launchpad-dev team mailing list archive

[Danilo Šegan <danilo@xxxxxxxxxxxxx>]

У уто, 04. 08 2009. у 16:53 -0300, Guilherme Salgado пише:

> > (2) Am I right that we would prefer a Storm branch that only had  
> > bugfixes, not new features or changes that required Launchpad changes  
> > (a hypothetical 0.14.1)?  If so, how important is that, as opposed to  
> > the Storm team releasing a 0.15 early?
> 
> Since we're already running fine against trunk, I think it's fine to
> keep that until they release a 0.15. I'd be against a 0.14.1 because
> 0.14 is already 6 months old.

I don't think we had staging updated since "trunk" was egg-ized and
included as LP dep, so I doubt we are "running fine" with trunk.  And
limited testing staging gives in no way equates to "running fine".

Also, we are not yet running fine with 0.14 + two revisions stub and you
cherrypicked either[1], and I wouldn't want to jump up on trunk right
away, before we can give it serious testing.

I am not arguing against doing the upgrade early, but we've got to
figure out existing stuff first. Fixing 2.2.7 production branch is going
to be very hard if we switch to new Storm version in development: 2.2.7
has not seen enough testing itself with 0.14 upgrade (note that 2.2.7
was the first one to roll out with 0.14 upgrade, and the next one will
be 3.0).  Let's be conservative and go for 0.14.1, because that will
give us due time to make sure 2.2.7 works fine, and nail down all the
small bugs before 3.0.

[1] Due to somehow incompatible change in SQLObjectResultSet.__nonzero__
(I can see how the code is different, but I don't see where exactly is
it incompatible in return values), we've got bug #408845 which causes
approximately 15% of our PO imports to fail.  I am still trying to find
a proper failing test case, but another upgrade in such a short
timeframe scares the shit out of me (is there anything else weird that
has broken with 0.14 like the above bug? who knows?). Yes, we have a
bunch of workarounds in place for things like missing .find() on
resultset, etc.

Cheers,
Danilo