[Aptitude-devel] Bug#787658: aptitude ignores available updates for a while after downgrades

[Manuel A. Fernandez Montecelo]

Control: tags -1 + pending - moreinfo


Both bugs #787658 and #714429 are manifestations of the same underlying
problem, and after thinking about this and diving into the code, I
decided that it is actually a bug and not a feature.


2015-09-19 12:50 To Christoph Anton Mitterer:
>>
>>Now the problem from aptitude side is, that after the packages have been downgraded
>>(in the example above the curl/libcurl packages) it doesn't offer them for upgrade
>>anymore, even though unstable is still enabled in sources.list and the newer packages
>>are still available.
>>[...]
>>
>>Neither "Update package list", nor "Clean package cache" or "Clean obsolete files"
>>resolves this situation.
>>But at least this time (I haven't tried that before), "Cancel pending actions" plus
>>restarting solved the issue, and the packages re-appeared for upgrade.
>>
>>So far (as said, I haven't tried the above before), the situation usually resolved
>>by itself after a while (I susupect it did when really new Package lists came in
>>from the repo).
>>
>>
>>I think I have seen the whole issue even when packages where downgraded, but when
>>I had already commented/disabled the "lower" repo (e.g. test) again.
>>
>>
>>Last but not least, since this may be "used" to accidentally hide security upgrades,
>>I selected important as severity. I'd guess a higher severity is not needed, since
>>downgrades typically don't happen automatically, so the admin has at least a clue
>>that he runs on an older version.
>
>I am not sure why it is behaving in this, if it's an explicit
>design/implementation decision or by chance, but I would not be
>surprised if this behaviour is intentional and it's there from the
>start, or comes after complaints along the lines of:
>
> "I want to keep this package downgraded and aptitude keeps nagging me
> to upgrade it all the time... please make it stop".
>
>I have seen lots of bug reports with similar complaints, not about a
>specific issue, but in the same sense "aptitude is not smart enough to
>second-guess my perfectly sensible intentions or follow my commands
>exactly, and I have a good reason to choose what I chose".

I still think that this might be a problem and that we will get reports
asking to revert the new behaviour.

But after investigating it and thinking about it, I thought that using
downgrade+holds is probably good enough aproximation if people ask about
this (and hopefully now they work better and across apt as well).


>After all, if one *actively* downgrades a package version, while one
>doesn't update the package list, aptitude is doing what was requested:
>keep the version that was specifically selected (against the default
>recommendation, that keeps being the same one until new become
>available).  Maybe it's because that version is broken, or makes you
>install 500GB of data that you don't want because of a mistake in the
>packaging, so you know for sure that you don't want the newer version
>until a new one fixing this becomes available.  And when a new version
>becomes available, it lets you know so you can decide again.
>
>The fact that it doesn't show in the set of Upgradable packages is also
>consistent with this "don't nag me all of the time".  Picture this
>user's rationale:
>
> "It is in the list of Upgradable all of the time, so after pressing U
> I still have to go to the N packages that I want to keep in the older
> versions and select the old version *one by one*, *every day* when I
> update testing / unstable / experimental.  *Annoying*".

I also think that this may be a problem for somebody's usecase :)


>So, in summary, I don't know if this is a bug or a feature (but I am
>leaning towards the latter).  It will need more investigation / thought
>in any case.

As I said above, I think that the fact that it's not updating the state
after performing the action is a bug; but even worse is the effect that
it makes difficult to search for packages that were downgraded a while
ago, even with search patterns and the command line (bug #714429), there
seems to be no alternative to search for them other than going package
by package, which is not very good.

So in the end, marking as +pending, to be included in the next release.


Cheers.
-- 
Manuel A. Fernandez Montecelo <manuel.montezelo at gmail.com>