GIT development branches for packagers?

[Andrew Lutomirski]

On Thu, Jan 16, 2014 at 10:15 AM, Adam Williamson <awilliam at redhat.com> wrote:
> On Wed, 2014-01-15 at 11:29 +0100, Tomas Mraz wrote:
>> On Út, 2014-01-14 at 13:13 -0800, Andrew Lutomirski wrote:
>> > On Tue, Jan 14, 2014 at 12:59 PM, Adam Williamson <awilliam at redhat.com> wrote:
>> > > On Tue, 2014-01-14 at 12:41 -0800, Andrew Lutomirski wrote:
>> > >> I have some trivial cleanups I want to make to a package a maintain.
>> > >> These cleanups are trivial enough that I don't think they're worth a
>> > >> new build.  Should I commit them to the master branch?  If so, I can
>> > >> imagine a couple of issues:
>> > >>
>> > >>  - A provenpackager could kick off a rebuild for whatever reason (e.g.
>> > >> dependency soname bump).  That will (I think) inadvertently include my
>> > >> changes.
>> > >
>> > > Yes, this will happen. Why do you think it's a problem, though? If your
>> > > changes are correct but you just don't think it's worth doing a new
>> > > build simply for them, why is it a problem if they get pulled in when
>> > > someone does another build for some *other* (presumably appropriate)
>> > > reason? It would seem like that's just what you'd want to happen.
>> >
>> > Depends how well I've tested.  I'd like to imagine that I never commit
>> > anything broken anywhere, but this is empirically incorrect -- I break
>> > development branches on a semi-regular basis.  I guess I'll just have
>> > to be more cautious w/ Fedora :)
>> >
>> > >
>> > >>  - I need to think about whether to add a changelog entry or not.  If
>> > >> not, those changes might be included silently.  If yes, then I need to
>> > >> think about what to do about the revision number.
>> > >
>> > > One thing I've seen done is to add the line that actually describes the
>> > > change, above the last date/builder/NEVR line, *without* adding a new
>> > > line identifying the new build, date and builder. That way when someone
>> > > comes along and does a new build, they ought to see what should happen -
>> > > they should roll your partial entry into the entry they add for the
>> > > build.
>> >
>> > That would work.
>>
>> I'd recommend rather the approach suggested by Kevin. Bump the release
>> and include a regular changelog entry. Just do not build. There is no
>> rule that all changeloged entries must be really built.
>
> I have found this kind of phantom release a bit annoying in some really
> esoteric situations - when the changelog indicates that there was, say,
> a 1.2-6 build, but there never was, only 1.2-5 and 1.2-7 - but most of
> the time it's not going to be a problem, yeah.

I can imagine this annoying anyone who does a mass rebuilt or a
similar set of rebuilds that aren't intended (by the one doing the
rebuilds) to change anything other than dependencies and, say, the
compiler version.  Sure, this *shouldn't* cause a problem if everyone
is appropriately careful, but I'm hesitant to trust things that
transparently deploy code when no has has explicitly asked for a
change to be deployed.

--Andy