Re: ANALYZE patch for review

"Mark Cave-Ayland" <m(dot)cave-ayland(at)webbased(dot)co(dot)uk> writes:
> Yup indeed. Please find enclosed the latest version of the analyze patch
> taking into account all the things we have discussed in the thread.

I've reviewed and applied this with some small changes. You did a good
job --- the only things you missed AFAICS were pg_dump support and
documentation.

I changed the API slightly for the typanalyze function and the
compute_stats subroutine. I don't think it's necessary or appropriate
to pass in the Relation pointer to typanalyze; that simplifies its
signature to a single INTERNAL parameter. (If we did want to pass the
Relation, I'd be inclined to make it a field in VacAttrStats.) Also
I tweaked compute_stats so that the attribute number of the column is
passed explicitly. I foresee needing this for functional-index stats
gathering --- it's likely that the accumulated sample rows will be built
on the fly and will have a tupdesc that includes both the main table
rows and the functional index columns. So compute_stats shouldn't
assume that the nominal column number is the right thing to use to
extract the column it wants.

> I also altered the examine_attribute() to allow the user-defined
> function to check attstattarget for the column manually (in case they
> want to do something special with minus or zero values),

As I've committed it, the convention that zero means "no stats" is
enforced by examine_attribute(), but the typanalyze function is
responsible for deciding what a negative value means. Seem reasonable?

Also, I put the following documentation about "kind" values into
pg_statistic.h.

/*
* The present allocation of "kind" codes is:
*
* 1-99: reserved for assignment by the core PostgreSQL project
* (values in this range will be documented in this file)
* 100-199: reserved for assignment by the PostGIS project
* (values to be documented in PostGIS documentation)
* 200-9999: reserved for future public assignments
*
* For private use you may choose a "kind" code at random in the range
* 10000-30000. However, for code that is to be widely disseminated it is
* better to obtain a publicly defined "kind" code by request from the
* PostgreSQL Global Development Group.
*/

regards, tom lane