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