New package metric: Number of contributors

[daveverwer]

Related to the work I’m doing on scoring, it would be great if we could get a number of contributors from GitHub’s API.

My thinking is that packages that have more contributors will generally be of higher quality.

[MaximBazarov]

I don't think it's a fair metric, people with more friends or more public, or more extravert, aren't necessarily produce better solutions.

[MaximBazarov]

To look at it from another angle, say I developed a best solution in the world for the problem at hand, package has some flaws but it's the best so far. It gets 5000 stars and some docs and tests, it has a lot of issues but, I resolve them from time to time. Now with the current system, it's the top 1 package, add to this a couple of copies of the same solution, and we eventually have top 7 packages that always will get 99% of attention.

Now a couple of years passed, there's a new solution, that doesn't have flaws, and so simple so it requires maintenance only major swift releases, there's no way for this package to be even discovered if we don't take the package quality into account.

[MaximBazarov]

in short, once you successful, you are the best solution, I don't think it's good for the evolution of the ecosystem.

[MaximBazarov]

I'm totally ok if you want to go with the current approach to see if I'm anywhere close to the reality btw.

[MaximBazarov]

👍

As an example, the Swift Server Work Group Incubation Process requires a minimum of two contributors for incubation.

I also like that, I think it makes sense to eliminate the "BUS factor"

[finestructure]

What I was getting at when saying in Discord that this doesn't compare to Google: you're not picking dependencies every day. Or if you are, you're doing something wrong. And when you do and you stop after looking at the first five, you're also doing it wrong. This is not a job an algorithm can do for you. You have to inspect those results in more depth and in more breadth than you would a shopping query.

I don't believe we should be spending too much time trying to come up with a Google-like page rank (which not everyone will agree with anyway) for those reasons.

Of course we'll try and add more metrics like tests and contributors and augment the score over time. But the point of that is mainly to make the detailed inspection of packages easier rather than come up with a "perfect" page rank.