On Sat, 2010-08-14 at 11:07 +0200, Kevin Kofler wrote:
David Malcolm wrote:
> I think that a distinction can be made between core packages that many
> different components depend upon versus "leaf" packages that do their
> own thing and no other component relies on. I do think we should be
> conservative when updating core components in released versions of
> Fedora; with rawhide much less so. But perhaps "leaf" packages can have
> a less conservative policy.
Well, a backwards-compatible update to a core library isn't normally a
problem. Of course it doesn't make sense to push a soname bump of something
like Boost to a stable release. An update of something guaranteeing
backwards binary compatibility, e.g. Qt or KDE, on the other hand, is quite
safe to push, after adequate testing. And "leaf" also needs to be qualified,
a library that's used by only a small number of applications can be updated
to a binary-incompatible version in a grouped update with the affected
applications: for example, this has often been done to add new hardware
support to libmtp and a few other such libraries, and those updates have
been very nice for the people with the affected hardware and didn't cause
any trouble for anyone else.
IMHO, labriaries should be updated in stable release only if there's no
backwards-incompatible change or if there's a really strong reason to
update that outweights the problems caused by API/ABI change. We could
probably be much more lenient with end-user apss, but proper testing is
always a must (and yes, more testers would be *very* helpful).
Seeing your mail, you more or less agree with this. So why exactly are
you against the policy explicitly requiring either positive karma or
some minimal time in testing (setting aside some current shrotcommings
of the implementation like resetting the timer on bug update when you
just add/remove fixed bug or edit update comment)?
Martin