Re: severity of 442668 is serious
On Sun, 2008-09-28 at 21:19 -0700, Steve Langasek wrote:
> Redundant versions of BDB in the archive unnecessarily bloat the release and
> Debian's install footprint, and impose a burden on the Debian DB packaging
> team. There are currently five versions of BDB in lenny, whereas there are
> only 7 packages in lenny that depend on db4.3 - *all* of which ought to have
> transitioned off of it at least a year ago (when db4.6 became available;
> db4.4 became available in 2005, but was not free of regressions for all use
> cases). And all but three of these are fixed in unstable. It is not
> justifiable for apps that make only the most basic use of BDB, as reprepro
> does, to get to keep their own copy of libdb for a release.
It's certainly a problem, and I'm completely in agreement with your
assessment of the problem.
At the same time, it's also the case that BDB seems to change
inordinately frequently. There seems to be a constant churn, and always
with vague statements about what exactly the regressions are. A
maintainer generally does not know exactly how a given application uses
the library, and the bug reports a developer gets usually don't contain
much information beyond "please upgrade to the new db; there are
regressions for some cases so be careful." When I see a bug report like
that, it goes to the very back of the pile.
> It's arrogant to think that your package needs special handling for the
> "disruptive" change from db4.3 to db4.6, when dozens of packages have
> already made this transition without incident. I've reviewed the reprepro
> source code, and there's nothing extraordinary about its use of BDB -
> nothing that should break when switching to db4.6, and nothing (such as
> transactions) that requires special upgrade handling.
So in this particular case, how was the user in question to know this?
Maybe Bernard Link is a BDB expert, but most of us are not. What I see
is that a requset was made a year ago, and he responded fairly promptly
asking for time. No further comment was made by anyone until Luk
declared that "we want to drop the package" makes the bug RC.
And Bernard expressed some concern that past BDB version changes have
been quite brittle; this has been my experience as well.
I completely agree that we don't want all those BDB versions; I
completely agree that Bernard should upload a fix for the bug; I
completely agree that this is a real problem.
But at the same time, I'd really appreciate some understanding for the
position of the poor developers who don't necessarily have the latest
info on the various regressions and broken compatibility associated with
upgrading a random library. I believe it is part of the BDB
maintainers' job to provide this information, and otherwise help out.
In the case of Bernard's package, all I can see is that they essentially
ignored his concern and his request for a year.
Thomas
Reply to: