I wrote:
>
> My Cygwin buildfarm member started failing (hanging, in fact)
> recently. It seems to hang consistently in join.sql and the only way I
> can get it to complete is to kill the backend fairly violently. See
> <http://www.pgbuildfarm.org/cgi-bin/show_log.pl?nm=brown_bat&dt=2009-07-22%2023:10:21>
> It looks like this is the result of the semi-join ordering bug fix Tom
> applied a few days ago, but it is building and running 8.4 quite
> happily, and I think that branch got the same changes, so I'm not
> quite sure about it.
>
> I'm going to try reversing that change locally to see if it fixes the
> problem. Unfortunately, this isn't an easy platform to debug.
>
>
Nope, it wasn't that. So my next candidate is the prior
join_is_legal/GEQO changes. But reverting that would get rid of the new
regression test where we seem to be failing, so I'm not quite sure where
to go on it.
cheers
andrew