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 /// 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 true if a TT_SelectorName should be indented when wrapped,
30 // false otherwise.
31 static bool shouldIndentWrappedSelectorName(const FormatStyle &Style,
32                                             LineType LineType) {
33   return Style.IndentWrappedFunctionNames || LineType == LT_ObjCMethodDecl;
34 }
35 
36 // Returns the length of everything up to the first possible line break after
37 // the ), ], } or > matching \c Tok.
38 static unsigned getLengthToMatchingParen(const FormatToken &Tok,
39                                          const std::vector<ParenState> &Stack) {
40   // Normally whether or not a break before T is possible is calculated and
41   // stored in T.CanBreakBefore. Braces, array initializers and text proto
42   // messages like `key: < ... >` are an exception: a break is possible
43   // before a closing brace R if a break was inserted after the corresponding
44   // opening brace. The information about whether or not a break is needed
45   // before a closing brace R is stored in the ParenState field
46   // S.BreakBeforeClosingBrace where S is the state that R closes.
47   //
48   // In order to decide whether there can be a break before encountered right
49   // braces, this implementation iterates over the sequence of tokens and over
50   // the paren stack in lockstep, keeping track of the stack level which visited
51   // right braces correspond to in MatchingStackIndex.
52   //
53   // For example, consider:
54   // L. <- line number
55   // 1. {
56   // 2. {1},
57   // 3. {2},
58   // 4. {{3}}}
59   //     ^ where we call this method with this token.
60   // The paren stack at this point contains 3 brace levels:
61   //  0. { at line 1, BreakBeforeClosingBrace: true
62   //  1. first { at line 4, BreakBeforeClosingBrace: false
63   //  2. second { at line 4, BreakBeforeClosingBrace: false,
64   //  where there might be fake parens levels in-between these levels.
65   // The algorithm will start at the first } on line 4, which is the matching
66   // brace of the initial left brace and at level 2 of the stack. Then,
67   // examining BreakBeforeClosingBrace: false at level 2, it will continue to
68   // the second } on line 4, and will traverse the stack downwards until it
69   // finds the matching { on level 1. Then, examining BreakBeforeClosingBrace:
70   // false at level 1, it will continue to the third } on line 4 and will
71   // traverse the stack downwards until it finds the matching { on level 0.
72   // Then, examining BreakBeforeClosingBrace: true at level 0, the algorithm
73   // will stop and will use the second } on line 4 to determine the length to
74   // return, as in this example the range will include the tokens: {3}}
75   //
76   // The algorithm will only traverse the stack if it encounters braces, array
77   // initializer squares or text proto angle brackets.
78   if (!Tok.MatchingParen)
79     return 0;
80   FormatToken *End = Tok.MatchingParen;
81   // Maintains a stack level corresponding to the current End token.
82   int MatchingStackIndex = Stack.size() - 1;
83   // Traverses the stack downwards, looking for the level to which LBrace
84   // corresponds. Returns either a pointer to the matching level or nullptr if
85   // LParen is not found in the initial portion of the stack up to
86   // MatchingStackIndex.
87   auto FindParenState = [&](const FormatToken *LBrace) -> const ParenState * {
88     while (MatchingStackIndex >= 0 && Stack[MatchingStackIndex].Tok != LBrace)
89       --MatchingStackIndex;
90     return MatchingStackIndex >= 0 ? &Stack[MatchingStackIndex] : nullptr;
91   };
92   for (; End->Next; End = End->Next) {
93     if (End->Next->CanBreakBefore)
94       break;
95     if (!End->Next->closesScope())
96       continue;
97     if (End->Next->MatchingParen &&
98         End->Next->MatchingParen->isOneOf(
99             tok::l_brace, TT_ArrayInitializerLSquare, tok::less)) {
100       const ParenState *State = FindParenState(End->Next->MatchingParen);
101       if (State && State->BreakBeforeClosingBrace)
102         break;
103     }
104   }
105   return End->TotalLength - Tok.TotalLength + 1;
106 }
107 
108 static unsigned getLengthToNextOperator(const FormatToken &Tok) {
109   if (!Tok.NextOperator)
110     return 0;
111   return Tok.NextOperator->TotalLength - Tok.TotalLength;
112 }
113 
114 // Returns \c true if \c Tok is the "." or "->" of a call and starts the next
115 // segment of a builder type call.
116 static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) {
117   return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope();
118 }
119 
120 // Returns \c true if \c Current starts a new parameter.
121 static bool startsNextParameter(const FormatToken &Current,
122                                 const FormatStyle &Style) {
123   const FormatToken &Previous = *Current.Previous;
124   if (Current.is(TT_CtorInitializerComma) &&
125       Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma)
126     return true;
127   if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName))
128     return true;
129   return Previous.is(tok::comma) && !Current.isTrailingComment() &&
130          ((Previous.isNot(TT_CtorInitializerComma) ||
131            Style.BreakConstructorInitializers !=
132                FormatStyle::BCIS_BeforeComma) &&
133           (Previous.isNot(TT_InheritanceComma) ||
134            Style.BreakInheritanceList != FormatStyle::BILS_BeforeComma));
135 }
136 
137 static bool opensProtoMessageField(const FormatToken &LessTok,
138                                    const FormatStyle &Style) {
139   if (LessTok.isNot(tok::less))
140     return false;
141   return Style.Language == FormatStyle::LK_TextProto ||
142          (Style.Language == FormatStyle::LK_Proto &&
143           (LessTok.NestingLevel > 0 ||
144            (LessTok.Previous && LessTok.Previous->is(tok::equal))));
145 }
146 
147 // Returns the delimiter of a raw string literal, or None if TokenText is not
148 // the text of a raw string literal. The delimiter could be the empty string.
149 // For example, the delimiter of R"deli(cont)deli" is deli.
150 static llvm::Optional<StringRef> getRawStringDelimiter(StringRef TokenText) {
151   if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'.
152       || !TokenText.startswith("R\"") || !TokenText.endswith("\""))
153     return None;
154 
155   // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has
156   // size at most 16 by the standard, so the first '(' must be among the first
157   // 19 bytes.
158   size_t LParenPos = TokenText.substr(0, 19).find_first_of('(');
159   if (LParenPos == StringRef::npos)
160     return None;
161   StringRef Delimiter = TokenText.substr(2, LParenPos - 2);
162 
163   // Check that the string ends in ')Delimiter"'.
164   size_t RParenPos = TokenText.size() - Delimiter.size() - 2;
165   if (TokenText[RParenPos] != ')')
166     return None;
167   if (!TokenText.substr(RParenPos + 1).startswith(Delimiter))
168     return None;
169   return Delimiter;
170 }
171 
172 // Returns the canonical delimiter for \p Language, or the empty string if no
173 // canonical delimiter is specified.
174 static StringRef
175 getCanonicalRawStringDelimiter(const FormatStyle &Style,
176                                FormatStyle::LanguageKind Language) {
177   for (const auto &Format : Style.RawStringFormats) {
178     if (Format.Language == Language)
179       return StringRef(Format.CanonicalDelimiter);
180   }
181   return "";
182 }
183 
184 RawStringFormatStyleManager::RawStringFormatStyleManager(
185     const FormatStyle &CodeStyle) {
186   for (const auto &RawStringFormat : CodeStyle.RawStringFormats) {
187     llvm::Optional<FormatStyle> LanguageStyle =
188         CodeStyle.GetLanguageStyle(RawStringFormat.Language);
189     if (!LanguageStyle) {
190       FormatStyle PredefinedStyle;
191       if (!getPredefinedStyle(RawStringFormat.BasedOnStyle,
192                               RawStringFormat.Language, &PredefinedStyle)) {
193         PredefinedStyle = getLLVMStyle();
194         PredefinedStyle.Language = RawStringFormat.Language;
195       }
196       LanguageStyle = PredefinedStyle;
197     }
198     LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit;
199     for (StringRef Delimiter : RawStringFormat.Delimiters) {
200       DelimiterStyle.insert({Delimiter, *LanguageStyle});
201     }
202     for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions) {
203       EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle});
204     }
205   }
206 }
207 
208 llvm::Optional<FormatStyle>
209 RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const {
210   auto It = DelimiterStyle.find(Delimiter);
211   if (It == DelimiterStyle.end())
212     return None;
213   return It->second;
214 }
215 
216 llvm::Optional<FormatStyle>
217 RawStringFormatStyleManager::getEnclosingFunctionStyle(
218     StringRef EnclosingFunction) const {
219   auto It = EnclosingFunctionStyle.find(EnclosingFunction);
220   if (It == EnclosingFunctionStyle.end())
221     return None;
222   return It->second;
223 }
224 
225 ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style,
226                                            const AdditionalKeywords &Keywords,
227                                            const SourceManager &SourceMgr,
228                                            WhitespaceManager &Whitespaces,
229                                            encoding::Encoding Encoding,
230                                            bool BinPackInconclusiveFunctions)
231     : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr),
232       Whitespaces(Whitespaces), Encoding(Encoding),
233       BinPackInconclusiveFunctions(BinPackInconclusiveFunctions),
234       CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {}
235 
236 LineState ContinuationIndenter::getInitialState(unsigned FirstIndent,
237                                                 unsigned FirstStartColumn,
238                                                 const AnnotatedLine *Line,
239                                                 bool DryRun) {
240   LineState State;
241   State.FirstIndent = FirstIndent;
242   if (FirstStartColumn && Line->First->NewlinesBefore == 0)
243     State.Column = FirstStartColumn;
244   else
245     State.Column = FirstIndent;
246   // With preprocessor directive indentation, the line starts on column 0
247   // since it's indented after the hash, but FirstIndent is set to the
248   // preprocessor indent.
249   if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
250       (Line->Type == LT_PreprocessorDirective ||
251        Line->Type == LT_ImportStatement))
252     State.Column = 0;
253   State.Line = Line;
254   State.NextToken = Line->First;
255   State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent,
256                                    /*AvoidBinPacking=*/false,
257                                    /*NoLineBreak=*/false));
258   State.LineContainsContinuedForLoopSection = false;
259   State.NoContinuation = false;
260   State.StartOfStringLiteral = 0;
261   State.StartOfLineLevel = 0;
262   State.LowestLevelOnLine = 0;
263   State.IgnoreStackForComparison = false;
264 
265   if (Style.Language == FormatStyle::LK_TextProto) {
266     // We need this in order to deal with the bin packing of text fields at
267     // global scope.
268     State.Stack.back().AvoidBinPacking = true;
269     State.Stack.back().BreakBeforeParameter = true;
270     State.Stack.back().AlignColons = false;
271   }
272 
273   // The first token has already been indented and thus consumed.
274   moveStateToNextToken(State, DryRun, /*Newline=*/false);
275   return State;
276 }
277 
278 bool ContinuationIndenter::canBreak(const LineState &State) {
279   const FormatToken &Current = *State.NextToken;
280   const FormatToken &Previous = *Current.Previous;
281   assert(&Previous == Current.Previous);
282   if (!Current.CanBreakBefore && !(State.Stack.back().BreakBeforeClosingBrace &&
283                                    Current.closesBlockOrBlockTypeList(Style)))
284     return false;
285   // The opening "{" of a braced list has to be on the same line as the first
286   // element if it is nested in another braced init list or function call.
287   if (!Current.MustBreakBefore && Previous.is(tok::l_brace) &&
288       Previous.isNot(TT_DictLiteral) && Previous.BlockKind == BK_BracedInit &&
289       Previous.Previous &&
290       Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma))
291     return false;
292   // This prevents breaks like:
293   //   ...
294   //   SomeParameter, OtherParameter).DoSomething(
295   //   ...
296   // As they hide "DoSomething" and are generally bad for readability.
297   if (Previous.opensScope() && Previous.isNot(tok::l_brace) &&
298       State.LowestLevelOnLine < State.StartOfLineLevel &&
299       State.LowestLevelOnLine < Current.NestingLevel)
300     return false;
301   if (Current.isMemberAccess() && State.Stack.back().ContainsUnwrappedBuilder)
302     return false;
303 
304   // Don't create a 'hanging' indent if there are multiple blocks in a single
305   // statement.
306   if (Previous.is(tok::l_brace) && State.Stack.size() > 1 &&
307       State.Stack[State.Stack.size() - 2].NestedBlockInlined &&
308       State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks)
309     return false;
310 
311   // Don't break after very short return types (e.g. "void") as that is often
312   // unexpected.
313   if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) {
314     if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None)
315       return false;
316   }
317 
318   // If binary operators are moved to the next line (including commas for some
319   // styles of constructor initializers), that's always ok.
320   if (!Current.isOneOf(TT_BinaryOperator, tok::comma) &&
321       State.Stack.back().NoLineBreakInOperand)
322     return false;
323 
324   return !State.Stack.back().NoLineBreak;
325 }
326 
327 bool ContinuationIndenter::mustBreak(const LineState &State) {
328   const FormatToken &Current = *State.NextToken;
329   const FormatToken &Previous = *Current.Previous;
330   if (Current.MustBreakBefore || Current.is(TT_InlineASMColon))
331     return true;
332   if (State.Stack.back().BreakBeforeClosingBrace &&
333       Current.closesBlockOrBlockTypeList(Style))
334     return true;
335   if (Previous.is(tok::semi) && State.LineContainsContinuedForLoopSection)
336     return true;
337   if (Style.Language == FormatStyle::LK_ObjC &&
338       Current.ObjCSelectorNameParts > 1 &&
339       Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) {
340     return true;
341   }
342   if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) ||
343        (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) &&
344         Style.isCpp() &&
345         // FIXME: This is a temporary workaround for the case where clang-format
346         // sets BreakBeforeParameter to avoid bin packing and this creates a
347         // completely unnecessary line break after a template type that isn't
348         // line-wrapped.
349         (Previous.NestingLevel == 1 || Style.BinPackParameters)) ||
350        (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) &&
351         Previous.isNot(tok::question)) ||
352        (!Style.BreakBeforeTernaryOperators &&
353         Previous.is(TT_ConditionalExpr))) &&
354       State.Stack.back().BreakBeforeParameter && !Current.isTrailingComment() &&
355       !Current.isOneOf(tok::r_paren, tok::r_brace))
356     return true;
357   if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) ||
358        (Previous.is(TT_ArrayInitializerLSquare) &&
359         Previous.ParameterCount > 1) ||
360        opensProtoMessageField(Previous, Style)) &&
361       Style.ColumnLimit > 0 &&
362       getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 >
363           getColumnLimit(State))
364     return true;
365 
366   const FormatToken &BreakConstructorInitializersToken =
367       Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon
368           ? Previous
369           : Current;
370   if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) &&
371       (State.Column + State.Line->Last->TotalLength - Previous.TotalLength >
372            getColumnLimit(State) ||
373        State.Stack.back().BreakBeforeParameter) &&
374       (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All ||
375        Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon ||
376        Style.ColumnLimit != 0))
377     return true;
378 
379   if (Current.is(TT_ObjCMethodExpr) && !Previous.is(TT_SelectorName) &&
380       State.Line->startsWith(TT_ObjCMethodSpecifier))
381     return true;
382   if (Current.is(TT_SelectorName) && !Previous.is(tok::at) &&
383       State.Stack.back().ObjCSelectorNameFound &&
384       State.Stack.back().BreakBeforeParameter)
385     return true;
386 
387   unsigned NewLineColumn = getNewLineColumn(State);
388   if (Current.isMemberAccess() && Style.ColumnLimit != 0 &&
389       State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit &&
390       (State.Column > NewLineColumn ||
391        Current.NestingLevel < State.StartOfLineLevel))
392     return true;
393 
394   if (startsSegmentOfBuilderTypeCall(Current) &&
395       (State.Stack.back().CallContinuation != 0 ||
396        State.Stack.back().BreakBeforeParameter) &&
397       // JavaScript is treated different here as there is a frequent pattern:
398       //   SomeFunction(function() {
399       //     ...
400       //   }.bind(...));
401       // FIXME: We should find a more generic solution to this problem.
402       !(State.Column <= NewLineColumn &&
403         Style.Language == FormatStyle::LK_JavaScript))
404     return true;
405 
406   // If the template declaration spans multiple lines, force wrap before the
407   // function/class declaration
408   if (Previous.ClosesTemplateDeclaration &&
409       State.Stack.back().BreakBeforeParameter && Current.CanBreakBefore)
410     return true;
411 
412   if (State.Column <= NewLineColumn)
413     return false;
414 
415   if (Style.AlwaysBreakBeforeMultilineStrings &&
416       (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth ||
417        Previous.is(tok::comma) || Current.NestingLevel < 2) &&
418       !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at) &&
419       !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) &&
420       nextIsMultilineString(State))
421     return true;
422 
423   // Using CanBreakBefore here and below takes care of the decision whether the
424   // current style uses wrapping before or after operators for the given
425   // operator.
426   if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) {
427     // If we need to break somewhere inside the LHS of a binary expression, we
428     // should also break after the operator. Otherwise, the formatting would
429     // hide the operator precedence, e.g. in:
430     //   if (aaaaaaaaaaaaaa ==
431     //           bbbbbbbbbbbbbb && c) {..
432     // For comparisons, we only apply this rule, if the LHS is a binary
433     // expression itself as otherwise, the line breaks seem superfluous.
434     // We need special cases for ">>" which we have split into two ">" while
435     // lexing in order to make template parsing easier.
436     bool IsComparison = (Previous.getPrecedence() == prec::Relational ||
437                          Previous.getPrecedence() == prec::Equality ||
438                          Previous.getPrecedence() == prec::Spaceship) &&
439                         Previous.Previous &&
440                         Previous.Previous->isNot(TT_BinaryOperator); // For >>.
441     bool LHSIsBinaryExpr =
442         Previous.Previous && Previous.Previous->EndsBinaryExpression;
443     if ((!IsComparison || LHSIsBinaryExpr) && !Current.isTrailingComment() &&
444         Previous.getPrecedence() != prec::Assignment &&
445         State.Stack.back().BreakBeforeParameter)
446       return true;
447   } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore &&
448              State.Stack.back().BreakBeforeParameter) {
449     return true;
450   }
451 
452   // Same as above, but for the first "<<" operator.
453   if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) &&
454       State.Stack.back().BreakBeforeParameter &&
455       State.Stack.back().FirstLessLess == 0)
456     return true;
457 
458   if (Current.NestingLevel == 0 && !Current.isTrailingComment()) {
459     // Always break after "template <...>" and leading annotations. This is only
460     // for cases where the entire line does not fit on a single line as a
461     // different LineFormatter would be used otherwise.
462     if (Previous.ClosesTemplateDeclaration)
463       return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No;
464     if (Previous.is(TT_FunctionAnnotationRParen))
465       return true;
466     if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) &&
467         Current.isNot(TT_LeadingJavaAnnotation))
468       return true;
469   }
470 
471   // If the return type spans multiple lines, wrap before the function name.
472   if ((Current.is(TT_FunctionDeclarationName) ||
473        (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) &&
474       !Previous.is(tok::kw_template) && State.Stack.back().BreakBeforeParameter)
475     return true;
476 
477   // The following could be precomputed as they do not depend on the state.
478   // However, as they should take effect only if the UnwrappedLine does not fit
479   // into the ColumnLimit, they are checked here in the ContinuationIndenter.
480   if (Style.ColumnLimit != 0 && Previous.BlockKind == BK_Block &&
481       Previous.is(tok::l_brace) && !Current.isOneOf(tok::r_brace, tok::comment))
482     return true;
483 
484   if (Current.is(tok::lessless) &&
485       ((Previous.is(tok::identifier) && Previous.TokenText == "endl") ||
486        (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") ||
487                                      Previous.TokenText == "\'\\n\'"))))
488     return true;
489 
490   if (Previous.is(TT_BlockComment) && Previous.IsMultiline)
491     return true;
492 
493   if (State.NoContinuation)
494     return true;
495 
496   return false;
497 }
498 
499 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline,
500                                                bool DryRun,
501                                                unsigned ExtraSpaces) {
502   const FormatToken &Current = *State.NextToken;
503 
504   assert(!State.Stack.empty());
505   State.NoContinuation = false;
506 
507   if ((Current.is(TT_ImplicitStringLiteral) &&
508        (Current.Previous->Tok.getIdentifierInfo() == nullptr ||
509         Current.Previous->Tok.getIdentifierInfo()->getPPKeywordID() ==
510             tok::pp_not_keyword))) {
511     unsigned EndColumn =
512         SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd());
513     if (Current.LastNewlineOffset != 0) {
514       // If there is a newline within this token, the final column will solely
515       // determined by the current end column.
516       State.Column = EndColumn;
517     } else {
518       unsigned StartColumn =
519           SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin());
520       assert(EndColumn >= StartColumn);
521       State.Column += EndColumn - StartColumn;
522     }
523     moveStateToNextToken(State, DryRun, /*Newline=*/false);
524     return 0;
525   }
526 
527   unsigned Penalty = 0;
528   if (Newline)
529     Penalty = addTokenOnNewLine(State, DryRun);
530   else
531     addTokenOnCurrentLine(State, DryRun, ExtraSpaces);
532 
533   return moveStateToNextToken(State, DryRun, Newline) + Penalty;
534 }
535 
536 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
537                                                  unsigned ExtraSpaces) {
538   FormatToken &Current = *State.NextToken;
539   const FormatToken &Previous = *State.NextToken->Previous;
540   if (Current.is(tok::equal) &&
541       (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) &&
542       State.Stack.back().VariablePos == 0) {
543     State.Stack.back().VariablePos = State.Column;
544     // Move over * and & if they are bound to the variable name.
545     const FormatToken *Tok = &Previous;
546     while (Tok && State.Stack.back().VariablePos >= Tok->ColumnWidth) {
547       State.Stack.back().VariablePos -= Tok->ColumnWidth;
548       if (Tok->SpacesRequiredBefore != 0)
549         break;
550       Tok = Tok->Previous;
551     }
552     if (Previous.PartOfMultiVariableDeclStmt)
553       State.Stack.back().LastSpace = State.Stack.back().VariablePos;
554   }
555 
556   unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces;
557 
558   // Indent preprocessor directives after the hash if required.
559   int PPColumnCorrection = 0;
560   if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
561       Previous.is(tok::hash) && State.FirstIndent > 0 &&
562       (State.Line->Type == LT_PreprocessorDirective ||
563        State.Line->Type == LT_ImportStatement)) {
564     Spaces += State.FirstIndent;
565 
566     // For preprocessor indent with tabs, State.Column will be 1 because of the
567     // hash. This causes second-level indents onward to have an extra space
568     // after the tabs. We avoid this misalignment by subtracting 1 from the
569     // column value passed to replaceWhitespace().
570     if (Style.UseTab != FormatStyle::UT_Never)
571       PPColumnCorrection = -1;
572   }
573 
574   if (!DryRun)
575     Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces,
576                                   State.Column + Spaces + PPColumnCorrection);
577 
578   // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance
579   // declaration unless there is multiple inheritance.
580   if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma &&
581       Current.is(TT_InheritanceColon))
582     State.Stack.back().NoLineBreak = true;
583   if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon &&
584       Previous.is(TT_InheritanceColon))
585     State.Stack.back().NoLineBreak = true;
586 
587   if (Current.is(TT_SelectorName) &&
588       !State.Stack.back().ObjCSelectorNameFound) {
589     unsigned MinIndent =
590         std::max(State.FirstIndent + Style.ContinuationIndentWidth,
591                  State.Stack.back().Indent);
592     unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth;
593     if (Current.LongestObjCSelectorName == 0)
594       State.Stack.back().AlignColons = false;
595     else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos)
596       State.Stack.back().ColonPos = MinIndent + Current.LongestObjCSelectorName;
597     else
598       State.Stack.back().ColonPos = FirstColonPos;
599   }
600 
601   // In "AlwaysBreak" mode, enforce wrapping directly after the parenthesis by
602   // disallowing any further line breaks if there is no line break after the
603   // opening parenthesis. Don't break if it doesn't conserve columns.
604   if (Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak &&
605       Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) &&
606       State.Column > getNewLineColumn(State) &&
607       (!Previous.Previous || !Previous.Previous->isOneOf(
608                                  tok::kw_for, tok::kw_while, tok::kw_switch)) &&
609       // Don't do this for simple (no expressions) one-argument function calls
610       // as that feels like needlessly wasting whitespace, e.g.:
611       //
612       //   caaaaaaaaaaaall(
613       //       caaaaaaaaaaaall(
614       //           caaaaaaaaaaaall(
615       //               caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa))));
616       Current.FakeLParens.size() > 0 &&
617       Current.FakeLParens.back() > prec::Unknown)
618     State.Stack.back().NoLineBreak = true;
619   if (Previous.is(TT_TemplateString) && Previous.opensScope())
620     State.Stack.back().NoLineBreak = true;
621 
622   if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign &&
623       Previous.opensScope() && Previous.isNot(TT_ObjCMethodExpr) &&
624       (Current.isNot(TT_LineComment) || Previous.BlockKind == BK_BracedInit))
625     State.Stack.back().Indent = State.Column + Spaces;
626   if (State.Stack.back().AvoidBinPacking && startsNextParameter(Current, Style))
627     State.Stack.back().NoLineBreak = true;
628   if (startsSegmentOfBuilderTypeCall(Current) &&
629       State.Column > getNewLineColumn(State))
630     State.Stack.back().ContainsUnwrappedBuilder = true;
631 
632   if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java)
633     State.Stack.back().NoLineBreak = true;
634   if (Current.isMemberAccess() && Previous.is(tok::r_paren) &&
635       (Previous.MatchingParen &&
636        (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10)))
637     // If there is a function call with long parameters, break before trailing
638     // calls. This prevents things like:
639     //   EXPECT_CALL(SomeLongParameter).Times(
640     //       2);
641     // We don't want to do this for short parameters as they can just be
642     // indexes.
643     State.Stack.back().NoLineBreak = true;
644 
645   // Don't allow the RHS of an operator to be split over multiple lines unless
646   // there is a line-break right after the operator.
647   // Exclude relational operators, as there, it is always more desirable to
648   // have the LHS 'left' of the RHS.
649   const FormatToken *P = Current.getPreviousNonComment();
650   if (!Current.is(tok::comment) && P &&
651       (P->isOneOf(TT_BinaryOperator, tok::comma) ||
652        (P->is(TT_ConditionalExpr) && P->is(tok::colon))) &&
653       !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) &&
654       P->getPrecedence() != prec::Assignment &&
655       P->getPrecedence() != prec::Relational &&
656       P->getPrecedence() != prec::Spaceship) {
657     bool BreakBeforeOperator =
658         P->MustBreakBefore || P->is(tok::lessless) ||
659         (P->is(TT_BinaryOperator) &&
660          Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) ||
661         (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators);
662     // Don't do this if there are only two operands. In these cases, there is
663     // always a nice vertical separation between them and the extra line break
664     // does not help.
665     bool HasTwoOperands =
666         P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr);
667     if ((!BreakBeforeOperator && !(HasTwoOperands && Style.AlignOperands)) ||
668         (!State.Stack.back().LastOperatorWrapped && BreakBeforeOperator))
669       State.Stack.back().NoLineBreakInOperand = true;
670   }
671 
672   State.Column += Spaces;
673   if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) &&
674       Previous.Previous &&
675       (Previous.Previous->isOneOf(tok::kw_if, tok::kw_for) ||
676        Previous.Previous->endsSequence(tok::kw_constexpr, tok::kw_if))) {
677     // Treat the condition inside an if as if it was a second function
678     // parameter, i.e. let nested calls have a continuation indent.
679     State.Stack.back().LastSpace = State.Column;
680     State.Stack.back().NestedBlockIndent = State.Column;
681   } else if (!Current.isOneOf(tok::comment, tok::caret) &&
682              ((Previous.is(tok::comma) &&
683                !Previous.is(TT_OverloadedOperator)) ||
684               (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) {
685     State.Stack.back().LastSpace = State.Column;
686   } else if (Previous.is(TT_CtorInitializerColon) &&
687              Style.BreakConstructorInitializers ==
688                  FormatStyle::BCIS_AfterColon) {
689     State.Stack.back().Indent = State.Column;
690     State.Stack.back().LastSpace = State.Column;
691   } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr,
692                                TT_CtorInitializerColon)) &&
693              ((Previous.getPrecedence() != prec::Assignment &&
694                (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 ||
695                 Previous.NextOperator)) ||
696               Current.StartsBinaryExpression)) {
697     // Indent relative to the RHS of the expression unless this is a simple
698     // assignment without binary expression on the RHS. Also indent relative to
699     // unary operators and the colons of constructor initializers.
700     State.Stack.back().LastSpace = State.Column;
701   } else if (Previous.is(TT_InheritanceColon)) {
702     State.Stack.back().Indent = State.Column;
703     State.Stack.back().LastSpace = State.Column;
704   } else if (Previous.opensScope()) {
705     // If a function has a trailing call, indent all parameters from the
706     // opening parenthesis. This avoids confusing indents like:
707     //   OuterFunction(InnerFunctionCall( // break
708     //       ParameterToInnerFunction))   // break
709     //       .SecondInnerFunctionCall();
710     bool HasTrailingCall = false;
711     if (Previous.MatchingParen) {
712       const FormatToken *Next = Previous.MatchingParen->getNextNonComment();
713       HasTrailingCall = Next && Next->isMemberAccess();
714     }
715     if (HasTrailingCall && State.Stack.size() > 1 &&
716         State.Stack[State.Stack.size() - 2].CallContinuation == 0)
717       State.Stack.back().LastSpace = State.Column;
718   }
719 }
720 
721 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State,
722                                                  bool DryRun) {
723   FormatToken &Current = *State.NextToken;
724   const FormatToken &Previous = *State.NextToken->Previous;
725 
726   // Extra penalty that needs to be added because of the way certain line
727   // breaks are chosen.
728   unsigned Penalty = 0;
729 
730   const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
731   const FormatToken *NextNonComment = Previous.getNextNonComment();
732   if (!NextNonComment)
733     NextNonComment = &Current;
734   // The first line break on any NestingLevel causes an extra penalty in order
735   // prefer similar line breaks.
736   if (!State.Stack.back().ContainsLineBreak)
737     Penalty += 15;
738   State.Stack.back().ContainsLineBreak = true;
739 
740   Penalty += State.NextToken->SplitPenalty;
741 
742   // Breaking before the first "<<" is generally not desirable if the LHS is
743   // short. Also always add the penalty if the LHS is split over multiple lines
744   // to avoid unnecessary line breaks that just work around this penalty.
745   if (NextNonComment->is(tok::lessless) &&
746       State.Stack.back().FirstLessLess == 0 &&
747       (State.Column <= Style.ColumnLimit / 3 ||
748        State.Stack.back().BreakBeforeParameter))
749     Penalty += Style.PenaltyBreakFirstLessLess;
750 
751   State.Column = getNewLineColumn(State);
752 
753   // Indent nested blocks relative to this column, unless in a very specific
754   // JavaScript special case where:
755   //
756   //   var loooooong_name =
757   //       function() {
758   //     // code
759   //   }
760   //
761   // is common and should be formatted like a free-standing function. The same
762   // goes for wrapping before the lambda return type arrow.
763   if (!Current.is(TT_LambdaArrow) &&
764       (Style.Language != FormatStyle::LK_JavaScript ||
765        Current.NestingLevel != 0 || !PreviousNonComment ||
766        !PreviousNonComment->is(tok::equal) ||
767        !Current.isOneOf(Keywords.kw_async, Keywords.kw_function)))
768     State.Stack.back().NestedBlockIndent = State.Column;
769 
770   if (NextNonComment->isMemberAccess()) {
771     if (State.Stack.back().CallContinuation == 0)
772       State.Stack.back().CallContinuation = State.Column;
773   } else if (NextNonComment->is(TT_SelectorName)) {
774     if (!State.Stack.back().ObjCSelectorNameFound) {
775       if (NextNonComment->LongestObjCSelectorName == 0) {
776         State.Stack.back().AlignColons = false;
777       } else {
778         State.Stack.back().ColonPos =
779             (shouldIndentWrappedSelectorName(Style, State.Line->Type)
780                  ? std::max(State.Stack.back().Indent,
781                             State.FirstIndent + Style.ContinuationIndentWidth)
782                  : State.Stack.back().Indent) +
783             std::max(NextNonComment->LongestObjCSelectorName,
784                      NextNonComment->ColumnWidth);
785       }
786     } else if (State.Stack.back().AlignColons &&
787                State.Stack.back().ColonPos <= NextNonComment->ColumnWidth) {
788       State.Stack.back().ColonPos = State.Column + NextNonComment->ColumnWidth;
789     }
790   } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
791              PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) {
792     // FIXME: This is hacky, find a better way. The problem is that in an ObjC
793     // method expression, the block should be aligned to the line starting it,
794     // e.g.:
795     //   [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason
796     //                        ^(int *i) {
797     //                            // ...
798     //                        }];
799     // Thus, we set LastSpace of the next higher NestingLevel, to which we move
800     // when we consume all of the "}"'s FakeRParens at the "{".
801     if (State.Stack.size() > 1)
802       State.Stack[State.Stack.size() - 2].LastSpace =
803           std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) +
804           Style.ContinuationIndentWidth;
805   }
806 
807   if ((PreviousNonComment &&
808        PreviousNonComment->isOneOf(tok::comma, tok::semi) &&
809        !State.Stack.back().AvoidBinPacking) ||
810       Previous.is(TT_BinaryOperator))
811     State.Stack.back().BreakBeforeParameter = false;
812   if (PreviousNonComment &&
813       PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) &&
814       Current.NestingLevel == 0)
815     State.Stack.back().BreakBeforeParameter = false;
816   if (NextNonComment->is(tok::question) ||
817       (PreviousNonComment && PreviousNonComment->is(tok::question)))
818     State.Stack.back().BreakBeforeParameter = true;
819   if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore)
820     State.Stack.back().BreakBeforeParameter = false;
821 
822   if (!DryRun) {
823     unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1;
824     if (Current.is(tok::r_brace) && Current.MatchingParen &&
825         // Only strip trailing empty lines for l_braces that have children, i.e.
826         // for function expressions (lambdas, arrows, etc).
827         !Current.MatchingParen->Children.empty()) {
828       // lambdas and arrow functions are expressions, thus their r_brace is not
829       // on its own line, and thus not covered by UnwrappedLineFormatter's logic
830       // about removing empty lines on closing blocks. Special case them here.
831       MaxEmptyLinesToKeep = 1;
832     }
833     unsigned Newlines = std::max(
834         1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep));
835     bool ContinuePPDirective =
836         State.Line->InPPDirective && State.Line->Type != LT_ImportStatement;
837     Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column,
838                                   ContinuePPDirective);
839   }
840 
841   if (!Current.isTrailingComment())
842     State.Stack.back().LastSpace = State.Column;
843   if (Current.is(tok::lessless))
844     // If we are breaking before a "<<", we always want to indent relative to
845     // RHS. This is necessary only for "<<", as we special-case it and don't
846     // always indent relative to the RHS.
847     State.Stack.back().LastSpace += 3; // 3 -> width of "<< ".
848 
849   State.StartOfLineLevel = Current.NestingLevel;
850   State.LowestLevelOnLine = Current.NestingLevel;
851 
852   // Any break on this level means that the parent level has been broken
853   // and we need to avoid bin packing there.
854   bool NestedBlockSpecialCase =
855       !Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 &&
856       State.Stack[State.Stack.size() - 2].NestedBlockInlined;
857   if (!NestedBlockSpecialCase)
858     for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i)
859       State.Stack[i].BreakBeforeParameter = true;
860 
861   if (PreviousNonComment &&
862       !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) &&
863       (PreviousNonComment->isNot(TT_TemplateCloser) ||
864        Current.NestingLevel != 0) &&
865       !PreviousNonComment->isOneOf(
866           TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation,
867           TT_LeadingJavaAnnotation) &&
868       Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope())
869     State.Stack.back().BreakBeforeParameter = true;
870 
871   // If we break after { or the [ of an array initializer, we should also break
872   // before the corresponding } or ].
873   if (PreviousNonComment &&
874       (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
875        opensProtoMessageField(*PreviousNonComment, Style)))
876     State.Stack.back().BreakBeforeClosingBrace = true;
877 
878   if (State.Stack.back().AvoidBinPacking) {
879     // If we are breaking after '(', '{', '<', this is not bin packing
880     // unless AllowAllParametersOfDeclarationOnNextLine is false or this is a
881     // dict/object literal.
882     if (!Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) ||
883         (!Style.AllowAllParametersOfDeclarationOnNextLine &&
884          State.Line->MustBeDeclaration) ||
885         Previous.is(TT_DictLiteral))
886       State.Stack.back().BreakBeforeParameter = true;
887   }
888 
889   return Penalty;
890 }
891 
892 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
893   if (!State.NextToken || !State.NextToken->Previous)
894     return 0;
895   FormatToken &Current = *State.NextToken;
896   const FormatToken &Previous = *Current.Previous;
897   // If we are continuing an expression, we want to use the continuation indent.
898   unsigned ContinuationIndent =
899       std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) +
900       Style.ContinuationIndentWidth;
901   const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
902   const FormatToken *NextNonComment = Previous.getNextNonComment();
903   if (!NextNonComment)
904     NextNonComment = &Current;
905 
906   // Java specific bits.
907   if (Style.Language == FormatStyle::LK_Java &&
908       Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends))
909     return std::max(State.Stack.back().LastSpace,
910                     State.Stack.back().Indent + Style.ContinuationIndentWidth);
911 
912   if (NextNonComment->is(tok::l_brace) && NextNonComment->BlockKind == BK_Block)
913     return Current.NestingLevel == 0 ? State.FirstIndent
914                                      : State.Stack.back().Indent;
915   if ((Current.isOneOf(tok::r_brace, tok::r_square) ||
916        (Current.is(tok::greater) &&
917         (Style.Language == FormatStyle::LK_Proto ||
918          Style.Language == FormatStyle::LK_TextProto))) &&
919       State.Stack.size() > 1) {
920     if (Current.closesBlockOrBlockTypeList(Style))
921       return State.Stack[State.Stack.size() - 2].NestedBlockIndent;
922     if (Current.MatchingParen &&
923         Current.MatchingParen->BlockKind == BK_BracedInit)
924       return State.Stack[State.Stack.size() - 2].LastSpace;
925     return State.FirstIndent;
926   }
927   // Indent a closing parenthesis at the previous level if followed by a semi or
928   // opening brace. This allows indentations such as:
929   //     foo(
930   //       a,
931   //     );
932   //     function foo(
933   //       a,
934   //     ) {
935   //       code(); //
936   //     }
937   if (Current.is(tok::r_paren) && State.Stack.size() > 1 &&
938       (!Current.Next || Current.Next->isOneOf(tok::semi, tok::l_brace)))
939     return State.Stack[State.Stack.size() - 2].LastSpace;
940   if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope())
941     return State.Stack[State.Stack.size() - 2].LastSpace;
942   if (Current.is(tok::identifier) && Current.Next &&
943       (Current.Next->is(TT_DictLiteral) ||
944        ((Style.Language == FormatStyle::LK_Proto ||
945          Style.Language == FormatStyle::LK_TextProto) &&
946         Current.Next->isOneOf(tok::less, tok::l_brace))))
947     return State.Stack.back().Indent;
948   if (NextNonComment->is(TT_ObjCStringLiteral) &&
949       State.StartOfStringLiteral != 0)
950     return State.StartOfStringLiteral - 1;
951   if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0)
952     return State.StartOfStringLiteral;
953   if (NextNonComment->is(tok::lessless) &&
954       State.Stack.back().FirstLessLess != 0)
955     return State.Stack.back().FirstLessLess;
956   if (NextNonComment->isMemberAccess()) {
957     if (State.Stack.back().CallContinuation == 0)
958       return ContinuationIndent;
959     return State.Stack.back().CallContinuation;
960   }
961   if (State.Stack.back().QuestionColumn != 0 &&
962       ((NextNonComment->is(tok::colon) &&
963         NextNonComment->is(TT_ConditionalExpr)) ||
964        Previous.is(TT_ConditionalExpr)))
965     return State.Stack.back().QuestionColumn;
966   if (Previous.is(tok::comma) && State.Stack.back().VariablePos != 0)
967     return State.Stack.back().VariablePos;
968   if ((PreviousNonComment &&
969        (PreviousNonComment->ClosesTemplateDeclaration ||
970         PreviousNonComment->isOneOf(
971             TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen,
972             TT_JavaAnnotation, TT_LeadingJavaAnnotation))) ||
973       (!Style.IndentWrappedFunctionNames &&
974        NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName)))
975     return std::max(State.Stack.back().LastSpace, State.Stack.back().Indent);
976   if (NextNonComment->is(TT_SelectorName)) {
977     if (!State.Stack.back().ObjCSelectorNameFound) {
978       unsigned MinIndent = State.Stack.back().Indent;
979       if (shouldIndentWrappedSelectorName(Style, State.Line->Type))
980         MinIndent = std::max(MinIndent,
981                              State.FirstIndent + Style.ContinuationIndentWidth);
982       // If LongestObjCSelectorName is 0, we are indenting the first
983       // part of an ObjC selector (or a selector component which is
984       // not colon-aligned due to block formatting).
985       //
986       // Otherwise, we are indenting a subsequent part of an ObjC
987       // selector which should be colon-aligned to the longest
988       // component of the ObjC selector.
989       //
990       // In either case, we want to respect Style.IndentWrappedFunctionNames.
991       return MinIndent +
992              std::max(NextNonComment->LongestObjCSelectorName,
993                       NextNonComment->ColumnWidth) -
994              NextNonComment->ColumnWidth;
995     }
996     if (!State.Stack.back().AlignColons)
997       return State.Stack.back().Indent;
998     if (State.Stack.back().ColonPos > NextNonComment->ColumnWidth)
999       return State.Stack.back().ColonPos - NextNonComment->ColumnWidth;
1000     return State.Stack.back().Indent;
1001   }
1002   if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr))
1003     return State.Stack.back().ColonPos;
1004   if (NextNonComment->is(TT_ArraySubscriptLSquare)) {
1005     if (State.Stack.back().StartOfArraySubscripts != 0)
1006       return State.Stack.back().StartOfArraySubscripts;
1007     return ContinuationIndent;
1008   }
1009 
1010   // This ensure that we correctly format ObjC methods calls without inputs,
1011   // i.e. where the last element isn't selector like: [callee method];
1012   if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 &&
1013       NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr))
1014     return State.Stack.back().Indent;
1015 
1016   if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) ||
1017       Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon))
1018     return ContinuationIndent;
1019   if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
1020       PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral))
1021     return ContinuationIndent;
1022   if (NextNonComment->is(TT_CtorInitializerComma))
1023     return State.Stack.back().Indent;
1024   if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) &&
1025       Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon)
1026     return State.Stack.back().Indent;
1027   if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) &&
1028       Style.BreakInheritanceList == FormatStyle::BILS_AfterColon)
1029     return State.Stack.back().Indent;
1030   if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon,
1031                               TT_InheritanceComma))
1032     return State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1033   if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() &&
1034       !Current.isOneOf(tok::colon, tok::comment))
1035     return ContinuationIndent;
1036   if (Current.is(TT_ProtoExtensionLSquare))
1037     return State.Stack.back().Indent;
1038   if (State.Stack.back().Indent == State.FirstIndent && PreviousNonComment &&
1039       PreviousNonComment->isNot(tok::r_brace))
1040     // Ensure that we fall back to the continuation indent width instead of
1041     // just flushing continuations left.
1042     return State.Stack.back().Indent + Style.ContinuationIndentWidth;
1043   return State.Stack.back().Indent;
1044 }
1045 
1046 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State,
1047                                                     bool DryRun, bool Newline) {
1048   assert(State.Stack.size());
1049   const FormatToken &Current = *State.NextToken;
1050 
1051   if (Current.isOneOf(tok::comma, TT_BinaryOperator))
1052     State.Stack.back().NoLineBreakInOperand = false;
1053   if (Current.is(TT_InheritanceColon))
1054     State.Stack.back().AvoidBinPacking = true;
1055   if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) {
1056     if (State.Stack.back().FirstLessLess == 0)
1057       State.Stack.back().FirstLessLess = State.Column;
1058     else
1059       State.Stack.back().LastOperatorWrapped = Newline;
1060   }
1061   if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless))
1062     State.Stack.back().LastOperatorWrapped = Newline;
1063   if (Current.is(TT_ConditionalExpr) && Current.Previous &&
1064       !Current.Previous->is(TT_ConditionalExpr))
1065     State.Stack.back().LastOperatorWrapped = Newline;
1066   if (Current.is(TT_ArraySubscriptLSquare) &&
1067       State.Stack.back().StartOfArraySubscripts == 0)
1068     State.Stack.back().StartOfArraySubscripts = State.Column;
1069   if (Style.BreakBeforeTernaryOperators && Current.is(tok::question))
1070     State.Stack.back().QuestionColumn = State.Column;
1071   if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) {
1072     const FormatToken *Previous = Current.Previous;
1073     while (Previous && Previous->isTrailingComment())
1074       Previous = Previous->Previous;
1075     if (Previous && Previous->is(tok::question))
1076       State.Stack.back().QuestionColumn = State.Column;
1077   }
1078   if (!Current.opensScope() && !Current.closesScope() &&
1079       !Current.is(TT_PointerOrReference))
1080     State.LowestLevelOnLine =
1081         std::min(State.LowestLevelOnLine, Current.NestingLevel);
1082   if (Current.isMemberAccess())
1083     State.Stack.back().StartOfFunctionCall =
1084         !Current.NextOperator ? 0 : State.Column;
1085   if (Current.is(TT_SelectorName))
1086     State.Stack.back().ObjCSelectorNameFound = true;
1087   if (Current.is(TT_CtorInitializerColon) &&
1088       Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) {
1089     // Indent 2 from the column, so:
1090     // SomeClass::SomeClass()
1091     //     : First(...), ...
1092     //       Next(...)
1093     //       ^ line up here.
1094     State.Stack.back().Indent =
1095         State.Column +
1096         (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma
1097              ? 0
1098              : 2);
1099     State.Stack.back().NestedBlockIndent = State.Stack.back().Indent;
1100     if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
1101       State.Stack.back().AvoidBinPacking = true;
1102     State.Stack.back().BreakBeforeParameter = false;
1103   }
1104   if (Current.is(TT_CtorInitializerColon) &&
1105       Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1106     State.Stack.back().Indent =
1107         State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1108     State.Stack.back().NestedBlockIndent = State.Stack.back().Indent;
1109     if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
1110       State.Stack.back().AvoidBinPacking = true;
1111   }
1112   if (Current.is(TT_InheritanceColon))
1113     State.Stack.back().Indent =
1114         State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1115   if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline)
1116     State.Stack.back().NestedBlockIndent =
1117         State.Column + Current.ColumnWidth + 1;
1118   if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow))
1119     State.Stack.back().LastSpace = State.Column;
1120 
1121   // Insert scopes created by fake parenthesis.
1122   const FormatToken *Previous = Current.getPreviousNonComment();
1123 
1124   // Add special behavior to support a format commonly used for JavaScript
1125   // closures:
1126   //   SomeFunction(function() {
1127   //     foo();
1128   //     bar();
1129   //   }, a, b, c);
1130   if (Current.isNot(tok::comment) && Previous &&
1131       Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
1132       !Previous->is(TT_DictLiteral) && State.Stack.size() > 1) {
1133     if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline)
1134       for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i)
1135         State.Stack[i].NoLineBreak = true;
1136     State.Stack[State.Stack.size() - 2].NestedBlockInlined = false;
1137   }
1138   if (Previous &&
1139       (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) ||
1140        Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr)) &&
1141       !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) {
1142     State.Stack.back().NestedBlockInlined =
1143         !Newline &&
1144         (Previous->isNot(tok::l_paren) || Previous->ParameterCount > 1);
1145   }
1146 
1147   moveStatePastFakeLParens(State, Newline);
1148   moveStatePastScopeCloser(State);
1149   bool AllowBreak = !State.Stack.back().NoLineBreak &&
1150                     !State.Stack.back().NoLineBreakInOperand;
1151   moveStatePastScopeOpener(State, Newline);
1152   moveStatePastFakeRParens(State);
1153 
1154   if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0)
1155     State.StartOfStringLiteral = State.Column + 1;
1156   else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0)
1157     State.StartOfStringLiteral = State.Column;
1158   else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) &&
1159            !Current.isStringLiteral())
1160     State.StartOfStringLiteral = 0;
1161 
1162   State.Column += Current.ColumnWidth;
1163   State.NextToken = State.NextToken->Next;
1164 
1165   unsigned Penalty =
1166       handleEndOfLine(Current, State, DryRun, AllowBreak);
1167 
1168   if (Current.Role)
1169     Current.Role->formatFromToken(State, this, DryRun);
1170   // If the previous has a special role, let it consume tokens as appropriate.
1171   // It is necessary to start at the previous token for the only implemented
1172   // role (comma separated list). That way, the decision whether or not to break
1173   // after the "{" is already done and both options are tried and evaluated.
1174   // FIXME: This is ugly, find a better way.
1175   if (Previous && Previous->Role)
1176     Penalty += Previous->Role->formatAfterToken(State, this, DryRun);
1177 
1178   return Penalty;
1179 }
1180 
1181 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State,
1182                                                     bool Newline) {
1183   const FormatToken &Current = *State.NextToken;
1184   const FormatToken *Previous = Current.getPreviousNonComment();
1185 
1186   // Don't add extra indentation for the first fake parenthesis after
1187   // 'return', assignments or opening <({[. The indentation for these cases
1188   // is special cased.
1189   bool SkipFirstExtraIndent =
1190       (Previous && (Previous->opensScope() ||
1191                     Previous->isOneOf(tok::semi, tok::kw_return) ||
1192                     (Previous->getPrecedence() == prec::Assignment &&
1193                      Style.AlignOperands) ||
1194                     Previous->is(TT_ObjCMethodExpr)));
1195   for (SmallVectorImpl<prec::Level>::const_reverse_iterator
1196            I = Current.FakeLParens.rbegin(),
1197            E = Current.FakeLParens.rend();
1198        I != E; ++I) {
1199     ParenState NewParenState = State.Stack.back();
1200     NewParenState.Tok = nullptr;
1201     NewParenState.ContainsLineBreak = false;
1202     NewParenState.LastOperatorWrapped = true;
1203     NewParenState.NoLineBreak =
1204         NewParenState.NoLineBreak || State.Stack.back().NoLineBreakInOperand;
1205 
1206     // Don't propagate AvoidBinPacking into subexpressions of arg/param lists.
1207     if (*I > prec::Comma)
1208       NewParenState.AvoidBinPacking = false;
1209 
1210     // Indent from 'LastSpace' unless these are fake parentheses encapsulating
1211     // a builder type call after 'return' or, if the alignment after opening
1212     // brackets is disabled.
1213     if (!Current.isTrailingComment() &&
1214         (Style.AlignOperands || *I < prec::Assignment) &&
1215         (!Previous || Previous->isNot(tok::kw_return) ||
1216          (Style.Language != FormatStyle::LK_Java && *I > 0)) &&
1217         (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign ||
1218          *I != prec::Comma || Current.NestingLevel == 0))
1219       NewParenState.Indent =
1220           std::max(std::max(State.Column, NewParenState.Indent),
1221                    State.Stack.back().LastSpace);
1222 
1223     // Do not indent relative to the fake parentheses inserted for "." or "->".
1224     // This is a special case to make the following to statements consistent:
1225     //   OuterFunction(InnerFunctionCall( // break
1226     //       ParameterToInnerFunction));
1227     //   OuterFunction(SomeObject.InnerFunctionCall( // break
1228     //       ParameterToInnerFunction));
1229     if (*I > prec::Unknown)
1230       NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column);
1231     if (*I != prec::Conditional && !Current.is(TT_UnaryOperator) &&
1232         Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign)
1233       NewParenState.StartOfFunctionCall = State.Column;
1234 
1235     // Always indent conditional expressions. Never indent expression where
1236     // the 'operator' is ',', ';' or an assignment (i.e. *I <=
1237     // prec::Assignment) as those have different indentation rules. Indent
1238     // other expression, unless the indentation needs to be skipped.
1239     if (*I == prec::Conditional ||
1240         (!SkipFirstExtraIndent && *I > prec::Assignment &&
1241          !Current.isTrailingComment()))
1242       NewParenState.Indent += Style.ContinuationIndentWidth;
1243     if ((Previous && !Previous->opensScope()) || *I != prec::Comma)
1244       NewParenState.BreakBeforeParameter = false;
1245     State.Stack.push_back(NewParenState);
1246     SkipFirstExtraIndent = false;
1247   }
1248 }
1249 
1250 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) {
1251   for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) {
1252     unsigned VariablePos = State.Stack.back().VariablePos;
1253     if (State.Stack.size() == 1) {
1254       // Do not pop the last element.
1255       break;
1256     }
1257     State.Stack.pop_back();
1258     State.Stack.back().VariablePos = VariablePos;
1259   }
1260 }
1261 
1262 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
1263                                                     bool Newline) {
1264   const FormatToken &Current = *State.NextToken;
1265   if (!Current.opensScope())
1266     return;
1267 
1268   if (Current.MatchingParen && Current.BlockKind == BK_Block) {
1269     moveStateToNewBlock(State);
1270     return;
1271   }
1272 
1273   unsigned NewIndent;
1274   unsigned LastSpace = State.Stack.back().LastSpace;
1275   bool AvoidBinPacking;
1276   bool BreakBeforeParameter = false;
1277   unsigned NestedBlockIndent = std::max(State.Stack.back().StartOfFunctionCall,
1278                                         State.Stack.back().NestedBlockIndent);
1279   if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
1280       opensProtoMessageField(Current, Style)) {
1281     if (Current.opensBlockOrBlockTypeList(Style)) {
1282       NewIndent = Style.IndentWidth +
1283                   std::min(State.Column, State.Stack.back().NestedBlockIndent);
1284     } else {
1285       NewIndent = State.Stack.back().LastSpace + Style.ContinuationIndentWidth;
1286     }
1287     const FormatToken *NextNoComment = Current.getNextNonComment();
1288     bool EndsInComma = Current.MatchingParen &&
1289                        Current.MatchingParen->Previous &&
1290                        Current.MatchingParen->Previous->is(tok::comma);
1291     AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) ||
1292                       Style.Language == FormatStyle::LK_Proto ||
1293                       Style.Language == FormatStyle::LK_TextProto ||
1294                       !Style.BinPackArguments ||
1295                       (NextNoComment &&
1296                        NextNoComment->isOneOf(TT_DesignatedInitializerPeriod,
1297                                               TT_DesignatedInitializerLSquare));
1298     BreakBeforeParameter = EndsInComma;
1299     if (Current.ParameterCount > 1)
1300       NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1);
1301   } else {
1302     NewIndent = Style.ContinuationIndentWidth +
1303                 std::max(State.Stack.back().LastSpace,
1304                          State.Stack.back().StartOfFunctionCall);
1305 
1306     // Ensure that different different brackets force relative alignment, e.g.:
1307     // void SomeFunction(vector<  // break
1308     //                       int> v);
1309     // FIXME: We likely want to do this for more combinations of brackets.
1310     if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) {
1311       NewIndent = std::max(NewIndent, State.Stack.back().Indent);
1312       LastSpace = std::max(LastSpace, State.Stack.back().Indent);
1313     }
1314 
1315     bool EndsInComma =
1316         Current.MatchingParen &&
1317         Current.MatchingParen->getPreviousNonComment() &&
1318         Current.MatchingParen->getPreviousNonComment()->is(tok::comma);
1319 
1320     // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters
1321     // for backwards compatibility.
1322     bool ObjCBinPackProtocolList =
1323         (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto &&
1324          Style.BinPackParameters) ||
1325         Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always;
1326 
1327     bool BinPackDeclaration =
1328         (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) ||
1329         (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList);
1330 
1331     AvoidBinPacking =
1332         (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) ||
1333         (State.Line->MustBeDeclaration && !BinPackDeclaration) ||
1334         (!State.Line->MustBeDeclaration && !Style.BinPackArguments) ||
1335         (Style.ExperimentalAutoDetectBinPacking &&
1336          (Current.PackingKind == PPK_OnePerLine ||
1337           (!BinPackInconclusiveFunctions &&
1338            Current.PackingKind == PPK_Inconclusive)));
1339 
1340     if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen) {
1341       if (Style.ColumnLimit) {
1342         // If this '[' opens an ObjC call, determine whether all parameters fit
1343         // into one line and put one per line if they don't.
1344         if (getLengthToMatchingParen(Current, State.Stack) + State.Column >
1345             getColumnLimit(State))
1346           BreakBeforeParameter = true;
1347       } else {
1348         // For ColumnLimit = 0, we have to figure out whether there is or has to
1349         // be a line break within this call.
1350         for (const FormatToken *Tok = &Current;
1351              Tok && Tok != Current.MatchingParen; Tok = Tok->Next) {
1352           if (Tok->MustBreakBefore ||
1353               (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) {
1354             BreakBeforeParameter = true;
1355             break;
1356           }
1357         }
1358       }
1359     }
1360 
1361     if (Style.Language == FormatStyle::LK_JavaScript && EndsInComma)
1362       BreakBeforeParameter = true;
1363   }
1364   // Generally inherit NoLineBreak from the current scope to nested scope.
1365   // However, don't do this for non-empty nested blocks, dict literals and
1366   // array literals as these follow different indentation rules.
1367   bool NoLineBreak =
1368       Current.Children.empty() &&
1369       !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) &&
1370       (State.Stack.back().NoLineBreak ||
1371        State.Stack.back().NoLineBreakInOperand ||
1372        (Current.is(TT_TemplateOpener) &&
1373         State.Stack.back().ContainsUnwrappedBuilder));
1374   State.Stack.push_back(
1375       ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak));
1376   State.Stack.back().NestedBlockIndent = NestedBlockIndent;
1377   State.Stack.back().BreakBeforeParameter = BreakBeforeParameter;
1378   State.Stack.back().HasMultipleNestedBlocks = Current.BlockParameterCount > 1;
1379   State.Stack.back().IsInsideObjCArrayLiteral =
1380       Current.is(TT_ArrayInitializerLSquare) && Current.Previous &&
1381       Current.Previous->is(tok::at);
1382 }
1383 
1384 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) {
1385   const FormatToken &Current = *State.NextToken;
1386   if (!Current.closesScope())
1387     return;
1388 
1389   // If we encounter a closing ), ], } or >, we can remove a level from our
1390   // stacks.
1391   if (State.Stack.size() > 1 &&
1392       (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) ||
1393        (Current.is(tok::r_brace) && State.NextToken != State.Line->First) ||
1394        State.NextToken->is(TT_TemplateCloser) ||
1395        (Current.is(tok::greater) && Current.is(TT_DictLiteral))))
1396     State.Stack.pop_back();
1397 
1398   if (Current.is(tok::r_square)) {
1399     // If this ends the array subscript expr, reset the corresponding value.
1400     const FormatToken *NextNonComment = Current.getNextNonComment();
1401     if (NextNonComment && NextNonComment->isNot(tok::l_square))
1402       State.Stack.back().StartOfArraySubscripts = 0;
1403   }
1404 }
1405 
1406 void ContinuationIndenter::moveStateToNewBlock(LineState &State) {
1407   unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent;
1408   // ObjC block sometimes follow special indentation rules.
1409   unsigned NewIndent =
1410       NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace)
1411                                ? Style.ObjCBlockIndentWidth
1412                                : Style.IndentWidth);
1413   State.Stack.push_back(ParenState(State.NextToken, NewIndent,
1414                                    State.Stack.back().LastSpace,
1415                                    /*AvoidBinPacking=*/true,
1416                                    /*NoLineBreak=*/false));
1417   State.Stack.back().NestedBlockIndent = NestedBlockIndent;
1418   State.Stack.back().BreakBeforeParameter = true;
1419 }
1420 
1421 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn,
1422                                      unsigned TabWidth,
1423                                      encoding::Encoding Encoding) {
1424   size_t LastNewlinePos = Text.find_last_of("\n");
1425   if (LastNewlinePos == StringRef::npos) {
1426     return StartColumn +
1427            encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding);
1428   } else {
1429     return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos),
1430                                          /*StartColumn=*/0, TabWidth, Encoding);
1431   }
1432 }
1433 
1434 unsigned ContinuationIndenter::reformatRawStringLiteral(
1435     const FormatToken &Current, LineState &State,
1436     const FormatStyle &RawStringStyle, bool DryRun) {
1437   unsigned StartColumn = State.Column - Current.ColumnWidth;
1438   StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText);
1439   StringRef NewDelimiter =
1440       getCanonicalRawStringDelimiter(Style, RawStringStyle.Language);
1441   if (NewDelimiter.empty() || OldDelimiter.empty())
1442     NewDelimiter = OldDelimiter;
1443   // The text of a raw string is between the leading 'R"delimiter(' and the
1444   // trailing 'delimiter)"'.
1445   unsigned OldPrefixSize = 3 + OldDelimiter.size();
1446   unsigned OldSuffixSize = 2 + OldDelimiter.size();
1447   // We create a virtual text environment which expects a null-terminated
1448   // string, so we cannot use StringRef.
1449   std::string RawText =
1450       Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize);
1451   if (NewDelimiter != OldDelimiter) {
1452     // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the
1453     // raw string.
1454     std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str();
1455     if (StringRef(RawText).contains(CanonicalDelimiterSuffix))
1456       NewDelimiter = OldDelimiter;
1457   }
1458 
1459   unsigned NewPrefixSize = 3 + NewDelimiter.size();
1460   unsigned NewSuffixSize = 2 + NewDelimiter.size();
1461 
1462   // The first start column is the column the raw text starts after formatting.
1463   unsigned FirstStartColumn = StartColumn + NewPrefixSize;
1464 
1465   // The next start column is the intended indentation a line break inside
1466   // the raw string at level 0. It is determined by the following rules:
1467   //   - if the content starts on newline, it is one level more than the current
1468   //     indent, and
1469   //   - if the content does not start on a newline, it is the first start
1470   //     column.
1471   // These rules have the advantage that the formatted content both does not
1472   // violate the rectangle rule and visually flows within the surrounding
1473   // source.
1474   bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n';
1475   unsigned NextStartColumn =
1476       ContentStartsOnNewline
1477           ? State.Stack.back().NestedBlockIndent + Style.IndentWidth
1478           : FirstStartColumn;
1479 
1480   // The last start column is the column the raw string suffix starts if it is
1481   // put on a newline.
1482   // The last start column is the intended indentation of the raw string postfix
1483   // if it is put on a newline. It is determined by the following rules:
1484   //   - if the raw string prefix starts on a newline, it is the column where
1485   //     that raw string prefix starts, and
1486   //   - if the raw string prefix does not start on a newline, it is the current
1487   //     indent.
1488   unsigned LastStartColumn = Current.NewlinesBefore
1489                                  ? FirstStartColumn - NewPrefixSize
1490                                  : State.Stack.back().NestedBlockIndent;
1491 
1492   std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat(
1493       RawStringStyle, RawText, {tooling::Range(0, RawText.size())},
1494       FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>",
1495       /*Status=*/nullptr);
1496 
1497   auto NewCode = applyAllReplacements(RawText, Fixes.first);
1498   tooling::Replacements NoFixes;
1499   if (!NewCode) {
1500     State.Column += Current.ColumnWidth;
1501     return 0;
1502   }
1503   if (!DryRun) {
1504     if (NewDelimiter != OldDelimiter) {
1505       // In 'R"delimiter(...', the delimiter starts 2 characters after the start
1506       // of the token.
1507       SourceLocation PrefixDelimiterStart =
1508           Current.Tok.getLocation().getLocWithOffset(2);
1509       auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement(
1510           SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter));
1511       if (PrefixErr) {
1512         llvm::errs()
1513             << "Failed to update the prefix delimiter of a raw string: "
1514             << llvm::toString(std::move(PrefixErr)) << "\n";
1515       }
1516       // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at
1517       // position length - 1 - |delimiter|.
1518       SourceLocation SuffixDelimiterStart =
1519           Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() -
1520                                                      1 - OldDelimiter.size());
1521       auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement(
1522           SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter));
1523       if (SuffixErr) {
1524         llvm::errs()
1525             << "Failed to update the suffix delimiter of a raw string: "
1526             << llvm::toString(std::move(SuffixErr)) << "\n";
1527       }
1528     }
1529     SourceLocation OriginLoc =
1530         Current.Tok.getLocation().getLocWithOffset(OldPrefixSize);
1531     for (const tooling::Replacement &Fix : Fixes.first) {
1532       auto Err = Whitespaces.addReplacement(tooling::Replacement(
1533           SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()),
1534           Fix.getLength(), Fix.getReplacementText()));
1535       if (Err) {
1536         llvm::errs() << "Failed to reformat raw string: "
1537                      << llvm::toString(std::move(Err)) << "\n";
1538       }
1539     }
1540   }
1541   unsigned RawLastLineEndColumn = getLastLineEndColumn(
1542       *NewCode, FirstStartColumn, Style.TabWidth, Encoding);
1543   State.Column = RawLastLineEndColumn + NewSuffixSize;
1544   // Since we're updating the column to after the raw string literal here, we
1545   // have to manually add the penalty for the prefix R"delim( over the column
1546   // limit.
1547   unsigned PrefixExcessCharacters =
1548       StartColumn + NewPrefixSize > Style.ColumnLimit ?
1549       StartColumn + NewPrefixSize - Style.ColumnLimit : 0;
1550   return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter;
1551 }
1552 
1553 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current,
1554                                                  LineState &State) {
1555   // Break before further function parameters on all levels.
1556   for (unsigned i = 0, e = State.Stack.size(); i != e; ++i)
1557     State.Stack[i].BreakBeforeParameter = true;
1558 
1559   unsigned ColumnsUsed = State.Column;
1560   // We can only affect layout of the first and the last line, so the penalty
1561   // for all other lines is constant, and we ignore it.
1562   State.Column = Current.LastLineColumnWidth;
1563 
1564   if (ColumnsUsed > getColumnLimit(State))
1565     return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State));
1566   return 0;
1567 }
1568 
1569 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current,
1570                                                LineState &State, bool DryRun,
1571                                                bool AllowBreak) {
1572   unsigned Penalty = 0;
1573   // Compute the raw string style to use in case this is a raw string literal
1574   // that can be reformatted.
1575   auto RawStringStyle = getRawStringStyle(Current, State);
1576   if (RawStringStyle && !Current.Finalized) {
1577     Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun);
1578   } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) {
1579     // Don't break multi-line tokens other than block comments and raw string
1580     // literals. Instead, just update the state.
1581     Penalty = addMultilineToken(Current, State);
1582   } else if (State.Line->Type != LT_ImportStatement) {
1583     // We generally don't break import statements.
1584     LineState OriginalState = State;
1585 
1586     // Whether we force the reflowing algorithm to stay strictly within the
1587     // column limit.
1588     bool Strict = false;
1589     // Whether the first non-strict attempt at reflowing did intentionally
1590     // exceed the column limit.
1591     bool Exceeded = false;
1592     std::tie(Penalty, Exceeded) = breakProtrudingToken(
1593         Current, State, AllowBreak, /*DryRun=*/true, Strict);
1594     if (Exceeded) {
1595       // If non-strict reflowing exceeds the column limit, try whether strict
1596       // reflowing leads to an overall lower penalty.
1597       LineState StrictState = OriginalState;
1598       unsigned StrictPenalty =
1599           breakProtrudingToken(Current, StrictState, AllowBreak,
1600                                /*DryRun=*/true, /*Strict=*/true)
1601               .first;
1602       Strict = StrictPenalty <= Penalty;
1603       if (Strict) {
1604         Penalty = StrictPenalty;
1605         State = StrictState;
1606       }
1607     }
1608     if (!DryRun) {
1609       // If we're not in dry-run mode, apply the changes with the decision on
1610       // strictness made above.
1611       breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false,
1612                            Strict);
1613     }
1614   }
1615   if (State.Column > getColumnLimit(State)) {
1616     unsigned ExcessCharacters = State.Column - getColumnLimit(State);
1617     Penalty += Style.PenaltyExcessCharacter * ExcessCharacters;
1618   }
1619   return Penalty;
1620 }
1621 
1622 // Returns the enclosing function name of a token, or the empty string if not
1623 // found.
1624 static StringRef getEnclosingFunctionName(const FormatToken &Current) {
1625   // Look for: 'function(' or 'function<templates>(' before Current.
1626   auto Tok = Current.getPreviousNonComment();
1627   if (!Tok || !Tok->is(tok::l_paren))
1628     return "";
1629   Tok = Tok->getPreviousNonComment();
1630   if (!Tok)
1631     return "";
1632   if (Tok->is(TT_TemplateCloser)) {
1633     Tok = Tok->MatchingParen;
1634     if (Tok)
1635       Tok = Tok->getPreviousNonComment();
1636   }
1637   if (!Tok || !Tok->is(tok::identifier))
1638     return "";
1639   return Tok->TokenText;
1640 }
1641 
1642 llvm::Optional<FormatStyle>
1643 ContinuationIndenter::getRawStringStyle(const FormatToken &Current,
1644                                         const LineState &State) {
1645   if (!Current.isStringLiteral())
1646     return None;
1647   auto Delimiter = getRawStringDelimiter(Current.TokenText);
1648   if (!Delimiter)
1649     return None;
1650   auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter);
1651   if (!RawStringStyle)
1652     RawStringStyle = RawStringFormats.getEnclosingFunctionStyle(
1653         getEnclosingFunctionName(Current));
1654   if (!RawStringStyle)
1655     return None;
1656   RawStringStyle->ColumnLimit = getColumnLimit(State);
1657   return RawStringStyle;
1658 }
1659 
1660 std::unique_ptr<BreakableToken> ContinuationIndenter::createBreakableToken(
1661     const FormatToken &Current, LineState &State, bool AllowBreak) {
1662   unsigned StartColumn = State.Column - Current.ColumnWidth;
1663   if (Current.isStringLiteral()) {
1664     // FIXME: String literal breaking is currently disabled for Java and JS, as
1665     // it requires strings to be merged using "+" which we don't support.
1666     if (Style.Language == FormatStyle::LK_Java ||
1667         Style.Language == FormatStyle::LK_JavaScript ||
1668         !Style.BreakStringLiterals ||
1669         !AllowBreak)
1670       return nullptr;
1671 
1672     // Don't break string literals inside preprocessor directives (except for
1673     // #define directives, as their contents are stored in separate lines and
1674     // are not affected by this check).
1675     // This way we avoid breaking code with line directives and unknown
1676     // preprocessor directives that contain long string literals.
1677     if (State.Line->Type == LT_PreprocessorDirective)
1678       return nullptr;
1679     // Exempts unterminated string literals from line breaking. The user will
1680     // likely want to terminate the string before any line breaking is done.
1681     if (Current.IsUnterminatedLiteral)
1682       return nullptr;
1683     // Don't break string literals inside Objective-C array literals (doing so
1684     // raises the warning -Wobjc-string-concatenation).
1685     if (State.Stack.back().IsInsideObjCArrayLiteral) {
1686       return nullptr;
1687     }
1688 
1689     StringRef Text = Current.TokenText;
1690     StringRef Prefix;
1691     StringRef Postfix;
1692     // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'.
1693     // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to
1694     // reduce the overhead) for each FormatToken, which is a string, so that we
1695     // don't run multiple checks here on the hot path.
1696     if ((Text.endswith(Postfix = "\"") &&
1697          (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") ||
1698           Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") ||
1699           Text.startswith(Prefix = "u8\"") ||
1700           Text.startswith(Prefix = "L\""))) ||
1701         (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) {
1702       // We need this to address the case where there is an unbreakable tail
1703       // only if certain other formatting decisions have been taken. The
1704       // UnbreakableTailLength of Current is an overapproximation is that case
1705       // and we need to be correct here.
1706       unsigned UnbreakableTailLength = (State.NextToken && canBreak(State))
1707                                            ? 0
1708                                            : Current.UnbreakableTailLength;
1709       return llvm::make_unique<BreakableStringLiteral>(
1710           Current, StartColumn, Prefix, Postfix, UnbreakableTailLength,
1711           State.Line->InPPDirective, Encoding, Style);
1712     }
1713   } else if (Current.is(TT_BlockComment)) {
1714     if (!Style.ReflowComments ||
1715         // If a comment token switches formatting, like
1716         // /* clang-format on */, we don't want to break it further,
1717         // but we may still want to adjust its indentation.
1718         switchesFormatting(Current)) {
1719       return nullptr;
1720     }
1721     return llvm::make_unique<BreakableBlockComment>(
1722         Current, StartColumn, Current.OriginalColumn, !Current.Previous,
1723         State.Line->InPPDirective, Encoding, Style);
1724   } else if (Current.is(TT_LineComment) &&
1725              (Current.Previous == nullptr ||
1726               Current.Previous->isNot(TT_ImplicitStringLiteral))) {
1727     if (!Style.ReflowComments ||
1728         CommentPragmasRegex.match(Current.TokenText.substr(2)) ||
1729         switchesFormatting(Current))
1730       return nullptr;
1731     return llvm::make_unique<BreakableLineCommentSection>(
1732         Current, StartColumn, Current.OriginalColumn, !Current.Previous,
1733         /*InPPDirective=*/false, Encoding, Style);
1734   }
1735   return nullptr;
1736 }
1737 
1738 std::pair<unsigned, bool>
1739 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current,
1740                                            LineState &State, bool AllowBreak,
1741                                            bool DryRun, bool Strict) {
1742   std::unique_ptr<const BreakableToken> Token =
1743       createBreakableToken(Current, State, AllowBreak);
1744   if (!Token)
1745     return {0, false};
1746   assert(Token->getLineCount() > 0);
1747   unsigned ColumnLimit = getColumnLimit(State);
1748   if (Current.is(TT_LineComment)) {
1749     // We don't insert backslashes when breaking line comments.
1750     ColumnLimit = Style.ColumnLimit;
1751   }
1752   if (Current.UnbreakableTailLength >= ColumnLimit)
1753     return {0, false};
1754   // ColumnWidth was already accounted into State.Column before calling
1755   // breakProtrudingToken.
1756   unsigned StartColumn = State.Column - Current.ColumnWidth;
1757   unsigned NewBreakPenalty = Current.isStringLiteral()
1758                                  ? Style.PenaltyBreakString
1759                                  : Style.PenaltyBreakComment;
1760   // Stores whether we intentionally decide to let a line exceed the column
1761   // limit.
1762   bool Exceeded = false;
1763   // Stores whether we introduce a break anywhere in the token.
1764   bool BreakInserted = Token->introducesBreakBeforeToken();
1765   // Store whether we inserted a new line break at the end of the previous
1766   // logical line.
1767   bool NewBreakBefore = false;
1768   // We use a conservative reflowing strategy. Reflow starts after a line is
1769   // broken or the corresponding whitespace compressed. Reflow ends as soon as a
1770   // line that doesn't get reflown with the previous line is reached.
1771   bool Reflow = false;
1772   // Keep track of where we are in the token:
1773   // Where we are in the content of the current logical line.
1774   unsigned TailOffset = 0;
1775   // The column number we're currently at.
1776   unsigned ContentStartColumn =
1777       Token->getContentStartColumn(0, /*Break=*/false);
1778   // The number of columns left in the current logical line after TailOffset.
1779   unsigned RemainingTokenColumns =
1780       Token->getRemainingLength(0, TailOffset, ContentStartColumn);
1781   // Adapt the start of the token, for example indent.
1782   if (!DryRun)
1783     Token->adaptStartOfLine(0, Whitespaces);
1784 
1785   unsigned Penalty = 0;
1786   LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column "
1787                           << StartColumn << ".\n");
1788   for (unsigned LineIndex = 0, EndIndex = Token->getLineCount();
1789        LineIndex != EndIndex; ++LineIndex) {
1790     LLVM_DEBUG(llvm::dbgs()
1791                << "  Line: " << LineIndex << " (Reflow: " << Reflow << ")\n");
1792     NewBreakBefore = false;
1793     // If we did reflow the previous line, we'll try reflowing again. Otherwise
1794     // we'll start reflowing if the current line is broken or whitespace is
1795     // compressed.
1796     bool TryReflow = Reflow;
1797     // Break the current token until we can fit the rest of the line.
1798     while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
1799       LLVM_DEBUG(llvm::dbgs() << "    Over limit, need: "
1800                               << (ContentStartColumn + RemainingTokenColumns)
1801                               << ", space: " << ColumnLimit
1802                               << ", reflown prefix: " << ContentStartColumn
1803                               << ", offset in line: " << TailOffset << "\n");
1804       // If the current token doesn't fit, find the latest possible split in the
1805       // current line so that breaking at it will be under the column limit.
1806       // FIXME: Use the earliest possible split while reflowing to correctly
1807       // compress whitespace within a line.
1808       BreakableToken::Split Split =
1809           Token->getSplit(LineIndex, TailOffset, ColumnLimit,
1810                           ContentStartColumn, CommentPragmasRegex);
1811       if (Split.first == StringRef::npos) {
1812         // No break opportunity - update the penalty and continue with the next
1813         // logical line.
1814         if (LineIndex < EndIndex - 1)
1815           // The last line's penalty is handled in addNextStateToQueue().
1816           Penalty += Style.PenaltyExcessCharacter *
1817                      (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
1818         LLVM_DEBUG(llvm::dbgs() << "    No break opportunity.\n");
1819         break;
1820       }
1821       assert(Split.first != 0);
1822 
1823       if (Token->supportsReflow()) {
1824         // Check whether the next natural split point after the current one can
1825         // still fit the line, either because we can compress away whitespace,
1826         // or because the penalty the excess characters introduce is lower than
1827         // the break penalty.
1828         // We only do this for tokens that support reflowing, and thus allow us
1829         // to change the whitespace arbitrarily (e.g. comments).
1830         // Other tokens, like string literals, can be broken on arbitrary
1831         // positions.
1832 
1833         // First, compute the columns from TailOffset to the next possible split
1834         // position.
1835         // For example:
1836         // ColumnLimit:     |
1837         // // Some text   that    breaks
1838         //    ^ tail offset
1839         //             ^-- split
1840         //    ^-------- to split columns
1841         //                    ^--- next split
1842         //    ^--------------- to next split columns
1843         unsigned ToSplitColumns = Token->getRangeLength(
1844             LineIndex, TailOffset, Split.first, ContentStartColumn);
1845         LLVM_DEBUG(llvm::dbgs() << "    ToSplit: " << ToSplitColumns << "\n");
1846 
1847         BreakableToken::Split NextSplit = Token->getSplit(
1848             LineIndex, TailOffset + Split.first + Split.second, ColumnLimit,
1849             ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex);
1850         // Compute the columns necessary to fit the next non-breakable sequence
1851         // into the current line.
1852         unsigned ToNextSplitColumns = 0;
1853         if (NextSplit.first == StringRef::npos) {
1854           ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset,
1855                                                          ContentStartColumn);
1856         } else {
1857           ToNextSplitColumns = Token->getRangeLength(
1858               LineIndex, TailOffset,
1859               Split.first + Split.second + NextSplit.first, ContentStartColumn);
1860         }
1861         // Compress the whitespace between the break and the start of the next
1862         // unbreakable sequence.
1863         ToNextSplitColumns =
1864             Token->getLengthAfterCompression(ToNextSplitColumns, Split);
1865         LLVM_DEBUG(llvm::dbgs()
1866                    << "    ContentStartColumn: " << ContentStartColumn << "\n");
1867         LLVM_DEBUG(llvm::dbgs()
1868                    << "    ToNextSplit: " << ToNextSplitColumns << "\n");
1869         // If the whitespace compression makes us fit, continue on the current
1870         // line.
1871         bool ContinueOnLine =
1872             ContentStartColumn + ToNextSplitColumns <= ColumnLimit;
1873         unsigned ExcessCharactersPenalty = 0;
1874         if (!ContinueOnLine && !Strict) {
1875           // Similarly, if the excess characters' penalty is lower than the
1876           // penalty of introducing a new break, continue on the current line.
1877           ExcessCharactersPenalty =
1878               (ContentStartColumn + ToNextSplitColumns - ColumnLimit) *
1879               Style.PenaltyExcessCharacter;
1880           LLVM_DEBUG(llvm::dbgs()
1881                      << "    Penalty excess: " << ExcessCharactersPenalty
1882                      << "\n            break : " << NewBreakPenalty << "\n");
1883           if (ExcessCharactersPenalty < NewBreakPenalty) {
1884             Exceeded = true;
1885             ContinueOnLine = true;
1886           }
1887         }
1888         if (ContinueOnLine) {
1889           LLVM_DEBUG(llvm::dbgs() << "    Continuing on line...\n");
1890           // The current line fits after compressing the whitespace - reflow
1891           // the next line into it if possible.
1892           TryReflow = true;
1893           if (!DryRun)
1894             Token->compressWhitespace(LineIndex, TailOffset, Split,
1895                                       Whitespaces);
1896           // When we continue on the same line, leave one space between content.
1897           ContentStartColumn += ToSplitColumns + 1;
1898           Penalty += ExcessCharactersPenalty;
1899           TailOffset += Split.first + Split.second;
1900           RemainingTokenColumns = Token->getRemainingLength(
1901               LineIndex, TailOffset, ContentStartColumn);
1902           continue;
1903         }
1904       }
1905       LLVM_DEBUG(llvm::dbgs() << "    Breaking...\n");
1906       ContentStartColumn =
1907           Token->getContentStartColumn(LineIndex, /*Break=*/true);
1908       unsigned NewRemainingTokenColumns = Token->getRemainingLength(
1909           LineIndex, TailOffset + Split.first + Split.second,
1910           ContentStartColumn);
1911 
1912       // When breaking before a tab character, it may be moved by a few columns,
1913       // but will still be expanded to the next tab stop, so we don't save any
1914       // columns.
1915       if (NewRemainingTokenColumns == RemainingTokenColumns) {
1916         // FIXME: Do we need to adjust the penalty?
1917         break;
1918       }
1919       assert(NewRemainingTokenColumns < RemainingTokenColumns);
1920 
1921       LLVM_DEBUG(llvm::dbgs() << "    Breaking at: " << TailOffset + Split.first
1922                               << ", " << Split.second << "\n");
1923       if (!DryRun)
1924         Token->insertBreak(LineIndex, TailOffset, Split, Whitespaces);
1925 
1926       Penalty += NewBreakPenalty;
1927       TailOffset += Split.first + Split.second;
1928       RemainingTokenColumns = NewRemainingTokenColumns;
1929       BreakInserted = true;
1930       NewBreakBefore = true;
1931     }
1932     // In case there's another line, prepare the state for the start of the next
1933     // line.
1934     if (LineIndex + 1 != EndIndex) {
1935       unsigned NextLineIndex = LineIndex + 1;
1936       if (NewBreakBefore)
1937         // After breaking a line, try to reflow the next line into the current
1938         // one once RemainingTokenColumns fits.
1939         TryReflow = true;
1940       if (TryReflow) {
1941         // We decided that we want to try reflowing the next line into the
1942         // current one.
1943         // We will now adjust the state as if the reflow is successful (in
1944         // preparation for the next line), and see whether that works. If we
1945         // decide that we cannot reflow, we will later reset the state to the
1946         // start of the next line.
1947         Reflow = false;
1948         // As we did not continue breaking the line, RemainingTokenColumns is
1949         // known to fit after ContentStartColumn. Adapt ContentStartColumn to
1950         // the position at which we want to format the next line if we do
1951         // actually reflow.
1952         // When we reflow, we need to add a space between the end of the current
1953         // line and the next line's start column.
1954         ContentStartColumn += RemainingTokenColumns + 1;
1955         // Get the split that we need to reflow next logical line into the end
1956         // of the current one; the split will include any leading whitespace of
1957         // the next logical line.
1958         BreakableToken::Split SplitBeforeNext =
1959             Token->getReflowSplit(NextLineIndex, CommentPragmasRegex);
1960         LLVM_DEBUG(llvm::dbgs()
1961                    << "    Size of reflown text: " << ContentStartColumn
1962                    << "\n    Potential reflow split: ");
1963         if (SplitBeforeNext.first != StringRef::npos) {
1964           LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", "
1965                                   << SplitBeforeNext.second << "\n");
1966           TailOffset = SplitBeforeNext.first + SplitBeforeNext.second;
1967           // If the rest of the next line fits into the current line below the
1968           // column limit, we can safely reflow.
1969           RemainingTokenColumns = Token->getRemainingLength(
1970               NextLineIndex, TailOffset, ContentStartColumn);
1971           Reflow = true;
1972           if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
1973             LLVM_DEBUG(llvm::dbgs()
1974                        << "    Over limit after reflow, need: "
1975                        << (ContentStartColumn + RemainingTokenColumns)
1976                        << ", space: " << ColumnLimit
1977                        << ", reflown prefix: " << ContentStartColumn
1978                        << ", offset in line: " << TailOffset << "\n");
1979             // If the whole next line does not fit, try to find a point in
1980             // the next line at which we can break so that attaching the part
1981             // of the next line to that break point onto the current line is
1982             // below the column limit.
1983             BreakableToken::Split Split =
1984                 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit,
1985                                 ContentStartColumn, CommentPragmasRegex);
1986             if (Split.first == StringRef::npos) {
1987               LLVM_DEBUG(llvm::dbgs() << "    Did not find later break\n");
1988               Reflow = false;
1989             } else {
1990               // Check whether the first split point gets us below the column
1991               // limit. Note that we will execute this split below as part of
1992               // the normal token breaking and reflow logic within the line.
1993               unsigned ToSplitColumns = Token->getRangeLength(
1994                   NextLineIndex, TailOffset, Split.first, ContentStartColumn);
1995               if (ContentStartColumn + ToSplitColumns > ColumnLimit) {
1996                 LLVM_DEBUG(llvm::dbgs() << "    Next split protrudes, need: "
1997                                         << (ContentStartColumn + ToSplitColumns)
1998                                         << ", space: " << ColumnLimit);
1999                 unsigned ExcessCharactersPenalty =
2000                     (ContentStartColumn + ToSplitColumns - ColumnLimit) *
2001                     Style.PenaltyExcessCharacter;
2002                 if (NewBreakPenalty < ExcessCharactersPenalty) {
2003                   Reflow = false;
2004                 }
2005               }
2006             }
2007           }
2008         } else {
2009           LLVM_DEBUG(llvm::dbgs() << "not found.\n");
2010         }
2011       }
2012       if (!Reflow) {
2013         // If we didn't reflow into the next line, the only space to consider is
2014         // the next logical line. Reset our state to match the start of the next
2015         // line.
2016         TailOffset = 0;
2017         ContentStartColumn =
2018             Token->getContentStartColumn(NextLineIndex, /*Break=*/false);
2019         RemainingTokenColumns = Token->getRemainingLength(
2020             NextLineIndex, TailOffset, ContentStartColumn);
2021         // Adapt the start of the token, for example indent.
2022         if (!DryRun)
2023           Token->adaptStartOfLine(NextLineIndex, Whitespaces);
2024       } else {
2025         // If we found a reflow split and have added a new break before the next
2026         // line, we are going to remove the line break at the start of the next
2027         // logical line. For example, here we'll add a new line break after
2028         // 'text', and subsequently delete the line break between 'that' and
2029         // 'reflows'.
2030         //   // some text that
2031         //   // reflows
2032         // ->
2033         //   // some text
2034         //   // that reflows
2035         // When adding the line break, we also added the penalty for it, so we
2036         // need to subtract that penalty again when we remove the line break due
2037         // to reflowing.
2038         if (NewBreakBefore) {
2039           assert(Penalty >= NewBreakPenalty);
2040           Penalty -= NewBreakPenalty;
2041         }
2042         if (!DryRun)
2043           Token->reflow(NextLineIndex, Whitespaces);
2044       }
2045     }
2046   }
2047 
2048   BreakableToken::Split SplitAfterLastLine =
2049       Token->getSplitAfterLastLine(TailOffset);
2050   if (SplitAfterLastLine.first != StringRef::npos) {
2051     LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n");
2052     if (!DryRun)
2053       Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine,
2054                                             Whitespaces);
2055     ContentStartColumn =
2056         Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true);
2057     RemainingTokenColumns = Token->getRemainingLength(
2058         Token->getLineCount() - 1,
2059         TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second,
2060         ContentStartColumn);
2061   }
2062 
2063   State.Column = ContentStartColumn + RemainingTokenColumns -
2064                  Current.UnbreakableTailLength;
2065 
2066   if (BreakInserted) {
2067     // If we break the token inside a parameter list, we need to break before
2068     // the next parameter on all levels, so that the next parameter is clearly
2069     // visible. Line comments already introduce a break.
2070     if (Current.isNot(TT_LineComment)) {
2071       for (unsigned i = 0, e = State.Stack.size(); i != e; ++i)
2072         State.Stack[i].BreakBeforeParameter = true;
2073     }
2074 
2075     if (Current.is(TT_BlockComment))
2076       State.NoContinuation = true;
2077 
2078     State.Stack.back().LastSpace = StartColumn;
2079   }
2080 
2081   Token->updateNextToken(State);
2082 
2083   return {Penalty, Exceeded};
2084 }
2085 
2086 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const {
2087   // In preprocessor directives reserve two chars for trailing " \"
2088   return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0);
2089 }
2090 
2091 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) {
2092   const FormatToken &Current = *State.NextToken;
2093   if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral))
2094     return false;
2095   // We never consider raw string literals "multiline" for the purpose of
2096   // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased
2097   // (see TokenAnnotator::mustBreakBefore().
2098   if (Current.TokenText.startswith("R\""))
2099     return false;
2100   if (Current.IsMultiline)
2101     return true;
2102   if (Current.getNextNonComment() &&
2103       Current.getNextNonComment()->isStringLiteral())
2104     return true; // Implicit concatenation.
2105   if (Style.ColumnLimit != 0 && Style.BreakStringLiterals &&
2106       State.Column + Current.ColumnWidth + Current.UnbreakableTailLength >
2107           Style.ColumnLimit)
2108     return true; // String will be split.
2109   return false;
2110 }
2111 
2112 } // namespace format
2113 } // namespace clang
2114