Re: [gentoo-project] How to improve detection of unmaintained packages?

[Alec Warner <antarus@gentoo.org>]

[-- Attachment #1: Type: text/plain, Size: 5686 bytes --]

On Sat, Mar 23, 2019 at 3:32 AM Michał Górny <mgorny@gentoo.org> wrote:

> Hi,
>
> Gentoo is still having a major problem of unmaintained packages.
> I'm not talking about pure 'maintainer-needed' here but packages that
> have apparent maintainers and stay under the radar for long, harming
> users in the process.  I'd like to query potential solutions as how we
> could improve this and look for new maintainers sooner.
>
>
> The current state
> =================
> The definition of an unmaintained package here is a bit blurry.  For our
> needs, let's say that an unmaintained package is a package that is not
> getting attention of any of the maintainers, whose bugs are not looked
> at, that does not receive version bumps or simply fails to build for
> a long time.
>
> This is especially the case with 'revived herds', i.e. projects that
> were formed from old herds.  Their main characteristic is that they
> 'maintain' a large number of loosely-related packages, and their
> developers take care of only a small subset of them.  Sadly, we still
> have people who cherish that model, and instead of taking packages they
> care about themselves, they shove it into one of 'their' herds.
>
> So far we're rarely catching such cases directly.  Sometimes it happens
> when another developer tries to use the package and notices the problem,
> then finds that it's been reported a long time ago and never received
> any attention.
>
> Sometimes, after retiring a developer we notice that he had 'maintained'
> packages that were broken for years and never received any attention.
> There are even real cases of developers taking over broken packages just
> to prevent them from being lastrited but without ever fixing them.
>
> Then, some of the packages are noticed as result of major API update
> trackers, such as the openssl-1.1+ tracker or ncurses[tinfo] tracker.
> Those API changes provoke build failures, and while investigating them
> we discover that some of the software hasn't seen any upstream attention
> since 2000 (!), not to mention maintainers that could actually patch
> the issues.
>
>
> Version bump-based inactivity?
> ==============================
> One of the options would be to monitor inactivity as negligence to bump
> packages.  With euscan and/or repology, we are at least able to
> partially monitor and report new versions of software (I think someone
> used to do that but I don't see those reports anymore).  While this
> still requires some manual processing (esp. given that repology results
> are sometimes mistaken), it would be a step forward.
>
> The counterarguments for doing this is that not all version bumps are
> meaningful to Gentoo.  We'd have to at least be able to filter out
> development releases if maintainers are not doing them.  Sometimes we
> also skip releases if they don't introduce anything meaningful to Gentoo
> users.  Finally, some developers reject new versions of software for
> various reasons.
>

I've also considered to just use time.

Many *packages* have not been touched in N time. While some software
doesn't get updates often, even routine maintenance should require edits on
a fairly regular basis.


>
>
> Bugzilla-based inactivity?
> ==========================
> I've noticed something interesting in Fedora lately.  They have a policy
> that if a package build failure is reported (note: they are reporting
> them automatically) and the maintainer does not update it from the 'NEW'
> state, it is automatically orphaned after 8 weeks.  Effectively,
> if the maintainer does not take care (or at least pretends to)
> of the package, it is orphaned automatically.
>
> I suppose we might be able to look for a similar policy in Gentoo.
> However, there are two obvious counterarguments.  Firstly, this would
> create 'busywork' that people would be required to do in order to
> prevent from orphaning their packages.  Secondly, a fair number of
> developers would just do this 'busywork' to every new bug just to avoid
> the problem, rendering the measure ineffective.
>

Avoid letting the perfect be the enemy of the good here. Any metric can be
gamed by developers; but it turns out we must choose some metric to drive
the organization. I'm fairly sure not *all* developers will automate this
busywork; because *some* of us want to see the number of unmaintained
packages reduced; resulting in a net-win.


>
>
> What can we actually do?
> ========================
> Do you have any specific ideas how we could actually improve
> the situation?  I'm particularly looking for things we could do at least
> semi-automatically, without having to spend tremendous effort looking
> through thousands of unhandled bugs manually.
>

So I'd recommend avoiding a specific implementation; which means don't
trigger off of a specific signal.

Signals:
1) euscan first; because its most accurate and plausible already
implemented.
2) Date-based scanning; its trivial to implement.

So now for each package, we have 2 straightforward signals. When was it
last touched, how many versions behind?

Rules:
A package is unmaintained if it:
  - Has not been touched in 5 years
  - Is behind 3 versions AND hasn't been touched in 2 years
  - Is behind 5 versions AND hasn't been touched in 1 years

As we add more signals (e.g. doesn't build, or unfixed bugs) we can add
additional rules.

We could generate a QA report per package on the qa reports page.
If there is an API for request the QA report, we could cross-link from
p.g.o.

-A



> --
> Best regards,
> Michał Górny
>
>

[-- Attachment #2: Type: text/html, Size: 7046 bytes --]