PFC <lists@boutiquenumerique.com> writes:
>> Hmm; the only occurrence of that string in the sources is in
>> nodes/read.c, which AFAIK would never be invoked for a query entered
>> straight from the client. What could trigger it would be trying to
>> read a rule that is misformatted for some reason. Maybe you have
>> a damaged pg_user view definition --- does "select * from pg_user"
>> fail the same way?
> I installed pg 8.0.0 over 8.0.1-r3 (without touching the data directory)
> and then it worked again, so I don't think it has to do with the data
> being broken. I didn't do createdb or anything. I used gentoo emerge.
Well, that implies some file-format-level incompatibility between 8.0.0
and 8.0.1-r3, which would be quite disturbing ... especially if we had
more than one report of it. As is, I suspect it means you had a broken
build of 8.0.1-r3 for some reason. One possibility is a different set
of configuration parameters from what was used to build 8.0.0 (though
offhand I can't think of anything that would yield this particular
failure).
You might try comparing the results of
select ev_action from pg_rewrite where ev_class = 'pg_user'::regclass;
from the two builds.
regards, tom lane