xref: /freebsd/contrib/llvm-project/clang/lib/Format/UnwrappedLineFormatter.cpp (revision d409305fa3838fb39b38c26fc085fb729b8766d5)
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       Indent = Line.Level * Style.IndentWidth + AdditionalIndent;
61     } else {
62       IndentForLevel.resize(Line.Level + 1);
63       Indent = getIndent(IndentForLevel, Line.Level);
64     }
65     if (static_cast<int>(Indent) + Offset >= 0)
66       Indent += Offset;
67     if (Line.First->is(TT_CSharpGenericTypeConstraint))
68       Indent = Line.Level * Style.IndentWidth + Style.ContinuationIndentWidth;
69   }
70 
71   /// Update the indent state given that \p Line indent should be
72   /// skipped.
73   void skipLine(const AnnotatedLine &Line) {
74     while (IndentForLevel.size() <= Line.Level)
75       IndentForLevel.push_back(Indent);
76   }
77 
78   /// Update the level indent to adapt to the given \p Line.
79   ///
80   /// When a line is not formatted, we move the subsequent lines on the same
81   /// level to the same indent.
82   /// Note that \c nextLine must have been called before this method.
83   void adjustToUnmodifiedLine(const AnnotatedLine &Line) {
84     unsigned LevelIndent = Line.First->OriginalColumn;
85     if (static_cast<int>(LevelIndent) - Offset >= 0)
86       LevelIndent -= Offset;
87     if ((!Line.First->is(tok::comment) || IndentForLevel[Line.Level] == -1) &&
88         !Line.InPPDirective)
89       IndentForLevel[Line.Level] = LevelIndent;
90   }
91 
92 private:
93   /// Get the offset of the line relatively to the level.
94   ///
95   /// For example, 'public:' labels in classes are offset by 1 or 2
96   /// characters to the left from their level.
97   int getIndentOffset(const FormatToken &RootToken) {
98     if (Style.Language == FormatStyle::LK_Java ||
99         Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp())
100       return 0;
101     if (RootToken.isAccessSpecifier(false) ||
102         RootToken.isObjCAccessSpecifier() ||
103         (RootToken.isOneOf(Keywords.kw_signals, Keywords.kw_qsignals) &&
104          RootToken.Next && RootToken.Next->is(tok::colon)))
105       return Style.AccessModifierOffset;
106     return 0;
107   }
108 
109   /// Get the indent of \p Level from \p IndentForLevel.
110   ///
111   /// \p IndentForLevel must contain the indent for the level \c l
112   /// at \p IndentForLevel[l], or a value < 0 if the indent for
113   /// that level is unknown.
114   unsigned getIndent(ArrayRef<int> IndentForLevel, unsigned Level) {
115     if (IndentForLevel[Level] != -1)
116       return IndentForLevel[Level];
117     if (Level == 0)
118       return 0;
119     return getIndent(IndentForLevel, Level - 1) + Style.IndentWidth;
120   }
121 
122   const FormatStyle &Style;
123   const AdditionalKeywords &Keywords;
124   const unsigned AdditionalIndent;
125 
126   /// The indent in characters for each level.
127   std::vector<int> IndentForLevel;
128 
129   /// Offset of the current line relative to the indent level.
130   ///
131   /// For example, the 'public' keywords is often indented with a negative
132   /// offset.
133   int Offset = 0;
134 
135   /// The current line's indent.
136   unsigned Indent = 0;
137 };
138 
139 const FormatToken *getMatchingNamespaceToken(
140     const AnnotatedLine *Line,
141     const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
142   if (!Line->startsWith(tok::r_brace))
143     return nullptr;
144   size_t StartLineIndex = Line->MatchingOpeningBlockLineIndex;
145   if (StartLineIndex == UnwrappedLine::kInvalidIndex)
146     return nullptr;
147   assert(StartLineIndex < AnnotatedLines.size());
148   return AnnotatedLines[StartLineIndex]->First->getNamespaceToken();
149 }
150 
151 StringRef getNamespaceTokenText(const AnnotatedLine *Line) {
152   const FormatToken *NamespaceToken = Line->First->getNamespaceToken();
153   return NamespaceToken ? NamespaceToken->TokenText : StringRef();
154 }
155 
156 StringRef getMatchingNamespaceTokenText(
157     const AnnotatedLine *Line,
158     const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
159   const FormatToken *NamespaceToken =
160       getMatchingNamespaceToken(Line, AnnotatedLines);
161   return NamespaceToken ? NamespaceToken->TokenText : StringRef();
162 }
163 
164 class LineJoiner {
165 public:
166   LineJoiner(const FormatStyle &Style, const AdditionalKeywords &Keywords,
167              const SmallVectorImpl<AnnotatedLine *> &Lines)
168       : Style(Style), Keywords(Keywords), End(Lines.end()), Next(Lines.begin()),
169         AnnotatedLines(Lines) {}
170 
171   /// Returns the next line, merging multiple lines into one if possible.
172   const AnnotatedLine *getNextMergedLine(bool DryRun,
173                                          LevelIndentTracker &IndentTracker) {
174     if (Next == End)
175       return nullptr;
176     const AnnotatedLine *Current = *Next;
177     IndentTracker.nextLine(*Current);
178     unsigned MergedLines = tryFitMultipleLinesInOne(IndentTracker, Next, End);
179     if (MergedLines > 0 && Style.ColumnLimit == 0)
180       // Disallow line merging if there is a break at the start of one of the
181       // input lines.
182       for (unsigned i = 0; i < MergedLines; ++i)
183         if (Next[i + 1]->First->NewlinesBefore > 0)
184           MergedLines = 0;
185     if (!DryRun)
186       for (unsigned i = 0; i < MergedLines; ++i)
187         join(*Next[0], *Next[i + 1]);
188     Next = Next + MergedLines + 1;
189     return Current;
190   }
191 
192 private:
193   /// Calculates how many lines can be merged into 1 starting at \p I.
194   unsigned
195   tryFitMultipleLinesInOne(LevelIndentTracker &IndentTracker,
196                            SmallVectorImpl<AnnotatedLine *>::const_iterator I,
197                            SmallVectorImpl<AnnotatedLine *>::const_iterator E) {
198     const unsigned Indent = IndentTracker.getIndent();
199 
200     // Can't join the last line with anything.
201     if (I + 1 == E)
202       return 0;
203     // We can never merge stuff if there are trailing line comments.
204     const AnnotatedLine *TheLine = *I;
205     if (TheLine->Last->is(TT_LineComment))
206       return 0;
207     if (I[1]->Type == LT_Invalid || I[1]->First->MustBreakBefore)
208       return 0;
209     if (TheLine->InPPDirective &&
210         (!I[1]->InPPDirective || I[1]->First->HasUnescapedNewline))
211       return 0;
212 
213     if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit)
214       return 0;
215 
216     unsigned Limit =
217         Style.ColumnLimit == 0 ? UINT_MAX : Style.ColumnLimit - Indent;
218     // If we already exceed the column limit, we set 'Limit' to 0. The different
219     // tryMerge..() functions can then decide whether to still do merging.
220     Limit = TheLine->Last->TotalLength > Limit
221                 ? 0
222                 : Limit - TheLine->Last->TotalLength;
223 
224     if (TheLine->Last->is(TT_FunctionLBrace) &&
225         TheLine->First == TheLine->Last &&
226         !Style.BraceWrapping.SplitEmptyFunction &&
227         I[1]->First->is(tok::r_brace))
228       return tryMergeSimpleBlock(I, E, Limit);
229 
230     // Handle empty record blocks where the brace has already been wrapped
231     if (TheLine->Last->is(tok::l_brace) && TheLine->First == TheLine->Last &&
232         I != AnnotatedLines.begin()) {
233       bool EmptyBlock = I[1]->First->is(tok::r_brace);
234 
235       const FormatToken *Tok = I[-1]->First;
236       if (Tok && Tok->is(tok::comment))
237         Tok = Tok->getNextNonComment();
238 
239       if (Tok && Tok->getNamespaceToken())
240         return !Style.BraceWrapping.SplitEmptyNamespace && EmptyBlock
241                    ? tryMergeSimpleBlock(I, E, Limit)
242                    : 0;
243 
244       if (Tok && Tok->is(tok::kw_typedef))
245         Tok = Tok->getNextNonComment();
246       if (Tok && Tok->isOneOf(tok::kw_class, tok::kw_struct, tok::kw_union,
247                               tok::kw_extern, Keywords.kw_interface))
248         return !Style.BraceWrapping.SplitEmptyRecord && EmptyBlock
249                    ? tryMergeSimpleBlock(I, E, Limit)
250                    : 0;
251 
252       if (Tok && Tok->is(tok::kw_template) &&
253           Style.BraceWrapping.SplitEmptyRecord && EmptyBlock) {
254         return 0;
255       }
256     }
257 
258     // FIXME: TheLine->Level != 0 might or might not be the right check to do.
259     // If necessary, change to something smarter.
260     bool MergeShortFunctions =
261         Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_All ||
262         (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
263          I[1]->First->is(tok::r_brace)) ||
264         (Style.AllowShortFunctionsOnASingleLine & FormatStyle::SFS_InlineOnly &&
265          TheLine->Level != 0);
266 
267     if (Style.CompactNamespaces) {
268       if (auto nsToken = TheLine->First->getNamespaceToken()) {
269         int i = 0;
270         unsigned closingLine = TheLine->MatchingClosingBlockLineIndex - 1;
271         for (; I + 1 + i != E &&
272                nsToken->TokenText == getNamespaceTokenText(I[i + 1]) &&
273                closingLine == I[i + 1]->MatchingClosingBlockLineIndex &&
274                I[i + 1]->Last->TotalLength < Limit;
275              i++, closingLine--) {
276           // No extra indent for compacted namespaces
277           IndentTracker.skipLine(*I[i + 1]);
278 
279           Limit -= I[i + 1]->Last->TotalLength;
280         }
281         return i;
282       }
283 
284       if (auto nsToken = getMatchingNamespaceToken(TheLine, AnnotatedLines)) {
285         int i = 0;
286         unsigned openingLine = TheLine->MatchingOpeningBlockLineIndex - 1;
287         for (; I + 1 + i != E &&
288                nsToken->TokenText ==
289                    getMatchingNamespaceTokenText(I[i + 1], AnnotatedLines) &&
290                openingLine == I[i + 1]->MatchingOpeningBlockLineIndex;
291              i++, openingLine--) {
292           // No space between consecutive braces
293           I[i + 1]->First->SpacesRequiredBefore = !I[i]->Last->is(tok::r_brace);
294 
295           // Indent like the outer-most namespace
296           IndentTracker.nextLine(*I[i + 1]);
297         }
298         return i;
299       }
300     }
301 
302     // Try to merge a function block with left brace unwrapped
303     if (TheLine->Last->is(TT_FunctionLBrace) &&
304         TheLine->First != TheLine->Last) {
305       return MergeShortFunctions ? tryMergeSimpleBlock(I, E, Limit) : 0;
306     }
307     // Try to merge a control statement block with left brace unwrapped
308     if (TheLine->Last->is(tok::l_brace) && TheLine->First != TheLine->Last &&
309         TheLine->First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for)) {
310       return Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never
311                  ? tryMergeSimpleBlock(I, E, Limit)
312                  : 0;
313     }
314     // Try to merge a control statement block with left brace wrapped
315     if (I[1]->First->is(tok::l_brace) &&
316         (TheLine->First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for,
317                                  tok::kw_switch, tok::kw_try, tok::kw_do,
318                                  TT_ForEachMacro) ||
319          (TheLine->First->is(tok::r_brace) && TheLine->First->Next &&
320           TheLine->First->Next->isOneOf(tok::kw_else, tok::kw_catch))) &&
321         Style.BraceWrapping.AfterControlStatement ==
322             FormatStyle::BWACS_MultiLine) {
323       // If possible, merge the next line's wrapped left brace with the current
324       // line. Otherwise, leave it on the next line, as this is a multi-line
325       // control statement.
326       return (Style.ColumnLimit == 0 ||
327               TheLine->Last->TotalLength <= Style.ColumnLimit)
328                  ? 1
329                  : 0;
330     } else if (I[1]->First->is(tok::l_brace) &&
331                TheLine->First->isOneOf(tok::kw_if, tok::kw_while,
332                                        tok::kw_for)) {
333       return (Style.BraceWrapping.AfterControlStatement ==
334               FormatStyle::BWACS_Always)
335                  ? tryMergeSimpleBlock(I, E, Limit)
336                  : 0;
337     } else if (I[1]->First->is(tok::l_brace) &&
338                TheLine->First->isOneOf(tok::kw_else, tok::kw_catch) &&
339                Style.BraceWrapping.AfterControlStatement ==
340                    FormatStyle::BWACS_MultiLine) {
341       // This case if different from the upper BWACS_MultiLine processing
342       // in that a preceding r_brace is not on the same line as else/catch
343       // most likely because of BeforeElse/BeforeCatch set to true.
344       // If the line length doesn't fit ColumnLimit, leave l_brace on the
345       // next line to respect the BWACS_MultiLine.
346       return (Style.ColumnLimit == 0 ||
347               TheLine->Last->TotalLength <= Style.ColumnLimit)
348                  ? 1
349                  : 0;
350     }
351     // Don't merge block with left brace wrapped after ObjC special blocks
352     if (TheLine->First->is(tok::l_brace) && I != AnnotatedLines.begin() &&
353         I[-1]->First->is(tok::at) && I[-1]->First->Next) {
354       tok::ObjCKeywordKind kwId = I[-1]->First->Next->Tok.getObjCKeywordID();
355       if (kwId == clang::tok::objc_autoreleasepool ||
356           kwId == clang::tok::objc_synchronized)
357         return 0;
358     }
359     // Don't merge block with left brace wrapped after case labels
360     if (TheLine->First->is(tok::l_brace) && I != AnnotatedLines.begin() &&
361         I[-1]->First->isOneOf(tok::kw_case, tok::kw_default))
362       return 0;
363 
364     // Don't merge an empty template class or struct if SplitEmptyRecords
365     // is defined.
366     if (Style.BraceWrapping.SplitEmptyRecord &&
367         TheLine->Last->is(tok::l_brace) && I != AnnotatedLines.begin() &&
368         I[-1]->Last) {
369       const FormatToken *Previous = I[-1]->Last;
370       if (Previous) {
371         if (Previous->is(tok::comment))
372           Previous = Previous->getPreviousNonComment();
373         if (Previous) {
374           if (Previous->is(tok::greater) && !I[-1]->InPPDirective)
375             return 0;
376           if (Previous->is(tok::identifier)) {
377             const FormatToken *PreviousPrevious =
378                 Previous->getPreviousNonComment();
379             if (PreviousPrevious &&
380                 PreviousPrevious->isOneOf(tok::kw_class, tok::kw_struct))
381               return 0;
382           }
383         }
384       }
385     }
386 
387     // Try to merge a block with left brace wrapped that wasn't yet covered
388     if (TheLine->Last->is(tok::l_brace)) {
389       return !Style.BraceWrapping.AfterFunction ||
390                      (I[1]->First->is(tok::r_brace) &&
391                       !Style.BraceWrapping.SplitEmptyRecord)
392                  ? tryMergeSimpleBlock(I, E, Limit)
393                  : 0;
394     }
395     // Try to merge a function block with left brace wrapped
396     if (I[1]->First->is(TT_FunctionLBrace) &&
397         Style.BraceWrapping.AfterFunction) {
398       if (I[1]->Last->is(TT_LineComment))
399         return 0;
400 
401       // Check for Limit <= 2 to account for the " {".
402       if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(TheLine)))
403         return 0;
404       Limit -= 2;
405 
406       unsigned MergedLines = 0;
407       if (MergeShortFunctions ||
408           (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
409            I[1]->First == I[1]->Last && I + 2 != E &&
410            I[2]->First->is(tok::r_brace))) {
411         MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
412         // If we managed to merge the block, count the function header, which is
413         // on a separate line.
414         if (MergedLines > 0)
415           ++MergedLines;
416       }
417       return MergedLines;
418     }
419     if (TheLine->First->is(tok::kw_if)) {
420       return Style.AllowShortIfStatementsOnASingleLine
421                  ? tryMergeSimpleControlStatement(I, E, Limit)
422                  : 0;
423     }
424     if (TheLine->First->isOneOf(tok::kw_for, tok::kw_while, tok::kw_do)) {
425       return Style.AllowShortLoopsOnASingleLine
426                  ? tryMergeSimpleControlStatement(I, E, Limit)
427                  : 0;
428     }
429     if (TheLine->First->isOneOf(tok::kw_case, tok::kw_default)) {
430       return Style.AllowShortCaseLabelsOnASingleLine
431                  ? tryMergeShortCaseLabels(I, E, Limit)
432                  : 0;
433     }
434     if (TheLine->InPPDirective &&
435         (TheLine->First->HasUnescapedNewline || TheLine->First->IsFirst)) {
436       return tryMergeSimplePPDirective(I, E, Limit);
437     }
438     return 0;
439   }
440 
441   unsigned
442   tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
443                             SmallVectorImpl<AnnotatedLine *>::const_iterator E,
444                             unsigned Limit) {
445     if (Limit == 0)
446       return 0;
447     if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline)
448       return 0;
449     if (1 + I[1]->Last->TotalLength > Limit)
450       return 0;
451     return 1;
452   }
453 
454   unsigned tryMergeSimpleControlStatement(
455       SmallVectorImpl<AnnotatedLine *>::const_iterator I,
456       SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) {
457     if (Limit == 0)
458       return 0;
459     if (Style.BraceWrapping.AfterControlStatement ==
460             FormatStyle::BWACS_Always &&
461         I[1]->First->is(tok::l_brace) &&
462         Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never)
463       return 0;
464     if (I[1]->InPPDirective != (*I)->InPPDirective ||
465         (I[1]->InPPDirective && I[1]->First->HasUnescapedNewline))
466       return 0;
467     Limit = limitConsideringMacros(I + 1, E, Limit);
468     AnnotatedLine &Line = **I;
469     if (!Line.First->is(tok::kw_do) && Line.Last->isNot(tok::r_paren))
470       return 0;
471     // Only merge do while if do is the only statement on the line.
472     if (Line.First->is(tok::kw_do) && !Line.Last->is(tok::kw_do))
473       return 0;
474     if (1 + I[1]->Last->TotalLength > Limit)
475       return 0;
476     if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for, tok::kw_while,
477                              TT_LineComment))
478       return 0;
479     // Only inline simple if's (no nested if or else), unless specified
480     if (Style.AllowShortIfStatementsOnASingleLine != FormatStyle::SIS_Always) {
481       if (I + 2 != E && Line.startsWith(tok::kw_if) &&
482           I[2]->First->is(tok::kw_else))
483         return 0;
484     }
485     return 1;
486   }
487 
488   unsigned
489   tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
490                           SmallVectorImpl<AnnotatedLine *>::const_iterator E,
491                           unsigned Limit) {
492     if (Limit == 0 || I + 1 == E ||
493         I[1]->First->isOneOf(tok::kw_case, tok::kw_default))
494       return 0;
495     if (I[0]->Last->is(tok::l_brace) || I[1]->First->is(tok::l_brace))
496       return 0;
497     unsigned NumStmts = 0;
498     unsigned Length = 0;
499     bool EndsWithComment = false;
500     bool InPPDirective = I[0]->InPPDirective;
501     const unsigned Level = I[0]->Level;
502     for (; NumStmts < 3; ++NumStmts) {
503       if (I + 1 + NumStmts == E)
504         break;
505       const AnnotatedLine *Line = I[1 + NumStmts];
506       if (Line->InPPDirective != InPPDirective)
507         break;
508       if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
509         break;
510       if (Line->First->isOneOf(tok::kw_if, tok::kw_for, tok::kw_switch,
511                                tok::kw_while) ||
512           EndsWithComment)
513         return 0;
514       if (Line->First->is(tok::comment)) {
515         if (Level != Line->Level)
516           return 0;
517         SmallVectorImpl<AnnotatedLine *>::const_iterator J = I + 2 + NumStmts;
518         for (; J != E; ++J) {
519           Line = *J;
520           if (Line->InPPDirective != InPPDirective)
521             break;
522           if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
523             break;
524           if (Line->First->isNot(tok::comment) || Level != Line->Level)
525             return 0;
526         }
527         break;
528       }
529       if (Line->Last->is(tok::comment))
530         EndsWithComment = true;
531       Length += I[1 + NumStmts]->Last->TotalLength + 1; // 1 for the space.
532     }
533     if (NumStmts == 0 || NumStmts == 3 || Length > Limit)
534       return 0;
535     return NumStmts;
536   }
537 
538   unsigned
539   tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
540                       SmallVectorImpl<AnnotatedLine *>::const_iterator E,
541                       unsigned Limit) {
542     AnnotatedLine &Line = **I;
543 
544     // Don't merge ObjC @ keywords and methods.
545     // FIXME: If an option to allow short exception handling clauses on a single
546     // line is added, change this to not return for @try and friends.
547     if (Style.Language != FormatStyle::LK_Java &&
548         Line.First->isOneOf(tok::at, tok::minus, tok::plus))
549       return 0;
550 
551     // Check that the current line allows merging. This depends on whether we
552     // are in a control flow statements as well as several style flags.
553     if (Line.First->isOneOf(tok::kw_else, tok::kw_case) ||
554         (Line.First->Next && Line.First->Next->is(tok::kw_else)))
555       return 0;
556     // default: in switch statement
557     if (Line.First->is(tok::kw_default)) {
558       const FormatToken *Tok = Line.First->getNextNonComment();
559       if (Tok && Tok->is(tok::colon))
560         return 0;
561     }
562     if (Line.First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_do, tok::kw_try,
563                             tok::kw___try, tok::kw_catch, tok::kw___finally,
564                             tok::kw_for, tok::r_brace, Keywords.kw___except)) {
565       if (Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never)
566         return 0;
567       // Don't merge when we can't except the case when
568       // the control statement block is empty
569       if (!Style.AllowShortIfStatementsOnASingleLine &&
570           Line.startsWith(tok::kw_if) &&
571           !Style.BraceWrapping.AfterControlStatement &&
572           !I[1]->First->is(tok::r_brace))
573         return 0;
574       if (!Style.AllowShortIfStatementsOnASingleLine &&
575           Line.startsWith(tok::kw_if) &&
576           Style.BraceWrapping.AfterControlStatement ==
577               FormatStyle::BWACS_Always &&
578           I + 2 != E && !I[2]->First->is(tok::r_brace))
579         return 0;
580       if (!Style.AllowShortLoopsOnASingleLine &&
581           Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for) &&
582           !Style.BraceWrapping.AfterControlStatement &&
583           !I[1]->First->is(tok::r_brace))
584         return 0;
585       if (!Style.AllowShortLoopsOnASingleLine &&
586           Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for) &&
587           Style.BraceWrapping.AfterControlStatement ==
588               FormatStyle::BWACS_Always &&
589           I + 2 != E && !I[2]->First->is(tok::r_brace))
590         return 0;
591       // FIXME: Consider an option to allow short exception handling clauses on
592       // a single line.
593       // FIXME: This isn't covered by tests.
594       // FIXME: For catch, __except, __finally the first token on the line
595       // is '}', so this isn't correct here.
596       if (Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch,
597                               Keywords.kw___except, tok::kw___finally))
598         return 0;
599     }
600 
601     if (Line.Last->is(tok::l_brace)) {
602       FormatToken *Tok = I[1]->First;
603       if (Tok->is(tok::r_brace) && !Tok->MustBreakBefore &&
604           (Tok->getNextNonComment() == nullptr ||
605            Tok->getNextNonComment()->is(tok::semi))) {
606         // We merge empty blocks even if the line exceeds the column limit.
607         Tok->SpacesRequiredBefore = Style.SpaceInEmptyBlock ? 1 : 0;
608         Tok->CanBreakBefore = true;
609         return 1;
610       } else if (Limit != 0 && !Line.startsWithNamespace() &&
611                  !startsExternCBlock(Line)) {
612         // We don't merge short records.
613         FormatToken *RecordTok = Line.First;
614         // Skip record modifiers.
615         while (RecordTok->Next &&
616                RecordTok->isOneOf(
617                    tok::kw_typedef, tok::kw_export, Keywords.kw_declare,
618                    Keywords.kw_abstract, tok::kw_default, tok::kw_public,
619                    tok::kw_private, tok::kw_protected, Keywords.kw_internal))
620           RecordTok = RecordTok->Next;
621         if (RecordTok &&
622             RecordTok->isOneOf(tok::kw_class, tok::kw_union, tok::kw_struct,
623                                Keywords.kw_interface))
624           return 0;
625 
626         // Check that we still have three lines and they fit into the limit.
627         if (I + 2 == E || I[2]->Type == LT_Invalid)
628           return 0;
629         Limit = limitConsideringMacros(I + 2, E, Limit);
630 
631         if (!nextTwoLinesFitInto(I, Limit))
632           return 0;
633 
634         // Second, check that the next line does not contain any braces - if it
635         // does, readability declines when putting it into a single line.
636         if (I[1]->Last->is(TT_LineComment))
637           return 0;
638         do {
639           if (Tok->is(tok::l_brace) && Tok->isNot(BK_BracedInit))
640             return 0;
641           Tok = Tok->Next;
642         } while (Tok);
643 
644         // Last, check that the third line starts with a closing brace.
645         Tok = I[2]->First;
646         if (Tok->isNot(tok::r_brace))
647           return 0;
648 
649         // Don't merge "if (a) { .. } else {".
650         if (Tok->Next && Tok->Next->is(tok::kw_else))
651           return 0;
652 
653         // Don't merge a trailing multi-line control statement block like:
654         // } else if (foo &&
655         //            bar)
656         // { <-- current Line
657         //   baz();
658         // }
659         if (Line.First == Line.Last &&
660             Style.BraceWrapping.AfterControlStatement ==
661                 FormatStyle::BWACS_MultiLine)
662           return 0;
663 
664         return 2;
665       }
666     } else if (I[1]->First->is(tok::l_brace)) {
667       if (I[1]->Last->is(TT_LineComment))
668         return 0;
669 
670       // Check for Limit <= 2 to account for the " {".
671       if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(*I)))
672         return 0;
673       Limit -= 2;
674       unsigned MergedLines = 0;
675       if (Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never ||
676           (I[1]->First == I[1]->Last && I + 2 != E &&
677            I[2]->First->is(tok::r_brace))) {
678         MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
679         // If we managed to merge the block, count the statement header, which
680         // is on a separate line.
681         if (MergedLines > 0)
682           ++MergedLines;
683       }
684       return MergedLines;
685     }
686     return 0;
687   }
688 
689   /// Returns the modified column limit for \p I if it is inside a macro and
690   /// needs a trailing '\'.
691   unsigned
692   limitConsideringMacros(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
693                          SmallVectorImpl<AnnotatedLine *>::const_iterator E,
694                          unsigned Limit) {
695     if (I[0]->InPPDirective && I + 1 != E &&
696         !I[1]->First->HasUnescapedNewline && !I[1]->First->is(tok::eof)) {
697       return Limit < 2 ? 0 : Limit - 2;
698     }
699     return Limit;
700   }
701 
702   bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
703                            unsigned Limit) {
704     if (I[1]->First->MustBreakBefore || I[2]->First->MustBreakBefore)
705       return false;
706     return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit;
707   }
708 
709   bool containsMustBreak(const AnnotatedLine *Line) {
710     for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) {
711       if (Tok->MustBreakBefore)
712         return true;
713     }
714     return false;
715   }
716 
717   void join(AnnotatedLine &A, const AnnotatedLine &B) {
718     assert(!A.Last->Next);
719     assert(!B.First->Previous);
720     if (B.Affected)
721       A.Affected = true;
722     A.Last->Next = B.First;
723     B.First->Previous = A.Last;
724     B.First->CanBreakBefore = true;
725     unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore;
726     for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) {
727       Tok->TotalLength += LengthA;
728       A.Last = Tok;
729     }
730   }
731 
732   const FormatStyle &Style;
733   const AdditionalKeywords &Keywords;
734   const SmallVectorImpl<AnnotatedLine *>::const_iterator End;
735 
736   SmallVectorImpl<AnnotatedLine *>::const_iterator Next;
737   const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines;
738 };
739 
740 static void markFinalized(FormatToken *Tok) {
741   for (; Tok; Tok = Tok->Next) {
742     Tok->Finalized = true;
743     for (AnnotatedLine *Child : Tok->Children)
744       markFinalized(Child->First);
745   }
746 }
747 
748 #ifndef NDEBUG
749 static void printLineState(const LineState &State) {
750   llvm::dbgs() << "State: ";
751   for (const ParenState &P : State.Stack) {
752     llvm::dbgs() << (P.Tok ? P.Tok->TokenText : "F") << "|" << P.Indent << "|"
753                  << P.LastSpace << "|" << P.NestedBlockIndent << " ";
754   }
755   llvm::dbgs() << State.NextToken->TokenText << "\n";
756 }
757 #endif
758 
759 /// Base class for classes that format one \c AnnotatedLine.
760 class LineFormatter {
761 public:
762   LineFormatter(ContinuationIndenter *Indenter, WhitespaceManager *Whitespaces,
763                 const FormatStyle &Style,
764                 UnwrappedLineFormatter *BlockFormatter)
765       : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style),
766         BlockFormatter(BlockFormatter) {}
767   virtual ~LineFormatter() {}
768 
769   /// Formats an \c AnnotatedLine and returns the penalty.
770   ///
771   /// If \p DryRun is \c false, directly applies the changes.
772   virtual unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
773                               unsigned FirstStartColumn, bool DryRun) = 0;
774 
775 protected:
776   /// If the \p State's next token is an r_brace closing a nested block,
777   /// format the nested block before it.
778   ///
779   /// Returns \c true if all children could be placed successfully and adapts
780   /// \p Penalty as well as \p State. If \p DryRun is false, also directly
781   /// creates changes using \c Whitespaces.
782   ///
783   /// The crucial idea here is that children always get formatted upon
784   /// encountering the closing brace right after the nested block. Now, if we
785   /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is
786   /// \c false), the entire block has to be kept on the same line (which is only
787   /// possible if it fits on the line, only contains a single statement, etc.
788   ///
789   /// If \p NewLine is true, we format the nested block on separate lines, i.e.
790   /// break after the "{", format all lines with correct indentation and the put
791   /// the closing "}" on yet another new line.
792   ///
793   /// This enables us to keep the simple structure of the
794   /// \c UnwrappedLineFormatter, where we only have two options for each token:
795   /// break or don't break.
796   bool formatChildren(LineState &State, bool NewLine, bool DryRun,
797                       unsigned &Penalty) {
798     const FormatToken *LBrace = State.NextToken->getPreviousNonComment();
799     FormatToken &Previous = *State.NextToken->Previous;
800     if (!LBrace || LBrace->isNot(tok::l_brace) || LBrace->isNot(BK_Block) ||
801         Previous.Children.size() == 0)
802       // The previous token does not open a block. Nothing to do. We don't
803       // assert so that we can simply call this function for all tokens.
804       return true;
805 
806     if (NewLine) {
807       int AdditionalIndent = State.Stack.back().Indent -
808                              Previous.Children[0]->Level * Style.IndentWidth;
809 
810       Penalty +=
811           BlockFormatter->format(Previous.Children, DryRun, AdditionalIndent,
812                                  /*FixBadIndentation=*/true);
813       return true;
814     }
815 
816     if (Previous.Children[0]->First->MustBreakBefore)
817       return false;
818 
819     // Cannot merge into one line if this line ends on a comment.
820     if (Previous.is(tok::comment))
821       return false;
822 
823     // Cannot merge multiple statements into a single line.
824     if (Previous.Children.size() > 1)
825       return false;
826 
827     const AnnotatedLine *Child = Previous.Children[0];
828     // We can't put the closing "}" on a line with a trailing comment.
829     if (Child->Last->isTrailingComment())
830       return false;
831 
832     // If the child line exceeds the column limit, we wouldn't want to merge it.
833     // We add +2 for the trailing " }".
834     if (Style.ColumnLimit > 0 &&
835         Child->Last->TotalLength + State.Column + 2 > Style.ColumnLimit)
836       return false;
837 
838     if (!DryRun) {
839       Whitespaces->replaceWhitespace(
840           *Child->First, /*Newlines=*/0, /*Spaces=*/1,
841           /*StartOfTokenColumn=*/State.Column, /*IsAligned=*/false,
842           State.Line->InPPDirective);
843     }
844     Penalty +=
845         formatLine(*Child, State.Column + 1, /*FirstStartColumn=*/0, DryRun);
846 
847     State.Column += 1 + Child->Last->TotalLength;
848     return true;
849   }
850 
851   ContinuationIndenter *Indenter;
852 
853 private:
854   WhitespaceManager *Whitespaces;
855   const FormatStyle &Style;
856   UnwrappedLineFormatter *BlockFormatter;
857 };
858 
859 /// Formatter that keeps the existing line breaks.
860 class NoColumnLimitLineFormatter : public LineFormatter {
861 public:
862   NoColumnLimitLineFormatter(ContinuationIndenter *Indenter,
863                              WhitespaceManager *Whitespaces,
864                              const FormatStyle &Style,
865                              UnwrappedLineFormatter *BlockFormatter)
866       : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
867 
868   /// Formats the line, simply keeping all of the input's line breaking
869   /// decisions.
870   unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
871                       unsigned FirstStartColumn, bool DryRun) override {
872     assert(!DryRun);
873     LineState State = Indenter->getInitialState(FirstIndent, FirstStartColumn,
874                                                 &Line, /*DryRun=*/false);
875     while (State.NextToken) {
876       bool Newline =
877           Indenter->mustBreak(State) ||
878           (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0);
879       unsigned Penalty = 0;
880       formatChildren(State, Newline, /*DryRun=*/false, Penalty);
881       Indenter->addTokenToState(State, Newline, /*DryRun=*/false);
882     }
883     return 0;
884   }
885 };
886 
887 /// Formatter that puts all tokens into a single line without breaks.
888 class NoLineBreakFormatter : public LineFormatter {
889 public:
890   NoLineBreakFormatter(ContinuationIndenter *Indenter,
891                        WhitespaceManager *Whitespaces, const FormatStyle &Style,
892                        UnwrappedLineFormatter *BlockFormatter)
893       : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
894 
895   /// Puts all tokens into a single line.
896   unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
897                       unsigned FirstStartColumn, bool DryRun) override {
898     unsigned Penalty = 0;
899     LineState State =
900         Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun);
901     while (State.NextToken) {
902       formatChildren(State, /*NewLine=*/false, DryRun, Penalty);
903       Indenter->addTokenToState(
904           State, /*Newline=*/State.NextToken->MustBreakBefore, DryRun);
905     }
906     return Penalty;
907   }
908 };
909 
910 /// Finds the best way to break lines.
911 class OptimizingLineFormatter : public LineFormatter {
912 public:
913   OptimizingLineFormatter(ContinuationIndenter *Indenter,
914                           WhitespaceManager *Whitespaces,
915                           const FormatStyle &Style,
916                           UnwrappedLineFormatter *BlockFormatter)
917       : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
918 
919   /// Formats the line by finding the best line breaks with line lengths
920   /// below the column limit.
921   unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
922                       unsigned FirstStartColumn, bool DryRun) override {
923     LineState State =
924         Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun);
925 
926     // If the ObjC method declaration does not fit on a line, we should format
927     // it with one arg per line.
928     if (State.Line->Type == LT_ObjCMethodDecl)
929       State.Stack.back().BreakBeforeParameter = true;
930 
931     // Find best solution in solution space.
932     return analyzeSolutionSpace(State, DryRun);
933   }
934 
935 private:
936   struct CompareLineStatePointers {
937     bool operator()(LineState *obj1, LineState *obj2) const {
938       return *obj1 < *obj2;
939     }
940   };
941 
942   /// A pair of <penalty, count> that is used to prioritize the BFS on.
943   ///
944   /// In case of equal penalties, we want to prefer states that were inserted
945   /// first. During state generation we make sure that we insert states first
946   /// that break the line as late as possible.
947   typedef std::pair<unsigned, unsigned> OrderedPenalty;
948 
949   /// An edge in the solution space from \c Previous->State to \c State,
950   /// inserting a newline dependent on the \c NewLine.
951   struct StateNode {
952     StateNode(const LineState &State, bool NewLine, StateNode *Previous)
953         : State(State), NewLine(NewLine), Previous(Previous) {}
954     LineState State;
955     bool NewLine;
956     StateNode *Previous;
957   };
958 
959   /// An item in the prioritized BFS search queue. The \c StateNode's
960   /// \c State has the given \c OrderedPenalty.
961   typedef std::pair<OrderedPenalty, StateNode *> QueueItem;
962 
963   /// The BFS queue type.
964   typedef std::priority_queue<QueueItem, std::vector<QueueItem>,
965                               std::greater<QueueItem>>
966       QueueType;
967 
968   /// Analyze the entire solution space starting from \p InitialState.
969   ///
970   /// This implements a variant of Dijkstra's algorithm on the graph that spans
971   /// the solution space (\c LineStates are the nodes). The algorithm tries to
972   /// find the shortest path (the one with lowest penalty) from \p InitialState
973   /// to a state where all tokens are placed. Returns the penalty.
974   ///
975   /// If \p DryRun is \c false, directly applies the changes.
976   unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun) {
977     std::set<LineState *, CompareLineStatePointers> Seen;
978 
979     // Increasing count of \c StateNode items we have created. This is used to
980     // create a deterministic order independent of the container.
981     unsigned Count = 0;
982     QueueType Queue;
983 
984     // Insert start element into queue.
985     StateNode *Node =
986         new (Allocator.Allocate()) StateNode(InitialState, false, nullptr);
987     Queue.push(QueueItem(OrderedPenalty(0, Count), Node));
988     ++Count;
989 
990     unsigned Penalty = 0;
991 
992     // While not empty, take first element and follow edges.
993     while (!Queue.empty()) {
994       Penalty = Queue.top().first.first;
995       StateNode *Node = Queue.top().second;
996       if (!Node->State.NextToken) {
997         LLVM_DEBUG(llvm::dbgs()
998                    << "\n---\nPenalty for line: " << Penalty << "\n");
999         break;
1000       }
1001       Queue.pop();
1002 
1003       // Cut off the analysis of certain solutions if the analysis gets too
1004       // complex. See description of IgnoreStackForComparison.
1005       if (Count > 50000)
1006         Node->State.IgnoreStackForComparison = true;
1007 
1008       if (!Seen.insert(&Node->State).second)
1009         // State already examined with lower penalty.
1010         continue;
1011 
1012       FormatDecision LastFormat = Node->State.NextToken->getDecision();
1013       if (LastFormat == FD_Unformatted || LastFormat == FD_Continue)
1014         addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue);
1015       if (LastFormat == FD_Unformatted || LastFormat == FD_Break)
1016         addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue);
1017     }
1018 
1019     if (Queue.empty()) {
1020       // We were unable to find a solution, do nothing.
1021       // FIXME: Add diagnostic?
1022       LLVM_DEBUG(llvm::dbgs() << "Could not find a solution.\n");
1023       return 0;
1024     }
1025 
1026     // Reconstruct the solution.
1027     if (!DryRun)
1028       reconstructPath(InitialState, Queue.top().second);
1029 
1030     LLVM_DEBUG(llvm::dbgs()
1031                << "Total number of analyzed states: " << Count << "\n");
1032     LLVM_DEBUG(llvm::dbgs() << "---\n");
1033 
1034     return Penalty;
1035   }
1036 
1037   /// Add the following state to the analysis queue \c Queue.
1038   ///
1039   /// Assume the current state is \p PreviousNode and has been reached with a
1040   /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true.
1041   void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode,
1042                            bool NewLine, unsigned *Count, QueueType *Queue) {
1043     if (NewLine && !Indenter->canBreak(PreviousNode->State))
1044       return;
1045     if (!NewLine && Indenter->mustBreak(PreviousNode->State))
1046       return;
1047 
1048     StateNode *Node = new (Allocator.Allocate())
1049         StateNode(PreviousNode->State, NewLine, PreviousNode);
1050     if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty))
1051       return;
1052 
1053     Penalty += Indenter->addTokenToState(Node->State, NewLine, true);
1054 
1055     Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node));
1056     ++(*Count);
1057   }
1058 
1059   /// Applies the best formatting by reconstructing the path in the
1060   /// solution space that leads to \c Best.
1061   void reconstructPath(LineState &State, StateNode *Best) {
1062     std::deque<StateNode *> Path;
1063     // We do not need a break before the initial token.
1064     while (Best->Previous) {
1065       Path.push_front(Best);
1066       Best = Best->Previous;
1067     }
1068     for (auto I = Path.begin(), E = Path.end(); I != E; ++I) {
1069       unsigned Penalty = 0;
1070       formatChildren(State, (*I)->NewLine, /*DryRun=*/false, Penalty);
1071       Penalty += Indenter->addTokenToState(State, (*I)->NewLine, false);
1072 
1073       LLVM_DEBUG({
1074         printLineState((*I)->Previous->State);
1075         if ((*I)->NewLine) {
1076           llvm::dbgs() << "Penalty for placing "
1077                        << (*I)->Previous->State.NextToken->Tok.getName()
1078                        << " on a new line: " << Penalty << "\n";
1079         }
1080       });
1081     }
1082   }
1083 
1084   llvm::SpecificBumpPtrAllocator<StateNode> Allocator;
1085 };
1086 
1087 } // anonymous namespace
1088 
1089 unsigned UnwrappedLineFormatter::format(
1090     const SmallVectorImpl<AnnotatedLine *> &Lines, bool DryRun,
1091     int AdditionalIndent, bool FixBadIndentation, unsigned FirstStartColumn,
1092     unsigned NextStartColumn, unsigned LastStartColumn) {
1093   LineJoiner Joiner(Style, Keywords, Lines);
1094 
1095   // Try to look up already computed penalty in DryRun-mode.
1096   std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned> CacheKey(
1097       &Lines, AdditionalIndent);
1098   auto CacheIt = PenaltyCache.find(CacheKey);
1099   if (DryRun && CacheIt != PenaltyCache.end())
1100     return CacheIt->second;
1101 
1102   assert(!Lines.empty());
1103   unsigned Penalty = 0;
1104   LevelIndentTracker IndentTracker(Style, Keywords, Lines[0]->Level,
1105                                    AdditionalIndent);
1106   const AnnotatedLine *PreviousLine = nullptr;
1107   const AnnotatedLine *NextLine = nullptr;
1108 
1109   // The minimum level of consecutive lines that have been formatted.
1110   unsigned RangeMinLevel = UINT_MAX;
1111 
1112   bool FirstLine = true;
1113   for (const AnnotatedLine *Line =
1114            Joiner.getNextMergedLine(DryRun, IndentTracker);
1115        Line; Line = NextLine, FirstLine = false) {
1116     const AnnotatedLine &TheLine = *Line;
1117     unsigned Indent = IndentTracker.getIndent();
1118 
1119     // We continue formatting unchanged lines to adjust their indent, e.g. if a
1120     // scope was added. However, we need to carefully stop doing this when we
1121     // exit the scope of affected lines to prevent indenting a the entire
1122     // remaining file if it currently missing a closing brace.
1123     bool PreviousRBrace =
1124         PreviousLine && PreviousLine->startsWith(tok::r_brace);
1125     bool ContinueFormatting =
1126         TheLine.Level > RangeMinLevel ||
1127         (TheLine.Level == RangeMinLevel && !PreviousRBrace &&
1128          !TheLine.startsWith(tok::r_brace));
1129 
1130     bool FixIndentation = (FixBadIndentation || ContinueFormatting) &&
1131                           Indent != TheLine.First->OriginalColumn;
1132     bool ShouldFormat = TheLine.Affected || FixIndentation;
1133     // We cannot format this line; if the reason is that the line had a
1134     // parsing error, remember that.
1135     if (ShouldFormat && TheLine.Type == LT_Invalid && Status) {
1136       Status->FormatComplete = false;
1137       Status->Line =
1138           SourceMgr.getSpellingLineNumber(TheLine.First->Tok.getLocation());
1139     }
1140 
1141     if (ShouldFormat && TheLine.Type != LT_Invalid) {
1142       if (!DryRun) {
1143         bool LastLine = Line->First->is(tok::eof);
1144         formatFirstToken(TheLine, PreviousLine, Lines, Indent,
1145                          LastLine ? LastStartColumn : NextStartColumn + Indent);
1146       }
1147 
1148       NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
1149       unsigned ColumnLimit = getColumnLimit(TheLine.InPPDirective, NextLine);
1150       bool FitsIntoOneLine =
1151           TheLine.Last->TotalLength + Indent <= ColumnLimit ||
1152           (TheLine.Type == LT_ImportStatement &&
1153            (Style.Language != FormatStyle::LK_JavaScript ||
1154             !Style.JavaScriptWrapImports)) ||
1155           (Style.isCSharp() &&
1156            TheLine.InPPDirective); // don't split #regions in C#
1157       if (Style.ColumnLimit == 0)
1158         NoColumnLimitLineFormatter(Indenter, Whitespaces, Style, this)
1159             .formatLine(TheLine, NextStartColumn + Indent,
1160                         FirstLine ? FirstStartColumn : 0, DryRun);
1161       else if (FitsIntoOneLine)
1162         Penalty += NoLineBreakFormatter(Indenter, Whitespaces, Style, this)
1163                        .formatLine(TheLine, NextStartColumn + Indent,
1164                                    FirstLine ? FirstStartColumn : 0, DryRun);
1165       else
1166         Penalty += OptimizingLineFormatter(Indenter, Whitespaces, Style, this)
1167                        .formatLine(TheLine, NextStartColumn + Indent,
1168                                    FirstLine ? FirstStartColumn : 0, DryRun);
1169       RangeMinLevel = std::min(RangeMinLevel, TheLine.Level);
1170     } else {
1171       // If no token in the current line is affected, we still need to format
1172       // affected children.
1173       if (TheLine.ChildrenAffected)
1174         for (const FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next)
1175           if (!Tok->Children.empty())
1176             format(Tok->Children, DryRun);
1177 
1178       // Adapt following lines on the current indent level to the same level
1179       // unless the current \c AnnotatedLine is not at the beginning of a line.
1180       bool StartsNewLine =
1181           TheLine.First->NewlinesBefore > 0 || TheLine.First->IsFirst;
1182       if (StartsNewLine)
1183         IndentTracker.adjustToUnmodifiedLine(TheLine);
1184       if (!DryRun) {
1185         bool ReformatLeadingWhitespace =
1186             StartsNewLine && ((PreviousLine && PreviousLine->Affected) ||
1187                               TheLine.LeadingEmptyLinesAffected);
1188         // Format the first token.
1189         if (ReformatLeadingWhitespace)
1190           formatFirstToken(TheLine, PreviousLine, Lines,
1191                            TheLine.First->OriginalColumn,
1192                            TheLine.First->OriginalColumn);
1193         else
1194           Whitespaces->addUntouchableToken(*TheLine.First,
1195                                            TheLine.InPPDirective);
1196 
1197         // Notify the WhitespaceManager about the unchanged whitespace.
1198         for (FormatToken *Tok = TheLine.First->Next; Tok; Tok = Tok->Next)
1199           Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective);
1200       }
1201       NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
1202       RangeMinLevel = UINT_MAX;
1203     }
1204     if (!DryRun)
1205       markFinalized(TheLine.First);
1206     PreviousLine = &TheLine;
1207   }
1208   PenaltyCache[CacheKey] = Penalty;
1209   return Penalty;
1210 }
1211 
1212 void UnwrappedLineFormatter::formatFirstToken(
1213     const AnnotatedLine &Line, const AnnotatedLine *PreviousLine,
1214     const SmallVectorImpl<AnnotatedLine *> &Lines, unsigned Indent,
1215     unsigned NewlineIndent) {
1216   FormatToken &RootToken = *Line.First;
1217   if (RootToken.is(tok::eof)) {
1218     unsigned Newlines = std::min(RootToken.NewlinesBefore, 1u);
1219     unsigned TokenIndent = Newlines ? NewlineIndent : 0;
1220     Whitespaces->replaceWhitespace(RootToken, Newlines, TokenIndent,
1221                                    TokenIndent);
1222     return;
1223   }
1224   unsigned Newlines =
1225       std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1);
1226   // Remove empty lines before "}" where applicable.
1227   if (RootToken.is(tok::r_brace) &&
1228       (!RootToken.Next ||
1229        (RootToken.Next->is(tok::semi) && !RootToken.Next->Next)) &&
1230       // Do not remove empty lines before namespace closing "}".
1231       !getNamespaceToken(&Line, Lines))
1232     Newlines = std::min(Newlines, 1u);
1233   // Remove empty lines at the start of nested blocks (lambdas/arrow functions)
1234   if (PreviousLine == nullptr && Line.Level > 0)
1235     Newlines = std::min(Newlines, 1u);
1236   if (Newlines == 0 && !RootToken.IsFirst)
1237     Newlines = 1;
1238   if (RootToken.IsFirst && !RootToken.HasUnescapedNewline)
1239     Newlines = 0;
1240 
1241   // Remove empty lines after "{".
1242   if (!Style.KeepEmptyLinesAtTheStartOfBlocks && PreviousLine &&
1243       PreviousLine->Last->is(tok::l_brace) &&
1244       !PreviousLine->startsWithNamespace() &&
1245       !startsExternCBlock(*PreviousLine))
1246     Newlines = 1;
1247 
1248   // Insert or remove empty line before access specifiers.
1249   if (PreviousLine && RootToken.isAccessSpecifier()) {
1250     switch (Style.EmptyLineBeforeAccessModifier) {
1251     case FormatStyle::ELBAMS_Never:
1252       if (RootToken.NewlinesBefore > 1)
1253         Newlines = 1;
1254       break;
1255     case FormatStyle::ELBAMS_Leave:
1256       Newlines = std::max(RootToken.NewlinesBefore, 1u);
1257       break;
1258     case FormatStyle::ELBAMS_LogicalBlock:
1259       if (PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) &&
1260           RootToken.NewlinesBefore <= 1)
1261         Newlines = 2;
1262       break;
1263     case FormatStyle::ELBAMS_Always: {
1264       const FormatToken *previousToken;
1265       if (PreviousLine->Last->is(tok::comment))
1266         previousToken = PreviousLine->Last->getPreviousNonComment();
1267       else
1268         previousToken = PreviousLine->Last;
1269       if ((!previousToken || !previousToken->is(tok::l_brace)) &&
1270           RootToken.NewlinesBefore <= 1)
1271         Newlines = 2;
1272     } break;
1273     }
1274   }
1275 
1276   // Remove empty lines after access specifiers.
1277   if (PreviousLine && PreviousLine->First->isAccessSpecifier() &&
1278       (!PreviousLine->InPPDirective || !RootToken.HasUnescapedNewline))
1279     Newlines = std::min(1u, Newlines);
1280 
1281   if (Newlines)
1282     Indent = NewlineIndent;
1283 
1284   // If in Whitemsmiths mode, indent start and end of blocks
1285   if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) {
1286     if (RootToken.isOneOf(tok::l_brace, tok::r_brace, tok::kw_case,
1287                           tok::kw_default))
1288       Indent += Style.IndentWidth;
1289   }
1290 
1291   // Preprocessor directives get indented before the hash only if specified
1292   if (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash &&
1293       (Line.Type == LT_PreprocessorDirective ||
1294        Line.Type == LT_ImportStatement))
1295     Indent = 0;
1296 
1297   Whitespaces->replaceWhitespace(RootToken, Newlines, Indent, Indent,
1298                                  /*IsAligned=*/false,
1299                                  Line.InPPDirective &&
1300                                      !RootToken.HasUnescapedNewline);
1301 }
1302 
1303 unsigned
1304 UnwrappedLineFormatter::getColumnLimit(bool InPPDirective,
1305                                        const AnnotatedLine *NextLine) const {
1306   // In preprocessor directives reserve two chars for trailing " \" if the
1307   // next line continues the preprocessor directive.
1308   bool ContinuesPPDirective =
1309       InPPDirective &&
1310       // If there is no next line, this is likely a child line and the parent
1311       // continues the preprocessor directive.
1312       (!NextLine ||
1313        (NextLine->InPPDirective &&
1314         // If there is an unescaped newline between this line and the next, the
1315         // next line starts a new preprocessor directive.
1316         !NextLine->First->HasUnescapedNewline));
1317   return Style.ColumnLimit - (ContinuesPPDirective ? 2 : 0);
1318 }
1319 
1320 } // namespace format
1321 } // namespace clang
1322