> I'd expect that to work OK, but with one pretty big "foot gun" in
> play, namely the fact that the XML type does validation.  That
> provides the risk that if you ever put invalid XML into the column in
> the 8.1 database, the attempt to INSERT/UPDATE it on the 8.3 instance
> will fail, and replication will fail (until/unless you delete the
> offending tuples from sl_log_[12] by hand).  That would be a nice way
> to really muss up the cluster.

That's great! I know every XML document is well-formed, and I have loaded
all the data into an xml type column using pg_restore without errors. So
I'll move forward with validating this approach (replicating using Slony).

-- m@