Rust API: UUID support in pola.rs Series

[caniko]

Problem description

I was trying to create a pola.rs DataFrame with a column of UUIDs from the uuid crate.

Something like this:

let ser = Series::new("ID", vec![Uuid::new_v4(), Uuid::new_v4()])

Gives the error: the trait bound polars::prelude::Series: polars::prelude::NamedFrom<Vec<uuid::Uuid>, _> is not satisfied

Explanation: the trait polars::prelude::NamedFrom<Vec<uuid::Uuid>, _> is not implemented for polars::prelude::Series

[jeremychone]

This would be great.

And in the meantime, I'd love to know how we can support this ourselves. Which Polars type should we translate the uuid to? It seems there's a Polars Decimal type, which is i128, but I'm not sure if that would work.

let ser = Series::new("ID", vec![Uuid::new_v4().to_what_type(), Uuid::new_v4().to_what_type())

[uditrana]

On a similar note, it seems that Apache supports a UUID type in parquet.

Would Polaris be able to read a parquet file with a column of that data type ? Seems like Polars would need a u128

[jeremychone]

After further reading, I believe there might be some issues with using u128 as a UUID representation due to endianness concerns.

It appears that the feature supporting this is FixedSizeBinary, and I came across a bug that seems related: #9373

Regardless, I hope that we'll get UUID support in the future. It's crucial for our use case, particularly for RequestLogLine analytics.

[mjclarke94]

+1! We're also running into issues with this. We're ending up with massive files due to us having to cast UUIDs to UTF8. The fixed size binary seems like a nice workaround, but given the UUID is a builtin type in python, it would be nice to be able to preserve the functionality attached to that type.

[scur-iolus]

Would be great to have this feature in v1!

[deanm0000]

UUID type is a subset of the larger Extension Type request which has been requested before #9112

I'm a big proponent of Extension Type support.

[pol-todo]

Yes please!