http://jira.nhibernate.org/browse/NH-300

I wonder why not just add some code in loader\loader.cs like this,

around line 316, modify like this:

The main reason is it will break existing code. Also, your code is not as efficient as using a set to collect results.

Do you mean it will not be compatible to existing code using NHibernate?

I suppose the existing code would take the duplicated results and loop them to find those distinct using the same way "object.ReferenceEquals()", cause who will need the duplicate object instance which are reference equal?

For the second reason, I am not quite understand. Could you please explain more?

Thanks

Yes, and it will not work like Hibernate. I can't just change the de-facto standard behavior, even if many people find it useless. The best solution would probably be to create a method named .Set instead of .List which would do the distinctificaton and return a set of results.

As for why your implementation is inefficient - you traverse the list of all previously loaded objects for each new object. This means something like quadratic complexity (n items traversed for each item loaded * n items loaded, bad).

If you use a set, then inserting an object in a set takes log n time, so the whole complexity would be n*log n (not so bad). Or it could be even faster if you just compared the new item to the last loaded item. This would be linear complexity (best), but it would only work if the incoming results were sorted.