stuartarchibald

Thanks for the report. As discussed at the Numba meeting today, the comment here: https://github.com/numba/numba/issues/8285#issuecomment-1195709069 covers the reasons and potential fix.

This sample code produces the wrong answer due to the above. ```python from numba import njit import numpy as np @njit(['(float32, float64)', '(float64, float32)']) def add(x, y): return x +...

Closing this issue due to lack of information. If more information becomes available please add it and reopen the issue. Thanks.

This needs more thought about uniquely capturing a `FunctionType` instance. * globals * closures * anything else in `.__code__` that is captured and can change. The file path is also...

I think most of the test failures in the above are because the cache file presence check is looking in a `__pycache__` based on the imported module containing the `exec(str)`...

Just following up on this PR. Is it still needed in the proposed form or was there some conclusion over a new functionality required?

@sklam is this intended for 0.57?

/AzurePipelines run

Thanks for the report. First thing is that this shows up a bug, there's a `LoweringError` in one of the implementations due to an invalid store of an `i1*` to...

I've tried to break this feature in a number of ways, including involved things like calling a CUDA intrinsic, from an `njit` function with no specified targets, from a CUDA...