Valentin Churavy

Would also be good to figure out why the error printing is messed up `ERROR: Enzyme.Compiler.EnzymeNoDerivativeError(Cstring(0x00007f4605d3885c))`

> However, I am sure there are many direct .file field uses which would all instead have to go through the depot resolve step instead. We should do that during...

So the overarching design consideration is: Users of Enzyme.jl/CUDA.jl/AMDGPU.jl should be able to "precompile" their code. Where can we store these precompilation results, while also ensuring that they get invalidated...

Just quickly documenting my recommedation here ``` julia> struct VTable{T} funcs::T end julia> @generated function (VT::VTable{T})(fidx, args...) where T N = length(T.parameters) quote Base.Cartesian.@nif $(N+1) d->fidx==d d->return VT.funcs[d](args...) d->error("fidx oob")...

I think ScopedValue is what we semantically want, but we would need to have https://github.com/JuliaLang/julia/issues/53584#issuecomment-1979089217 fixed, potentially by https://github.com/JuliaLang/julia/pull/55045

Yeah that would make cases like https://github.com/JuliaLang/julia/discussions/48378 workable, and maybe better the MPI situation as well.

`libsig` is GPLv2 licensed as part of the OpenJDK, but the idea is fairly simple for someone to implement.

Hm should be defined here https://github.com/EnzymeAD/Enzyme.jl/blob/5baf187dd2086aa47954e6e564b1bb244307a696/lib/EnzymeCore/ext/AdaptExt.jl#L7 edit: https://github.com/EnzymeAD/Enzyme.jl/pull/1232#issuecomment-2022847656 @maleadt was your test on 1.8?

We should also standardize the extension name https://github.com/JuliaGPU/CUDA.jl/pull/2281

The generated headers we could ship, the generated cpp get compiled with the rest of the code. We don't check them into the source tree