Coming to this thread a bit late as I've been out of email
connectivity for the past week...
On Tue, Jun 17, 2008 at 2:43 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> In any case, trying to define a module as a schema doesn't help at all
> to solve the hard problem, which is how to get this stuff to play nice
> with pg_dump. I think that the agreed-on solution was that pg_dump
> should emit some kind of "LOAD MODULE foo" command, and *not* dump any
> of the individual objects in the module. We can't have that if we try
> to equate modules with schemas instead of making them a new kind of
> object.
This is certainly the end result that I'd like, and intend to work
towards. My main concern has been cases where a module-owned table
gets updated with data that would not be recreated/updated by the LOAD
MODULE in the dump. PostGIS support tables are one example of this,
PL/Java classpath / function information is another. There are
probably many more.
I see two potential solutions:
a) explicitly mark such tables as requiring data to be dumped somehow,
and have pg_dump emit "upsert" statements for all rows in the table.
b) allow modules to define a function that can pg_dump can call to
emit appropriate extra restore commands, above whatever LOAD MODULE
foo does. This has the downside of requiring more work from module
owners (though perhaps a default function that effectively does option
a) could be provided), with a potential upside of allowing module
dumps to become upgrade-friendly by not being tied to a particular
version's table layout.
Thoughts?
Tom
