Federico Caselli (CaselIT) wrote:

we are adding python side objects to ease usage of more advantage types, like Range or BitString in postgresql. links https://docs.sqlalchemy.org/en/20/dialects/postgresql.html#sqlalchemy.dialects.postgresql.Range and https://docs.sqlalchemy.org/en/21/dialects/postgresql.html#sqlalchemy.dialects.postgresql.BitString

Maybe we could provide a class for this? Or a named tuple? Something like

class SparseValue(NamedTuple):
  num_dimensions: int
  indexes: Sequence[int]
  values: array.array[Any] | np.ndarray  # assuming this is supported by the driver

This could be used also as the return type of sparse array columns?

Threat this as a suggestion, it's likely ok as is

View this in Gerrit at https://gerrit.sqlalchemy.org/c/sqlalchemy/sqlalchemy/+/6023

Federico Caselli (CaselIT) wrote:

we could even just reuse the value defined in the driver? https://github.com/oracle/python-oracledb/blob/2325059859ca9b4a517b8bfae7fa3c4f5b536ab2/src/oracledb/sparse_vector.py#L41

View this in Gerrit at https://gerrit.sqlalchemy.org/c/sqlalchemy/sqlalchemy/+/6023

@CaselIT Including the SparseVector object in sqlalchemy.dialects.oracle seems reasonable to me.

Also importing the value directly from oracledb also feels like a sensible approach—I’d go ahead with that change. The only consideration is that, with this setup, users will need to explicitly import oracledb and then create the object using oracledb.SparseVector().

We can probably alias that in the oracle dialect module.

@zzzeek would you be ok doing a try/expect at module lever to try and import that oracledb class and otherwise use a function that would error when invoked?
I know that's not how we usually do this, but it may make sense in this case?

MMMmmmm I'm a pretty strong -1 on having SQLAlchemy users need to import symbols from the oracledb DBAPI module directly. as you noted, this is exactly the thing we've been trying to move away from in the few cases where it was happening.

I would much prefer if we made our own SparseVector object in sqlalchemy.dialects.oracle , which would likely look very similar to oracledb's. But I think having our own object presented to the user is best.

is this a value object returned per-row? even then I'd be +1 marshall

yes, it's a value object returned per row.

@CaselIT Based on the points raised, I see a couple of possible approaches we could take. Before proceeding, I wanted to check with you on which approach you’d prefer I follow.

So the easiest thing it likely to just make a simple class in sqlalchemy and marshal to-from that in the type, like mike suggested.

I've uploaded the latest changes based on your suggestion. Could you please review and let me know if anything needs to be updated?