Getting popcon back in shape

Senko Rasic senko at
Fri Jan 6 18:56:05 GMT 2006


On 1/6/06, Matt Zimmerman <mdz at> wrote:
> On Fri, Jan 06, 2006 at 10:55:21AM +0100, Lucas Nussbaum wrote:
> > I personally find that is a very useful resource for
> > developers : it can help answer questions like :
> >
> > However, it seems that we get very few popcon results, mostly from
> > old-timers (was popcon once enabled by default during the warty release
> > cycle ?). Also, the pages have not been updated in a very long time (see
> > Bugzilla bug 12761).
> >
> > Is there something planned to get all this back in shape, or should we
> > use Debian's popcon results instead ?

We at Ubuntu-hr team have started a somewhat related project - a DVD installer
which would also serve as a repository for popular packages (only those that
can be legally put on DVD, to avoid legal issues). We've considered using
popcon data, but we feel that many users don't enable it, so we
created a "poll",
in which users can upload list of installed packages, and we generate statistics
from that.

So far we've got 76 package lists (we're keeping a low profile until
we can come up
with a first release of the .iso), and we're using this in conjunction
with the popcon
data. You can visit for more details.

> In general, popcon doesn't answer these questions as effectively for us as
> it does for Debian, due to the fact that most users use the preselected
> desktop environment.  These packages dominate the results, and the variance
> in the remaining packages doesn't seem very informative.

This is true, but I believe that can be solved if we ignore all the
packages from
the preselected environment (or, ignoring all packages from the
installation cd).
The remaining packages *do* provide information - the problem is just in
gathering enough samples so that statistics make sense.

The other problem that we're facing is with users who upgraded from previous
versions. They have packages which were useful in the previous versions but
are obsoleted in the new one. These packages stay on the system because the
upgrade installed new packages in parallel. And these show up in the lists
although practically nobody is using them any more.

> general, genuine application use is more interesting than package
> installation and file access, and this is more difficult to measure.

I agree. With so many apps to chose from in universe/multiverse, and with
synaptic as user-friendly as it is, it's easy for someone to install a dozen
variants of an application to test it out, and "forget" to deinstall them after
testing. These aren't used at all, but still show up in lists.

> Surveys, while they have their own problem, would seem to give us more
> useful results.
> For example, a Fridge poll about Epiphany vs. Firefox would give some idea
> of what the user community prefers today.

Maybe that could be expanded a bit. We could create a big poll, divided into
sections (each section describing one application type, or use case, for example
"Web Browser"). The poll could look like a "wizard", asking questions for each
section, like:

(Page 1)
"What web browser do you prefer?" Firefox/Epiphany/Mozilla/Lynx
(Page 2)
"What's your favorite word processor?" ooo2/abiword/emacs

This would provide a general overview of what the "typical user" (yeah,
I know there's no such person) wants.


Senko Rasic <senko at senko dot net>

More information about the ubuntu-devel mailing list