Fix function in call is S3 matching

[mschubert]

If b is a do.call(function(...)), the is_S3_user_generic will check a function in the next recursion.
Because this can not be indexed by b[[1]], we get the following error:

Error in b[[1]] : object of type 'closure' is not subsettable

The PR additionally checks if b is a function, and gets its body if it is.

Alternatively, if b is a call and not S3, do we need to check its contents for S3?

[klmr]

Could you provide an MWE that illustrates the failure? … ideally as a unit test that fails without the PR and greenlights afterwards, but this isn’t necessary.

At any rate, the body call in https://github.com/klmr/modules/blob/9581658ad69622d4091dc2dcd01356398c82f9c2/R/S3.r#L57 can probably be omitted now, right?

[mschubert]

I was afraid you'd say that 😄

I tried. And after an hour I gave up. It's got something to do with a function in an environment attached to a closure, but I couldn't easily reproduce it. However, that is the reason why currently import('ebits/stats') fails (which is a mess by the way, I should rewrite that 😇).

And yes, the outer body call could be omitted then.

[klmr]

Argh, that’s unfortunate. I’ll take a look as soon as I get a minute.

[klmr]

Alright, here’s an MWE:

# x.r
f = function () NULL
body(f) = substitute(do.call(g, list()), list(g = function () 24))

I don’t think there’s a way of achieving the same without manipulating the function’s body at runtime (but I’d love to be proven wrong). Now on to create a test case.