1 //===--- WhitespaceManager.cpp - Format C++ code --------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 ///
9 /// \file
10 /// This file implements WhitespaceManager class.
11 ///
12 //===----------------------------------------------------------------------===//
13 
14 #include "WhitespaceManager.h"
15 #include "llvm/ADT/STLExtras.h"
16 #include "llvm/ADT/SmallVector.h"
17 #include <algorithm>
18 
19 namespace clang {
20 namespace format {
21 
22 bool WhitespaceManager::Change::IsBeforeInFile::operator()(
23     const Change &C1, const Change &C2) const {
24   return SourceMgr.isBeforeInTranslationUnit(
25       C1.OriginalWhitespaceRange.getBegin(),
26       C2.OriginalWhitespaceRange.getBegin());
27 }
28 
29 WhitespaceManager::Change::Change(const FormatToken &Tok,
30                                   bool CreateReplacement,
31                                   SourceRange OriginalWhitespaceRange,
32                                   int Spaces, unsigned StartOfTokenColumn,
33                                   unsigned NewlinesBefore,
34                                   StringRef PreviousLinePostfix,
35                                   StringRef CurrentLinePrefix, bool IsAligned,
36                                   bool ContinuesPPDirective, bool IsInsideToken)
37     : Tok(&Tok), CreateReplacement(CreateReplacement),
38       OriginalWhitespaceRange(OriginalWhitespaceRange),
39       StartOfTokenColumn(StartOfTokenColumn), NewlinesBefore(NewlinesBefore),
40       PreviousLinePostfix(PreviousLinePostfix),
41       CurrentLinePrefix(CurrentLinePrefix), IsAligned(IsAligned),
42       ContinuesPPDirective(ContinuesPPDirective), Spaces(Spaces),
43       IsInsideToken(IsInsideToken), IsTrailingComment(false), TokenLength(0),
44       PreviousEndOfTokenColumn(0), EscapedNewlineColumn(0),
45       StartOfBlockComment(nullptr), IndentationOffset(0), ConditionalsLevel(0) {
46 }
47 
48 void WhitespaceManager::replaceWhitespace(FormatToken &Tok, unsigned Newlines,
49                                           unsigned Spaces,
50                                           unsigned StartOfTokenColumn,
51                                           bool IsAligned, bool InPPDirective) {
52   if (Tok.Finalized)
53     return;
54   Tok.setDecision((Newlines > 0) ? FD_Break : FD_Continue);
55   Changes.push_back(Change(Tok, /*CreateReplacement=*/true, Tok.WhitespaceRange,
56                            Spaces, StartOfTokenColumn, Newlines, "", "",
57                            IsAligned, InPPDirective && !Tok.IsFirst,
58                            /*IsInsideToken=*/false));
59 }
60 
61 void WhitespaceManager::addUntouchableToken(const FormatToken &Tok,
62                                             bool InPPDirective) {
63   if (Tok.Finalized)
64     return;
65   Changes.push_back(Change(Tok, /*CreateReplacement=*/false,
66                            Tok.WhitespaceRange, /*Spaces=*/0,
67                            Tok.OriginalColumn, Tok.NewlinesBefore, "", "",
68                            /*IsAligned=*/false, InPPDirective && !Tok.IsFirst,
69                            /*IsInsideToken=*/false));
70 }
71 
72 llvm::Error
73 WhitespaceManager::addReplacement(const tooling::Replacement &Replacement) {
74   return Replaces.add(Replacement);
75 }
76 
77 bool WhitespaceManager::inputUsesCRLF(StringRef Text, bool DefaultToCRLF) {
78   size_t LF = Text.count('\n');
79   size_t CR = Text.count('\r') * 2;
80   return LF == CR ? DefaultToCRLF : CR > LF;
81 }
82 
83 void WhitespaceManager::replaceWhitespaceInToken(
84     const FormatToken &Tok, unsigned Offset, unsigned ReplaceChars,
85     StringRef PreviousPostfix, StringRef CurrentPrefix, bool InPPDirective,
86     unsigned Newlines, int Spaces) {
87   if (Tok.Finalized)
88     return;
89   SourceLocation Start = Tok.getStartOfNonWhitespace().getLocWithOffset(Offset);
90   Changes.push_back(
91       Change(Tok, /*CreateReplacement=*/true,
92              SourceRange(Start, Start.getLocWithOffset(ReplaceChars)), Spaces,
93              std::max(0, Spaces), Newlines, PreviousPostfix, CurrentPrefix,
94              /*IsAligned=*/true, InPPDirective && !Tok.IsFirst,
95              /*IsInsideToken=*/true));
96 }
97 
98 const tooling::Replacements &WhitespaceManager::generateReplacements() {
99   if (Changes.empty())
100     return Replaces;
101 
102   llvm::sort(Changes, Change::IsBeforeInFile(SourceMgr));
103   calculateLineBreakInformation();
104   alignConsecutiveMacros();
105   alignConsecutiveDeclarations();
106   alignConsecutiveBitFields();
107   alignConsecutiveAssignments();
108   alignChainedConditionals();
109   alignTrailingComments();
110   alignEscapedNewlines();
111   alignArrayInitializers();
112   generateChanges();
113 
114   return Replaces;
115 }
116 
117 void WhitespaceManager::calculateLineBreakInformation() {
118   Changes[0].PreviousEndOfTokenColumn = 0;
119   Change *LastOutsideTokenChange = &Changes[0];
120   for (unsigned i = 1, e = Changes.size(); i != e; ++i) {
121     SourceLocation OriginalWhitespaceStart =
122         Changes[i].OriginalWhitespaceRange.getBegin();
123     SourceLocation PreviousOriginalWhitespaceEnd =
124         Changes[i - 1].OriginalWhitespaceRange.getEnd();
125     unsigned OriginalWhitespaceStartOffset =
126         SourceMgr.getFileOffset(OriginalWhitespaceStart);
127     unsigned PreviousOriginalWhitespaceEndOffset =
128         SourceMgr.getFileOffset(PreviousOriginalWhitespaceEnd);
129     assert(PreviousOriginalWhitespaceEndOffset <=
130            OriginalWhitespaceStartOffset);
131     const char *const PreviousOriginalWhitespaceEndData =
132         SourceMgr.getCharacterData(PreviousOriginalWhitespaceEnd);
133     StringRef Text(PreviousOriginalWhitespaceEndData,
134                    SourceMgr.getCharacterData(OriginalWhitespaceStart) -
135                        PreviousOriginalWhitespaceEndData);
136     // Usually consecutive changes would occur in consecutive tokens. This is
137     // not the case however when analyzing some preprocessor runs of the
138     // annotated lines. For example, in this code:
139     //
140     // #if A // line 1
141     // int i = 1;
142     // #else B // line 2
143     // int i = 2;
144     // #endif // line 3
145     //
146     // one of the runs will produce the sequence of lines marked with line 1, 2
147     // and 3. So the two consecutive whitespace changes just before '// line 2'
148     // and before '#endif // line 3' span multiple lines and tokens:
149     //
150     // #else B{change X}[// line 2
151     // int i = 2;
152     // ]{change Y}#endif // line 3
153     //
154     // For this reason, if the text between consecutive changes spans multiple
155     // newlines, the token length must be adjusted to the end of the original
156     // line of the token.
157     auto NewlinePos = Text.find_first_of('\n');
158     if (NewlinePos == StringRef::npos) {
159       Changes[i - 1].TokenLength = OriginalWhitespaceStartOffset -
160                                    PreviousOriginalWhitespaceEndOffset +
161                                    Changes[i].PreviousLinePostfix.size() +
162                                    Changes[i - 1].CurrentLinePrefix.size();
163     } else {
164       Changes[i - 1].TokenLength =
165           NewlinePos + Changes[i - 1].CurrentLinePrefix.size();
166     }
167 
168     // If there are multiple changes in this token, sum up all the changes until
169     // the end of the line.
170     if (Changes[i - 1].IsInsideToken && Changes[i - 1].NewlinesBefore == 0)
171       LastOutsideTokenChange->TokenLength +=
172           Changes[i - 1].TokenLength + Changes[i - 1].Spaces;
173     else
174       LastOutsideTokenChange = &Changes[i - 1];
175 
176     Changes[i].PreviousEndOfTokenColumn =
177         Changes[i - 1].StartOfTokenColumn + Changes[i - 1].TokenLength;
178 
179     Changes[i - 1].IsTrailingComment =
180         (Changes[i].NewlinesBefore > 0 || Changes[i].Tok->is(tok::eof) ||
181          (Changes[i].IsInsideToken && Changes[i].Tok->is(tok::comment))) &&
182         Changes[i - 1].Tok->is(tok::comment) &&
183         // FIXME: This is a dirty hack. The problem is that
184         // BreakableLineCommentSection does comment reflow changes and here is
185         // the aligning of trailing comments. Consider the case where we reflow
186         // the second line up in this example:
187         //
188         // // line 1
189         // // line 2
190         //
191         // That amounts to 2 changes by BreakableLineCommentSection:
192         //  - the first, delimited by (), for the whitespace between the tokens,
193         //  - and second, delimited by [], for the whitespace at the beginning
194         //  of the second token:
195         //
196         // // line 1(
197         // )[// ]line 2
198         //
199         // So in the end we have two changes like this:
200         //
201         // // line1()[ ]line 2
202         //
203         // Note that the OriginalWhitespaceStart of the second change is the
204         // same as the PreviousOriginalWhitespaceEnd of the first change.
205         // In this case, the below check ensures that the second change doesn't
206         // get treated as a trailing comment change here, since this might
207         // trigger additional whitespace to be wrongly inserted before "line 2"
208         // by the comment aligner here.
209         //
210         // For a proper solution we need a mechanism to say to WhitespaceManager
211         // that a particular change breaks the current sequence of trailing
212         // comments.
213         OriginalWhitespaceStart != PreviousOriginalWhitespaceEnd;
214   }
215   // FIXME: The last token is currently not always an eof token; in those
216   // cases, setting TokenLength of the last token to 0 is wrong.
217   Changes.back().TokenLength = 0;
218   Changes.back().IsTrailingComment = Changes.back().Tok->is(tok::comment);
219 
220   const WhitespaceManager::Change *LastBlockComment = nullptr;
221   for (auto &Change : Changes) {
222     // Reset the IsTrailingComment flag for changes inside of trailing comments
223     // so they don't get realigned later. Comment line breaks however still need
224     // to be aligned.
225     if (Change.IsInsideToken && Change.NewlinesBefore == 0)
226       Change.IsTrailingComment = false;
227     Change.StartOfBlockComment = nullptr;
228     Change.IndentationOffset = 0;
229     if (Change.Tok->is(tok::comment)) {
230       if (Change.Tok->is(TT_LineComment) || !Change.IsInsideToken)
231         LastBlockComment = &Change;
232       else {
233         if ((Change.StartOfBlockComment = LastBlockComment))
234           Change.IndentationOffset =
235               Change.StartOfTokenColumn -
236               Change.StartOfBlockComment->StartOfTokenColumn;
237       }
238     } else {
239       LastBlockComment = nullptr;
240     }
241   }
242 
243   // Compute conditional nesting level
244   // Level is increased for each conditional, unless this conditional continues
245   // a chain of conditional, i.e. starts immediately after the colon of another
246   // conditional.
247   SmallVector<bool, 16> ScopeStack;
248   int ConditionalsLevel = 0;
249   for (auto &Change : Changes) {
250     for (unsigned i = 0, e = Change.Tok->FakeLParens.size(); i != e; ++i) {
251       bool isNestedConditional =
252           Change.Tok->FakeLParens[e - 1 - i] == prec::Conditional &&
253           !(i == 0 && Change.Tok->Previous &&
254             Change.Tok->Previous->is(TT_ConditionalExpr) &&
255             Change.Tok->Previous->is(tok::colon));
256       if (isNestedConditional)
257         ++ConditionalsLevel;
258       ScopeStack.push_back(isNestedConditional);
259     }
260 
261     Change.ConditionalsLevel = ConditionalsLevel;
262 
263     for (unsigned i = Change.Tok->FakeRParens; i > 0 && ScopeStack.size(); --i)
264       if (ScopeStack.pop_back_val())
265         --ConditionalsLevel;
266   }
267 }
268 
269 // Align a single sequence of tokens, see AlignTokens below.
270 // Column - The token for which Matches returns true is moved to this column.
271 // RightJustify - Whether it is the token's right end or left end that gets
272 // moved to that column.
273 template <typename F>
274 static void
275 AlignTokenSequence(const FormatStyle &Style, unsigned Start, unsigned End,
276                    unsigned Column, bool RightJustify, F &&Matches,
277                    SmallVector<WhitespaceManager::Change, 16> &Changes) {
278   bool FoundMatchOnLine = false;
279   int Shift = 0;
280 
281   // ScopeStack keeps track of the current scope depth. It contains indices of
282   // the first token on each scope.
283   // We only run the "Matches" function on tokens from the outer-most scope.
284   // However, we do need to pay special attention to one class of tokens
285   // that are not in the outer-most scope, and that is function parameters
286   // which are split across multiple lines, as illustrated by this example:
287   //   double a(int x);
288   //   int    b(int  y,
289   //          double z);
290   // In the above example, we need to take special care to ensure that
291   // 'double z' is indented along with it's owning function 'b'.
292   // The same holds for calling a function:
293   //   double a = foo(x);
294   //   int    b = bar(foo(y),
295   //            foor(z));
296   // Similar for broken string literals:
297   //   double x = 3.14;
298   //   auto s   = "Hello"
299   //          "World";
300   // Special handling is required for 'nested' ternary operators.
301   SmallVector<unsigned, 16> ScopeStack;
302 
303   for (unsigned i = Start; i != End; ++i) {
304     if (ScopeStack.size() != 0 &&
305         Changes[i].indentAndNestingLevel() <
306             Changes[ScopeStack.back()].indentAndNestingLevel())
307       ScopeStack.pop_back();
308 
309     // Compare current token to previous non-comment token to ensure whether
310     // it is in a deeper scope or not.
311     unsigned PreviousNonComment = i - 1;
312     while (PreviousNonComment > Start &&
313            Changes[PreviousNonComment].Tok->is(tok::comment))
314       --PreviousNonComment;
315     if (i != Start && Changes[i].indentAndNestingLevel() >
316                           Changes[PreviousNonComment].indentAndNestingLevel())
317       ScopeStack.push_back(i);
318 
319     bool InsideNestedScope = ScopeStack.size() != 0;
320     bool ContinuedStringLiteral = i > Start &&
321                                   Changes[i].Tok->is(tok::string_literal) &&
322                                   Changes[i - 1].Tok->is(tok::string_literal);
323     bool SkipMatchCheck = InsideNestedScope || ContinuedStringLiteral;
324 
325     if (Changes[i].NewlinesBefore > 0 && !SkipMatchCheck) {
326       Shift = 0;
327       FoundMatchOnLine = false;
328     }
329 
330     // If this is the first matching token to be aligned, remember by how many
331     // spaces it has to be shifted, so the rest of the changes on the line are
332     // shifted by the same amount
333     if (!FoundMatchOnLine && !SkipMatchCheck && Matches(Changes[i])) {
334       FoundMatchOnLine = true;
335       Shift = Column - (RightJustify ? Changes[i].TokenLength : 0) -
336               Changes[i].StartOfTokenColumn;
337       Changes[i].Spaces += Shift;
338       // FIXME: This is a workaround that should be removed when we fix
339       // http://llvm.org/PR53699. An assertion later below verifies this.
340       if (Changes[i].NewlinesBefore == 0)
341         Changes[i].Spaces =
342             std::max(Changes[i].Spaces,
343                      static_cast<int>(Changes[i].Tok->SpacesRequiredBefore));
344     }
345 
346     // This is for function parameters that are split across multiple lines,
347     // as mentioned in the ScopeStack comment.
348     if (InsideNestedScope && Changes[i].NewlinesBefore > 0) {
349       unsigned ScopeStart = ScopeStack.back();
350       auto ShouldShiftBeAdded = [&] {
351         // Function declaration
352         if (Changes[ScopeStart - 1].Tok->is(TT_FunctionDeclarationName))
353           return true;
354 
355         // Lambda.
356         if (Changes[ScopeStart - 1].Tok->is(TT_LambdaLBrace))
357           return false;
358 
359         // Continued function declaration
360         if (ScopeStart > Start + 1 &&
361             Changes[ScopeStart - 2].Tok->is(TT_FunctionDeclarationName))
362           return true;
363 
364         // Continued function call
365         if (ScopeStart > Start + 1 &&
366             Changes[ScopeStart - 2].Tok->is(tok::identifier) &&
367             Changes[ScopeStart - 1].Tok->is(tok::l_paren) &&
368             Changes[ScopeStart].Tok->isNot(TT_LambdaLSquare)) {
369           if (Changes[i].Tok->MatchingParen &&
370               Changes[i].Tok->MatchingParen->is(TT_LambdaLBrace))
371             return false;
372           if (Changes[ScopeStart].NewlinesBefore > 0)
373             return false;
374           return Style.BinPackArguments;
375         }
376 
377         // Ternary operator
378         if (Changes[i].Tok->is(TT_ConditionalExpr))
379           return true;
380 
381         // Period Initializer .XXX = 1.
382         if (Changes[i].Tok->is(TT_DesignatedInitializerPeriod))
383           return true;
384 
385         // Continued ternary operator
386         if (Changes[i].Tok->Previous &&
387             Changes[i].Tok->Previous->is(TT_ConditionalExpr))
388           return true;
389 
390         // Continued braced list.
391         if (ScopeStart > Start + 1 &&
392             Changes[ScopeStart - 2].Tok->isNot(tok::identifier) &&
393             Changes[ScopeStart - 1].Tok->is(tok::l_brace) &&
394             Changes[i].Tok->isNot(tok::r_brace)) {
395           for (unsigned OuterScopeStart : llvm::reverse(ScopeStack)) {
396             // Lambda.
397             if (OuterScopeStart > Start &&
398                 Changes[OuterScopeStart - 1].Tok->is(TT_LambdaLBrace))
399               return false;
400           }
401           return true;
402         }
403 
404         return false;
405       };
406 
407       if (ShouldShiftBeAdded())
408         Changes[i].Spaces += Shift;
409     }
410 
411     if (ContinuedStringLiteral)
412       Changes[i].Spaces += Shift;
413 
414     // We should not remove required spaces unless we break the line before.
415     assert(Shift >= 0 || Changes[i].NewlinesBefore > 0 ||
416            Changes[i].Spaces >=
417                static_cast<int>(Changes[i].Tok->SpacesRequiredBefore) ||
418            Changes[i].Tok->is(tok::eof));
419 
420     Changes[i].StartOfTokenColumn += Shift;
421     if (i + 1 != Changes.size())
422       Changes[i + 1].PreviousEndOfTokenColumn += Shift;
423 
424     // If PointerAlignment is PAS_Right, keep *s or &s next to the token
425     if (Style.PointerAlignment == FormatStyle::PAS_Right &&
426         Changes[i].Spaces != 0) {
427       for (int Previous = i - 1;
428            Previous >= 0 &&
429            Changes[Previous].Tok->getType() == TT_PointerOrReference;
430            --Previous) {
431         Changes[Previous + 1].Spaces -= Shift;
432         Changes[Previous].Spaces += Shift;
433       }
434     }
435   }
436 }
437 
438 // Walk through a subset of the changes, starting at StartAt, and find
439 // sequences of matching tokens to align. To do so, keep track of the lines and
440 // whether or not a matching token was found on a line. If a matching token is
441 // found, extend the current sequence. If the current line cannot be part of a
442 // sequence, e.g. because there is an empty line before it or it contains only
443 // non-matching tokens, finalize the previous sequence.
444 // The value returned is the token on which we stopped, either because we
445 // exhausted all items inside Changes, or because we hit a scope level higher
446 // than our initial scope.
447 // This function is recursive. Each invocation processes only the scope level
448 // equal to the initial level, which is the level of Changes[StartAt].
449 // If we encounter a scope level greater than the initial level, then we call
450 // ourselves recursively, thereby avoiding the pollution of the current state
451 // with the alignment requirements of the nested sub-level. This recursive
452 // behavior is necessary for aligning function prototypes that have one or more
453 // arguments.
454 // If this function encounters a scope level less than the initial level,
455 // it returns the current position.
456 // There is a non-obvious subtlety in the recursive behavior: Even though we
457 // defer processing of nested levels to recursive invocations of this
458 // function, when it comes time to align a sequence of tokens, we run the
459 // alignment on the entire sequence, including the nested levels.
460 // When doing so, most of the nested tokens are skipped, because their
461 // alignment was already handled by the recursive invocations of this function.
462 // However, the special exception is that we do NOT skip function parameters
463 // that are split across multiple lines. See the test case in FormatTest.cpp
464 // that mentions "split function parameter alignment" for an example of this.
465 // When the parameter RightJustify is true, the operator will be
466 // right-justified. It is used to align compound assignments like `+=` and `=`.
467 // When RightJustify and ACS.PadOperators are true, operators in each block to
468 // be aligned will be padded on the left to the same length before aligning.
469 template <typename F>
470 static unsigned AlignTokens(const FormatStyle &Style, F &&Matches,
471                             SmallVector<WhitespaceManager::Change, 16> &Changes,
472                             unsigned StartAt,
473                             const FormatStyle::AlignConsecutiveStyle &ACS = {},
474                             bool RightJustify = false) {
475   // We arrange each line in 3 parts. The operator to be aligned (the anchor),
476   // and text to its left and right. In the aligned text the width of each part
477   // will be the maximum of that over the block that has been aligned. Maximum
478   // widths of each part so far. When RightJustify is true and ACS.PadOperators
479   // is false, the part from start of line to the right end of the anchor.
480   // Otherwise, only the part to the left of the anchor. Including the space
481   // that exists on its left from the start. Not including the padding added on
482   // the left to right-justify the anchor.
483   unsigned WidthLeft = 0;
484   // The operator to be aligned when RightJustify is true and ACS.PadOperators
485   // is false. 0 otherwise.
486   unsigned WidthAnchor = 0;
487   // Width to the right of the anchor. Plus width of the anchor when
488   // RightJustify is false.
489   unsigned WidthRight = 0;
490 
491   // Line number of the start and the end of the current token sequence.
492   unsigned StartOfSequence = 0;
493   unsigned EndOfSequence = 0;
494 
495   // Measure the scope level (i.e. depth of (), [], {}) of the first token, and
496   // abort when we hit any token in a higher scope than the starting one.
497   auto IndentAndNestingLevel = StartAt < Changes.size()
498                                    ? Changes[StartAt].indentAndNestingLevel()
499                                    : std::tuple<unsigned, unsigned, unsigned>();
500 
501   // Keep track of the number of commas before the matching tokens, we will only
502   // align a sequence of matching tokens if they are preceded by the same number
503   // of commas.
504   unsigned CommasBeforeLastMatch = 0;
505   unsigned CommasBeforeMatch = 0;
506 
507   // Whether a matching token has been found on the current line.
508   bool FoundMatchOnLine = false;
509 
510   // Whether the current line consists purely of comments.
511   bool LineIsComment = true;
512 
513   // Aligns a sequence of matching tokens, on the MinColumn column.
514   //
515   // Sequences start from the first matching token to align, and end at the
516   // first token of the first line that doesn't need to be aligned.
517   //
518   // We need to adjust the StartOfTokenColumn of each Change that is on a line
519   // containing any matching token to be aligned and located after such token.
520   auto AlignCurrentSequence = [&] {
521     if (StartOfSequence > 0 && StartOfSequence < EndOfSequence)
522       AlignTokenSequence(Style, StartOfSequence, EndOfSequence,
523                          WidthLeft + WidthAnchor, RightJustify, Matches,
524                          Changes);
525     WidthLeft = 0;
526     WidthAnchor = 0;
527     WidthRight = 0;
528     StartOfSequence = 0;
529     EndOfSequence = 0;
530   };
531 
532   unsigned i = StartAt;
533   for (unsigned e = Changes.size(); i != e; ++i) {
534     if (Changes[i].indentAndNestingLevel() < IndentAndNestingLevel)
535       break;
536 
537     if (Changes[i].NewlinesBefore != 0) {
538       CommasBeforeMatch = 0;
539       EndOfSequence = i;
540 
541       // Whether to break the alignment sequence because of an empty line.
542       bool EmptyLineBreak =
543           (Changes[i].NewlinesBefore > 1) && !ACS.AcrossEmptyLines;
544 
545       // Whether to break the alignment sequence because of a line without a
546       // match.
547       bool NoMatchBreak =
548           !FoundMatchOnLine && !(LineIsComment && ACS.AcrossComments);
549 
550       if (EmptyLineBreak || NoMatchBreak)
551         AlignCurrentSequence();
552 
553       // A new line starts, re-initialize line status tracking bools.
554       // Keep the match state if a string literal is continued on this line.
555       if (i == 0 || !Changes[i].Tok->is(tok::string_literal) ||
556           !Changes[i - 1].Tok->is(tok::string_literal))
557         FoundMatchOnLine = false;
558       LineIsComment = true;
559     }
560 
561     if (!Changes[i].Tok->is(tok::comment))
562       LineIsComment = false;
563 
564     if (Changes[i].Tok->is(tok::comma)) {
565       ++CommasBeforeMatch;
566     } else if (Changes[i].indentAndNestingLevel() > IndentAndNestingLevel) {
567       // Call AlignTokens recursively, skipping over this scope block.
568       unsigned StoppedAt = AlignTokens(Style, Matches, Changes, i, ACS);
569       i = StoppedAt - 1;
570       continue;
571     }
572 
573     if (!Matches(Changes[i]))
574       continue;
575 
576     // If there is more than one matching token per line, or if the number of
577     // preceding commas, do not match anymore, end the sequence.
578     if (FoundMatchOnLine || CommasBeforeMatch != CommasBeforeLastMatch)
579       AlignCurrentSequence();
580 
581     CommasBeforeLastMatch = CommasBeforeMatch;
582     FoundMatchOnLine = true;
583 
584     if (StartOfSequence == 0)
585       StartOfSequence = i;
586 
587     unsigned ChangeWidthLeft = Changes[i].StartOfTokenColumn;
588     unsigned ChangeWidthAnchor = 0;
589     unsigned ChangeWidthRight = 0;
590     if (RightJustify) {
591       if (ACS.PadOperators)
592         ChangeWidthAnchor = Changes[i].TokenLength;
593       else
594         ChangeWidthLeft += Changes[i].TokenLength;
595     } else
596       ChangeWidthRight = Changes[i].TokenLength;
597     for (unsigned j = i + 1; j != e && Changes[j].NewlinesBefore == 0; ++j) {
598       ChangeWidthRight += Changes[j].Spaces;
599       // Changes are generally 1:1 with the tokens, but a change could also be
600       // inside of a token, in which case it's counted more than once: once for
601       // the whitespace surrounding the token (!IsInsideToken) and once for
602       // each whitespace change within it (IsInsideToken).
603       // Therefore, changes inside of a token should only count the space.
604       if (!Changes[j].IsInsideToken)
605         ChangeWidthRight += Changes[j].TokenLength;
606     }
607 
608     // If we are restricted by the maximum column width, end the sequence.
609     unsigned NewLeft = std::max(ChangeWidthLeft, WidthLeft);
610     unsigned NewAnchor = std::max(ChangeWidthAnchor, WidthAnchor);
611     unsigned NewRight = std::max(ChangeWidthRight, WidthRight);
612     // `ColumnLimit == 0` means there is no column limit.
613     if (Style.ColumnLimit != 0 &&
614         Style.ColumnLimit < NewLeft + NewAnchor + NewRight) {
615       AlignCurrentSequence();
616       StartOfSequence = i;
617       WidthLeft = ChangeWidthLeft;
618       WidthAnchor = ChangeWidthAnchor;
619       WidthRight = ChangeWidthRight;
620     } else {
621       WidthLeft = NewLeft;
622       WidthAnchor = NewAnchor;
623       WidthRight = NewRight;
624     }
625   }
626 
627   EndOfSequence = i;
628   AlignCurrentSequence();
629   return i;
630 }
631 
632 // Aligns a sequence of matching tokens, on the MinColumn column.
633 //
634 // Sequences start from the first matching token to align, and end at the
635 // first token of the first line that doesn't need to be aligned.
636 //
637 // We need to adjust the StartOfTokenColumn of each Change that is on a line
638 // containing any matching token to be aligned and located after such token.
639 static void AlignMacroSequence(
640     unsigned &StartOfSequence, unsigned &EndOfSequence, unsigned &MinColumn,
641     unsigned &MaxColumn, bool &FoundMatchOnLine,
642     std::function<bool(const WhitespaceManager::Change &C)> AlignMacrosMatches,
643     SmallVector<WhitespaceManager::Change, 16> &Changes) {
644   if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) {
645 
646     FoundMatchOnLine = false;
647     int Shift = 0;
648 
649     for (unsigned I = StartOfSequence; I != EndOfSequence; ++I) {
650       if (Changes[I].NewlinesBefore > 0) {
651         Shift = 0;
652         FoundMatchOnLine = false;
653       }
654 
655       // If this is the first matching token to be aligned, remember by how many
656       // spaces it has to be shifted, so the rest of the changes on the line are
657       // shifted by the same amount
658       if (!FoundMatchOnLine && AlignMacrosMatches(Changes[I])) {
659         FoundMatchOnLine = true;
660         Shift = MinColumn - Changes[I].StartOfTokenColumn;
661         Changes[I].Spaces += Shift;
662       }
663 
664       assert(Shift >= 0);
665       Changes[I].StartOfTokenColumn += Shift;
666       if (I + 1 != Changes.size())
667         Changes[I + 1].PreviousEndOfTokenColumn += Shift;
668     }
669   }
670 
671   MinColumn = 0;
672   MaxColumn = UINT_MAX;
673   StartOfSequence = 0;
674   EndOfSequence = 0;
675 }
676 
677 void WhitespaceManager::alignConsecutiveMacros() {
678   if (!Style.AlignConsecutiveMacros.Enabled)
679     return;
680 
681   auto AlignMacrosMatches = [](const Change &C) {
682     const FormatToken *Current = C.Tok;
683     unsigned SpacesRequiredBefore = 1;
684 
685     if (Current->SpacesRequiredBefore == 0 || !Current->Previous)
686       return false;
687 
688     Current = Current->Previous;
689 
690     // If token is a ")", skip over the parameter list, to the
691     // token that precedes the "("
692     if (Current->is(tok::r_paren) && Current->MatchingParen) {
693       Current = Current->MatchingParen->Previous;
694       SpacesRequiredBefore = 0;
695     }
696 
697     if (!Current || !Current->is(tok::identifier))
698       return false;
699 
700     if (!Current->Previous || !Current->Previous->is(tok::pp_define))
701       return false;
702 
703     // For a macro function, 0 spaces are required between the
704     // identifier and the lparen that opens the parameter list.
705     // For a simple macro, 1 space is required between the
706     // identifier and the first token of the defined value.
707     return Current->Next->SpacesRequiredBefore == SpacesRequiredBefore;
708   };
709 
710   unsigned MinColumn = 0;
711   unsigned MaxColumn = UINT_MAX;
712 
713   // Start and end of the token sequence we're processing.
714   unsigned StartOfSequence = 0;
715   unsigned EndOfSequence = 0;
716 
717   // Whether a matching token has been found on the current line.
718   bool FoundMatchOnLine = false;
719 
720   // Whether the current line consists only of comments
721   bool LineIsComment = true;
722 
723   unsigned I = 0;
724   for (unsigned E = Changes.size(); I != E; ++I) {
725     if (Changes[I].NewlinesBefore != 0) {
726       EndOfSequence = I;
727 
728       // Whether to break the alignment sequence because of an empty line.
729       bool EmptyLineBreak = (Changes[I].NewlinesBefore > 1) &&
730                             !Style.AlignConsecutiveMacros.AcrossEmptyLines;
731 
732       // Whether to break the alignment sequence because of a line without a
733       // match.
734       bool NoMatchBreak =
735           !FoundMatchOnLine &&
736           !(LineIsComment && Style.AlignConsecutiveMacros.AcrossComments);
737 
738       if (EmptyLineBreak || NoMatchBreak)
739         AlignMacroSequence(StartOfSequence, EndOfSequence, MinColumn, MaxColumn,
740                            FoundMatchOnLine, AlignMacrosMatches, Changes);
741 
742       // A new line starts, re-initialize line status tracking bools.
743       FoundMatchOnLine = false;
744       LineIsComment = true;
745     }
746 
747     if (!Changes[I].Tok->is(tok::comment))
748       LineIsComment = false;
749 
750     if (!AlignMacrosMatches(Changes[I]))
751       continue;
752 
753     FoundMatchOnLine = true;
754 
755     if (StartOfSequence == 0)
756       StartOfSequence = I;
757 
758     unsigned ChangeMinColumn = Changes[I].StartOfTokenColumn;
759     int LineLengthAfter = -Changes[I].Spaces;
760     for (unsigned j = I; j != E && Changes[j].NewlinesBefore == 0; ++j)
761       LineLengthAfter += Changes[j].Spaces + Changes[j].TokenLength;
762     unsigned ChangeMaxColumn = Style.ColumnLimit - LineLengthAfter;
763 
764     MinColumn = std::max(MinColumn, ChangeMinColumn);
765     MaxColumn = std::min(MaxColumn, ChangeMaxColumn);
766   }
767 
768   EndOfSequence = I;
769   AlignMacroSequence(StartOfSequence, EndOfSequence, MinColumn, MaxColumn,
770                      FoundMatchOnLine, AlignMacrosMatches, Changes);
771 }
772 
773 void WhitespaceManager::alignConsecutiveAssignments() {
774   if (!Style.AlignConsecutiveAssignments.Enabled)
775     return;
776 
777   AlignTokens(
778       Style,
779       [&](const Change &C) {
780         // Do not align on equal signs that are first on a line.
781         if (C.NewlinesBefore > 0)
782           return false;
783 
784         // Do not align on equal signs that are last on a line.
785         if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
786           return false;
787 
788         // Do not align operator= overloads.
789         FormatToken *Previous = C.Tok->getPreviousNonComment();
790         if (Previous && Previous->is(tok::kw_operator))
791           return false;
792 
793         return Style.AlignConsecutiveAssignments.AlignCompound
794                    ? C.Tok->getPrecedence() == prec::Assignment
795                    : C.Tok->is(tok::equal);
796       },
797       Changes, /*StartAt=*/0, Style.AlignConsecutiveAssignments,
798       /*RightJustify=*/true);
799 }
800 
801 void WhitespaceManager::alignConsecutiveBitFields() {
802   if (!Style.AlignConsecutiveBitFields.Enabled)
803     return;
804 
805   AlignTokens(
806       Style,
807       [&](Change const &C) {
808         // Do not align on ':' that is first on a line.
809         if (C.NewlinesBefore > 0)
810           return false;
811 
812         // Do not align on ':' that is last on a line.
813         if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
814           return false;
815 
816         return C.Tok->is(TT_BitFieldColon);
817       },
818       Changes, /*StartAt=*/0, Style.AlignConsecutiveBitFields);
819 }
820 
821 void WhitespaceManager::alignConsecutiveDeclarations() {
822   if (!Style.AlignConsecutiveDeclarations.Enabled)
823     return;
824 
825   AlignTokens(
826       Style,
827       [](Change const &C) {
828         // tok::kw_operator is necessary for aligning operator overload
829         // definitions.
830         if (C.Tok->isOneOf(TT_FunctionDeclarationName, tok::kw_operator))
831           return true;
832         if (C.Tok->isNot(TT_StartOfName))
833           return false;
834         if (C.Tok->Previous &&
835             C.Tok->Previous->is(TT_StatementAttributeLikeMacro))
836           return false;
837         // Check if there is a subsequent name that starts the same declaration.
838         for (FormatToken *Next = C.Tok->Next; Next; Next = Next->Next) {
839           if (Next->is(tok::comment))
840             continue;
841           if (Next->is(TT_PointerOrReference))
842             return false;
843           if (!Next->Tok.getIdentifierInfo())
844             break;
845           if (Next->isOneOf(TT_StartOfName, TT_FunctionDeclarationName,
846                             tok::kw_operator))
847             return false;
848         }
849         return true;
850       },
851       Changes, /*StartAt=*/0, Style.AlignConsecutiveDeclarations);
852 }
853 
854 void WhitespaceManager::alignChainedConditionals() {
855   if (Style.BreakBeforeTernaryOperators) {
856     AlignTokens(
857         Style,
858         [](Change const &C) {
859           // Align question operators and last colon
860           return C.Tok->is(TT_ConditionalExpr) &&
861                  ((C.Tok->is(tok::question) && !C.NewlinesBefore) ||
862                   (C.Tok->is(tok::colon) && C.Tok->Next &&
863                    (C.Tok->Next->FakeLParens.size() == 0 ||
864                     C.Tok->Next->FakeLParens.back() != prec::Conditional)));
865         },
866         Changes, /*StartAt=*/0);
867   } else {
868     static auto AlignWrappedOperand = [](Change const &C) {
869       FormatToken *Previous = C.Tok->getPreviousNonComment();
870       return C.NewlinesBefore && Previous && Previous->is(TT_ConditionalExpr) &&
871              (Previous->is(tok::colon) &&
872               (C.Tok->FakeLParens.size() == 0 ||
873                C.Tok->FakeLParens.back() != prec::Conditional));
874     };
875     // Ensure we keep alignment of wrapped operands with non-wrapped operands
876     // Since we actually align the operators, the wrapped operands need the
877     // extra offset to be properly aligned.
878     for (Change &C : Changes)
879       if (AlignWrappedOperand(C))
880         C.StartOfTokenColumn -= 2;
881     AlignTokens(
882         Style,
883         [this](Change const &C) {
884           // Align question operators if next operand is not wrapped, as
885           // well as wrapped operands after question operator or last
886           // colon in conditional sequence
887           return (C.Tok->is(TT_ConditionalExpr) && C.Tok->is(tok::question) &&
888                   &C != &Changes.back() && (&C + 1)->NewlinesBefore == 0 &&
889                   !(&C + 1)->IsTrailingComment) ||
890                  AlignWrappedOperand(C);
891         },
892         Changes, /*StartAt=*/0);
893   }
894 }
895 
896 void WhitespaceManager::alignTrailingComments() {
897   unsigned MinColumn = 0;
898   unsigned MaxColumn = UINT_MAX;
899   unsigned StartOfSequence = 0;
900   bool BreakBeforeNext = false;
901   unsigned Newlines = 0;
902   for (unsigned i = 0, e = Changes.size(); i != e; ++i) {
903     if (Changes[i].StartOfBlockComment)
904       continue;
905     Newlines += Changes[i].NewlinesBefore;
906     if (!Changes[i].IsTrailingComment)
907       continue;
908 
909     unsigned ChangeMinColumn = Changes[i].StartOfTokenColumn;
910     unsigned ChangeMaxColumn;
911 
912     if (Style.ColumnLimit == 0)
913       ChangeMaxColumn = UINT_MAX;
914     else if (Style.ColumnLimit >= Changes[i].TokenLength)
915       ChangeMaxColumn = Style.ColumnLimit - Changes[i].TokenLength;
916     else
917       ChangeMaxColumn = ChangeMinColumn;
918 
919     // If we don't create a replacement for this change, we have to consider
920     // it to be immovable.
921     if (!Changes[i].CreateReplacement)
922       ChangeMaxColumn = ChangeMinColumn;
923 
924     if (i + 1 != e && Changes[i + 1].ContinuesPPDirective)
925       ChangeMaxColumn -= 2;
926     // If this comment follows an } in column 0, it probably documents the
927     // closing of a namespace and we don't want to align it.
928     bool FollowsRBraceInColumn0 = i > 0 && Changes[i].NewlinesBefore == 0 &&
929                                   Changes[i - 1].Tok->is(tok::r_brace) &&
930                                   Changes[i - 1].StartOfTokenColumn == 0;
931     bool WasAlignedWithStartOfNextLine = false;
932     if (Changes[i].NewlinesBefore == 1) { // A comment on its own line.
933       unsigned CommentColumn = SourceMgr.getSpellingColumnNumber(
934           Changes[i].OriginalWhitespaceRange.getEnd());
935       for (unsigned j = i + 1; j != e; ++j) {
936         if (Changes[j].Tok->is(tok::comment))
937           continue;
938 
939         unsigned NextColumn = SourceMgr.getSpellingColumnNumber(
940             Changes[j].OriginalWhitespaceRange.getEnd());
941         // The start of the next token was previously aligned with the
942         // start of this comment.
943         WasAlignedWithStartOfNextLine =
944             CommentColumn == NextColumn ||
945             CommentColumn == NextColumn + Style.IndentWidth;
946         break;
947       }
948     }
949     if (!Style.AlignTrailingComments || FollowsRBraceInColumn0) {
950       alignTrailingComments(StartOfSequence, i, MinColumn);
951       MinColumn = ChangeMinColumn;
952       MaxColumn = ChangeMinColumn;
953       StartOfSequence = i;
954     } else if (BreakBeforeNext || Newlines > 1 ||
955                (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn) ||
956                // Break the comment sequence if the previous line did not end
957                // in a trailing comment.
958                (Changes[i].NewlinesBefore == 1 && i > 0 &&
959                 !Changes[i - 1].IsTrailingComment) ||
960                WasAlignedWithStartOfNextLine) {
961       alignTrailingComments(StartOfSequence, i, MinColumn);
962       MinColumn = ChangeMinColumn;
963       MaxColumn = ChangeMaxColumn;
964       StartOfSequence = i;
965     } else {
966       MinColumn = std::max(MinColumn, ChangeMinColumn);
967       MaxColumn = std::min(MaxColumn, ChangeMaxColumn);
968     }
969     BreakBeforeNext = (i == 0) || (Changes[i].NewlinesBefore > 1) ||
970                       // Never start a sequence with a comment at the beginning
971                       // of the line.
972                       (Changes[i].NewlinesBefore == 1 && StartOfSequence == i);
973     Newlines = 0;
974   }
975   alignTrailingComments(StartOfSequence, Changes.size(), MinColumn);
976 }
977 
978 void WhitespaceManager::alignTrailingComments(unsigned Start, unsigned End,
979                                               unsigned Column) {
980   for (unsigned i = Start; i != End; ++i) {
981     int Shift = 0;
982     if (Changes[i].IsTrailingComment)
983       Shift = Column - Changes[i].StartOfTokenColumn;
984     if (Changes[i].StartOfBlockComment) {
985       Shift = Changes[i].IndentationOffset +
986               Changes[i].StartOfBlockComment->StartOfTokenColumn -
987               Changes[i].StartOfTokenColumn;
988     }
989     if (Shift < 0)
990       continue;
991     Changes[i].Spaces += Shift;
992     if (i + 1 != Changes.size())
993       Changes[i + 1].PreviousEndOfTokenColumn += Shift;
994     Changes[i].StartOfTokenColumn += Shift;
995   }
996 }
997 
998 void WhitespaceManager::alignEscapedNewlines() {
999   if (Style.AlignEscapedNewlines == FormatStyle::ENAS_DontAlign)
1000     return;
1001 
1002   bool AlignLeft = Style.AlignEscapedNewlines == FormatStyle::ENAS_Left;
1003   unsigned MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit;
1004   unsigned StartOfMacro = 0;
1005   for (unsigned i = 1, e = Changes.size(); i < e; ++i) {
1006     Change &C = Changes[i];
1007     if (C.NewlinesBefore > 0) {
1008       if (C.ContinuesPPDirective) {
1009         MaxEndOfLine = std::max(C.PreviousEndOfTokenColumn + 2, MaxEndOfLine);
1010       } else {
1011         alignEscapedNewlines(StartOfMacro + 1, i, MaxEndOfLine);
1012         MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit;
1013         StartOfMacro = i;
1014       }
1015     }
1016   }
1017   alignEscapedNewlines(StartOfMacro + 1, Changes.size(), MaxEndOfLine);
1018 }
1019 
1020 void WhitespaceManager::alignEscapedNewlines(unsigned Start, unsigned End,
1021                                              unsigned Column) {
1022   for (unsigned i = Start; i < End; ++i) {
1023     Change &C = Changes[i];
1024     if (C.NewlinesBefore > 0) {
1025       assert(C.ContinuesPPDirective);
1026       if (C.PreviousEndOfTokenColumn + 1 > Column)
1027         C.EscapedNewlineColumn = 0;
1028       else
1029         C.EscapedNewlineColumn = Column;
1030     }
1031   }
1032 }
1033 
1034 void WhitespaceManager::alignArrayInitializers() {
1035   if (Style.AlignArrayOfStructures == FormatStyle::AIAS_None)
1036     return;
1037 
1038   for (unsigned ChangeIndex = 1U, ChangeEnd = Changes.size();
1039        ChangeIndex < ChangeEnd; ++ChangeIndex) {
1040     auto &C = Changes[ChangeIndex];
1041     if (C.Tok->IsArrayInitializer) {
1042       bool FoundComplete = false;
1043       for (unsigned InsideIndex = ChangeIndex + 1; InsideIndex < ChangeEnd;
1044            ++InsideIndex) {
1045         if (Changes[InsideIndex].Tok == C.Tok->MatchingParen) {
1046           alignArrayInitializers(ChangeIndex, InsideIndex + 1);
1047           ChangeIndex = InsideIndex + 1;
1048           FoundComplete = true;
1049           break;
1050         }
1051       }
1052       if (!FoundComplete)
1053         ChangeIndex = ChangeEnd;
1054     }
1055   }
1056 }
1057 
1058 void WhitespaceManager::alignArrayInitializers(unsigned Start, unsigned End) {
1059 
1060   if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Right)
1061     alignArrayInitializersRightJustified(getCells(Start, End));
1062   else if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Left)
1063     alignArrayInitializersLeftJustified(getCells(Start, End));
1064 }
1065 
1066 void WhitespaceManager::alignArrayInitializersRightJustified(
1067     CellDescriptions &&CellDescs) {
1068   if (!CellDescs.isRectangular())
1069     return;
1070 
1071   auto &Cells = CellDescs.Cells;
1072   // Now go through and fixup the spaces.
1073   auto *CellIter = Cells.begin();
1074   for (auto i = 0U; i < CellDescs.CellCounts[0]; ++i, ++CellIter) {
1075     unsigned NetWidth = 0U;
1076     if (isSplitCell(*CellIter))
1077       NetWidth = getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces);
1078     auto CellWidth = getMaximumCellWidth(CellIter, NetWidth);
1079 
1080     if (Changes[CellIter->Index].Tok->is(tok::r_brace)) {
1081       // So in here we want to see if there is a brace that falls
1082       // on a line that was split. If so on that line we make sure that
1083       // the spaces in front of the brace are enough.
1084       const auto *Next = CellIter;
1085       do {
1086         const FormatToken *Previous = Changes[Next->Index].Tok->Previous;
1087         if (Previous && Previous->isNot(TT_LineComment)) {
1088           Changes[Next->Index].Spaces = 0;
1089           Changes[Next->Index].NewlinesBefore = 0;
1090         }
1091         Next = Next->NextColumnElement;
1092       } while (Next);
1093       // Unless the array is empty, we need the position of all the
1094       // immediately adjacent cells
1095       if (CellIter != Cells.begin()) {
1096         auto ThisNetWidth =
1097             getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces);
1098         auto MaxNetWidth = getMaximumNetWidth(
1099             Cells.begin(), CellIter, CellDescs.InitialSpaces,
1100             CellDescs.CellCounts[0], CellDescs.CellCounts.size());
1101         if (ThisNetWidth < MaxNetWidth)
1102           Changes[CellIter->Index].Spaces = (MaxNetWidth - ThisNetWidth);
1103         auto RowCount = 1U;
1104         auto Offset = std::distance(Cells.begin(), CellIter);
1105         for (const auto *Next = CellIter->NextColumnElement; Next != nullptr;
1106              Next = Next->NextColumnElement) {
1107           auto *Start = (Cells.begin() + RowCount * CellDescs.CellCounts[0]);
1108           auto *End = Start + Offset;
1109           ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces);
1110           if (ThisNetWidth < MaxNetWidth)
1111             Changes[Next->Index].Spaces = (MaxNetWidth - ThisNetWidth);
1112           ++RowCount;
1113         }
1114       }
1115     } else {
1116       auto ThisWidth =
1117           calculateCellWidth(CellIter->Index, CellIter->EndIndex, true) +
1118           NetWidth;
1119       if (Changes[CellIter->Index].NewlinesBefore == 0) {
1120         Changes[CellIter->Index].Spaces = (CellWidth - (ThisWidth + NetWidth));
1121         Changes[CellIter->Index].Spaces += (i > 0) ? 1 : 0;
1122       }
1123       alignToStartOfCell(CellIter->Index, CellIter->EndIndex);
1124       for (const auto *Next = CellIter->NextColumnElement; Next != nullptr;
1125            Next = Next->NextColumnElement) {
1126         ThisWidth =
1127             calculateCellWidth(Next->Index, Next->EndIndex, true) + NetWidth;
1128         if (Changes[Next->Index].NewlinesBefore == 0) {
1129           Changes[Next->Index].Spaces = (CellWidth - ThisWidth);
1130           Changes[Next->Index].Spaces += (i > 0) ? 1 : 0;
1131         }
1132         alignToStartOfCell(Next->Index, Next->EndIndex);
1133       }
1134     }
1135   }
1136 }
1137 
1138 void WhitespaceManager::alignArrayInitializersLeftJustified(
1139     CellDescriptions &&CellDescs) {
1140 
1141   if (!CellDescs.isRectangular())
1142     return;
1143 
1144   auto &Cells = CellDescs.Cells;
1145   // Now go through and fixup the spaces.
1146   auto *CellIter = Cells.begin();
1147   // The first cell needs to be against the left brace.
1148   if (Changes[CellIter->Index].NewlinesBefore == 0)
1149     Changes[CellIter->Index].Spaces = 0;
1150   else
1151     Changes[CellIter->Index].Spaces = CellDescs.InitialSpaces;
1152   ++CellIter;
1153   for (auto i = 1U; i < CellDescs.CellCounts[0]; i++, ++CellIter) {
1154     auto MaxNetWidth = getMaximumNetWidth(
1155         Cells.begin(), CellIter, CellDescs.InitialSpaces,
1156         CellDescs.CellCounts[0], CellDescs.CellCounts.size());
1157     auto ThisNetWidth =
1158         getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces);
1159     if (Changes[CellIter->Index].NewlinesBefore == 0) {
1160       Changes[CellIter->Index].Spaces =
1161           MaxNetWidth - ThisNetWidth +
1162           (Changes[CellIter->Index].Tok->isNot(tok::r_brace) ? 1 : 0);
1163     }
1164     auto RowCount = 1U;
1165     auto Offset = std::distance(Cells.begin(), CellIter);
1166     for (const auto *Next = CellIter->NextColumnElement; Next != nullptr;
1167          Next = Next->NextColumnElement) {
1168       if (RowCount > CellDescs.CellCounts.size())
1169         break;
1170       auto *Start = (Cells.begin() + RowCount * CellDescs.CellCounts[0]);
1171       auto *End = Start + Offset;
1172       auto ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces);
1173       if (Changes[Next->Index].NewlinesBefore == 0) {
1174         Changes[Next->Index].Spaces =
1175             MaxNetWidth - ThisNetWidth +
1176             (Changes[Next->Index].Tok->isNot(tok::r_brace) ? 1 : 0);
1177       }
1178       ++RowCount;
1179     }
1180   }
1181 }
1182 
1183 bool WhitespaceManager::isSplitCell(const CellDescription &Cell) {
1184   if (Cell.HasSplit)
1185     return true;
1186   for (const auto *Next = Cell.NextColumnElement; Next != nullptr;
1187        Next = Next->NextColumnElement)
1188     if (Next->HasSplit)
1189       return true;
1190   return false;
1191 }
1192 
1193 WhitespaceManager::CellDescriptions WhitespaceManager::getCells(unsigned Start,
1194                                                                 unsigned End) {
1195 
1196   unsigned Depth = 0;
1197   unsigned Cell = 0;
1198   SmallVector<unsigned> CellCounts;
1199   unsigned InitialSpaces = 0;
1200   unsigned InitialTokenLength = 0;
1201   unsigned EndSpaces = 0;
1202   SmallVector<CellDescription> Cells;
1203   const FormatToken *MatchingParen = nullptr;
1204   for (unsigned i = Start; i < End; ++i) {
1205     auto &C = Changes[i];
1206     if (C.Tok->is(tok::l_brace))
1207       ++Depth;
1208     else if (C.Tok->is(tok::r_brace))
1209       --Depth;
1210     if (Depth == 2) {
1211       if (C.Tok->is(tok::l_brace)) {
1212         Cell = 0;
1213         MatchingParen = C.Tok->MatchingParen;
1214         if (InitialSpaces == 0) {
1215           InitialSpaces = C.Spaces + C.TokenLength;
1216           InitialTokenLength = C.TokenLength;
1217           auto j = i - 1;
1218           for (; Changes[j].NewlinesBefore == 0 && j > Start; --j) {
1219             InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength;
1220             InitialTokenLength += Changes[j].TokenLength;
1221           }
1222           if (C.NewlinesBefore == 0) {
1223             InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength;
1224             InitialTokenLength += Changes[j].TokenLength;
1225           }
1226         }
1227       } else if (C.Tok->is(tok::comma)) {
1228         if (!Cells.empty())
1229           Cells.back().EndIndex = i;
1230         if (C.Tok->getNextNonComment()->isNot(tok::r_brace)) // dangling comma
1231           ++Cell;
1232       }
1233     } else if (Depth == 1) {
1234       if (C.Tok == MatchingParen) {
1235         if (!Cells.empty())
1236           Cells.back().EndIndex = i;
1237         Cells.push_back(CellDescription{i, ++Cell, i + 1, false, nullptr});
1238         CellCounts.push_back(C.Tok->Previous->isNot(tok::comma) ? Cell + 1
1239                                                                 : Cell);
1240         // Go to the next non-comment and ensure there is a break in front
1241         const auto *NextNonComment = C.Tok->getNextNonComment();
1242         while (NextNonComment->is(tok::comma))
1243           NextNonComment = NextNonComment->getNextNonComment();
1244         auto j = i;
1245         while (Changes[j].Tok != NextNonComment && j < End)
1246           ++j;
1247         if (j < End && Changes[j].NewlinesBefore == 0 &&
1248             Changes[j].Tok->isNot(tok::r_brace)) {
1249           Changes[j].NewlinesBefore = 1;
1250           // Account for the added token lengths
1251           Changes[j].Spaces = InitialSpaces - InitialTokenLength;
1252         }
1253       } else if (C.Tok->is(tok::comment)) {
1254         // Trailing comments stay at a space past the last token
1255         C.Spaces = Changes[i - 1].Tok->is(tok::comma) ? 1 : 2;
1256       } else if (C.Tok->is(tok::l_brace)) {
1257         // We need to make sure that the ending braces is aligned to the
1258         // start of our initializer
1259         auto j = i - 1;
1260         for (; j > 0 && !Changes[j].Tok->ArrayInitializerLineStart; --j)
1261           ; // Nothing the loop does the work
1262         EndSpaces = Changes[j].Spaces;
1263       }
1264     } else if (Depth == 0 && C.Tok->is(tok::r_brace)) {
1265       C.NewlinesBefore = 1;
1266       C.Spaces = EndSpaces;
1267     }
1268     if (C.Tok->StartsColumn) {
1269       // This gets us past tokens that have been split over multiple
1270       // lines
1271       bool HasSplit = false;
1272       if (Changes[i].NewlinesBefore > 0) {
1273         // So if we split a line previously and the tail line + this token is
1274         // less then the column limit we remove the split here and just put
1275         // the column start at a space past the comma
1276         //
1277         // FIXME This if branch covers the cases where the column is not
1278         // the first column. This leads to weird pathologies like the formatting
1279         // auto foo = Items{
1280         //     Section{
1281         //             0, bar(),
1282         //     }
1283         // };
1284         // Well if it doesn't lead to that it's indicative that the line
1285         // breaking should be revisited. Unfortunately alot of other options
1286         // interact with this
1287         auto j = i - 1;
1288         if ((j - 1) > Start && Changes[j].Tok->is(tok::comma) &&
1289             Changes[j - 1].NewlinesBefore > 0) {
1290           --j;
1291           auto LineLimit = Changes[j].Spaces + Changes[j].TokenLength;
1292           if (LineLimit < Style.ColumnLimit) {
1293             Changes[i].NewlinesBefore = 0;
1294             Changes[i].Spaces = 1;
1295           }
1296         }
1297       }
1298       while (Changes[i].NewlinesBefore > 0 && Changes[i].Tok == C.Tok) {
1299         Changes[i].Spaces = InitialSpaces;
1300         ++i;
1301         HasSplit = true;
1302       }
1303       if (Changes[i].Tok != C.Tok)
1304         --i;
1305       Cells.push_back(CellDescription{i, Cell, i, HasSplit, nullptr});
1306     }
1307   }
1308 
1309   return linkCells({Cells, CellCounts, InitialSpaces});
1310 }
1311 
1312 unsigned WhitespaceManager::calculateCellWidth(unsigned Start, unsigned End,
1313                                                bool WithSpaces) const {
1314   unsigned CellWidth = 0;
1315   for (auto i = Start; i < End; i++) {
1316     if (Changes[i].NewlinesBefore > 0)
1317       CellWidth = 0;
1318     CellWidth += Changes[i].TokenLength;
1319     CellWidth += (WithSpaces ? Changes[i].Spaces : 0);
1320   }
1321   return CellWidth;
1322 }
1323 
1324 void WhitespaceManager::alignToStartOfCell(unsigned Start, unsigned End) {
1325   if ((End - Start) <= 1)
1326     return;
1327   // If the line is broken anywhere in there make sure everything
1328   // is aligned to the parent
1329   for (auto i = Start + 1; i < End; i++)
1330     if (Changes[i].NewlinesBefore > 0)
1331       Changes[i].Spaces = Changes[Start].Spaces;
1332 }
1333 
1334 WhitespaceManager::CellDescriptions
1335 WhitespaceManager::linkCells(CellDescriptions &&CellDesc) {
1336   auto &Cells = CellDesc.Cells;
1337   for (auto *CellIter = Cells.begin(); CellIter != Cells.end(); ++CellIter) {
1338     if (CellIter->NextColumnElement == nullptr &&
1339         ((CellIter + 1) != Cells.end())) {
1340       for (auto *NextIter = CellIter + 1; NextIter != Cells.end(); ++NextIter) {
1341         if (NextIter->Cell == CellIter->Cell) {
1342           CellIter->NextColumnElement = &(*NextIter);
1343           break;
1344         }
1345       }
1346     }
1347   }
1348   return std::move(CellDesc);
1349 }
1350 
1351 void WhitespaceManager::generateChanges() {
1352   for (unsigned i = 0, e = Changes.size(); i != e; ++i) {
1353     const Change &C = Changes[i];
1354     if (i > 0 && Changes[i - 1].OriginalWhitespaceRange.getBegin() ==
1355                      C.OriginalWhitespaceRange.getBegin()) {
1356       // Do not generate two replacements for the same location.
1357       continue;
1358     }
1359     if (C.CreateReplacement) {
1360       std::string ReplacementText = C.PreviousLinePostfix;
1361       if (C.ContinuesPPDirective)
1362         appendEscapedNewlineText(ReplacementText, C.NewlinesBefore,
1363                                  C.PreviousEndOfTokenColumn,
1364                                  C.EscapedNewlineColumn);
1365       else
1366         appendNewlineText(ReplacementText, C.NewlinesBefore);
1367       // FIXME: This assert should hold if we computed the column correctly.
1368       // assert((int)C.StartOfTokenColumn >= C.Spaces);
1369       appendIndentText(
1370           ReplacementText, C.Tok->IndentLevel, std::max(0, C.Spaces),
1371           std::max((int)C.StartOfTokenColumn, C.Spaces) - std::max(0, C.Spaces),
1372           C.IsAligned);
1373       ReplacementText.append(C.CurrentLinePrefix);
1374       storeReplacement(C.OriginalWhitespaceRange, ReplacementText);
1375     }
1376   }
1377 }
1378 
1379 void WhitespaceManager::storeReplacement(SourceRange Range, StringRef Text) {
1380   unsigned WhitespaceLength = SourceMgr.getFileOffset(Range.getEnd()) -
1381                               SourceMgr.getFileOffset(Range.getBegin());
1382   // Don't create a replacement, if it does not change anything.
1383   if (StringRef(SourceMgr.getCharacterData(Range.getBegin()),
1384                 WhitespaceLength) == Text)
1385     return;
1386   auto Err = Replaces.add(tooling::Replacement(
1387       SourceMgr, CharSourceRange::getCharRange(Range), Text));
1388   // FIXME: better error handling. For now, just print an error message in the
1389   // release version.
1390   if (Err) {
1391     llvm::errs() << llvm::toString(std::move(Err)) << "\n";
1392     assert(false);
1393   }
1394 }
1395 
1396 void WhitespaceManager::appendNewlineText(std::string &Text,
1397                                           unsigned Newlines) {
1398   if (UseCRLF) {
1399     Text.reserve(Text.size() + 2 * Newlines);
1400     for (unsigned i = 0; i < Newlines; ++i)
1401       Text.append("\r\n");
1402   } else {
1403     Text.append(Newlines, '\n');
1404   }
1405 }
1406 
1407 void WhitespaceManager::appendEscapedNewlineText(
1408     std::string &Text, unsigned Newlines, unsigned PreviousEndOfTokenColumn,
1409     unsigned EscapedNewlineColumn) {
1410   if (Newlines > 0) {
1411     unsigned Spaces =
1412         std::max<int>(1, EscapedNewlineColumn - PreviousEndOfTokenColumn - 1);
1413     for (unsigned i = 0; i < Newlines; ++i) {
1414       Text.append(Spaces, ' ');
1415       Text.append(UseCRLF ? "\\\r\n" : "\\\n");
1416       Spaces = std::max<int>(0, EscapedNewlineColumn - 1);
1417     }
1418   }
1419 }
1420 
1421 void WhitespaceManager::appendIndentText(std::string &Text,
1422                                          unsigned IndentLevel, unsigned Spaces,
1423                                          unsigned WhitespaceStartColumn,
1424                                          bool IsAligned) {
1425   switch (Style.UseTab) {
1426   case FormatStyle::UT_Never:
1427     Text.append(Spaces, ' ');
1428     break;
1429   case FormatStyle::UT_Always: {
1430     if (Style.TabWidth) {
1431       unsigned FirstTabWidth =
1432           Style.TabWidth - WhitespaceStartColumn % Style.TabWidth;
1433 
1434       // Insert only spaces when we want to end up before the next tab.
1435       if (Spaces < FirstTabWidth || Spaces == 1) {
1436         Text.append(Spaces, ' ');
1437         break;
1438       }
1439       // Align to the next tab.
1440       Spaces -= FirstTabWidth;
1441       Text.append("\t");
1442 
1443       Text.append(Spaces / Style.TabWidth, '\t');
1444       Text.append(Spaces % Style.TabWidth, ' ');
1445     } else if (Spaces == 1) {
1446       Text.append(Spaces, ' ');
1447     }
1448     break;
1449   }
1450   case FormatStyle::UT_ForIndentation:
1451     if (WhitespaceStartColumn == 0) {
1452       unsigned Indentation = IndentLevel * Style.IndentWidth;
1453       Spaces = appendTabIndent(Text, Spaces, Indentation);
1454     }
1455     Text.append(Spaces, ' ');
1456     break;
1457   case FormatStyle::UT_ForContinuationAndIndentation:
1458     if (WhitespaceStartColumn == 0)
1459       Spaces = appendTabIndent(Text, Spaces, Spaces);
1460     Text.append(Spaces, ' ');
1461     break;
1462   case FormatStyle::UT_AlignWithSpaces:
1463     if (WhitespaceStartColumn == 0) {
1464       unsigned Indentation =
1465           IsAligned ? IndentLevel * Style.IndentWidth : Spaces;
1466       Spaces = appendTabIndent(Text, Spaces, Indentation);
1467     }
1468     Text.append(Spaces, ' ');
1469     break;
1470   }
1471 }
1472 
1473 unsigned WhitespaceManager::appendTabIndent(std::string &Text, unsigned Spaces,
1474                                             unsigned Indentation) {
1475   // This happens, e.g. when a line in a block comment is indented less than the
1476   // first one.
1477   if (Indentation > Spaces)
1478     Indentation = Spaces;
1479   if (Style.TabWidth) {
1480     unsigned Tabs = Indentation / Style.TabWidth;
1481     Text.append(Tabs, '\t');
1482     Spaces -= Tabs * Style.TabWidth;
1483   }
1484   return Spaces;
1485 }
1486 
1487 } // namespace format
1488 } // namespace clang
1489