8234805: (dc) Remove JNI upcall from DatagramChannel.receive implementation

[Alan Bateman]

On 28/11/2019 12:02, Chris Hegarty wrote:
> :
> Just a question. Is the use of ArraysSupport::vectorizedMismatch here
> guaranteed to never return a tail (of elements to be checked)? If a
> value other than -1 (no tail) is returned, it will not be seen by the
> current code. If there is a guarantee of no tail (given the usage), then
> maybe capture the value returned by vectorizedMismatch, and assert that
> it is -1?
That is a good observation and it's probably best that it tests the 
remaining bytes. I did have this initially it wasn't used due to the 
sizeof struct sockaddr_in6.

I've created a webrev, unfortunately I fat fingered the scp location so 
we copied over the original. The only change is to the method to compare 
the two SOCKETADDESSS structures and the comment on DCI.socketAddress 
that Daniel asked to be expanded to say that it guarded by readLock. 
Daniel asked for the test to print the network configuration but we 
decided this wasn't necessary, instead the right thing is for the jtreg 
failure handler to run ifconfig/equivalent to avoid every test needing 
to do this. Daniel also suggested the test should skip when there are no 
IPv4 addresses but this isn't necessary as it doesn't require an IPv4 
address. The only change to the test is to filter out the wildcard 
address to avoid surprising configurations on Solaris where network 
interfaces appear to be plumbed with ::0. This test is only interesting 
in specific local addresses to make sure that the code in the change 
gets a bit of a work out (existing tests will exercise it too of course).

http://cr.openjdk.java.net/~alanb/8234805/webrev/

-Alan.