Jeff Janes <jeff.janes@gmail.com> writes:
> I'm getting seg faults on contrib/bloom when updating a tuple which
> was found via a bloom index.
> It does not happen on every update, but it does happen within a few
> seconds of run time, so it is readily reproducible. The test harness
> is a bit of a mess, I'll try to clean it up and post it if no one
> spots the bug soon via looking at the stack trace below.
> Obviously scan->opaque is null, but I don't know why it is null and
> whether blendscan is obliged to deal with nulls, or if index_endscan
> is obliged not to send them.
I'm guessing there is a code path that can get from blbeginscan to
blendscan without going through blrescan. I'm not sure why it'd
only happen sometimes, though.
regards, tom lane