Michael Abbott

`atan2` was in Base a long long time ago, and doesn't exist anymore, really that should be deleted entirely. For `gamma`, when were these methods added to SpecialFunctions? I don't...

Maybe it's enough to test `hasmethod(gamma, Tuple{Number, Number})` in the tests. It doesn't quite belong in `non_numeric_arg_functions` , and unlike `rem2pi` you don't really want a different test, you just...

Same test in Symbolics.jl failing on CI for an unrelated PR: https://github.com/JuliaSymbolics/Symbolics.jl/pull/373/checks?check_run_id=3563295246 Ditto for ModelingToolkit.jl failure: https://github.com/SciML/ModelingToolkit.jl/pull/1252/checks?check_run_id=3528813219

I think the more useful result is to pick a something like a sub-gradient. The use case to imagine is some optimisation problem, where equality might be the best or...

This sounds OK to me, better than not having the rules, or unconditionally loading. I don't see a way to make a very friendly error message, sadly... it's a pity...

Thanks for looking. I thought so but indeed we should re-check. ForwardDiff v0.10.18 and below have the error if there is a trinary rule. This is the line upper-bounding their...

Oh no, I thought I was linking to a commit, not just a line number. No, Fix1 is another thing. Ideally it would perturb it, as you say, and deduce...

If breaking things, you could consider a pattern like ``` @test_rrule cat(rand(1), rand(3, 2, 1); dims=(1,2)) check_inferred=false ``` instead of ``` test_rrule(cat, rand(1), rand(3, 2, 1); fkwargs=(dims=(1,2),), check_inferred=false) ``` Then...

Testing generic code more deeply is a great idea, but I'd vote for a less adversarial phrasing phrasing here. We're not trying to keep hackers away from some automated bug...

Here's a sketch of how a testing thing might work: https://github.com/mcabbott/ChainRulesTestUtils.jl/commit/6964a693642f3639ec8158b192fdccd027f625e3 Maybe someone with more of the test set loaded into cortex can say whether this seems like the right...