Buildfarm code rewrite - status update

[Maciej (Matchek) Bliziński]

If you don't care about buildfarm code, you can stop reading now.

Since it's been slowly going on for 10 months, I'll give you a quick
update of what's going on with the rewrite.

Problems: Buildfarm tools can't do what they should, e.g. display
package metadata via www; impossible to set up a buildfarm using the
code from Subversion
Cause: Too much data in a single JSON blob; the installation procedure
not tested
Objective: Rewrite buildfarm code to split the JSON blob into smaller
blobs; test the deployment procedure

Glossary:

* "the old code":
  What is currently in Subversion.
  The buildfarm runs it at the moment.
  http://gar.svn.sourceforge.net/svnroot/gar/csw/mgar
* "the new code":
  The code being written, available on github:
  https://github.com/opencsw/gar/tree/blob-split

As of October 2013, the new code is working enough to set up a local
build host and check packages. It's reflected in the buildfarm setup
instructions[1].

I haven't tested the rest of the package life cycle with uploading and
catalog generation. I'll need to do that before I release the new code
and make the buildfarm run it.

I'm currently in a little bit of yak shaving: The RESTful web apps are
an increasingly important piece of the infrastructure, so they need to
be tested. During testing, we need to set up and tear down a test
database for each test, so we're sure that tests don't influence each
other. The problem is that our ORM, SQLObject, is always returning a
connection to the same in-memory SQLite database. So I'm currently
trying to figure out[2] why isn't SQLObject freeing the memory when
closing an in-memory SQLite database.

Maciej

[1] http://www.opencsw.org/manual/for-maintainers/buildfarm-setup.html
[2] https://sourceforge.net/mailarchive/forum.php?thread_name=20131019161004.GA2227%40iskra.aviel.ru&forum_name=sqlobject-discuss