(Sorry if you receive this twice, but I sent it Saturday evening and it
never made it to the list)
Thanks for your thorough review and comments, Tom.
Here's a new patch for review. Summary of changes/response to earlier
comments:
- add a routine for NullTest nodes -- done.
- declare selectivity functions without fmgr notation -- done.
- create selfuncs.h for declarations -- done, but I didn't move anything
else out of builtins.h
- use DatumGetBool() and adjust style -- done
- create better default selectivities -- done: - DEFAULT_UNK_SEL = 0.005 - DEFAULT_NOT_UNK_SEL = 1 -
DEFAULT_UNK_SEL - DEFAULT_BOOL_SEL = 0.5
- recurse clause_selectivity() for non-Var input -- done
- simplify MCV logic -- done, used 2nd approach (always use the first most
common val's frequency)
Questions:
- I added a debug define (BOOLTESTDEBUG) to selfuncs.h, and a corresponding
ifdef/elog NOTICE to clause_selectivity(). This was to help me debug/verify
the calculations. Should this be left in the code when I create a patch (it
is in this one), and if so, is there a preferred "standard" approach to this
type of debug code?
- Using the debug code mentioned above, I noted that clause_selectivity()
did not seem to get called at all for clauses like "where myfield = 0" or
"where myfield > 0". I haven't looked too closely at it yet, but I was
wondering if this is expected behavior?
Thanks,
-- Joe
