NFD

Change is moved to feature-lp branch http://gerrit.named-data.net/2436

As of patchset6, I have completed implementation of forwarding pipelines. I'll move on to best-route strategy.

There's a design flaw: Interest loop pipeline sends Nack by entering outgoing Nack pipeline, which won't work, because outgoing Nack pipeline expects a PIT in-record, which is not yet inserted. I'll update the design to directly send Nack via the face.

As of patchset10, all test cases for best-route2 are added. Implementation of "after receive Interest" trigger is complete.

I'm changing strategy API "send Nack" action to take const Face& instead of shared_ptr<Face> to make strategy cleaner, at the cost of having a const_cast in Forwarder; the alternate quick solution (before #3205) is to have const_pointer_cast in strategy, which is worse.

commit:b5684b0bd83814af5f804eaf52e1bf651a4e4e69 is causing IntegrationTests to fail at test_multicast.test_udp_multicast test case to fail.

As revealed in logs, the steps leading to this failure are:

1. A,B,C are connected on the same multi-access link, and all nodes are using best-route strategy.
2. A is the consumer. C is the producer. B is neither a consumer nor a producer.
3. A sends an Interest onto the link. It's received by both B and C.
4. B returns a Nack. A returns this Nack to the consumer.
5. C returns the Data, but A has already returned a Nack to the consumer.
6. The consumer does not receive Data (from C), and thus fails the scenario.

Since we haven't defined the semantics of a Nack on a multi-access link, I propose to forbid Nack on any multi-access link.

• In incoming Nack pipeline, ignore the Nack if it comes from a multi-access face.
• In outgoing Nack pipeline and Interest loop pipeline, do not send a Nack if it goes to a multi-access face.

20150924 conference call approves note-15 solution.