[Flang] remove whole-archive option for AIX linker (#76039)
[llvm-project.git] / clang / lib / Format / WhitespaceManager.cpp
blob3bc6915b8df0a7083902cfa049c2b153d0e3235a
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(Changes[Previous].Tok->isPointerOrReference());
474 if (Changes[Previous].Tok->isNot(tok::star)) {
475 if (ReferenceNotRightAligned)
476 continue;
477 } else if (Style.PointerAlignment != FormatStyle::PAS_Right) {
478 continue;
480 Changes[Previous + 1].Spaces -= Shift;
481 Changes[Previous].Spaces += Shift;
482 Changes[Previous].StartOfTokenColumn += Shift;
488 // Walk through a subset of the changes, starting at StartAt, and find
489 // sequences of matching tokens to align. To do so, keep track of the lines and
490 // whether or not a matching token was found on a line. If a matching token is
491 // found, extend the current sequence. If the current line cannot be part of a
492 // sequence, e.g. because there is an empty line before it or it contains only
493 // non-matching tokens, finalize the previous sequence.
494 // The value returned is the token on which we stopped, either because we
495 // exhausted all items inside Changes, or because we hit a scope level higher
496 // than our initial scope.
497 // This function is recursive. Each invocation processes only the scope level
498 // equal to the initial level, which is the level of Changes[StartAt].
499 // If we encounter a scope level greater than the initial level, then we call
500 // ourselves recursively, thereby avoiding the pollution of the current state
501 // with the alignment requirements of the nested sub-level. This recursive
502 // behavior is necessary for aligning function prototypes that have one or more
503 // arguments.
504 // If this function encounters a scope level less than the initial level,
505 // it returns the current position.
506 // There is a non-obvious subtlety in the recursive behavior: Even though we
507 // defer processing of nested levels to recursive invocations of this
508 // function, when it comes time to align a sequence of tokens, we run the
509 // alignment on the entire sequence, including the nested levels.
510 // When doing so, most of the nested tokens are skipped, because their
511 // alignment was already handled by the recursive invocations of this function.
512 // However, the special exception is that we do NOT skip function parameters
513 // that are split across multiple lines. See the test case in FormatTest.cpp
514 // that mentions "split function parameter alignment" for an example of this.
515 // When the parameter RightJustify is true, the operator will be
516 // right-justified. It is used to align compound assignments like `+=` and `=`.
517 // When RightJustify and ACS.PadOperators are true, operators in each block to
518 // be aligned will be padded on the left to the same length before aligning.
519 template <typename F>
520 static unsigned AlignTokens(const FormatStyle &Style, F &&Matches,
521 SmallVector<WhitespaceManager::Change, 16> &Changes,
522 unsigned StartAt,
523 const FormatStyle::AlignConsecutiveStyle &ACS = {},
524 bool RightJustify = false) {
525 // We arrange each line in 3 parts. The operator to be aligned (the anchor),
526 // and text to its left and right. In the aligned text the width of each part
527 // will be the maximum of that over the block that has been aligned. Maximum
528 // widths of each part so far. When RightJustify is true and ACS.PadOperators
529 // is false, the part from start of line to the right end of the anchor.
530 // Otherwise, only the part to the left of the anchor. Including the space
531 // that exists on its left from the start. Not including the padding added on
532 // the left to right-justify the anchor.
533 unsigned WidthLeft = 0;
534 // The operator to be aligned when RightJustify is true and ACS.PadOperators
535 // is false. 0 otherwise.
536 unsigned WidthAnchor = 0;
537 // Width to the right of the anchor. Plus width of the anchor when
538 // RightJustify is false.
539 unsigned WidthRight = 0;
541 // Line number of the start and the end of the current token sequence.
542 unsigned StartOfSequence = 0;
543 unsigned EndOfSequence = 0;
545 // Measure the scope level (i.e. depth of (), [], {}) of the first token, and
546 // abort when we hit any token in a higher scope than the starting one.
547 auto IndentAndNestingLevel = StartAt < Changes.size()
548 ? Changes[StartAt].indentAndNestingLevel()
549 : std::tuple<unsigned, unsigned, unsigned>();
551 // Keep track of the number of commas before the matching tokens, we will only
552 // align a sequence of matching tokens if they are preceded by the same number
553 // of commas.
554 unsigned CommasBeforeLastMatch = 0;
555 unsigned CommasBeforeMatch = 0;
557 // Whether a matching token has been found on the current line.
558 bool FoundMatchOnLine = false;
560 // Whether the current line consists purely of comments.
561 bool LineIsComment = true;
563 // Aligns a sequence of matching tokens, on the MinColumn column.
565 // Sequences start from the first matching token to align, and end at the
566 // first token of the first line that doesn't need to be aligned.
568 // We need to adjust the StartOfTokenColumn of each Change that is on a line
569 // containing any matching token to be aligned and located after such token.
570 auto AlignCurrentSequence = [&] {
571 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) {
572 AlignTokenSequence(Style, StartOfSequence, EndOfSequence,
573 WidthLeft + WidthAnchor, RightJustify, Matches,
574 Changes);
576 WidthLeft = 0;
577 WidthAnchor = 0;
578 WidthRight = 0;
579 StartOfSequence = 0;
580 EndOfSequence = 0;
583 unsigned i = StartAt;
584 for (unsigned e = Changes.size(); i != e; ++i) {
585 auto &CurrentChange = Changes[i];
586 if (CurrentChange.indentAndNestingLevel() < IndentAndNestingLevel)
587 break;
589 if (CurrentChange.NewlinesBefore != 0) {
590 CommasBeforeMatch = 0;
591 EndOfSequence = i;
593 // Whether to break the alignment sequence because of an empty line.
594 bool EmptyLineBreak =
595 (CurrentChange.NewlinesBefore > 1) && !ACS.AcrossEmptyLines;
597 // Whether to break the alignment sequence because of a line without a
598 // match.
599 bool NoMatchBreak =
600 !FoundMatchOnLine && !(LineIsComment && ACS.AcrossComments);
602 if (EmptyLineBreak || NoMatchBreak)
603 AlignCurrentSequence();
605 // A new line starts, re-initialize line status tracking bools.
606 // Keep the match state if a string literal is continued on this line.
607 if (i == 0 || CurrentChange.Tok->isNot(tok::string_literal) ||
608 Changes[i - 1].Tok->isNot(tok::string_literal)) {
609 FoundMatchOnLine = false;
611 LineIsComment = true;
614 if (CurrentChange.Tok->isNot(tok::comment))
615 LineIsComment = false;
617 if (CurrentChange.Tok->is(tok::comma)) {
618 ++CommasBeforeMatch;
619 } else if (CurrentChange.indentAndNestingLevel() > IndentAndNestingLevel) {
620 // Call AlignTokens recursively, skipping over this scope block.
621 unsigned StoppedAt =
622 AlignTokens(Style, Matches, Changes, i, ACS, RightJustify);
623 i = StoppedAt - 1;
624 continue;
627 if (!Matches(CurrentChange))
628 continue;
630 // If there is more than one matching token per line, or if the number of
631 // preceding commas, do not match anymore, end the sequence.
632 if (FoundMatchOnLine || CommasBeforeMatch != CommasBeforeLastMatch)
633 AlignCurrentSequence();
635 CommasBeforeLastMatch = CommasBeforeMatch;
636 FoundMatchOnLine = true;
638 if (StartOfSequence == 0)
639 StartOfSequence = i;
641 unsigned ChangeWidthLeft = CurrentChange.StartOfTokenColumn;
642 unsigned ChangeWidthAnchor = 0;
643 unsigned ChangeWidthRight = 0;
644 if (RightJustify)
645 if (ACS.PadOperators)
646 ChangeWidthAnchor = CurrentChange.TokenLength;
647 else
648 ChangeWidthLeft += CurrentChange.TokenLength;
649 else
650 ChangeWidthRight = CurrentChange.TokenLength;
651 for (unsigned j = i + 1; j != e && Changes[j].NewlinesBefore == 0; ++j) {
652 ChangeWidthRight += Changes[j].Spaces;
653 // Changes are generally 1:1 with the tokens, but a change could also be
654 // inside of a token, in which case it's counted more than once: once for
655 // the whitespace surrounding the token (!IsInsideToken) and once for
656 // each whitespace change within it (IsInsideToken).
657 // Therefore, changes inside of a token should only count the space.
658 if (!Changes[j].IsInsideToken)
659 ChangeWidthRight += Changes[j].TokenLength;
662 // If we are restricted by the maximum column width, end the sequence.
663 unsigned NewLeft = std::max(ChangeWidthLeft, WidthLeft);
664 unsigned NewAnchor = std::max(ChangeWidthAnchor, WidthAnchor);
665 unsigned NewRight = std::max(ChangeWidthRight, WidthRight);
666 // `ColumnLimit == 0` means there is no column limit.
667 if (Style.ColumnLimit != 0 &&
668 Style.ColumnLimit < NewLeft + NewAnchor + NewRight) {
669 AlignCurrentSequence();
670 StartOfSequence = i;
671 WidthLeft = ChangeWidthLeft;
672 WidthAnchor = ChangeWidthAnchor;
673 WidthRight = ChangeWidthRight;
674 } else {
675 WidthLeft = NewLeft;
676 WidthAnchor = NewAnchor;
677 WidthRight = NewRight;
681 EndOfSequence = i;
682 AlignCurrentSequence();
683 return i;
686 // Aligns a sequence of matching tokens, on the MinColumn column.
688 // Sequences start from the first matching token to align, and end at the
689 // first token of the first line that doesn't need to be aligned.
691 // We need to adjust the StartOfTokenColumn of each Change that is on a line
692 // containing any matching token to be aligned and located after such token.
693 static void AlignMatchingTokenSequence(
694 unsigned &StartOfSequence, unsigned &EndOfSequence, unsigned &MinColumn,
695 std::function<bool(const WhitespaceManager::Change &C)> Matches,
696 SmallVector<WhitespaceManager::Change, 16> &Changes) {
697 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) {
698 bool FoundMatchOnLine = false;
699 int Shift = 0;
701 for (unsigned I = StartOfSequence; I != EndOfSequence; ++I) {
702 if (Changes[I].NewlinesBefore > 0) {
703 Shift = 0;
704 FoundMatchOnLine = false;
707 // If this is the first matching token to be aligned, remember by how many
708 // spaces it has to be shifted, so the rest of the changes on the line are
709 // shifted by the same amount.
710 if (!FoundMatchOnLine && Matches(Changes[I])) {
711 FoundMatchOnLine = true;
712 Shift = MinColumn - Changes[I].StartOfTokenColumn;
713 Changes[I].Spaces += Shift;
716 assert(Shift >= 0);
717 Changes[I].StartOfTokenColumn += Shift;
718 if (I + 1 != Changes.size())
719 Changes[I + 1].PreviousEndOfTokenColumn += Shift;
723 MinColumn = 0;
724 StartOfSequence = 0;
725 EndOfSequence = 0;
728 void WhitespaceManager::alignConsecutiveMacros() {
729 if (!Style.AlignConsecutiveMacros.Enabled)
730 return;
732 auto AlignMacrosMatches = [](const Change &C) {
733 const FormatToken *Current = C.Tok;
734 unsigned SpacesRequiredBefore = 1;
736 if (Current->SpacesRequiredBefore == 0 || !Current->Previous)
737 return false;
739 Current = Current->Previous;
741 // If token is a ")", skip over the parameter list, to the
742 // token that precedes the "("
743 if (Current->is(tok::r_paren) && Current->MatchingParen) {
744 Current = Current->MatchingParen->Previous;
745 SpacesRequiredBefore = 0;
748 if (!Current || Current->isNot(tok::identifier))
749 return false;
751 if (!Current->Previous || Current->Previous->isNot(tok::pp_define))
752 return false;
754 // For a macro function, 0 spaces are required between the
755 // identifier and the lparen that opens the parameter list.
756 // For a simple macro, 1 space is required between the
757 // identifier and the first token of the defined value.
758 return Current->Next->SpacesRequiredBefore == SpacesRequiredBefore;
761 unsigned MinColumn = 0;
763 // Start and end of the token sequence we're processing.
764 unsigned StartOfSequence = 0;
765 unsigned EndOfSequence = 0;
767 // Whether a matching token has been found on the current line.
768 bool FoundMatchOnLine = false;
770 // Whether the current line consists only of comments
771 bool LineIsComment = true;
773 unsigned I = 0;
774 for (unsigned E = Changes.size(); I != E; ++I) {
775 if (Changes[I].NewlinesBefore != 0) {
776 EndOfSequence = I;
778 // Whether to break the alignment sequence because of an empty line.
779 bool EmptyLineBreak = (Changes[I].NewlinesBefore > 1) &&
780 !Style.AlignConsecutiveMacros.AcrossEmptyLines;
782 // Whether to break the alignment sequence because of a line without a
783 // match.
784 bool NoMatchBreak =
785 !FoundMatchOnLine &&
786 !(LineIsComment && Style.AlignConsecutiveMacros.AcrossComments);
788 if (EmptyLineBreak || NoMatchBreak) {
789 AlignMatchingTokenSequence(StartOfSequence, EndOfSequence, MinColumn,
790 AlignMacrosMatches, Changes);
793 // A new line starts, re-initialize line status tracking bools.
794 FoundMatchOnLine = false;
795 LineIsComment = true;
798 if (Changes[I].Tok->isNot(tok::comment))
799 LineIsComment = false;
801 if (!AlignMacrosMatches(Changes[I]))
802 continue;
804 FoundMatchOnLine = true;
806 if (StartOfSequence == 0)
807 StartOfSequence = I;
809 unsigned ChangeMinColumn = Changes[I].StartOfTokenColumn;
810 MinColumn = std::max(MinColumn, ChangeMinColumn);
813 EndOfSequence = I;
814 AlignMatchingTokenSequence(StartOfSequence, EndOfSequence, MinColumn,
815 AlignMacrosMatches, Changes);
818 void WhitespaceManager::alignConsecutiveAssignments() {
819 if (!Style.AlignConsecutiveAssignments.Enabled)
820 return;
822 AlignTokens(
823 Style,
824 [&](const Change &C) {
825 // Do not align on equal signs that are first on a line.
826 if (C.NewlinesBefore > 0)
827 return false;
829 // Do not align on equal signs that are last on a line.
830 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
831 return false;
833 // Do not align operator= overloads.
834 FormatToken *Previous = C.Tok->getPreviousNonComment();
835 if (Previous && Previous->is(tok::kw_operator))
836 return false;
838 return Style.AlignConsecutiveAssignments.AlignCompound
839 ? C.Tok->getPrecedence() == prec::Assignment
840 : (C.Tok->is(tok::equal) ||
841 // In Verilog the '<=' is not a compound assignment, thus
842 // it is aligned even when the AlignCompound option is not
843 // set.
844 (Style.isVerilog() && C.Tok->is(tok::lessequal) &&
845 C.Tok->getPrecedence() == prec::Assignment));
847 Changes, /*StartAt=*/0, Style.AlignConsecutiveAssignments,
848 /*RightJustify=*/true);
851 void WhitespaceManager::alignConsecutiveBitFields() {
852 if (!Style.AlignConsecutiveBitFields.Enabled)
853 return;
855 AlignTokens(
856 Style,
857 [&](Change const &C) {
858 // Do not align on ':' that is first on a line.
859 if (C.NewlinesBefore > 0)
860 return false;
862 // Do not align on ':' that is last on a line.
863 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
864 return false;
866 return C.Tok->is(TT_BitFieldColon);
868 Changes, /*StartAt=*/0, Style.AlignConsecutiveBitFields);
871 void WhitespaceManager::alignConsecutiveShortCaseStatements() {
872 if (!Style.AlignConsecutiveShortCaseStatements.Enabled ||
873 !Style.AllowShortCaseLabelsOnASingleLine) {
874 return;
877 auto Matches = [&](const Change &C) {
878 if (Style.AlignConsecutiveShortCaseStatements.AlignCaseColons)
879 return C.Tok->is(TT_CaseLabelColon);
881 // Ignore 'IsInsideToken' to allow matching trailing comments which
882 // need to be reflowed as that causes the token to appear in two
883 // different changes, which will cause incorrect alignment as we'll
884 // reflow early due to detecting multiple aligning tokens per line.
885 return !C.IsInsideToken && C.Tok->Previous &&
886 C.Tok->Previous->is(TT_CaseLabelColon);
889 unsigned MinColumn = 0;
891 // Empty case statements don't break the alignment, but don't necessarily
892 // match our predicate, so we need to track their column so they can push out
893 // our alignment.
894 unsigned MinEmptyCaseColumn = 0;
896 // Start and end of the token sequence we're processing.
897 unsigned StartOfSequence = 0;
898 unsigned EndOfSequence = 0;
900 // Whether a matching token has been found on the current line.
901 bool FoundMatchOnLine = false;
903 bool LineIsComment = true;
904 bool LineIsEmptyCase = false;
906 unsigned I = 0;
907 for (unsigned E = Changes.size(); I != E; ++I) {
908 if (Changes[I].NewlinesBefore != 0) {
909 // Whether to break the alignment sequence because of an empty line.
910 bool EmptyLineBreak =
911 (Changes[I].NewlinesBefore > 1) &&
912 !Style.AlignConsecutiveShortCaseStatements.AcrossEmptyLines;
914 // Whether to break the alignment sequence because of a line without a
915 // match.
916 bool NoMatchBreak =
917 !FoundMatchOnLine &&
918 !(LineIsComment &&
919 Style.AlignConsecutiveShortCaseStatements.AcrossComments) &&
920 !LineIsEmptyCase;
922 if (EmptyLineBreak || NoMatchBreak) {
923 AlignMatchingTokenSequence(StartOfSequence, EndOfSequence, MinColumn,
924 Matches, Changes);
925 MinEmptyCaseColumn = 0;
928 // A new line starts, re-initialize line status tracking bools.
929 FoundMatchOnLine = false;
930 LineIsComment = true;
931 LineIsEmptyCase = false;
934 if (Changes[I].Tok->isNot(tok::comment))
935 LineIsComment = false;
937 if (Changes[I].Tok->is(TT_CaseLabelColon)) {
938 LineIsEmptyCase =
939 !Changes[I].Tok->Next || Changes[I].Tok->Next->isTrailingComment();
941 if (LineIsEmptyCase) {
942 if (Style.AlignConsecutiveShortCaseStatements.AlignCaseColons) {
943 MinEmptyCaseColumn =
944 std::max(MinEmptyCaseColumn, Changes[I].StartOfTokenColumn);
945 } else {
946 MinEmptyCaseColumn =
947 std::max(MinEmptyCaseColumn, Changes[I].StartOfTokenColumn + 2);
952 if (!Matches(Changes[I]))
953 continue;
955 if (LineIsEmptyCase)
956 continue;
958 FoundMatchOnLine = true;
960 if (StartOfSequence == 0)
961 StartOfSequence = I;
963 EndOfSequence = I + 1;
965 MinColumn = std::max(MinColumn, Changes[I].StartOfTokenColumn);
967 // Allow empty case statements to push out our alignment.
968 MinColumn = std::max(MinColumn, MinEmptyCaseColumn);
971 AlignMatchingTokenSequence(StartOfSequence, EndOfSequence, MinColumn, Matches,
972 Changes);
975 void WhitespaceManager::alignConsecutiveDeclarations() {
976 if (!Style.AlignConsecutiveDeclarations.Enabled)
977 return;
979 AlignTokens(
980 Style,
981 [](Change const &C) {
982 if (C.Tok->is(TT_FunctionDeclarationName))
983 return true;
984 if (C.Tok->isNot(TT_StartOfName))
985 return false;
986 if (C.Tok->Previous &&
987 C.Tok->Previous->is(TT_StatementAttributeLikeMacro))
988 return false;
989 // Check if there is a subsequent name that starts the same declaration.
990 for (FormatToken *Next = C.Tok->Next; Next; Next = Next->Next) {
991 if (Next->is(tok::comment))
992 continue;
993 if (Next->is(TT_PointerOrReference))
994 return false;
995 if (!Next->Tok.getIdentifierInfo())
996 break;
997 if (Next->isOneOf(TT_StartOfName, TT_FunctionDeclarationName,
998 tok::kw_operator)) {
999 return false;
1002 return true;
1004 Changes, /*StartAt=*/0, Style.AlignConsecutiveDeclarations);
1007 void WhitespaceManager::alignChainedConditionals() {
1008 if (Style.BreakBeforeTernaryOperators) {
1009 AlignTokens(
1010 Style,
1011 [](Change const &C) {
1012 // Align question operators and last colon
1013 return C.Tok->is(TT_ConditionalExpr) &&
1014 ((C.Tok->is(tok::question) && !C.NewlinesBefore) ||
1015 (C.Tok->is(tok::colon) && C.Tok->Next &&
1016 (C.Tok->Next->FakeLParens.size() == 0 ||
1017 C.Tok->Next->FakeLParens.back() != prec::Conditional)));
1019 Changes, /*StartAt=*/0);
1020 } else {
1021 static auto AlignWrappedOperand = [](Change const &C) {
1022 FormatToken *Previous = C.Tok->getPreviousNonComment();
1023 return C.NewlinesBefore && Previous && Previous->is(TT_ConditionalExpr) &&
1024 (Previous->is(tok::colon) &&
1025 (C.Tok->FakeLParens.size() == 0 ||
1026 C.Tok->FakeLParens.back() != prec::Conditional));
1028 // Ensure we keep alignment of wrapped operands with non-wrapped operands
1029 // Since we actually align the operators, the wrapped operands need the
1030 // extra offset to be properly aligned.
1031 for (Change &C : Changes)
1032 if (AlignWrappedOperand(C))
1033 C.StartOfTokenColumn -= 2;
1034 AlignTokens(
1035 Style,
1036 [this](Change const &C) {
1037 // Align question operators if next operand is not wrapped, as
1038 // well as wrapped operands after question operator or last
1039 // colon in conditional sequence
1040 return (C.Tok->is(TT_ConditionalExpr) && C.Tok->is(tok::question) &&
1041 &C != &Changes.back() && (&C + 1)->NewlinesBefore == 0 &&
1042 !(&C + 1)->IsTrailingComment) ||
1043 AlignWrappedOperand(C);
1045 Changes, /*StartAt=*/0);
1049 void WhitespaceManager::alignTrailingComments() {
1050 if (Style.AlignTrailingComments.Kind == FormatStyle::TCAS_Never)
1051 return;
1053 const int Size = Changes.size();
1054 int MinColumn = 0;
1055 int StartOfSequence = 0;
1056 bool BreakBeforeNext = false;
1057 int NewLineThreshold = 1;
1058 if (Style.AlignTrailingComments.Kind == FormatStyle::TCAS_Always)
1059 NewLineThreshold = Style.AlignTrailingComments.OverEmptyLines + 1;
1061 for (int I = 0, MaxColumn = INT_MAX, Newlines = 0; I < Size; ++I) {
1062 auto &C = Changes[I];
1063 if (C.StartOfBlockComment)
1064 continue;
1065 Newlines += C.NewlinesBefore;
1066 if (!C.IsTrailingComment)
1067 continue;
1069 if (Style.AlignTrailingComments.Kind == FormatStyle::TCAS_Leave) {
1070 const int OriginalSpaces =
1071 C.OriginalWhitespaceRange.getEnd().getRawEncoding() -
1072 C.OriginalWhitespaceRange.getBegin().getRawEncoding() -
1073 C.Tok->LastNewlineOffset;
1074 assert(OriginalSpaces >= 0);
1075 const auto RestoredLineLength =
1076 C.StartOfTokenColumn + C.TokenLength + OriginalSpaces;
1077 // If leaving comments makes the line exceed the column limit, give up to
1078 // leave the comments.
1079 if (RestoredLineLength >= Style.ColumnLimit && Style.ColumnLimit > 0)
1080 break;
1081 C.Spaces = OriginalSpaces;
1082 continue;
1085 const int ChangeMinColumn = C.StartOfTokenColumn;
1086 int ChangeMaxColumn;
1088 // If we don't create a replacement for this change, we have to consider
1089 // it to be immovable.
1090 if (!C.CreateReplacement)
1091 ChangeMaxColumn = ChangeMinColumn;
1092 else if (Style.ColumnLimit == 0)
1093 ChangeMaxColumn = INT_MAX;
1094 else if (Style.ColumnLimit >= C.TokenLength)
1095 ChangeMaxColumn = Style.ColumnLimit - C.TokenLength;
1096 else
1097 ChangeMaxColumn = ChangeMinColumn;
1099 if (I + 1 < Size && Changes[I + 1].ContinuesPPDirective &&
1100 ChangeMaxColumn >= 2) {
1101 ChangeMaxColumn -= 2;
1104 bool WasAlignedWithStartOfNextLine = false;
1105 if (C.NewlinesBefore >= 1) { // A comment on its own line.
1106 const auto CommentColumn =
1107 SourceMgr.getSpellingColumnNumber(C.OriginalWhitespaceRange.getEnd());
1108 for (int J = I + 1; J < Size; ++J) {
1109 if (Changes[J].Tok->is(tok::comment))
1110 continue;
1112 const auto NextColumn = SourceMgr.getSpellingColumnNumber(
1113 Changes[J].OriginalWhitespaceRange.getEnd());
1114 // The start of the next token was previously aligned with the
1115 // start of this comment.
1116 WasAlignedWithStartOfNextLine =
1117 CommentColumn == NextColumn ||
1118 CommentColumn == NextColumn + Style.IndentWidth;
1119 break;
1123 // We don't want to align comments which end a scope, which are here
1124 // identified by most closing braces.
1125 auto DontAlignThisComment = [](const auto *Tok) {
1126 if (Tok->is(tok::semi)) {
1127 Tok = Tok->getPreviousNonComment();
1128 if (!Tok)
1129 return false;
1131 if (Tok->is(tok::r_paren)) {
1132 // Back up past the parentheses and a `TT_DoWhile` that may precede.
1133 Tok = Tok->MatchingParen;
1134 if (!Tok)
1135 return false;
1136 Tok = Tok->getPreviousNonComment();
1137 if (!Tok)
1138 return false;
1139 if (Tok->is(TT_DoWhile)) {
1140 const auto *Prev = Tok->getPreviousNonComment();
1141 if (!Prev) {
1142 // A do-while-loop without braces.
1143 return true;
1145 Tok = Prev;
1149 if (Tok->isNot(tok::r_brace))
1150 return false;
1152 while (Tok->Previous && Tok->Previous->is(tok::r_brace))
1153 Tok = Tok->Previous;
1154 return Tok->NewlinesBefore > 0;
1157 if (I > 0 && C.NewlinesBefore == 0 &&
1158 DontAlignThisComment(Changes[I - 1].Tok)) {
1159 alignTrailingComments(StartOfSequence, I, MinColumn);
1160 // Reset to initial values, but skip this change for the next alignment
1161 // pass.
1162 MinColumn = 0;
1163 MaxColumn = INT_MAX;
1164 StartOfSequence = I + 1;
1165 } else if (BreakBeforeNext || Newlines > NewLineThreshold ||
1166 (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn) ||
1167 // Break the comment sequence if the previous line did not end
1168 // in a trailing comment.
1169 (C.NewlinesBefore == 1 && I > 0 &&
1170 !Changes[I - 1].IsTrailingComment) ||
1171 WasAlignedWithStartOfNextLine) {
1172 alignTrailingComments(StartOfSequence, I, MinColumn);
1173 MinColumn = ChangeMinColumn;
1174 MaxColumn = ChangeMaxColumn;
1175 StartOfSequence = I;
1176 } else {
1177 MinColumn = std::max(MinColumn, ChangeMinColumn);
1178 MaxColumn = std::min(MaxColumn, ChangeMaxColumn);
1180 BreakBeforeNext = (I == 0) || (C.NewlinesBefore > 1) ||
1181 // Never start a sequence with a comment at the beginning
1182 // of the line.
1183 (C.NewlinesBefore == 1 && StartOfSequence == I);
1184 Newlines = 0;
1186 alignTrailingComments(StartOfSequence, Size, MinColumn);
1189 void WhitespaceManager::alignTrailingComments(unsigned Start, unsigned End,
1190 unsigned Column) {
1191 for (unsigned i = Start; i != End; ++i) {
1192 int Shift = 0;
1193 if (Changes[i].IsTrailingComment)
1194 Shift = Column - Changes[i].StartOfTokenColumn;
1195 if (Changes[i].StartOfBlockComment) {
1196 Shift = Changes[i].IndentationOffset +
1197 Changes[i].StartOfBlockComment->StartOfTokenColumn -
1198 Changes[i].StartOfTokenColumn;
1200 if (Shift <= 0)
1201 continue;
1202 Changes[i].Spaces += Shift;
1203 if (i + 1 != Changes.size())
1204 Changes[i + 1].PreviousEndOfTokenColumn += Shift;
1205 Changes[i].StartOfTokenColumn += Shift;
1209 void WhitespaceManager::alignEscapedNewlines() {
1210 if (Style.AlignEscapedNewlines == FormatStyle::ENAS_DontAlign)
1211 return;
1213 bool AlignLeft = Style.AlignEscapedNewlines == FormatStyle::ENAS_Left;
1214 unsigned MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit;
1215 unsigned StartOfMacro = 0;
1216 for (unsigned i = 1, e = Changes.size(); i < e; ++i) {
1217 Change &C = Changes[i];
1218 if (C.NewlinesBefore > 0) {
1219 if (C.ContinuesPPDirective) {
1220 MaxEndOfLine = std::max(C.PreviousEndOfTokenColumn + 2, MaxEndOfLine);
1221 } else {
1222 alignEscapedNewlines(StartOfMacro + 1, i, MaxEndOfLine);
1223 MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit;
1224 StartOfMacro = i;
1228 alignEscapedNewlines(StartOfMacro + 1, Changes.size(), MaxEndOfLine);
1231 void WhitespaceManager::alignEscapedNewlines(unsigned Start, unsigned End,
1232 unsigned Column) {
1233 for (unsigned i = Start; i < End; ++i) {
1234 Change &C = Changes[i];
1235 if (C.NewlinesBefore > 0) {
1236 assert(C.ContinuesPPDirective);
1237 if (C.PreviousEndOfTokenColumn + 1 > Column)
1238 C.EscapedNewlineColumn = 0;
1239 else
1240 C.EscapedNewlineColumn = Column;
1245 void WhitespaceManager::alignArrayInitializers() {
1246 if (Style.AlignArrayOfStructures == FormatStyle::AIAS_None)
1247 return;
1249 for (unsigned ChangeIndex = 1U, ChangeEnd = Changes.size();
1250 ChangeIndex < ChangeEnd; ++ChangeIndex) {
1251 auto &C = Changes[ChangeIndex];
1252 if (C.Tok->IsArrayInitializer) {
1253 bool FoundComplete = false;
1254 for (unsigned InsideIndex = ChangeIndex + 1; InsideIndex < ChangeEnd;
1255 ++InsideIndex) {
1256 if (Changes[InsideIndex].Tok == C.Tok->MatchingParen) {
1257 alignArrayInitializers(ChangeIndex, InsideIndex + 1);
1258 ChangeIndex = InsideIndex + 1;
1259 FoundComplete = true;
1260 break;
1263 if (!FoundComplete)
1264 ChangeIndex = ChangeEnd;
1269 void WhitespaceManager::alignArrayInitializers(unsigned Start, unsigned End) {
1271 if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Right)
1272 alignArrayInitializersRightJustified(getCells(Start, End));
1273 else if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Left)
1274 alignArrayInitializersLeftJustified(getCells(Start, End));
1277 void WhitespaceManager::alignArrayInitializersRightJustified(
1278 CellDescriptions &&CellDescs) {
1279 if (!CellDescs.isRectangular())
1280 return;
1282 const int BracePadding = Style.Cpp11BracedListStyle ? 0 : 1;
1283 auto &Cells = CellDescs.Cells;
1284 // Now go through and fixup the spaces.
1285 auto *CellIter = Cells.begin();
1286 for (auto i = 0U; i < CellDescs.CellCounts[0]; ++i, ++CellIter) {
1287 unsigned NetWidth = 0U;
1288 if (isSplitCell(*CellIter))
1289 NetWidth = getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces);
1290 auto CellWidth = getMaximumCellWidth(CellIter, NetWidth);
1292 if (Changes[CellIter->Index].Tok->is(tok::r_brace)) {
1293 // So in here we want to see if there is a brace that falls
1294 // on a line that was split. If so on that line we make sure that
1295 // the spaces in front of the brace are enough.
1296 const auto *Next = CellIter;
1297 do {
1298 const FormatToken *Previous = Changes[Next->Index].Tok->Previous;
1299 if (Previous && Previous->isNot(TT_LineComment)) {
1300 Changes[Next->Index].Spaces = BracePadding;
1301 Changes[Next->Index].NewlinesBefore = 0;
1303 Next = Next->NextColumnElement;
1304 } while (Next);
1305 // Unless the array is empty, we need the position of all the
1306 // immediately adjacent cells
1307 if (CellIter != Cells.begin()) {
1308 auto ThisNetWidth =
1309 getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces);
1310 auto MaxNetWidth = getMaximumNetWidth(
1311 Cells.begin(), CellIter, CellDescs.InitialSpaces,
1312 CellDescs.CellCounts[0], CellDescs.CellCounts.size());
1313 if (ThisNetWidth < MaxNetWidth)
1314 Changes[CellIter->Index].Spaces = (MaxNetWidth - ThisNetWidth);
1315 auto RowCount = 1U;
1316 auto Offset = std::distance(Cells.begin(), CellIter);
1317 for (const auto *Next = CellIter->NextColumnElement; Next;
1318 Next = Next->NextColumnElement) {
1319 if (RowCount >= CellDescs.CellCounts.size())
1320 break;
1321 auto *Start = (Cells.begin() + RowCount * CellDescs.CellCounts[0]);
1322 auto *End = Start + Offset;
1323 ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces);
1324 if (ThisNetWidth < MaxNetWidth)
1325 Changes[Next->Index].Spaces = (MaxNetWidth - ThisNetWidth);
1326 ++RowCount;
1329 } else {
1330 auto ThisWidth =
1331 calculateCellWidth(CellIter->Index, CellIter->EndIndex, true) +
1332 NetWidth;
1333 if (Changes[CellIter->Index].NewlinesBefore == 0) {
1334 Changes[CellIter->Index].Spaces = (CellWidth - (ThisWidth + NetWidth));
1335 Changes[CellIter->Index].Spaces += (i > 0) ? 1 : BracePadding;
1337 alignToStartOfCell(CellIter->Index, CellIter->EndIndex);
1338 for (const auto *Next = CellIter->NextColumnElement; Next;
1339 Next = Next->NextColumnElement) {
1340 ThisWidth =
1341 calculateCellWidth(Next->Index, Next->EndIndex, true) + NetWidth;
1342 if (Changes[Next->Index].NewlinesBefore == 0) {
1343 Changes[Next->Index].Spaces = (CellWidth - ThisWidth);
1344 Changes[Next->Index].Spaces += (i > 0) ? 1 : BracePadding;
1346 alignToStartOfCell(Next->Index, Next->EndIndex);
1352 void WhitespaceManager::alignArrayInitializersLeftJustified(
1353 CellDescriptions &&CellDescs) {
1355 if (!CellDescs.isRectangular())
1356 return;
1358 const int BracePadding = Style.Cpp11BracedListStyle ? 0 : 1;
1359 auto &Cells = CellDescs.Cells;
1360 // Now go through and fixup the spaces.
1361 auto *CellIter = Cells.begin();
1362 // The first cell needs to be against the left brace.
1363 if (Changes[CellIter->Index].NewlinesBefore == 0)
1364 Changes[CellIter->Index].Spaces = BracePadding;
1365 else
1366 Changes[CellIter->Index].Spaces = CellDescs.InitialSpaces;
1367 ++CellIter;
1368 for (auto i = 1U; i < CellDescs.CellCounts[0]; i++, ++CellIter) {
1369 auto MaxNetWidth = getMaximumNetWidth(
1370 Cells.begin(), CellIter, CellDescs.InitialSpaces,
1371 CellDescs.CellCounts[0], CellDescs.CellCounts.size());
1372 auto ThisNetWidth =
1373 getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces);
1374 if (Changes[CellIter->Index].NewlinesBefore == 0) {
1375 Changes[CellIter->Index].Spaces =
1376 MaxNetWidth - ThisNetWidth +
1377 (Changes[CellIter->Index].Tok->isNot(tok::r_brace) ? 1
1378 : BracePadding);
1380 auto RowCount = 1U;
1381 auto Offset = std::distance(Cells.begin(), CellIter);
1382 for (const auto *Next = CellIter->NextColumnElement; Next;
1383 Next = Next->NextColumnElement) {
1384 if (RowCount >= CellDescs.CellCounts.size())
1385 break;
1386 auto *Start = (Cells.begin() + RowCount * CellDescs.CellCounts[0]);
1387 auto *End = Start + Offset;
1388 auto ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces);
1389 if (Changes[Next->Index].NewlinesBefore == 0) {
1390 Changes[Next->Index].Spaces =
1391 MaxNetWidth - ThisNetWidth +
1392 (Changes[Next->Index].Tok->isNot(tok::r_brace) ? 1 : BracePadding);
1394 ++RowCount;
1399 bool WhitespaceManager::isSplitCell(const CellDescription &Cell) {
1400 if (Cell.HasSplit)
1401 return true;
1402 for (const auto *Next = Cell.NextColumnElement; Next;
1403 Next = Next->NextColumnElement) {
1404 if (Next->HasSplit)
1405 return true;
1407 return false;
1410 WhitespaceManager::CellDescriptions WhitespaceManager::getCells(unsigned Start,
1411 unsigned End) {
1413 unsigned Depth = 0;
1414 unsigned Cell = 0;
1415 SmallVector<unsigned> CellCounts;
1416 unsigned InitialSpaces = 0;
1417 unsigned InitialTokenLength = 0;
1418 unsigned EndSpaces = 0;
1419 SmallVector<CellDescription> Cells;
1420 const FormatToken *MatchingParen = nullptr;
1421 for (unsigned i = Start; i < End; ++i) {
1422 auto &C = Changes[i];
1423 if (C.Tok->is(tok::l_brace))
1424 ++Depth;
1425 else if (C.Tok->is(tok::r_brace))
1426 --Depth;
1427 if (Depth == 2) {
1428 if (C.Tok->is(tok::l_brace)) {
1429 Cell = 0;
1430 MatchingParen = C.Tok->MatchingParen;
1431 if (InitialSpaces == 0) {
1432 InitialSpaces = C.Spaces + C.TokenLength;
1433 InitialTokenLength = C.TokenLength;
1434 auto j = i - 1;
1435 for (; Changes[j].NewlinesBefore == 0 && j > Start; --j) {
1436 InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength;
1437 InitialTokenLength += Changes[j].TokenLength;
1439 if (C.NewlinesBefore == 0) {
1440 InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength;
1441 InitialTokenLength += Changes[j].TokenLength;
1444 } else if (C.Tok->is(tok::comma)) {
1445 if (!Cells.empty())
1446 Cells.back().EndIndex = i;
1447 if (C.Tok->getNextNonComment()->isNot(tok::r_brace)) // dangling comma
1448 ++Cell;
1450 } else if (Depth == 1) {
1451 if (C.Tok == MatchingParen) {
1452 if (!Cells.empty())
1453 Cells.back().EndIndex = i;
1454 Cells.push_back(CellDescription{i, ++Cell, i + 1, false, nullptr});
1455 CellCounts.push_back(C.Tok->Previous->isNot(tok::comma) ? Cell + 1
1456 : Cell);
1457 // Go to the next non-comment and ensure there is a break in front
1458 const auto *NextNonComment = C.Tok->getNextNonComment();
1459 while (NextNonComment->is(tok::comma))
1460 NextNonComment = NextNonComment->getNextNonComment();
1461 auto j = i;
1462 while (Changes[j].Tok != NextNonComment && j < End)
1463 ++j;
1464 if (j < End && Changes[j].NewlinesBefore == 0 &&
1465 Changes[j].Tok->isNot(tok::r_brace)) {
1466 Changes[j].NewlinesBefore = 1;
1467 // Account for the added token lengths
1468 Changes[j].Spaces = InitialSpaces - InitialTokenLength;
1470 } else if (C.Tok->is(tok::comment) && C.Tok->NewlinesBefore == 0) {
1471 // Trailing comments stay at a space past the last token
1472 C.Spaces = Changes[i - 1].Tok->is(tok::comma) ? 1 : 2;
1473 } else if (C.Tok->is(tok::l_brace)) {
1474 // We need to make sure that the ending braces is aligned to the
1475 // start of our initializer
1476 auto j = i - 1;
1477 for (; j > 0 && !Changes[j].Tok->ArrayInitializerLineStart; --j)
1478 ; // Nothing the loop does the work
1479 EndSpaces = Changes[j].Spaces;
1481 } else if (Depth == 0 && C.Tok->is(tok::r_brace)) {
1482 C.NewlinesBefore = 1;
1483 C.Spaces = EndSpaces;
1485 if (C.Tok->StartsColumn) {
1486 // This gets us past tokens that have been split over multiple
1487 // lines
1488 bool HasSplit = false;
1489 if (Changes[i].NewlinesBefore > 0) {
1490 // So if we split a line previously and the tail line + this token is
1491 // less then the column limit we remove the split here and just put
1492 // the column start at a space past the comma
1494 // FIXME This if branch covers the cases where the column is not
1495 // the first column. This leads to weird pathologies like the formatting
1496 // auto foo = Items{
1497 // Section{
1498 // 0, bar(),
1499 // }
1500 // };
1501 // Well if it doesn't lead to that it's indicative that the line
1502 // breaking should be revisited. Unfortunately alot of other options
1503 // interact with this
1504 auto j = i - 1;
1505 if ((j - 1) > Start && Changes[j].Tok->is(tok::comma) &&
1506 Changes[j - 1].NewlinesBefore > 0) {
1507 --j;
1508 auto LineLimit = Changes[j].Spaces + Changes[j].TokenLength;
1509 if (LineLimit < Style.ColumnLimit) {
1510 Changes[i].NewlinesBefore = 0;
1511 Changes[i].Spaces = 1;
1515 while (Changes[i].NewlinesBefore > 0 && Changes[i].Tok == C.Tok) {
1516 Changes[i].Spaces = InitialSpaces;
1517 ++i;
1518 HasSplit = true;
1520 if (Changes[i].Tok != C.Tok)
1521 --i;
1522 Cells.push_back(CellDescription{i, Cell, i, HasSplit, nullptr});
1526 return linkCells({Cells, CellCounts, InitialSpaces});
1529 unsigned WhitespaceManager::calculateCellWidth(unsigned Start, unsigned End,
1530 bool WithSpaces) const {
1531 unsigned CellWidth = 0;
1532 for (auto i = Start; i < End; i++) {
1533 if (Changes[i].NewlinesBefore > 0)
1534 CellWidth = 0;
1535 CellWidth += Changes[i].TokenLength;
1536 CellWidth += (WithSpaces ? Changes[i].Spaces : 0);
1538 return CellWidth;
1541 void WhitespaceManager::alignToStartOfCell(unsigned Start, unsigned End) {
1542 if ((End - Start) <= 1)
1543 return;
1544 // If the line is broken anywhere in there make sure everything
1545 // is aligned to the parent
1546 for (auto i = Start + 1; i < End; i++)
1547 if (Changes[i].NewlinesBefore > 0)
1548 Changes[i].Spaces = Changes[Start].Spaces;
1551 WhitespaceManager::CellDescriptions
1552 WhitespaceManager::linkCells(CellDescriptions &&CellDesc) {
1553 auto &Cells = CellDesc.Cells;
1554 for (auto *CellIter = Cells.begin(); CellIter != Cells.end(); ++CellIter) {
1555 if (!CellIter->NextColumnElement && (CellIter + 1) != Cells.end()) {
1556 for (auto *NextIter = CellIter + 1; NextIter != Cells.end(); ++NextIter) {
1557 if (NextIter->Cell == CellIter->Cell) {
1558 CellIter->NextColumnElement = &(*NextIter);
1559 break;
1564 return std::move(CellDesc);
1567 void WhitespaceManager::generateChanges() {
1568 for (unsigned i = 0, e = Changes.size(); i != e; ++i) {
1569 const Change &C = Changes[i];
1570 if (i > 0) {
1571 auto Last = Changes[i - 1].OriginalWhitespaceRange;
1572 auto New = Changes[i].OriginalWhitespaceRange;
1573 // Do not generate two replacements for the same location. As a special
1574 // case, it is allowed if there is a replacement for the empty range
1575 // between 2 tokens and another non-empty range at the start of the second
1576 // token. We didn't implement logic to combine replacements for 2
1577 // consecutive source ranges into a single replacement, because the
1578 // program works fine without it.
1580 // We can't eliminate empty original whitespace ranges. They appear when
1581 // 2 tokens have no whitespace in between in the input. It does not
1582 // matter whether whitespace is to be added. If no whitespace is to be
1583 // added, the replacement will be empty, and it gets eliminated after this
1584 // step in storeReplacement. For example, if the input is `foo();`,
1585 // there will be a replacement for the range between every consecutive
1586 // pair of tokens.
1588 // A replacement at the start of a token can be added by
1589 // BreakableStringLiteralUsingOperators::insertBreak when it adds braces
1590 // around the string literal. Say Verilog code is being formatted and the
1591 // first line is to become the next 2 lines.
1592 // x("long string");
1593 // x({"long ",
1594 // "string"});
1595 // There will be a replacement for the empty range between the parenthesis
1596 // and the string and another replacement for the quote character. The
1597 // replacement for the empty range between the parenthesis and the quote
1598 // comes from ContinuationIndenter::addTokenOnCurrentLine when it changes
1599 // the original empty range between the parenthesis and the string to
1600 // another empty one. The replacement for the quote character comes from
1601 // BreakableStringLiteralUsingOperators::insertBreak when it adds the
1602 // brace. In the example, the replacement for the empty range is the same
1603 // as the original text. However, eliminating replacements that are same
1604 // as the original does not help in general. For example, a newline can
1605 // be inserted, causing the first line to become the next 3 lines.
1606 // xxxxxxxxxxx("long string");
1607 // xxxxxxxxxxx(
1608 // {"long ",
1609 // "string"});
1610 // In that case, the empty range between the parenthesis and the string
1611 // will be replaced by a newline and 4 spaces. So we will still have to
1612 // deal with a replacement for an empty source range followed by a
1613 // replacement for a non-empty source range.
1614 if (Last.getBegin() == New.getBegin() &&
1615 (Last.getEnd() != Last.getBegin() ||
1616 New.getEnd() == New.getBegin())) {
1617 continue;
1620 if (C.CreateReplacement) {
1621 std::string ReplacementText = C.PreviousLinePostfix;
1622 if (C.ContinuesPPDirective) {
1623 appendEscapedNewlineText(ReplacementText, C.NewlinesBefore,
1624 C.PreviousEndOfTokenColumn,
1625 C.EscapedNewlineColumn);
1626 } else {
1627 appendNewlineText(ReplacementText, C.NewlinesBefore);
1629 // FIXME: This assert should hold if we computed the column correctly.
1630 // assert((int)C.StartOfTokenColumn >= C.Spaces);
1631 appendIndentText(
1632 ReplacementText, C.Tok->IndentLevel, std::max(0, C.Spaces),
1633 std::max((int)C.StartOfTokenColumn, C.Spaces) - std::max(0, C.Spaces),
1634 C.IsAligned);
1635 ReplacementText.append(C.CurrentLinePrefix);
1636 storeReplacement(C.OriginalWhitespaceRange, ReplacementText);
1641 void WhitespaceManager::storeReplacement(SourceRange Range, StringRef Text) {
1642 unsigned WhitespaceLength = SourceMgr.getFileOffset(Range.getEnd()) -
1643 SourceMgr.getFileOffset(Range.getBegin());
1644 // Don't create a replacement, if it does not change anything.
1645 if (StringRef(SourceMgr.getCharacterData(Range.getBegin()),
1646 WhitespaceLength) == Text) {
1647 return;
1649 auto Err = Replaces.add(tooling::Replacement(
1650 SourceMgr, CharSourceRange::getCharRange(Range), Text));
1651 // FIXME: better error handling. For now, just print an error message in the
1652 // release version.
1653 if (Err) {
1654 llvm::errs() << llvm::toString(std::move(Err)) << "\n";
1655 assert(false);
1659 void WhitespaceManager::appendNewlineText(std::string &Text,
1660 unsigned Newlines) {
1661 if (UseCRLF) {
1662 Text.reserve(Text.size() + 2 * Newlines);
1663 for (unsigned i = 0; i < Newlines; ++i)
1664 Text.append("\r\n");
1665 } else {
1666 Text.append(Newlines, '\n');
1670 void WhitespaceManager::appendEscapedNewlineText(
1671 std::string &Text, unsigned Newlines, unsigned PreviousEndOfTokenColumn,
1672 unsigned EscapedNewlineColumn) {
1673 if (Newlines > 0) {
1674 unsigned Spaces =
1675 std::max<int>(1, EscapedNewlineColumn - PreviousEndOfTokenColumn - 1);
1676 for (unsigned i = 0; i < Newlines; ++i) {
1677 Text.append(Spaces, ' ');
1678 Text.append(UseCRLF ? "\\\r\n" : "\\\n");
1679 Spaces = std::max<int>(0, EscapedNewlineColumn - 1);
1684 void WhitespaceManager::appendIndentText(std::string &Text,
1685 unsigned IndentLevel, unsigned Spaces,
1686 unsigned WhitespaceStartColumn,
1687 bool IsAligned) {
1688 switch (Style.UseTab) {
1689 case FormatStyle::UT_Never:
1690 Text.append(Spaces, ' ');
1691 break;
1692 case FormatStyle::UT_Always: {
1693 if (Style.TabWidth) {
1694 unsigned FirstTabWidth =
1695 Style.TabWidth - WhitespaceStartColumn % Style.TabWidth;
1697 // Insert only spaces when we want to end up before the next tab.
1698 if (Spaces < FirstTabWidth || Spaces == 1) {
1699 Text.append(Spaces, ' ');
1700 break;
1702 // Align to the next tab.
1703 Spaces -= FirstTabWidth;
1704 Text.append("\t");
1706 Text.append(Spaces / Style.TabWidth, '\t');
1707 Text.append(Spaces % Style.TabWidth, ' ');
1708 } else if (Spaces == 1) {
1709 Text.append(Spaces, ' ');
1711 break;
1713 case FormatStyle::UT_ForIndentation:
1714 if (WhitespaceStartColumn == 0) {
1715 unsigned Indentation = IndentLevel * Style.IndentWidth;
1716 Spaces = appendTabIndent(Text, Spaces, Indentation);
1718 Text.append(Spaces, ' ');
1719 break;
1720 case FormatStyle::UT_ForContinuationAndIndentation:
1721 if (WhitespaceStartColumn == 0)
1722 Spaces = appendTabIndent(Text, Spaces, Spaces);
1723 Text.append(Spaces, ' ');
1724 break;
1725 case FormatStyle::UT_AlignWithSpaces:
1726 if (WhitespaceStartColumn == 0) {
1727 unsigned Indentation =
1728 IsAligned ? IndentLevel * Style.IndentWidth : Spaces;
1729 Spaces = appendTabIndent(Text, Spaces, Indentation);
1731 Text.append(Spaces, ' ');
1732 break;
1736 unsigned WhitespaceManager::appendTabIndent(std::string &Text, unsigned Spaces,
1737 unsigned Indentation) {
1738 // This happens, e.g. when a line in a block comment is indented less than the
1739 // first one.
1740 if (Indentation > Spaces)
1741 Indentation = Spaces;
1742 if (Style.TabWidth) {
1743 unsigned Tabs = Indentation / Style.TabWidth;
1744 Text.append(Tabs, '\t');
1745 Spaces -= Tabs * Style.TabWidth;
1747 return Spaces;
1750 } // namespace format
1751 } // namespace clang