1 //===--- UnwrappedLineFormatter.cpp - Format C++ code ---------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #include "UnwrappedLineFormatter.h"
11 #include "WhitespaceManager.h"
12 #include "llvm/Support/Debug.h"
15 #define DEBUG_TYPE "format-formatter"
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);
29 /// \brief Tracks the indent level of \c AnnotatedLines across levels.
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
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
38 class LevelIndentTracker {
40 LevelIndentTracker(const FormatStyle &Style,
41 const AdditionalKeywords &Keywords, unsigned StartLevel,
43 : Style(Style), Keywords(Keywords), AdditionalIndent(AdditionalIndent) {
44 for (unsigned i = 0; i != StartLevel; ++i)
45 IndentForLevel.push_back(Style.IndentWidth * i + AdditionalIndent);
48 /// \brief Returns the indent for the current line.
49 unsigned getIndent() const { return Indent; }
51 /// \brief Update the indent state given that \p Line is going to be formatted
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;
62 IndentForLevel.resize(Line.Level + 1);
63 Indent = getIndent(IndentForLevel, Line.Level);
65 if (static_cast<int>(Indent) + Offset >= 0)
69 /// \brief Update the indent state given that \p Line indent should be
71 void skipLine(const AnnotatedLine &Line) {
72 while (IndentForLevel.size() <= Line.Level)
73 IndentForLevel.push_back(Indent);
76 /// \brief Update the level indent to adapt to the given \p Line.
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.
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) &&
87 IndentForLevel[Line.Level] = LevelIndent;
91 /// \brief Get the offset of the line relatively to the level.
93 /// For example, 'public:' labels in classes are offset by 1 or 2
94 /// characters to the left from their level.
95 int getIndentOffset(const FormatToken &RootToken) {
96 if (Style.Language == FormatStyle::LK_Java ||
97 Style.Language == FormatStyle::LK_JavaScript)
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;
107 /// \brief Get the indent of \p Level from \p IndentForLevel.
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.
112 unsigned getIndent(ArrayRef<int> IndentForLevel, unsigned Level) {
113 if (IndentForLevel[Level] != -1)
114 return IndentForLevel[Level];
117 return getIndent(IndentForLevel, Level - 1) + Style.IndentWidth;
120 const FormatStyle &Style;
121 const AdditionalKeywords &Keywords;
122 const unsigned AdditionalIndent;
124 /// \brief The indent in characters for each level.
125 std::vector<int> IndentForLevel;
127 /// \brief Offset of the current line relative to the indent level.
129 /// For example, the 'public' keywords is often indented with a negative
133 /// \brief The current line's indent.
137 bool isNamespaceDeclaration(const AnnotatedLine *Line) {
138 const FormatToken *NamespaceTok = Line->First;
139 return NamespaceTok && NamespaceTok->getNamespaceToken();
142 bool isEndOfNamespace(const AnnotatedLine *Line,
143 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
144 if (!Line->startsWith(tok::r_brace))
146 size_t StartLineIndex = Line->MatchingOpeningBlockLineIndex;
147 if (StartLineIndex == UnwrappedLine::kInvalidIndex)
149 assert(StartLineIndex < AnnotatedLines.size());
150 return isNamespaceDeclaration(AnnotatedLines[StartLineIndex]);
155 LineJoiner(const FormatStyle &Style, const AdditionalKeywords &Keywords,
156 const SmallVectorImpl<AnnotatedLine *> &Lines)
157 : Style(Style), Keywords(Keywords), End(Lines.end()), Next(Lines.begin()),
158 AnnotatedLines(Lines) {}
160 /// \brief Returns the next line, merging multiple lines into one if possible.
161 const AnnotatedLine *getNextMergedLine(bool DryRun,
162 LevelIndentTracker &IndentTracker) {
165 const AnnotatedLine *Current = *Next;
166 IndentTracker.nextLine(*Current);
167 unsigned MergedLines = tryFitMultipleLinesInOne(IndentTracker, Next, End);
168 if (MergedLines > 0 && Style.ColumnLimit == 0)
169 // Disallow line merging if there is a break at the start of one of the
171 for (unsigned i = 0; i < MergedLines; ++i)
172 if (Next[i + 1]->First->NewlinesBefore > 0)
175 for (unsigned i = 0; i < MergedLines; ++i)
176 join(*Next[0], *Next[i + 1]);
177 Next = Next + MergedLines + 1;
182 /// \brief Calculates how many lines can be merged into 1 starting at \p I.
184 tryFitMultipleLinesInOne(LevelIndentTracker &IndentTracker,
185 SmallVectorImpl<AnnotatedLine *>::const_iterator I,
186 SmallVectorImpl<AnnotatedLine *>::const_iterator E) {
187 const unsigned Indent = IndentTracker.getIndent();
189 // Can't join the last line with anything.
192 // We can never merge stuff if there are trailing line comments.
193 const AnnotatedLine *TheLine = *I;
194 if (TheLine->Last->is(TT_LineComment))
196 if (I[1]->Type == LT_Invalid || I[1]->First->MustBreakBefore)
198 if (TheLine->InPPDirective &&
199 (!I[1]->InPPDirective || I[1]->First->HasUnescapedNewline))
202 if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit)
206 Style.ColumnLimit == 0 ? UINT_MAX : Style.ColumnLimit - Indent;
207 // If we already exceed the column limit, we set 'Limit' to 0. The different
208 // tryMerge..() functions can then decide whether to still do merging.
209 Limit = TheLine->Last->TotalLength > Limit
211 : Limit - TheLine->Last->TotalLength;
213 if (TheLine->Last->is(TT_FunctionLBrace) &&
214 TheLine->First == TheLine->Last &&
215 !Style.BraceWrapping.SplitEmptyFunction &&
216 I[1]->First->is(tok::r_brace))
217 return tryMergeSimpleBlock(I, E, Limit);
219 // Handle empty record blocks where the brace has already been wrapped
220 if (TheLine->Last->is(tok::l_brace) && TheLine->First == TheLine->Last &&
221 I != AnnotatedLines.begin()) {
222 bool EmptyBlock = I[1]->First->is(tok::r_brace);
224 const FormatToken *Tok = I[-1]->First;
225 if (Tok && Tok->is(tok::comment))
226 Tok = Tok->getNextNonComment();
228 if (Tok && Tok->getNamespaceToken())
229 return !Style.BraceWrapping.SplitEmptyNamespace && EmptyBlock
230 ? tryMergeSimpleBlock(I, E, Limit)
233 if (Tok && Tok->is(tok::kw_typedef))
234 Tok = Tok->getNextNonComment();
235 if (Tok && Tok->isOneOf(tok::kw_class, tok::kw_struct, tok::kw_union,
236 tok::kw_extern, Keywords.kw_interface))
237 return !Style.BraceWrapping.SplitEmptyRecord && EmptyBlock
238 ? tryMergeSimpleBlock(I, E, Limit)
242 // FIXME: TheLine->Level != 0 might or might not be the right check to do.
243 // If necessary, change to something smarter.
244 bool MergeShortFunctions =
245 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_All ||
246 (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
247 I[1]->First->is(tok::r_brace)) ||
248 (Style.AllowShortFunctionsOnASingleLine & FormatStyle::SFS_InlineOnly &&
249 TheLine->Level != 0);
251 if (Style.CompactNamespaces) {
252 if (isNamespaceDeclaration(TheLine)) {
254 unsigned closingLine = TheLine->MatchingOpeningBlockLineIndex - 1;
255 for (; I + 1 + i != E && isNamespaceDeclaration(I[i + 1]) &&
256 closingLine == I[i + 1]->MatchingOpeningBlockLineIndex &&
257 I[i + 1]->Last->TotalLength < Limit;
258 i++, closingLine--) {
259 // No extra indent for compacted namespaces
260 IndentTracker.skipLine(*I[i + 1]);
262 Limit -= I[i + 1]->Last->TotalLength;
267 if (isEndOfNamespace(TheLine, AnnotatedLines)) {
269 unsigned openingLine = TheLine->MatchingOpeningBlockLineIndex - 1;
270 for (; I + 1 + i != E && isEndOfNamespace(I[i + 1], AnnotatedLines) &&
271 openingLine == I[i + 1]->MatchingOpeningBlockLineIndex;
272 i++, openingLine--) {
273 // No space between consecutive braces
274 I[i + 1]->First->SpacesRequiredBefore = !I[i]->Last->is(tok::r_brace);
276 // Indent like the outer-most namespace
277 IndentTracker.nextLine(*I[i + 1]);
283 // Try to merge a function block with left brace unwrapped
284 if (TheLine->Last->is(TT_FunctionLBrace) &&
285 TheLine->First != TheLine->Last) {
286 return MergeShortFunctions ? tryMergeSimpleBlock(I, E, Limit) : 0;
288 // Try to merge a control statement block with left brace unwrapped
289 if (TheLine->Last->is(tok::l_brace) && TheLine->First != TheLine->Last &&
290 TheLine->First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for)) {
291 return Style.AllowShortBlocksOnASingleLine
292 ? tryMergeSimpleBlock(I, E, Limit)
295 // Try to merge a control statement block with left brace wrapped
296 if (I[1]->First->is(tok::l_brace) &&
297 TheLine->First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for)) {
298 return Style.BraceWrapping.AfterControlStatement
299 ? tryMergeSimpleBlock(I, E, Limit)
302 // Try to merge either empty or one-line block if is precedeed by control
304 if (TheLine->First->is(tok::l_brace) && TheLine->First == TheLine->Last &&
305 I != AnnotatedLines.begin() &&
306 I[-1]->First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for)) {
307 return Style.AllowShortBlocksOnASingleLine
308 ? tryMergeSimpleBlock(I - 1, E, Limit)
311 // Try to merge a block with left brace wrapped that wasn't yet covered
312 if (TheLine->Last->is(tok::l_brace)) {
313 return !Style.BraceWrapping.AfterFunction ||
314 (I[1]->First->is(tok::r_brace) &&
315 !Style.BraceWrapping.SplitEmptyRecord)
316 ? tryMergeSimpleBlock(I, E, Limit)
319 // Try to merge a function block with left brace wrapped
320 if (I[1]->First->is(TT_FunctionLBrace) &&
321 Style.BraceWrapping.AfterFunction) {
322 if (I[1]->Last->is(TT_LineComment))
325 // Check for Limit <= 2 to account for the " {".
326 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(TheLine)))
330 unsigned MergedLines = 0;
331 if (MergeShortFunctions ||
332 (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
333 I[1]->First == I[1]->Last && I + 2 != E &&
334 I[2]->First->is(tok::r_brace))) {
335 MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
336 // If we managed to merge the block, count the function header, which is
337 // on a separate line.
343 if (TheLine->First->is(tok::kw_if)) {
344 return Style.AllowShortIfStatementsOnASingleLine
345 ? tryMergeSimpleControlStatement(I, E, Limit)
348 if (TheLine->First->isOneOf(tok::kw_for, tok::kw_while)) {
349 return Style.AllowShortLoopsOnASingleLine
350 ? tryMergeSimpleControlStatement(I, E, Limit)
353 if (TheLine->First->isOneOf(tok::kw_case, tok::kw_default)) {
354 return Style.AllowShortCaseLabelsOnASingleLine
355 ? tryMergeShortCaseLabels(I, E, Limit)
358 if (TheLine->InPPDirective &&
359 (TheLine->First->HasUnescapedNewline || TheLine->First->IsFirst)) {
360 return tryMergeSimplePPDirective(I, E, Limit);
366 tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
367 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
371 if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline)
373 if (1 + I[1]->Last->TotalLength > Limit)
378 unsigned tryMergeSimpleControlStatement(
379 SmallVectorImpl<AnnotatedLine *>::const_iterator I,
380 SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) {
383 if (Style.BraceWrapping.AfterControlStatement &&
384 (I[1]->First->is(tok::l_brace) && !Style.AllowShortBlocksOnASingleLine))
386 if (I[1]->InPPDirective != (*I)->InPPDirective ||
387 (I[1]->InPPDirective && I[1]->First->HasUnescapedNewline))
389 Limit = limitConsideringMacros(I + 1, E, Limit);
390 AnnotatedLine &Line = **I;
391 if (Line.Last->isNot(tok::r_paren))
393 if (1 + I[1]->Last->TotalLength > Limit)
395 if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for, tok::kw_while,
398 // Only inline simple if's (no nested if or else).
399 if (I + 2 != E && Line.startsWith(tok::kw_if) &&
400 I[2]->First->is(tok::kw_else))
406 tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
407 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
409 if (Limit == 0 || I + 1 == E ||
410 I[1]->First->isOneOf(tok::kw_case, tok::kw_default))
412 unsigned NumStmts = 0;
414 bool EndsWithComment = false;
415 bool InPPDirective = I[0]->InPPDirective;
416 const unsigned Level = I[0]->Level;
417 for (; NumStmts < 3; ++NumStmts) {
418 if (I + 1 + NumStmts == E)
420 const AnnotatedLine *Line = I[1 + NumStmts];
421 if (Line->InPPDirective != InPPDirective)
423 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
425 if (Line->First->isOneOf(tok::kw_if, tok::kw_for, tok::kw_switch,
429 if (Line->First->is(tok::comment)) {
430 if (Level != Line->Level)
432 SmallVectorImpl<AnnotatedLine *>::const_iterator J = I + 2 + NumStmts;
433 for (; J != E; ++J) {
435 if (Line->InPPDirective != InPPDirective)
437 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
439 if (Line->First->isNot(tok::comment) || Level != Line->Level)
444 if (Line->Last->is(tok::comment))
445 EndsWithComment = true;
446 Length += I[1 + NumStmts]->Last->TotalLength + 1; // 1 for the space.
448 if (NumStmts == 0 || NumStmts == 3 || Length > Limit)
454 tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
455 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
457 AnnotatedLine &Line = **I;
459 // Don't merge ObjC @ keywords and methods.
460 // FIXME: If an option to allow short exception handling clauses on a single
461 // line is added, change this to not return for @try and friends.
462 if (Style.Language != FormatStyle::LK_Java &&
463 Line.First->isOneOf(tok::at, tok::minus, tok::plus))
466 // Check that the current line allows merging. This depends on whether we
467 // are in a control flow statements as well as several style flags.
468 if (Line.First->isOneOf(tok::kw_else, tok::kw_case) ||
469 (Line.First->Next && Line.First->Next->is(tok::kw_else)))
471 if (Line.First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_do, tok::kw_try,
472 tok::kw___try, tok::kw_catch, tok::kw___finally,
473 tok::kw_for, tok::r_brace, Keywords.kw___except)) {
474 if (!Style.AllowShortBlocksOnASingleLine)
476 // Don't merge when we can't except the case when
477 // the control statement block is empty
478 if (!Style.AllowShortIfStatementsOnASingleLine &&
479 Line.startsWith(tok::kw_if) &&
480 !Style.BraceWrapping.AfterControlStatement &&
481 !I[1]->First->is(tok::r_brace))
483 if (!Style.AllowShortIfStatementsOnASingleLine &&
484 Line.startsWith(tok::kw_if) &&
485 Style.BraceWrapping.AfterControlStatement && I + 2 != E &&
486 !I[2]->First->is(tok::r_brace))
488 if (!Style.AllowShortLoopsOnASingleLine &&
489 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for) &&
490 !Style.BraceWrapping.AfterControlStatement &&
491 !I[1]->First->is(tok::r_brace))
493 if (!Style.AllowShortLoopsOnASingleLine &&
494 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for) &&
495 Style.BraceWrapping.AfterControlStatement && I + 2 != E &&
496 !I[2]->First->is(tok::r_brace))
498 // FIXME: Consider an option to allow short exception handling clauses on
500 // FIXME: This isn't covered by tests.
501 // FIXME: For catch, __except, __finally the first token on the line
502 // is '}', so this isn't correct here.
503 if (Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch,
504 Keywords.kw___except, tok::kw___finally))
508 if (Line.Last->is(tok::l_brace)) {
509 FormatToken *Tok = I[1]->First;
510 if (Tok->is(tok::r_brace) && !Tok->MustBreakBefore &&
511 (Tok->getNextNonComment() == nullptr ||
512 Tok->getNextNonComment()->is(tok::semi))) {
513 // We merge empty blocks even if the line exceeds the column limit.
514 Tok->SpacesRequiredBefore = 0;
515 Tok->CanBreakBefore = true;
517 } else if (Limit != 0 && !Line.startsWith(tok::kw_namespace) &&
518 !startsExternCBlock(Line)) {
519 // We don't merge short records.
520 FormatToken *RecordTok = Line.First;
521 // Skip record modifiers.
522 while (RecordTok->Next &&
523 RecordTok->isOneOf(tok::kw_typedef, tok::kw_export,
524 Keywords.kw_declare, Keywords.kw_abstract,
526 RecordTok = RecordTok->Next;
528 RecordTok->isOneOf(tok::kw_class, tok::kw_union, tok::kw_struct,
529 Keywords.kw_interface))
532 // Check that we still have three lines and they fit into the limit.
533 if (I + 2 == E || I[2]->Type == LT_Invalid)
535 Limit = limitConsideringMacros(I + 2, E, Limit);
537 if (!nextTwoLinesFitInto(I, Limit))
540 // Second, check that the next line does not contain any braces - if it
541 // does, readability declines when putting it into a single line.
542 if (I[1]->Last->is(TT_LineComment))
545 if (Tok->is(tok::l_brace) && Tok->BlockKind != BK_BracedInit)
550 // Last, check that the third line starts with a closing brace.
552 if (Tok->isNot(tok::r_brace))
555 // Don't merge "if (a) { .. } else {".
556 if (Tok->Next && Tok->Next->is(tok::kw_else))
561 } else if (I[1]->First->is(tok::l_brace)) {
562 if (I[1]->Last->is(TT_LineComment))
565 // Check for Limit <= 2 to account for the " {".
566 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(*I)))
569 unsigned MergedLines = 0;
570 if (Style.AllowShortBlocksOnASingleLine ||
571 (I[1]->First == I[1]->Last && I + 2 != E &&
572 I[2]->First->is(tok::r_brace))) {
573 MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
574 // If we managed to merge the block, count the statement header, which
575 // is on a separate line.
584 /// Returns the modified column limit for \p I if it is inside a macro and
585 /// needs a trailing '\'.
587 limitConsideringMacros(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
588 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
590 if (I[0]->InPPDirective && I + 1 != E &&
591 !I[1]->First->HasUnescapedNewline && !I[1]->First->is(tok::eof)) {
592 return Limit < 2 ? 0 : Limit - 2;
597 bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
599 if (I[1]->First->MustBreakBefore || I[2]->First->MustBreakBefore)
601 return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit;
604 bool containsMustBreak(const AnnotatedLine *Line) {
605 for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) {
606 if (Tok->MustBreakBefore)
612 void join(AnnotatedLine &A, const AnnotatedLine &B) {
613 assert(!A.Last->Next);
614 assert(!B.First->Previous);
617 A.Last->Next = B.First;
618 B.First->Previous = A.Last;
619 B.First->CanBreakBefore = true;
620 unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore;
621 for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) {
622 Tok->TotalLength += LengthA;
627 const FormatStyle &Style;
628 const AdditionalKeywords &Keywords;
629 const SmallVectorImpl<AnnotatedLine *>::const_iterator End;
631 SmallVectorImpl<AnnotatedLine *>::const_iterator Next;
632 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines;
635 static void markFinalized(FormatToken *Tok) {
636 for (; Tok; Tok = Tok->Next) {
637 Tok->Finalized = true;
638 for (AnnotatedLine *Child : Tok->Children)
639 markFinalized(Child->First);
644 static void printLineState(const LineState &State) {
645 llvm::dbgs() << "State: ";
646 for (const ParenState &P : State.Stack) {
647 llvm::dbgs() << P.Indent << "|" << P.LastSpace << "|" << P.NestedBlockIndent
650 llvm::dbgs() << State.NextToken->TokenText << "\n";
654 /// \brief Base class for classes that format one \c AnnotatedLine.
655 class LineFormatter {
657 LineFormatter(ContinuationIndenter *Indenter, WhitespaceManager *Whitespaces,
658 const FormatStyle &Style,
659 UnwrappedLineFormatter *BlockFormatter)
660 : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style),
661 BlockFormatter(BlockFormatter) {}
662 virtual ~LineFormatter() {}
664 /// \brief Formats an \c AnnotatedLine and returns the penalty.
666 /// If \p DryRun is \c false, directly applies the changes.
667 virtual unsigned formatLine(const AnnotatedLine &Line,
668 unsigned FirstIndent,
669 unsigned FirstStartColumn,
673 /// \brief If the \p State's next token is an r_brace closing a nested block,
674 /// format the nested block before it.
676 /// Returns \c true if all children could be placed successfully and adapts
677 /// \p Penalty as well as \p State. If \p DryRun is false, also directly
678 /// creates changes using \c Whitespaces.
680 /// The crucial idea here is that children always get formatted upon
681 /// encountering the closing brace right after the nested block. Now, if we
682 /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is
683 /// \c false), the entire block has to be kept on the same line (which is only
684 /// possible if it fits on the line, only contains a single statement, etc.
686 /// If \p NewLine is true, we format the nested block on separate lines, i.e.
687 /// break after the "{", format all lines with correct indentation and the put
688 /// the closing "}" on yet another new line.
690 /// This enables us to keep the simple structure of the
691 /// \c UnwrappedLineFormatter, where we only have two options for each token:
692 /// break or don't break.
693 bool formatChildren(LineState &State, bool NewLine, bool DryRun,
695 const FormatToken *LBrace = State.NextToken->getPreviousNonComment();
696 FormatToken &Previous = *State.NextToken->Previous;
697 if (!LBrace || LBrace->isNot(tok::l_brace) ||
698 LBrace->BlockKind != BK_Block || Previous.Children.size() == 0)
699 // The previous token does not open a block. Nothing to do. We don't
700 // assert so that we can simply call this function for all tokens.
704 int AdditionalIndent = State.Stack.back().Indent -
705 Previous.Children[0]->Level * Style.IndentWidth;
708 BlockFormatter->format(Previous.Children, DryRun, AdditionalIndent,
709 /*FixBadIndentation=*/true);
713 if (Previous.Children[0]->First->MustBreakBefore)
716 // Cannot merge into one line if this line ends on a comment.
717 if (Previous.is(tok::comment))
720 // Cannot merge multiple statements into a single line.
721 if (Previous.Children.size() > 1)
724 const AnnotatedLine *Child = Previous.Children[0];
725 // We can't put the closing "}" on a line with a trailing comment.
726 if (Child->Last->isTrailingComment())
729 // If the child line exceeds the column limit, we wouldn't want to merge it.
730 // We add +2 for the trailing " }".
731 if (Style.ColumnLimit > 0 &&
732 Child->Last->TotalLength + State.Column + 2 > Style.ColumnLimit)
736 Whitespaces->replaceWhitespace(
737 *Child->First, /*Newlines=*/0, /*Spaces=*/1,
738 /*StartOfTokenColumn=*/State.Column, State.Line->InPPDirective);
741 formatLine(*Child, State.Column + 1, /*FirstStartColumn=*/0, DryRun);
743 State.Column += 1 + Child->Last->TotalLength;
747 ContinuationIndenter *Indenter;
750 WhitespaceManager *Whitespaces;
751 const FormatStyle &Style;
752 UnwrappedLineFormatter *BlockFormatter;
755 /// \brief Formatter that keeps the existing line breaks.
756 class NoColumnLimitLineFormatter : public LineFormatter {
758 NoColumnLimitLineFormatter(ContinuationIndenter *Indenter,
759 WhitespaceManager *Whitespaces,
760 const FormatStyle &Style,
761 UnwrappedLineFormatter *BlockFormatter)
762 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
764 /// \brief Formats the line, simply keeping all of the input's line breaking
766 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
767 unsigned FirstStartColumn, bool DryRun) override {
769 LineState State = Indenter->getInitialState(FirstIndent, FirstStartColumn,
770 &Line, /*DryRun=*/false);
771 while (State.NextToken) {
773 Indenter->mustBreak(State) ||
774 (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0);
775 unsigned Penalty = 0;
776 formatChildren(State, Newline, /*DryRun=*/false, Penalty);
777 Indenter->addTokenToState(State, Newline, /*DryRun=*/false);
783 /// \brief Formatter that puts all tokens into a single line without breaks.
784 class NoLineBreakFormatter : public LineFormatter {
786 NoLineBreakFormatter(ContinuationIndenter *Indenter,
787 WhitespaceManager *Whitespaces, const FormatStyle &Style,
788 UnwrappedLineFormatter *BlockFormatter)
789 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
791 /// \brief Puts all tokens into a single line.
792 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
793 unsigned FirstStartColumn, bool DryRun) override {
794 unsigned Penalty = 0;
796 Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun);
797 while (State.NextToken) {
798 formatChildren(State, /*Newline=*/false, DryRun, Penalty);
799 Indenter->addTokenToState(
800 State, /*Newline=*/State.NextToken->MustBreakBefore, DryRun);
806 /// \brief Finds the best way to break lines.
807 class OptimizingLineFormatter : public LineFormatter {
809 OptimizingLineFormatter(ContinuationIndenter *Indenter,
810 WhitespaceManager *Whitespaces,
811 const FormatStyle &Style,
812 UnwrappedLineFormatter *BlockFormatter)
813 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
815 /// \brief Formats the line by finding the best line breaks with line lengths
816 /// below the column limit.
817 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
818 unsigned FirstStartColumn, bool DryRun) override {
820 Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun);
822 // If the ObjC method declaration does not fit on a line, we should format
823 // it with one arg per line.
824 if (State.Line->Type == LT_ObjCMethodDecl)
825 State.Stack.back().BreakBeforeParameter = true;
827 // Find best solution in solution space.
828 return analyzeSolutionSpace(State, DryRun);
832 struct CompareLineStatePointers {
833 bool operator()(LineState *obj1, LineState *obj2) const {
834 return *obj1 < *obj2;
838 /// \brief A pair of <penalty, count> that is used to prioritize the BFS on.
840 /// In case of equal penalties, we want to prefer states that were inserted
841 /// first. During state generation we make sure that we insert states first
842 /// that break the line as late as possible.
843 typedef std::pair<unsigned, unsigned> OrderedPenalty;
845 /// \brief An edge in the solution space from \c Previous->State to \c State,
846 /// inserting a newline dependent on the \c NewLine.
848 StateNode(const LineState &State, bool NewLine, StateNode *Previous)
849 : State(State), NewLine(NewLine), Previous(Previous) {}
855 /// \brief An item in the prioritized BFS search queue. The \c StateNode's
856 /// \c State has the given \c OrderedPenalty.
857 typedef std::pair<OrderedPenalty, StateNode *> QueueItem;
859 /// \brief The BFS queue type.
860 typedef std::priority_queue<QueueItem, std::vector<QueueItem>,
861 std::greater<QueueItem>>
864 /// \brief Analyze the entire solution space starting from \p InitialState.
866 /// This implements a variant of Dijkstra's algorithm on the graph that spans
867 /// the solution space (\c LineStates are the nodes). The algorithm tries to
868 /// find the shortest path (the one with lowest penalty) from \p InitialState
869 /// to a state where all tokens are placed. Returns the penalty.
871 /// If \p DryRun is \c false, directly applies the changes.
872 unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun) {
873 std::set<LineState *, CompareLineStatePointers> Seen;
875 // Increasing count of \c StateNode items we have created. This is used to
876 // create a deterministic order independent of the container.
880 // Insert start element into queue.
882 new (Allocator.Allocate()) StateNode(InitialState, false, nullptr);
883 Queue.push(QueueItem(OrderedPenalty(0, Count), Node));
886 unsigned Penalty = 0;
888 // While not empty, take first element and follow edges.
889 while (!Queue.empty()) {
890 Penalty = Queue.top().first.first;
891 StateNode *Node = Queue.top().second;
892 if (!Node->State.NextToken) {
893 DEBUG(llvm::dbgs() << "\n---\nPenalty for line: " << Penalty << "\n");
898 // Cut off the analysis of certain solutions if the analysis gets too
899 // complex. See description of IgnoreStackForComparison.
901 Node->State.IgnoreStackForComparison = true;
903 if (!Seen.insert(&Node->State).second)
904 // State already examined with lower penalty.
907 FormatDecision LastFormat = Node->State.NextToken->Decision;
908 if (LastFormat == FD_Unformatted || LastFormat == FD_Continue)
909 addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue);
910 if (LastFormat == FD_Unformatted || LastFormat == FD_Break)
911 addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue);
915 // We were unable to find a solution, do nothing.
916 // FIXME: Add diagnostic?
917 DEBUG(llvm::dbgs() << "Could not find a solution.\n");
921 // Reconstruct the solution.
923 reconstructPath(InitialState, Queue.top().second);
925 DEBUG(llvm::dbgs() << "Total number of analyzed states: " << Count << "\n");
926 DEBUG(llvm::dbgs() << "---\n");
931 /// \brief Add the following state to the analysis queue \c Queue.
933 /// Assume the current state is \p PreviousNode and has been reached with a
934 /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true.
935 void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode,
936 bool NewLine, unsigned *Count, QueueType *Queue) {
937 if (NewLine && !Indenter->canBreak(PreviousNode->State))
939 if (!NewLine && Indenter->mustBreak(PreviousNode->State))
942 StateNode *Node = new (Allocator.Allocate())
943 StateNode(PreviousNode->State, NewLine, PreviousNode);
944 if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty))
947 Penalty += Indenter->addTokenToState(Node->State, NewLine, true);
949 Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node));
953 /// \brief Applies the best formatting by reconstructing the path in the
954 /// solution space that leads to \c Best.
955 void reconstructPath(LineState &State, StateNode *Best) {
956 std::deque<StateNode *> Path;
957 // We do not need a break before the initial token.
958 while (Best->Previous) {
959 Path.push_front(Best);
960 Best = Best->Previous;
962 for (std::deque<StateNode *>::iterator I = Path.begin(), E = Path.end();
964 unsigned Penalty = 0;
965 formatChildren(State, (*I)->NewLine, /*DryRun=*/false, Penalty);
966 Penalty += Indenter->addTokenToState(State, (*I)->NewLine, false);
969 printLineState((*I)->Previous->State);
971 llvm::dbgs() << "Penalty for placing "
972 << (*I)->Previous->State.NextToken->Tok.getName() << ": "
979 llvm::SpecificBumpPtrAllocator<StateNode> Allocator;
982 } // anonymous namespace
985 UnwrappedLineFormatter::format(const SmallVectorImpl<AnnotatedLine *> &Lines,
986 bool DryRun, int AdditionalIndent,
987 bool FixBadIndentation,
988 unsigned FirstStartColumn,
989 unsigned NextStartColumn,
990 unsigned LastStartColumn) {
991 LineJoiner Joiner(Style, Keywords, Lines);
993 // Try to look up already computed penalty in DryRun-mode.
994 std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned> CacheKey(
995 &Lines, AdditionalIndent);
996 auto CacheIt = PenaltyCache.find(CacheKey);
997 if (DryRun && CacheIt != PenaltyCache.end())
998 return CacheIt->second;
1000 assert(!Lines.empty());
1001 unsigned Penalty = 0;
1002 LevelIndentTracker IndentTracker(Style, Keywords, Lines[0]->Level,
1004 const AnnotatedLine *PreviousLine = nullptr;
1005 const AnnotatedLine *NextLine = nullptr;
1007 // The minimum level of consecutive lines that have been formatted.
1008 unsigned RangeMinLevel = UINT_MAX;
1010 bool FirstLine = true;
1011 for (const AnnotatedLine *Line =
1012 Joiner.getNextMergedLine(DryRun, IndentTracker);
1013 Line; Line = NextLine, FirstLine = false) {
1014 const AnnotatedLine &TheLine = *Line;
1015 unsigned Indent = IndentTracker.getIndent();
1017 // We continue formatting unchanged lines to adjust their indent, e.g. if a
1018 // scope was added. However, we need to carefully stop doing this when we
1019 // exit the scope of affected lines to prevent indenting a the entire
1020 // remaining file if it currently missing a closing brace.
1021 bool ContinueFormatting =
1022 TheLine.Level > RangeMinLevel ||
1023 (TheLine.Level == RangeMinLevel && !TheLine.startsWith(tok::r_brace));
1025 bool FixIndentation = (FixBadIndentation || ContinueFormatting) &&
1026 Indent != TheLine.First->OriginalColumn;
1027 bool ShouldFormat = TheLine.Affected || FixIndentation;
1028 // We cannot format this line; if the reason is that the line had a
1029 // parsing error, remember that.
1030 if (ShouldFormat && TheLine.Type == LT_Invalid && Status) {
1031 Status->FormatComplete = false;
1033 SourceMgr.getSpellingLineNumber(TheLine.First->Tok.getLocation());
1036 if (ShouldFormat && TheLine.Type != LT_Invalid) {
1038 bool LastLine = Line->First->is(tok::eof);
1039 formatFirstToken(TheLine, PreviousLine,
1041 LastLine ? LastStartColumn : NextStartColumn + Indent);
1044 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
1045 unsigned ColumnLimit = getColumnLimit(TheLine.InPPDirective, NextLine);
1046 bool FitsIntoOneLine =
1047 TheLine.Last->TotalLength + Indent <= ColumnLimit ||
1048 (TheLine.Type == LT_ImportStatement &&
1049 (Style.Language != FormatStyle::LK_JavaScript ||
1050 !Style.JavaScriptWrapImports));
1051 if (Style.ColumnLimit == 0)
1052 NoColumnLimitLineFormatter(Indenter, Whitespaces, Style, this)
1053 .formatLine(TheLine, NextStartColumn + Indent,
1054 FirstLine ? FirstStartColumn : 0, DryRun);
1055 else if (FitsIntoOneLine)
1056 Penalty += NoLineBreakFormatter(Indenter, Whitespaces, Style, this)
1057 .formatLine(TheLine, NextStartColumn + Indent,
1058 FirstLine ? FirstStartColumn : 0, DryRun);
1060 Penalty += OptimizingLineFormatter(Indenter, Whitespaces, Style, this)
1061 .formatLine(TheLine, NextStartColumn + Indent,
1062 FirstLine ? FirstStartColumn : 0, DryRun);
1063 RangeMinLevel = std::min(RangeMinLevel, TheLine.Level);
1065 // If no token in the current line is affected, we still need to format
1066 // affected children.
1067 if (TheLine.ChildrenAffected)
1068 for (const FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next)
1069 if (!Tok->Children.empty())
1070 format(Tok->Children, DryRun);
1072 // Adapt following lines on the current indent level to the same level
1073 // unless the current \c AnnotatedLine is not at the beginning of a line.
1074 bool StartsNewLine =
1075 TheLine.First->NewlinesBefore > 0 || TheLine.First->IsFirst;
1077 IndentTracker.adjustToUnmodifiedLine(TheLine);
1079 bool ReformatLeadingWhitespace =
1080 StartsNewLine && ((PreviousLine && PreviousLine->Affected) ||
1081 TheLine.LeadingEmptyLinesAffected);
1082 // Format the first token.
1083 if (ReformatLeadingWhitespace)
1084 formatFirstToken(TheLine, PreviousLine,
1085 TheLine.First->OriginalColumn,
1086 TheLine.First->OriginalColumn);
1088 Whitespaces->addUntouchableToken(*TheLine.First,
1089 TheLine.InPPDirective);
1091 // Notify the WhitespaceManager about the unchanged whitespace.
1092 for (FormatToken *Tok = TheLine.First->Next; Tok; Tok = Tok->Next)
1093 Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective);
1095 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
1096 RangeMinLevel = UINT_MAX;
1099 markFinalized(TheLine.First);
1100 PreviousLine = &TheLine;
1102 PenaltyCache[CacheKey] = Penalty;
1106 void UnwrappedLineFormatter::formatFirstToken(const AnnotatedLine &Line,
1107 const AnnotatedLine *PreviousLine,
1109 unsigned NewlineIndent) {
1110 FormatToken &RootToken = *Line.First;
1111 if (RootToken.is(tok::eof)) {
1112 unsigned Newlines = std::min(RootToken.NewlinesBefore, 1u);
1113 unsigned TokenIndent = Newlines ? NewlineIndent : 0;
1114 Whitespaces->replaceWhitespace(RootToken, Newlines, TokenIndent,
1119 std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1);
1120 // Remove empty lines before "}" where applicable.
1121 if (RootToken.is(tok::r_brace) &&
1123 (RootToken.Next->is(tok::semi) && !RootToken.Next->Next)))
1124 Newlines = std::min(Newlines, 1u);
1125 // Remove empty lines at the start of nested blocks (lambdas/arrow functions)
1126 if (PreviousLine == nullptr && Line.Level > 0)
1127 Newlines = std::min(Newlines, 1u);
1128 if (Newlines == 0 && !RootToken.IsFirst)
1130 if (RootToken.IsFirst && !RootToken.HasUnescapedNewline)
1133 // Remove empty lines after "{".
1134 if (!Style.KeepEmptyLinesAtTheStartOfBlocks && PreviousLine &&
1135 PreviousLine->Last->is(tok::l_brace) &&
1136 PreviousLine->First->isNot(tok::kw_namespace) &&
1137 !startsExternCBlock(*PreviousLine))
1140 // Insert extra new line before access specifiers.
1141 if (PreviousLine && PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) &&
1142 RootToken.isAccessSpecifier() && RootToken.NewlinesBefore == 1)
1145 // Remove empty lines after access specifiers.
1146 if (PreviousLine && PreviousLine->First->isAccessSpecifier() &&
1147 (!PreviousLine->InPPDirective || !RootToken.HasUnescapedNewline))
1148 Newlines = std::min(1u, Newlines);
1151 Indent = NewlineIndent;
1153 // Preprocessor directives get indented after the hash, if indented.
1154 if (Line.Type == LT_PreprocessorDirective || Line.Type == LT_ImportStatement)
1157 Whitespaces->replaceWhitespace(RootToken, Newlines, Indent, Indent,
1158 Line.InPPDirective &&
1159 !RootToken.HasUnescapedNewline);
1163 UnwrappedLineFormatter::getColumnLimit(bool InPPDirective,
1164 const AnnotatedLine *NextLine) const {
1165 // In preprocessor directives reserve two chars for trailing " \" if the
1166 // next line continues the preprocessor directive.
1167 bool ContinuesPPDirective =
1169 // If there is no next line, this is likely a child line and the parent
1170 // continues the preprocessor directive.
1172 (NextLine->InPPDirective &&
1173 // If there is an unescaped newline between this line and the next, the
1174 // next line starts a new preprocessor directive.
1175 !NextLine->First->HasUnescapedNewline));
1176 return Style.ColumnLimit - (ContinuesPPDirective ? 2 : 0);
1179 } // namespace format
1180 } // namespace clang