* HARD (lots of boring work): provide unit tests

* MEDIUM: provide options in MANIFEST.cfg to limit buildouts to some
  PG versions, to ease the load due to useless cases (known
  incompatibilites don't have to be tried over and over; some PG
  versions may be of interest in a few cases only). We need a global
  setting, and per-buildout settings.

* MEDIUM: harmonize psql calls: stop using Debian-specific PGCLUSTER,
  allow setting of psql executable (has the correct default for
  unix-socket domains). Use PGHOST environment var with empty value to
  solve the problem of uniformizing defaults for Unix-socket domains,
  instead of passing a host.

* MEDIUM: have the slaves retrieve ``build_utils`` directly from VCS. This
  will probably complicate the scheduling setup.

* EASY: use (also) test log analyzer and test launching commands from
  openerp-runbot.

* MEDIUM: add a build property for human readability of branches being
  tested. Ideally, do that on the Change object if that works.

* HARD: change conf (slaves.cfg) so that one
  slave per host is enough. Will spare lots of disk and bw thanks to sharing of
  branches, eggs and downloads at slave level, and work for slave
  admins, too. This is tricky, since we can't use the pg_port
  etc. properties directly.

* HARD: true reproduceability and better blame list through setting of
  revs in build properties together with new options in buildout
  recipe to apply them. NB: it seems that property values can be dicts
