Allow the construction of self-referential Refs

[garyb]

No description provided.

[natefaubion]

I don't really like that new requires an unnecessary closure allocation with const on top of the ref allocation and then Effect thunk allocation.

[garyb]

Yeah, I guess the whole thing is FFI already so one more function won't do any harm.

[JordanMartinez]

What's the status of this PR?

[garyb]

Is "forgotten entirely about" a valid status? 😄

It sounds like the review concern was addressed, but there's no explicit approval.

[hdgarrood]

This looks good to me, although I think I’d prefer the function was called newSelf too.

[JordanMartinez]

newSelf

Same. @garyb How quickly would you be able to update this PR? Or should someone else submit a PR to your branch to get the name changed?

[garyb]

Done! I wonder if there's still a better name though, as newSelf is a bit weird. Something like knot or newKnot perhaps.

[JordanMartinez]

I wonder if there's still a better name though, as newSelf is a bit weird. Something like knot or newKnot perhaps.

Why? What about knot or newKnot makes you think it's better?

[JordanMartinez]

Looks like CI is failing because the tests weren't updated:

* Building project in /home/travis/build/purescript/purescript-refs

[1/1 UnknownName] test/Main.purs:39:10

  39    ref <- Ref.self \ref -> RefBox { ref, value: 0 }

               ^^^^^^^^

  

  Unknown value Ref.self

[hdgarrood]

People use the phrase “tying the knot” to mean constructing a structure which references itself: https://wiki.haskell.org/Tying_the_Knot

[JordanMartinez]

Let's assume you come to a language and codebase you've never seen before and you only see this: ref.newSelf and later ref.newKnot.
By just looking at those isolated instances, what immediately comes to mind?

IMHO, I think the newSelf makes me think more of what this PR is doing. The newKnot would require me to know about the "Tying the Knot" phrase and I would need to connect that concept with this usage rather than assume that knot is some domain concept and I'm creating a new one.

[natefaubion]

newWithSelf?

[JordanMartinez]

Sure.

[garyb]

Renamed again, and updated the tests this time!

@JordanMartinez re: people coming to the library looking for this function, realistically I'm not sure how often that will happen. I needed to do this in Halogen, and used an unsafeCoerce hack to make it work, but outside of that case I can't think of another occasion I've needed it. In Halogen it's in a situation where something is stored in an existential form, but internally it needs to be able to do things with its full set of type parameters, so... pretty niche.

Maybe there are some other uses cases for it though, I dunno. I've definitely used self referential AVars more often. But they don't have this problem since they can be created as empty-but-not-Maybe due to their nature, unlike Ref.

I think providing a way to do it in the library rather than having to do an ad-hoc unsafeCoerce hack for the rare cases it's needed is still worthwhile, even if it is rare, since it's not possible to misuse this the way I theoretically could in Halogen.