On Wed, Jan 5, 2011 at 19:24, Shigeru HANADA <hanada@metrosystems.co.jp> wrote:
> 2) foreign_scan.patch includes ForeignScan executor node and
> FdwRoutine interface
I can see now Iterate() callback is called in per-tuple memory context.
I'll adjust copy from API for the change. We don't need to export the
executor state in CopyState.
ForeignNext() still needs to materialize the slot. It seems reasonable
for me to add tts_tableoid to TupleTableSlot and modify slot_getattr()
to return the field for virtual and minimal tuples. Am I missing any
problems here?
Even if we still materialize tuples in 9.1, we would be better to use
ExecStoreVirtualTuple() in file_fdw and pgsql_fdw for future optimization.
--
Itagaki Takahiro