Frames White

Odds are as a work around you need to either implement `FiniteDifferences.to_vec` or maybe `rand_tangent` (https://github.com/JuliaDiff/ChainRulesTestUtils.jl/blob/6925da14c12e3d743c8d3620db8a8bee1433d5c3/src/rand_tangent.jl#L8) for your type. I would need to see the stack trace to know which....

So the problem with the second one is just that it isn't inferrable. So **ChainRulesTestUtils is working correctly there.** `Tuple(totalpullback))` is not an inferable operation, because it can't tell how...

For the first case what is happening seems to to be that finite differencing is returning a result that contans a `tanh` element. Which is weird. Something is going wrong...

This makes sense. It should be a negligible change, and we can use the type of the input to determine the type we should use for `rtol`, at least for...

will review tomorrow

Is this ready for rereview?

I think this is actually do do with the way the Test stdlib rewrites stack-traces. I have gone digging before but i don't recall what my conclusion was. I think...

I also would like to get rid of `to_vec` per https://github.com/JuliaDiff/FiniteDifferences.jl/issues/97 by making FiniteDifferences stop using it and use ChainRules types instead. It may be that we need to incorporate...

Give this it's own page in the docs? How to Test an AD system using this. And put a subheading `Zygote`

Similar there are some other cases like `AbstractVector{