I think we should look into handling the different page types better.
The hash_page_stats function was copied from btree, which only has one
type. It's not clear whether all the values apply to each page type.
At least they should be null if they don't apply. BRIN has a separate
function for each page type, which might make more sense. I suggest
taken the test suite that I posted and expanding the tests so that we
see output for each different page type.
Besides that, I would still like better data types for some of the
output columns, as previously discussed. In addition to what I already
pointed out, the ctid column can be of type ctid instead of text.
Since the commit fest is drawing to a close, I'll set this patch as
returned with feedback. Please continue working on it, since there is
clearly renewed interest in hash indexes, and we'll need this type of
functionality for that.
--
Peter Eisentraut http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services