On 2018/09/25 01:53, Daniel Dickman wrote: > On Tue, Sep 25, 2018 at 1:10 AM, Björn Ketelaars > <bjorn.ketela...@hydroxide.nl> wrote: > > However, I'm not confident that there is enough time to test the above > > before the 6.4 freeze. Therefore I would like to propose to revert the > > recent update of py-zmq. Diff enclosed.
OK sthen@ > let me ponder this for a few days before we revert py-zmq. I've done > most of the same updates locally on my end so I'll compare my work > with yours. We can always put it back in later if we're confident in the other updates (there's not a massive chain of dependencies). I don't see a downside, and at this point leaving it broken for a few days risks it being broken for release. (half-serious suggestion, maybe we should set all ports to EPOCH=0 to avoid the barrier of people not wanting to add EPOCH where it didn't exist before..) > Question in the meantime. Does a slightly older version of matplotlib > fix the problem for you? For example, 2.0.2? That version can avoid > some of the risk of doing such a big update. For example, it doesn't > depend on kiwisolver and subprocess32 is allegedly optional. I think > if that version fixes the problem it would really minimize the risk of > doing an update. Adding new dependencies are fairly low-risk (the only real problem is if some existing port starts picking them up and either breaking as a result, or switching on some code which makes it a new dependency), updates are riskier.
