On Tue, Jan 11, 2011 at 05:55, IOhannes m zmoelnig <zmoel...@iem.at> wrote:
> On 2011-01-11 09:42, IOhannes m zmoelnig wrote:
>> On 2011-01-11 00:06, ales...@users.alioth.debian.org wrote:
>>>  Package: libgmerlin-dev
>>> -Architecture: all
>>> +Architecture: any
>>>  Section: libdevel
>>> -Depends: libgmerlin0 (>= ${source:Version}),
>>> - libgmerlin0 (<< ${source:Upstream-Version}+1~),
>>> +Depends: libgmerlin0 (= ${binary:Version}),
>>>   pkg-config,
>>>   libxml2-dev,
>>>   libgavl-dev (>= 1.2.0),
>> i see that you also reverted the Depends section to match the exact
>> binary version of the installed library.
>> i'm curious on why and about the (dis)advantages of the 2 approaches.
>> and how they are tied to the Architecture (as you seem to atomically
>> commit both the architecture and the depends field)
> i guess, in Architecture:all you cannot use the ${binary:Version}.
> but afaict, you could use the ${source:Version} in Architecture:any, no?

The problem is with binNMUs. When a package is rebuilt with no source
changes, the arch:all and arch:any parts of the package get "out of
sync": they have different versions. Thus you cannot have such a
strict relationship when mixing arch:all and any packages. Ensuring an
appropriately tight relationship in the arch:all case is hard and can
lead to errors. That's why Reinhard said it's not really worth it
(because the space savings are tiny).


Felipe Sateler

pkg-multimedia-maintainers mailing list

Reply via email to