The system-wide change for F26 in regards to upgrading Python to 3.6 is accepted for Python 3.6 release candidate 1, which will be released at 2016-12-05 according to PEP 494.
Currently the third beta is released, which I already built in copr. I tested the package and found no issues with this build.
However, there are some changes that will require more extensive early testing. In particular, on Python-SIG we agreed to try changing the default behavior of os.urandom() to be more more suitable for Fedora than the upstream. For details, see PEP 524 for the upstream specification and Nick Coghlan's proposal for Fedora.
Also since a lot of packages, use cases, etc. rely on Python, it would be great to gather feedback and address any issues early.
Thus I would like to deploy either the third or the fourth beta of Python 3.6 in Rawhide using a side-tag.
I'm a big fan of doing integration tasks like this early so we have plenty of recovery time.
So, for my part, the sooner this lands in Rawhide, the better.
Agreed.
While I like the idea there is a lot of logistical issues involved with long term maintenance of side tags. anytime someone does a build in rawhide you need to commit and do another in the side tag in order to keep up. side tags are really only designed for short term use. If you are willing to commit to the work that comes up then I am okay with it, I just wanted to make you aware of the extra work it would entail.
I am aware of the work that this will require (for which I will devote most of my time), so I was also wondering for how long a side tag can be active. Also I don't have proven packager access in order to rebuild the important packages so that might slow things down a bit (although maybe I should apply for that).
I'd encourage you to apply for pp. Doing such an upgrade without that just doesn't make sense (it involves many more people and is then much more work).
@cstratak is now a proven packager as approved in #1643
Also just some notes to add.
The 4th beta will be released at 2016-11-21.
My preliminary plan, taken that date into consideration, is to deploy the 4th beta in rawhide, build/fix the most important packages manually using a side tag, and then file a ticket for rel-eng to do a targeted mass rebuild for all the python packages, cleaning up at the same time whatever issues might appear.
After that, deploy Python 3.6.0 RC1 and the normal mass rebuild should take care of having python packages built against 3.6.0 RC1, as there shouldn't (hopefully) be too many changes between the 4th beta and the rc1.
The above plan looks good to me, +1 to deploy latest upstream release of Python 3.6 in Fedora rawhide.
Python 3.6.0 beta 4 is now available.
As discussed at the latest FESCo meeting, the next one will be held at 2016-12-02 and the voting should be continued at the tickets.
Currently I plan to start testing in copr and do some rebuilding of important packages there for some days and then deploy python 3.6 in a side tag.
As far as I can deduce, this issue is +3 so far. Could a decision be reached through the ticket?
+1, good plan. Makes sense to do the rebuild in a side tag and then get it to rawhide as soon as possible.
I'll just note that historically python maintainers have done a rebuild of all python packages in a side tag when updating python to a new major version. I think it would probably make sense to do that this time around too, now that you have provenpackager access.
The stated contingency mechanism covers only the case when python 3.6 is not built in koji yet, just in a side tag and there's no deadline specified. Hence, I'm -1 unless you're going to do the rawhide koji builds AFTER branching f26, which doesn't make sense because you'd miss the mass rebuild. Please clarify and fill the missing information on the wiki page.
Edited the wiki page, it should be clearer now, could you review it?
Looks fine now, +1.
Also I would like to mention that I am currently blocked for that due to openssl issues [0]. Investigating it, hopefully it will be resolved in a reasonable time frame.
[0] http://bugs.python.org/issue28689
Unfortunately due to the latest release of openssl, which was released and then deployed in rawhide, that broke python tests (among other things), the narrow time frame between the beta and rc1 was lost. The correct openssl fix is backported now in rawhide, however it doesn't make sense now to deploy the beta, as the release candidate will be available on the fifth of December and the only critical issues can be fixed now upstream.
Thus this ticket is not valid anymore. The plan will move on as described at the system wide change.
@cstratak changed the status to Closed
Closed
What's the status of rebuilds of dependent packages? Copr repo cstratak/python3.6 seems to be empty. Are any packages failing to build? Is there a tracker for problems that need solving?
The latest openssl build synced today so as of there are no packages in copr. From today since the middle of next week I should have rebuilt in copr the most important ones.
Login to comment on this ticket.