Claudio Natoli <claudio(dot)natoli(at)memetrics(dot)com> writes:
> As an attempt at quantification, I'll say this. We are currently using a
> test file with about 5000 pairs of create/drop. It (almost?) never makes it
> through a pass of this file. Your mileage may vary.
FWIW, I ran 10000 create/drop pairs through 7.3.2 (well, actually REL7_3
branch tip, but I'll be *real* surprised if it differs on this point)
on HPUX 10.20. No sign of a problem.
You did say you tried already on Linux, right? Which version exactly?
Can someone try the experiment on some *BSD platform? Three data points
would make me reasonably comfortable that it doesn't fail on Unixen, and
then I'll say Cygwin gets the blame...
regards, tom lane