[AMDGPU] Infer amdgpu-no-flat-scratch-init attribute in AMDGPUAttributor (#94647)
[llvm-project.git] / clang / lib / Format / UnwrappedLineFormatter.cpp
blob1804c1437fd41d4c19fa4fec7c51f172280e8e78
1 //===--- UnwrappedLineFormatter.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 //===----------------------------------------------------------------------===//
9 #include "UnwrappedLineFormatter.h"
10 #include "FormatToken.h"
11 #include "NamespaceEndCommentsFixer.h"
12 #include "WhitespaceManager.h"
13 #include "llvm/Support/Debug.h"
14 #include <queue>
16 #define DEBUG_TYPE "format-formatter"
18 namespace clang {
19 namespace format {
21 namespace {
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.
36 ///
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
39 /// with.
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
43 /// given line.
44 class LevelIndentTracker {
45 public:
46 LevelIndentTracker(const FormatStyle &Style,
47 const AdditionalKeywords &Keywords, unsigned StartLevel,
48 int AdditionalIndent)
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
58 /// next.
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;
76 } else {
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)
87 Indent += Offset;
88 if (Line.IsContinuation)
89 Indent = Line.Level * Style.IndentWidth + Style.ContinuationIndentWidth;
92 /// Update the level indent to adapt to the given \p Line.
93 ///
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)
99 return;
100 assert(Line.Level < IndentForLevel.size());
101 if (Line.First->is(tok::comment) && IndentForLevel[Line.Level] != -1)
102 return;
103 unsigned LevelIndent = Line.First->OriginalColumn;
104 if (static_cast<int>(LevelIndent) - Offset >= 0)
105 LevelIndent -= Offset;
106 IndentForLevel[Line.Level] = LevelIndent;
109 private:
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() ||
116 Style.isCSharp()) {
117 return 0;
120 auto IsAccessModifier = [&](const FormatToken &RootToken) {
121 if (Line.Type == LT_AccessModifier || RootToken.isObjCAccessSpecifier())
122 return true;
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)) {
129 return true;
132 if (Next && Next->isOneOf(Keywords.kw_slots, Keywords.kw_qslots) &&
133 Next->Next && Next->Next->is(tok::colon)) {
134 return true;
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;
149 return 0;
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];
161 if (Level == 0)
162 return 0;
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
179 /// offset.
180 int Offset = 0;
182 /// The current line's indent.
183 unsigned Indent = 0;
186 const FormatToken *getMatchingNamespaceToken(
187 const AnnotatedLine *Line,
188 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
189 if (!Line->startsWith(tok::r_brace))
190 return nullptr;
191 size_t StartLineIndex = Line->MatchingOpeningBlockLineIndex;
192 if (StartLineIndex == UnwrappedLine::kInvalidIndex)
193 return nullptr;
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();
211 class LineJoiner {
212 public:
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) {
221 if (Next == End)
222 return nullptr;
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
228 // input lines.
229 for (unsigned i = 0; i < MergedLines; ++i)
230 if (Next[i + 1]->First->NewlinesBefore > 0)
231 MergedLines = 0;
233 if (!DryRun)
234 for (unsigned i = 0; i < MergedLines; ++i)
235 join(*Next[0], *Next[i + 1]);
236 Next = Next + MergedLines + 1;
237 return Current;
240 private:
241 /// Calculates how many lines can be merged into 1 starting at \p I.
242 unsigned
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.
249 if (I + 1 == E)
250 return 0;
251 // We can never merge stuff if there are trailing line comments.
252 const AnnotatedLine *TheLine = *I;
253 if (TheLine->Last->is(TT_LineComment))
254 return 0;
255 const auto &NextLine = *I[1];
256 if (NextLine.Type == LT_Invalid || NextLine.First->MustBreakBefore)
257 return 0;
258 if (TheLine->InPPDirective &&
259 (!NextLine.InPPDirective || NextLine.First->HasUnescapedNewline)) {
260 return 0;
263 if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit)
264 return 0;
266 unsigned Limit =
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)
294 : 0;
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)
303 : 0;
306 if (Tok && Tok->is(tok::kw_template) &&
307 Style.BraceWrapping.SplitEmptyRecord && EmptyBlock) {
308 return 0;
312 auto ShouldMergeShortFunctions = [this, &I, &NextLine, PreviousLine,
313 TheLine]() {
314 if (Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_All)
315 return true;
316 if (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
317 NextLine.First->is(tok::r_brace)) {
318 return true;
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))
326 return true;
328 if (TheLine->Level != 0) {
329 if (!PreviousLine)
330 return false;
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) {
336 assert(*J);
337 if (!(*J)->InPPDirective && !(*J)->isComment() &&
338 (*J)->Level < TheLine->Level) {
339 Line = *J;
340 break;
344 if (!Line)
345 return false;
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);
356 return false;
359 bool MergeShortFunctions = ShouldMergeShortFunctions();
361 const auto *FirstNonComment = TheLine->getFirstNonComment();
362 if (!FirstNonComment)
363 return 0;
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()) {
369 int J = 1;
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;
384 ++CompactedLine) {
385 if (!(*CompactedLine)->InPPDirective) {
386 const int Level = (*CompactedLine)->Level;
387 (*CompactedLine)->Level = std::max(Level - OutdentBy, 0);
391 return J - 1;
394 if (auto nsToken = getMatchingNamespaceToken(TheLine, AnnotatedLines)) {
395 int i = 0;
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]);
409 return i;
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,
426 TT_ForEachMacro)) {
427 return Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never
428 ? tryMergeSimpleBlock(I, E, Limit)
429 : 0;
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 <=
445 Style.ColumnLimit)
447 : 0;
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)
454 : 0;
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)
467 : 0;
470 if (PreviousLine && TheLine->First->is(tok::l_brace)) {
471 switch (PreviousLine->First->Tok.getKind()) {
472 case tok::at:
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) {
479 return 0;
482 break;
484 case tok::kw_case:
485 case tok::kw_default:
486 // Don't merge block with left brace wrapped after case labels.
487 return 0;
489 default:
490 break;
494 // Don't merge an empty template class or struct if SplitEmptyRecords
495 // is defined.
496 if (PreviousLine && Style.BraceWrapping.SplitEmptyRecord &&
497 TheLine->Last->is(tok::l_brace) && PreviousLine->Last) {
498 const FormatToken *Previous = PreviousLine->Last;
499 if (Previous) {
500 if (Previous->is(tok::comment))
501 Previous = Previous->getPreviousNonComment();
502 if (Previous) {
503 if (Previous->is(tok::greater) && !PreviousLine->InPPDirective)
504 return 0;
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)) {
510 return 0;
517 if (TheLine->First->is(TT_SwitchExpressionLabel)) {
518 return Style.AllowShortCaseExpressionOnASingleLine
519 ? tryMergeShortCaseLabels(I, E, Limit)
520 : 0;
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
533 // elsewhere.
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,
539 tok::kw_struct)) {
540 // Try to merge a block with left brace unwrapped that wasn't yet
541 // covered.
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))
553 return 0;
555 // Check for Limit <= 2 to account for the " {".
556 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(TheLine)))
557 return 0;
558 Limit -= 2;
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.
568 if (MergedLines > 0)
569 ++MergedLines;
571 return MergedLines;
573 auto IsElseLine = [&TheLine]() -> bool {
574 const FormatToken *First = TheLine->First;
575 if (First->is(tok::kw_else))
576 return true;
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)
586 : 0;
588 if (TheLine->First->isOneOf(tok::kw_for, tok::kw_while, tok::kw_do,
589 TT_ForEachMacro)) {
590 return Style.AllowShortLoopsOnASingleLine
591 ? tryMergeSimpleControlStatement(I, E, Limit)
592 : 0;
594 if (TheLine->First->isOneOf(tok::kw_case, tok::kw_default)) {
595 return Style.AllowShortCaseLabelsOnASingleLine
596 ? tryMergeShortCaseLabels(I, E, Limit)
597 : 0;
599 if (TheLine->InPPDirective &&
600 (TheLine->First->HasUnescapedNewline || TheLine->First->IsFirst)) {
601 return tryMergeSimplePPDirective(I, E, Limit);
603 return 0;
606 unsigned
607 tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
608 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
609 unsigned Limit) {
610 if (Limit == 0)
611 return 0;
612 if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline)
613 return 0;
614 if (1 + I[1]->Last->TotalLength > Limit)
615 return 0;
616 return 1;
619 unsigned tryMergeSimpleControlStatement(
620 SmallVectorImpl<AnnotatedLine *>::const_iterator I,
621 SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) {
622 if (Limit == 0)
623 return 0;
624 if (Style.BraceWrapping.AfterControlStatement ==
625 FormatStyle::BWACS_Always &&
626 I[1]->First->is(tok::l_brace) &&
627 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) {
628 return 0;
630 if (I[1]->InPPDirective != (*I)->InPPDirective ||
631 (I[1]->InPPDirective && I[1]->First->HasUnescapedNewline)) {
632 return 0;
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)) {
638 return 0;
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))
642 return 0;
643 if (1 + I[1]->Last->TotalLength > Limit)
644 return 0;
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)) {
648 return 0;
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)) {
655 return 0;
658 return 1;
661 unsigned
662 tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
663 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
664 unsigned Limit) {
665 if (Limit == 0 || I + 1 == E ||
666 I[1]->First->isOneOf(tok::kw_case, tok::kw_default)) {
667 return 0;
669 if (I[0]->Last->is(tok::l_brace) || I[1]->First->is(tok::l_brace))
670 return 0;
671 unsigned NumStmts = 0;
672 unsigned Length = 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)
679 break;
680 const AnnotatedLine *Line = I[1 + NumStmts];
681 if (Line->InPPDirective != InPPDirective)
682 break;
683 if (Line->InMacroBody != InMacroBody)
684 break;
685 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
686 break;
687 if (Line->First->isOneOf(tok::kw_if, tok::kw_for, tok::kw_switch,
688 tok::kw_while) ||
689 EndsWithComment) {
690 return 0;
692 if (Line->First->is(tok::comment)) {
693 if (Level != Line->Level)
694 return 0;
695 SmallVectorImpl<AnnotatedLine *>::const_iterator J = I + 2 + NumStmts;
696 for (; J != E; ++J) {
697 Line = *J;
698 if (Line->InPPDirective != InPPDirective)
699 break;
700 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
701 break;
702 if (Line->First->isNot(tok::comment) || Level != Line->Level)
703 return 0;
705 break;
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)
712 return 0;
713 return NumStmts;
716 unsigned
717 tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
718 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
719 unsigned Limit) {
720 // Don't merge with a preprocessor directive.
721 if (I[1]->Type == LT_PreprocessorDirective)
722 return 0;
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)) {
731 return 0;
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))) {
738 return 0;
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))
744 return 0;
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)) {
761 if (IsSplitBlock)
762 return 0;
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)) {
769 return 0;
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)) {
776 return 0;
778 if (!Style.AllowShortLoopsOnASingleLine &&
779 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for,
780 TT_ForEachMacro) &&
781 !Style.BraceWrapping.AfterControlStatement &&
782 I[1]->First->isNot(tok::r_brace)) {
783 return 0;
785 if (!Style.AllowShortLoopsOnASingleLine &&
786 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for,
787 TT_ForEachMacro) &&
788 Style.BraceWrapping.AfterControlStatement ==
789 FormatStyle::BWACS_Always &&
790 I + 2 != E && I[2]->First->isNot(tok::r_brace)) {
791 return 0;
793 // FIXME: Consider an option to allow short exception handling clauses on
794 // a single line.
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)) {
800 return 0;
804 if (Line.endsWith(tok::l_brace)) {
805 if (Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never &&
806 Line.First->is(TT_BlockLBrace)) {
807 return 0;
810 if (IsSplitBlock && Line.First == Line.Last &&
811 I > AnnotatedLines.begin() &&
812 (I[-1]->endsWith(tok::kw_else) || IsCtrlStmt(*I[-1]))) {
813 return 0;
815 FormatToken *Tok = I[1]->First;
816 auto ShouldMerge = [Tok]() {
817 if (Tok->isNot(tok::r_brace) || Tok->MustBreakBefore)
818 return false;
819 const FormatToken *Next = Tok->getNextNonComment();
820 return !Next || Next->is(tok::semi);
823 if (ShouldMerge()) {
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;
828 return 1;
829 } else if (Limit != 0 && !Line.startsWithNamespace() &&
830 !startsExternCBlock(Line)) {
831 // We don't merge short records.
832 if (isRecordLBrace(*Line.Last))
833 return 0;
835 // Check that we still have three lines and they fit into the limit.
836 if (I + 2 == E || I[2]->Type == LT_Invalid)
837 return 0;
838 Limit = limitConsideringMacros(I + 2, E, Limit);
840 if (!nextTwoLinesFitInto(I, Limit))
841 return 0;
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))
846 return 0;
847 do {
848 if (Tok->is(tok::l_brace) && Tok->isNot(BK_BracedInit))
849 return 0;
850 Tok = Tok->Next;
851 } while (Tok);
853 // Last, check that the third line starts with a closing brace.
854 Tok = I[2]->First;
855 if (Tok->isNot(tok::r_brace))
856 return 0;
858 // Don't merge "if (a) { .. } else {".
859 if (Tok->Next && Tok->Next->is(tok::kw_else))
860 return 0;
862 // Don't merge a trailing multi-line control statement block like:
863 // } else if (foo &&
864 // bar)
865 // { <-- current Line
866 // baz();
867 // }
868 if (Line.First == Line.Last && Line.First->isNot(TT_FunctionLBrace) &&
869 Style.BraceWrapping.AfterControlStatement ==
870 FormatStyle::BWACS_MultiLine) {
871 return 0;
874 return 2;
876 } else if (I[1]->First->is(tok::l_brace)) {
877 if (I[1]->Last->is(TT_LineComment))
878 return 0;
880 // Check for Limit <= 2 to account for the " {".
881 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(*I)))
882 return 0;
883 Limit -= 2;
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.
891 if (MergedLines > 0)
892 ++MergedLines;
894 return MergedLines;
896 return 0;
899 /// Returns the modified column limit for \p I if it is inside a macro and
900 /// needs a trailing '\'.
901 unsigned
902 limitConsideringMacros(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
903 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
904 unsigned Limit) {
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;
909 return Limit;
912 bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
913 unsigned Limit) {
914 if (I[1]->First->MustBreakBefore || I[2]->First->MustBreakBefore)
915 return false;
916 return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit;
919 bool containsMustBreak(const AnnotatedLine *Line) {
920 assert(Line->First);
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)
925 return true;
926 return false;
929 void join(AnnotatedLine &A, const AnnotatedLine &B) {
930 assert(!A.Last->Next);
931 assert(!B.First->Previous);
932 if (B.Affected)
933 A.Affected = true;
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;
940 A.Last = Tok;
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)) {
957 Tok = Tok->Next;
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;
975 } else {
976 Tok->Finalized = true;
981 #ifndef NDEBUG
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";
990 #endif
992 /// Base class for classes that format one \c AnnotatedLine.
993 class LineFormatter {
994 public:
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;
1008 protected:
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.
1037 return true;
1040 if (NewLine || Previous.MacroParent) {
1041 const ParenState &P = State.Stack.back();
1043 int AdditionalIndent =
1044 P.Indent - Previous.Children[0]->Level * Style.IndentWidth;
1045 Penalty +=
1046 BlockFormatter->format(Previous.Children, DryRun, AdditionalIndent,
1047 /*FixBadIndentation=*/true);
1048 return true;
1051 if (Previous.Children[0]->First->MustBreakBefore)
1052 return false;
1054 // Cannot merge into one line if this line ends on a comment.
1055 if (Previous.is(tok::comment))
1056 return false;
1058 // Cannot merge multiple statements into a single line.
1059 if (Previous.Children.size() > 1)
1060 return false;
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())
1065 return false;
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) {
1071 return false;
1074 if (!DryRun) {
1075 Whitespaces->replaceWhitespace(
1076 *Child->First, /*Newlines=*/0, /*Spaces=*/1,
1077 /*StartOfTokenColumn=*/State.Column, /*IsAligned=*/false,
1078 State.Line->InPPDirective);
1080 Penalty +=
1081 formatLine(*Child, State.Column + 1, /*FirstStartColumn=*/0, DryRun);
1082 if (!DryRun)
1083 markFinalized(Child->First);
1085 State.Column += 1 + Child->Last->TotalLength;
1086 return true;
1089 ContinuationIndenter *Indenter;
1091 private:
1092 WhitespaceManager *Whitespaces;
1093 const FormatStyle &Style;
1094 UnwrappedLineFormatter *BlockFormatter;
1097 /// Formatter that keeps the existing line breaks.
1098 class NoColumnLimitLineFormatter : public LineFormatter {
1099 public:
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
1107 /// decisions.
1108 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
1109 unsigned FirstStartColumn, bool DryRun) override {
1110 assert(!DryRun);
1111 LineState State = Indenter->getInitialState(FirstIndent, FirstStartColumn,
1112 &Line, /*DryRun=*/false);
1113 while (State.NextToken) {
1114 bool Newline =
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);
1121 return 0;
1125 /// Formatter that puts all tokens into a single line without breaks.
1126 class NoLineBreakFormatter : public LineFormatter {
1127 public:
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;
1137 LineState State =
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);
1144 return Penalty;
1148 /// Finds the best way to break lines.
1149 class OptimizingLineFormatter : public LineFormatter {
1150 public:
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 {
1161 LineState State =
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);
1173 private:
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.
1189 struct StateNode {
1190 StateNode(const LineState &State, bool NewLine, StateNode *Previous)
1191 : State(State), NewLine(NewLine), Previous(Previous) {}
1192 LineState State;
1193 bool NewLine;
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>>
1204 QueueType;
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.
1219 unsigned Count = 0;
1220 QueueType Queue;
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));
1226 ++Count;
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)
1234 return 0;
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");
1241 break;
1243 Queue.pop();
1245 // Cut off the analysis of certain solutions if the analysis gets too
1246 // complex. See description of IgnoreStackForComparison.
1247 if (Count > 50'000)
1248 Node->State.IgnoreStackForComparison = true;
1250 if (!Seen.insert(&Node->State).second) {
1251 // State already examined with lower penalty.
1252 continue;
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");
1266 return 0;
1269 // Reconstruct the solution.
1270 if (!DryRun)
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");
1277 return Penalty;
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))
1287 return;
1288 if (!NewLine && Indenter->mustBreak(PreviousNode->State))
1289 return;
1291 StateNode *Node = new (Allocator.Allocate())
1292 StateNode(PreviousNode->State, NewLine, PreviousNode);
1293 if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty))
1294 return;
1296 Penalty += Indenter->addTokenToState(Node->State, NewLine, true);
1298 Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node));
1299 ++(*Count);
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);
1316 LLVM_DEBUG({
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,
1348 AdditionalIndent);
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;
1383 Status->Line =
1384 SourceMgr.getSpellingLineNumber(TheLine.First->Tok.getLocation());
1387 if (ShouldFormat && TheLine.Type != LT_Invalid) {
1388 if (!DryRun) {
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);
1411 } else {
1412 Penalty += OptimizingLineFormatter(Indenter, Whitespaces, Style, this)
1413 .formatLine(TheLine, NextStartColumn + Indent,
1414 FirstLine ? FirstStartColumn : 0, DryRun);
1416 RangeMinLevel = std::min(RangeMinLevel, TheLine.Level);
1417 } else {
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;
1430 if (StartsNewLine)
1431 IndentTracker.adjustToUnmodifiedLine(TheLine);
1432 if (!DryRun) {
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);
1441 } else {
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;
1453 if (!DryRun)
1454 markFinalized(TheLine.First);
1456 PenaltyCache[CacheKey] = Penalty;
1457 return 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;
1466 auto Newlines =
1467 std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1);
1468 // Remove empty lines before "}" where applicable.
1469 if (RootToken.is(tok::r_brace) &&
1470 (!RootToken.Next ||
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)
1480 Newlines = 1;
1481 if (RootToken.IsFirst &&
1482 (!Style.KeepEmptyLines.AtStartOfFile || !RootToken.HasUnescapedNewline)) {
1483 Newlines = 0;
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)) {
1493 Newlines = 1;
1496 // Insert or remove empty line before access specifiers.
1497 if (PreviousLine && RootToken.isAccessSpecifier()) {
1498 switch (Style.EmptyLineBeforeAccessModifier) {
1499 case FormatStyle::ELBAMS_Never:
1500 if (Newlines > 1)
1501 Newlines = 1;
1502 break;
1503 case FormatStyle::ELBAMS_Leave:
1504 Newlines = std::max(RootToken.NewlinesBefore, 1u);
1505 break;
1506 case FormatStyle::ELBAMS_LogicalBlock:
1507 if (PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) && Newlines <= 1)
1508 Newlines = 2;
1509 if (PreviousLine->First->isAccessSpecifier())
1510 Newlines = 1; // Previous is an access modifier remove all new lines.
1511 break;
1512 case FormatStyle::ELBAMS_Always: {
1513 const FormatToken *previousToken;
1514 if (PreviousLine->Last->is(tok::comment))
1515 previousToken = PreviousLine->Last->getPreviousNonComment();
1516 else
1517 previousToken = PreviousLine->Last;
1518 if ((!previousToken || previousToken->isNot(tok::l_brace)) &&
1519 Newlines <= 1) {
1520 Newlines = 2;
1522 } break;
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:
1534 Newlines = 1;
1535 break;
1536 case FormatStyle::ELAAMS_Leave:
1537 Newlines = std::max(Newlines, 1u);
1538 break;
1539 case FormatStyle::ELAAMS_Always:
1540 if (RootToken.is(tok::r_brace)) // Do not add at end of class.
1541 Newlines = 1u;
1542 else
1543 Newlines = std::max(Newlines, 2u);
1544 break;
1549 return Newlines;
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,
1564 TokenIndent);
1565 return;
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)) {
1583 Indent = 0;
1586 Whitespaces->replaceWhitespace(RootToken, RootToken.Newlines, Indent, Indent,
1587 /*IsAligned=*/false,
1588 Line.InPPDirective &&
1589 !RootToken.HasUnescapedNewline);
1592 unsigned
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 =
1598 InPPDirective &&
1599 // If there is no next line, this is likely a child line and the parent
1600 // continues the preprocessor directive.
1601 (!NextLine ||
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