This looks promising however Opaque field is odd type and doesn't work? about confluent-kafka-go

[edenhill]

You are right, the pointer type is not needed.

[ljackson]

I nearly have a working solution for both produce and produce_batch I will attach a proposed diff for comments when I have something working

[ljackson]

Well this now works for p.Produce calls see note about opaque param and opaque in kafka.Message the msg one overrides the param for now. This panics at runtime for batch version but this is a POC for your consideration. Also the cgomap type as uint64 so you don't have worrie about overflow IMO also should make the map faster.

This is based on @edenhill your testing branch with master merged in and with your change above

opaque_test.patch.txt

[ljackson]

I just looked at your changes from int to uintptr so that isn't taken into account above. But I do think you want to make the map in setup() rather than the few other places your doing it now. Thx

[edenhill]

uint64 would be very convenient but since we need to cast it to a pointer it wont work on 32-bit systems.
We're thus limited to a 32-bit id and the avoid-zero wrapping code at this time.
(I could it would be possible to provide a 64-bit implementaton supported platforms with build tags, but that's a future micro optimization excercise)

[edenhill]

Thanks @ljackson, the opaque fix will be incorporated on the kafkatesting branch that will soon be meregd.