Ian Rose

> Hmm...looks like the issue persists ( [#9161 (comment)](https://github.com/dask/dask/pull/9161#issuecomment-1146213734) ) Shucks

I took a look at the last month of CI runs on `main`, and this has happened twice in the last month: https://github.com/dask/dask/runs/7291695922?check_suite_focus=true https://github.com/dask/dask/runs/7043653348?check_suite_focus=true There was also an hdf-related segfault...

> I presume the problem case is if someone mixed say datetime.date and str within a column and different partitions could see all date, or all str, and get a...

Understood, thanks. To your earlier question: the downside of `schema=None` is that there is no checking that the schemas are at all consistent. So you can run into issues with...

It's not necessarily as simple as sampling a single row, since that row might be downstream of a lot of computations. If your `to_parquet` call is at the end of...

I think this is what [PEP 681](https://www.python.org/dev/peps/pep-0681/) is intended to address. It isn't accepted, but an experimental version is available in pyright today.

I very much agree with this.

I love this idea.

Is there anything more to be done here, or is this ready to merge?

Thanks for the review @gjoseph92 !