xref: /freebsd/contrib/llvm-project/clang/lib/Format/WhitespaceManager.cpp (revision 6c4b055cfb6bf549e9145dde6454cc6b178c35e4)
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 
operator ()(const Change & C1,const Change & C2) const22 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          (C1.OriginalWhitespaceRange.getBegin() ==
28               C2.OriginalWhitespaceRange.getBegin() &&
29           SourceMgr.isBeforeInTranslationUnit(
30               C1.OriginalWhitespaceRange.getEnd(),
31               C2.OriginalWhitespaceRange.getEnd()));
32 }
33 
Change(const FormatToken & Tok,bool CreateReplacement,SourceRange OriginalWhitespaceRange,int Spaces,unsigned StartOfTokenColumn,unsigned NewlinesBefore,StringRef PreviousLinePostfix,StringRef CurrentLinePrefix,bool IsAligned,bool ContinuesPPDirective,bool IsInsideToken)34 WhitespaceManager::Change::Change(const FormatToken &Tok,
35                                   bool CreateReplacement,
36                                   SourceRange OriginalWhitespaceRange,
37                                   int Spaces, unsigned StartOfTokenColumn,
38                                   unsigned NewlinesBefore,
39                                   StringRef PreviousLinePostfix,
40                                   StringRef CurrentLinePrefix, bool IsAligned,
41                                   bool ContinuesPPDirective, bool IsInsideToken)
42     : Tok(&Tok), CreateReplacement(CreateReplacement),
43       OriginalWhitespaceRange(OriginalWhitespaceRange),
44       StartOfTokenColumn(StartOfTokenColumn), NewlinesBefore(NewlinesBefore),
45       PreviousLinePostfix(PreviousLinePostfix),
46       CurrentLinePrefix(CurrentLinePrefix), IsAligned(IsAligned),
47       ContinuesPPDirective(ContinuesPPDirective), Spaces(Spaces),
48       IsInsideToken(IsInsideToken), IsTrailingComment(false), TokenLength(0),
49       PreviousEndOfTokenColumn(0), EscapedNewlineColumn(0),
50       StartOfBlockComment(nullptr), IndentationOffset(0), ConditionalsLevel(0) {
51 }
52 
replaceWhitespace(FormatToken & Tok,unsigned Newlines,unsigned Spaces,unsigned StartOfTokenColumn,bool IsAligned,bool InPPDirective)53 void WhitespaceManager::replaceWhitespace(FormatToken &Tok, unsigned Newlines,
54                                           unsigned Spaces,
55                                           unsigned StartOfTokenColumn,
56                                           bool IsAligned, bool InPPDirective) {
57   if (Tok.Finalized || (Tok.MacroCtx && Tok.MacroCtx->Role == MR_ExpandedArg))
58     return;
59   Tok.setDecision((Newlines > 0) ? FD_Break : FD_Continue);
60   Changes.push_back(Change(Tok, /*CreateReplacement=*/true, Tok.WhitespaceRange,
61                            Spaces, StartOfTokenColumn, Newlines, "", "",
62                            IsAligned, InPPDirective && !Tok.IsFirst,
63                            /*IsInsideToken=*/false));
64 }
65 
addUntouchableToken(const FormatToken & Tok,bool InPPDirective)66 void WhitespaceManager::addUntouchableToken(const FormatToken &Tok,
67                                             bool InPPDirective) {
68   if (Tok.Finalized || (Tok.MacroCtx && Tok.MacroCtx->Role == MR_ExpandedArg))
69     return;
70   Changes.push_back(Change(Tok, /*CreateReplacement=*/false,
71                            Tok.WhitespaceRange, /*Spaces=*/0,
72                            Tok.OriginalColumn, Tok.NewlinesBefore, "", "",
73                            /*IsAligned=*/false, InPPDirective && !Tok.IsFirst,
74                            /*IsInsideToken=*/false));
75 }
76 
77 llvm::Error
addReplacement(const tooling::Replacement & Replacement)78 WhitespaceManager::addReplacement(const tooling::Replacement &Replacement) {
79   return Replaces.add(Replacement);
80 }
81 
inputUsesCRLF(StringRef Text,bool DefaultToCRLF)82 bool WhitespaceManager::inputUsesCRLF(StringRef Text, bool DefaultToCRLF) {
83   size_t LF = Text.count('\n');
84   size_t CR = Text.count('\r') * 2;
85   return LF == CR ? DefaultToCRLF : CR > LF;
86 }
87 
replaceWhitespaceInToken(const FormatToken & Tok,unsigned Offset,unsigned ReplaceChars,StringRef PreviousPostfix,StringRef CurrentPrefix,bool InPPDirective,unsigned Newlines,int Spaces)88 void WhitespaceManager::replaceWhitespaceInToken(
89     const FormatToken &Tok, unsigned Offset, unsigned ReplaceChars,
90     StringRef PreviousPostfix, StringRef CurrentPrefix, bool InPPDirective,
91     unsigned Newlines, int Spaces) {
92   if (Tok.Finalized || (Tok.MacroCtx && Tok.MacroCtx->Role == MR_ExpandedArg))
93     return;
94   SourceLocation Start = Tok.getStartOfNonWhitespace().getLocWithOffset(Offset);
95   Changes.push_back(
96       Change(Tok, /*CreateReplacement=*/true,
97              SourceRange(Start, Start.getLocWithOffset(ReplaceChars)), Spaces,
98              std::max(0, Spaces), Newlines, PreviousPostfix, CurrentPrefix,
99              /*IsAligned=*/true, InPPDirective && !Tok.IsFirst,
100              /*IsInsideToken=*/true));
101 }
102 
generateReplacements()103 const tooling::Replacements &WhitespaceManager::generateReplacements() {
104   if (Changes.empty())
105     return Replaces;
106 
107   llvm::sort(Changes, Change::IsBeforeInFile(SourceMgr));
108   calculateLineBreakInformation();
109   alignConsecutiveMacros();
110   alignConsecutiveShortCaseStatements(/*IsExpr=*/true);
111   alignConsecutiveShortCaseStatements(/*IsExpr=*/false);
112   alignConsecutiveDeclarations();
113   alignConsecutiveBitFields();
114   alignConsecutiveAssignments();
115   if (Style.isTableGen()) {
116     alignConsecutiveTableGenBreakingDAGArgColons();
117     alignConsecutiveTableGenCondOperatorColons();
118     alignConsecutiveTableGenDefinitions();
119   }
120   alignChainedConditionals();
121   alignTrailingComments();
122   alignEscapedNewlines();
123   alignArrayInitializers();
124   generateChanges();
125 
126   return Replaces;
127 }
128 
calculateLineBreakInformation()129 void WhitespaceManager::calculateLineBreakInformation() {
130   Changes[0].PreviousEndOfTokenColumn = 0;
131   Change *LastOutsideTokenChange = &Changes[0];
132   for (unsigned I = 1, e = Changes.size(); I != e; ++I) {
133     auto &C = Changes[I];
134     auto &P = Changes[I - 1];
135     auto &PrevTokLength = P.TokenLength;
136     SourceLocation OriginalWhitespaceStart =
137         C.OriginalWhitespaceRange.getBegin();
138     SourceLocation PreviousOriginalWhitespaceEnd =
139         P.OriginalWhitespaceRange.getEnd();
140     unsigned OriginalWhitespaceStartOffset =
141         SourceMgr.getFileOffset(OriginalWhitespaceStart);
142     unsigned PreviousOriginalWhitespaceEndOffset =
143         SourceMgr.getFileOffset(PreviousOriginalWhitespaceEnd);
144     assert(PreviousOriginalWhitespaceEndOffset <=
145            OriginalWhitespaceStartOffset);
146     const char *const PreviousOriginalWhitespaceEndData =
147         SourceMgr.getCharacterData(PreviousOriginalWhitespaceEnd);
148     StringRef Text(PreviousOriginalWhitespaceEndData,
149                    SourceMgr.getCharacterData(OriginalWhitespaceStart) -
150                        PreviousOriginalWhitespaceEndData);
151     // Usually consecutive changes would occur in consecutive tokens. This is
152     // not the case however when analyzing some preprocessor runs of the
153     // annotated lines. For example, in this code:
154     //
155     // #if A // line 1
156     // int i = 1;
157     // #else B // line 2
158     // int i = 2;
159     // #endif // line 3
160     //
161     // one of the runs will produce the sequence of lines marked with line 1, 2
162     // and 3. So the two consecutive whitespace changes just before '// line 2'
163     // and before '#endif // line 3' span multiple lines and tokens:
164     //
165     // #else B{change X}[// line 2
166     // int i = 2;
167     // ]{change Y}#endif // line 3
168     //
169     // For this reason, if the text between consecutive changes spans multiple
170     // newlines, the token length must be adjusted to the end of the original
171     // line of the token.
172     auto NewlinePos = Text.find_first_of('\n');
173     if (NewlinePos == StringRef::npos) {
174       PrevTokLength = OriginalWhitespaceStartOffset -
175                       PreviousOriginalWhitespaceEndOffset +
176                       C.PreviousLinePostfix.size() + P.CurrentLinePrefix.size();
177       if (!P.IsInsideToken)
178         PrevTokLength = std::min(PrevTokLength, P.Tok->ColumnWidth);
179     } else {
180       PrevTokLength = NewlinePos + P.CurrentLinePrefix.size();
181     }
182 
183     // If there are multiple changes in this token, sum up all the changes until
184     // the end of the line.
185     if (P.IsInsideToken && P.NewlinesBefore == 0)
186       LastOutsideTokenChange->TokenLength += PrevTokLength + P.Spaces;
187     else
188       LastOutsideTokenChange = &P;
189 
190     C.PreviousEndOfTokenColumn = P.StartOfTokenColumn + PrevTokLength;
191 
192     P.IsTrailingComment =
193         (C.NewlinesBefore > 0 || C.Tok->is(tok::eof) ||
194          (C.IsInsideToken && C.Tok->is(tok::comment))) &&
195         P.Tok->is(tok::comment) &&
196         // FIXME: This is a dirty hack. The problem is that
197         // BreakableLineCommentSection does comment reflow changes and here is
198         // the aligning of trailing comments. Consider the case where we reflow
199         // the second line up in this example:
200         //
201         // // line 1
202         // // line 2
203         //
204         // That amounts to 2 changes by BreakableLineCommentSection:
205         //  - the first, delimited by (), for the whitespace between the tokens,
206         //  - and second, delimited by [], for the whitespace at the beginning
207         //  of the second token:
208         //
209         // // line 1(
210         // )[// ]line 2
211         //
212         // So in the end we have two changes like this:
213         //
214         // // line1()[ ]line 2
215         //
216         // Note that the OriginalWhitespaceStart of the second change is the
217         // same as the PreviousOriginalWhitespaceEnd of the first change.
218         // In this case, the below check ensures that the second change doesn't
219         // get treated as a trailing comment change here, since this might
220         // trigger additional whitespace to be wrongly inserted before "line 2"
221         // by the comment aligner here.
222         //
223         // For a proper solution we need a mechanism to say to WhitespaceManager
224         // that a particular change breaks the current sequence of trailing
225         // comments.
226         OriginalWhitespaceStart != PreviousOriginalWhitespaceEnd;
227   }
228   // FIXME: The last token is currently not always an eof token; in those
229   // cases, setting TokenLength of the last token to 0 is wrong.
230   Changes.back().TokenLength = 0;
231   Changes.back().IsTrailingComment = Changes.back().Tok->is(tok::comment);
232 
233   const WhitespaceManager::Change *LastBlockComment = nullptr;
234   for (auto &Change : Changes) {
235     // Reset the IsTrailingComment flag for changes inside of trailing comments
236     // so they don't get realigned later. Comment line breaks however still need
237     // to be aligned.
238     if (Change.IsInsideToken && Change.NewlinesBefore == 0)
239       Change.IsTrailingComment = false;
240     Change.StartOfBlockComment = nullptr;
241     Change.IndentationOffset = 0;
242     if (Change.Tok->is(tok::comment)) {
243       if (Change.Tok->is(TT_LineComment) || !Change.IsInsideToken) {
244         LastBlockComment = &Change;
245       } else if ((Change.StartOfBlockComment = LastBlockComment)) {
246         Change.IndentationOffset =
247             Change.StartOfTokenColumn -
248             Change.StartOfBlockComment->StartOfTokenColumn;
249       }
250     } else {
251       LastBlockComment = nullptr;
252     }
253   }
254 
255   // Compute conditional nesting level
256   // Level is increased for each conditional, unless this conditional continues
257   // a chain of conditional, i.e. starts immediately after the colon of another
258   // conditional.
259   SmallVector<bool, 16> ScopeStack;
260   int ConditionalsLevel = 0;
261   for (auto &Change : Changes) {
262     for (unsigned i = 0, e = Change.Tok->FakeLParens.size(); i != e; ++i) {
263       bool isNestedConditional =
264           Change.Tok->FakeLParens[e - 1 - i] == prec::Conditional &&
265           !(i == 0 && Change.Tok->Previous &&
266             Change.Tok->Previous->is(TT_ConditionalExpr) &&
267             Change.Tok->Previous->is(tok::colon));
268       if (isNestedConditional)
269         ++ConditionalsLevel;
270       ScopeStack.push_back(isNestedConditional);
271     }
272 
273     Change.ConditionalsLevel = ConditionalsLevel;
274 
275     for (unsigned i = Change.Tok->FakeRParens; i > 0 && ScopeStack.size(); --i)
276       if (ScopeStack.pop_back_val())
277         --ConditionalsLevel;
278   }
279 }
280 
281 // Align a single sequence of tokens, see AlignTokens below.
282 // Column - The token for which Matches returns true is moved to this column.
283 // RightJustify - Whether it is the token's right end or left end that gets
284 // moved to that column.
285 template <typename F>
286 static void
AlignTokenSequence(const FormatStyle & Style,unsigned Start,unsigned End,unsigned Column,bool RightJustify,F && Matches,SmallVector<WhitespaceManager::Change,16> & Changes)287 AlignTokenSequence(const FormatStyle &Style, unsigned Start, unsigned End,
288                    unsigned Column, bool RightJustify, F &&Matches,
289                    SmallVector<WhitespaceManager::Change, 16> &Changes) {
290   bool FoundMatchOnLine = false;
291   int Shift = 0;
292 
293   // ScopeStack keeps track of the current scope depth. It contains indices of
294   // the first token on each scope.
295   // We only run the "Matches" function on tokens from the outer-most scope.
296   // However, we do need to pay special attention to one class of tokens
297   // that are not in the outer-most scope, and that is function parameters
298   // which are split across multiple lines, as illustrated by this example:
299   //   double a(int x);
300   //   int    b(int  y,
301   //          double z);
302   // In the above example, we need to take special care to ensure that
303   // 'double z' is indented along with it's owning function 'b'.
304   // The same holds for calling a function:
305   //   double a = foo(x);
306   //   int    b = bar(foo(y),
307   //            foor(z));
308   // Similar for broken string literals:
309   //   double x = 3.14;
310   //   auto s   = "Hello"
311   //          "World";
312   // Special handling is required for 'nested' ternary operators.
313   SmallVector<unsigned, 16> ScopeStack;
314 
315   for (unsigned i = Start; i != End; ++i) {
316     auto &CurrentChange = Changes[i];
317     if (ScopeStack.size() != 0 &&
318         CurrentChange.indentAndNestingLevel() <
319             Changes[ScopeStack.back()].indentAndNestingLevel()) {
320       ScopeStack.pop_back();
321     }
322 
323     // Compare current token to previous non-comment token to ensure whether
324     // it is in a deeper scope or not.
325     unsigned PreviousNonComment = i - 1;
326     while (PreviousNonComment > Start &&
327            Changes[PreviousNonComment].Tok->is(tok::comment)) {
328       --PreviousNonComment;
329     }
330     if (i != Start && CurrentChange.indentAndNestingLevel() >
331                           Changes[PreviousNonComment].indentAndNestingLevel()) {
332       ScopeStack.push_back(i);
333     }
334 
335     bool InsideNestedScope = ScopeStack.size() != 0;
336     bool ContinuedStringLiteral = i > Start &&
337                                   CurrentChange.Tok->is(tok::string_literal) &&
338                                   Changes[i - 1].Tok->is(tok::string_literal);
339     bool SkipMatchCheck = InsideNestedScope || ContinuedStringLiteral;
340 
341     if (CurrentChange.NewlinesBefore > 0 && !SkipMatchCheck) {
342       Shift = 0;
343       FoundMatchOnLine = false;
344     }
345 
346     // If this is the first matching token to be aligned, remember by how many
347     // spaces it has to be shifted, so the rest of the changes on the line are
348     // shifted by the same amount
349     if (!FoundMatchOnLine && !SkipMatchCheck && Matches(CurrentChange)) {
350       FoundMatchOnLine = true;
351       Shift = Column - (RightJustify ? CurrentChange.TokenLength : 0) -
352               CurrentChange.StartOfTokenColumn;
353       CurrentChange.Spaces += Shift;
354       // FIXME: This is a workaround that should be removed when we fix
355       // http://llvm.org/PR53699. An assertion later below verifies this.
356       if (CurrentChange.NewlinesBefore == 0) {
357         CurrentChange.Spaces =
358             std::max(CurrentChange.Spaces,
359                      static_cast<int>(CurrentChange.Tok->SpacesRequiredBefore));
360       }
361     }
362 
363     if (Shift == 0)
364       continue;
365 
366     // This is for function parameters that are split across multiple lines,
367     // as mentioned in the ScopeStack comment.
368     if (InsideNestedScope && CurrentChange.NewlinesBefore > 0) {
369       unsigned ScopeStart = ScopeStack.back();
370       auto ShouldShiftBeAdded = [&] {
371         // Function declaration
372         if (Changes[ScopeStart - 1].Tok->is(TT_FunctionDeclarationName))
373           return true;
374 
375         // Lambda.
376         if (Changes[ScopeStart - 1].Tok->is(TT_LambdaLBrace))
377           return false;
378 
379         // Continued function declaration
380         if (ScopeStart > Start + 1 &&
381             Changes[ScopeStart - 2].Tok->is(TT_FunctionDeclarationName)) {
382           return true;
383         }
384 
385         // Continued (template) function call.
386         if (ScopeStart > Start + 1 &&
387             Changes[ScopeStart - 2].Tok->isOneOf(tok::identifier,
388                                                  TT_TemplateCloser) &&
389             Changes[ScopeStart - 1].Tok->is(tok::l_paren) &&
390             Changes[ScopeStart].Tok->isNot(TT_LambdaLSquare)) {
391           if (CurrentChange.Tok->MatchingParen &&
392               CurrentChange.Tok->MatchingParen->is(TT_LambdaLBrace)) {
393             return false;
394           }
395           if (Changes[ScopeStart].NewlinesBefore > 0)
396             return false;
397           if (CurrentChange.Tok->is(tok::l_brace) &&
398               CurrentChange.Tok->is(BK_BracedInit)) {
399             return true;
400           }
401           return Style.BinPackArguments;
402         }
403 
404         // Ternary operator
405         if (CurrentChange.Tok->is(TT_ConditionalExpr))
406           return true;
407 
408         // Period Initializer .XXX = 1.
409         if (CurrentChange.Tok->is(TT_DesignatedInitializerPeriod))
410           return true;
411 
412         // Continued ternary operator
413         if (CurrentChange.Tok->Previous &&
414             CurrentChange.Tok->Previous->is(TT_ConditionalExpr)) {
415           return true;
416         }
417 
418         // Continued direct-list-initialization using braced list.
419         if (ScopeStart > Start + 1 &&
420             Changes[ScopeStart - 2].Tok->is(tok::identifier) &&
421             Changes[ScopeStart - 1].Tok->is(tok::l_brace) &&
422             CurrentChange.Tok->is(tok::l_brace) &&
423             CurrentChange.Tok->is(BK_BracedInit)) {
424           return true;
425         }
426 
427         // Continued braced list.
428         if (ScopeStart > Start + 1 &&
429             Changes[ScopeStart - 2].Tok->isNot(tok::identifier) &&
430             Changes[ScopeStart - 1].Tok->is(tok::l_brace) &&
431             CurrentChange.Tok->isNot(tok::r_brace)) {
432           for (unsigned OuterScopeStart : llvm::reverse(ScopeStack)) {
433             // Lambda.
434             if (OuterScopeStart > Start &&
435                 Changes[OuterScopeStart - 1].Tok->is(TT_LambdaLBrace)) {
436               return false;
437             }
438           }
439           if (Changes[ScopeStart].NewlinesBefore > 0)
440             return false;
441           return true;
442         }
443 
444         // Continued template parameter.
445         if (Changes[ScopeStart - 1].Tok->is(TT_TemplateOpener))
446           return true;
447 
448         return false;
449       };
450 
451       if (ShouldShiftBeAdded())
452         CurrentChange.Spaces += Shift;
453     }
454 
455     if (ContinuedStringLiteral)
456       CurrentChange.Spaces += Shift;
457 
458     // We should not remove required spaces unless we break the line before.
459     assert(Shift > 0 || Changes[i].NewlinesBefore > 0 ||
460            CurrentChange.Spaces >=
461                static_cast<int>(Changes[i].Tok->SpacesRequiredBefore) ||
462            CurrentChange.Tok->is(tok::eof));
463 
464     CurrentChange.StartOfTokenColumn += Shift;
465     if (i + 1 != Changes.size())
466       Changes[i + 1].PreviousEndOfTokenColumn += Shift;
467 
468     // If PointerAlignment is PAS_Right, keep *s or &s next to the token,
469     // except if the token is equal, then a space is needed.
470     if ((Style.PointerAlignment == FormatStyle::PAS_Right ||
471          Style.ReferenceAlignment == FormatStyle::RAS_Right) &&
472         CurrentChange.Spaces != 0 &&
473         !CurrentChange.Tok->isOneOf(tok::equal, tok::r_paren,
474                                     TT_TemplateCloser)) {
475       const bool ReferenceNotRightAligned =
476           Style.ReferenceAlignment != FormatStyle::RAS_Right &&
477           Style.ReferenceAlignment != FormatStyle::RAS_Pointer;
478       for (int Previous = i - 1;
479            Previous >= 0 && Changes[Previous].Tok->is(TT_PointerOrReference);
480            --Previous) {
481         assert(Changes[Previous].Tok->isPointerOrReference());
482         if (Changes[Previous].Tok->isNot(tok::star)) {
483           if (ReferenceNotRightAligned)
484             continue;
485         } else if (Style.PointerAlignment != FormatStyle::PAS_Right) {
486           continue;
487         }
488         Changes[Previous + 1].Spaces -= Shift;
489         Changes[Previous].Spaces += Shift;
490         Changes[Previous].StartOfTokenColumn += Shift;
491       }
492     }
493   }
494 }
495 
496 // Walk through a subset of the changes, starting at StartAt, and find
497 // sequences of matching tokens to align. To do so, keep track of the lines and
498 // whether or not a matching token was found on a line. If a matching token is
499 // found, extend the current sequence. If the current line cannot be part of a
500 // sequence, e.g. because there is an empty line before it or it contains only
501 // non-matching tokens, finalize the previous sequence.
502 // The value returned is the token on which we stopped, either because we
503 // exhausted all items inside Changes, or because we hit a scope level higher
504 // than our initial scope.
505 // This function is recursive. Each invocation processes only the scope level
506 // equal to the initial level, which is the level of Changes[StartAt].
507 // If we encounter a scope level greater than the initial level, then we call
508 // ourselves recursively, thereby avoiding the pollution of the current state
509 // with the alignment requirements of the nested sub-level. This recursive
510 // behavior is necessary for aligning function prototypes that have one or more
511 // arguments.
512 // If this function encounters a scope level less than the initial level,
513 // it returns the current position.
514 // There is a non-obvious subtlety in the recursive behavior: Even though we
515 // defer processing of nested levels to recursive invocations of this
516 // function, when it comes time to align a sequence of tokens, we run the
517 // alignment on the entire sequence, including the nested levels.
518 // When doing so, most of the nested tokens are skipped, because their
519 // alignment was already handled by the recursive invocations of this function.
520 // However, the special exception is that we do NOT skip function parameters
521 // that are split across multiple lines. See the test case in FormatTest.cpp
522 // that mentions "split function parameter alignment" for an example of this.
523 // When the parameter RightJustify is true, the operator will be
524 // right-justified. It is used to align compound assignments like `+=` and `=`.
525 // When RightJustify and ACS.PadOperators are true, operators in each block to
526 // be aligned will be padded on the left to the same length before aligning.
527 template <typename F>
AlignTokens(const FormatStyle & Style,F && Matches,SmallVector<WhitespaceManager::Change,16> & Changes,unsigned StartAt,const FormatStyle::AlignConsecutiveStyle & ACS={},bool RightJustify=false)528 static unsigned AlignTokens(const FormatStyle &Style, F &&Matches,
529                             SmallVector<WhitespaceManager::Change, 16> &Changes,
530                             unsigned StartAt,
531                             const FormatStyle::AlignConsecutiveStyle &ACS = {},
532                             bool RightJustify = false) {
533   // We arrange each line in 3 parts. The operator to be aligned (the anchor),
534   // and text to its left and right. In the aligned text the width of each part
535   // will be the maximum of that over the block that has been aligned. Maximum
536   // widths of each part so far. When RightJustify is true and ACS.PadOperators
537   // is false, the part from start of line to the right end of the anchor.
538   // Otherwise, only the part to the left of the anchor. Including the space
539   // that exists on its left from the start. Not including the padding added on
540   // the left to right-justify the anchor.
541   unsigned WidthLeft = 0;
542   // The operator to be aligned when RightJustify is true and ACS.PadOperators
543   // is false. 0 otherwise.
544   unsigned WidthAnchor = 0;
545   // Width to the right of the anchor. Plus width of the anchor when
546   // RightJustify is false.
547   unsigned WidthRight = 0;
548 
549   // Line number of the start and the end of the current token sequence.
550   unsigned StartOfSequence = 0;
551   unsigned EndOfSequence = 0;
552 
553   // Measure the scope level (i.e. depth of (), [], {}) of the first token, and
554   // abort when we hit any token in a higher scope than the starting one.
555   auto IndentAndNestingLevel = StartAt < Changes.size()
556                                    ? Changes[StartAt].indentAndNestingLevel()
557                                    : std::tuple<unsigned, unsigned, unsigned>();
558 
559   // Keep track of the number of commas before the matching tokens, we will only
560   // align a sequence of matching tokens if they are preceded by the same number
561   // of commas.
562   unsigned CommasBeforeLastMatch = 0;
563   unsigned CommasBeforeMatch = 0;
564 
565   // Whether a matching token has been found on the current line.
566   bool FoundMatchOnLine = false;
567 
568   // Whether the current line consists purely of comments.
569   bool LineIsComment = true;
570 
571   // Aligns a sequence of matching tokens, on the MinColumn column.
572   //
573   // Sequences start from the first matching token to align, and end at the
574   // first token of the first line that doesn't need to be aligned.
575   //
576   // We need to adjust the StartOfTokenColumn of each Change that is on a line
577   // containing any matching token to be aligned and located after such token.
__anonfa4fec7a0202null578   auto AlignCurrentSequence = [&] {
579     if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) {
580       AlignTokenSequence(Style, StartOfSequence, EndOfSequence,
581                          WidthLeft + WidthAnchor, RightJustify, Matches,
582                          Changes);
583     }
584     WidthLeft = 0;
585     WidthAnchor = 0;
586     WidthRight = 0;
587     StartOfSequence = 0;
588     EndOfSequence = 0;
589   };
590 
591   unsigned i = StartAt;
592   for (unsigned e = Changes.size(); i != e; ++i) {
593     auto &CurrentChange = Changes[i];
594     if (CurrentChange.indentAndNestingLevel() < IndentAndNestingLevel)
595       break;
596 
597     if (CurrentChange.NewlinesBefore != 0) {
598       CommasBeforeMatch = 0;
599       EndOfSequence = i;
600 
601       // Whether to break the alignment sequence because of an empty line.
602       bool EmptyLineBreak =
603           (CurrentChange.NewlinesBefore > 1) && !ACS.AcrossEmptyLines;
604 
605       // Whether to break the alignment sequence because of a line without a
606       // match.
607       bool NoMatchBreak =
608           !FoundMatchOnLine && !(LineIsComment && ACS.AcrossComments);
609 
610       if (EmptyLineBreak || NoMatchBreak)
611         AlignCurrentSequence();
612 
613       // A new line starts, re-initialize line status tracking bools.
614       // Keep the match state if a string literal is continued on this line.
615       if (i == 0 || CurrentChange.Tok->isNot(tok::string_literal) ||
616           Changes[i - 1].Tok->isNot(tok::string_literal)) {
617         FoundMatchOnLine = false;
618       }
619       LineIsComment = true;
620     }
621 
622     if (CurrentChange.Tok->isNot(tok::comment))
623       LineIsComment = false;
624 
625     if (CurrentChange.Tok->is(tok::comma)) {
626       ++CommasBeforeMatch;
627     } else if (CurrentChange.indentAndNestingLevel() > IndentAndNestingLevel) {
628       // Call AlignTokens recursively, skipping over this scope block.
629       unsigned StoppedAt =
630           AlignTokens(Style, Matches, Changes, i, ACS, RightJustify);
631       i = StoppedAt - 1;
632       continue;
633     }
634 
635     if (!Matches(CurrentChange))
636       continue;
637 
638     // If there is more than one matching token per line, or if the number of
639     // preceding commas, do not match anymore, end the sequence.
640     if (FoundMatchOnLine || CommasBeforeMatch != CommasBeforeLastMatch)
641       AlignCurrentSequence();
642 
643     CommasBeforeLastMatch = CommasBeforeMatch;
644     FoundMatchOnLine = true;
645 
646     if (StartOfSequence == 0)
647       StartOfSequence = i;
648 
649     unsigned ChangeWidthLeft = CurrentChange.StartOfTokenColumn;
650     unsigned ChangeWidthAnchor = 0;
651     unsigned ChangeWidthRight = 0;
652     if (RightJustify)
653       if (ACS.PadOperators)
654         ChangeWidthAnchor = CurrentChange.TokenLength;
655       else
656         ChangeWidthLeft += CurrentChange.TokenLength;
657     else
658       ChangeWidthRight = CurrentChange.TokenLength;
659     for (unsigned j = i + 1; j != e && Changes[j].NewlinesBefore == 0; ++j) {
660       ChangeWidthRight += Changes[j].Spaces;
661       // Changes are generally 1:1 with the tokens, but a change could also be
662       // inside of a token, in which case it's counted more than once: once for
663       // the whitespace surrounding the token (!IsInsideToken) and once for
664       // each whitespace change within it (IsInsideToken).
665       // Therefore, changes inside of a token should only count the space.
666       if (!Changes[j].IsInsideToken)
667         ChangeWidthRight += Changes[j].TokenLength;
668     }
669 
670     // If we are restricted by the maximum column width, end the sequence.
671     unsigned NewLeft = std::max(ChangeWidthLeft, WidthLeft);
672     unsigned NewAnchor = std::max(ChangeWidthAnchor, WidthAnchor);
673     unsigned NewRight = std::max(ChangeWidthRight, WidthRight);
674     // `ColumnLimit == 0` means there is no column limit.
675     if (Style.ColumnLimit != 0 &&
676         Style.ColumnLimit < NewLeft + NewAnchor + NewRight) {
677       AlignCurrentSequence();
678       StartOfSequence = i;
679       WidthLeft = ChangeWidthLeft;
680       WidthAnchor = ChangeWidthAnchor;
681       WidthRight = ChangeWidthRight;
682     } else {
683       WidthLeft = NewLeft;
684       WidthAnchor = NewAnchor;
685       WidthRight = NewRight;
686     }
687   }
688 
689   EndOfSequence = i;
690   AlignCurrentSequence();
691   return i;
692 }
693 
694 // Aligns a sequence of matching tokens, on the MinColumn column.
695 //
696 // Sequences start from the first matching token to align, and end at the
697 // first token of the first line that doesn't need to be aligned.
698 //
699 // We need to adjust the StartOfTokenColumn of each Change that is on a line
700 // containing any matching token to be aligned and located after such token.
AlignMatchingTokenSequence(unsigned & StartOfSequence,unsigned & EndOfSequence,unsigned & MinColumn,std::function<bool (const WhitespaceManager::Change & C)> Matches,SmallVector<WhitespaceManager::Change,16> & Changes)701 static void AlignMatchingTokenSequence(
702     unsigned &StartOfSequence, unsigned &EndOfSequence, unsigned &MinColumn,
703     std::function<bool(const WhitespaceManager::Change &C)> Matches,
704     SmallVector<WhitespaceManager::Change, 16> &Changes) {
705   if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) {
706     bool FoundMatchOnLine = false;
707     int Shift = 0;
708 
709     for (unsigned I = StartOfSequence; I != EndOfSequence; ++I) {
710       if (Changes[I].NewlinesBefore > 0) {
711         Shift = 0;
712         FoundMatchOnLine = false;
713       }
714 
715       // If this is the first matching token to be aligned, remember by how many
716       // spaces it has to be shifted, so the rest of the changes on the line are
717       // shifted by the same amount.
718       if (!FoundMatchOnLine && Matches(Changes[I])) {
719         FoundMatchOnLine = true;
720         Shift = MinColumn - Changes[I].StartOfTokenColumn;
721         Changes[I].Spaces += Shift;
722       }
723 
724       assert(Shift >= 0);
725       Changes[I].StartOfTokenColumn += Shift;
726       if (I + 1 != Changes.size())
727         Changes[I + 1].PreviousEndOfTokenColumn += Shift;
728     }
729   }
730 
731   MinColumn = 0;
732   StartOfSequence = 0;
733   EndOfSequence = 0;
734 }
735 
alignConsecutiveMacros()736 void WhitespaceManager::alignConsecutiveMacros() {
737   if (!Style.AlignConsecutiveMacros.Enabled)
738     return;
739 
740   auto AlignMacrosMatches = [](const Change &C) {
741     const FormatToken *Current = C.Tok;
742     unsigned SpacesRequiredBefore = 1;
743 
744     if (Current->SpacesRequiredBefore == 0 || !Current->Previous)
745       return false;
746 
747     Current = Current->Previous;
748 
749     // If token is a ")", skip over the parameter list, to the
750     // token that precedes the "("
751     if (Current->is(tok::r_paren) && Current->MatchingParen) {
752       Current = Current->MatchingParen->Previous;
753       SpacesRequiredBefore = 0;
754     }
755 
756     if (!Current || Current->isNot(tok::identifier))
757       return false;
758 
759     if (!Current->Previous || Current->Previous->isNot(tok::pp_define))
760       return false;
761 
762     // For a macro function, 0 spaces are required between the
763     // identifier and the lparen that opens the parameter list.
764     // For a simple macro, 1 space is required between the
765     // identifier and the first token of the defined value.
766     return Current->Next->SpacesRequiredBefore == SpacesRequiredBefore;
767   };
768 
769   unsigned MinColumn = 0;
770 
771   // Start and end of the token sequence we're processing.
772   unsigned StartOfSequence = 0;
773   unsigned EndOfSequence = 0;
774 
775   // Whether a matching token has been found on the current line.
776   bool FoundMatchOnLine = false;
777 
778   // Whether the current line consists only of comments
779   bool LineIsComment = true;
780 
781   unsigned I = 0;
782   for (unsigned E = Changes.size(); I != E; ++I) {
783     if (Changes[I].NewlinesBefore != 0) {
784       EndOfSequence = I;
785 
786       // Whether to break the alignment sequence because of an empty line.
787       bool EmptyLineBreak = (Changes[I].NewlinesBefore > 1) &&
788                             !Style.AlignConsecutiveMacros.AcrossEmptyLines;
789 
790       // Whether to break the alignment sequence because of a line without a
791       // match.
792       bool NoMatchBreak =
793           !FoundMatchOnLine &&
794           !(LineIsComment && Style.AlignConsecutiveMacros.AcrossComments);
795 
796       if (EmptyLineBreak || NoMatchBreak) {
797         AlignMatchingTokenSequence(StartOfSequence, EndOfSequence, MinColumn,
798                                    AlignMacrosMatches, Changes);
799       }
800 
801       // A new line starts, re-initialize line status tracking bools.
802       FoundMatchOnLine = false;
803       LineIsComment = true;
804     }
805 
806     if (Changes[I].Tok->isNot(tok::comment))
807       LineIsComment = false;
808 
809     if (!AlignMacrosMatches(Changes[I]))
810       continue;
811 
812     FoundMatchOnLine = true;
813 
814     if (StartOfSequence == 0)
815       StartOfSequence = I;
816 
817     unsigned ChangeMinColumn = Changes[I].StartOfTokenColumn;
818     MinColumn = std::max(MinColumn, ChangeMinColumn);
819   }
820 
821   EndOfSequence = I;
822   AlignMatchingTokenSequence(StartOfSequence, EndOfSequence, MinColumn,
823                              AlignMacrosMatches, Changes);
824 }
825 
alignConsecutiveAssignments()826 void WhitespaceManager::alignConsecutiveAssignments() {
827   if (!Style.AlignConsecutiveAssignments.Enabled)
828     return;
829 
830   AlignTokens(
831       Style,
832       [&](const Change &C) {
833         // Do not align on equal signs that are first on a line.
834         if (C.NewlinesBefore > 0)
835           return false;
836 
837         // Do not align on equal signs that are last on a line.
838         if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
839           return false;
840 
841         // Do not align operator= overloads.
842         FormatToken *Previous = C.Tok->getPreviousNonComment();
843         if (Previous && Previous->is(tok::kw_operator))
844           return false;
845 
846         return Style.AlignConsecutiveAssignments.AlignCompound
847                    ? C.Tok->getPrecedence() == prec::Assignment
848                    : (C.Tok->is(tok::equal) ||
849                       // In Verilog the '<=' is not a compound assignment, thus
850                       // it is aligned even when the AlignCompound option is not
851                       // set.
852                       (Style.isVerilog() && C.Tok->is(tok::lessequal) &&
853                        C.Tok->getPrecedence() == prec::Assignment));
854       },
855       Changes, /*StartAt=*/0, Style.AlignConsecutiveAssignments,
856       /*RightJustify=*/true);
857 }
858 
alignConsecutiveBitFields()859 void WhitespaceManager::alignConsecutiveBitFields() {
860   alignConsecutiveColons(Style.AlignConsecutiveBitFields, TT_BitFieldColon);
861 }
862 
alignConsecutiveColons(const FormatStyle::AlignConsecutiveStyle & AlignStyle,TokenType Type)863 void WhitespaceManager::alignConsecutiveColons(
864     const FormatStyle::AlignConsecutiveStyle &AlignStyle, TokenType Type) {
865   if (!AlignStyle.Enabled)
866     return;
867 
868   AlignTokens(
869       Style,
870       [&](Change const &C) {
871         // Do not align on ':' that is first on a line.
872         if (C.NewlinesBefore > 0)
873           return false;
874 
875         // Do not align on ':' that is last on a line.
876         if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
877           return false;
878 
879         return C.Tok->is(Type);
880       },
881       Changes, /*StartAt=*/0, AlignStyle);
882 }
883 
alignConsecutiveShortCaseStatements(bool IsExpr)884 void WhitespaceManager::alignConsecutiveShortCaseStatements(bool IsExpr) {
885   if (!Style.AlignConsecutiveShortCaseStatements.Enabled ||
886       !(IsExpr ? Style.AllowShortCaseExpressionOnASingleLine
887                : Style.AllowShortCaseLabelsOnASingleLine)) {
888     return;
889   }
890 
891   const auto Type = IsExpr ? TT_CaseLabelArrow : TT_CaseLabelColon;
892   const auto &Option = Style.AlignConsecutiveShortCaseStatements;
893   const bool AlignArrowOrColon =
894       IsExpr ? Option.AlignCaseArrows : Option.AlignCaseColons;
895 
896   auto Matches = [&](const Change &C) {
897     if (AlignArrowOrColon)
898       return C.Tok->is(Type);
899 
900     // Ignore 'IsInsideToken' to allow matching trailing comments which
901     // need to be reflowed as that causes the token to appear in two
902     // different changes, which will cause incorrect alignment as we'll
903     // reflow early due to detecting multiple aligning tokens per line.
904     return !C.IsInsideToken && C.Tok->Previous && C.Tok->Previous->is(Type);
905   };
906 
907   unsigned MinColumn = 0;
908 
909   // Empty case statements don't break the alignment, but don't necessarily
910   // match our predicate, so we need to track their column so they can push out
911   // our alignment.
912   unsigned MinEmptyCaseColumn = 0;
913 
914   // Start and end of the token sequence we're processing.
915   unsigned StartOfSequence = 0;
916   unsigned EndOfSequence = 0;
917 
918   // Whether a matching token has been found on the current line.
919   bool FoundMatchOnLine = false;
920 
921   bool LineIsComment = true;
922   bool LineIsEmptyCase = false;
923 
924   unsigned I = 0;
925   for (unsigned E = Changes.size(); I != E; ++I) {
926     if (Changes[I].NewlinesBefore != 0) {
927       // Whether to break the alignment sequence because of an empty line.
928       bool EmptyLineBreak =
929           (Changes[I].NewlinesBefore > 1) &&
930           !Style.AlignConsecutiveShortCaseStatements.AcrossEmptyLines;
931 
932       // Whether to break the alignment sequence because of a line without a
933       // match.
934       bool NoMatchBreak =
935           !FoundMatchOnLine &&
936           !(LineIsComment &&
937             Style.AlignConsecutiveShortCaseStatements.AcrossComments) &&
938           !LineIsEmptyCase;
939 
940       if (EmptyLineBreak || NoMatchBreak) {
941         AlignMatchingTokenSequence(StartOfSequence, EndOfSequence, MinColumn,
942                                    Matches, Changes);
943         MinEmptyCaseColumn = 0;
944       }
945 
946       // A new line starts, re-initialize line status tracking bools.
947       FoundMatchOnLine = false;
948       LineIsComment = true;
949       LineIsEmptyCase = false;
950     }
951 
952     if (Changes[I].Tok->isNot(tok::comment))
953       LineIsComment = false;
954 
955     if (Changes[I].Tok->is(Type)) {
956       LineIsEmptyCase =
957           !Changes[I].Tok->Next || Changes[I].Tok->Next->isTrailingComment();
958 
959       if (LineIsEmptyCase) {
960         if (Style.AlignConsecutiveShortCaseStatements.AlignCaseColons) {
961           MinEmptyCaseColumn =
962               std::max(MinEmptyCaseColumn, Changes[I].StartOfTokenColumn);
963         } else {
964           MinEmptyCaseColumn =
965               std::max(MinEmptyCaseColumn, Changes[I].StartOfTokenColumn + 2);
966         }
967       }
968     }
969 
970     if (!Matches(Changes[I]))
971       continue;
972 
973     if (LineIsEmptyCase)
974       continue;
975 
976     FoundMatchOnLine = true;
977 
978     if (StartOfSequence == 0)
979       StartOfSequence = I;
980 
981     EndOfSequence = I + 1;
982 
983     MinColumn = std::max(MinColumn, Changes[I].StartOfTokenColumn);
984 
985     // Allow empty case statements to push out our alignment.
986     MinColumn = std::max(MinColumn, MinEmptyCaseColumn);
987   }
988 
989   AlignMatchingTokenSequence(StartOfSequence, EndOfSequence, MinColumn, Matches,
990                              Changes);
991 }
992 
alignConsecutiveTableGenBreakingDAGArgColons()993 void WhitespaceManager::alignConsecutiveTableGenBreakingDAGArgColons() {
994   alignConsecutiveColons(Style.AlignConsecutiveTableGenBreakingDAGArgColons,
995                          TT_TableGenDAGArgListColonToAlign);
996 }
997 
alignConsecutiveTableGenCondOperatorColons()998 void WhitespaceManager::alignConsecutiveTableGenCondOperatorColons() {
999   alignConsecutiveColons(Style.AlignConsecutiveTableGenCondOperatorColons,
1000                          TT_TableGenCondOperatorColon);
1001 }
1002 
alignConsecutiveTableGenDefinitions()1003 void WhitespaceManager::alignConsecutiveTableGenDefinitions() {
1004   alignConsecutiveColons(Style.AlignConsecutiveTableGenDefinitionColons,
1005                          TT_InheritanceColon);
1006 }
1007 
alignConsecutiveDeclarations()1008 void WhitespaceManager::alignConsecutiveDeclarations() {
1009   if (!Style.AlignConsecutiveDeclarations.Enabled)
1010     return;
1011 
1012   AlignTokens(
1013       Style,
1014       [&](Change const &C) {
1015         if (Style.AlignConsecutiveDeclarations.AlignFunctionPointers) {
1016           for (const auto *Prev = C.Tok->Previous; Prev; Prev = Prev->Previous)
1017             if (Prev->is(tok::equal))
1018               return false;
1019           if (C.Tok->is(TT_FunctionTypeLParen))
1020             return true;
1021         }
1022         if (C.Tok->is(TT_FunctionDeclarationName))
1023           return true;
1024         if (C.Tok->isNot(TT_StartOfName))
1025           return false;
1026         if (C.Tok->Previous &&
1027             C.Tok->Previous->is(TT_StatementAttributeLikeMacro))
1028           return false;
1029         // Check if there is a subsequent name that starts the same declaration.
1030         for (FormatToken *Next = C.Tok->Next; Next; Next = Next->Next) {
1031           if (Next->is(tok::comment))
1032             continue;
1033           if (Next->is(TT_PointerOrReference))
1034             return false;
1035           if (!Next->Tok.getIdentifierInfo())
1036             break;
1037           if (Next->isOneOf(TT_StartOfName, TT_FunctionDeclarationName,
1038                             tok::kw_operator)) {
1039             return false;
1040           }
1041         }
1042         return true;
1043       },
1044       Changes, /*StartAt=*/0, Style.AlignConsecutiveDeclarations);
1045 }
1046 
alignChainedConditionals()1047 void WhitespaceManager::alignChainedConditionals() {
1048   if (Style.BreakBeforeTernaryOperators) {
1049     AlignTokens(
1050         Style,
1051         [](Change const &C) {
1052           // Align question operators and last colon
1053           return C.Tok->is(TT_ConditionalExpr) &&
1054                  ((C.Tok->is(tok::question) && !C.NewlinesBefore) ||
1055                   (C.Tok->is(tok::colon) && C.Tok->Next &&
1056                    (C.Tok->Next->FakeLParens.size() == 0 ||
1057                     C.Tok->Next->FakeLParens.back() != prec::Conditional)));
1058         },
1059         Changes, /*StartAt=*/0);
1060   } else {
1061     static auto AlignWrappedOperand = [](Change const &C) {
1062       FormatToken *Previous = C.Tok->getPreviousNonComment();
1063       return C.NewlinesBefore && Previous && Previous->is(TT_ConditionalExpr) &&
1064              (Previous->is(tok::colon) &&
1065               (C.Tok->FakeLParens.size() == 0 ||
1066                C.Tok->FakeLParens.back() != prec::Conditional));
1067     };
1068     // Ensure we keep alignment of wrapped operands with non-wrapped operands
1069     // Since we actually align the operators, the wrapped operands need the
1070     // extra offset to be properly aligned.
1071     for (Change &C : Changes)
1072       if (AlignWrappedOperand(C))
1073         C.StartOfTokenColumn -= 2;
1074     AlignTokens(
1075         Style,
1076         [this](Change const &C) {
1077           // Align question operators if next operand is not wrapped, as
1078           // well as wrapped operands after question operator or last
1079           // colon in conditional sequence
1080           return (C.Tok->is(TT_ConditionalExpr) && C.Tok->is(tok::question) &&
1081                   &C != &Changes.back() && (&C + 1)->NewlinesBefore == 0 &&
1082                   !(&C + 1)->IsTrailingComment) ||
1083                  AlignWrappedOperand(C);
1084         },
1085         Changes, /*StartAt=*/0);
1086   }
1087 }
1088 
alignTrailingComments()1089 void WhitespaceManager::alignTrailingComments() {
1090   if (Style.AlignTrailingComments.Kind == FormatStyle::TCAS_Never)
1091     return;
1092 
1093   const int Size = Changes.size();
1094   int MinColumn = 0;
1095   int StartOfSequence = 0;
1096   bool BreakBeforeNext = false;
1097   int NewLineThreshold = 1;
1098   if (Style.AlignTrailingComments.Kind == FormatStyle::TCAS_Always)
1099     NewLineThreshold = Style.AlignTrailingComments.OverEmptyLines + 1;
1100 
1101   for (int I = 0, MaxColumn = INT_MAX, Newlines = 0; I < Size; ++I) {
1102     auto &C = Changes[I];
1103     if (C.StartOfBlockComment)
1104       continue;
1105     Newlines += C.NewlinesBefore;
1106     if (!C.IsTrailingComment)
1107       continue;
1108 
1109     if (Style.AlignTrailingComments.Kind == FormatStyle::TCAS_Leave) {
1110       const int OriginalSpaces =
1111           C.OriginalWhitespaceRange.getEnd().getRawEncoding() -
1112           C.OriginalWhitespaceRange.getBegin().getRawEncoding() -
1113           C.Tok->LastNewlineOffset;
1114       assert(OriginalSpaces >= 0);
1115       const auto RestoredLineLength =
1116           C.StartOfTokenColumn + C.TokenLength + OriginalSpaces;
1117       // If leaving comments makes the line exceed the column limit, give up to
1118       // leave the comments.
1119       if (RestoredLineLength >= Style.ColumnLimit && Style.ColumnLimit > 0)
1120         break;
1121       C.Spaces = C.NewlinesBefore > 0 ? C.Tok->OriginalColumn : OriginalSpaces;
1122       continue;
1123     }
1124 
1125     const int ChangeMinColumn = C.StartOfTokenColumn;
1126     int ChangeMaxColumn;
1127 
1128     // If we don't create a replacement for this change, we have to consider
1129     // it to be immovable.
1130     if (!C.CreateReplacement)
1131       ChangeMaxColumn = ChangeMinColumn;
1132     else if (Style.ColumnLimit == 0)
1133       ChangeMaxColumn = INT_MAX;
1134     else if (Style.ColumnLimit >= C.TokenLength)
1135       ChangeMaxColumn = Style.ColumnLimit - C.TokenLength;
1136     else
1137       ChangeMaxColumn = ChangeMinColumn;
1138 
1139     if (I + 1 < Size && Changes[I + 1].ContinuesPPDirective &&
1140         ChangeMaxColumn >= 2) {
1141       ChangeMaxColumn -= 2;
1142     }
1143 
1144     bool WasAlignedWithStartOfNextLine = false;
1145     if (C.NewlinesBefore >= 1) { // A comment on its own line.
1146       const auto CommentColumn =
1147           SourceMgr.getSpellingColumnNumber(C.OriginalWhitespaceRange.getEnd());
1148       for (int J = I + 1; J < Size; ++J) {
1149         if (Changes[J].Tok->is(tok::comment))
1150           continue;
1151 
1152         const auto NextColumn = SourceMgr.getSpellingColumnNumber(
1153             Changes[J].OriginalWhitespaceRange.getEnd());
1154         // The start of the next token was previously aligned with the
1155         // start of this comment.
1156         WasAlignedWithStartOfNextLine =
1157             CommentColumn == NextColumn ||
1158             CommentColumn == NextColumn + Style.IndentWidth;
1159         break;
1160       }
1161     }
1162 
1163     // We don't want to align comments which end a scope, which are here
1164     // identified by most closing braces.
1165     auto DontAlignThisComment = [](const auto *Tok) {
1166       if (Tok->is(tok::semi)) {
1167         Tok = Tok->getPreviousNonComment();
1168         if (!Tok)
1169           return false;
1170       }
1171       if (Tok->is(tok::r_paren)) {
1172         // Back up past the parentheses and a `TT_DoWhile` that may precede.
1173         Tok = Tok->MatchingParen;
1174         if (!Tok)
1175           return false;
1176         Tok = Tok->getPreviousNonComment();
1177         if (!Tok)
1178           return false;
1179         if (Tok->is(TT_DoWhile)) {
1180           const auto *Prev = Tok->getPreviousNonComment();
1181           if (!Prev) {
1182             // A do-while-loop without braces.
1183             return true;
1184           }
1185           Tok = Prev;
1186         }
1187       }
1188 
1189       if (Tok->isNot(tok::r_brace))
1190         return false;
1191 
1192       while (Tok->Previous && Tok->Previous->is(tok::r_brace))
1193         Tok = Tok->Previous;
1194       return Tok->NewlinesBefore > 0;
1195     };
1196 
1197     if (I > 0 && C.NewlinesBefore == 0 &&
1198         DontAlignThisComment(Changes[I - 1].Tok)) {
1199       alignTrailingComments(StartOfSequence, I, MinColumn);
1200       // Reset to initial values, but skip this change for the next alignment
1201       // pass.
1202       MinColumn = 0;
1203       MaxColumn = INT_MAX;
1204       StartOfSequence = I + 1;
1205     } else if (BreakBeforeNext || Newlines > NewLineThreshold ||
1206                (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn) ||
1207                // Break the comment sequence if the previous line did not end
1208                // in a trailing comment.
1209                (C.NewlinesBefore == 1 && I > 0 &&
1210                 !Changes[I - 1].IsTrailingComment) ||
1211                WasAlignedWithStartOfNextLine) {
1212       alignTrailingComments(StartOfSequence, I, MinColumn);
1213       MinColumn = ChangeMinColumn;
1214       MaxColumn = ChangeMaxColumn;
1215       StartOfSequence = I;
1216     } else {
1217       MinColumn = std::max(MinColumn, ChangeMinColumn);
1218       MaxColumn = std::min(MaxColumn, ChangeMaxColumn);
1219     }
1220     BreakBeforeNext = (I == 0) || (C.NewlinesBefore > 1) ||
1221                       // Never start a sequence with a comment at the beginning
1222                       // of the line.
1223                       (C.NewlinesBefore == 1 && StartOfSequence == I);
1224     Newlines = 0;
1225   }
1226   alignTrailingComments(StartOfSequence, Size, MinColumn);
1227 }
1228 
alignTrailingComments(unsigned Start,unsigned End,unsigned Column)1229 void WhitespaceManager::alignTrailingComments(unsigned Start, unsigned End,
1230                                               unsigned Column) {
1231   for (unsigned i = Start; i != End; ++i) {
1232     int Shift = 0;
1233     if (Changes[i].IsTrailingComment)
1234       Shift = Column - Changes[i].StartOfTokenColumn;
1235     if (Changes[i].StartOfBlockComment) {
1236       Shift = Changes[i].IndentationOffset +
1237               Changes[i].StartOfBlockComment->StartOfTokenColumn -
1238               Changes[i].StartOfTokenColumn;
1239     }
1240     if (Shift <= 0)
1241       continue;
1242     Changes[i].Spaces += Shift;
1243     if (i + 1 != Changes.size())
1244       Changes[i + 1].PreviousEndOfTokenColumn += Shift;
1245     Changes[i].StartOfTokenColumn += Shift;
1246   }
1247 }
1248 
alignEscapedNewlines()1249 void WhitespaceManager::alignEscapedNewlines() {
1250   const auto Align = Style.AlignEscapedNewlines;
1251   if (Align == FormatStyle::ENAS_DontAlign)
1252     return;
1253 
1254   const bool WithLastLine = Align == FormatStyle::ENAS_LeftWithLastLine;
1255   const bool AlignLeft = Align == FormatStyle::ENAS_Left || WithLastLine;
1256   const auto MaxColumn = Style.ColumnLimit;
1257   unsigned MaxEndOfLine = AlignLeft ? 0 : MaxColumn;
1258   unsigned StartOfMacro = 0;
1259   for (unsigned i = 1, e = Changes.size(); i < e; ++i) {
1260     Change &C = Changes[i];
1261     if (C.NewlinesBefore == 0 && (!WithLastLine || C.Tok->isNot(tok::eof)))
1262       continue;
1263     const bool InPPDirective = C.ContinuesPPDirective;
1264     const auto BackslashColumn = C.PreviousEndOfTokenColumn + 2;
1265     if (InPPDirective ||
1266         (WithLastLine && (MaxColumn == 0 || BackslashColumn <= MaxColumn))) {
1267       MaxEndOfLine = std::max(BackslashColumn, MaxEndOfLine);
1268     }
1269     if (!InPPDirective) {
1270       alignEscapedNewlines(StartOfMacro + 1, i, MaxEndOfLine);
1271       MaxEndOfLine = AlignLeft ? 0 : MaxColumn;
1272       StartOfMacro = i;
1273     }
1274   }
1275   alignEscapedNewlines(StartOfMacro + 1, Changes.size(), MaxEndOfLine);
1276 }
1277 
alignEscapedNewlines(unsigned Start,unsigned End,unsigned Column)1278 void WhitespaceManager::alignEscapedNewlines(unsigned Start, unsigned End,
1279                                              unsigned Column) {
1280   for (unsigned i = Start; i < End; ++i) {
1281     Change &C = Changes[i];
1282     if (C.NewlinesBefore > 0) {
1283       assert(C.ContinuesPPDirective);
1284       if (C.PreviousEndOfTokenColumn + 1 > Column)
1285         C.EscapedNewlineColumn = 0;
1286       else
1287         C.EscapedNewlineColumn = Column;
1288     }
1289   }
1290 }
1291 
alignArrayInitializers()1292 void WhitespaceManager::alignArrayInitializers() {
1293   if (Style.AlignArrayOfStructures == FormatStyle::AIAS_None)
1294     return;
1295 
1296   for (unsigned ChangeIndex = 1U, ChangeEnd = Changes.size();
1297        ChangeIndex < ChangeEnd; ++ChangeIndex) {
1298     auto &C = Changes[ChangeIndex];
1299     if (C.Tok->IsArrayInitializer) {
1300       bool FoundComplete = false;
1301       for (unsigned InsideIndex = ChangeIndex + 1; InsideIndex < ChangeEnd;
1302            ++InsideIndex) {
1303         if (Changes[InsideIndex].Tok == C.Tok->MatchingParen) {
1304           alignArrayInitializers(ChangeIndex, InsideIndex + 1);
1305           ChangeIndex = InsideIndex + 1;
1306           FoundComplete = true;
1307           break;
1308         }
1309       }
1310       if (!FoundComplete)
1311         ChangeIndex = ChangeEnd;
1312     }
1313   }
1314 }
1315 
alignArrayInitializers(unsigned Start,unsigned End)1316 void WhitespaceManager::alignArrayInitializers(unsigned Start, unsigned End) {
1317 
1318   if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Right)
1319     alignArrayInitializersRightJustified(getCells(Start, End));
1320   else if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Left)
1321     alignArrayInitializersLeftJustified(getCells(Start, End));
1322 }
1323 
alignArrayInitializersRightJustified(CellDescriptions && CellDescs)1324 void WhitespaceManager::alignArrayInitializersRightJustified(
1325     CellDescriptions &&CellDescs) {
1326   if (!CellDescs.isRectangular())
1327     return;
1328 
1329   const int BracePadding = Style.Cpp11BracedListStyle ? 0 : 1;
1330   auto &Cells = CellDescs.Cells;
1331   // Now go through and fixup the spaces.
1332   auto *CellIter = Cells.begin();
1333   for (auto i = 0U; i < CellDescs.CellCounts[0]; ++i, ++CellIter) {
1334     unsigned NetWidth = 0U;
1335     if (isSplitCell(*CellIter))
1336       NetWidth = getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces);
1337     auto CellWidth = getMaximumCellWidth(CellIter, NetWidth);
1338 
1339     if (Changes[CellIter->Index].Tok->is(tok::r_brace)) {
1340       // So in here we want to see if there is a brace that falls
1341       // on a line that was split. If so on that line we make sure that
1342       // the spaces in front of the brace are enough.
1343       const auto *Next = CellIter;
1344       do {
1345         const FormatToken *Previous = Changes[Next->Index].Tok->Previous;
1346         if (Previous && Previous->isNot(TT_LineComment)) {
1347           Changes[Next->Index].Spaces = BracePadding;
1348           Changes[Next->Index].NewlinesBefore = 0;
1349         }
1350         Next = Next->NextColumnElement;
1351       } while (Next);
1352       // Unless the array is empty, we need the position of all the
1353       // immediately adjacent cells
1354       if (CellIter != Cells.begin()) {
1355         auto ThisNetWidth =
1356             getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces);
1357         auto MaxNetWidth = getMaximumNetWidth(
1358             Cells.begin(), CellIter, CellDescs.InitialSpaces,
1359             CellDescs.CellCounts[0], CellDescs.CellCounts.size());
1360         if (ThisNetWidth < MaxNetWidth)
1361           Changes[CellIter->Index].Spaces = (MaxNetWidth - ThisNetWidth);
1362         auto RowCount = 1U;
1363         auto Offset = std::distance(Cells.begin(), CellIter);
1364         for (const auto *Next = CellIter->NextColumnElement; Next;
1365              Next = Next->NextColumnElement) {
1366           if (RowCount >= CellDescs.CellCounts.size())
1367             break;
1368           auto *Start = (Cells.begin() + RowCount * CellDescs.CellCounts[0]);
1369           auto *End = Start + Offset;
1370           ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces);
1371           if (ThisNetWidth < MaxNetWidth)
1372             Changes[Next->Index].Spaces = (MaxNetWidth - ThisNetWidth);
1373           ++RowCount;
1374         }
1375       }
1376     } else {
1377       auto ThisWidth =
1378           calculateCellWidth(CellIter->Index, CellIter->EndIndex, true) +
1379           NetWidth;
1380       if (Changes[CellIter->Index].NewlinesBefore == 0) {
1381         Changes[CellIter->Index].Spaces = (CellWidth - (ThisWidth + NetWidth));
1382         Changes[CellIter->Index].Spaces += (i > 0) ? 1 : BracePadding;
1383       }
1384       alignToStartOfCell(CellIter->Index, CellIter->EndIndex);
1385       for (const auto *Next = CellIter->NextColumnElement; Next;
1386            Next = Next->NextColumnElement) {
1387         ThisWidth =
1388             calculateCellWidth(Next->Index, Next->EndIndex, true) + NetWidth;
1389         if (Changes[Next->Index].NewlinesBefore == 0) {
1390           Changes[Next->Index].Spaces = (CellWidth - ThisWidth);
1391           Changes[Next->Index].Spaces += (i > 0) ? 1 : BracePadding;
1392         }
1393         alignToStartOfCell(Next->Index, Next->EndIndex);
1394       }
1395     }
1396   }
1397 }
1398 
alignArrayInitializersLeftJustified(CellDescriptions && CellDescs)1399 void WhitespaceManager::alignArrayInitializersLeftJustified(
1400     CellDescriptions &&CellDescs) {
1401 
1402   if (!CellDescs.isRectangular())
1403     return;
1404 
1405   const int BracePadding = Style.Cpp11BracedListStyle ? 0 : 1;
1406   auto &Cells = CellDescs.Cells;
1407   // Now go through and fixup the spaces.
1408   auto *CellIter = Cells.begin();
1409   // The first cell of every row needs to be against the left brace.
1410   for (const auto *Next = CellIter; Next; Next = Next->NextColumnElement) {
1411     auto &Change = Changes[Next->Index];
1412     Change.Spaces =
1413         Change.NewlinesBefore == 0 ? BracePadding : CellDescs.InitialSpaces;
1414   }
1415   ++CellIter;
1416   for (auto i = 1U; i < CellDescs.CellCounts[0]; i++, ++CellIter) {
1417     auto MaxNetWidth = getMaximumNetWidth(
1418         Cells.begin(), CellIter, CellDescs.InitialSpaces,
1419         CellDescs.CellCounts[0], CellDescs.CellCounts.size());
1420     auto ThisNetWidth =
1421         getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces);
1422     if (Changes[CellIter->Index].NewlinesBefore == 0) {
1423       Changes[CellIter->Index].Spaces =
1424           MaxNetWidth - ThisNetWidth +
1425           (Changes[CellIter->Index].Tok->isNot(tok::r_brace) ? 1
1426                                                              : BracePadding);
1427     }
1428     auto RowCount = 1U;
1429     auto Offset = std::distance(Cells.begin(), CellIter);
1430     for (const auto *Next = CellIter->NextColumnElement; Next;
1431          Next = Next->NextColumnElement) {
1432       if (RowCount >= CellDescs.CellCounts.size())
1433         break;
1434       auto *Start = (Cells.begin() + RowCount * CellDescs.CellCounts[0]);
1435       auto *End = Start + Offset;
1436       auto ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces);
1437       if (Changes[Next->Index].NewlinesBefore == 0) {
1438         Changes[Next->Index].Spaces =
1439             MaxNetWidth - ThisNetWidth +
1440             (Changes[Next->Index].Tok->isNot(tok::r_brace) ? 1 : BracePadding);
1441       }
1442       ++RowCount;
1443     }
1444   }
1445 }
1446 
isSplitCell(const CellDescription & Cell)1447 bool WhitespaceManager::isSplitCell(const CellDescription &Cell) {
1448   if (Cell.HasSplit)
1449     return true;
1450   for (const auto *Next = Cell.NextColumnElement; Next;
1451        Next = Next->NextColumnElement) {
1452     if (Next->HasSplit)
1453       return true;
1454   }
1455   return false;
1456 }
1457 
getCells(unsigned Start,unsigned End)1458 WhitespaceManager::CellDescriptions WhitespaceManager::getCells(unsigned Start,
1459                                                                 unsigned End) {
1460 
1461   unsigned Depth = 0;
1462   unsigned Cell = 0;
1463   SmallVector<unsigned> CellCounts;
1464   unsigned InitialSpaces = 0;
1465   unsigned InitialTokenLength = 0;
1466   unsigned EndSpaces = 0;
1467   SmallVector<CellDescription> Cells;
1468   const FormatToken *MatchingParen = nullptr;
1469   for (unsigned i = Start; i < End; ++i) {
1470     auto &C = Changes[i];
1471     if (C.Tok->is(tok::l_brace))
1472       ++Depth;
1473     else if (C.Tok->is(tok::r_brace))
1474       --Depth;
1475     if (Depth == 2) {
1476       if (C.Tok->is(tok::l_brace)) {
1477         Cell = 0;
1478         MatchingParen = C.Tok->MatchingParen;
1479         if (InitialSpaces == 0) {
1480           InitialSpaces = C.Spaces + C.TokenLength;
1481           InitialTokenLength = C.TokenLength;
1482           auto j = i - 1;
1483           for (; Changes[j].NewlinesBefore == 0 && j > Start; --j) {
1484             InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength;
1485             InitialTokenLength += Changes[j].TokenLength;
1486           }
1487           if (C.NewlinesBefore == 0) {
1488             InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength;
1489             InitialTokenLength += Changes[j].TokenLength;
1490           }
1491         }
1492       } else if (C.Tok->is(tok::comma)) {
1493         if (!Cells.empty())
1494           Cells.back().EndIndex = i;
1495         if (const auto *Next = C.Tok->getNextNonComment();
1496             Next && Next->isNot(tok::r_brace)) { // dangling comma
1497           ++Cell;
1498         }
1499       }
1500     } else if (Depth == 1) {
1501       if (C.Tok == MatchingParen) {
1502         if (!Cells.empty())
1503           Cells.back().EndIndex = i;
1504         Cells.push_back(CellDescription{i, ++Cell, i + 1, false, nullptr});
1505         CellCounts.push_back(C.Tok->Previous->isNot(tok::comma) ? Cell + 1
1506                                                                 : Cell);
1507         // Go to the next non-comment and ensure there is a break in front
1508         const auto *NextNonComment = C.Tok->getNextNonComment();
1509         while (NextNonComment && NextNonComment->is(tok::comma))
1510           NextNonComment = NextNonComment->getNextNonComment();
1511         auto j = i;
1512         while (j < End && Changes[j].Tok != NextNonComment)
1513           ++j;
1514         if (j < End && Changes[j].NewlinesBefore == 0 &&
1515             Changes[j].Tok->isNot(tok::r_brace)) {
1516           Changes[j].NewlinesBefore = 1;
1517           // Account for the added token lengths
1518           Changes[j].Spaces = InitialSpaces - InitialTokenLength;
1519         }
1520       } else if (C.Tok->is(tok::comment) && C.Tok->NewlinesBefore == 0) {
1521         // Trailing comments stay at a space past the last token
1522         C.Spaces = Changes[i - 1].Tok->is(tok::comma) ? 1 : 2;
1523       } else if (C.Tok->is(tok::l_brace)) {
1524         // We need to make sure that the ending braces is aligned to the
1525         // start of our initializer
1526         auto j = i - 1;
1527         for (; j > 0 && !Changes[j].Tok->ArrayInitializerLineStart; --j)
1528           ; // Nothing the loop does the work
1529         EndSpaces = Changes[j].Spaces;
1530       }
1531     } else if (Depth == 0 && C.Tok->is(tok::r_brace)) {
1532       C.NewlinesBefore = 1;
1533       C.Spaces = EndSpaces;
1534     }
1535     if (C.Tok->StartsColumn) {
1536       // This gets us past tokens that have been split over multiple
1537       // lines
1538       bool HasSplit = false;
1539       if (Changes[i].NewlinesBefore > 0) {
1540         // So if we split a line previously and the tail line + this token is
1541         // less then the column limit we remove the split here and just put
1542         // the column start at a space past the comma
1543         //
1544         // FIXME This if branch covers the cases where the column is not
1545         // the first column. This leads to weird pathologies like the formatting
1546         // auto foo = Items{
1547         //     Section{
1548         //             0, bar(),
1549         //     }
1550         // };
1551         // Well if it doesn't lead to that it's indicative that the line
1552         // breaking should be revisited. Unfortunately alot of other options
1553         // interact with this
1554         auto j = i - 1;
1555         if ((j - 1) > Start && Changes[j].Tok->is(tok::comma) &&
1556             Changes[j - 1].NewlinesBefore > 0) {
1557           --j;
1558           auto LineLimit = Changes[j].Spaces + Changes[j].TokenLength;
1559           if (LineLimit < Style.ColumnLimit) {
1560             Changes[i].NewlinesBefore = 0;
1561             Changes[i].Spaces = 1;
1562           }
1563         }
1564       }
1565       while (Changes[i].NewlinesBefore > 0 && Changes[i].Tok == C.Tok) {
1566         Changes[i].Spaces = InitialSpaces;
1567         ++i;
1568         HasSplit = true;
1569       }
1570       if (Changes[i].Tok != C.Tok)
1571         --i;
1572       Cells.push_back(CellDescription{i, Cell, i, HasSplit, nullptr});
1573     }
1574   }
1575 
1576   return linkCells({Cells, CellCounts, InitialSpaces});
1577 }
1578 
calculateCellWidth(unsigned Start,unsigned End,bool WithSpaces) const1579 unsigned WhitespaceManager::calculateCellWidth(unsigned Start, unsigned End,
1580                                                bool WithSpaces) const {
1581   unsigned CellWidth = 0;
1582   for (auto i = Start; i < End; i++) {
1583     if (Changes[i].NewlinesBefore > 0)
1584       CellWidth = 0;
1585     CellWidth += Changes[i].TokenLength;
1586     CellWidth += (WithSpaces ? Changes[i].Spaces : 0);
1587   }
1588   return CellWidth;
1589 }
1590 
alignToStartOfCell(unsigned Start,unsigned End)1591 void WhitespaceManager::alignToStartOfCell(unsigned Start, unsigned End) {
1592   if ((End - Start) <= 1)
1593     return;
1594   // If the line is broken anywhere in there make sure everything
1595   // is aligned to the parent
1596   for (auto i = Start + 1; i < End; i++)
1597     if (Changes[i].NewlinesBefore > 0)
1598       Changes[i].Spaces = Changes[Start].Spaces;
1599 }
1600 
1601 WhitespaceManager::CellDescriptions
linkCells(CellDescriptions && CellDesc)1602 WhitespaceManager::linkCells(CellDescriptions &&CellDesc) {
1603   auto &Cells = CellDesc.Cells;
1604   for (auto *CellIter = Cells.begin(); CellIter != Cells.end(); ++CellIter) {
1605     if (!CellIter->NextColumnElement && (CellIter + 1) != Cells.end()) {
1606       for (auto *NextIter = CellIter + 1; NextIter != Cells.end(); ++NextIter) {
1607         if (NextIter->Cell == CellIter->Cell) {
1608           CellIter->NextColumnElement = &(*NextIter);
1609           break;
1610         }
1611       }
1612     }
1613   }
1614   return std::move(CellDesc);
1615 }
1616 
generateChanges()1617 void WhitespaceManager::generateChanges() {
1618   for (unsigned i = 0, e = Changes.size(); i != e; ++i) {
1619     const Change &C = Changes[i];
1620     if (i > 0) {
1621       auto Last = Changes[i - 1].OriginalWhitespaceRange;
1622       auto New = Changes[i].OriginalWhitespaceRange;
1623       // Do not generate two replacements for the same location.  As a special
1624       // case, it is allowed if there is a replacement for the empty range
1625       // between 2 tokens and another non-empty range at the start of the second
1626       // token.  We didn't implement logic to combine replacements for 2
1627       // consecutive source ranges into a single replacement, because the
1628       // program works fine without it.
1629       //
1630       // We can't eliminate empty original whitespace ranges.  They appear when
1631       // 2 tokens have no whitespace in between in the input.  It does not
1632       // matter whether whitespace is to be added.  If no whitespace is to be
1633       // added, the replacement will be empty, and it gets eliminated after this
1634       // step in storeReplacement.  For example, if the input is `foo();`,
1635       // there will be a replacement for the range between every consecutive
1636       // pair of tokens.
1637       //
1638       // A replacement at the start of a token can be added by
1639       // BreakableStringLiteralUsingOperators::insertBreak when it adds braces
1640       // around the string literal.  Say Verilog code is being formatted and the
1641       // first line is to become the next 2 lines.
1642       //     x("long string");
1643       //     x({"long ",
1644       //        "string"});
1645       // There will be a replacement for the empty range between the parenthesis
1646       // and the string and another replacement for the quote character.  The
1647       // replacement for the empty range between the parenthesis and the quote
1648       // comes from ContinuationIndenter::addTokenOnCurrentLine when it changes
1649       // the original empty range between the parenthesis and the string to
1650       // another empty one.  The replacement for the quote character comes from
1651       // BreakableStringLiteralUsingOperators::insertBreak when it adds the
1652       // brace.  In the example, the replacement for the empty range is the same
1653       // as the original text.  However, eliminating replacements that are same
1654       // as the original does not help in general.  For example, a newline can
1655       // be inserted, causing the first line to become the next 3 lines.
1656       //     xxxxxxxxxxx("long string");
1657       //     xxxxxxxxxxx(
1658       //         {"long ",
1659       //          "string"});
1660       // In that case, the empty range between the parenthesis and the string
1661       // will be replaced by a newline and 4 spaces.  So we will still have to
1662       // deal with a replacement for an empty source range followed by a
1663       // replacement for a non-empty source range.
1664       if (Last.getBegin() == New.getBegin() &&
1665           (Last.getEnd() != Last.getBegin() ||
1666            New.getEnd() == New.getBegin())) {
1667         continue;
1668       }
1669     }
1670     if (C.CreateReplacement) {
1671       std::string ReplacementText = C.PreviousLinePostfix;
1672       if (C.ContinuesPPDirective) {
1673         appendEscapedNewlineText(ReplacementText, C.NewlinesBefore,
1674                                  C.PreviousEndOfTokenColumn,
1675                                  C.EscapedNewlineColumn);
1676       } else {
1677         appendNewlineText(ReplacementText, C.NewlinesBefore);
1678       }
1679       // FIXME: This assert should hold if we computed the column correctly.
1680       // assert((int)C.StartOfTokenColumn >= C.Spaces);
1681       appendIndentText(
1682           ReplacementText, C.Tok->IndentLevel, std::max(0, C.Spaces),
1683           std::max((int)C.StartOfTokenColumn, C.Spaces) - std::max(0, C.Spaces),
1684           C.IsAligned);
1685       ReplacementText.append(C.CurrentLinePrefix);
1686       storeReplacement(C.OriginalWhitespaceRange, ReplacementText);
1687     }
1688   }
1689 }
1690 
storeReplacement(SourceRange Range,StringRef Text)1691 void WhitespaceManager::storeReplacement(SourceRange Range, StringRef Text) {
1692   unsigned WhitespaceLength = SourceMgr.getFileOffset(Range.getEnd()) -
1693                               SourceMgr.getFileOffset(Range.getBegin());
1694   // Don't create a replacement, if it does not change anything.
1695   if (StringRef(SourceMgr.getCharacterData(Range.getBegin()),
1696                 WhitespaceLength) == Text) {
1697     return;
1698   }
1699   auto Err = Replaces.add(tooling::Replacement(
1700       SourceMgr, CharSourceRange::getCharRange(Range), Text));
1701   // FIXME: better error handling. For now, just print an error message in the
1702   // release version.
1703   if (Err) {
1704     llvm::errs() << llvm::toString(std::move(Err)) << "\n";
1705     assert(false);
1706   }
1707 }
1708 
appendNewlineText(std::string & Text,unsigned Newlines)1709 void WhitespaceManager::appendNewlineText(std::string &Text,
1710                                           unsigned Newlines) {
1711   if (UseCRLF) {
1712     Text.reserve(Text.size() + 2 * Newlines);
1713     for (unsigned i = 0; i < Newlines; ++i)
1714       Text.append("\r\n");
1715   } else {
1716     Text.append(Newlines, '\n');
1717   }
1718 }
1719 
appendEscapedNewlineText(std::string & Text,unsigned Newlines,unsigned PreviousEndOfTokenColumn,unsigned EscapedNewlineColumn)1720 void WhitespaceManager::appendEscapedNewlineText(
1721     std::string &Text, unsigned Newlines, unsigned PreviousEndOfTokenColumn,
1722     unsigned EscapedNewlineColumn) {
1723   if (Newlines > 0) {
1724     unsigned Spaces =
1725         std::max<int>(1, EscapedNewlineColumn - PreviousEndOfTokenColumn - 1);
1726     for (unsigned i = 0; i < Newlines; ++i) {
1727       Text.append(Spaces, ' ');
1728       Text.append(UseCRLF ? "\\\r\n" : "\\\n");
1729       Spaces = std::max<int>(0, EscapedNewlineColumn - 1);
1730     }
1731   }
1732 }
1733 
appendIndentText(std::string & Text,unsigned IndentLevel,unsigned Spaces,unsigned WhitespaceStartColumn,bool IsAligned)1734 void WhitespaceManager::appendIndentText(std::string &Text,
1735                                          unsigned IndentLevel, unsigned Spaces,
1736                                          unsigned WhitespaceStartColumn,
1737                                          bool IsAligned) {
1738   switch (Style.UseTab) {
1739   case FormatStyle::UT_Never:
1740     Text.append(Spaces, ' ');
1741     break;
1742   case FormatStyle::UT_Always: {
1743     if (Style.TabWidth) {
1744       unsigned FirstTabWidth =
1745           Style.TabWidth - WhitespaceStartColumn % Style.TabWidth;
1746 
1747       // Insert only spaces when we want to end up before the next tab.
1748       if (Spaces < FirstTabWidth || Spaces == 1) {
1749         Text.append(Spaces, ' ');
1750         break;
1751       }
1752       // Align to the next tab.
1753       Spaces -= FirstTabWidth;
1754       Text.append("\t");
1755 
1756       Text.append(Spaces / Style.TabWidth, '\t');
1757       Text.append(Spaces % Style.TabWidth, ' ');
1758     } else if (Spaces == 1) {
1759       Text.append(Spaces, ' ');
1760     }
1761     break;
1762   }
1763   case FormatStyle::UT_ForIndentation:
1764     if (WhitespaceStartColumn == 0) {
1765       unsigned Indentation = IndentLevel * Style.IndentWidth;
1766       Spaces = appendTabIndent(Text, Spaces, Indentation);
1767     }
1768     Text.append(Spaces, ' ');
1769     break;
1770   case FormatStyle::UT_ForContinuationAndIndentation:
1771     if (WhitespaceStartColumn == 0)
1772       Spaces = appendTabIndent(Text, Spaces, Spaces);
1773     Text.append(Spaces, ' ');
1774     break;
1775   case FormatStyle::UT_AlignWithSpaces:
1776     if (WhitespaceStartColumn == 0) {
1777       unsigned Indentation =
1778           IsAligned ? IndentLevel * Style.IndentWidth : Spaces;
1779       Spaces = appendTabIndent(Text, Spaces, Indentation);
1780     }
1781     Text.append(Spaces, ' ');
1782     break;
1783   }
1784 }
1785 
appendTabIndent(std::string & Text,unsigned Spaces,unsigned Indentation)1786 unsigned WhitespaceManager::appendTabIndent(std::string &Text, unsigned Spaces,
1787                                             unsigned Indentation) {
1788   // This happens, e.g. when a line in a block comment is indented less than the
1789   // first one.
1790   if (Indentation > Spaces)
1791     Indentation = Spaces;
1792   if (Style.TabWidth) {
1793     unsigned Tabs = Indentation / Style.TabWidth;
1794     Text.append(Tabs, '\t');
1795     Spaces -= Tabs * Style.TabWidth;
1796   }
1797   return Spaces;
1798 }
1799 
1800 } // namespace format
1801 } // namespace clang
1802