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,const FormatStyle::AlignConsecutiveStyle & ACS=FormatStyle::ACS_None)364 static unsigned AlignTokens(
365 const FormatStyle &Style, F &&Matches,
366 SmallVector<WhitespaceManager::Change, 16> &Changes, unsigned StartAt,
367 const FormatStyle::AlignConsecutiveStyle &ACS = FormatStyle::ACS_None) {
368 unsigned MinColumn = 0;
369 unsigned MaxColumn = UINT_MAX;
370
371 // Line number of the start and the end of the current token sequence.
372 unsigned StartOfSequence = 0;
373 unsigned EndOfSequence = 0;
374
375 // Measure the scope level (i.e. depth of (), [], {}) of the first token, and
376 // abort when we hit any token in a higher scope than the starting one.
377 auto IndentAndNestingLevel = StartAt < Changes.size()
378 ? Changes[StartAt].indentAndNestingLevel()
379 : std::tuple<unsigned, unsigned, unsigned>();
380
381 // Keep track of the number of commas before the matching tokens, we will only
382 // align a sequence of matching tokens if they are preceded by the same number
383 // of commas.
384 unsigned CommasBeforeLastMatch = 0;
385 unsigned CommasBeforeMatch = 0;
386
387 // Whether a matching token has been found on the current line.
388 bool FoundMatchOnLine = false;
389
390 // Whether the current line consists purely of comments.
391 bool LineIsComment = true;
392
393 // Aligns a sequence of matching tokens, on the MinColumn column.
394 //
395 // Sequences start from the first matching token to align, and end at the
396 // first token of the first line that doesn't need to be aligned.
397 //
398 // We need to adjust the StartOfTokenColumn of each Change that is on a line
399 // containing any matching token to be aligned and located after such token.
400 auto AlignCurrentSequence = [&] {
401 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence)
402 AlignTokenSequence(StartOfSequence, EndOfSequence, MinColumn, Matches,
403 Changes);
404 MinColumn = 0;
405 MaxColumn = UINT_MAX;
406 StartOfSequence = 0;
407 EndOfSequence = 0;
408 };
409
410 unsigned i = StartAt;
411 for (unsigned e = Changes.size(); i != e; ++i) {
412 if (Changes[i].indentAndNestingLevel() < IndentAndNestingLevel)
413 break;
414
415 if (Changes[i].NewlinesBefore != 0) {
416 CommasBeforeMatch = 0;
417 EndOfSequence = i;
418
419 // Whether to break the alignment sequence because of an empty line.
420 bool EmptyLineBreak =
421 (Changes[i].NewlinesBefore > 1) &&
422 (ACS != FormatStyle::ACS_AcrossEmptyLines) &&
423 (ACS != FormatStyle::ACS_AcrossEmptyLinesAndComments);
424
425 // Whether to break the alignment sequence because of a line without a
426 // match.
427 bool NoMatchBreak =
428 !FoundMatchOnLine &&
429 !(LineIsComment &&
430 ((ACS == FormatStyle::ACS_AcrossComments) ||
431 (ACS == FormatStyle::ACS_AcrossEmptyLinesAndComments)));
432
433 if (EmptyLineBreak || NoMatchBreak)
434 AlignCurrentSequence();
435
436 // A new line starts, re-initialize line status tracking bools.
437 FoundMatchOnLine = false;
438 LineIsComment = true;
439 }
440
441 if (!Changes[i].Tok->is(tok::comment)) {
442 LineIsComment = false;
443 }
444
445 if (Changes[i].Tok->is(tok::comma)) {
446 ++CommasBeforeMatch;
447 } else if (Changes[i].indentAndNestingLevel() > IndentAndNestingLevel) {
448 // Call AlignTokens recursively, skipping over this scope block.
449 unsigned StoppedAt = AlignTokens(Style, Matches, Changes, i, ACS);
450 i = StoppedAt - 1;
451 continue;
452 }
453
454 if (!Matches(Changes[i]))
455 continue;
456
457 // If there is more than one matching token per line, or if the number of
458 // preceding commas, do not match anymore, end the sequence.
459 if (FoundMatchOnLine || CommasBeforeMatch != CommasBeforeLastMatch)
460 AlignCurrentSequence();
461
462 CommasBeforeLastMatch = CommasBeforeMatch;
463 FoundMatchOnLine = true;
464
465 if (StartOfSequence == 0)
466 StartOfSequence = i;
467
468 unsigned ChangeMinColumn = Changes[i].StartOfTokenColumn;
469 int LineLengthAfter = Changes[i].TokenLength;
470 for (unsigned j = i + 1; j != e && Changes[j].NewlinesBefore == 0; ++j) {
471 LineLengthAfter += Changes[j].Spaces;
472 // Changes are generally 1:1 with the tokens, but a change could also be
473 // inside of a token, in which case it's counted more than once: once for
474 // the whitespace surrounding the token (!IsInsideToken) and once for
475 // each whitespace change within it (IsInsideToken).
476 // Therefore, changes inside of a token should only count the space.
477 if (!Changes[j].IsInsideToken)
478 LineLengthAfter += Changes[j].TokenLength;
479 }
480 unsigned ChangeMaxColumn = Style.ColumnLimit - LineLengthAfter;
481
482 // If we are restricted by the maximum column width, end the sequence.
483 if (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn ||
484 CommasBeforeLastMatch != CommasBeforeMatch) {
485 AlignCurrentSequence();
486 StartOfSequence = i;
487 }
488
489 MinColumn = std::max(MinColumn, ChangeMinColumn);
490 MaxColumn = std::min(MaxColumn, ChangeMaxColumn);
491 }
492
493 EndOfSequence = i;
494 AlignCurrentSequence();
495 return i;
496 }
497
498 // Aligns a sequence of matching tokens, on the MinColumn column.
499 //
500 // Sequences start from the first matching token to align, and end at the
501 // first token of the first line that doesn't need to be aligned.
502 //
503 // We need to adjust the StartOfTokenColumn of each Change that is on a line
504 // 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)505 static void AlignMacroSequence(
506 unsigned &StartOfSequence, unsigned &EndOfSequence, unsigned &MinColumn,
507 unsigned &MaxColumn, bool &FoundMatchOnLine,
508 std::function<bool(const WhitespaceManager::Change &C)> AlignMacrosMatches,
509 SmallVector<WhitespaceManager::Change, 16> &Changes) {
510 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) {
511
512 FoundMatchOnLine = false;
513 int Shift = 0;
514
515 for (unsigned I = StartOfSequence; I != EndOfSequence; ++I) {
516 if (Changes[I].NewlinesBefore > 0) {
517 Shift = 0;
518 FoundMatchOnLine = false;
519 }
520
521 // If this is the first matching token to be aligned, remember by how many
522 // spaces it has to be shifted, so the rest of the changes on the line are
523 // shifted by the same amount
524 if (!FoundMatchOnLine && AlignMacrosMatches(Changes[I])) {
525 FoundMatchOnLine = true;
526 Shift = MinColumn - Changes[I].StartOfTokenColumn;
527 Changes[I].Spaces += Shift;
528 }
529
530 assert(Shift >= 0);
531 Changes[I].StartOfTokenColumn += Shift;
532 if (I + 1 != Changes.size())
533 Changes[I + 1].PreviousEndOfTokenColumn += Shift;
534 }
535 }
536
537 MinColumn = 0;
538 MaxColumn = UINT_MAX;
539 StartOfSequence = 0;
540 EndOfSequence = 0;
541 }
542
alignConsecutiveMacros()543 void WhitespaceManager::alignConsecutiveMacros() {
544 if (Style.AlignConsecutiveMacros == FormatStyle::ACS_None)
545 return;
546
547 auto AlignMacrosMatches = [](const Change &C) {
548 const FormatToken *Current = C.Tok;
549 unsigned SpacesRequiredBefore = 1;
550
551 if (Current->SpacesRequiredBefore == 0 || !Current->Previous)
552 return false;
553
554 Current = Current->Previous;
555
556 // If token is a ")", skip over the parameter list, to the
557 // token that precedes the "("
558 if (Current->is(tok::r_paren) && Current->MatchingParen) {
559 Current = Current->MatchingParen->Previous;
560 SpacesRequiredBefore = 0;
561 }
562
563 if (!Current || !Current->is(tok::identifier))
564 return false;
565
566 if (!Current->Previous || !Current->Previous->is(tok::pp_define))
567 return false;
568
569 // For a macro function, 0 spaces are required between the
570 // identifier and the lparen that opens the parameter list.
571 // For a simple macro, 1 space is required between the
572 // identifier and the first token of the defined value.
573 return Current->Next->SpacesRequiredBefore == SpacesRequiredBefore;
574 };
575
576 unsigned MinColumn = 0;
577 unsigned MaxColumn = UINT_MAX;
578
579 // Start and end of the token sequence we're processing.
580 unsigned StartOfSequence = 0;
581 unsigned EndOfSequence = 0;
582
583 // Whether a matching token has been found on the current line.
584 bool FoundMatchOnLine = false;
585
586 // Whether the current line consists only of comments
587 bool LineIsComment = true;
588
589 unsigned I = 0;
590 for (unsigned E = Changes.size(); I != E; ++I) {
591 if (Changes[I].NewlinesBefore != 0) {
592 EndOfSequence = I;
593
594 // Whether to break the alignment sequence because of an empty line.
595 bool EmptyLineBreak =
596 (Changes[I].NewlinesBefore > 1) &&
597 (Style.AlignConsecutiveMacros != FormatStyle::ACS_AcrossEmptyLines) &&
598 (Style.AlignConsecutiveMacros !=
599 FormatStyle::ACS_AcrossEmptyLinesAndComments);
600
601 // Whether to break the alignment sequence because of a line without a
602 // match.
603 bool NoMatchBreak =
604 !FoundMatchOnLine &&
605 !(LineIsComment && ((Style.AlignConsecutiveMacros ==
606 FormatStyle::ACS_AcrossComments) ||
607 (Style.AlignConsecutiveMacros ==
608 FormatStyle::ACS_AcrossEmptyLinesAndComments)));
609
610 if (EmptyLineBreak || NoMatchBreak)
611 AlignMacroSequence(StartOfSequence, EndOfSequence, MinColumn, MaxColumn,
612 FoundMatchOnLine, AlignMacrosMatches, Changes);
613
614 // A new line starts, re-initialize line status tracking bools.
615 FoundMatchOnLine = false;
616 LineIsComment = true;
617 }
618
619 if (!Changes[I].Tok->is(tok::comment)) {
620 LineIsComment = false;
621 }
622
623 if (!AlignMacrosMatches(Changes[I]))
624 continue;
625
626 FoundMatchOnLine = true;
627
628 if (StartOfSequence == 0)
629 StartOfSequence = I;
630
631 unsigned ChangeMinColumn = Changes[I].StartOfTokenColumn;
632 int LineLengthAfter = -Changes[I].Spaces;
633 for (unsigned j = I; j != E && Changes[j].NewlinesBefore == 0; ++j)
634 LineLengthAfter += Changes[j].Spaces + Changes[j].TokenLength;
635 unsigned ChangeMaxColumn = Style.ColumnLimit - LineLengthAfter;
636
637 MinColumn = std::max(MinColumn, ChangeMinColumn);
638 MaxColumn = std::min(MaxColumn, ChangeMaxColumn);
639 }
640
641 EndOfSequence = I;
642 AlignMacroSequence(StartOfSequence, EndOfSequence, MinColumn, MaxColumn,
643 FoundMatchOnLine, AlignMacrosMatches, Changes);
644 }
645
alignConsecutiveAssignments()646 void WhitespaceManager::alignConsecutiveAssignments() {
647 if (Style.AlignConsecutiveAssignments == FormatStyle::ACS_None)
648 return;
649
650 AlignTokens(
651 Style,
652 [&](const Change &C) {
653 // Do not align on equal signs that are first on a line.
654 if (C.NewlinesBefore > 0)
655 return false;
656
657 // Do not align on equal signs that are last on a line.
658 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
659 return false;
660
661 return C.Tok->is(tok::equal);
662 },
663 Changes, /*StartAt=*/0, Style.AlignConsecutiveAssignments);
664 }
665
alignConsecutiveBitFields()666 void WhitespaceManager::alignConsecutiveBitFields() {
667 if (Style.AlignConsecutiveBitFields == FormatStyle::ACS_None)
668 return;
669
670 AlignTokens(
671 Style,
672 [&](Change const &C) {
673 // Do not align on ':' that is first on a line.
674 if (C.NewlinesBefore > 0)
675 return false;
676
677 // Do not align on ':' that is last on a line.
678 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
679 return false;
680
681 return C.Tok->is(TT_BitFieldColon);
682 },
683 Changes, /*StartAt=*/0, Style.AlignConsecutiveBitFields);
684 }
685
alignConsecutiveDeclarations()686 void WhitespaceManager::alignConsecutiveDeclarations() {
687 if (Style.AlignConsecutiveDeclarations == FormatStyle::ACS_None)
688 return;
689
690 // FIXME: Currently we don't handle properly the PointerAlignment: Right
691 // The * and & are not aligned and are left dangling. Something has to be done
692 // about it, but it raises the question of alignment of code like:
693 // const char* const* v1;
694 // float const* v2;
695 // SomeVeryLongType const& v3;
696 AlignTokens(
697 Style,
698 [](Change const &C) {
699 // tok::kw_operator is necessary for aligning operator overload
700 // definitions.
701 if (C.Tok->isOneOf(TT_FunctionDeclarationName, tok::kw_operator))
702 return true;
703 if (C.Tok->isNot(TT_StartOfName))
704 return false;
705 if (C.Tok->Previous &&
706 C.Tok->Previous->is(TT_StatementAttributeLikeMacro))
707 return false;
708 // Check if there is a subsequent name that starts the same declaration.
709 for (FormatToken *Next = C.Tok->Next; Next; Next = Next->Next) {
710 if (Next->is(tok::comment))
711 continue;
712 if (!Next->Tok.getIdentifierInfo())
713 break;
714 if (Next->isOneOf(TT_StartOfName, TT_FunctionDeclarationName,
715 tok::kw_operator))
716 return false;
717 }
718 return true;
719 },
720 Changes, /*StartAt=*/0, Style.AlignConsecutiveDeclarations);
721 }
722
alignChainedConditionals()723 void WhitespaceManager::alignChainedConditionals() {
724 if (Style.BreakBeforeTernaryOperators) {
725 AlignTokens(
726 Style,
727 [](Change const &C) {
728 // Align question operators and last colon
729 return C.Tok->is(TT_ConditionalExpr) &&
730 ((C.Tok->is(tok::question) && !C.NewlinesBefore) ||
731 (C.Tok->is(tok::colon) && C.Tok->Next &&
732 (C.Tok->Next->FakeLParens.size() == 0 ||
733 C.Tok->Next->FakeLParens.back() != prec::Conditional)));
734 },
735 Changes, /*StartAt=*/0);
736 } else {
737 static auto AlignWrappedOperand = [](Change const &C) {
738 auto Previous = C.Tok->getPreviousNonComment(); // Previous;
739 return C.NewlinesBefore && Previous && Previous->is(TT_ConditionalExpr) &&
740 (Previous->is(tok::question) ||
741 (Previous->is(tok::colon) &&
742 (C.Tok->FakeLParens.size() == 0 ||
743 C.Tok->FakeLParens.back() != prec::Conditional)));
744 };
745 // Ensure we keep alignment of wrapped operands with non-wrapped operands
746 // Since we actually align the operators, the wrapped operands need the
747 // extra offset to be properly aligned.
748 for (Change &C : Changes) {
749 if (AlignWrappedOperand(C))
750 C.StartOfTokenColumn -= 2;
751 }
752 AlignTokens(
753 Style,
754 [this](Change const &C) {
755 // Align question operators if next operand is not wrapped, as
756 // well as wrapped operands after question operator or last
757 // colon in conditional sequence
758 return (C.Tok->is(TT_ConditionalExpr) && C.Tok->is(tok::question) &&
759 &C != &Changes.back() && (&C + 1)->NewlinesBefore == 0 &&
760 !(&C + 1)->IsTrailingComment) ||
761 AlignWrappedOperand(C);
762 },
763 Changes, /*StartAt=*/0);
764 }
765 }
766
alignTrailingComments()767 void WhitespaceManager::alignTrailingComments() {
768 unsigned MinColumn = 0;
769 unsigned MaxColumn = UINT_MAX;
770 unsigned StartOfSequence = 0;
771 bool BreakBeforeNext = false;
772 unsigned Newlines = 0;
773 for (unsigned i = 0, e = Changes.size(); i != e; ++i) {
774 if (Changes[i].StartOfBlockComment)
775 continue;
776 Newlines += Changes[i].NewlinesBefore;
777 if (!Changes[i].IsTrailingComment)
778 continue;
779
780 unsigned ChangeMinColumn = Changes[i].StartOfTokenColumn;
781 unsigned ChangeMaxColumn;
782
783 if (Style.ColumnLimit == 0)
784 ChangeMaxColumn = UINT_MAX;
785 else if (Style.ColumnLimit >= Changes[i].TokenLength)
786 ChangeMaxColumn = Style.ColumnLimit - Changes[i].TokenLength;
787 else
788 ChangeMaxColumn = ChangeMinColumn;
789
790 // If we don't create a replacement for this change, we have to consider
791 // it to be immovable.
792 if (!Changes[i].CreateReplacement)
793 ChangeMaxColumn = ChangeMinColumn;
794
795 if (i + 1 != e && Changes[i + 1].ContinuesPPDirective)
796 ChangeMaxColumn -= 2;
797 // If this comment follows an } in column 0, it probably documents the
798 // closing of a namespace and we don't want to align it.
799 bool FollowsRBraceInColumn0 = i > 0 && Changes[i].NewlinesBefore == 0 &&
800 Changes[i - 1].Tok->is(tok::r_brace) &&
801 Changes[i - 1].StartOfTokenColumn == 0;
802 bool WasAlignedWithStartOfNextLine = false;
803 if (Changes[i].NewlinesBefore == 1) { // A comment on its own line.
804 unsigned CommentColumn = SourceMgr.getSpellingColumnNumber(
805 Changes[i].OriginalWhitespaceRange.getEnd());
806 for (unsigned j = i + 1; j != e; ++j) {
807 if (Changes[j].Tok->is(tok::comment))
808 continue;
809
810 unsigned NextColumn = SourceMgr.getSpellingColumnNumber(
811 Changes[j].OriginalWhitespaceRange.getEnd());
812 // The start of the next token was previously aligned with the
813 // start of this comment.
814 WasAlignedWithStartOfNextLine =
815 CommentColumn == NextColumn ||
816 CommentColumn == NextColumn + Style.IndentWidth;
817 break;
818 }
819 }
820 if (!Style.AlignTrailingComments || FollowsRBraceInColumn0) {
821 alignTrailingComments(StartOfSequence, i, MinColumn);
822 MinColumn = ChangeMinColumn;
823 MaxColumn = ChangeMinColumn;
824 StartOfSequence = i;
825 } else if (BreakBeforeNext || Newlines > 1 ||
826 (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn) ||
827 // Break the comment sequence if the previous line did not end
828 // in a trailing comment.
829 (Changes[i].NewlinesBefore == 1 && i > 0 &&
830 !Changes[i - 1].IsTrailingComment) ||
831 WasAlignedWithStartOfNextLine) {
832 alignTrailingComments(StartOfSequence, i, MinColumn);
833 MinColumn = ChangeMinColumn;
834 MaxColumn = ChangeMaxColumn;
835 StartOfSequence = i;
836 } else {
837 MinColumn = std::max(MinColumn, ChangeMinColumn);
838 MaxColumn = std::min(MaxColumn, ChangeMaxColumn);
839 }
840 BreakBeforeNext = (i == 0) || (Changes[i].NewlinesBefore > 1) ||
841 // Never start a sequence with a comment at the beginning
842 // of the line.
843 (Changes[i].NewlinesBefore == 1 && StartOfSequence == i);
844 Newlines = 0;
845 }
846 alignTrailingComments(StartOfSequence, Changes.size(), MinColumn);
847 }
848
alignTrailingComments(unsigned Start,unsigned End,unsigned Column)849 void WhitespaceManager::alignTrailingComments(unsigned Start, unsigned End,
850 unsigned Column) {
851 for (unsigned i = Start; i != End; ++i) {
852 int Shift = 0;
853 if (Changes[i].IsTrailingComment) {
854 Shift = Column - Changes[i].StartOfTokenColumn;
855 }
856 if (Changes[i].StartOfBlockComment) {
857 Shift = Changes[i].IndentationOffset +
858 Changes[i].StartOfBlockComment->StartOfTokenColumn -
859 Changes[i].StartOfTokenColumn;
860 }
861 assert(Shift >= 0);
862 Changes[i].Spaces += Shift;
863 if (i + 1 != Changes.size())
864 Changes[i + 1].PreviousEndOfTokenColumn += Shift;
865 Changes[i].StartOfTokenColumn += Shift;
866 }
867 }
868
alignEscapedNewlines()869 void WhitespaceManager::alignEscapedNewlines() {
870 if (Style.AlignEscapedNewlines == FormatStyle::ENAS_DontAlign)
871 return;
872
873 bool AlignLeft = Style.AlignEscapedNewlines == FormatStyle::ENAS_Left;
874 unsigned MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit;
875 unsigned StartOfMacro = 0;
876 for (unsigned i = 1, e = Changes.size(); i < e; ++i) {
877 Change &C = Changes[i];
878 if (C.NewlinesBefore > 0) {
879 if (C.ContinuesPPDirective) {
880 MaxEndOfLine = std::max(C.PreviousEndOfTokenColumn + 2, MaxEndOfLine);
881 } else {
882 alignEscapedNewlines(StartOfMacro + 1, i, MaxEndOfLine);
883 MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit;
884 StartOfMacro = i;
885 }
886 }
887 }
888 alignEscapedNewlines(StartOfMacro + 1, Changes.size(), MaxEndOfLine);
889 }
890
alignEscapedNewlines(unsigned Start,unsigned End,unsigned Column)891 void WhitespaceManager::alignEscapedNewlines(unsigned Start, unsigned End,
892 unsigned Column) {
893 for (unsigned i = Start; i < End; ++i) {
894 Change &C = Changes[i];
895 if (C.NewlinesBefore > 0) {
896 assert(C.ContinuesPPDirective);
897 if (C.PreviousEndOfTokenColumn + 1 > Column)
898 C.EscapedNewlineColumn = 0;
899 else
900 C.EscapedNewlineColumn = Column;
901 }
902 }
903 }
904
generateChanges()905 void WhitespaceManager::generateChanges() {
906 for (unsigned i = 0, e = Changes.size(); i != e; ++i) {
907 const Change &C = Changes[i];
908 if (i > 0) {
909 assert(Changes[i - 1].OriginalWhitespaceRange.getBegin() !=
910 C.OriginalWhitespaceRange.getBegin() &&
911 "Generating two replacements for the same location");
912 }
913 if (C.CreateReplacement) {
914 std::string ReplacementText = C.PreviousLinePostfix;
915 if (C.ContinuesPPDirective)
916 appendEscapedNewlineText(ReplacementText, C.NewlinesBefore,
917 C.PreviousEndOfTokenColumn,
918 C.EscapedNewlineColumn);
919 else
920 appendNewlineText(ReplacementText, C.NewlinesBefore);
921 appendIndentText(
922 ReplacementText, C.Tok->IndentLevel, std::max(0, C.Spaces),
923 C.StartOfTokenColumn - std::max(0, C.Spaces), C.IsAligned);
924 ReplacementText.append(C.CurrentLinePrefix);
925 storeReplacement(C.OriginalWhitespaceRange, ReplacementText);
926 }
927 }
928 }
929
storeReplacement(SourceRange Range,StringRef Text)930 void WhitespaceManager::storeReplacement(SourceRange Range, StringRef Text) {
931 unsigned WhitespaceLength = SourceMgr.getFileOffset(Range.getEnd()) -
932 SourceMgr.getFileOffset(Range.getBegin());
933 // Don't create a replacement, if it does not change anything.
934 if (StringRef(SourceMgr.getCharacterData(Range.getBegin()),
935 WhitespaceLength) == Text)
936 return;
937 auto Err = Replaces.add(tooling::Replacement(
938 SourceMgr, CharSourceRange::getCharRange(Range), Text));
939 // FIXME: better error handling. For now, just print an error message in the
940 // release version.
941 if (Err) {
942 llvm::errs() << llvm::toString(std::move(Err)) << "\n";
943 assert(false);
944 }
945 }
946
appendNewlineText(std::string & Text,unsigned Newlines)947 void WhitespaceManager::appendNewlineText(std::string &Text,
948 unsigned Newlines) {
949 for (unsigned i = 0; i < Newlines; ++i)
950 Text.append(UseCRLF ? "\r\n" : "\n");
951 }
952
appendEscapedNewlineText(std::string & Text,unsigned Newlines,unsigned PreviousEndOfTokenColumn,unsigned EscapedNewlineColumn)953 void WhitespaceManager::appendEscapedNewlineText(
954 std::string &Text, unsigned Newlines, unsigned PreviousEndOfTokenColumn,
955 unsigned EscapedNewlineColumn) {
956 if (Newlines > 0) {
957 unsigned Spaces =
958 std::max<int>(1, EscapedNewlineColumn - PreviousEndOfTokenColumn - 1);
959 for (unsigned i = 0; i < Newlines; ++i) {
960 Text.append(Spaces, ' ');
961 Text.append(UseCRLF ? "\\\r\n" : "\\\n");
962 Spaces = std::max<int>(0, EscapedNewlineColumn - 1);
963 }
964 }
965 }
966
appendIndentText(std::string & Text,unsigned IndentLevel,unsigned Spaces,unsigned WhitespaceStartColumn,bool IsAligned)967 void WhitespaceManager::appendIndentText(std::string &Text,
968 unsigned IndentLevel, unsigned Spaces,
969 unsigned WhitespaceStartColumn,
970 bool IsAligned) {
971 switch (Style.UseTab) {
972 case FormatStyle::UT_Never:
973 Text.append(Spaces, ' ');
974 break;
975 case FormatStyle::UT_Always: {
976 if (Style.TabWidth) {
977 unsigned FirstTabWidth =
978 Style.TabWidth - WhitespaceStartColumn % Style.TabWidth;
979
980 // Insert only spaces when we want to end up before the next tab.
981 if (Spaces < FirstTabWidth || Spaces == 1) {
982 Text.append(Spaces, ' ');
983 break;
984 }
985 // Align to the next tab.
986 Spaces -= FirstTabWidth;
987 Text.append("\t");
988
989 Text.append(Spaces / Style.TabWidth, '\t');
990 Text.append(Spaces % Style.TabWidth, ' ');
991 } else if (Spaces == 1) {
992 Text.append(Spaces, ' ');
993 }
994 break;
995 }
996 case FormatStyle::UT_ForIndentation:
997 if (WhitespaceStartColumn == 0) {
998 unsigned Indentation = IndentLevel * Style.IndentWidth;
999 Spaces = appendTabIndent(Text, Spaces, Indentation);
1000 }
1001 Text.append(Spaces, ' ');
1002 break;
1003 case FormatStyle::UT_ForContinuationAndIndentation:
1004 if (WhitespaceStartColumn == 0)
1005 Spaces = appendTabIndent(Text, Spaces, Spaces);
1006 Text.append(Spaces, ' ');
1007 break;
1008 case FormatStyle::UT_AlignWithSpaces:
1009 if (WhitespaceStartColumn == 0) {
1010 unsigned Indentation =
1011 IsAligned ? IndentLevel * Style.IndentWidth : Spaces;
1012 Spaces = appendTabIndent(Text, Spaces, Indentation);
1013 }
1014 Text.append(Spaces, ' ');
1015 break;
1016 }
1017 }
1018
appendTabIndent(std::string & Text,unsigned Spaces,unsigned Indentation)1019 unsigned WhitespaceManager::appendTabIndent(std::string &Text, unsigned Spaces,
1020 unsigned Indentation) {
1021 // This happens, e.g. when a line in a block comment is indented less than the
1022 // first one.
1023 if (Indentation > Spaces)
1024 Indentation = Spaces;
1025 if (Style.TabWidth) {
1026 unsigned Tabs = Indentation / Style.TabWidth;
1027 Text.append(Tabs, '\t');
1028 Spaces -= Tabs * Style.TabWidth;
1029 }
1030 return Spaces;
1031 }
1032
1033 } // namespace format
1034 } // namespace clang
1035