BINARY column type support about databricks-sql-python HOT 1 OPEN

Following up on this now that we've implemented SQLAlchemy 2 on main. We have a much better understanding of how SQLAlchemy handles types now and there are some limitations of Databricks itself that make this unfeasible for the moment. Namely, databricks-sql-connector doesn't have a way to bind parameter values of BINARY type. sqla needs every data type to bindable for parameterisation. Since DBR doesn't support binding BINARY values, our dialect doesn't either. This same limitation applies to ARRAY, MAP, and STRUCT types.

That said, where there is a will there is a way. While DBR doesn't support parameterisation of these types, you can write a string SQL query that includes them. DBR understands hexadecimal strings and can translate them to BINARY types during ingestion. And pysql will deliver them as io.Buffer's during egress.

So if this is necessary for your use-case, you can implement a workaround by using your own custom SQLAlchemy type class (which extends TypeDecorator). You'll need to implement the process_bind_param, process_literal_param, and process_result_value methods for this custom type in such a way that the ...param methods accept an io.Buffer and convert it to a hexadecimal string and render it inline with the query text.

We've marked this as a FutureFeature to be added to this dialect here. If DBR adds support for binding these parameter types then it would be very easy to implement as part of our repository. If not, then inlining these values could work. Although would not be strictly safe from SQL injection, since an arbitrary value would be rendered in the resulting SQL string.

Although this may not be as risky, since hexadecimal strings cannot include any of the dangerous SQL commands: the hexadecimal charset doesn't include ; or enough consonants to spell DROP, UPDATE, CREATE etc.

from databricks-sql-python.