>      > Another that it returns a value corresponding to how many non-keyword
>      > arguments are allowed (assuming no keywords arguments are required).
>
>     This sounds like the right thing.
>
> This makes the attempts to portably implement keyword arguments in SRFI
> 171 impossible, doesn't it?

As far as I can tell, it would still work. The portable implementation
passes the keyword args using a rest argument, so `procedure-arity-mask`
would return all 1 bits. If there are also N positional arguments, the
lowest N bits would be carved out as zeros. So it would give slightly
misleading information about the rest argument in that case but nothing
catastrophic.