Skip to content

Potentially remove TestF77Mismatch? #20445

Closed

Description

@h-vetinari

Conda-forge has been more or less forced recently by the CI provider landscape to cross-compile packages for aarch64 & ppc64le. The tests generally still get run (using emulation through QEMU), but that means that a few special things cannot work.

As such an example, we had to skip TestF77Mismatch on the scipy-feedstock - @rgommers later removed it in scipy/scipy#15010, writing:

We no longer need this test, because g77 is not a thing anymore
(the build will fail or other things will break if an ancient g77
is still used somehow).

This test is also fragile, as seen on conda-forge for the v1.7.2 builds,
the test will fail when the tests run under QEMU with:

[snip]

That seems to be because ldd may incorrectly report something is not a
dynamic executable if the correct loader cannot be found.

Seeing that these tests doing the same between scipy & numpy, I wanted to propose whether this test could also be removed here. It would save us a further test to skip in the build recipe (which would be beneficial because test skips are easily overlooked later and prone to bitrot).

That said, if the test stays, we can work around it pretty easily.

PS. The failure showed up while testing 1.22.0rc1, though that has nothing to do with the numpy version, and is due to the intervening infra changes in conda-forge.

Activity

  1. mattip commented on Nov 24, 2021

    @mattip
    Member

    I think you are talking about this test TestF77Mismatch? Indeed if g77 is gone it makes sense to delete the test, thanks.

  2. rgommers commented on Nov 24, 2021

    @rgommers
    Member

    agreed, +1 for deleting. want to make a PR @h-vetinari?

  3. h-vetinari commented on Nov 24, 2021

    @h-vetinari
    ContributorAuthor

    agreed, +1 for deleting. want to make a PR @h-vetinari?

    Done: #20457

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    32 - InstallationProblems installing or compiling NumPy

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions