Zdenek Kotala wrote:
> 2) data type is unknown
>
> Unfortunately, in low function is no clue what data type is really
> stored in a chunks.
Do we need to know? We haven't changed the on-disk format of any data
types between 8.3 and 8.4, have we?
The other idea that was suggested earlier is to retoast all toast datums
referenced from the heap page, when the heap page is read in. We know
the data types at that point. That generates a lot more random I/O on
the first access to the unconverted heap page, but I think it would be ok.
--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com