brintos

brintos / llvm-project-archived public Read only

0
0
Text · 13.1 KiB · 28fdbcb Raw
359 lines · 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-exception6//7//===----------------------------------------------------------------------===//8///9/// \file10/// This file implements specific functions of \c FormatTokens and their11/// roles.12///13//===----------------------------------------------------------------------===//14 15#include "FormatToken.h"16#include "ContinuationIndenter.h"17#include "llvm/ADT/SmallVector.h"18#include <climits>19 20namespace clang {21namespace format {22 23const char *getTokenTypeName(TokenType Type) {24  static const char *const TokNames[] = {25#define TYPE(X) #X,26      LIST_TOKEN_TYPES27#undef TYPE28      nullptr};29 30  if (Type < NUM_TOKEN_TYPES)31    return TokNames[Type];32  llvm_unreachable("unknown TokenType");33  return nullptr;34}35 36static constexpr std::array<StringRef, 14> QtPropertyKeywords = {37    "BINDABLE",   "CONSTANT", "DESIGNABLE", "FINAL", "MEMBER",38    "NOTIFY",     "READ",     "REQUIRED",   "RESET", "REVISION",39    "SCRIPTABLE", "STORED",   "USER",       "WRITE",40};41 42bool FormatToken::isQtProperty() const {43  assert(llvm::is_sorted(QtPropertyKeywords));44  return llvm::binary_search(QtPropertyKeywords, TokenText);45}46 47// Sorted common C++ non-keyword types.48static constexpr std::array<StringRef, 14> CppNonKeywordTypes = {49    "clock_t",  "int16_t",   "int32_t", "int64_t",   "int8_t",50    "intptr_t", "ptrdiff_t", "size_t",  "time_t",    "uint16_t",51    "uint32_t", "uint64_t",  "uint8_t", "uintptr_t",52};53 54bool FormatToken::isTypeName(const LangOptions &LangOpts) const {55  if (is(TT_TypeName) || Tok.isSimpleTypeSpecifier(LangOpts))56    return true;57  assert(llvm::is_sorted(CppNonKeywordTypes));58  return (LangOpts.CXXOperatorNames || LangOpts.C11) && is(tok::identifier) &&59         llvm::binary_search(CppNonKeywordTypes, TokenText);60}61 62bool FormatToken::isTypeOrIdentifier(const LangOptions &LangOpts) const {63  return isTypeName(LangOpts) || isOneOf(tok::kw_auto, tok::identifier);64}65 66bool FormatToken::isBlockIndentedInitRBrace(const FormatStyle &Style) const {67  assert(is(tok::r_brace));68  assert(MatchingParen);69  assert(MatchingParen->is(tok::l_brace));70  if (Style.Cpp11BracedListStyle == FormatStyle::BLS_Block ||71      !Style.BreakBeforeCloseBracketBracedList) {72    return false;73  }74  const auto *LBrace = MatchingParen;75  if (LBrace->is(BK_BracedInit))76    return true;77  if (LBrace->Previous && LBrace->Previous->is(tok::equal))78    return true;79  return false;80}81 82bool FormatToken::opensBlockOrBlockTypeList(const FormatStyle &Style) const {83  // C# Does not indent object initialisers as continuations.84  if (is(tok::l_brace) && getBlockKind() == BK_BracedInit && Style.isCSharp())85    return true;86  if (is(TT_TemplateString) && opensScope())87    return true;88  return is(TT_ArrayInitializerLSquare) || is(TT_ProtoExtensionLSquare) ||89         (is(tok::l_brace) &&90          (getBlockKind() == BK_Block || is(TT_DictLiteral) ||91           (Style.Cpp11BracedListStyle == FormatStyle::BLS_Block &&92            NestingLevel == 0))) ||93         (is(tok::less) && Style.isProto());94}95 96TokenRole::~TokenRole() {}97 98void TokenRole::precomputeFormattingInfos(const FormatToken *Token) {}99 100unsigned CommaSeparatedList::formatAfterToken(LineState &State,101                                              ContinuationIndenter *Indenter,102                                              bool DryRun) {103  if (!State.NextToken || !State.NextToken->Previous)104    return 0;105 106  if (Formats.size() <= 1)107    return 0; // Handled by formatFromToken (1) or avoid severe penalty (0).108 109  // Ensure that we start on the opening brace.110  const FormatToken *LBrace =111      State.NextToken->Previous->getPreviousNonComment();112  if (!LBrace || LBrace->isNoneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||113      LBrace->is(BK_Block) || LBrace->is(TT_DictLiteral) ||114      LBrace->Next->is(TT_DesignatedInitializerPeriod)) {115    return 0;116  }117 118  // Calculate the number of code points we have to format this list. As the119  // first token is already placed, we have to subtract it.120  unsigned RemainingCodePoints =121      Style.ColumnLimit - State.Column + State.NextToken->Previous->ColumnWidth;122 123  // Find the best ColumnFormat, i.e. the best number of columns to use.124  const ColumnFormat *Format = getColumnFormat(RemainingCodePoints);125 126  // If no ColumnFormat can be used, the braced list would generally be127  // bin-packed. Add a severe penalty to this so that column layouts are128  // preferred if possible.129  if (!Format)130    return 10'000;131 132  // Format the entire list.133  unsigned Penalty = 0;134  unsigned Column = 0;135  unsigned Item = 0;136  while (State.NextToken != LBrace->MatchingParen) {137    bool NewLine = false;138    unsigned ExtraSpaces = 0;139 140    // If the previous token was one of our commas, we are now on the next item.141    if (Item < Commas.size() && State.NextToken->Previous == Commas[Item]) {142      if (!State.NextToken->isTrailingComment()) {143        ExtraSpaces += Format->ColumnSizes[Column] - ItemLengths[Item];144        ++Column;145      }146      ++Item;147    }148 149    if (Column == Format->Columns || State.NextToken->MustBreakBefore) {150      Column = 0;151      NewLine = true;152    }153 154    // Place token using the continuation indenter and store the penalty.155    Penalty += Indenter->addTokenToState(State, NewLine, DryRun, ExtraSpaces);156  }157  return Penalty;158}159 160unsigned CommaSeparatedList::formatFromToken(LineState &State,161                                             ContinuationIndenter *Indenter,162                                             bool DryRun) {163  // Formatting with 1 Column isn't really a column layout, so we don't need the164  // special logic here. We can just avoid bin packing any of the parameters.165  if (Formats.size() == 1 || HasNestedBracedList)166    State.Stack.back().AvoidBinPacking = true;167  return 0;168}169 170// Returns the lengths in code points between Begin and End (both included),171// assuming that the entire sequence is put on a single line.172static unsigned CodePointsBetween(const FormatToken *Begin,173                                  const FormatToken *End) {174  assert(End->TotalLength >= Begin->TotalLength);175  return End->TotalLength - Begin->TotalLength + Begin->ColumnWidth;176}177 178void CommaSeparatedList::precomputeFormattingInfos(const FormatToken *Token) {179  // FIXME: At some point we might want to do this for other lists, too.180  if (!Token->MatchingParen ||181      Token->isNoneOf(tok::l_brace, TT_ArrayInitializerLSquare)) {182    return;183  }184 185  // In C++11 braced list style, we should not format in columns unless they186  // have many items (20 or more) or we allow bin-packing of function call187  // arguments.188  if (Style.Cpp11BracedListStyle != FormatStyle::BLS_Block &&189      !Style.BinPackArguments &&190      (Commas.size() < 19 || !Style.BinPackLongBracedList)) {191    return;192  }193 194  // Limit column layout for JavaScript array initializers to 20 or more items195  // for now to introduce it carefully. We can become more aggressive if this196  // necessary.197  if (Token->is(TT_ArrayInitializerLSquare) && Commas.size() < 19)198    return;199 200  // Column format doesn't really make sense if we don't align after brackets.201  if (!Style.AlignAfterOpenBracket)202    return;203 204  FormatToken *ItemBegin = Token->Next;205  while (ItemBegin->isTrailingComment())206    ItemBegin = ItemBegin->Next;207  SmallVector<bool, 8> MustBreakBeforeItem;208 209  // The lengths of an item if it is put at the end of the line. This includes210  // trailing comments which are otherwise ignored for column alignment.211  SmallVector<unsigned, 8> EndOfLineItemLength;212  MustBreakBeforeItem.reserve(Commas.size() + 1);213  EndOfLineItemLength.reserve(Commas.size() + 1);214  ItemLengths.reserve(Commas.size() + 1);215 216  bool HasSeparatingComment = false;217  for (unsigned i = 0, e = Commas.size() + 1; i != e; ++i) {218    assert(ItemBegin);219    // Skip comments on their own line.220    while (ItemBegin->HasUnescapedNewline && ItemBegin->isTrailingComment()) {221      ItemBegin = ItemBegin->Next;222      HasSeparatingComment = i > 0;223    }224 225    MustBreakBeforeItem.push_back(ItemBegin->MustBreakBefore);226    if (ItemBegin->is(tok::l_brace))227      HasNestedBracedList = true;228    const FormatToken *ItemEnd = nullptr;229    if (i == Commas.size()) {230      ItemEnd = Token->MatchingParen;231      const FormatToken *NonCommentEnd = ItemEnd->getPreviousNonComment();232      ItemLengths.push_back(CodePointsBetween(ItemBegin, NonCommentEnd));233      if (Style.Cpp11BracedListStyle != FormatStyle::BLS_Block &&234          !ItemEnd->Previous->isTrailingComment()) {235        // In Cpp11 braced list style, the } and possibly other subsequent236        // tokens will need to stay on a line with the last element.237        while (ItemEnd->Next && !ItemEnd->Next->CanBreakBefore)238          ItemEnd = ItemEnd->Next;239      } else {240        // In other braced lists styles, the "}" can be wrapped to the new line.241        ItemEnd = Token->MatchingParen->Previous;242      }243    } else {244      ItemEnd = Commas[i];245      // The comma is counted as part of the item when calculating the length.246      ItemLengths.push_back(CodePointsBetween(ItemBegin, ItemEnd));247 248      // Consume trailing comments so the are included in EndOfLineItemLength.249      if (ItemEnd->Next && !ItemEnd->Next->HasUnescapedNewline &&250          ItemEnd->Next->isTrailingComment()) {251        ItemEnd = ItemEnd->Next;252      }253    }254    EndOfLineItemLength.push_back(CodePointsBetween(ItemBegin, ItemEnd));255    // If there is a trailing comma in the list, the next item will start at the256    // closing brace. Don't create an extra item for this.257    if (ItemEnd->getNextNonComment() == Token->MatchingParen)258      break;259    ItemBegin = ItemEnd->Next;260  }261 262  // Don't use column layout for lists with few elements and in presence of263  // separating comments.264  if (Commas.size() < 5 || HasSeparatingComment)265    return;266 267  if (Token->NestingLevel != 0 && Token->is(tok::l_brace) && Commas.size() < 19)268    return;269 270  // We can never place more than ColumnLimit / 3 items in a row (because of the271  // spaces and the comma).272  unsigned MaxItems = Style.ColumnLimit / 3;273  SmallVector<unsigned> MinSizeInColumn;274  MinSizeInColumn.reserve(MaxItems);275  for (unsigned Columns = 1; Columns <= MaxItems; ++Columns) {276    ColumnFormat Format;277    Format.Columns = Columns;278    Format.ColumnSizes.resize(Columns);279    MinSizeInColumn.assign(Columns, UINT_MAX);280    Format.LineCount = 1;281    bool HasRowWithSufficientColumns = false;282    unsigned Column = 0;283    for (unsigned i = 0, e = ItemLengths.size(); i != e; ++i) {284      assert(i < MustBreakBeforeItem.size());285      if (MustBreakBeforeItem[i] || Column == Columns) {286        ++Format.LineCount;287        Column = 0;288      }289      if (Column == Columns - 1)290        HasRowWithSufficientColumns = true;291      unsigned Length =292          (Column == Columns - 1) ? EndOfLineItemLength[i] : ItemLengths[i];293      Format.ColumnSizes[Column] = std::max(Format.ColumnSizes[Column], Length);294      MinSizeInColumn[Column] = std::min(MinSizeInColumn[Column], Length);295      ++Column;296    }297    // If all rows are terminated early (e.g. by trailing comments), we don't298    // need to look further.299    if (!HasRowWithSufficientColumns)300      break;301    Format.TotalWidth = Columns - 1; // Width of the N-1 spaces.302 303    for (unsigned i = 0; i < Columns; ++i)304      Format.TotalWidth += Format.ColumnSizes[i];305 306    // Don't use this Format, if the difference between the longest and shortest307    // element in a column exceeds a threshold to avoid excessive spaces.308    if ([&] {309          for (unsigned i = 0; i < Columns - 1; ++i)310            if (Format.ColumnSizes[i] - MinSizeInColumn[i] > 10)311              return true;312          return false;313        }()) {314      continue;315    }316 317    // Ignore layouts that are bound to violate the column limit.318    if (Format.TotalWidth > Style.ColumnLimit && Columns > 1)319      continue;320 321    Formats.push_back(Format);322  }323}324 325const CommaSeparatedList::ColumnFormat *326CommaSeparatedList::getColumnFormat(unsigned RemainingCharacters) const {327  const ColumnFormat *BestFormat = nullptr;328  for (const ColumnFormat &Format : llvm::reverse(Formats)) {329    if (Format.TotalWidth <= RemainingCharacters || Format.Columns == 1) {330      if (BestFormat && Format.LineCount > BestFormat->LineCount)331        break;332      BestFormat = &Format;333    }334  }335  return BestFormat;336}337 338bool startsNextParameter(const FormatToken &Current, const FormatStyle &Style) {339  assert(Current.Previous);340  const auto &Previous = *Current.Previous;341  if (Current.is(TT_CtorInitializerComma) &&342      Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) {343    return true;344  }345  if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName))346    return true;347  if (Current.is(TT_QtProperty))348    return true;349  return Previous.is(tok::comma) && !Current.isTrailingComment() &&350         ((Previous.isNot(TT_CtorInitializerComma) ||351           Style.BreakConstructorInitializers !=352               FormatStyle::BCIS_BeforeComma) &&353          (Previous.isNot(TT_InheritanceComma) ||354           Style.BreakInheritanceList != FormatStyle::BILS_BeforeComma));355}356 357} // namespace format358} // namespace clang359