Tom Lane <tgl@sss.pgh.pa.us> writes:
> Gregory Stark <stark@enterprisedb.com> writes:
>> Incidentally the visibility bugs are indeed entirely fixed in 8.3. In 8.2 and
>> before cluster and alter table rewrites can both cause tuples to not appear
>> for transactions which were started before the cluster or alter table such as
>> a long-running pg_dump.
>
> AFAIK that's true only for CLUSTER, not ALTER TABLE. There would be a
> bunch of logical inconsistencies in altering rows and then pretending
> you hadn't.
Uh, what's true? That the bugs are fixed or that we ever had them to begin
with?
Oh, are you saying that ALTER TABLE doesn't go through the rewrite code to
carry along old versions of the tuples because it can't alter the old
versions? That is a bit annoying. That means there's no safe way to do a table
rewrite without clustering the table?
--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com
Get trained by Bruce Momjian - ask me about EnterpriseDB's PostgreSQL training!