Graham Markall

Many thanks for the report and your detailed investigation! I can reproduce similar behaviour with the script in the report, getting: ``` $ ipython repro.ipy 848 µs ± 5.75 µs...

This may be due to some refcount operations when using the dict inside a jitted function, or due to some extra pointer chasing, but some more investigation is needed to...

Thanks for the report and nice reproducer - I can reproduce similar performance (with slightly different multiples): ``` $ python repro.py sum_squared 0.074 ms sum_squared_compact 4.553 ms pure numpy 3.112...

Thanks for opening an issue. I note that a similar error has been mentioned on the Discourse forum: https://numba.discourse.group/t/underlying-object-has-vanished/1836 Are you using `literal_unroll` as well?

Forgot to mention earlier - providing the full stacktrace and the output of `numba -s` will also be helpful.

> I don't think so. I don't use it explicitly. Maybe it is hidden somewhere? Thanks - I think you're probably not using it, I was only asking because the...

Potentially https://github.com/numba/numba/pull/5971 would address this (note that it got closed due to staleness) - cc @luk-f-a

> Did I do something wrong? No, the rule the stale bot uses is to close things that needed triage with no action for more than 30 days.

Just in case you saw it through a notification - I erroneously asked you for a full traceback, when there is one above, in https://github.com/numba/numba/issues/9348#issuecomment-1851400127 - I deleted the comment...

Also spotted: https://github.com/polakowo/vectorbt/issues/218