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