I happened to start a long discussion thread on the Fedora devel mailing list. My first message was (abridged):
While we’re on it (in the form “how many devs do we have”): How hard/impossible/unsuitable would it be to get a usable estimate on the # of users, per package?
Here are so many problems, technical, policy, resources, (others?). That said, feedback in the form “How many users uses/installs my package” would IMHO be a great input for any packager. It would be the difference between dropping the package in a black hole vs getting a message back from the community.
Somewhere down this road, Ben Cotton remarked that before we can resolve how this should be done we need to define the questions to ask. Here, an attempt to summarize this, some 50 messages later.
We will probably not be able get useful absolute numbers. However, relative numbers and trends should be both possible and useful.
Privacy concerns blocks collecting of otherwise useful info. Based on this, I have more or less limited the questions to what can be deduced from installed software and yum/dnf logs.
Q: How much are products and spins are actually used?
Metthew has indicated that we need some usage counts on our products in bz #1156007. Now, shouldn’t spins be treated the same way?
Q: How is the package I created actually used?
As a packager, it would be great to have some feedback if users actually installs “my” package. Seemingly trivial, this is about motivating packagers. The most interesting question here is likely How many users have installed my package directly? (i. e., not as a dependency). Here is also the negative feedback How many users have actively removed my package?
Here are a lot of alternatives see e. g. this message. I suggest we limit this to what might add some motivation and feedback to a packager.
Q Which package does users prefer?
Info on the relative popularity of alternative implementations like postfix/sendmail/exim4 or KDE/Gnome would probably improve some discussions and decision making.
Q: Which non-Fedora packages does users need to install?
When users needs to install applications which not are in Fedora repos, this is an argument to package those. This falls into two categories: sw packaged in non-fedora repos like COPR and rpmfusion or unpackaged software. Trying to probe for specific unpackaged software e. g., chrome is technically possible but might raise privacy concerns.
Q: How are packages updated?
This question is partly about how the overall update process works., so we can improve it over time. But it’s also hint for a packager whether a certain feature/fix needs to be back-ported to older branches.
Q: How many users uses platform x/architecture Y?
Knowing usage of platforms (KVM, bare metal, AWS, etc.) and architectures (i386, x86_64, amd64, etc.) is important input when analysing the impact of not supporting any of these in some respect.
Questions not asked due to privacy concerns:
- Default browser
- Number of installed “real” users (i. e., # true multi-user installations).
- Default desktop
- How many monitors are attached?
- What daemons are running on the system?