On Fri, Jun 14, 2024 at 11:40 AM Andres Freund <andres@anarazel.de> wrote:
Hi,
On June 14, 2024 8:09:43 AM PDT, Andrew Dunstan <andrew@dunslane.net> wrote: >Over at [1] Andres expressed enthusiasm for enabling TAP tests to call LibPQ directly via FFI, and there was some support from others as well. Attached is a very rough POC for just that.There are two perl modules, one which wraps libpq (or almost all of it) in perl, and another which uses that module to create a session object that can be used to run SQL. Also in the patch is a modification of one TAP test (arbitrarily chosen as src/bin/pg_amcheck/t/004_verify_heapam.p) to use the new interface, so it doesn't use psql at all. > >There's a bunch of work to do here, but for a morning's work it's not too bad :-) Luckily I had most of the first file already to hand.
Yay!
>Next I plan to look at some of the recovery tests and other uses of background_psql, which might be more challenging,a dn require extension of the session API. Also there's a lot of error checking and documentation that need to be added.
I'd suggest trying to convert the various looping constructs first, they're responsible for a large number of spawned shells. And I vaguely recall that there were none/very few that depend on actually being run via psql.
Yeah, here's a new version with a few more scripts modified, and also poll_query_until() adjusted. That seems to be the biggest looping construct.
The biggest remaining unadjusted script users of psql are all in the subscription and recovery tests.
