Re: ':any' syntax in package names in jessie/sid Packages
+++ Stuart Prescott [2014-04-18 17:25 +1000]:
> Hi Eugene,
>
> > It seems that in jessie (and similar in sid) a number of packages [1]
> > started to use ':' symbol in their dependency lists as part of package
> > names. This is, if I'm not misreading the Debian Policy §7.1 and §5.6.1,
> > is not allowed.
> >
> > Suggestions for issue's severity and how to proceed?
>
> I think you've found yet another "multiarch is not documented in policy"
> bug.
You are right that there is a collection of policy changes needed to
reflect the multiarch changes.
> Unfortunately, the people who understand multiarch well enough to write it
> up for policy haven't done so which leaves us with no normative
> documentation in policy for the the Multi-Arch field in Packages, no
> description of how the package manager should deal with multi-arch packages
> and their dependencies and no documentation of best practices for -dev
> packages etc.
I agree it's not all (any?) in policy. I don't agree that there is no
documentation for those things. The multiarch spec is pretty good:
https://wiki.ubuntu.com/MultiarchSpec The :any and :native usage, and
how package-managers should implemnt it is in:
https://wiki.ubuntu.com/MultiarchCross
along with some info on -dev packages, but I agree we could do with
some more/better now that we have enough experience to say something
useful about best practice.
> As with the rest of multiarch, the documentation of python:any is at:
>
> https://wiki.ubuntu.com/MultiarchSpec#Extended_semantics_of_per-
> architecture_package_relationships
>
> (I believe that there are some aspects of that document that have not been
> implemented in Debian or have been implemented differently in Debian -- it's
> not a substitute for having this in Policy)
So far as I know that spec doc is correct for Debian and Ubuntu. The
only significant difference is that Ubuntu has patched apt to assume
that :all packages are M-A:foreign by default. Debian has not, and
requires all packages to be so marked explicitly.
I agree that it's time to merge those two docs into policy (or at
least into each other to make one consistent spec).
I've never made a policy change, but I guess I do know enough about
multiarch to help with that. Does one just send patches as bugs aginst
policy? I've never quite understood how things get from 'something
someone hacked up' to 'policy'.
As xnox says there is still some pending changes around the interpreter problem, as described here:
https://wiki.debian.org/HelmutGrohne/MultiarchSpecChanges
And that debate is part of the reason this stuff hasn't been
considered 'finalised' and thus ready for policy. But I think the core
stuff is now well-enough used that at the very least policy should not
be inconsistent with it.
Wookey
--
Principal hats: Linaro, Emdebian, Wookware, Balloonboard, ARM
http://wookware.org/
Reply to: