> pg_dump already has rudimentary dependency tracking (one level
> deep); each
> item can have a list of oid's it depends on. You *could* patch it to add
> the types to the table dependencies.
>
> In the future I'd imagine we'll just dump the OIDs of all first level
> dependencies for each object, then at restore-time, process them in
> whatever order the user requests (defaulting to dependency-order).
Well, the problem is that you can add a new type and then add a column to a
really old table that uses that type - that causes pain. Lots of other
people have also reported the "view dumped before table it is based on"
problem.
Chris