Feature: Add RFC for new gencap, #write.

[jemc]

Rendered.

[Praetonus]

👍

[plietar]

Sounds reasonable. Just a few points,

Do we also need an extra gencap for #write! = { ref, box, tag } ? (I can't think of any meaningful name for that set)
https://github.com/ponylang/ponyc/blob/master/src/libponyc/type/cap.c seems like we would need it

#read does not allow iso and trn even though they are readable, but #write would. This could be surprising (I was really confused as to why #read did not allow these when I started looking into generics)

Do we want a way to write the following, where this could be one of ref, trn or iso, depending on the receiver capability ?

fun #write replace(x: A) : this->A^ =>
  this.contents = consume x

[Praetonus]

Do we want a way to write the following, where this could be one of ref, trn or iso, depending on the receiver capability ?

I think cases like that are already covered by automatic receiver recovery.

[jemc]

#read does not allow iso and trn even though they are readable, but #write would. This could be surprising (I was really confused as to why #read did not allow these when I started looking into generics)

Yeah, I remember being struck surprised by this as well. However, I'm not sure what could/should be done 🤔

Do we also need an extra gencap for #write! = { ref, box, tag } ? (I can't think of any meaningful name for that set)
https://github.com/ponylang/ponyc/blob/master/src/libponyc/type/cap.c seems like we would need it

Hm, I'm not sure where you think we need it. You're probably right, and I'm probably wrong, but I already implemented my first draft of this in a branch, and didn't use such a set. We should go over this again in code review for the implementation.

[plietar]

Do we want a way to write the following, where this could be one of ref, trn or iso, depending on the receiver capability ?

I think cases like that are already covered by automatic receiver recovery.

I was thinking more about reflecting the receiver capability in the this viewpoint, but I've realised ref is the most flexible viewpoint anyway, so there's not much point doing that.

Do we also need an extra gencap for #write! = { ref, box, tag } ? (I can't think of any meaningful name for that set) https://github.com/ponylang/ponyc/blob/master/src/libponyc/type/cap.c seems like we would need it

Hm, I'm not sure where you think we need it. You're probably right, and I'm probably wrong, but I already implemented my first draft of this in a branch, and didn't use such a set. We should go over this again in code review for the implementation

I'm talking about the cap_aliasing function, which takes a (gen)cap and returns its alias. Today, all existing gencap have a alias (#any->#alias, #send->#share, others don't change), but #write would not.

This can be seen as an implementation detail, with a #writealiased gencap if it is indeed necessary, which wouldn't have to be expose to source programs.

I've always seen this is the reason #alias exists. The documentation says something about it being only useful internally in the compiler.

[jemc]

I'm putting this into final comment period.

Will address @plietar's concern in the implementation, and we can discuss it more in the PR - I don't think we want to add another user-facing gencap to accomodate #writealiased - if we need another such gencap it should be an implementation detail as he suggested.