UnwrappedLineFormatter.cpp revision 355940
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
17namespace clang {
18namespace format {
19
20namespace {
21
22bool 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.
38class LevelIndentTracker {
39public:
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  }
68
69  /// Update the indent state given that \p Line indent should be
70  /// skipped.
71  void skipLine(const AnnotatedLine &Line) {
72    while (IndentForLevel.size() <= Line.Level)
73      IndentForLevel.push_back(Indent);
74  }
75
76  /// Update the level indent to adapt to the given \p Line.
77  ///
78  /// When a line is not formatted, we move the subsequent lines on the same
79  /// level to the same indent.
80  /// Note that \c nextLine must have been called before this method.
81  void adjustToUnmodifiedLine(const AnnotatedLine &Line) {
82    unsigned LevelIndent = Line.First->OriginalColumn;
83    if (static_cast<int>(LevelIndent) - Offset >= 0)
84      LevelIndent -= Offset;
85    if ((!Line.First->is(tok::comment) || IndentForLevel[Line.Level] == -1) &&
86        !Line.InPPDirective)
87      IndentForLevel[Line.Level] = LevelIndent;
88  }
89
90private:
91  /// Get the offset of the line relatively to the level.
92  ///
93  /// For example, 'public:' labels in classes are offset by 1 or 2
94  /// characters to the left from their level.
95  int getIndentOffset(const FormatToken &RootToken) {
96    if (Style.Language == FormatStyle::LK_Java ||
97        Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp())
98      return 0;
99    if (RootToken.isAccessSpecifier(false) ||
100        RootToken.isObjCAccessSpecifier() ||
101        (RootToken.isOneOf(Keywords.kw_signals, Keywords.kw_qsignals) &&
102         RootToken.Next && RootToken.Next->is(tok::colon)))
103      return Style.AccessModifierOffset;
104    return 0;
105  }
106
107  /// Get the indent of \p Level from \p IndentForLevel.
108  ///
109  /// \p IndentForLevel must contain the indent for the level \c l
110  /// at \p IndentForLevel[l], or a value < 0 if the indent for
111  /// that level is unknown.
112  unsigned getIndent(ArrayRef<int> IndentForLevel, unsigned Level) {
113    if (IndentForLevel[Level] != -1)
114      return IndentForLevel[Level];
115    if (Level == 0)
116      return 0;
117    return getIndent(IndentForLevel, Level - 1) + Style.IndentWidth;
118  }
119
120  const FormatStyle &Style;
121  const AdditionalKeywords &Keywords;
122  const unsigned AdditionalIndent;
123
124  /// The indent in characters for each level.
125  std::vector<int> IndentForLevel;
126
127  /// Offset of the current line relative to the indent level.
128  ///
129  /// For example, the 'public' keywords is often indented with a negative
130  /// offset.
131  int Offset = 0;
132
133  /// The current line's indent.
134  unsigned Indent = 0;
135};
136
137const FormatToken *getMatchingNamespaceToken(
138    const AnnotatedLine *Line,
139    const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
140  if (!Line->startsWith(tok::r_brace))
141    return nullptr;
142  size_t StartLineIndex = Line->MatchingOpeningBlockLineIndex;
143  if (StartLineIndex == UnwrappedLine::kInvalidIndex)
144    return nullptr;
145  assert(StartLineIndex < AnnotatedLines.size());
146  return AnnotatedLines[StartLineIndex]->First->getNamespaceToken();
147}
148
149StringRef getNamespaceTokenText(const AnnotatedLine *Line) {
150  const FormatToken *NamespaceToken = Line->First->getNamespaceToken();
151  return NamespaceToken ? NamespaceToken->TokenText : StringRef();
152}
153
154StringRef getMatchingNamespaceTokenText(
155    const AnnotatedLine *Line,
156    const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
157  const FormatToken *NamespaceToken =
158      getMatchingNamespaceToken(Line, AnnotatedLines);
159  return NamespaceToken ? NamespaceToken->TokenText : StringRef();
160}
161
162class LineJoiner {
163public:
164  LineJoiner(const FormatStyle &Style, const AdditionalKeywords &Keywords,
165             const SmallVectorImpl<AnnotatedLine *> &Lines)
166      : Style(Style), Keywords(Keywords), End(Lines.end()), Next(Lines.begin()),
167        AnnotatedLines(Lines) {}
168
169  /// Returns the next line, merging multiple lines into one if possible.
170  const AnnotatedLine *getNextMergedLine(bool DryRun,
171                                         LevelIndentTracker &IndentTracker) {
172    if (Next == End)
173      return nullptr;
174    const AnnotatedLine *Current = *Next;
175    IndentTracker.nextLine(*Current);
176    unsigned MergedLines = tryFitMultipleLinesInOne(IndentTracker, Next, End);
177    if (MergedLines > 0 && Style.ColumnLimit == 0)
178      // Disallow line merging if there is a break at the start of one of the
179      // input lines.
180      for (unsigned i = 0; i < MergedLines; ++i)
181        if (Next[i + 1]->First->NewlinesBefore > 0)
182          MergedLines = 0;
183    if (!DryRun)
184      for (unsigned i = 0; i < MergedLines; ++i)
185        join(*Next[0], *Next[i + 1]);
186    Next = Next + MergedLines + 1;
187    return Current;
188  }
189
190private:
191  /// Calculates how many lines can be merged into 1 starting at \p I.
192  unsigned
193  tryFitMultipleLinesInOne(LevelIndentTracker &IndentTracker,
194                           SmallVectorImpl<AnnotatedLine *>::const_iterator I,
195                           SmallVectorImpl<AnnotatedLine *>::const_iterator E) {
196    const unsigned Indent = IndentTracker.getIndent();
197
198    // Can't join the last line with anything.
199    if (I + 1 == E)
200      return 0;
201    // We can never merge stuff if there are trailing line comments.
202    const AnnotatedLine *TheLine = *I;
203    if (TheLine->Last->is(TT_LineComment))
204      return 0;
205    if (I[1]->Type == LT_Invalid || I[1]->First->MustBreakBefore)
206      return 0;
207    if (TheLine->InPPDirective &&
208        (!I[1]->InPPDirective || I[1]->First->HasUnescapedNewline))
209      return 0;
210
211    if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit)
212      return 0;
213
214    unsigned Limit =
215        Style.ColumnLimit == 0 ? UINT_MAX : Style.ColumnLimit - Indent;
216    // If we already exceed the column limit, we set 'Limit' to 0. The different
217    // tryMerge..() functions can then decide whether to still do merging.
218    Limit = TheLine->Last->TotalLength > Limit
219                ? 0
220                : Limit - TheLine->Last->TotalLength;
221
222    if (TheLine->Last->is(TT_FunctionLBrace) &&
223        TheLine->First == TheLine->Last &&
224        !Style.BraceWrapping.SplitEmptyFunction &&
225        I[1]->First->is(tok::r_brace))
226      return tryMergeSimpleBlock(I, E, Limit);
227
228    // Handle empty record blocks where the brace has already been wrapped
229    if (TheLine->Last->is(tok::l_brace) && TheLine->First == TheLine->Last &&
230        I != AnnotatedLines.begin()) {
231      bool EmptyBlock = I[1]->First->is(tok::r_brace);
232
233      const FormatToken *Tok = I[-1]->First;
234      if (Tok && Tok->is(tok::comment))
235        Tok = Tok->getNextNonComment();
236
237      if (Tok && Tok->getNamespaceToken())
238        return !Style.BraceWrapping.SplitEmptyNamespace && EmptyBlock
239                   ? tryMergeSimpleBlock(I, E, Limit)
240                   : 0;
241
242      if (Tok && Tok->is(tok::kw_typedef))
243        Tok = Tok->getNextNonComment();
244      if (Tok && Tok->isOneOf(tok::kw_class, tok::kw_struct, tok::kw_union,
245                              tok::kw_extern, Keywords.kw_interface))
246        return !Style.BraceWrapping.SplitEmptyRecord && EmptyBlock
247                   ? tryMergeSimpleBlock(I, E, Limit)
248                   : 0;
249    }
250
251    // FIXME: TheLine->Level != 0 might or might not be the right check to do.
252    // If necessary, change to something smarter.
253    bool MergeShortFunctions =
254        Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_All ||
255        (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
256         I[1]->First->is(tok::r_brace)) ||
257        (Style.AllowShortFunctionsOnASingleLine & FormatStyle::SFS_InlineOnly &&
258         TheLine->Level != 0);
259
260    if (Style.CompactNamespaces) {
261      if (auto nsToken = TheLine->First->getNamespaceToken()) {
262        int i = 0;
263        unsigned closingLine = TheLine->MatchingClosingBlockLineIndex - 1;
264        for (; I + 1 + i != E &&
265               nsToken->TokenText == getNamespaceTokenText(I[i + 1]) &&
266               closingLine == I[i + 1]->MatchingClosingBlockLineIndex &&
267               I[i + 1]->Last->TotalLength < Limit;
268             i++, closingLine--) {
269          // No extra indent for compacted namespaces
270          IndentTracker.skipLine(*I[i + 1]);
271
272          Limit -= I[i + 1]->Last->TotalLength;
273        }
274        return i;
275      }
276
277      if (auto nsToken = getMatchingNamespaceToken(TheLine, AnnotatedLines)) {
278        int i = 0;
279        unsigned openingLine = TheLine->MatchingOpeningBlockLineIndex - 1;
280        for (; I + 1 + i != E &&
281               nsToken->TokenText ==
282                   getMatchingNamespaceTokenText(I[i + 1], AnnotatedLines) &&
283               openingLine == I[i + 1]->MatchingOpeningBlockLineIndex;
284             i++, openingLine--) {
285          // No space between consecutive braces
286          I[i + 1]->First->SpacesRequiredBefore = !I[i]->Last->is(tok::r_brace);
287
288          // Indent like the outer-most namespace
289          IndentTracker.nextLine(*I[i + 1]);
290        }
291        return i;
292      }
293    }
294
295    // Try to merge a function block with left brace unwrapped
296    if (TheLine->Last->is(TT_FunctionLBrace) &&
297        TheLine->First != TheLine->Last) {
298      return MergeShortFunctions ? tryMergeSimpleBlock(I, E, Limit) : 0;
299    }
300    // Try to merge a control statement block with left brace unwrapped
301    if (TheLine->Last->is(tok::l_brace) && TheLine->First != TheLine->Last &&
302        TheLine->First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for)) {
303      return Style.AllowShortBlocksOnASingleLine
304                 ? tryMergeSimpleBlock(I, E, Limit)
305                 : 0;
306    }
307    // Try to merge a control statement block with left brace wrapped
308    if (I[1]->First->is(tok::l_brace) &&
309        TheLine->First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for)) {
310      return Style.BraceWrapping.AfterControlStatement
311                 ? tryMergeSimpleBlock(I, E, Limit)
312                 : 0;
313    }
314    // Try to merge either empty or one-line block if is precedeed by control
315    // statement token
316    if (TheLine->First->is(tok::l_brace) && TheLine->First == TheLine->Last &&
317        I != AnnotatedLines.begin() &&
318        I[-1]->First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for)) {
319      unsigned MergedLines = 0;
320      if (Style.AllowShortBlocksOnASingleLine) {
321        MergedLines = tryMergeSimpleBlock(I - 1, E, Limit);
322        // If we managed to merge the block, discard the first merged line
323        // since we are merging starting from I.
324        if (MergedLines > 0)
325          --MergedLines;
326      }
327      return MergedLines;
328    }
329    // Don't merge block with left brace wrapped after ObjC special blocks
330    if (TheLine->First->is(tok::l_brace) && I != AnnotatedLines.begin() &&
331        I[-1]->First->is(tok::at) && I[-1]->First->Next) {
332      tok::ObjCKeywordKind kwId = I[-1]->First->Next->Tok.getObjCKeywordID();
333      if (kwId == clang::tok::objc_autoreleasepool ||
334          kwId == clang::tok::objc_synchronized)
335        return 0;
336    }
337    // Don't merge block with left brace wrapped after case labels
338    if (TheLine->First->is(tok::l_brace) && I != AnnotatedLines.begin() &&
339        I[-1]->First->isOneOf(tok::kw_case, tok::kw_default))
340      return 0;
341    // Try to merge a block with left brace wrapped that wasn't yet covered
342    if (TheLine->Last->is(tok::l_brace)) {
343      return !Style.BraceWrapping.AfterFunction ||
344                     (I[1]->First->is(tok::r_brace) &&
345                      !Style.BraceWrapping.SplitEmptyRecord)
346                 ? tryMergeSimpleBlock(I, E, Limit)
347                 : 0;
348    }
349    // Try to merge a function block with left brace wrapped
350    if (I[1]->First->is(TT_FunctionLBrace) &&
351        Style.BraceWrapping.AfterFunction) {
352      if (I[1]->Last->is(TT_LineComment))
353        return 0;
354
355      // Check for Limit <= 2 to account for the " {".
356      if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(TheLine)))
357        return 0;
358      Limit -= 2;
359
360      unsigned MergedLines = 0;
361      if (MergeShortFunctions ||
362          (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
363           I[1]->First == I[1]->Last && I + 2 != E &&
364           I[2]->First->is(tok::r_brace))) {
365        MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
366        // If we managed to merge the block, count the function header, which is
367        // on a separate line.
368        if (MergedLines > 0)
369          ++MergedLines;
370      }
371      return MergedLines;
372    }
373    if (TheLine->First->is(tok::kw_if)) {
374      return Style.AllowShortIfStatementsOnASingleLine
375                 ? tryMergeSimpleControlStatement(I, E, Limit)
376                 : 0;
377    }
378    if (TheLine->First->isOneOf(tok::kw_for, tok::kw_while)) {
379      return Style.AllowShortLoopsOnASingleLine
380                 ? tryMergeSimpleControlStatement(I, E, Limit)
381                 : 0;
382    }
383    if (TheLine->First->isOneOf(tok::kw_case, tok::kw_default)) {
384      return Style.AllowShortCaseLabelsOnASingleLine
385                 ? tryMergeShortCaseLabels(I, E, Limit)
386                 : 0;
387    }
388    if (TheLine->InPPDirective &&
389        (TheLine->First->HasUnescapedNewline || TheLine->First->IsFirst)) {
390      return tryMergeSimplePPDirective(I, E, Limit);
391    }
392    return 0;
393  }
394
395  unsigned
396  tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
397                            SmallVectorImpl<AnnotatedLine *>::const_iterator E,
398                            unsigned Limit) {
399    if (Limit == 0)
400      return 0;
401    if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline)
402      return 0;
403    if (1 + I[1]->Last->TotalLength > Limit)
404      return 0;
405    return 1;
406  }
407
408  unsigned tryMergeSimpleControlStatement(
409      SmallVectorImpl<AnnotatedLine *>::const_iterator I,
410      SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) {
411    if (Limit == 0)
412      return 0;
413    if (Style.BraceWrapping.AfterControlStatement &&
414        (I[1]->First->is(tok::l_brace) && !Style.AllowShortBlocksOnASingleLine))
415      return 0;
416    if (I[1]->InPPDirective != (*I)->InPPDirective ||
417        (I[1]->InPPDirective && I[1]->First->HasUnescapedNewline))
418      return 0;
419    Limit = limitConsideringMacros(I + 1, E, Limit);
420    AnnotatedLine &Line = **I;
421    if (Line.Last->isNot(tok::r_paren))
422      return 0;
423    if (1 + I[1]->Last->TotalLength > Limit)
424      return 0;
425    if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for, tok::kw_while,
426                             TT_LineComment))
427      return 0;
428    // Only inline simple if's (no nested if or else), unless specified
429    if (Style.AllowShortIfStatementsOnASingleLine != FormatStyle::SIS_Always) {
430      if (I + 2 != E && Line.startsWith(tok::kw_if) &&
431          I[2]->First->is(tok::kw_else))
432        return 0;
433    }
434    return 1;
435  }
436
437  unsigned
438  tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
439                          SmallVectorImpl<AnnotatedLine *>::const_iterator E,
440                          unsigned Limit) {
441    if (Limit == 0 || I + 1 == E ||
442        I[1]->First->isOneOf(tok::kw_case, tok::kw_default))
443      return 0;
444    if (I[0]->Last->is(tok::l_brace) || I[1]->First->is(tok::l_brace))
445      return 0;
446    unsigned NumStmts = 0;
447    unsigned Length = 0;
448    bool EndsWithComment = false;
449    bool InPPDirective = I[0]->InPPDirective;
450    const unsigned Level = I[0]->Level;
451    for (; NumStmts < 3; ++NumStmts) {
452      if (I + 1 + NumStmts == E)
453        break;
454      const AnnotatedLine *Line = I[1 + NumStmts];
455      if (Line->InPPDirective != InPPDirective)
456        break;
457      if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
458        break;
459      if (Line->First->isOneOf(tok::kw_if, tok::kw_for, tok::kw_switch,
460                               tok::kw_while) ||
461          EndsWithComment)
462        return 0;
463      if (Line->First->is(tok::comment)) {
464        if (Level != Line->Level)
465          return 0;
466        SmallVectorImpl<AnnotatedLine *>::const_iterator J = I + 2 + NumStmts;
467        for (; J != E; ++J) {
468          Line = *J;
469          if (Line->InPPDirective != InPPDirective)
470            break;
471          if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
472            break;
473          if (Line->First->isNot(tok::comment) || Level != Line->Level)
474            return 0;
475        }
476        break;
477      }
478      if (Line->Last->is(tok::comment))
479        EndsWithComment = true;
480      Length += I[1 + NumStmts]->Last->TotalLength + 1; // 1 for the space.
481    }
482    if (NumStmts == 0 || NumStmts == 3 || Length > Limit)
483      return 0;
484    return NumStmts;
485  }
486
487  unsigned
488  tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
489                      SmallVectorImpl<AnnotatedLine *>::const_iterator E,
490                      unsigned Limit) {
491    AnnotatedLine &Line = **I;
492
493    // Don't merge ObjC @ keywords and methods.
494    // FIXME: If an option to allow short exception handling clauses on a single
495    // line is added, change this to not return for @try and friends.
496    if (Style.Language != FormatStyle::LK_Java &&
497        Line.First->isOneOf(tok::at, tok::minus, tok::plus))
498      return 0;
499
500    // Check that the current line allows merging. This depends on whether we
501    // are in a control flow statements as well as several style flags.
502    if (Line.First->isOneOf(tok::kw_else, tok::kw_case) ||
503        (Line.First->Next && Line.First->Next->is(tok::kw_else)))
504      return 0;
505    // default: in switch statement
506    if (Line.First->is(tok::kw_default)) {
507      const FormatToken *Tok = Line.First->getNextNonComment();
508      if (Tok && Tok->is(tok::colon))
509        return 0;
510    }
511    if (Line.First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_do, tok::kw_try,
512                            tok::kw___try, tok::kw_catch, tok::kw___finally,
513                            tok::kw_for, tok::r_brace, Keywords.kw___except)) {
514      if (!Style.AllowShortBlocksOnASingleLine)
515        return 0;
516      // Don't merge when we can't except the case when
517      // the control statement block is empty
518      if (!Style.AllowShortIfStatementsOnASingleLine &&
519          Line.startsWith(tok::kw_if) &&
520          !Style.BraceWrapping.AfterControlStatement &&
521          !I[1]->First->is(tok::r_brace))
522        return 0;
523      if (!Style.AllowShortIfStatementsOnASingleLine &&
524          Line.startsWith(tok::kw_if) &&
525          Style.BraceWrapping.AfterControlStatement && I + 2 != E &&
526          !I[2]->First->is(tok::r_brace))
527        return 0;
528      if (!Style.AllowShortLoopsOnASingleLine &&
529          Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for) &&
530          !Style.BraceWrapping.AfterControlStatement &&
531          !I[1]->First->is(tok::r_brace))
532        return 0;
533      if (!Style.AllowShortLoopsOnASingleLine &&
534          Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for) &&
535          Style.BraceWrapping.AfterControlStatement && I + 2 != E &&
536          !I[2]->First->is(tok::r_brace))
537        return 0;
538      // FIXME: Consider an option to allow short exception handling clauses on
539      // a single line.
540      // FIXME: This isn't covered by tests.
541      // FIXME: For catch, __except, __finally the first token on the line
542      // is '}', so this isn't correct here.
543      if (Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch,
544                              Keywords.kw___except, tok::kw___finally))
545        return 0;
546    }
547
548    if (Line.Last->is(tok::l_brace)) {
549      FormatToken *Tok = I[1]->First;
550      if (Tok->is(tok::r_brace) && !Tok->MustBreakBefore &&
551          (Tok->getNextNonComment() == nullptr ||
552           Tok->getNextNonComment()->is(tok::semi))) {
553        // We merge empty blocks even if the line exceeds the column limit.
554        Tok->SpacesRequiredBefore = 0;
555        Tok->CanBreakBefore = true;
556        return 1;
557      } else if (Limit != 0 && !Line.startsWithNamespace() &&
558                 !startsExternCBlock(Line)) {
559        // We don't merge short records.
560        FormatToken *RecordTok = Line.First;
561        // Skip record modifiers.
562        while (RecordTok->Next &&
563               RecordTok->isOneOf(tok::kw_typedef, tok::kw_export,
564                                  Keywords.kw_declare, Keywords.kw_abstract,
565                                  tok::kw_default))
566          RecordTok = RecordTok->Next;
567        if (RecordTok &&
568            RecordTok->isOneOf(tok::kw_class, tok::kw_union, tok::kw_struct,
569                               Keywords.kw_interface))
570          return 0;
571
572        // Check that we still have three lines and they fit into the limit.
573        if (I + 2 == E || I[2]->Type == LT_Invalid)
574          return 0;
575        Limit = limitConsideringMacros(I + 2, E, Limit);
576
577        if (!nextTwoLinesFitInto(I, Limit))
578          return 0;
579
580        // Second, check that the next line does not contain any braces - if it
581        // does, readability declines when putting it into a single line.
582        if (I[1]->Last->is(TT_LineComment))
583          return 0;
584        do {
585          if (Tok->is(tok::l_brace) && Tok->BlockKind != BK_BracedInit)
586            return 0;
587          Tok = Tok->Next;
588        } while (Tok);
589
590        // Last, check that the third line starts with a closing brace.
591        Tok = I[2]->First;
592        if (Tok->isNot(tok::r_brace))
593          return 0;
594
595        // Don't merge "if (a) { .. } else {".
596        if (Tok->Next && Tok->Next->is(tok::kw_else))
597          return 0;
598
599        return 2;
600      }
601    } else if (I[1]->First->is(tok::l_brace)) {
602      if (I[1]->Last->is(TT_LineComment))
603        return 0;
604
605      // Check for Limit <= 2 to account for the " {".
606      if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(*I)))
607        return 0;
608      Limit -= 2;
609      unsigned MergedLines = 0;
610      if (Style.AllowShortBlocksOnASingleLine ||
611          (I[1]->First == I[1]->Last && I + 2 != E &&
612           I[2]->First->is(tok::r_brace))) {
613        MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
614        // If we managed to merge the block, count the statement header, which
615        // is on a separate line.
616        if (MergedLines > 0)
617          ++MergedLines;
618      }
619      return MergedLines;
620    }
621    return 0;
622  }
623
624  /// Returns the modified column limit for \p I if it is inside a macro and
625  /// needs a trailing '\'.
626  unsigned
627  limitConsideringMacros(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
628                         SmallVectorImpl<AnnotatedLine *>::const_iterator E,
629                         unsigned Limit) {
630    if (I[0]->InPPDirective && I + 1 != E &&
631        !I[1]->First->HasUnescapedNewline && !I[1]->First->is(tok::eof)) {
632      return Limit < 2 ? 0 : Limit - 2;
633    }
634    return Limit;
635  }
636
637  bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
638                           unsigned Limit) {
639    if (I[1]->First->MustBreakBefore || I[2]->First->MustBreakBefore)
640      return false;
641    return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit;
642  }
643
644  bool containsMustBreak(const AnnotatedLine *Line) {
645    for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) {
646      if (Tok->MustBreakBefore)
647        return true;
648    }
649    return false;
650  }
651
652  void join(AnnotatedLine &A, const AnnotatedLine &B) {
653    assert(!A.Last->Next);
654    assert(!B.First->Previous);
655    if (B.Affected)
656      A.Affected = true;
657    A.Last->Next = B.First;
658    B.First->Previous = A.Last;
659    B.First->CanBreakBefore = true;
660    unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore;
661    for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) {
662      Tok->TotalLength += LengthA;
663      A.Last = Tok;
664    }
665  }
666
667  const FormatStyle &Style;
668  const AdditionalKeywords &Keywords;
669  const SmallVectorImpl<AnnotatedLine *>::const_iterator End;
670
671  SmallVectorImpl<AnnotatedLine *>::const_iterator Next;
672  const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines;
673};
674
675static void markFinalized(FormatToken *Tok) {
676  for (; Tok; Tok = Tok->Next) {
677    Tok->Finalized = true;
678    for (AnnotatedLine *Child : Tok->Children)
679      markFinalized(Child->First);
680  }
681}
682
683#ifndef NDEBUG
684static void printLineState(const LineState &State) {
685  llvm::dbgs() << "State: ";
686  for (const ParenState &P : State.Stack) {
687    llvm::dbgs() << (P.Tok ? P.Tok->TokenText : "F") << "|" << P.Indent << "|"
688                 << P.LastSpace << "|" << P.NestedBlockIndent << " ";
689  }
690  llvm::dbgs() << State.NextToken->TokenText << "\n";
691}
692#endif
693
694/// Base class for classes that format one \c AnnotatedLine.
695class LineFormatter {
696public:
697  LineFormatter(ContinuationIndenter *Indenter, WhitespaceManager *Whitespaces,
698                const FormatStyle &Style,
699                UnwrappedLineFormatter *BlockFormatter)
700      : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style),
701        BlockFormatter(BlockFormatter) {}
702  virtual ~LineFormatter() {}
703
704  /// Formats an \c AnnotatedLine and returns the penalty.
705  ///
706  /// If \p DryRun is \c false, directly applies the changes.
707  virtual unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
708                              unsigned FirstStartColumn, bool DryRun) = 0;
709
710protected:
711  /// If the \p State's next token is an r_brace closing a nested block,
712  /// format the nested block before it.
713  ///
714  /// Returns \c true if all children could be placed successfully and adapts
715  /// \p Penalty as well as \p State. If \p DryRun is false, also directly
716  /// creates changes using \c Whitespaces.
717  ///
718  /// The crucial idea here is that children always get formatted upon
719  /// encountering the closing brace right after the nested block. Now, if we
720  /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is
721  /// \c false), the entire block has to be kept on the same line (which is only
722  /// possible if it fits on the line, only contains a single statement, etc.
723  ///
724  /// If \p NewLine is true, we format the nested block on separate lines, i.e.
725  /// break after the "{", format all lines with correct indentation and the put
726  /// the closing "}" on yet another new line.
727  ///
728  /// This enables us to keep the simple structure of the
729  /// \c UnwrappedLineFormatter, where we only have two options for each token:
730  /// break or don't break.
731  bool formatChildren(LineState &State, bool NewLine, bool DryRun,
732                      unsigned &Penalty) {
733    const FormatToken *LBrace = State.NextToken->getPreviousNonComment();
734    FormatToken &Previous = *State.NextToken->Previous;
735    if (!LBrace || LBrace->isNot(tok::l_brace) ||
736        LBrace->BlockKind != BK_Block || Previous.Children.size() == 0)
737      // The previous token does not open a block. Nothing to do. We don't
738      // assert so that we can simply call this function for all tokens.
739      return true;
740
741    if (NewLine) {
742      int AdditionalIndent = State.Stack.back().Indent -
743                             Previous.Children[0]->Level * Style.IndentWidth;
744
745      Penalty +=
746          BlockFormatter->format(Previous.Children, DryRun, AdditionalIndent,
747                                 /*FixBadIndentation=*/true);
748      return true;
749    }
750
751    if (Previous.Children[0]->First->MustBreakBefore)
752      return false;
753
754    // Cannot merge into one line if this line ends on a comment.
755    if (Previous.is(tok::comment))
756      return false;
757
758    // Cannot merge multiple statements into a single line.
759    if (Previous.Children.size() > 1)
760      return false;
761
762    const AnnotatedLine *Child = Previous.Children[0];
763    // We can't put the closing "}" on a line with a trailing comment.
764    if (Child->Last->isTrailingComment())
765      return false;
766
767    // If the child line exceeds the column limit, we wouldn't want to merge it.
768    // We add +2 for the trailing " }".
769    if (Style.ColumnLimit > 0 &&
770        Child->Last->TotalLength + State.Column + 2 > Style.ColumnLimit)
771      return false;
772
773    if (!DryRun) {
774      Whitespaces->replaceWhitespace(
775          *Child->First, /*Newlines=*/0, /*Spaces=*/1,
776          /*StartOfTokenColumn=*/State.Column, State.Line->InPPDirective);
777    }
778    Penalty +=
779        formatLine(*Child, State.Column + 1, /*FirstStartColumn=*/0, DryRun);
780
781    State.Column += 1 + Child->Last->TotalLength;
782    return true;
783  }
784
785  ContinuationIndenter *Indenter;
786
787private:
788  WhitespaceManager *Whitespaces;
789  const FormatStyle &Style;
790  UnwrappedLineFormatter *BlockFormatter;
791};
792
793/// Formatter that keeps the existing line breaks.
794class NoColumnLimitLineFormatter : public LineFormatter {
795public:
796  NoColumnLimitLineFormatter(ContinuationIndenter *Indenter,
797                             WhitespaceManager *Whitespaces,
798                             const FormatStyle &Style,
799                             UnwrappedLineFormatter *BlockFormatter)
800      : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
801
802  /// Formats the line, simply keeping all of the input's line breaking
803  /// decisions.
804  unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
805                      unsigned FirstStartColumn, bool DryRun) override {
806    assert(!DryRun);
807    LineState State = Indenter->getInitialState(FirstIndent, FirstStartColumn,
808                                                &Line, /*DryRun=*/false);
809    while (State.NextToken) {
810      bool Newline =
811          Indenter->mustBreak(State) ||
812          (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0);
813      unsigned Penalty = 0;
814      formatChildren(State, Newline, /*DryRun=*/false, Penalty);
815      Indenter->addTokenToState(State, Newline, /*DryRun=*/false);
816    }
817    return 0;
818  }
819};
820
821/// Formatter that puts all tokens into a single line without breaks.
822class NoLineBreakFormatter : public LineFormatter {
823public:
824  NoLineBreakFormatter(ContinuationIndenter *Indenter,
825                       WhitespaceManager *Whitespaces, const FormatStyle &Style,
826                       UnwrappedLineFormatter *BlockFormatter)
827      : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
828
829  /// Puts all tokens into a single line.
830  unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
831                      unsigned FirstStartColumn, bool DryRun) override {
832    unsigned Penalty = 0;
833    LineState State =
834        Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun);
835    while (State.NextToken) {
836      formatChildren(State, /*NewLine=*/false, DryRun, Penalty);
837      Indenter->addTokenToState(
838          State, /*Newline=*/State.NextToken->MustBreakBefore, DryRun);
839    }
840    return Penalty;
841  }
842};
843
844/// Finds the best way to break lines.
845class OptimizingLineFormatter : public LineFormatter {
846public:
847  OptimizingLineFormatter(ContinuationIndenter *Indenter,
848                          WhitespaceManager *Whitespaces,
849                          const FormatStyle &Style,
850                          UnwrappedLineFormatter *BlockFormatter)
851      : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
852
853  /// Formats the line by finding the best line breaks with line lengths
854  /// below the column limit.
855  unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
856                      unsigned FirstStartColumn, bool DryRun) override {
857    LineState State =
858        Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun);
859
860    // If the ObjC method declaration does not fit on a line, we should format
861    // it with one arg per line.
862    if (State.Line->Type == LT_ObjCMethodDecl)
863      State.Stack.back().BreakBeforeParameter = true;
864
865    // Find best solution in solution space.
866    return analyzeSolutionSpace(State, DryRun);
867  }
868
869private:
870  struct CompareLineStatePointers {
871    bool operator()(LineState *obj1, LineState *obj2) const {
872      return *obj1 < *obj2;
873    }
874  };
875
876  /// A pair of <penalty, count> that is used to prioritize the BFS on.
877  ///
878  /// In case of equal penalties, we want to prefer states that were inserted
879  /// first. During state generation we make sure that we insert states first
880  /// that break the line as late as possible.
881  typedef std::pair<unsigned, unsigned> OrderedPenalty;
882
883  /// An edge in the solution space from \c Previous->State to \c State,
884  /// inserting a newline dependent on the \c NewLine.
885  struct StateNode {
886    StateNode(const LineState &State, bool NewLine, StateNode *Previous)
887        : State(State), NewLine(NewLine), Previous(Previous) {}
888    LineState State;
889    bool NewLine;
890    StateNode *Previous;
891  };
892
893  /// An item in the prioritized BFS search queue. The \c StateNode's
894  /// \c State has the given \c OrderedPenalty.
895  typedef std::pair<OrderedPenalty, StateNode *> QueueItem;
896
897  /// The BFS queue type.
898  typedef std::priority_queue<QueueItem, std::vector<QueueItem>,
899                              std::greater<QueueItem>>
900      QueueType;
901
902  /// Analyze the entire solution space starting from \p InitialState.
903  ///
904  /// This implements a variant of Dijkstra's algorithm on the graph that spans
905  /// the solution space (\c LineStates are the nodes). The algorithm tries to
906  /// find the shortest path (the one with lowest penalty) from \p InitialState
907  /// to a state where all tokens are placed. Returns the penalty.
908  ///
909  /// If \p DryRun is \c false, directly applies the changes.
910  unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun) {
911    std::set<LineState *, CompareLineStatePointers> Seen;
912
913    // Increasing count of \c StateNode items we have created. This is used to
914    // create a deterministic order independent of the container.
915    unsigned Count = 0;
916    QueueType Queue;
917
918    // Insert start element into queue.
919    StateNode *Node =
920        new (Allocator.Allocate()) StateNode(InitialState, false, nullptr);
921    Queue.push(QueueItem(OrderedPenalty(0, Count), Node));
922    ++Count;
923
924    unsigned Penalty = 0;
925
926    // While not empty, take first element and follow edges.
927    while (!Queue.empty()) {
928      Penalty = Queue.top().first.first;
929      StateNode *Node = Queue.top().second;
930      if (!Node->State.NextToken) {
931        LLVM_DEBUG(llvm::dbgs()
932                   << "\n---\nPenalty for line: " << Penalty << "\n");
933        break;
934      }
935      Queue.pop();
936
937      // Cut off the analysis of certain solutions if the analysis gets too
938      // complex. See description of IgnoreStackForComparison.
939      if (Count > 50000)
940        Node->State.IgnoreStackForComparison = true;
941
942      if (!Seen.insert(&Node->State).second)
943        // State already examined with lower penalty.
944        continue;
945
946      FormatDecision LastFormat = Node->State.NextToken->Decision;
947      if (LastFormat == FD_Unformatted || LastFormat == FD_Continue)
948        addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue);
949      if (LastFormat == FD_Unformatted || LastFormat == FD_Break)
950        addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue);
951    }
952
953    if (Queue.empty()) {
954      // We were unable to find a solution, do nothing.
955      // FIXME: Add diagnostic?
956      LLVM_DEBUG(llvm::dbgs() << "Could not find a solution.\n");
957      return 0;
958    }
959
960    // Reconstruct the solution.
961    if (!DryRun)
962      reconstructPath(InitialState, Queue.top().second);
963
964    LLVM_DEBUG(llvm::dbgs()
965               << "Total number of analyzed states: " << Count << "\n");
966    LLVM_DEBUG(llvm::dbgs() << "---\n");
967
968    return Penalty;
969  }
970
971  /// Add the following state to the analysis queue \c Queue.
972  ///
973  /// Assume the current state is \p PreviousNode and has been reached with a
974  /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true.
975  void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode,
976                           bool NewLine, unsigned *Count, QueueType *Queue) {
977    if (NewLine && !Indenter->canBreak(PreviousNode->State))
978      return;
979    if (!NewLine && Indenter->mustBreak(PreviousNode->State))
980      return;
981
982    StateNode *Node = new (Allocator.Allocate())
983        StateNode(PreviousNode->State, NewLine, PreviousNode);
984    if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty))
985      return;
986
987    Penalty += Indenter->addTokenToState(Node->State, NewLine, true);
988
989    Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node));
990    ++(*Count);
991  }
992
993  /// Applies the best formatting by reconstructing the path in the
994  /// solution space that leads to \c Best.
995  void reconstructPath(LineState &State, StateNode *Best) {
996    std::deque<StateNode *> Path;
997    // We do not need a break before the initial token.
998    while (Best->Previous) {
999      Path.push_front(Best);
1000      Best = Best->Previous;
1001    }
1002    for (auto I = Path.begin(), E = Path.end(); I != E; ++I) {
1003      unsigned Penalty = 0;
1004      formatChildren(State, (*I)->NewLine, /*DryRun=*/false, Penalty);
1005      Penalty += Indenter->addTokenToState(State, (*I)->NewLine, false);
1006
1007      LLVM_DEBUG({
1008        printLineState((*I)->Previous->State);
1009        if ((*I)->NewLine) {
1010          llvm::dbgs() << "Penalty for placing "
1011                       << (*I)->Previous->State.NextToken->Tok.getName()
1012                       << " on a new line: " << Penalty << "\n";
1013        }
1014      });
1015    }
1016  }
1017
1018  llvm::SpecificBumpPtrAllocator<StateNode> Allocator;
1019};
1020
1021} // anonymous namespace
1022
1023unsigned UnwrappedLineFormatter::format(
1024    const SmallVectorImpl<AnnotatedLine *> &Lines, bool DryRun,
1025    int AdditionalIndent, bool FixBadIndentation, unsigned FirstStartColumn,
1026    unsigned NextStartColumn, unsigned LastStartColumn) {
1027  LineJoiner Joiner(Style, Keywords, Lines);
1028
1029  // Try to look up already computed penalty in DryRun-mode.
1030  std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned> CacheKey(
1031      &Lines, AdditionalIndent);
1032  auto CacheIt = PenaltyCache.find(CacheKey);
1033  if (DryRun && CacheIt != PenaltyCache.end())
1034    return CacheIt->second;
1035
1036  assert(!Lines.empty());
1037  unsigned Penalty = 0;
1038  LevelIndentTracker IndentTracker(Style, Keywords, Lines[0]->Level,
1039                                   AdditionalIndent);
1040  const AnnotatedLine *PreviousLine = nullptr;
1041  const AnnotatedLine *NextLine = nullptr;
1042
1043  // The minimum level of consecutive lines that have been formatted.
1044  unsigned RangeMinLevel = UINT_MAX;
1045
1046  bool FirstLine = true;
1047  for (const AnnotatedLine *Line =
1048           Joiner.getNextMergedLine(DryRun, IndentTracker);
1049       Line; Line = NextLine, FirstLine = false) {
1050    const AnnotatedLine &TheLine = *Line;
1051    unsigned Indent = IndentTracker.getIndent();
1052
1053    // We continue formatting unchanged lines to adjust their indent, e.g. if a
1054    // scope was added. However, we need to carefully stop doing this when we
1055    // exit the scope of affected lines to prevent indenting a the entire
1056    // remaining file if it currently missing a closing brace.
1057    bool PreviousRBrace =
1058        PreviousLine && PreviousLine->startsWith(tok::r_brace);
1059    bool ContinueFormatting =
1060        TheLine.Level > RangeMinLevel ||
1061        (TheLine.Level == RangeMinLevel && !PreviousRBrace &&
1062         !TheLine.startsWith(tok::r_brace));
1063
1064    bool FixIndentation = (FixBadIndentation || ContinueFormatting) &&
1065                          Indent != TheLine.First->OriginalColumn;
1066    bool ShouldFormat = TheLine.Affected || FixIndentation;
1067    // We cannot format this line; if the reason is that the line had a
1068    // parsing error, remember that.
1069    if (ShouldFormat && TheLine.Type == LT_Invalid && Status) {
1070      Status->FormatComplete = false;
1071      Status->Line =
1072          SourceMgr.getSpellingLineNumber(TheLine.First->Tok.getLocation());
1073    }
1074
1075    if (ShouldFormat && TheLine.Type != LT_Invalid) {
1076      if (!DryRun) {
1077        bool LastLine = Line->First->is(tok::eof);
1078        formatFirstToken(TheLine, PreviousLine, Lines, Indent,
1079                         LastLine ? LastStartColumn : NextStartColumn + Indent);
1080      }
1081
1082      NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
1083      unsigned ColumnLimit = getColumnLimit(TheLine.InPPDirective, NextLine);
1084      bool FitsIntoOneLine =
1085          TheLine.Last->TotalLength + Indent <= ColumnLimit ||
1086          (TheLine.Type == LT_ImportStatement &&
1087           (Style.Language != FormatStyle::LK_JavaScript ||
1088            !Style.JavaScriptWrapImports)) ||
1089          (Style.isCSharp() &&
1090           TheLine.InPPDirective); // don't split #regions in C#
1091      if (Style.ColumnLimit == 0)
1092        NoColumnLimitLineFormatter(Indenter, Whitespaces, Style, this)
1093            .formatLine(TheLine, NextStartColumn + Indent,
1094                        FirstLine ? FirstStartColumn : 0, DryRun);
1095      else if (FitsIntoOneLine)
1096        Penalty += NoLineBreakFormatter(Indenter, Whitespaces, Style, this)
1097                       .formatLine(TheLine, NextStartColumn + Indent,
1098                                   FirstLine ? FirstStartColumn : 0, DryRun);
1099      else
1100        Penalty += OptimizingLineFormatter(Indenter, Whitespaces, Style, this)
1101                       .formatLine(TheLine, NextStartColumn + Indent,
1102                                   FirstLine ? FirstStartColumn : 0, DryRun);
1103      RangeMinLevel = std::min(RangeMinLevel, TheLine.Level);
1104    } else {
1105      // If no token in the current line is affected, we still need to format
1106      // affected children.
1107      if (TheLine.ChildrenAffected)
1108        for (const FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next)
1109          if (!Tok->Children.empty())
1110            format(Tok->Children, DryRun);
1111
1112      // Adapt following lines on the current indent level to the same level
1113      // unless the current \c AnnotatedLine is not at the beginning of a line.
1114      bool StartsNewLine =
1115          TheLine.First->NewlinesBefore > 0 || TheLine.First->IsFirst;
1116      if (StartsNewLine)
1117        IndentTracker.adjustToUnmodifiedLine(TheLine);
1118      if (!DryRun) {
1119        bool ReformatLeadingWhitespace =
1120            StartsNewLine && ((PreviousLine && PreviousLine->Affected) ||
1121                              TheLine.LeadingEmptyLinesAffected);
1122        // Format the first token.
1123        if (ReformatLeadingWhitespace)
1124          formatFirstToken(TheLine, PreviousLine, Lines,
1125                           TheLine.First->OriginalColumn,
1126                           TheLine.First->OriginalColumn);
1127        else
1128          Whitespaces->addUntouchableToken(*TheLine.First,
1129                                           TheLine.InPPDirective);
1130
1131        // Notify the WhitespaceManager about the unchanged whitespace.
1132        for (FormatToken *Tok = TheLine.First->Next; Tok; Tok = Tok->Next)
1133          Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective);
1134      }
1135      NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
1136      RangeMinLevel = UINT_MAX;
1137    }
1138    if (!DryRun)
1139      markFinalized(TheLine.First);
1140    PreviousLine = &TheLine;
1141  }
1142  PenaltyCache[CacheKey] = Penalty;
1143  return Penalty;
1144}
1145
1146void UnwrappedLineFormatter::formatFirstToken(
1147    const AnnotatedLine &Line, const AnnotatedLine *PreviousLine,
1148    const SmallVectorImpl<AnnotatedLine *> &Lines, unsigned Indent,
1149    unsigned NewlineIndent) {
1150  FormatToken &RootToken = *Line.First;
1151  if (RootToken.is(tok::eof)) {
1152    unsigned Newlines = std::min(RootToken.NewlinesBefore, 1u);
1153    unsigned TokenIndent = Newlines ? NewlineIndent : 0;
1154    Whitespaces->replaceWhitespace(RootToken, Newlines, TokenIndent,
1155                                   TokenIndent);
1156    return;
1157  }
1158  unsigned Newlines =
1159      std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1);
1160  // Remove empty lines before "}" where applicable.
1161  if (RootToken.is(tok::r_brace) &&
1162      (!RootToken.Next ||
1163       (RootToken.Next->is(tok::semi) && !RootToken.Next->Next)) &&
1164      // Do not remove empty lines before namespace closing "}".
1165      !getNamespaceToken(&Line, Lines))
1166    Newlines = std::min(Newlines, 1u);
1167  // Remove empty lines at the start of nested blocks (lambdas/arrow functions)
1168  if (PreviousLine == nullptr && Line.Level > 0)
1169    Newlines = std::min(Newlines, 1u);
1170  if (Newlines == 0 && !RootToken.IsFirst)
1171    Newlines = 1;
1172  if (RootToken.IsFirst && !RootToken.HasUnescapedNewline)
1173    Newlines = 0;
1174
1175  // Remove empty lines after "{".
1176  if (!Style.KeepEmptyLinesAtTheStartOfBlocks && PreviousLine &&
1177      PreviousLine->Last->is(tok::l_brace) &&
1178      !PreviousLine->startsWithNamespace() &&
1179      !startsExternCBlock(*PreviousLine))
1180    Newlines = 1;
1181
1182  // Insert extra new line before access specifiers.
1183  if (PreviousLine && PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) &&
1184      RootToken.isAccessSpecifier() && RootToken.NewlinesBefore == 1)
1185    ++Newlines;
1186
1187  // Remove empty lines after access specifiers.
1188  if (PreviousLine && PreviousLine->First->isAccessSpecifier() &&
1189      (!PreviousLine->InPPDirective || !RootToken.HasUnescapedNewline))
1190    Newlines = std::min(1u, Newlines);
1191
1192  if (Newlines)
1193    Indent = NewlineIndent;
1194
1195  // Preprocessor directives get indented before the hash only if specified
1196  if (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash &&
1197      (Line.Type == LT_PreprocessorDirective ||
1198       Line.Type == LT_ImportStatement))
1199    Indent = 0;
1200
1201  Whitespaces->replaceWhitespace(RootToken, Newlines, Indent, Indent,
1202                                 Line.InPPDirective &&
1203                                     !RootToken.HasUnescapedNewline);
1204}
1205
1206unsigned
1207UnwrappedLineFormatter::getColumnLimit(bool InPPDirective,
1208                                       const AnnotatedLine *NextLine) const {
1209  // In preprocessor directives reserve two chars for trailing " \" if the
1210  // next line continues the preprocessor directive.
1211  bool ContinuesPPDirective =
1212      InPPDirective &&
1213      // If there is no next line, this is likely a child line and the parent
1214      // continues the preprocessor directive.
1215      (!NextLine ||
1216       (NextLine->InPPDirective &&
1217        // If there is an unescaped newline between this line and the next, the
1218        // next line starts a new preprocessor directive.
1219        !NextLine->First->HasUnescapedNewline));
1220  return Style.ColumnLimit - (ContinuesPPDirective ? 2 : 0);
1221}
1222
1223} // namespace format
1224} // namespace clang
1225