timo there's also the opportunity to have a "proto sub" for a variadic C (or other native) function that has the information about "variadicness" on it, and then one or many multi candidates for specific annotations going from specific to general 00:09
could even be the same trait for proto sub vs only sub
japhb Oh now that's an interesting usage (variadicity as a proto trait). 00:27
timo i think it makes sense from a "hook to hang it on" but it'd be annoying to have to make a proto + at least one multi for any vararg native function 00:28
proto and multi do have the syntax to put a "don't use anything to the right for deciding the right multi candidate" which is kind of the opposite of what we want here 00:29
i think it's ;; in the parameter list
and of course we might want to have a multi where each sub is bound to a different native symbol, so forcing the trait to be on the proto wouldn't be nice for that 00:30
japhb Hmmm, decent point. 00:35
I feel like it would be useful to get as close to `sub foo(*@args) is native {}` as we can would be nice, because slurpy arrays are the closest equivalent I can think of for variadics. 00:39
One possible way to think of it might be a sub-Signature applied to a Capture. 00:40
`sub foo(|c (Str, Str, Int)) is native { }` 00:41
nine Variadic calls don't differ from normal calls very much. In the end, it only comes down to the number of 64 bit integer and 128 bit float arguments respectively. There are no other argument types (on x86-64 but from a quick glance, ARM is similar). 07:46
We may need some extra information for Raku though like the mentioned is rw or string encodings. 07:48
In any case, these are certainly properties of the call site. After all the point of variadic functions is that the argument list can be different for each call.
patrickb Can we move this discussion to github.com/Raku/problem-solving/issues/473 12:21
I hope that makes keeping track of the different ideas easier. 12:22
nine: Yes, it's not much different. In dyncall it's two additional calls to dcMode() to set the mode before the first fixed arg and before the first dynamic arg. 12:23
patrickb But the question I think needs to be solved is where the types of the dynamic args come from. I agree it's a thing of the call site as long as we don't want to reduce the system in so far as having to specify each different type variant in a new declaration. (That's approach 1. in that p-s issue). 12:26
timo ah, amd64 system V abi has the number of floating point arguments inside the rax register 15:59

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!