Move extensibility to use-site

[robinheghan]

Copying this idea from elm slack, by Changlin Li

[…] RE Roc, I agree I've been meaning to try it out. There's one small UX thing about composability that I wish Roc did differently (essentially I don't think that type synonyms themselves should express whether they are extensible, but the extensibility should be delayed to the call site, so that e.g. in hypothetical Elm syntax type MyRecord = {a : Int, b : String} could be used at the callsite as myFunctionWithExtensibleRecord :: a ++ MyRecord -> Int, which would also allow type synonyms to be nicely merged together (e.g. type MyBigRecord = MyRecord ++ MyOtherRecord) without needing to decide beforehand which type synonyms are meant to be extensible and which aren't, since that's usually practically ends up being more relevant for callsites

[boxed]

Minor quibble: I find ++ to be a weird syntax. That's the append operator right? Append isn't commutative.

I like this a lot. It's a lot nicer than going around copy-pasting {a | foo: int, bar: baz} in many places.

[blaix]

Minor quibble: I find ++ to be a weird syntax. That's the append operator right? Append isn't commutative.

What about javsacript's spread syntax?

Something like: { a : String, ...OtherRecord }

Could be used in pattern matching too: { a, ...everythingElse }, [ "blog", "post", ...path ] :)

I like this a lot.

Me too!

My gut reaction was wondering if this could be abused. In other words, is keeping it explicit per record protecting us from ourselves? But I don't think so.

[boxed]

@blaix it sounds like you are talking's about two different things.

[blaix]

Very likely! I'm easily confused. Happy to delete if I'm talking nonsense.

[robinheghan]

If I'm understanding @blaix correctly, you'd be able to extend a record literal type like { ...Record1, ...Record2 } in addition to doing it inline, which his first example is doing. Is that right, @blaix ?

For records it works fine. For Custom Types/Tags, I guess it could work like | Tag1 | Tag2 | ...TagAlias and | ...Alias1 | ...Alias2?

I do like the idea of matching this up with pattern matching, but I think that's a discussion we can have on Zulip.

[blaix]

Right, I think I focused too much on the A = B ++ C part of the example. I see now where I misunderstood the function signature part of the suggestion. I guess that would look something like greet : { Named, ... } -> String.