HOTFIX: Use a true sentinel for UseDefaultAcls

[ijuma]

In 67fc2a9, we are using an empty collection and comparing via
value equality, so if a user passes an empty collection, they will
get the default ACLs instead of no ACLs. We fix that issue here.

[ijuma]

cc @rajinisivaram @gwenshap

[asfbot]

Refer to this link for build results (access rights to CI server needed):
https://builds.apache.org/job/kafka-pr-jdk7-scala2.10/2840/
Test PASSed (JDK 7 and Scala 2.10).

[asfbot]

Refer to this link for build results (access rights to CI server needed):
https://builds.apache.org/job/kafka-pr-jdk8-scala2.12/2840/
Test PASSed (JDK 8 and Scala 2.12).

[asfbot]

Refer to this link for build results (access rights to CI server needed):
https://builds.apache.org/job/kafka-pr-jdk8-scala2.11/2844/
Test PASSed (JDK 8 and Scala 2.11).

[rajinisivaram]

@ijuma Thanks for the PR. LGTM.

[gwenshap]

Just to clarify - we need to merge this into both trunk and 0.10.2 and roll out a new RC?
If we need a new RC, can you please comment on the email vote thread?

[ijuma]

I looked into this some more and I think we don't need to do another RC. The reason is that passing an empty list of ACLs throws an error in ZkClient.create():

if (acl == null || acl.size() == 0) {
  throw new NullPointerException("Missing value for ACL");
}

So 67fc2a9 only affects the behaviour of code that caused an exception previously. It seems that, in the worst case, it may cause test code that was checking this exceptional condition to break. Given that, it doesn't seem worth another RC by itself.

The next question is whether this PR should be merged at all. It seems safer to throw an exception in case an empty list is passed than to assume the caller wanted the default ACLs to be applied. So, I'd say we should probably merge to trunk and 0.10.2 so that we have consistent behaviour in both branches. If another RC for 0.10.2.1 is required, then we'll include it there, but otherwise, we won't.

Thoughts?

[gwenshap]

Sounds good. I'll merge to both branches