Trends in the fetchmail project's growth
The scattergram below was made with Gnuplot 3.7 from data pulled
directly out of the project NEWS file using two custom shellscripts,
timeseries and growthplot. If you see a broken-image icon, upgrade
to a browser that
can view PNGs.
The graph shows the population growth of the fetchmail project. The
horizontal scale is days since baseline, which is when I started
collecting statistics in October 1996 at version 1.9.0. Left vertical
scale is number of participants. There is one data point for each
release; therefore, the changes in density of marks indicate release
frequency.
The peak in the earliest part of the graph (before the note "Bad
addresses dropped") seems to be an artifact; I was not regularly
dropping addresses that became invalid at the time. Turnover on the
list seems to be about 5% per month (but that's just my estimate, I
don't have numbers on this).
The blue scatter of squares is total
participants. The green scatter of crosses is
the count of people on fetchmail-friends after I split the list. The
violet scatter of x marks is the population
of fetchmail-announce after the split.
The brown scatter of asterisks tracks project
size in lines of code (right vertical axis). The scale relationship
between this scatter and the other three is arbitrary.
This graph is quite revealing. Several trends stand out:
-
Over time, the project population displays rather consistent linear growth.
-
The key event in the project's lifetime was release 4.3.0 in October
1997, when I declared the code to be out of development and in
maintainance mode, and split the fetchmail list.
-
The run-up to 4.3.0 saw the most intensive spate of releases in the
project's history (the gap in that run happened when I took a two-week
vacation). It was followed by a significant slowdown.
-
After 4.3.0, the developer population remained fairly stable around
an average of about 250 participants.
-
Essentially all population growth after 4.3.0 happened on the announce list,
among people using fetchmail but not active co-developers.
-
The growth trend in code size looks sublinear, perhaps logarithmic.
The linear growth trend in population is particularly interesting; a
priori we might expect geometric or logistic growth, given that the
project spreads by word of mouth. I have not yet been able to
plausibly imagine a growth model that would produce these numbers.
There are some other pages doing similar things:
-
Here
are growth statistics on the debhelper packaging utility.
-
Here is a page on the
vocabulary of the Linux kernel.
Eric S. Raymond <esr@thyrsus.com>