Re: HeapTuple header changes cause core dumps in CVS tip

On Mon, 15 Jul 2002 16:46:44 -0400, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
wrote:
>regression=# update foo set f1 = 'qq';
>server closed the connection unexpectedly

Same with DELETE FROM foo;

>I am not sure if this is a bug introduced by the patch, or if it's
>exposed a previously lurking bug.

I suspect the former :-(

>It seems that the HEAP_MOVED bits
>should be cleared before re-using cmax for something else, but I have
>not dug through the old logic to see how it was done before.

AFAICS from a quick look at tqual it didn't matter before. Once the
vacuum transaction had committed, on the next access to the tuple
MOVED_IN caused XMIN_COMMITTED to be set, and after that the MOVED
bits were never looked at again. MOVED_OFF is not an issue.

I'll take a closer look at tqual and the vacuum source code tomorrow.
For now the attached patch cures both symptoms (UPDATE and DELETE) and
passes all regression tests. A regression test for this case will
follow.

Servus
Manfred