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 level indent to adapt to the given \p Line.
71 /// When a line is not formatted, we move the subsequent lines on the same
72 /// level to the same indent.
73 /// Note that \c nextLine must have been called before this method.
74 void adjustToUnmodifiedLine(const AnnotatedLine &Line) {
75 unsigned LevelIndent = Line.First->OriginalColumn;
76 if (static_cast<int>(LevelIndent) - Offset >= 0)
77 LevelIndent -= Offset;
78 if ((!Line.First->is(tok::comment) || IndentForLevel[Line.Level] == -1) &&
80 IndentForLevel[Line.Level] = LevelIndent;
84 /// \brief Get the offset of the line relatively to the level.
86 /// For example, 'public:' labels in classes are offset by 1 or 2
87 /// characters to the left from their level.
88 int getIndentOffset(const FormatToken &RootToken) {
89 if (Style.Language == FormatStyle::LK_Java ||
90 Style.Language == FormatStyle::LK_JavaScript)
92 if (RootToken.isAccessSpecifier(false) ||
93 RootToken.isObjCAccessSpecifier() ||
94 (RootToken.isOneOf(Keywords.kw_signals, Keywords.kw_qsignals) &&
95 RootToken.Next && RootToken.Next->is(tok::colon)))
96 return Style.AccessModifierOffset;
100 /// \brief Get the indent of \p Level from \p IndentForLevel.
102 /// \p IndentForLevel must contain the indent for the level \c l
103 /// at \p IndentForLevel[l], or a value < 0 if the indent for
104 /// that level is unknown.
105 unsigned getIndent(ArrayRef<int> IndentForLevel, unsigned Level) {
106 if (IndentForLevel[Level] != -1)
107 return IndentForLevel[Level];
110 return getIndent(IndentForLevel, Level - 1) + Style.IndentWidth;
113 const FormatStyle &Style;
114 const AdditionalKeywords &Keywords;
115 const unsigned AdditionalIndent;
117 /// \brief The indent in characters for each level.
118 std::vector<int> IndentForLevel;
120 /// \brief Offset of the current line relative to the indent level.
122 /// For example, the 'public' keywords is often indented with a negative
126 /// \brief The current line's indent.
132 LineJoiner(const FormatStyle &Style, const AdditionalKeywords &Keywords,
133 const SmallVectorImpl<AnnotatedLine *> &Lines)
134 : Style(Style), Keywords(Keywords), End(Lines.end()),
135 Next(Lines.begin()) {}
137 /// \brief Returns the next line, merging multiple lines into one if possible.
138 const AnnotatedLine *getNextMergedLine(bool DryRun,
139 LevelIndentTracker &IndentTracker) {
142 const AnnotatedLine *Current = *Next;
143 IndentTracker.nextLine(*Current);
144 unsigned MergedLines =
145 tryFitMultipleLinesInOne(IndentTracker.getIndent(), Next, End);
146 if (MergedLines > 0 && Style.ColumnLimit == 0)
147 // Disallow line merging if there is a break at the start of one of the
149 for (unsigned i = 0; i < MergedLines; ++i)
150 if (Next[i + 1]->First->NewlinesBefore > 0)
153 for (unsigned i = 0; i < MergedLines; ++i)
154 join(*Next[0], *Next[i + 1]);
155 Next = Next + MergedLines + 1;
160 /// \brief Calculates how many lines can be merged into 1 starting at \p I.
162 tryFitMultipleLinesInOne(unsigned Indent,
163 SmallVectorImpl<AnnotatedLine *>::const_iterator I,
164 SmallVectorImpl<AnnotatedLine *>::const_iterator E) {
165 // Can't join the last line with anything.
168 // We can never merge stuff if there are trailing line comments.
169 const AnnotatedLine *TheLine = *I;
170 if (TheLine->Last->is(TT_LineComment))
172 if (I[1]->Type == LT_Invalid || I[1]->First->MustBreakBefore)
174 if (TheLine->InPPDirective &&
175 (!I[1]->InPPDirective || I[1]->First->HasUnescapedNewline))
178 if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit)
182 Style.ColumnLimit == 0 ? UINT_MAX : Style.ColumnLimit - Indent;
183 // If we already exceed the column limit, we set 'Limit' to 0. The different
184 // tryMerge..() functions can then decide whether to still do merging.
185 Limit = TheLine->Last->TotalLength > Limit
187 : Limit - TheLine->Last->TotalLength;
189 // FIXME: TheLine->Level != 0 might or might not be the right check to do.
190 // If necessary, change to something smarter.
191 bool MergeShortFunctions =
192 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_All ||
193 (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty &&
194 I[1]->First->is(tok::r_brace)) ||
195 (Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Inline &&
196 TheLine->Level != 0);
198 if (TheLine->Last->is(TT_FunctionLBrace) &&
199 TheLine->First != TheLine->Last) {
200 return MergeShortFunctions ? tryMergeSimpleBlock(I, E, Limit) : 0;
202 if (TheLine->Last->is(tok::l_brace)) {
203 return !Style.BraceWrapping.AfterFunction
204 ? tryMergeSimpleBlock(I, E, Limit)
207 if (I[1]->First->is(TT_FunctionLBrace) &&
208 Style.BraceWrapping.AfterFunction) {
209 if (I[1]->Last->is(TT_LineComment))
212 // Check for Limit <= 2 to account for the " {".
213 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(TheLine)))
217 unsigned MergedLines = 0;
218 if (MergeShortFunctions) {
219 MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
220 // If we managed to merge the block, count the function header, which is
221 // on a separate line.
227 if (TheLine->First->is(tok::kw_if)) {
228 return Style.AllowShortIfStatementsOnASingleLine
229 ? tryMergeSimpleControlStatement(I, E, Limit)
232 if (TheLine->First->isOneOf(tok::kw_for, tok::kw_while)) {
233 return Style.AllowShortLoopsOnASingleLine
234 ? tryMergeSimpleControlStatement(I, E, Limit)
237 if (TheLine->First->isOneOf(tok::kw_case, tok::kw_default)) {
238 return Style.AllowShortCaseLabelsOnASingleLine
239 ? tryMergeShortCaseLabels(I, E, Limit)
242 if (TheLine->InPPDirective &&
243 (TheLine->First->HasUnescapedNewline || TheLine->First->IsFirst)) {
244 return tryMergeSimplePPDirective(I, E, Limit);
250 tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
251 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
255 if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline)
257 if (1 + I[1]->Last->TotalLength > Limit)
262 unsigned tryMergeSimpleControlStatement(
263 SmallVectorImpl<AnnotatedLine *>::const_iterator I,
264 SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) {
267 if (Style.BraceWrapping.AfterControlStatement &&
268 (I[1]->First->is(tok::l_brace) && !Style.AllowShortBlocksOnASingleLine))
270 if (I[1]->InPPDirective != (*I)->InPPDirective ||
271 (I[1]->InPPDirective && I[1]->First->HasUnescapedNewline))
273 Limit = limitConsideringMacros(I + 1, E, Limit);
274 AnnotatedLine &Line = **I;
275 if (Line.Last->isNot(tok::r_paren))
277 if (1 + I[1]->Last->TotalLength > Limit)
279 if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for, tok::kw_while,
282 // Only inline simple if's (no nested if or else).
283 if (I + 2 != E && Line.startsWith(tok::kw_if) &&
284 I[2]->First->is(tok::kw_else))
290 tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
291 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
293 if (Limit == 0 || I + 1 == E ||
294 I[1]->First->isOneOf(tok::kw_case, tok::kw_default))
296 unsigned NumStmts = 0;
298 bool InPPDirective = I[0]->InPPDirective;
299 for (; NumStmts < 3; ++NumStmts) {
300 if (I + 1 + NumStmts == E)
302 const AnnotatedLine *Line = I[1 + NumStmts];
303 if (Line->InPPDirective != InPPDirective)
305 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace))
307 if (Line->First->isOneOf(tok::kw_if, tok::kw_for, tok::kw_switch,
308 tok::kw_while, tok::comment) ||
309 Line->Last->is(tok::comment))
311 Length += I[1 + NumStmts]->Last->TotalLength + 1; // 1 for the space.
313 if (NumStmts == 0 || NumStmts == 3 || Length > Limit)
319 tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
320 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
322 AnnotatedLine &Line = **I;
324 // Don't merge ObjC @ keywords and methods.
325 // FIXME: If an option to allow short exception handling clauses on a single
326 // line is added, change this to not return for @try and friends.
327 if (Style.Language != FormatStyle::LK_Java &&
328 Line.First->isOneOf(tok::at, tok::minus, tok::plus))
331 // Check that the current line allows merging. This depends on whether we
332 // are in a control flow statements as well as several style flags.
333 if (Line.First->isOneOf(tok::kw_else, tok::kw_case) ||
334 (Line.First->Next && Line.First->Next->is(tok::kw_else)))
336 if (Line.First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_do, tok::kw_try,
337 tok::kw___try, tok::kw_catch, tok::kw___finally,
338 tok::kw_for, tok::r_brace, Keywords.kw___except)) {
339 if (!Style.AllowShortBlocksOnASingleLine)
341 if (!Style.AllowShortIfStatementsOnASingleLine &&
342 Line.startsWith(tok::kw_if))
344 if (!Style.AllowShortLoopsOnASingleLine &&
345 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for))
347 // FIXME: Consider an option to allow short exception handling clauses on
349 // FIXME: This isn't covered by tests.
350 // FIXME: For catch, __except, __finally the first token on the line
351 // is '}', so this isn't correct here.
352 if (Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch,
353 Keywords.kw___except, tok::kw___finally))
357 FormatToken *Tok = I[1]->First;
358 if (Tok->is(tok::r_brace) && !Tok->MustBreakBefore &&
359 (Tok->getNextNonComment() == nullptr ||
360 Tok->getNextNonComment()->is(tok::semi))) {
361 // We merge empty blocks even if the line exceeds the column limit.
362 Tok->SpacesRequiredBefore = 0;
363 Tok->CanBreakBefore = true;
365 } else if (Limit != 0 && !Line.startsWith(tok::kw_namespace) &&
366 !startsExternCBlock(Line)) {
367 // We don't merge short records.
368 if (Line.First->isOneOf(tok::kw_class, tok::kw_union, tok::kw_struct,
369 Keywords.kw_interface))
372 // Check that we still have three lines and they fit into the limit.
373 if (I + 2 == E || I[2]->Type == LT_Invalid)
375 Limit = limitConsideringMacros(I + 2, E, Limit);
377 if (!nextTwoLinesFitInto(I, Limit))
380 // Second, check that the next line does not contain any braces - if it
381 // does, readability declines when putting it into a single line.
382 if (I[1]->Last->is(TT_LineComment))
385 if (Tok->is(tok::l_brace) && Tok->BlockKind != BK_BracedInit)
390 // Last, check that the third line starts with a closing brace.
392 if (Tok->isNot(tok::r_brace))
395 // Don't merge "if (a) { .. } else {".
396 if (Tok->Next && Tok->Next->is(tok::kw_else))
404 /// Returns the modified column limit for \p I if it is inside a macro and
405 /// needs a trailing '\'.
407 limitConsideringMacros(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
408 SmallVectorImpl<AnnotatedLine *>::const_iterator E,
410 if (I[0]->InPPDirective && I + 1 != E &&
411 !I[1]->First->HasUnescapedNewline && !I[1]->First->is(tok::eof)) {
412 return Limit < 2 ? 0 : Limit - 2;
417 bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
419 if (I[1]->First->MustBreakBefore || I[2]->First->MustBreakBefore)
421 return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit;
424 bool containsMustBreak(const AnnotatedLine *Line) {
425 for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) {
426 if (Tok->MustBreakBefore)
432 void join(AnnotatedLine &A, const AnnotatedLine &B) {
433 assert(!A.Last->Next);
434 assert(!B.First->Previous);
437 A.Last->Next = B.First;
438 B.First->Previous = A.Last;
439 B.First->CanBreakBefore = true;
440 unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore;
441 for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) {
442 Tok->TotalLength += LengthA;
447 const FormatStyle &Style;
448 const AdditionalKeywords &Keywords;
449 const SmallVectorImpl<AnnotatedLine *>::const_iterator End;
451 SmallVectorImpl<AnnotatedLine *>::const_iterator Next;
454 static void markFinalized(FormatToken *Tok) {
455 for (; Tok; Tok = Tok->Next) {
456 Tok->Finalized = true;
457 for (AnnotatedLine *Child : Tok->Children)
458 markFinalized(Child->First);
463 static void printLineState(const LineState &State) {
464 llvm::dbgs() << "State: ";
465 for (const ParenState &P : State.Stack) {
466 llvm::dbgs() << P.Indent << "|" << P.LastSpace << "|" << P.NestedBlockIndent
469 llvm::dbgs() << State.NextToken->TokenText << "\n";
473 /// \brief Base class for classes that format one \c AnnotatedLine.
474 class LineFormatter {
476 LineFormatter(ContinuationIndenter *Indenter, WhitespaceManager *Whitespaces,
477 const FormatStyle &Style,
478 UnwrappedLineFormatter *BlockFormatter)
479 : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style),
480 BlockFormatter(BlockFormatter) {}
481 virtual ~LineFormatter() {}
483 /// \brief Formats an \c AnnotatedLine and returns the penalty.
485 /// If \p DryRun is \c false, directly applies the changes.
486 virtual unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
490 /// \brief If the \p State's next token is an r_brace closing a nested block,
491 /// format the nested block before it.
493 /// Returns \c true if all children could be placed successfully and adapts
494 /// \p Penalty as well as \p State. If \p DryRun is false, also directly
495 /// creates changes using \c Whitespaces.
497 /// The crucial idea here is that children always get formatted upon
498 /// encountering the closing brace right after the nested block. Now, if we
499 /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is
500 /// \c false), the entire block has to be kept on the same line (which is only
501 /// possible if it fits on the line, only contains a single statement, etc.
503 /// If \p NewLine is true, we format the nested block on separate lines, i.e.
504 /// break after the "{", format all lines with correct indentation and the put
505 /// the closing "}" on yet another new line.
507 /// This enables us to keep the simple structure of the
508 /// \c UnwrappedLineFormatter, where we only have two options for each token:
509 /// break or don't break.
510 bool formatChildren(LineState &State, bool NewLine, bool DryRun,
512 const FormatToken *LBrace = State.NextToken->getPreviousNonComment();
513 FormatToken &Previous = *State.NextToken->Previous;
514 if (!LBrace || LBrace->isNot(tok::l_brace) ||
515 LBrace->BlockKind != BK_Block || Previous.Children.size() == 0)
516 // The previous token does not open a block. Nothing to do. We don't
517 // assert so that we can simply call this function for all tokens.
521 int AdditionalIndent = State.Stack.back().Indent -
522 Previous.Children[0]->Level * Style.IndentWidth;
525 BlockFormatter->format(Previous.Children, DryRun, AdditionalIndent,
526 /*FixBadIndentation=*/true);
530 if (Previous.Children[0]->First->MustBreakBefore)
533 // Cannot merge into one line if this line ends on a comment.
534 if (Previous.is(tok::comment))
537 // Cannot merge multiple statements into a single line.
538 if (Previous.Children.size() > 1)
541 const AnnotatedLine *Child = Previous.Children[0];
542 // We can't put the closing "}" on a line with a trailing comment.
543 if (Child->Last->isTrailingComment())
546 // If the child line exceeds the column limit, we wouldn't want to merge it.
547 // We add +2 for the trailing " }".
548 if (Style.ColumnLimit > 0 &&
549 Child->Last->TotalLength + State.Column + 2 > Style.ColumnLimit)
553 Whitespaces->replaceWhitespace(
554 *Child->First, /*Newlines=*/0, /*Spaces=*/1,
555 /*StartOfTokenColumn=*/State.Column, State.Line->InPPDirective);
557 Penalty += formatLine(*Child, State.Column + 1, DryRun);
559 State.Column += 1 + Child->Last->TotalLength;
563 ContinuationIndenter *Indenter;
566 WhitespaceManager *Whitespaces;
567 const FormatStyle &Style;
568 UnwrappedLineFormatter *BlockFormatter;
571 /// \brief Formatter that keeps the existing line breaks.
572 class NoColumnLimitLineFormatter : public LineFormatter {
574 NoColumnLimitLineFormatter(ContinuationIndenter *Indenter,
575 WhitespaceManager *Whitespaces,
576 const FormatStyle &Style,
577 UnwrappedLineFormatter *BlockFormatter)
578 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
580 /// \brief Formats the line, simply keeping all of the input's line breaking
582 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
583 bool DryRun) override {
586 Indenter->getInitialState(FirstIndent, &Line, /*DryRun=*/false);
587 while (State.NextToken) {
589 Indenter->mustBreak(State) ||
590 (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0);
591 unsigned Penalty = 0;
592 formatChildren(State, Newline, /*DryRun=*/false, Penalty);
593 Indenter->addTokenToState(State, Newline, /*DryRun=*/false);
599 /// \brief Formatter that puts all tokens into a single line without breaks.
600 class NoLineBreakFormatter : public LineFormatter {
602 NoLineBreakFormatter(ContinuationIndenter *Indenter,
603 WhitespaceManager *Whitespaces, const FormatStyle &Style,
604 UnwrappedLineFormatter *BlockFormatter)
605 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
607 /// \brief Puts all tokens into a single line.
608 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
609 bool DryRun) override {
610 unsigned Penalty = 0;
611 LineState State = Indenter->getInitialState(FirstIndent, &Line, DryRun);
612 while (State.NextToken) {
613 formatChildren(State, /*Newline=*/false, DryRun, Penalty);
614 Indenter->addTokenToState(State, /*Newline=*/false, DryRun);
620 /// \brief Finds the best way to break lines.
621 class OptimizingLineFormatter : public LineFormatter {
623 OptimizingLineFormatter(ContinuationIndenter *Indenter,
624 WhitespaceManager *Whitespaces,
625 const FormatStyle &Style,
626 UnwrappedLineFormatter *BlockFormatter)
627 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
629 /// \brief Formats the line by finding the best line breaks with line lengths
630 /// below the column limit.
631 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
632 bool DryRun) override {
633 LineState State = Indenter->getInitialState(FirstIndent, &Line, DryRun);
635 // If the ObjC method declaration does not fit on a line, we should format
636 // it with one arg per line.
637 if (State.Line->Type == LT_ObjCMethodDecl)
638 State.Stack.back().BreakBeforeParameter = true;
640 // Find best solution in solution space.
641 return analyzeSolutionSpace(State, DryRun);
645 struct CompareLineStatePointers {
646 bool operator()(LineState *obj1, LineState *obj2) const {
647 return *obj1 < *obj2;
651 /// \brief A pair of <penalty, count> that is used to prioritize the BFS on.
653 /// In case of equal penalties, we want to prefer states that were inserted
654 /// first. During state generation we make sure that we insert states first
655 /// that break the line as late as possible.
656 typedef std::pair<unsigned, unsigned> OrderedPenalty;
658 /// \brief An edge in the solution space from \c Previous->State to \c State,
659 /// inserting a newline dependent on the \c NewLine.
661 StateNode(const LineState &State, bool NewLine, StateNode *Previous)
662 : State(State), NewLine(NewLine), Previous(Previous) {}
668 /// \brief An item in the prioritized BFS search queue. The \c StateNode's
669 /// \c State has the given \c OrderedPenalty.
670 typedef std::pair<OrderedPenalty, StateNode *> QueueItem;
672 /// \brief The BFS queue type.
673 typedef std::priority_queue<QueueItem, std::vector<QueueItem>,
674 std::greater<QueueItem>> QueueType;
676 /// \brief Analyze the entire solution space starting from \p InitialState.
678 /// This implements a variant of Dijkstra's algorithm on the graph that spans
679 /// the solution space (\c LineStates are the nodes). The algorithm tries to
680 /// find the shortest path (the one with lowest penalty) from \p InitialState
681 /// to a state where all tokens are placed. Returns the penalty.
683 /// If \p DryRun is \c false, directly applies the changes.
684 unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun) {
685 std::set<LineState *, CompareLineStatePointers> Seen;
687 // Increasing count of \c StateNode items we have created. This is used to
688 // create a deterministic order independent of the container.
692 // Insert start element into queue.
694 new (Allocator.Allocate()) StateNode(InitialState, false, nullptr);
695 Queue.push(QueueItem(OrderedPenalty(0, Count), Node));
698 unsigned Penalty = 0;
700 // While not empty, take first element and follow edges.
701 while (!Queue.empty()) {
702 Penalty = Queue.top().first.first;
703 StateNode *Node = Queue.top().second;
704 if (!Node->State.NextToken) {
705 DEBUG(llvm::dbgs() << "\n---\nPenalty for line: " << Penalty << "\n");
710 // Cut off the analysis of certain solutions if the analysis gets too
711 // complex. See description of IgnoreStackForComparison.
713 Node->State.IgnoreStackForComparison = true;
715 if (!Seen.insert(&Node->State).second)
716 // State already examined with lower penalty.
719 FormatDecision LastFormat = Node->State.NextToken->Decision;
720 if (LastFormat == FD_Unformatted || LastFormat == FD_Continue)
721 addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue);
722 if (LastFormat == FD_Unformatted || LastFormat == FD_Break)
723 addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue);
727 // We were unable to find a solution, do nothing.
728 // FIXME: Add diagnostic?
729 DEBUG(llvm::dbgs() << "Could not find a solution.\n");
733 // Reconstruct the solution.
735 reconstructPath(InitialState, Queue.top().second);
737 DEBUG(llvm::dbgs() << "Total number of analyzed states: " << Count << "\n");
738 DEBUG(llvm::dbgs() << "---\n");
743 /// \brief Add the following state to the analysis queue \c Queue.
745 /// Assume the current state is \p PreviousNode and has been reached with a
746 /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true.
747 void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode,
748 bool NewLine, unsigned *Count, QueueType *Queue) {
749 if (NewLine && !Indenter->canBreak(PreviousNode->State))
751 if (!NewLine && Indenter->mustBreak(PreviousNode->State))
754 StateNode *Node = new (Allocator.Allocate())
755 StateNode(PreviousNode->State, NewLine, PreviousNode);
756 if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty))
759 Penalty += Indenter->addTokenToState(Node->State, NewLine, true);
761 Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node));
765 /// \brief Applies the best formatting by reconstructing the path in the
766 /// solution space that leads to \c Best.
767 void reconstructPath(LineState &State, StateNode *Best) {
768 std::deque<StateNode *> Path;
769 // We do not need a break before the initial token.
770 while (Best->Previous) {
771 Path.push_front(Best);
772 Best = Best->Previous;
774 for (std::deque<StateNode *>::iterator I = Path.begin(), E = Path.end();
776 unsigned Penalty = 0;
777 formatChildren(State, (*I)->NewLine, /*DryRun=*/false, Penalty);
778 Penalty += Indenter->addTokenToState(State, (*I)->NewLine, false);
781 printLineState((*I)->Previous->State);
783 llvm::dbgs() << "Penalty for placing "
784 << (*I)->Previous->State.NextToken->Tok.getName() << ": "
791 llvm::SpecificBumpPtrAllocator<StateNode> Allocator;
794 } // anonymous namespace
797 UnwrappedLineFormatter::format(const SmallVectorImpl<AnnotatedLine *> &Lines,
798 bool DryRun, int AdditionalIndent,
799 bool FixBadIndentation) {
800 LineJoiner Joiner(Style, Keywords, Lines);
802 // Try to look up already computed penalty in DryRun-mode.
803 std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned> CacheKey(
804 &Lines, AdditionalIndent);
805 auto CacheIt = PenaltyCache.find(CacheKey);
806 if (DryRun && CacheIt != PenaltyCache.end())
807 return CacheIt->second;
809 assert(!Lines.empty());
810 unsigned Penalty = 0;
811 LevelIndentTracker IndentTracker(Style, Keywords, Lines[0]->Level,
813 const AnnotatedLine *PreviousLine = nullptr;
814 const AnnotatedLine *NextLine = nullptr;
816 // The minimum level of consecutive lines that have been formatted.
817 unsigned RangeMinLevel = UINT_MAX;
819 for (const AnnotatedLine *Line =
820 Joiner.getNextMergedLine(DryRun, IndentTracker);
821 Line; Line = NextLine) {
822 const AnnotatedLine &TheLine = *Line;
823 unsigned Indent = IndentTracker.getIndent();
825 // We continue formatting unchanged lines to adjust their indent, e.g. if a
826 // scope was added. However, we need to carefully stop doing this when we
827 // exit the scope of affected lines to prevent indenting a the entire
828 // remaining file if it currently missing a closing brace.
829 bool ContinueFormatting =
830 TheLine.Level > RangeMinLevel ||
831 (TheLine.Level == RangeMinLevel && !TheLine.startsWith(tok::r_brace));
833 bool FixIndentation = (FixBadIndentation || ContinueFormatting) &&
834 Indent != TheLine.First->OriginalColumn;
835 bool ShouldFormat = TheLine.Affected || FixIndentation;
836 // We cannot format this line; if the reason is that the line had a
837 // parsing error, remember that.
838 if (ShouldFormat && TheLine.Type == LT_Invalid && Status) {
839 Status->FormatComplete = false;
841 SourceMgr.getSpellingLineNumber(TheLine.First->Tok.getLocation());
844 if (ShouldFormat && TheLine.Type != LT_Invalid) {
846 formatFirstToken(TheLine, PreviousLine, Indent);
848 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
849 unsigned ColumnLimit = getColumnLimit(TheLine.InPPDirective, NextLine);
850 bool FitsIntoOneLine =
851 TheLine.Last->TotalLength + Indent <= ColumnLimit ||
852 (TheLine.Type == LT_ImportStatement &&
853 (Style.Language != FormatStyle::LK_JavaScript ||
854 !Style.JavaScriptWrapImports));
856 if (Style.ColumnLimit == 0)
857 NoColumnLimitLineFormatter(Indenter, Whitespaces, Style, this)
858 .formatLine(TheLine, Indent, DryRun);
859 else if (FitsIntoOneLine)
860 Penalty += NoLineBreakFormatter(Indenter, Whitespaces, Style, this)
861 .formatLine(TheLine, Indent, DryRun);
863 Penalty += OptimizingLineFormatter(Indenter, Whitespaces, Style, this)
864 .formatLine(TheLine, Indent, DryRun);
865 RangeMinLevel = std::min(RangeMinLevel, TheLine.Level);
867 // If no token in the current line is affected, we still need to format
868 // affected children.
869 if (TheLine.ChildrenAffected)
870 for (const FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next)
871 if (!Tok->Children.empty())
872 format(Tok->Children, DryRun);
874 // Adapt following lines on the current indent level to the same level
875 // unless the current \c AnnotatedLine is not at the beginning of a line.
877 TheLine.First->NewlinesBefore > 0 || TheLine.First->IsFirst;
879 IndentTracker.adjustToUnmodifiedLine(TheLine);
881 bool ReformatLeadingWhitespace =
882 StartsNewLine && ((PreviousLine && PreviousLine->Affected) ||
883 TheLine.LeadingEmptyLinesAffected);
884 // Format the first token.
885 if (ReformatLeadingWhitespace)
886 formatFirstToken(TheLine, PreviousLine,
887 TheLine.First->OriginalColumn);
889 Whitespaces->addUntouchableToken(*TheLine.First,
890 TheLine.InPPDirective);
892 // Notify the WhitespaceManager about the unchanged whitespace.
893 for (FormatToken *Tok = TheLine.First->Next; Tok; Tok = Tok->Next)
894 Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective);
896 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
897 RangeMinLevel = UINT_MAX;
900 markFinalized(TheLine.First);
901 PreviousLine = &TheLine;
903 PenaltyCache[CacheKey] = Penalty;
907 void UnwrappedLineFormatter::formatFirstToken(const AnnotatedLine &Line,
908 const AnnotatedLine *PreviousLine,
910 FormatToken& RootToken = *Line.First;
911 if (RootToken.is(tok::eof)) {
912 unsigned Newlines = std::min(RootToken.NewlinesBefore, 1u);
913 Whitespaces->replaceWhitespace(RootToken, Newlines, /*Spaces=*/0,
914 /*StartOfTokenColumn=*/0);
918 std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1);
919 // Remove empty lines before "}" where applicable.
920 if (RootToken.is(tok::r_brace) &&
922 (RootToken.Next->is(tok::semi) && !RootToken.Next->Next)))
923 Newlines = std::min(Newlines, 1u);
924 if (Newlines == 0 && !RootToken.IsFirst)
926 if (RootToken.IsFirst && !RootToken.HasUnescapedNewline)
929 // Remove empty lines after "{".
930 if (!Style.KeepEmptyLinesAtTheStartOfBlocks && PreviousLine &&
931 PreviousLine->Last->is(tok::l_brace) &&
932 PreviousLine->First->isNot(tok::kw_namespace) &&
933 !startsExternCBlock(*PreviousLine))
936 // Insert extra new line before access specifiers.
937 if (PreviousLine && PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) &&
938 RootToken.isAccessSpecifier() && RootToken.NewlinesBefore == 1)
941 // Remove empty lines after access specifiers.
942 if (PreviousLine && PreviousLine->First->isAccessSpecifier() &&
943 (!PreviousLine->InPPDirective || !RootToken.HasUnescapedNewline))
944 Newlines = std::min(1u, Newlines);
946 Whitespaces->replaceWhitespace(RootToken, Newlines, Indent, Indent,
947 Line.InPPDirective &&
948 !RootToken.HasUnescapedNewline);
952 UnwrappedLineFormatter::getColumnLimit(bool InPPDirective,
953 const AnnotatedLine *NextLine) const {
954 // In preprocessor directives reserve two chars for trailing " \" if the
955 // next line continues the preprocessor directive.
956 bool ContinuesPPDirective =
958 // If there is no next line, this is likely a child line and the parent
959 // continues the preprocessor directive.
961 (NextLine->InPPDirective &&
962 // If there is an unescaped newline between this line and the next, the
963 // next line starts a new preprocessor directive.
964 !NextLine->First->HasUnescapedNewline));
965 return Style.ColumnLimit - (ContinuesPPDirective ? 2 : 0);
968 } // namespace format