Stephen Frost <sfrost@snowman.net> writes:
> Patch attached. Passes the regression tests and our internal testing
> shows that it eliminates the error we were seeing (and doesn't cause
> blocking, which is even better).
> We have a workaround in place for our build system (more-or-less
> "don't do that" approach), but it'd really be great if this was
> back-patched and in the next round of point releases. Glancing
> through the branches, looks like it should apply pretty cleanly.
It looks like it will work back to 8.4; before that we didn't have
RegisterSnapshot. The patch could be adjusted for 8.3 if anyone
is sufficiently excited about it, but personally I'm not.
regards, tom lane