[Popcon-developers] popcon & derivatives

[Bill Allombert]

On Sun, Jan 19, 2014 at 09:45:55AM +0800, Paul Wise wrote:
> Hi all,
> 
> On the derivatives list we got a question about the contents of the
> derivatives guidelines being out-of-date with respect to popcon. I
> replied saying I would followup with popcon devs.
> 
> https://lists.debian.org/debian-derivatives/2014/01/msg00004.html
> https://wiki.debian.org/Derivatives/Guidelines
> https://lists.debian.org/debian-derivatives/2014/01/msg00007.html
> 
> Do we have any estimate about when a derivative should avoid sending
> reports to Debian because it might overload the server? For example I
> guess Ubuntu would be above this limit since it has an order of
> magnitude more popcon submitters than Debian.

Sorry I did not reply the last time.

The issue is not so much the number of derivative subscribers, but how far the
extra data will help rather than disrupt the statistic.
This is difficult to gauge, but is linked to how much the derivates Package
file and default installed set are different from Debian.
The computation of the popcon report relies on the correct Packages files.
For that purpose we require a local debian mirror on the popcon.debian.org.
And obviously the default set of installed package in a distribution has a
great influence on the average popcon report.

So derivatives using very different Packages files should probably set up
their own popcon server. It is theroretically possible to host it on popcon.d.o
but this would require the Package files of the derivatives and requires
significantly more processing resources.

On the other hand, distributions that take a normal debian install and extend it
using new packages and backports should be OK I think
(even though we would get more accurate results if we knew the dependencies of the
extra packages).

> The guidelines talk about adding a version number extension. AFAIK
> popcon.d.o doesn't use that, except to differentiate that Ubuntu
> popcon version that sent reports to Debian. 

Debian rejects reports with 1.56ubuntu1 due to a bug in this version.
Otherwise, it handles all versions in the same way.

> I note that there are
> several version extensions already listed on the popcon site, which
> appear to correspond to Canaima, Ubuntu and Emdebian. 
> Would it be a
> good idea to extend popcon.d.o to also use popcon version extensions
> as an indicator of derivatives and mention this as an alternative in
> the derivatives guidelines, at least until most derivatives are using
> that version of popcon?

I do not think this is a good idea. The number of such report is minuscule
and it seems much better to use the dpkg Vendor field. In particular it does
not require the derivative to change the popularity-contest package.

> If necessary I could extend the derivatives
> census scripts to output a list of current and historical popcon
> versions and which derivatives they correspond to.

Yes, this could be useful, but only if we receive a significant number of
such submissions to have meaningful statistics.
We could then cross the Vendor field with the popcon version to get the
derivative version.

(categories with a single report are not only statistically useless but
actually dangerous because the anonymity can often be broken).

I hope this answers your questions.

Cheers,
-- 
Bill. <ballombe at debian.org>

Imagine a large red swirl here.