views:

494

answers:

1

I've been having trouble articulating the differences between ILookup<TKey, TVal> and IGrouping<TKey, TVal>, and am curious if I understand it correctly now. LINQ compounded the issue by producing sequences of IGrouping items while also giving me a ToLookup extension method. So it felt like they were the same until I looked more closely.

var q1 = 
    from n in N
    group n by n.MyKey into g
    select g;
// q1 is IEnumerable<IGrouping<TKey, TVal>>

Which is equivalent to:

var q2 = N.GroupBy(n => n.MyKey, n => n);
// q2 is IEnumerable<IGrouping<TKey, TVal>>

Which looks a lot like:

var q3 = N.ToLookup(n => n.MyKey, n => n);
// q3 is ILookup<TKey, TVal>

Am I correct in the following analogies?

  1. An IGrouping<TKey, TVal> is a single group (i.e. a keyed sequence), analogous to KeyValue<TKey, TVal> where the value is actually a sequence of elements (rather than a single element)
  2. An IEnumerable<IGrouping<TKey, TVal>> is a sequence of those (similar to what you get when iterating over an IDictionary<TKey, TVal>
  3. An ILookup<TKey, TVal> is more like a IDictionary<TKey, TVal> where the value is actually a sequence of elements
+10  A: 

Yes, all of those are correct.

And ILookup<TKey, TValue> also extends IEnumerable<IGrouping<TKey, TValue>> so you can iterate over all the key/collection pairs as well as (or instead of) just looking up particular keys.

I basically think of ILookup<TKey,TValue> as being like IDictionary<TKey, IEnumerable<TValue>>.

Bear in mind that ToLookup is a "do it now" operation (immediate execution) whereas a GroupJoin is deferred. As it happens, with the way that "pull LINQ" works, when you start pulling IGroupings from the result of a GroupJoin, it has to read all the data anyway (because you can't switch group midway through) whereas in other implementations it may be able to produce a streaming result. (It does in Push LINQ; I would expect LINQ to Events to be the same.)

Jon Skeet
Thanks for the reply. I haven't come across thinking of Linq in terms of push/pull before. When I went to Google it I came across one of your blogs so I'll check that out. Sounds like an interesting way to think about it.
McKAMEY
I still think that these differences don't justify 2 different interfaces. I think that the Library designer should have decided for just one interface. Maybe just a personal taste that here is something not explicit enough.
rudimenter