I noticed that real(::Missing) and imag(::Missing) weren't implemented (threw an error) and so I added them.

Unless they were left off intentionally.

Also reminds me we need to finish #25502

@dlfivefifty, any interest in just adding #25502 (comment) into this PR? These changes are so small...

any interest in just adding #25502 (comment)

Isn't that PR fine as is?

At this rate of adding special missing methods I feel like we should instead have something like (::Function)(::Missing) = missing instead of go digging for methods that dont have their missing method defined.

At this rate of adding special missing methods I feel like we should instead have something like (::Function)(::Missing) = missing instead of go digging for methods that dont have their missing method defined.

That's been evoked at least since Nullable was introduced. Indeed that would make a lot of sense for one-argument methods, but then it's more difficult when there are several arguments as we have no way currently to express "at least one argument is missing". We would also need a special mechanism to prevent ambiguities from arising everywhere. In theory this could be introduced in 1.x without breaking code, at least if methods with arguments marked as ::Any take precedence over the missing fallback.

There's also the objection that missing generally only makes sense where a scalar would be passed. It's probably better to have det(missing) or permutedims(missing) throw an error than return missing (though in practice you'll likely get an error soon enough anyway).

I think a less dangerous approach would be to keep an official list of "math" functions a la Calculus.symbolic_derivatives_1arg that can be looped over.

I'll remind everybody here of the terms of the missing deal: I was assured that those who wanted missing would content themselves with a fixed set of commonly-used functions (as in e.g. C#), and use some kind of manual lifting for everything else.