Re: 8.1 and syntax checking at create time

Michael Fuhr <mike@fuhr.org> writes:
> Are you using a database that was restored from an earlier version
> of PostgreSQL?  I wonder if you're not getting the lanvalidator
> function.

Ah-hah, that sounds like a good theory.  He'd have had to have carried
the DB forward from 7.4 or before, though, since plpgsql had a validator
in 8.0.

We've had repeated problems with PL languages stemming from the fact
that pg_dump dumps them at a pretty low semantic level.  Aside from this
problem with adding a validator, we used to have issues with hardwired
paths to the shared libraries in the CREATE FUNCTION commands.  And in
8.1, whether the functions are in "public" or "pg_catalog" is going to
vary across installations depending on whether the language was restored
from a dump or not.

I wonder if we could change the dump representation to abstract out the
knowledge encapsulated in "createlang".  I don't suppose this would
work:
\! createlang plpgsql <dbname>

but it'd be nice if the dump didn't know any more about the language
than its name, and didn't mention the implementation functions at all.
        regards, tom lane