delphi - 使用 TDictionary "for...in"

Question

我正在使用 Delphi Berlin 10.1(最新更新)，并且我的应用程序中的 TDictionary 遇到一些特定值的问题。 “for..in”无法正确循环。

下面是一个示例代码，其中“for...in”不循环遍历所有值，另一个示例则循环遍历所有值。

在第一种情况下，“for...in”循环仅执行两个步骤，而在第二种情况下，它会执行所有步骤。

procedure TForm1.btn1Click(Sender: TObject);
var
  tmpPar: TPair<Integer, Integer>;
  tmpDictionary: TDictionary<Integer, Integer>;
begin
  // NOT WORKING
  tmpDictionary := TDictionary<Integer, Integer>.Create;
  try
      tmpDictionary.Add(631, 40832);
      tmpDictionary.Add(1312, 40837);
      tmpDictionary.Add(5947, 40842);

      for tmpPar in tmpDictionary do
      begin
          tmpDictionary.Remove(tmpPar.Key);
      end;
  finally
      tmpDictionary.Free;
  end;

  // WORKING
  tmpDictionary := TDictionary<Integer, Integer>.Create;
  try
      tmpDictionary.Add(123, 5432);
      tmpDictionary.Add(453, 23);
      tmpDictionary.Add(76, 2334);

      for tmpPar in tmpDictionary do
      begin
          tmpDictionary.Remove(tmpPar.Key);
      end;
  finally
      tmpDictionary.Free;
  end;
end;

第一种情况有问题吗？

提前致谢!

Answer 1

你的例子只是运气好而已 - 不应该期望这个构造会表现良好。如果您单步执行示例，您会发现第一种情况在删除时调用列表重新排序，但第二个示例则不会。

要了解发生了什么，请检查从字典中删除项目的代码:

function TDictionary<TKey,TValue>.DoRemove(const Key: TKey; HashCode: Integer;
  Notification: TCollectionNotification): TValue;
var
  gap, index, hc, bucket: Integer;
  LKey: TKey;
begin
  index := GetBucketIndex(Key, HashCode);
  if index < 0 then
    Exit(Default(TValue));

  // Removing item from linear probe hash table is moderately
  // tricky. We need to fill in gaps, which will involve moving items
  // which may not even hash to the same location.
  // Knuth covers it well enough in Vol III. 6.4.; but beware, Algorithm R
  // (2nd ed) has a bug: step R4 should go to step R1, not R2 (already errata'd).
  // My version does linear probing forward, not backward, however.

  // gap refers to the hole that needs filling-in by shifting items down.
  // index searches for items that have been probed out of their slot,
  // but being careful not to move items if their bucket is between
  // our gap and our index (so that they'd be moved before their bucket).
  // We move the item at index into the gap, whereupon the new gap is
  // at the index. If the index hits a hole, then we're done.

  // If our load factor was exactly 1, we'll need to hit this hole
  // in order to terminate. Shouldn't normally be necessary, though.
  {...   etc   ...}

您会看到，实现了一种算法，该算法决定在删除项目时何时以及如何对基础列表进行重新排序(这是为了尝试优化已分配内存块中间隙的位置以优化将来的插入)。枚举只是在基础列表中的索引中移动，因此一旦您从列表中删除一个项目，枚举器就不再有效，因为它只会将您移动到基础列表中的下一个索引，该索引已更改。

对于普通列表，删除时通常会反向迭代。但是，对于字典，您必须首先构建要在第一次枚举过程中删除的键列表，然后枚举该列表以将它们从字典中删除。