Lies, Damn Lies, and Statistics
Aldo Cortesi contacted me privately via email with a response to this post. With his permission I've updated this post with the substance of his response and my own rebuttal. The updated sections are marked with UPDATE/END blocks.
So after chromatic linked to Aldo Cortesi's post The impact of
language choice on github projects I was curious. After reading the
post, I was shocked and dismayed at some of the analysis Mr. Cortesi
attempts to derive from his data. I brought it up on IRC and decided at
rjbs's advice is best, "just ignore it."
Obviously that didn't stand, because I saw Flavio Poletti and osfameron both comment about Cortesi's post. Upon second reflection I realized that perhaps there was some nice value to be taken out of the data he presents.
At the time Cortesi extracted his data from github, Perl was the 6th most popular language.1 Cortesi's first two illustrations look particularly nice for Perl. First he pulled out the median number of contributors (emphasis is his).
This means in general that Perl projects have more developers than the other languages measured except for C++. This is a good sign. Next he pulled out the median number of commits (again emphasis his).
Most projects have around 75 commits.
Perl however in his data (again like C++) has a much higher media, nearly 200 commits median. His explanation for this was (to me) very unsatisfying.
The Perl and C++ data, however, seems significant - projects in these languages on average have a much longer commit history. I suspect that this is due to a decline in popularity in these languages. Recall that I collected data only for projects that had recent commits. If fewer new projects are created in C++ and Perl, we would expect projects in these languages to be older, on average.
I think that the sheer fact that Perl (and C++) are both much older languages is enough to explain the discrepancy. Perl has many projects that have converted from one or more version control systems (Moose for example has a revision history that is only a year younger than Git itself, and is considered a young project in the Perl world).
The Perl community also tends to get behind a few good projects. This isn't really obvious unless you're active in the Perl community, there are many many things that aren't obvious unless you're in the Perl community, it actually is a problem but one for a later blog post.
So to extrapolate further, the Perl community tries to engender people rather than create a new project, to use one of the 21 thousand existing projects to solve their problem. This explains why we have a higher median number of contributors to our projects as well as why our projects are on a whole older.
Cortesi pointed out the following.
The first point where you strongly disagree with me is my conjecture that the higher median commit count for Perl and C++ is due to a decline in popluarity in the languages, which means that my sample is skewed towards "older" projects. You counter that it might just be that Perl and C++ are older, and that this is reflected in the project data. This doesn't sound right to me - for one thing, Perl was first releaed in 1987, and Python in 1990. I just don't think that a 3 year difference over a 20+ year history can explain the size of the effect. One way to approach this might be to look at the rate of project creation over the last few years, which would be easy enough to do.
He's right, I'm simply parroting the stereotype that Perl is older than Python (or even Ruby which came into existence in the mid 90s) when I rightly know better. The stereotype however does have some merit. It's generally accepted that Perl had it's second wind during the dot-com era and that was when Python had it's first wind.
I'm honestly not sure that the data can be conclusive argued one way or the other without perhaps correlating the age of the data (time since first commit) in there as well.
Next Cortesi measured number of files touched per commit (emphasis his).
Most commits touch about 4 files, with C++ touching somewhat more, and Perl, Python and Ruby somewhat less.
This is most easily explained by the nature of the dynamic languages that more is done in smaller modules, blah blah blah. Neither Cortesi nor I really found anything interesting in this because it upholds the standard stereotypes.
Next Cortesi talks about Contributors (emphasis his).
Here Perl clocks in at exactly the median. If I were to guess I think the meaning here is that on average Perl projects have on average a larger core group of contributers with smaller one off commits, and that these two facts offset each other. This story reflects the information above (higher median contributer count, older more established projects), as well as the next point Cortesi makes (emphasis his).
For all languages, a small fraction of the committers do the vast majority of the work. This won't be news to anyone in the Open Source community. More interesting, though, is the fact that C, C++ and Perl projects are significantly more "top-heavy" than those in other languages, with a smaller core of contributors doing more of the work.
This really just reinforces the previous point that Perl's projects are older and more established, with a stronger commitment from it's core contributors. This could be read as Perl being a dying language, just like C and C++. This could also be read as these three languages not being fad languages, where people swoop in, commit a whole bunch and then drift off. Really there isn't enough information here to tease out which is true.
Finally Cortesi tries to make a correlation between the number of committers for projects and something. I think however there may be a problem with his analysis. First he measures commits vs committer as a measure of how well languages recruit and retain contributors. Perl here shows up fairly well actually, third only to Python and Ruby. I strongly suspect that the significant discrepancy in data between these languages (there are more than twice as many Python projects included as Perl, and Ruby is four times again as much as Python) may come into play, but I've no evidence to back that up.
Cortesi noted the following, which is a fair point I overlooked.
Here, I'd like to point out that the graph shows the total commits vs the total committers per project. As such, it doesn't tell us anything about the retention of committers - a committer who makes a single commit and leaves the project is still counted. I've been thinking about how to measure committer retention - perhaps looking at the timespan bracketed by the first and last commits of a committer, as a percentage of the lifespan of the project.
The next set of graphs shows the number of commits per day over the first 300 days of a project. Perl here shows a strong decline, the strongest actually. Coming in second is C, then Ruby being the last obviously show a decline (Python and C++ could arguably show a decline, but they just as arguably are flat). The other languages are all flat, or show a slight increase.
A suggestion came up in the comments to Cortesi's that Perl projects tend to be "complete" and move into maintenance mode. Cortesi actively rejects this idea. While I disagree with his analysis2 I think he's correct in refuting this. I suspect that instead Perl projects quickly move to a point where they are obviously inferior to another project and are abandoned.
This story would play out with the earlier arguments regarding the
commit count, and the age of Perl projects. The story behind
DBIx::Class is a good example (even though it works counter to my
argument). Originally Matt Trout used and actively contributed to
DBIx::Class was intended to research ideas for moving
Class::DBI's codebase forward. It wasn't until fundamental design
DBIx::Class were discovered that
the fork of the community really occured, and even then for the longest
DBIx::Class maintained a CDBI compatible API.
This story plays out more often in the small, and in many of those the research project is found to be fundamentally flawed, or folded back into the parent project, or abandoned due to lack of community interest. The fact that this story plays out again and again in the Perl community is in some ways very ironic considering the Perl motto of TIMTOWDI, but this is the basis of the new motto TIMTOWDI BSCINABTE (pronounced Timtoady Bicarbonate and meaning There Is More Than One Way To Do It, But Sometimes Consistence Is Not A Bad Thing Either).
Finally, it's obvious in the analysis and comments that Cortesi dislikes Perl. He is allowed to have his opinion, obviously to anyone who reads more than one post on this blog I don't share it. I found it particularly dismaying3 however he let his opinion of Perl ruin his analysis, and lead to what he calls "the venomousness of some of the Perl commentators".
I can only assume that the venomousness he is talking about was removed or particularly bad in email, because the comments that stand now appear relatively rational and inline with the tone set in the body. I can't pretend that the Perl community isn't full of loudmouth assholes, I'm one of them, however if people find themselves feeling venomous, they should take it to their own blog. Better yet, take a deep breath and start that new project that is really cool and will change the world, and upload it to github so that the world beyond Perl can see it.
This is pure conjecture on my part, any opinion expressed here is entirely my own and has not been approved or vetted by Mr. Cortesi.
Aldo seems genuinely surprised at the strength of reaction he had, and in speaking with him his comments weren't intended to be as inflammatory as they came off. I've expressed that the Perl of the past isn't the Modern Enlightened Perl we all know and love (YOU DO LOVE IT RIGHT?!?), and tried to show how things really are in the Perl universe I see rather than the one he's obviously experienced in the past.
By: Chris Prather on 2009-12-15T18:40:00
Today it is the second most popular language on github. The increase in "popularity" is entirely based upon the
gitpanproject which uploaded the entire Backpan repository as git repos. However the 21,766 repositories in
gitpanare not followed by at least 3 people each, so thankfully wouldn't affect the outcome of the experiment. ↩
The conclusion that Cortesi makes is that Perl is that Perl is somehow much more difficult a language to contribute to after 300 days, but this is contrary to the other evidence he presents that perl projects are in general older with a larger median contributor size, and a larger core of contributors over time. ↩
I for one found the "offhand speculation" didn't even follow the data presented accurately enough to justify its existence without a qualifier (simply stating "my opinion is that ... "). A stronger distinction between the analysis of the data, and the sterotypical opinions he thought it backed up might have helped tone down the venom. ↩