
30 // the ), ], } or > matching \c Tok.
31 static unsigned getLengthToMatchingParen(const FormatToken &Tok) {
32   if (!Tok.MatchingParen)
33     return 0;
34   FormatToken *End = Tok.MatchingParen;
35   while (End->Next && !End->Next->CanBreakBefore) {
36     End = End->Next;
37   }
38   return End->TotalLength - Tok.TotalLength + 1;
39 }
40 
41 static unsigned getLengthToNextOperator(const FormatToken &Tok) {
42   if (!Tok.NextOperator)
43     return 0;
44   return Tok.NextOperator->TotalLength - Tok.TotalLength;
45 }
46 
47 // Returns \c true if \c Tok is the "." or "->" of a call and starts the next
48 // segment of a builder type call.
49 static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) {
50   return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope();
51 }
52 
53 // Returns \c true if \c Current starts a new parameter.
54 static bool startsNextParameter(const FormatToken &Current,
55                                 const FormatStyle &Style) {
56   const FormatToken &Previous = *Current.Previous;
57   if (Current.is(TT_CtorInitializerComma) &&
58       Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma)
59     return true;
60   if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName))
61     return true;
62   return Previous.is(tok::comma) && !Current.isTrailingComment() &&
63          ((Previous.isNot(TT_CtorInitializerComma) ||
64            Style.BreakConstructorInitializers !=
65                FormatStyle::BCIS_BeforeComma) &&
66           (Previous.isNot(TT_InheritanceComma) ||
67            !Style.BreakBeforeInheritanceComma));
68 }
69 
70 static bool opensProtoMessageField(const FormatToken &LessTok,
71                                    const FormatStyle &Style) {
72   if (LessTok.isNot(tok::less))
73     return false;
74   return Style.Language == FormatStyle::LK_TextProto ||
75          (Style.Language == FormatStyle::LK_Proto &&
76           (LessTok.NestingLevel > 0 ||
77            (LessTok.Previous && LessTok.Previous->is(tok::equal))));
78 }
79 
80 // Returns the delimiter of a raw string literal, or None if TokenText is not
81 // the text of a raw string literal. The delimiter could be the empty string.
82 // For example, the delimiter of R"deli(cont)deli" is deli.
83 static llvm::Optional<StringRef> getRawStringDelimiter(StringRef TokenText) {
84   if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'.
85       || !TokenText.startswith("R\"") || !TokenText.endswith("\""))
86     return None;
87 
88   // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has
89   // size at most 16 by the standard, so the first '(' must be among the first
90   // 19 bytes.
91   size_t LParenPos = TokenText.substr(0, 19).find_first_of('(');
92   if (LParenPos == StringRef::npos)
93     return None;
94   StringRef Delimiter = TokenText.substr(2, LParenPos - 2);
95 
96   // Check that the string ends in ')Delimiter"'.
97   size_t RParenPos = TokenText.size() - Delimiter.size() - 2;
98   if (TokenText[RParenPos] != ')')
99     return None;
100   if (!TokenText.substr(RParenPos + 1).startswith(Delimiter))
101     return None;
102   return Delimiter;
103 }
104 
105 // Returns the canonical delimiter for \p Language, or the empty string if no
106 // canonical delimiter is specified.
107 static StringRef
108 getCanonicalRawStringDelimiter(const FormatStyle &Style,
109                                FormatStyle::LanguageKind Language) {
110   for (const auto &Format : Style.RawStringFormats) {
111     if (Format.Language == Language)
112       return StringRef(Format.CanonicalDelimiter);
113   }
114   return "";
115 }
116 
117 RawStringFormatStyleManager::RawStringFormatStyleManager(
118     const FormatStyle &CodeStyle) {
119   for (const auto &RawStringFormat : CodeStyle.RawStringFormats) {
120     llvm::Optional<FormatStyle> LanguageStyle =
121         CodeStyle.GetLanguageStyle(RawStringFormat.Language);
122     if (!LanguageStyle) {
123       FormatStyle PredefinedStyle;
124       if (!getPredefinedStyle(RawStringFormat.BasedOnStyle,
125                               RawStringFormat.Language, &PredefinedStyle)) {
126         PredefinedStyle = getLLVMStyle();
127         PredefinedStyle.Language = RawStringFormat.Language;
128       }
129       LanguageStyle = PredefinedStyle;
130     }
131     LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit;
132     for (StringRef Delimiter : RawStringFormat.Delimiters) {
133       DelimiterStyle.insert({Delimiter, *LanguageStyle});
134     }
135     for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions) {
136       EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle});
137     }
138   }
139 }
140 
141 llvm::Optional<FormatStyle>
142 RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const {
143   auto It = DelimiterStyle.find(Delimiter);
144   if (It == DelimiterStyle.end())
145     return None;
146   return It->second;
147 }
148 
149 llvm::Optional<FormatStyle>
150 RawStringFormatStyleManager::getEnclosingFunctionStyle(
151     StringRef EnclosingFunction) const {
152   auto It = EnclosingFunctionStyle.find(EnclosingFunction);
153   if (It == EnclosingFunctionStyle.end())
154     return None;
155   return It->second;
156 }
157 
158 ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style,
159                                            const AdditionalKeywords &Keywords,
160                                            const SourceManager &SourceMgr,
161                                            WhitespaceManager &Whitespaces,
162                                            encoding::Encoding Encoding,
163                                            bool BinPackInconclusiveFunctions)
164     : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr),
165       Whitespaces(Whitespaces), Encoding(Encoding),
166       BinPackInconclusiveFunctions(BinPackInconclusiveFunctions),
167       CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {}
168 
169 LineState ContinuationIndenter::getInitialState(unsigned FirstIndent,
170                                                 unsigned FirstStartColumn,
171                                                 const AnnotatedLine *Line,
172                                                 bool DryRun) {
173   LineState State;
174   State.FirstIndent = FirstIndent;
175   if (FirstStartColumn && Line->First->NewlinesBefore == 0)
176     State.Column = FirstStartColumn;
177   else
178     State.Column = FirstIndent;
179   // With preprocessor directive indentation, the line starts on column 0
180   // since it's indented after the hash, but FirstIndent is set to the
181   // preprocessor indent.
182   if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
183       (Line->Type == LT_PreprocessorDirective ||
184        Line->Type == LT_ImportStatement))
185     State.Column = 0;
186   State.Line = Line;
187   State.NextToken = Line->First;
188   State.Stack.push_back(ParenState(FirstIndent, FirstIndent,
189                                    /*AvoidBinPacking=*/false,
190                                    /*NoLineBreak=*/false));
191   State.LineContainsContinuedForLoopSection = false;
192   State.NoContinuation = false;
193   State.StartOfStringLiteral = 0;
194   State.StartOfLineLevel = 0;
195   State.LowestLevelOnLine = 0;
196   State.IgnoreStackForComparison = false;
197 
198   if (Style.Language == FormatStyle::LK_TextProto) {
199     // We need this in order to deal with the bin packing of text fields at
200     // global scope.
201     State.Stack.back().AvoidBinPacking = true;
202     State.Stack.back().BreakBeforeParameter = true;
203     State.Stack.back().AlignColons = false;
204   }
205 
206   // The first token has already been indented and thus consumed.
207   moveStateToNextToken(State, DryRun, /*Newline=*/false);
208   return State;
209 }
210 
211 bool ContinuationIndenter::canBreak(const LineState &State) {
212   const FormatToken &Current = *State.NextToken;
213   const FormatToken &Previous = *Current.Previous;
214   assert(&Previous == Current.Previous);
215   if (!Current.CanBreakBefore && !(State.Stack.back().BreakBeforeClosingBrace &&
216                                    Current.closesBlockOrBlockTypeList(Style)))
217     return false;
218   // The opening "{" of a braced list has to be on the same line as the first
219   // element if it is nested in another braced init list or function call.
220   if (!Current.MustBreakBefore && Previous.is(tok::l_brace) &&
221       Previous.isNot(TT_DictLiteral) && Previous.BlockKind == BK_BracedInit &&
222       Previous.Previous &&
223       Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma))
224     return false;
225   // This prevents breaks like:
226   //   ...
227   //   SomeParameter, OtherParameter).DoSomething(
228   //   ...
229   // As they hide "DoSomething" and are generally bad for readability.
230   if (Previous.opensScope() && Previous.isNot(tok::l_brace) &&
231       State.LowestLevelOnLine < State.StartOfLineLevel &&
232       State.LowestLevelOnLine < Current.NestingLevel)
233     return false;
234   if (Current.isMemberAccess() && State.Stack.back().ContainsUnwrappedBuilder)
235     return false;
236 
237   // Don't create a 'hanging' indent if there are multiple blocks in a single
238   // statement.
239   if (Previous.is(tok::l_brace) && State.Stack.size() > 1 &&
240       State.Stack[State.Stack.size() - 2].NestedBlockInlined &&
241       State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks)
242     return false;
243 
244   // Don't break after very short return types (e.g. "void") as that is often
245   // unexpected.
246   if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) {
247     if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None)
248       return false;
249   }
250 
251   // If binary operators are moved to the next line (including commas for some
252   // styles of constructor initializers), that's always ok.
253   if (!Current.isOneOf(TT_BinaryOperator, tok::comma) &&
254       State.Stack.back().NoLineBreakInOperand)
255     return false;
256 
257   return !State.Stack.back().NoLineBreak;
258 }
259 
260 bool ContinuationIndenter::mustBreak(const LineState &State) {
261   const FormatToken &Current = *State.NextToken;
262   const FormatToken &Previous = *Current.Previous;
263   if (Current.MustBreakBefore || Current.is(TT_InlineASMColon))
264     return true;
265   if (State.Stack.back().BreakBeforeClosingBrace &&
266       Current.closesBlockOrBlockTypeList(Style))
267     return true;
268   if (Previous.is(tok::semi) && State.LineContainsContinuedForLoopSection)
269     return true;
270   if (Style.Language == FormatStyle::LK_ObjC &&
271       Current.ObjCSelectorNameParts > 1 &&
272       Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) {
273     return true;
274   }
275   if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) ||
276        (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) &&
277         Style.isCpp() &&
278         // FIXME: This is a temporary workaround for the case where clang-format
279         // sets BreakBeforeParameter to avoid bin packing and this creates a
280         // completely unnecessary line break after a template type that isn't
281         // line-wrapped.
282         (Previous.NestingLevel == 1 || Style.BinPackParameters)) ||
283        (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) &&
284         Previous.isNot(tok::question)) ||
285        (!Style.BreakBeforeTernaryOperators &&
286         Previous.is(TT_ConditionalExpr))) &&
287       State.Stack.back().BreakBeforeParameter && !Current.isTrailingComment() &&
288       !Current.isOneOf(tok::r_paren, tok::r_brace))
289     return true;
290   if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) ||
291        (Previous.is(TT_ArrayInitializerLSquare) &&
292         Previous.ParameterCount > 1) ||
293        opensProtoMessageField(Previous, Style)) &&
294       Style.ColumnLimit > 0 &&
295       getLengthToMatchingParen(Previous) + State.Column - 1 >
296           getColumnLimit(State))
297     return true;
298 
299   const FormatToken &BreakConstructorInitializersToken =
300       Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon
301           ? Previous
302           : Current;
303   if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) &&
304       (State.Column + State.Line->Last->TotalLength - Previous.TotalLength >
305            getColumnLimit(State) ||
306        State.Stack.back().BreakBeforeParameter) &&
307       (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All ||
308        Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon ||
309        Style.ColumnLimit != 0))
310     return true;
311 
312   if (Current.is(TT_ObjCMethodExpr) && !Previous.is(TT_SelectorName) &&
313       State.Line->startsWith(TT_ObjCMethodSpecifier))
314     return true;
315   if (Current.is(TT_SelectorName) && State.Stack.back().ObjCSelectorNameFound &&
316       State.Stack.back().BreakBeforeParameter)
317     return true;
318 
319   unsigned NewLineColumn = getNewLineColumn(State);
320   if (Current.isMemberAccess() && Style.ColumnLimit != 0 &&
321       State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit &&
322       (State.Column > NewLineColumn ||
323        Current.NestingLevel < State.StartOfLineLevel))
324     return true;
325 
326   if (startsSegmentOfBuilderTypeCall(Current) &&
327       (State.Stack.back().CallContinuation != 0 ||
328        State.Stack.back().BreakBeforeParameter) &&
329       // JavaScript is treated different here as there is a frequent pattern:
330       //   SomeFunction(function() {
331       //     ...
332       //   }.bind(...));
333       // FIXME: We should find a more generic solution to this problem.
334       !(State.Column <= NewLineColumn &&
335         Style.Language == FormatStyle::LK_JavaScript))
336     return true;
337 
338   if (State.Column <= NewLineColumn)
339     return false;
340 
341   if (Style.AlwaysBreakBeforeMultilineStrings &&
342       (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth ||
343        Previous.is(tok::comma) || Current.NestingLevel < 2) &&
344       !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at) &&
345       !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) &&
346       nextIsMultilineString(State))
347     return true;
348 
349   // Using CanBreakBefore here and below takes care of the decision whether the
350   // current style uses wrapping before or after operators for the given
351   // operator.
352   if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) {
353     // If we need to break somewhere inside the LHS of a binary expression, we
354     // should also break after the operator. Otherwise, the formatting would
355     // hide the operator precedence, e.g. in:
356     //   if (aaaaaaaaaaaaaa ==
357     //           bbbbbbbbbbbbbb && c) {..
358     // For comparisons, we only apply this rule, if the LHS is a binary
359     // expression itself as otherwise, the line breaks seem superfluous.
360     // We need special cases for ">>" which we have split into two ">" while
361     // lexing in order to make template parsing easier.
362     bool IsComparison = (Previous.getPrecedence() == prec::Relational ||
363                          Previous.getPrecedence() == prec::Equality ||
364                          Previous.getPrecedence() == prec::Spaceship) &&
365                         Previous.Previous &&
366                         Previous.Previous->isNot(TT_BinaryOperator); // For >>.
367     bool LHSIsBinaryExpr =
368         Previous.Previous && Previous.Previous->EndsBinaryExpression;
369     if ((!IsComparison || LHSIsBinaryExpr) && !Current.isTrailingComment() &&
370         Previous.getPrecedence() != prec::Assignment &&
371         State.Stack.back().BreakBeforeParameter)
372       return true;
373   } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore &&
374              State.Stack.back().BreakBeforeParameter) {
375     return true;
376   }
377 
378   // Same as above, but for the first "<<" operator.
379   if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) &&
380       State.Stack.back().BreakBeforeParameter &&
381       State.Stack.back().FirstLessLess == 0)
382     return true;
383 
384   if (Current.NestingLevel == 0 && !Current.isTrailingComment()) {
385     // Always break after "template <...>" and leading annotations. This is only
386     // for cases where the entire line does not fit on a single line as a
387     // different LineFormatter would be used otherwise.
388     if (Previous.ClosesTemplateDeclaration)
389       return true;
390     if (Previous.is(TT_FunctionAnnotationRParen))
391       return true;
392     if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) &&
393         Current.isNot(TT_LeadingJavaAnnotation))
394       return true;
395   }
396 
397   // If the return type spans multiple lines, wrap before the function name.
398   if ((Current.is(TT_FunctionDeclarationName) ||
399        (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) &&
400       !Previous.is(tok::kw_template) && State.Stack.back().BreakBeforeParameter)
401     return true;
402 
403   // The following could be precomputed as they do not depend on the state.
404   // However, as they should take effect only if the UnwrappedLine does not fit
405   // into the ColumnLimit, they are checked here in the ContinuationIndenter.
406   if (Style.ColumnLimit != 0 && Previous.BlockKind == BK_Block &&
407       Previous.is(tok::l_brace) && !Current.isOneOf(tok::r_brace, tok::comment))
408     return true;
409 
410   if (Current.is(tok::lessless) &&
411       ((Previous.is(tok::identifier) && Previous.TokenText == "endl") ||
412        (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") ||
413                                      Previous.TokenText == "\'\\n\'"))))
414     return true;
415 
416   if (Previous.is(TT_BlockComment) && Previous.IsMultiline)
417     return true;
418 
419   if (State.NoContinuation)
420     return true;
421 
422   return false;
423 }
424 
425 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline,
426                                                bool DryRun,
427                                                unsigned ExtraSpaces) {
428   const FormatToken &Current = *State.NextToken;
429 
430   assert(!State.Stack.empty());
431   State.NoContinuation = false;
432 
433   if ((Current.is(TT_ImplicitStringLiteral) &&
434        (Current.Previous->Tok.getIdentifierInfo() == nullptr ||
435         Current.Previous->Tok.getIdentifierInfo()->getPPKeywordID() ==
436             tok::pp_not_keyword))) {
437     unsigned EndColumn =
438         SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd());
439     if (Current.LastNewlineOffset != 0) {
440       // If there is a newline within this token, the final column will solely
441       // determined by the current end column.
442       State.Column = EndColumn;
443     } else {
444       unsigned StartColumn =
445           SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin());
446       assert(EndColumn >= StartColumn);
447       State.Column += EndColumn - StartColumn;
448     }
449     moveStateToNextToken(State, DryRun, /*Newline=*/false);
450     return 0;
451   }
452 
453   unsigned Penalty = 0;
454   if (Newline)
455     Penalty = addTokenOnNewLine(State, DryRun);
456   else
457     addTokenOnCurrentLine(State, DryRun, ExtraSpaces);
458 
459   return moveStateToNextToken(State, DryRun, Newline) + Penalty;
460 }
461 
462 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
463                                                  unsigned ExtraSpaces) {
464   FormatToken &Current = *State.NextToken;
465   const FormatToken &Previous = *State.NextToken->Previous;
466   if (Current.is(tok::equal) &&
467       (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) &&
468       State.Stack.back().VariablePos == 0) {
469     State.Stack.back().VariablePos = State.Column;
470     // Move over * and & if they are bound to the variable name.
471     const FormatToken *Tok = &Previous;
472     while (Tok && State.Stack.back().VariablePos >= Tok->ColumnWidth) {
473       State.Stack.back().VariablePos -= Tok->ColumnWidth;
474       if (Tok->SpacesRequiredBefore != 0)
475         break;
476       Tok = Tok->Previous;
477     }
478     if (Previous.PartOfMultiVariableDeclStmt)
479       State.Stack.back().LastSpace = State.Stack.back().VariablePos;
480   }
481 
482   unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces;
483 
484   // Indent preprocessor directives after the hash if required.
485   int PPColumnCorrection = 0;
486   if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
487       Previous.is(tok::hash) && State.FirstIndent > 0 &&
488       (State.Line->Type == LT_PreprocessorDirective ||
489        State.Line->Type == LT_ImportStatement)) {
490     Spaces += State.FirstIndent;
491 
492     // For preprocessor indent with tabs, State.Column will be 1 because of the
493     // hash. This causes second-level indents onward to have an extra space
494     // after the tabs. We avoid this misalignment by subtracting 1 from the
495     // column value passed to replaceWhitespace().
496     if (Style.UseTab != FormatStyle::UT_Never)
497       PPColumnCorrection = -1;
498   }
499 
500   if (!DryRun)
501     Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces,
502                                   State.Column + Spaces + PPColumnCorrection);
503 
504   // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance
505   // declaration unless there is multiple inheritance.
506   if (Style.BreakBeforeInheritanceComma && Current.is(TT_InheritanceColon))
507     State.Stack.back().NoLineBreak = true;
508 
509   if (Current.is(TT_SelectorName) &&
510       !State.Stack.back().ObjCSelectorNameFound) {
511     unsigned MinIndent =
512         std::max(State.FirstIndent + Style.ContinuationIndentWidth,
513                  State.Stack.back().Indent);
514     unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth;
515     if (Current.LongestObjCSelectorName == 0)
516       State.Stack.back().AlignColons = false;
517     else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos)
518       State.Stack.back().ColonPos = MinIndent + Current.LongestObjCSelectorName;
519     else
520       State.Stack.back().ColonPos = FirstColonPos;
521   }
522 
523   // In "AlwaysBreak" mode, enforce wrapping directly after the parenthesis by
524   // disallowing any further line breaks if there is no line break after the
525   // opening parenthesis. Don't break if it doesn't conserve columns.
526   if (Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak &&
527       Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) &&
528       State.Column > getNewLineColumn(State) &&
529       (!Previous.Previous || !Previous.Previous->isOneOf(
530                                  tok::kw_for, tok::kw_while, tok::kw_switch)) &&
531       // Don't do this for simple (no expressions) one-argument function calls
532       // as that feels like needlessly wasting whitespace, e.g.:
533       //
534       //   caaaaaaaaaaaall(
535       //       caaaaaaaaaaaall(
536       //           caaaaaaaaaaaall(
537       //               caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa))));
538       Current.FakeLParens.size() > 0 &&
539       Current.FakeLParens.back() > prec::Unknown)
540     State.Stack.back().NoLineBreak = true;
541   if (Previous.is(TT_TemplateString) && Previous.opensScope())
542     State.Stack.back().NoLineBreak = true;
543 
544   if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign &&
545       Previous.opensScope() && Previous.isNot(TT_ObjCMethodExpr) &&
546       (Current.isNot(TT_LineComment) || Previous.BlockKind == BK_BracedInit))
547     State.Stack.back().Indent = State.Column + Spaces;
548   if (State.Stack.back().AvoidBinPacking && startsNextParameter(Current, Style))
549     State.Stack.back().NoLineBreak = true;
550   if (startsSegmentOfBuilderTypeCall(Current) &&
551       State.Column > getNewLineColumn(State))
552     State.Stack.back().ContainsUnwrappedBuilder = true;
553 
554   if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java)
555     State.Stack.back().NoLineBreak = true;
556   if (Current.isMemberAccess() && Previous.is(tok::r_paren) &&
557       (Previous.MatchingParen &&
558        (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10)))
559     // If there is a function call with long parameters, break before trailing
560     // calls. This prevents things like:
561     //   EXPECT_CALL(SomeLongParameter).Times(
562     //       2);
563     // We don't want to do this for short parameters as they can just be
564     // indexes.
565     State.Stack.back().NoLineBreak = true;
566 
567   // Don't allow the RHS of an operator to be split over multiple lines unless
568   // there is a line-break right after the operator.
569   // Exclude relational operators, as there, it is always more desirable to
570   // have the LHS 'left' of the RHS.
571   const FormatToken *P = Current.getPreviousNonComment();
572   if (!Current.is(tok::comment) && P &&
573       (P->isOneOf(TT_BinaryOperator, tok::comma) ||
574        (P->is(TT_ConditionalExpr) && P->is(tok::colon))) &&
575       !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) &&
576       P->getPrecedence() != prec::Assignment &&
577       P->getPrecedence() != prec::Relational &&
578       P->getPrecedence() != prec::Spaceship) {
579     bool BreakBeforeOperator =
580         P->MustBreakBefore || P->is(tok::lessless) ||
581         (P->is(TT_BinaryOperator) &&
582          Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) ||
583         (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators);
584     // Don't do this if there are only two operands. In these cases, there is
585     // always a nice vertical separation between them and the extra line break
586     // does not help.
587     bool HasTwoOperands =
588         P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr);
589     if ((!BreakBeforeOperator && !(HasTwoOperands && Style.AlignOperands)) ||
590         (!State.Stack.back().LastOperatorWrapped && BreakBeforeOperator))
591       State.Stack.back().NoLineBreakInOperand = true;
592   }
593 
594   State.Column += Spaces;
595   if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) &&
596       Previous.Previous &&
597       (Previous.Previous->isOneOf(tok::kw_if, tok::kw_for) ||
598        Previous.Previous->endsSequence(tok::kw_constexpr, tok::kw_if))) {
599     // Treat the condition inside an if as if it was a second function
600     // parameter, i.e. let nested calls have a continuation indent.
601     State.Stack.back().LastSpace = State.Column;
602     State.Stack.back().NestedBlockIndent = State.Column;
603   } else if (!Current.isOneOf(tok::comment, tok::caret) &&
604              ((Previous.is(tok::comma) &&
605                !Previous.is(TT_OverloadedOperator)) ||
606               (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) {
607     State.Stack.back().LastSpace = State.Column;
608   } else if (Previous.is(TT_CtorInitializerColon) &&
609              Style.BreakConstructorInitializers ==
610                  FormatStyle::BCIS_AfterColon) {
611     State.Stack.back().Indent = State.Column;
612     State.Stack.back().LastSpace = State.Column;
613   } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr,
614                                TT_CtorInitializerColon)) &&
615              ((Previous.getPrecedence() != prec::Assignment &&
616                (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 ||
617                 Previous.NextOperator)) ||
618               Current.StartsBinaryExpression)) {
619     // Indent relative to the RHS of the expression unless this is a simple
620     // assignment without binary expression on the RHS. Also indent relative to
621     // unary operators and the colons of constructor initializers.
622     State.Stack.back().LastSpace = State.Column;
623   } else if (Previous.is(TT_InheritanceColon)) {
624     State.Stack.back().Indent = State.Column;
625     State.Stack.back().LastSpace = State.Column;
626   } else if (Previous.opensScope()) {
627     // If a function has a trailing call, indent all parameters from the
628     // opening parenthesis. This avoids confusing indents like:
629     //   OuterFunction(InnerFunctionCall( // break
630     //       ParameterToInnerFunction))   // break
631     //       .SecondInnerFunctionCall();
632     bool HasTrailingCall = false;
633     if (Previous.MatchingParen) {
634       const FormatToken *Next = Previous.MatchingParen->getNextNonComment();
635       HasTrailingCall = Next && Next->isMemberAccess();
636     }
637     if (HasTrailingCall && State.Stack.size() > 1 &&
638         State.Stack[State.Stack.size() - 2].CallContinuation == 0)
639       State.Stack.back().LastSpace = State.Column;
640   }
641 }
642 
643 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State,
644                                                  bool DryRun) {
645   FormatToken &Current = *State.NextToken;
646   const FormatToken &Previous = *State.NextToken->Previous;
647 
648   // Extra penalty that needs to be added because of the way certain line
649   // breaks are chosen.
650   unsigned Penalty = 0;
651 
652   const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
653   const FormatToken *NextNonComment = Previous.getNextNonComment();
654   if (!NextNonComment)
655     NextNonComment = &Current;
656   // The first line break on any NestingLevel causes an extra penalty in order
657   // prefer similar line breaks.
658   if (!State.Stack.back().ContainsLineBreak)
659     Penalty += 15;
660   State.Stack.back().ContainsLineBreak = true;
661 
662   Penalty += State.NextToken->SplitPenalty;
663 
664   // Breaking before the first "<<" is generally not desirable if the LHS is
665   // short. Also always add the penalty if the LHS is split over multiple lines
666   // to avoid unnecessary line breaks that just work around this penalty.
667   if (NextNonComment->is(tok::lessless) &&
668       State.Stack.back().FirstLessLess == 0 &&
669       (State.Column <= Style.ColumnLimit / 3 ||
670        State.Stack.back().BreakBeforeParameter))
671     Penalty += Style.PenaltyBreakFirstLessLess;
672 
673   State.Column = getNewLineColumn(State);
674 
675   // Indent nested blocks relative to this column, unless in a very specific
676   // JavaScript special case where:
677   //
678   //   var loooooong_name =
679   //       function() {
680   //     // code
681   //   }
682   //
683   // is common and should be formatted like a free-standing function. The same
684   // goes for wrapping before the lambda return type arrow.
685   if (!Current.is(TT_LambdaArrow) &&
686       (Style.Language != FormatStyle::LK_JavaScript ||
687        Current.NestingLevel != 0 || !PreviousNonComment ||
688        !PreviousNonComment->is(tok::equal) ||
689        !Current.isOneOf(Keywords.kw_async, Keywords.kw_function)))
690     State.Stack.back().NestedBlockIndent = State.Column;
691 
692   if (NextNonComment->isMemberAccess()) {
693     if (State.Stack.back().CallContinuation == 0)
694       State.Stack.back().CallContinuation = State.Column;
695   } else if (NextNonComment->is(TT_SelectorName)) {
696     if (!State.Stack.back().ObjCSelectorNameFound) {
697       if (NextNonComment->LongestObjCSelectorName == 0) {
698         State.Stack.back().AlignColons = false;
699       } else {
700         State.Stack.back().ColonPos =
701             (Style.IndentWrappedFunctionNames
702                  ? std::max(State.Stack.back().Indent,
703                             State.FirstIndent + Style.ContinuationIndentWidth)
704                  : State.Stack.back().Indent) +
705             std::max(NextNonComment->LongestObjCSelectorName,
706                      NextNonComment->ColumnWidth);
707       }
708     } else if (State.Stack.back().AlignColons &&
709                State.Stack.back().ColonPos <= NextNonComment->ColumnWidth) {
710       State.Stack.back().ColonPos = State.Column + NextNonComment->ColumnWidth;
711     }
712   } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
713              PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) {
714     // FIXME: This is hacky, find a better way. The problem is that in an ObjC
715     // method expression, the block should be aligned to the line starting it,
716     // e.g.:
717     //   [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason
718     //                        ^(int *i) {
719     //                            // ...
720     //                        }];
721     // Thus, we set LastSpace of the next higher NestingLevel, to which we move
722     // when we consume all of the "}"'s FakeRParens at the "{".
723     if (State.Stack.size() > 1)
724       State.Stack[State.Stack.size() - 2].LastSpace =
725           std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) +
726           Style.ContinuationIndentWidth;
727   }
728 
729   if ((PreviousNonComment &&
730        PreviousNonComment->isOneOf(tok::comma, tok::semi) &&
731        !State.Stack.back().AvoidBinPacking) ||
732       Previous.is(TT_BinaryOperator))
733     State.Stack.back().BreakBeforeParameter = false;
734   if (Previous.isOneOf(TT_TemplateCloser, TT_JavaAnnotation) &&
735       Current.NestingLevel == 0)
736     State.Stack.back().BreakBeforeParameter = false;
737   if (NextNonComment->is(tok::question) ||
738       (PreviousNonComment && PreviousNonComment->is(tok::question)))
739     State.Stack.back().BreakBeforeParameter = true;
740   if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore)
741     State.Stack.back().BreakBeforeParameter = false;
742 
743   if (!DryRun) {
744     unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1;
745     if (Current.is(tok::r_brace) && Current.MatchingParen &&
746         // Only strip trailing empty lines for l_braces that have children, i.e.
747         // for function expressions (lambdas, arrows, etc).
748         !Current.MatchingParen->Children.empty()) {
749       // lambdas and arrow functions are expressions, thus their r_brace is not
750       // on its own line, and thus not covered by UnwrappedLineFormatter's logic
751       // about removing empty lines on closing blocks. Special case them here.
752       MaxEmptyLinesToKeep = 1;
753     }
754     unsigned Newlines = std::max(
755         1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep));
756     bool ContinuePPDirective =
757         State.Line->InPPDirective && State.Line->Type != LT_ImportStatement;
758     Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column,
759                                   ContinuePPDirective);
760   }
761 
762   if (!Current.isTrailingComment())
763     State.Stack.back().LastSpace = State.Column;
764   if (Current.is(tok::lessless))
765     // If we are breaking before a "<<", we always want to indent relative to
766     // RHS. This is necessary only for "<<", as we special-case it and don't
767     // always indent relative to the RHS.
768     State.Stack.back().LastSpace += 3; // 3 -> width of "<< ".
769 
770   State.StartOfLineLevel = Current.NestingLevel;
771   State.LowestLevelOnLine = Current.NestingLevel;
772 
773   // Any break on this level means that the parent level has been broken
774   // and we need to avoid bin packing there.
775   bool NestedBlockSpecialCase =
776       !Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 &&
777       State.Stack[State.Stack.size() - 2].NestedBlockInlined;
778   if (!NestedBlockSpecialCase)
779     for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i)
780       State.Stack[i].BreakBeforeParameter = true;
781 
782   if (PreviousNonComment &&
783       !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) &&
784       (PreviousNonComment->isNot(TT_TemplateCloser) ||
785        Current.NestingLevel != 0) &&
786       !PreviousNonComment->isOneOf(
787           TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation,
788           TT_LeadingJavaAnnotation) &&
789       Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope())
790     State.Stack.back().BreakBeforeParameter = true;
791 
792   // If we break after { or the [ of an array initializer, we should also break
793   // before the corresponding } or ].
794   if (PreviousNonComment &&
795       (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
796        opensProtoMessageField(*PreviousNonComment, Style)))
797     State.Stack.back().BreakBeforeClosingBrace = true;
798 
799   if (State.Stack.back().AvoidBinPacking) {
800     // If we are breaking after '(', '{', '<', this is not bin packing
801     // unless AllowAllParametersOfDeclarationOnNextLine is false or this is a
802     // dict/object literal.
803     if (!Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) ||
804         (!Style.AllowAllParametersOfDeclarationOnNextLine &&
805          State.Line->MustBeDeclaration) ||
806         Previous.is(TT_DictLiteral))
807       State.Stack.back().BreakBeforeParameter = true;
808   }
809 
810   return Penalty;
811 }
812 
813 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
814   if (!State.NextToken || !State.NextToken->Previous)
815     return 0;
816   FormatToken &Current = *State.NextToken;
817   const FormatToken &Previous = *Current.Previous;
818   // If we are continuing an expression, we want to use the continuation indent.
819   unsigned ContinuationIndent =
820       std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) +
821       Style.ContinuationIndentWidth;
822   const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
823   const FormatToken *NextNonComment = Previous.getNextNonComment();
824   if (!NextNonComment)
825     NextNonComment = &Current;
826 
827   // Java specific bits.
828   if (Style.Language == FormatStyle::LK_Java &&
829       Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends))
830     return std::max(State.Stack.back().LastSpace,
831                     State.Stack.back().Indent + Style.ContinuationIndentWidth);
832 
833   if (NextNonComment->is(tok::l_brace) && NextNonComment->BlockKind == BK_Block)
834     return Current.NestingLevel == 0 ? State.FirstIndent
835                                      : State.Stack.back().Indent;
836   if ((Current.isOneOf(tok::r_brace, tok::r_square) ||
837        (Current.is(tok::greater) &&
838         (Style.Language == FormatStyle::LK_Proto ||
839          Style.Language == FormatStyle::LK_TextProto))) &&
840       State.Stack.size() > 1) {
841     if (Current.closesBlockOrBlockTypeList(Style))
842       return State.Stack[State.Stack.size() - 2].NestedBlockIndent;
843     if (Current.MatchingParen &&
844         Current.MatchingParen->BlockKind == BK_BracedInit)
845       return State.Stack[State.Stack.size() - 2].LastSpace;
846     return State.FirstIndent;
847   }
848   // Indent a closing parenthesis at the previous level if followed by a semi or
849   // opening brace. This allows indentations such as:
850   //     foo(
851   //       a,
852   //     );
853   //     function foo(
854   //       a,
855   //     ) {
856   //       code(); //
857   //     }
858   if (Current.is(tok::r_paren) && State.Stack.size() > 1 &&
859       (!Current.Next || Current.Next->isOneOf(tok::semi, tok::l_brace)))
860     return State.Stack[State.Stack.size() - 2].LastSpace;
861   if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope())
862     return State.Stack[State.Stack.size() - 2].LastSpace;
863   if (Current.is(tok::identifier) && Current.Next &&
864       (Current.Next->is(TT_DictLiteral) ||
865        ((Style.Language == FormatStyle::LK_Proto ||
866          Style.Language == FormatStyle::LK_TextProto) &&
867         Current.Next->isOneOf(tok::less, tok::l_brace))))
868     return State.Stack.back().Indent;
869   if (NextNonComment->is(TT_ObjCStringLiteral) &&
870       State.StartOfStringLiteral != 0)
871     return State.StartOfStringLiteral - 1;
872   if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0)
873     return State.StartOfStringLiteral;
874   if (NextNonComment->is(tok::lessless) &&
875       State.Stack.back().FirstLessLess != 0)
876     return State.Stack.back().FirstLessLess;
877   if (NextNonComment->isMemberAccess()) {
878     if (State.Stack.back().CallContinuation == 0)
879       return ContinuationIndent;
880     return State.Stack.back().CallContinuation;
881   }
882   if (State.Stack.back().QuestionColumn != 0 &&
883       ((NextNonComment->is(tok::colon) &&
884         NextNonComment->is(TT_ConditionalExpr)) ||
885        Previous.is(TT_ConditionalExpr)))
886     return State.Stack.back().QuestionColumn;
887   if (Previous.is(tok::comma) && State.Stack.back().VariablePos != 0)
888     return State.Stack.back().VariablePos;
889   if ((PreviousNonComment &&
890        (PreviousNonComment->ClosesTemplateDeclaration ||
891         PreviousNonComment->isOneOf(
892             TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen,
893             TT_JavaAnnotation, TT_LeadingJavaAnnotation))) ||
894       (!Style.IndentWrappedFunctionNames &&
895        NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName)))
896     return std::max(State.Stack.back().LastSpace, State.Stack.back().Indent);
897   if (NextNonComment->is(TT_SelectorName)) {
898     if (!State.Stack.back().ObjCSelectorNameFound) {
899       unsigned MinIndent = State.Stack.back().Indent;
900       if (Style.IndentWrappedFunctionNames)
901         MinIndent = std::max(MinIndent,
902                              State.FirstIndent + Style.ContinuationIndentWidth);
903       // If LongestObjCSelectorName is 0, we are indenting the first
904       // part of an ObjC selector (or a selector component which is
905       // not colon-aligned due to block formatting).
906       //
907       // Otherwise, we are indenting a subsequent part of an ObjC
908       // selector which should be colon-aligned to the longest
909       // component of the ObjC selector.
910       //
911       // In either case, we want to respect Style.IndentWrappedFunctionNames.
912       return MinIndent +
913              std::max(NextNonComment->LongestObjCSelectorName,
914                       NextNonComment->ColumnWidth) -
915              NextNonComment->ColumnWidth;
916     }
917     if (!State.Stack.back().AlignColons)
918       return State.Stack.back().Indent;
919     if (State.Stack.back().ColonPos > NextNonComment->ColumnWidth)
920       return State.Stack.back().ColonPos - NextNonComment->ColumnWidth;
921     return State.Stack.back().Indent;
922   }
923   if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr))
924     return State.Stack.back().ColonPos;
925   if (NextNonComment->is(TT_ArraySubscriptLSquare)) {
926     if (State.Stack.back().StartOfArraySubscripts != 0)
927       return State.Stack.back().StartOfArraySubscripts;
928     return ContinuationIndent;
929   }
930 
931   // This ensure that we correctly format ObjC methods calls without inputs,
932   // i.e. where the last element isn't selector like: [callee method];
933   if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 &&
934       NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr))
935     return State.Stack.back().Indent;
936 
937   if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) ||
938       Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon))
939     return ContinuationIndent;
940   if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
941       PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral))
942     return ContinuationIndent;
943   if (NextNonComment->is(TT_CtorInitializerComma))
944     return State.Stack.back().Indent;
945   if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) &&
946       Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon)
947     return State.Stack.back().Indent;
948   if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon,
949                               TT_InheritanceComma))
950     return State.FirstIndent + Style.ConstructorInitializerIndentWidth;
951   if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() &&
952       !Current.isOneOf(tok::colon, tok::comment))
953     return ContinuationIndent;
954   if (Current.is(TT_ProtoExtensionLSquare))
955     return State.Stack.back().Indent;
956   if (State.Stack.back().Indent == State.FirstIndent && PreviousNonComment &&
957       PreviousNonComment->isNot(tok::r_brace))
958     // Ensure that we fall back to the continuation indent width instead of
959     // just flushing continuations left.
960     return State.Stack.back().Indent + Style.ContinuationIndentWidth;
961   return State.Stack.back().Indent;
962 }
963 
964 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State,
965                                                     bool DryRun, bool Newline) {
966   assert(State.Stack.size());
967   const FormatToken &Current = *State.NextToken;
968 
969   if (Current.isOneOf(tok::comma, TT_BinaryOperator))
970     State.Stack.back().NoLineBreakInOperand = false;
971   if (Current.is(TT_InheritanceColon))
972     State.Stack.back().AvoidBinPacking = true;
973   if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) {
974     if (State.Stack.back().FirstLessLess == 0)
975       State.Stack.back().FirstLessLess = State.Column;
976     else
977       State.Stack.back().LastOperatorWrapped = Newline;
978   }
979   if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless))
980     State.Stack.back().LastOperatorWrapped = Newline;
981   if (Current.is(TT_ConditionalExpr) && Current.Previous &&
982       !Current.Previous->is(TT_ConditionalExpr))
983     State.Stack.back().LastOperatorWrapped = Newline;
984   if (Current.is(TT_ArraySubscriptLSquare) &&
985       State.Stack.back().StartOfArraySubscripts == 0)
986     State.Stack.back().StartOfArraySubscripts = State.Column;
987   if (Style.BreakBeforeTernaryOperators && Current.is(tok::question))
988     State.Stack.back().QuestionColumn = State.Column;
989   if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) {
990     const FormatToken *Previous = Current.Previous;
991     while (Previous && Previous->isTrailingComment())
992       Previous = Previous->Previous;
993     if (Previous && Previous->is(tok::question))
994       State.Stack.back().QuestionColumn = State.Column;
995   }
996   if (!Current.opensScope() && !Current.closesScope() &&
997       !Current.is(TT_PointerOrReference))
998     State.LowestLevelOnLine =
999         std::min(State.LowestLevelOnLine, Current.NestingLevel);
1000   if (Current.isMemberAccess())
1001     State.Stack.back().StartOfFunctionCall =
1002         !Current.NextOperator ? 0 : State.Column;
1003   if (Current.is(TT_SelectorName)) {
1004     State.Stack.back().ObjCSelectorNameFound = true;
1005     if (Style.IndentWrappedFunctionNames) {
1006       State.Stack.back().Indent =
1007           State.FirstIndent + Style.ContinuationIndentWidth;
1008     }
1009   }
1010   if (Current.is(TT_CtorInitializerColon) &&
1011       Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) {
1012     // Indent 2 from the column, so:
1013     // SomeClass::SomeClass()
1014     //     : First(...), ...
1015     //       Next(...)
1016     //       ^ line up here.
1017     State.Stack.back().Indent =
1018         State.Column +
1019         (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma
1020              ? 0
1021              : 2);
1022     State.Stack.back().NestedBlockIndent = State.Stack.back().Indent;
1023     if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
1024       State.Stack.back().AvoidBinPacking = true;
1025     State.Stack.back().BreakBeforeParameter = false;
1026   }
1027   if (Current.is(TT_CtorInitializerColon) &&
1028       Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1029     State.Stack.back().Indent =
1030         State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1031     State.Stack.back().NestedBlockIndent = State.Stack.back().Indent;
1032     if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
1033       State.Stack.back().AvoidBinPacking = true;
1034   }
1035   if (Current.is(TT_InheritanceColon))
1036     State.Stack.back().Indent =
1037         State.FirstIndent + Style.ContinuationIndentWidth;
1038   if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline)
1039     State.Stack.back().NestedBlockIndent =
1040         State.Column + Current.ColumnWidth + 1;
1041   if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow))
1042     State.Stack.back().LastSpace = State.Column;
1043 
1044   // Insert scopes created by fake parenthesis.
1045   const FormatToken *Previous = Current.getPreviousNonComment();
1046 
1047   // Add special behavior to support a format commonly used for JavaScript
1048   // closures:
1049   //   SomeFunction(function() {
1050   //     foo();
1051   //     bar();
1052   //   }, a, b, c);
1053   if (Current.isNot(tok::comment) && Previous &&
1054       Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
1055       !Previous->is(TT_DictLiteral) && State.Stack.size() > 1) {
1056     if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline)
1057       for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i)
1058         State.Stack[i].NoLineBreak = true;
1059     State.Stack[State.Stack.size() - 2].NestedBlockInlined = false;
1060   }
1061   if (Previous &&
1062       (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) ||
1063        Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr)) &&
1064       !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) {
1065     State.Stack.back().NestedBlockInlined =
1066         !Newline &&
1067         (Previous->isNot(tok::l_paren) || Previous->ParameterCount > 1);
1068   }
1069 
1070   moveStatePastFakeLParens(State, Newline);
1071   moveStatePastScopeCloser(State);
1072   bool AllowBreak = !State.Stack.back().NoLineBreak &&
1073                     !State.Stack.back().NoLineBreakInOperand;
1074   moveStatePastScopeOpener(State, Newline);
1075   moveStatePastFakeRParens(State);
1076 
1077   if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0)
1078     State.StartOfStringLiteral = State.Column + 1;
1079   else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0)
1080     State.StartOfStringLiteral = State.Column;
1081   else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) &&
1082            !Current.isStringLiteral())
1083     State.StartOfStringLiteral = 0;
1084 
1085   State.Column += Current.ColumnWidth;
1086   State.NextToken = State.NextToken->Next;
1087 
1088   unsigned Penalty =
1089       handleEndOfLine(Current, State, DryRun, AllowBreak);
1090 
1091   if (Current.Role)
1092     Current.Role->formatFromToken(State, this, DryRun);
1093   // If the previous has a special role, let it consume tokens as appropriate.
1094   // It is necessary to start at the previous token for the only implemented
1095   // role (comma separated list). That way, the decision whether or not to break
1096   // after the "{" is already done and both options are tried and evaluated.
1097   // FIXME: This is ugly, find a better way.
1098   if (Previous && Previous->Role)
1099     Penalty += Previous->Role->formatAfterToken(State, this, DryRun);
1100 
1101   return Penalty;
1102 }
1103 
1104 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State,
1105                                                     bool Newline) {
1106   const FormatToken &Current = *State.NextToken;
1107   const FormatToken *Previous = Current.getPreviousNonComment();
1108 
1109   // Don't add extra indentation for the first fake parenthesis after
1110   // 'return', assignments or opening <({[. The indentation for these cases
1111   // is special cased.
1112   bool SkipFirstExtraIndent =
1113       (Previous && (Previous->opensScope() ||
1114                     Previous->isOneOf(tok::semi, tok::kw_return) ||
1115                     (Previous->getPrecedence() == prec::Assignment &&
1116                      Style.AlignOperands) ||
1117                     Previous->is(TT_ObjCMethodExpr)));
1118   for (SmallVectorImpl<prec::Level>::const_reverse_iterator
1119            I = Current.FakeLParens.rbegin(),
1120            E = Current.FakeLParens.rend();
1121        I != E; ++I) {
1122     ParenState NewParenState = State.Stack.back();
1123     NewParenState.ContainsLineBreak = false;
1124     NewParenState.LastOperatorWrapped = true;
1125     NewParenState.NoLineBreak =
1126         NewParenState.NoLineBreak || State.Stack.back().NoLineBreakInOperand;
1127 
1128     // Don't propagate AvoidBinPacking into subexpressions of arg/param lists.
1129     if (*I > prec::Comma)
1130       NewParenState.AvoidBinPacking = false;
1131 
1132     // Indent from 'LastSpace' unless these are fake parentheses encapsulating
1133     // a builder type call after 'return' or, if the alignment after opening
1134     // brackets is disabled.
1135     if (!Current.isTrailingComment() &&
1136         (Style.AlignOperands || *I < prec::Assignment) &&
1137         (!Previous || Previous->isNot(tok::kw_return) ||
1138          (Style.Language != FormatStyle::LK_Java && *I > 0)) &&
1139         (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign ||
1140          *I != prec::Comma || Current.NestingLevel == 0))
1141       NewParenState.Indent =
1142           std::max(std::max(State.Column, NewParenState.Indent),
1143                    State.Stack.back().LastSpace);
1144 
1145     // Do not indent relative to the fake parentheses inserted for "." or "->".
1146     // This is a special case to make the following to statements consistent:
1147     //   OuterFunction(InnerFunctionCall( // break
1148     //       ParameterToInnerFunction));
1149     //   OuterFunction(SomeObject.InnerFunctionCall( // break
1150     //       ParameterToInnerFunction));
1151     if (*I > prec::Unknown)
1152       NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column);
1153     if (*I != prec::Conditional && !Current.is(TT_UnaryOperator) &&
1154         Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign)
1155       NewParenState.StartOfFunctionCall = State.Column;
1156 
1157     // Always indent conditional expressions. Never indent expression where
1158     // the 'operator' is ',', ';' or an assignment (i.e. *I <=
1159     // prec::Assignment) as those have different indentation rules. Indent
1160     // other expression, unless the indentation needs to be skipped.
1161     if (*I == prec::Conditional ||
1162         (!SkipFirstExtraIndent && *I > prec::Assignment &&
1163          !Current.isTrailingComment()))
1164       NewParenState.Indent += Style.ContinuationIndentWidth;
1165     if ((Previous && !Previous->opensScope()) || *I != prec::Comma)
1166       NewParenState.BreakBeforeParameter = false;
1167     State.Stack.push_back(NewParenState);
1168     SkipFirstExtraIndent = false;
1169   }
1170 }
1171 
1172 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) {
1173   for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) {
1174     unsigned VariablePos = State.Stack.back().VariablePos;
1175     if (State.Stack.size() == 1) {
1176       // Do not pop the last element.
1177       break;
1178     }
1179     State.Stack.pop_back();
1180     State.Stack.back().VariablePos = VariablePos;
1181   }
1182 }
1183 
1184 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
1185                                                     bool Newline) {
1186   const FormatToken &Current = *State.NextToken;
1187   if (!Current.opensScope())
1188     return;
1189 
1190   if (Current.MatchingParen && Current.BlockKind == BK_Block) {
1191     moveStateToNewBlock(State);
1192     return;
1193   }
1194 
1195   unsigned NewIndent;
1196   unsigned LastSpace = State.Stack.back().LastSpace;
1197   bool AvoidBinPacking;
1198   bool BreakBeforeParameter = false;
1199   unsigned NestedBlockIndent = std::max(State.Stack.back().StartOfFunctionCall,
1200                                         State.Stack.back().NestedBlockIndent);
1201   if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
1202       opensProtoMessageField(Current, Style)) {
1203     if (Current.opensBlockOrBlockTypeList(Style)) {
1204       NewIndent = Style.IndentWidth +
1205                   std::min(State.Column, State.Stack.back().NestedBlockIndent);
1206     } else {
1207       NewIndent = State.Stack.back().LastSpace + Style.ContinuationIndentWidth;
1208     }
1209     const FormatToken *NextNoComment = Current.getNextNonComment();
1210     bool EndsInComma = Current.MatchingParen &&
1211                        Current.MatchingParen->Previous &&
1212                        Current.MatchingParen->Previous->is(tok::comma);
1213     AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) ||
1214                       Style.Language == FormatStyle::LK_Proto ||
1215                       Style.Language == FormatStyle::LK_TextProto ||
1216                       !Style.BinPackArguments ||
1217                       (NextNoComment &&
1218                        NextNoComment->isOneOf(TT_DesignatedInitializerPeriod,
1219                                               TT_DesignatedInitializerLSquare));
1220     BreakBeforeParameter = EndsInComma;
1221     if (Current.ParameterCount > 1)
1222       NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1);
1223   } else {
1224     NewIndent = Style.ContinuationIndentWidth +
1225                 std::max(State.Stack.back().LastSpace,
1226                          State.Stack.back().StartOfFunctionCall);
1227 
1228     // Ensure that different different brackets force relative alignment, e.g.:
1229     // void SomeFunction(vector<  // break
1230     //                       int> v);
1231     // FIXME: We likely want to do this for more combinations of brackets.
1232     if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) {
1233       NewIndent = std::max(NewIndent, State.Stack.back().Indent);
1234       LastSpace = std::max(LastSpace, State.Stack.back().Indent);
1235     }
1236 
1237     bool EndsInComma =
1238         Current.MatchingParen &&
1239         Current.MatchingParen->getPreviousNonComment() &&
1240         Current.MatchingParen->getPreviousNonComment()->is(tok::comma);
1241 
1242     // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters
1243     // for backwards compatibility.
1244     bool ObjCBinPackProtocolList =
1245         (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto &&
1246          Style.BinPackParameters) ||
1247         Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always;
1248 
1249     bool BinPackDeclaration =
1250         (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) ||
1251         (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList);
1252 
1253     AvoidBinPacking =
1254         (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) ||
1255         (State.Line->MustBeDeclaration && !BinPackDeclaration) ||
1256         (!State.Line->MustBeDeclaration && !Style.BinPackArguments) ||
1257         (Style.ExperimentalAutoDetectBinPacking &&
1258          (Current.PackingKind == PPK_OnePerLine ||
1259           (!BinPackInconclusiveFunctions &&
1260            Current.PackingKind == PPK_Inconclusive)));
1261 
1262     if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen) {
1263       if (Style.ColumnLimit) {
1264         // If this '[' opens an ObjC call, determine whether all parameters fit
1265         // into one line and put one per line if they don't.
1266         if (getLengthToMatchingParen(Current) + State.Column >
1267             getColumnLimit(State))
1268           BreakBeforeParameter = true;
1269       } else {
1270         // For ColumnLimit = 0, we have to figure out whether there is or has to
1271         // be a line break within this call.
1272         for (const FormatToken *Tok = &Current;
1273              Tok && Tok != Current.MatchingParen; Tok = Tok->Next) {
1274           if (Tok->MustBreakBefore ||
1275               (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) {
1276             BreakBeforeParameter = true;
1277             break;
1278           }
1279         }
1280       }
1281     }
1282 
1283     if (Style.Language == FormatStyle::LK_JavaScript && EndsInComma)
1284       BreakBeforeParameter = true;
1285   }
1286   // Generally inherit NoLineBreak from the current scope to nested scope.
1287   // However, don't do this for non-empty nested blocks, dict literals and
1288   // array literals as these follow different indentation rules.
1289   bool NoLineBreak =
1290       Current.Children.empty() &&
1291       !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) &&
1292       (State.Stack.back().NoLineBreak ||
1293        State.Stack.back().NoLineBreakInOperand ||
1294        (Current.is(TT_TemplateOpener) &&
1295         State.Stack.back().ContainsUnwrappedBuilder));
1296   State.Stack.push_back(
1297       ParenState(NewIndent, LastSpace, AvoidBinPacking, NoLineBreak));
1298   State.Stack.back().NestedBlockIndent = NestedBlockIndent;
1299   State.Stack.back().BreakBeforeParameter = BreakBeforeParameter;
1300   State.Stack.back().HasMultipleNestedBlocks = Current.BlockParameterCount > 1;
1301   State.Stack.back().IsInsideObjCArrayLiteral =
1302       Current.is(TT_ArrayInitializerLSquare) && Current.Previous &&
1303       Current.Previous->is(tok::at);
1304 }
1305 
1306 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) {
1307   const FormatToken &Current = *State.NextToken;
1308   if (!Current.closesScope())
1309     return;
1310 
1311   // If we encounter a closing ), ], } or >, we can remove a level from our
1312   // stacks.
1313   if (State.Stack.size() > 1 &&
1314       (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) ||
1315        (Current.is(tok::r_brace) && State.NextToken != State.Line->First) ||
1316        State.NextToken->is(TT_TemplateCloser) ||
1317        (Current.is(tok::greater) && Current.is(TT_DictLiteral))))
1318     State.Stack.pop_back();
1319 
1320   if (Current.is(tok::r_square)) {
1321     // If this ends the array subscript expr, reset the corresponding value.
1322     const FormatToken *NextNonComment = Current.getNextNonComment();
1323     if (NextNonComment && NextNonComment->isNot(tok::l_square))
1324       State.Stack.back().StartOfArraySubscripts = 0;
1325   }
1326 }
1327 
1328 void ContinuationIndenter::moveStateToNewBlock(LineState &State) {
1329   unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent;
1330   // ObjC block sometimes follow special indentation rules.
1331   unsigned NewIndent =
1332       NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace)
1333                                ? Style.ObjCBlockIndentWidth
1334                                : Style.IndentWidth);
1335   State.Stack.push_back(ParenState(NewIndent, State.Stack.back().LastSpace,
1336                                    /*AvoidBinPacking=*/true,
1337                                    /*NoLineBreak=*/false));
1338   State.Stack.back().NestedBlockIndent = NestedBlockIndent;
1339   State.Stack.back().BreakBeforeParameter = true;
1340 }
1341 
1342 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn,
1343                                      unsigned TabWidth,
1344                                      encoding::Encoding Encoding) {
1345   size_t LastNewlinePos = Text.find_last_of("\n");
1346   if (LastNewlinePos == StringRef::npos) {
1347     return StartColumn +
1348            encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding);
1349   } else {
1350     return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos),
1351                                          /*StartColumn=*/0, TabWidth, Encoding);
1352   }
1353 }
1354 
1355 unsigned ContinuationIndenter::reformatRawStringLiteral(
1356     const FormatToken &Current, LineState &State,
1357     const FormatStyle &RawStringStyle, bool DryRun) {
1358   unsigned StartColumn = State.Column - Current.ColumnWidth;
1359   StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText);
1360   StringRef NewDelimiter =
1361       getCanonicalRawStringDelimiter(Style, RawStringStyle.Language);
1362   if (NewDelimiter.empty() || OldDelimiter.empty())
1363     NewDelimiter = OldDelimiter;
1364   // The text of a raw string is between the leading 'R"delimiter(' and the
1365   // trailing 'delimiter)"'.
1366   unsigned OldPrefixSize = 3 + OldDelimiter.size();
1367   unsigned OldSuffixSize = 2 + OldDelimiter.size();
1368   // We create a virtual text environment which expects a null-terminated
1369   // string, so we cannot use StringRef.
1370   std::string RawText =
1371       Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize);
1372   if (NewDelimiter != OldDelimiter) {
1373     // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the
1374     // raw string.
1375     std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str();
1376     if (StringRef(RawText).contains(CanonicalDelimiterSuffix))
1377       NewDelimiter = OldDelimiter;
1378   }
1379 
1380   unsigned NewPrefixSize = 3 + NewDelimiter.size();
1381   unsigned NewSuffixSize = 2 + NewDelimiter.size();
1382 
1383   // The first start column is the column the raw text starts after formatting.
1384   unsigned FirstStartColumn = StartColumn + NewPrefixSize;
1385 
1386   // The next start column is the intended indentation a line break inside
1387   // the raw string at level 0. It is determined by the following rules:
1388   //   - if the content starts on newline, it is one level more than the current
1389   //     indent, and
1390   //   - if the content does not start on a newline, it is the first start
1391   //     column.
1392   // These rules have the advantage that the formatted content both does not
1393   // violate the rectangle rule and visually flows within the surrounding
1394   // source.
1395   bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n';
1396   unsigned NextStartColumn =
1397       ContentStartsOnNewline
1398           ? State.Stack.back().NestedBlockIndent + Style.IndentWidth
1399           : FirstStartColumn;
1400 
1401   // The last start column is the column the raw string suffix starts if it is
1402   // put on a newline.
1403   // The last start column is the intended indentation of the raw string postfix
1404   // if it is put on a newline. It is determined by the following rules:
1405   //   - if the raw string prefix starts on a newline, it is the column where
1406   //     that raw string prefix starts, and
1407   //   - if the raw string prefix does not start on a newline, it is the current
1408   //     indent.
1409   unsigned LastStartColumn = Current.NewlinesBefore
1410                                  ? FirstStartColumn - NewPrefixSize
1411                                  : State.Stack.back().NestedBlockIndent;
1412 
1413   std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat(
1414       RawStringStyle, RawText, {tooling::Range(0, RawText.size())},
1415       FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>",
1416       /*Status=*/nullptr);
1417 
1418   auto NewCode = applyAllReplacements(RawText, Fixes.first);
1419   tooling::Replacements NoFixes;
1420   if (!NewCode) {
1421     State.Column += Current.ColumnWidth;
1422     return 0;
1423   }
1424   if (!DryRun) {
1425     if (NewDelimiter != OldDelimiter) {
1426       // In 'R"delimiter(...', the delimiter starts 2 characters after the start
1427       // of the token.
1428       SourceLocation PrefixDelimiterStart =
1429           Current.Tok.getLocation().getLocWithOffset(2);
1430       auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement(
1431           SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter));
1432       if (PrefixErr) {
1433         llvm::errs()
1434             << "Failed to update the prefix delimiter of a raw string: "
1435             << llvm::toString(std::move(PrefixErr)) << "\n";
1436       }
1437       // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at
1438       // position length - 1 - |delimiter|.
1439       SourceLocation SuffixDelimiterStart =
1440           Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() -
1441                                                      1 - OldDelimiter.size());
1442       auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement(
1443           SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter));
1444       if (SuffixErr) {
1445         llvm::errs()
1446             << "Failed to update the suffix delimiter of a raw string: "
1447             << llvm::toString(std::move(SuffixErr)) << "\n";
1448       }
1449     }
1450     SourceLocation OriginLoc =
1451         Current.Tok.getLocation().getLocWithOffset(OldPrefixSize);
1452     for (const tooling::Replacement &Fix : Fixes.first) {
1453       auto Err = Whitespaces.addReplacement(tooling::Replacement(
1454           SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()),
1455           Fix.getLength(), Fix.getReplacementText()));
1456       if (Err) {
1457         llvm::errs() << "Failed to reformat raw string: "
1458                      << llvm::toString(std::move(Err)) << "\n";
1459       }
1460     }
1461   }
1462   unsigned RawLastLineEndColumn = getLastLineEndColumn(
1463       *NewCode, FirstStartColumn, Style.TabWidth, Encoding);
1464   State.Column = RawLastLineEndColumn + NewSuffixSize;
1465   // Since we're updating the column to after the raw string literal here, we
1466   // have to manually add the penalty for the prefix R"delim( over the column
1467   // limit.
1468   unsigned PrefixExcessCharacters =
1469       StartColumn + NewPrefixSize > Style.ColumnLimit ?
1470       StartColumn + NewPrefixSize - Style.ColumnLimit : 0;
1471   return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter;
1472 }
1473 
1474 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current,
1475                                                  LineState &State) {
1476   // Break before further function parameters on all levels.
1477   for (unsigned i = 0, e = State.Stack.size(); i != e; ++i)
1478     State.Stack[i].BreakBeforeParameter = true;
1479 
1480   unsigned ColumnsUsed = State.Column;
1481   // We can only affect layout of the first and the last line, so the penalty
1482   // for all other lines is constant, and we ignore it.
1483   State.Column = Current.LastLineColumnWidth;
1484 
1485   if (ColumnsUsed > getColumnLimit(State))
1486     return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State));
1487   return 0;
1488 }
1489 
1490 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current,
1491                                                LineState &State, bool DryRun,
1492                                                bool AllowBreak) {
1493   unsigned Penalty = 0;
1494   // Compute the raw string style to use in case this is a raw string literal
1495   // that can be reformatted.
1496   auto RawStringStyle = getRawStringStyle(Current, State);
1497   if (RawStringStyle && !Current.Finalized) {
1498     Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun);
1499   } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) {
1500     // Don't break multi-line tokens other than block comments and raw string
1501     // literals. Instead, just update the state.
1502     Penalty = addMultilineToken(Current, State);
1503   } else if (State.Line->Type != LT_ImportStatement) {
1504     // We generally don't break import statements.
1505     LineState OriginalState = State;
1506 
1507     // Whether we force the reflowing algorithm to stay strictly within the
1508     // column limit.
1509     bool Strict = false;
1510     // Whether the first non-strict attempt at reflowing did intentionally
1511     // exceed the column limit.
1512     bool Exceeded = false;
1513     std::tie(Penalty, Exceeded) = breakProtrudingToken(
1514         Current, State, AllowBreak, /*DryRun=*/true, Strict);
1515     if (Exceeded) {
1516       // If non-strict reflowing exceeds the column limit, try whether strict
1517       // reflowing leads to an overall lower penalty.
1518       LineState StrictState = OriginalState;
1519       unsigned StrictPenalty =
1520           breakProtrudingToken(Current, StrictState, AllowBreak,
1521                                /*DryRun=*/true, /*Strict=*/true)
1522               .first;
1523       Strict = StrictPenalty <= Penalty;
1524       if (Strict) {
1525         Penalty = StrictPenalty;
1526         State = StrictState;
1527       }
1528     }
1529     if (!DryRun) {
1530       // If we're not in dry-run mode, apply the changes with the decision on
1531       // strictness made above.
1532       breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false,
1533                            Strict);
1534     }
1535   }
1536   if (State.Column > getColumnLimit(State)) {
1537     unsigned ExcessCharacters = State.Column - getColumnLimit(State);
1538     Penalty += Style.PenaltyExcessCharacter * ExcessCharacters;
1539   }
1540   return Penalty;
1541 }
1542 
1543 // Returns the enclosing function name of a token, or the empty string if not
1544 // found.
1545 static StringRef getEnclosingFunctionName(const FormatToken &Current) {
1546   // Look for: 'function(' or 'function<templates>(' before Current.
1547   auto Tok = Current.getPreviousNonComment();
1548   if (!Tok || !Tok->is(tok::l_paren))
1549     return "";
1550   Tok = Tok->getPreviousNonComment();
1551   if (!Tok)
1552     return "";
1553   if (Tok->is(TT_TemplateCloser)) {
1554     Tok = Tok->MatchingParen;
1555     if (Tok)
1556       Tok = Tok->getPreviousNonComment();
1557   }
1558   if (!Tok || !Tok->is(tok::identifier))
1559     return "";
1560   return Tok->TokenText;
1561 }
1562 
1563 llvm::Optional<FormatStyle>
1564 ContinuationIndenter::getRawStringStyle(const FormatToken &Current,
1565                                         const LineState &State) {
1566   if (!Current.isStringLiteral())
1567     return None;
1568   auto Delimiter = getRawStringDelimiter(Current.TokenText);
1569   if (!Delimiter)
1570     return None;
1571   auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter);
1572   if (!RawStringStyle)
1573     RawStringStyle = RawStringFormats.getEnclosingFunctionStyle(
1574         getEnclosingFunctionName(Current));
1575   if (!RawStringStyle)
1576     return None;
1577   RawStringStyle->ColumnLimit = getColumnLimit(State);
1578   return RawStringStyle;
1579 }
1580 
1581 std::unique_ptr<BreakableToken> ContinuationIndenter::createBreakableToken(
1582     const FormatToken &Current, LineState &State, bool AllowBreak) {
1583   unsigned StartColumn = State.Column - Current.ColumnWidth;
1584   if (Current.isStringLiteral()) {
1585     // FIXME: String literal breaking is currently disabled for Java and JS, as
1586     // it requires strings to be merged using "+" which we don't support.
1587     if (Style.Language == FormatStyle::LK_Java ||
1588         Style.Language == FormatStyle::LK_JavaScript ||
1589         !Style.BreakStringLiterals ||
1590         !AllowBreak)
1591       return nullptr;
1592 
1593     // Don't break string literals inside preprocessor directives (except for
1594     // #define directives, as their contents are stored in separate lines and
1595     // are not affected by this check).
1596     // This way we avoid breaking code with line directives and unknown
1597     // preprocessor directives that contain long string literals.
1598     if (State.Line->Type == LT_PreprocessorDirective)
1599       return nullptr;
1600     // Exempts unterminated string literals from line breaking. The user will
1601     // likely want to terminate the string before any line breaking is done.
1602     if (Current.IsUnterminatedLiteral)
1603       return nullptr;
1604     // Don't break string literals inside Objective-C array literals (doing so
1605     // raises the warning -Wobjc-string-concatenation).
1606     if (State.Stack.back().IsInsideObjCArrayLiteral) {
1607       return nullptr;
1608     }
1609 
1610     StringRef Text = Current.TokenText;
1611     StringRef Prefix;
1612     StringRef Postfix;
1613     // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'.
1614     // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to
1615     // reduce the overhead) for each FormatToken, which is a string, so that we
1616     // don't run multiple checks here on the hot path.
1617     if ((Text.endswith(Postfix = "\"") &&
1618          (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") ||
1619           Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") ||
1620           Text.startswith(Prefix = "u8\"") ||
1621           Text.startswith(Prefix = "L\""))) ||
1622         (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) {
1623       // We need this to address the case where there is an unbreakable tail
1624       // only if certain other formatting decisions have been taken. The
1625       // UnbreakableTailLength of Current is an overapproximation is that case
1626       // and we need to be correct here.
1627       unsigned UnbreakableTailLength = (State.NextToken && canBreak(State))
1628                                            ? 0
1629                                            : Current.UnbreakableTailLength;
1630       return llvm::make_unique<BreakableStringLiteral>(
1631           Current, StartColumn, Prefix, Postfix, UnbreakableTailLength,
1632           State.Line->InPPDirective, Encoding, Style);
1633     }
1634   } else if (Current.is(TT_BlockComment)) {
1635     if (!Style.ReflowComments ||
1636         // If a comment token switches formatting, like
1637         // /* clang-format on */, we don't want to break it further,
1638         // but we may still want to adjust its indentation.
1639         switchesFormatting(Current)) {
1640       return nullptr;
1641     }
1642     return llvm::make_unique<BreakableBlockComment>(
1643         Current, StartColumn, Current.OriginalColumn, !Current.Previous,
1644         State.Line->InPPDirective, Encoding, Style);
1645   } else if (Current.is(TT_LineComment) &&
1646              (Current.Previous == nullptr ||
1647               Current.Previous->isNot(TT_ImplicitStringLiteral))) {
1648     if (!Style.ReflowComments ||
1649         CommentPragmasRegex.match(Current.TokenText.substr(2)) ||
1650         switchesFormatting(Current))
1651       return nullptr;
1652     return llvm::make_unique<BreakableLineCommentSection>(
1653         Current, StartColumn, Current.OriginalColumn, !Current.Previous,
1654         /*InPPDirective=*/false, Encoding, Style);
1655   }
1656   return nullptr;
1657 }
1658 
1659 std::pair<unsigned, bool>
1660 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current,
1661                                            LineState &State, bool AllowBreak,
1662                                            bool DryRun, bool Strict) {
1663   std::unique_ptr<const BreakableToken> Token =
1664       createBreakableToken(Current, State, AllowBreak);
1665   if (!Token)
1666     return {0, false};
1667   assert(Token->getLineCount() > 0);
1668   unsigned ColumnLimit = getColumnLimit(State);
1669   if (Current.is(TT_LineComment)) {
1670     // We don't insert backslashes when breaking line comments.
1671     ColumnLimit = Style.ColumnLimit;
1672   }
1673   if (Current.UnbreakableTailLength >= ColumnLimit)
1674     return {0, false};
1675   // ColumnWidth was already accounted into State.Column before calling
1676   // breakProtrudingToken.
1677   unsigned StartColumn = State.Column - Current.ColumnWidth;
1678   unsigned NewBreakPenalty = Current.isStringLiteral()
1679                                  ? Style.PenaltyBreakString
1680                                  : Style.PenaltyBreakComment;
1681   // Stores whether we intentionally decide to let a line exceed the column
1682   // limit.
1683   bool Exceeded = false;
1684   // Stores whether we introduce a break anywhere in the token.
1685   bool BreakInserted = Token->introducesBreakBeforeToken();
1686   // Store whether we inserted a new line break at the end of the previous
1687   // logical line.
1688   bool NewBreakBefore = false;
1689   // We use a conservative reflowing strategy. Reflow starts after a line is
1690   // broken or the corresponding whitespace compressed. Reflow ends as soon as a
1691   // line that doesn't get reflown with the previous line is reached.
1692   bool Reflow = false;
1693   // Keep track of where we are in the token:
1694   // Where we are in the content of the current logical line.
1695   unsigned TailOffset = 0;
1696   // The column number we're currently at.
1697   unsigned ContentStartColumn =
1698       Token->getContentStartColumn(0, /*Break=*/false);
1699   // The number of columns left in the current logical line after TailOffset.
1700   unsigned RemainingTokenColumns =
1701       Token->getRemainingLength(0, TailOffset, ContentStartColumn);
1702   // Adapt the start of the token, for example indent.
1703   if (!DryRun)
1704     Token->adaptStartOfLine(0, Whitespaces);
1705 
1706   unsigned Penalty = 0;
1707   DEBUG(llvm::dbgs() << "Breaking protruding token at column " << StartColumn
1708                      << ".\n");
1709   for (unsigned LineIndex = 0, EndIndex = Token->getLineCount();
1710        LineIndex != EndIndex; ++LineIndex) {
1711     DEBUG(llvm::dbgs() << "  Line: " << LineIndex << " (Reflow: " << Reflow
1712                        << ")\n");
1713     NewBreakBefore = false;
1714     // If we did reflow the previous line, we'll try reflowing again. Otherwise
1715     // we'll start reflowing if the current line is broken or whitespace is
1716     // compressed.
1717     bool TryReflow = Reflow;
1718     // Break the current token until we can fit the rest of the line.
1719     while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
1720       DEBUG(llvm::dbgs() << "    Over limit, need: "
1721                          << (ContentStartColumn + RemainingTokenColumns)
1722                          << ", space: " << ColumnLimit
1723                          << ", reflown prefix: " << ContentStartColumn
1724                          << ", offset in line: " << TailOffset << "\n");
1725       // If the current token doesn't fit, find the latest possible split in the
1726       // current line so that breaking at it will be under the column limit.
1727       // FIXME: Use the earliest possible split while reflowing to correctly
1728       // compress whitespace within a line.
1729       BreakableToken::Split Split =
1730           Token->getSplit(LineIndex, TailOffset, ColumnLimit,
1731                           ContentStartColumn, CommentPragmasRegex);
1732       if (Split.first == StringRef::npos) {
1733         // No break opportunity - update the penalty and continue with the next
1734         // logical line.
1735         if (LineIndex < EndIndex - 1)
1736           // The last line's penalty is handled in addNextStateToQueue().
1737           Penalty += Style.PenaltyExcessCharacter *
1738                      (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
1739         DEBUG(llvm::dbgs() << "    No break opportunity.\n");
1740         break;
1741       }
1742       assert(Split.first != 0);
1743 
1744       if (Token->supportsReflow()) {
1745         // Check whether the next natural split point after the current one can
1746         // still fit the line, either because we can compress away whitespace,
1747         // or because the penalty the excess characters introduce is lower than
1748         // the break penalty.
1749         // We only do this for tokens that support reflowing, and thus allow us
1750         // to change the whitespace arbitrarily (e.g. comments).
1751         // Other tokens, like string literals, can be broken on arbitrary
1752         // positions.
1753 
1754         // First, compute the columns from TailOffset to the next possible split
1755         // position.
1756         // For example:
1757         // ColumnLimit:     |
1758         // // Some text   that    breaks
1759         //    ^ tail offset
1760         //             ^-- split
1761         //    ^-------- to split columns
1762         //                    ^--- next split
1763         //    ^--------------- to next split columns
1764         unsigned ToSplitColumns = Token->getRangeLength(
1765             LineIndex, TailOffset, Split.first, ContentStartColumn);
1766         DEBUG(llvm::dbgs() << "    ToSplit: " << ToSplitColumns << "\n");
1767 
1768         BreakableToken::Split NextSplit = Token->getSplit(
1769             LineIndex, TailOffset + Split.first + Split.second, ColumnLimit,
1770             ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex);
1771         // Compute the columns necessary to fit the next non-breakable sequence
1772         // into the current line.
1773         unsigned ToNextSplitColumns = 0;
1774         if (NextSplit.first == StringRef::npos) {
1775           ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset,
1776                                                          ContentStartColumn);
1777         } else {
1778           ToNextSplitColumns = Token->getRangeLength(
1779               LineIndex, TailOffset,
1780               Split.first + Split.second + NextSplit.first, ContentStartColumn);
1781         }
1782         // Compress the whitespace between the break and the start of the next
1783         // unbreakable sequence.
1784         ToNextSplitColumns =
1785             Token->getLengthAfterCompression(ToNextSplitColumns, Split);
1786         DEBUG(llvm::dbgs() << "    ContentStartColumn: " << ContentStartColumn
1787                            << "\n");
1788         DEBUG(llvm::dbgs() << "    ToNextSplit: " << ToNextSplitColumns << "\n");
1789         // If the whitespace compression makes us fit, continue on the current
1790         // line.
1791         bool ContinueOnLine =
1792             ContentStartColumn + ToNextSplitColumns <= ColumnLimit;
1793         unsigned ExcessCharactersPenalty = 0;
1794         if (!ContinueOnLine && !Strict) {
1795           // Similarly, if the excess characters' penalty is lower than the
1796           // penalty of introducing a new break, continue on the current line.
1797           ExcessCharactersPenalty =
1798               (ContentStartColumn + ToNextSplitColumns - ColumnLimit) *
1799               Style.PenaltyExcessCharacter;
1800           DEBUG(llvm::dbgs()
1801                 << "    Penalty excess: " << ExcessCharactersPenalty
1802                 << "\n            break : " << NewBreakPenalty << "\n");
1803           if (ExcessCharactersPenalty < NewBreakPenalty) {
1804             Exceeded = true;
1805             ContinueOnLine = true;
1806           }
1807         }
1808         if (ContinueOnLine) {
1809           DEBUG(llvm::dbgs() << "    Continuing on line...\n");
1810           // The current line fits after compressing the whitespace - reflow
1811           // the next line into it if possible.
1812           TryReflow = true;
1813           if (!DryRun)
1814             Token->compressWhitespace(LineIndex, TailOffset, Split,
1815                                       Whitespaces);
1816           // When we continue on the same line, leave one space between content.
1817           ContentStartColumn += ToSplitColumns + 1;
1818           Penalty += ExcessCharactersPenalty;
1819           TailOffset += Split.first + Split.second;
1820           RemainingTokenColumns = Token->getRemainingLength(
1821               LineIndex, TailOffset, ContentStartColumn);
1822           continue;
1823         }
1824       }
1825       DEBUG(llvm::dbgs() << "    Breaking...\n");
1826       ContentStartColumn =
1827           Token->getContentStartColumn(LineIndex, /*Break=*/true);
1828       unsigned NewRemainingTokenColumns = Token->getRemainingLength(
1829           LineIndex, TailOffset + Split.first + Split.second,
1830           ContentStartColumn);
1831 
1832       // When breaking before a tab character, it may be moved by a few columns,
1833       // but will still be expanded to the next tab stop, so we don't save any
1834       // columns.
1835       if (NewRemainingTokenColumns == RemainingTokenColumns) {
1836         // FIXME: Do we need to adjust the penalty?
1837         break;
1838       }
1839       assert(NewRemainingTokenColumns < RemainingTokenColumns);
1840 
1841       DEBUG(llvm::dbgs() << "    Breaking at: " << TailOffset + Split.first
1842                          << ", " << Split.second << "\n");
1843       if (!DryRun)
1844         Token->insertBreak(LineIndex, TailOffset, Split, Whitespaces);
1845 
1846       Penalty += NewBreakPenalty;
1847       TailOffset += Split.first + Split.second;
1848       RemainingTokenColumns = NewRemainingTokenColumns;
1849       BreakInserted = true;
1850       NewBreakBefore = true;
1851     }
1852     // In case there's another line, prepare the state for the start of the next
1853     // line.
1854     if (LineIndex + 1 != EndIndex) {
1855       unsigned NextLineIndex = LineIndex + 1;
1856       if (NewBreakBefore)
1857         // After breaking a line, try to reflow the next line into the current
1858         // one once RemainingTokenColumns fits.
1859         TryReflow = true;
1860       if (TryReflow) {
1861         // We decided that we want to try reflowing the next line into the
1862         // current one.
1863         // We will now adjust the state as if the reflow is successful (in
1864         // preparation for the next line), and see whether that works. If we
1865         // decide that we cannot reflow, we will later reset the state to the
1866         // start of the next line.
1867         Reflow = false;
1868         // As we did not continue breaking the line, RemainingTokenColumns is
1869         // known to fit after ContentStartColumn. Adapt ContentStartColumn to
1870         // the position at which we want to format the next line if we do
1871         // actually reflow.
1872         // When we reflow, we need to add a space between the end of the current
1873         // line and the next line's start column.
1874         ContentStartColumn += RemainingTokenColumns + 1;
1875         // Get the split that we need to reflow next logical line into the end
1876         // of the current one; the split will include any leading whitespace of
1877         // the next logical line.
1878         BreakableToken::Split SplitBeforeNext =
1879             Token->getReflowSplit(NextLineIndex, CommentPragmasRegex);
1880         DEBUG(llvm::dbgs() << "    Size of reflown text: " << ContentStartColumn
1881                            << "\n    Potential reflow split: ");
1882         if (SplitBeforeNext.first != StringRef::npos) {
1883           DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", "
1884                              << SplitBeforeNext.second << "\n");
1885           TailOffset = SplitBeforeNext.first + SplitBeforeNext.second;
1886           // If the rest of the next line fits into the current line below the
1887           // column limit, we can safely reflow.
1888           RemainingTokenColumns = Token->getRemainingLength(
1889               NextLineIndex, TailOffset, ContentStartColumn);
1890           Reflow = true;
1891           if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
1892             DEBUG(llvm::dbgs() << "    Over limit after reflow, need: "
1893                                << (ContentStartColumn + RemainingTokenColumns)
1894                                << ", space: " << ColumnLimit
1895                                << ", reflown prefix: " << ContentStartColumn
1896                                << ", offset in line: " << TailOffset << "\n");
1897             // If the whole next line does not fit, try to find a point in
1898             // the next line at which we can break so that attaching the part
1899             // of the next line to that break point onto the current line is
1900             // below the column limit.
1901             BreakableToken::Split Split =
1902                 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit,
1903                                 ContentStartColumn, CommentPragmasRegex);
1904             if (Split.first == StringRef::npos) {
1905               DEBUG(llvm::dbgs() << "    Did not find later break\n");
1906               Reflow = false;
1907             } else {
1908               // Check whether the first split point gets us below the column
1909               // limit. Note that we will execute this split below as part of
1910               // the normal token breaking and reflow logic within the line.
1911               unsigned ToSplitColumns = Token->getRangeLength(
1912                   NextLineIndex, TailOffset, Split.first, ContentStartColumn);
1913               if (ContentStartColumn + ToSplitColumns > ColumnLimit) {
1914                 DEBUG(llvm::dbgs() << "    Next split protrudes, need: "
1915                                    << (ContentStartColumn + ToSplitColumns)
1916                                    << ", space: " << ColumnLimit);
1917                 unsigned ExcessCharactersPenalty =
1918                     (ContentStartColumn + ToSplitColumns - ColumnLimit) *
1919                     Style.PenaltyExcessCharacter;
1920                 if (NewBreakPenalty < ExcessCharactersPenalty) {
1921                   Reflow = false;
1922                 }
1923               }
1924             }
1925           }
1926         } else {
1927           DEBUG(llvm::dbgs() << "not found.\n");
1928         }
1929       }
1930       if (!Reflow) {
1931         // If we didn't reflow into the next line, the only space to consider is
1932         // the next logical line. Reset our state to match the start of the next
1933         // line.
1934         TailOffset = 0;
1935         ContentStartColumn =
1936             Token->getContentStartColumn(NextLineIndex, /*Break=*/false);
1937         RemainingTokenColumns = Token->getRemainingLength(
1938             NextLineIndex, TailOffset, ContentStartColumn);
1939         // Adapt the start of the token, for example indent.
1940         if (!DryRun)
1941           Token->adaptStartOfLine(NextLineIndex, Whitespaces);
1942       } else {
1943         // If we found a reflow split and have added a new break before the next
1944         // line, we are going to remove the line break at the start of the next
1945         // logical line. For example, here we'll add a new line break after
1946         // 'text', and subsequently delete the line break between 'that' and
1947         // 'reflows'.
1948         //   // some text that
1949         //   // reflows
1950         // ->
1951         //   // some text
1952         //   // that reflows
1953         // When adding the line break, we also added the penalty for it, so we
1954         // need to subtract that penalty again when we remove the line break due
1955         // to reflowing.
1956         if (NewBreakBefore) {
1957           assert(Penalty >= NewBreakPenalty);
1958           Penalty -= NewBreakPenalty;
1959         }
1960         if (!DryRun)
1961           Token->reflow(NextLineIndex, Whitespaces);
1962       }
1963     }
1964   }
1965 
1966   BreakableToken::Split SplitAfterLastLine =
1967       Token->getSplitAfterLastLine(TailOffset);
1968   if (SplitAfterLastLine.first != StringRef::npos) {
1969     DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n");
1970     if (!DryRun)
1971       Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine,
1972                                             Whitespaces);
1973     ContentStartColumn =
1974         Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true);
1975     RemainingTokenColumns = Token->getRemainingLength(
1976         Token->getLineCount() - 1,
1977         TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second,
1978         ContentStartColumn);
1979   }
1980 
1981   State.Column = ContentStartColumn + RemainingTokenColumns -
1982                  Current.UnbreakableTailLength;
1983 
1984   if (BreakInserted) {
1985     // If we break the token inside a parameter list, we need to break before
1986     // the next parameter on all levels, so that the next parameter is clearly
1987     // visible. Line comments already introduce a break.
1988     if (Current.isNot(TT_LineComment)) {
1989       for (unsigned i = 0, e = State.Stack.size(); i != e; ++i)
1990         State.Stack[i].BreakBeforeParameter = true;
1991     }
1992 
1993     if (Current.is(TT_BlockComment))
1994       State.NoContinuation = true;
1995 
1996     State.Stack.back().LastSpace = StartColumn;
1997   }
1998 
1999   Token->updateNextToken(State);
2000 
2001   return {Penalty, Exceeded};
2002 }
2003 
2004 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const {
2005   // In preprocessor directives reserve two chars for trailing " \"
2006   return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0);
2007 }
2008 
2009 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) {
2010   const FormatToken &Current = *State.NextToken;
2011   if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral))
2012     return false;
2013   // We never consider raw string literals "multiline" for the purpose of
2014   // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased
2015   // (see TokenAnnotator::mustBreakBefore().
2016   if (Current.TokenText.startswith("R\""))
2017     return false;
2018   if (Current.IsMultiline)
2019     return true;
2020   if (Current.getNextNonComment() &&
2021       Current.getNextNonComment()->isStringLiteral())
2022     return true; // Implicit concatenation.
2023   if (Style.ColumnLimit != 0 && Style.BreakStringLiterals &&
2024       State.Column + Current.ColumnWidth + Current.UnbreakableTailLength >
2025           Style.ColumnLimit)
2026     return true; // String will be split.
2027   return false;
2028 }
2029 
2030 } // namespace format
2031 } // namespace clang
2032