Re: Cooker upgrade still broken

On Jan 23, 2012, at 5:48 PM, [email protected] wrote:

> 
>  Now it appears to be on its way to convert a 2010.2 to cooker...
> 
>  What is required is somebody to work on these kinds of updates
> frequently, and correct any issues that may arise. Manually I can
> fix it, and rpm/urpmi has enough machinery to allow it to be done
> automatically.
> 

Trying to get there …

There's a question
	What is the upgrade QA matrix?
which needs definite answers first:

Its unclear if ROSA <-> Mandriva have the same upgrade matrix or not.
(I personally think the matrix should include 2010.2, and I also think Mageia <-> Mandriva
upgrades are useful interoperability checks).

There is a blueprint (still mostly nebulous and undiscussed,
but I was running buildbots for rpm-5.3 all the way back to 2007 for several months)
for automated continuous integration for rpm here:
	https://blueprints.launchpad.net/rpm/+spec/rosa-continuous-integration
RPM's make check did test the rpmdb conversion.

URPMI "make check" should be added to buildbot's, including what URPMI
does for conversion (which is different tool than just doing the conversion).

There's also dbconvert.c, which is Yet Another converter executable
instead of brute force scriptie with Berkeley DB utils, and similar
in perl as part of URPMI.

Once rpm-5.4 is in harness on a system with a "native" /bin/rpm
of some persuasion, then the upgrade matrix can be covered for
some minimal package set including rpm (and in yr case glibc
upgrades).

But the fundamental problem with CI is the assumption that
the process Just Works. IOW, CI is good for catching regressions
and automatating a traversal of a upgrade matrix iff the tools
are in place first.

This isn't true when glibc is being reworked because rpm won't be upgraded
(because glibc will break first).

Still: trying to get there …

73 de Jeff