Regression in type inferrability

[gdalle]

When I bumped AbstractDifferentiation compat to 0.6 on ImplicitDifferentiation, I got type inference errors in ChainRulesTestUtils that were not there before. Probably a regression due to the latest changes. Any idea what might have worsened inference @devmotion ?
https://github.com/gdalle/ImplicitDifferentiation.jl/actions/runs/6659445596/job/18098532825

[devmotion]

Are you sure that these are caused by AbstractDifferentiation? Was the AbstractDifferentiation update the only change compared with e.g. runs on the master branch? I wonder because, e.g., the latest Zygote release (0.6.67) also contains a few more fundamental changes (getindex adjoints were removed in favour of ChainRules).

[gdalle]

I'll see if the tests still fail with the previous Zygote, thanks for the tip

[gdalle]

I just checked, reverting to Zygote v0.6.66 doesn't fix things, but reverting to AbstractDifferentiation v0.5.3 does.

[mohamed82008]

Can you post an MWE?

[gdalle]

Unfortunately it was buried inside the ImplicitDifferentiation.jl test suite, and I didn't have time to investigate, so I just deactivated the inference tests to be able to update to AD@0.6 in the compat