Re: [bug fix] Savepoint-related statements terminates connection

On Thu, Sep 7, 2017 at 8:07 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> I've pushed up an attempt at this:
>
>
https://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=b976499480bdbab6d69a11e47991febe53865adc
>
> Feel free to suggest improvements.

Thank you, this helps a lot. Especially since some of the behavior is a bit
surprising, for example stopping on error leading to ROLLBACK not being
done and the retroactive upgrade of preceding commands in an implicit block
to a transaction block when a BEGIN appears.

When reading this I also realized that the backend does send responses for
every individual query in a multi-query request, it's only libpq's PQexec
that throws away the intermediate results and only provides access to the
last one. I always thought the backend did that. The docs hinted that it's
the frontend ("psql only prints the last one", "PGresult describes the
result of the last command") but to assure myself I looked with tcpdump.

It's a pity that the underlying protocol has 2 ways to do batching of
queries but the official library hides both. I guess I should go review the
"Batch/pipelining support for libpq" patch rather than complaining.