ForwardDiff Optimization test failing

[penelopeysm]

Turing.jl/test/optimisation/Optimisation.jl

Lines 634 to 639 in 0f46682

	elseif adbackend isa AutoForwardDiff
	# TODO: Figure out why this is happening.
	# https://github.com/TuringLang/Turing.jl/issues/2369
	@test_throws DivideError maximum_likelihood(m; adtype=adbackend)
	@test_throws DivideError maximum_a_posteriori(m; adtype=adbackend)
	else

Marking as @test_throws for now, but we should try to get to the bottom of it and re-enable the test.

[penelopeysm]

Note that this didn't happen in older versions of Turing, #2341 seems to have introduced this and it's not clear why (it might not be code changed in that PR, it might just generally be newer versions of deps).

It also seems very likely that this is an upstream error, not a Turing one, although I haven't looked at it in any depth.

[penelopeysm]

This failure has magically disappeared!

If I had to guess, it's something to do with how we were using AutoForwardDiff(; chunksize=0). As long as we don't do that, we should be fine. I think the new code in DynamicPPL will catch it: https://github.com/TuringLang/DynamicPPL.jl/blob/90c7b26c852b7d0ae87dee0a5a0010b097a0c1d3/ext/DynamicPPLForwardDiffExt.jl#L30-L35 Or to be precise, it should have caught it before, but somehow it wasn't happening.