Guillem Jover wrote:
> On Fri, 2009-11-27 at 21:09:57 +0200, Eugene V. Lyubimkin wrote:
>> Guillem Jover wrote:
>>> Something like this could do (on a clean lenny chroot):
>>>
>>> r...@gaara:~# export LC_ALL=C
>>> r...@gaara:~# dpkg --auto-deconfigure --unpack libc-bin_2.10.2-2_amd64.deb
>> This:
>>
>> a) may break the planned changeset by allowing dpkg to do unwanted actions;
> 
> This is already the case, dpkg might disappear a package in some
> situations, and the front-ends should be able to cope with those
> cases, that's why those events are reported via the status-fd pipe.
Personally I consider this as bad design, so I'm trying to minimize the impact
whenever possible.

>> b) is package(libc)-specific and not generic solution, because the action
>> sequence proposed in the first message of the thread will probably work
>> flawlessly for almost all other packages, it's only dpkg is "smart" enough to
>> break the upgrade with no actual reason, and the front-end has no way to
>> predict it.
> 
> Always using --auto-deconfigure should be safe, and is the generic
> solution. Your “flawless” solution involves several force options,
No, this is not true. The only force option required today for this action is
"turning off" PATH check by --force-path. The '--auto-deconfigure' is a hack
IMO, the right way is to provide, finally, a way to request several different
actions at once from dpkg.

> making files unavaiable from the file system for a time window, that
> could be not short if there was an error during the transaction that
> made dpkg exit, it's certainly not the correct solution.
I guess if there will be some unpredictable error during transaction, dpkg
database of 'what files belong to what package' may be messed by this in-place
replacing. Am I wrong?

> And this case can be spotted easily when a package replaces another
> one and that other breaks this first, the replacing one should always
> be installed first, otherwise the system will end up with missing files
> for a period of time.
There may be much worse intermediate situations during upgrade.

> Also your continuous use of force options in cupt leaves to be desired
> (just saw the new --force-bad-path), pointing as excuse a breakage in
> dpkg when it seemed pretty clear the ordering chosen was wrong
From the point of view of the high-level package manager - no.

> and
> can leave the system in a broken state. Not to mention as I've said
> several times already that apt manages to handle all those situations
> with *no* force option at all.
I've said already that I don't care what can apt do.

-- 
Eugene V. Lyubimkin aka JackYF, JID: jackyf.devel(maildog)gmail.com
C++/Perl developer, Debian Developer

Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to