Mohamed Tarek

>Also note that the reason I'm mentioning ComponentArrays is because the original plan was never for DI to support an arbitrary number of arguments: that's what we agreed AbstractDifferentation would...

>@mohamed82008 I don't think I have write access. Fixed :)

>How about we put both the AbstractDifferentiation paper and the future DI paper in the CITATION.bib here? That might be a reasonable way to give credit to the paper's authors...

> What do you think of the wording here? https://github.com/gdalle/DifferentiationInterface.jl?tab=readme-ov-file#citation Looks good. Thanks. I think the reference to AD.jl as "the inspiration" sufficiently gives credit to the AD.jl contributors and...

Not a fan of supporting multiple arguments with ForwardDiff. ForwardDiff's API does not support that.

hmm on second thought, I changed my mind

NonconvexUtils already supports this https://github.com/JuliaNonconvex/NonconvexUtils.jl/blob/main/src/forwarddiff_frule.jl#L1. You can just assume the inputs are all real/arrays and not recursive containers and simplify the implementation. I use flatten/unflatten but that's because I wanted...

someone can open a PR to check if the output of frule is nothing, before unpacking it https://github.com/ThummeTo/ForwardDiffChainRules.jl/blob/master/src/ForwardDiffChainRules.jl#L62

if it is nothing, throw an informative error saying that no frule is defined for f

Steps to reproduce it in Julia 1.5.3: 1. `] dev AlgebraicMultigrid` 2. Change the compat entry for `IterativeSolvers` in AMG's Project.toml file from `IterativeSolvers = "0.8.3"` to `IterativeSolvers = "0.9"`...