Re: Extend injection_points_attach() to accept a user-defined function

On Tue, Nov 04, 2025 at 02:17:44PM +0530, Rahila Syed wrote:
> PFA a rebased patch that contains the test.
> The tests use the newly added SQL function to attach the injection_notice
> function to an injection point

+ if (injection_point_local)
+ {
+ condition.type = INJ_CONDITION_PID;
+ condition.pid = MyProcPid;
+ }

Hmm. Is there a point in registering a condition that's linked to
the shared library injection_points? The automatic drop is kind of
nice to have, I guess, but it gives the illusion that an attached
callback will not be run. However, a callback from an entirely
different library *will* run anyway because it cannot look at the
condition registered, or the other library has an equivalent able to
treat a local condition with the same format and same structure as
what's in injection_points.c.

Different idea: we could allow one to pass a bytea that could be given
directly to InjectionPointAttach()? Without a use-case, I cannot get
much excited about that yet, but if someone has a use for it, why not.

Maybe we should also discard the pgstat_create_inj() call in this
path? The existing injection_points_detach() would be able to deal
with a point attached with a callback from a different library anyway.
Keeping pgstat_report_inj_fixed() feels OK in this new attach path.
--
Michael