Curt Sampson <cjs(at)cynic(dot)net> writes:
> I'm not actually sure why there would be a problem (aside from
> performance) in removing and rebuilding an index on a system table,
All the hard-coded low-level stuff that expects the index to be there
blows up.
This might possibly be made to work for indexes on inessential catalogs
like pg_statistic, but I can't imagine it working for pg_attribute...
regards, tom lane