Tatsuo Ishii <t-ishii@sra.co.jp> writes:
>> Yes, and no. I think the patch would be too risky to drop into 7.0.*
>> with no beta testing...
> I think we need beta testing then. Otherwise we would have to tell
> people to disable merge join until 7.1 gets released, I guess.
Well ... mumble ... as far as I've heard, you and Frank are the only
ones to get bit by it so far. So I'm not sure it deserves a fire-drill
response. There is at least one other "fatal" bug in 7.0 mergejoin
(planner will accept an inner plan of a type that doesn't support mark/
restore; this is still not fixed in current sources) not to mention a
fatal bug in hashjoin (hash table portal may get released before cursor
portal; this is fixed in current sources but I see no practical way to
fix it under 7.0 memory management). Is it worth trying to beta-test
our way to a 7.0.3 release? I'd vote for pushing up the 7.1 release
schedule instead, if we are sufficiently worried about these issues.
regards, tom lane