Open musicinmybrain opened 10 months ago
The “flaky”
x86_64
failures are inTestSLSQP::test_minimize_unbounded_approximated
; I’ll report more detail on them in a separate issue.
A quick note is that we have never officially supported these architectures, but if these failures are only tied to the coblya method then it would more likely be a scipy issue. These new tests are associated with the new support for passing any valid scipy minimize method to our function to make it compatible with the scipy api.
One more note: we exclude this test from the source release with https://github.com/mechmotum/cyipopt/blob/master/MANIFEST.in#L10. So it isn't clear why you'd be running this test when doing packaging downstream. The PyPi source release should be used for all downstream packaging.
Sorry, that prior comment doesn't apply to this issue, but would apply to #238
One more note: we exclude this test from the source release with https://github.com/mechmotum/cyipopt/blob/master/MANIFEST.in#L10. So it isn't clear why you'd be running this test when doing packaging downstream. The PyPi source release should be used for all downstream packaging.
Good point. It looks like the only reason I have been preferring the GitHub archive has been to get the examples/
directory. Currently, I package these along with the documentation, and I run them along with the tests for additional confidence. Neither of these is strictly necessary. On the other hand, would you consider adding the examples/
directory to the PyPI sdist?
On the other hand, would you consider adding the examples/ directory to the PyPI sdist?
Yes, they are very small so we could do this.
While working on updating the
python-cyipopt
package in Fedora Linux to 1.3.0, I observed consistent failures of oftest_minimize_ipopt_jac_with_scipy_methods[cobyla]
onppc64le
ands390x
.In five scratch-builds, I observed:
x86_64
: 2/5 builds failedaarch64
: 0/5 builds failedppc64le
: 5/5 builds faileds390x
: 5/5 builds failedThe
ppc64le
ands390x
failures all looked like this, with identical numeric values:The “flaky”
x86_64
failures are inTestSLSQP::test_minimize_unbounded_approximated
; I’ll report more detail on them in a separate issue.The
scipy
package is at version 1.11.1. I’m happy to run any tests or add any other details that might be helpful.