Richard Feldman (Jan 29 2022 at 17:27): Richard Feldman (Jan 29 2022 at 17:27):so for https://github.com/rtfeldman/roc/issues/2214 I was thinking about using the word "newtype" instead of "opaque type" - that's the term both Rust and Haskell use for this - see https://rust-unofficial.github.io/patterns/patterns/behavioural/newtype.html and https://wiki.haskell.org/Newtype - and it seems intuitive, since literally what you're doing is making a new type (even if you decide to expose everything about it, thus making it not really opaque per se)
Richard Feldman (Jan 29 2022 at 17:27):also @Folkert de Vries mentioned that "opaque" is not a word people non-native English speakers are familiar with
Richard Feldman (Jan 29 2022 at 17:27):@Chad Stearns I remember you had thoughts on this!
Anton (Jan 29 2022 at 19:00):I'm a bit confused, how would you "expose everything about it"?
Richard Feldman (Jan 29 2022 at 19:55):like if I make toStr : WrappedStr -> Str and fromStr : Str -> WrappedStr, and neither of those loses any information, then I'm not really hiding anything :big_smile:
Richard Feldman (Jan 29 2022 at 19:55):although then again, I suppose a counterargument to that would be something like StrBuffer
Richard Feldman (Jan 29 2022 at 19:56):where internally it holds a linked list of strings or something (not saying that's a good idea, but someone could do it)
Richard Feldman (Jan 29 2022 at 19:57):so maybe it's actually fair to call that opaque
Ayaz Hafiz (Jan 29 2022 at 20:11):I like newtype. "nominal" or "nameonly" may work too, but the first is likely not familiar either
jan kili (Jan 29 2022 at 20:17):Regarding opacity, yup:
toStr : CustomerId -> Str
fromStr : Str -> CustomerId
toInt : CustomerId -> I32
fromInt : Int * -> CustomerId
How's it implemented? Nobody knows :shrug:
Richard Feldman (Jan 29 2022 at 21:01):good point!
Zeljko Nesic (Jan 30 2022 at 03:02):Me gusta new type but my favorite term for this is wrapped.
CustomerId is just a wrapped Int64.
Being wrapped means it has it's own shape.
And I think Dutch also wrap things, right? @Folkert de Vries :D
Anton (Jan 30 2022 at 08:27):wrapped seems better to me as well. The name newtype sort of hides its primary purpose.
Richard Feldman (Jan 30 2022 at 15:42):a thing that's kinda weird to me about "wrapped" as the name for this is that there are other ways to wrap types in other types, e.g. I can wrap a Foo type in a record with { value : Foo } or in a single-tag union, like [ Value Foo ], but that's not the same as defining a whole new type that's incompatible with others
Richard Feldman (Jan 30 2022 at 15:43):"unique type" is another option to consider, since it's...well, unique :big_smile:
Jeroen Engels (Jan 30 2022 at 20:49):I was reading the thread and was thinking of unique as well. Or maybe "separate". I like unique more, but it should be fairly clear that it's not about the uniqueness of the value or something
jan kili (Jan 30 2022 at 21:07):bedroc
jan kili (Jan 30 2022 at 21:10):Jokes aside, though, that's the one conceptualization I have for it - something like basetype might be universally clear
jan kili (Jan 30 2022 at 21:16):Or maybe atom for succinctness?
Richard Feldman (Jan 30 2022 at 21:28):when I searched for "unique type" https://en.m.wikipedia.org/wiki/Uniqueness_type came up :laughing:
Richard Feldman (Jan 30 2022 at 21:30):I think there's definitely value in using a standard name for something as long as it's a reasonable and easy to learn name
Lucas Rosa (Jan 31 2022 at 02:57):FWIW I prefer opaque because that word more aligns with their intended usage
Chad Stearns (Feb 06 2022 at 00:28):Yeah I had a negative view about "newtype". Like, I hear ya when you say " it seems intuitive, since literally what you're doing is making a new type" @Richard Feldman , but then my criticism is that description also fits with non-newtypes as well. Like when one programs they are "literally making new types" just normally.
But I guess its intuitive because a newtype is _only_ new. Thats the _only_ thing thats different about it compared to what its a newtype _of_.
Chad Stearns (Feb 06 2022 at 00:28):Thats my view. My thoughts would also apply to "unique" type, since most types are unique as well. (Color = Red | Blue is unique compared to Direction = Up | Down).
Chad Stearns (Feb 06 2022 at 00:32):Maybe "wrapped" too, since most types "wrap" something else.. altho I kind of like "wrapped" because thats usually how I describe types like these "..DrawingId just wraps a normal Id.."
Chad Stearns (Feb 06 2022 at 00:38):More ideas?
"Contained type"
"Controlled type"
"Managed type"
Idk. I am trying to most directly reference the purpose of these types as I see it, which is to take an ordinary value and force it to be treated differently so that its usage can be controlled by an api.
Anton (Feb 06 2022 at 10:31):I like "Contained type" as well.
Zeljko Nesic (Feb 06 2022 at 16:07):a idea: Hiding type
jan kili (Feb 06 2022 at 16:48):I like that! We could save three characters by going with shy foo = bar :laughing:
Last updated: Jun 16 2026 at 16:19 UTC