Phillip Cloud

Can you break down the timing of: `CREATE VIEW v AS FROM read_csv('slow_url')` and `CREATE TABLE t AS FROM read_csv('slow_url')` ? It's not clear that there's a "double fetch" (as...

I'm sympathetic to problem here, but can we perhaps try to get this performance difference fixed upstream in DuckDB so that we don't have to alter the Ibis API?

You can add `ops.Mode: "mode"` to the `SIMPLE_OPS` mapping in the PySpark compiler, then run the test suite with `pytest -m pyspark -k mode` and remove the `"pyspark"` string from...

Yep, this is an issue. There's some work here https://github.com/ibis-project/ibis/pull/10914, but I haven't had time to go through it fully yet.

Thanks for the issue. It seems like we're generating incorrect code (technically sqlglot is). It also seems like it's not possible to tell Trino not to flatten structs, which is...

> I am not sure if it helps, since the output would still be different from other backends: with trino you would have one column per struct field, while with...

The `sequence` function is also kneecapped at 10k elements so there's really no free lunch here.

There are a number of issues with making an API for this. See [here](https://github.com/ibis-project/ibis/issues/8054) for a discussion. In the meantime, for Impala, you should be able to do this: ```...

Thanks for the issue! Do you have a file or some example data that doesn't work as you'd like it to that you could upload here?

A short term workaround would be to compile the SQL and pass that to the `SparkContext.sql` method: ```python expr = ... # an ibis expression spark_sql = expr.compile() spark_df =...