1 //===--- FormatToken.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 /// \file
10 /// This file implements specific functions of \c FormatTokens and their
11 /// roles.
12 ///
13 //===----------------------------------------------------------------------===//
14 
15 #include "FormatToken.h"
16 #include "ContinuationIndenter.h"
17 #include "llvm/ADT/SmallVector.h"
18 #include "llvm/Support/Debug.h"
19 #include <climits>
20 
21 namespace clang {
22 namespace format {
23 
24 const char *getTokenTypeName(TokenType Type) {
25   static const char *const TokNames[] = {
26 #define TYPE(X) #X,
27       LIST_TOKEN_TYPES
28 #undef TYPE
29       nullptr};
30 
31   if (Type < NUM_TOKEN_TYPES)
32     return TokNames[Type];
33   llvm_unreachable("unknown TokenType");
34   return nullptr;
35 }
36 
37 // FIXME: This is copy&pasted from Sema. Put it in a common place and remove
38 // duplication.
39 bool FormatToken::isSimpleTypeSpecifier() const {
40   switch (Tok.getKind()) {
41   case tok::kw_short:
42   case tok::kw_long:
43   case tok::kw___int64:
44   case tok::kw___int128:
45   case tok::kw_signed:
46   case tok::kw_unsigned:
47   case tok::kw_void:
48   case tok::kw_char:
49   case tok::kw_int:
50   case tok::kw_half:
51   case tok::kw_float:
52   case tok::kw_double:
53   case tok::kw___bf16:
54   case tok::kw__Float16:
55   case tok::kw___float128:
56   case tok::kw___ibm128:
57   case tok::kw_wchar_t:
58   case tok::kw_bool:
59   case tok::kw___underlying_type:
60   case tok::annot_typename:
61   case tok::kw_char8_t:
62   case tok::kw_char16_t:
63   case tok::kw_char32_t:
64   case tok::kw_typeof:
65   case tok::kw_decltype:
66   case tok::kw__Atomic:
67     return true;
68   default:
69     return false;
70   }
71 }
72 
73 bool FormatToken::isTypeOrIdentifier() const {
74   return isSimpleTypeSpecifier() || Tok.isOneOf(tok::kw_auto, tok::identifier);
75 }
76 
77 bool FormatToken::opensBlockOrBlockTypeList(const FormatStyle &Style) const {
78   // C# Does not indent object initialisers as continuations.
79   if (is(tok::l_brace) && getBlockKind() == BK_BracedInit && Style.isCSharp())
80     return true;
81   if (is(TT_TemplateString) && opensScope())
82     return true;
83   return is(TT_ArrayInitializerLSquare) || is(TT_ProtoExtensionLSquare) ||
84          (is(tok::l_brace) &&
85           (getBlockKind() == BK_Block || is(TT_DictLiteral) ||
86            (!Style.Cpp11BracedListStyle && NestingLevel == 0))) ||
87          (is(tok::less) && (Style.Language == FormatStyle::LK_Proto ||
88                             Style.Language == FormatStyle::LK_TextProto));
89 }
90 
91 TokenRole::~TokenRole() {}
92 
93 void TokenRole::precomputeFormattingInfos(const FormatToken *Token) {}
94 
95 unsigned CommaSeparatedList::formatAfterToken(LineState &State,
96                                               ContinuationIndenter *Indenter,
97                                               bool DryRun) {
98   if (State.NextToken == nullptr || !State.NextToken->Previous)
99     return 0;
100 
101   if (Formats.size() == 1)
102     return 0; // Handled by formatFromToken
103 
104   // Ensure that we start on the opening brace.
105   const FormatToken *LBrace =
106       State.NextToken->Previous->getPreviousNonComment();
107   if (!LBrace || !LBrace->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
108       LBrace->is(BK_Block) || LBrace->is(TT_DictLiteral) ||
109       LBrace->Next->is(TT_DesignatedInitializerPeriod)) {
110     return 0;
111   }
112 
113   // Calculate the number of code points we have to format this list. As the
114   // first token is already placed, we have to subtract it.
115   unsigned RemainingCodePoints =
116       Style.ColumnLimit - State.Column + State.NextToken->Previous->ColumnWidth;
117 
118   // Find the best ColumnFormat, i.e. the best number of columns to use.
119   const ColumnFormat *Format = getColumnFormat(RemainingCodePoints);
120 
121   // If no ColumnFormat can be used, the braced list would generally be
122   // bin-packed. Add a severe penalty to this so that column layouts are
123   // preferred if possible.
124   if (!Format)
125     return 10000;
126 
127   // Format the entire list.
128   unsigned Penalty = 0;
129   unsigned Column = 0;
130   unsigned Item = 0;
131   while (State.NextToken != LBrace->MatchingParen) {
132     bool NewLine = false;
133     unsigned ExtraSpaces = 0;
134 
135     // If the previous token was one of our commas, we are now on the next item.
136     if (Item < Commas.size() && State.NextToken->Previous == Commas[Item]) {
137       if (!State.NextToken->isTrailingComment()) {
138         ExtraSpaces += Format->ColumnSizes[Column] - ItemLengths[Item];
139         ++Column;
140       }
141       ++Item;
142     }
143 
144     if (Column == Format->Columns || State.NextToken->MustBreakBefore) {
145       Column = 0;
146       NewLine = true;
147     }
148 
149     // Place token using the continuation indenter and store the penalty.
150     Penalty += Indenter->addTokenToState(State, NewLine, DryRun, ExtraSpaces);
151   }
152   return Penalty;
153 }
154 
155 unsigned CommaSeparatedList::formatFromToken(LineState &State,
156                                              ContinuationIndenter *Indenter,
157                                              bool DryRun) {
158   // Formatting with 1 Column isn't really a column layout, so we don't need the
159   // special logic here. We can just avoid bin packing any of the parameters.
160   if (Formats.size() == 1 || HasNestedBracedList)
161     State.Stack.back().AvoidBinPacking = true;
162   return 0;
163 }
164 
165 // Returns the lengths in code points between Begin and End (both included),
166 // assuming that the entire sequence is put on a single line.
167 static unsigned CodePointsBetween(const FormatToken *Begin,
168                                   const FormatToken *End) {
169   assert(End->TotalLength >= Begin->TotalLength);
170   return End->TotalLength - Begin->TotalLength + Begin->ColumnWidth;
171 }
172 
173 void CommaSeparatedList::precomputeFormattingInfos(const FormatToken *Token) {
174   // FIXME: At some point we might want to do this for other lists, too.
175   if (!Token->MatchingParen ||
176       !Token->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare)) {
177     return;
178   }
179 
180   // In C++11 braced list style, we should not format in columns unless they
181   // have many items (20 or more) or we allow bin-packing of function call
182   // arguments.
183   if (Style.Cpp11BracedListStyle && !Style.BinPackArguments &&
184       Commas.size() < 19) {
185     return;
186   }
187 
188   // Limit column layout for JavaScript array initializers to 20 or more items
189   // for now to introduce it carefully. We can become more aggressive if this
190   // necessary.
191   if (Token->is(TT_ArrayInitializerLSquare) && Commas.size() < 19)
192     return;
193 
194   // Column format doesn't really make sense if we don't align after brackets.
195   if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign)
196     return;
197 
198   FormatToken *ItemBegin = Token->Next;
199   while (ItemBegin->isTrailingComment())
200     ItemBegin = ItemBegin->Next;
201   SmallVector<bool, 8> MustBreakBeforeItem;
202 
203   // The lengths of an item if it is put at the end of the line. This includes
204   // trailing comments which are otherwise ignored for column alignment.
205   SmallVector<unsigned, 8> EndOfLineItemLength;
206   MustBreakBeforeItem.reserve(Commas.size() + 1);
207   EndOfLineItemLength.reserve(Commas.size() + 1);
208   ItemLengths.reserve(Commas.size() + 1);
209 
210   bool HasSeparatingComment = false;
211   for (unsigned i = 0, e = Commas.size() + 1; i != e; ++i) {
212     assert(ItemBegin);
213     // Skip comments on their own line.
214     while (ItemBegin->HasUnescapedNewline && ItemBegin->isTrailingComment()) {
215       ItemBegin = ItemBegin->Next;
216       HasSeparatingComment = i > 0;
217     }
218 
219     MustBreakBeforeItem.push_back(ItemBegin->MustBreakBefore);
220     if (ItemBegin->is(tok::l_brace))
221       HasNestedBracedList = true;
222     const FormatToken *ItemEnd = nullptr;
223     if (i == Commas.size()) {
224       ItemEnd = Token->MatchingParen;
225       const FormatToken *NonCommentEnd = ItemEnd->getPreviousNonComment();
226       ItemLengths.push_back(CodePointsBetween(ItemBegin, NonCommentEnd));
227       if (Style.Cpp11BracedListStyle &&
228           !ItemEnd->Previous->isTrailingComment()) {
229         // In Cpp11 braced list style, the } and possibly other subsequent
230         // tokens will need to stay on a line with the last element.
231         while (ItemEnd->Next && !ItemEnd->Next->CanBreakBefore)
232           ItemEnd = ItemEnd->Next;
233       } else {
234         // In other braced lists styles, the "}" can be wrapped to the new line.
235         ItemEnd = Token->MatchingParen->Previous;
236       }
237     } else {
238       ItemEnd = Commas[i];
239       // The comma is counted as part of the item when calculating the length.
240       ItemLengths.push_back(CodePointsBetween(ItemBegin, ItemEnd));
241 
242       // Consume trailing comments so the are included in EndOfLineItemLength.
243       if (ItemEnd->Next && !ItemEnd->Next->HasUnescapedNewline &&
244           ItemEnd->Next->isTrailingComment()) {
245         ItemEnd = ItemEnd->Next;
246       }
247     }
248     EndOfLineItemLength.push_back(CodePointsBetween(ItemBegin, ItemEnd));
249     // If there is a trailing comma in the list, the next item will start at the
250     // closing brace. Don't create an extra item for this.
251     if (ItemEnd->getNextNonComment() == Token->MatchingParen)
252       break;
253     ItemBegin = ItemEnd->Next;
254   }
255 
256   // Don't use column layout for lists with few elements and in presence of
257   // separating comments.
258   if (Commas.size() < 5 || HasSeparatingComment)
259     return;
260 
261   if (Token->NestingLevel != 0 && Token->is(tok::l_brace) && Commas.size() < 19)
262     return;
263 
264   // We can never place more than ColumnLimit / 3 items in a row (because of the
265   // spaces and the comma).
266   unsigned MaxItems = Style.ColumnLimit / 3;
267   SmallVector<unsigned> MinSizeInColumn;
268   MinSizeInColumn.reserve(MaxItems);
269   for (unsigned Columns = 1; Columns <= MaxItems; ++Columns) {
270     ColumnFormat Format;
271     Format.Columns = Columns;
272     Format.ColumnSizes.resize(Columns);
273     MinSizeInColumn.assign(Columns, UINT_MAX);
274     Format.LineCount = 1;
275     bool HasRowWithSufficientColumns = false;
276     unsigned Column = 0;
277     for (unsigned i = 0, e = ItemLengths.size(); i != e; ++i) {
278       assert(i < MustBreakBeforeItem.size());
279       if (MustBreakBeforeItem[i] || Column == Columns) {
280         ++Format.LineCount;
281         Column = 0;
282       }
283       if (Column == Columns - 1)
284         HasRowWithSufficientColumns = true;
285       unsigned Length =
286           (Column == Columns - 1) ? EndOfLineItemLength[i] : ItemLengths[i];
287       Format.ColumnSizes[Column] = std::max(Format.ColumnSizes[Column], Length);
288       MinSizeInColumn[Column] = std::min(MinSizeInColumn[Column], Length);
289       ++Column;
290     }
291     // If all rows are terminated early (e.g. by trailing comments), we don't
292     // need to look further.
293     if (!HasRowWithSufficientColumns)
294       break;
295     Format.TotalWidth = Columns - 1; // Width of the N-1 spaces.
296 
297     for (unsigned i = 0; i < Columns; ++i)
298       Format.TotalWidth += Format.ColumnSizes[i];
299 
300     // Don't use this Format, if the difference between the longest and shortest
301     // element in a column exceeds a threshold to avoid excessive spaces.
302     if ([&] {
303           for (unsigned i = 0; i < Columns - 1; ++i)
304             if (Format.ColumnSizes[i] - MinSizeInColumn[i] > 10)
305               return true;
306           return false;
307         }()) {
308       continue;
309     }
310 
311     // Ignore layouts that are bound to violate the column limit.
312     if (Format.TotalWidth > Style.ColumnLimit && Columns > 1)
313       continue;
314 
315     Formats.push_back(Format);
316   }
317 }
318 
319 const CommaSeparatedList::ColumnFormat *
320 CommaSeparatedList::getColumnFormat(unsigned RemainingCharacters) const {
321   const ColumnFormat *BestFormat = nullptr;
322   for (const ColumnFormat &Format : llvm::reverse(Formats)) {
323     if (Format.TotalWidth <= RemainingCharacters || Format.Columns == 1) {
324       if (BestFormat && Format.LineCount > BestFormat->LineCount)
325         break;
326       BestFormat = &Format;
327     }
328   }
329   return BestFormat;
330 }
331 
332 } // namespace format
333 } // namespace clang
334