Is a generic newtype possible?
atezet opened this issue · comments
Is it possible to create a generic newtype with nutype
? When I do, let's say:
use nutype::nutype;
#[nutype(validate(predicate = |v| !v.is_empty()))]
struct NonEmptyVec<T>(Vec<T>);
I get:
--> src/main.rs:4:27
|
4 | struct NonEmptyVec<T>(Vec<T>);
| ^ not found in this scope
|
help: you might be missing a type parameter
|
4 | struct NonEmptyVec<T><T>(Vec<T>);
| +++
Hi @atezet
It's not possible at the moment.
But I have this in mind and want to make it possible in the future.
I already like using the library as-is a lot, but definitely looking forward to that! Thanks
@atezet Btw, if you need only non empty vector, I recommend taking a look at nonempty
: https://github.com/cloudhead/nonempty
Thanks for the suggestion! We tried that at first, and iirc it worked for the cases where we needed a vector, but we also need IndexMap
and some others. Unfortunately nonempty
only supports Vec
, even though its name would suggest otherwise.
I also suggested just using nutype
and fixing the type of the stored values, but my colleague went ahead and implemented a newtype wrapping a generic Iterator
for now.
Would love to see this to be possible with nutype
or nonempty
.
Hey there, I just briefly checked out the interface (no time to check the code), and my example works as expected! I think your observations regarding support for trait bounds and auto derives are very correct. It would be cool if something like (there are better predicates to do this):
#[nutype(validate(predicate = |i| i.into_iter().count() != 0))]
struct NonEmpty<T: IntoIterator>(T);
would work as well, but that should be covered by #142
@atezet Hi, thanks for the feedback!
@atezet I just finished the work with generics and published 0.4.3-beta.1
. Let me know if that works for you.
I am planning to release 0.4.3
in a week.
@atezet So in 0.4.3 it's possible to use generics and to set type bounds.
However, to properly implement your use case it would require setting type bounds with where
clause (this one is not supported yet):
#[nutype(
validate(predicate = |c| c.into_iter().next().is_some()),
)]
struct NonEmpty<C>(C)
where
for<'a> &'a C: IntoIterator;
The problem here is that we need to set the &C: IntoIterator
boundary, but within <C>
it's only possible to set the boundary on C
, (not &C
).
As a workaround, it's possible to clone. But of course it's not optimal from the performance perspective:
#[nutype(
validate(predicate = |c| c.clone().into_iter().next().is_some())
)]
struct NonEmpty<C: Clone + IntoIterator>(C);
@atezet I've found a good solution for you.
We can introduce a helping IsEmpty
trait and with that workaround the limitation without cloning:
use nutype::nutype;
trait IsEmpty {
fn is_empty(&self) -> bool;
}
impl<T> IsEmpty for T
where
for<'a> &'a T: IntoIterator,
{
fn is_empty(&self) -> bool {
self.into_iter().next().is_none()
}
}
#[nutype(
validate(predicate = |c| !c.is_empty()),
)]
struct NonEmpty<C: IsEmpty>(C);
fn main() {
let number_vector = NonEmpty::try_new(vec![1, 2, 3]).unwrap();
let chars: std::collections::BTreeSet<char> = "abc".chars().collect();
let char_set = NonEmpty::try_new(chars).unwrap();
}
Hi @greyblake. Sorry for not replying earlier. We're in a final sprint before the code is audited, so I've some quite busy times and will look at using your solution in the next phase. Thanks for putting in the work, I think it looks quite clean already! I really like what nutype
can do for typesafe type driven development with minimal boilerplate
@atezet No worries! Good luck with the sprint!
Looking forward for any feedback from you, if you'll have any.