Maximilian Linhoff comments

Results 344 comments of


                                            Maximilian Linhoff

Empty configuration option for client_x509_proxy, key, certificate results in cwd being used as client cert path

I think the PR is already there but stuck as it touched to many parts of the auth system at once: https://github.com/rucio/rucio/pull/6848

Replace cython code by jitted code

To add to what @kosack wrote, some caveats we found: - enable caching for all methods where it is possible (i.e add `cache=True` to all jit decorators). - Do you...

Replace cython code by jitted code

This will always be the case. numba lags behind numpy releases. If you use numba, it does not make sense to test against numpy dev releases and only sometime after...

Replace cython code by jitted code

> Thanks for the feedback ! > About this point, what are the cascades of implications in being a no-arch package ? > Easier conda distributions ? (no more compiled...

Replace cython code by jitted code

Note that using `prange` has implications also in case of `parallel=False`, as it might help the compiler to vectorize the loop. Also, fastmath can be an option, depending on precision...

display unknown option error

Would also _very_ welcome this feature!

Cannot create scoped token to upload organisation package, but newly personal general token works.

We also just ran into this in our organization. Could this be addressed? As a workaround, we had to add individual people back to the organization-owned project as "external collaborators".

BUG: IntEnum behaves differently in bitwise_or from numpy 1.26 to numpy 2+

> class DqEnum(np.uint32, Enum) That works for `Enum` instead of `IntEnum` but not for `IntFlag`

Improving performance of compiled code by using hardware instructions available on the runtime hardware

The manylinux docker images that are used to build the wheels come with gcc 10 (manylinux2014 based on CentOS 7, which is end of life) and gcc 12 (manylinux_2_28, based...

NaT Time

I don't mind, we can work here on a solution or merge the tests as xfail for future PRs. Maybe @mhvk could describe the idea why this should be easier...