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   // Calculate the number of code points we have to format this list. As the
113   // first token is already placed, we have to subtract it.
114   unsigned RemainingCodePoints =
115       Style.ColumnLimit - State.Column + State.NextToken->Previous->ColumnWidth;
116 
117   // Find the best ColumnFormat, i.e. the best number of columns to use.
118   const ColumnFormat *Format = getColumnFormat(RemainingCodePoints);
119 
120   // If no ColumnFormat can be used, the braced list would generally be
121   // bin-packed. Add a severe penalty to this so that column layouts are
122   // preferred if possible.
123   if (!Format)
124     return 10000;
125 
126   // Format the entire list.
127   unsigned Penalty = 0;
128   unsigned Column = 0;
129   unsigned Item = 0;
130   while (State.NextToken != LBrace->MatchingParen) {
131     bool NewLine = false;
132     unsigned ExtraSpaces = 0;
133 
134     // If the previous token was one of our commas, we are now on the next item.
135     if (Item < Commas.size() && State.NextToken->Previous == Commas[Item]) {
136       if (!State.NextToken->isTrailingComment()) {
137         ExtraSpaces += Format->ColumnSizes[Column] - ItemLengths[Item];
138         ++Column;
139       }
140       ++Item;
141     }
142 
143     if (Column == Format->Columns || State.NextToken->MustBreakBefore) {
144       Column = 0;
145       NewLine = true;
146     }
147 
148     // Place token using the continuation indenter and store the penalty.
149     Penalty += Indenter->addTokenToState(State, NewLine, DryRun, ExtraSpaces);
150   }
151   return Penalty;
152 }
153 
154 unsigned CommaSeparatedList::formatFromToken(LineState &State,
155                                              ContinuationIndenter *Indenter,
156                                              bool DryRun) {
157   // Formatting with 1 Column isn't really a column layout, so we don't need the
158   // special logic here. We can just avoid bin packing any of the parameters.
159   if (Formats.size() == 1 || HasNestedBracedList)
160     State.Stack.back().AvoidBinPacking = true;
161   return 0;
162 }
163 
164 // Returns the lengths in code points between Begin and End (both included),
165 // assuming that the entire sequence is put on a single line.
166 static unsigned CodePointsBetween(const FormatToken *Begin,
167                                   const FormatToken *End) {
168   assert(End->TotalLength >= Begin->TotalLength);
169   return End->TotalLength - Begin->TotalLength + Begin->ColumnWidth;
170 }
171 
172 void CommaSeparatedList::precomputeFormattingInfos(const FormatToken *Token) {
173   // FIXME: At some point we might want to do this for other lists, too.
174   if (!Token->MatchingParen ||
175       !Token->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare))
176     return;
177 
178   // In C++11 braced list style, we should not format in columns unless they
179   // have many items (20 or more) or we allow bin-packing of function call
180   // arguments.
181   if (Style.Cpp11BracedListStyle && !Style.BinPackArguments &&
182       Commas.size() < 19)
183     return;
184 
185   // Limit column layout for JavaScript array initializers to 20 or more items
186   // for now to introduce it carefully. We can become more aggressive if this
187   // necessary.
188   if (Token->is(TT_ArrayInitializerLSquare) && Commas.size() < 19)
189     return;
190 
191   // Column format doesn't really make sense if we don't align after brackets.
192   if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign)
193     return;
194 
195   FormatToken *ItemBegin = Token->Next;
196   while (ItemBegin->isTrailingComment())
197     ItemBegin = ItemBegin->Next;
198   SmallVector<bool, 8> MustBreakBeforeItem;
199 
200   // The lengths of an item if it is put at the end of the line. This includes
201   // trailing comments which are otherwise ignored for column alignment.
202   SmallVector<unsigned, 8> EndOfLineItemLength;
203   MustBreakBeforeItem.reserve(Commas.size() + 1);
204   EndOfLineItemLength.reserve(Commas.size() + 1);
205   ItemLengths.reserve(Commas.size() + 1);
206 
207   bool HasSeparatingComment = false;
208   for (unsigned i = 0, e = Commas.size() + 1; i != e; ++i) {
209     assert(ItemBegin);
210     // Skip comments on their own line.
211     while (ItemBegin->HasUnescapedNewline && ItemBegin->isTrailingComment()) {
212       ItemBegin = ItemBegin->Next;
213       HasSeparatingComment = i > 0;
214     }
215 
216     MustBreakBeforeItem.push_back(ItemBegin->MustBreakBefore);
217     if (ItemBegin->is(tok::l_brace))
218       HasNestedBracedList = true;
219     const FormatToken *ItemEnd = nullptr;
220     if (i == Commas.size()) {
221       ItemEnd = Token->MatchingParen;
222       const FormatToken *NonCommentEnd = ItemEnd->getPreviousNonComment();
223       ItemLengths.push_back(CodePointsBetween(ItemBegin, NonCommentEnd));
224       if (Style.Cpp11BracedListStyle &&
225           !ItemEnd->Previous->isTrailingComment()) {
226         // In Cpp11 braced list style, the } and possibly other subsequent
227         // tokens will need to stay on a line with the last element.
228         while (ItemEnd->Next && !ItemEnd->Next->CanBreakBefore)
229           ItemEnd = ItemEnd->Next;
230       } else {
231         // In other braced lists styles, the "}" can be wrapped to the new line.
232         ItemEnd = Token->MatchingParen->Previous;
233       }
234     } else {
235       ItemEnd = Commas[i];
236       // The comma is counted as part of the item when calculating the length.
237       ItemLengths.push_back(CodePointsBetween(ItemBegin, ItemEnd));
238 
239       // Consume trailing comments so the are included in EndOfLineItemLength.
240       if (ItemEnd->Next && !ItemEnd->Next->HasUnescapedNewline &&
241           ItemEnd->Next->isTrailingComment())
242         ItemEnd = ItemEnd->Next;
243     }
244     EndOfLineItemLength.push_back(CodePointsBetween(ItemBegin, ItemEnd));
245     // If there is a trailing comma in the list, the next item will start at the
246     // closing brace. Don't create an extra item for this.
247     if (ItemEnd->getNextNonComment() == Token->MatchingParen)
248       break;
249     ItemBegin = ItemEnd->Next;
250   }
251 
252   // Don't use column layout for lists with few elements and in presence of
253   // separating comments.
254   if (Commas.size() < 5 || HasSeparatingComment)
255     return;
256 
257   if (Token->NestingLevel != 0 && Token->is(tok::l_brace) && Commas.size() < 19)
258     return;
259 
260   // We can never place more than ColumnLimit / 3 items in a row (because of the
261   // spaces and the comma).
262   unsigned MaxItems = Style.ColumnLimit / 3;
263   std::vector<unsigned> MinSizeInColumn;
264   MinSizeInColumn.reserve(MaxItems);
265   for (unsigned Columns = 1; Columns <= MaxItems; ++Columns) {
266     ColumnFormat Format;
267     Format.Columns = Columns;
268     Format.ColumnSizes.resize(Columns);
269     MinSizeInColumn.assign(Columns, UINT_MAX);
270     Format.LineCount = 1;
271     bool HasRowWithSufficientColumns = false;
272     unsigned Column = 0;
273     for (unsigned i = 0, e = ItemLengths.size(); i != e; ++i) {
274       assert(i < MustBreakBeforeItem.size());
275       if (MustBreakBeforeItem[i] || Column == Columns) {
276         ++Format.LineCount;
277         Column = 0;
278       }
279       if (Column == Columns - 1)
280         HasRowWithSufficientColumns = true;
281       unsigned Length =
282           (Column == Columns - 1) ? EndOfLineItemLength[i] : ItemLengths[i];
283       Format.ColumnSizes[Column] = std::max(Format.ColumnSizes[Column], Length);
284       MinSizeInColumn[Column] = std::min(MinSizeInColumn[Column], Length);
285       ++Column;
286     }
287     // If all rows are terminated early (e.g. by trailing comments), we don't
288     // need to look further.
289     if (!HasRowWithSufficientColumns)
290       break;
291     Format.TotalWidth = Columns - 1; // Width of the N-1 spaces.
292 
293     for (unsigned i = 0; i < Columns; ++i)
294       Format.TotalWidth += Format.ColumnSizes[i];
295 
296     // Don't use this Format, if the difference between the longest and shortest
297     // element in a column exceeds a threshold to avoid excessive spaces.
298     if ([&] {
299           for (unsigned i = 0; i < Columns - 1; ++i)
300             if (Format.ColumnSizes[i] - MinSizeInColumn[i] > 10)
301               return true;
302           return false;
303         }())
304       continue;
305 
306     // Ignore layouts that are bound to violate the column limit.
307     if (Format.TotalWidth > Style.ColumnLimit && Columns > 1)
308       continue;
309 
310     Formats.push_back(Format);
311   }
312 }
313 
314 const CommaSeparatedList::ColumnFormat *
315 CommaSeparatedList::getColumnFormat(unsigned RemainingCharacters) const {
316   const ColumnFormat *BestFormat = nullptr;
317   for (const ColumnFormat &Format : llvm::reverse(Formats)) {
318     if (Format.TotalWidth <= RemainingCharacters || Format.Columns == 1) {
319       if (BestFormat && Format.LineCount > BestFormat->LineCount)
320         break;
321       BestFormat = &Format;
322     }
323   }
324   return BestFormat;
325 }
326 
327 } // namespace format
328 } // namespace clang
329