On Wed, Feb 02, 2011 at 02:51:15PM +0100, Andreas Tille wrote: > A second idea would be to push the package test issue currently > discussed on debian-devel foreward. I specifically consider this a > science issue because I could perfectly imagine a process where we > could feed test data into a just compiled programm and compare the > result with some expectations.
While I was quite a bit dubious about this in Travemünde, I (re-)read parts of the -devel thread now and changed my mind. Having a way to install a scientific blend and then fire off a "verify everything (important) actually produces correct results" on the installed packages seems to be an important thing we could offer the scientific developer/user community. Possibly have "minimal test"/"full test" options, so the cluster admin can either check things work in general (e.g. on every newly installed node), or run a full check once before replicating the cluster nodes. We might even want to think about tying this into a more elaborate re-building architecture. E.g. retrieving all available Fortran/C/C++ compilers and basic computional science library packages via debtags and letting the user/cluster admin decide which they want, install those, then recompile their applications against those and finally verify they work through autopkgtest. For the case of scientific packages, it is important to semi-intelligently parse the output and catch unimportant rounding-differences in the nth decimal place without declaring everything broken. I guess there are a couple of dead sourceforge projects for this. > A third issue which comes to mind is the request to use *defined* > versions of programms and not just a version which is by chance packaged > in any Debian release, for instance by using packages from > snapshots.debian.org. Can you elaborate on this? I am not sure what you mean here. It doesn't sound like it would take somebody 3+ months of full-time work to do this in any case, though. Michael -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected] Archive: http://lists.debian.org/[email protected]
