Commit "nl: Return -NLE_AGAIN if non-blocking socket would block" breaks nl_cache_mngr_data_ready?

[Andrew Collins]

I'm upgrading an application to libnl 3.2.22, and I ran into an issue
with my usual poll()/nl_cache_mngr_data_ready() method for watching
for link updates.

Every time I get a link update and poll wakes up, I call
nl_cache_mngr_data_ready() which consumes data on a non-blocking
socket (created through the cache mgr) until nl_recv returns something
<= 0.

Prior to 3.2.22, -EAGAIN would be handled within nl_recv and send back
0, stopping nl_cache_mngr_data_ready() which would then return
success.

Since "nl: Return -NLE_AGAIN if non-blocking socket would block"
however, nl_recv returns an actual error and nl_cache_mngr_data_ready
interprets this as a failure.  Users of
nl_cache_mngr_poll will likely see the same issue.

What's the right thing to do here?  Should nl_cache_mngr_data_ready be
handling -NLE_AGAIN or is my application expected to?