[Soc-coordination] GSOC idea for next year: undusting some packages

[Michael Vogt]

On Wed, Oct 01, 2008 at 06:41:51AM -0700, Daniel Burrows wrote:
> On Wed, Oct 01, 2008 at 07:10:49AM +0200, Christian Perrier <bubulle at debian.org> was heard to say:
> > ===============================================================
> > Maintaining our key packages
> > 
> > Has anyone noticed how loosely APT is maintained since about...several years?
> > 
> > Michael Vogt is currently the person doing uploads and work but, being
> > committed to other things (IIRC in Ubuntu...or in real life), he
> > certainly can't devote enough time for this.
[...]

I agree here and I'm sorry that I have trouble devoting more time to
apt. With a full time job (more than full time very often) and a life
its not easy to manage.

>   I don't know if GSoC would help, but I agree with what Christian said.
> The state of apt right now is a scandal.  This is a core part of the
> package management system for our project as well as several commercial
> distributions, and yet it's been neglected on and off ever since I
> started working with it nine years ago.  Sometimes someone picks it up
> for a while and fixes a few of the problems, but it's been a long time
> since it had a really active primary maintainer.  (I hope no-one is
> offended if I say that: Michael and Otavio have been doing fine
> work [0], but mostly acting as caretakers, not project leaders.  I
> think they would agree with me.)

I'm not quite sure if "scandal" is the right word. But I agree that it
has been neglected and really needs more work.
 
>   There isn't anyone right now who has enough time to stay on top of
> the project even to the extent that I do on aptitude (for instance, I
> can't because aptitude takes all my free time).  The codebase is quite
> crufty and a pain to work on, and there are serious design issues that
> lead to bugs which users bump into on a regular basis.  No-one has the
> time or confidence to do the redesign and cleanup work that is necessary
> for the long-term health of the project.  We also have people off
> writing wrapper libraries for apt because the API is a pain to use and
> doesn't use modern C++.  It seems to me that it would be better for us
> to improve the API instead, but we don't have time to do anything but
> fix the worst bugs as they come up.

Agreed.

>   I'm not sure exactly why this is so.  I tend to think that it's a
> combination of people not being very interested in package management
> (because it's not something that directly solves problems for them, but
> more of an obstacle to getting their work done [1]), and perhaps the
> fact that Debian doesn't attract people who want to write medium-large
> pieces of C++ software.  But more the former than the latter.

I think the reason is that apt is "good enough". It works most of the
time for most people. Anohter problem might be because people just
assume that because it is a key package it is certainly well
maintained and has a good code base (that was my assumption at least a
long time back :). And a third problem could be that its such a core
package that it might be a bit scary for some contributors.

For short term improvements I think just having someone doing bug
triage, testing and confirming issues would already help a lot. Just
having someone to talk to over irc or jabber about things that I have
missed would help me a lot already.

Cheers,
 Michael