James Liu

> Is it feasible to cull empty archetypes, or re-order archetypes so that the full ones are always at the end (i.e. so we can make the check for emptyness...

Revisiting this, it may make sense to add this only for the parallel case, where the work per non-empty archetype is definitively non-trivial since it needs to allocate for each...

> I think this should be done before merging, which would make this non-controversial IMO. This was done before this comment. Should be ready to go.

I'm also seeing the same thing for on a Sony Bravia TV with the app as well. Unchecked the aforementioned boxes and it is still showing suggestions from the hidden...

Ultimately this is reliant on the following: - Seeing if Epic has support for purely P2P connections in Epic Online Services, - Are these features accessible in the current C#...

A correct implementation isn't particularly simple to achieve. We're still ironing out the details of how to use it ourselves. We'll have an updated package and documentation once we confirm...

Sorry, there has basically be no new developments since then. I haven't had too much time to dedicate to this right now, though I do intend to flesh out the...

Unfortunately, I've had zero bandwidth in the previous few months. To answer your immediate question, Backroll is P2P only. However, if the underlying transport layer utilizes relay networks via HouraiNetworking...

Reran the benchmarks. Some insights: - Contains is definitely slower due to the extra on-the-stack vs allocated check smallvec has to do. - As expected a nearly 2x speedup on...

@Calandiel ended up removing the smallvec changes, it's indeed an unacceptable overhead. Reran benchmarks, seems to address the `insert` and `all_ones` cases. ``` name master ns/iter usize ns/iter diff ns/iter...