I know … a lot has happened with Numba since Fall 2021 when numba 0.54.1 was released but it has worked well for the dev work I have been doing.
Earlier this week, I upgraded my environment to 0.57 and noticed some different behavior when compiling a bunch of the jitclasses I’m using.
First, I’m getting the following warning:
:3: NumbaExperimentalFeatureWarning: Use of isinstance() detected. This is an experimental feature.
I scoured my code looking for a call to ‘isinstance’ and simply could not find one. When I compile this code in 0.54.1, I don’t get this warning. I do use a decent amount of numpy functions in nopython mode and am wondering if there is something there. Any thoughts?
Second, I’m getting the following warnings when compiling:
:3: NumbaDebugInfoWarning: Could not find source for function: <function __numba_array_expr_0x7eff78564d60 at 0x7eff7e4afa30>. Debug line information may be inaccurate.
I’m getting this numerous times … probably upwards of 40. Again, I don’t get this warning when compiling under 0.54.1.
My code still works but I have noticed that it is about 10-20% slower with 0.57 that with 0.54.1. I’m not well versed in the ways of numba debugging but would be happy to dive into figuring out what is going on if someone can point me in the right direction via a tutorial. I’m doing my experimentation within jupyter, FWIW.
Thanks in advance for any help/guidance/potential insight!
David