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 //===----------------------------------------------------------------------===//
8
9 #include "UnwrappedLineFormatter.h"
10 #include "NamespaceEndCommentsFixer.h"
11 #include "WhitespaceManager.h"
12 #include "llvm/Support/Debug.h"
13 #include <queue>
14
15 #define DEBUG_TYPE "format-formatter"
16
17 namespace clang {
18 namespace format {
19
20 namespace {
21
startsExternCBlock(const AnnotatedLine & Line)22 bool startsExternCBlock(const AnnotatedLine &Line) {
23 const FormatToken *Next = Line.First->getNextNonComment();
24 const FormatToken *NextNext = Next ? Next->getNextNonComment() : nullptr;
25 return Line.startsWith(tok::kw_extern) && Next && Next->isStringLiteral() &&
26 NextNext && NextNext->is(tok::l_brace);
27 }
28
isRecordLBrace(const FormatToken & Tok)29 bool isRecordLBrace(const FormatToken &Tok) {
30 return Tok.isOneOf(TT_ClassLBrace, TT_EnumLBrace, TT_RecordLBrace,
31 TT_StructLBrace, TT_UnionLBrace);
32 }
33
34 /// Tracks the indent level of \c AnnotatedLines across levels.
35 ///
36 /// \c nextLine must be called for each \c AnnotatedLine, after which \c
37 /// getIndent() will return the indent for the last line \c nextLine was called
38 /// with.
39 /// If the line is not formatted (and thus the indent does not change), calling
40 /// \c adjustToUnmodifiedLine after the call to \c nextLine will cause
41 /// subsequent lines on the same level to be indented at the same level as the
42 /// given line.
43 class LevelIndentTracker {
44 public:
LevelIndentTracker(const FormatStyle & Style,const AdditionalKeywords & Keywords,unsigned StartLevel,int AdditionalIndent)45 LevelIndentTracker(const FormatStyle &Style,
46 const AdditionalKeywords &Keywords, unsigned StartLevel,
47 int AdditionalIndent)
48 : Style(Style), Keywords(Keywords), AdditionalIndent(AdditionalIndent) {
49 for (unsigned i = 0; i != StartLevel; ++i)
50 IndentForLevel.push_back(Style.IndentWidth * i + AdditionalIndent);
51 }
52
53 /// Returns the indent for the current line.
getIndent() const54 unsigned getIndent() const { return Indent; }
55
56 /// Update the indent state given that \p Line is going to be formatted
57 /// next.
nextLine(const AnnotatedLine & Line)58 void nextLine(const AnnotatedLine &Line) {
59 Offset = getIndentOffset(*Line.First);
60 // Update the indent level cache size so that we can rely on it
61 // having the right size in adjustToUnmodifiedline.
62 skipLine(Line, /*UnknownIndent=*/true);
63 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None &&
64 (Line.InPPDirective ||
65 (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash &&
66 Line.Type == LT_CommentAbovePPDirective))) {
67 unsigned PPIndentWidth =
68 (Style.PPIndentWidth >= 0) ? Style.PPIndentWidth : Style.IndentWidth;
69 Indent = Line.InMacroBody
70 ? Line.PPLevel * PPIndentWidth +
71 (Line.Level - Line.PPLevel) * Style.IndentWidth
72 : Line.Level * PPIndentWidth;
73 Indent += AdditionalIndent;
74 } else {
75 Indent = getIndent(Line.Level);
76 }
77 if (static_cast<int>(Indent) + Offset >= 0)
78 Indent += Offset;
79 if (Line.IsContinuation)
80 Indent = Line.Level * Style.IndentWidth + Style.ContinuationIndentWidth;
81 }
82
83 /// Update the indent state given that \p Line indent should be
84 /// skipped.
skipLine(const AnnotatedLine & Line,bool UnknownIndent=false)85 void skipLine(const AnnotatedLine &Line, bool UnknownIndent = false) {
86 if (Line.Level >= IndentForLevel.size())
87 IndentForLevel.resize(Line.Level + 1, UnknownIndent ? -1 : Indent);
88 }
89
90 /// Update the level indent to adapt to the given \p Line.
91 ///
92 /// When a line is not formatted, we move the subsequent lines on the same
93 /// level to the same indent.
94 /// Note that \c nextLine must have been called before this method.
adjustToUnmodifiedLine(const AnnotatedLine & Line)95 void adjustToUnmodifiedLine(const AnnotatedLine &Line) {
96 unsigned LevelIndent = Line.First->OriginalColumn;
97 if (static_cast<int>(LevelIndent) - Offset >= 0)
98 LevelIndent -= Offset;
99 assert(Line.Level < IndentForLevel.size());
100 if ((!Line.First->is(tok::comment) || IndentForLevel[Line.Level] == -1) &&
101 !Line.InPPDirective) {
102 IndentForLevel[Line.Level] = LevelIndent;
103 }
104 }
105
106 private:
107 /// Get the offset of the line relatively to the level.
108 ///
109 /// For example, 'public:' labels in classes are offset by 1 or 2
110 /// characters to the left from their level.
getIndentOffset(const FormatToken & RootToken)111 int getIndentOffset(const FormatToken &RootToken) {
112 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() ||
113 Style.isCSharp()) {
114 return 0;
115 }
116
117 auto IsAccessModifier = [this, &RootToken]() {
118 if (RootToken.isAccessSpecifier(Style.isCpp())) {
119 return true;
120 } else if (RootToken.isObjCAccessSpecifier()) {
121 return true;
122 }
123 // Handle Qt signals.
124 else if ((RootToken.isOneOf(Keywords.kw_signals, Keywords.kw_qsignals) &&
125 RootToken.Next && RootToken.Next->is(tok::colon))) {
126 return true;
127 } else if (RootToken.Next &&
128 RootToken.Next->isOneOf(Keywords.kw_slots,
129 Keywords.kw_qslots) &&
130 RootToken.Next->Next && RootToken.Next->Next->is(tok::colon)) {
131 return true;
132 }
133 // Handle malformed access specifier e.g. 'private' without trailing ':'.
134 else if (!RootToken.Next && RootToken.isAccessSpecifier(false)) {
135 return true;
136 }
137 return false;
138 };
139
140 if (IsAccessModifier()) {
141 // The AccessModifierOffset may be overridden by IndentAccessModifiers,
142 // in which case we take a negative value of the IndentWidth to simulate
143 // the upper indent level.
144 return Style.IndentAccessModifiers ? -Style.IndentWidth
145 : Style.AccessModifierOffset;
146 }
147 return 0;
148 }
149
150 /// Get the indent of \p Level from \p IndentForLevel.
151 ///
152 /// \p IndentForLevel must contain the indent for the level \c l
153 /// at \p IndentForLevel[l], or a value < 0 if the indent for
154 /// that level is unknown.
getIndent(unsigned Level) const155 unsigned getIndent(unsigned Level) const {
156 if (IndentForLevel[Level] != -1)
157 return IndentForLevel[Level];
158 if (Level == 0)
159 return 0;
160 return getIndent(Level - 1) + Style.IndentWidth;
161 }
162
163 const FormatStyle &Style;
164 const AdditionalKeywords &Keywords;
165 const unsigned AdditionalIndent;
166
167 /// The indent in characters for each level.
168 SmallVector<int> IndentForLevel;
169
170 /// Offset of the current line relative to the indent level.
171 ///
172 /// For example, the 'public' keywords is often indented with a negative
173 /// offset.
174 int Offset = 0;
175
176 /// The current line's indent.
177 unsigned Indent = 0;
178 };
179
getMatchingNamespaceToken(const AnnotatedLine * Line,const SmallVectorImpl<AnnotatedLine * > & AnnotatedLines)180 const FormatToken *getMatchingNamespaceToken(
181 const AnnotatedLine *Line,
182 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
183 if (!Line->startsWith(tok::r_brace))
184 return nullptr;
185 size_t StartLineIndex = Line->MatchingOpeningBlockLineIndex;
186 if (StartLineIndex == UnwrappedLine::kInvalidIndex)
187 return nullptr;
188 assert(StartLineIndex < AnnotatedLines.size());
189 return AnnotatedLines[StartLineIndex]->First->getNamespaceToken();
190 }
191
getNamespaceTokenText(const AnnotatedLine * Line)192 StringRef getNamespaceTokenText(const AnnotatedLine *Line) {
193 const FormatToken *NamespaceToken = Line->First->getNamespaceToken();
194 return NamespaceToken ? NamespaceToken->TokenText : StringRef();
195 }
196
getMatchingNamespaceTokenText(const AnnotatedLine * Line,const SmallVectorImpl<AnnotatedLine * > & AnnotatedLines)197 StringRef getMatchingNamespaceTokenText(
198 const AnnotatedLine *Line,
199 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
200 const FormatToken *NamespaceToken =
201 getMatchingNamespaceToken(Line, AnnotatedLines);
202 return NamespaceToken ? NamespaceToken->TokenText : StringRef();
203 }
204
205 class LineJoiner {
206 public:
LineJoiner(const FormatStyle & Style,const AdditionalKeywords & Keywords,const SmallVectorImpl<AnnotatedLine * > & Lines)207 LineJoiner(const FormatStyle &Style, const AdditionalKeywords &Keywords,
208 const SmallVectorImpl<AnnotatedLine *> &Lines)
209 : Style(Style), Keywords(Keywords), End(Lines.end()), Next(Lines.begin()),
210 AnnotatedLines(Lines) {}
211
212 /// Returns the next line, merging multiple lines into one if possible.
getNextMergedLine(bool DryRun,LevelIndentTracker & IndentTracker)213 const AnnotatedLine *getNextMergedLine(bool DryRun,
214 LevelIndentTracker &IndentTracker) {
215 if (Next == End)
216 return nullptr;
217 const AnnotatedLine *Current = *Next;
218 IndentTracker.nextLine(*Current);
219 unsigned MergedLines = tryFitMultipleLinesInOne(IndentTracker, Next, End);
220 if (MergedLines > 0 && Style.ColumnLimit == 0) {
221 // Disallow line merging if there is a break at the start of one of the
222 // input lines.
223 for (unsigned i = 0; i < MergedLines; ++i)
224 if (Next[i + 1]->First->NewlinesBefore > 0)
225 MergedLines = 0;
226 }
227 if (!DryRun)
228 for (unsigned i = 0; i < MergedLines; ++i)
229 join(*Next[0], *Next[i + 1]);
230 Next = Next + MergedLines + 1;
231 return Current;
232 }
233
234 private:
235 /// Calculates how many lines can be merged into 1 starting at \p I.
236 unsigned
tryFitMultipleLinesInOne(LevelIndentTracker & IndentTracker,SmallVectorImpl<AnnotatedLine * >::const_iterator I,SmallVectorImpl<AnnotatedLine * >::const_iterator E)237 tryFitMultipleLinesInOne(LevelIndentTracker &IndentTracker,
238 SmallVectorImpl<AnnotatedLine *>::const_iterator I,
239 SmallVectorImpl<AnnotatedLine *>::const_iterator E) {
240 const unsigned Indent = IndentTracker.getIndent();
241
242 // Can't join the last line with anything.
243 if (I + 1 == E)
244 return 0;
245 // We can never merge stuff if there are trailing line comments.
246 const AnnotatedLine *TheLine = *I;
247 if (TheLine->Last->is(TT_LineComment))
248 return 0;
249 const auto &NextLine = *I[1];
250 if (NextLine.Type == LT_Invalid || NextLine.First->MustBreakBefore)
251 return 0;
252 if (TheLine->InPPDirective &&
253 (!NextLine.InPPDirective || NextLine.First->HasUnescapedNewline)) {
254 return 0;
255 }
256
257 if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit)
258 return 0;
259
260 unsigned Limit =
261 Style.ColumnLimit == 0 ? UINT_MAX : Style.ColumnLimit - Indent;
262 // If we already exceed the column limit, we set 'Limit' to 0. The different
263 // tryMerge..() functions can then decide whether to still do merging.
264 Limit = TheLine->Last->TotalLength > Limit
265 ? 0
266 : Limit - TheLine->Last->TotalLength;
267
268 if (TheLine->Last->is(TT_FunctionLBrace) &&
269 TheLine->First == TheLine->Last &&
270 !Style.BraceWrapping.SplitEmptyFunction &&
271 NextLine.First->is(tok::r_brace)) {
272 return tryMergeSimpleBlock(I, E, Limit);
273 }
274
275 const auto *PreviousLine = I != AnnotatedLines.begin() ? I[-1] : nullptr;
276 // Handle empty record blocks where the brace has already been wrapped.
277 if (PreviousLine && TheLine->Last->is(tok::l_brace) &&
278 TheLine->First == TheLine->Last) {
279 bool EmptyBlock = NextLine.First->is(tok::r_brace);
280
281 const FormatToken *Tok = PreviousLine->First;
282 if (Tok && Tok->is(tok::comment))
283 Tok = Tok->getNextNonComment();
284
285 if (Tok && Tok->getNamespaceToken()) {
286 return !Style.BraceWrapping.SplitEmptyNamespace && EmptyBlock
287 ? tryMergeSimpleBlock(I, E, Limit)
288 : 0;
289 }
290
291 if (Tok && Tok->is(tok::kw_typedef))
292 Tok = Tok->getNextNonComment();
293 if (Tok && Tok->isOneOf(tok::kw_class, tok::kw_struct, tok::kw_union,
294 tok::kw_extern, Keywords.kw_interface)) {
295 return !Style.BraceWrapping.SplitEmptyRecord && EmptyBlock
296 ? tryMergeSimpleBlock(I, E, Limit)
297 : 0;
298 }
299
300 if (Tok && Tok->is(tok::kw_template) &&
301 Style.BraceWrapping.SplitEmptyRecord && EmptyBlock) {
302 return 0;
303 }
304 }
305
306 auto ShouldMergeShortFunctions = [this, &I, &NextLine, PreviousLine,
307 TheLine]() {
308 if (Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_All)
309 return true;
310 if (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
311 NextLine.First->is(tok::r_brace)) {
312 return true;
313 }
314
315 if (Style.AllowShortFunctionsOnASingleLine &
316 FormatStyle::SFS_InlineOnly) {
317 // Just checking TheLine->Level != 0 is not enough, because it
318 // provokes treating functions inside indented namespaces as short.
319 if (Style.isJavaScript() && TheLine->Last->is(TT_FunctionLBrace))
320 return true;
321
322 if (TheLine->Level != 0) {
323 if (!PreviousLine)
324 return false;
325
326 // TODO: Use IndentTracker to avoid loop?
327 // Find the last line with lower level.
328 const AnnotatedLine *Line = nullptr;
329 for (auto J = I - 1; J >= AnnotatedLines.begin(); --J) {
330 assert(*J);
331 if (!(*J)->InPPDirective && !(*J)->isComment() &&
332 (*J)->Level < TheLine->Level) {
333 Line = *J;
334 break;
335 }
336 }
337
338 if (!Line)
339 return false;
340
341 // Check if the found line starts a record.
342 const FormatToken *LastNonComment = Line->Last;
343 assert(LastNonComment);
344 if (LastNonComment->is(tok::comment)) {
345 LastNonComment = LastNonComment->getPreviousNonComment();
346 // There must be another token (usually `{`), because we chose a
347 // non-PPDirective and non-comment line that has a smaller level.
348 assert(LastNonComment);
349 }
350 return isRecordLBrace(*LastNonComment);
351 }
352 }
353
354 return false;
355 };
356
357 bool MergeShortFunctions = ShouldMergeShortFunctions();
358
359 const FormatToken *FirstNonComment = TheLine->First;
360 if (FirstNonComment->is(tok::comment)) {
361 FirstNonComment = FirstNonComment->getNextNonComment();
362 if (!FirstNonComment)
363 return 0;
364 }
365 // FIXME: There are probably cases where we should use FirstNonComment
366 // instead of TheLine->First.
367
368 if (Style.CompactNamespaces) {
369 if (auto nsToken = TheLine->First->getNamespaceToken()) {
370 int i = 0;
371 unsigned closingLine = TheLine->MatchingClosingBlockLineIndex - 1;
372 for (; I + 1 + i != E &&
373 nsToken->TokenText == getNamespaceTokenText(I[i + 1]) &&
374 closingLine == I[i + 1]->MatchingClosingBlockLineIndex &&
375 I[i + 1]->Last->TotalLength < Limit;
376 i++, --closingLine) {
377 // No extra indent for compacted namespaces.
378 IndentTracker.skipLine(*I[i + 1]);
379
380 Limit -= I[i + 1]->Last->TotalLength;
381 }
382 return i;
383 }
384
385 if (auto nsToken = getMatchingNamespaceToken(TheLine, AnnotatedLines)) {
386 int i = 0;
387 unsigned openingLine = TheLine->MatchingOpeningBlockLineIndex - 1;
388 for (; I + 1 + i != E &&
389 nsToken->TokenText ==
390 getMatchingNamespaceTokenText(I[i + 1], AnnotatedLines) &&
391 openingLine == I[i + 1]->MatchingOpeningBlockLineIndex;
392 i++, --openingLine) {
393 // No space between consecutive braces.
394 I[i + 1]->First->SpacesRequiredBefore = !I[i]->Last->is(tok::r_brace);
395
396 // Indent like the outer-most namespace.
397 IndentTracker.nextLine(*I[i + 1]);
398 }
399 return i;
400 }
401 }
402
403 // Try to merge a function block with left brace unwrapped.
404 if (TheLine->Last->is(TT_FunctionLBrace) && TheLine->First != TheLine->Last)
405 return MergeShortFunctions ? tryMergeSimpleBlock(I, E, Limit) : 0;
406 // Try to merge a control statement block with left brace unwrapped.
407 if (TheLine->Last->is(tok::l_brace) && FirstNonComment != TheLine->Last &&
408 FirstNonComment->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for,
409 TT_ForEachMacro)) {
410 return Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never
411 ? tryMergeSimpleBlock(I, E, Limit)
412 : 0;
413 }
414 // Try to merge a control statement block with left brace wrapped.
415 if (NextLine.First->is(tok::l_brace)) {
416 if ((TheLine->First->isOneOf(tok::kw_if, tok::kw_else, tok::kw_while,
417 tok::kw_for, tok::kw_switch, tok::kw_try,
418 tok::kw_do, TT_ForEachMacro) ||
419 (TheLine->First->is(tok::r_brace) && TheLine->First->Next &&
420 TheLine->First->Next->isOneOf(tok::kw_else, tok::kw_catch))) &&
421 Style.BraceWrapping.AfterControlStatement ==
422 FormatStyle::BWACS_MultiLine) {
423 // If possible, merge the next line's wrapped left brace with the
424 // current line. Otherwise, leave it on the next line, as this is a
425 // multi-line control statement.
426 return (Style.ColumnLimit == 0 || TheLine->Level * Style.IndentWidth +
427 TheLine->Last->TotalLength <=
428 Style.ColumnLimit)
429 ? 1
430 : 0;
431 }
432 if (TheLine->First->isOneOf(tok::kw_if, tok::kw_else, tok::kw_while,
433 tok::kw_for, TT_ForEachMacro)) {
434 return (Style.BraceWrapping.AfterControlStatement ==
435 FormatStyle::BWACS_Always)
436 ? tryMergeSimpleBlock(I, E, Limit)
437 : 0;
438 }
439 if (TheLine->First->isOneOf(tok::kw_else, tok::kw_catch) &&
440 Style.BraceWrapping.AfterControlStatement ==
441 FormatStyle::BWACS_MultiLine) {
442 // This case if different from the upper BWACS_MultiLine processing
443 // in that a preceding r_brace is not on the same line as else/catch
444 // most likely because of BeforeElse/BeforeCatch set to true.
445 // If the line length doesn't fit ColumnLimit, leave l_brace on the
446 // next line to respect the BWACS_MultiLine.
447 return (Style.ColumnLimit == 0 ||
448 TheLine->Last->TotalLength <= Style.ColumnLimit)
449 ? 1
450 : 0;
451 }
452 }
453 if (PreviousLine && TheLine->First->is(tok::l_brace)) {
454 switch (PreviousLine->First->Tok.getKind()) {
455 case tok::at:
456 // Don't merge block with left brace wrapped after ObjC special blocks.
457 if (PreviousLine->First->Next) {
458 tok::ObjCKeywordKind kwId =
459 PreviousLine->First->Next->Tok.getObjCKeywordID();
460 if (kwId == tok::objc_autoreleasepool ||
461 kwId == tok::objc_synchronized) {
462 return 0;
463 }
464 }
465 break;
466
467 case tok::kw_case:
468 case tok::kw_default:
469 // Don't merge block with left brace wrapped after case labels.
470 return 0;
471
472 default:
473 break;
474 }
475 }
476
477 // Don't merge an empty template class or struct if SplitEmptyRecords
478 // is defined.
479 if (PreviousLine && Style.BraceWrapping.SplitEmptyRecord &&
480 TheLine->Last->is(tok::l_brace) && PreviousLine->Last) {
481 const FormatToken *Previous = PreviousLine->Last;
482 if (Previous) {
483 if (Previous->is(tok::comment))
484 Previous = Previous->getPreviousNonComment();
485 if (Previous) {
486 if (Previous->is(tok::greater) && !PreviousLine->InPPDirective)
487 return 0;
488 if (Previous->is(tok::identifier)) {
489 const FormatToken *PreviousPrevious =
490 Previous->getPreviousNonComment();
491 if (PreviousPrevious &&
492 PreviousPrevious->isOneOf(tok::kw_class, tok::kw_struct)) {
493 return 0;
494 }
495 }
496 }
497 }
498 }
499
500 if (TheLine->Last->is(tok::l_brace)) {
501 bool ShouldMerge = false;
502 // Try to merge records.
503 if (TheLine->Last->is(TT_EnumLBrace)) {
504 ShouldMerge = Style.AllowShortEnumsOnASingleLine;
505 } else if (TheLine->Last->isOneOf(TT_ClassLBrace, TT_StructLBrace)) {
506 // NOTE: We use AfterClass (whereas AfterStruct exists) for both classes
507 // and structs, but it seems that wrapping is still handled correctly
508 // elsewhere.
509 ShouldMerge = !Style.BraceWrapping.AfterClass ||
510 (NextLine.First->is(tok::r_brace) &&
511 !Style.BraceWrapping.SplitEmptyRecord);
512 } else {
513 // Try to merge a block with left brace unwrapped that wasn't yet
514 // covered.
515 assert(TheLine->InPPDirective ||
516 !TheLine->First->isOneOf(tok::kw_class, tok::kw_enum,
517 tok::kw_struct));
518 ShouldMerge = !Style.BraceWrapping.AfterFunction ||
519 (NextLine.First->is(tok::r_brace) &&
520 !Style.BraceWrapping.SplitEmptyFunction);
521 }
522 return ShouldMerge ? tryMergeSimpleBlock(I, E, Limit) : 0;
523 }
524
525 // Try to merge a function block with left brace wrapped.
526 if (NextLine.First->is(TT_FunctionLBrace) &&
527 Style.BraceWrapping.AfterFunction) {
528 if (NextLine.Last->is(TT_LineComment))
529 return 0;
530
531 // Check for Limit <= 2 to account for the " {".
532 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(TheLine)))
533 return 0;
534 Limit -= 2;
535
536 unsigned MergedLines = 0;
537 if (MergeShortFunctions ||
538 (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
539 NextLine.First == NextLine.Last && I + 2 != E &&
540 I[2]->First->is(tok::r_brace))) {
541 MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
542 // If we managed to merge the block, count the function header, which is
543 // on a separate line.
544 if (MergedLines > 0)
545 ++MergedLines;
546 }
547 return MergedLines;
548 }
549 auto IsElseLine = [&TheLine]() -> bool {
550 const FormatToken *First = TheLine->First;
551 if (First->is(tok::kw_else))
552 return true;
553
554 return First->is(tok::r_brace) && First->Next &&
555 First->Next->is(tok::kw_else);
556 };
557 if (TheLine->First->is(tok::kw_if) ||
558 (IsElseLine() && (Style.AllowShortIfStatementsOnASingleLine ==
559 FormatStyle::SIS_AllIfsAndElse))) {
560 return Style.AllowShortIfStatementsOnASingleLine
561 ? tryMergeSimpleControlStatement(I, E, Limit)
562 : 0;
563 }
564 if (TheLine->First->isOneOf(tok::kw_for, tok::kw_while, tok::kw_do,
565 TT_ForEachMacro)) {
566 return Style.AllowShortLoopsOnASingleLine
567 ? tryMergeSimpleControlStatement(I, E, Limit)
568 : 0;
569 }
570 if (TheLine->First->isOneOf(tok::kw_case, tok::kw_default)) {
571 return Style.AllowShortCaseLabelsOnASingleLine
572 ? tryMergeShortCaseLabels(I, E, Limit)
573 : 0;
574 }
575 if (TheLine->InPPDirective &&
576 (TheLine->First->HasUnescapedNewline || TheLine->First->IsFirst)) {
577 return tryMergeSimplePPDirective(I, E, Limit);
578 }
579 return 0;
580 }
581
582 unsigned
tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine * >::const_iterator I,SmallVectorImpl<AnnotatedLine * >::const_iterator E,unsigned Limit)583 tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
584 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
585 unsigned Limit) {
586 if (Limit == 0)
587 return 0;
588 if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline)
589 return 0;
590 if (1 + I[1]->Last->TotalLength > Limit)
591 return 0;
592 return 1;
593 }
594
tryMergeSimpleControlStatement(SmallVectorImpl<AnnotatedLine * >::const_iterator I,SmallVectorImpl<AnnotatedLine * >::const_iterator E,unsigned Limit)595 unsigned tryMergeSimpleControlStatement(
596 SmallVectorImpl<AnnotatedLine *>::const_iterator I,
597 SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) {
598 if (Limit == 0)
599 return 0;
600 if (Style.BraceWrapping.AfterControlStatement ==
601 FormatStyle::BWACS_Always &&
602 I[1]->First->is(tok::l_brace) &&
603 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) {
604 return 0;
605 }
606 if (I[1]->InPPDirective != (*I)->InPPDirective ||
607 (I[1]->InPPDirective && I[1]->First->HasUnescapedNewline)) {
608 return 0;
609 }
610 Limit = limitConsideringMacros(I + 1, E, Limit);
611 AnnotatedLine &Line = **I;
612 if (!Line.First->is(tok::kw_do) && !Line.First->is(tok::kw_else) &&
613 !Line.Last->is(tok::kw_else) && Line.Last->isNot(tok::r_paren)) {
614 return 0;
615 }
616 // Only merge `do while` if `do` is the only statement on the line.
617 if (Line.First->is(tok::kw_do) && !Line.Last->is(tok::kw_do))
618 return 0;
619 if (1 + I[1]->Last->TotalLength > Limit)
620 return 0;
621 // Don't merge with loops, ifs, a single semicolon or a line comment.
622 if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for, tok::kw_while,
623 TT_ForEachMacro, TT_LineComment)) {
624 return 0;
625 }
626 // Only inline simple if's (no nested if or else), unless specified
627 if (Style.AllowShortIfStatementsOnASingleLine ==
628 FormatStyle::SIS_WithoutElse) {
629 if (I + 2 != E && Line.startsWith(tok::kw_if) &&
630 I[2]->First->is(tok::kw_else)) {
631 return 0;
632 }
633 }
634 return 1;
635 }
636
637 unsigned
tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine * >::const_iterator I,SmallVectorImpl<AnnotatedLine * >::const_iterator E,unsigned Limit)638 tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
639 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
640 unsigned Limit) {
641 if (Limit == 0 || I + 1 == E ||
642 I[1]->First->isOneOf(tok::kw_case, tok::kw_default)) {
643 return 0;
644 }
645 if (I[0]->Last->is(tok::l_brace) || I[1]->First->is(tok::l_brace))
646 return 0;
647 unsigned NumStmts = 0;
648 unsigned Length = 0;
649 bool EndsWithComment = false;
650 bool InPPDirective = I[0]->InPPDirective;
651 bool InMacroBody = I[0]->InMacroBody;
652 const unsigned Level = I[0]->Level;
653 for (; NumStmts < 3; ++NumStmts) {
654 if (I + 1 + NumStmts == E)
655 break;
656 const AnnotatedLine *Line = I[1 + NumStmts];
657 if (Line->InPPDirective != InPPDirective)
658 break;
659 if (Line->InMacroBody != InMacroBody)
660 break;
661 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
662 break;
663 if (Line->First->isOneOf(tok::kw_if, tok::kw_for, tok::kw_switch,
664 tok::kw_while) ||
665 EndsWithComment) {
666 return 0;
667 }
668 if (Line->First->is(tok::comment)) {
669 if (Level != Line->Level)
670 return 0;
671 SmallVectorImpl<AnnotatedLine *>::const_iterator J = I + 2 + NumStmts;
672 for (; J != E; ++J) {
673 Line = *J;
674 if (Line->InPPDirective != InPPDirective)
675 break;
676 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
677 break;
678 if (Line->First->isNot(tok::comment) || Level != Line->Level)
679 return 0;
680 }
681 break;
682 }
683 if (Line->Last->is(tok::comment))
684 EndsWithComment = true;
685 Length += I[1 + NumStmts]->Last->TotalLength + 1; // 1 for the space.
686 }
687 if (NumStmts == 0 || NumStmts == 3 || Length > Limit)
688 return 0;
689 return NumStmts;
690 }
691
692 unsigned
tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine * >::const_iterator I,SmallVectorImpl<AnnotatedLine * >::const_iterator E,unsigned Limit)693 tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
694 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
695 unsigned Limit) {
696 // Don't merge with a preprocessor directive.
697 if (I[1]->Type == LT_PreprocessorDirective)
698 return 0;
699
700 AnnotatedLine &Line = **I;
701
702 // Don't merge ObjC @ keywords and methods.
703 // FIXME: If an option to allow short exception handling clauses on a single
704 // line is added, change this to not return for @try and friends.
705 if (Style.Language != FormatStyle::LK_Java &&
706 Line.First->isOneOf(tok::at, tok::minus, tok::plus)) {
707 return 0;
708 }
709
710 // Check that the current line allows merging. This depends on whether we
711 // are in a control flow statements as well as several style flags.
712 if (Line.First->is(tok::kw_case) ||
713 (Line.First->Next && Line.First->Next->is(tok::kw_else))) {
714 return 0;
715 }
716 // default: in switch statement
717 if (Line.First->is(tok::kw_default)) {
718 const FormatToken *Tok = Line.First->getNextNonComment();
719 if (Tok && Tok->is(tok::colon))
720 return 0;
721 }
722
723 auto IsCtrlStmt = [](const auto &Line) {
724 return Line.First->isOneOf(tok::kw_if, tok::kw_else, tok::kw_while,
725 tok::kw_do, tok::kw_for, TT_ForEachMacro);
726 };
727
728 const bool IsSplitBlock =
729 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never ||
730 (Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Empty &&
731 I[1]->First->isNot(tok::r_brace));
732
733 if (IsCtrlStmt(Line) ||
734 Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch,
735 tok::kw___finally, tok::r_brace,
736 Keywords.kw___except)) {
737 if (IsSplitBlock)
738 return 0;
739 // Don't merge when we can't except the case when
740 // the control statement block is empty
741 if (!Style.AllowShortIfStatementsOnASingleLine &&
742 Line.First->isOneOf(tok::kw_if, tok::kw_else) &&
743 !Style.BraceWrapping.AfterControlStatement &&
744 !I[1]->First->is(tok::r_brace)) {
745 return 0;
746 }
747 if (!Style.AllowShortIfStatementsOnASingleLine &&
748 Line.First->isOneOf(tok::kw_if, tok::kw_else) &&
749 Style.BraceWrapping.AfterControlStatement ==
750 FormatStyle::BWACS_Always &&
751 I + 2 != E && !I[2]->First->is(tok::r_brace)) {
752 return 0;
753 }
754 if (!Style.AllowShortLoopsOnASingleLine &&
755 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for,
756 TT_ForEachMacro) &&
757 !Style.BraceWrapping.AfterControlStatement &&
758 !I[1]->First->is(tok::r_brace)) {
759 return 0;
760 }
761 if (!Style.AllowShortLoopsOnASingleLine &&
762 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for,
763 TT_ForEachMacro) &&
764 Style.BraceWrapping.AfterControlStatement ==
765 FormatStyle::BWACS_Always &&
766 I + 2 != E && !I[2]->First->is(tok::r_brace)) {
767 return 0;
768 }
769 // FIXME: Consider an option to allow short exception handling clauses on
770 // a single line.
771 // FIXME: This isn't covered by tests.
772 // FIXME: For catch, __except, __finally the first token on the line
773 // is '}', so this isn't correct here.
774 if (Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch,
775 Keywords.kw___except, tok::kw___finally)) {
776 return 0;
777 }
778 }
779
780 if (Line.Last->is(tok::l_brace)) {
781 if (IsSplitBlock && Line.First == Line.Last &&
782 I > AnnotatedLines.begin() &&
783 (I[-1]->endsWith(tok::kw_else) || IsCtrlStmt(*I[-1]))) {
784 return 0;
785 }
786 FormatToken *Tok = I[1]->First;
787 auto ShouldMerge = [Tok]() {
788 if (Tok->isNot(tok::r_brace) || Tok->MustBreakBefore)
789 return false;
790 const FormatToken *Next = Tok->getNextNonComment();
791 return !Next || Next->is(tok::semi);
792 };
793
794 if (ShouldMerge()) {
795 // We merge empty blocks even if the line exceeds the column limit.
796 Tok->SpacesRequiredBefore = Style.SpaceInEmptyBlock ? 1 : 0;
797 Tok->CanBreakBefore = true;
798 return 1;
799 } else if (Limit != 0 && !Line.startsWithNamespace() &&
800 !startsExternCBlock(Line)) {
801 // We don't merge short records.
802 if (isRecordLBrace(*Line.Last))
803 return 0;
804
805 // Check that we still have three lines and they fit into the limit.
806 if (I + 2 == E || I[2]->Type == LT_Invalid)
807 return 0;
808 Limit = limitConsideringMacros(I + 2, E, Limit);
809
810 if (!nextTwoLinesFitInto(I, Limit))
811 return 0;
812
813 // Second, check that the next line does not contain any braces - if it
814 // does, readability declines when putting it into a single line.
815 if (I[1]->Last->is(TT_LineComment))
816 return 0;
817 do {
818 if (Tok->is(tok::l_brace) && Tok->isNot(BK_BracedInit))
819 return 0;
820 Tok = Tok->Next;
821 } while (Tok);
822
823 // Last, check that the third line starts with a closing brace.
824 Tok = I[2]->First;
825 if (Tok->isNot(tok::r_brace))
826 return 0;
827
828 // Don't merge "if (a) { .. } else {".
829 if (Tok->Next && Tok->Next->is(tok::kw_else))
830 return 0;
831
832 // Don't merge a trailing multi-line control statement block like:
833 // } else if (foo &&
834 // bar)
835 // { <-- current Line
836 // baz();
837 // }
838 if (Line.First == Line.Last && Line.First->isNot(TT_FunctionLBrace) &&
839 Style.BraceWrapping.AfterControlStatement ==
840 FormatStyle::BWACS_MultiLine) {
841 return 0;
842 }
843
844 return 2;
845 }
846 } else if (I[1]->First->is(tok::l_brace)) {
847 if (I[1]->Last->is(TT_LineComment))
848 return 0;
849
850 // Check for Limit <= 2 to account for the " {".
851 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(*I)))
852 return 0;
853 Limit -= 2;
854 unsigned MergedLines = 0;
855 if (Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never ||
856 (I[1]->First == I[1]->Last && I + 2 != E &&
857 I[2]->First->is(tok::r_brace))) {
858 MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
859 // If we managed to merge the block, count the statement header, which
860 // is on a separate line.
861 if (MergedLines > 0)
862 ++MergedLines;
863 }
864 return MergedLines;
865 }
866 return 0;
867 }
868
869 /// Returns the modified column limit for \p I if it is inside a macro and
870 /// needs a trailing '\'.
871 unsigned
limitConsideringMacros(SmallVectorImpl<AnnotatedLine * >::const_iterator I,SmallVectorImpl<AnnotatedLine * >::const_iterator E,unsigned Limit)872 limitConsideringMacros(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
873 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
874 unsigned Limit) {
875 if (I[0]->InPPDirective && I + 1 != E &&
876 !I[1]->First->HasUnescapedNewline && !I[1]->First->is(tok::eof)) {
877 return Limit < 2 ? 0 : Limit - 2;
878 }
879 return Limit;
880 }
881
nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine * >::const_iterator I,unsigned Limit)882 bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
883 unsigned Limit) {
884 if (I[1]->First->MustBreakBefore || I[2]->First->MustBreakBefore)
885 return false;
886 return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit;
887 }
888
containsMustBreak(const AnnotatedLine * Line)889 bool containsMustBreak(const AnnotatedLine *Line) {
890 for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next)
891 if (Tok->MustBreakBefore)
892 return true;
893 return false;
894 }
895
join(AnnotatedLine & A,const AnnotatedLine & B)896 void join(AnnotatedLine &A, const AnnotatedLine &B) {
897 assert(!A.Last->Next);
898 assert(!B.First->Previous);
899 if (B.Affected)
900 A.Affected = true;
901 A.Last->Next = B.First;
902 B.First->Previous = A.Last;
903 B.First->CanBreakBefore = true;
904 unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore;
905 for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) {
906 Tok->TotalLength += LengthA;
907 A.Last = Tok;
908 }
909 }
910
911 const FormatStyle &Style;
912 const AdditionalKeywords &Keywords;
913 const SmallVectorImpl<AnnotatedLine *>::const_iterator End;
914
915 SmallVectorImpl<AnnotatedLine *>::const_iterator Next;
916 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines;
917 };
918
markFinalized(FormatToken * Tok)919 static void markFinalized(FormatToken *Tok) {
920 for (; Tok; Tok = Tok->Next) {
921 Tok->Finalized = true;
922 for (AnnotatedLine *Child : Tok->Children)
923 markFinalized(Child->First);
924 }
925 }
926
927 #ifndef NDEBUG
printLineState(const LineState & State)928 static void printLineState(const LineState &State) {
929 llvm::dbgs() << "State: ";
930 for (const ParenState &P : State.Stack) {
931 llvm::dbgs() << (P.Tok ? P.Tok->TokenText : "F") << "|" << P.Indent << "|"
932 << P.LastSpace << "|" << P.NestedBlockIndent << " ";
933 }
934 llvm::dbgs() << State.NextToken->TokenText << "\n";
935 }
936 #endif
937
938 /// Base class for classes that format one \c AnnotatedLine.
939 class LineFormatter {
940 public:
LineFormatter(ContinuationIndenter * Indenter,WhitespaceManager * Whitespaces,const FormatStyle & Style,UnwrappedLineFormatter * BlockFormatter)941 LineFormatter(ContinuationIndenter *Indenter, WhitespaceManager *Whitespaces,
942 const FormatStyle &Style,
943 UnwrappedLineFormatter *BlockFormatter)
944 : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style),
945 BlockFormatter(BlockFormatter) {}
~LineFormatter()946 virtual ~LineFormatter() {}
947
948 /// Formats an \c AnnotatedLine and returns the penalty.
949 ///
950 /// If \p DryRun is \c false, directly applies the changes.
951 virtual unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
952 unsigned FirstStartColumn, bool DryRun) = 0;
953
954 protected:
955 /// If the \p State's next token is an r_brace closing a nested block,
956 /// format the nested block before it.
957 ///
958 /// Returns \c true if all children could be placed successfully and adapts
959 /// \p Penalty as well as \p State. If \p DryRun is false, also directly
960 /// creates changes using \c Whitespaces.
961 ///
962 /// The crucial idea here is that children always get formatted upon
963 /// encountering the closing brace right after the nested block. Now, if we
964 /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is
965 /// \c false), the entire block has to be kept on the same line (which is only
966 /// possible if it fits on the line, only contains a single statement, etc.
967 ///
968 /// If \p NewLine is true, we format the nested block on separate lines, i.e.
969 /// break after the "{", format all lines with correct indentation and the put
970 /// the closing "}" on yet another new line.
971 ///
972 /// This enables us to keep the simple structure of the
973 /// \c UnwrappedLineFormatter, where we only have two options for each token:
974 /// break or don't break.
formatChildren(LineState & State,bool NewLine,bool DryRun,unsigned & Penalty)975 bool formatChildren(LineState &State, bool NewLine, bool DryRun,
976 unsigned &Penalty) {
977 const FormatToken *LBrace = State.NextToken->getPreviousNonComment();
978 FormatToken &Previous = *State.NextToken->Previous;
979 if (!LBrace || LBrace->isNot(tok::l_brace) || LBrace->isNot(BK_Block) ||
980 Previous.Children.size() == 0) {
981 // The previous token does not open a block. Nothing to do. We don't
982 // assert so that we can simply call this function for all tokens.
983 return true;
984 }
985
986 if (NewLine) {
987 const ParenState &P = State.Stack.back();
988
989 int AdditionalIndent =
990 P.Indent - Previous.Children[0]->Level * Style.IndentWidth;
991
992 if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
993 P.NestedBlockIndent == P.LastSpace) {
994 if (State.NextToken->MatchingParen &&
995 State.NextToken->MatchingParen->is(TT_LambdaLBrace)) {
996 State.Stack.pop_back();
997 }
998 if (LBrace->is(TT_LambdaLBrace))
999 AdditionalIndent = 0;
1000 }
1001
1002 Penalty +=
1003 BlockFormatter->format(Previous.Children, DryRun, AdditionalIndent,
1004 /*FixBadIndentation=*/true);
1005 return true;
1006 }
1007
1008 if (Previous.Children[0]->First->MustBreakBefore)
1009 return false;
1010
1011 // Cannot merge into one line if this line ends on a comment.
1012 if (Previous.is(tok::comment))
1013 return false;
1014
1015 // Cannot merge multiple statements into a single line.
1016 if (Previous.Children.size() > 1)
1017 return false;
1018
1019 const AnnotatedLine *Child = Previous.Children[0];
1020 // We can't put the closing "}" on a line with a trailing comment.
1021 if (Child->Last->isTrailingComment())
1022 return false;
1023
1024 // If the child line exceeds the column limit, we wouldn't want to merge it.
1025 // We add +2 for the trailing " }".
1026 if (Style.ColumnLimit > 0 &&
1027 Child->Last->TotalLength + State.Column + 2 > Style.ColumnLimit) {
1028 return false;
1029 }
1030
1031 if (!DryRun) {
1032 Whitespaces->replaceWhitespace(
1033 *Child->First, /*Newlines=*/0, /*Spaces=*/1,
1034 /*StartOfTokenColumn=*/State.Column, /*IsAligned=*/false,
1035 State.Line->InPPDirective);
1036 }
1037 Penalty +=
1038 formatLine(*Child, State.Column + 1, /*FirstStartColumn=*/0, DryRun);
1039
1040 State.Column += 1 + Child->Last->TotalLength;
1041 return true;
1042 }
1043
1044 ContinuationIndenter *Indenter;
1045
1046 private:
1047 WhitespaceManager *Whitespaces;
1048 const FormatStyle &Style;
1049 UnwrappedLineFormatter *BlockFormatter;
1050 };
1051
1052 /// Formatter that keeps the existing line breaks.
1053 class NoColumnLimitLineFormatter : public LineFormatter {
1054 public:
NoColumnLimitLineFormatter(ContinuationIndenter * Indenter,WhitespaceManager * Whitespaces,const FormatStyle & Style,UnwrappedLineFormatter * BlockFormatter)1055 NoColumnLimitLineFormatter(ContinuationIndenter *Indenter,
1056 WhitespaceManager *Whitespaces,
1057 const FormatStyle &Style,
1058 UnwrappedLineFormatter *BlockFormatter)
1059 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
1060
1061 /// Formats the line, simply keeping all of the input's line breaking
1062 /// decisions.
formatLine(const AnnotatedLine & Line,unsigned FirstIndent,unsigned FirstStartColumn,bool DryRun)1063 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
1064 unsigned FirstStartColumn, bool DryRun) override {
1065 assert(!DryRun);
1066 LineState State = Indenter->getInitialState(FirstIndent, FirstStartColumn,
1067 &Line, /*DryRun=*/false);
1068 while (State.NextToken) {
1069 bool Newline =
1070 Indenter->mustBreak(State) ||
1071 (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0);
1072 unsigned Penalty = 0;
1073 formatChildren(State, Newline, /*DryRun=*/false, Penalty);
1074 Indenter->addTokenToState(State, Newline, /*DryRun=*/false);
1075 }
1076 return 0;
1077 }
1078 };
1079
1080 /// Formatter that puts all tokens into a single line without breaks.
1081 class NoLineBreakFormatter : public LineFormatter {
1082 public:
NoLineBreakFormatter(ContinuationIndenter * Indenter,WhitespaceManager * Whitespaces,const FormatStyle & Style,UnwrappedLineFormatter * BlockFormatter)1083 NoLineBreakFormatter(ContinuationIndenter *Indenter,
1084 WhitespaceManager *Whitespaces, const FormatStyle &Style,
1085 UnwrappedLineFormatter *BlockFormatter)
1086 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
1087
1088 /// Puts all tokens into a single line.
formatLine(const AnnotatedLine & Line,unsigned FirstIndent,unsigned FirstStartColumn,bool DryRun)1089 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
1090 unsigned FirstStartColumn, bool DryRun) override {
1091 unsigned Penalty = 0;
1092 LineState State =
1093 Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun);
1094 while (State.NextToken) {
1095 formatChildren(State, /*NewLine=*/false, DryRun, Penalty);
1096 Indenter->addTokenToState(
1097 State, /*Newline=*/State.NextToken->MustBreakBefore, DryRun);
1098 }
1099 return Penalty;
1100 }
1101 };
1102
1103 /// Finds the best way to break lines.
1104 class OptimizingLineFormatter : public LineFormatter {
1105 public:
OptimizingLineFormatter(ContinuationIndenter * Indenter,WhitespaceManager * Whitespaces,const FormatStyle & Style,UnwrappedLineFormatter * BlockFormatter)1106 OptimizingLineFormatter(ContinuationIndenter *Indenter,
1107 WhitespaceManager *Whitespaces,
1108 const FormatStyle &Style,
1109 UnwrappedLineFormatter *BlockFormatter)
1110 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
1111
1112 /// Formats the line by finding the best line breaks with line lengths
1113 /// below the column limit.
formatLine(const AnnotatedLine & Line,unsigned FirstIndent,unsigned FirstStartColumn,bool DryRun)1114 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
1115 unsigned FirstStartColumn, bool DryRun) override {
1116 LineState State =
1117 Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun);
1118
1119 // If the ObjC method declaration does not fit on a line, we should format
1120 // it with one arg per line.
1121 if (State.Line->Type == LT_ObjCMethodDecl)
1122 State.Stack.back().BreakBeforeParameter = true;
1123
1124 // Find best solution in solution space.
1125 return analyzeSolutionSpace(State, DryRun);
1126 }
1127
1128 private:
1129 struct CompareLineStatePointers {
operator ()clang::format::__anon5560ac9b0111::OptimizingLineFormatter::CompareLineStatePointers1130 bool operator()(LineState *obj1, LineState *obj2) const {
1131 return *obj1 < *obj2;
1132 }
1133 };
1134
1135 /// A pair of <penalty, count> that is used to prioritize the BFS on.
1136 ///
1137 /// In case of equal penalties, we want to prefer states that were inserted
1138 /// first. During state generation we make sure that we insert states first
1139 /// that break the line as late as possible.
1140 typedef std::pair<unsigned, unsigned> OrderedPenalty;
1141
1142 /// An edge in the solution space from \c Previous->State to \c State,
1143 /// inserting a newline dependent on the \c NewLine.
1144 struct StateNode {
StateNodeclang::format::__anon5560ac9b0111::OptimizingLineFormatter::StateNode1145 StateNode(const LineState &State, bool NewLine, StateNode *Previous)
1146 : State(State), NewLine(NewLine), Previous(Previous) {}
1147 LineState State;
1148 bool NewLine;
1149 StateNode *Previous;
1150 };
1151
1152 /// An item in the prioritized BFS search queue. The \c StateNode's
1153 /// \c State has the given \c OrderedPenalty.
1154 typedef std::pair<OrderedPenalty, StateNode *> QueueItem;
1155
1156 /// The BFS queue type.
1157 typedef std::priority_queue<QueueItem, SmallVector<QueueItem>,
1158 std::greater<QueueItem>>
1159 QueueType;
1160
1161 /// Analyze the entire solution space starting from \p InitialState.
1162 ///
1163 /// This implements a variant of Dijkstra's algorithm on the graph that spans
1164 /// the solution space (\c LineStates are the nodes). The algorithm tries to
1165 /// find the shortest path (the one with lowest penalty) from \p InitialState
1166 /// to a state where all tokens are placed. Returns the penalty.
1167 ///
1168 /// If \p DryRun is \c false, directly applies the changes.
analyzeSolutionSpace(LineState & InitialState,bool DryRun)1169 unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun) {
1170 std::set<LineState *, CompareLineStatePointers> Seen;
1171
1172 // Increasing count of \c StateNode items we have created. This is used to
1173 // create a deterministic order independent of the container.
1174 unsigned Count = 0;
1175 QueueType Queue;
1176
1177 // Insert start element into queue.
1178 StateNode *RootNode =
1179 new (Allocator.Allocate()) StateNode(InitialState, false, nullptr);
1180 Queue.push(QueueItem(OrderedPenalty(0, Count), RootNode));
1181 ++Count;
1182
1183 unsigned Penalty = 0;
1184
1185 // While not empty, take first element and follow edges.
1186 while (!Queue.empty()) {
1187 // Quit if we still haven't found a solution by now.
1188 if (Count > 25000000)
1189 return 0;
1190
1191 Penalty = Queue.top().first.first;
1192 StateNode *Node = Queue.top().second;
1193 if (!Node->State.NextToken) {
1194 LLVM_DEBUG(llvm::dbgs()
1195 << "\n---\nPenalty for line: " << Penalty << "\n");
1196 break;
1197 }
1198 Queue.pop();
1199
1200 // Cut off the analysis of certain solutions if the analysis gets too
1201 // complex. See description of IgnoreStackForComparison.
1202 if (Count > 50000)
1203 Node->State.IgnoreStackForComparison = true;
1204
1205 if (!Seen.insert(&Node->State).second) {
1206 // State already examined with lower penalty.
1207 continue;
1208 }
1209
1210 FormatDecision LastFormat = Node->State.NextToken->getDecision();
1211 if (LastFormat == FD_Unformatted || LastFormat == FD_Continue)
1212 addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue);
1213 if (LastFormat == FD_Unformatted || LastFormat == FD_Break)
1214 addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue);
1215 }
1216
1217 if (Queue.empty()) {
1218 // We were unable to find a solution, do nothing.
1219 // FIXME: Add diagnostic?
1220 LLVM_DEBUG(llvm::dbgs() << "Could not find a solution.\n");
1221 return 0;
1222 }
1223
1224 // Reconstruct the solution.
1225 if (!DryRun)
1226 reconstructPath(InitialState, Queue.top().second);
1227
1228 LLVM_DEBUG(llvm::dbgs()
1229 << "Total number of analyzed states: " << Count << "\n");
1230 LLVM_DEBUG(llvm::dbgs() << "---\n");
1231
1232 return Penalty;
1233 }
1234
1235 /// Add the following state to the analysis queue \c Queue.
1236 ///
1237 /// Assume the current state is \p PreviousNode and has been reached with a
1238 /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true.
addNextStateToQueue(unsigned Penalty,StateNode * PreviousNode,bool NewLine,unsigned * Count,QueueType * Queue)1239 void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode,
1240 bool NewLine, unsigned *Count, QueueType *Queue) {
1241 if (NewLine && !Indenter->canBreak(PreviousNode->State))
1242 return;
1243 if (!NewLine && Indenter->mustBreak(PreviousNode->State))
1244 return;
1245
1246 StateNode *Node = new (Allocator.Allocate())
1247 StateNode(PreviousNode->State, NewLine, PreviousNode);
1248 if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty))
1249 return;
1250
1251 Penalty += Indenter->addTokenToState(Node->State, NewLine, true);
1252
1253 Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node));
1254 ++(*Count);
1255 }
1256
1257 /// Applies the best formatting by reconstructing the path in the
1258 /// solution space that leads to \c Best.
reconstructPath(LineState & State,StateNode * Best)1259 void reconstructPath(LineState &State, StateNode *Best) {
1260 llvm::SmallVector<StateNode *> Path;
1261 // We do not need a break before the initial token.
1262 while (Best->Previous) {
1263 Path.push_back(Best);
1264 Best = Best->Previous;
1265 }
1266 for (const auto &Node : llvm::reverse(Path)) {
1267 unsigned Penalty = 0;
1268 formatChildren(State, Node->NewLine, /*DryRun=*/false, Penalty);
1269 Penalty += Indenter->addTokenToState(State, Node->NewLine, false);
1270
1271 LLVM_DEBUG({
1272 printLineState(Node->Previous->State);
1273 if (Node->NewLine) {
1274 llvm::dbgs() << "Penalty for placing "
1275 << Node->Previous->State.NextToken->Tok.getName()
1276 << " on a new line: " << Penalty << "\n";
1277 }
1278 });
1279 }
1280 }
1281
1282 llvm::SpecificBumpPtrAllocator<StateNode> Allocator;
1283 };
1284
1285 } // anonymous namespace
1286
format(const SmallVectorImpl<AnnotatedLine * > & Lines,bool DryRun,int AdditionalIndent,bool FixBadIndentation,unsigned FirstStartColumn,unsigned NextStartColumn,unsigned LastStartColumn)1287 unsigned UnwrappedLineFormatter::format(
1288 const SmallVectorImpl<AnnotatedLine *> &Lines, bool DryRun,
1289 int AdditionalIndent, bool FixBadIndentation, unsigned FirstStartColumn,
1290 unsigned NextStartColumn, unsigned LastStartColumn) {
1291 LineJoiner Joiner(Style, Keywords, Lines);
1292
1293 // Try to look up already computed penalty in DryRun-mode.
1294 std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned> CacheKey(
1295 &Lines, AdditionalIndent);
1296 auto CacheIt = PenaltyCache.find(CacheKey);
1297 if (DryRun && CacheIt != PenaltyCache.end())
1298 return CacheIt->second;
1299
1300 assert(!Lines.empty());
1301 unsigned Penalty = 0;
1302 LevelIndentTracker IndentTracker(Style, Keywords, Lines[0]->Level,
1303 AdditionalIndent);
1304 const AnnotatedLine *PrevPrevLine = nullptr;
1305 const AnnotatedLine *PreviousLine = nullptr;
1306 const AnnotatedLine *NextLine = nullptr;
1307
1308 // The minimum level of consecutive lines that have been formatted.
1309 unsigned RangeMinLevel = UINT_MAX;
1310
1311 bool FirstLine = true;
1312 for (const AnnotatedLine *Line =
1313 Joiner.getNextMergedLine(DryRun, IndentTracker);
1314 Line; PrevPrevLine = PreviousLine, PreviousLine = Line, Line = NextLine,
1315 FirstLine = false) {
1316 assert(Line->First);
1317 const AnnotatedLine &TheLine = *Line;
1318 unsigned Indent = IndentTracker.getIndent();
1319
1320 // We continue formatting unchanged lines to adjust their indent, e.g. if a
1321 // scope was added. However, we need to carefully stop doing this when we
1322 // exit the scope of affected lines to prevent indenting the entire
1323 // remaining file if it currently missing a closing brace.
1324 bool PreviousRBrace =
1325 PreviousLine && PreviousLine->startsWith(tok::r_brace);
1326 bool ContinueFormatting =
1327 TheLine.Level > RangeMinLevel ||
1328 (TheLine.Level == RangeMinLevel && !PreviousRBrace &&
1329 !TheLine.startsWith(tok::r_brace));
1330
1331 bool FixIndentation = (FixBadIndentation || ContinueFormatting) &&
1332 Indent != TheLine.First->OriginalColumn;
1333 bool ShouldFormat = TheLine.Affected || FixIndentation;
1334 // We cannot format this line; if the reason is that the line had a
1335 // parsing error, remember that.
1336 if (ShouldFormat && TheLine.Type == LT_Invalid && Status) {
1337 Status->FormatComplete = false;
1338 Status->Line =
1339 SourceMgr.getSpellingLineNumber(TheLine.First->Tok.getLocation());
1340 }
1341
1342 if (ShouldFormat && TheLine.Type != LT_Invalid) {
1343 if (!DryRun) {
1344 bool LastLine = TheLine.First->is(tok::eof);
1345 formatFirstToken(TheLine, PreviousLine, PrevPrevLine, Lines, Indent,
1346 LastLine ? LastStartColumn : NextStartColumn + Indent);
1347 }
1348
1349 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
1350 unsigned ColumnLimit = getColumnLimit(TheLine.InPPDirective, NextLine);
1351 bool FitsIntoOneLine =
1352 TheLine.Last->TotalLength + Indent <= ColumnLimit ||
1353 (TheLine.Type == LT_ImportStatement &&
1354 (!Style.isJavaScript() || !Style.JavaScriptWrapImports)) ||
1355 (Style.isCSharp() &&
1356 TheLine.InPPDirective); // don't split #regions in C#
1357 if (Style.ColumnLimit == 0) {
1358 NoColumnLimitLineFormatter(Indenter, Whitespaces, Style, this)
1359 .formatLine(TheLine, NextStartColumn + Indent,
1360 FirstLine ? FirstStartColumn : 0, DryRun);
1361 } else if (FitsIntoOneLine) {
1362 Penalty += NoLineBreakFormatter(Indenter, Whitespaces, Style, this)
1363 .formatLine(TheLine, NextStartColumn + Indent,
1364 FirstLine ? FirstStartColumn : 0, DryRun);
1365 } else {
1366 Penalty += OptimizingLineFormatter(Indenter, Whitespaces, Style, this)
1367 .formatLine(TheLine, NextStartColumn + Indent,
1368 FirstLine ? FirstStartColumn : 0, DryRun);
1369 }
1370 RangeMinLevel = std::min(RangeMinLevel, TheLine.Level);
1371 } else {
1372 // If no token in the current line is affected, we still need to format
1373 // affected children.
1374 if (TheLine.ChildrenAffected) {
1375 for (const FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next)
1376 if (!Tok->Children.empty())
1377 format(Tok->Children, DryRun);
1378 }
1379
1380 // Adapt following lines on the current indent level to the same level
1381 // unless the current \c AnnotatedLine is not at the beginning of a line.
1382 bool StartsNewLine =
1383 TheLine.First->NewlinesBefore > 0 || TheLine.First->IsFirst;
1384 if (StartsNewLine)
1385 IndentTracker.adjustToUnmodifiedLine(TheLine);
1386 if (!DryRun) {
1387 bool ReformatLeadingWhitespace =
1388 StartsNewLine && ((PreviousLine && PreviousLine->Affected) ||
1389 TheLine.LeadingEmptyLinesAffected);
1390 // Format the first token.
1391 if (ReformatLeadingWhitespace) {
1392 formatFirstToken(TheLine, PreviousLine, PrevPrevLine, Lines,
1393 TheLine.First->OriginalColumn,
1394 TheLine.First->OriginalColumn);
1395 } else {
1396 Whitespaces->addUntouchableToken(*TheLine.First,
1397 TheLine.InPPDirective);
1398 }
1399
1400 // Notify the WhitespaceManager about the unchanged whitespace.
1401 for (FormatToken *Tok = TheLine.First->Next; Tok; Tok = Tok->Next)
1402 Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective);
1403 }
1404 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
1405 RangeMinLevel = UINT_MAX;
1406 }
1407 if (!DryRun)
1408 markFinalized(TheLine.First);
1409 }
1410 PenaltyCache[CacheKey] = Penalty;
1411 return Penalty;
1412 }
1413
formatFirstToken(const AnnotatedLine & Line,const AnnotatedLine * PreviousLine,const AnnotatedLine * PrevPrevLine,const SmallVectorImpl<AnnotatedLine * > & Lines,unsigned Indent,unsigned NewlineIndent)1414 void UnwrappedLineFormatter::formatFirstToken(
1415 const AnnotatedLine &Line, const AnnotatedLine *PreviousLine,
1416 const AnnotatedLine *PrevPrevLine,
1417 const SmallVectorImpl<AnnotatedLine *> &Lines, unsigned Indent,
1418 unsigned NewlineIndent) {
1419 FormatToken &RootToken = *Line.First;
1420 if (RootToken.is(tok::eof)) {
1421 unsigned Newlines = std::min(RootToken.NewlinesBefore, 1u);
1422 unsigned TokenIndent = Newlines ? NewlineIndent : 0;
1423 Whitespaces->replaceWhitespace(RootToken, Newlines, TokenIndent,
1424 TokenIndent);
1425 return;
1426 }
1427 unsigned Newlines =
1428 std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1);
1429 // Remove empty lines before "}" where applicable.
1430 if (RootToken.is(tok::r_brace) &&
1431 (!RootToken.Next ||
1432 (RootToken.Next->is(tok::semi) && !RootToken.Next->Next)) &&
1433 // Do not remove empty lines before namespace closing "}".
1434 !getNamespaceToken(&Line, Lines)) {
1435 Newlines = std::min(Newlines, 1u);
1436 }
1437 // Remove empty lines at the start of nested blocks (lambdas/arrow functions)
1438 if (PreviousLine == nullptr && Line.Level > 0)
1439 Newlines = std::min(Newlines, 1u);
1440 if (Newlines == 0 && !RootToken.IsFirst)
1441 Newlines = 1;
1442 if (RootToken.IsFirst && !RootToken.HasUnescapedNewline)
1443 Newlines = 0;
1444
1445 // Remove empty lines after "{".
1446 if (!Style.KeepEmptyLinesAtTheStartOfBlocks && PreviousLine &&
1447 PreviousLine->Last->is(tok::l_brace) &&
1448 !PreviousLine->startsWithNamespace() &&
1449 !(PrevPrevLine && PrevPrevLine->startsWithNamespace() &&
1450 PreviousLine->startsWith(tok::l_brace)) &&
1451 !startsExternCBlock(*PreviousLine)) {
1452 Newlines = 1;
1453 }
1454
1455 // Insert or remove empty line before access specifiers.
1456 if (PreviousLine && RootToken.isAccessSpecifier()) {
1457 switch (Style.EmptyLineBeforeAccessModifier) {
1458 case FormatStyle::ELBAMS_Never:
1459 if (Newlines > 1)
1460 Newlines = 1;
1461 break;
1462 case FormatStyle::ELBAMS_Leave:
1463 Newlines = std::max(RootToken.NewlinesBefore, 1u);
1464 break;
1465 case FormatStyle::ELBAMS_LogicalBlock:
1466 if (PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) && Newlines <= 1)
1467 Newlines = 2;
1468 if (PreviousLine->First->isAccessSpecifier())
1469 Newlines = 1; // Previous is an access modifier remove all new lines.
1470 break;
1471 case FormatStyle::ELBAMS_Always: {
1472 const FormatToken *previousToken;
1473 if (PreviousLine->Last->is(tok::comment))
1474 previousToken = PreviousLine->Last->getPreviousNonComment();
1475 else
1476 previousToken = PreviousLine->Last;
1477 if ((!previousToken || !previousToken->is(tok::l_brace)) && Newlines <= 1)
1478 Newlines = 2;
1479 } break;
1480 }
1481 }
1482
1483 // Insert or remove empty line after access specifiers.
1484 if (PreviousLine && PreviousLine->First->isAccessSpecifier() &&
1485 (!PreviousLine->InPPDirective || !RootToken.HasUnescapedNewline)) {
1486 // EmptyLineBeforeAccessModifier is handling the case when two access
1487 // modifiers follow each other.
1488 if (!RootToken.isAccessSpecifier()) {
1489 switch (Style.EmptyLineAfterAccessModifier) {
1490 case FormatStyle::ELAAMS_Never:
1491 Newlines = 1;
1492 break;
1493 case FormatStyle::ELAAMS_Leave:
1494 Newlines = std::max(Newlines, 1u);
1495 break;
1496 case FormatStyle::ELAAMS_Always:
1497 if (RootToken.is(tok::r_brace)) // Do not add at end of class.
1498 Newlines = 1u;
1499 else
1500 Newlines = std::max(Newlines, 2u);
1501 break;
1502 }
1503 }
1504 }
1505
1506 if (Newlines)
1507 Indent = NewlineIndent;
1508
1509 // Preprocessor directives get indented before the hash only if specified. In
1510 // Javascript import statements are indented like normal statements.
1511 if (!Style.isJavaScript() &&
1512 Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash &&
1513 (Line.Type == LT_PreprocessorDirective ||
1514 Line.Type == LT_ImportStatement)) {
1515 Indent = 0;
1516 }
1517
1518 Whitespaces->replaceWhitespace(RootToken, Newlines, Indent, Indent,
1519 /*IsAligned=*/false,
1520 Line.InPPDirective &&
1521 !RootToken.HasUnescapedNewline);
1522 }
1523
1524 unsigned
getColumnLimit(bool InPPDirective,const AnnotatedLine * NextLine) const1525 UnwrappedLineFormatter::getColumnLimit(bool InPPDirective,
1526 const AnnotatedLine *NextLine) const {
1527 // In preprocessor directives reserve two chars for trailing " \" if the
1528 // next line continues the preprocessor directive.
1529 bool ContinuesPPDirective =
1530 InPPDirective &&
1531 // If there is no next line, this is likely a child line and the parent
1532 // continues the preprocessor directive.
1533 (!NextLine ||
1534 (NextLine->InPPDirective &&
1535 // If there is an unescaped newline between this line and the next, the
1536 // next line starts a new preprocessor directive.
1537 !NextLine->First->HasUnescapedNewline));
1538 return Style.ColumnLimit - (ContinuesPPDirective ? 2 : 0);
1539 }
1540
1541 } // namespace format
1542 } // namespace clang
1543