pg_dump sorts its output first by object type, then by object name, and
then processes all that for dependencies. This works well, but for
overloaded functions this still gives a random sort order that can
produce annoying diffs in the dump.
Would it be acceptable to introduce a secondary sort key field into the
DumpableObject struct that functions would fill with, say, the argument
types (maybe something like "text,int,int" -- need to play with this a
little)?