Ritchie Vink

It gets even worse. The granularity may differ for every element. In polars, the output of an operation should always depend on the type, never on the data.

> Hence my desire for the ability to grab the column dtype in an expression :D Yep, we can add that under the `meta` namespace.

This depends on the context the expression runs in. So you would need to pass a schema and a context. I am not sure this is what you wanted?

Added in #23167

This needs a rebase.

> Otherwise I can simplify this PR by removing the requirements changes and the run-pandas-gpu target (I'll just leave the *-no-env targets). I think I'd prefer that. Can you also...

Have you tried setting the `retry` to non-zero?

Is there a repro on a public dataset that I could use to trigger this? I want to fix this.

What is your query? Eager/ streaming/ projection pushdown. Please, help me a bit. As the issue description states, try to make a code snippet that reproduces it.

Thanks. I have some vacation days, but I hope to get to this.