Re: edos analysis for debian-installer
Hi Ralf,
On Friday 12 February 2010, Ralf Treinen wrote:
> I have just enabled daily (from now on) runs of the edos installability
> analysis of debian-installer:
> http://edos.debian.net/edos-debcheck/installer.php
Nice idea and thanks for thinking of it.
> Does this make sense? I have to admit that I do not understand the
> process that leads to that distribution and Packages file, so I just
> applied what we were already doing for the unstable, testing, and stable
> standard debian distributions.
For sid it makes some sense, but it's only of very limited value. Some
uninstallables are always going to show up, simply because a lot of udebs
are arch:all, but may not have their depends met for a specific arch as
the installation method they are used in is simply not relevant for that
arch.
Also, a lot of dependencies that do exist are not registered in the control
file. Reason for that is that the dependencies are also used to order
components in the main menu of the installer (they help determine the
order of execution of installation steps). So the picture will always be
incomplete to some extend.
OTOH, I do now see a structural problem for armel where a lot of packages
have a dependency on libgcc1 (the regular package, not a udeb as that does
not exist). It won't break anything, but it's something I have been
working to improve. (We used to have loads of dependencies on regular libs
instead of udebs because the packaging toolchain did not support udebs. We
finally got proper dependencies for libc6 for *all* udebs only very
recently.)
And the IA64 kernel udeb should probably have a provides for ext2-modules
[1]; and maybe similar for fat-modules on hppa. There's a few more minor
issues we could follow up on to clean things up a bit.
> Would it make sense to do the same
> analysis for testing/debian-installer and stable/debian-installer?
I doubt it's worth the trouble. It won't give any real indication of actual
problems and the change rate is close to zero.
We generally find any real dependency problems that do exist in sid in the
daily builds we have for all arches.
As far as I'm concerned it's up to you whether to keep generating and
publish the info or not, but I do hope nobody will be filing BRs from it
[2]. I'm not sure if I'll ever think of checking back :-P
Hope the above information gives you sufficient context. Looking through it
this time was certainly useful.
Cheers,
FJP
[1] After checking it looks as if we should add an ext2-modules udeb
instead.
[2] Maybe add some kind of disclaimer?
Reply to: