One interesting question is how we can get efficient implementations for different kinds of collections.

<br>

<br>STL has the concept of Forward, Bidirectional and Random access iterators, and different implementations of algorithm depending on the capabilities of the iterators (and underlying collection).

<br>

<br>Should collections use marker interfaces or something similar to indicate their capabilities? Attributes seems cool, but will they be fast enough?

<br>

<br>/Staffan

We simply use interfaces and normal overloading. VERY VERY ROUGHLY, IEnumerable&lt;T&gt; corresponds to InputIterator, ICollection&lt;T&gt; corresponds to ForwardIterator, and IList&lt;T&gt; corresponds to RandomAccessIterator.

<br>We use overloading when the same operation has different algorithms: for example, look at ToArray or RemoveWhenTrue.

I would like to see a partial sort - I use this extensively in statistical calculations where I need the 3% point or the median and can avoid the time for a full sort.

Why does FindFirst&lt;T&gt; return the found item as an out parameter? Wouldn't it be more convenient to define it &quot;public static T FindFirst&lt;T&gt;(IEnumerable&lt;T&gt; collection, Predicate&lt;T&gt; predicate)&quot;, and return null if no item matches the condition?

Thomas:

<br>

<br>if T is a value type (say int), then there is no reasonable value to return. One could return the default value (all zeros), but a) that's might be a real value in the collection, and b) for a user-defined struct, testing the return value for the all-zero struct could be quite inconvenient.