Closed paleolimbot closed 2 months ago
Not clear if this started with this PR, but on main the merged commit from this PR is failing in the wheel building: for Python 3.12, running the tests is segfaulting (https://github.com/apache/arrow-nanoarrow/actions/runs/8689330033/job/23826851854)
I think I have seen that one time before but I was confused because the backtrace seems to suggest that it happens while collecting the tests? (Or is that backtrace generally known to be unreliable?)
I experimented with a few different methods here...I think this one has a good balance of speed and not messing with the global precision.
This seems to be vaguely on par with pyarrow convert: