[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Release management and testing problems

On Sat, 3 Aug 2002, Brian May wrote:
> On Fri, Aug 02, 2002 at 04:20:52PM +0200, Santiago Vila wrote:
> > If we have libfoo1 and libfoo1-dev generated by source package libfoo
> > (or libfoo1) in testing, and we upload to unstable new libraries
> > libfoo2 and libfoo2-dev using libfoo2 as the source package name, what
> > exactly prevent those new packages to enter testing in ten days provided
> > they are bug-free? Nothing.
> (note: libfoo1-dev and libfoo2-dev are often called just libfoo-dev;
> this means that the build-depends does not need to be changed)
> If libfoo1 and libfoo2 conflict....
> (yes, this is not good; I am not sure if any libraries are still
> guilty of this or not).
> If some packages depend on libfoo1 in testing, and some depend on
> libfoo2, and libfoo1 conflicts with libfoo2, then this isn't very good
> for testing. I am not sure how this case is currently handled.

Packages in testing often conflict with other packages in testing.
The testing scripts ensure that a package may be installed in testing
before it enters testing (i.e their dependencies may be fulfilled) not
that it does not conflict with any other package in testing.
[ Think about mutually conflicting MTAs, for example ].

Having libfoo1 and libfoo2 to conflict at each other would be such a
serious bug that we should not even be discussing about how to "solve"
whatever problem it creates. The conflict would have to be removed first.

It may happen, of course, that libfoo1-dev and libfoo2-dev are both
called libfoo-dev. In such case:

> How would you force packages in testing to be recompiled with libfoo2?

We don't force that. We would just wait for libfoo2 and the new
libfoo-dev to enter testing, and then new packages will be linked
against libfoo2.

> Would this require a new upload? However, no source code has changed.

Old packages linked against libfoo1 would not require a new upload.
A new upload would be desirable, but not required, that's what oldlibs
is for.

But if we wanted all packages to be recompiled against libfoo2 as a
release goal, then we would submit serious bugs against old packages
still using libfoo1.

Reply to: