ISPN-1803

[Sanne]

This is lowering the cost for initializing Marshaller instances by a good 50%, as the array-size reducing configuration was not applied to most of the Marshallers we actually create.

https://issues.jboss.org/browse/ISPN-1803

[Sanne]

[FIXED]

[danberindei]

GenericJBossMarshaller is only called in the HotRod client and server - I put a breakpoint in getMarshaller() and it was never hit during a Transactional run.

And even if you were testing HotRod, GenericJBossMarshaller caches a thread-local instance of Marshaller + it doesn't have MarshalledValues - so it would create only one Marshaller instance per thread.

[Sanne]

@danberindei sorry, I messed up here. Thank you so much for checking!
The pull request is missing the most important changes - where you see FIXED in the comment above, I had realized I had many failures just after sending the pull request, and hastily removed the offending commit. Consider this a preview for now, I still need to fix it. Maybe I should have closed the pull request?

But the changes up to this point are fine I think. WDYT ?

I didn't really realize this was going to affect Hot Rod as well.

[danberindei]

@Sanne Your changes look fine - I just didn't see where the big speed-up came from :)

GenericJBossMarshaller is used by default in HotRod, but if you want to use it from other languages you have to configure ApacheAvroMarshaller instead. So perhaps we should configure ApacheAvroMarshaller to be the default instead (unless it's much slower).