Re: List of packages which should probably be Architecture: all

[Colin Watson <cjwatson@debian.org>]

On Wed, Jan 02, 2008 at 03:06:21PM -0500, Hubert Chathi wrote:
> On Wed, 02 Jan 2008 13:17:24 -0600, Raphael Geissert <atomo64+debian@gmail.com> said:
> > Hello all, I've written a script which tries to detect packages which
> > should be architecture all based on the fact that they don't contain a
> > Depends field.  This is usually bug either because of a missing
> > Depends or because the package should be Architecture: all.
> 
> Maybe you want to make this into a lintian test?

One thing I feel is worth mentioning is that it is more harmful for a
package to be mistakenly Architecture: all than mistakenly Architecture:
any. The former merely wastes some disk space, while the latter will
cause actual broken packages. While the breakage would be obvious in the
case of packages containing ELF binaries, in the case of packages like
os-prober that include different scripts depending on the build
architecture, the breakage would be more subtle and time-consuming: it
will simply fail to detect things that should have been detected.

In light of this, and that there's no straightforward way I can think of
for Lintian to detect this situation given a binary package, I feel that
a Lintian test risks prompting inexperienced maintainers to err on the
side of incaution and set an incorrect Architecture field. I appreciate
the zeal involved in cleaning up those packages which are any when they
should be all, but is a Lintian test for this worth the potential
breakage?

Cheers,

-- 
Colin Watson                                       [cjwatson@debian.org]