[Mageia-dev] Proposal of a backporting process

[Samuel Verschelde]

Le vendredi 24 juin 2011 21:39:51, Ahmad Samir a écrit :
> On 24 June 2011 02:09, Michael Scherer <misc at zarb.org> wrote:
> > Hi,
> > 
> > as said in the thread of firefox 5, and in the meeting of packager
> > sooner this week, this is the first mail about backports ( on 3 ).
> > 
> > So here is the proposal of a process, based on the feedback of people,
> > and the idea of some packagers ( mainly stormi ).
> > 
> > 
> > - Someone request a backport ( by bugzilla, by madb, by a email, by
> > taking a packager family in hostage, whatever ). I would prefer use
> > bugzilla but this may not be very user friendly, or too heavy.
> 
> How would the packager get notified of backports requests via madb?

There are several options :
- option 1 : maintainers prefer to have all backports requests in bugzilla. 
Madb will then create backports requests via XML-RPC, with the original 
reporter in CC maybe, and regularly watch bug report status. This will be 
extra work on madb's side and force those users (who maybe don't know how to 
use bugzilla) to use 1 tool for the request and a different tool for testing 
reports, but why not.
- option 2 : maintainers are OK to use bugzilla for bugs and madb for package 
requests => madb will query the maintainers database and notify the 
maintainer(s) by mail. It could, like bugzilla, send notifications to a ML too, 
and provide a simple yet sufficient tracking system (status, comments).

> 
> Would you elaborate on how bugzilla is heavy for a backports request?

Heavy I don't know, but I think that we can give users a better tool to 
request backports, see what backports already have been requested, etc.

> 
> > - a packager decide to do it. Based on the policy ( outlined in another
> > mail ), and maybe seeing with the maintainer first about that for non
> > trivial applications, the backport can be done, or not. The criterias
> > for being backported or not are not important to the process, just
> > assume that they exist for now ( and look at next mail ). So based on
> > criteria, someone say "it can be backported, so I do it".
> 
> [...]
> 
> > - I am not sure on this part, but basically, we have 2 choices :
> >  - the packager take the cauldron package and push to backport testing
> >  - the packager move the cauldron package in svn to backport, and there
> > send it to backport testing.
> > 
> > Proposal 1 mean less work duplication, but proposal 2 let us do more
> > customization.
> 
> Option 1 doesn't only mean not duplicating work, but also that the the
> spec in backports svn isn't ever out-dated; the only reason I see a
> package being in stable distro SVN is if it's in /release|updates, not
> backports...

I'm not sure I understand your point. What do you mean with out-dated specs in 
backports ? 
I favor option 2 (with all needed useful shortcuts in mgarepo and BS to make 
it simple for packagers) because it allows to cope with the following 
situation :
- foo is in version 1.2.2 in release|updates
- foo is in version 2.0alpha in cauldron, full of bugs but hopefully ready for 
the next stable release
- the latest release in the 1.x branch, 1.3.0, brings many features requested 
by some users, we want to provide it as a backport : with option 1 we can't, 
with option 2 we can. 

or : 
- foo is in version 1.2.2 in release|updates
- foo is in version 2.0alpha in cauldron, full of bugs but hopefully ready for 
the next stable release
- we had backported version 1.2.6 before switching to 2.0alpha in cauldron
- the backported version 1.2.6 has a big bug we hadn't spotted during tests 
and we want to fix in the backport : with option 1 we can't, with option 2 we 
can.

So, for me, this is definitely option 2.

However, I think it must be made a painless as possible to packagers :
- in the common case, allow to submit directly from cauldron to the backports 
media, but let the BS detect that and automatically do the SVN copy part.
- for the situations I described above, work with the backport branch 
similarly as we work on updates (technically speaking : SVN, BS...). 


> 
> > if the package doesn't build, the packager fix ( or drop the idea if
> > this requires too much work )
> > 
> > - the packager send requesting feedback about the backport from the
> > people who requested it, and test it as well.
> 
> Probably off-topic, but how will that work with madb? i.e. how will
> the maintainer get the feedback?

I partially answered above : either via bugzilla, or via a simple tracking 
system included in madb for that need. It will depend on the chosen process, 
we'll try to adapt the tool to the situation.


Best regards

Samuel Verschelde