Re: SQL/MED - file_fdw

On Fri, 7 Jan 2011 10:57:17 +0900
Itagaki Takahiro <itagaki.takahiro@gmail.com> wrote:
> I updated the COPY FROM API patch.
> - GetCopyExecutorState() is removed because FDWs will use their own context.
>
> The patch just rearranges codes for COPY FROM to export those functions.
> It also modifies some of COPY TO codes internally for code readability.
> - BeginCopyFrom(rel, filename, attnamelist, options)
> - EndCopyFrom(cstate)
> - NextCopyFrom(cstate, OUT values, OUT nulls, OUT tupleOid)
> - CopyFromErrorCallback(arg)

For the purpose of file_fdw, additional ResetCopyFrom() would be
necessary. I'm planning to include such changes in file_fdw patch.
Please find attached partial patch for ResetCopyFrom(). Is there
anything else which should be done at reset?

> Some items to be considered:
> - BeginCopyFrom() could receive filename as an option instead of a separated
> argument. If do so, file_fdw would be more simple, but it's a change only for
> file_fdw. COPY commands in the core won't be improved at all.

> - NextCopyFrom() returns values/nulls arrays rather than a HeapTuple. I expect
> the caller store the result into tupletableslot with ExecStoreVirtualTuple().
> It is designed for performance, but if the caller always needs an materialized
> HeapTuple, HeapTuple is better for the result type.

IIUC, materizlizing is for tableoid system column. If we could add
tts_tableoid into TupleTableSlot, virtual tuple would be enough.  In
this design, caller can receive results with tts_values/tts_isnull
arrays.

Regards,
--
Shigeru Hanada