1 //===--- UnwrappedLineFormatter.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 #include "UnwrappedLineFormatter.h"
10 #include "NamespaceEndCommentsFixer.h"
11 #include "WhitespaceManager.h"
12 #include "llvm/Support/Debug.h"
13 #include <queue>
14 
15 #define DEBUG_TYPE "format-formatter"
16 
17 namespace clang {
18 namespace format {
19 
20 namespace {
21 
22 bool startsExternCBlock(const AnnotatedLine &Line) {
23   const FormatToken *Next = Line.First->getNextNonComment();
24   const FormatToken *NextNext = Next ? Next->getNextNonComment() : nullptr;
25   return Line.startsWith(tok::kw_extern) && Next && Next->isStringLiteral() &&
26          NextNext && NextNext->is(tok::l_brace);
27 }
28 
29 /// Tracks the indent level of \c AnnotatedLines across levels.
30 ///
31 /// \c nextLine must be called for each \c AnnotatedLine, after which \c
32 /// getIndent() will return the indent for the last line \c nextLine was called
33 /// with.
34 /// If the line is not formatted (and thus the indent does not change), calling
35 /// \c adjustToUnmodifiedLine after the call to \c nextLine will cause
36 /// subsequent lines on the same level to be indented at the same level as the
37 /// given line.
38 class LevelIndentTracker {
39 public:
40   LevelIndentTracker(const FormatStyle &Style,
41                      const AdditionalKeywords &Keywords, unsigned StartLevel,
42                      int AdditionalIndent)
43       : Style(Style), Keywords(Keywords), AdditionalIndent(AdditionalIndent) {
44     for (unsigned i = 0; i != StartLevel; ++i)
45       IndentForLevel.push_back(Style.IndentWidth * i + AdditionalIndent);
46   }
47 
48   /// Returns the indent for the current line.
49   unsigned getIndent() const { return Indent; }
50 
51   /// Update the indent state given that \p Line is going to be formatted
52   /// next.
53   void nextLine(const AnnotatedLine &Line) {
54     Offset = getIndentOffset(*Line.First);
55     // Update the indent level cache size so that we can rely on it
56     // having the right size in adjustToUnmodifiedline.
57     while (IndentForLevel.size() <= Line.Level)
58       IndentForLevel.push_back(-1);
59     if (Line.InPPDirective) {
60       unsigned IndentWidth =
61           (Style.PPIndentWidth >= 0) ? Style.PPIndentWidth : Style.IndentWidth;
62       Indent = Line.Level * IndentWidth + AdditionalIndent;
63     } else {
64       IndentForLevel.resize(Line.Level + 1);
65       Indent = getIndent(Line.Level);
66     }
67     if (static_cast<int>(Indent) + Offset >= 0)
68       Indent += Offset;
69     if (Line.First->is(TT_CSharpGenericTypeConstraint))
70       Indent = Line.Level * Style.IndentWidth + Style.ContinuationIndentWidth;
71   }
72 
73   /// Update the indent state given that \p Line indent should be
74   /// skipped.
75   void skipLine(const AnnotatedLine &Line) {
76     while (IndentForLevel.size() <= Line.Level)
77       IndentForLevel.push_back(Indent);
78   }
79 
80   /// Update the level indent to adapt to the given \p Line.
81   ///
82   /// When a line is not formatted, we move the subsequent lines on the same
83   /// level to the same indent.
84   /// Note that \c nextLine must have been called before this method.
85   void adjustToUnmodifiedLine(const AnnotatedLine &Line) {
86     unsigned LevelIndent = Line.First->OriginalColumn;
87     if (static_cast<int>(LevelIndent) - Offset >= 0)
88       LevelIndent -= Offset;
89     if ((!Line.First->is(tok::comment) || IndentForLevel[Line.Level] == -1) &&
90         !Line.InPPDirective)
91       IndentForLevel[Line.Level] = LevelIndent;
92   }
93 
94 private:
95   /// Get the offset of the line relatively to the level.
96   ///
97   /// For example, 'public:' labels in classes are offset by 1 or 2
98   /// characters to the left from their level.
99   int getIndentOffset(const FormatToken &RootToken) {
100     if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() ||
101         Style.isCSharp())
102       return 0;
103     if (RootToken.isAccessSpecifier(false) ||
104         RootToken.isObjCAccessSpecifier() ||
105         (RootToken.isOneOf(Keywords.kw_signals, Keywords.kw_qsignals) &&
106          RootToken.Next && RootToken.Next->is(tok::colon))) {
107       // The AccessModifierOffset may be overridden by IndentAccessModifiers,
108       // in which case we take a negative value of the IndentWidth to simulate
109       // the upper indent level.
110       return Style.IndentAccessModifiers ? -Style.IndentWidth
111                                          : Style.AccessModifierOffset;
112     }
113     return 0;
114   }
115 
116   /// Get the indent of \p Level from \p IndentForLevel.
117   ///
118   /// \p IndentForLevel must contain the indent for the level \c l
119   /// at \p IndentForLevel[l], or a value < 0 if the indent for
120   /// that level is unknown.
121   unsigned getIndent(unsigned Level) const {
122     if (IndentForLevel[Level] != -1)
123       return IndentForLevel[Level];
124     if (Level == 0)
125       return 0;
126     return getIndent(Level - 1) + Style.IndentWidth;
127   }
128 
129   const FormatStyle &Style;
130   const AdditionalKeywords &Keywords;
131   const unsigned AdditionalIndent;
132 
133   /// The indent in characters for each level.
134   std::vector<int> IndentForLevel;
135 
136   /// Offset of the current line relative to the indent level.
137   ///
138   /// For example, the 'public' keywords is often indented with a negative
139   /// offset.
140   int Offset = 0;
141 
142   /// The current line's indent.
143   unsigned Indent = 0;
144 };
145 
146 const FormatToken *getMatchingNamespaceToken(
147     const AnnotatedLine *Line,
148     const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
149   if (!Line->startsWith(tok::r_brace))
150     return nullptr;
151   size_t StartLineIndex = Line->MatchingOpeningBlockLineIndex;
152   if (StartLineIndex == UnwrappedLine::kInvalidIndex)
153     return nullptr;
154   assert(StartLineIndex < AnnotatedLines.size());
155   return AnnotatedLines[StartLineIndex]->First->getNamespaceToken();
156 }
157 
158 StringRef getNamespaceTokenText(const AnnotatedLine *Line) {
159   const FormatToken *NamespaceToken = Line->First->getNamespaceToken();
160   return NamespaceToken ? NamespaceToken->TokenText : StringRef();
161 }
162 
163 StringRef getMatchingNamespaceTokenText(
164     const AnnotatedLine *Line,
165     const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
166   const FormatToken *NamespaceToken =
167       getMatchingNamespaceToken(Line, AnnotatedLines);
168   return NamespaceToken ? NamespaceToken->TokenText : StringRef();
169 }
170 
171 class LineJoiner {
172 public:
173   LineJoiner(const FormatStyle &Style, const AdditionalKeywords &Keywords,
174              const SmallVectorImpl<AnnotatedLine *> &Lines)
175       : Style(Style), Keywords(Keywords), End(Lines.end()), Next(Lines.begin()),
176         AnnotatedLines(Lines) {}
177 
178   /// Returns the next line, merging multiple lines into one if possible.
179   const AnnotatedLine *getNextMergedLine(bool DryRun,
180                                          LevelIndentTracker &IndentTracker) {
181     if (Next == End)
182       return nullptr;
183     const AnnotatedLine *Current = *Next;
184     IndentTracker.nextLine(*Current);
185     unsigned MergedLines = tryFitMultipleLinesInOne(IndentTracker, Next, End);
186     if (MergedLines > 0 && Style.ColumnLimit == 0)
187       // Disallow line merging if there is a break at the start of one of the
188       // input lines.
189       for (unsigned i = 0; i < MergedLines; ++i)
190         if (Next[i + 1]->First->NewlinesBefore > 0)
191           MergedLines = 0;
192     if (!DryRun)
193       for (unsigned i = 0; i < MergedLines; ++i)
194         join(*Next[0], *Next[i + 1]);
195     Next = Next + MergedLines + 1;
196     return Current;
197   }
198 
199 private:
200   /// Calculates how many lines can be merged into 1 starting at \p I.
201   unsigned
202   tryFitMultipleLinesInOne(LevelIndentTracker &IndentTracker,
203                            SmallVectorImpl<AnnotatedLine *>::const_iterator I,
204                            SmallVectorImpl<AnnotatedLine *>::const_iterator E) {
205     const unsigned Indent = IndentTracker.getIndent();
206 
207     // Can't join the last line with anything.
208     if (I + 1 == E)
209       return 0;
210     // We can never merge stuff if there are trailing line comments.
211     const AnnotatedLine *TheLine = *I;
212     if (TheLine->Last->is(TT_LineComment))
213       return 0;
214     if (I[1]->Type == LT_Invalid || I[1]->First->MustBreakBefore)
215       return 0;
216     if (TheLine->InPPDirective &&
217         (!I[1]->InPPDirective || I[1]->First->HasUnescapedNewline))
218       return 0;
219 
220     if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit)
221       return 0;
222 
223     unsigned Limit =
224         Style.ColumnLimit == 0 ? UINT_MAX : Style.ColumnLimit - Indent;
225     // If we already exceed the column limit, we set 'Limit' to 0. The different
226     // tryMerge..() functions can then decide whether to still do merging.
227     Limit = TheLine->Last->TotalLength > Limit
228                 ? 0
229                 : Limit - TheLine->Last->TotalLength;
230 
231     if (TheLine->Last->is(TT_FunctionLBrace) &&
232         TheLine->First == TheLine->Last &&
233         !Style.BraceWrapping.SplitEmptyFunction &&
234         I[1]->First->is(tok::r_brace))
235       return tryMergeSimpleBlock(I, E, Limit);
236 
237     // Handle empty record blocks where the brace has already been wrapped
238     if (TheLine->Last->is(tok::l_brace) && TheLine->First == TheLine->Last &&
239         I != AnnotatedLines.begin()) {
240       bool EmptyBlock = I[1]->First->is(tok::r_brace);
241 
242       const FormatToken *Tok = I[-1]->First;
243       if (Tok && Tok->is(tok::comment))
244         Tok = Tok->getNextNonComment();
245 
246       if (Tok && Tok->getNamespaceToken())
247         return !Style.BraceWrapping.SplitEmptyNamespace && EmptyBlock
248                    ? tryMergeSimpleBlock(I, E, Limit)
249                    : 0;
250 
251       if (Tok && Tok->is(tok::kw_typedef))
252         Tok = Tok->getNextNonComment();
253       if (Tok && Tok->isOneOf(tok::kw_class, tok::kw_struct, tok::kw_union,
254                               tok::kw_extern, Keywords.kw_interface))
255         return !Style.BraceWrapping.SplitEmptyRecord && EmptyBlock
256                    ? tryMergeSimpleBlock(I, E, Limit)
257                    : 0;
258 
259       if (Tok && Tok->is(tok::kw_template) &&
260           Style.BraceWrapping.SplitEmptyRecord && EmptyBlock) {
261         return 0;
262       }
263     }
264 
265     auto ShouldMergeShortFunctions =
266         [this, B = AnnotatedLines.begin()](
267             SmallVectorImpl<AnnotatedLine *>::const_iterator I) {
268           if (Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_All)
269             return true;
270           if (Style.AllowShortFunctionsOnASingleLine >=
271                   FormatStyle::SFS_Empty &&
272               I[1]->First->is(tok::r_brace))
273             return true;
274 
275           if (Style.AllowShortFunctionsOnASingleLine &
276               FormatStyle::SFS_InlineOnly) {
277             // Just checking TheLine->Level != 0 is not enough, because it
278             // provokes treating functions inside indented namespaces as short.
279             if ((*I)->Level != 0) {
280               if (I == B)
281                 return false;
282 
283               // TODO: Use IndentTracker to avoid loop?
284               // Find the last line with lower level.
285               auto J = I - 1;
286               for (; J != B; --J)
287                 if ((*J)->Level < (*I)->Level)
288                   break;
289 
290               // Check if the found line starts a record.
291               auto *RecordTok = (*J)->First;
292               while (RecordTok) {
293                 // TODO: Refactor to isRecord(RecordTok).
294                 if (RecordTok->isOneOf(tok::kw_class, tok::kw_struct))
295                   return true;
296                 if (Style.isCpp() && RecordTok->is(tok::kw_union))
297                   return true;
298                 if (Style.isCSharp() && RecordTok->is(Keywords.kw_interface))
299                   return true;
300                 if (Style.Language == FormatStyle::LK_Java &&
301                     RecordTok->is(tok::kw_enum))
302                   return true;
303                 if (Style.isJavaScript() && RecordTok->is(Keywords.kw_function))
304                   return true;
305 
306                 RecordTok = RecordTok->Next;
307               }
308 
309               return false;
310             }
311           }
312 
313           return false;
314         };
315 
316     bool MergeShortFunctions = ShouldMergeShortFunctions(I);
317 
318     if (Style.CompactNamespaces) {
319       if (auto nsToken = TheLine->First->getNamespaceToken()) {
320         int i = 0;
321         unsigned closingLine = TheLine->MatchingClosingBlockLineIndex - 1;
322         for (; I + 1 + i != E &&
323                nsToken->TokenText == getNamespaceTokenText(I[i + 1]) &&
324                closingLine == I[i + 1]->MatchingClosingBlockLineIndex &&
325                I[i + 1]->Last->TotalLength < Limit;
326              i++, closingLine--) {
327           // No extra indent for compacted namespaces
328           IndentTracker.skipLine(*I[i + 1]);
329 
330           Limit -= I[i + 1]->Last->TotalLength;
331         }
332         return i;
333       }
334 
335       if (auto nsToken = getMatchingNamespaceToken(TheLine, AnnotatedLines)) {
336         int i = 0;
337         unsigned openingLine = TheLine->MatchingOpeningBlockLineIndex - 1;
338         for (; I + 1 + i != E &&
339                nsToken->TokenText ==
340                    getMatchingNamespaceTokenText(I[i + 1], AnnotatedLines) &&
341                openingLine == I[i + 1]->MatchingOpeningBlockLineIndex;
342              i++, openingLine--) {
343           // No space between consecutive braces
344           I[i + 1]->First->SpacesRequiredBefore = !I[i]->Last->is(tok::r_brace);
345 
346           // Indent like the outer-most namespace
347           IndentTracker.nextLine(*I[i + 1]);
348         }
349         return i;
350       }
351     }
352 
353     // Try to merge a function block with left brace unwrapped
354     if (TheLine->Last->is(TT_FunctionLBrace) &&
355         TheLine->First != TheLine->Last) {
356       return MergeShortFunctions ? tryMergeSimpleBlock(I, E, Limit) : 0;
357     }
358     // Try to merge a control statement block with left brace unwrapped
359     if (TheLine->Last->is(tok::l_brace) && TheLine->First != TheLine->Last &&
360         TheLine->First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for)) {
361       return Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never
362                  ? tryMergeSimpleBlock(I, E, Limit)
363                  : 0;
364     }
365     // Try to merge a control statement block with left brace wrapped
366     if (I[1]->First->is(tok::l_brace) &&
367         (TheLine->First->isOneOf(tok::kw_if, tok::kw_else, tok::kw_while,
368                                  tok::kw_for, tok::kw_switch, tok::kw_try,
369                                  tok::kw_do, TT_ForEachMacro) ||
370          (TheLine->First->is(tok::r_brace) && TheLine->First->Next &&
371           TheLine->First->Next->isOneOf(tok::kw_else, tok::kw_catch))) &&
372         Style.BraceWrapping.AfterControlStatement ==
373             FormatStyle::BWACS_MultiLine) {
374       // If possible, merge the next line's wrapped left brace with the current
375       // line. Otherwise, leave it on the next line, as this is a multi-line
376       // control statement.
377       return (Style.ColumnLimit == 0 ||
378               TheLine->Last->TotalLength <= Style.ColumnLimit)
379                  ? 1
380                  : 0;
381     } else if (I[1]->First->is(tok::l_brace) &&
382                TheLine->First->isOneOf(tok::kw_if, tok::kw_else, tok::kw_while,
383                                        tok::kw_for)) {
384       return (Style.BraceWrapping.AfterControlStatement ==
385               FormatStyle::BWACS_Always)
386                  ? tryMergeSimpleBlock(I, E, Limit)
387                  : 0;
388     } else if (I[1]->First->is(tok::l_brace) &&
389                TheLine->First->isOneOf(tok::kw_else, tok::kw_catch) &&
390                Style.BraceWrapping.AfterControlStatement ==
391                    FormatStyle::BWACS_MultiLine) {
392       // This case if different from the upper BWACS_MultiLine processing
393       // in that a preceding r_brace is not on the same line as else/catch
394       // most likely because of BeforeElse/BeforeCatch set to true.
395       // If the line length doesn't fit ColumnLimit, leave l_brace on the
396       // next line to respect the BWACS_MultiLine.
397       return (Style.ColumnLimit == 0 ||
398               TheLine->Last->TotalLength <= Style.ColumnLimit)
399                  ? 1
400                  : 0;
401     }
402     // Don't merge block with left brace wrapped after ObjC special blocks
403     if (TheLine->First->is(tok::l_brace) && I != AnnotatedLines.begin() &&
404         I[-1]->First->is(tok::at) && I[-1]->First->Next) {
405       tok::ObjCKeywordKind kwId = I[-1]->First->Next->Tok.getObjCKeywordID();
406       if (kwId == clang::tok::objc_autoreleasepool ||
407           kwId == clang::tok::objc_synchronized)
408         return 0;
409     }
410     // Don't merge block with left brace wrapped after case labels
411     if (TheLine->First->is(tok::l_brace) && I != AnnotatedLines.begin() &&
412         I[-1]->First->isOneOf(tok::kw_case, tok::kw_default))
413       return 0;
414 
415     // Don't merge an empty template class or struct if SplitEmptyRecords
416     // is defined.
417     if (Style.BraceWrapping.SplitEmptyRecord &&
418         TheLine->Last->is(tok::l_brace) && I != AnnotatedLines.begin() &&
419         I[-1]->Last) {
420       const FormatToken *Previous = I[-1]->Last;
421       if (Previous) {
422         if (Previous->is(tok::comment))
423           Previous = Previous->getPreviousNonComment();
424         if (Previous) {
425           if (Previous->is(tok::greater) && !I[-1]->InPPDirective)
426             return 0;
427           if (Previous->is(tok::identifier)) {
428             const FormatToken *PreviousPrevious =
429                 Previous->getPreviousNonComment();
430             if (PreviousPrevious &&
431                 PreviousPrevious->isOneOf(tok::kw_class, tok::kw_struct))
432               return 0;
433           }
434         }
435       }
436     }
437 
438     // Try to merge a block with left brace unwrapped that wasn't yet covered
439     if (TheLine->Last->is(tok::l_brace)) {
440       const FormatToken *Tok = TheLine->First;
441       bool ShouldMerge = false;
442       if (Tok->is(tok::kw_typedef)) {
443         Tok = Tok->getNextNonComment();
444         assert(Tok);
445       }
446       if (Tok->isOneOf(tok::kw_class, tok::kw_struct)) {
447         ShouldMerge = !Style.BraceWrapping.AfterClass ||
448                       (I[1]->First->is(tok::r_brace) &&
449                        !Style.BraceWrapping.SplitEmptyRecord);
450       } else if (Tok->is(tok::kw_enum)) {
451         ShouldMerge = Style.AllowShortEnumsOnASingleLine;
452       } else {
453         ShouldMerge = !Style.BraceWrapping.AfterFunction ||
454                       (I[1]->First->is(tok::r_brace) &&
455                        !Style.BraceWrapping.SplitEmptyFunction);
456       }
457       return ShouldMerge ? tryMergeSimpleBlock(I, E, Limit) : 0;
458     }
459     // Try to merge a function block with left brace wrapped
460     if (I[1]->First->is(TT_FunctionLBrace) &&
461         Style.BraceWrapping.AfterFunction) {
462       if (I[1]->Last->is(TT_LineComment))
463         return 0;
464 
465       // Check for Limit <= 2 to account for the " {".
466       if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(TheLine)))
467         return 0;
468       Limit -= 2;
469 
470       unsigned MergedLines = 0;
471       if (MergeShortFunctions ||
472           (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
473            I[1]->First == I[1]->Last && I + 2 != E &&
474            I[2]->First->is(tok::r_brace))) {
475         MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
476         // If we managed to merge the block, count the function header, which is
477         // on a separate line.
478         if (MergedLines > 0)
479           ++MergedLines;
480       }
481       return MergedLines;
482     }
483     auto IsElseLine = [&TheLine]() -> bool {
484       const FormatToken *First = TheLine->First;
485       if (First->is(tok::kw_else))
486         return true;
487 
488       return First->is(tok::r_brace) && First->Next &&
489              First->Next->is(tok::kw_else);
490     };
491     if (TheLine->First->is(tok::kw_if) ||
492         (IsElseLine() && (Style.AllowShortIfStatementsOnASingleLine ==
493                           FormatStyle::SIS_AllIfsAndElse))) {
494       return Style.AllowShortIfStatementsOnASingleLine
495                  ? tryMergeSimpleControlStatement(I, E, Limit)
496                  : 0;
497     }
498     if (TheLine->First->isOneOf(tok::kw_for, tok::kw_while, tok::kw_do)) {
499       return Style.AllowShortLoopsOnASingleLine
500                  ? tryMergeSimpleControlStatement(I, E, Limit)
501                  : 0;
502     }
503     if (TheLine->First->isOneOf(tok::kw_case, tok::kw_default)) {
504       return Style.AllowShortCaseLabelsOnASingleLine
505                  ? tryMergeShortCaseLabels(I, E, Limit)
506                  : 0;
507     }
508     if (TheLine->InPPDirective &&
509         (TheLine->First->HasUnescapedNewline || TheLine->First->IsFirst)) {
510       return tryMergeSimplePPDirective(I, E, Limit);
511     }
512     return 0;
513   }
514 
515   unsigned
516   tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
517                             SmallVectorImpl<AnnotatedLine *>::const_iterator E,
518                             unsigned Limit) {
519     if (Limit == 0)
520       return 0;
521     if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline)
522       return 0;
523     if (1 + I[1]->Last->TotalLength > Limit)
524       return 0;
525     return 1;
526   }
527 
528   unsigned tryMergeSimpleControlStatement(
529       SmallVectorImpl<AnnotatedLine *>::const_iterator I,
530       SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) {
531     if (Limit == 0)
532       return 0;
533     if (Style.BraceWrapping.AfterControlStatement ==
534             FormatStyle::BWACS_Always &&
535         I[1]->First->is(tok::l_brace) &&
536         Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never)
537       return 0;
538     if (I[1]->InPPDirective != (*I)->InPPDirective ||
539         (I[1]->InPPDirective && I[1]->First->HasUnescapedNewline))
540       return 0;
541     Limit = limitConsideringMacros(I + 1, E, Limit);
542     AnnotatedLine &Line = **I;
543     if (!Line.First->is(tok::kw_do) && !Line.First->is(tok::kw_else) &&
544         !Line.Last->is(tok::kw_else) && Line.Last->isNot(tok::r_paren))
545       return 0;
546     // Only merge do while if do is the only statement on the line.
547     if (Line.First->is(tok::kw_do) && !Line.Last->is(tok::kw_do))
548       return 0;
549     if (1 + I[1]->Last->TotalLength > Limit)
550       return 0;
551     if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for, tok::kw_while,
552                              TT_LineComment))
553       return 0;
554     // Only inline simple if's (no nested if or else), unless specified
555     if (Style.AllowShortIfStatementsOnASingleLine ==
556         FormatStyle::SIS_WithoutElse) {
557       if (I + 2 != E && Line.startsWith(tok::kw_if) &&
558           I[2]->First->is(tok::kw_else))
559         return 0;
560     }
561     return 1;
562   }
563 
564   unsigned
565   tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
566                           SmallVectorImpl<AnnotatedLine *>::const_iterator E,
567                           unsigned Limit) {
568     if (Limit == 0 || I + 1 == E ||
569         I[1]->First->isOneOf(tok::kw_case, tok::kw_default))
570       return 0;
571     if (I[0]->Last->is(tok::l_brace) || I[1]->First->is(tok::l_brace))
572       return 0;
573     unsigned NumStmts = 0;
574     unsigned Length = 0;
575     bool EndsWithComment = false;
576     bool InPPDirective = I[0]->InPPDirective;
577     const unsigned Level = I[0]->Level;
578     for (; NumStmts < 3; ++NumStmts) {
579       if (I + 1 + NumStmts == E)
580         break;
581       const AnnotatedLine *Line = I[1 + NumStmts];
582       if (Line->InPPDirective != InPPDirective)
583         break;
584       if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
585         break;
586       if (Line->First->isOneOf(tok::kw_if, tok::kw_for, tok::kw_switch,
587                                tok::kw_while) ||
588           EndsWithComment)
589         return 0;
590       if (Line->First->is(tok::comment)) {
591         if (Level != Line->Level)
592           return 0;
593         SmallVectorImpl<AnnotatedLine *>::const_iterator J = I + 2 + NumStmts;
594         for (; J != E; ++J) {
595           Line = *J;
596           if (Line->InPPDirective != InPPDirective)
597             break;
598           if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
599             break;
600           if (Line->First->isNot(tok::comment) || Level != Line->Level)
601             return 0;
602         }
603         break;
604       }
605       if (Line->Last->is(tok::comment))
606         EndsWithComment = true;
607       Length += I[1 + NumStmts]->Last->TotalLength + 1; // 1 for the space.
608     }
609     if (NumStmts == 0 || NumStmts == 3 || Length > Limit)
610       return 0;
611     return NumStmts;
612   }
613 
614   unsigned
615   tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
616                       SmallVectorImpl<AnnotatedLine *>::const_iterator E,
617                       unsigned Limit) {
618     AnnotatedLine &Line = **I;
619 
620     // Don't merge ObjC @ keywords and methods.
621     // FIXME: If an option to allow short exception handling clauses on a single
622     // line is added, change this to not return for @try and friends.
623     if (Style.Language != FormatStyle::LK_Java &&
624         Line.First->isOneOf(tok::at, tok::minus, tok::plus))
625       return 0;
626 
627     // Check that the current line allows merging. This depends on whether we
628     // are in a control flow statements as well as several style flags.
629     if (Line.First->is(tok::kw_case) ||
630         (Line.First->Next && Line.First->Next->is(tok::kw_else)))
631       return 0;
632     // default: in switch statement
633     if (Line.First->is(tok::kw_default)) {
634       const FormatToken *Tok = Line.First->getNextNonComment();
635       if (Tok && Tok->is(tok::colon))
636         return 0;
637     }
638     if (Line.First->isOneOf(tok::kw_if, tok::kw_else, tok::kw_while, tok::kw_do,
639                             tok::kw_try, tok::kw___try, tok::kw_catch,
640                             tok::kw___finally, tok::kw_for, tok::r_brace,
641                             Keywords.kw___except)) {
642       if (Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never)
643         return 0;
644       if (Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Empty &&
645           !I[1]->First->is(tok::r_brace))
646         return 0;
647       // Don't merge when we can't except the case when
648       // the control statement block is empty
649       if (!Style.AllowShortIfStatementsOnASingleLine &&
650           Line.First->isOneOf(tok::kw_if, tok::kw_else) &&
651           !Style.BraceWrapping.AfterControlStatement &&
652           !I[1]->First->is(tok::r_brace))
653         return 0;
654       if (!Style.AllowShortIfStatementsOnASingleLine &&
655           Line.First->isOneOf(tok::kw_if, tok::kw_else) &&
656           Style.BraceWrapping.AfterControlStatement ==
657               FormatStyle::BWACS_Always &&
658           I + 2 != E && !I[2]->First->is(tok::r_brace))
659         return 0;
660       if (!Style.AllowShortLoopsOnASingleLine &&
661           Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for) &&
662           !Style.BraceWrapping.AfterControlStatement &&
663           !I[1]->First->is(tok::r_brace))
664         return 0;
665       if (!Style.AllowShortLoopsOnASingleLine &&
666           Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for) &&
667           Style.BraceWrapping.AfterControlStatement ==
668               FormatStyle::BWACS_Always &&
669           I + 2 != E && !I[2]->First->is(tok::r_brace))
670         return 0;
671       // FIXME: Consider an option to allow short exception handling clauses on
672       // a single line.
673       // FIXME: This isn't covered by tests.
674       // FIXME: For catch, __except, __finally the first token on the line
675       // is '}', so this isn't correct here.
676       if (Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch,
677                               Keywords.kw___except, tok::kw___finally))
678         return 0;
679     }
680 
681     if (Line.Last->is(tok::l_brace)) {
682       FormatToken *Tok = I[1]->First;
683       if (Tok->is(tok::r_brace) && !Tok->MustBreakBefore &&
684           (Tok->getNextNonComment() == nullptr ||
685            Tok->getNextNonComment()->is(tok::semi))) {
686         // We merge empty blocks even if the line exceeds the column limit.
687         Tok->SpacesRequiredBefore = Style.SpaceInEmptyBlock ? 1 : 0;
688         Tok->CanBreakBefore = true;
689         return 1;
690       } else if (Limit != 0 && !Line.startsWithNamespace() &&
691                  !startsExternCBlock(Line)) {
692         // We don't merge short records.
693         FormatToken *RecordTok = Line.First;
694         // Skip record modifiers.
695         while (RecordTok->Next &&
696                RecordTok->isOneOf(tok::kw_typedef, tok::kw_export,
697                                   Keywords.kw_declare, Keywords.kw_abstract,
698                                   tok::kw_default, Keywords.kw_override,
699                                   tok::kw_public, tok::kw_private,
700                                   tok::kw_protected, Keywords.kw_internal))
701           RecordTok = RecordTok->Next;
702         if (RecordTok &&
703             RecordTok->isOneOf(tok::kw_class, tok::kw_union, tok::kw_struct,
704                                Keywords.kw_interface))
705           return 0;
706 
707         // Check that we still have three lines and they fit into the limit.
708         if (I + 2 == E || I[2]->Type == LT_Invalid)
709           return 0;
710         Limit = limitConsideringMacros(I + 2, E, Limit);
711 
712         if (!nextTwoLinesFitInto(I, Limit))
713           return 0;
714 
715         // Second, check that the next line does not contain any braces - if it
716         // does, readability declines when putting it into a single line.
717         if (I[1]->Last->is(TT_LineComment))
718           return 0;
719         do {
720           if (Tok->is(tok::l_brace) && Tok->isNot(BK_BracedInit))
721             return 0;
722           Tok = Tok->Next;
723         } while (Tok);
724 
725         // Last, check that the third line starts with a closing brace.
726         Tok = I[2]->First;
727         if (Tok->isNot(tok::r_brace))
728           return 0;
729 
730         // Don't merge "if (a) { .. } else {".
731         if (Tok->Next && Tok->Next->is(tok::kw_else))
732           return 0;
733 
734         // Don't merge a trailing multi-line control statement block like:
735         // } else if (foo &&
736         //            bar)
737         // { <-- current Line
738         //   baz();
739         // }
740         if (Line.First == Line.Last && Line.First->isNot(TT_FunctionLBrace) &&
741             Style.BraceWrapping.AfterControlStatement ==
742                 FormatStyle::BWACS_MultiLine)
743           return 0;
744 
745         return 2;
746       }
747     } else if (I[1]->First->is(tok::l_brace)) {
748       if (I[1]->Last->is(TT_LineComment))
749         return 0;
750 
751       // Check for Limit <= 2 to account for the " {".
752       if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(*I)))
753         return 0;
754       Limit -= 2;
755       unsigned MergedLines = 0;
756       if (Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never ||
757           (I[1]->First == I[1]->Last && I + 2 != E &&
758            I[2]->First->is(tok::r_brace))) {
759         MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
760         // If we managed to merge the block, count the statement header, which
761         // is on a separate line.
762         if (MergedLines > 0)
763           ++MergedLines;
764       }
765       return MergedLines;
766     }
767     return 0;
768   }
769 
770   /// Returns the modified column limit for \p I if it is inside a macro and
771   /// needs a trailing '\'.
772   unsigned
773   limitConsideringMacros(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
774                          SmallVectorImpl<AnnotatedLine *>::const_iterator E,
775                          unsigned Limit) {
776     if (I[0]->InPPDirective && I + 1 != E &&
777         !I[1]->First->HasUnescapedNewline && !I[1]->First->is(tok::eof)) {
778       return Limit < 2 ? 0 : Limit - 2;
779     }
780     return Limit;
781   }
782 
783   bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
784                            unsigned Limit) {
785     if (I[1]->First->MustBreakBefore || I[2]->First->MustBreakBefore)
786       return false;
787     return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit;
788   }
789 
790   bool containsMustBreak(const AnnotatedLine *Line) {
791     for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) {
792       if (Tok->MustBreakBefore)
793         return true;
794     }
795     return false;
796   }
797 
798   void join(AnnotatedLine &A, const AnnotatedLine &B) {
799     assert(!A.Last->Next);
800     assert(!B.First->Previous);
801     if (B.Affected)
802       A.Affected = true;
803     A.Last->Next = B.First;
804     B.First->Previous = A.Last;
805     B.First->CanBreakBefore = true;
806     unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore;
807     for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) {
808       Tok->TotalLength += LengthA;
809       A.Last = Tok;
810     }
811   }
812 
813   const FormatStyle &Style;
814   const AdditionalKeywords &Keywords;
815   const SmallVectorImpl<AnnotatedLine *>::const_iterator End;
816 
817   SmallVectorImpl<AnnotatedLine *>::const_iterator Next;
818   const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines;
819 };
820 
821 static void markFinalized(FormatToken *Tok) {
822   for (; Tok; Tok = Tok->Next) {
823     Tok->Finalized = true;
824     for (AnnotatedLine *Child : Tok->Children)
825       markFinalized(Child->First);
826   }
827 }
828 
829 #ifndef NDEBUG
830 static void printLineState(const LineState &State) {
831   llvm::dbgs() << "State: ";
832   for (const ParenState &P : State.Stack) {
833     llvm::dbgs() << (P.Tok ? P.Tok->TokenText : "F") << "|" << P.Indent << "|"
834                  << P.LastSpace << "|" << P.NestedBlockIndent << " ";
835   }
836   llvm::dbgs() << State.NextToken->TokenText << "\n";
837 }
838 #endif
839 
840 /// Base class for classes that format one \c AnnotatedLine.
841 class LineFormatter {
842 public:
843   LineFormatter(ContinuationIndenter *Indenter, WhitespaceManager *Whitespaces,
844                 const FormatStyle &Style,
845                 UnwrappedLineFormatter *BlockFormatter)
846       : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style),
847         BlockFormatter(BlockFormatter) {}
848   virtual ~LineFormatter() {}
849 
850   /// Formats an \c AnnotatedLine and returns the penalty.
851   ///
852   /// If \p DryRun is \c false, directly applies the changes.
853   virtual unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
854                               unsigned FirstStartColumn, bool DryRun) = 0;
855 
856 protected:
857   /// If the \p State's next token is an r_brace closing a nested block,
858   /// format the nested block before it.
859   ///
860   /// Returns \c true if all children could be placed successfully and adapts
861   /// \p Penalty as well as \p State. If \p DryRun is false, also directly
862   /// creates changes using \c Whitespaces.
863   ///
864   /// The crucial idea here is that children always get formatted upon
865   /// encountering the closing brace right after the nested block. Now, if we
866   /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is
867   /// \c false), the entire block has to be kept on the same line (which is only
868   /// possible if it fits on the line, only contains a single statement, etc.
869   ///
870   /// If \p NewLine is true, we format the nested block on separate lines, i.e.
871   /// break after the "{", format all lines with correct indentation and the put
872   /// the closing "}" on yet another new line.
873   ///
874   /// This enables us to keep the simple structure of the
875   /// \c UnwrappedLineFormatter, where we only have two options for each token:
876   /// break or don't break.
877   bool formatChildren(LineState &State, bool NewLine, bool DryRun,
878                       unsigned &Penalty) {
879     const FormatToken *LBrace = State.NextToken->getPreviousNonComment();
880     FormatToken &Previous = *State.NextToken->Previous;
881     if (!LBrace || LBrace->isNot(tok::l_brace) || LBrace->isNot(BK_Block) ||
882         Previous.Children.size() == 0)
883       // The previous token does not open a block. Nothing to do. We don't
884       // assert so that we can simply call this function for all tokens.
885       return true;
886 
887     if (NewLine) {
888       const ParenState &P = State.Stack.back();
889 
890       int AdditionalIndent =
891           P.Indent - Previous.Children[0]->Level * Style.IndentWidth;
892 
893       if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
894           P.NestedBlockIndent == P.LastSpace) {
895         if (State.NextToken->MatchingParen &&
896             State.NextToken->MatchingParen->is(TT_LambdaLBrace)) {
897           State.Stack.pop_back();
898         }
899         if (LBrace->is(TT_LambdaLBrace))
900           AdditionalIndent = 0;
901       }
902 
903       Penalty +=
904           BlockFormatter->format(Previous.Children, DryRun, AdditionalIndent,
905                                  /*FixBadIndentation=*/true);
906       return true;
907     }
908 
909     if (Previous.Children[0]->First->MustBreakBefore)
910       return false;
911 
912     // Cannot merge into one line if this line ends on a comment.
913     if (Previous.is(tok::comment))
914       return false;
915 
916     // Cannot merge multiple statements into a single line.
917     if (Previous.Children.size() > 1)
918       return false;
919 
920     const AnnotatedLine *Child = Previous.Children[0];
921     // We can't put the closing "}" on a line with a trailing comment.
922     if (Child->Last->isTrailingComment())
923       return false;
924 
925     // If the child line exceeds the column limit, we wouldn't want to merge it.
926     // We add +2 for the trailing " }".
927     if (Style.ColumnLimit > 0 &&
928         Child->Last->TotalLength + State.Column + 2 > Style.ColumnLimit)
929       return false;
930 
931     if (!DryRun) {
932       Whitespaces->replaceWhitespace(
933           *Child->First, /*Newlines=*/0, /*Spaces=*/1,
934           /*StartOfTokenColumn=*/State.Column, /*IsAligned=*/false,
935           State.Line->InPPDirective);
936     }
937     Penalty +=
938         formatLine(*Child, State.Column + 1, /*FirstStartColumn=*/0, DryRun);
939 
940     State.Column += 1 + Child->Last->TotalLength;
941     return true;
942   }
943 
944   ContinuationIndenter *Indenter;
945 
946 private:
947   WhitespaceManager *Whitespaces;
948   const FormatStyle &Style;
949   UnwrappedLineFormatter *BlockFormatter;
950 };
951 
952 /// Formatter that keeps the existing line breaks.
953 class NoColumnLimitLineFormatter : public LineFormatter {
954 public:
955   NoColumnLimitLineFormatter(ContinuationIndenter *Indenter,
956                              WhitespaceManager *Whitespaces,
957                              const FormatStyle &Style,
958                              UnwrappedLineFormatter *BlockFormatter)
959       : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
960 
961   /// Formats the line, simply keeping all of the input's line breaking
962   /// decisions.
963   unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
964                       unsigned FirstStartColumn, bool DryRun) override {
965     assert(!DryRun);
966     LineState State = Indenter->getInitialState(FirstIndent, FirstStartColumn,
967                                                 &Line, /*DryRun=*/false);
968     while (State.NextToken) {
969       bool Newline =
970           Indenter->mustBreak(State) ||
971           (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0);
972       unsigned Penalty = 0;
973       formatChildren(State, Newline, /*DryRun=*/false, Penalty);
974       Indenter->addTokenToState(State, Newline, /*DryRun=*/false);
975     }
976     return 0;
977   }
978 };
979 
980 /// Formatter that puts all tokens into a single line without breaks.
981 class NoLineBreakFormatter : public LineFormatter {
982 public:
983   NoLineBreakFormatter(ContinuationIndenter *Indenter,
984                        WhitespaceManager *Whitespaces, const FormatStyle &Style,
985                        UnwrappedLineFormatter *BlockFormatter)
986       : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
987 
988   /// Puts all tokens into a single line.
989   unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
990                       unsigned FirstStartColumn, bool DryRun) override {
991     unsigned Penalty = 0;
992     LineState State =
993         Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun);
994     while (State.NextToken) {
995       formatChildren(State, /*NewLine=*/false, DryRun, Penalty);
996       Indenter->addTokenToState(
997           State, /*Newline=*/State.NextToken->MustBreakBefore, DryRun);
998     }
999     return Penalty;
1000   }
1001 };
1002 
1003 /// Finds the best way to break lines.
1004 class OptimizingLineFormatter : public LineFormatter {
1005 public:
1006   OptimizingLineFormatter(ContinuationIndenter *Indenter,
1007                           WhitespaceManager *Whitespaces,
1008                           const FormatStyle &Style,
1009                           UnwrappedLineFormatter *BlockFormatter)
1010       : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
1011 
1012   /// Formats the line by finding the best line breaks with line lengths
1013   /// below the column limit.
1014   unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
1015                       unsigned FirstStartColumn, bool DryRun) override {
1016     LineState State =
1017         Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun);
1018 
1019     // If the ObjC method declaration does not fit on a line, we should format
1020     // it with one arg per line.
1021     if (State.Line->Type == LT_ObjCMethodDecl)
1022       State.Stack.back().BreakBeforeParameter = true;
1023 
1024     // Find best solution in solution space.
1025     return analyzeSolutionSpace(State, DryRun);
1026   }
1027 
1028 private:
1029   struct CompareLineStatePointers {
1030     bool operator()(LineState *obj1, LineState *obj2) const {
1031       return *obj1 < *obj2;
1032     }
1033   };
1034 
1035   /// A pair of <penalty, count> that is used to prioritize the BFS on.
1036   ///
1037   /// In case of equal penalties, we want to prefer states that were inserted
1038   /// first. During state generation we make sure that we insert states first
1039   /// that break the line as late as possible.
1040   typedef std::pair<unsigned, unsigned> OrderedPenalty;
1041 
1042   /// An edge in the solution space from \c Previous->State to \c State,
1043   /// inserting a newline dependent on the \c NewLine.
1044   struct StateNode {
1045     StateNode(const LineState &State, bool NewLine, StateNode *Previous)
1046         : State(State), NewLine(NewLine), Previous(Previous) {}
1047     LineState State;
1048     bool NewLine;
1049     StateNode *Previous;
1050   };
1051 
1052   /// An item in the prioritized BFS search queue. The \c StateNode's
1053   /// \c State has the given \c OrderedPenalty.
1054   typedef std::pair<OrderedPenalty, StateNode *> QueueItem;
1055 
1056   /// The BFS queue type.
1057   typedef std::priority_queue<QueueItem, std::vector<QueueItem>,
1058                               std::greater<QueueItem>>
1059       QueueType;
1060 
1061   /// Analyze the entire solution space starting from \p InitialState.
1062   ///
1063   /// This implements a variant of Dijkstra's algorithm on the graph that spans
1064   /// the solution space (\c LineStates are the nodes). The algorithm tries to
1065   /// find the shortest path (the one with lowest penalty) from \p InitialState
1066   /// to a state where all tokens are placed. Returns the penalty.
1067   ///
1068   /// If \p DryRun is \c false, directly applies the changes.
1069   unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun) {
1070     std::set<LineState *, CompareLineStatePointers> Seen;
1071 
1072     // Increasing count of \c StateNode items we have created. This is used to
1073     // create a deterministic order independent of the container.
1074     unsigned Count = 0;
1075     QueueType Queue;
1076 
1077     // Insert start element into queue.
1078     StateNode *RootNode =
1079         new (Allocator.Allocate()) StateNode(InitialState, false, nullptr);
1080     Queue.push(QueueItem(OrderedPenalty(0, Count), RootNode));
1081     ++Count;
1082 
1083     unsigned Penalty = 0;
1084 
1085     // While not empty, take first element and follow edges.
1086     while (!Queue.empty()) {
1087       Penalty = Queue.top().first.first;
1088       StateNode *Node = Queue.top().second;
1089       if (!Node->State.NextToken) {
1090         LLVM_DEBUG(llvm::dbgs()
1091                    << "\n---\nPenalty for line: " << Penalty << "\n");
1092         break;
1093       }
1094       Queue.pop();
1095 
1096       // Cut off the analysis of certain solutions if the analysis gets too
1097       // complex. See description of IgnoreStackForComparison.
1098       if (Count > 50000)
1099         Node->State.IgnoreStackForComparison = true;
1100 
1101       if (!Seen.insert(&Node->State).second)
1102         // State already examined with lower penalty.
1103         continue;
1104 
1105       FormatDecision LastFormat = Node->State.NextToken->getDecision();
1106       if (LastFormat == FD_Unformatted || LastFormat == FD_Continue)
1107         addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue);
1108       if (LastFormat == FD_Unformatted || LastFormat == FD_Break)
1109         addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue);
1110     }
1111 
1112     if (Queue.empty()) {
1113       // We were unable to find a solution, do nothing.
1114       // FIXME: Add diagnostic?
1115       LLVM_DEBUG(llvm::dbgs() << "Could not find a solution.\n");
1116       return 0;
1117     }
1118 
1119     // Reconstruct the solution.
1120     if (!DryRun)
1121       reconstructPath(InitialState, Queue.top().second);
1122 
1123     LLVM_DEBUG(llvm::dbgs()
1124                << "Total number of analyzed states: " << Count << "\n");
1125     LLVM_DEBUG(llvm::dbgs() << "---\n");
1126 
1127     return Penalty;
1128   }
1129 
1130   /// Add the following state to the analysis queue \c Queue.
1131   ///
1132   /// Assume the current state is \p PreviousNode and has been reached with a
1133   /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true.
1134   void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode,
1135                            bool NewLine, unsigned *Count, QueueType *Queue) {
1136     if (NewLine && !Indenter->canBreak(PreviousNode->State))
1137       return;
1138     if (!NewLine && Indenter->mustBreak(PreviousNode->State))
1139       return;
1140 
1141     StateNode *Node = new (Allocator.Allocate())
1142         StateNode(PreviousNode->State, NewLine, PreviousNode);
1143     if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty))
1144       return;
1145 
1146     Penalty += Indenter->addTokenToState(Node->State, NewLine, true);
1147 
1148     Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node));
1149     ++(*Count);
1150   }
1151 
1152   /// Applies the best formatting by reconstructing the path in the
1153   /// solution space that leads to \c Best.
1154   void reconstructPath(LineState &State, StateNode *Best) {
1155     llvm::SmallVector<StateNode *> Path;
1156     // We do not need a break before the initial token.
1157     while (Best->Previous) {
1158       Path.push_back(Best);
1159       Best = Best->Previous;
1160     }
1161     for (const auto &Node : llvm::reverse(Path)) {
1162       unsigned Penalty = 0;
1163       formatChildren(State, Node->NewLine, /*DryRun=*/false, Penalty);
1164       Penalty += Indenter->addTokenToState(State, Node->NewLine, false);
1165 
1166       LLVM_DEBUG({
1167         printLineState(Node->Previous->State);
1168         if (Node->NewLine) {
1169           llvm::dbgs() << "Penalty for placing "
1170                        << Node->Previous->State.NextToken->Tok.getName()
1171                        << " on a new line: " << Penalty << "\n";
1172         }
1173       });
1174     }
1175   }
1176 
1177   llvm::SpecificBumpPtrAllocator<StateNode> Allocator;
1178 };
1179 
1180 } // anonymous namespace
1181 
1182 unsigned UnwrappedLineFormatter::format(
1183     const SmallVectorImpl<AnnotatedLine *> &Lines, bool DryRun,
1184     int AdditionalIndent, bool FixBadIndentation, unsigned FirstStartColumn,
1185     unsigned NextStartColumn, unsigned LastStartColumn) {
1186   LineJoiner Joiner(Style, Keywords, Lines);
1187 
1188   // Try to look up already computed penalty in DryRun-mode.
1189   std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned> CacheKey(
1190       &Lines, AdditionalIndent);
1191   auto CacheIt = PenaltyCache.find(CacheKey);
1192   if (DryRun && CacheIt != PenaltyCache.end())
1193     return CacheIt->second;
1194 
1195   assert(!Lines.empty());
1196   unsigned Penalty = 0;
1197   LevelIndentTracker IndentTracker(Style, Keywords, Lines[0]->Level,
1198                                    AdditionalIndent);
1199   const AnnotatedLine *PrevPrevLine = nullptr;
1200   const AnnotatedLine *PreviousLine = nullptr;
1201   const AnnotatedLine *NextLine = nullptr;
1202 
1203   // The minimum level of consecutive lines that have been formatted.
1204   unsigned RangeMinLevel = UINT_MAX;
1205 
1206   bool FirstLine = true;
1207   for (const AnnotatedLine *Line =
1208            Joiner.getNextMergedLine(DryRun, IndentTracker);
1209        Line; PrevPrevLine = PreviousLine, PreviousLine = Line, Line = NextLine,
1210                            FirstLine = false) {
1211     const AnnotatedLine &TheLine = *Line;
1212     unsigned Indent = IndentTracker.getIndent();
1213 
1214     // We continue formatting unchanged lines to adjust their indent, e.g. if a
1215     // scope was added. However, we need to carefully stop doing this when we
1216     // exit the scope of affected lines to prevent indenting a the entire
1217     // remaining file if it currently missing a closing brace.
1218     bool PreviousRBrace =
1219         PreviousLine && PreviousLine->startsWith(tok::r_brace);
1220     bool ContinueFormatting =
1221         TheLine.Level > RangeMinLevel ||
1222         (TheLine.Level == RangeMinLevel && !PreviousRBrace &&
1223          !TheLine.startsWith(tok::r_brace));
1224 
1225     bool FixIndentation = (FixBadIndentation || ContinueFormatting) &&
1226                           Indent != TheLine.First->OriginalColumn;
1227     bool ShouldFormat = TheLine.Affected || FixIndentation;
1228     // We cannot format this line; if the reason is that the line had a
1229     // parsing error, remember that.
1230     if (ShouldFormat && TheLine.Type == LT_Invalid && Status) {
1231       Status->FormatComplete = false;
1232       Status->Line =
1233           SourceMgr.getSpellingLineNumber(TheLine.First->Tok.getLocation());
1234     }
1235 
1236     if (ShouldFormat && TheLine.Type != LT_Invalid) {
1237       if (!DryRun) {
1238         bool LastLine = Line->First->is(tok::eof);
1239         formatFirstToken(TheLine, PreviousLine, PrevPrevLine, Lines, Indent,
1240                          LastLine ? LastStartColumn : NextStartColumn + Indent);
1241       }
1242 
1243       NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
1244       unsigned ColumnLimit = getColumnLimit(TheLine.InPPDirective, NextLine);
1245       bool FitsIntoOneLine =
1246           TheLine.Last->TotalLength + Indent <= ColumnLimit ||
1247           (TheLine.Type == LT_ImportStatement &&
1248            (!Style.isJavaScript() || !Style.JavaScriptWrapImports)) ||
1249           (Style.isCSharp() &&
1250            TheLine.InPPDirective); // don't split #regions in C#
1251       if (Style.ColumnLimit == 0)
1252         NoColumnLimitLineFormatter(Indenter, Whitespaces, Style, this)
1253             .formatLine(TheLine, NextStartColumn + Indent,
1254                         FirstLine ? FirstStartColumn : 0, DryRun);
1255       else if (FitsIntoOneLine)
1256         Penalty += NoLineBreakFormatter(Indenter, Whitespaces, Style, this)
1257                        .formatLine(TheLine, NextStartColumn + Indent,
1258                                    FirstLine ? FirstStartColumn : 0, DryRun);
1259       else
1260         Penalty += OptimizingLineFormatter(Indenter, Whitespaces, Style, this)
1261                        .formatLine(TheLine, NextStartColumn + Indent,
1262                                    FirstLine ? FirstStartColumn : 0, DryRun);
1263       RangeMinLevel = std::min(RangeMinLevel, TheLine.Level);
1264     } else {
1265       // If no token in the current line is affected, we still need to format
1266       // affected children.
1267       if (TheLine.ChildrenAffected)
1268         for (const FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next)
1269           if (!Tok->Children.empty())
1270             format(Tok->Children, DryRun);
1271 
1272       // Adapt following lines on the current indent level to the same level
1273       // unless the current \c AnnotatedLine is not at the beginning of a line.
1274       bool StartsNewLine =
1275           TheLine.First->NewlinesBefore > 0 || TheLine.First->IsFirst;
1276       if (StartsNewLine)
1277         IndentTracker.adjustToUnmodifiedLine(TheLine);
1278       if (!DryRun) {
1279         bool ReformatLeadingWhitespace =
1280             StartsNewLine && ((PreviousLine && PreviousLine->Affected) ||
1281                               TheLine.LeadingEmptyLinesAffected);
1282         // Format the first token.
1283         if (ReformatLeadingWhitespace)
1284           formatFirstToken(TheLine, PreviousLine, PrevPrevLine, Lines,
1285                            TheLine.First->OriginalColumn,
1286                            TheLine.First->OriginalColumn);
1287         else
1288           Whitespaces->addUntouchableToken(*TheLine.First,
1289                                            TheLine.InPPDirective);
1290 
1291         // Notify the WhitespaceManager about the unchanged whitespace.
1292         for (FormatToken *Tok = TheLine.First->Next; Tok; Tok = Tok->Next)
1293           Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective);
1294       }
1295       NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
1296       RangeMinLevel = UINT_MAX;
1297     }
1298     if (!DryRun)
1299       markFinalized(TheLine.First);
1300   }
1301   PenaltyCache[CacheKey] = Penalty;
1302   return Penalty;
1303 }
1304 
1305 void UnwrappedLineFormatter::formatFirstToken(
1306     const AnnotatedLine &Line, const AnnotatedLine *PreviousLine,
1307     const AnnotatedLine *PrevPrevLine,
1308     const SmallVectorImpl<AnnotatedLine *> &Lines, unsigned Indent,
1309     unsigned NewlineIndent) {
1310   FormatToken &RootToken = *Line.First;
1311   if (RootToken.is(tok::eof)) {
1312     unsigned Newlines = std::min(RootToken.NewlinesBefore, 1u);
1313     unsigned TokenIndent = Newlines ? NewlineIndent : 0;
1314     Whitespaces->replaceWhitespace(RootToken, Newlines, TokenIndent,
1315                                    TokenIndent);
1316     return;
1317   }
1318   unsigned Newlines =
1319       std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1);
1320   // Remove empty lines before "}" where applicable.
1321   if (RootToken.is(tok::r_brace) &&
1322       (!RootToken.Next ||
1323        (RootToken.Next->is(tok::semi) && !RootToken.Next->Next)) &&
1324       // Do not remove empty lines before namespace closing "}".
1325       !getNamespaceToken(&Line, Lines))
1326     Newlines = std::min(Newlines, 1u);
1327   // Remove empty lines at the start of nested blocks (lambdas/arrow functions)
1328   if (PreviousLine == nullptr && Line.Level > 0)
1329     Newlines = std::min(Newlines, 1u);
1330   if (Newlines == 0 && !RootToken.IsFirst)
1331     Newlines = 1;
1332   if (RootToken.IsFirst && !RootToken.HasUnescapedNewline)
1333     Newlines = 0;
1334 
1335   // Remove empty lines after "{".
1336   if (!Style.KeepEmptyLinesAtTheStartOfBlocks && PreviousLine &&
1337       PreviousLine->Last->is(tok::l_brace) &&
1338       !PreviousLine->startsWithNamespace() &&
1339       !(PrevPrevLine && PrevPrevLine->startsWithNamespace() &&
1340         PreviousLine->startsWith(tok::l_brace)) &&
1341       !startsExternCBlock(*PreviousLine))
1342     Newlines = 1;
1343 
1344   // Insert or remove empty line before access specifiers.
1345   if (PreviousLine && RootToken.isAccessSpecifier()) {
1346     switch (Style.EmptyLineBeforeAccessModifier) {
1347     case FormatStyle::ELBAMS_Never:
1348       if (Newlines > 1)
1349         Newlines = 1;
1350       break;
1351     case FormatStyle::ELBAMS_Leave:
1352       Newlines = std::max(RootToken.NewlinesBefore, 1u);
1353       break;
1354     case FormatStyle::ELBAMS_LogicalBlock:
1355       if (PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) && Newlines <= 1)
1356         Newlines = 2;
1357       if (PreviousLine->First->isAccessSpecifier())
1358         Newlines = 1; // Previous is an access modifier remove all new lines.
1359       break;
1360     case FormatStyle::ELBAMS_Always: {
1361       const FormatToken *previousToken;
1362       if (PreviousLine->Last->is(tok::comment))
1363         previousToken = PreviousLine->Last->getPreviousNonComment();
1364       else
1365         previousToken = PreviousLine->Last;
1366       if ((!previousToken || !previousToken->is(tok::l_brace)) && Newlines <= 1)
1367         Newlines = 2;
1368     } break;
1369     }
1370   }
1371 
1372   // Insert or remove empty line after access specifiers.
1373   if (PreviousLine && PreviousLine->First->isAccessSpecifier() &&
1374       (!PreviousLine->InPPDirective || !RootToken.HasUnescapedNewline)) {
1375     // EmptyLineBeforeAccessModifier is handling the case when two access
1376     // modifiers follow each other.
1377     if (!RootToken.isAccessSpecifier()) {
1378       switch (Style.EmptyLineAfterAccessModifier) {
1379       case FormatStyle::ELAAMS_Never:
1380         Newlines = 1;
1381         break;
1382       case FormatStyle::ELAAMS_Leave:
1383         Newlines = std::max(Newlines, 1u);
1384         break;
1385       case FormatStyle::ELAAMS_Always:
1386         if (RootToken.is(tok::r_brace)) // Do not add at end of class.
1387           Newlines = 1u;
1388         else
1389           Newlines = std::max(Newlines, 2u);
1390         break;
1391       }
1392     }
1393   }
1394 
1395   if (Newlines)
1396     Indent = NewlineIndent;
1397 
1398   // Preprocessor directives get indented before the hash only if specified
1399   if (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash &&
1400       (Line.Type == LT_PreprocessorDirective ||
1401        Line.Type == LT_ImportStatement))
1402     Indent = 0;
1403 
1404   Whitespaces->replaceWhitespace(RootToken, Newlines, Indent, Indent,
1405                                  /*IsAligned=*/false,
1406                                  Line.InPPDirective &&
1407                                      !RootToken.HasUnescapedNewline);
1408 }
1409 
1410 unsigned
1411 UnwrappedLineFormatter::getColumnLimit(bool InPPDirective,
1412                                        const AnnotatedLine *NextLine) const {
1413   // In preprocessor directives reserve two chars for trailing " \" if the
1414   // next line continues the preprocessor directive.
1415   bool ContinuesPPDirective =
1416       InPPDirective &&
1417       // If there is no next line, this is likely a child line and the parent
1418       // continues the preprocessor directive.
1419       (!NextLine ||
1420        (NextLine->InPPDirective &&
1421         // If there is an unescaped newline between this line and the next, the
1422         // next line starts a new preprocessor directive.
1423         !NextLine->First->HasUnescapedNewline));
1424   return Style.ColumnLimit - (ContinuesPPDirective ? 2 : 0);
1425 }
1426 
1427 } // namespace format
1428 } // namespace clang
1429