On Tue, May 20, 2014 at 11:18:29AM -0700, Jeff Davis wrote:
> I think this can be done without breaking upgrade compatibility, because
> I think the structure already satisfies the invariants I mentioned in
> the other email (aside from the special case of a root tuple with two
> nodes and no prefix). That being said, it's a little scary to refactor
> indexing code while trying to keep it upgrade-compatible.
We can make pg_upgrade mark such indexes as invalid and create a user
script to reindex all the indexes after the upgrade. We have done
that in the past.
-- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB
http://enterprisedb.com
+ Everyone has their own god. +
