Re: [Statcvs-users] Paper on performance indicators in CVS repositories

[James R. <sj...@jd...>]

Hi Richard,

> Hi all,
>
> I came across an interesting paper on using information from CVS
> repositories to predict student performance. Summary: they found no
> good performance indicators. The best one was plain ol' lines of code.

In the case of my final year university group project, the lecturers are
explicitly basing some of our assessment on the number of lines contributed
to CVS. I think that this is a sorry situation, given how poorly a simple
line count can represent somebody's input to a software engineering project.

> Link and some commentary on my weblog:
> http://dowhatimean.net/2005/10/cvs-and-performance

You mention that the lack of correlation between code metrics and access
patterns does not bode well for StatCVS. I don't think that it's
necessarily a bad thing... Finding correlations between student grades
and CVS activity is only one possible application
for a good tool such as StatCVS.

For example, in an industry project, the project manager may use a whole
bunch of statistics from various sources (including, but not limited to,
StatCVS) to evaluate the health and progress of their project.

I guess what I'm trying to say is that by not providing more metrics, we'll
never see if any interesting correlations come up in unexpected places.

I'd certainly like to see some sort of Java code metrics as an "add-on"
for StatCVS, like the ability to measure (for example) method fan-in
and fan-out in a Java project over time.

The more stats, the merrier!

> Best,
> Richard
>

Thanks,
James
-- 
James Ring