I wrote that test not too long ago :) Probably it was a cut-n-paste mistake.

The azure failure was the same sporadic InvalidValue error we have seen before on Python 3.7. I think something has changed in error suppression and is comes from elsewhere than the test in which it appears.

I may kick these fixes off to 1.16.1, but this one looks pretty safe.

Thanks Allan.

I'm running a test with this on sparc, but it will take a while.

Incidentally, I ran the test suite for the maintenance/1.16.x branch on ppc64be, with and without USE_DEBUG.

Without USE_DEBUG, this is the only test that fails.

With USE_DEBUG, I am getting uint alignment issues like in #12626. I'm investigating.

I suggest running a build with -Wcast-align. I can send you my sparc log, but it's a bit large.

OK, I found a stupid typo in #12626. I'll try cast-align soon.

Seems to help.