xref: /freebsd/contrib/llvm-project/clang/lib/Format/ContinuationIndenter.cpp (revision 1db9f3b21e39176dd5b67cf8ac378633b172463e)
1 //===--- ContinuationIndenter.cpp - Format C++ code -----------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 ///
9 /// \file
10 /// This file implements the continuation indenter.
11 ///
12 //===----------------------------------------------------------------------===//
13 
14 #include "ContinuationIndenter.h"
15 #include "BreakableToken.h"
16 #include "FormatInternal.h"
17 #include "FormatToken.h"
18 #include "WhitespaceManager.h"
19 #include "clang/Basic/OperatorPrecedence.h"
20 #include "clang/Basic/SourceManager.h"
21 #include "clang/Basic/TokenKinds.h"
22 #include "clang/Format/Format.h"
23 #include "llvm/ADT/StringSet.h"
24 #include "llvm/Support/Debug.h"
25 #include <optional>
26 
27 #define DEBUG_TYPE "format-indenter"
28 
29 namespace clang {
30 namespace format {
31 
32 // Returns true if a TT_SelectorName should be indented when wrapped,
33 // false otherwise.
34 static bool shouldIndentWrappedSelectorName(const FormatStyle &Style,
35                                             LineType LineType) {
36   return Style.IndentWrappedFunctionNames || LineType == LT_ObjCMethodDecl;
37 }
38 
39 // Returns true if a binary operator following \p Tok should be unindented when
40 // the style permits it.
41 static bool shouldUnindentNextOperator(const FormatToken &Tok) {
42   const FormatToken *Previous = Tok.getPreviousNonComment();
43   return Previous && (Previous->getPrecedence() == prec::Assignment ||
44                       Previous->isOneOf(tok::kw_return, TT_RequiresClause));
45 }
46 
47 // Returns the length of everything up to the first possible line break after
48 // the ), ], } or > matching \c Tok.
49 static unsigned getLengthToMatchingParen(const FormatToken &Tok,
50                                          ArrayRef<ParenState> Stack) {
51   // Normally whether or not a break before T is possible is calculated and
52   // stored in T.CanBreakBefore. Braces, array initializers and text proto
53   // messages like `key: < ... >` are an exception: a break is possible
54   // before a closing brace R if a break was inserted after the corresponding
55   // opening brace. The information about whether or not a break is needed
56   // before a closing brace R is stored in the ParenState field
57   // S.BreakBeforeClosingBrace where S is the state that R closes.
58   //
59   // In order to decide whether there can be a break before encountered right
60   // braces, this implementation iterates over the sequence of tokens and over
61   // the paren stack in lockstep, keeping track of the stack level which visited
62   // right braces correspond to in MatchingStackIndex.
63   //
64   // For example, consider:
65   // L. <- line number
66   // 1. {
67   // 2. {1},
68   // 3. {2},
69   // 4. {{3}}}
70   //     ^ where we call this method with this token.
71   // The paren stack at this point contains 3 brace levels:
72   //  0. { at line 1, BreakBeforeClosingBrace: true
73   //  1. first { at line 4, BreakBeforeClosingBrace: false
74   //  2. second { at line 4, BreakBeforeClosingBrace: false,
75   //  where there might be fake parens levels in-between these levels.
76   // The algorithm will start at the first } on line 4, which is the matching
77   // brace of the initial left brace and at level 2 of the stack. Then,
78   // examining BreakBeforeClosingBrace: false at level 2, it will continue to
79   // the second } on line 4, and will traverse the stack downwards until it
80   // finds the matching { on level 1. Then, examining BreakBeforeClosingBrace:
81   // false at level 1, it will continue to the third } on line 4 and will
82   // traverse the stack downwards until it finds the matching { on level 0.
83   // Then, examining BreakBeforeClosingBrace: true at level 0, the algorithm
84   // will stop and will use the second } on line 4 to determine the length to
85   // return, as in this example the range will include the tokens: {3}}
86   //
87   // The algorithm will only traverse the stack if it encounters braces, array
88   // initializer squares or text proto angle brackets.
89   if (!Tok.MatchingParen)
90     return 0;
91   FormatToken *End = Tok.MatchingParen;
92   // Maintains a stack level corresponding to the current End token.
93   int MatchingStackIndex = Stack.size() - 1;
94   // Traverses the stack downwards, looking for the level to which LBrace
95   // corresponds. Returns either a pointer to the matching level or nullptr if
96   // LParen is not found in the initial portion of the stack up to
97   // MatchingStackIndex.
98   auto FindParenState = [&](const FormatToken *LBrace) -> const ParenState * {
99     while (MatchingStackIndex >= 0 && Stack[MatchingStackIndex].Tok != LBrace)
100       --MatchingStackIndex;
101     return MatchingStackIndex >= 0 ? &Stack[MatchingStackIndex] : nullptr;
102   };
103   for (; End->Next; End = End->Next) {
104     if (End->Next->CanBreakBefore)
105       break;
106     if (!End->Next->closesScope())
107       continue;
108     if (End->Next->MatchingParen &&
109         End->Next->MatchingParen->isOneOf(
110             tok::l_brace, TT_ArrayInitializerLSquare, tok::less)) {
111       const ParenState *State = FindParenState(End->Next->MatchingParen);
112       if (State && State->BreakBeforeClosingBrace)
113         break;
114     }
115   }
116   return End->TotalLength - Tok.TotalLength + 1;
117 }
118 
119 static unsigned getLengthToNextOperator(const FormatToken &Tok) {
120   if (!Tok.NextOperator)
121     return 0;
122   return Tok.NextOperator->TotalLength - Tok.TotalLength;
123 }
124 
125 // Returns \c true if \c Tok is the "." or "->" of a call and starts the next
126 // segment of a builder type call.
127 static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) {
128   return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope();
129 }
130 
131 // Returns \c true if \c Current starts a new parameter.
132 static bool startsNextParameter(const FormatToken &Current,
133                                 const FormatStyle &Style) {
134   const FormatToken &Previous = *Current.Previous;
135   if (Current.is(TT_CtorInitializerComma) &&
136       Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) {
137     return true;
138   }
139   if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName))
140     return true;
141   return Previous.is(tok::comma) && !Current.isTrailingComment() &&
142          ((Previous.isNot(TT_CtorInitializerComma) ||
143            Style.BreakConstructorInitializers !=
144                FormatStyle::BCIS_BeforeComma) &&
145           (Previous.isNot(TT_InheritanceComma) ||
146            Style.BreakInheritanceList != FormatStyle::BILS_BeforeComma));
147 }
148 
149 static bool opensProtoMessageField(const FormatToken &LessTok,
150                                    const FormatStyle &Style) {
151   if (LessTok.isNot(tok::less))
152     return false;
153   return Style.Language == FormatStyle::LK_TextProto ||
154          (Style.Language == FormatStyle::LK_Proto &&
155           (LessTok.NestingLevel > 0 ||
156            (LessTok.Previous && LessTok.Previous->is(tok::equal))));
157 }
158 
159 // Returns the delimiter of a raw string literal, or std::nullopt if TokenText
160 // is not the text of a raw string literal. The delimiter could be the empty
161 // string.  For example, the delimiter of R"deli(cont)deli" is deli.
162 static std::optional<StringRef> getRawStringDelimiter(StringRef TokenText) {
163   if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'.
164       || !TokenText.starts_with("R\"") || !TokenText.ends_with("\"")) {
165     return std::nullopt;
166   }
167 
168   // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has
169   // size at most 16 by the standard, so the first '(' must be among the first
170   // 19 bytes.
171   size_t LParenPos = TokenText.substr(0, 19).find_first_of('(');
172   if (LParenPos == StringRef::npos)
173     return std::nullopt;
174   StringRef Delimiter = TokenText.substr(2, LParenPos - 2);
175 
176   // Check that the string ends in ')Delimiter"'.
177   size_t RParenPos = TokenText.size() - Delimiter.size() - 2;
178   if (TokenText[RParenPos] != ')')
179     return std::nullopt;
180   if (!TokenText.substr(RParenPos + 1).starts_with(Delimiter))
181     return std::nullopt;
182   return Delimiter;
183 }
184 
185 // Returns the canonical delimiter for \p Language, or the empty string if no
186 // canonical delimiter is specified.
187 static StringRef
188 getCanonicalRawStringDelimiter(const FormatStyle &Style,
189                                FormatStyle::LanguageKind Language) {
190   for (const auto &Format : Style.RawStringFormats)
191     if (Format.Language == Language)
192       return StringRef(Format.CanonicalDelimiter);
193   return "";
194 }
195 
196 RawStringFormatStyleManager::RawStringFormatStyleManager(
197     const FormatStyle &CodeStyle) {
198   for (const auto &RawStringFormat : CodeStyle.RawStringFormats) {
199     std::optional<FormatStyle> LanguageStyle =
200         CodeStyle.GetLanguageStyle(RawStringFormat.Language);
201     if (!LanguageStyle) {
202       FormatStyle PredefinedStyle;
203       if (!getPredefinedStyle(RawStringFormat.BasedOnStyle,
204                               RawStringFormat.Language, &PredefinedStyle)) {
205         PredefinedStyle = getLLVMStyle();
206         PredefinedStyle.Language = RawStringFormat.Language;
207       }
208       LanguageStyle = PredefinedStyle;
209     }
210     LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit;
211     for (StringRef Delimiter : RawStringFormat.Delimiters)
212       DelimiterStyle.insert({Delimiter, *LanguageStyle});
213     for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions)
214       EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle});
215   }
216 }
217 
218 std::optional<FormatStyle>
219 RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const {
220   auto It = DelimiterStyle.find(Delimiter);
221   if (It == DelimiterStyle.end())
222     return std::nullopt;
223   return It->second;
224 }
225 
226 std::optional<FormatStyle>
227 RawStringFormatStyleManager::getEnclosingFunctionStyle(
228     StringRef EnclosingFunction) const {
229   auto It = EnclosingFunctionStyle.find(EnclosingFunction);
230   if (It == EnclosingFunctionStyle.end())
231     return std::nullopt;
232   return It->second;
233 }
234 
235 ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style,
236                                            const AdditionalKeywords &Keywords,
237                                            const SourceManager &SourceMgr,
238                                            WhitespaceManager &Whitespaces,
239                                            encoding::Encoding Encoding,
240                                            bool BinPackInconclusiveFunctions)
241     : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr),
242       Whitespaces(Whitespaces), Encoding(Encoding),
243       BinPackInconclusiveFunctions(BinPackInconclusiveFunctions),
244       CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {}
245 
246 LineState ContinuationIndenter::getInitialState(unsigned FirstIndent,
247                                                 unsigned FirstStartColumn,
248                                                 const AnnotatedLine *Line,
249                                                 bool DryRun) {
250   LineState State;
251   State.FirstIndent = FirstIndent;
252   if (FirstStartColumn && Line->First->NewlinesBefore == 0)
253     State.Column = FirstStartColumn;
254   else
255     State.Column = FirstIndent;
256   // With preprocessor directive indentation, the line starts on column 0
257   // since it's indented after the hash, but FirstIndent is set to the
258   // preprocessor indent.
259   if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
260       (Line->Type == LT_PreprocessorDirective ||
261        Line->Type == LT_ImportStatement)) {
262     State.Column = 0;
263   }
264   State.Line = Line;
265   State.NextToken = Line->First;
266   State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent,
267                                    /*AvoidBinPacking=*/false,
268                                    /*NoLineBreak=*/false));
269   State.NoContinuation = false;
270   State.StartOfStringLiteral = 0;
271   State.NoLineBreak = false;
272   State.StartOfLineLevel = 0;
273   State.LowestLevelOnLine = 0;
274   State.IgnoreStackForComparison = false;
275 
276   if (Style.Language == FormatStyle::LK_TextProto) {
277     // We need this in order to deal with the bin packing of text fields at
278     // global scope.
279     auto &CurrentState = State.Stack.back();
280     CurrentState.AvoidBinPacking = true;
281     CurrentState.BreakBeforeParameter = true;
282     CurrentState.AlignColons = false;
283   }
284 
285   // The first token has already been indented and thus consumed.
286   moveStateToNextToken(State, DryRun, /*Newline=*/false);
287   return State;
288 }
289 
290 bool ContinuationIndenter::canBreak(const LineState &State) {
291   const FormatToken &Current = *State.NextToken;
292   const FormatToken &Previous = *Current.Previous;
293   const auto &CurrentState = State.Stack.back();
294   assert(&Previous == Current.Previous);
295   if (!Current.CanBreakBefore && !(CurrentState.BreakBeforeClosingBrace &&
296                                    Current.closesBlockOrBlockTypeList(Style))) {
297     return false;
298   }
299   // The opening "{" of a braced list has to be on the same line as the first
300   // element if it is nested in another braced init list or function call.
301   if (!Current.MustBreakBefore && Previous.is(tok::l_brace) &&
302       Previous.isNot(TT_DictLiteral) && Previous.is(BK_BracedInit) &&
303       Previous.Previous &&
304       Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma)) {
305     return false;
306   }
307   // This prevents breaks like:
308   //   ...
309   //   SomeParameter, OtherParameter).DoSomething(
310   //   ...
311   // As they hide "DoSomething" and are generally bad for readability.
312   if (Previous.opensScope() && Previous.isNot(tok::l_brace) &&
313       State.LowestLevelOnLine < State.StartOfLineLevel &&
314       State.LowestLevelOnLine < Current.NestingLevel) {
315     return false;
316   }
317   if (Current.isMemberAccess() && CurrentState.ContainsUnwrappedBuilder)
318     return false;
319 
320   // Don't create a 'hanging' indent if there are multiple blocks in a single
321   // statement and we are aligning lambda blocks to their signatures.
322   if (Previous.is(tok::l_brace) && State.Stack.size() > 1 &&
323       State.Stack[State.Stack.size() - 2].NestedBlockInlined &&
324       State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks &&
325       Style.LambdaBodyIndentation == FormatStyle::LBI_Signature) {
326     return false;
327   }
328 
329   // Don't break after very short return types (e.g. "void") as that is often
330   // unexpected.
331   if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) {
332     if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None)
333       return false;
334   }
335 
336   // If binary operators are moved to the next line (including commas for some
337   // styles of constructor initializers), that's always ok.
338   if (!Current.isOneOf(TT_BinaryOperator, tok::comma) &&
339       // Allow breaking opening brace of lambdas (when passed as function
340       // arguments) to a new line when BeforeLambdaBody brace wrapping is
341       // enabled.
342       (!Style.BraceWrapping.BeforeLambdaBody ||
343        Current.isNot(TT_LambdaLBrace)) &&
344       CurrentState.NoLineBreakInOperand) {
345     return false;
346   }
347 
348   if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr))
349     return false;
350 
351   if (Current.is(TT_ConditionalExpr) && Previous.is(tok::r_paren) &&
352       Previous.MatchingParen && Previous.MatchingParen->Previous &&
353       Previous.MatchingParen->Previous->MatchingParen &&
354       Previous.MatchingParen->Previous->MatchingParen->is(TT_LambdaLBrace)) {
355     // We have a lambda within a conditional expression, allow breaking here.
356     assert(Previous.MatchingParen->Previous->is(tok::r_brace));
357     return true;
358   }
359 
360   return !State.NoLineBreak && !CurrentState.NoLineBreak;
361 }
362 
363 bool ContinuationIndenter::mustBreak(const LineState &State) {
364   const FormatToken &Current = *State.NextToken;
365   const FormatToken &Previous = *Current.Previous;
366   const auto &CurrentState = State.Stack.back();
367   if (Style.BraceWrapping.BeforeLambdaBody && Current.CanBreakBefore &&
368       Current.is(TT_LambdaLBrace) && Previous.isNot(TT_LineComment)) {
369     auto LambdaBodyLength = getLengthToMatchingParen(Current, State.Stack);
370     return LambdaBodyLength > getColumnLimit(State);
371   }
372   if (Current.MustBreakBefore ||
373       (Current.is(TT_InlineASMColon) &&
374        (Style.BreakBeforeInlineASMColon == FormatStyle::BBIAS_Always ||
375         (Style.BreakBeforeInlineASMColon == FormatStyle::BBIAS_OnlyMultiline &&
376          Style.ColumnLimit > 0)))) {
377     return true;
378   }
379   if (CurrentState.BreakBeforeClosingBrace &&
380       (Current.closesBlockOrBlockTypeList(Style) ||
381        (Current.is(tok::r_brace) &&
382         Current.isBlockIndentedInitRBrace(Style)))) {
383     return true;
384   }
385   if (CurrentState.BreakBeforeClosingParen && Current.is(tok::r_paren))
386     return true;
387   if (Style.Language == FormatStyle::LK_ObjC &&
388       Style.ObjCBreakBeforeNestedBlockParam &&
389       Current.ObjCSelectorNameParts > 1 &&
390       Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) {
391     return true;
392   }
393   // Avoid producing inconsistent states by requiring breaks where they are not
394   // permitted for C# generic type constraints.
395   if (CurrentState.IsCSharpGenericTypeConstraint &&
396       Previous.isNot(TT_CSharpGenericTypeConstraintComma)) {
397     return false;
398   }
399   if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) ||
400        (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) &&
401         State.Line->First->isNot(TT_AttributeSquare) && Style.isCpp() &&
402         // FIXME: This is a temporary workaround for the case where clang-format
403         // sets BreakBeforeParameter to avoid bin packing and this creates a
404         // completely unnecessary line break after a template type that isn't
405         // line-wrapped.
406         (Previous.NestingLevel == 1 || Style.BinPackParameters)) ||
407        (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) &&
408         Previous.isNot(tok::question)) ||
409        (!Style.BreakBeforeTernaryOperators &&
410         Previous.is(TT_ConditionalExpr))) &&
411       CurrentState.BreakBeforeParameter && !Current.isTrailingComment() &&
412       !Current.isOneOf(tok::r_paren, tok::r_brace)) {
413     return true;
414   }
415   if (CurrentState.IsChainedConditional &&
416       ((Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) &&
417         Current.is(tok::colon)) ||
418        (!Style.BreakBeforeTernaryOperators && Previous.is(TT_ConditionalExpr) &&
419         Previous.is(tok::colon)))) {
420     return true;
421   }
422   if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) ||
423        (Previous.is(TT_ArrayInitializerLSquare) &&
424         Previous.ParameterCount > 1) ||
425        opensProtoMessageField(Previous, Style)) &&
426       Style.ColumnLimit > 0 &&
427       getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 >
428           getColumnLimit(State)) {
429     return true;
430   }
431 
432   const FormatToken &BreakConstructorInitializersToken =
433       Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon
434           ? Previous
435           : Current;
436   if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) &&
437       (State.Column + State.Line->Last->TotalLength - Previous.TotalLength >
438            getColumnLimit(State) ||
439        CurrentState.BreakBeforeParameter) &&
440       (!Current.isTrailingComment() || Current.NewlinesBefore > 0) &&
441       (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All ||
442        Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon ||
443        Style.ColumnLimit != 0)) {
444     return true;
445   }
446 
447   if (Current.is(TT_ObjCMethodExpr) && Previous.isNot(TT_SelectorName) &&
448       State.Line->startsWith(TT_ObjCMethodSpecifier)) {
449     return true;
450   }
451   if (Current.is(TT_SelectorName) && Previous.isNot(tok::at) &&
452       CurrentState.ObjCSelectorNameFound && CurrentState.BreakBeforeParameter &&
453       (Style.ObjCBreakBeforeNestedBlockParam ||
454        !Current.startsSequence(TT_SelectorName, tok::colon, tok::caret))) {
455     return true;
456   }
457 
458   unsigned NewLineColumn = getNewLineColumn(State);
459   if (Current.isMemberAccess() && Style.ColumnLimit != 0 &&
460       State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit &&
461       (State.Column > NewLineColumn ||
462        Current.NestingLevel < State.StartOfLineLevel)) {
463     return true;
464   }
465 
466   if (startsSegmentOfBuilderTypeCall(Current) &&
467       (CurrentState.CallContinuation != 0 ||
468        CurrentState.BreakBeforeParameter) &&
469       // JavaScript is treated different here as there is a frequent pattern:
470       //   SomeFunction(function() {
471       //     ...
472       //   }.bind(...));
473       // FIXME: We should find a more generic solution to this problem.
474       !(State.Column <= NewLineColumn && Style.isJavaScript()) &&
475       !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) {
476     return true;
477   }
478 
479   // If the template declaration spans multiple lines, force wrap before the
480   // function/class declaration.
481   if (Previous.ClosesTemplateDeclaration && CurrentState.BreakBeforeParameter &&
482       Current.CanBreakBefore) {
483     return true;
484   }
485 
486   if (State.Line->First->isNot(tok::kw_enum) && State.Column <= NewLineColumn)
487     return false;
488 
489   if (Style.AlwaysBreakBeforeMultilineStrings &&
490       (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth ||
491        Previous.is(tok::comma) || Current.NestingLevel < 2) &&
492       !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at,
493                         Keywords.kw_dollar) &&
494       !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) &&
495       nextIsMultilineString(State)) {
496     return true;
497   }
498 
499   // Using CanBreakBefore here and below takes care of the decision whether the
500   // current style uses wrapping before or after operators for the given
501   // operator.
502   if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) {
503     const auto PreviousPrecedence = Previous.getPrecedence();
504     if (PreviousPrecedence != prec::Assignment &&
505         CurrentState.BreakBeforeParameter && !Current.isTrailingComment()) {
506       const bool LHSIsBinaryExpr =
507           Previous.Previous && Previous.Previous->EndsBinaryExpression;
508       if (LHSIsBinaryExpr)
509         return true;
510       // If we need to break somewhere inside the LHS of a binary expression, we
511       // should also break after the operator. Otherwise, the formatting would
512       // hide the operator precedence, e.g. in:
513       //   if (aaaaaaaaaaaaaa ==
514       //           bbbbbbbbbbbbbb && c) {..
515       // For comparisons, we only apply this rule, if the LHS is a binary
516       // expression itself as otherwise, the line breaks seem superfluous.
517       // We need special cases for ">>" which we have split into two ">" while
518       // lexing in order to make template parsing easier.
519       const bool IsComparison =
520           (PreviousPrecedence == prec::Relational ||
521            PreviousPrecedence == prec::Equality ||
522            PreviousPrecedence == prec::Spaceship) &&
523           Previous.Previous &&
524           Previous.Previous->isNot(TT_BinaryOperator); // For >>.
525       if (!IsComparison)
526         return true;
527     }
528   } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore &&
529              CurrentState.BreakBeforeParameter) {
530     return true;
531   }
532 
533   // Same as above, but for the first "<<" operator.
534   if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) &&
535       CurrentState.BreakBeforeParameter && CurrentState.FirstLessLess == 0) {
536     return true;
537   }
538 
539   if (Current.NestingLevel == 0 && !Current.isTrailingComment()) {
540     // Always break after "template <...>"(*) and leading annotations. This is
541     // only for cases where the entire line does not fit on a single line as a
542     // different LineFormatter would be used otherwise.
543     // *: Except when another option interferes with that, like concepts.
544     if (Previous.ClosesTemplateDeclaration) {
545       if (Current.is(tok::kw_concept)) {
546         switch (Style.BreakBeforeConceptDeclarations) {
547         case FormatStyle::BBCDS_Allowed:
548           break;
549         case FormatStyle::BBCDS_Always:
550           return true;
551         case FormatStyle::BBCDS_Never:
552           return false;
553         }
554       }
555       if (Current.is(TT_RequiresClause)) {
556         switch (Style.RequiresClausePosition) {
557         case FormatStyle::RCPS_SingleLine:
558         case FormatStyle::RCPS_WithPreceding:
559           return false;
560         default:
561           return true;
562         }
563       }
564       return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No;
565     }
566     if (Previous.is(TT_FunctionAnnotationRParen) &&
567         State.Line->Type != LT_PreprocessorDirective) {
568       return true;
569     }
570     if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) &&
571         Current.isNot(TT_LeadingJavaAnnotation)) {
572       return true;
573     }
574   }
575 
576   if (Style.isJavaScript() && Previous.is(tok::r_paren) &&
577       Previous.is(TT_JavaAnnotation)) {
578     // Break after the closing parenthesis of TypeScript decorators before
579     // functions, getters and setters.
580     static const llvm::StringSet<> BreakBeforeDecoratedTokens = {"get", "set",
581                                                                  "function"};
582     if (BreakBeforeDecoratedTokens.contains(Current.TokenText))
583       return true;
584   }
585 
586   if (Current.is(TT_FunctionDeclarationName) &&
587       !State.Line->ReturnTypeWrapped &&
588       // Don't break before a C# function when no break after return type.
589       (!Style.isCSharp() ||
590        Style.AlwaysBreakAfterReturnType != FormatStyle::RTBS_None) &&
591       // Don't always break between a JavaScript `function` and the function
592       // name.
593       !Style.isJavaScript() && Previous.isNot(tok::kw_template) &&
594       CurrentState.BreakBeforeParameter) {
595     return true;
596   }
597 
598   // The following could be precomputed as they do not depend on the state.
599   // However, as they should take effect only if the UnwrappedLine does not fit
600   // into the ColumnLimit, they are checked here in the ContinuationIndenter.
601   if (Style.ColumnLimit != 0 && Previous.is(BK_Block) &&
602       Previous.is(tok::l_brace) &&
603       !Current.isOneOf(tok::r_brace, tok::comment)) {
604     return true;
605   }
606 
607   if (Current.is(tok::lessless) &&
608       ((Previous.is(tok::identifier) && Previous.TokenText == "endl") ||
609        (Previous.Tok.isLiteral() && (Previous.TokenText.ends_with("\\n\"") ||
610                                      Previous.TokenText == "\'\\n\'")))) {
611     return true;
612   }
613 
614   if (Previous.is(TT_BlockComment) && Previous.IsMultiline)
615     return true;
616 
617   if (State.NoContinuation)
618     return true;
619 
620   return false;
621 }
622 
623 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline,
624                                                bool DryRun,
625                                                unsigned ExtraSpaces) {
626   const FormatToken &Current = *State.NextToken;
627   assert(State.NextToken->Previous);
628   const FormatToken &Previous = *State.NextToken->Previous;
629 
630   assert(!State.Stack.empty());
631   State.NoContinuation = false;
632 
633   if (Current.is(TT_ImplicitStringLiteral) &&
634       (!Previous.Tok.getIdentifierInfo() ||
635        Previous.Tok.getIdentifierInfo()->getPPKeywordID() ==
636            tok::pp_not_keyword)) {
637     unsigned EndColumn =
638         SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd());
639     if (Current.LastNewlineOffset != 0) {
640       // If there is a newline within this token, the final column will solely
641       // determined by the current end column.
642       State.Column = EndColumn;
643     } else {
644       unsigned StartColumn =
645           SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin());
646       assert(EndColumn >= StartColumn);
647       State.Column += EndColumn - StartColumn;
648     }
649     moveStateToNextToken(State, DryRun, /*Newline=*/false);
650     return 0;
651   }
652 
653   unsigned Penalty = 0;
654   if (Newline)
655     Penalty = addTokenOnNewLine(State, DryRun);
656   else
657     addTokenOnCurrentLine(State, DryRun, ExtraSpaces);
658 
659   return moveStateToNextToken(State, DryRun, Newline) + Penalty;
660 }
661 
662 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
663                                                  unsigned ExtraSpaces) {
664   FormatToken &Current = *State.NextToken;
665   assert(State.NextToken->Previous);
666   const FormatToken &Previous = *State.NextToken->Previous;
667   auto &CurrentState = State.Stack.back();
668 
669   bool DisallowLineBreaksOnThisLine =
670       Style.LambdaBodyIndentation == FormatStyle::LBI_Signature &&
671       Style.isCpp() && [&Current] {
672         // Deal with lambda arguments in C++. The aim here is to ensure that we
673         // don't over-indent lambda function bodies when lambdas are passed as
674         // arguments to function calls. We do this by ensuring that either all
675         // arguments (including any lambdas) go on the same line as the function
676         // call, or we break before the first argument.
677         auto PrevNonComment = Current.getPreviousNonComment();
678         if (!PrevNonComment || PrevNonComment->isNot(tok::l_paren))
679           return false;
680         if (Current.isOneOf(tok::comment, tok::l_paren, TT_LambdaLSquare))
681           return false;
682         auto BlockParameterCount = PrevNonComment->BlockParameterCount;
683         if (BlockParameterCount == 0)
684           return false;
685 
686         // Multiple lambdas in the same function call.
687         if (BlockParameterCount > 1)
688           return true;
689 
690         // A lambda followed by another arg.
691         if (!PrevNonComment->Role)
692           return false;
693         auto Comma = PrevNonComment->Role->lastComma();
694         if (!Comma)
695           return false;
696         auto Next = Comma->getNextNonComment();
697         return Next &&
698                !Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret);
699       }();
700 
701   if (DisallowLineBreaksOnThisLine)
702     State.NoLineBreak = true;
703 
704   if (Current.is(tok::equal) &&
705       (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) &&
706       CurrentState.VariablePos == 0) {
707     CurrentState.VariablePos = State.Column;
708     // Move over * and & if they are bound to the variable name.
709     const FormatToken *Tok = &Previous;
710     while (Tok && CurrentState.VariablePos >= Tok->ColumnWidth) {
711       CurrentState.VariablePos -= Tok->ColumnWidth;
712       if (Tok->SpacesRequiredBefore != 0)
713         break;
714       Tok = Tok->Previous;
715     }
716     if (Previous.PartOfMultiVariableDeclStmt)
717       CurrentState.LastSpace = CurrentState.VariablePos;
718   }
719 
720   unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces;
721 
722   // Indent preprocessor directives after the hash if required.
723   int PPColumnCorrection = 0;
724   if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
725       Previous.is(tok::hash) && State.FirstIndent > 0 &&
726       &Previous == State.Line->First &&
727       (State.Line->Type == LT_PreprocessorDirective ||
728        State.Line->Type == LT_ImportStatement)) {
729     Spaces += State.FirstIndent;
730 
731     // For preprocessor indent with tabs, State.Column will be 1 because of the
732     // hash. This causes second-level indents onward to have an extra space
733     // after the tabs. We avoid this misalignment by subtracting 1 from the
734     // column value passed to replaceWhitespace().
735     if (Style.UseTab != FormatStyle::UT_Never)
736       PPColumnCorrection = -1;
737   }
738 
739   if (!DryRun) {
740     Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces,
741                                   State.Column + Spaces + PPColumnCorrection);
742   }
743 
744   // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance
745   // declaration unless there is multiple inheritance.
746   if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma &&
747       Current.is(TT_InheritanceColon)) {
748     CurrentState.NoLineBreak = true;
749   }
750   if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon &&
751       Previous.is(TT_InheritanceColon)) {
752     CurrentState.NoLineBreak = true;
753   }
754 
755   if (Current.is(TT_SelectorName) && !CurrentState.ObjCSelectorNameFound) {
756     unsigned MinIndent = std::max(
757         State.FirstIndent + Style.ContinuationIndentWidth, CurrentState.Indent);
758     unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth;
759     if (Current.LongestObjCSelectorName == 0)
760       CurrentState.AlignColons = false;
761     else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos)
762       CurrentState.ColonPos = MinIndent + Current.LongestObjCSelectorName;
763     else
764       CurrentState.ColonPos = FirstColonPos;
765   }
766 
767   // In "AlwaysBreak" or "BlockIndent" mode, enforce wrapping directly after the
768   // parenthesis by disallowing any further line breaks if there is no line
769   // break after the opening parenthesis. Don't break if it doesn't conserve
770   // columns.
771   if ((Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak ||
772        Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent) &&
773       (Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) ||
774        (Previous.is(tok::l_brace) && Previous.isNot(BK_Block) &&
775         Style.Cpp11BracedListStyle)) &&
776       State.Column > getNewLineColumn(State) &&
777       (!Previous.Previous ||
778        !Previous.Previous->isOneOf(TT_CastRParen, tok::kw_for, tok::kw_while,
779                                    tok::kw_switch)) &&
780       // Don't do this for simple (no expressions) one-argument function calls
781       // as that feels like needlessly wasting whitespace, e.g.:
782       //
783       //   caaaaaaaaaaaall(
784       //       caaaaaaaaaaaall(
785       //           caaaaaaaaaaaall(
786       //               caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa))));
787       Current.FakeLParens.size() > 0 &&
788       Current.FakeLParens.back() > prec::Unknown) {
789     CurrentState.NoLineBreak = true;
790   }
791   if (Previous.is(TT_TemplateString) && Previous.opensScope())
792     CurrentState.NoLineBreak = true;
793 
794   // Align following lines within parentheses / brackets if configured.
795   // Note: This doesn't apply to macro expansion lines, which are MACRO( , , )
796   // with args as children of the '(' and ',' tokens. It does not make sense to
797   // align the commas with the opening paren.
798   if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign &&
799       !CurrentState.IsCSharpGenericTypeConstraint && Previous.opensScope() &&
800       Previous.isNot(TT_ObjCMethodExpr) && Previous.isNot(TT_RequiresClause) &&
801       !(Current.MacroParent && Previous.MacroParent) &&
802       (Current.isNot(TT_LineComment) ||
803        Previous.isOneOf(BK_BracedInit, TT_VerilogMultiLineListLParen))) {
804     CurrentState.Indent = State.Column + Spaces;
805     CurrentState.IsAligned = true;
806   }
807   if (CurrentState.AvoidBinPacking && startsNextParameter(Current, Style))
808     CurrentState.NoLineBreak = true;
809   if (startsSegmentOfBuilderTypeCall(Current) &&
810       State.Column > getNewLineColumn(State)) {
811     CurrentState.ContainsUnwrappedBuilder = true;
812   }
813 
814   if (Current.is(TT_TrailingReturnArrow) &&
815       Style.Language == FormatStyle::LK_Java) {
816     CurrentState.NoLineBreak = true;
817   }
818   if (Current.isMemberAccess() && Previous.is(tok::r_paren) &&
819       (Previous.MatchingParen &&
820        (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) {
821     // If there is a function call with long parameters, break before trailing
822     // calls. This prevents things like:
823     //   EXPECT_CALL(SomeLongParameter).Times(
824     //       2);
825     // We don't want to do this for short parameters as they can just be
826     // indexes.
827     CurrentState.NoLineBreak = true;
828   }
829 
830   // Don't allow the RHS of an operator to be split over multiple lines unless
831   // there is a line-break right after the operator.
832   // Exclude relational operators, as there, it is always more desirable to
833   // have the LHS 'left' of the RHS.
834   const FormatToken *P = Current.getPreviousNonComment();
835   if (Current.isNot(tok::comment) && P &&
836       (P->isOneOf(TT_BinaryOperator, tok::comma) ||
837        (P->is(TT_ConditionalExpr) && P->is(tok::colon))) &&
838       !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) &&
839       P->getPrecedence() != prec::Assignment &&
840       P->getPrecedence() != prec::Relational &&
841       P->getPrecedence() != prec::Spaceship) {
842     bool BreakBeforeOperator =
843         P->MustBreakBefore || P->is(tok::lessless) ||
844         (P->is(TT_BinaryOperator) &&
845          Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) ||
846         (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators);
847     // Don't do this if there are only two operands. In these cases, there is
848     // always a nice vertical separation between them and the extra line break
849     // does not help.
850     bool HasTwoOperands = P->OperatorIndex == 0 && !P->NextOperator &&
851                           P->isNot(TT_ConditionalExpr);
852     if ((!BreakBeforeOperator &&
853          !(HasTwoOperands &&
854            Style.AlignOperands != FormatStyle::OAS_DontAlign)) ||
855         (!CurrentState.LastOperatorWrapped && BreakBeforeOperator)) {
856       CurrentState.NoLineBreakInOperand = true;
857     }
858   }
859 
860   State.Column += Spaces;
861   if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) &&
862       Previous.Previous &&
863       (Previous.Previous->is(tok::kw_for) || Previous.Previous->isIf())) {
864     // Treat the condition inside an if as if it was a second function
865     // parameter, i.e. let nested calls have a continuation indent.
866     CurrentState.LastSpace = State.Column;
867     CurrentState.NestedBlockIndent = State.Column;
868   } else if (!Current.isOneOf(tok::comment, tok::caret) &&
869              ((Previous.is(tok::comma) &&
870                Previous.isNot(TT_OverloadedOperator)) ||
871               (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) {
872     CurrentState.LastSpace = State.Column;
873   } else if (Previous.is(TT_CtorInitializerColon) &&
874              (!Current.isTrailingComment() || Current.NewlinesBefore > 0) &&
875              Style.BreakConstructorInitializers ==
876                  FormatStyle::BCIS_AfterColon) {
877     CurrentState.Indent = State.Column;
878     CurrentState.LastSpace = State.Column;
879   } else if (Previous.isOneOf(TT_ConditionalExpr, TT_CtorInitializerColon)) {
880     CurrentState.LastSpace = State.Column;
881   } else if (Previous.is(TT_BinaryOperator) &&
882              ((Previous.getPrecedence() != prec::Assignment &&
883                (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 ||
884                 Previous.NextOperator)) ||
885               Current.StartsBinaryExpression)) {
886     // Indent relative to the RHS of the expression unless this is a simple
887     // assignment without binary expression on the RHS.
888     if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None)
889       CurrentState.LastSpace = State.Column;
890   } else if (Previous.is(TT_InheritanceColon)) {
891     CurrentState.Indent = State.Column;
892     CurrentState.LastSpace = State.Column;
893   } else if (Current.is(TT_CSharpGenericTypeConstraintColon)) {
894     CurrentState.ColonPos = State.Column;
895   } else if (Previous.opensScope()) {
896     // If a function has a trailing call, indent all parameters from the
897     // opening parenthesis. This avoids confusing indents like:
898     //   OuterFunction(InnerFunctionCall( // break
899     //       ParameterToInnerFunction))   // break
900     //       .SecondInnerFunctionCall();
901     if (Previous.MatchingParen) {
902       const FormatToken *Next = Previous.MatchingParen->getNextNonComment();
903       if (Next && Next->isMemberAccess() && State.Stack.size() > 1 &&
904           State.Stack[State.Stack.size() - 2].CallContinuation == 0) {
905         CurrentState.LastSpace = State.Column;
906       }
907     }
908   }
909 }
910 
911 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State,
912                                                  bool DryRun) {
913   FormatToken &Current = *State.NextToken;
914   assert(State.NextToken->Previous);
915   const FormatToken &Previous = *State.NextToken->Previous;
916   auto &CurrentState = State.Stack.back();
917 
918   // Extra penalty that needs to be added because of the way certain line
919   // breaks are chosen.
920   unsigned Penalty = 0;
921 
922   const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
923   const FormatToken *NextNonComment = Previous.getNextNonComment();
924   if (!NextNonComment)
925     NextNonComment = &Current;
926   // The first line break on any NestingLevel causes an extra penalty in order
927   // prefer similar line breaks.
928   if (!CurrentState.ContainsLineBreak)
929     Penalty += 15;
930   CurrentState.ContainsLineBreak = true;
931 
932   Penalty += State.NextToken->SplitPenalty;
933 
934   // Breaking before the first "<<" is generally not desirable if the LHS is
935   // short. Also always add the penalty if the LHS is split over multiple lines
936   // to avoid unnecessary line breaks that just work around this penalty.
937   if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess == 0 &&
938       (State.Column <= Style.ColumnLimit / 3 ||
939        CurrentState.BreakBeforeParameter)) {
940     Penalty += Style.PenaltyBreakFirstLessLess;
941   }
942 
943   State.Column = getNewLineColumn(State);
944 
945   // Add Penalty proportional to amount of whitespace away from FirstColumn
946   // This tends to penalize several lines that are far-right indented,
947   // and prefers a line-break prior to such a block, e.g:
948   //
949   // Constructor() :
950   //   member(value), looooooooooooooooong_member(
951   //                      looooooooooong_call(param_1, param_2, param_3))
952   // would then become
953   // Constructor() :
954   //   member(value),
955   //   looooooooooooooooong_member(
956   //       looooooooooong_call(param_1, param_2, param_3))
957   if (State.Column > State.FirstIndent) {
958     Penalty +=
959         Style.PenaltyIndentedWhitespace * (State.Column - State.FirstIndent);
960   }
961 
962   // Indent nested blocks relative to this column, unless in a very specific
963   // JavaScript special case where:
964   //
965   //   var loooooong_name =
966   //       function() {
967   //     // code
968   //   }
969   //
970   // is common and should be formatted like a free-standing function. The same
971   // goes for wrapping before the lambda return type arrow.
972   if (Current.isNot(TT_TrailingReturnArrow) &&
973       (!Style.isJavaScript() || Current.NestingLevel != 0 ||
974        !PreviousNonComment || PreviousNonComment->isNot(tok::equal) ||
975        !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) {
976     CurrentState.NestedBlockIndent = State.Column;
977   }
978 
979   if (NextNonComment->isMemberAccess()) {
980     if (CurrentState.CallContinuation == 0)
981       CurrentState.CallContinuation = State.Column;
982   } else if (NextNonComment->is(TT_SelectorName)) {
983     if (!CurrentState.ObjCSelectorNameFound) {
984       if (NextNonComment->LongestObjCSelectorName == 0) {
985         CurrentState.AlignColons = false;
986       } else {
987         CurrentState.ColonPos =
988             (shouldIndentWrappedSelectorName(Style, State.Line->Type)
989                  ? std::max(CurrentState.Indent,
990                             State.FirstIndent + Style.ContinuationIndentWidth)
991                  : CurrentState.Indent) +
992             std::max(NextNonComment->LongestObjCSelectorName,
993                      NextNonComment->ColumnWidth);
994       }
995     } else if (CurrentState.AlignColons &&
996                CurrentState.ColonPos <= NextNonComment->ColumnWidth) {
997       CurrentState.ColonPos = State.Column + NextNonComment->ColumnWidth;
998     }
999   } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
1000              PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) {
1001     // FIXME: This is hacky, find a better way. The problem is that in an ObjC
1002     // method expression, the block should be aligned to the line starting it,
1003     // e.g.:
1004     //   [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason
1005     //                        ^(int *i) {
1006     //                            // ...
1007     //                        }];
1008     // Thus, we set LastSpace of the next higher NestingLevel, to which we move
1009     // when we consume all of the "}"'s FakeRParens at the "{".
1010     if (State.Stack.size() > 1) {
1011       State.Stack[State.Stack.size() - 2].LastSpace =
1012           std::max(CurrentState.LastSpace, CurrentState.Indent) +
1013           Style.ContinuationIndentWidth;
1014     }
1015   }
1016 
1017   if ((PreviousNonComment &&
1018        PreviousNonComment->isOneOf(tok::comma, tok::semi) &&
1019        !CurrentState.AvoidBinPacking) ||
1020       Previous.is(TT_BinaryOperator)) {
1021     CurrentState.BreakBeforeParameter = false;
1022   }
1023   if (PreviousNonComment &&
1024       (PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) ||
1025        PreviousNonComment->ClosesRequiresClause) &&
1026       Current.NestingLevel == 0) {
1027     CurrentState.BreakBeforeParameter = false;
1028   }
1029   if (NextNonComment->is(tok::question) ||
1030       (PreviousNonComment && PreviousNonComment->is(tok::question))) {
1031     CurrentState.BreakBeforeParameter = true;
1032   }
1033   if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore)
1034     CurrentState.BreakBeforeParameter = false;
1035 
1036   if (!DryRun) {
1037     unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1;
1038     if (Current.is(tok::r_brace) && Current.MatchingParen &&
1039         // Only strip trailing empty lines for l_braces that have children, i.e.
1040         // for function expressions (lambdas, arrows, etc).
1041         !Current.MatchingParen->Children.empty()) {
1042       // lambdas and arrow functions are expressions, thus their r_brace is not
1043       // on its own line, and thus not covered by UnwrappedLineFormatter's logic
1044       // about removing empty lines on closing blocks. Special case them here.
1045       MaxEmptyLinesToKeep = 1;
1046     }
1047     unsigned Newlines =
1048         std::max(1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep));
1049     bool ContinuePPDirective =
1050         State.Line->InPPDirective && State.Line->Type != LT_ImportStatement;
1051     Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column,
1052                                   CurrentState.IsAligned, ContinuePPDirective);
1053   }
1054 
1055   if (!Current.isTrailingComment())
1056     CurrentState.LastSpace = State.Column;
1057   if (Current.is(tok::lessless)) {
1058     // If we are breaking before a "<<", we always want to indent relative to
1059     // RHS. This is necessary only for "<<", as we special-case it and don't
1060     // always indent relative to the RHS.
1061     CurrentState.LastSpace += 3; // 3 -> width of "<< ".
1062   }
1063 
1064   State.StartOfLineLevel = Current.NestingLevel;
1065   State.LowestLevelOnLine = Current.NestingLevel;
1066 
1067   // Any break on this level means that the parent level has been broken
1068   // and we need to avoid bin packing there.
1069   bool NestedBlockSpecialCase =
1070       (!Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 &&
1071        State.Stack[State.Stack.size() - 2].NestedBlockInlined) ||
1072       (Style.Language == FormatStyle::LK_ObjC && Current.is(tok::r_brace) &&
1073        State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam);
1074   // Do not force parameter break for statements with requires expressions.
1075   NestedBlockSpecialCase =
1076       NestedBlockSpecialCase ||
1077       (Current.MatchingParen &&
1078        Current.MatchingParen->is(TT_RequiresExpressionLBrace));
1079   if (!NestedBlockSpecialCase) {
1080     auto ParentLevelIt = std::next(State.Stack.rbegin());
1081     if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
1082         Current.MatchingParen && Current.MatchingParen->is(TT_LambdaLBrace)) {
1083       // If the first character on the new line is a lambda's closing brace, the
1084       // stack still contains that lambda's parenthesis. As such, we need to
1085       // recurse further down the stack than usual to find the parenthesis level
1086       // containing the lambda, which is where we want to set
1087       // BreakBeforeParameter.
1088       //
1089       // We specifically special case "OuterScope"-formatted lambdas here
1090       // because, when using that setting, breaking before the parameter
1091       // directly following the lambda is particularly unsightly. However, when
1092       // "OuterScope" is not set, the logic to find the parent parenthesis level
1093       // still appears to be sometimes incorrect. It has not been fixed yet
1094       // because it would lead to significant changes in existing behaviour.
1095       //
1096       // TODO: fix the non-"OuterScope" case too.
1097       auto FindCurrentLevel = [&](const auto &It) {
1098         return std::find_if(It, State.Stack.rend(), [](const auto &PState) {
1099           return PState.Tok != nullptr; // Ignore fake parens.
1100         });
1101       };
1102       auto MaybeIncrement = [&](const auto &It) {
1103         return It != State.Stack.rend() ? std::next(It) : It;
1104       };
1105       auto LambdaLevelIt = FindCurrentLevel(State.Stack.rbegin());
1106       auto LevelContainingLambdaIt =
1107           FindCurrentLevel(MaybeIncrement(LambdaLevelIt));
1108       ParentLevelIt = MaybeIncrement(LevelContainingLambdaIt);
1109     }
1110     for (auto I = ParentLevelIt, E = State.Stack.rend(); I != E; ++I)
1111       I->BreakBeforeParameter = true;
1112   }
1113 
1114   if (PreviousNonComment &&
1115       !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) &&
1116       ((PreviousNonComment->isNot(TT_TemplateCloser) &&
1117         !PreviousNonComment->ClosesRequiresClause) ||
1118        Current.NestingLevel != 0) &&
1119       !PreviousNonComment->isOneOf(
1120           TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation,
1121           TT_LeadingJavaAnnotation) &&
1122       Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope() &&
1123       // We don't want to enforce line breaks for subsequent arguments just
1124       // because we have been forced to break before a lambda body.
1125       (!Style.BraceWrapping.BeforeLambdaBody ||
1126        Current.isNot(TT_LambdaLBrace))) {
1127     CurrentState.BreakBeforeParameter = true;
1128   }
1129 
1130   // If we break after { or the [ of an array initializer, we should also break
1131   // before the corresponding } or ].
1132   if (PreviousNonComment &&
1133       (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
1134        opensProtoMessageField(*PreviousNonComment, Style))) {
1135     CurrentState.BreakBeforeClosingBrace = true;
1136   }
1137 
1138   if (PreviousNonComment && PreviousNonComment->is(tok::l_paren)) {
1139     CurrentState.BreakBeforeClosingParen =
1140         Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent;
1141   }
1142 
1143   if (CurrentState.AvoidBinPacking) {
1144     // If we are breaking after '(', '{', '<', or this is the break after a ':'
1145     // to start a member initializer list in a constructor, this should not
1146     // be considered bin packing unless the relevant AllowAll option is false or
1147     // this is a dict/object literal.
1148     bool PreviousIsBreakingCtorInitializerColon =
1149         PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) &&
1150         Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon;
1151     bool AllowAllConstructorInitializersOnNextLine =
1152         Style.PackConstructorInitializers == FormatStyle::PCIS_NextLine ||
1153         Style.PackConstructorInitializers == FormatStyle::PCIS_NextLineOnly;
1154     if (!(Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) ||
1155           PreviousIsBreakingCtorInitializerColon) ||
1156         (!Style.AllowAllParametersOfDeclarationOnNextLine &&
1157          State.Line->MustBeDeclaration) ||
1158         (!Style.AllowAllArgumentsOnNextLine &&
1159          !State.Line->MustBeDeclaration) ||
1160         (!AllowAllConstructorInitializersOnNextLine &&
1161          PreviousIsBreakingCtorInitializerColon) ||
1162         Previous.is(TT_DictLiteral)) {
1163       CurrentState.BreakBeforeParameter = true;
1164     }
1165 
1166     // If we are breaking after a ':' to start a member initializer list,
1167     // and we allow all arguments on the next line, we should not break
1168     // before the next parameter.
1169     if (PreviousIsBreakingCtorInitializerColon &&
1170         AllowAllConstructorInitializersOnNextLine) {
1171       CurrentState.BreakBeforeParameter = false;
1172     }
1173   }
1174 
1175   return Penalty;
1176 }
1177 
1178 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
1179   if (!State.NextToken || !State.NextToken->Previous)
1180     return 0;
1181 
1182   FormatToken &Current = *State.NextToken;
1183   const auto &CurrentState = State.Stack.back();
1184 
1185   if (CurrentState.IsCSharpGenericTypeConstraint &&
1186       Current.isNot(TT_CSharpGenericTypeConstraint)) {
1187     return CurrentState.ColonPos + 2;
1188   }
1189 
1190   const FormatToken &Previous = *Current.Previous;
1191   // If we are continuing an expression, we want to use the continuation indent.
1192   unsigned ContinuationIndent =
1193       std::max(CurrentState.LastSpace, CurrentState.Indent) +
1194       Style.ContinuationIndentWidth;
1195   const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
1196   const FormatToken *NextNonComment = Previous.getNextNonComment();
1197   if (!NextNonComment)
1198     NextNonComment = &Current;
1199 
1200   // Java specific bits.
1201   if (Style.Language == FormatStyle::LK_Java &&
1202       Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) {
1203     return std::max(CurrentState.LastSpace,
1204                     CurrentState.Indent + Style.ContinuationIndentWidth);
1205   }
1206 
1207   // Indentation of the statement following a Verilog case label is taken care
1208   // of in moveStateToNextToken.
1209   if (Style.isVerilog() && PreviousNonComment &&
1210       Keywords.isVerilogEndOfLabel(*PreviousNonComment)) {
1211     return State.FirstIndent;
1212   }
1213 
1214   if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths &&
1215       State.Line->First->is(tok::kw_enum)) {
1216     return (Style.IndentWidth * State.Line->First->IndentLevel) +
1217            Style.IndentWidth;
1218   }
1219 
1220   if ((NextNonComment->is(tok::l_brace) && NextNonComment->is(BK_Block)) ||
1221       (Style.isVerilog() && Keywords.isVerilogBegin(*NextNonComment))) {
1222     if (Current.NestingLevel == 0 ||
1223         (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
1224          State.NextToken->is(TT_LambdaLBrace))) {
1225       return State.FirstIndent;
1226     }
1227     return CurrentState.Indent;
1228   }
1229   if ((Current.isOneOf(tok::r_brace, tok::r_square) ||
1230        (Current.is(tok::greater) && Style.isProto())) &&
1231       State.Stack.size() > 1) {
1232     if (Current.closesBlockOrBlockTypeList(Style))
1233       return State.Stack[State.Stack.size() - 2].NestedBlockIndent;
1234     if (Current.MatchingParen && Current.MatchingParen->is(BK_BracedInit))
1235       return State.Stack[State.Stack.size() - 2].LastSpace;
1236     return State.FirstIndent;
1237   }
1238   // Indent a closing parenthesis at the previous level if followed by a semi,
1239   // const, or opening brace. This allows indentations such as:
1240   //     foo(
1241   //       a,
1242   //     );
1243   //     int Foo::getter(
1244   //         //
1245   //     ) const {
1246   //       return foo;
1247   //     }
1248   //     function foo(
1249   //       a,
1250   //     ) {
1251   //       code(); //
1252   //     }
1253   if (Current.is(tok::r_paren) && State.Stack.size() > 1 &&
1254       (!Current.Next ||
1255        Current.Next->isOneOf(tok::semi, tok::kw_const, tok::l_brace))) {
1256     return State.Stack[State.Stack.size() - 2].LastSpace;
1257   }
1258   if (Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent &&
1259       (Current.is(tok::r_paren) ||
1260        (Current.is(tok::r_brace) && Current.MatchingParen &&
1261         Current.MatchingParen->is(BK_BracedInit))) &&
1262       State.Stack.size() > 1) {
1263     return State.Stack[State.Stack.size() - 2].LastSpace;
1264   }
1265   if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope())
1266     return State.Stack[State.Stack.size() - 2].LastSpace;
1267   // Field labels in a nested type should be aligned to the brace. For example
1268   // in ProtoBuf:
1269   //   optional int32 b = 2 [(foo_options) = {aaaaaaaaaaaaaaaaaaa: 123,
1270   //                                          bbbbbbbbbbbbbbbbbbbbbbbb:"baz"}];
1271   // For Verilog, a quote following a brace is treated as an identifier.  And
1272   // Both braces and colons get annotated as TT_DictLiteral.  So we have to
1273   // check.
1274   if (Current.is(tok::identifier) && Current.Next &&
1275       (!Style.isVerilog() || Current.Next->is(tok::colon)) &&
1276       (Current.Next->is(TT_DictLiteral) ||
1277        (Style.isProto() && Current.Next->isOneOf(tok::less, tok::l_brace)))) {
1278     return CurrentState.Indent;
1279   }
1280   if (NextNonComment->is(TT_ObjCStringLiteral) &&
1281       State.StartOfStringLiteral != 0) {
1282     return State.StartOfStringLiteral - 1;
1283   }
1284   if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0)
1285     return State.StartOfStringLiteral;
1286   if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess != 0)
1287     return CurrentState.FirstLessLess;
1288   if (NextNonComment->isMemberAccess()) {
1289     if (CurrentState.CallContinuation == 0)
1290       return ContinuationIndent;
1291     return CurrentState.CallContinuation;
1292   }
1293   if (CurrentState.QuestionColumn != 0 &&
1294       ((NextNonComment->is(tok::colon) &&
1295         NextNonComment->is(TT_ConditionalExpr)) ||
1296        Previous.is(TT_ConditionalExpr))) {
1297     if (((NextNonComment->is(tok::colon) && NextNonComment->Next &&
1298           !NextNonComment->Next->FakeLParens.empty() &&
1299           NextNonComment->Next->FakeLParens.back() == prec::Conditional) ||
1300          (Previous.is(tok::colon) && !Current.FakeLParens.empty() &&
1301           Current.FakeLParens.back() == prec::Conditional)) &&
1302         !CurrentState.IsWrappedConditional) {
1303       // NOTE: we may tweak this slightly:
1304       //    * not remove the 'lead' ContinuationIndentWidth
1305       //    * always un-indent by the operator when
1306       //    BreakBeforeTernaryOperators=true
1307       unsigned Indent = CurrentState.Indent;
1308       if (Style.AlignOperands != FormatStyle::OAS_DontAlign)
1309         Indent -= Style.ContinuationIndentWidth;
1310       if (Style.BreakBeforeTernaryOperators && CurrentState.UnindentOperator)
1311         Indent -= 2;
1312       return Indent;
1313     }
1314     return CurrentState.QuestionColumn;
1315   }
1316   if (Previous.is(tok::comma) && CurrentState.VariablePos != 0)
1317     return CurrentState.VariablePos;
1318   if (Current.is(TT_RequiresClause)) {
1319     if (Style.IndentRequiresClause)
1320       return CurrentState.Indent + Style.IndentWidth;
1321     switch (Style.RequiresClausePosition) {
1322     case FormatStyle::RCPS_OwnLine:
1323     case FormatStyle::RCPS_WithFollowing:
1324       return CurrentState.Indent;
1325     default:
1326       break;
1327     }
1328   }
1329   if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon,
1330                               TT_InheritanceComma)) {
1331     return State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1332   }
1333   if ((PreviousNonComment &&
1334        (PreviousNonComment->ClosesTemplateDeclaration ||
1335         PreviousNonComment->ClosesRequiresClause ||
1336         (PreviousNonComment->is(TT_AttributeMacro) &&
1337          Current.isNot(tok::l_paren)) ||
1338         PreviousNonComment->isOneOf(
1339             TT_AttributeRParen, TT_AttributeSquare, TT_FunctionAnnotationRParen,
1340             TT_JavaAnnotation, TT_LeadingJavaAnnotation))) ||
1341       (!Style.IndentWrappedFunctionNames &&
1342        NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) {
1343     return std::max(CurrentState.LastSpace, CurrentState.Indent);
1344   }
1345   if (NextNonComment->is(TT_SelectorName)) {
1346     if (!CurrentState.ObjCSelectorNameFound) {
1347       unsigned MinIndent = CurrentState.Indent;
1348       if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) {
1349         MinIndent = std::max(MinIndent,
1350                              State.FirstIndent + Style.ContinuationIndentWidth);
1351       }
1352       // If LongestObjCSelectorName is 0, we are indenting the first
1353       // part of an ObjC selector (or a selector component which is
1354       // not colon-aligned due to block formatting).
1355       //
1356       // Otherwise, we are indenting a subsequent part of an ObjC
1357       // selector which should be colon-aligned to the longest
1358       // component of the ObjC selector.
1359       //
1360       // In either case, we want to respect Style.IndentWrappedFunctionNames.
1361       return MinIndent +
1362              std::max(NextNonComment->LongestObjCSelectorName,
1363                       NextNonComment->ColumnWidth) -
1364              NextNonComment->ColumnWidth;
1365     }
1366     if (!CurrentState.AlignColons)
1367       return CurrentState.Indent;
1368     if (CurrentState.ColonPos > NextNonComment->ColumnWidth)
1369       return CurrentState.ColonPos - NextNonComment->ColumnWidth;
1370     return CurrentState.Indent;
1371   }
1372   if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr))
1373     return CurrentState.ColonPos;
1374   if (NextNonComment->is(TT_ArraySubscriptLSquare)) {
1375     if (CurrentState.StartOfArraySubscripts != 0) {
1376       return CurrentState.StartOfArraySubscripts;
1377     } else if (Style.isCSharp()) { // C# allows `["key"] = value` inside object
1378                                    // initializers.
1379       return CurrentState.Indent;
1380     }
1381     return ContinuationIndent;
1382   }
1383 
1384   // OpenMP clauses want to get additional indentation when they are pushed onto
1385   // the next line.
1386   if (State.Line->InPragmaDirective) {
1387     FormatToken *PragmaType = State.Line->First->Next->Next;
1388     if (PragmaType && PragmaType->TokenText.equals("omp"))
1389       return CurrentState.Indent + Style.ContinuationIndentWidth;
1390   }
1391 
1392   // This ensure that we correctly format ObjC methods calls without inputs,
1393   // i.e. where the last element isn't selector like: [callee method];
1394   if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 &&
1395       NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) {
1396     return CurrentState.Indent;
1397   }
1398 
1399   if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) ||
1400       Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) {
1401     return ContinuationIndent;
1402   }
1403   if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
1404       PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) {
1405     return ContinuationIndent;
1406   }
1407   if (NextNonComment->is(TT_CtorInitializerComma))
1408     return CurrentState.Indent;
1409   if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) &&
1410       Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1411     return CurrentState.Indent;
1412   }
1413   if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) &&
1414       Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) {
1415     return CurrentState.Indent;
1416   }
1417   if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() &&
1418       !Current.isOneOf(tok::colon, tok::comment)) {
1419     return ContinuationIndent;
1420   }
1421   if (Current.is(TT_ProtoExtensionLSquare))
1422     return CurrentState.Indent;
1423   if (Current.isBinaryOperator() && CurrentState.UnindentOperator) {
1424     return CurrentState.Indent - Current.Tok.getLength() -
1425            Current.SpacesRequiredBefore;
1426   }
1427   if (Current.is(tok::comment) && NextNonComment->isBinaryOperator() &&
1428       CurrentState.UnindentOperator) {
1429     return CurrentState.Indent - NextNonComment->Tok.getLength() -
1430            NextNonComment->SpacesRequiredBefore;
1431   }
1432   if (CurrentState.Indent == State.FirstIndent && PreviousNonComment &&
1433       !PreviousNonComment->isOneOf(tok::r_brace, TT_CtorInitializerComma)) {
1434     // Ensure that we fall back to the continuation indent width instead of
1435     // just flushing continuations left.
1436     return CurrentState.Indent + Style.ContinuationIndentWidth;
1437   }
1438   return CurrentState.Indent;
1439 }
1440 
1441 static bool hasNestedBlockInlined(const FormatToken *Previous,
1442                                   const FormatToken &Current,
1443                                   const FormatStyle &Style) {
1444   if (Previous->isNot(tok::l_paren))
1445     return true;
1446   if (Previous->ParameterCount > 1)
1447     return true;
1448 
1449   // Also a nested block if contains a lambda inside function with 1 parameter.
1450   return Style.BraceWrapping.BeforeLambdaBody && Current.is(TT_LambdaLSquare);
1451 }
1452 
1453 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State,
1454                                                     bool DryRun, bool Newline) {
1455   assert(State.Stack.size());
1456   const FormatToken &Current = *State.NextToken;
1457   auto &CurrentState = State.Stack.back();
1458 
1459   if (Current.is(TT_CSharpGenericTypeConstraint))
1460     CurrentState.IsCSharpGenericTypeConstraint = true;
1461   if (Current.isOneOf(tok::comma, TT_BinaryOperator))
1462     CurrentState.NoLineBreakInOperand = false;
1463   if (Current.isOneOf(TT_InheritanceColon, TT_CSharpGenericTypeConstraintColon))
1464     CurrentState.AvoidBinPacking = true;
1465   if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) {
1466     if (CurrentState.FirstLessLess == 0)
1467       CurrentState.FirstLessLess = State.Column;
1468     else
1469       CurrentState.LastOperatorWrapped = Newline;
1470   }
1471   if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless))
1472     CurrentState.LastOperatorWrapped = Newline;
1473   if (Current.is(TT_ConditionalExpr) && Current.Previous &&
1474       Current.Previous->isNot(TT_ConditionalExpr)) {
1475     CurrentState.LastOperatorWrapped = Newline;
1476   }
1477   if (Current.is(TT_ArraySubscriptLSquare) &&
1478       CurrentState.StartOfArraySubscripts == 0) {
1479     CurrentState.StartOfArraySubscripts = State.Column;
1480   }
1481 
1482   auto IsWrappedConditional = [](const FormatToken &Tok) {
1483     if (!(Tok.is(TT_ConditionalExpr) && Tok.is(tok::question)))
1484       return false;
1485     if (Tok.MustBreakBefore)
1486       return true;
1487 
1488     const FormatToken *Next = Tok.getNextNonComment();
1489     return Next && Next->MustBreakBefore;
1490   };
1491   if (IsWrappedConditional(Current))
1492     CurrentState.IsWrappedConditional = true;
1493   if (Style.BreakBeforeTernaryOperators && Current.is(tok::question))
1494     CurrentState.QuestionColumn = State.Column;
1495   if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) {
1496     const FormatToken *Previous = Current.Previous;
1497     while (Previous && Previous->isTrailingComment())
1498       Previous = Previous->Previous;
1499     if (Previous && Previous->is(tok::question))
1500       CurrentState.QuestionColumn = State.Column;
1501   }
1502   if (!Current.opensScope() && !Current.closesScope() &&
1503       Current.isNot(TT_PointerOrReference)) {
1504     State.LowestLevelOnLine =
1505         std::min(State.LowestLevelOnLine, Current.NestingLevel);
1506   }
1507   if (Current.isMemberAccess())
1508     CurrentState.StartOfFunctionCall = !Current.NextOperator ? 0 : State.Column;
1509   if (Current.is(TT_SelectorName))
1510     CurrentState.ObjCSelectorNameFound = true;
1511   if (Current.is(TT_CtorInitializerColon) &&
1512       Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) {
1513     // Indent 2 from the column, so:
1514     // SomeClass::SomeClass()
1515     //     : First(...), ...
1516     //       Next(...)
1517     //       ^ line up here.
1518     CurrentState.Indent = State.Column + (Style.BreakConstructorInitializers ==
1519                                                   FormatStyle::BCIS_BeforeComma
1520                                               ? 0
1521                                               : 2);
1522     CurrentState.NestedBlockIndent = CurrentState.Indent;
1523     if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) {
1524       CurrentState.AvoidBinPacking = true;
1525       CurrentState.BreakBeforeParameter =
1526           Style.ColumnLimit > 0 &&
1527           Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine &&
1528           Style.PackConstructorInitializers != FormatStyle::PCIS_NextLineOnly;
1529     } else {
1530       CurrentState.BreakBeforeParameter = false;
1531     }
1532   }
1533   if (Current.is(TT_CtorInitializerColon) &&
1534       Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1535     CurrentState.Indent =
1536         State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1537     CurrentState.NestedBlockIndent = CurrentState.Indent;
1538     if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack)
1539       CurrentState.AvoidBinPacking = true;
1540     else
1541       CurrentState.BreakBeforeParameter = false;
1542   }
1543   if (Current.is(TT_InheritanceColon)) {
1544     CurrentState.Indent =
1545         State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1546   }
1547   if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline)
1548     CurrentState.NestedBlockIndent = State.Column + Current.ColumnWidth + 1;
1549   if (Current.isOneOf(TT_LambdaLSquare, TT_TrailingReturnArrow))
1550     CurrentState.LastSpace = State.Column;
1551   if (Current.is(TT_RequiresExpression) &&
1552       Style.RequiresExpressionIndentation == FormatStyle::REI_Keyword) {
1553     CurrentState.NestedBlockIndent = State.Column;
1554   }
1555 
1556   // Insert scopes created by fake parenthesis.
1557   const FormatToken *Previous = Current.getPreviousNonComment();
1558 
1559   // Add special behavior to support a format commonly used for JavaScript
1560   // closures:
1561   //   SomeFunction(function() {
1562   //     foo();
1563   //     bar();
1564   //   }, a, b, c);
1565   if (Current.isNot(tok::comment) && !Current.ClosesRequiresClause &&
1566       Previous && Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
1567       Previous->isNot(TT_DictLiteral) && State.Stack.size() > 1 &&
1568       !CurrentState.HasMultipleNestedBlocks) {
1569     if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline)
1570       for (ParenState &PState : llvm::drop_end(State.Stack))
1571         PState.NoLineBreak = true;
1572     State.Stack[State.Stack.size() - 2].NestedBlockInlined = false;
1573   }
1574   if (Previous && (Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr) ||
1575                    (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) &&
1576                     !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)))) {
1577     CurrentState.NestedBlockInlined =
1578         !Newline && hasNestedBlockInlined(Previous, Current, Style);
1579   }
1580 
1581   moveStatePastFakeLParens(State, Newline);
1582   moveStatePastScopeCloser(State);
1583   // Do not use CurrentState here, since the two functions before may change the
1584   // Stack.
1585   bool AllowBreak = !State.Stack.back().NoLineBreak &&
1586                     !State.Stack.back().NoLineBreakInOperand;
1587   moveStatePastScopeOpener(State, Newline);
1588   moveStatePastFakeRParens(State);
1589 
1590   if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0)
1591     State.StartOfStringLiteral = State.Column + 1;
1592   if (Current.is(TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) {
1593     State.StartOfStringLiteral = State.Column + 1;
1594   } else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) {
1595     State.StartOfStringLiteral = State.Column;
1596   } else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) &&
1597              !Current.isStringLiteral()) {
1598     State.StartOfStringLiteral = 0;
1599   }
1600 
1601   State.Column += Current.ColumnWidth;
1602   State.NextToken = State.NextToken->Next;
1603   // Verilog case labels are on the same unwrapped lines as the statements that
1604   // follow. TokenAnnotator identifies them and sets MustBreakBefore.
1605   // Indentation is taken care of here. A case label can only have 1 statement
1606   // in Verilog, so we don't have to worry about lines that follow.
1607   if (Style.isVerilog() && State.NextToken &&
1608       State.NextToken->MustBreakBefore &&
1609       Keywords.isVerilogEndOfLabel(Current)) {
1610     State.FirstIndent += Style.IndentWidth;
1611     CurrentState.Indent = State.FirstIndent;
1612   }
1613 
1614   unsigned Penalty =
1615       handleEndOfLine(Current, State, DryRun, AllowBreak, Newline);
1616 
1617   if (Current.Role)
1618     Current.Role->formatFromToken(State, this, DryRun);
1619   // If the previous has a special role, let it consume tokens as appropriate.
1620   // It is necessary to start at the previous token for the only implemented
1621   // role (comma separated list). That way, the decision whether or not to break
1622   // after the "{" is already done and both options are tried and evaluated.
1623   // FIXME: This is ugly, find a better way.
1624   if (Previous && Previous->Role)
1625     Penalty += Previous->Role->formatAfterToken(State, this, DryRun);
1626 
1627   return Penalty;
1628 }
1629 
1630 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State,
1631                                                     bool Newline) {
1632   const FormatToken &Current = *State.NextToken;
1633   if (Current.FakeLParens.empty())
1634     return;
1635 
1636   const FormatToken *Previous = Current.getPreviousNonComment();
1637 
1638   // Don't add extra indentation for the first fake parenthesis after
1639   // 'return', assignments, opening <({[, or requires clauses. The indentation
1640   // for these cases is special cased.
1641   bool SkipFirstExtraIndent =
1642       Previous &&
1643       (Previous->opensScope() ||
1644        Previous->isOneOf(tok::semi, tok::kw_return, TT_RequiresClause) ||
1645        (Previous->getPrecedence() == prec::Assignment &&
1646         Style.AlignOperands != FormatStyle::OAS_DontAlign) ||
1647        Previous->is(TT_ObjCMethodExpr));
1648   for (const auto &PrecedenceLevel : llvm::reverse(Current.FakeLParens)) {
1649     const auto &CurrentState = State.Stack.back();
1650     ParenState NewParenState = CurrentState;
1651     NewParenState.Tok = nullptr;
1652     NewParenState.ContainsLineBreak = false;
1653     NewParenState.LastOperatorWrapped = true;
1654     NewParenState.IsChainedConditional = false;
1655     NewParenState.IsWrappedConditional = false;
1656     NewParenState.UnindentOperator = false;
1657     NewParenState.NoLineBreak =
1658         NewParenState.NoLineBreak || CurrentState.NoLineBreakInOperand;
1659 
1660     // Don't propagate AvoidBinPacking into subexpressions of arg/param lists.
1661     if (PrecedenceLevel > prec::Comma)
1662       NewParenState.AvoidBinPacking = false;
1663 
1664     // Indent from 'LastSpace' unless these are fake parentheses encapsulating
1665     // a builder type call after 'return' or, if the alignment after opening
1666     // brackets is disabled.
1667     if (!Current.isTrailingComment() &&
1668         (Style.AlignOperands != FormatStyle::OAS_DontAlign ||
1669          PrecedenceLevel < prec::Assignment) &&
1670         (!Previous || Previous->isNot(tok::kw_return) ||
1671          (Style.Language != FormatStyle::LK_Java && PrecedenceLevel > 0)) &&
1672         (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign ||
1673          PrecedenceLevel != prec::Comma || Current.NestingLevel == 0)) {
1674       NewParenState.Indent = std::max(
1675           std::max(State.Column, NewParenState.Indent), CurrentState.LastSpace);
1676     }
1677 
1678     // Special case for generic selection expressions, its comma-separated
1679     // expressions are not aligned to the opening paren like regular calls, but
1680     // rather continuation-indented relative to the _Generic keyword.
1681     if (Previous && Previous->endsSequence(tok::l_paren, tok::kw__Generic))
1682       NewParenState.Indent = CurrentState.LastSpace;
1683 
1684     if ((shouldUnindentNextOperator(Current) ||
1685          (Previous &&
1686           (PrecedenceLevel == prec::Conditional &&
1687            Previous->is(tok::question) && Previous->is(TT_ConditionalExpr)))) &&
1688         !Newline) {
1689       // If BreakBeforeBinaryOperators is set, un-indent a bit to account for
1690       // the operator and keep the operands aligned.
1691       if (Style.AlignOperands == FormatStyle::OAS_AlignAfterOperator)
1692         NewParenState.UnindentOperator = true;
1693       // Mark indentation as alignment if the expression is aligned.
1694       if (Style.AlignOperands != FormatStyle::OAS_DontAlign)
1695         NewParenState.IsAligned = true;
1696     }
1697 
1698     // Do not indent relative to the fake parentheses inserted for "." or "->".
1699     // This is a special case to make the following to statements consistent:
1700     //   OuterFunction(InnerFunctionCall( // break
1701     //       ParameterToInnerFunction));
1702     //   OuterFunction(SomeObject.InnerFunctionCall( // break
1703     //       ParameterToInnerFunction));
1704     if (PrecedenceLevel > prec::Unknown)
1705       NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column);
1706     if (PrecedenceLevel != prec::Conditional &&
1707         Current.isNot(TT_UnaryOperator) &&
1708         Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) {
1709       NewParenState.StartOfFunctionCall = State.Column;
1710     }
1711 
1712     // Indent conditional expressions, unless they are chained "else-if"
1713     // conditionals. Never indent expression where the 'operator' is ',', ';' or
1714     // an assignment (i.e. *I <= prec::Assignment) as those have different
1715     // indentation rules. Indent other expression, unless the indentation needs
1716     // to be skipped.
1717     if (PrecedenceLevel == prec::Conditional && Previous &&
1718         Previous->is(tok::colon) && Previous->is(TT_ConditionalExpr) &&
1719         &PrecedenceLevel == &Current.FakeLParens.back() &&
1720         !CurrentState.IsWrappedConditional) {
1721       NewParenState.IsChainedConditional = true;
1722       NewParenState.UnindentOperator = State.Stack.back().UnindentOperator;
1723     } else if (PrecedenceLevel == prec::Conditional ||
1724                (!SkipFirstExtraIndent && PrecedenceLevel > prec::Assignment &&
1725                 !Current.isTrailingComment())) {
1726       NewParenState.Indent += Style.ContinuationIndentWidth;
1727     }
1728     if ((Previous && !Previous->opensScope()) || PrecedenceLevel != prec::Comma)
1729       NewParenState.BreakBeforeParameter = false;
1730     State.Stack.push_back(NewParenState);
1731     SkipFirstExtraIndent = false;
1732   }
1733 }
1734 
1735 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) {
1736   for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) {
1737     unsigned VariablePos = State.Stack.back().VariablePos;
1738     if (State.Stack.size() == 1) {
1739       // Do not pop the last element.
1740       break;
1741     }
1742     State.Stack.pop_back();
1743     State.Stack.back().VariablePos = VariablePos;
1744   }
1745 
1746   if (State.NextToken->ClosesRequiresClause && Style.IndentRequiresClause) {
1747     // Remove the indentation of the requires clauses (which is not in Indent,
1748     // but in LastSpace).
1749     State.Stack.back().LastSpace -= Style.IndentWidth;
1750   }
1751 }
1752 
1753 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
1754                                                     bool Newline) {
1755   const FormatToken &Current = *State.NextToken;
1756   if (!Current.opensScope())
1757     return;
1758 
1759   const auto &CurrentState = State.Stack.back();
1760 
1761   // Don't allow '<' or '(' in C# generic type constraints to start new scopes.
1762   if (Current.isOneOf(tok::less, tok::l_paren) &&
1763       CurrentState.IsCSharpGenericTypeConstraint) {
1764     return;
1765   }
1766 
1767   if (Current.MatchingParen && Current.is(BK_Block)) {
1768     moveStateToNewBlock(State);
1769     return;
1770   }
1771 
1772   unsigned NewIndent;
1773   unsigned LastSpace = CurrentState.LastSpace;
1774   bool AvoidBinPacking;
1775   bool BreakBeforeParameter = false;
1776   unsigned NestedBlockIndent = std::max(CurrentState.StartOfFunctionCall,
1777                                         CurrentState.NestedBlockIndent);
1778   if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
1779       opensProtoMessageField(Current, Style)) {
1780     if (Current.opensBlockOrBlockTypeList(Style)) {
1781       NewIndent = Style.IndentWidth +
1782                   std::min(State.Column, CurrentState.NestedBlockIndent);
1783     } else if (Current.is(tok::l_brace)) {
1784       NewIndent =
1785           CurrentState.LastSpace + Style.BracedInitializerIndentWidth.value_or(
1786                                        Style.ContinuationIndentWidth);
1787     } else {
1788       NewIndent = CurrentState.LastSpace + Style.ContinuationIndentWidth;
1789     }
1790     const FormatToken *NextNonComment = Current.getNextNonComment();
1791     bool EndsInComma = Current.MatchingParen &&
1792                        Current.MatchingParen->Previous &&
1793                        Current.MatchingParen->Previous->is(tok::comma);
1794     AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) ||
1795                       Style.isProto() || !Style.BinPackArguments ||
1796                       (NextNonComment && NextNonComment->isOneOf(
1797                                              TT_DesignatedInitializerPeriod,
1798                                              TT_DesignatedInitializerLSquare));
1799     BreakBeforeParameter = EndsInComma;
1800     if (Current.ParameterCount > 1)
1801       NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1);
1802   } else {
1803     NewIndent =
1804         Style.ContinuationIndentWidth +
1805         std::max(CurrentState.LastSpace, CurrentState.StartOfFunctionCall);
1806 
1807     // Ensure that different different brackets force relative alignment, e.g.:
1808     // void SomeFunction(vector<  // break
1809     //                       int> v);
1810     // FIXME: We likely want to do this for more combinations of brackets.
1811     if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) {
1812       NewIndent = std::max(NewIndent, CurrentState.Indent);
1813       LastSpace = std::max(LastSpace, CurrentState.Indent);
1814     }
1815 
1816     bool EndsInComma =
1817         Current.MatchingParen &&
1818         Current.MatchingParen->getPreviousNonComment() &&
1819         Current.MatchingParen->getPreviousNonComment()->is(tok::comma);
1820 
1821     // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters
1822     // for backwards compatibility.
1823     bool ObjCBinPackProtocolList =
1824         (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto &&
1825          Style.BinPackParameters) ||
1826         Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always;
1827 
1828     bool BinPackDeclaration =
1829         (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) ||
1830         (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList);
1831 
1832     bool GenericSelection =
1833         Current.getPreviousNonComment() &&
1834         Current.getPreviousNonComment()->is(tok::kw__Generic);
1835 
1836     AvoidBinPacking =
1837         (CurrentState.IsCSharpGenericTypeConstraint) || GenericSelection ||
1838         (Style.isJavaScript() && EndsInComma) ||
1839         (State.Line->MustBeDeclaration && !BinPackDeclaration) ||
1840         (!State.Line->MustBeDeclaration && !Style.BinPackArguments) ||
1841         (Style.ExperimentalAutoDetectBinPacking &&
1842          (Current.is(PPK_OnePerLine) ||
1843           (!BinPackInconclusiveFunctions && Current.is(PPK_Inconclusive))));
1844 
1845     if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen &&
1846         Style.ObjCBreakBeforeNestedBlockParam) {
1847       if (Style.ColumnLimit) {
1848         // If this '[' opens an ObjC call, determine whether all parameters fit
1849         // into one line and put one per line if they don't.
1850         if (getLengthToMatchingParen(Current, State.Stack) + State.Column >
1851             getColumnLimit(State)) {
1852           BreakBeforeParameter = true;
1853         }
1854       } else {
1855         // For ColumnLimit = 0, we have to figure out whether there is or has to
1856         // be a line break within this call.
1857         for (const FormatToken *Tok = &Current;
1858              Tok && Tok != Current.MatchingParen; Tok = Tok->Next) {
1859           if (Tok->MustBreakBefore ||
1860               (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) {
1861             BreakBeforeParameter = true;
1862             break;
1863           }
1864         }
1865       }
1866     }
1867 
1868     if (Style.isJavaScript() && EndsInComma)
1869       BreakBeforeParameter = true;
1870   }
1871   // Generally inherit NoLineBreak from the current scope to nested scope.
1872   // However, don't do this for non-empty nested blocks, dict literals and
1873   // array literals as these follow different indentation rules.
1874   bool NoLineBreak =
1875       Current.Children.empty() &&
1876       !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) &&
1877       (CurrentState.NoLineBreak || CurrentState.NoLineBreakInOperand ||
1878        (Current.is(TT_TemplateOpener) &&
1879         CurrentState.ContainsUnwrappedBuilder));
1880   State.Stack.push_back(
1881       ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak));
1882   auto &NewState = State.Stack.back();
1883   NewState.NestedBlockIndent = NestedBlockIndent;
1884   NewState.BreakBeforeParameter = BreakBeforeParameter;
1885   NewState.HasMultipleNestedBlocks = (Current.BlockParameterCount > 1);
1886 
1887   if (Style.BraceWrapping.BeforeLambdaBody && Current.Next &&
1888       Current.is(tok::l_paren)) {
1889     // Search for any parameter that is a lambda.
1890     FormatToken const *next = Current.Next;
1891     while (next) {
1892       if (next->is(TT_LambdaLSquare)) {
1893         NewState.HasMultipleNestedBlocks = true;
1894         break;
1895       }
1896       next = next->Next;
1897     }
1898   }
1899 
1900   NewState.IsInsideObjCArrayLiteral = Current.is(TT_ArrayInitializerLSquare) &&
1901                                       Current.Previous &&
1902                                       Current.Previous->is(tok::at);
1903 }
1904 
1905 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) {
1906   const FormatToken &Current = *State.NextToken;
1907   if (!Current.closesScope())
1908     return;
1909 
1910   // If we encounter a closing ), ], } or >, we can remove a level from our
1911   // stacks.
1912   if (State.Stack.size() > 1 &&
1913       (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) ||
1914        (Current.is(tok::r_brace) && State.NextToken != State.Line->First) ||
1915        State.NextToken->is(TT_TemplateCloser) ||
1916        (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) {
1917     State.Stack.pop_back();
1918   }
1919 
1920   auto &CurrentState = State.Stack.back();
1921 
1922   // Reevaluate whether ObjC message arguments fit into one line.
1923   // If a receiver spans multiple lines, e.g.:
1924   //   [[object block:^{
1925   //     return 42;
1926   //   }] a:42 b:42];
1927   // BreakBeforeParameter is calculated based on an incorrect assumption
1928   // (it is checked whether the whole expression fits into one line without
1929   // considering a line break inside a message receiver).
1930   // We check whether arguments fit after receiver scope closer (into the same
1931   // line).
1932   if (CurrentState.BreakBeforeParameter && Current.MatchingParen &&
1933       Current.MatchingParen->Previous) {
1934     const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous;
1935     if (CurrentScopeOpener.is(TT_ObjCMethodExpr) &&
1936         CurrentScopeOpener.MatchingParen) {
1937       int NecessarySpaceInLine =
1938           getLengthToMatchingParen(CurrentScopeOpener, State.Stack) +
1939           CurrentScopeOpener.TotalLength - Current.TotalLength - 1;
1940       if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <=
1941           Style.ColumnLimit) {
1942         CurrentState.BreakBeforeParameter = false;
1943       }
1944     }
1945   }
1946 
1947   if (Current.is(tok::r_square)) {
1948     // If this ends the array subscript expr, reset the corresponding value.
1949     const FormatToken *NextNonComment = Current.getNextNonComment();
1950     if (NextNonComment && NextNonComment->isNot(tok::l_square))
1951       CurrentState.StartOfArraySubscripts = 0;
1952   }
1953 }
1954 
1955 void ContinuationIndenter::moveStateToNewBlock(LineState &State) {
1956   if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
1957       State.NextToken->is(TT_LambdaLBrace) &&
1958       !State.Line->MightBeFunctionDecl) {
1959     State.Stack.back().NestedBlockIndent = State.FirstIndent;
1960   }
1961   unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent;
1962   // ObjC block sometimes follow special indentation rules.
1963   unsigned NewIndent =
1964       NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace)
1965                                ? Style.ObjCBlockIndentWidth
1966                                : Style.IndentWidth);
1967   State.Stack.push_back(ParenState(State.NextToken, NewIndent,
1968                                    State.Stack.back().LastSpace,
1969                                    /*AvoidBinPacking=*/true,
1970                                    /*NoLineBreak=*/false));
1971   State.Stack.back().NestedBlockIndent = NestedBlockIndent;
1972   State.Stack.back().BreakBeforeParameter = true;
1973 }
1974 
1975 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn,
1976                                      unsigned TabWidth,
1977                                      encoding::Encoding Encoding) {
1978   size_t LastNewlinePos = Text.find_last_of("\n");
1979   if (LastNewlinePos == StringRef::npos) {
1980     return StartColumn +
1981            encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding);
1982   } else {
1983     return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos),
1984                                          /*StartColumn=*/0, TabWidth, Encoding);
1985   }
1986 }
1987 
1988 unsigned ContinuationIndenter::reformatRawStringLiteral(
1989     const FormatToken &Current, LineState &State,
1990     const FormatStyle &RawStringStyle, bool DryRun, bool Newline) {
1991   unsigned StartColumn = State.Column - Current.ColumnWidth;
1992   StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText);
1993   StringRef NewDelimiter =
1994       getCanonicalRawStringDelimiter(Style, RawStringStyle.Language);
1995   if (NewDelimiter.empty())
1996     NewDelimiter = OldDelimiter;
1997   // The text of a raw string is between the leading 'R"delimiter(' and the
1998   // trailing 'delimiter)"'.
1999   unsigned OldPrefixSize = 3 + OldDelimiter.size();
2000   unsigned OldSuffixSize = 2 + OldDelimiter.size();
2001   // We create a virtual text environment which expects a null-terminated
2002   // string, so we cannot use StringRef.
2003   std::string RawText = std::string(
2004       Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize));
2005   if (NewDelimiter != OldDelimiter) {
2006     // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the
2007     // raw string.
2008     std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str();
2009     if (StringRef(RawText).contains(CanonicalDelimiterSuffix))
2010       NewDelimiter = OldDelimiter;
2011   }
2012 
2013   unsigned NewPrefixSize = 3 + NewDelimiter.size();
2014   unsigned NewSuffixSize = 2 + NewDelimiter.size();
2015 
2016   // The first start column is the column the raw text starts after formatting.
2017   unsigned FirstStartColumn = StartColumn + NewPrefixSize;
2018 
2019   // The next start column is the intended indentation a line break inside
2020   // the raw string at level 0. It is determined by the following rules:
2021   //   - if the content starts on newline, it is one level more than the current
2022   //     indent, and
2023   //   - if the content does not start on a newline, it is the first start
2024   //     column.
2025   // These rules have the advantage that the formatted content both does not
2026   // violate the rectangle rule and visually flows within the surrounding
2027   // source.
2028   bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n';
2029   // If this token is the last parameter (checked by looking if it's followed by
2030   // `)` and is not on a newline, the base the indent off the line's nested
2031   // block indent. Otherwise, base the indent off the arguments indent, so we
2032   // can achieve:
2033   //
2034   // fffffffffff(1, 2, 3, R"pb(
2035   //     key1: 1  #
2036   //     key2: 2)pb");
2037   //
2038   // fffffffffff(1, 2, 3,
2039   //             R"pb(
2040   //               key1: 1  #
2041   //               key2: 2
2042   //             )pb");
2043   //
2044   // fffffffffff(1, 2, 3,
2045   //             R"pb(
2046   //               key1: 1  #
2047   //               key2: 2
2048   //             )pb",
2049   //             5);
2050   unsigned CurrentIndent =
2051       (!Newline && Current.Next && Current.Next->is(tok::r_paren))
2052           ? State.Stack.back().NestedBlockIndent
2053           : State.Stack.back().Indent;
2054   unsigned NextStartColumn = ContentStartsOnNewline
2055                                  ? CurrentIndent + Style.IndentWidth
2056                                  : FirstStartColumn;
2057 
2058   // The last start column is the column the raw string suffix starts if it is
2059   // put on a newline.
2060   // The last start column is the intended indentation of the raw string postfix
2061   // if it is put on a newline. It is determined by the following rules:
2062   //   - if the raw string prefix starts on a newline, it is the column where
2063   //     that raw string prefix starts, and
2064   //   - if the raw string prefix does not start on a newline, it is the current
2065   //     indent.
2066   unsigned LastStartColumn =
2067       Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent;
2068 
2069   std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat(
2070       RawStringStyle, RawText, {tooling::Range(0, RawText.size())},
2071       FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>",
2072       /*Status=*/nullptr);
2073 
2074   auto NewCode = applyAllReplacements(RawText, Fixes.first);
2075   tooling::Replacements NoFixes;
2076   if (!NewCode)
2077     return addMultilineToken(Current, State);
2078   if (!DryRun) {
2079     if (NewDelimiter != OldDelimiter) {
2080       // In 'R"delimiter(...', the delimiter starts 2 characters after the start
2081       // of the token.
2082       SourceLocation PrefixDelimiterStart =
2083           Current.Tok.getLocation().getLocWithOffset(2);
2084       auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement(
2085           SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter));
2086       if (PrefixErr) {
2087         llvm::errs()
2088             << "Failed to update the prefix delimiter of a raw string: "
2089             << llvm::toString(std::move(PrefixErr)) << "\n";
2090       }
2091       // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at
2092       // position length - 1 - |delimiter|.
2093       SourceLocation SuffixDelimiterStart =
2094           Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() -
2095                                                      1 - OldDelimiter.size());
2096       auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement(
2097           SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter));
2098       if (SuffixErr) {
2099         llvm::errs()
2100             << "Failed to update the suffix delimiter of a raw string: "
2101             << llvm::toString(std::move(SuffixErr)) << "\n";
2102       }
2103     }
2104     SourceLocation OriginLoc =
2105         Current.Tok.getLocation().getLocWithOffset(OldPrefixSize);
2106     for (const tooling::Replacement &Fix : Fixes.first) {
2107       auto Err = Whitespaces.addReplacement(tooling::Replacement(
2108           SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()),
2109           Fix.getLength(), Fix.getReplacementText()));
2110       if (Err) {
2111         llvm::errs() << "Failed to reformat raw string: "
2112                      << llvm::toString(std::move(Err)) << "\n";
2113       }
2114     }
2115   }
2116   unsigned RawLastLineEndColumn = getLastLineEndColumn(
2117       *NewCode, FirstStartColumn, Style.TabWidth, Encoding);
2118   State.Column = RawLastLineEndColumn + NewSuffixSize;
2119   // Since we're updating the column to after the raw string literal here, we
2120   // have to manually add the penalty for the prefix R"delim( over the column
2121   // limit.
2122   unsigned PrefixExcessCharacters =
2123       StartColumn + NewPrefixSize > Style.ColumnLimit
2124           ? StartColumn + NewPrefixSize - Style.ColumnLimit
2125           : 0;
2126   bool IsMultiline =
2127       ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos);
2128   if (IsMultiline) {
2129     // Break before further function parameters on all levels.
2130     for (ParenState &Paren : State.Stack)
2131       Paren.BreakBeforeParameter = true;
2132   }
2133   return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter;
2134 }
2135 
2136 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current,
2137                                                  LineState &State) {
2138   // Break before further function parameters on all levels.
2139   for (ParenState &Paren : State.Stack)
2140     Paren.BreakBeforeParameter = true;
2141 
2142   unsigned ColumnsUsed = State.Column;
2143   // We can only affect layout of the first and the last line, so the penalty
2144   // for all other lines is constant, and we ignore it.
2145   State.Column = Current.LastLineColumnWidth;
2146 
2147   if (ColumnsUsed > getColumnLimit(State))
2148     return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State));
2149   return 0;
2150 }
2151 
2152 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current,
2153                                                LineState &State, bool DryRun,
2154                                                bool AllowBreak, bool Newline) {
2155   unsigned Penalty = 0;
2156   // Compute the raw string style to use in case this is a raw string literal
2157   // that can be reformatted.
2158   auto RawStringStyle = getRawStringStyle(Current, State);
2159   if (RawStringStyle && !Current.Finalized) {
2160     Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun,
2161                                        Newline);
2162   } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) {
2163     // Don't break multi-line tokens other than block comments and raw string
2164     // literals. Instead, just update the state.
2165     Penalty = addMultilineToken(Current, State);
2166   } else if (State.Line->Type != LT_ImportStatement) {
2167     // We generally don't break import statements.
2168     LineState OriginalState = State;
2169 
2170     // Whether we force the reflowing algorithm to stay strictly within the
2171     // column limit.
2172     bool Strict = false;
2173     // Whether the first non-strict attempt at reflowing did intentionally
2174     // exceed the column limit.
2175     bool Exceeded = false;
2176     std::tie(Penalty, Exceeded) = breakProtrudingToken(
2177         Current, State, AllowBreak, /*DryRun=*/true, Strict);
2178     if (Exceeded) {
2179       // If non-strict reflowing exceeds the column limit, try whether strict
2180       // reflowing leads to an overall lower penalty.
2181       LineState StrictState = OriginalState;
2182       unsigned StrictPenalty =
2183           breakProtrudingToken(Current, StrictState, AllowBreak,
2184                                /*DryRun=*/true, /*Strict=*/true)
2185               .first;
2186       Strict = StrictPenalty <= Penalty;
2187       if (Strict) {
2188         Penalty = StrictPenalty;
2189         State = StrictState;
2190       }
2191     }
2192     if (!DryRun) {
2193       // If we're not in dry-run mode, apply the changes with the decision on
2194       // strictness made above.
2195       breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false,
2196                            Strict);
2197     }
2198   }
2199   if (State.Column > getColumnLimit(State)) {
2200     unsigned ExcessCharacters = State.Column - getColumnLimit(State);
2201     Penalty += Style.PenaltyExcessCharacter * ExcessCharacters;
2202   }
2203   return Penalty;
2204 }
2205 
2206 // Returns the enclosing function name of a token, or the empty string if not
2207 // found.
2208 static StringRef getEnclosingFunctionName(const FormatToken &Current) {
2209   // Look for: 'function(' or 'function<templates>(' before Current.
2210   auto Tok = Current.getPreviousNonComment();
2211   if (!Tok || Tok->isNot(tok::l_paren))
2212     return "";
2213   Tok = Tok->getPreviousNonComment();
2214   if (!Tok)
2215     return "";
2216   if (Tok->is(TT_TemplateCloser)) {
2217     Tok = Tok->MatchingParen;
2218     if (Tok)
2219       Tok = Tok->getPreviousNonComment();
2220   }
2221   if (!Tok || Tok->isNot(tok::identifier))
2222     return "";
2223   return Tok->TokenText;
2224 }
2225 
2226 std::optional<FormatStyle>
2227 ContinuationIndenter::getRawStringStyle(const FormatToken &Current,
2228                                         const LineState &State) {
2229   if (!Current.isStringLiteral())
2230     return std::nullopt;
2231   auto Delimiter = getRawStringDelimiter(Current.TokenText);
2232   if (!Delimiter)
2233     return std::nullopt;
2234   auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter);
2235   if (!RawStringStyle && Delimiter->empty()) {
2236     RawStringStyle = RawStringFormats.getEnclosingFunctionStyle(
2237         getEnclosingFunctionName(Current));
2238   }
2239   if (!RawStringStyle)
2240     return std::nullopt;
2241   RawStringStyle->ColumnLimit = getColumnLimit(State);
2242   return RawStringStyle;
2243 }
2244 
2245 std::unique_ptr<BreakableToken>
2246 ContinuationIndenter::createBreakableToken(const FormatToken &Current,
2247                                            LineState &State, bool AllowBreak) {
2248   unsigned StartColumn = State.Column - Current.ColumnWidth;
2249   if (Current.isStringLiteral()) {
2250     // Strings in JSON cannot be broken. Breaking strings in JavaScript is
2251     // disabled for now.
2252     if (Style.isJson() || Style.isJavaScript() || !Style.BreakStringLiterals ||
2253         !AllowBreak) {
2254       return nullptr;
2255     }
2256 
2257     // Don't break string literals inside preprocessor directives (except for
2258     // #define directives, as their contents are stored in separate lines and
2259     // are not affected by this check).
2260     // This way we avoid breaking code with line directives and unknown
2261     // preprocessor directives that contain long string literals.
2262     if (State.Line->Type == LT_PreprocessorDirective)
2263       return nullptr;
2264     // Exempts unterminated string literals from line breaking. The user will
2265     // likely want to terminate the string before any line breaking is done.
2266     if (Current.IsUnterminatedLiteral)
2267       return nullptr;
2268     // Don't break string literals inside Objective-C array literals (doing so
2269     // raises the warning -Wobjc-string-concatenation).
2270     if (State.Stack.back().IsInsideObjCArrayLiteral)
2271       return nullptr;
2272 
2273     // The "DPI"/"DPI-C" in SystemVerilog direct programming interface
2274     // imports/exports cannot be split, e.g.
2275     // `import "DPI" function foo();`
2276     // FIXME: make this use same infra as C++ import checks
2277     if (Style.isVerilog() && Current.Previous &&
2278         Current.Previous->isOneOf(tok::kw_export, Keywords.kw_import)) {
2279       return nullptr;
2280     }
2281     StringRef Text = Current.TokenText;
2282 
2283     // We need this to address the case where there is an unbreakable tail only
2284     // if certain other formatting decisions have been taken. The
2285     // UnbreakableTailLength of Current is an overapproximation in that case and
2286     // we need to be correct here.
2287     unsigned UnbreakableTailLength = (State.NextToken && canBreak(State))
2288                                          ? 0
2289                                          : Current.UnbreakableTailLength;
2290 
2291     if (Style.isVerilog() || Style.Language == FormatStyle::LK_Java ||
2292         Style.isJavaScript() || Style.isCSharp()) {
2293       BreakableStringLiteralUsingOperators::QuoteStyleType QuoteStyle;
2294       if (Style.isJavaScript() && Text.starts_with("'") &&
2295           Text.ends_with("'")) {
2296         QuoteStyle = BreakableStringLiteralUsingOperators::SingleQuotes;
2297       } else if (Style.isCSharp() && Text.starts_with("@\"") &&
2298                  Text.ends_with("\"")) {
2299         QuoteStyle = BreakableStringLiteralUsingOperators::AtDoubleQuotes;
2300       } else if (Text.starts_with("\"") && Text.ends_with("\"")) {
2301         QuoteStyle = BreakableStringLiteralUsingOperators::DoubleQuotes;
2302       } else {
2303         return nullptr;
2304       }
2305       return std::make_unique<BreakableStringLiteralUsingOperators>(
2306           Current, QuoteStyle,
2307           /*UnindentPlus=*/shouldUnindentNextOperator(Current), StartColumn,
2308           UnbreakableTailLength, State.Line->InPPDirective, Encoding, Style);
2309     }
2310 
2311     StringRef Prefix;
2312     StringRef Postfix;
2313     // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'.
2314     // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to
2315     // reduce the overhead) for each FormatToken, which is a string, so that we
2316     // don't run multiple checks here on the hot path.
2317     if ((Text.ends_with(Postfix = "\"") &&
2318          (Text.starts_with(Prefix = "@\"") || Text.starts_with(Prefix = "\"") ||
2319           Text.starts_with(Prefix = "u\"") ||
2320           Text.starts_with(Prefix = "U\"") ||
2321           Text.starts_with(Prefix = "u8\"") ||
2322           Text.starts_with(Prefix = "L\""))) ||
2323         (Text.starts_with(Prefix = "_T(\"") &&
2324          Text.ends_with(Postfix = "\")"))) {
2325       return std::make_unique<BreakableStringLiteral>(
2326           Current, StartColumn, Prefix, Postfix, UnbreakableTailLength,
2327           State.Line->InPPDirective, Encoding, Style);
2328     }
2329   } else if (Current.is(TT_BlockComment)) {
2330     if (!Style.ReflowComments ||
2331         // If a comment token switches formatting, like
2332         // /* clang-format on */, we don't want to break it further,
2333         // but we may still want to adjust its indentation.
2334         switchesFormatting(Current)) {
2335       return nullptr;
2336     }
2337     return std::make_unique<BreakableBlockComment>(
2338         Current, StartColumn, Current.OriginalColumn, !Current.Previous,
2339         State.Line->InPPDirective, Encoding, Style, Whitespaces.useCRLF());
2340   } else if (Current.is(TT_LineComment) &&
2341              (!Current.Previous ||
2342               Current.Previous->isNot(TT_ImplicitStringLiteral))) {
2343     bool RegularComments = [&]() {
2344       for (const FormatToken *T = &Current; T && T->is(TT_LineComment);
2345            T = T->Next) {
2346         if (!(T->TokenText.starts_with("//") || T->TokenText.starts_with("#")))
2347           return false;
2348       }
2349       return true;
2350     }();
2351     if (!Style.ReflowComments ||
2352         CommentPragmasRegex.match(Current.TokenText.substr(2)) ||
2353         switchesFormatting(Current) || !RegularComments) {
2354       return nullptr;
2355     }
2356     return std::make_unique<BreakableLineCommentSection>(
2357         Current, StartColumn, /*InPPDirective=*/false, Encoding, Style);
2358   }
2359   return nullptr;
2360 }
2361 
2362 std::pair<unsigned, bool>
2363 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current,
2364                                            LineState &State, bool AllowBreak,
2365                                            bool DryRun, bool Strict) {
2366   std::unique_ptr<const BreakableToken> Token =
2367       createBreakableToken(Current, State, AllowBreak);
2368   if (!Token)
2369     return {0, false};
2370   assert(Token->getLineCount() > 0);
2371   unsigned ColumnLimit = getColumnLimit(State);
2372   if (Current.is(TT_LineComment)) {
2373     // We don't insert backslashes when breaking line comments.
2374     ColumnLimit = Style.ColumnLimit;
2375   }
2376   if (ColumnLimit == 0) {
2377     // To make the rest of the function easier set the column limit to the
2378     // maximum, if there should be no limit.
2379     ColumnLimit = std::numeric_limits<decltype(ColumnLimit)>::max();
2380   }
2381   if (Current.UnbreakableTailLength >= ColumnLimit)
2382     return {0, false};
2383   // ColumnWidth was already accounted into State.Column before calling
2384   // breakProtrudingToken.
2385   unsigned StartColumn = State.Column - Current.ColumnWidth;
2386   unsigned NewBreakPenalty = Current.isStringLiteral()
2387                                  ? Style.PenaltyBreakString
2388                                  : Style.PenaltyBreakComment;
2389   // Stores whether we intentionally decide to let a line exceed the column
2390   // limit.
2391   bool Exceeded = false;
2392   // Stores whether we introduce a break anywhere in the token.
2393   bool BreakInserted = Token->introducesBreakBeforeToken();
2394   // Store whether we inserted a new line break at the end of the previous
2395   // logical line.
2396   bool NewBreakBefore = false;
2397   // We use a conservative reflowing strategy. Reflow starts after a line is
2398   // broken or the corresponding whitespace compressed. Reflow ends as soon as a
2399   // line that doesn't get reflown with the previous line is reached.
2400   bool Reflow = false;
2401   // Keep track of where we are in the token:
2402   // Where we are in the content of the current logical line.
2403   unsigned TailOffset = 0;
2404   // The column number we're currently at.
2405   unsigned ContentStartColumn =
2406       Token->getContentStartColumn(0, /*Break=*/false);
2407   // The number of columns left in the current logical line after TailOffset.
2408   unsigned RemainingTokenColumns =
2409       Token->getRemainingLength(0, TailOffset, ContentStartColumn);
2410   // Adapt the start of the token, for example indent.
2411   if (!DryRun)
2412     Token->adaptStartOfLine(0, Whitespaces);
2413 
2414   unsigned ContentIndent = 0;
2415   unsigned Penalty = 0;
2416   LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column "
2417                           << StartColumn << ".\n");
2418   for (unsigned LineIndex = 0, EndIndex = Token->getLineCount();
2419        LineIndex != EndIndex; ++LineIndex) {
2420     LLVM_DEBUG(llvm::dbgs()
2421                << "  Line: " << LineIndex << " (Reflow: " << Reflow << ")\n");
2422     NewBreakBefore = false;
2423     // If we did reflow the previous line, we'll try reflowing again. Otherwise
2424     // we'll start reflowing if the current line is broken or whitespace is
2425     // compressed.
2426     bool TryReflow = Reflow;
2427     // Break the current token until we can fit the rest of the line.
2428     while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
2429       LLVM_DEBUG(llvm::dbgs() << "    Over limit, need: "
2430                               << (ContentStartColumn + RemainingTokenColumns)
2431                               << ", space: " << ColumnLimit
2432                               << ", reflown prefix: " << ContentStartColumn
2433                               << ", offset in line: " << TailOffset << "\n");
2434       // If the current token doesn't fit, find the latest possible split in the
2435       // current line so that breaking at it will be under the column limit.
2436       // FIXME: Use the earliest possible split while reflowing to correctly
2437       // compress whitespace within a line.
2438       BreakableToken::Split Split =
2439           Token->getSplit(LineIndex, TailOffset, ColumnLimit,
2440                           ContentStartColumn, CommentPragmasRegex);
2441       if (Split.first == StringRef::npos) {
2442         // No break opportunity - update the penalty and continue with the next
2443         // logical line.
2444         if (LineIndex < EndIndex - 1) {
2445           // The last line's penalty is handled in addNextStateToQueue() or when
2446           // calling replaceWhitespaceAfterLastLine below.
2447           Penalty += Style.PenaltyExcessCharacter *
2448                      (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
2449         }
2450         LLVM_DEBUG(llvm::dbgs() << "    No break opportunity.\n");
2451         break;
2452       }
2453       assert(Split.first != 0);
2454 
2455       if (Token->supportsReflow()) {
2456         // Check whether the next natural split point after the current one can
2457         // still fit the line, either because we can compress away whitespace,
2458         // or because the penalty the excess characters introduce is lower than
2459         // the break penalty.
2460         // We only do this for tokens that support reflowing, and thus allow us
2461         // to change the whitespace arbitrarily (e.g. comments).
2462         // Other tokens, like string literals, can be broken on arbitrary
2463         // positions.
2464 
2465         // First, compute the columns from TailOffset to the next possible split
2466         // position.
2467         // For example:
2468         // ColumnLimit:     |
2469         // // Some text   that    breaks
2470         //    ^ tail offset
2471         //             ^-- split
2472         //    ^-------- to split columns
2473         //                    ^--- next split
2474         //    ^--------------- to next split columns
2475         unsigned ToSplitColumns = Token->getRangeLength(
2476             LineIndex, TailOffset, Split.first, ContentStartColumn);
2477         LLVM_DEBUG(llvm::dbgs() << "    ToSplit: " << ToSplitColumns << "\n");
2478 
2479         BreakableToken::Split NextSplit = Token->getSplit(
2480             LineIndex, TailOffset + Split.first + Split.second, ColumnLimit,
2481             ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex);
2482         // Compute the columns necessary to fit the next non-breakable sequence
2483         // into the current line.
2484         unsigned ToNextSplitColumns = 0;
2485         if (NextSplit.first == StringRef::npos) {
2486           ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset,
2487                                                          ContentStartColumn);
2488         } else {
2489           ToNextSplitColumns = Token->getRangeLength(
2490               LineIndex, TailOffset,
2491               Split.first + Split.second + NextSplit.first, ContentStartColumn);
2492         }
2493         // Compress the whitespace between the break and the start of the next
2494         // unbreakable sequence.
2495         ToNextSplitColumns =
2496             Token->getLengthAfterCompression(ToNextSplitColumns, Split);
2497         LLVM_DEBUG(llvm::dbgs()
2498                    << "    ContentStartColumn: " << ContentStartColumn << "\n");
2499         LLVM_DEBUG(llvm::dbgs()
2500                    << "    ToNextSplit: " << ToNextSplitColumns << "\n");
2501         // If the whitespace compression makes us fit, continue on the current
2502         // line.
2503         bool ContinueOnLine =
2504             ContentStartColumn + ToNextSplitColumns <= ColumnLimit;
2505         unsigned ExcessCharactersPenalty = 0;
2506         if (!ContinueOnLine && !Strict) {
2507           // Similarly, if the excess characters' penalty is lower than the
2508           // penalty of introducing a new break, continue on the current line.
2509           ExcessCharactersPenalty =
2510               (ContentStartColumn + ToNextSplitColumns - ColumnLimit) *
2511               Style.PenaltyExcessCharacter;
2512           LLVM_DEBUG(llvm::dbgs()
2513                      << "    Penalty excess: " << ExcessCharactersPenalty
2514                      << "\n            break : " << NewBreakPenalty << "\n");
2515           if (ExcessCharactersPenalty < NewBreakPenalty) {
2516             Exceeded = true;
2517             ContinueOnLine = true;
2518           }
2519         }
2520         if (ContinueOnLine) {
2521           LLVM_DEBUG(llvm::dbgs() << "    Continuing on line...\n");
2522           // The current line fits after compressing the whitespace - reflow
2523           // the next line into it if possible.
2524           TryReflow = true;
2525           if (!DryRun) {
2526             Token->compressWhitespace(LineIndex, TailOffset, Split,
2527                                       Whitespaces);
2528           }
2529           // When we continue on the same line, leave one space between content.
2530           ContentStartColumn += ToSplitColumns + 1;
2531           Penalty += ExcessCharactersPenalty;
2532           TailOffset += Split.first + Split.second;
2533           RemainingTokenColumns = Token->getRemainingLength(
2534               LineIndex, TailOffset, ContentStartColumn);
2535           continue;
2536         }
2537       }
2538       LLVM_DEBUG(llvm::dbgs() << "    Breaking...\n");
2539       // Update the ContentIndent only if the current line was not reflown with
2540       // the previous line, since in that case the previous line should still
2541       // determine the ContentIndent. Also never intent the last line.
2542       if (!Reflow)
2543         ContentIndent = Token->getContentIndent(LineIndex);
2544       LLVM_DEBUG(llvm::dbgs()
2545                  << "    ContentIndent: " << ContentIndent << "\n");
2546       ContentStartColumn = ContentIndent + Token->getContentStartColumn(
2547                                                LineIndex, /*Break=*/true);
2548 
2549       unsigned NewRemainingTokenColumns = Token->getRemainingLength(
2550           LineIndex, TailOffset + Split.first + Split.second,
2551           ContentStartColumn);
2552       if (NewRemainingTokenColumns == 0) {
2553         // No content to indent.
2554         ContentIndent = 0;
2555         ContentStartColumn =
2556             Token->getContentStartColumn(LineIndex, /*Break=*/true);
2557         NewRemainingTokenColumns = Token->getRemainingLength(
2558             LineIndex, TailOffset + Split.first + Split.second,
2559             ContentStartColumn);
2560       }
2561 
2562       // When breaking before a tab character, it may be moved by a few columns,
2563       // but will still be expanded to the next tab stop, so we don't save any
2564       // columns.
2565       if (NewRemainingTokenColumns >= RemainingTokenColumns) {
2566         // FIXME: Do we need to adjust the penalty?
2567         break;
2568       }
2569 
2570       LLVM_DEBUG(llvm::dbgs() << "    Breaking at: " << TailOffset + Split.first
2571                               << ", " << Split.second << "\n");
2572       if (!DryRun) {
2573         Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent,
2574                            Whitespaces);
2575       }
2576 
2577       Penalty += NewBreakPenalty;
2578       TailOffset += Split.first + Split.second;
2579       RemainingTokenColumns = NewRemainingTokenColumns;
2580       BreakInserted = true;
2581       NewBreakBefore = true;
2582     }
2583     // In case there's another line, prepare the state for the start of the next
2584     // line.
2585     if (LineIndex + 1 != EndIndex) {
2586       unsigned NextLineIndex = LineIndex + 1;
2587       if (NewBreakBefore) {
2588         // After breaking a line, try to reflow the next line into the current
2589         // one once RemainingTokenColumns fits.
2590         TryReflow = true;
2591       }
2592       if (TryReflow) {
2593         // We decided that we want to try reflowing the next line into the
2594         // current one.
2595         // We will now adjust the state as if the reflow is successful (in
2596         // preparation for the next line), and see whether that works. If we
2597         // decide that we cannot reflow, we will later reset the state to the
2598         // start of the next line.
2599         Reflow = false;
2600         // As we did not continue breaking the line, RemainingTokenColumns is
2601         // known to fit after ContentStartColumn. Adapt ContentStartColumn to
2602         // the position at which we want to format the next line if we do
2603         // actually reflow.
2604         // When we reflow, we need to add a space between the end of the current
2605         // line and the next line's start column.
2606         ContentStartColumn += RemainingTokenColumns + 1;
2607         // Get the split that we need to reflow next logical line into the end
2608         // of the current one; the split will include any leading whitespace of
2609         // the next logical line.
2610         BreakableToken::Split SplitBeforeNext =
2611             Token->getReflowSplit(NextLineIndex, CommentPragmasRegex);
2612         LLVM_DEBUG(llvm::dbgs()
2613                    << "    Size of reflown text: " << ContentStartColumn
2614                    << "\n    Potential reflow split: ");
2615         if (SplitBeforeNext.first != StringRef::npos) {
2616           LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", "
2617                                   << SplitBeforeNext.second << "\n");
2618           TailOffset = SplitBeforeNext.first + SplitBeforeNext.second;
2619           // If the rest of the next line fits into the current line below the
2620           // column limit, we can safely reflow.
2621           RemainingTokenColumns = Token->getRemainingLength(
2622               NextLineIndex, TailOffset, ContentStartColumn);
2623           Reflow = true;
2624           if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
2625             LLVM_DEBUG(llvm::dbgs()
2626                        << "    Over limit after reflow, need: "
2627                        << (ContentStartColumn + RemainingTokenColumns)
2628                        << ", space: " << ColumnLimit
2629                        << ", reflown prefix: " << ContentStartColumn
2630                        << ", offset in line: " << TailOffset << "\n");
2631             // If the whole next line does not fit, try to find a point in
2632             // the next line at which we can break so that attaching the part
2633             // of the next line to that break point onto the current line is
2634             // below the column limit.
2635             BreakableToken::Split Split =
2636                 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit,
2637                                 ContentStartColumn, CommentPragmasRegex);
2638             if (Split.first == StringRef::npos) {
2639               LLVM_DEBUG(llvm::dbgs() << "    Did not find later break\n");
2640               Reflow = false;
2641             } else {
2642               // Check whether the first split point gets us below the column
2643               // limit. Note that we will execute this split below as part of
2644               // the normal token breaking and reflow logic within the line.
2645               unsigned ToSplitColumns = Token->getRangeLength(
2646                   NextLineIndex, TailOffset, Split.first, ContentStartColumn);
2647               if (ContentStartColumn + ToSplitColumns > ColumnLimit) {
2648                 LLVM_DEBUG(llvm::dbgs() << "    Next split protrudes, need: "
2649                                         << (ContentStartColumn + ToSplitColumns)
2650                                         << ", space: " << ColumnLimit);
2651                 unsigned ExcessCharactersPenalty =
2652                     (ContentStartColumn + ToSplitColumns - ColumnLimit) *
2653                     Style.PenaltyExcessCharacter;
2654                 if (NewBreakPenalty < ExcessCharactersPenalty)
2655                   Reflow = false;
2656               }
2657             }
2658           }
2659         } else {
2660           LLVM_DEBUG(llvm::dbgs() << "not found.\n");
2661         }
2662       }
2663       if (!Reflow) {
2664         // If we didn't reflow into the next line, the only space to consider is
2665         // the next logical line. Reset our state to match the start of the next
2666         // line.
2667         TailOffset = 0;
2668         ContentStartColumn =
2669             Token->getContentStartColumn(NextLineIndex, /*Break=*/false);
2670         RemainingTokenColumns = Token->getRemainingLength(
2671             NextLineIndex, TailOffset, ContentStartColumn);
2672         // Adapt the start of the token, for example indent.
2673         if (!DryRun)
2674           Token->adaptStartOfLine(NextLineIndex, Whitespaces);
2675       } else {
2676         // If we found a reflow split and have added a new break before the next
2677         // line, we are going to remove the line break at the start of the next
2678         // logical line. For example, here we'll add a new line break after
2679         // 'text', and subsequently delete the line break between 'that' and
2680         // 'reflows'.
2681         //   // some text that
2682         //   // reflows
2683         // ->
2684         //   // some text
2685         //   // that reflows
2686         // When adding the line break, we also added the penalty for it, so we
2687         // need to subtract that penalty again when we remove the line break due
2688         // to reflowing.
2689         if (NewBreakBefore) {
2690           assert(Penalty >= NewBreakPenalty);
2691           Penalty -= NewBreakPenalty;
2692         }
2693         if (!DryRun)
2694           Token->reflow(NextLineIndex, Whitespaces);
2695       }
2696     }
2697   }
2698 
2699   BreakableToken::Split SplitAfterLastLine =
2700       Token->getSplitAfterLastLine(TailOffset);
2701   if (SplitAfterLastLine.first != StringRef::npos) {
2702     LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n");
2703 
2704     // We add the last line's penalty here, since that line is going to be split
2705     // now.
2706     Penalty += Style.PenaltyExcessCharacter *
2707                (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
2708 
2709     if (!DryRun) {
2710       Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine,
2711                                             Whitespaces);
2712     }
2713     ContentStartColumn =
2714         Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true);
2715     RemainingTokenColumns = Token->getRemainingLength(
2716         Token->getLineCount() - 1,
2717         TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second,
2718         ContentStartColumn);
2719   }
2720 
2721   State.Column = ContentStartColumn + RemainingTokenColumns -
2722                  Current.UnbreakableTailLength;
2723 
2724   if (BreakInserted) {
2725     if (!DryRun)
2726       Token->updateAfterBroken(Whitespaces);
2727 
2728     // If we break the token inside a parameter list, we need to break before
2729     // the next parameter on all levels, so that the next parameter is clearly
2730     // visible. Line comments already introduce a break.
2731     if (Current.isNot(TT_LineComment))
2732       for (ParenState &Paren : State.Stack)
2733         Paren.BreakBeforeParameter = true;
2734 
2735     if (Current.is(TT_BlockComment))
2736       State.NoContinuation = true;
2737 
2738     State.Stack.back().LastSpace = StartColumn;
2739   }
2740 
2741   Token->updateNextToken(State);
2742 
2743   return {Penalty, Exceeded};
2744 }
2745 
2746 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const {
2747   // In preprocessor directives reserve two chars for trailing " \".
2748   return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0);
2749 }
2750 
2751 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) {
2752   const FormatToken &Current = *State.NextToken;
2753   if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral))
2754     return false;
2755   // We never consider raw string literals "multiline" for the purpose of
2756   // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased
2757   // (see TokenAnnotator::mustBreakBefore().
2758   if (Current.TokenText.starts_with("R\""))
2759     return false;
2760   if (Current.IsMultiline)
2761     return true;
2762   if (Current.getNextNonComment() &&
2763       Current.getNextNonComment()->isStringLiteral()) {
2764     return true; // Implicit concatenation.
2765   }
2766   if (Style.ColumnLimit != 0 && Style.BreakStringLiterals &&
2767       State.Column + Current.ColumnWidth + Current.UnbreakableTailLength >
2768           Style.ColumnLimit) {
2769     return true; // String will be split.
2770   }
2771   return false;
2772 }
2773 
2774 } // namespace format
2775 } // namespace clang
2776