JSKenyon
JSKenyon
> After I disabled zoom, it works fine on my laptop (two fingers touch). This seems to have worked for me too.
Thanks @gjoseph92 - those are some really interesting results! I definitely think that the improvements in memory footprint are compelling. That said, I would be really interested to see these...
Just an upvote from me - I am also playing around with nested parallelism using dask and numba and run into a similar issue when attempting to cache parallel functions....
I tried out the fix by editing my local `caching.py`. I can confirm that it resolves the crash but it also seems to effectively prevent caching. I see the cost...
I put a print statement inside the loading code here: https://github.com/stuartarchibald/numba/blob/af64272017f1aeaba36c0c177b736c609ea2908b/numba/caching.py#L495-L507 and found that `data_name` is always `None` once the fix is applied. Is this the expected behaviour @stuartarchibald?
I have finally managed to generate a "reliable" reproducer that I believe might also serve as a template for a unit test. The following dies reliably when you run it...
Oh, and thanks @crusaderky for your example. It helped me immensely.
Incidentally, the same problem occurs with a `ThreadPoolExecutor`. ## Traceback `[1] 24245 segmentation fault (core dumped) python parralel_numba_again.py`
Just wanting to check in here - this is still causing me some trouble. Is there workaround?
Just checking in - it has been a while but it seems as though this is still a problem. @stuartarchibald, @esc does this look like something that can be solved?...