test ChainRules 1

[ChrisRackauckas]

No description provided.

[ChrisRackauckas]

Held back by TuringLang/DistributionsAD.jl#184

[ChrisRackauckas]

hmm. Maybe there's still something in the Flux stack holding back ChainRules 1 @DhairyaLGandhi @oxinabox ? It's getting pretty hard to track this one down haha.

[devmotion]

DistributionsAD is not compatible with CR1 yet. There's an open PR though: TuringLang/DistributionsAD.jl#198 The main issue is that the Zygote tests fail. It's discussed in more detail in TuringLang/DistributionsAD.jl#197 which is the basis of TuringLang/DistributionsAD.jl#198; I had hoped that the issues would be fixed by moving to a more recent Zygote version and CR1 but this was not the case. I wonder if we should just make another broken release - I assume the Zygote tests have been broken for a while but nobody noticed it since tests were canceled due to some other test failures caused by recent changes in Distributions (these issues are also fixed in the PR).

[DhairyaLGandhi]

I think @oxinabox is tracking the changes in CR 1 that broke the tests in DistributionsAD.

[oxinabox]

I am. Though my time for digging into them is limited. I hope to have time next week.

I agree it probably was not a new change, in ChainRulesCore or in Zygote.
It might have been a new rule in ChainRules, exposing a underlying problem in the Zygote-ChainRules interface.
Or just that the Zygote tests hadn't been run in so long.

[ChrisRackauckas]

Awesome.