On Wed, Feb 28, 2018 at 9:46 AM Nikolay Shaplov <dh...@nataraj.su> wrote: > 1. I've seen you've added a new attribute into pg_index. Why??!! > As far as I can get, if have index built on several columns (A1, A2, A3) you > can set, own opclass for each column. And set individual options for each > opclass if we are speaking about options. So I would expect to have these > options not in pg_index, but in pg_attribute. And we already have one there: > attoptions.I just do not get how you have come to per-index options. May be I > should read code more attentively...
It seems sensible to have both per-column options and per-index options. For example, we've got the fastupdate option for GIN, which is a property of the index as a whole, not any individual column. But you could also want to specify some column-specific options, which seems to be what this patch is about, since an opclass is associated with an individual column. And since an index can have more than one column, I agree that it seems more appropriate to store this information in pg_attribute than pg_index. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company