ldb_msg schema aware operations (was Re: [Release Planning 4.8] Samba 4.8.0)

[Douglas Bagnall]

hi Matthias,

> I had a look at your LDB patch which theoretically seems correct. What
> was the reason to have abandoned it?

I don't remember exactly, but something more urgent came up and I forgot
I'd even started (Andrew had to remind me).

> One observation: in each comparator you call:
>> +               const struct ldb_schema_attribute *a;
>> +               a = ldb_schema_attribute_by_name(ldb, el->name);
> Are we sure that the impact is negligible? In my patches for instance I
> passed the attribute around for not delaying too much.

Yeah, it seems to be looking the schema up more times than is
necessary, though (as far as I recall) I didn't have a test workload
where it showed up to at all. We have since done some work on
generating more realistic loads but I haven't looked at the data for
this.

I actually had a plan for making the schema lookups faster (using a
case-insensitive perfect or near-perfect hash), which also languished
on the same lack of evidence.

Douglas