launchpad-dev team mailing list archive
-
launchpad-dev team
-
Mailing list archive
-
Message #05080
Re: Stop the line: Test failures in *stable*
FWIW, a few thoughts.
I'm going to post separately on how we manage change in the LP
project, which is closely related to this particular thread, but
covers more area.
In this specific thread:
- our test suite has many isolation and order-sensitive issues.
Things that will help this:
- less doctests
- clearer 'layers'
- parallel testing (the pain we encounter will go directly to test
suite reliability
- the migration from hardy to lucid allowed a great deal of skew to
arise, it would be nice to be a bit tighter on that in future, and
ideally to decouple the platform, python, & postgresql version
changes. I've mentioned the issue we had in the UDS planning threads
w.r.t. Python, which was positively received - we may have have an
easier time going to python 2.7 in 18 months.
- Going back to precommit testing will be positive I think; we should
be able to get landings down to 20 minutes or so sometime in 2011.
- Test suite failures *will* occur whenever we change the platform,
the stack or test infrastructure. We should deal and move on - its a
normal thing : as long as we quickly rectify things, there's no need
to panic. (Much the same as 100% test coverage still not preventing
all bugs). Our ability to react and fix is the primary tool we have
when something does go wrong.
- reintroducing the ORDER BY random() for queries without an ORDER BY
clause would be wonderful. We put that in for sqlobject and it was
darn useful.
-Rob
References