@Fokko follow up of our discussion on Slack: https://apache-iceberg.slack.com/archives/C029EE6HQ5D/p1730134036731089?thread_ts=1730122956.980119&cid=C029EE6HQ5D

Pyarrow 17 installed numpy too, but starting from pyarrow 18 they removed the dependency.

apache/arrow#44148

the io/pyarrow.py file imports numpy, so it can happens that the import of the pyarrow io strategy fails and it falls back to the s3fs strategy, hoping the user has the package installed on his system.

@bigluck Thanks, that's a great catch. We only use the positional deletes to combine the positional deletes (when there are more positional deletes per file). It would be great to see if we can remove this and also make the numpy dependency optional. It is quite a big one and would be nice to get rid of.

opened #1259 to continue the numpy deprecation conversation.
Optionally, we can temporary bring in numpy as a project dependency before exploring its deprecation

Keep in mind that the CI passes here because we have numpy as a PySpark dependency :)