On Sat, Oct 05, 2013 at 01:01:07PM +0100, Colin Watson wrote:
> On Sat, Oct 05, 2013 at 01:46:04PM +0200, Jonas Smedegaard wrote:
> > I don't see how it can clutter package managers for our users (except on
> > unstable which arguably are developers/testers, regular users): AIUI
> > packages only enter testing when their binary dependencies can be
> > satisfied.
> Sorry, that's factually incorrect. Architecture: all packages only have
> their dependencies checked on i386 - due to essentially this class of
> problem whereby many Architecture: all packages in fact mean
> "Architecture: some".
Incidentally, I have some data on this which perhaps may convince you if
you think it's bad that uninstallable packages currently effectively
have to be allowed into testing (as I do).
http://ftp-master.debian.org/testing/testing_uninst.txt shows the
current number of uninstallable packages in testing if you only check
Architecture: all dependencies on i386 - very low across the board, as
you can see - while
http://ftp-master.debian.org/testing/testing_uninst_full.txt shows the
current number of uninstallable packages in testing if you check
Architecture: all dependencies on all architectures. You can see that
the numbers are of course much higher. But how much of this is a
Well, if I look at powerpc (the architecture I was considering first),
then binary packages whose names contain the substring "node" make up
118 out of the total count of 201 uninstallables: 59%. Now, powerpc is
better off than some other architectures, admittedly; but that still
means that just fixing all the node packages would make a big dent in
this problem, and get much closer to what I think should be a sensible
goal: check Architecture: all dependencies on all architectures, so that
testing can actually be dependency-clean rather than just
dependency-clean if you ignore certain kinds of problems.
Colin Watson [cjwat...@debian.org]