Graham Markall

IIUC, we include the test suite with installs of Numba, with the expectation that you can run the tests on any installed Numba version - so if the tests require...

Many thanks for the PR! > CUDA tests may need to be run, but I simply cannot download the CUDA drivers from Nvidia today for some reason, so maybe a...

gpuci run tests

(Marking as RFR because the test fail is a CI issue, not an issue with this PR)

> Some [co-authorship](https://docs.github.com/en/pull-requests/committing-changes-to-your-project/creating-and-editing-commits/creating-a-commit-with-multiple-authors) attribution here would be appreciated based on my commit to drop the 32-bit Linux support from the CI back on August 11th ([#8319 (comment)](https://github.com/numba/numba/pull/8319#discussion_r943924791)). Many thanks for...

Thanks for the issue report and nice reproducer! I can reproduce the issue. The first `readonly array` is `a`, so copying any of the subsequent values breaks the "chain" of...

Looks like `array.flatten` erroneously types its result as `readonly` if its input is `readonly`, which is unnecessary as it returns a copy. With: ```diff diff --git a/numba/core/typing/arraydecl.py b/numba/core/typing/arraydecl.py index 629263aad..edb4ca103...

> So you mean broadcast_to producing a readonly array is the expected behaviour My current understanding is that this is a bug, and it shouldn't be readonly. > but under...

Sorry, I may be being confusing - I do think `flatten` is the cause, but the error in `flatten` is in the Numba implementation of it - I don't suspect...

(For clarity, I also don't think there's anything wrong with your code - that looks like it should work)