[plug] [OT] After Year 12

[James Devenish]

In message <20040804110052.GA24238 at cp.yi.org>
on Wed, Aug 04, 2004 at 07:00:52PM +0800, Cameron Patrick wrote:
> I got 75% on one machine I tried that on, 64% on another, 76% on yet
> another, 62% on a fourth, and so on. (The two with higher proportions
> are desktop machines, the two with lower proportions are servers.)
> I'd say that the GPL is almost certainly the single most popular free
> software licence -- and having a "market share" of somewhere between
> half and three quarters of the installed software is quite an
> achievement, certainly not something that I'd describe as "only".

Err...oh dear, I obviously have to explain this in more detail. The
basic problem (as it appears to me) is working out how frequently each
licence is used. I think there would be many competing views about how
to define the criteria for inclusion in such a survey. For example, I
would not include vapourware or software that was never distributed.
(That is, it 1000 people thought "hey, I'll write GPL software" and if
they then lose interest during the pre-alpha stage, I would not count
those 1000 projects.) I also think it would lack meaning if I included
software that "no one" uses. So, your argument about the popularity
contest is valid in my view. But it would be difficult to evaluate this
outside of captive communities. So, I don't know how I could achieve a
meaningful world-wide result. So, I tried the next-best thing.

It could be suggested that if we go to Freshmeat or SourceForge or
VersionTracker, we could take measurements from those communities.
(Although we we would first have to exclude software for which source
code is not available.) However, even if we did this, we could not
extrapolate the results to the wider world.

I would expect that if I did a survey of software licences on a OpenBSD
system (for example), the environment would be biased towards BSD-style
licensing. I could use the results in this environment to say something
about the penetration of BSD-type licences into a BSD-based system. (But
that does not reflect the relative presence or absence of BSD-type
software in the wider world.) It would simply be a curiosity about the
BSD-style environment. My intent with the GPL/Debian example was merely
an analogue of this.

The purpose of my Debian demonstration was to provide samples from an
environment that is heavily weighted towards GPL software. That is, to
ask "what is the penetration of the GPL into a realistic GNU-based
system"? The results indicated that 1-in-4 and 1-in-2 packages on our
machines were non-GPL despite the bias of Debian towards GPL libraries
and UNIX tools. That is, 24-45% of packages were non-GPL. To draw an
analogy: if I was in a room where 24-45% people wore red shirts, and
55-76% of people wore blue shirts, I would not say that "everyone" was
wearing blue shirts, even though blue shirts were certainly be the most
popular shirts in use.

> I'd be curious to see what the the results were like if you weighted
> them by popcon results, but not curious enough to actually script it
> myself.