performance issues with ldapsearch vs l.search(...)

[<bjo...@it...>]

For synchronization with some relational authorative datastore, I have
to search the ldap-server now and then (from cron). From the same
computer, to the same ldap-server, with no load (not production yet) I
get very different use of time.

ldapsearch (from openldap's ldapsearch.c):
$time ldapsearch -h ldap -x -b "ou=3Dusers,dc=3Dntnu,dc=3Dno" \
"objectClass=3DposixAccount" uid > /dev/null=20

# numResponses: 26876
# numEntries: 26875

real    0m14.560s
user    0m3.390s
sys     0m0.820s

python-ldap:
# start-script
import ldap
l =3D ldap.open("ldap")
l.simple_bind_s("","")
res =3D l.search_s("ou=3Dusers,dc=3Dexample,dc=3Dcom",ldap.SCOPE_SUBTREE,=
"objectClass=3DposixAccount",["uid"])
print len(res), ' records returned'
l.unbind()
# end script

$ time python ldap-test.py
26875  records returned

real    1m6.524s
user    0m59.580s
sys     0m1.640s

second run with python-ldap:
real    0m58.190s
user    0m52.390s
sys     0m1.620s

Why the big difference? Running truss and strace on the searches, I find
that ldapsearch fetch all entries in one retrieve, while python-ldap
fetch one entry by one.=20

ldapsearch is behaving the same way as pythonldap if I add e.g. '-S uid'=20
(sort by uid)

In python-ldap:
LDAPObject.c: line 684
in function: l_ldap_result( LDAPObject* self, PyObject *args )
    int all =3D 1;
        ^^^^^^^^
does this actually mean anything, and if so - what?

Any suggestions on how to fix LDAPObject.c to fetch the result all in
one chunk? Or am I looking in the wrong file or something?

--
Regards

Bj=F8rn Ove Gr=F8tan