Any ideas how to best continue with this? As a workaround I've tried execution the unit tests in a debian-docker container with

(venv) auth@803d93942a80:~/src$ slapd -V
@(#) $OpenLDAP: slapd  (Feb 14 2021 18:32:34) $
	Debian OpenLDAP Maintainers <pkg-openldap-devel@lists.alioth.debian.org>
(venv) auth@803d93942a80:~/src$ apt info slapd
Package: slapd
Version: 2.4.47+dfsg-3+deb10u6

This seems to work - insofar as I'm now able to start the slap daemon.

I would like to run the test suite natively too if possible as it helps me bring additional debugging-tools to play that are harder to use inside the docker container. Any pointers?

Regarding the first error:

611d131f /path/to/testdir/slapd.conf: line 12: <database> failed init (mdb)

Build of OpenLDAP that comes with macOS doesn't have mdb backend (you can verify it with running /usr/libexec/slapd -VVV).

Looking at your last error:

611d1725.21951778 0x104d79e00 config_setup_ldif: expected directory /path/to/testdir/python-ldap-test-60577 to be empty!

It's a result of the fix for ITS#9016
This is how slapd was started in python-ldap-3.3.1:

Since testrundir is not empty and has files, slapd complains and fails to start.

In #382 the configuration was changed from using slapd.conf to use slapd.d. Unfortunately, it doesn't work with the version of OpenLDAP that comes from homebrew, because it is built without modules support, and the test fails with:

E           RuntimeError: ['/usr/local/opt/openldap/sbin/slapadd', '-F', '/Users/user/src/test/python-ldap-test-51436/slapd.d', '-n0'] process failed:
E           b''
E           b'slapadd: dn="cn=module,cn=config" (line=14): (65) unrecognized objectClass \'olcModuleList\'\n'

So, you can either patch python-ldap-3.3.1 that comes from pypi with something like this:

diff --git a/Lib/slapdtest/_slapdtest.py b/Lib/slapdtest/_slapdtest.py
index afd5304..a7a5c63 100644
--- a/Lib/slapdtest/_slapdtest.py
+++ b/Lib/slapdtest/_slapdtest.py
@@ -190,6 +190,7 @@ class SlapdObject(object):
     local_host = LOCALHOST
     testrunsubdirs = (
         'schema',
+        'slapd.d',
     )
     openldap_schema_files = (
         'core.schema',
@@ -220,6 +221,7 @@ class SlapdObject(object):
         self.server_id = self._port % 4096
         self.testrundir = os.path.join(self.TMPDIR, 'python-ldap-test-%d' % self._port)
         self._schema_prefix = os.path.join(self.testrundir, 'schema')
+        self._slapd_d = os.path.join(self.testrundir, 'slapd.d')
         self._slapd_conf = os.path.join(self.testrundir, 'slapd.conf')
         self._db_directory = os.path.join(self.testrundir, "openldap-data")
         self.ldap_uri = "ldap://%s:%d/" % (self.local_host, self._port)
@@ -418,7 +420,7 @@ class SlapdObject(object):
         slapd_args = [
             self.PATH_SLAPD,
             '-f', self._slapd_conf,
-            '-F', self.testrundir,
+            '-F', self._slapd_d,
             '-h', ' '.join(urls),
         ]
         if self._log.isEnabledFor(logging.DEBUG):

or add --enable-modules in openldap.rb formulae and rebuild it from source.

Thanks for the detailed reply, will try asap.

For reference: SlapdObject patching at runtime, no code modifications in python-ldap

@contextlib.contextmanager
def __patch_slapd_start(server: SlapdObject):
    start_routine_orig = SlapdObject._start_slapd

    def _start_slapd_patched(self: SlapdObject) -> None:
        with patch.object(self, 'testrundir', self._slapd_d) as m:
            start_routine_orig(self)

    with patch.object(SlapdObject, '_start_slapd', _start_slapd_patched):
        yield


@pytest.fixture
def ldap_server():
    server = SlapdObject()
    server.database = 'ldif'
    server.testrunsubdirs += ('slapd.d', )
    server._slapd_d = str(Path(server.testrundir, 'slapd.d'))

    with __patch_slapd_start(server):
        server.start()

    yield server
    server.stop()

I’ve encountered the same problem on Alpine Linux. We do provide MDB backend, but it’s built as a dynamic module.

I’m trying vashirov’s patch from the comment above (#422 (comment)). It fixed most of the test failures, but not all. There are still expected directory … to be empty! errors from tests that restart LDAP (e.g. Tests.t_ldapobject.Test01_ReconnectLDAPObject). Okay, so I’m adding cleanup of _slapd_d directory, it fixes another bunch of failures, but again, not all of them.

I ended up removing -F option from _start_slapd. Now all tests pass.

This is also a problem on Debian: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1008355

@vashirov could you post your patch as a PR? Thanks

@vashirov could you post your patch as a PR? Thanks

@vashirov or I can create it myself crediting you as author if you consent

@mistotebe, my patch was for 3.3.x. It is not required for 3.4.x that has changes from #382.
Test from the issue description runs fine with python-ldap-3.4.0 on macOS 12.4 with openldap-2.6.2 from homebrew.
It doesn't work with slapd that is shipped with macOS, because it's quite old (2.4.28) and it's not built with mdb support, but slapdtest uses it in the template:

So I'm not sure if anything else should be done here.

Oh, I see I missed that because superficially, your patch seemed it should still apply and definitely looked much cleaner than the current version... Thanks for the response @vashirov!

At least that's one less thing to worry about and we can unblock 3.4.1. Still, not sure what to do about defaults on Mac OS except everyone passing database='hdb'/'ldif' to the SlapdObject constructor but we'll have to figure that out later.