Greetings, Greg,
Dr. Johnson:
I thought you might be interested in hearing about my experience with Hackystat
on a
commercial project. I have started introducing Hackystat to my team in a very
limited
fashion. So far, we have been using Hackystat to improve our usage of our Ant
build
system. For example, we have been able to figure out how much time team members
spend
performing builds. This has allowed us to put together some economic arguments
for
improving the build process.
Very cool! Cedric Zhang in our lab is doing a lot of work with build analyses right now
as part of his Ph.D. research. Cedric, once you get your analyses online, please send
Greg an email with a quick summary of what you're able to do. Perhaps it will be helpful
and/or interesting.
I would like to share a couple of hurdles we have faced so far. I hope you can
address
these in a future version of Hackystat:
First, we noticed that you packaged several third-party components into the
sensorshell.jar file. Unfortunately, the third-party components did not work
well with
our build script, because sensorshell's versions conflicted with our own. We
got around
this problem by decomposing the sensorshell.jar into its constituent parts and
using
our own versions where appropriate. Now that you have created a Hackystat
Installer,
are there still benefits of packaging up third-party components into the
sensorshell.jar? If not, then I hope you can reconsider the organization of
senshorshell.jar.
Good idea, and as you say, now that we have hackyInstaller it is much more feasible. I've
posted this issue at:
<http://hackydev.ics.hawaii.edu:8080/browse/HACK-371>
The scheduled release is 7.1, i.e. the first stable release after our major architectural
reorganization. We're currently postponing smaller development tasks to make way for this
major piece of work, which will vastly improve the lives of hackystat developers and
those that depend upon them. :-)
Second, we noticed that the current Hackystat Build Sensor recognizes Java
compilation
warnings as build failures. Thus, even though Ant may recognize the build as
successsful, Hackystat will still record it as a failure. Interpreting the
build data
has been quite difficult. We have modified the source code of our Build Sensor
to
ignore any warnings. I can send you the modified code if you agree that this is
a bug.
The ability to 'filter' the types of data sent by the build sensor on the client side as
well as the types of data considered by build analyses on the server side has been
requested by multiple users. I've spoke with Cedric and he has volunteered to work on it:
<http://hackydev.ics.hawaii.edu:8080/browse/HACK-372>
That's about it for now. I have so many ideas on applying this tool to our
team, and of
course have a long wishlist of features I'd like to see. If you are interested
in any
further feedback, I'd be glad to be of assistance.
That would be great. We are just entering a period of substantial reorganization of the
system to support external users and developers, so this is a perfect time to submit your
wishlist. :-)
Cheers,
Philip
