KalmanFilter gives wrong results on armhf

[Rebecca Palmer]

Test log:
https://autopkgtest.ubuntu.com/packages/statsmodels/disco/armhf

(This is probably *not* a regression: upstream tests weren't run in 0.8.0-7.)

As noted in the changelog, I attempted to disable this feature on armhf as I'd rather have an explicit error than silent nonsense. However, as it was an arch-specific patch applied at build time to a part of the package that is arch:all, it made the build-time tests pass but didn't actually do anything to the as-installed package.

[Graham Inggs (ginggs)]

One option here is to ask in #ubuntu-release for this test failure to be ignored.

force-badtest statsmodels/0.8.0-9/armhf
added to
https://code.launchpad.net/~ubuntu-release/britney/hints-ubuntu

Another option is to detect the architecture and skip the tests at runtime, similar to what was done in debian/patches/deb_skip_test_ons390

[Steve Langasek (vorlon)]

the autopkgtests pass on armhf in recent releases.

[Rebecca Palmer (rebecca-palmer)]

It's _mostly_ fixed, but not completely: two variants of varmax
test_bse_oim, and test_pickle_fit_sarimax, still fail.

It's no longer a build failure as the affected tests are xfailed.

[Rebecca Palmer (rebecca-palmer)]

0.13.5+dfsg-1 removes the xfail patch, and builds successfully. (I don't know when it was actually fixed, or whether it ever was in statsmodels rather than some lower-level library.)