Graham Markall

`test_target_extension.TestTargetHierarchySelection.test_overload_allocation` fails when run in isolation

4

I suspect there is some relationship with #9954 - perhaps why things appeared to work without that change. Running `TestTargetHierarchySelection` as a whole: ``` $ python -m numba.runtests numba.tests.test_target_extension.TestTargetHierarchySelection ............

[WIP] LLVM 20 llvmdev recipe

3

Using this PR for development / notes. ## Build / installation I built for linux-64 with the GHA builder (https://github.com/numba/llvmlite/actions/runs/15325329108) Installation requires: ``` + liblzma-devel 5.8.1 hb9d3cd8_1 conda-forge 442kB +...

`_call_function_ex_replace_args_large()` expects that the `vararg_stmt` is an `ir.Assign` whose value is an `ir.Var`. This is not the case when the arguments come from a generator expression, and it is an...

#8783, Namedtuple fixes with merged `main`

1

I can't push to #8783, so this PR is testing that PR with `main` merged.

Our bool type isn't an integer (previous discussions have concluded this is an oversight) so it has no bitwidth. Adding a bitwidth of 8 (the number of bits used to...

[WIP] Use the NVIDIA CUDA target by default

2

This PR switches to using the NVIDIA CUDA target by default, incorporating the logic that presently exists in numba-cuda for shadowing the built-in `numba.cuda` module (https://github.com/NVIDIA/numba-cuda/blob/main/site-packages/_numba_cuda_redirector.py). If it is not...