On Sun, 2017-02-19 at 11:23 -0700, stan wrote: > > I downgraded to the last version of dnf that uses the python 3.5 ABI, > and dnf works again, with occasional segfaults. But it won't update > anything because there are too many conflicts with going to python > 3.6. How is this not python 4? If the ABI changes so much between 3.5 > and 3.6 that they are backward incompatible, then it isn't python3 > any more. What are the python developers thinking? It's nothing new, this is how Python has always worked, all the way back to 2.x (not sure about 1.x). Minor release versions always bump the ABI. > I ran a dnf --best update and there are 317 errors, and dnf skips > everything. :-) > > Most of the errors are related to the 3.6 ABI update, and packages that > aren't upgraded yet. They propagate back up the libsolv chain, and > prevent everything else from updating. > > I suppose I'll be forced to --allowerasing if I want to do a dnf > update. Or force install the dnf dependencies of the new 3.6 ABI so I > can then run dnf update and have it update everything else. I'd just do a --best --allowerasing and see what it's actually blocked on. There aren't actually many things left that aren't rebuilt for Python 3.6 at this point, and most of the ones that are left are pretty obscure. You can see the list in every 'rawhide compose report' mail, as it lists all packages whose dependencies cannot be resolved. Most of the issues in the current Rawhide compose are actually with boost. -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net _______________________________________________ test mailing list -- test@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to test-leave@xxxxxxxxxxxxxxxxxxxxxxx