Zulip Chat Archive
Stream: std4
Topic: From/TryFrom as in Rust
Hanting Zhang (Jan 19 2023 at 22:19):
Hello!
Are there any thoughts on having stuff from Rust's std::convert
? i.e. From
, TryFrom
, Into
, etc. Is this a direction Std wants to move in? If not, what would be the alternative way of encoding conversions?
Mario Carneiro (Jan 19 2023 at 22:27):
hm that seems like more of a language design question, I would rather take it to the core team
Mario Carneiro (Jan 19 2023 at 22:30):
Right now conversions seem to be expressed either via typeclasses that fix one side like ToString
, ToExpr
, FromJson
, or via intrinsic methods like UInt32.toUInt64
Mario Carneiro (Jan 19 2023 at 22:31):
there is also the coercion system of course, which we use a lot more expansively than rust
Mario Carneiro (Jan 19 2023 at 22:31):
it doesn't seem like a good idea to have fallible conversions as coercions to option/except though
James Gallicchio (Jan 19 2023 at 22:44):
There are the various String.toXXX?
functions that go into Option, but I'm not sure how well those generalize. Are those typeclasses used in Rust in the way we use coercions?
Hanting Zhang (Jan 19 2023 at 22:57):
Ok, thanks. For #xy, the original problem is you want to write things like
class Field (F : Type _) where
Repr : Type _
asBytes : ToByteArray Repr -- ideally this can be resolved as an instance when constructing Fields
toRepr : F \to Repr
Here Field
is to denote a large prime field for cryptographic computations. Ideally, we would want to have some representation of field elements as bytes -- but we also may want to represent them with something with more information than just ByteArray
. Hence, we have the need to say instead "we have some Repr
, and Repr
can be converted to ByteArray
."
I look at this code though, and I feel like there's something off.
Hanting Zhang (Jan 19 2023 at 23:01):
The rust code would look something like
pub trait Field {
Repr : AsRef<u8> + AsMut<u8>
fn to_repr(&self) -> Self::Repr;
}
Mario Carneiro (Jan 19 2023 at 23:31):
which, interestingly, does not use {Try}{From,Into}
Mario Carneiro (Jan 19 2023 at 23:32):
I don't think we really have any equivalent of AsRef
, since that is making an assertion about the actual byte representation of the value
Hanting Zhang (Jan 19 2023 at 23:40):
Oh, yes :man_facepalming: I guess here the example is using AsRef
.
Hanting Zhang (Jan 19 2023 at 23:40):
What would be the best way to replicate the Rust code in Lean?
Mario Carneiro (Jan 19 2023 at 23:41):
what do you want to do with the bytes?
François G. Dorais (Jan 20 2023 at 00:29):
An answer to Mario's question is what is needed. However, if I'm right about your use case, Leo's "views" idea is something to look into. See:
<https://leanprover.zulipchat.com/#narrow/stream/270676-lean4/topic/.E2.9C.94.20Error.3A.20unknown.20free.20variable.20'_uniq.2E402'/near/275522380>
PS: I vaguely recall a better reference on Zulip but I couldn't find it and I might have dreamed it. In any case, the test that Leo links to in the thread above is useful to grasp the idea.
Siddharth Bhat (Jan 20 2023 at 03:02):
For reference, the idea of a view is from the paper "the view from the left" by Conor mc bride
Last updated: Dec 20 2023 at 11:08 UTC