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

[Peter Samuelson <peter@p12n.org>]

First, +1 to the Pre-Depends.  I think it's well-justified here.

[Eugene V. Lyubimkin]
> 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].

> [1] If we count the situation for resuming broken upgrade, there is a
> some chance you'll have to call dpkg manually or some hacks
> to proceed anyway.

Not always.  There are states dpkg goes through that 'apt-get install'
can "recover" from on its own.  You don't always have to go to dpkg.

Also, what if apt wants to call one of its auxilliary binaries during
the install/upgrade?  I imagine it's not implemented that way _now_,
but a Pre-Depends would make such a thing a lot safer if they want it.
(Same is true if they want to dlopen a library during the install, but
that's somewhat less likely.)

> I object to this change.

On what grounds?  So far we have:

Pro:
  - Would make upgrades more robust
  - Would make apt implementation more flexible

Con:
  - In general, Pre-Depends enforces stricter constraints
  - ???

I thought Julian gave a good answer to the issue of the stricter
constraints: apt is quite low on the dependency chain, depending only
on some quite low-level libraries, so the impact should be minimal.

-- 
Peter Samuelson | org-tld!p12n!peter | http://p12n.org/