RFR: 8222968 ByteArrayPublisher is not thread-safe resulting in broken re-use of HttpRequests

[Daniel Fuchs]

Hi Micahel, Chris,

On 27/06/2019 11:50, Chris Hegarty wrote:
>> http://cr.openjdk.java.net/~michaelm/8222968/webrev.1/index.html
> 
> If I’m not mistaken, then after this change there will be no explicit
> reference to the PullPublisher. There is an implicit reference from the
> nested Subscription. I think this should be ok, but just want to ensure
> that this has been considered.

I had similar concerns. Note that the FilePublisher (later in that
same file) already uses a local variable for the PullPublisher,
so there's a precedent.
Hopefully we would have seen some failures already if that had been
an issue. Once the subscription has been forwarded to the subscriber
then the publisher has served its role, and it might not matter
if it's GC'ed or not as long as the subscription itself keeps
operating. I have convinced myself that it would not be an issue
after looking at PullPublisher.java.


> On the test:
> 

[...]

> 
> 2) Given the teststabilization work that is ongoing, the test should
>     probably use the loopback, for both binding and connecting. Also,
>     rather than using the IPv4 loopback literal, defer to the loopback
>     InetAddress to retrieve the host address ( which may be IPv6 ).

Yes please!!

Otherwise looks good to me too, Michael!

cheers,

-- daniel