The Hermit Hacker <scrappy@hub.org> writes:
> On Thu, 7 Sep 2000, Tom Lane wrote:
>> This is the planner bug that I was just alluding to in other email ---
>> the planner is trying to use a nestloop as the inner input to a
>> mergejoin, and that doesn't work :-(. But you only see the problem if
>> the outer side contains multiple matches to a single inside tuple.
>>
>> I have a fix for current sources; let me see if I can retrofit it for
>> 7.0.*.
> that would be perfect ... if we can get that retrofit'd, I'd be quite
> tempted to put out a 7.0.3 for this, considering that its obviously not an
> isolated incident ;(
I have committed a fix into REL7_0 branch. Although it seems to work,
I don't trust it really far because it depends on heap_markpos() and
heap_restrpos(), which haven't been used in a long time and are full
of alarmed-sounding comments. (The equivalent fix in current sources
does not use these routines, but that's because nodeMaterial.c has been
completely rewritten, so back-patching that code doesn't seem like a
risk-free choice either.)
I'd suggest running the REL7_0 sources on your machine for awhile before
deciding it's safe to call it 7.0.3.
regards, tom lane