Michael Abbott

Maybe it's ideal to do this at 1.1? Then in case anyone still wants 1.6, there will be a 1.0 release they can install. I wouldn't put much work into...

My fictitious user is someone trying to update bounds so that their CI works! Not creating tighter coupling than we have to seems polite, when it's really easy. Agree real...

Rearrange does `reshape` and `permutedims`. I guess this package does only `permutedims` at present. You might be interested in TensorCast.jl, which does such things. It even has tests here https://github.com/mcabbott/TensorCast.jl/pull/54/files...

Please include the error messages for both failures. Do they fail on CPU too?

I think the `getindex` error is a dup of https://github.com/FluxML/OneHotArrays.jl/issues/28 . Close if you agree?

Isn't this just how arrays work? E.g. one role of `Transpose` is to dispatch to the BLAS with a `'T'`, easy. The second role is to plug into generic code...

Can you clarify what `a` is, and what the dot in `((r[k]).a[j])[i] ` means? `r[1] isa Vector{Vector{Float64}}` which doesn't seem to have useful fields. From your second message, I think...

OK. This works the same way, in that the code to stack it seems happy with this level of recursion, but the code to infer the ranges of indices seems...

To be clear the method which copies to CPU should only be for `inds` which are arrays, which is where you have to worry about races. For simpler things like...

I think Kyle said he had a branch doing something like this, using ConstructionBase. That appears to be able to reconstruct closures which is neat: ```julia julia> adder = let...