Re: Minimum selectivity estimate for LIKE 'prefix%'

Am Donnerstag, 6. März 2008 schrieb Tom Lane:
> I attach a proposed patch against HEAD for this.  It's a bit long
> but most of the bulk is refactoring eqsel() to make it easy to use from
> prefix_selectivity().  The intellectual content is just in the last
> diff hunk.

Looking at the patch as committed here
<http://anoncvs.postgresql.org/cvsweb.cgi/pgsql/src/backend/utils/adt/selfuncs.c.diff?r1=1.243;r2=1.244>,
I am confused by one hunk:

*************** prefix_selectivity(VariableStatData *var
*** 4452,4468 ****
* "x < greaterstr".
*-------
*/
- cmpopr = get_opfamily_member(opfamily, vartype, vartype,
- BTLessStrategyNumber);
- if (cmpopr == InvalidOid)
- elog(ERROR, "no < operator for opfamily %u", opfamily);
- fmgr_info(get_opcode(cmpopr), &opproc);
-
greaterstrcon = make_greater_string(prefixcon, &opproc);
if (greaterstrcon)
{
Selectivity topsel;

topsel = ineq_histogram_selectivity(vardata, &opproc, false,
greaterstrcon->constvalue,
greaterstrcon->consttype);
--- 4503,4519 ----
* "x < greaterstr".
*-------
*/
greaterstrcon = make_greater_string(prefixcon, &opproc);
if (greaterstrcon)
{
Selectivity topsel;

+ cmpopr = get_opfamily_member(opfamily, vartype, vartype,
+ BTLessStrategyNumber);
+ if (cmpopr == InvalidOid)
+ elog(ERROR, "no < operator for opfamily %u", opfamily);
+ fmgr_info(get_opcode(cmpopr), &opproc);
+
topsel = ineq_histogram_selectivity(vardata, &opproc, false,
greaterstrcon->constvalue,
greaterstrcon->consttype);

make_greater_string() is documented as

* The caller must provide the appropriate "less than" comparison function
* for testing the strings.

but using the new code flow it appears that opproc would still refer to
the >= operator looked up earlier.

The 8.1 code calls this variable ltproc because it didn't need the >= operator.
Perhaps using two variables would also be clearer here.