Bazaar back into Pybots?

[Martin Pool]

On 5/8/07, Alexander Belchenko <bialix at ukr.net> wrote:
> I time-to-time run selftest on my machine, and I can say, that selftest
> *never* run without fails or errors on Windows. Even when John fix (almost)
> all win32-related issues before 0.8 release, it degrades very quickly.
>
> Last time (before 0.16 release) there was less than 20 errors on Python 2.5
> with all dependencies installed (medusa, paramiko, pycurl, diffutils,
> gettext). Most of the failing tests was because of locking problem.
> And this test is completely ugly:
> test_lockdir.TestLockDir.test_35_wait_lock_changing -- it fails only
> when running in the middle of test suite, but *never* when running
> alone or with --first flag. Probably it should say something important?

OK, well automated running is probably only going to help us stay
clean once we are clean.  How can we get there?  Just pay good
attention to Alexander, or have other people run the tests
interactively themselves?  If so, maybe automated running is not a
priority yet?

> Also side effect of regular running bzr selftest: it read/write very
> big amount of data (files, directories) on HDD, as result HDD degrades
> very quickly. Usual disks, especially in laptops, is really not designed
> for instant loading. Recently I read about machines that used for
> regular testing of pypy project: they have broken and exchanged during
> warranty period. If we can reduce our disk I/O load and do most of testing
> in memory itself -- it will be good, IMO.

To start with we should finish off Andrew's test performance branch,
which had some review comments.  Andrew's away this week, but i might
look at it.  That should greatly reduce the amount of physical IO at
least on linux, because the files are deleted so quickly they're not
written to disk.

-- 
Martin