Truly bizarre behavior with VACUUM FULL?

Folks,

I have a system on a 7.2.4 database.  Let me tell you what appears to be
happening:

I have a bug-fixed version of a function, which replaced a buggy version of
the function via CREATE OR REPLACE.  Each night, when the database goes
through VACUUM FULL, the old (buggy) version of the function seems to
overwrite the new (correct) version of the function.

Is it possible that VACUUM FULL is restoring the discarded tuple in the system
table and dropping the valid one?

I can think of ways to fix this bizarre behavior, but thought that someone
might want to look at it first.

--
-Josh BerkusAglio Database SolutionsSan Francisco