clang/lib/Format/FormatToken.cpp

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