njlr (Jan 11 2024 at 18:15): njlr (Jan 11 2024 at 18:15):Do I need to define all of the abilities supported by a type when I define that type?
Brendan Hansknecht (Jan 11 2024 at 18:16):Yep
njlr (Jan 11 2024 at 18:17):Hmm I wonder if this will become a problem.
F# does not have traits / type-classes but at least it has interfaces to fall back on in the style of Java/C#
I assume Roc does not have interfaces due to heap allocation / general slowness.
Is there a plan in Roc to handle these situations?
Richard Feldman (Jan 11 2024 at 18:27):you don't necessarily have to write out the whole implementation
Richard Feldman (Jan 11 2024 at 18:28):like you can say implements [Eq, Hash, Inspect] and it will infer an implementation based on the inner type the opaque type is wrapping
njlr (Jan 11 2024 at 18:28):If the type comes from somewhere else (like say in a library), I guess there is no way to "extend" it?
Richard Feldman (Jan 11 2024 at 18:28):that's correct, and that's by design
Richard Feldman (Jan 11 2024 at 18:29):part of the reason for defining all of the abilities explicitly is that it's important for opaque types to be opaque by default (that is, not exposing their internal implementations) so that backwards-compatible changes can be made to them later
njlr (Jan 11 2024 at 18:29):Fair enough!
I do wonder if this will lead to lots of boilerplate situations, however.
Richard Feldman (Jan 11 2024 at 18:29):part of this is not letting others implement those abilities, because then if the original implementor later decides to add an ability, it might conflict with the userspace implementations and cause problems
Richard Feldman (Jan 11 2024 at 18:30):I suspect not, but an important consideration is that editor plugins can make it very quick to add them :big_smile:
Richard Feldman (Jan 11 2024 at 18:30):also keep in mind that this is just for opaque types and not type aliases
Richard Feldman (Jan 11 2024 at 18:30):so for example if you just want a User record and you do User : { email : Str, ... } then it gets all the abilities inferred and you don't need to write anything
Richard Feldman (Jan 11 2024 at 18:30):it's only if you want to make a new custom type that this comes up
njlr (Jan 11 2024 at 18:32):I'm thinking of a situation like...
- Library A implements operations on tree-like things and defines an ability
- Library B implements some type T I want to use
- To use T instances with Library A functions, I need to write my own type that has the abilities defined in Library A and I also need to convert T instances T into my type
Richard Feldman (Jan 11 2024 at 18:35):to me, in this scenario the author of Library A made a mistake :big_smile:
Richard Feldman (Jan 11 2024 at 18:35):I think "implementing operations on tree-like things" is not a good idea for a library
Richard Feldman (Jan 11 2024 at 18:35):it's better to implement operations on one specific tree implementation
Richard Feldman (Jan 11 2024 at 18:36):I think this is something Java encouraged that hasn't worked out well in practice (at least based on my past career as a Java developer)
Richard Feldman (Jan 11 2024 at 18:36):that is, "build lots of things in terms of interfaces and then that way you can swap out implementations" - I don't think the benefits of doing everything in that style have outweighed the drawbacks in practice, so I don't want to encourage that style in Roc
njlr (Jan 11 2024 at 18:36):I was more thinking about Haskell libraries but I can understand if this is not the direction for Roc
Richard Feldman (Jan 11 2024 at 18:36):yeah same consideration there! :big_smile:
Brendan Hansknecht (Jan 11 2024 at 18:37):You can easily get this feature by passing in a record of needed functions
njlr (Jan 11 2024 at 18:37):This was the reponse Dom Syme (F#) gave too
Brendan Hansknecht (Jan 11 2024 at 18:37):Just a more explicit form of the interface
Notification Bot (Jan 11 2024 at 18:38):njlr has marked this topic as resolved.
Last updated: Jul 06 2025 at 12:14 UTC