PS, going to ignore tests for now.

I just fixed uploading code coverage for the GPU tests at#2767, so hopefully when I merge inmain here it should improve the patch coverage 🤞

Hi@QuLogic and@dstansby, thanks for working on this! Could I help at all with resolving the codecov issue? I'd be super excited to see this completed to enable further trying out imagecodecs with Zarr V3 (e.g.,https://github.com/maxrjones/virtual-tiff/pull/4).

(clicking update branch in case that fixes it)

Closing and reopening to hopefully fix RTD (it has agit checkout error)

Ok better RTD checked out the code 😌

Looks like Codecov got stuck somewhere. Let's try another close/reopen

I don't think that the codecov/patch target will be hit unless the overall coverage actually increases, rather than remains constant. The issue is that two of the lines changed (L87 and L110) are not tested in main either -https://app.codecov.io/gh/zarr-developers/zarr-python/blob/main/src%2Fzarr%2Fcore%2Fbuffer%2Fgpu.py, so the PR isn't decreasing code coverage but still will not hit the 80% target for code coverage for patches (i.e., all new/modified lines of code). Do you prefer to ignore the codecov/patch target or to include unit tests for adding and creating zero length buffers?

If someone can add some tests, that would be welcome

What was concerning to me was one Codecov job says "failing after 1s". Though this may be a non-issue and just the report percentage (as you suggested)

If someone can add some tests, that would be welcome

I added test coverage for the GPU buffer in#2978. However, the test for the GPU buffer with the sharding codec is failing. Debugging that failure would require a cuda-compatable GPU but I don't have one easily accessible. I am wondering if you'd consider merging this PR prior to addressing the failures in#2978 for a few reasons:

1. The impact on the CPU buffer is fully tested and much more commonly used
2. The sharding codec failure with the GPU buffer is independent of this PR, such that these changes would not introduce new bugs (Add more unit tests for GPU buffer #2978)
3. Using unsigned bytes in the arrays in the buffer class would unblock imagecodecs usage for both V2 arrays (Buffer uses signed bytes with v2 compressors #2735) and V3 arrays (e.g.,https://github.com/maxrjones/virtual-tiff/blob/main/src/virtual_tiff/imagecodecs.py, which currently requires installing from this PR)

Thanks for your consideration, and@d-v-b for the helpful discussion and recommendation to post this suggestion at the Zarr-Python dev call today!

thanks for pushing this along@maxrjones. Based on the changes in this PR, I'm happy merging this regardless of what codecov says at the moment.