[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [RFC] Changing APT to pre-depend on ${shlibs:Depends}

On Wed, Apr 27, 2011 at 18:34, Eugene V. Lyubimkin <jackyf@debian.org> wrote:
> On 2011-04-27 14:53, Julian Andres Klode wrote:
>> I hereby request comments on changing APT to pre-depend on
>> ${shlibs:Depends}.
>>         When we upload a new version of APT, depending on a newer
>>         library version (due to new symbols, whatever), and APT gets
>>         unpacked before the library, the system's ability to upgrade is
>>         broken, unless you fix it manually via calls to dpkg.
> First, this statement is not true because other package managers exist.

Exactly my argument as this would properly be true for synaptic or
update-manager, too, so these should have their dependencies as
pre-depends, too, as a user used to gui will be unable to do it by hand
as much as a user used to APT will be able to do it with dpkg…
And if that is too much over the top i don't want to be the person who
has to decide if cupt, smart, rpm and alike are now in group 1 or 2…

> Second, why the APT's ability to upgrade is broken under these
> conditions? Unless I'm missing something, the upgrade cannot be
> started in the middle of another upgrade [1].

I see two chances for this:
1) unpack apt with dpkg - your problem. A user who is able to install apt
   with dpkg should be able to fix it with dpkg. APT is not essential
   after all…
2) power loose - you will properly have bigger problems than a maybe
   malfunction APT at this point if you will get into this problem at all.

The point Julian has is:
If APT is used to handle his own upgrade it will set itself nearly pseudo-
essential - featuring immediate configuration - so the difference of
this change would be in reality none existent.
But yeah, if it is already done in APT anyway, why forcing
it on dpkg or any other package manager not using APT…

Even without that the current order implementation would properly not
allow the unpack of apt before the unpack of libstdc++ even through
it would be allowed by policy (it would be just silly as it would break
the chance of a working 'dpkg --configure -a' after a power loose or so,
so it would be limited to places it would be impossible to do otherwise).

Best regards

David Kalnischkies

P.S.: The title says RFC not FYI so could we avoid mentioning ctte here?

Reply to: