Matt Bauman

@nanosoldier `runbenchmarks(ALL, vs=":master")` @nanosoldier `runtests()` Can we do both in the same comment?

I mean, the point of this PR was precisely to do this experiment. It is one possible fix, but not necessarily _the_ fix. > is there even any extant nontrivially...

Looking through [current uses](https://juliahub.com/ui/Search?q=CodeUnits&type=code) of the type specifically, it seems like `CodeUnits` has two purposes — or, rather, is commonly used in two contexts. It's either used to make a...

I just like to play with them during trainings and such to demonstrate what flux can do and the general structure of the models.

> scalar indices that are supported by array A This is exactly why the machinery passes `A` itself so deeply into the call stack here. Simple fix would be something...

Yes, that would work just as well with a lot less work. Since we can't specify versions in REQUIRE for lazy (and non-required) modules loaded by `@require`, I just wanted...

I think the biggest question is how you want to run the tests — is it part of a separate CI infrastructure? Or run via JuliaHub?

Yes, but this is a simpler case and could/should be handled directly and separately. It's likely that we're stashing the function into a tuple somewhere and just need a `::Type{T}`...

Somewhat related is https://github.com/JuliaLang/julia/issues/15648, but I like how this turns the problem on its head: instead of trying to wedge sparse matrices into a generic API, we simply "lift" dense...

No, unfortunately not. The default dense case is fixed, but the sparse infrastructure is still built around relying on inference. It's a pretty major refactor to fix it as I...