1 //===--- UnwrappedLineFormatter.cpp - Format C++ code ---------------------===//
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
7 //===----------------------------------------------------------------------===//
9 #include "UnwrappedLineFormatter.h"
10 #include "FormatToken.h"
11 #include "NamespaceEndCommentsFixer.h"
12 #include "WhitespaceManager.h"
13 #include "llvm/Support/Debug.h"
16 #define DEBUG_TYPE "format-formatter"
23 bool startsExternCBlock(const AnnotatedLine
&Line
) {
24 const FormatToken
*Next
= Line
.First
->getNextNonComment();
25 const FormatToken
*NextNext
= Next
? Next
->getNextNonComment() : nullptr;
26 return Line
.startsWith(tok::kw_extern
) && Next
&& Next
->isStringLiteral() &&
27 NextNext
&& NextNext
->is(tok::l_brace
);
30 bool isRecordLBrace(const FormatToken
&Tok
) {
31 return Tok
.isOneOf(TT_ClassLBrace
, TT_EnumLBrace
, TT_RecordLBrace
,
32 TT_StructLBrace
, TT_UnionLBrace
);
35 /// Tracks the indent level of \c AnnotatedLines across levels.
37 /// \c nextLine must be called for each \c AnnotatedLine, after which \c
38 /// getIndent() will return the indent for the last line \c nextLine was called
40 /// If the line is not formatted (and thus the indent does not change), calling
41 /// \c adjustToUnmodifiedLine after the call to \c nextLine will cause
42 /// subsequent lines on the same level to be indented at the same level as the
44 class LevelIndentTracker
{
46 LevelIndentTracker(const FormatStyle
&Style
,
47 const AdditionalKeywords
&Keywords
, unsigned StartLevel
,
49 : Style(Style
), Keywords(Keywords
), AdditionalIndent(AdditionalIndent
) {
50 for (unsigned i
= 0; i
!= StartLevel
; ++i
)
51 IndentForLevel
.push_back(Style
.IndentWidth
* i
+ AdditionalIndent
);
54 /// Returns the indent for the current line.
55 unsigned getIndent() const { return Indent
; }
57 /// Update the indent state given that \p Line is going to be formatted
59 void nextLine(const AnnotatedLine
&Line
) {
60 Offset
= getIndentOffset(Line
);
61 // Update the indent level cache size so that we can rely on it
62 // having the right size in adjustToUnmodifiedline.
63 if (Line
.Level
>= IndentForLevel
.size())
64 IndentForLevel
.resize(Line
.Level
+ 1, -1);
65 if (Style
.IndentPPDirectives
!= FormatStyle::PPDIS_None
&&
66 (Line
.InPPDirective
||
67 (Style
.IndentPPDirectives
== FormatStyle::PPDIS_BeforeHash
&&
68 Line
.Type
== LT_CommentAbovePPDirective
))) {
69 unsigned PPIndentWidth
=
70 (Style
.PPIndentWidth
>= 0) ? Style
.PPIndentWidth
: Style
.IndentWidth
;
71 Indent
= Line
.InMacroBody
72 ? Line
.PPLevel
* PPIndentWidth
+
73 (Line
.Level
- Line
.PPLevel
) * Style
.IndentWidth
74 : Line
.Level
* PPIndentWidth
;
75 Indent
+= AdditionalIndent
;
77 // When going to lower levels, forget previous higher levels so that we
78 // recompute future higher levels. But don't forget them if we enter a PP
79 // directive, since these do not terminate a C++ code block.
80 if (!Line
.InPPDirective
) {
81 assert(Line
.Level
<= IndentForLevel
.size());
82 IndentForLevel
.resize(Line
.Level
+ 1);
84 Indent
= getIndent(Line
.Level
);
86 if (static_cast<int>(Indent
) + Offset
>= 0)
88 if (Line
.IsContinuation
)
89 Indent
= Line
.Level
* Style
.IndentWidth
+ Style
.ContinuationIndentWidth
;
92 /// Update the level indent to adapt to the given \p Line.
94 /// When a line is not formatted, we move the subsequent lines on the same
95 /// level to the same indent.
96 /// Note that \c nextLine must have been called before this method.
97 void adjustToUnmodifiedLine(const AnnotatedLine
&Line
) {
98 if (Line
.InPPDirective
|| Line
.IsContinuation
)
100 assert(Line
.Level
< IndentForLevel
.size());
101 if (Line
.First
->is(tok::comment
) && IndentForLevel
[Line
.Level
] != -1)
103 unsigned LevelIndent
= Line
.First
->OriginalColumn
;
104 if (static_cast<int>(LevelIndent
) - Offset
>= 0)
105 LevelIndent
-= Offset
;
106 IndentForLevel
[Line
.Level
] = LevelIndent
;
110 /// Get the offset of the line relatively to the level.
112 /// For example, 'public:' labels in classes are offset by 1 or 2
113 /// characters to the left from their level.
114 int getIndentOffset(const AnnotatedLine
&Line
) {
115 if (Style
.Language
== FormatStyle::LK_Java
|| Style
.isJavaScript() ||
120 auto IsAccessModifier
= [&](const FormatToken
&RootToken
) {
121 if (Line
.Type
== LT_AccessModifier
|| RootToken
.isObjCAccessSpecifier())
124 const auto *Next
= RootToken
.Next
;
126 // Handle Qt signals.
127 if (RootToken
.isOneOf(Keywords
.kw_signals
, Keywords
.kw_qsignals
) &&
128 Next
&& Next
->is(tok::colon
)) {
132 if (Next
&& Next
->isOneOf(Keywords
.kw_slots
, Keywords
.kw_qslots
) &&
133 Next
->Next
&& Next
->Next
->is(tok::colon
)) {
137 // Handle malformed access specifier e.g. 'private' without trailing ':'.
138 return !Next
&& RootToken
.isAccessSpecifier(false);
141 if (IsAccessModifier(*Line
.First
)) {
142 // The AccessModifierOffset may be overridden by IndentAccessModifiers,
143 // in which case we take a negative value of the IndentWidth to simulate
144 // the upper indent level.
145 return Style
.IndentAccessModifiers
? -Style
.IndentWidth
146 : Style
.AccessModifierOffset
;
152 /// Get the indent of \p Level from \p IndentForLevel.
154 /// \p IndentForLevel must contain the indent for the level \c l
155 /// at \p IndentForLevel[l], or a value < 0 if the indent for
156 /// that level is unknown.
157 unsigned getIndent(unsigned Level
) const {
158 assert(Level
< IndentForLevel
.size());
159 if (IndentForLevel
[Level
] != -1)
160 return IndentForLevel
[Level
];
163 return getIndent(Level
- 1) + Style
.IndentWidth
;
166 const FormatStyle
&Style
;
167 const AdditionalKeywords
&Keywords
;
168 const unsigned AdditionalIndent
;
170 /// The indent in characters for each level. It remembers the indent of
171 /// previous lines (that are not PP directives) of equal or lower levels. This
172 /// is used to align formatted lines to the indent of previous non-formatted
173 /// lines. Think about the --lines parameter of clang-format.
174 SmallVector
<int> IndentForLevel
;
176 /// Offset of the current line relative to the indent level.
178 /// For example, the 'public' keywords is often indented with a negative
182 /// The current line's indent.
186 const FormatToken
*getMatchingNamespaceToken(
187 const AnnotatedLine
*Line
,
188 const SmallVectorImpl
<AnnotatedLine
*> &AnnotatedLines
) {
189 if (!Line
->startsWith(tok::r_brace
))
191 size_t StartLineIndex
= Line
->MatchingOpeningBlockLineIndex
;
192 if (StartLineIndex
== UnwrappedLine::kInvalidIndex
)
194 assert(StartLineIndex
< AnnotatedLines
.size());
195 return AnnotatedLines
[StartLineIndex
]->First
->getNamespaceToken();
198 StringRef
getNamespaceTokenText(const AnnotatedLine
*Line
) {
199 const FormatToken
*NamespaceToken
= Line
->First
->getNamespaceToken();
200 return NamespaceToken
? NamespaceToken
->TokenText
: StringRef();
203 StringRef
getMatchingNamespaceTokenText(
204 const AnnotatedLine
*Line
,
205 const SmallVectorImpl
<AnnotatedLine
*> &AnnotatedLines
) {
206 const FormatToken
*NamespaceToken
=
207 getMatchingNamespaceToken(Line
, AnnotatedLines
);
208 return NamespaceToken
? NamespaceToken
->TokenText
: StringRef();
213 LineJoiner(const FormatStyle
&Style
, const AdditionalKeywords
&Keywords
,
214 const SmallVectorImpl
<AnnotatedLine
*> &Lines
)
215 : Style(Style
), Keywords(Keywords
), End(Lines
.end()), Next(Lines
.begin()),
216 AnnotatedLines(Lines
) {}
218 /// Returns the next line, merging multiple lines into one if possible.
219 const AnnotatedLine
*getNextMergedLine(bool DryRun
,
220 LevelIndentTracker
&IndentTracker
) {
223 const AnnotatedLine
*Current
= *Next
;
224 IndentTracker
.nextLine(*Current
);
225 unsigned MergedLines
= tryFitMultipleLinesInOne(IndentTracker
, Next
, End
);
226 if (MergedLines
> 0 && Style
.ColumnLimit
== 0) {
227 // Disallow line merging if there is a break at the start of one of the
229 for (unsigned i
= 0; i
< MergedLines
; ++i
)
230 if (Next
[i
+ 1]->First
->NewlinesBefore
> 0)
234 for (unsigned i
= 0; i
< MergedLines
; ++i
)
235 join(*Next
[0], *Next
[i
+ 1]);
236 Next
= Next
+ MergedLines
+ 1;
241 /// Calculates how many lines can be merged into 1 starting at \p I.
243 tryFitMultipleLinesInOne(LevelIndentTracker
&IndentTracker
,
244 SmallVectorImpl
<AnnotatedLine
*>::const_iterator I
,
245 SmallVectorImpl
<AnnotatedLine
*>::const_iterator E
) {
246 const unsigned Indent
= IndentTracker
.getIndent();
248 // Can't join the last line with anything.
251 // We can never merge stuff if there are trailing line comments.
252 const AnnotatedLine
*TheLine
= *I
;
253 if (TheLine
->Last
->is(TT_LineComment
))
255 const auto &NextLine
= *I
[1];
256 if (NextLine
.Type
== LT_Invalid
|| NextLine
.First
->MustBreakBefore
)
258 if (TheLine
->InPPDirective
&&
259 (!NextLine
.InPPDirective
|| NextLine
.First
->HasUnescapedNewline
)) {
263 if (Style
.ColumnLimit
> 0 && Indent
> Style
.ColumnLimit
)
267 Style
.ColumnLimit
== 0 ? UINT_MAX
: Style
.ColumnLimit
- Indent
;
268 // If we already exceed the column limit, we set 'Limit' to 0. The different
269 // tryMerge..() functions can then decide whether to still do merging.
270 Limit
= TheLine
->Last
->TotalLength
> Limit
272 : Limit
- TheLine
->Last
->TotalLength
;
274 if (TheLine
->Last
->is(TT_FunctionLBrace
) &&
275 TheLine
->First
== TheLine
->Last
&&
276 !Style
.BraceWrapping
.SplitEmptyFunction
&&
277 NextLine
.First
->is(tok::r_brace
)) {
278 return tryMergeSimpleBlock(I
, E
, Limit
);
281 const auto *PreviousLine
= I
!= AnnotatedLines
.begin() ? I
[-1] : nullptr;
282 // Handle empty record blocks where the brace has already been wrapped.
283 if (PreviousLine
&& TheLine
->Last
->is(tok::l_brace
) &&
284 TheLine
->First
== TheLine
->Last
) {
285 bool EmptyBlock
= NextLine
.First
->is(tok::r_brace
);
287 const FormatToken
*Tok
= PreviousLine
->First
;
288 if (Tok
&& Tok
->is(tok::comment
))
289 Tok
= Tok
->getNextNonComment();
291 if (Tok
&& Tok
->getNamespaceToken()) {
292 return !Style
.BraceWrapping
.SplitEmptyNamespace
&& EmptyBlock
293 ? tryMergeSimpleBlock(I
, E
, Limit
)
297 if (Tok
&& Tok
->is(tok::kw_typedef
))
298 Tok
= Tok
->getNextNonComment();
299 if (Tok
&& Tok
->isOneOf(tok::kw_class
, tok::kw_struct
, tok::kw_union
,
300 tok::kw_extern
, Keywords
.kw_interface
)) {
301 return !Style
.BraceWrapping
.SplitEmptyRecord
&& EmptyBlock
302 ? tryMergeSimpleBlock(I
, E
, Limit
)
306 if (Tok
&& Tok
->is(tok::kw_template
) &&
307 Style
.BraceWrapping
.SplitEmptyRecord
&& EmptyBlock
) {
312 auto ShouldMergeShortFunctions
= [this, &I
, &NextLine
, PreviousLine
,
314 if (Style
.AllowShortFunctionsOnASingleLine
== FormatStyle::SFS_All
)
316 if (Style
.AllowShortFunctionsOnASingleLine
>= FormatStyle::SFS_Empty
&&
317 NextLine
.First
->is(tok::r_brace
)) {
321 if (Style
.AllowShortFunctionsOnASingleLine
&
322 FormatStyle::SFS_InlineOnly
) {
323 // Just checking TheLine->Level != 0 is not enough, because it
324 // provokes treating functions inside indented namespaces as short.
325 if (Style
.isJavaScript() && TheLine
->Last
->is(TT_FunctionLBrace
))
328 if (TheLine
->Level
!= 0) {
332 // TODO: Use IndentTracker to avoid loop?
333 // Find the last line with lower level.
334 const AnnotatedLine
*Line
= nullptr;
335 for (auto J
= I
- 1; J
>= AnnotatedLines
.begin(); --J
) {
337 if (!(*J
)->InPPDirective
&& !(*J
)->isComment() &&
338 (*J
)->Level
< TheLine
->Level
) {
347 // Check if the found line starts a record.
348 const auto *LastNonComment
= Line
->getLastNonComment();
349 // There must be another token (usually `{`), because we chose a
350 // non-PPDirective and non-comment line that has a smaller level.
351 assert(LastNonComment
);
352 return isRecordLBrace(*LastNonComment
);
359 bool MergeShortFunctions
= ShouldMergeShortFunctions();
361 const auto *FirstNonComment
= TheLine
->getFirstNonComment();
362 if (!FirstNonComment
)
364 // FIXME: There are probably cases where we should use FirstNonComment
365 // instead of TheLine->First.
367 if (Style
.CompactNamespaces
) {
368 if (const auto *NSToken
= TheLine
->First
->getNamespaceToken()) {
370 assert(TheLine
->MatchingClosingBlockLineIndex
> 0);
371 for (auto ClosingLineIndex
= TheLine
->MatchingClosingBlockLineIndex
- 1;
372 I
+ J
!= E
&& NSToken
->TokenText
== getNamespaceTokenText(I
[J
]) &&
373 ClosingLineIndex
== I
[J
]->MatchingClosingBlockLineIndex
&&
374 I
[J
]->Last
->TotalLength
< Limit
;
375 ++J
, --ClosingLineIndex
) {
376 Limit
-= I
[J
]->Last
->TotalLength
;
378 // Reduce indent level for bodies of namespaces which were compacted,
379 // but only if their content was indented in the first place.
380 auto *ClosingLine
= AnnotatedLines
.begin() + ClosingLineIndex
+ 1;
381 const int OutdentBy
= I
[J
]->Level
- TheLine
->Level
;
382 assert(OutdentBy
>= 0);
383 for (auto *CompactedLine
= I
+ J
; CompactedLine
<= ClosingLine
;
385 if (!(*CompactedLine
)->InPPDirective
) {
386 const int Level
= (*CompactedLine
)->Level
;
387 (*CompactedLine
)->Level
= std::max(Level
- OutdentBy
, 0);
394 if (auto nsToken
= getMatchingNamespaceToken(TheLine
, AnnotatedLines
)) {
396 unsigned openingLine
= TheLine
->MatchingOpeningBlockLineIndex
- 1;
397 for (; I
+ 1 + i
!= E
&&
398 nsToken
->TokenText
==
399 getMatchingNamespaceTokenText(I
[i
+ 1], AnnotatedLines
) &&
400 openingLine
== I
[i
+ 1]->MatchingOpeningBlockLineIndex
;
401 i
++, --openingLine
) {
402 // No space between consecutive braces.
403 I
[i
+ 1]->First
->SpacesRequiredBefore
=
404 I
[i
]->Last
->isNot(tok::r_brace
);
406 // Indent like the outer-most namespace.
407 IndentTracker
.nextLine(*I
[i
+ 1]);
413 const auto *LastNonComment
= TheLine
->getLastNonComment();
414 assert(LastNonComment
);
415 // FIXME: There are probably cases where we should use LastNonComment
416 // instead of TheLine->Last.
418 // Try to merge a function block with left brace unwrapped.
419 if (LastNonComment
->is(TT_FunctionLBrace
) &&
420 TheLine
->First
!= LastNonComment
) {
421 return MergeShortFunctions
? tryMergeSimpleBlock(I
, E
, Limit
) : 0;
423 // Try to merge a control statement block with left brace unwrapped.
424 if (TheLine
->Last
->is(tok::l_brace
) && FirstNonComment
!= TheLine
->Last
&&
425 FirstNonComment
->isOneOf(tok::kw_if
, tok::kw_while
, tok::kw_for
,
427 return Style
.AllowShortBlocksOnASingleLine
!= FormatStyle::SBS_Never
428 ? tryMergeSimpleBlock(I
, E
, Limit
)
431 // Try to merge a control statement block with left brace wrapped.
432 if (NextLine
.First
->is(tok::l_brace
)) {
433 if ((TheLine
->First
->isOneOf(tok::kw_if
, tok::kw_else
, tok::kw_while
,
434 tok::kw_for
, tok::kw_switch
, tok::kw_try
,
435 tok::kw_do
, TT_ForEachMacro
) ||
436 (TheLine
->First
->is(tok::r_brace
) && TheLine
->First
->Next
&&
437 TheLine
->First
->Next
->isOneOf(tok::kw_else
, tok::kw_catch
))) &&
438 Style
.BraceWrapping
.AfterControlStatement
==
439 FormatStyle::BWACS_MultiLine
) {
440 // If possible, merge the next line's wrapped left brace with the
441 // current line. Otherwise, leave it on the next line, as this is a
442 // multi-line control statement.
443 return (Style
.ColumnLimit
== 0 || TheLine
->Level
* Style
.IndentWidth
+
444 TheLine
->Last
->TotalLength
<=
449 if (TheLine
->First
->isOneOf(tok::kw_if
, tok::kw_else
, tok::kw_while
,
450 tok::kw_for
, TT_ForEachMacro
)) {
451 return (Style
.BraceWrapping
.AfterControlStatement
==
452 FormatStyle::BWACS_Always
)
453 ? tryMergeSimpleBlock(I
, E
, Limit
)
456 if (TheLine
->First
->isOneOf(tok::kw_else
, tok::kw_catch
) &&
457 Style
.BraceWrapping
.AfterControlStatement
==
458 FormatStyle::BWACS_MultiLine
) {
459 // This case if different from the upper BWACS_MultiLine processing
460 // in that a preceding r_brace is not on the same line as else/catch
461 // most likely because of BeforeElse/BeforeCatch set to true.
462 // If the line length doesn't fit ColumnLimit, leave l_brace on the
463 // next line to respect the BWACS_MultiLine.
464 return (Style
.ColumnLimit
== 0 ||
465 TheLine
->Last
->TotalLength
<= Style
.ColumnLimit
)
470 if (PreviousLine
&& TheLine
->First
->is(tok::l_brace
)) {
471 switch (PreviousLine
->First
->Tok
.getKind()) {
473 // Don't merge block with left brace wrapped after ObjC special blocks.
474 if (PreviousLine
->First
->Next
) {
475 tok::ObjCKeywordKind kwId
=
476 PreviousLine
->First
->Next
->Tok
.getObjCKeywordID();
477 if (kwId
== tok::objc_autoreleasepool
||
478 kwId
== tok::objc_synchronized
) {
485 case tok::kw_default
:
486 // Don't merge block with left brace wrapped after case labels.
494 // Don't merge an empty template class or struct if SplitEmptyRecords
496 if (PreviousLine
&& Style
.BraceWrapping
.SplitEmptyRecord
&&
497 TheLine
->Last
->is(tok::l_brace
) && PreviousLine
->Last
) {
498 const FormatToken
*Previous
= PreviousLine
->Last
;
500 if (Previous
->is(tok::comment
))
501 Previous
= Previous
->getPreviousNonComment();
503 if (Previous
->is(tok::greater
) && !PreviousLine
->InPPDirective
)
505 if (Previous
->is(tok::identifier
)) {
506 const FormatToken
*PreviousPrevious
=
507 Previous
->getPreviousNonComment();
508 if (PreviousPrevious
&&
509 PreviousPrevious
->isOneOf(tok::kw_class
, tok::kw_struct
)) {
517 if (TheLine
->First
->is(TT_SwitchExpressionLabel
)) {
518 return Style
.AllowShortCaseExpressionOnASingleLine
519 ? tryMergeShortCaseLabels(I
, E
, Limit
)
523 if (TheLine
->Last
->is(tok::l_brace
)) {
524 bool ShouldMerge
= false;
525 // Try to merge records.
526 if (TheLine
->Last
->is(TT_EnumLBrace
)) {
527 ShouldMerge
= Style
.AllowShortEnumsOnASingleLine
;
528 } else if (TheLine
->Last
->is(TT_RequiresExpressionLBrace
)) {
529 ShouldMerge
= Style
.AllowShortCompoundRequirementOnASingleLine
;
530 } else if (TheLine
->Last
->isOneOf(TT_ClassLBrace
, TT_StructLBrace
)) {
531 // NOTE: We use AfterClass (whereas AfterStruct exists) for both classes
532 // and structs, but it seems that wrapping is still handled correctly
534 ShouldMerge
= !Style
.BraceWrapping
.AfterClass
||
535 (NextLine
.First
->is(tok::r_brace
) &&
536 !Style
.BraceWrapping
.SplitEmptyRecord
);
537 } else if (TheLine
->InPPDirective
||
538 !TheLine
->First
->isOneOf(tok::kw_class
, tok::kw_enum
,
540 // Try to merge a block with left brace unwrapped that wasn't yet
542 ShouldMerge
= !Style
.BraceWrapping
.AfterFunction
||
543 (NextLine
.First
->is(tok::r_brace
) &&
544 !Style
.BraceWrapping
.SplitEmptyFunction
);
546 return ShouldMerge
? tryMergeSimpleBlock(I
, E
, Limit
) : 0;
549 // Try to merge a function block with left brace wrapped.
550 if (NextLine
.First
->is(TT_FunctionLBrace
) &&
551 Style
.BraceWrapping
.AfterFunction
) {
552 if (NextLine
.Last
->is(TT_LineComment
))
555 // Check for Limit <= 2 to account for the " {".
556 if (Limit
<= 2 || (Style
.ColumnLimit
== 0 && containsMustBreak(TheLine
)))
560 unsigned MergedLines
= 0;
561 if (MergeShortFunctions
||
562 (Style
.AllowShortFunctionsOnASingleLine
>= FormatStyle::SFS_Empty
&&
563 NextLine
.First
== NextLine
.Last
&& I
+ 2 != E
&&
564 I
[2]->First
->is(tok::r_brace
))) {
565 MergedLines
= tryMergeSimpleBlock(I
+ 1, E
, Limit
);
566 // If we managed to merge the block, count the function header, which is
567 // on a separate line.
573 auto IsElseLine
= [&TheLine
]() -> bool {
574 const FormatToken
*First
= TheLine
->First
;
575 if (First
->is(tok::kw_else
))
578 return First
->is(tok::r_brace
) && First
->Next
&&
579 First
->Next
->is(tok::kw_else
);
581 if (TheLine
->First
->is(tok::kw_if
) ||
582 (IsElseLine() && (Style
.AllowShortIfStatementsOnASingleLine
==
583 FormatStyle::SIS_AllIfsAndElse
))) {
584 return Style
.AllowShortIfStatementsOnASingleLine
585 ? tryMergeSimpleControlStatement(I
, E
, Limit
)
588 if (TheLine
->First
->isOneOf(tok::kw_for
, tok::kw_while
, tok::kw_do
,
590 return Style
.AllowShortLoopsOnASingleLine
591 ? tryMergeSimpleControlStatement(I
, E
, Limit
)
594 if (TheLine
->First
->isOneOf(tok::kw_case
, tok::kw_default
)) {
595 return Style
.AllowShortCaseLabelsOnASingleLine
596 ? tryMergeShortCaseLabels(I
, E
, Limit
)
599 if (TheLine
->InPPDirective
&&
600 (TheLine
->First
->HasUnescapedNewline
|| TheLine
->First
->IsFirst
)) {
601 return tryMergeSimplePPDirective(I
, E
, Limit
);
607 tryMergeSimplePPDirective(SmallVectorImpl
<AnnotatedLine
*>::const_iterator I
,
608 SmallVectorImpl
<AnnotatedLine
*>::const_iterator E
,
612 if (I
+ 2 != E
&& I
[2]->InPPDirective
&& !I
[2]->First
->HasUnescapedNewline
)
614 if (1 + I
[1]->Last
->TotalLength
> Limit
)
619 unsigned tryMergeSimpleControlStatement(
620 SmallVectorImpl
<AnnotatedLine
*>::const_iterator I
,
621 SmallVectorImpl
<AnnotatedLine
*>::const_iterator E
, unsigned Limit
) {
624 if (Style
.BraceWrapping
.AfterControlStatement
==
625 FormatStyle::BWACS_Always
&&
626 I
[1]->First
->is(tok::l_brace
) &&
627 Style
.AllowShortBlocksOnASingleLine
== FormatStyle::SBS_Never
) {
630 if (I
[1]->InPPDirective
!= (*I
)->InPPDirective
||
631 (I
[1]->InPPDirective
&& I
[1]->First
->HasUnescapedNewline
)) {
634 Limit
= limitConsideringMacros(I
+ 1, E
, Limit
);
635 AnnotatedLine
&Line
= **I
;
636 if (Line
.First
->isNot(tok::kw_do
) && Line
.First
->isNot(tok::kw_else
) &&
637 Line
.Last
->isNot(tok::kw_else
) && Line
.Last
->isNot(tok::r_paren
)) {
640 // Only merge `do while` if `do` is the only statement on the line.
641 if (Line
.First
->is(tok::kw_do
) && Line
.Last
->isNot(tok::kw_do
))
643 if (1 + I
[1]->Last
->TotalLength
> Limit
)
645 // Don't merge with loops, ifs, a single semicolon or a line comment.
646 if (I
[1]->First
->isOneOf(tok::semi
, tok::kw_if
, tok::kw_for
, tok::kw_while
,
647 TT_ForEachMacro
, TT_LineComment
)) {
650 // Only inline simple if's (no nested if or else), unless specified
651 if (Style
.AllowShortIfStatementsOnASingleLine
==
652 FormatStyle::SIS_WithoutElse
) {
653 if (I
+ 2 != E
&& Line
.startsWith(tok::kw_if
) &&
654 I
[2]->First
->is(tok::kw_else
)) {
662 tryMergeShortCaseLabels(SmallVectorImpl
<AnnotatedLine
*>::const_iterator I
,
663 SmallVectorImpl
<AnnotatedLine
*>::const_iterator E
,
665 if (Limit
== 0 || I
+ 1 == E
||
666 I
[1]->First
->isOneOf(tok::kw_case
, tok::kw_default
)) {
669 if (I
[0]->Last
->is(tok::l_brace
) || I
[1]->First
->is(tok::l_brace
))
671 unsigned NumStmts
= 0;
673 bool EndsWithComment
= false;
674 bool InPPDirective
= I
[0]->InPPDirective
;
675 bool InMacroBody
= I
[0]->InMacroBody
;
676 const unsigned Level
= I
[0]->Level
;
677 for (; NumStmts
< 3; ++NumStmts
) {
678 if (I
+ 1 + NumStmts
== E
)
680 const AnnotatedLine
*Line
= I
[1 + NumStmts
];
681 if (Line
->InPPDirective
!= InPPDirective
)
683 if (Line
->InMacroBody
!= InMacroBody
)
685 if (Line
->First
->isOneOf(tok::kw_case
, tok::kw_default
, tok::r_brace
))
687 if (Line
->First
->isOneOf(tok::kw_if
, tok::kw_for
, tok::kw_switch
,
692 if (Line
->First
->is(tok::comment
)) {
693 if (Level
!= Line
->Level
)
695 SmallVectorImpl
<AnnotatedLine
*>::const_iterator J
= I
+ 2 + NumStmts
;
696 for (; J
!= E
; ++J
) {
698 if (Line
->InPPDirective
!= InPPDirective
)
700 if (Line
->First
->isOneOf(tok::kw_case
, tok::kw_default
, tok::r_brace
))
702 if (Line
->First
->isNot(tok::comment
) || Level
!= Line
->Level
)
707 if (Line
->Last
->is(tok::comment
))
708 EndsWithComment
= true;
709 Length
+= I
[1 + NumStmts
]->Last
->TotalLength
+ 1; // 1 for the space.
711 if (NumStmts
== 0 || NumStmts
== 3 || Length
> Limit
)
717 tryMergeSimpleBlock(SmallVectorImpl
<AnnotatedLine
*>::const_iterator I
,
718 SmallVectorImpl
<AnnotatedLine
*>::const_iterator E
,
720 // Don't merge with a preprocessor directive.
721 if (I
[1]->Type
== LT_PreprocessorDirective
)
724 AnnotatedLine
&Line
= **I
;
726 // Don't merge ObjC @ keywords and methods.
727 // FIXME: If an option to allow short exception handling clauses on a single
728 // line is added, change this to not return for @try and friends.
729 if (Style
.Language
!= FormatStyle::LK_Java
&&
730 Line
.First
->isOneOf(tok::at
, tok::minus
, tok::plus
)) {
734 // Check that the current line allows merging. This depends on whether we
735 // are in a control flow statements as well as several style flags.
736 if (Line
.First
->is(tok::kw_case
) ||
737 (Line
.First
->Next
&& Line
.First
->Next
->is(tok::kw_else
))) {
740 // default: in switch statement
741 if (Line
.First
->is(tok::kw_default
)) {
742 const FormatToken
*Tok
= Line
.First
->getNextNonComment();
743 if (Tok
&& Tok
->is(tok::colon
))
747 auto IsCtrlStmt
= [](const auto &Line
) {
748 return Line
.First
->isOneOf(tok::kw_if
, tok::kw_else
, tok::kw_while
,
749 tok::kw_do
, tok::kw_for
, TT_ForEachMacro
);
752 const bool IsSplitBlock
=
753 Style
.AllowShortBlocksOnASingleLine
== FormatStyle::SBS_Never
||
754 (Style
.AllowShortBlocksOnASingleLine
== FormatStyle::SBS_Empty
&&
755 I
[1]->First
->isNot(tok::r_brace
));
757 if (IsCtrlStmt(Line
) ||
758 Line
.First
->isOneOf(tok::kw_try
, tok::kw___try
, tok::kw_catch
,
759 tok::kw___finally
, tok::r_brace
,
760 Keywords
.kw___except
)) {
763 // Don't merge when we can't except the case when
764 // the control statement block is empty
765 if (!Style
.AllowShortIfStatementsOnASingleLine
&&
766 Line
.First
->isOneOf(tok::kw_if
, tok::kw_else
) &&
767 !Style
.BraceWrapping
.AfterControlStatement
&&
768 I
[1]->First
->isNot(tok::r_brace
)) {
771 if (!Style
.AllowShortIfStatementsOnASingleLine
&&
772 Line
.First
->isOneOf(tok::kw_if
, tok::kw_else
) &&
773 Style
.BraceWrapping
.AfterControlStatement
==
774 FormatStyle::BWACS_Always
&&
775 I
+ 2 != E
&& I
[2]->First
->isNot(tok::r_brace
)) {
778 if (!Style
.AllowShortLoopsOnASingleLine
&&
779 Line
.First
->isOneOf(tok::kw_while
, tok::kw_do
, tok::kw_for
,
781 !Style
.BraceWrapping
.AfterControlStatement
&&
782 I
[1]->First
->isNot(tok::r_brace
)) {
785 if (!Style
.AllowShortLoopsOnASingleLine
&&
786 Line
.First
->isOneOf(tok::kw_while
, tok::kw_do
, tok::kw_for
,
788 Style
.BraceWrapping
.AfterControlStatement
==
789 FormatStyle::BWACS_Always
&&
790 I
+ 2 != E
&& I
[2]->First
->isNot(tok::r_brace
)) {
793 // FIXME: Consider an option to allow short exception handling clauses on
795 // FIXME: This isn't covered by tests.
796 // FIXME: For catch, __except, __finally the first token on the line
797 // is '}', so this isn't correct here.
798 if (Line
.First
->isOneOf(tok::kw_try
, tok::kw___try
, tok::kw_catch
,
799 Keywords
.kw___except
, tok::kw___finally
)) {
804 if (Line
.endsWith(tok::l_brace
)) {
805 if (Style
.AllowShortBlocksOnASingleLine
== FormatStyle::SBS_Never
&&
806 Line
.First
->is(TT_BlockLBrace
)) {
810 if (IsSplitBlock
&& Line
.First
== Line
.Last
&&
811 I
> AnnotatedLines
.begin() &&
812 (I
[-1]->endsWith(tok::kw_else
) || IsCtrlStmt(*I
[-1]))) {
815 FormatToken
*Tok
= I
[1]->First
;
816 auto ShouldMerge
= [Tok
]() {
817 if (Tok
->isNot(tok::r_brace
) || Tok
->MustBreakBefore
)
819 const FormatToken
*Next
= Tok
->getNextNonComment();
820 return !Next
|| Next
->is(tok::semi
);
824 // We merge empty blocks even if the line exceeds the column limit.
825 Tok
->SpacesRequiredBefore
=
826 (Style
.SpaceInEmptyBlock
|| Line
.Last
->is(tok::comment
)) ? 1 : 0;
827 Tok
->CanBreakBefore
= true;
829 } else if (Limit
!= 0 && !Line
.startsWithNamespace() &&
830 !startsExternCBlock(Line
)) {
831 // We don't merge short records.
832 if (isRecordLBrace(*Line
.Last
))
835 // Check that we still have three lines and they fit into the limit.
836 if (I
+ 2 == E
|| I
[2]->Type
== LT_Invalid
)
838 Limit
= limitConsideringMacros(I
+ 2, E
, Limit
);
840 if (!nextTwoLinesFitInto(I
, Limit
))
843 // Second, check that the next line does not contain any braces - if it
844 // does, readability declines when putting it into a single line.
845 if (I
[1]->Last
->is(TT_LineComment
))
848 if (Tok
->is(tok::l_brace
) && Tok
->isNot(BK_BracedInit
))
853 // Last, check that the third line starts with a closing brace.
855 if (Tok
->isNot(tok::r_brace
))
858 // Don't merge "if (a) { .. } else {".
859 if (Tok
->Next
&& Tok
->Next
->is(tok::kw_else
))
862 // Don't merge a trailing multi-line control statement block like:
865 // { <-- current Line
868 if (Line
.First
== Line
.Last
&& Line
.First
->isNot(TT_FunctionLBrace
) &&
869 Style
.BraceWrapping
.AfterControlStatement
==
870 FormatStyle::BWACS_MultiLine
) {
876 } else if (I
[1]->First
->is(tok::l_brace
)) {
877 if (I
[1]->Last
->is(TT_LineComment
))
880 // Check for Limit <= 2 to account for the " {".
881 if (Limit
<= 2 || (Style
.ColumnLimit
== 0 && containsMustBreak(*I
)))
884 unsigned MergedLines
= 0;
885 if (Style
.AllowShortBlocksOnASingleLine
!= FormatStyle::SBS_Never
||
886 (I
[1]->First
== I
[1]->Last
&& I
+ 2 != E
&&
887 I
[2]->First
->is(tok::r_brace
))) {
888 MergedLines
= tryMergeSimpleBlock(I
+ 1, E
, Limit
);
889 // If we managed to merge the block, count the statement header, which
890 // is on a separate line.
899 /// Returns the modified column limit for \p I if it is inside a macro and
900 /// needs a trailing '\'.
902 limitConsideringMacros(SmallVectorImpl
<AnnotatedLine
*>::const_iterator I
,
903 SmallVectorImpl
<AnnotatedLine
*>::const_iterator E
,
905 if (I
[0]->InPPDirective
&& I
+ 1 != E
&&
906 !I
[1]->First
->HasUnescapedNewline
&& I
[1]->First
->isNot(tok::eof
)) {
907 return Limit
< 2 ? 0 : Limit
- 2;
912 bool nextTwoLinesFitInto(SmallVectorImpl
<AnnotatedLine
*>::const_iterator I
,
914 if (I
[1]->First
->MustBreakBefore
|| I
[2]->First
->MustBreakBefore
)
916 return 1 + I
[1]->Last
->TotalLength
+ 1 + I
[2]->Last
->TotalLength
<= Limit
;
919 bool containsMustBreak(const AnnotatedLine
*Line
) {
921 // Ignore the first token, because in this situation, it applies more to the
922 // last token of the previous line.
923 for (const FormatToken
*Tok
= Line
->First
->Next
; Tok
; Tok
= Tok
->Next
)
924 if (Tok
->MustBreakBefore
)
929 void join(AnnotatedLine
&A
, const AnnotatedLine
&B
) {
930 assert(!A
.Last
->Next
);
931 assert(!B
.First
->Previous
);
934 A
.Last
->Next
= B
.First
;
935 B
.First
->Previous
= A
.Last
;
936 B
.First
->CanBreakBefore
= true;
937 unsigned LengthA
= A
.Last
->TotalLength
+ B
.First
->SpacesRequiredBefore
;
938 for (FormatToken
*Tok
= B
.First
; Tok
; Tok
= Tok
->Next
) {
939 Tok
->TotalLength
+= LengthA
;
944 const FormatStyle
&Style
;
945 const AdditionalKeywords
&Keywords
;
946 const SmallVectorImpl
<AnnotatedLine
*>::const_iterator End
;
948 SmallVectorImpl
<AnnotatedLine
*>::const_iterator Next
;
949 const SmallVectorImpl
<AnnotatedLine
*> &AnnotatedLines
;
952 static void markFinalized(FormatToken
*Tok
) {
953 if (Tok
->is(tok::hash
) && !Tok
->Previous
&& Tok
->Next
&&
954 Tok
->Next
->isOneOf(tok::pp_if
, tok::pp_ifdef
, tok::pp_ifndef
,
955 tok::pp_elif
, tok::pp_elifdef
, tok::pp_elifndef
,
956 tok::pp_else
, tok::pp_endif
)) {
959 for (; Tok
; Tok
= Tok
->Next
) {
960 if (Tok
->MacroCtx
&& Tok
->MacroCtx
->Role
== MR_ExpandedArg
) {
961 // In the first pass we format all macro arguments in the expanded token
962 // stream. Instead of finalizing the macro arguments, we mark that they
963 // will be modified as unexpanded arguments (as part of the macro call
964 // formatting) in the next pass.
965 Tok
->MacroCtx
->Role
= MR_UnexpandedArg
;
966 // Reset whether spaces or a line break are required before this token, as
967 // that is context dependent, and that context may change when formatting
968 // the macro call. For example, given M(x) -> 2 * x, and the macro call
969 // M(var), the token 'var' will have SpacesRequiredBefore = 1 after being
970 // formatted as part of the expanded macro, but SpacesRequiredBefore = 0
971 // for its position within the macro call.
972 Tok
->SpacesRequiredBefore
= 0;
973 if (!Tok
->MustBreakBeforeFinalized
)
974 Tok
->MustBreakBefore
= 0;
976 Tok
->Finalized
= true;
982 static void printLineState(const LineState
&State
) {
983 llvm::dbgs() << "State: ";
984 for (const ParenState
&P
: State
.Stack
) {
985 llvm::dbgs() << (P
.Tok
? P
.Tok
->TokenText
: "F") << "|" << P
.Indent
<< "|"
986 << P
.LastSpace
<< "|" << P
.NestedBlockIndent
<< " ";
988 llvm::dbgs() << State
.NextToken
->TokenText
<< "\n";
992 /// Base class for classes that format one \c AnnotatedLine.
993 class LineFormatter
{
995 LineFormatter(ContinuationIndenter
*Indenter
, WhitespaceManager
*Whitespaces
,
996 const FormatStyle
&Style
,
997 UnwrappedLineFormatter
*BlockFormatter
)
998 : Indenter(Indenter
), Whitespaces(Whitespaces
), Style(Style
),
999 BlockFormatter(BlockFormatter
) {}
1000 virtual ~LineFormatter() {}
1002 /// Formats an \c AnnotatedLine and returns the penalty.
1004 /// If \p DryRun is \c false, directly applies the changes.
1005 virtual unsigned formatLine(const AnnotatedLine
&Line
, unsigned FirstIndent
,
1006 unsigned FirstStartColumn
, bool DryRun
) = 0;
1009 /// If the \p State's next token is an r_brace closing a nested block,
1010 /// format the nested block before it.
1012 /// Returns \c true if all children could be placed successfully and adapts
1013 /// \p Penalty as well as \p State. If \p DryRun is false, also directly
1014 /// creates changes using \c Whitespaces.
1016 /// The crucial idea here is that children always get formatted upon
1017 /// encountering the closing brace right after the nested block. Now, if we
1018 /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is
1019 /// \c false), the entire block has to be kept on the same line (which is only
1020 /// possible if it fits on the line, only contains a single statement, etc.
1022 /// If \p NewLine is true, we format the nested block on separate lines, i.e.
1023 /// break after the "{", format all lines with correct indentation and the put
1024 /// the closing "}" on yet another new line.
1026 /// This enables us to keep the simple structure of the
1027 /// \c UnwrappedLineFormatter, where we only have two options for each token:
1028 /// break or don't break.
1029 bool formatChildren(LineState
&State
, bool NewLine
, bool DryRun
,
1030 unsigned &Penalty
) {
1031 const FormatToken
*LBrace
= State
.NextToken
->getPreviousNonComment();
1032 bool HasLBrace
= LBrace
&& LBrace
->is(tok::l_brace
) && LBrace
->is(BK_Block
);
1033 FormatToken
&Previous
= *State
.NextToken
->Previous
;
1034 if (Previous
.Children
.size() == 0 || (!HasLBrace
&& !LBrace
->MacroParent
)) {
1035 // The previous token does not open a block. Nothing to do. We don't
1036 // assert so that we can simply call this function for all tokens.
1040 if (NewLine
|| Previous
.MacroParent
) {
1041 const ParenState
&P
= State
.Stack
.back();
1043 int AdditionalIndent
=
1044 P
.Indent
- Previous
.Children
[0]->Level
* Style
.IndentWidth
;
1046 BlockFormatter
->format(Previous
.Children
, DryRun
, AdditionalIndent
,
1047 /*FixBadIndentation=*/true);
1051 if (Previous
.Children
[0]->First
->MustBreakBefore
)
1054 // Cannot merge into one line if this line ends on a comment.
1055 if (Previous
.is(tok::comment
))
1058 // Cannot merge multiple statements into a single line.
1059 if (Previous
.Children
.size() > 1)
1062 const AnnotatedLine
*Child
= Previous
.Children
[0];
1063 // We can't put the closing "}" on a line with a trailing comment.
1064 if (Child
->Last
->isTrailingComment())
1067 // If the child line exceeds the column limit, we wouldn't want to merge it.
1068 // We add +2 for the trailing " }".
1069 if (Style
.ColumnLimit
> 0 &&
1070 Child
->Last
->TotalLength
+ State
.Column
+ 2 > Style
.ColumnLimit
) {
1075 Whitespaces
->replaceWhitespace(
1076 *Child
->First
, /*Newlines=*/0, /*Spaces=*/1,
1077 /*StartOfTokenColumn=*/State
.Column
, /*IsAligned=*/false,
1078 State
.Line
->InPPDirective
);
1081 formatLine(*Child
, State
.Column
+ 1, /*FirstStartColumn=*/0, DryRun
);
1083 markFinalized(Child
->First
);
1085 State
.Column
+= 1 + Child
->Last
->TotalLength
;
1089 ContinuationIndenter
*Indenter
;
1092 WhitespaceManager
*Whitespaces
;
1093 const FormatStyle
&Style
;
1094 UnwrappedLineFormatter
*BlockFormatter
;
1097 /// Formatter that keeps the existing line breaks.
1098 class NoColumnLimitLineFormatter
: public LineFormatter
{
1100 NoColumnLimitLineFormatter(ContinuationIndenter
*Indenter
,
1101 WhitespaceManager
*Whitespaces
,
1102 const FormatStyle
&Style
,
1103 UnwrappedLineFormatter
*BlockFormatter
)
1104 : LineFormatter(Indenter
, Whitespaces
, Style
, BlockFormatter
) {}
1106 /// Formats the line, simply keeping all of the input's line breaking
1108 unsigned formatLine(const AnnotatedLine
&Line
, unsigned FirstIndent
,
1109 unsigned FirstStartColumn
, bool DryRun
) override
{
1111 LineState State
= Indenter
->getInitialState(FirstIndent
, FirstStartColumn
,
1112 &Line
, /*DryRun=*/false);
1113 while (State
.NextToken
) {
1115 Indenter
->mustBreak(State
) ||
1116 (Indenter
->canBreak(State
) && State
.NextToken
->NewlinesBefore
> 0);
1117 unsigned Penalty
= 0;
1118 formatChildren(State
, Newline
, /*DryRun=*/false, Penalty
);
1119 Indenter
->addTokenToState(State
, Newline
, /*DryRun=*/false);
1125 /// Formatter that puts all tokens into a single line without breaks.
1126 class NoLineBreakFormatter
: public LineFormatter
{
1128 NoLineBreakFormatter(ContinuationIndenter
*Indenter
,
1129 WhitespaceManager
*Whitespaces
, const FormatStyle
&Style
,
1130 UnwrappedLineFormatter
*BlockFormatter
)
1131 : LineFormatter(Indenter
, Whitespaces
, Style
, BlockFormatter
) {}
1133 /// Puts all tokens into a single line.
1134 unsigned formatLine(const AnnotatedLine
&Line
, unsigned FirstIndent
,
1135 unsigned FirstStartColumn
, bool DryRun
) override
{
1136 unsigned Penalty
= 0;
1138 Indenter
->getInitialState(FirstIndent
, FirstStartColumn
, &Line
, DryRun
);
1139 while (State
.NextToken
) {
1140 formatChildren(State
, /*NewLine=*/false, DryRun
, Penalty
);
1141 Indenter
->addTokenToState(
1142 State
, /*Newline=*/State
.NextToken
->MustBreakBefore
, DryRun
);
1148 /// Finds the best way to break lines.
1149 class OptimizingLineFormatter
: public LineFormatter
{
1151 OptimizingLineFormatter(ContinuationIndenter
*Indenter
,
1152 WhitespaceManager
*Whitespaces
,
1153 const FormatStyle
&Style
,
1154 UnwrappedLineFormatter
*BlockFormatter
)
1155 : LineFormatter(Indenter
, Whitespaces
, Style
, BlockFormatter
) {}
1157 /// Formats the line by finding the best line breaks with line lengths
1158 /// below the column limit.
1159 unsigned formatLine(const AnnotatedLine
&Line
, unsigned FirstIndent
,
1160 unsigned FirstStartColumn
, bool DryRun
) override
{
1162 Indenter
->getInitialState(FirstIndent
, FirstStartColumn
, &Line
, DryRun
);
1164 // If the ObjC method declaration does not fit on a line, we should format
1165 // it with one arg per line.
1166 if (State
.Line
->Type
== LT_ObjCMethodDecl
)
1167 State
.Stack
.back().BreakBeforeParameter
= true;
1169 // Find best solution in solution space.
1170 return analyzeSolutionSpace(State
, DryRun
);
1174 struct CompareLineStatePointers
{
1175 bool operator()(LineState
*obj1
, LineState
*obj2
) const {
1176 return *obj1
< *obj2
;
1180 /// A pair of <penalty, count> that is used to prioritize the BFS on.
1182 /// In case of equal penalties, we want to prefer states that were inserted
1183 /// first. During state generation we make sure that we insert states first
1184 /// that break the line as late as possible.
1185 typedef std::pair
<unsigned, unsigned> OrderedPenalty
;
1187 /// An edge in the solution space from \c Previous->State to \c State,
1188 /// inserting a newline dependent on the \c NewLine.
1190 StateNode(const LineState
&State
, bool NewLine
, StateNode
*Previous
)
1191 : State(State
), NewLine(NewLine
), Previous(Previous
) {}
1194 StateNode
*Previous
;
1197 /// An item in the prioritized BFS search queue. The \c StateNode's
1198 /// \c State has the given \c OrderedPenalty.
1199 typedef std::pair
<OrderedPenalty
, StateNode
*> QueueItem
;
1201 /// The BFS queue type.
1202 typedef std::priority_queue
<QueueItem
, SmallVector
<QueueItem
>,
1203 std::greater
<QueueItem
>>
1206 /// Analyze the entire solution space starting from \p InitialState.
1208 /// This implements a variant of Dijkstra's algorithm on the graph that spans
1209 /// the solution space (\c LineStates are the nodes). The algorithm tries to
1210 /// find the shortest path (the one with lowest penalty) from \p InitialState
1211 /// to a state where all tokens are placed. Returns the penalty.
1213 /// If \p DryRun is \c false, directly applies the changes.
1214 unsigned analyzeSolutionSpace(LineState
&InitialState
, bool DryRun
) {
1215 std::set
<LineState
*, CompareLineStatePointers
> Seen
;
1217 // Increasing count of \c StateNode items we have created. This is used to
1218 // create a deterministic order independent of the container.
1222 // Insert start element into queue.
1223 StateNode
*RootNode
=
1224 new (Allocator
.Allocate()) StateNode(InitialState
, false, nullptr);
1225 Queue
.push(QueueItem(OrderedPenalty(0, Count
), RootNode
));
1228 unsigned Penalty
= 0;
1230 // While not empty, take first element and follow edges.
1231 while (!Queue
.empty()) {
1232 // Quit if we still haven't found a solution by now.
1233 if (Count
> 25'000'000)
1236 Penalty
= Queue
.top().first
.first
;
1237 StateNode
*Node
= Queue
.top().second
;
1238 if (!Node
->State
.NextToken
) {
1239 LLVM_DEBUG(llvm::dbgs()
1240 << "\n---\nPenalty for line: " << Penalty
<< "\n");
1245 // Cut off the analysis of certain solutions if the analysis gets too
1246 // complex. See description of IgnoreStackForComparison.
1248 Node
->State
.IgnoreStackForComparison
= true;
1250 if (!Seen
.insert(&Node
->State
).second
) {
1251 // State already examined with lower penalty.
1255 FormatDecision LastFormat
= Node
->State
.NextToken
->getDecision();
1256 if (LastFormat
== FD_Unformatted
|| LastFormat
== FD_Continue
)
1257 addNextStateToQueue(Penalty
, Node
, /*NewLine=*/false, &Count
, &Queue
);
1258 if (LastFormat
== FD_Unformatted
|| LastFormat
== FD_Break
)
1259 addNextStateToQueue(Penalty
, Node
, /*NewLine=*/true, &Count
, &Queue
);
1262 if (Queue
.empty()) {
1263 // We were unable to find a solution, do nothing.
1264 // FIXME: Add diagnostic?
1265 LLVM_DEBUG(llvm::dbgs() << "Could not find a solution.\n");
1269 // Reconstruct the solution.
1271 reconstructPath(InitialState
, Queue
.top().second
);
1273 LLVM_DEBUG(llvm::dbgs()
1274 << "Total number of analyzed states: " << Count
<< "\n");
1275 LLVM_DEBUG(llvm::dbgs() << "---\n");
1280 /// Add the following state to the analysis queue \c Queue.
1282 /// Assume the current state is \p PreviousNode and has been reached with a
1283 /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true.
1284 void addNextStateToQueue(unsigned Penalty
, StateNode
*PreviousNode
,
1285 bool NewLine
, unsigned *Count
, QueueType
*Queue
) {
1286 if (NewLine
&& !Indenter
->canBreak(PreviousNode
->State
))
1288 if (!NewLine
&& Indenter
->mustBreak(PreviousNode
->State
))
1291 StateNode
*Node
= new (Allocator
.Allocate())
1292 StateNode(PreviousNode
->State
, NewLine
, PreviousNode
);
1293 if (!formatChildren(Node
->State
, NewLine
, /*DryRun=*/true, Penalty
))
1296 Penalty
+= Indenter
->addTokenToState(Node
->State
, NewLine
, true);
1298 Queue
->push(QueueItem(OrderedPenalty(Penalty
, *Count
), Node
));
1302 /// Applies the best formatting by reconstructing the path in the
1303 /// solution space that leads to \c Best.
1304 void reconstructPath(LineState
&State
, StateNode
*Best
) {
1305 llvm::SmallVector
<StateNode
*> Path
;
1306 // We do not need a break before the initial token.
1307 while (Best
->Previous
) {
1308 Path
.push_back(Best
);
1309 Best
= Best
->Previous
;
1311 for (const auto &Node
: llvm::reverse(Path
)) {
1312 unsigned Penalty
= 0;
1313 formatChildren(State
, Node
->NewLine
, /*DryRun=*/false, Penalty
);
1314 Penalty
+= Indenter
->addTokenToState(State
, Node
->NewLine
, false);
1317 printLineState(Node
->Previous
->State
);
1318 if (Node
->NewLine
) {
1319 llvm::dbgs() << "Penalty for placing "
1320 << Node
->Previous
->State
.NextToken
->Tok
.getName()
1321 << " on a new line: " << Penalty
<< "\n";
1327 llvm::SpecificBumpPtrAllocator
<StateNode
> Allocator
;
1330 } // anonymous namespace
1332 unsigned UnwrappedLineFormatter::format(
1333 const SmallVectorImpl
<AnnotatedLine
*> &Lines
, bool DryRun
,
1334 int AdditionalIndent
, bool FixBadIndentation
, unsigned FirstStartColumn
,
1335 unsigned NextStartColumn
, unsigned LastStartColumn
) {
1336 LineJoiner
Joiner(Style
, Keywords
, Lines
);
1338 // Try to look up already computed penalty in DryRun-mode.
1339 std::pair
<const SmallVectorImpl
<AnnotatedLine
*> *, unsigned> CacheKey(
1340 &Lines
, AdditionalIndent
);
1341 auto CacheIt
= PenaltyCache
.find(CacheKey
);
1342 if (DryRun
&& CacheIt
!= PenaltyCache
.end())
1343 return CacheIt
->second
;
1345 assert(!Lines
.empty());
1346 unsigned Penalty
= 0;
1347 LevelIndentTracker
IndentTracker(Style
, Keywords
, Lines
[0]->Level
,
1349 const AnnotatedLine
*PrevPrevLine
= nullptr;
1350 const AnnotatedLine
*PreviousLine
= nullptr;
1351 const AnnotatedLine
*NextLine
= nullptr;
1353 // The minimum level of consecutive lines that have been formatted.
1354 unsigned RangeMinLevel
= UINT_MAX
;
1356 bool FirstLine
= true;
1357 for (const AnnotatedLine
*Line
=
1358 Joiner
.getNextMergedLine(DryRun
, IndentTracker
);
1359 Line
; PrevPrevLine
= PreviousLine
, PreviousLine
= Line
, Line
= NextLine
,
1360 FirstLine
= false) {
1361 assert(Line
->First
);
1362 const AnnotatedLine
&TheLine
= *Line
;
1363 unsigned Indent
= IndentTracker
.getIndent();
1365 // We continue formatting unchanged lines to adjust their indent, e.g. if a
1366 // scope was added. However, we need to carefully stop doing this when we
1367 // exit the scope of affected lines to prevent indenting the entire
1368 // remaining file if it currently missing a closing brace.
1369 bool PreviousRBrace
=
1370 PreviousLine
&& PreviousLine
->startsWith(tok::r_brace
);
1371 bool ContinueFormatting
=
1372 TheLine
.Level
> RangeMinLevel
||
1373 (TheLine
.Level
== RangeMinLevel
&& !PreviousRBrace
&&
1374 !TheLine
.startsWith(tok::r_brace
));
1376 bool FixIndentation
= (FixBadIndentation
|| ContinueFormatting
) &&
1377 Indent
!= TheLine
.First
->OriginalColumn
;
1378 bool ShouldFormat
= TheLine
.Affected
|| FixIndentation
;
1379 // We cannot format this line; if the reason is that the line had a
1380 // parsing error, remember that.
1381 if (ShouldFormat
&& TheLine
.Type
== LT_Invalid
&& Status
) {
1382 Status
->FormatComplete
= false;
1384 SourceMgr
.getSpellingLineNumber(TheLine
.First
->Tok
.getLocation());
1387 if (ShouldFormat
&& TheLine
.Type
!= LT_Invalid
) {
1389 bool LastLine
= TheLine
.First
->is(tok::eof
);
1390 formatFirstToken(TheLine
, PreviousLine
, PrevPrevLine
, Lines
, Indent
,
1391 LastLine
? LastStartColumn
: NextStartColumn
+ Indent
);
1394 NextLine
= Joiner
.getNextMergedLine(DryRun
, IndentTracker
);
1395 unsigned ColumnLimit
= getColumnLimit(TheLine
.InPPDirective
, NextLine
);
1396 bool FitsIntoOneLine
=
1397 !TheLine
.ContainsMacroCall
&&
1398 (TheLine
.Last
->TotalLength
+ Indent
<= ColumnLimit
||
1399 (TheLine
.Type
== LT_ImportStatement
&&
1400 (!Style
.isJavaScript() || !Style
.JavaScriptWrapImports
)) ||
1401 (Style
.isCSharp() &&
1402 TheLine
.InPPDirective
)); // don't split #regions in C#
1403 if (Style
.ColumnLimit
== 0) {
1404 NoColumnLimitLineFormatter(Indenter
, Whitespaces
, Style
, this)
1405 .formatLine(TheLine
, NextStartColumn
+ Indent
,
1406 FirstLine
? FirstStartColumn
: 0, DryRun
);
1407 } else if (FitsIntoOneLine
) {
1408 Penalty
+= NoLineBreakFormatter(Indenter
, Whitespaces
, Style
, this)
1409 .formatLine(TheLine
, NextStartColumn
+ Indent
,
1410 FirstLine
? FirstStartColumn
: 0, DryRun
);
1412 Penalty
+= OptimizingLineFormatter(Indenter
, Whitespaces
, Style
, this)
1413 .formatLine(TheLine
, NextStartColumn
+ Indent
,
1414 FirstLine
? FirstStartColumn
: 0, DryRun
);
1416 RangeMinLevel
= std::min(RangeMinLevel
, TheLine
.Level
);
1418 // If no token in the current line is affected, we still need to format
1419 // affected children.
1420 if (TheLine
.ChildrenAffected
) {
1421 for (const FormatToken
*Tok
= TheLine
.First
; Tok
; Tok
= Tok
->Next
)
1422 if (!Tok
->Children
.empty())
1423 format(Tok
->Children
, DryRun
);
1426 // Adapt following lines on the current indent level to the same level
1427 // unless the current \c AnnotatedLine is not at the beginning of a line.
1428 bool StartsNewLine
=
1429 TheLine
.First
->NewlinesBefore
> 0 || TheLine
.First
->IsFirst
;
1431 IndentTracker
.adjustToUnmodifiedLine(TheLine
);
1433 bool ReformatLeadingWhitespace
=
1434 StartsNewLine
&& ((PreviousLine
&& PreviousLine
->Affected
) ||
1435 TheLine
.LeadingEmptyLinesAffected
);
1436 // Format the first token.
1437 if (ReformatLeadingWhitespace
) {
1438 formatFirstToken(TheLine
, PreviousLine
, PrevPrevLine
, Lines
,
1439 TheLine
.First
->OriginalColumn
,
1440 TheLine
.First
->OriginalColumn
);
1442 Whitespaces
->addUntouchableToken(*TheLine
.First
,
1443 TheLine
.InPPDirective
);
1446 // Notify the WhitespaceManager about the unchanged whitespace.
1447 for (FormatToken
*Tok
= TheLine
.First
->Next
; Tok
; Tok
= Tok
->Next
)
1448 Whitespaces
->addUntouchableToken(*Tok
, TheLine
.InPPDirective
);
1450 NextLine
= Joiner
.getNextMergedLine(DryRun
, IndentTracker
);
1451 RangeMinLevel
= UINT_MAX
;
1454 markFinalized(TheLine
.First
);
1456 PenaltyCache
[CacheKey
] = Penalty
;
1460 static auto computeNewlines(const AnnotatedLine
&Line
,
1461 const AnnotatedLine
*PreviousLine
,
1462 const AnnotatedLine
*PrevPrevLine
,
1463 const SmallVectorImpl
<AnnotatedLine
*> &Lines
,
1464 const FormatStyle
&Style
) {
1465 const auto &RootToken
= *Line
.First
;
1467 std::min(RootToken
.NewlinesBefore
, Style
.MaxEmptyLinesToKeep
+ 1);
1468 // Remove empty lines before "}" where applicable.
1469 if (RootToken
.is(tok::r_brace
) &&
1471 (RootToken
.Next
->is(tok::semi
) && !RootToken
.Next
->Next
)) &&
1472 // Do not remove empty lines before namespace closing "}".
1473 !getNamespaceToken(&Line
, Lines
)) {
1474 Newlines
= std::min(Newlines
, 1u);
1476 // Remove empty lines at the start of nested blocks (lambdas/arrow functions)
1477 if (!PreviousLine
&& Line
.Level
> 0)
1478 Newlines
= std::min(Newlines
, 1u);
1479 if (Newlines
== 0 && !RootToken
.IsFirst
)
1481 if (RootToken
.IsFirst
&&
1482 (!Style
.KeepEmptyLines
.AtStartOfFile
|| !RootToken
.HasUnescapedNewline
)) {
1486 // Remove empty lines after "{".
1487 if (!Style
.KeepEmptyLines
.AtStartOfBlock
&& PreviousLine
&&
1488 PreviousLine
->Last
->is(tok::l_brace
) &&
1489 !PreviousLine
->startsWithNamespace() &&
1490 !(PrevPrevLine
&& PrevPrevLine
->startsWithNamespace() &&
1491 PreviousLine
->startsWith(tok::l_brace
)) &&
1492 !startsExternCBlock(*PreviousLine
)) {
1496 // Insert or remove empty line before access specifiers.
1497 if (PreviousLine
&& RootToken
.isAccessSpecifier()) {
1498 switch (Style
.EmptyLineBeforeAccessModifier
) {
1499 case FormatStyle::ELBAMS_Never
:
1503 case FormatStyle::ELBAMS_Leave
:
1504 Newlines
= std::max(RootToken
.NewlinesBefore
, 1u);
1506 case FormatStyle::ELBAMS_LogicalBlock
:
1507 if (PreviousLine
->Last
->isOneOf(tok::semi
, tok::r_brace
) && Newlines
<= 1)
1509 if (PreviousLine
->First
->isAccessSpecifier())
1510 Newlines
= 1; // Previous is an access modifier remove all new lines.
1512 case FormatStyle::ELBAMS_Always
: {
1513 const FormatToken
*previousToken
;
1514 if (PreviousLine
->Last
->is(tok::comment
))
1515 previousToken
= PreviousLine
->Last
->getPreviousNonComment();
1517 previousToken
= PreviousLine
->Last
;
1518 if ((!previousToken
|| previousToken
->isNot(tok::l_brace
)) &&
1526 // Insert or remove empty line after access specifiers.
1527 if (PreviousLine
&& PreviousLine
->First
->isAccessSpecifier() &&
1528 (!PreviousLine
->InPPDirective
|| !RootToken
.HasUnescapedNewline
)) {
1529 // EmptyLineBeforeAccessModifier is handling the case when two access
1530 // modifiers follow each other.
1531 if (!RootToken
.isAccessSpecifier()) {
1532 switch (Style
.EmptyLineAfterAccessModifier
) {
1533 case FormatStyle::ELAAMS_Never
:
1536 case FormatStyle::ELAAMS_Leave
:
1537 Newlines
= std::max(Newlines
, 1u);
1539 case FormatStyle::ELAAMS_Always
:
1540 if (RootToken
.is(tok::r_brace
)) // Do not add at end of class.
1543 Newlines
= std::max(Newlines
, 2u);
1552 void UnwrappedLineFormatter::formatFirstToken(
1553 const AnnotatedLine
&Line
, const AnnotatedLine
*PreviousLine
,
1554 const AnnotatedLine
*PrevPrevLine
,
1555 const SmallVectorImpl
<AnnotatedLine
*> &Lines
, unsigned Indent
,
1556 unsigned NewlineIndent
) {
1557 FormatToken
&RootToken
= *Line
.First
;
1558 if (RootToken
.is(tok::eof
)) {
1559 unsigned Newlines
= std::min(
1560 RootToken
.NewlinesBefore
,
1561 Style
.KeepEmptyLines
.AtEndOfFile
? Style
.MaxEmptyLinesToKeep
+ 1 : 1);
1562 unsigned TokenIndent
= Newlines
? NewlineIndent
: 0;
1563 Whitespaces
->replaceWhitespace(RootToken
, Newlines
, TokenIndent
,
1568 if (RootToken
.Newlines
< 0) {
1569 RootToken
.Newlines
=
1570 computeNewlines(Line
, PreviousLine
, PrevPrevLine
, Lines
, Style
);
1571 assert(RootToken
.Newlines
>= 0);
1574 if (RootToken
.Newlines
> 0)
1575 Indent
= NewlineIndent
;
1577 // Preprocessor directives get indented before the hash only if specified. In
1578 // Javascript import statements are indented like normal statements.
1579 if (!Style
.isJavaScript() &&
1580 Style
.IndentPPDirectives
!= FormatStyle::PPDIS_BeforeHash
&&
1581 (Line
.Type
== LT_PreprocessorDirective
||
1582 Line
.Type
== LT_ImportStatement
)) {
1586 Whitespaces
->replaceWhitespace(RootToken
, RootToken
.Newlines
, Indent
, Indent
,
1587 /*IsAligned=*/false,
1588 Line
.InPPDirective
&&
1589 !RootToken
.HasUnescapedNewline
);
1593 UnwrappedLineFormatter::getColumnLimit(bool InPPDirective
,
1594 const AnnotatedLine
*NextLine
) const {
1595 // In preprocessor directives reserve two chars for trailing " \" if the
1596 // next line continues the preprocessor directive.
1597 bool ContinuesPPDirective
=
1599 // If there is no next line, this is likely a child line and the parent
1600 // continues the preprocessor directive.
1602 (NextLine
->InPPDirective
&&
1603 // If there is an unescaped newline between this line and the next, the
1604 // next line starts a new preprocessor directive.
1605 !NextLine
->First
->HasUnescapedNewline
));
1606 return Style
.ColumnLimit
- (ContinuesPPDirective
? 2 : 0);
1609 } // namespace format
1610 } // namespace clang