stuartarchibald

Buildfarm ID: `numba_smoketest_cpu_yaml_81`.

> Re: reproducing the fail may be possible on an x86 box with many cores. > > I have a multi-core x86_64 system. The deprecation notice is appearing on Ubuntu...

IIRC 0.54 was the first version using LLVM 11 and there were some things to do with missed vectorization that Numba encountered when the upgrade was undertaken (which were fixed,...

Out of interest, why not use Numba's `ctypes` support, it's more idiomatic python?

The issue with binding to C libraries as demonstrated in this example is that if the `@njit` decorator is removed, the code stops working correctly. Whereas the use of `ctypes`...

@eric-wieser review and discuss adding to the extension API as it seems like a useful thing? Hiatus is due to https://groups.google.com/a/continuum.io/forum/#!topic/numba-users/CY7oXibUwXA, thanks for your patience.

I think that this PR acts as the motivation for converting the extending `Interval` example into an actual unit test so as to make sure the code runs, and then...

Thanks for the report. Unfortunately I cannot reproduce this on linux. I'm a little puzzled as to a) the use case, because this isn't how `literal_unroll` is intended to be...

Presume this is the function that's causing the issue? https://github.com/aesara-devs/aesara/blob/d09e222b06c0bc45d25c77d2ededbf25fe980c71/aesara/link/numba/dispatch/tensor_basic.py#L199-L212 if so I think the way it's written is actually hitting two Numba bugs and it's silently not doing what...

> We simplified it from a function where the literal_unroll actually did something useful, this was just the shortest example we could find that still resulted in a segfault. And...