Merlin Moncure <mmonc...@gmail.com> writes:
> On Tue, Jun 26, 2012 at 3:05 AM, Magnus Hagander <mag...@hagander.net> wrote:
>> But with a small change to psql they could, without the need for a
>> whole new type of variable. For example, psql could set all those
>> variable as "psql.<commandlinevarname>", which could then be accessed
>> from the DO PL code just fine.
> That's a really neat idea.
I do see a problem with this client-push idea, which is what happens if
psql sends a SET and later the active transaction gets rolled back.
psql does not have enough knowledge to be sure whether it lost the SET
or not. It could hack things by always resending all variables after
any rollback, but ugh.
We could address that by inventing a non-transactional variant of SET,
perhaps. Not sure it's worth the complication though --- I don't think
I want to have to define how that would interact with other variants
of SET in the same transaction ...
Another approach would be to define such variables as being truly
shared, in the spirit of last-update-wins multi master replication.
The backend sends over its values using the existing GUC_REPORT
mechanism. So a rollback would cause the psql-side variable to revert
as well. Not actually sure if that behavior would be more or less
useful than a simpler definition, but it's worth thinking about.
In this connection, there was some recent discussion in the jdbc list
of wanting clients to be able to set the GUC_REPORT flag on any GUC
variable, because the jdbc driver would like to track some settings we
have not seen fit to mark that way. Not sure if anybody mentioned that
on -hackers yet, but it's coming. If we had that ability then a
shared-variable behavior like this could be built entirely on the psql
side: the push part is just SET, and the pull part is GUC_REPORT.
regards, tom lane
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
