mgravell / avro.net Goto Github PK

I've pushed rebased version of my branch at : 3f1645d The results vary. For 2 string fields, 24 chars each, it gives 2-5% perf boost. I didn't run any profiler yet.
We may consider:

• moving reader up, instead of stream which would lead to no allocks on the reader instances. I've noticed that reader is created for each item deserialization.
• introducing next types like AvroIntArray etc, using the same unmanaged memory chunk

Do you want to orient the serializer towards this, unsafe no-alloc scenarios as well?