Andy Ferris

Yes that's a pragmatic approach. The other thing I've been doing in my application code is using a finite `Union` of the concrete types the app _actually uses_ instead of...

Ok I think we are getting to the heart of matters. I'm not sure how helpful I can be though! > I'm curious to hear your thoughts on what the...

The other thing is AFAIK there are few advantages to `AbstractMatrix{T}` compared to `Any`. Except for `ndims` and `eltype`, Julia will still need to read the actual type for every...

Hi @jariji, I suppose this was intentional behavior, though I haven't thought too strongly about `UnorderedDictionary` _in particular_. The package exports `isdictequal` that you can use in this case, and...

> Unordered dictionaries with unordered comparison are a nice data type for certain use cases. 100% agreed. Julia has `issetequal` and I added `isdictequal` (because by default `AbstractDictionary` does not...

> but we would need a matching hash method The `hash` funciton has to match `isequal`, not `==`. I'd be pretty keen to keep `isequal` behaving as it is. >...

Okay, this one is because internally we use `zip` and iterate that. Originally `zip` would error out when the collections had different lengths, but now it is more forgiving (I...

For sure - we can just do that check a little earlier with `HasLength` than without.

You want a dictionary where you can choose the AbstractVector holding the values? (And leave Indices as-is?) I think we could do something. The other thing we need to do...

Yes - though likely it will be a handful of token-based methods (like `deletetokenvalue!` or something, which directly gives you the index to your `AbstractVector` rather than the key). Note...