[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: commit: ldap/servers/slapd config.c proto-slap.h schema_init.c slap.h

Howard Chu writes:
> Your concern makes no sense.

Perhaps I misunderstood the suggestion.  I thought you were suggesting
to turn a (mail=*foo*) into (mail=*) - either if there is no substring
index for it or if it returned ALLIDS; I'm not sure.
I'm talking about the situation where you have a presence index but not
a substring index: The unchecked limit would make (mail=*foo*) fail at
once, while (mail=*) might narrow it down enough that the server would
then trawl through a lot of entries - often only to fail to find

If that's not it, please ignore the following:

> You've taken measures to compensate for a
> particular type of search being too expensive in server time,

The user's time is more important, but yes.

> and now that the possibility exists to make those searches more
> efficient you perceive this as a problem?

Failing at once is more efficient than to search for a while and then
fail anyway.

> You already said that your unchecked limit is uncomfortably high;
> using this proposed fallback would allow you to set a much lower
> unchecked limit.

That would be very nice, but I don't see why it would help.

>>> Of course, in practice, I never use 
>>> presence indices. I wonder how many people do.
>> Probably quite a lot, since I think some OpenLDAP documentation says or
>> said that if one has an equality index (I think), one might just as well
>> add a presence index, since that takes very little extra space.  I
>> believe that's why we had presence indices.
> Well, if there is any documentation that says this, it is wrong. The 
> presence and equality indices are about equal in size.

I'll try to find it again.  Hopefully we just misunderstood.