[WIP] bus-factor automation

[pigoz]

This tries to give some numbers on the actual distribution of diffstats for
last year's contributors.

do not merge yet discussion is needed

[dhinus]

Awesome! 👍

I would like an algorithm to generate a single number from that data. A naive implementation could be
bus_factor = number of people with > x% contributions

For example for x = 5, you would get:
clojure/clojurescript bus factor 1 (only 1 contributor has more than 5%)
Microsoft/TypeScript bus factor 8 (8 contributors have more than 5%)

A better implementation should take the exact percentages into account and could return a float instead of an int. I need more coffee to think of how it could be implemented, though.

[caesarsol]

In my opinion this bus-factor index should be an index ranging from 0% to 100% where:

• 0% means the total of commits are done by 1 person, which is bad
• 100% means every person made an equal percentage of commits, which is good (and utopic)

what do you think?

[pigoz]

Yes, I prefer @caesarsol suggestion.

With @dhinus approach, for any number of x you will encounter bad cases. For instance for X=5, in

82.46% – wm4 – (1713809)
7.73% – pigoz – (160735)
3.10% – haasn – (64531)
0.84% – giselher – (17462)
0.83% – ubitux – (17186)
0.76% – kevmitch – (15867)
0.65% – divVerent – (13551)
0.65% – Kovensky – (13456)
0.60% – lachs0r – (12520)
0.46% – rossy – (9504)

you would get bus-factor = 2, but you are going to look at a totally different project if the first or the second maintainer leaves

[Granze]

Probably the one suggested by @caesarsol is a better approach.

The problem with the @dhinus formula is that we have to choose what x should equal to. And we could discuss this for hours, days, forever. :)

[mfirry]

Should we give Truck-Factor a try?

[dhinus]

@caesarsol I like the idea, how would you compute the index?

@mfirry that looks very interesting but seems very conservative and I'm afraid all compile-to-js languages would be TF=1 or TF=2. I tried downloading it and analysing a few repos, I got TF=2 for clojurescript and scala-js, and an error for Typescript ("Too much columns.", which sounds like a Dogescript error :P). It's also a bit of a pain to use, it requires you to manually clone the repos you want to analyse, and if you don't run it from the correct directory it fails with weird errors.

[mfirry]

So number-of-commits / number-of-committers?

[caesarsol]

Hey, I have currently no idea how to implement a similar index 😄

But I'm experimenting a little with different metrics and visualizing graph data, take a look at
https://gist.github.com/caesarsol/05d3bee9f75aa35fe7e2390fd49a2f58
(with the help of @YeasterEgg)

For a quick view, paste it in http://requirebin.com/