Tom Lane writes:
> The view also tries to be slightly more user-friendly than a raw look at
> pg_statistic. It joins to pg_class and pg_attribute to get the names of
> tables and columns, and it presents the various kinds of statistics
> intelligently labeled. (Thus, the view has wired into it exactly the
> sort of assumptions we agreed not to make in pg_statistic itself.
> Peter E. may be unhappy with me here. But it seems to me that we can
> change/extend the view definition when we add new kinds of statistics
> entries.)
For informational views this is okay AFAICS; the SQL information schema is
full of this kind of thing. When you want to present alternative
information you add a new column and return null in the old ones.
--
Peter Eisentraut peter_e@gmx.net http://funkturm.homeip.net/~peter
