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
17 namespace clang {
18 namespace format {
19
operator ()(const Change & C1,const Change & C2) const20 bool WhitespaceManager::Change::IsBeforeInFile::operator()(
21 const Change &C1, const Change &C2) const {
22 return SourceMgr.isBeforeInTranslationUnit(
23 C1.OriginalWhitespaceRange.getBegin(),
24 C2.OriginalWhitespaceRange.getBegin());
25 }
26
Change(const FormatToken & Tok,bool CreateReplacement,SourceRange OriginalWhitespaceRange,int Spaces,unsigned StartOfTokenColumn,unsigned NewlinesBefore,StringRef PreviousLinePostfix,StringRef CurrentLinePrefix,bool IsAligned,bool ContinuesPPDirective,bool IsInsideToken)27 WhitespaceManager::Change::Change(const FormatToken &Tok,
28 bool CreateReplacement,
29 SourceRange OriginalWhitespaceRange,
30 int Spaces, unsigned StartOfTokenColumn,
31 unsigned NewlinesBefore,
32 StringRef PreviousLinePostfix,
33 StringRef CurrentLinePrefix, bool IsAligned,
34 bool ContinuesPPDirective, bool IsInsideToken)
35 : Tok(&Tok), CreateReplacement(CreateReplacement),
36 OriginalWhitespaceRange(OriginalWhitespaceRange),
37 StartOfTokenColumn(StartOfTokenColumn), NewlinesBefore(NewlinesBefore),
38 PreviousLinePostfix(PreviousLinePostfix),
39 CurrentLinePrefix(CurrentLinePrefix), IsAligned(IsAligned),
40 ContinuesPPDirective(ContinuesPPDirective), Spaces(Spaces),
41 IsInsideToken(IsInsideToken), IsTrailingComment(false), TokenLength(0),
42 PreviousEndOfTokenColumn(0), EscapedNewlineColumn(0),
43 StartOfBlockComment(nullptr), IndentationOffset(0), ConditionalsLevel(0) {
44 }
45
replaceWhitespace(FormatToken & Tok,unsigned Newlines,unsigned Spaces,unsigned StartOfTokenColumn,bool IsAligned,bool InPPDirective)46 void WhitespaceManager::replaceWhitespace(FormatToken &Tok, unsigned Newlines,
47 unsigned Spaces,
48 unsigned StartOfTokenColumn,
49 bool IsAligned, bool InPPDirective) {
50 if (Tok.Finalized)
51 return;
52 Tok.setDecision((Newlines > 0) ? FD_Break : FD_Continue);
53 Changes.push_back(Change(Tok, /*CreateReplacement=*/true, Tok.WhitespaceRange,
54 Spaces, StartOfTokenColumn, Newlines, "", "",
55 IsAligned, InPPDirective && !Tok.IsFirst,
56 /*IsInsideToken=*/false));
57 }
58
addUntouchableToken(const FormatToken & Tok,bool InPPDirective)59 void WhitespaceManager::addUntouchableToken(const FormatToken &Tok,
60 bool InPPDirective) {
61 if (Tok.Finalized)
62 return;
63 Changes.push_back(Change(Tok, /*CreateReplacement=*/false,
64 Tok.WhitespaceRange, /*Spaces=*/0,
65 Tok.OriginalColumn, Tok.NewlinesBefore, "", "",
66 /*IsAligned=*/false, InPPDirective && !Tok.IsFirst,
67 /*IsInsideToken=*/false));
68 }
69
70 llvm::Error
addReplacement(const tooling::Replacement & Replacement)71 WhitespaceManager::addReplacement(const tooling::Replacement &Replacement) {
72 return Replaces.add(Replacement);
73 }
74
replaceWhitespaceInToken(const FormatToken & Tok,unsigned Offset,unsigned ReplaceChars,StringRef PreviousPostfix,StringRef CurrentPrefix,bool InPPDirective,unsigned Newlines,int Spaces)75 void WhitespaceManager::replaceWhitespaceInToken(
76 const FormatToken &Tok, unsigned Offset, unsigned ReplaceChars,
77 StringRef PreviousPostfix, StringRef CurrentPrefix, bool InPPDirective,
78 unsigned Newlines, int Spaces) {
79 if (Tok.Finalized)
80 return;
81 SourceLocation Start = Tok.getStartOfNonWhitespace().getLocWithOffset(Offset);
82 Changes.push_back(
83 Change(Tok, /*CreateReplacement=*/true,
84 SourceRange(Start, Start.getLocWithOffset(ReplaceChars)), Spaces,
85 std::max(0, Spaces), Newlines, PreviousPostfix, CurrentPrefix,
86 /*IsAligned=*/true, InPPDirective && !Tok.IsFirst,
87 /*IsInsideToken=*/true));
88 }
89
generateReplacements()90 const tooling::Replacements &WhitespaceManager::generateReplacements() {
91 if (Changes.empty())
92 return Replaces;
93
94 llvm::sort(Changes, Change::IsBeforeInFile(SourceMgr));
95 calculateLineBreakInformation();
96 alignConsecutiveMacros();
97 alignConsecutiveDeclarations();
98 alignConsecutiveBitFields();
99 alignConsecutiveAssignments();
100 alignChainedConditionals();
101 alignTrailingComments();
102 alignEscapedNewlines();
103 generateChanges();
104
105 return Replaces;
106 }
107
calculateLineBreakInformation()108 void WhitespaceManager::calculateLineBreakInformation() {
109 Changes[0].PreviousEndOfTokenColumn = 0;
110 Change *LastOutsideTokenChange = &Changes[0];
111 for (unsigned i = 1, e = Changes.size(); i != e; ++i) {
112 SourceLocation OriginalWhitespaceStart =
113 Changes[i].OriginalWhitespaceRange.getBegin();
114 SourceLocation PreviousOriginalWhitespaceEnd =
115 Changes[i - 1].OriginalWhitespaceRange.getEnd();
116 unsigned OriginalWhitespaceStartOffset =
117 SourceMgr.getFileOffset(OriginalWhitespaceStart);
118 unsigned PreviousOriginalWhitespaceEndOffset =
119 SourceMgr.getFileOffset(PreviousOriginalWhitespaceEnd);
120 assert(PreviousOriginalWhitespaceEndOffset <=
121 OriginalWhitespaceStartOffset);
122 const char *const PreviousOriginalWhitespaceEndData =
123 SourceMgr.getCharacterData(PreviousOriginalWhitespaceEnd);
124 StringRef Text(PreviousOriginalWhitespaceEndData,
125 SourceMgr.getCharacterData(OriginalWhitespaceStart) -
126 PreviousOriginalWhitespaceEndData);
127 // Usually consecutive changes would occur in consecutive tokens. This is
128 // not the case however when analyzing some preprocessor runs of the
129 // annotated lines. For example, in this code:
130 //
131 // #if A // line 1
132 // int i = 1;
133 // #else B // line 2
134 // int i = 2;
135 // #endif // line 3
136 //
137 // one of the runs will produce the sequence of lines marked with line 1, 2
138 // and 3. So the two consecutive whitespace changes just before '// line 2'
139 // and before '#endif // line 3' span multiple lines and tokens:
140 //
141 // #else B{change X}[// line 2
142 // int i = 2;
143 // ]{change Y}#endif // line 3
144 //
145 // For this reason, if the text between consecutive changes spans multiple
146 // newlines, the token length must be adjusted to the end of the original
147 // line of the token.
148 auto NewlinePos = Text.find_first_of('\n');
149 if (NewlinePos == StringRef::npos) {
150 Changes[i - 1].TokenLength = OriginalWhitespaceStartOffset -
151 PreviousOriginalWhitespaceEndOffset +
152 Changes[i].PreviousLinePostfix.size() +
153 Changes[i - 1].CurrentLinePrefix.size();
154 } else {
155 Changes[i - 1].TokenLength =
156 NewlinePos + Changes[i - 1].CurrentLinePrefix.size();
157 }
158
159 // If there are multiple changes in this token, sum up all the changes until
160 // the end of the line.
161 if (Changes[i - 1].IsInsideToken && Changes[i - 1].NewlinesBefore == 0)
162 LastOutsideTokenChange->TokenLength +=
163 Changes[i - 1].TokenLength + Changes[i - 1].Spaces;
164 else
165 LastOutsideTokenChange = &Changes[i - 1];
166
167 Changes[i].PreviousEndOfTokenColumn =
168 Changes[i - 1].StartOfTokenColumn + Changes[i - 1].TokenLength;
169
170 Changes[i - 1].IsTrailingComment =
171 (Changes[i].NewlinesBefore > 0 || Changes[i].Tok->is(tok::eof) ||
172 (Changes[i].IsInsideToken && Changes[i].Tok->is(tok::comment))) &&
173 Changes[i - 1].Tok->is(tok::comment) &&
174 // FIXME: This is a dirty hack. The problem is that
175 // BreakableLineCommentSection does comment reflow changes and here is
176 // the aligning of trailing comments. Consider the case where we reflow
177 // the second line up in this example:
178 //
179 // // line 1
180 // // line 2
181 //
182 // That amounts to 2 changes by BreakableLineCommentSection:
183 // - the first, delimited by (), for the whitespace between the tokens,
184 // - and second, delimited by [], for the whitespace at the beginning
185 // of the second token:
186 //
187 // // line 1(
188 // )[// ]line 2
189 //
190 // So in the end we have two changes like this:
191 //
192 // // line1()[ ]line 2
193 //
194 // Note that the OriginalWhitespaceStart of the second change is the
195 // same as the PreviousOriginalWhitespaceEnd of the first change.
196 // In this case, the below check ensures that the second change doesn't
197 // get treated as a trailing comment change here, since this might
198 // trigger additional whitespace to be wrongly inserted before "line 2"
199 // by the comment aligner here.
200 //
201 // For a proper solution we need a mechanism to say to WhitespaceManager
202 // that a particular change breaks the current sequence of trailing
203 // comments.
204 OriginalWhitespaceStart != PreviousOriginalWhitespaceEnd;
205 }
206 // FIXME: The last token is currently not always an eof token; in those
207 // cases, setting TokenLength of the last token to 0 is wrong.
208 Changes.back().TokenLength = 0;
209 Changes.back().IsTrailingComment = Changes.back().Tok->is(tok::comment);
210
211 const WhitespaceManager::Change *LastBlockComment = nullptr;
212 for (auto &Change : Changes) {
213 // Reset the IsTrailingComment flag for changes inside of trailing comments
214 // so they don't get realigned later. Comment line breaks however still need
215 // to be aligned.
216 if (Change.IsInsideToken && Change.NewlinesBefore == 0)
217 Change.IsTrailingComment = false;
218 Change.StartOfBlockComment = nullptr;
219 Change.IndentationOffset = 0;
220 if (Change.Tok->is(tok::comment)) {
221 if (Change.Tok->is(TT_LineComment) || !Change.IsInsideToken)
222 LastBlockComment = &Change;
223 else {
224 if ((Change.StartOfBlockComment = LastBlockComment))
225 Change.IndentationOffset =
226 Change.StartOfTokenColumn -
227 Change.StartOfBlockComment->StartOfTokenColumn;
228 }
229 } else {
230 LastBlockComment = nullptr;
231 }
232 }
233
234 // Compute conditional nesting level
235 // Level is increased for each conditional, unless this conditional continues
236 // a chain of conditional, i.e. starts immediately after the colon of another
237 // conditional.
238 SmallVector<bool, 16> ScopeStack;
239 int ConditionalsLevel = 0;
240 for (auto &Change : Changes) {
241 for (unsigned i = 0, e = Change.Tok->FakeLParens.size(); i != e; ++i) {
242 bool isNestedConditional =
243 Change.Tok->FakeLParens[e - 1 - i] == prec::Conditional &&
244 !(i == 0 && Change.Tok->Previous &&
245 Change.Tok->Previous->is(TT_ConditionalExpr) &&
246 Change.Tok->Previous->is(tok::colon));
247 if (isNestedConditional)
248 ++ConditionalsLevel;
249 ScopeStack.push_back(isNestedConditional);
250 }
251
252 Change.ConditionalsLevel = ConditionalsLevel;
253
254 for (unsigned i = Change.Tok->FakeRParens; i > 0 && ScopeStack.size();
255 --i) {
256 if (ScopeStack.pop_back_val())
257 --ConditionalsLevel;
258 }
259 }
260 }
261
262 // Align a single sequence of tokens, see AlignTokens below.
263 template <typename F>
264 static void
AlignTokenSequence(unsigned Start,unsigned End,unsigned Column,F && Matches,SmallVector<WhitespaceManager::Change,16> & Changes)265 AlignTokenSequence(unsigned Start, unsigned End, unsigned Column, F &&Matches,
266 SmallVector<WhitespaceManager::Change, 16> &Changes) {
267 bool FoundMatchOnLine = false;
268 int Shift = 0;
269
270 // ScopeStack keeps track of the current scope depth. It contains indices of
271 // the first token on each scope.
272 // We only run the "Matches" function on tokens from the outer-most scope.
273 // However, we do need to pay special attention to one class of tokens
274 // that are not in the outer-most scope, and that is function parameters
275 // which are split across multiple lines, as illustrated by this example:
276 // double a(int x);
277 // int b(int y,
278 // double z);
279 // In the above example, we need to take special care to ensure that
280 // 'double z' is indented along with it's owning function 'b'.
281 // Special handling is required for 'nested' ternary operators.
282 SmallVector<unsigned, 16> ScopeStack;
283
284 for (unsigned i = Start; i != End; ++i) {
285 if (ScopeStack.size() != 0 &&
286 Changes[i].indentAndNestingLevel() <
287 Changes[ScopeStack.back()].indentAndNestingLevel())
288 ScopeStack.pop_back();
289
290 // Compare current token to previous non-comment token to ensure whether
291 // it is in a deeper scope or not.
292 unsigned PreviousNonComment = i - 1;
293 while (PreviousNonComment > Start &&
294 Changes[PreviousNonComment].Tok->is(tok::comment))
295 PreviousNonComment--;
296 if (i != Start && Changes[i].indentAndNestingLevel() >
297 Changes[PreviousNonComment].indentAndNestingLevel())
298 ScopeStack.push_back(i);
299
300 bool InsideNestedScope = ScopeStack.size() != 0;
301
302 if (Changes[i].NewlinesBefore > 0 && !InsideNestedScope) {
303 Shift = 0;
304 FoundMatchOnLine = false;
305 }
306
307 // If this is the first matching token to be aligned, remember by how many
308 // spaces it has to be shifted, so the rest of the changes on the line are
309 // shifted by the same amount
310 if (!FoundMatchOnLine && !InsideNestedScope && Matches(Changes[i])) {
311 FoundMatchOnLine = true;
312 Shift = Column - Changes[i].StartOfTokenColumn;
313 Changes[i].Spaces += Shift;
314 }
315
316 // This is for function parameters that are split across multiple lines,
317 // as mentioned in the ScopeStack comment.
318 if (InsideNestedScope && Changes[i].NewlinesBefore > 0) {
319 unsigned ScopeStart = ScopeStack.back();
320 if (Changes[ScopeStart - 1].Tok->is(TT_FunctionDeclarationName) ||
321 (ScopeStart > Start + 1 &&
322 Changes[ScopeStart - 2].Tok->is(TT_FunctionDeclarationName)) ||
323 Changes[i].Tok->is(TT_ConditionalExpr) ||
324 (Changes[i].Tok->Previous &&
325 Changes[i].Tok->Previous->is(TT_ConditionalExpr)))
326 Changes[i].Spaces += Shift;
327 }
328
329 assert(Shift >= 0);
330 Changes[i].StartOfTokenColumn += Shift;
331 if (i + 1 != Changes.size())
332 Changes[i + 1].PreviousEndOfTokenColumn += Shift;
333 }
334 }
335
336 // Walk through a subset of the changes, starting at StartAt, and find
337 // sequences of matching tokens to align. To do so, keep track of the lines and
338 // whether or not a matching token was found on a line. If a matching token is
339 // found, extend the current sequence. If the current line cannot be part of a
340 // sequence, e.g. because there is an empty line before it or it contains only
341 // non-matching tokens, finalize the previous sequence.
342 // The value returned is the token on which we stopped, either because we
343 // exhausted all items inside Changes, or because we hit a scope level higher
344 // than our initial scope.
345 // This function is recursive. Each invocation processes only the scope level
346 // equal to the initial level, which is the level of Changes[StartAt].
347 // If we encounter a scope level greater than the initial level, then we call
348 // ourselves recursively, thereby avoiding the pollution of the current state
349 // with the alignment requirements of the nested sub-level. This recursive
350 // behavior is necessary for aligning function prototypes that have one or more
351 // arguments.
352 // If this function encounters a scope level less than the initial level,
353 // it returns the current position.
354 // There is a non-obvious subtlety in the recursive behavior: Even though we
355 // defer processing of nested levels to recursive invocations of this
356 // function, when it comes time to align a sequence of tokens, we run the
357 // alignment on the entire sequence, including the nested levels.
358 // When doing so, most of the nested tokens are skipped, because their
359 // alignment was already handled by the recursive invocations of this function.
360 // However, the special exception is that we do NOT skip function parameters
361 // that are split across multiple lines. See the test case in FormatTest.cpp
362 // that mentions "split function parameter alignment" for an example of this.
363 template <typename F>
AlignTokens(const FormatStyle & Style,F && Matches,SmallVector<WhitespaceManager::Change,16> & Changes,unsigned StartAt)364 static unsigned AlignTokens(const FormatStyle &Style, F &&Matches,
365 SmallVector<WhitespaceManager::Change, 16> &Changes,
366 unsigned StartAt) {
367 unsigned MinColumn = 0;
368 unsigned MaxColumn = UINT_MAX;
369
370 // Line number of the start and the end of the current token sequence.
371 unsigned StartOfSequence = 0;
372 unsigned EndOfSequence = 0;
373
374 // Measure the scope level (i.e. depth of (), [], {}) of the first token, and
375 // abort when we hit any token in a higher scope than the starting one.
376 auto IndentAndNestingLevel = StartAt < Changes.size()
377 ? Changes[StartAt].indentAndNestingLevel()
378 : std::tuple<unsigned, unsigned, unsigned>();
379
380 // Keep track of the number of commas before the matching tokens, we will only
381 // align a sequence of matching tokens if they are preceded by the same number
382 // of commas.
383 unsigned CommasBeforeLastMatch = 0;
384 unsigned CommasBeforeMatch = 0;
385
386 // Whether a matching token has been found on the current line.
387 bool FoundMatchOnLine = false;
388
389 // Aligns a sequence of matching tokens, on the MinColumn column.
390 //
391 // Sequences start from the first matching token to align, and end at the
392 // first token of the first line that doesn't need to be aligned.
393 //
394 // We need to adjust the StartOfTokenColumn of each Change that is on a line
395 // containing any matching token to be aligned and located after such token.
396 auto AlignCurrentSequence = [&] {
397 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence)
398 AlignTokenSequence(StartOfSequence, EndOfSequence, MinColumn, Matches,
399 Changes);
400 MinColumn = 0;
401 MaxColumn = UINT_MAX;
402 StartOfSequence = 0;
403 EndOfSequence = 0;
404 };
405
406 unsigned i = StartAt;
407 for (unsigned e = Changes.size(); i != e; ++i) {
408 if (Changes[i].indentAndNestingLevel() < IndentAndNestingLevel)
409 break;
410
411 if (Changes[i].NewlinesBefore != 0) {
412 CommasBeforeMatch = 0;
413 EndOfSequence = i;
414 // If there is a blank line, there is a forced-align-break (eg,
415 // preprocessor), or if the last line didn't contain any matching token,
416 // the sequence ends here.
417 if (Changes[i].NewlinesBefore > 1 ||
418 Changes[i].Tok->MustBreakAlignBefore || !FoundMatchOnLine)
419 AlignCurrentSequence();
420
421 FoundMatchOnLine = false;
422 }
423
424 if (Changes[i].Tok->is(tok::comma)) {
425 ++CommasBeforeMatch;
426 } else if (Changes[i].indentAndNestingLevel() > IndentAndNestingLevel) {
427 // Call AlignTokens recursively, skipping over this scope block.
428 unsigned StoppedAt = AlignTokens(Style, Matches, Changes, i);
429 i = StoppedAt - 1;
430 continue;
431 }
432
433 if (!Matches(Changes[i]))
434 continue;
435
436 // If there is more than one matching token per line, or if the number of
437 // preceding commas, do not match anymore, end the sequence.
438 if (FoundMatchOnLine || CommasBeforeMatch != CommasBeforeLastMatch)
439 AlignCurrentSequence();
440
441 CommasBeforeLastMatch = CommasBeforeMatch;
442 FoundMatchOnLine = true;
443
444 if (StartOfSequence == 0)
445 StartOfSequence = i;
446
447 unsigned ChangeMinColumn = Changes[i].StartOfTokenColumn;
448 int LineLengthAfter = Changes[i].TokenLength;
449 for (unsigned j = i + 1; j != e && Changes[j].NewlinesBefore == 0; ++j) {
450 LineLengthAfter += Changes[j].Spaces;
451 // Changes are generally 1:1 with the tokens, but a change could also be
452 // inside of a token, in which case it's counted more than once: once for
453 // the whitespace surrounding the token (!IsInsideToken) and once for
454 // each whitespace change within it (IsInsideToken).
455 // Therefore, changes inside of a token should only count the space.
456 if (!Changes[j].IsInsideToken)
457 LineLengthAfter += Changes[j].TokenLength;
458 }
459 unsigned ChangeMaxColumn = Style.ColumnLimit - LineLengthAfter;
460
461 // If we are restricted by the maximum column width, end the sequence.
462 if (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn ||
463 CommasBeforeLastMatch != CommasBeforeMatch) {
464 AlignCurrentSequence();
465 StartOfSequence = i;
466 }
467
468 MinColumn = std::max(MinColumn, ChangeMinColumn);
469 MaxColumn = std::min(MaxColumn, ChangeMaxColumn);
470 }
471
472 EndOfSequence = i;
473 AlignCurrentSequence();
474 return i;
475 }
476
477 // Aligns a sequence of matching tokens, on the MinColumn column.
478 //
479 // Sequences start from the first matching token to align, and end at the
480 // first token of the first line that doesn't need to be aligned.
481 //
482 // We need to adjust the StartOfTokenColumn of each Change that is on a line
483 // containing any matching token to be aligned and located after such token.
AlignMacroSequence(unsigned & StartOfSequence,unsigned & EndOfSequence,unsigned & MinColumn,unsigned & MaxColumn,bool & FoundMatchOnLine,std::function<bool (const WhitespaceManager::Change & C)> AlignMacrosMatches,SmallVector<WhitespaceManager::Change,16> & Changes)484 static void AlignMacroSequence(
485 unsigned &StartOfSequence, unsigned &EndOfSequence, unsigned &MinColumn,
486 unsigned &MaxColumn, bool &FoundMatchOnLine,
487 std::function<bool(const WhitespaceManager::Change &C)> AlignMacrosMatches,
488 SmallVector<WhitespaceManager::Change, 16> &Changes) {
489 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) {
490
491 FoundMatchOnLine = false;
492 int Shift = 0;
493
494 for (unsigned I = StartOfSequence; I != EndOfSequence; ++I) {
495 if (Changes[I].NewlinesBefore > 0) {
496 Shift = 0;
497 FoundMatchOnLine = false;
498 }
499
500 // If this is the first matching token to be aligned, remember by how many
501 // spaces it has to be shifted, so the rest of the changes on the line are
502 // shifted by the same amount
503 if (!FoundMatchOnLine && AlignMacrosMatches(Changes[I])) {
504 FoundMatchOnLine = true;
505 Shift = MinColumn - Changes[I].StartOfTokenColumn;
506 Changes[I].Spaces += Shift;
507 }
508
509 assert(Shift >= 0);
510 Changes[I].StartOfTokenColumn += Shift;
511 if (I + 1 != Changes.size())
512 Changes[I + 1].PreviousEndOfTokenColumn += Shift;
513 }
514 }
515
516 MinColumn = 0;
517 MaxColumn = UINT_MAX;
518 StartOfSequence = 0;
519 EndOfSequence = 0;
520 }
521
alignConsecutiveMacros()522 void WhitespaceManager::alignConsecutiveMacros() {
523 if (!Style.AlignConsecutiveMacros)
524 return;
525
526 auto AlignMacrosMatches = [](const Change &C) {
527 const FormatToken *Current = C.Tok;
528 unsigned SpacesRequiredBefore = 1;
529
530 if (Current->SpacesRequiredBefore == 0 || !Current->Previous)
531 return false;
532
533 Current = Current->Previous;
534
535 // If token is a ")", skip over the parameter list, to the
536 // token that precedes the "("
537 if (Current->is(tok::r_paren) && Current->MatchingParen) {
538 Current = Current->MatchingParen->Previous;
539 SpacesRequiredBefore = 0;
540 }
541
542 if (!Current || !Current->is(tok::identifier))
543 return false;
544
545 if (!Current->Previous || !Current->Previous->is(tok::pp_define))
546 return false;
547
548 // For a macro function, 0 spaces are required between the
549 // identifier and the lparen that opens the parameter list.
550 // For a simple macro, 1 space is required between the
551 // identifier and the first token of the defined value.
552 return Current->Next->SpacesRequiredBefore == SpacesRequiredBefore;
553 };
554
555 unsigned MinColumn = 0;
556 unsigned MaxColumn = UINT_MAX;
557
558 // Start and end of the token sequence we're processing.
559 unsigned StartOfSequence = 0;
560 unsigned EndOfSequence = 0;
561
562 // Whether a matching token has been found on the current line.
563 bool FoundMatchOnLine = false;
564
565 unsigned I = 0;
566 for (unsigned E = Changes.size(); I != E; ++I) {
567 if (Changes[I].NewlinesBefore != 0) {
568 EndOfSequence = I;
569 // If there is a blank line, or if the last line didn't contain any
570 // matching token, the sequence ends here.
571 if (Changes[I].NewlinesBefore > 1 || !FoundMatchOnLine)
572 AlignMacroSequence(StartOfSequence, EndOfSequence, MinColumn, MaxColumn,
573 FoundMatchOnLine, AlignMacrosMatches, Changes);
574
575 FoundMatchOnLine = false;
576 }
577
578 if (!AlignMacrosMatches(Changes[I]))
579 continue;
580
581 FoundMatchOnLine = true;
582
583 if (StartOfSequence == 0)
584 StartOfSequence = I;
585
586 unsigned ChangeMinColumn = Changes[I].StartOfTokenColumn;
587 int LineLengthAfter = -Changes[I].Spaces;
588 for (unsigned j = I; j != E && Changes[j].NewlinesBefore == 0; ++j)
589 LineLengthAfter += Changes[j].Spaces + Changes[j].TokenLength;
590 unsigned ChangeMaxColumn = Style.ColumnLimit - LineLengthAfter;
591
592 MinColumn = std::max(MinColumn, ChangeMinColumn);
593 MaxColumn = std::min(MaxColumn, ChangeMaxColumn);
594 }
595
596 EndOfSequence = I;
597 AlignMacroSequence(StartOfSequence, EndOfSequence, MinColumn, MaxColumn,
598 FoundMatchOnLine, AlignMacrosMatches, Changes);
599 }
600
alignConsecutiveAssignments()601 void WhitespaceManager::alignConsecutiveAssignments() {
602 if (!Style.AlignConsecutiveAssignments)
603 return;
604
605 AlignTokens(
606 Style,
607 [&](const Change &C) {
608 // Do not align on equal signs that are first on a line.
609 if (C.NewlinesBefore > 0)
610 return false;
611
612 // Do not align on equal signs that are last on a line.
613 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
614 return false;
615
616 return C.Tok->is(tok::equal);
617 },
618 Changes, /*StartAt=*/0);
619 }
620
alignConsecutiveBitFields()621 void WhitespaceManager::alignConsecutiveBitFields() {
622 if (!Style.AlignConsecutiveBitFields)
623 return;
624
625 AlignTokens(
626 Style,
627 [&](Change const &C) {
628 // Do not align on ':' that is first on a line.
629 if (C.NewlinesBefore > 0)
630 return false;
631
632 // Do not align on ':' that is last on a line.
633 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
634 return false;
635
636 return C.Tok->is(TT_BitFieldColon);
637 },
638 Changes, /*StartAt=*/0);
639 }
640
alignConsecutiveDeclarations()641 void WhitespaceManager::alignConsecutiveDeclarations() {
642 if (!Style.AlignConsecutiveDeclarations)
643 return;
644
645 // FIXME: Currently we don't handle properly the PointerAlignment: Right
646 // The * and & are not aligned and are left dangling. Something has to be done
647 // about it, but it raises the question of alignment of code like:
648 // const char* const* v1;
649 // float const* v2;
650 // SomeVeryLongType const& v3;
651 AlignTokens(
652 Style,
653 [](Change const &C) {
654 // tok::kw_operator is necessary for aligning operator overload
655 // definitions.
656 if (C.Tok->isOneOf(TT_FunctionDeclarationName, tok::kw_operator))
657 return true;
658 if (C.Tok->isNot(TT_StartOfName))
659 return false;
660 // Check if there is a subsequent name that starts the same declaration.
661 for (FormatToken *Next = C.Tok->Next; Next; Next = Next->Next) {
662 if (Next->is(tok::comment))
663 continue;
664 if (!Next->Tok.getIdentifierInfo())
665 break;
666 if (Next->isOneOf(TT_StartOfName, TT_FunctionDeclarationName,
667 tok::kw_operator))
668 return false;
669 }
670 return true;
671 },
672 Changes, /*StartAt=*/0);
673 }
674
alignChainedConditionals()675 void WhitespaceManager::alignChainedConditionals() {
676 if (Style.BreakBeforeTernaryOperators) {
677 AlignTokens(
678 Style,
679 [](Change const &C) {
680 // Align question operators and last colon
681 return C.Tok->is(TT_ConditionalExpr) &&
682 ((C.Tok->is(tok::question) && !C.NewlinesBefore) ||
683 (C.Tok->is(tok::colon) && C.Tok->Next &&
684 (C.Tok->Next->FakeLParens.size() == 0 ||
685 C.Tok->Next->FakeLParens.back() != prec::Conditional)));
686 },
687 Changes, /*StartAt=*/0);
688 } else {
689 static auto AlignWrappedOperand = [](Change const &C) {
690 auto Previous = C.Tok->getPreviousNonComment(); // Previous;
691 return C.NewlinesBefore && Previous && Previous->is(TT_ConditionalExpr) &&
692 (Previous->is(tok::question) ||
693 (Previous->is(tok::colon) &&
694 (C.Tok->FakeLParens.size() == 0 ||
695 C.Tok->FakeLParens.back() != prec::Conditional)));
696 };
697 // Ensure we keep alignment of wrapped operands with non-wrapped operands
698 // Since we actually align the operators, the wrapped operands need the
699 // extra offset to be properly aligned.
700 for (Change &C : Changes) {
701 if (AlignWrappedOperand(C))
702 C.StartOfTokenColumn -= 2;
703 }
704 AlignTokens(
705 Style,
706 [this](Change const &C) {
707 // Align question operators if next operand is not wrapped, as
708 // well as wrapped operands after question operator or last
709 // colon in conditional sequence
710 return (C.Tok->is(TT_ConditionalExpr) && C.Tok->is(tok::question) &&
711 &C != &Changes.back() && (&C + 1)->NewlinesBefore == 0 &&
712 !(&C + 1)->IsTrailingComment) ||
713 AlignWrappedOperand(C);
714 },
715 Changes, /*StartAt=*/0);
716 }
717 }
718
alignTrailingComments()719 void WhitespaceManager::alignTrailingComments() {
720 unsigned MinColumn = 0;
721 unsigned MaxColumn = UINT_MAX;
722 unsigned StartOfSequence = 0;
723 bool BreakBeforeNext = false;
724 unsigned Newlines = 0;
725 for (unsigned i = 0, e = Changes.size(); i != e; ++i) {
726 if (Changes[i].StartOfBlockComment)
727 continue;
728 Newlines += Changes[i].NewlinesBefore;
729 if (Changes[i].Tok->MustBreakAlignBefore)
730 BreakBeforeNext = true;
731 if (!Changes[i].IsTrailingComment)
732 continue;
733
734 unsigned ChangeMinColumn = Changes[i].StartOfTokenColumn;
735 unsigned ChangeMaxColumn;
736
737 if (Style.ColumnLimit == 0)
738 ChangeMaxColumn = UINT_MAX;
739 else if (Style.ColumnLimit >= Changes[i].TokenLength)
740 ChangeMaxColumn = Style.ColumnLimit - Changes[i].TokenLength;
741 else
742 ChangeMaxColumn = ChangeMinColumn;
743
744 // If we don't create a replacement for this change, we have to consider
745 // it to be immovable.
746 if (!Changes[i].CreateReplacement)
747 ChangeMaxColumn = ChangeMinColumn;
748
749 if (i + 1 != e && Changes[i + 1].ContinuesPPDirective)
750 ChangeMaxColumn -= 2;
751 // If this comment follows an } in column 0, it probably documents the
752 // closing of a namespace and we don't want to align it.
753 bool FollowsRBraceInColumn0 = i > 0 && Changes[i].NewlinesBefore == 0 &&
754 Changes[i - 1].Tok->is(tok::r_brace) &&
755 Changes[i - 1].StartOfTokenColumn == 0;
756 bool WasAlignedWithStartOfNextLine = false;
757 if (Changes[i].NewlinesBefore == 1) { // A comment on its own line.
758 unsigned CommentColumn = SourceMgr.getSpellingColumnNumber(
759 Changes[i].OriginalWhitespaceRange.getEnd());
760 for (unsigned j = i + 1; j != e; ++j) {
761 if (Changes[j].Tok->is(tok::comment))
762 continue;
763
764 unsigned NextColumn = SourceMgr.getSpellingColumnNumber(
765 Changes[j].OriginalWhitespaceRange.getEnd());
766 // The start of the next token was previously aligned with the
767 // start of this comment.
768 WasAlignedWithStartOfNextLine =
769 CommentColumn == NextColumn ||
770 CommentColumn == NextColumn + Style.IndentWidth;
771 break;
772 }
773 }
774 if (!Style.AlignTrailingComments || FollowsRBraceInColumn0) {
775 alignTrailingComments(StartOfSequence, i, MinColumn);
776 MinColumn = ChangeMinColumn;
777 MaxColumn = ChangeMinColumn;
778 StartOfSequence = i;
779 } else if (BreakBeforeNext || Newlines > 1 ||
780 (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn) ||
781 // Break the comment sequence if the previous line did not end
782 // in a trailing comment.
783 (Changes[i].NewlinesBefore == 1 && i > 0 &&
784 !Changes[i - 1].IsTrailingComment) ||
785 WasAlignedWithStartOfNextLine) {
786 alignTrailingComments(StartOfSequence, i, MinColumn);
787 MinColumn = ChangeMinColumn;
788 MaxColumn = ChangeMaxColumn;
789 StartOfSequence = i;
790 } else {
791 MinColumn = std::max(MinColumn, ChangeMinColumn);
792 MaxColumn = std::min(MaxColumn, ChangeMaxColumn);
793 }
794 BreakBeforeNext = (i == 0) || (Changes[i].NewlinesBefore > 1) ||
795 // Never start a sequence with a comment at the beginning
796 // of the line.
797 (Changes[i].NewlinesBefore == 1 && StartOfSequence == i);
798 Newlines = 0;
799 }
800 alignTrailingComments(StartOfSequence, Changes.size(), MinColumn);
801 }
802
alignTrailingComments(unsigned Start,unsigned End,unsigned Column)803 void WhitespaceManager::alignTrailingComments(unsigned Start, unsigned End,
804 unsigned Column) {
805 for (unsigned i = Start; i != End; ++i) {
806 int Shift = 0;
807 if (Changes[i].IsTrailingComment) {
808 Shift = Column - Changes[i].StartOfTokenColumn;
809 }
810 if (Changes[i].StartOfBlockComment) {
811 Shift = Changes[i].IndentationOffset +
812 Changes[i].StartOfBlockComment->StartOfTokenColumn -
813 Changes[i].StartOfTokenColumn;
814 }
815 assert(Shift >= 0);
816 Changes[i].Spaces += Shift;
817 if (i + 1 != Changes.size())
818 Changes[i + 1].PreviousEndOfTokenColumn += Shift;
819 Changes[i].StartOfTokenColumn += Shift;
820 }
821 }
822
alignEscapedNewlines()823 void WhitespaceManager::alignEscapedNewlines() {
824 if (Style.AlignEscapedNewlines == FormatStyle::ENAS_DontAlign)
825 return;
826
827 bool AlignLeft = Style.AlignEscapedNewlines == FormatStyle::ENAS_Left;
828 unsigned MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit;
829 unsigned StartOfMacro = 0;
830 for (unsigned i = 1, e = Changes.size(); i < e; ++i) {
831 Change &C = Changes[i];
832 if (C.NewlinesBefore > 0) {
833 if (C.ContinuesPPDirective) {
834 MaxEndOfLine = std::max(C.PreviousEndOfTokenColumn + 2, MaxEndOfLine);
835 } else {
836 alignEscapedNewlines(StartOfMacro + 1, i, MaxEndOfLine);
837 MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit;
838 StartOfMacro = i;
839 }
840 }
841 }
842 alignEscapedNewlines(StartOfMacro + 1, Changes.size(), MaxEndOfLine);
843 }
844
alignEscapedNewlines(unsigned Start,unsigned End,unsigned Column)845 void WhitespaceManager::alignEscapedNewlines(unsigned Start, unsigned End,
846 unsigned Column) {
847 for (unsigned i = Start; i < End; ++i) {
848 Change &C = Changes[i];
849 if (C.NewlinesBefore > 0) {
850 assert(C.ContinuesPPDirective);
851 if (C.PreviousEndOfTokenColumn + 1 > Column)
852 C.EscapedNewlineColumn = 0;
853 else
854 C.EscapedNewlineColumn = Column;
855 }
856 }
857 }
858
generateChanges()859 void WhitespaceManager::generateChanges() {
860 for (unsigned i = 0, e = Changes.size(); i != e; ++i) {
861 const Change &C = Changes[i];
862 if (i > 0) {
863 assert(Changes[i - 1].OriginalWhitespaceRange.getBegin() !=
864 C.OriginalWhitespaceRange.getBegin() &&
865 "Generating two replacements for the same location");
866 }
867 if (C.CreateReplacement) {
868 std::string ReplacementText = C.PreviousLinePostfix;
869 if (C.ContinuesPPDirective)
870 appendEscapedNewlineText(ReplacementText, C.NewlinesBefore,
871 C.PreviousEndOfTokenColumn,
872 C.EscapedNewlineColumn);
873 else
874 appendNewlineText(ReplacementText, C.NewlinesBefore);
875 appendIndentText(
876 ReplacementText, C.Tok->IndentLevel, std::max(0, C.Spaces),
877 C.StartOfTokenColumn - std::max(0, C.Spaces), C.IsAligned);
878 ReplacementText.append(C.CurrentLinePrefix);
879 storeReplacement(C.OriginalWhitespaceRange, ReplacementText);
880 }
881 }
882 }
883
storeReplacement(SourceRange Range,StringRef Text)884 void WhitespaceManager::storeReplacement(SourceRange Range, StringRef Text) {
885 unsigned WhitespaceLength = SourceMgr.getFileOffset(Range.getEnd()) -
886 SourceMgr.getFileOffset(Range.getBegin());
887 // Don't create a replacement, if it does not change anything.
888 if (StringRef(SourceMgr.getCharacterData(Range.getBegin()),
889 WhitespaceLength) == Text)
890 return;
891 auto Err = Replaces.add(tooling::Replacement(
892 SourceMgr, CharSourceRange::getCharRange(Range), Text));
893 // FIXME: better error handling. For now, just print an error message in the
894 // release version.
895 if (Err) {
896 llvm::errs() << llvm::toString(std::move(Err)) << "\n";
897 assert(false);
898 }
899 }
900
appendNewlineText(std::string & Text,unsigned Newlines)901 void WhitespaceManager::appendNewlineText(std::string &Text,
902 unsigned Newlines) {
903 for (unsigned i = 0; i < Newlines; ++i)
904 Text.append(UseCRLF ? "\r\n" : "\n");
905 }
906
appendEscapedNewlineText(std::string & Text,unsigned Newlines,unsigned PreviousEndOfTokenColumn,unsigned EscapedNewlineColumn)907 void WhitespaceManager::appendEscapedNewlineText(
908 std::string &Text, unsigned Newlines, unsigned PreviousEndOfTokenColumn,
909 unsigned EscapedNewlineColumn) {
910 if (Newlines > 0) {
911 unsigned Spaces =
912 std::max<int>(1, EscapedNewlineColumn - PreviousEndOfTokenColumn - 1);
913 for (unsigned i = 0; i < Newlines; ++i) {
914 Text.append(Spaces, ' ');
915 Text.append(UseCRLF ? "\\\r\n" : "\\\n");
916 Spaces = std::max<int>(0, EscapedNewlineColumn - 1);
917 }
918 }
919 }
920
appendIndentText(std::string & Text,unsigned IndentLevel,unsigned Spaces,unsigned WhitespaceStartColumn,bool IsAligned)921 void WhitespaceManager::appendIndentText(std::string &Text,
922 unsigned IndentLevel, unsigned Spaces,
923 unsigned WhitespaceStartColumn,
924 bool IsAligned) {
925 switch (Style.UseTab) {
926 case FormatStyle::UT_Never:
927 Text.append(Spaces, ' ');
928 break;
929 case FormatStyle::UT_Always: {
930 if (Style.TabWidth) {
931 unsigned FirstTabWidth =
932 Style.TabWidth - WhitespaceStartColumn % Style.TabWidth;
933
934 // Insert only spaces when we want to end up before the next tab.
935 if (Spaces < FirstTabWidth || Spaces == 1) {
936 Text.append(Spaces, ' ');
937 break;
938 }
939 // Align to the next tab.
940 Spaces -= FirstTabWidth;
941 Text.append("\t");
942
943 Text.append(Spaces / Style.TabWidth, '\t');
944 Text.append(Spaces % Style.TabWidth, ' ');
945 } else if (Spaces == 1) {
946 Text.append(Spaces, ' ');
947 }
948 break;
949 }
950 case FormatStyle::UT_ForIndentation:
951 if (WhitespaceStartColumn == 0) {
952 unsigned Indentation = IndentLevel * Style.IndentWidth;
953 Spaces = appendTabIndent(Text, Spaces, Indentation);
954 }
955 Text.append(Spaces, ' ');
956 break;
957 case FormatStyle::UT_ForContinuationAndIndentation:
958 if (WhitespaceStartColumn == 0)
959 Spaces = appendTabIndent(Text, Spaces, Spaces);
960 Text.append(Spaces, ' ');
961 break;
962 case FormatStyle::UT_AlignWithSpaces:
963 if (WhitespaceStartColumn == 0) {
964 unsigned Indentation =
965 IsAligned ? IndentLevel * Style.IndentWidth : Spaces;
966 Spaces = appendTabIndent(Text, Spaces, Indentation);
967 }
968 Text.append(Spaces, ' ');
969 break;
970 }
971 }
972
appendTabIndent(std::string & Text,unsigned Spaces,unsigned Indentation)973 unsigned WhitespaceManager::appendTabIndent(std::string &Text, unsigned Spaces,
974 unsigned Indentation) {
975 // This happens, e.g. when a line in a block comment is indented less than the
976 // first one.
977 if (Indentation > Spaces)
978 Indentation = Spaces;
979 if (Style.TabWidth) {
980 unsigned Tabs = Indentation / Style.TabWidth;
981 Text.append(Tabs, '\t');
982 Spaces -= Tabs * Style.TabWidth;
983 }
984 return Spaces;
985 }
986
987 } // namespace format
988 } // namespace clang
989