How many users does Fedora have?

[Zbigniew Jędrzejewski-Szmek]

On Tue, Dec 02, 2014 at 09:42:24AM -0500, Ben Cotton wrote:
> Knowing people use what your make is a very powerful motivator.
> 
> > So, a revised/enhanced/even worse set of questions:
> >
> > - How many users have installed product X/spin Y?
> > - How many users have installed package X?
> > - How many users have actively removed package X?
> > - What packages are installed from non-Fedora repos?
> > - Count on non-packaged specific applications.
> > - How often are updates performed? Which packages are excluded?
> > - How often is package X updated from updates-testing?
Some of those questions go too far imo and expose too much about the
individual machines and users. Actually even more interesting than
uninstalled packages would be masked or disabled *units*, since it
is often more effective to get rid of a daemon by masking, then by
package unistallation, when the package might be brought back by a
dependency. But despite being interesting, I'd consider this too much
of a breach of privacy. I think we should stick to the list of
installed packages. Nothing more, even the fact that packages
are uninstalled should be inferred from agregate data.

> This is a good start, though I'm still not sure it addresses the "why
> are we asking?" component. The answers may be of value to individual
> maintainers/developers, but are they of use to Fedora as a whole?
> (Aside: what does "Fedora as a whole" even mean?)
I think it is useful to look at debian-devel and what popcon numbers
are useful for. From what I have seen:
- time trends show how/when people upgrade after a new version is released

  (Generally useful, but specifically when working on a package, this will
   tell me if it is at all useful to fix something in F19 branch at this point
   or not.)

- what architectures are used. This was instrumental in the Debian's
  systemd debate, since without popcon nobody would know what Debian/kfreebsd
  usage is low double digits, and Debian/hurd even smaller, so the
  fact that they are not supported by systemd is not that important.

  For Fedora, it would be useful to know which architectures are used,
  and relative popularity of spins. I think the discussions about the default
  desktop environment would be more productive if they were based on
  actual usage levels.

- comparisons of trends of alternative packages over time (postfix/sendmail/exim4,
  kde vs gnome vs xfce, etc.)

Notice that absolute numbers are not on this list. Popcon absolute numbers
are dubious, but trends and relative numbers are useful.

[...]
> - What percentage of installs explicitly installed package X versus
> installing as a dependency?
> - What is the default browser on the system?
> - How many local users have been created? How many of them have real
> shells (as opposed to nologin)?
> - What is the default desktop on the system?
> - What daemons are running on the system?
> - How many monitors are attached?
I don't think we should be asking any of those.

> - What's the hardware platform (bare metal, AWS, KVM, etc)?
Yes. We would get the architecture from installed packages, but
not the platform. It would be nice to include 'hostnamectl chassis'
and 'systemd-detect-virt' style output.

Zbyszek