[DFAJumpThreading] Remove incoming StartBlock from all phis when unfolding select...
[llvm-project.git] / clang / lib / Format / WhitespaceManager.cpp
blobff8b1e6e13a3f7730861f685788ab369ef8b89df
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 //===----------------------------------------------------------------------===//
14 #include "WhitespaceManager.h"
15 #include "llvm/ADT/STLExtras.h"
16 #include "llvm/ADT/SmallVector.h"
17 #include <algorithm>
19 namespace clang {
20 namespace format {
22 bool WhitespaceManager::Change::IsBeforeInFile::operator()(
23 const Change &C1, const Change &C2) const {
24 return SourceMgr.isBeforeInTranslationUnit(
25 C1.OriginalWhitespaceRange.getBegin(),
26 C2.OriginalWhitespaceRange.getBegin()) ||
27 (C1.OriginalWhitespaceRange.getBegin() ==
28 C2.OriginalWhitespaceRange.getBegin() &&
29 SourceMgr.isBeforeInTranslationUnit(
30 C1.OriginalWhitespaceRange.getEnd(),
31 C2.OriginalWhitespaceRange.getEnd()));
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) {
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));
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));
77 llvm::Error
78 WhitespaceManager::addReplacement(const tooling::Replacement &Replacement) {
79 return Replaces.add(Replacement);
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;
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));
103 const tooling::Replacements &WhitespaceManager::generateReplacements() {
104 if (Changes.empty())
105 return Replaces;
107 llvm::sort(Changes, Change::IsBeforeInFile(SourceMgr));
108 calculateLineBreakInformation();
109 alignConsecutiveMacros();
110 alignConsecutiveShortCaseStatements();
111 alignConsecutiveDeclarations();
112 alignConsecutiveBitFields();
113 alignConsecutiveAssignments();
114 alignChainedConditionals();
115 alignTrailingComments();
116 alignEscapedNewlines();
117 alignArrayInitializers();
118 generateChanges();
120 return Replaces;
123 void WhitespaceManager::calculateLineBreakInformation() {
124 Changes[0].PreviousEndOfTokenColumn = 0;
125 Change *LastOutsideTokenChange = &Changes[0];
126 for (unsigned i = 1, e = Changes.size(); i != e; ++i) {
127 SourceLocation OriginalWhitespaceStart =
128 Changes[i].OriginalWhitespaceRange.getBegin();
129 SourceLocation PreviousOriginalWhitespaceEnd =
130 Changes[i - 1].OriginalWhitespaceRange.getEnd();
131 unsigned OriginalWhitespaceStartOffset =
132 SourceMgr.getFileOffset(OriginalWhitespaceStart);
133 unsigned PreviousOriginalWhitespaceEndOffset =
134 SourceMgr.getFileOffset(PreviousOriginalWhitespaceEnd);
135 assert(PreviousOriginalWhitespaceEndOffset <=
136 OriginalWhitespaceStartOffset);
137 const char *const PreviousOriginalWhitespaceEndData =
138 SourceMgr.getCharacterData(PreviousOriginalWhitespaceEnd);
139 StringRef Text(PreviousOriginalWhitespaceEndData,
140 SourceMgr.getCharacterData(OriginalWhitespaceStart) -
141 PreviousOriginalWhitespaceEndData);
142 // Usually consecutive changes would occur in consecutive tokens. This is
143 // not the case however when analyzing some preprocessor runs of the
144 // annotated lines. For example, in this code:
146 // #if A // line 1
147 // int i = 1;
148 // #else B // line 2
149 // int i = 2;
150 // #endif // line 3
152 // one of the runs will produce the sequence of lines marked with line 1, 2
153 // and 3. So the two consecutive whitespace changes just before '// line 2'
154 // and before '#endif // line 3' span multiple lines and tokens:
156 // #else B{change X}[// line 2
157 // int i = 2;
158 // ]{change Y}#endif // line 3
160 // For this reason, if the text between consecutive changes spans multiple
161 // newlines, the token length must be adjusted to the end of the original
162 // line of the token.
163 auto NewlinePos = Text.find_first_of('\n');
164 if (NewlinePos == StringRef::npos) {
165 Changes[i - 1].TokenLength = OriginalWhitespaceStartOffset -
166 PreviousOriginalWhitespaceEndOffset +
167 Changes[i].PreviousLinePostfix.size() +
168 Changes[i - 1].CurrentLinePrefix.size();
169 } else {
170 Changes[i - 1].TokenLength =
171 NewlinePos + Changes[i - 1].CurrentLinePrefix.size();
174 // If there are multiple changes in this token, sum up all the changes until
175 // the end of the line.
176 if (Changes[i - 1].IsInsideToken && Changes[i - 1].NewlinesBefore == 0) {
177 LastOutsideTokenChange->TokenLength +=
178 Changes[i - 1].TokenLength + Changes[i - 1].Spaces;
179 } else {
180 LastOutsideTokenChange = &Changes[i - 1];
183 Changes[i].PreviousEndOfTokenColumn =
184 Changes[i - 1].StartOfTokenColumn + Changes[i - 1].TokenLength;
186 Changes[i - 1].IsTrailingComment =
187 (Changes[i].NewlinesBefore > 0 || Changes[i].Tok->is(tok::eof) ||
188 (Changes[i].IsInsideToken && Changes[i].Tok->is(tok::comment))) &&
189 Changes[i - 1].Tok->is(tok::comment) &&
190 // FIXME: This is a dirty hack. The problem is that
191 // BreakableLineCommentSection does comment reflow changes and here is
192 // the aligning of trailing comments. Consider the case where we reflow
193 // the second line up in this example:
195 // // line 1
196 // // line 2
198 // That amounts to 2 changes by BreakableLineCommentSection:
199 // - the first, delimited by (), for the whitespace between the tokens,
200 // - and second, delimited by [], for the whitespace at the beginning
201 // of the second token:
203 // // line 1(
204 // )[// ]line 2
206 // So in the end we have two changes like this:
208 // // line1()[ ]line 2
210 // Note that the OriginalWhitespaceStart of the second change is the
211 // same as the PreviousOriginalWhitespaceEnd of the first change.
212 // In this case, the below check ensures that the second change doesn't
213 // get treated as a trailing comment change here, since this might
214 // trigger additional whitespace to be wrongly inserted before "line 2"
215 // by the comment aligner here.
217 // For a proper solution we need a mechanism to say to WhitespaceManager
218 // that a particular change breaks the current sequence of trailing
219 // comments.
220 OriginalWhitespaceStart != PreviousOriginalWhitespaceEnd;
222 // FIXME: The last token is currently not always an eof token; in those
223 // cases, setting TokenLength of the last token to 0 is wrong.
224 Changes.back().TokenLength = 0;
225 Changes.back().IsTrailingComment = Changes.back().Tok->is(tok::comment);
227 const WhitespaceManager::Change *LastBlockComment = nullptr;
228 for (auto &Change : Changes) {
229 // Reset the IsTrailingComment flag for changes inside of trailing comments
230 // so they don't get realigned later. Comment line breaks however still need
231 // to be aligned.
232 if (Change.IsInsideToken && Change.NewlinesBefore == 0)
233 Change.IsTrailingComment = false;
234 Change.StartOfBlockComment = nullptr;
235 Change.IndentationOffset = 0;
236 if (Change.Tok->is(tok::comment)) {
237 if (Change.Tok->is(TT_LineComment) || !Change.IsInsideToken) {
238 LastBlockComment = &Change;
239 } else if ((Change.StartOfBlockComment = LastBlockComment)) {
240 Change.IndentationOffset =
241 Change.StartOfTokenColumn -
242 Change.StartOfBlockComment->StartOfTokenColumn;
244 } else {
245 LastBlockComment = nullptr;
249 // Compute conditional nesting level
250 // Level is increased for each conditional, unless this conditional continues
251 // a chain of conditional, i.e. starts immediately after the colon of another
252 // conditional.
253 SmallVector<bool, 16> ScopeStack;
254 int ConditionalsLevel = 0;
255 for (auto &Change : Changes) {
256 for (unsigned i = 0, e = Change.Tok->FakeLParens.size(); i != e; ++i) {
257 bool isNestedConditional =
258 Change.Tok->FakeLParens[e - 1 - i] == prec::Conditional &&
259 !(i == 0 && Change.Tok->Previous &&
260 Change.Tok->Previous->is(TT_ConditionalExpr) &&
261 Change.Tok->Previous->is(tok::colon));
262 if (isNestedConditional)
263 ++ConditionalsLevel;
264 ScopeStack.push_back(isNestedConditional);
267 Change.ConditionalsLevel = ConditionalsLevel;
269 for (unsigned i = Change.Tok->FakeRParens; i > 0 && ScopeStack.size(); --i)
270 if (ScopeStack.pop_back_val())
271 --ConditionalsLevel;
275 // Align a single sequence of tokens, see AlignTokens below.
276 // Column - The token for which Matches returns true is moved to this column.
277 // RightJustify - Whether it is the token's right end or left end that gets
278 // moved to that column.
279 template <typename F>
280 static void
281 AlignTokenSequence(const FormatStyle &Style, unsigned Start, unsigned End,
282 unsigned Column, bool RightJustify, F &&Matches,
283 SmallVector<WhitespaceManager::Change, 16> &Changes) {
284 bool FoundMatchOnLine = false;
285 int Shift = 0;
287 // ScopeStack keeps track of the current scope depth. It contains indices of
288 // the first token on each scope.
289 // We only run the "Matches" function on tokens from the outer-most scope.
290 // However, we do need to pay special attention to one class of tokens
291 // that are not in the outer-most scope, and that is function parameters
292 // which are split across multiple lines, as illustrated by this example:
293 // double a(int x);
294 // int b(int y,
295 // double z);
296 // In the above example, we need to take special care to ensure that
297 // 'double z' is indented along with it's owning function 'b'.
298 // The same holds for calling a function:
299 // double a = foo(x);
300 // int b = bar(foo(y),
301 // foor(z));
302 // Similar for broken string literals:
303 // double x = 3.14;
304 // auto s = "Hello"
305 // "World";
306 // Special handling is required for 'nested' ternary operators.
307 SmallVector<unsigned, 16> ScopeStack;
309 for (unsigned i = Start; i != End; ++i) {
310 auto &CurrentChange = Changes[i];
311 if (ScopeStack.size() != 0 &&
312 CurrentChange.indentAndNestingLevel() <
313 Changes[ScopeStack.back()].indentAndNestingLevel()) {
314 ScopeStack.pop_back();
317 // Compare current token to previous non-comment token to ensure whether
318 // it is in a deeper scope or not.
319 unsigned PreviousNonComment = i - 1;
320 while (PreviousNonComment > Start &&
321 Changes[PreviousNonComment].Tok->is(tok::comment)) {
322 --PreviousNonComment;
324 if (i != Start && CurrentChange.indentAndNestingLevel() >
325 Changes[PreviousNonComment].indentAndNestingLevel()) {
326 ScopeStack.push_back(i);
329 bool InsideNestedScope = ScopeStack.size() != 0;
330 bool ContinuedStringLiteral = i > Start &&
331 CurrentChange.Tok->is(tok::string_literal) &&
332 Changes[i - 1].Tok->is(tok::string_literal);
333 bool SkipMatchCheck = InsideNestedScope || ContinuedStringLiteral;
335 if (CurrentChange.NewlinesBefore > 0 && !SkipMatchCheck) {
336 Shift = 0;
337 FoundMatchOnLine = false;
340 // If this is the first matching token to be aligned, remember by how many
341 // spaces it has to be shifted, so the rest of the changes on the line are
342 // shifted by the same amount
343 if (!FoundMatchOnLine && !SkipMatchCheck && Matches(CurrentChange)) {
344 FoundMatchOnLine = true;
345 Shift = Column - (RightJustify ? CurrentChange.TokenLength : 0) -
346 CurrentChange.StartOfTokenColumn;
347 CurrentChange.Spaces += Shift;
348 // FIXME: This is a workaround that should be removed when we fix
349 // http://llvm.org/PR53699. An assertion later below verifies this.
350 if (CurrentChange.NewlinesBefore == 0) {
351 CurrentChange.Spaces =
352 std::max(CurrentChange.Spaces,
353 static_cast<int>(CurrentChange.Tok->SpacesRequiredBefore));
357 if (Shift == 0)
358 continue;
360 // This is for function parameters that are split across multiple lines,
361 // as mentioned in the ScopeStack comment.
362 if (InsideNestedScope && CurrentChange.NewlinesBefore > 0) {
363 unsigned ScopeStart = ScopeStack.back();
364 auto ShouldShiftBeAdded = [&] {
365 // Function declaration
366 if (Changes[ScopeStart - 1].Tok->is(TT_FunctionDeclarationName))
367 return true;
369 // Lambda.
370 if (Changes[ScopeStart - 1].Tok->is(TT_LambdaLBrace))
371 return false;
373 // Continued function declaration
374 if (ScopeStart > Start + 1 &&
375 Changes[ScopeStart - 2].Tok->is(TT_FunctionDeclarationName)) {
376 return true;
379 // Continued (template) function call.
380 if (ScopeStart > Start + 1 &&
381 Changes[ScopeStart - 2].Tok->isOneOf(tok::identifier,
382 TT_TemplateCloser) &&
383 Changes[ScopeStart - 1].Tok->is(tok::l_paren) &&
384 Changes[ScopeStart].Tok->isNot(TT_LambdaLSquare)) {
385 if (CurrentChange.Tok->MatchingParen &&
386 CurrentChange.Tok->MatchingParen->is(TT_LambdaLBrace)) {
387 return false;
389 if (Changes[ScopeStart].NewlinesBefore > 0)
390 return false;
391 if (CurrentChange.Tok->is(tok::l_brace) &&
392 CurrentChange.Tok->is(BK_BracedInit)) {
393 return true;
395 return Style.BinPackArguments;
398 // Ternary operator
399 if (CurrentChange.Tok->is(TT_ConditionalExpr))
400 return true;
402 // Period Initializer .XXX = 1.
403 if (CurrentChange.Tok->is(TT_DesignatedInitializerPeriod))
404 return true;
406 // Continued ternary operator
407 if (CurrentChange.Tok->Previous &&
408 CurrentChange.Tok->Previous->is(TT_ConditionalExpr)) {
409 return true;
412 // Continued direct-list-initialization using braced list.
413 if (ScopeStart > Start + 1 &&
414 Changes[ScopeStart - 2].Tok->is(tok::identifier) &&
415 Changes[ScopeStart - 1].Tok->is(tok::l_brace) &&
416 CurrentChange.Tok->is(tok::l_brace) &&
417 CurrentChange.Tok->is(BK_BracedInit)) {
418 return true;
421 // Continued braced list.
422 if (ScopeStart > Start + 1 &&
423 Changes[ScopeStart - 2].Tok->isNot(tok::identifier) &&
424 Changes[ScopeStart - 1].Tok->is(tok::l_brace) &&
425 CurrentChange.Tok->isNot(tok::r_brace)) {
426 for (unsigned OuterScopeStart : llvm::reverse(ScopeStack)) {
427 // Lambda.
428 if (OuterScopeStart > Start &&
429 Changes[OuterScopeStart - 1].Tok->is(TT_LambdaLBrace)) {
430 return false;
433 if (Changes[ScopeStart].NewlinesBefore > 0)
434 return false;
435 return true;
438 // Continued template parameter.
439 if (Changes[ScopeStart - 1].Tok->is(TT_TemplateOpener))
440 return true;
442 return false;
445 if (ShouldShiftBeAdded())
446 CurrentChange.Spaces += Shift;
449 if (ContinuedStringLiteral)
450 CurrentChange.Spaces += Shift;
452 // We should not remove required spaces unless we break the line before.
453 assert(Shift > 0 || Changes[i].NewlinesBefore > 0 ||
454 CurrentChange.Spaces >=
455 static_cast<int>(Changes[i].Tok->SpacesRequiredBefore) ||
456 CurrentChange.Tok->is(tok::eof));
458 CurrentChange.StartOfTokenColumn += Shift;
459 if (i + 1 != Changes.size())
460 Changes[i + 1].PreviousEndOfTokenColumn += Shift;
462 // If PointerAlignment is PAS_Right, keep *s or &s next to the token
463 if ((Style.PointerAlignment == FormatStyle::PAS_Right ||
464 Style.ReferenceAlignment == FormatStyle::RAS_Right) &&
465 CurrentChange.Spaces != 0) {
466 const bool ReferenceNotRightAligned =
467 Style.ReferenceAlignment != FormatStyle::RAS_Right &&
468 Style.ReferenceAlignment != FormatStyle::RAS_Pointer;
469 for (int Previous = i - 1;
470 Previous >= 0 &&
471 Changes[Previous].Tok->getType() == TT_PointerOrReference;
472 --Previous) {
473 assert(
474 Changes[Previous].Tok->isOneOf(tok::star, tok::amp, tok::ampamp));
475 if (Changes[Previous].Tok->isNot(tok::star)) {
476 if (ReferenceNotRightAligned)
477 continue;
478 } else if (Style.PointerAlignment != FormatStyle::PAS_Right) {
479 continue;
481 Changes[Previous + 1].Spaces -= Shift;
482 Changes[Previous].Spaces += Shift;
483 Changes[Previous].StartOfTokenColumn += Shift;
489 // Walk through a subset of the changes, starting at StartAt, and find
490 // sequences of matching tokens to align. To do so, keep track of the lines and
491 // whether or not a matching token was found on a line. If a matching token is
492 // found, extend the current sequence. If the current line cannot be part of a
493 // sequence, e.g. because there is an empty line before it or it contains only
494 // non-matching tokens, finalize the previous sequence.
495 // The value returned is the token on which we stopped, either because we
496 // exhausted all items inside Changes, or because we hit a scope level higher
497 // than our initial scope.
498 // This function is recursive. Each invocation processes only the scope level
499 // equal to the initial level, which is the level of Changes[StartAt].
500 // If we encounter a scope level greater than the initial level, then we call
501 // ourselves recursively, thereby avoiding the pollution of the current state
502 // with the alignment requirements of the nested sub-level. This recursive
503 // behavior is necessary for aligning function prototypes that have one or more
504 // arguments.
505 // If this function encounters a scope level less than the initial level,
506 // it returns the current position.
507 // There is a non-obvious subtlety in the recursive behavior: Even though we
508 // defer processing of nested levels to recursive invocations of this
509 // function, when it comes time to align a sequence of tokens, we run the
510 // alignment on the entire sequence, including the nested levels.
511 // When doing so, most of the nested tokens are skipped, because their
512 // alignment was already handled by the recursive invocations of this function.
513 // However, the special exception is that we do NOT skip function parameters
514 // that are split across multiple lines. See the test case in FormatTest.cpp
515 // that mentions "split function parameter alignment" for an example of this.
516 // When the parameter RightJustify is true, the operator will be
517 // right-justified. It is used to align compound assignments like `+=` and `=`.
518 // When RightJustify and ACS.PadOperators are true, operators in each block to
519 // be aligned will be padded on the left to the same length before aligning.
520 template <typename F>
521 static unsigned AlignTokens(const FormatStyle &Style, F &&Matches,
522 SmallVector<WhitespaceManager::Change, 16> &Changes,
523 unsigned StartAt,
524 const FormatStyle::AlignConsecutiveStyle &ACS = {},
525 bool RightJustify = false) {
526 // We arrange each line in 3 parts. The operator to be aligned (the anchor),
527 // and text to its left and right. In the aligned text the width of each part
528 // will be the maximum of that over the block that has been aligned. Maximum
529 // widths of each part so far. When RightJustify is true and ACS.PadOperators
530 // is false, the part from start of line to the right end of the anchor.
531 // Otherwise, only the part to the left of the anchor. Including the space
532 // that exists on its left from the start. Not including the padding added on
533 // the left to right-justify the anchor.
534 unsigned WidthLeft = 0;
535 // The operator to be aligned when RightJustify is true and ACS.PadOperators
536 // is false. 0 otherwise.
537 unsigned WidthAnchor = 0;
538 // Width to the right of the anchor. Plus width of the anchor when
539 // RightJustify is false.
540 unsigned WidthRight = 0;
542 // Line number of the start and the end of the current token sequence.
543 unsigned StartOfSequence = 0;
544 unsigned EndOfSequence = 0;
546 // Measure the scope level (i.e. depth of (), [], {}) of the first token, and
547 // abort when we hit any token in a higher scope than the starting one.
548 auto IndentAndNestingLevel = StartAt < Changes.size()
549 ? Changes[StartAt].indentAndNestingLevel()
550 : std::tuple<unsigned, unsigned, unsigned>();
552 // Keep track of the number of commas before the matching tokens, we will only
553 // align a sequence of matching tokens if they are preceded by the same number
554 // of commas.
555 unsigned CommasBeforeLastMatch = 0;
556 unsigned CommasBeforeMatch = 0;
558 // Whether a matching token has been found on the current line.
559 bool FoundMatchOnLine = false;
561 // Whether the current line consists purely of comments.
562 bool LineIsComment = true;
564 // Aligns a sequence of matching tokens, on the MinColumn column.
566 // Sequences start from the first matching token to align, and end at the
567 // first token of the first line that doesn't need to be aligned.
569 // We need to adjust the StartOfTokenColumn of each Change that is on a line
570 // containing any matching token to be aligned and located after such token.
571 auto AlignCurrentSequence = [&] {
572 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) {
573 AlignTokenSequence(Style, StartOfSequence, EndOfSequence,
574 WidthLeft + WidthAnchor, RightJustify, Matches,
575 Changes);
577 WidthLeft = 0;
578 WidthAnchor = 0;
579 WidthRight = 0;
580 StartOfSequence = 0;
581 EndOfSequence = 0;
584 unsigned i = StartAt;
585 for (unsigned e = Changes.size(); i != e; ++i) {
586 auto &CurrentChange = Changes[i];
587 if (CurrentChange.indentAndNestingLevel() < IndentAndNestingLevel)
588 break;
590 if (CurrentChange.NewlinesBefore != 0) {
591 CommasBeforeMatch = 0;
592 EndOfSequence = i;
594 // Whether to break the alignment sequence because of an empty line.
595 bool EmptyLineBreak =
596 (CurrentChange.NewlinesBefore > 1) && !ACS.AcrossEmptyLines;
598 // Whether to break the alignment sequence because of a line without a
599 // match.
600 bool NoMatchBreak =
601 !FoundMatchOnLine && !(LineIsComment && ACS.AcrossComments);
603 if (EmptyLineBreak || NoMatchBreak)
604 AlignCurrentSequence();
606 // A new line starts, re-initialize line status tracking bools.
607 // Keep the match state if a string literal is continued on this line.
608 if (i == 0 || CurrentChange.Tok->isNot(tok::string_literal) ||
609 Changes[i - 1].Tok->isNot(tok::string_literal)) {
610 FoundMatchOnLine = false;
612 LineIsComment = true;
615 if (CurrentChange.Tok->isNot(tok::comment))
616 LineIsComment = false;
618 if (CurrentChange.Tok->is(tok::comma)) {
619 ++CommasBeforeMatch;
620 } else if (CurrentChange.indentAndNestingLevel() > IndentAndNestingLevel) {
621 // Call AlignTokens recursively, skipping over this scope block.
622 unsigned StoppedAt =
623 AlignTokens(Style, Matches, Changes, i, ACS, RightJustify);
624 i = StoppedAt - 1;
625 continue;
628 if (!Matches(CurrentChange))
629 continue;
631 // If there is more than one matching token per line, or if the number of
632 // preceding commas, do not match anymore, end the sequence.
633 if (FoundMatchOnLine || CommasBeforeMatch != CommasBeforeLastMatch)
634 AlignCurrentSequence();
636 CommasBeforeLastMatch = CommasBeforeMatch;
637 FoundMatchOnLine = true;
639 if (StartOfSequence == 0)
640 StartOfSequence = i;
642 unsigned ChangeWidthLeft = CurrentChange.StartOfTokenColumn;
643 unsigned ChangeWidthAnchor = 0;
644 unsigned ChangeWidthRight = 0;
645 if (RightJustify)
646 if (ACS.PadOperators)
647 ChangeWidthAnchor = CurrentChange.TokenLength;
648 else
649 ChangeWidthLeft += CurrentChange.TokenLength;
650 else
651 ChangeWidthRight = CurrentChange.TokenLength;
652 for (unsigned j = i + 1; j != e && Changes[j].NewlinesBefore == 0; ++j) {
653 ChangeWidthRight += Changes[j].Spaces;
654 // Changes are generally 1:1 with the tokens, but a change could also be
655 // inside of a token, in which case it's counted more than once: once for
656 // the whitespace surrounding the token (!IsInsideToken) and once for
657 // each whitespace change within it (IsInsideToken).
658 // Therefore, changes inside of a token should only count the space.
659 if (!Changes[j].IsInsideToken)
660 ChangeWidthRight += Changes[j].TokenLength;
663 // If we are restricted by the maximum column width, end the sequence.
664 unsigned NewLeft = std::max(ChangeWidthLeft, WidthLeft);
665 unsigned NewAnchor = std::max(ChangeWidthAnchor, WidthAnchor);
666 unsigned NewRight = std::max(ChangeWidthRight, WidthRight);
667 // `ColumnLimit == 0` means there is no column limit.
668 if (Style.ColumnLimit != 0 &&
669 Style.ColumnLimit < NewLeft + NewAnchor + NewRight) {
670 AlignCurrentSequence();
671 StartOfSequence = i;
672 WidthLeft = ChangeWidthLeft;
673 WidthAnchor = ChangeWidthAnchor;
674 WidthRight = ChangeWidthRight;
675 } else {
676 WidthLeft = NewLeft;
677 WidthAnchor = NewAnchor;
678 WidthRight = NewRight;
682 EndOfSequence = i;
683 AlignCurrentSequence();
684 return i;
687 // Aligns a sequence of matching tokens, on the MinColumn column.
689 // Sequences start from the first matching token to align, and end at the
690 // first token of the first line that doesn't need to be aligned.
692 // We need to adjust the StartOfTokenColumn of each Change that is on a line
693 // containing any matching token to be aligned and located after such token.
694 static void AlignMatchingTokenSequence(
695 unsigned &StartOfSequence, unsigned &EndOfSequence, unsigned &MinColumn,
696 std::function<bool(const WhitespaceManager::Change &C)> Matches,
697 SmallVector<WhitespaceManager::Change, 16> &Changes) {
698 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) {
699 bool FoundMatchOnLine = false;
700 int Shift = 0;
702 for (unsigned I = StartOfSequence; I != EndOfSequence; ++I) {
703 if (Changes[I].NewlinesBefore > 0) {
704 Shift = 0;
705 FoundMatchOnLine = false;
708 // If this is the first matching token to be aligned, remember by how many
709 // spaces it has to be shifted, so the rest of the changes on the line are
710 // shifted by the same amount.
711 if (!FoundMatchOnLine && Matches(Changes[I])) {
712 FoundMatchOnLine = true;
713 Shift = MinColumn - Changes[I].StartOfTokenColumn;
714 Changes[I].Spaces += Shift;
717 assert(Shift >= 0);
718 Changes[I].StartOfTokenColumn += Shift;
719 if (I + 1 != Changes.size())
720 Changes[I + 1].PreviousEndOfTokenColumn += Shift;
724 MinColumn = 0;
725 StartOfSequence = 0;
726 EndOfSequence = 0;
729 void WhitespaceManager::alignConsecutiveMacros() {
730 if (!Style.AlignConsecutiveMacros.Enabled)
731 return;
733 auto AlignMacrosMatches = [](const Change &C) {
734 const FormatToken *Current = C.Tok;
735 unsigned SpacesRequiredBefore = 1;
737 if (Current->SpacesRequiredBefore == 0 || !Current->Previous)
738 return false;
740 Current = Current->Previous;
742 // If token is a ")", skip over the parameter list, to the
743 // token that precedes the "("
744 if (Current->is(tok::r_paren) && Current->MatchingParen) {
745 Current = Current->MatchingParen->Previous;
746 SpacesRequiredBefore = 0;
749 if (!Current || Current->isNot(tok::identifier))
750 return false;
752 if (!Current->Previous || Current->Previous->isNot(tok::pp_define))
753 return false;
755 // For a macro function, 0 spaces are required between the
756 // identifier and the lparen that opens the parameter list.
757 // For a simple macro, 1 space is required between the
758 // identifier and the first token of the defined value.
759 return Current->Next->SpacesRequiredBefore == SpacesRequiredBefore;
762 unsigned MinColumn = 0;
764 // Start and end of the token sequence we're processing.
765 unsigned StartOfSequence = 0;
766 unsigned EndOfSequence = 0;
768 // Whether a matching token has been found on the current line.
769 bool FoundMatchOnLine = false;
771 // Whether the current line consists only of comments
772 bool LineIsComment = true;
774 unsigned I = 0;
775 for (unsigned E = Changes.size(); I != E; ++I) {
776 if (Changes[I].NewlinesBefore != 0) {
777 EndOfSequence = I;
779 // Whether to break the alignment sequence because of an empty line.
780 bool EmptyLineBreak = (Changes[I].NewlinesBefore > 1) &&
781 !Style.AlignConsecutiveMacros.AcrossEmptyLines;
783 // Whether to break the alignment sequence because of a line without a
784 // match.
785 bool NoMatchBreak =
786 !FoundMatchOnLine &&
787 !(LineIsComment && Style.AlignConsecutiveMacros.AcrossComments);
789 if (EmptyLineBreak || NoMatchBreak) {
790 AlignMatchingTokenSequence(StartOfSequence, EndOfSequence, MinColumn,
791 AlignMacrosMatches, Changes);
794 // A new line starts, re-initialize line status tracking bools.
795 FoundMatchOnLine = false;
796 LineIsComment = true;
799 if (Changes[I].Tok->isNot(tok::comment))
800 LineIsComment = false;
802 if (!AlignMacrosMatches(Changes[I]))
803 continue;
805 FoundMatchOnLine = true;
807 if (StartOfSequence == 0)
808 StartOfSequence = I;
810 unsigned ChangeMinColumn = Changes[I].StartOfTokenColumn;
811 MinColumn = std::max(MinColumn, ChangeMinColumn);
814 EndOfSequence = I;
815 AlignMatchingTokenSequence(StartOfSequence, EndOfSequence, MinColumn,
816 AlignMacrosMatches, Changes);
819 void WhitespaceManager::alignConsecutiveAssignments() {
820 if (!Style.AlignConsecutiveAssignments.Enabled)
821 return;
823 AlignTokens(
824 Style,
825 [&](const Change &C) {
826 // Do not align on equal signs that are first on a line.
827 if (C.NewlinesBefore > 0)
828 return false;
830 // Do not align on equal signs that are last on a line.
831 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
832 return false;
834 // Do not align operator= overloads.
835 FormatToken *Previous = C.Tok->getPreviousNonComment();
836 if (Previous && Previous->is(tok::kw_operator))
837 return false;
839 return Style.AlignConsecutiveAssignments.AlignCompound
840 ? C.Tok->getPrecedence() == prec::Assignment
841 : (C.Tok->is(tok::equal) ||
842 // In Verilog the '<=' is not a compound assignment, thus
843 // it is aligned even when the AlignCompound option is not
844 // set.
845 (Style.isVerilog() && C.Tok->is(tok::lessequal) &&
846 C.Tok->getPrecedence() == prec::Assignment));
848 Changes, /*StartAt=*/0, Style.AlignConsecutiveAssignments,
849 /*RightJustify=*/true);
852 void WhitespaceManager::alignConsecutiveBitFields() {
853 if (!Style.AlignConsecutiveBitFields.Enabled)
854 return;
856 AlignTokens(
857 Style,
858 [&](Change const &C) {
859 // Do not align on ':' that is first on a line.
860 if (C.NewlinesBefore > 0)
861 return false;
863 // Do not align on ':' that is last on a line.
864 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
865 return false;
867 return C.Tok->is(TT_BitFieldColon);
869 Changes, /*StartAt=*/0, Style.AlignConsecutiveBitFields);
872 void WhitespaceManager::alignConsecutiveShortCaseStatements() {
873 if (!Style.AlignConsecutiveShortCaseStatements.Enabled ||
874 !Style.AllowShortCaseLabelsOnASingleLine) {
875 return;
878 auto Matches = [&](const Change &C) {
879 if (Style.AlignConsecutiveShortCaseStatements.AlignCaseColons)
880 return C.Tok->is(TT_CaseLabelColon);
882 // Ignore 'IsInsideToken' to allow matching trailing comments which
883 // need to be reflowed as that causes the token to appear in two
884 // different changes, which will cause incorrect alignment as we'll
885 // reflow early due to detecting multiple aligning tokens per line.
886 return !C.IsInsideToken && C.Tok->Previous &&
887 C.Tok->Previous->is(TT_CaseLabelColon);
890 unsigned MinColumn = 0;
892 // Empty case statements don't break the alignment, but don't necessarily
893 // match our predicate, so we need to track their column so they can push out
894 // our alignment.
895 unsigned MinEmptyCaseColumn = 0;
897 // Start and end of the token sequence we're processing.
898 unsigned StartOfSequence = 0;
899 unsigned EndOfSequence = 0;
901 // Whether a matching token has been found on the current line.
902 bool FoundMatchOnLine = false;
904 bool LineIsComment = true;
905 bool LineIsEmptyCase = false;
907 unsigned I = 0;
908 for (unsigned E = Changes.size(); I != E; ++I) {
909 if (Changes[I].NewlinesBefore != 0) {
910 // Whether to break the alignment sequence because of an empty line.
911 bool EmptyLineBreak =
912 (Changes[I].NewlinesBefore > 1) &&
913 !Style.AlignConsecutiveShortCaseStatements.AcrossEmptyLines;
915 // Whether to break the alignment sequence because of a line without a
916 // match.
917 bool NoMatchBreak =
918 !FoundMatchOnLine &&
919 !(LineIsComment &&
920 Style.AlignConsecutiveShortCaseStatements.AcrossComments) &&
921 !LineIsEmptyCase;
923 if (EmptyLineBreak || NoMatchBreak) {
924 AlignMatchingTokenSequence(StartOfSequence, EndOfSequence, MinColumn,
925 Matches, Changes);
926 MinEmptyCaseColumn = 0;
929 // A new line starts, re-initialize line status tracking bools.
930 FoundMatchOnLine = false;
931 LineIsComment = true;
932 LineIsEmptyCase = false;
935 if (Changes[I].Tok->isNot(tok::comment))
936 LineIsComment = false;
938 if (Changes[I].Tok->is(TT_CaseLabelColon)) {
939 LineIsEmptyCase =
940 !Changes[I].Tok->Next || Changes[I].Tok->Next->isTrailingComment();
942 if (LineIsEmptyCase) {
943 if (Style.AlignConsecutiveShortCaseStatements.AlignCaseColons) {
944 MinEmptyCaseColumn =
945 std::max(MinEmptyCaseColumn, Changes[I].StartOfTokenColumn);
946 } else {
947 MinEmptyCaseColumn =
948 std::max(MinEmptyCaseColumn, Changes[I].StartOfTokenColumn + 2);
953 if (!Matches(Changes[I]))
954 continue;
956 if (LineIsEmptyCase)
957 continue;
959 FoundMatchOnLine = true;
961 if (StartOfSequence == 0)
962 StartOfSequence = I;
964 EndOfSequence = I + 1;
966 MinColumn = std::max(MinColumn, Changes[I].StartOfTokenColumn);
968 // Allow empty case statements to push out our alignment.
969 MinColumn = std::max(MinColumn, MinEmptyCaseColumn);
972 AlignMatchingTokenSequence(StartOfSequence, EndOfSequence, MinColumn, Matches,
973 Changes);
976 void WhitespaceManager::alignConsecutiveDeclarations() {
977 if (!Style.AlignConsecutiveDeclarations.Enabled)
978 return;
980 AlignTokens(
981 Style,
982 [](Change const &C) {
983 if (C.Tok->is(TT_FunctionDeclarationName))
984 return true;
985 if (C.Tok->isNot(TT_StartOfName))
986 return false;
987 if (C.Tok->Previous &&
988 C.Tok->Previous->is(TT_StatementAttributeLikeMacro))
989 return false;
990 // Check if there is a subsequent name that starts the same declaration.
991 for (FormatToken *Next = C.Tok->Next; Next; Next = Next->Next) {
992 if (Next->is(tok::comment))
993 continue;
994 if (Next->is(TT_PointerOrReference))
995 return false;
996 if (!Next->Tok.getIdentifierInfo())
997 break;
998 if (Next->isOneOf(TT_StartOfName, TT_FunctionDeclarationName,
999 tok::kw_operator)) {
1000 return false;
1003 return true;
1005 Changes, /*StartAt=*/0, Style.AlignConsecutiveDeclarations);
1008 void WhitespaceManager::alignChainedConditionals() {
1009 if (Style.BreakBeforeTernaryOperators) {
1010 AlignTokens(
1011 Style,
1012 [](Change const &C) {
1013 // Align question operators and last colon
1014 return C.Tok->is(TT_ConditionalExpr) &&
1015 ((C.Tok->is(tok::question) && !C.NewlinesBefore) ||
1016 (C.Tok->is(tok::colon) && C.Tok->Next &&
1017 (C.Tok->Next->FakeLParens.size() == 0 ||
1018 C.Tok->Next->FakeLParens.back() != prec::Conditional)));
1020 Changes, /*StartAt=*/0);
1021 } else {
1022 static auto AlignWrappedOperand = [](Change const &C) {
1023 FormatToken *Previous = C.Tok->getPreviousNonComment();
1024 return C.NewlinesBefore && Previous && Previous->is(TT_ConditionalExpr) &&
1025 (Previous->is(tok::colon) &&
1026 (C.Tok->FakeLParens.size() == 0 ||
1027 C.Tok->FakeLParens.back() != prec::Conditional));
1029 // Ensure we keep alignment of wrapped operands with non-wrapped operands
1030 // Since we actually align the operators, the wrapped operands need the
1031 // extra offset to be properly aligned.
1032 for (Change &C : Changes)
1033 if (AlignWrappedOperand(C))
1034 C.StartOfTokenColumn -= 2;
1035 AlignTokens(
1036 Style,
1037 [this](Change const &C) {
1038 // Align question operators if next operand is not wrapped, as
1039 // well as wrapped operands after question operator or last
1040 // colon in conditional sequence
1041 return (C.Tok->is(TT_ConditionalExpr) && C.Tok->is(tok::question) &&
1042 &C != &Changes.back() && (&C + 1)->NewlinesBefore == 0 &&
1043 !(&C + 1)->IsTrailingComment) ||
1044 AlignWrappedOperand(C);
1046 Changes, /*StartAt=*/0);
1050 void WhitespaceManager::alignTrailingComments() {
1051 if (Style.AlignTrailingComments.Kind == FormatStyle::TCAS_Never)
1052 return;
1054 const int Size = Changes.size();
1055 int MinColumn = 0;
1056 int StartOfSequence = 0;
1057 bool BreakBeforeNext = false;
1058 int NewLineThreshold = 1;
1059 if (Style.AlignTrailingComments.Kind == FormatStyle::TCAS_Always)
1060 NewLineThreshold = Style.AlignTrailingComments.OverEmptyLines + 1;
1062 for (int I = 0, MaxColumn = INT_MAX, Newlines = 0; I < Size; ++I) {
1063 auto &C = Changes[I];
1064 if (C.StartOfBlockComment)
1065 continue;
1066 Newlines += C.NewlinesBefore;
1067 if (!C.IsTrailingComment)
1068 continue;
1070 if (Style.AlignTrailingComments.Kind == FormatStyle::TCAS_Leave) {
1071 const int OriginalSpaces =
1072 C.OriginalWhitespaceRange.getEnd().getRawEncoding() -
1073 C.OriginalWhitespaceRange.getBegin().getRawEncoding() -
1074 C.Tok->LastNewlineOffset;
1075 assert(OriginalSpaces >= 0);
1076 const auto RestoredLineLength =
1077 C.StartOfTokenColumn + C.TokenLength + OriginalSpaces;
1078 // If leaving comments makes the line exceed the column limit, give up to
1079 // leave the comments.
1080 if (RestoredLineLength >= Style.ColumnLimit && Style.ColumnLimit > 0)
1081 break;
1082 C.Spaces = OriginalSpaces;
1083 continue;
1086 const int ChangeMinColumn = C.StartOfTokenColumn;
1087 int ChangeMaxColumn;
1089 // If we don't create a replacement for this change, we have to consider
1090 // it to be immovable.
1091 if (!C.CreateReplacement)
1092 ChangeMaxColumn = ChangeMinColumn;
1093 else if (Style.ColumnLimit == 0)
1094 ChangeMaxColumn = INT_MAX;
1095 else if (Style.ColumnLimit >= C.TokenLength)
1096 ChangeMaxColumn = Style.ColumnLimit - C.TokenLength;
1097 else
1098 ChangeMaxColumn = ChangeMinColumn;
1100 if (I + 1 < Size && Changes[I + 1].ContinuesPPDirective &&
1101 ChangeMaxColumn >= 2) {
1102 ChangeMaxColumn -= 2;
1105 bool WasAlignedWithStartOfNextLine = false;
1106 if (C.NewlinesBefore >= 1) { // A comment on its own line.
1107 const auto CommentColumn =
1108 SourceMgr.getSpellingColumnNumber(C.OriginalWhitespaceRange.getEnd());
1109 for (int J = I + 1; J < Size; ++J) {
1110 if (Changes[J].Tok->is(tok::comment))
1111 continue;
1113 const auto NextColumn = SourceMgr.getSpellingColumnNumber(
1114 Changes[J].OriginalWhitespaceRange.getEnd());
1115 // The start of the next token was previously aligned with the
1116 // start of this comment.
1117 WasAlignedWithStartOfNextLine =
1118 CommentColumn == NextColumn ||
1119 CommentColumn == NextColumn + Style.IndentWidth;
1120 break;
1124 // We don't want to align comments which end a scope, which are here
1125 // identified by most closing braces.
1126 auto DontAlignThisComment = [](const auto *Tok) {
1127 if (Tok->is(tok::semi)) {
1128 Tok = Tok->getPreviousNonComment();
1129 if (!Tok)
1130 return false;
1132 if (Tok->is(tok::r_paren)) {
1133 // Back up past the parentheses and a `TT_DoWhile` that may precede.
1134 Tok = Tok->MatchingParen;
1135 if (!Tok)
1136 return false;
1137 Tok = Tok->getPreviousNonComment();
1138 if (!Tok)
1139 return false;
1140 if (Tok->is(TT_DoWhile)) {
1141 const auto *Prev = Tok->getPreviousNonComment();
1142 if (!Prev) {
1143 // A do-while-loop without braces.
1144 return true;
1146 Tok = Prev;
1150 if (Tok->isNot(tok::r_brace))
1151 return false;
1153 while (Tok->Previous && Tok->Previous->is(tok::r_brace))
1154 Tok = Tok->Previous;
1155 return Tok->NewlinesBefore > 0;
1158 if (I > 0 && C.NewlinesBefore == 0 &&
1159 DontAlignThisComment(Changes[I - 1].Tok)) {
1160 alignTrailingComments(StartOfSequence, I, MinColumn);
1161 // Reset to initial values, but skip this change for the next alignment
1162 // pass.
1163 MinColumn = 0;
1164 MaxColumn = INT_MAX;
1165 StartOfSequence = I + 1;
1166 } else if (BreakBeforeNext || Newlines > NewLineThreshold ||
1167 (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn) ||
1168 // Break the comment sequence if the previous line did not end
1169 // in a trailing comment.
1170 (C.NewlinesBefore == 1 && I > 0 &&
1171 !Changes[I - 1].IsTrailingComment) ||
1172 WasAlignedWithStartOfNextLine) {
1173 alignTrailingComments(StartOfSequence, I, MinColumn);
1174 MinColumn = ChangeMinColumn;
1175 MaxColumn = ChangeMaxColumn;
1176 StartOfSequence = I;
1177 } else {
1178 MinColumn = std::max(MinColumn, ChangeMinColumn);
1179 MaxColumn = std::min(MaxColumn, ChangeMaxColumn);
1181 BreakBeforeNext = (I == 0) || (C.NewlinesBefore > 1) ||
1182 // Never start a sequence with a comment at the beginning
1183 // of the line.
1184 (C.NewlinesBefore == 1 && StartOfSequence == I);
1185 Newlines = 0;
1187 alignTrailingComments(StartOfSequence, Size, MinColumn);
1190 void WhitespaceManager::alignTrailingComments(unsigned Start, unsigned End,
1191 unsigned Column) {
1192 for (unsigned i = Start; i != End; ++i) {
1193 int Shift = 0;
1194 if (Changes[i].IsTrailingComment)
1195 Shift = Column - Changes[i].StartOfTokenColumn;
1196 if (Changes[i].StartOfBlockComment) {
1197 Shift = Changes[i].IndentationOffset +
1198 Changes[i].StartOfBlockComment->StartOfTokenColumn -
1199 Changes[i].StartOfTokenColumn;
1201 if (Shift <= 0)
1202 continue;
1203 Changes[i].Spaces += Shift;
1204 if (i + 1 != Changes.size())
1205 Changes[i + 1].PreviousEndOfTokenColumn += Shift;
1206 Changes[i].StartOfTokenColumn += Shift;
1210 void WhitespaceManager::alignEscapedNewlines() {
1211 if (Style.AlignEscapedNewlines == FormatStyle::ENAS_DontAlign)
1212 return;
1214 bool AlignLeft = Style.AlignEscapedNewlines == FormatStyle::ENAS_Left;
1215 unsigned MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit;
1216 unsigned StartOfMacro = 0;
1217 for (unsigned i = 1, e = Changes.size(); i < e; ++i) {
1218 Change &C = Changes[i];
1219 if (C.NewlinesBefore > 0) {
1220 if (C.ContinuesPPDirective) {
1221 MaxEndOfLine = std::max(C.PreviousEndOfTokenColumn + 2, MaxEndOfLine);
1222 } else {
1223 alignEscapedNewlines(StartOfMacro + 1, i, MaxEndOfLine);
1224 MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit;
1225 StartOfMacro = i;
1229 alignEscapedNewlines(StartOfMacro + 1, Changes.size(), MaxEndOfLine);
1232 void WhitespaceManager::alignEscapedNewlines(unsigned Start, unsigned End,
1233 unsigned Column) {
1234 for (unsigned i = Start; i < End; ++i) {
1235 Change &C = Changes[i];
1236 if (C.NewlinesBefore > 0) {
1237 assert(C.ContinuesPPDirective);
1238 if (C.PreviousEndOfTokenColumn + 1 > Column)
1239 C.EscapedNewlineColumn = 0;
1240 else
1241 C.EscapedNewlineColumn = Column;
1246 void WhitespaceManager::alignArrayInitializers() {
1247 if (Style.AlignArrayOfStructures == FormatStyle::AIAS_None)
1248 return;
1250 for (unsigned ChangeIndex = 1U, ChangeEnd = Changes.size();
1251 ChangeIndex < ChangeEnd; ++ChangeIndex) {
1252 auto &C = Changes[ChangeIndex];
1253 if (C.Tok->IsArrayInitializer) {
1254 bool FoundComplete = false;
1255 for (unsigned InsideIndex = ChangeIndex + 1; InsideIndex < ChangeEnd;
1256 ++InsideIndex) {
1257 if (Changes[InsideIndex].Tok == C.Tok->MatchingParen) {
1258 alignArrayInitializers(ChangeIndex, InsideIndex + 1);
1259 ChangeIndex = InsideIndex + 1;
1260 FoundComplete = true;
1261 break;
1264 if (!FoundComplete)
1265 ChangeIndex = ChangeEnd;
1270 void WhitespaceManager::alignArrayInitializers(unsigned Start, unsigned End) {
1272 if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Right)
1273 alignArrayInitializersRightJustified(getCells(Start, End));
1274 else if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Left)
1275 alignArrayInitializersLeftJustified(getCells(Start, End));
1278 void WhitespaceManager::alignArrayInitializersRightJustified(
1279 CellDescriptions &&CellDescs) {
1280 if (!CellDescs.isRectangular())
1281 return;
1283 const int BracePadding = Style.Cpp11BracedListStyle ? 0 : 1;
1284 auto &Cells = CellDescs.Cells;
1285 // Now go through and fixup the spaces.
1286 auto *CellIter = Cells.begin();
1287 for (auto i = 0U; i < CellDescs.CellCounts[0]; ++i, ++CellIter) {
1288 unsigned NetWidth = 0U;
1289 if (isSplitCell(*CellIter))
1290 NetWidth = getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces);
1291 auto CellWidth = getMaximumCellWidth(CellIter, NetWidth);
1293 if (Changes[CellIter->Index].Tok->is(tok::r_brace)) {
1294 // So in here we want to see if there is a brace that falls
1295 // on a line that was split. If so on that line we make sure that
1296 // the spaces in front of the brace are enough.
1297 const auto *Next = CellIter;
1298 do {
1299 const FormatToken *Previous = Changes[Next->Index].Tok->Previous;
1300 if (Previous && Previous->isNot(TT_LineComment)) {
1301 Changes[Next->Index].Spaces = BracePadding;
1302 Changes[Next->Index].NewlinesBefore = 0;
1304 Next = Next->NextColumnElement;
1305 } while (Next);
1306 // Unless the array is empty, we need the position of all the
1307 // immediately adjacent cells
1308 if (CellIter != Cells.begin()) {
1309 auto ThisNetWidth =
1310 getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces);
1311 auto MaxNetWidth = getMaximumNetWidth(
1312 Cells.begin(), CellIter, CellDescs.InitialSpaces,
1313 CellDescs.CellCounts[0], CellDescs.CellCounts.size());
1314 if (ThisNetWidth < MaxNetWidth)
1315 Changes[CellIter->Index].Spaces = (MaxNetWidth - ThisNetWidth);
1316 auto RowCount = 1U;
1317 auto Offset = std::distance(Cells.begin(), CellIter);
1318 for (const auto *Next = CellIter->NextColumnElement; Next;
1319 Next = Next->NextColumnElement) {
1320 auto *Start = (Cells.begin() + RowCount * CellDescs.CellCounts[0]);
1321 auto *End = Start + Offset;
1322 ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces);
1323 if (ThisNetWidth < MaxNetWidth)
1324 Changes[Next->Index].Spaces = (MaxNetWidth - ThisNetWidth);
1325 ++RowCount;
1328 } else {
1329 auto ThisWidth =
1330 calculateCellWidth(CellIter->Index, CellIter->EndIndex, true) +
1331 NetWidth;
1332 if (Changes[CellIter->Index].NewlinesBefore == 0) {
1333 Changes[CellIter->Index].Spaces = (CellWidth - (ThisWidth + NetWidth));
1334 Changes[CellIter->Index].Spaces += (i > 0) ? 1 : BracePadding;
1336 alignToStartOfCell(CellIter->Index, CellIter->EndIndex);
1337 for (const auto *Next = CellIter->NextColumnElement; Next;
1338 Next = Next->NextColumnElement) {
1339 ThisWidth =
1340 calculateCellWidth(Next->Index, Next->EndIndex, true) + NetWidth;
1341 if (Changes[Next->Index].NewlinesBefore == 0) {
1342 Changes[Next->Index].Spaces = (CellWidth - ThisWidth);
1343 Changes[Next->Index].Spaces += (i > 0) ? 1 : BracePadding;
1345 alignToStartOfCell(Next->Index, Next->EndIndex);
1351 void WhitespaceManager::alignArrayInitializersLeftJustified(
1352 CellDescriptions &&CellDescs) {
1354 if (!CellDescs.isRectangular())
1355 return;
1357 const int BracePadding = Style.Cpp11BracedListStyle ? 0 : 1;
1358 auto &Cells = CellDescs.Cells;
1359 // Now go through and fixup the spaces.
1360 auto *CellIter = Cells.begin();
1361 // The first cell needs to be against the left brace.
1362 if (Changes[CellIter->Index].NewlinesBefore == 0)
1363 Changes[CellIter->Index].Spaces = BracePadding;
1364 else
1365 Changes[CellIter->Index].Spaces = CellDescs.InitialSpaces;
1366 ++CellIter;
1367 for (auto i = 1U; i < CellDescs.CellCounts[0]; i++, ++CellIter) {
1368 auto MaxNetWidth = getMaximumNetWidth(
1369 Cells.begin(), CellIter, CellDescs.InitialSpaces,
1370 CellDescs.CellCounts[0], CellDescs.CellCounts.size());
1371 auto ThisNetWidth =
1372 getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces);
1373 if (Changes[CellIter->Index].NewlinesBefore == 0) {
1374 Changes[CellIter->Index].Spaces =
1375 MaxNetWidth - ThisNetWidth +
1376 (Changes[CellIter->Index].Tok->isNot(tok::r_brace) ? 1
1377 : BracePadding);
1379 auto RowCount = 1U;
1380 auto Offset = std::distance(Cells.begin(), CellIter);
1381 for (const auto *Next = CellIter->NextColumnElement; Next;
1382 Next = Next->NextColumnElement) {
1383 if (RowCount > CellDescs.CellCounts.size())
1384 break;
1385 auto *Start = (Cells.begin() + RowCount * CellDescs.CellCounts[0]);
1386 auto *End = Start + Offset;
1387 auto ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces);
1388 if (Changes[Next->Index].NewlinesBefore == 0) {
1389 Changes[Next->Index].Spaces =
1390 MaxNetWidth - ThisNetWidth +
1391 (Changes[Next->Index].Tok->isNot(tok::r_brace) ? 1 : BracePadding);
1393 ++RowCount;
1398 bool WhitespaceManager::isSplitCell(const CellDescription &Cell) {
1399 if (Cell.HasSplit)
1400 return true;
1401 for (const auto *Next = Cell.NextColumnElement; Next;
1402 Next = Next->NextColumnElement) {
1403 if (Next->HasSplit)
1404 return true;
1406 return false;
1409 WhitespaceManager::CellDescriptions WhitespaceManager::getCells(unsigned Start,
1410 unsigned End) {
1412 unsigned Depth = 0;
1413 unsigned Cell = 0;
1414 SmallVector<unsigned> CellCounts;
1415 unsigned InitialSpaces = 0;
1416 unsigned InitialTokenLength = 0;
1417 unsigned EndSpaces = 0;
1418 SmallVector<CellDescription> Cells;
1419 const FormatToken *MatchingParen = nullptr;
1420 for (unsigned i = Start; i < End; ++i) {
1421 auto &C = Changes[i];
1422 if (C.Tok->is(tok::l_brace))
1423 ++Depth;
1424 else if (C.Tok->is(tok::r_brace))
1425 --Depth;
1426 if (Depth == 2) {
1427 if (C.Tok->is(tok::l_brace)) {
1428 Cell = 0;
1429 MatchingParen = C.Tok->MatchingParen;
1430 if (InitialSpaces == 0) {
1431 InitialSpaces = C.Spaces + C.TokenLength;
1432 InitialTokenLength = C.TokenLength;
1433 auto j = i - 1;
1434 for (; Changes[j].NewlinesBefore == 0 && j > Start; --j) {
1435 InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength;
1436 InitialTokenLength += Changes[j].TokenLength;
1438 if (C.NewlinesBefore == 0) {
1439 InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength;
1440 InitialTokenLength += Changes[j].TokenLength;
1443 } else if (C.Tok->is(tok::comma)) {
1444 if (!Cells.empty())
1445 Cells.back().EndIndex = i;
1446 if (C.Tok->getNextNonComment()->isNot(tok::r_brace)) // dangling comma
1447 ++Cell;
1449 } else if (Depth == 1) {
1450 if (C.Tok == MatchingParen) {
1451 if (!Cells.empty())
1452 Cells.back().EndIndex = i;
1453 Cells.push_back(CellDescription{i, ++Cell, i + 1, false, nullptr});
1454 CellCounts.push_back(C.Tok->Previous->isNot(tok::comma) ? Cell + 1
1455 : Cell);
1456 // Go to the next non-comment and ensure there is a break in front
1457 const auto *NextNonComment = C.Tok->getNextNonComment();
1458 while (NextNonComment->is(tok::comma))
1459 NextNonComment = NextNonComment->getNextNonComment();
1460 auto j = i;
1461 while (Changes[j].Tok != NextNonComment && j < End)
1462 ++j;
1463 if (j < End && Changes[j].NewlinesBefore == 0 &&
1464 Changes[j].Tok->isNot(tok::r_brace)) {
1465 Changes[j].NewlinesBefore = 1;
1466 // Account for the added token lengths
1467 Changes[j].Spaces = InitialSpaces - InitialTokenLength;
1469 } else if (C.Tok->is(tok::comment)) {
1470 // Trailing comments stay at a space past the last token
1471 C.Spaces = Changes[i - 1].Tok->is(tok::comma) ? 1 : 2;
1472 } else if (C.Tok->is(tok::l_brace)) {
1473 // We need to make sure that the ending braces is aligned to the
1474 // start of our initializer
1475 auto j = i - 1;
1476 for (; j > 0 && !Changes[j].Tok->ArrayInitializerLineStart; --j)
1477 ; // Nothing the loop does the work
1478 EndSpaces = Changes[j].Spaces;
1480 } else if (Depth == 0 && C.Tok->is(tok::r_brace)) {
1481 C.NewlinesBefore = 1;
1482 C.Spaces = EndSpaces;
1484 if (C.Tok->StartsColumn) {
1485 // This gets us past tokens that have been split over multiple
1486 // lines
1487 bool HasSplit = false;
1488 if (Changes[i].NewlinesBefore > 0) {
1489 // So if we split a line previously and the tail line + this token is
1490 // less then the column limit we remove the split here and just put
1491 // the column start at a space past the comma
1493 // FIXME This if branch covers the cases where the column is not
1494 // the first column. This leads to weird pathologies like the formatting
1495 // auto foo = Items{
1496 // Section{
1497 // 0, bar(),
1498 // }
1499 // };
1500 // Well if it doesn't lead to that it's indicative that the line
1501 // breaking should be revisited. Unfortunately alot of other options
1502 // interact with this
1503 auto j = i - 1;
1504 if ((j - 1) > Start && Changes[j].Tok->is(tok::comma) &&
1505 Changes[j - 1].NewlinesBefore > 0) {
1506 --j;
1507 auto LineLimit = Changes[j].Spaces + Changes[j].TokenLength;
1508 if (LineLimit < Style.ColumnLimit) {
1509 Changes[i].NewlinesBefore = 0;
1510 Changes[i].Spaces = 1;
1514 while (Changes[i].NewlinesBefore > 0 && Changes[i].Tok == C.Tok) {
1515 Changes[i].Spaces = InitialSpaces;
1516 ++i;
1517 HasSplit = true;
1519 if (Changes[i].Tok != C.Tok)
1520 --i;
1521 Cells.push_back(CellDescription{i, Cell, i, HasSplit, nullptr});
1525 return linkCells({Cells, CellCounts, InitialSpaces});
1528 unsigned WhitespaceManager::calculateCellWidth(unsigned Start, unsigned End,
1529 bool WithSpaces) const {
1530 unsigned CellWidth = 0;
1531 for (auto i = Start; i < End; i++) {
1532 if (Changes[i].NewlinesBefore > 0)
1533 CellWidth = 0;
1534 CellWidth += Changes[i].TokenLength;
1535 CellWidth += (WithSpaces ? Changes[i].Spaces : 0);
1537 return CellWidth;
1540 void WhitespaceManager::alignToStartOfCell(unsigned Start, unsigned End) {
1541 if ((End - Start) <= 1)
1542 return;
1543 // If the line is broken anywhere in there make sure everything
1544 // is aligned to the parent
1545 for (auto i = Start + 1; i < End; i++)
1546 if (Changes[i].NewlinesBefore > 0)
1547 Changes[i].Spaces = Changes[Start].Spaces;
1550 WhitespaceManager::CellDescriptions
1551 WhitespaceManager::linkCells(CellDescriptions &&CellDesc) {
1552 auto &Cells = CellDesc.Cells;
1553 for (auto *CellIter = Cells.begin(); CellIter != Cells.end(); ++CellIter) {
1554 if (!CellIter->NextColumnElement && (CellIter + 1) != Cells.end()) {
1555 for (auto *NextIter = CellIter + 1; NextIter != Cells.end(); ++NextIter) {
1556 if (NextIter->Cell == CellIter->Cell) {
1557 CellIter->NextColumnElement = &(*NextIter);
1558 break;
1563 return std::move(CellDesc);
1566 void WhitespaceManager::generateChanges() {
1567 for (unsigned i = 0, e = Changes.size(); i != e; ++i) {
1568 const Change &C = Changes[i];
1569 if (i > 0) {
1570 auto Last = Changes[i - 1].OriginalWhitespaceRange;
1571 auto New = Changes[i].OriginalWhitespaceRange;
1572 // Do not generate two replacements for the same location. As a special
1573 // case, it is allowed if there is a replacement for the empty range
1574 // between 2 tokens and another non-empty range at the start of the second
1575 // token. We didn't implement logic to combine replacements for 2
1576 // consecutive source ranges into a single replacement, because the
1577 // program works fine without it.
1579 // We can't eliminate empty original whitespace ranges. They appear when
1580 // 2 tokens have no whitespace in between in the input. It does not
1581 // matter whether whitespace is to be added. If no whitespace is to be
1582 // added, the replacement will be empty, and it gets eliminated after this
1583 // step in storeReplacement. For example, if the input is `foo();`,
1584 // there will be a replacement for the range between every consecutive
1585 // pair of tokens.
1587 // A replacement at the start of a token can be added by
1588 // BreakableStringLiteralUsingOperators::insertBreak when it adds braces
1589 // around the string literal. Say Verilog code is being formatted and the
1590 // first line is to become the next 2 lines.
1591 // x("long string");
1592 // x({"long ",
1593 // "string"});
1594 // There will be a replacement for the empty range between the parenthesis
1595 // and the string and another replacement for the quote character. The
1596 // replacement for the empty range between the parenthesis and the quote
1597 // comes from ContinuationIndenter::addTokenOnCurrentLine when it changes
1598 // the original empty range between the parenthesis and the string to
1599 // another empty one. The replacement for the quote character comes from
1600 // BreakableStringLiteralUsingOperators::insertBreak when it adds the
1601 // brace. In the example, the replacement for the empty range is the same
1602 // as the original text. However, eliminating replacements that are same
1603 // as the original does not help in general. For example, a newline can
1604 // be inserted, causing the first line to become the next 3 lines.
1605 // xxxxxxxxxxx("long string");
1606 // xxxxxxxxxxx(
1607 // {"long ",
1608 // "string"});
1609 // In that case, the empty range between the parenthesis and the string
1610 // will be replaced by a newline and 4 spaces. So we will still have to
1611 // deal with a replacement for an empty source range followed by a
1612 // replacement for a non-empty source range.
1613 if (Last.getBegin() == New.getBegin() &&
1614 (Last.getEnd() != Last.getBegin() ||
1615 New.getEnd() == New.getBegin())) {
1616 continue;
1619 if (C.CreateReplacement) {
1620 std::string ReplacementText = C.PreviousLinePostfix;
1621 if (C.ContinuesPPDirective) {
1622 appendEscapedNewlineText(ReplacementText, C.NewlinesBefore,
1623 C.PreviousEndOfTokenColumn,
1624 C.EscapedNewlineColumn);
1625 } else {
1626 appendNewlineText(ReplacementText, C.NewlinesBefore);
1628 // FIXME: This assert should hold if we computed the column correctly.
1629 // assert((int)C.StartOfTokenColumn >= C.Spaces);
1630 appendIndentText(
1631 ReplacementText, C.Tok->IndentLevel, std::max(0, C.Spaces),
1632 std::max((int)C.StartOfTokenColumn, C.Spaces) - std::max(0, C.Spaces),
1633 C.IsAligned);
1634 ReplacementText.append(C.CurrentLinePrefix);
1635 storeReplacement(C.OriginalWhitespaceRange, ReplacementText);
1640 void WhitespaceManager::storeReplacement(SourceRange Range, StringRef Text) {
1641 unsigned WhitespaceLength = SourceMgr.getFileOffset(Range.getEnd()) -
1642 SourceMgr.getFileOffset(Range.getBegin());
1643 // Don't create a replacement, if it does not change anything.
1644 if (StringRef(SourceMgr.getCharacterData(Range.getBegin()),
1645 WhitespaceLength) == Text) {
1646 return;
1648 auto Err = Replaces.add(tooling::Replacement(
1649 SourceMgr, CharSourceRange::getCharRange(Range), Text));
1650 // FIXME: better error handling. For now, just print an error message in the
1651 // release version.
1652 if (Err) {
1653 llvm::errs() << llvm::toString(std::move(Err)) << "\n";
1654 assert(false);
1658 void WhitespaceManager::appendNewlineText(std::string &Text,
1659 unsigned Newlines) {
1660 if (UseCRLF) {
1661 Text.reserve(Text.size() + 2 * Newlines);
1662 for (unsigned i = 0; i < Newlines; ++i)
1663 Text.append("\r\n");
1664 } else {
1665 Text.append(Newlines, '\n');
1669 void WhitespaceManager::appendEscapedNewlineText(
1670 std::string &Text, unsigned Newlines, unsigned PreviousEndOfTokenColumn,
1671 unsigned EscapedNewlineColumn) {
1672 if (Newlines > 0) {
1673 unsigned Spaces =
1674 std::max<int>(1, EscapedNewlineColumn - PreviousEndOfTokenColumn - 1);
1675 for (unsigned i = 0; i < Newlines; ++i) {
1676 Text.append(Spaces, ' ');
1677 Text.append(UseCRLF ? "\\\r\n" : "\\\n");
1678 Spaces = std::max<int>(0, EscapedNewlineColumn - 1);
1683 void WhitespaceManager::appendIndentText(std::string &Text,
1684 unsigned IndentLevel, unsigned Spaces,
1685 unsigned WhitespaceStartColumn,
1686 bool IsAligned) {
1687 switch (Style.UseTab) {
1688 case FormatStyle::UT_Never:
1689 Text.append(Spaces, ' ');
1690 break;
1691 case FormatStyle::UT_Always: {
1692 if (Style.TabWidth) {
1693 unsigned FirstTabWidth =
1694 Style.TabWidth - WhitespaceStartColumn % Style.TabWidth;
1696 // Insert only spaces when we want to end up before the next tab.
1697 if (Spaces < FirstTabWidth || Spaces == 1) {
1698 Text.append(Spaces, ' ');
1699 break;
1701 // Align to the next tab.
1702 Spaces -= FirstTabWidth;
1703 Text.append("\t");
1705 Text.append(Spaces / Style.TabWidth, '\t');
1706 Text.append(Spaces % Style.TabWidth, ' ');
1707 } else if (Spaces == 1) {
1708 Text.append(Spaces, ' ');
1710 break;
1712 case FormatStyle::UT_ForIndentation:
1713 if (WhitespaceStartColumn == 0) {
1714 unsigned Indentation = IndentLevel * Style.IndentWidth;
1715 Spaces = appendTabIndent(Text, Spaces, Indentation);
1717 Text.append(Spaces, ' ');
1718 break;
1719 case FormatStyle::UT_ForContinuationAndIndentation:
1720 if (WhitespaceStartColumn == 0)
1721 Spaces = appendTabIndent(Text, Spaces, Spaces);
1722 Text.append(Spaces, ' ');
1723 break;
1724 case FormatStyle::UT_AlignWithSpaces:
1725 if (WhitespaceStartColumn == 0) {
1726 unsigned Indentation =
1727 IsAligned ? IndentLevel * Style.IndentWidth : Spaces;
1728 Spaces = appendTabIndent(Text, Spaces, Indentation);
1730 Text.append(Spaces, ' ');
1731 break;
1735 unsigned WhitespaceManager::appendTabIndent(std::string &Text, unsigned Spaces,
1736 unsigned Indentation) {
1737 // This happens, e.g. when a line in a block comment is indented less than the
1738 // first one.
1739 if (Indentation > Spaces)
1740 Indentation = Spaces;
1741 if (Style.TabWidth) {
1742 unsigned Tabs = Indentation / Style.TabWidth;
1743 Text.append(Tabs, '\t');
1744 Spaces -= Tabs * Style.TabWidth;
1746 return Spaces;
1749 } // namespace format
1750 } // namespace clang