Rebased against head Here's my summary of the long thread above.
This change is in keeping with the SQL spec. There is an argument (Tom) that says that this will annoy more people than it will please. I presume this is due to the fact that libpq behaviour will change. As the author of the JDBC driver, and I believe I speak for other driver (NPGSQL for one) authors as well that have implemented the protocol I would argue that the current behaviour is more annoying. We currently have to keep state and determine if COMMIT actually failed or it ROLLED BACK. There are a number of async drivers that would also benefit from not having to keep state in the session. Regards, Dave Cramer www.postgres.rocks On Tue, 10 Nov 2020 at 11:53, Dave Cramer <davecramer@postgres.rocks> wrote: > > > On Mon, 9 Nov 2020 at 16:26, Dave Cramer <davecramer@postgres.rocks> > wrote: > >> >> >> On Wed, 30 Sep 2020 at 18:14, Andrew Dunstan < >> andrew.duns...@2ndquadrant.com> wrote: >> >>> >>> On 8/4/20 12:19 PM, Dave Cramer wrote: >>> > Attached is the rebased patch for consideration. >>> > >>> > >>> >>> >>> It's a bit sad this has been hanging around so long without attention. >>> >>> >>> The previous discussion seems to give the patch a clean bill of health >>> for most/all of the native drivers. Are there any implications for libpq >>> based drivers such as DBD::Pg and psycopg2? How about for ecpg? >>> >>> >>> cheers >>> >>> >>> andrew >>> >>> >>> -- >>> Andrew Dunstan https://www.2ndQuadrant.com >>> PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services >>> >> >> >> Attached is a rebased patch with fixes for the isolation tests >> >> > >> >> Dave Cramer >> www.postgres.rocks >> >
0001-Throw-error-and-rollback-on-a-failed-transaction-ins.patch
Description: Binary data
