Thread: Traversing the catalog using heap_open, systable_beginscan, ...
Hello,
I am writing a foreign data wrapper and following the oracle example. Are there docs that discuss heap_open, systable_beginscan and systable_getnext?
I currently just need to do simple stuff, like translate an Oid that is received via a FDW callback into the text name of a table. I see all of the catalog headers, pg_class.h, etc... but not seeing the pattern of which args to pass into ScanKeyInit to traverse the catalog correctly.
Hints greatly appreciated.
If there is a better list, please point me to it.
Thanks,
Mark
Mark Crosland <mark.crosland@gmail.com> writes:
> Hello,
> I am writing a foreign data wrapper and following the oracle example. Are
> there docs that discuss heap_open, systable_beginscan and systable_getnext?
> I currently just need to do simple stuff, like translate an Oid that is
> received via a FDW callback into the text name of a table. I see all of the
> catalog headers, pg_class.h, etc... but not seeing the pattern of which
> args to pass into ScanKeyInit to traverse the catalog correctly.
Well, you *can* look that sort of stuff up that way, but usually there's
a better way; in particular, if at all possible you should consult the
catalog caches instead. Look into src/backend/utils/cache/lsyscache.c
for convenience routines that do this sort of thing. The particular need
you cite above is addressed by get_rel_name(), and even if you don't see
an exact fit, the routines provide good examples for fetching fields out
of catalog entries.
Also, once you've got a Relation pointer, much of what you might want
to know about that table is available directly out of the relcache
entry.
regards, tom lane
Thanks, that helps. I can get table and column info from the catalog now.
What is the recommended way (structs to access via one of the scan/modify callbacks?) to retrieve the column, operation and value, "col", "=" and "val" in this example. And is the same method used for both queries and updates?
select * from table where col=val
A lot of the FDW examples seem to use table options, which isn't quite what we need. We are going for more of a mapping between tables in different DBs.
Thanks,
On Sat, Nov 9, 2013 at 8:12 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
Mark Crosland <mark.crosland@gmail.com> writes:
> Hello,
> I am writing a foreign data wrapper and following the oracle example. Are
> there docs that discuss heap_open, systable_beginscan and systable_getnext?
> I currently just need to do simple stuff, like translate an Oid that is
> received via a FDW callback into the text name of a table. I see all of the
> catalog headers, pg_class.h, etc... but not seeing the pattern of which
> args to pass into ScanKeyInit to traverse the catalog correctly.
Well, you *can* look that sort of stuff up that way, but usually there's
a better way; in particular, if at all possible you should consult the
catalog caches instead. Look into src/backend/utils/cache/lsyscache.c
for convenience routines that do this sort of thing. The particular need
you cite above is addressed by get_rel_name(), and even if you don't see
an exact fit, the routines provide good examples for fetching fields out
of catalog entries.
Also, once you've got a Relation pointer, much of what you might want
to know about that table is available directly out of the relcache
entry.
regards, tom lane
Mark Crosland <mark.crosland@gmail.com> writes:
> Thanks, that helps. I can get table and column info from the catalog now.
> What is the recommended way (structs to access via one of the scan/modify
> callbacks?) to retrieve the column, operation and value, "col", "=" and
> "val" in this example. And is the same method used for both queries and
> updates?
> select * from table where col=val
contrib/postgres_fdw might be a useful example for you. A lot of this
stuff isn't documented terribly well, you just have to look for some
example code that does what you want to do.
regards, tom lane