Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A justified change, with apologies for missing the subtlety in question during code review for #494. All good, please merge!

To note, I'm noticing the occasional segmentation fault on the test_read_write test (it's hard to estimate at what frequency as, all in the same environment e.g. desktop or laptop, they tend to happen often over short periods and then not at all for some reason). I did check by running it on the lama-to-dask branch too that it isn't something related to this change, and indeed I see the seg faults intermittently on that too. (I'm not sure if I was seeing those on there recently so far as I remember, but don't think I was, so perhaps a recent change has led to this? Are you seeing seg faults at all?)