Tom Lane wrote:
> Andrew Dunstan <andrew@dunslane.net> writes:
>
>> The attached patch sets the OID to InvalidOid for the duration of the
>> memcmp if the HEAP_HASOID flag is set, and restores it afterwards.
>>
>
> This method is utterly, utterly unacceptable; you're probably trashing
> the contents of a disk buffer there. Even assuming that there's zero
> risk of a failure between the set and the restore, what if someone is in
> process of writing the buffer to disk? Or even just examining the old
> tuple?
>
>
>
OK, I guess I assumed we had a private copy.
Next thought is to split the memcmp() into two to avoid the Oid field,
where it's present. I was hoping to avoid the kinda ugly arithmetic.
cheers
andrew