Michael Abbott

I'm not sure I follow, but I was not envisaging providing a way to specify an arbitrary list of types. You could still test any one particular weird thing as...

Sorry, by "strange types" I mean the various non-Array types this tests (from LinearAlgebra, plus Base's view/permuted/reshaped etc.). Those are the main goal here, but perhaps a second mode which...

OK. The reason I wasn't so keen on the automated approach is that accidentally writing the signature more narrowly than you intended seems like an easy mistake to make (e.g....

To summarise what this does, tagged version is like so: ```julia julia> using ForwardDiff: Dual julia> pr(x, op, y) = println("x ", op, " y = ", op(x,y)); julia> pr.(Dual(1,1),...

Sorry just went by your pictures, where after used it for two kinds of change. x to x_eps, and x_eps to f(x_eps). It seems the original commits the same sin,...

Could we just fix IntervalSets to use `isless` etc, and bump its version number? I doubt that anyone was deeply reliant on these differences of behaviour, and cohesion is worth...

> I think the basic interface needed is a nice gradient function. Enzyme's own `gradient` should now do this, as `make_zero` understands nested structures: ```julia julia> sh = [1f0, 2f0];...

What is this used for? And what is the intended behaviour, on arrays of various dimensions? I'm not sure the code matches the words, nor the example given. ```julia julia>...

What output would you like here? There is no data, of course: ``` julia> x = randn(Float32, 4, 0, 6) 4×0×6 Array{Float32, 3} ``` If you're trying to propagate size...

Yes that's not crazy, I suspect that this one could be made to work without too much effort. But there are likely to be many other places where similar things...