Re: posgres 12 bug (partitioned table)
Amit Langote <amitlangote09@gmail.com>
From: Amit Langote <amitlangote09@gmail.com>
To: Pavel Biryukov <79166341370@yandex.ru>
Cc: Andres Freund <andres@anarazel.de>, Tom Lane <tgl@sss.pgh.pa.us>, PostgreSQL mailing lists <pgsql-bugs@lists.postgresql.org>
Date: 2020-10-28T14:01:16Z
Lists: pgsql-bugs, pgsql-hackers
Attachments
- parted-insert-returning_syscols_HEAD_v2.patch (application/octet-stream) patch v2
- parted-insert-returning_syscols_BACKPATCH_v2.patch (application/octet-stream) patch v2
On Tue, Oct 27, 2020 at 10:55 PM Pavel Biryukov <79166341370@yandex.ru> wrote: > > Hi! > > What's the state with this issue? I think that the patch that Andres Freund posted earlier on this thread [1] would be fine as a workaround at least for stable releases v12 and v13. I have attached with this email a rebased version of that patch, although I also made a few changes. The idea of the patch is to allocate and use a partition-specific *non-virtual* slot, one that is capable of providing system columns when the RETURNING projection needs them. Andres' patch would allocate such a slot even if RETURNING contained no system columns, whereas I changed the slot creation code stanza to also check that RETURNING indeed contains system columns. I've attached 2 patch files: one for HEAD and another for v12 and v13 branches. That said, the discussion on what to do going forward to *cleanly* support accessing system columns through partitioned tables is pending, but maybe the "workaround" fix will be enough in the meantime (at least v12 and v13 can only get a workaround fix). -- Amit Langote EDB: http://www.enterprisedb.com [1] https://www.postgresql.org/message-id/20200811180629.zx57llliqcmcgfyr%40alap3.anarazel.de
Commits
-
Don't crash on reference to an un-available system column.
- d479d0028525 14.0 landed
- 2602ee4689c7 13.3 landed
- 05ce4bf8b1d4 12.7 landed