Graham Markall

@sklam Many thanks! Can this go to RTM?

At first I was surprised this is expected to work, but then I found that (presumably) it's implemented with `__setattr__` and even has documented examples: https://numpy.org/doc/2.2/reference/generated/numpy.ndarray.flat.html Marking this as a...

Thanks for the report! I can reproduce this issue. A smaller test also seems to reproduce the ``` NotImplementedError: ('use load_from_data_pointer() instead' ``` error message: ```python from numba import njit,...

> So you mean after a fix, we could meet use-after-free bugs? Yes - that would probably explain the random value of the record when it is boxed. > If...

From discussion in the dev meeting: probably a reasonable and lower effort way to address this issue for now is to fix Numba such that when this use of a...

The Python version of the LLVM Kaleidoscope tutorial is a bit old now (I don't think it's maintained) but it might be a good source of examples: https://github.com/eliben/pykaleidoscope/

cc @yxdragon for updates as #10004 reports a similar issue.

When I run the reproducer (with the class name changed to `release_me`, as it contained a space above) I get: ``` $ python repro_orig.py release release before this ``` Do...

Thanks for the clarification. A couple of observations... Adding a `tp_clear` method that clears `defargs` seems insufficient. I tried: ```diff diff --git a/numba/_dispatcher.cpp b/numba/_dispatcher.cpp index a0ed6b830..206869b54 100644 --- a/numba/_dispatcher.cpp +++...

My gut feeling is that this is likely to be related to the Dispatcher holding a reference to the namespace in which `@jit` was executed, which may not be in...