Update is_platform_arm() to detect 32-bit arm and other variants #44225
Conversation
|
https://app.circleci.com/pipelines/github/pandas-dev/pandas/1865/workflows/32e3f6d0-8386-40e9-9fe1-1a540ae30de5/jobs/32707 looks like 1 test fails on this |
jreback
added
ARM
- When running an arm32 ("linux32'd") chroot on an arm64 machine,
Python's platform.machine() will return "armv8l".
- In other cases, on "real" arm32, it'll return whatever uname says
(just like in the first case) which might be e.g. armv7a.
Keeping the other options ("aarch64", "arm64") given that Windows
or other kernels might choose to return different values and these
were added for a reason, but at least this fixes detection on Linux.
This allows tests like test_subtype_integer_errors to be skipped
as intended on arm.
Bug: https://bugs.gentoo.org/818964
Signed-off-by: Sam James <sam@gentoo.org>
Thanks! I wonder why, actually. That test isn't being run conditionally anyway and this PR is supposed to reduce the number of tests we run on ARM, but that doesn't mean it's doing that. Will poke if it fails again after rebasing on master. |
|
The test from before now passes, so I think it must either be flaky or was fixed on master since (thank goodness!). The other failing CI just timed out. |
|
thanks @thesamesam |
|
Thanks a lot @jreback! By the way, is there a procedure for requesting a backport for this? But no problem if we can't, it is only a small patch for us to carry downstream for a bit. (This function is only used for checking if we should skip tests) Cheers! |
|
we only backport regressions |
|
No problem at all, thanks again! |
When running an arm32 ("linux32'd") chroot on an arm64 machine,
Python's platform.machine() will return "armv8l".
In other cases, on "real" arm32, it'll return whatever uname says
(just like in the first case) which might be e.g. armv7a.
Keeping the other options ("aarch64", "arm64") given that Windows
or other kernels might choose to return different values and these
were added for a reason, but at least this fixes detection on Linux.
This allows tests like test_subtype_integer_errors to be skipped
as intended on arm.
Bug: https://bugs.gentoo.org/818964
Signed-off-by: Sam James sam@gentoo.org