the idea to convert BasicAttributes to our BasicAttributesImpl was a good idea, but which turned out wrong ... The problem now, if we do that, is that *all* the data will be corrupted, as we have serialized BasicAttributes, not our version.
This is something we can fix in 1.5 version, because it's not supposed to be a stable version, but for 1.0.2, this is definitively a bad situation.
I guess that you might have to educate your developpers about how bad is JNDI, and that they should avoid new BasicAttributes() like plague...
GuysI think I understand and agree with the changes you've made... as you tighten up the implementation of DS, old code that used to 'get away with it' needs to break...But you can't blame me for trying to reverse this change... I have a couple of sad faced developers back here that now need to change and retest their code ;-)I'll await any final fix and in the mean time change our use of BasicAttributes().SimonTSorry (unfortunately) no. We cannot do that. The old behavior was just plain wrong.
On 6/1/07, Simon.Temple@saaconsultants.com <Simon.Temple@saaconsultants.com> wrote:Hi AlexYour suggestion makes good sense, a developer would get early visibility of any attempt to use the "wrong type" of BasicAttributes.However, I now have code that used to work with DS 1.0.0 that no longer works with DS 1.0.2, furthermore I have code that will work if used with a remote context (i.e. via LDAP communications) and won't if I run with an embedded directory context.It's the inconsistent behaviour of the same interface between versions and remote/local contexts that I have found confusing.Is there any way we can have the old behaviour back where it didn't see to matter what type of BasicAttributes() we used? :o)Kind RegardsSimon Temple+1 on all of Emmanuel's points. Let me add one additional option.
We could write some protective code to detect when the BasicAttributes is created without the
boolean argument for ignoring case. We should have some kind of RuntimeException for it.
Could be a ConfigurationException maybe? Don't know exactly but basically we need:
BasicAttributes attrs = ...
if ( ! attrs.isCaseIgnored() )
throw new "something that tells the user to correct this miss configuration of BasicAttributes";
On 6/1/07, Emmanuel Lecharny <firstname.lastname@example.org> wrote:Hi Simon,
you can raise a JIRA, because we should not allow NPE in the server.
Now, regarding Attributes(), you should *always* pass true as an
argument. This is a major drawback of JNDI as a generic API for
directory, because it does not take into account the underlying
directory it has to deal with. There is nothing we can do about it,
except fixing NPEs, when it comes to use ADS embedded (meaning :
without the network layer which automatically substitutes
BasicAttributes to a more ldap compliant BasicAttributesImpl...)
On 6/1/07, Simon.Temple@saaconsultants.com
< Simon.Temple@saaconsultants.com> wrote:
> We've found a problem with DS 1.0.2. This problem only exists when running
> with DS embedded in the same VM.
> Running the same code remotely (outside of DS VM) works fine.
> Example code:
> Attributes attrs = new BasicAttributes();
> attrs.put("objectClass", "organizationalUnit");
> attrs.put("description", "Test OU");
> DirContext subContext = context.createSubcontext("ou=Test",
> Exception from createSubcontext():
> Caused by: java.lang.NullPointerException
> org.apache.directory.server.core.schema.SchemaService.assertAllAttributesAllowed (SchemaService.java:1806)
> org.apache.directory.server.core.schema.SchemaService.add(SchemaService.java :1636)
> ... 130 more
> If you change the BasicAttributes() constructor call to:
> Attributes attrs = new BasicAttributes( true );
> it works fine.
> This issue means we cannot use DS 1.0.2. Should I raise a JIRA entry yet or
> should I hold off until you guys have had chance to check my findings?
> Many Thanks