> This is an updated version of the PL instrumentation plugin patch that I
> submitted on July-28.  The new version re-implements the plugin loader
> code to use "rendezvous variables" as suggested by Tom Lane (thanks Tom,
> very elegant design).

Applied with a few small changes --- I renamed the GUC variables after a
suggestion by Simon Riggs, and fixed things so that
backend_load_libraries could actually do something useful (you had it as
PGC_POSTMASTER, making it effectively no more flexible than the existing
preload_libraries list).

The only change that will directly impact your code is that I thought
it'd be better to provide plpgsql_exec_error_callback and
exec_assign_expr as separate fields instead of arguments to func_setup,

        if (*plugin_ptr)
                (*plugin_ptr)->error_callback = plpgsql_exec_error_callback;
                (*plugin_ptr)->assign_expr = exec_assign_expr;

                if ((*plugin_ptr)->func_setup)
                        ((*plugin_ptr)->func_setup)(estate, func);

I'm not totally wedded to this if you don't like it, but my thought was
that passing these as arguments to func_setup would mean a lot of pain
anytime we wanted to change the set of function pointers provided:
every plugin would need textual changes whether it actually used these
functions or not.

> I have not implemented any support for unloading shared libraries.  Once
> we've finalized the design for rendezvous variables, I'll submit a
> separate documentation patch.

I added docs for the GUC variables but didn't do more than that.  I
think the code comments are probably sufficient as far as rendezvous
variables and PLpgSQL_plugin go ... did you have something else in mind?

                        regards, tom lane

---------------------------(end of broadcast)---------------------------
TIP 9: In versions below 8.0, the planner will ignore your desire to
       choose an index scan if your joining column's datatypes do not

Reply via email to