On Sun, Jan 03, 2016 at 03:18:23PM +0100, Ludovic Courtès wrote: > Leo Famulari <l...@famulari.name> skribis: > > > On Thu, Dec 10, 2015 at 02:05:56AM -0500, Leo Famulari wrote: > >> python-urwid-1.3.0 fails to build on x86_64 during the > >> "AsyncioEventLoopTest" test with the error "KeyError: '5 is not > >> registered'". It has failed repeatedly for some time now. It fails in > >> the same way when updated to python-urwid-1.3.1. > >> > >> I looked for interesting changes made between the last successful build > >> and the first failing build. Notably, this range includes the upgrade > >> from python-3.3.5 to python-3.4.3 (08c04509). Asyncio was integrated > >> into the Python standard library in 3.4 — previously it had been an > >> external library. [0] Our python-3.4.3 package passes its 'test_asyncio' > >> test, FWIW. > >> > >> I entered the failed build tree and successfully ran the tests using the > >> python-3.4.3-7 [1] installed by Debian Stretch. That only tells us so > >> much, but I think it does indicate either a bug in our python-3.4.3, or > >> some problem with python-urwid caused by the unfamiliar Guix build > >> environment. > >> > >> Here's the hydra.gnu.org page: > >> http://hydra.gnu.org/build/861615 > > > > BTW, I filed a bug upstream: > > https://github.com/urwid/urwid/issues/164 > > > > No response yet, although I should add some more information to the bug > > report. > > > > In the meantime, what about downgrading python-urwid to 1.2.2 and > > leaving python2-urwid at 1.3.0? > > I think it’s best to avoid introducing version differences. > > What about disabling tests in python2-urwid in the meantime, with a > comment pointing to the above bug report?
I assume you mean "disabling tests in python-urwid"? In that case, how about just disabling that test by changing the failing procedure's name, as done in python-pyopenssl? > > Thanks, > Ludo’.
