Thijs

I opted to turn this into a setting, I don't think that regression can be solved, pandas conversion logic is too convoluted to make this work, and they don't offer...

I think we want the option internally anyways for `fetchnumpy`, I can change it to default to `True` for `fetchdf`, but I think there will be people that don't want...

Here is fine, we should just fix the `__init__.py` stub for `create_function`. What I'm confused about is this: > We also get a typing error since return_type should be a...

> If I set return type to a scalar: > > ```python > duckdb.create_function( > "my_double", my_double, parameters=INTEGER, return_type=INTEGER > ) > ``` > > I get this error: >...

Actually I understand the confusion @bnm3k, the stubs have `return_type` and `parameters` in the wrong order, my apologies.

Very related to #13543 Your sequence was not created in the temp catalog, so it cant be found/referenced by the temp table

After a quick look through the udf wrapper code, this just looks broken. There is no overload for `string_t` to perform a call to `StringVector::AddString(result, returned_string)` which makes this inherently...

I'm not talking about your comments, I'm talking about the issue at hand, the system should be making sure the strings returned by the UDF are owned by the Vector....

(without reading fully) I wonder if is relevant to mention here ? EDIT: No I don't think so

Thanks for the report, that is a known limitation, but admittedly undocumented. Dependencies don't have that level of granularity yet, the dependency is part of the table as a whole,...