I wrote:
> Well, if there are no other comments, I'll push forward with the fix
> proposed here:
> http://archives.postgresql.org/pgsql-hackers/2009-11/msg00531.php
Done. I did not make the change I speculated about of allowing
completely unknown variables (those that don't even match
custom_variable_classes) to be set by superusers. It would be a very
minor tweak to the committed code to allow that, but I'm not convinced
that making a corner case in dump/restore slightly easier is worth the
loss of error checking. In practice, if you have ALTER ... SETs for
custom variables, you'd better list their modules in
custom_variable_classes, or it won't work nicely. I see no really
strong reason not to fix that parameter before you restore instead of
after.
regards, tom lane