David Widmann
David Widmann
Was fixed by #163.
I guess this is still waiting for #78? I think it would be good to also make ChainRulesCore a weak dependency on new Julia versions (see, e.g., how it is...
I updated the PR (and fixed a few issues with types and functions that were not available in the extension). Since in-place plans are not supported (or at least not...
@sethaxen since you were interested in differentiation rules as well (#103), I guess it might be valuable if you would have a look at the PR before I go ahead...
> Now that adjoints are defined at the *(::Plan, ::AbstractArray) level, can the rules for fft, rfft etc be removed? I'd argue no, they should be kept. Zygote also defined...
Yeah, I was thinking about that. Given that generally the ecosystem tries to keep dependencies minimal, I agree that it's maybe not completely optimal to include and precompile them even...
Hmm, I came to realize that it won't work this way even if tests would pass on Github since it means you would have to add AbstractFFTsTestUtils manually every time...
Actually, I started to dislike this subpackage approach more and more. Therefore I commented on https://github.com/JuliaIO/TranscodingStreams.jl/pull/133 to get some additional input.
I think a completely separate package is a suboptimal approach since it will be problematic to keep them in sync - changes to the API here mean tests could start...
> And in practice we would likely make the two PRs concurrently and at least via local testing we would know that the tests are all going to pass after...