Add rrule for Diagonal * AbstractVector

[tkf]

ref #52 (comment)

Reading #103 and other source code, I suppose we are only defining rrules for Real element types at the moment so didn't put adjoint in the definition. (By the way, if that's the case, I think it makes sense to mention it in #103.)

[oxinabox]

I suppose we are only defining rrules for Real element types at the moment so didn't put adjoint in the definition. (By the way, if that's the case, I think it makes sense to mention it in #103.)

it is https://github.com/JuliaDiff/ChainRules.jl/pull/103/files#diff-493d89206c11fea99259d0771034d26aR283

We explictly don't conjugate inside rules.

[oxinabox]

For consistancy with the rest of the code, the intentation style is:

        rrule_test(
            *, randn(rng, N),
            (Diagonal(randn(rng, N)), Diagonal(randn(rng, N))),
            (randn(rng, N), randn(rng, N)),
        )

i.e. always start and end a multiline call with a line that has no arguments on it,
and always ident by 4, rather than indenting to line-up with the brace

[nickrobinson251]

actually it would be with one argument per line

rrule_test(
    *,
    randn(rng, N),
    (Diagonal(randn(rng, N)), Diagonal(randn(rng, N))),
    (randn(rng, N), randn(rng, N)),
)

[oxinabox]

Yes, but I feel like this is one of the cases where the clear relationship between
*, randn(rng, N), as inputs to the function makes it worth putting them on one-line.

I'ld be happy with either

[tkf]

Good to know that ChainRules.jl is using a sane code style!

[tkf]

OK. I think I reflected the comments in the new version. Travis is happy with it.

[oxinabox]

Thanks