James William Pye <pgsql@jwp.name> writes:
> In PL/Py, I had the bright idea of storing bytecode in the probin field of the
> function's pg_proc row. However, this idea has lately become rather dim as I
> have recently rediscovered(thanks Adrian) that this breaks dumps; pg_dump outputs
> a PL/Py function as "CREATE FUNCTION x() RETURNS y LANGUAGE python AS
> '<bytecode>', '<source>'". Of course, when loading this, it fails:
> 'ERROR: only one AS item needed for language "python"'.
> So is this "fix your broken PL" or "pg_dump should only be doing that for C
> language functions"?
Offhand it seems to me that pg_dump is behaving reasonably: it's storing
probin if it sees something there to be stored. The asymmetry is in the
backend, specifically functioncmds.c's interpret_AS_clause(): it has a
hardwired assumption that probin is only relevant to C functions.
Feel free to propose a saner definition. AFAICS the current coding
makes probin useless for all except C functions, so I think it could
be improved.
regards, tom lane
