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
29 /// Tracks the indent level of \c AnnotatedLines across levels.
30 ///
31 /// \c nextLine must be called for each \c AnnotatedLine, after which \c
32 /// getIndent() will return the indent for the last line \c nextLine was called
33 /// with.
34 /// If the line is not formatted (and thus the indent does not change), calling
35 /// \c adjustToUnmodifiedLine after the call to \c nextLine will cause
36 /// subsequent lines on the same level to be indented at the same level as the
37 /// given line.
38 class LevelIndentTracker {
39 public:
LevelIndentTracker(const FormatStyle & Style,const AdditionalKeywords & Keywords,unsigned StartLevel,int AdditionalIndent)40 LevelIndentTracker(const FormatStyle &Style,
41 const AdditionalKeywords &Keywords, unsigned StartLevel,
42 int AdditionalIndent)
43 : Style(Style), Keywords(Keywords), AdditionalIndent(AdditionalIndent) {
44 for (unsigned i = 0; i != StartLevel; ++i)
45 IndentForLevel.push_back(Style.IndentWidth * i + AdditionalIndent);
46 }
47
48 /// Returns the indent for the current line.
getIndent() const49 unsigned getIndent() const { return Indent; }
50
51 /// Update the indent state given that \p Line is going to be formatted
52 /// next.
nextLine(const AnnotatedLine & Line)53 void nextLine(const AnnotatedLine &Line) {
54 Offset = getIndentOffset(*Line.First);
55 // Update the indent level cache size so that we can rely on it
56 // having the right size in adjustToUnmodifiedline.
57 while (IndentForLevel.size() <= Line.Level)
58 IndentForLevel.push_back(-1);
59 if (Line.InPPDirective) {
60 Indent = Line.Level * Style.IndentWidth + AdditionalIndent;
61 } else {
62 IndentForLevel.resize(Line.Level + 1);
63 Indent = getIndent(IndentForLevel, Line.Level);
64 }
65 if (static_cast<int>(Indent) + Offset >= 0)
66 Indent += Offset;
67 }
68
69 /// Update the indent state given that \p Line indent should be
70 /// skipped.
skipLine(const AnnotatedLine & Line)71 void skipLine(const AnnotatedLine &Line) {
72 while (IndentForLevel.size() <= Line.Level)
73 IndentForLevel.push_back(Indent);
74 }
75
76 /// Update the level indent to adapt to the given \p Line.
77 ///
78 /// When a line is not formatted, we move the subsequent lines on the same
79 /// level to the same indent.
80 /// Note that \c nextLine must have been called before this method.
adjustToUnmodifiedLine(const AnnotatedLine & Line)81 void adjustToUnmodifiedLine(const AnnotatedLine &Line) {
82 unsigned LevelIndent = Line.First->OriginalColumn;
83 if (static_cast<int>(LevelIndent) - Offset >= 0)
84 LevelIndent -= Offset;
85 if ((!Line.First->is(tok::comment) || IndentForLevel[Line.Level] == -1) &&
86 !Line.InPPDirective)
87 IndentForLevel[Line.Level] = LevelIndent;
88 }
89
90 private:
91 /// Get the offset of the line relatively to the level.
92 ///
93 /// For example, 'public:' labels in classes are offset by 1 or 2
94 /// characters to the left from their level.
getIndentOffset(const FormatToken & RootToken)95 int getIndentOffset(const FormatToken &RootToken) {
96 if (Style.Language == FormatStyle::LK_Java ||
97 Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp())
98 return 0;
99 if (RootToken.isAccessSpecifier(false) ||
100 RootToken.isObjCAccessSpecifier() ||
101 (RootToken.isOneOf(Keywords.kw_signals, Keywords.kw_qsignals) &&
102 RootToken.Next && RootToken.Next->is(tok::colon)))
103 return Style.AccessModifierOffset;
104 return 0;
105 }
106
107 /// Get the indent of \p Level from \p IndentForLevel.
108 ///
109 /// \p IndentForLevel must contain the indent for the level \c l
110 /// at \p IndentForLevel[l], or a value < 0 if the indent for
111 /// that level is unknown.
getIndent(ArrayRef<int> IndentForLevel,unsigned Level)112 unsigned getIndent(ArrayRef<int> IndentForLevel, unsigned Level) {
113 if (IndentForLevel[Level] != -1)
114 return IndentForLevel[Level];
115 if (Level == 0)
116 return 0;
117 return getIndent(IndentForLevel, Level - 1) + Style.IndentWidth;
118 }
119
120 const FormatStyle &Style;
121 const AdditionalKeywords &Keywords;
122 const unsigned AdditionalIndent;
123
124 /// The indent in characters for each level.
125 std::vector<int> IndentForLevel;
126
127 /// Offset of the current line relative to the indent level.
128 ///
129 /// For example, the 'public' keywords is often indented with a negative
130 /// offset.
131 int Offset = 0;
132
133 /// The current line's indent.
134 unsigned Indent = 0;
135 };
136
getMatchingNamespaceToken(const AnnotatedLine * Line,const SmallVectorImpl<AnnotatedLine * > & AnnotatedLines)137 const FormatToken *getMatchingNamespaceToken(
138 const AnnotatedLine *Line,
139 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
140 if (!Line->startsWith(tok::r_brace))
141 return nullptr;
142 size_t StartLineIndex = Line->MatchingOpeningBlockLineIndex;
143 if (StartLineIndex == UnwrappedLine::kInvalidIndex)
144 return nullptr;
145 assert(StartLineIndex < AnnotatedLines.size());
146 return AnnotatedLines[StartLineIndex]->First->getNamespaceToken();
147 }
148
getNamespaceTokenText(const AnnotatedLine * Line)149 StringRef getNamespaceTokenText(const AnnotatedLine *Line) {
150 const FormatToken *NamespaceToken = Line->First->getNamespaceToken();
151 return NamespaceToken ? NamespaceToken->TokenText : StringRef();
152 }
153
getMatchingNamespaceTokenText(const AnnotatedLine * Line,const SmallVectorImpl<AnnotatedLine * > & AnnotatedLines)154 StringRef getMatchingNamespaceTokenText(
155 const AnnotatedLine *Line,
156 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
157 const FormatToken *NamespaceToken =
158 getMatchingNamespaceToken(Line, AnnotatedLines);
159 return NamespaceToken ? NamespaceToken->TokenText : StringRef();
160 }
161
162 class LineJoiner {
163 public:
LineJoiner(const FormatStyle & Style,const AdditionalKeywords & Keywords,const SmallVectorImpl<AnnotatedLine * > & Lines)164 LineJoiner(const FormatStyle &Style, const AdditionalKeywords &Keywords,
165 const SmallVectorImpl<AnnotatedLine *> &Lines)
166 : Style(Style), Keywords(Keywords), End(Lines.end()), Next(Lines.begin()),
167 AnnotatedLines(Lines) {}
168
169 /// Returns the next line, merging multiple lines into one if possible.
getNextMergedLine(bool DryRun,LevelIndentTracker & IndentTracker)170 const AnnotatedLine *getNextMergedLine(bool DryRun,
171 LevelIndentTracker &IndentTracker) {
172 if (Next == End)
173 return nullptr;
174 const AnnotatedLine *Current = *Next;
175 IndentTracker.nextLine(*Current);
176 unsigned MergedLines = tryFitMultipleLinesInOne(IndentTracker, Next, End);
177 if (MergedLines > 0 && Style.ColumnLimit == 0)
178 // Disallow line merging if there is a break at the start of one of the
179 // input lines.
180 for (unsigned i = 0; i < MergedLines; ++i)
181 if (Next[i + 1]->First->NewlinesBefore > 0)
182 MergedLines = 0;
183 if (!DryRun)
184 for (unsigned i = 0; i < MergedLines; ++i)
185 join(*Next[0], *Next[i + 1]);
186 Next = Next + MergedLines + 1;
187 return Current;
188 }
189
190 private:
191 /// Calculates how many lines can be merged into 1 starting at \p I.
192 unsigned
tryFitMultipleLinesInOne(LevelIndentTracker & IndentTracker,SmallVectorImpl<AnnotatedLine * >::const_iterator I,SmallVectorImpl<AnnotatedLine * >::const_iterator E)193 tryFitMultipleLinesInOne(LevelIndentTracker &IndentTracker,
194 SmallVectorImpl<AnnotatedLine *>::const_iterator I,
195 SmallVectorImpl<AnnotatedLine *>::const_iterator E) {
196 const unsigned Indent = IndentTracker.getIndent();
197
198 // Can't join the last line with anything.
199 if (I + 1 == E)
200 return 0;
201 // We can never merge stuff if there are trailing line comments.
202 const AnnotatedLine *TheLine = *I;
203 if (TheLine->Last->is(TT_LineComment))
204 return 0;
205 if (I[1]->Type == LT_Invalid || I[1]->First->MustBreakBefore)
206 return 0;
207 if (TheLine->InPPDirective &&
208 (!I[1]->InPPDirective || I[1]->First->HasUnescapedNewline))
209 return 0;
210
211 if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit)
212 return 0;
213
214 unsigned Limit =
215 Style.ColumnLimit == 0 ? UINT_MAX : Style.ColumnLimit - Indent;
216 // If we already exceed the column limit, we set 'Limit' to 0. The different
217 // tryMerge..() functions can then decide whether to still do merging.
218 Limit = TheLine->Last->TotalLength > Limit
219 ? 0
220 : Limit - TheLine->Last->TotalLength;
221
222 if (TheLine->Last->is(TT_FunctionLBrace) &&
223 TheLine->First == TheLine->Last &&
224 !Style.BraceWrapping.SplitEmptyFunction &&
225 I[1]->First->is(tok::r_brace))
226 return tryMergeSimpleBlock(I, E, Limit);
227
228 // Handle empty record blocks where the brace has already been wrapped
229 if (TheLine->Last->is(tok::l_brace) && TheLine->First == TheLine->Last &&
230 I != AnnotatedLines.begin()) {
231 bool EmptyBlock = I[1]->First->is(tok::r_brace);
232
233 const FormatToken *Tok = I[-1]->First;
234 if (Tok && Tok->is(tok::comment))
235 Tok = Tok->getNextNonComment();
236
237 if (Tok && Tok->getNamespaceToken())
238 return !Style.BraceWrapping.SplitEmptyNamespace && EmptyBlock
239 ? tryMergeSimpleBlock(I, E, Limit)
240 : 0;
241
242 if (Tok && Tok->is(tok::kw_typedef))
243 Tok = Tok->getNextNonComment();
244 if (Tok && Tok->isOneOf(tok::kw_class, tok::kw_struct, tok::kw_union,
245 tok::kw_extern, Keywords.kw_interface))
246 return !Style.BraceWrapping.SplitEmptyRecord && EmptyBlock
247 ? tryMergeSimpleBlock(I, E, Limit)
248 : 0;
249 }
250
251 // FIXME: TheLine->Level != 0 might or might not be the right check to do.
252 // If necessary, change to something smarter.
253 bool MergeShortFunctions =
254 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_All ||
255 (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
256 I[1]->First->is(tok::r_brace)) ||
257 (Style.AllowShortFunctionsOnASingleLine & FormatStyle::SFS_InlineOnly &&
258 TheLine->Level != 0);
259
260 if (Style.CompactNamespaces) {
261 if (auto nsToken = TheLine->First->getNamespaceToken()) {
262 int i = 0;
263 unsigned closingLine = TheLine->MatchingClosingBlockLineIndex - 1;
264 for (; I + 1 + i != E &&
265 nsToken->TokenText == getNamespaceTokenText(I[i + 1]) &&
266 closingLine == I[i + 1]->MatchingClosingBlockLineIndex &&
267 I[i + 1]->Last->TotalLength < Limit;
268 i++, closingLine--) {
269 // No extra indent for compacted namespaces
270 IndentTracker.skipLine(*I[i + 1]);
271
272 Limit -= I[i + 1]->Last->TotalLength;
273 }
274 return i;
275 }
276
277 if (auto nsToken = getMatchingNamespaceToken(TheLine, AnnotatedLines)) {
278 int i = 0;
279 unsigned openingLine = TheLine->MatchingOpeningBlockLineIndex - 1;
280 for (; I + 1 + i != E &&
281 nsToken->TokenText ==
282 getMatchingNamespaceTokenText(I[i + 1], AnnotatedLines) &&
283 openingLine == I[i + 1]->MatchingOpeningBlockLineIndex;
284 i++, openingLine--) {
285 // No space between consecutive braces
286 I[i + 1]->First->SpacesRequiredBefore = !I[i]->Last->is(tok::r_brace);
287
288 // Indent like the outer-most namespace
289 IndentTracker.nextLine(*I[i + 1]);
290 }
291 return i;
292 }
293 }
294
295 // Try to merge a function block with left brace unwrapped
296 if (TheLine->Last->is(TT_FunctionLBrace) &&
297 TheLine->First != TheLine->Last) {
298 return MergeShortFunctions ? tryMergeSimpleBlock(I, E, Limit) : 0;
299 }
300 // Try to merge a control statement block with left brace unwrapped
301 if (TheLine->Last->is(tok::l_brace) && TheLine->First != TheLine->Last &&
302 TheLine->First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for)) {
303 return Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never
304 ? tryMergeSimpleBlock(I, E, Limit)
305 : 0;
306 }
307 // Try to merge a control statement block with left brace wrapped
308 if (I[1]->First->is(tok::l_brace) &&
309 (TheLine->First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for,
310 tok::kw_switch, tok::kw_try, tok::kw_do) ||
311 (TheLine->First->is(tok::r_brace) && TheLine->First->Next &&
312 TheLine->First->Next->isOneOf(tok::kw_else, tok::kw_catch))) &&
313 Style.BraceWrapping.AfterControlStatement ==
314 FormatStyle::BWACS_MultiLine) {
315 // If possible, merge the next line's wrapped left brace with the current
316 // line. Otherwise, leave it on the next line, as this is a multi-line
317 // control statement.
318 return (Style.ColumnLimit == 0 ||
319 TheLine->Last->TotalLength <= Style.ColumnLimit)
320 ? 1
321 : 0;
322 } else if (I[1]->First->is(tok::l_brace) &&
323 TheLine->First->isOneOf(tok::kw_if, tok::kw_while,
324 tok::kw_for)) {
325 return (Style.BraceWrapping.AfterControlStatement ==
326 FormatStyle::BWACS_Always)
327 ? tryMergeSimpleBlock(I, E, Limit)
328 : 0;
329 } else if (I[1]->First->is(tok::l_brace) &&
330 TheLine->First->isOneOf(tok::kw_else, tok::kw_catch) &&
331 Style.BraceWrapping.AfterControlStatement ==
332 FormatStyle::BWACS_MultiLine) {
333 // This case if different from the upper BWACS_MultiLine processing
334 // in that a preceding r_brace is not on the same line as else/catch
335 // most likely because of BeforeElse/BeforeCatch set to true.
336 // If the line length doesn't fit ColumnLimit, leave l_brace on the
337 // next line to respect the BWACS_MultiLine.
338 return (Style.ColumnLimit == 0 ||
339 TheLine->Last->TotalLength <= Style.ColumnLimit)
340 ? 1
341 : 0;
342 }
343 // Try to merge either empty or one-line block if is precedeed by control
344 // statement token
345 if (TheLine->First->is(tok::l_brace) && TheLine->First == TheLine->Last &&
346 I != AnnotatedLines.begin() &&
347 I[-1]->First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for)) {
348 unsigned MergedLines = 0;
349 if (Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never) {
350 MergedLines = tryMergeSimpleBlock(I - 1, E, Limit);
351 // If we managed to merge the block, discard the first merged line
352 // since we are merging starting from I.
353 if (MergedLines > 0)
354 --MergedLines;
355 }
356 return MergedLines;
357 }
358 // Don't merge block with left brace wrapped after ObjC special blocks
359 if (TheLine->First->is(tok::l_brace) && I != AnnotatedLines.begin() &&
360 I[-1]->First->is(tok::at) && I[-1]->First->Next) {
361 tok::ObjCKeywordKind kwId = I[-1]->First->Next->Tok.getObjCKeywordID();
362 if (kwId == clang::tok::objc_autoreleasepool ||
363 kwId == clang::tok::objc_synchronized)
364 return 0;
365 }
366 // Don't merge block with left brace wrapped after case labels
367 if (TheLine->First->is(tok::l_brace) && I != AnnotatedLines.begin() &&
368 I[-1]->First->isOneOf(tok::kw_case, tok::kw_default))
369 return 0;
370 // Try to merge a block with left brace wrapped that wasn't yet covered
371 if (TheLine->Last->is(tok::l_brace)) {
372 return !Style.BraceWrapping.AfterFunction ||
373 (I[1]->First->is(tok::r_brace) &&
374 !Style.BraceWrapping.SplitEmptyRecord)
375 ? tryMergeSimpleBlock(I, E, Limit)
376 : 0;
377 }
378 // Try to merge a function block with left brace wrapped
379 if (I[1]->First->is(TT_FunctionLBrace) &&
380 Style.BraceWrapping.AfterFunction) {
381 if (I[1]->Last->is(TT_LineComment))
382 return 0;
383
384 // Check for Limit <= 2 to account for the " {".
385 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(TheLine)))
386 return 0;
387 Limit -= 2;
388
389 unsigned MergedLines = 0;
390 if (MergeShortFunctions ||
391 (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
392 I[1]->First == I[1]->Last && I + 2 != E &&
393 I[2]->First->is(tok::r_brace))) {
394 MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
395 // If we managed to merge the block, count the function header, which is
396 // on a separate line.
397 if (MergedLines > 0)
398 ++MergedLines;
399 }
400 return MergedLines;
401 }
402 if (TheLine->First->is(tok::kw_if)) {
403 return Style.AllowShortIfStatementsOnASingleLine
404 ? tryMergeSimpleControlStatement(I, E, Limit)
405 : 0;
406 }
407 if (TheLine->First->isOneOf(tok::kw_for, tok::kw_while)) {
408 return Style.AllowShortLoopsOnASingleLine
409 ? tryMergeSimpleControlStatement(I, E, Limit)
410 : 0;
411 }
412 if (TheLine->First->isOneOf(tok::kw_case, tok::kw_default)) {
413 return Style.AllowShortCaseLabelsOnASingleLine
414 ? tryMergeShortCaseLabels(I, E, Limit)
415 : 0;
416 }
417 if (TheLine->InPPDirective &&
418 (TheLine->First->HasUnescapedNewline || TheLine->First->IsFirst)) {
419 return tryMergeSimplePPDirective(I, E, Limit);
420 }
421 return 0;
422 }
423
424 unsigned
tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine * >::const_iterator I,SmallVectorImpl<AnnotatedLine * >::const_iterator E,unsigned Limit)425 tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
426 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
427 unsigned Limit) {
428 if (Limit == 0)
429 return 0;
430 if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline)
431 return 0;
432 if (1 + I[1]->Last->TotalLength > Limit)
433 return 0;
434 return 1;
435 }
436
tryMergeSimpleControlStatement(SmallVectorImpl<AnnotatedLine * >::const_iterator I,SmallVectorImpl<AnnotatedLine * >::const_iterator E,unsigned Limit)437 unsigned tryMergeSimpleControlStatement(
438 SmallVectorImpl<AnnotatedLine *>::const_iterator I,
439 SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) {
440 if (Limit == 0)
441 return 0;
442 if (Style.BraceWrapping.AfterControlStatement ==
443 FormatStyle::BWACS_Always &&
444 I[1]->First->is(tok::l_brace) &&
445 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never)
446 return 0;
447 if (I[1]->InPPDirective != (*I)->InPPDirective ||
448 (I[1]->InPPDirective && I[1]->First->HasUnescapedNewline))
449 return 0;
450 Limit = limitConsideringMacros(I + 1, E, Limit);
451 AnnotatedLine &Line = **I;
452 if (Line.Last->isNot(tok::r_paren))
453 return 0;
454 if (1 + I[1]->Last->TotalLength > Limit)
455 return 0;
456 if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for, tok::kw_while,
457 TT_LineComment))
458 return 0;
459 // Only inline simple if's (no nested if or else), unless specified
460 if (Style.AllowShortIfStatementsOnASingleLine != FormatStyle::SIS_Always) {
461 if (I + 2 != E && Line.startsWith(tok::kw_if) &&
462 I[2]->First->is(tok::kw_else))
463 return 0;
464 }
465 return 1;
466 }
467
468 unsigned
tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine * >::const_iterator I,SmallVectorImpl<AnnotatedLine * >::const_iterator E,unsigned Limit)469 tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
470 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
471 unsigned Limit) {
472 if (Limit == 0 || I + 1 == E ||
473 I[1]->First->isOneOf(tok::kw_case, tok::kw_default))
474 return 0;
475 if (I[0]->Last->is(tok::l_brace) || I[1]->First->is(tok::l_brace))
476 return 0;
477 unsigned NumStmts = 0;
478 unsigned Length = 0;
479 bool EndsWithComment = false;
480 bool InPPDirective = I[0]->InPPDirective;
481 const unsigned Level = I[0]->Level;
482 for (; NumStmts < 3; ++NumStmts) {
483 if (I + 1 + NumStmts == E)
484 break;
485 const AnnotatedLine *Line = I[1 + NumStmts];
486 if (Line->InPPDirective != InPPDirective)
487 break;
488 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
489 break;
490 if (Line->First->isOneOf(tok::kw_if, tok::kw_for, tok::kw_switch,
491 tok::kw_while) ||
492 EndsWithComment)
493 return 0;
494 if (Line->First->is(tok::comment)) {
495 if (Level != Line->Level)
496 return 0;
497 SmallVectorImpl<AnnotatedLine *>::const_iterator J = I + 2 + NumStmts;
498 for (; J != E; ++J) {
499 Line = *J;
500 if (Line->InPPDirective != InPPDirective)
501 break;
502 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
503 break;
504 if (Line->First->isNot(tok::comment) || Level != Line->Level)
505 return 0;
506 }
507 break;
508 }
509 if (Line->Last->is(tok::comment))
510 EndsWithComment = true;
511 Length += I[1 + NumStmts]->Last->TotalLength + 1; // 1 for the space.
512 }
513 if (NumStmts == 0 || NumStmts == 3 || Length > Limit)
514 return 0;
515 return NumStmts;
516 }
517
518 unsigned
tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine * >::const_iterator I,SmallVectorImpl<AnnotatedLine * >::const_iterator E,unsigned Limit)519 tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
520 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
521 unsigned Limit) {
522 AnnotatedLine &Line = **I;
523
524 // Don't merge ObjC @ keywords and methods.
525 // FIXME: If an option to allow short exception handling clauses on a single
526 // line is added, change this to not return for @try and friends.
527 if (Style.Language != FormatStyle::LK_Java &&
528 Line.First->isOneOf(tok::at, tok::minus, tok::plus))
529 return 0;
530
531 // Check that the current line allows merging. This depends on whether we
532 // are in a control flow statements as well as several style flags.
533 if (Line.First->isOneOf(tok::kw_else, tok::kw_case) ||
534 (Line.First->Next && Line.First->Next->is(tok::kw_else)))
535 return 0;
536 // default: in switch statement
537 if (Line.First->is(tok::kw_default)) {
538 const FormatToken *Tok = Line.First->getNextNonComment();
539 if (Tok && Tok->is(tok::colon))
540 return 0;
541 }
542 if (Line.First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_do, tok::kw_try,
543 tok::kw___try, tok::kw_catch, tok::kw___finally,
544 tok::kw_for, tok::r_brace, Keywords.kw___except)) {
545 if (Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never)
546 return 0;
547 // Don't merge when we can't except the case when
548 // the control statement block is empty
549 if (!Style.AllowShortIfStatementsOnASingleLine &&
550 Line.startsWith(tok::kw_if) &&
551 !Style.BraceWrapping.AfterControlStatement &&
552 !I[1]->First->is(tok::r_brace))
553 return 0;
554 if (!Style.AllowShortIfStatementsOnASingleLine &&
555 Line.startsWith(tok::kw_if) &&
556 Style.BraceWrapping.AfterControlStatement ==
557 FormatStyle::BWACS_Always &&
558 I + 2 != E && !I[2]->First->is(tok::r_brace))
559 return 0;
560 if (!Style.AllowShortLoopsOnASingleLine &&
561 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for) &&
562 !Style.BraceWrapping.AfterControlStatement &&
563 !I[1]->First->is(tok::r_brace))
564 return 0;
565 if (!Style.AllowShortLoopsOnASingleLine &&
566 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for) &&
567 Style.BraceWrapping.AfterControlStatement ==
568 FormatStyle::BWACS_Always &&
569 I + 2 != E && !I[2]->First->is(tok::r_brace))
570 return 0;
571 // FIXME: Consider an option to allow short exception handling clauses on
572 // a single line.
573 // FIXME: This isn't covered by tests.
574 // FIXME: For catch, __except, __finally the first token on the line
575 // is '}', so this isn't correct here.
576 if (Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch,
577 Keywords.kw___except, tok::kw___finally))
578 return 0;
579 }
580
581 if (Line.Last->is(tok::l_brace)) {
582 FormatToken *Tok = I[1]->First;
583 if (Tok->is(tok::r_brace) && !Tok->MustBreakBefore &&
584 (Tok->getNextNonComment() == nullptr ||
585 Tok->getNextNonComment()->is(tok::semi))) {
586 // We merge empty blocks even if the line exceeds the column limit.
587 Tok->SpacesRequiredBefore = Style.SpaceInEmptyBlock ? 1 : 0;
588 Tok->CanBreakBefore = true;
589 return 1;
590 } else if (Limit != 0 && !Line.startsWithNamespace() &&
591 !startsExternCBlock(Line)) {
592 // We don't merge short records.
593 FormatToken *RecordTok = Line.First;
594 // Skip record modifiers.
595 while (RecordTok->Next &&
596 RecordTok->isOneOf(tok::kw_typedef, tok::kw_export,
597 Keywords.kw_declare, Keywords.kw_abstract,
598 tok::kw_default))
599 RecordTok = RecordTok->Next;
600 if (RecordTok &&
601 RecordTok->isOneOf(tok::kw_class, tok::kw_union, tok::kw_struct,
602 Keywords.kw_interface))
603 return 0;
604
605 // Check that we still have three lines and they fit into the limit.
606 if (I + 2 == E || I[2]->Type == LT_Invalid)
607 return 0;
608 Limit = limitConsideringMacros(I + 2, E, Limit);
609
610 if (!nextTwoLinesFitInto(I, Limit))
611 return 0;
612
613 // Second, check that the next line does not contain any braces - if it
614 // does, readability declines when putting it into a single line.
615 if (I[1]->Last->is(TT_LineComment))
616 return 0;
617 do {
618 if (Tok->is(tok::l_brace) && Tok->BlockKind != BK_BracedInit)
619 return 0;
620 Tok = Tok->Next;
621 } while (Tok);
622
623 // Last, check that the third line starts with a closing brace.
624 Tok = I[2]->First;
625 if (Tok->isNot(tok::r_brace))
626 return 0;
627
628 // Don't merge "if (a) { .. } else {".
629 if (Tok->Next && Tok->Next->is(tok::kw_else))
630 return 0;
631
632 // Don't merge a trailing multi-line control statement block like:
633 // } else if (foo &&
634 // bar)
635 // { <-- current Line
636 // baz();
637 // }
638 if (Line.First == Line.Last &&
639 Style.BraceWrapping.AfterControlStatement ==
640 FormatStyle::BWACS_MultiLine)
641 return 0;
642
643 return 2;
644 }
645 } else if (I[1]->First->is(tok::l_brace)) {
646 if (I[1]->Last->is(TT_LineComment))
647 return 0;
648
649 // Check for Limit <= 2 to account for the " {".
650 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(*I)))
651 return 0;
652 Limit -= 2;
653 unsigned MergedLines = 0;
654 if (Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never ||
655 (I[1]->First == I[1]->Last && I + 2 != E &&
656 I[2]->First->is(tok::r_brace))) {
657 MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
658 // If we managed to merge the block, count the statement header, which
659 // is on a separate line.
660 if (MergedLines > 0)
661 ++MergedLines;
662 }
663 return MergedLines;
664 }
665 return 0;
666 }
667
668 /// Returns the modified column limit for \p I if it is inside a macro and
669 /// needs a trailing '\'.
670 unsigned
limitConsideringMacros(SmallVectorImpl<AnnotatedLine * >::const_iterator I,SmallVectorImpl<AnnotatedLine * >::const_iterator E,unsigned Limit)671 limitConsideringMacros(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
672 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
673 unsigned Limit) {
674 if (I[0]->InPPDirective && I + 1 != E &&
675 !I[1]->First->HasUnescapedNewline && !I[1]->First->is(tok::eof)) {
676 return Limit < 2 ? 0 : Limit - 2;
677 }
678 return Limit;
679 }
680
nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine * >::const_iterator I,unsigned Limit)681 bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
682 unsigned Limit) {
683 if (I[1]->First->MustBreakBefore || I[2]->First->MustBreakBefore)
684 return false;
685 return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit;
686 }
687
containsMustBreak(const AnnotatedLine * Line)688 bool containsMustBreak(const AnnotatedLine *Line) {
689 for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) {
690 if (Tok->MustBreakBefore)
691 return true;
692 }
693 return false;
694 }
695
join(AnnotatedLine & A,const AnnotatedLine & B)696 void join(AnnotatedLine &A, const AnnotatedLine &B) {
697 assert(!A.Last->Next);
698 assert(!B.First->Previous);
699 if (B.Affected)
700 A.Affected = true;
701 A.Last->Next = B.First;
702 B.First->Previous = A.Last;
703 B.First->CanBreakBefore = true;
704 unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore;
705 for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) {
706 Tok->TotalLength += LengthA;
707 A.Last = Tok;
708 }
709 }
710
711 const FormatStyle &Style;
712 const AdditionalKeywords &Keywords;
713 const SmallVectorImpl<AnnotatedLine *>::const_iterator End;
714
715 SmallVectorImpl<AnnotatedLine *>::const_iterator Next;
716 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines;
717 };
718
markFinalized(FormatToken * Tok)719 static void markFinalized(FormatToken *Tok) {
720 for (; Tok; Tok = Tok->Next) {
721 Tok->Finalized = true;
722 for (AnnotatedLine *Child : Tok->Children)
723 markFinalized(Child->First);
724 }
725 }
726
727 #ifndef NDEBUG
printLineState(const LineState & State)728 static void printLineState(const LineState &State) {
729 llvm::dbgs() << "State: ";
730 for (const ParenState &P : State.Stack) {
731 llvm::dbgs() << (P.Tok ? P.Tok->TokenText : "F") << "|" << P.Indent << "|"
732 << P.LastSpace << "|" << P.NestedBlockIndent << " ";
733 }
734 llvm::dbgs() << State.NextToken->TokenText << "\n";
735 }
736 #endif
737
738 /// Base class for classes that format one \c AnnotatedLine.
739 class LineFormatter {
740 public:
LineFormatter(ContinuationIndenter * Indenter,WhitespaceManager * Whitespaces,const FormatStyle & Style,UnwrappedLineFormatter * BlockFormatter)741 LineFormatter(ContinuationIndenter *Indenter, WhitespaceManager *Whitespaces,
742 const FormatStyle &Style,
743 UnwrappedLineFormatter *BlockFormatter)
744 : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style),
745 BlockFormatter(BlockFormatter) {}
~LineFormatter()746 virtual ~LineFormatter() {}
747
748 /// Formats an \c AnnotatedLine and returns the penalty.
749 ///
750 /// If \p DryRun is \c false, directly applies the changes.
751 virtual unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
752 unsigned FirstStartColumn, bool DryRun) = 0;
753
754 protected:
755 /// If the \p State's next token is an r_brace closing a nested block,
756 /// format the nested block before it.
757 ///
758 /// Returns \c true if all children could be placed successfully and adapts
759 /// \p Penalty as well as \p State. If \p DryRun is false, also directly
760 /// creates changes using \c Whitespaces.
761 ///
762 /// The crucial idea here is that children always get formatted upon
763 /// encountering the closing brace right after the nested block. Now, if we
764 /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is
765 /// \c false), the entire block has to be kept on the same line (which is only
766 /// possible if it fits on the line, only contains a single statement, etc.
767 ///
768 /// If \p NewLine is true, we format the nested block on separate lines, i.e.
769 /// break after the "{", format all lines with correct indentation and the put
770 /// the closing "}" on yet another new line.
771 ///
772 /// This enables us to keep the simple structure of the
773 /// \c UnwrappedLineFormatter, where we only have two options for each token:
774 /// break or don't break.
formatChildren(LineState & State,bool NewLine,bool DryRun,unsigned & Penalty)775 bool formatChildren(LineState &State, bool NewLine, bool DryRun,
776 unsigned &Penalty) {
777 const FormatToken *LBrace = State.NextToken->getPreviousNonComment();
778 FormatToken &Previous = *State.NextToken->Previous;
779 if (!LBrace || LBrace->isNot(tok::l_brace) ||
780 LBrace->BlockKind != BK_Block || Previous.Children.size() == 0)
781 // The previous token does not open a block. Nothing to do. We don't
782 // assert so that we can simply call this function for all tokens.
783 return true;
784
785 if (NewLine) {
786 int AdditionalIndent = State.Stack.back().Indent -
787 Previous.Children[0]->Level * Style.IndentWidth;
788
789 Penalty +=
790 BlockFormatter->format(Previous.Children, DryRun, AdditionalIndent,
791 /*FixBadIndentation=*/true);
792 return true;
793 }
794
795 if (Previous.Children[0]->First->MustBreakBefore)
796 return false;
797
798 // Cannot merge into one line if this line ends on a comment.
799 if (Previous.is(tok::comment))
800 return false;
801
802 // Cannot merge multiple statements into a single line.
803 if (Previous.Children.size() > 1)
804 return false;
805
806 const AnnotatedLine *Child = Previous.Children[0];
807 // We can't put the closing "}" on a line with a trailing comment.
808 if (Child->Last->isTrailingComment())
809 return false;
810
811 // If the child line exceeds the column limit, we wouldn't want to merge it.
812 // We add +2 for the trailing " }".
813 if (Style.ColumnLimit > 0 &&
814 Child->Last->TotalLength + State.Column + 2 > Style.ColumnLimit)
815 return false;
816
817 if (!DryRun) {
818 Whitespaces->replaceWhitespace(
819 *Child->First, /*Newlines=*/0, /*Spaces=*/1,
820 /*StartOfTokenColumn=*/State.Column, State.Line->InPPDirective);
821 }
822 Penalty +=
823 formatLine(*Child, State.Column + 1, /*FirstStartColumn=*/0, DryRun);
824
825 State.Column += 1 + Child->Last->TotalLength;
826 return true;
827 }
828
829 ContinuationIndenter *Indenter;
830
831 private:
832 WhitespaceManager *Whitespaces;
833 const FormatStyle &Style;
834 UnwrappedLineFormatter *BlockFormatter;
835 };
836
837 /// Formatter that keeps the existing line breaks.
838 class NoColumnLimitLineFormatter : public LineFormatter {
839 public:
NoColumnLimitLineFormatter(ContinuationIndenter * Indenter,WhitespaceManager * Whitespaces,const FormatStyle & Style,UnwrappedLineFormatter * BlockFormatter)840 NoColumnLimitLineFormatter(ContinuationIndenter *Indenter,
841 WhitespaceManager *Whitespaces,
842 const FormatStyle &Style,
843 UnwrappedLineFormatter *BlockFormatter)
844 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
845
846 /// Formats the line, simply keeping all of the input's line breaking
847 /// decisions.
formatLine(const AnnotatedLine & Line,unsigned FirstIndent,unsigned FirstStartColumn,bool DryRun)848 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
849 unsigned FirstStartColumn, bool DryRun) override {
850 assert(!DryRun);
851 LineState State = Indenter->getInitialState(FirstIndent, FirstStartColumn,
852 &Line, /*DryRun=*/false);
853 while (State.NextToken) {
854 bool Newline =
855 Indenter->mustBreak(State) ||
856 (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0);
857 unsigned Penalty = 0;
858 formatChildren(State, Newline, /*DryRun=*/false, Penalty);
859 Indenter->addTokenToState(State, Newline, /*DryRun=*/false);
860 }
861 return 0;
862 }
863 };
864
865 /// Formatter that puts all tokens into a single line without breaks.
866 class NoLineBreakFormatter : public LineFormatter {
867 public:
NoLineBreakFormatter(ContinuationIndenter * Indenter,WhitespaceManager * Whitespaces,const FormatStyle & Style,UnwrappedLineFormatter * BlockFormatter)868 NoLineBreakFormatter(ContinuationIndenter *Indenter,
869 WhitespaceManager *Whitespaces, const FormatStyle &Style,
870 UnwrappedLineFormatter *BlockFormatter)
871 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
872
873 /// Puts all tokens into a single line.
formatLine(const AnnotatedLine & Line,unsigned FirstIndent,unsigned FirstStartColumn,bool DryRun)874 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
875 unsigned FirstStartColumn, bool DryRun) override {
876 unsigned Penalty = 0;
877 LineState State =
878 Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun);
879 while (State.NextToken) {
880 formatChildren(State, /*NewLine=*/false, DryRun, Penalty);
881 Indenter->addTokenToState(
882 State, /*Newline=*/State.NextToken->MustBreakBefore, DryRun);
883 }
884 return Penalty;
885 }
886 };
887
888 /// Finds the best way to break lines.
889 class OptimizingLineFormatter : public LineFormatter {
890 public:
OptimizingLineFormatter(ContinuationIndenter * Indenter,WhitespaceManager * Whitespaces,const FormatStyle & Style,UnwrappedLineFormatter * BlockFormatter)891 OptimizingLineFormatter(ContinuationIndenter *Indenter,
892 WhitespaceManager *Whitespaces,
893 const FormatStyle &Style,
894 UnwrappedLineFormatter *BlockFormatter)
895 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
896
897 /// Formats the line by finding the best line breaks with line lengths
898 /// below the column limit.
formatLine(const AnnotatedLine & Line,unsigned FirstIndent,unsigned FirstStartColumn,bool DryRun)899 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
900 unsigned FirstStartColumn, bool DryRun) override {
901 LineState State =
902 Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun);
903
904 // If the ObjC method declaration does not fit on a line, we should format
905 // it with one arg per line.
906 if (State.Line->Type == LT_ObjCMethodDecl)
907 State.Stack.back().BreakBeforeParameter = true;
908
909 // Find best solution in solution space.
910 return analyzeSolutionSpace(State, DryRun);
911 }
912
913 private:
914 struct CompareLineStatePointers {
operator ()clang::format::__anon1f81125b0111::OptimizingLineFormatter::CompareLineStatePointers915 bool operator()(LineState *obj1, LineState *obj2) const {
916 return *obj1 < *obj2;
917 }
918 };
919
920 /// A pair of <penalty, count> that is used to prioritize the BFS on.
921 ///
922 /// In case of equal penalties, we want to prefer states that were inserted
923 /// first. During state generation we make sure that we insert states first
924 /// that break the line as late as possible.
925 typedef std::pair<unsigned, unsigned> OrderedPenalty;
926
927 /// An edge in the solution space from \c Previous->State to \c State,
928 /// inserting a newline dependent on the \c NewLine.
929 struct StateNode {
StateNodeclang::format::__anon1f81125b0111::OptimizingLineFormatter::StateNode930 StateNode(const LineState &State, bool NewLine, StateNode *Previous)
931 : State(State), NewLine(NewLine), Previous(Previous) {}
932 LineState State;
933 bool NewLine;
934 StateNode *Previous;
935 };
936
937 /// An item in the prioritized BFS search queue. The \c StateNode's
938 /// \c State has the given \c OrderedPenalty.
939 typedef std::pair<OrderedPenalty, StateNode *> QueueItem;
940
941 /// The BFS queue type.
942 typedef std::priority_queue<QueueItem, std::vector<QueueItem>,
943 std::greater<QueueItem>>
944 QueueType;
945
946 /// Analyze the entire solution space starting from \p InitialState.
947 ///
948 /// This implements a variant of Dijkstra's algorithm on the graph that spans
949 /// the solution space (\c LineStates are the nodes). The algorithm tries to
950 /// find the shortest path (the one with lowest penalty) from \p InitialState
951 /// to a state where all tokens are placed. Returns the penalty.
952 ///
953 /// If \p DryRun is \c false, directly applies the changes.
analyzeSolutionSpace(LineState & InitialState,bool DryRun)954 unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun) {
955 std::set<LineState *, CompareLineStatePointers> Seen;
956
957 // Increasing count of \c StateNode items we have created. This is used to
958 // create a deterministic order independent of the container.
959 unsigned Count = 0;
960 QueueType Queue;
961
962 // Insert start element into queue.
963 StateNode *Node =
964 new (Allocator.Allocate()) StateNode(InitialState, false, nullptr);
965 Queue.push(QueueItem(OrderedPenalty(0, Count), Node));
966 ++Count;
967
968 unsigned Penalty = 0;
969
970 // While not empty, take first element and follow edges.
971 while (!Queue.empty()) {
972 Penalty = Queue.top().first.first;
973 StateNode *Node = Queue.top().second;
974 if (!Node->State.NextToken) {
975 LLVM_DEBUG(llvm::dbgs()
976 << "\n---\nPenalty for line: " << Penalty << "\n");
977 break;
978 }
979 Queue.pop();
980
981 // Cut off the analysis of certain solutions if the analysis gets too
982 // complex. See description of IgnoreStackForComparison.
983 if (Count > 50000)
984 Node->State.IgnoreStackForComparison = true;
985
986 if (!Seen.insert(&Node->State).second)
987 // State already examined with lower penalty.
988 continue;
989
990 FormatDecision LastFormat = Node->State.NextToken->Decision;
991 if (LastFormat == FD_Unformatted || LastFormat == FD_Continue)
992 addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue);
993 if (LastFormat == FD_Unformatted || LastFormat == FD_Break)
994 addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue);
995 }
996
997 if (Queue.empty()) {
998 // We were unable to find a solution, do nothing.
999 // FIXME: Add diagnostic?
1000 LLVM_DEBUG(llvm::dbgs() << "Could not find a solution.\n");
1001 return 0;
1002 }
1003
1004 // Reconstruct the solution.
1005 if (!DryRun)
1006 reconstructPath(InitialState, Queue.top().second);
1007
1008 LLVM_DEBUG(llvm::dbgs()
1009 << "Total number of analyzed states: " << Count << "\n");
1010 LLVM_DEBUG(llvm::dbgs() << "---\n");
1011
1012 return Penalty;
1013 }
1014
1015 /// Add the following state to the analysis queue \c Queue.
1016 ///
1017 /// Assume the current state is \p PreviousNode and has been reached with a
1018 /// 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)1019 void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode,
1020 bool NewLine, unsigned *Count, QueueType *Queue) {
1021 if (NewLine && !Indenter->canBreak(PreviousNode->State))
1022 return;
1023 if (!NewLine && Indenter->mustBreak(PreviousNode->State))
1024 return;
1025
1026 StateNode *Node = new (Allocator.Allocate())
1027 StateNode(PreviousNode->State, NewLine, PreviousNode);
1028 if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty))
1029 return;
1030
1031 Penalty += Indenter->addTokenToState(Node->State, NewLine, true);
1032
1033 Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node));
1034 ++(*Count);
1035 }
1036
1037 /// Applies the best formatting by reconstructing the path in the
1038 /// solution space that leads to \c Best.
reconstructPath(LineState & State,StateNode * Best)1039 void reconstructPath(LineState &State, StateNode *Best) {
1040 std::deque<StateNode *> Path;
1041 // We do not need a break before the initial token.
1042 while (Best->Previous) {
1043 Path.push_front(Best);
1044 Best = Best->Previous;
1045 }
1046 for (auto I = Path.begin(), E = Path.end(); I != E; ++I) {
1047 unsigned Penalty = 0;
1048 formatChildren(State, (*I)->NewLine, /*DryRun=*/false, Penalty);
1049 Penalty += Indenter->addTokenToState(State, (*I)->NewLine, false);
1050
1051 LLVM_DEBUG({
1052 printLineState((*I)->Previous->State);
1053 if ((*I)->NewLine) {
1054 llvm::dbgs() << "Penalty for placing "
1055 << (*I)->Previous->State.NextToken->Tok.getName()
1056 << " on a new line: " << Penalty << "\n";
1057 }
1058 });
1059 }
1060 }
1061
1062 llvm::SpecificBumpPtrAllocator<StateNode> Allocator;
1063 };
1064
1065 } // anonymous namespace
1066
format(const SmallVectorImpl<AnnotatedLine * > & Lines,bool DryRun,int AdditionalIndent,bool FixBadIndentation,unsigned FirstStartColumn,unsigned NextStartColumn,unsigned LastStartColumn)1067 unsigned UnwrappedLineFormatter::format(
1068 const SmallVectorImpl<AnnotatedLine *> &Lines, bool DryRun,
1069 int AdditionalIndent, bool FixBadIndentation, unsigned FirstStartColumn,
1070 unsigned NextStartColumn, unsigned LastStartColumn) {
1071 LineJoiner Joiner(Style, Keywords, Lines);
1072
1073 // Try to look up already computed penalty in DryRun-mode.
1074 std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned> CacheKey(
1075 &Lines, AdditionalIndent);
1076 auto CacheIt = PenaltyCache.find(CacheKey);
1077 if (DryRun && CacheIt != PenaltyCache.end())
1078 return CacheIt->second;
1079
1080 assert(!Lines.empty());
1081 unsigned Penalty = 0;
1082 LevelIndentTracker IndentTracker(Style, Keywords, Lines[0]->Level,
1083 AdditionalIndent);
1084 const AnnotatedLine *PreviousLine = nullptr;
1085 const AnnotatedLine *NextLine = nullptr;
1086
1087 // The minimum level of consecutive lines that have been formatted.
1088 unsigned RangeMinLevel = UINT_MAX;
1089
1090 bool FirstLine = true;
1091 for (const AnnotatedLine *Line =
1092 Joiner.getNextMergedLine(DryRun, IndentTracker);
1093 Line; Line = NextLine, FirstLine = false) {
1094 const AnnotatedLine &TheLine = *Line;
1095 unsigned Indent = IndentTracker.getIndent();
1096
1097 // We continue formatting unchanged lines to adjust their indent, e.g. if a
1098 // scope was added. However, we need to carefully stop doing this when we
1099 // exit the scope of affected lines to prevent indenting a the entire
1100 // remaining file if it currently missing a closing brace.
1101 bool PreviousRBrace =
1102 PreviousLine && PreviousLine->startsWith(tok::r_brace);
1103 bool ContinueFormatting =
1104 TheLine.Level > RangeMinLevel ||
1105 (TheLine.Level == RangeMinLevel && !PreviousRBrace &&
1106 !TheLine.startsWith(tok::r_brace));
1107
1108 bool FixIndentation = (FixBadIndentation || ContinueFormatting) &&
1109 Indent != TheLine.First->OriginalColumn;
1110 bool ShouldFormat = TheLine.Affected || FixIndentation;
1111 // We cannot format this line; if the reason is that the line had a
1112 // parsing error, remember that.
1113 if (ShouldFormat && TheLine.Type == LT_Invalid && Status) {
1114 Status->FormatComplete = false;
1115 Status->Line =
1116 SourceMgr.getSpellingLineNumber(TheLine.First->Tok.getLocation());
1117 }
1118
1119 if (ShouldFormat && TheLine.Type != LT_Invalid) {
1120 if (!DryRun) {
1121 bool LastLine = Line->First->is(tok::eof);
1122 formatFirstToken(TheLine, PreviousLine, Lines, Indent,
1123 LastLine ? LastStartColumn : NextStartColumn + Indent);
1124 }
1125
1126 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
1127 unsigned ColumnLimit = getColumnLimit(TheLine.InPPDirective, NextLine);
1128 bool FitsIntoOneLine =
1129 TheLine.Last->TotalLength + Indent <= ColumnLimit ||
1130 (TheLine.Type == LT_ImportStatement &&
1131 (Style.Language != FormatStyle::LK_JavaScript ||
1132 !Style.JavaScriptWrapImports)) ||
1133 (Style.isCSharp() &&
1134 TheLine.InPPDirective); // don't split #regions in C#
1135 if (Style.ColumnLimit == 0)
1136 NoColumnLimitLineFormatter(Indenter, Whitespaces, Style, this)
1137 .formatLine(TheLine, NextStartColumn + Indent,
1138 FirstLine ? FirstStartColumn : 0, DryRun);
1139 else if (FitsIntoOneLine)
1140 Penalty += NoLineBreakFormatter(Indenter, Whitespaces, Style, this)
1141 .formatLine(TheLine, NextStartColumn + Indent,
1142 FirstLine ? FirstStartColumn : 0, DryRun);
1143 else
1144 Penalty += OptimizingLineFormatter(Indenter, Whitespaces, Style, this)
1145 .formatLine(TheLine, NextStartColumn + Indent,
1146 FirstLine ? FirstStartColumn : 0, DryRun);
1147 RangeMinLevel = std::min(RangeMinLevel, TheLine.Level);
1148 } else {
1149 // If no token in the current line is affected, we still need to format
1150 // affected children.
1151 if (TheLine.ChildrenAffected)
1152 for (const FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next)
1153 if (!Tok->Children.empty())
1154 format(Tok->Children, DryRun);
1155
1156 // Adapt following lines on the current indent level to the same level
1157 // unless the current \c AnnotatedLine is not at the beginning of a line.
1158 bool StartsNewLine =
1159 TheLine.First->NewlinesBefore > 0 || TheLine.First->IsFirst;
1160 if (StartsNewLine)
1161 IndentTracker.adjustToUnmodifiedLine(TheLine);
1162 if (!DryRun) {
1163 bool ReformatLeadingWhitespace =
1164 StartsNewLine && ((PreviousLine && PreviousLine->Affected) ||
1165 TheLine.LeadingEmptyLinesAffected);
1166 // Format the first token.
1167 if (ReformatLeadingWhitespace)
1168 formatFirstToken(TheLine, PreviousLine, Lines,
1169 TheLine.First->OriginalColumn,
1170 TheLine.First->OriginalColumn);
1171 else
1172 Whitespaces->addUntouchableToken(*TheLine.First,
1173 TheLine.InPPDirective);
1174
1175 // Notify the WhitespaceManager about the unchanged whitespace.
1176 for (FormatToken *Tok = TheLine.First->Next; Tok; Tok = Tok->Next)
1177 Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective);
1178 }
1179 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
1180 RangeMinLevel = UINT_MAX;
1181 }
1182 if (!DryRun)
1183 markFinalized(TheLine.First);
1184 PreviousLine = &TheLine;
1185 }
1186 PenaltyCache[CacheKey] = Penalty;
1187 return Penalty;
1188 }
1189
formatFirstToken(const AnnotatedLine & Line,const AnnotatedLine * PreviousLine,const SmallVectorImpl<AnnotatedLine * > & Lines,unsigned Indent,unsigned NewlineIndent)1190 void UnwrappedLineFormatter::formatFirstToken(
1191 const AnnotatedLine &Line, const AnnotatedLine *PreviousLine,
1192 const SmallVectorImpl<AnnotatedLine *> &Lines, unsigned Indent,
1193 unsigned NewlineIndent) {
1194 FormatToken &RootToken = *Line.First;
1195 if (RootToken.is(tok::eof)) {
1196 unsigned Newlines = std::min(RootToken.NewlinesBefore, 1u);
1197 unsigned TokenIndent = Newlines ? NewlineIndent : 0;
1198 Whitespaces->replaceWhitespace(RootToken, Newlines, TokenIndent,
1199 TokenIndent);
1200 return;
1201 }
1202 unsigned Newlines =
1203 std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1);
1204 // Remove empty lines before "}" where applicable.
1205 if (RootToken.is(tok::r_brace) &&
1206 (!RootToken.Next ||
1207 (RootToken.Next->is(tok::semi) && !RootToken.Next->Next)) &&
1208 // Do not remove empty lines before namespace closing "}".
1209 !getNamespaceToken(&Line, Lines))
1210 Newlines = std::min(Newlines, 1u);
1211 // Remove empty lines at the start of nested blocks (lambdas/arrow functions)
1212 if (PreviousLine == nullptr && Line.Level > 0)
1213 Newlines = std::min(Newlines, 1u);
1214 if (Newlines == 0 && !RootToken.IsFirst)
1215 Newlines = 1;
1216 if (RootToken.IsFirst && !RootToken.HasUnescapedNewline)
1217 Newlines = 0;
1218
1219 // Remove empty lines after "{".
1220 if (!Style.KeepEmptyLinesAtTheStartOfBlocks && PreviousLine &&
1221 PreviousLine->Last->is(tok::l_brace) &&
1222 !PreviousLine->startsWithNamespace() &&
1223 !startsExternCBlock(*PreviousLine))
1224 Newlines = 1;
1225
1226 // Insert extra new line before access specifiers.
1227 if (PreviousLine && PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) &&
1228 RootToken.isAccessSpecifier() && RootToken.NewlinesBefore == 1)
1229 ++Newlines;
1230
1231 // Remove empty lines after access specifiers.
1232 if (PreviousLine && PreviousLine->First->isAccessSpecifier() &&
1233 (!PreviousLine->InPPDirective || !RootToken.HasUnescapedNewline))
1234 Newlines = std::min(1u, Newlines);
1235
1236 if (Newlines)
1237 Indent = NewlineIndent;
1238
1239 // If in Whitemsmiths mode, indent start and end of blocks
1240 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) {
1241 if (RootToken.isOneOf(tok::l_brace, tok::r_brace, tok::kw_case))
1242 Indent += Style.IndentWidth;
1243 }
1244
1245 // Preprocessor directives get indented before the hash only if specified
1246 if (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash &&
1247 (Line.Type == LT_PreprocessorDirective ||
1248 Line.Type == LT_ImportStatement))
1249 Indent = 0;
1250
1251 Whitespaces->replaceWhitespace(RootToken, Newlines, Indent, Indent,
1252 Line.InPPDirective &&
1253 !RootToken.HasUnescapedNewline);
1254 }
1255
1256 unsigned
getColumnLimit(bool InPPDirective,const AnnotatedLine * NextLine) const1257 UnwrappedLineFormatter::getColumnLimit(bool InPPDirective,
1258 const AnnotatedLine *NextLine) const {
1259 // In preprocessor directives reserve two chars for trailing " \" if the
1260 // next line continues the preprocessor directive.
1261 bool ContinuesPPDirective =
1262 InPPDirective &&
1263 // If there is no next line, this is likely a child line and the parent
1264 // continues the preprocessor directive.
1265 (!NextLine ||
1266 (NextLine->InPPDirective &&
1267 // If there is an unescaped newline between this line and the next, the
1268 // next line starts a new preprocessor directive.
1269 !NextLine->First->HasUnescapedNewline));
1270 return Style.ColumnLimit - (ContinuesPPDirective ? 2 : 0);
1271 }
1272
1273 } // namespace format
1274 } // namespace clang
1275