Using Span and Memory in custom serializers

[vilinski]

Hi all,

I'm tweaking my custom serializer right now. We use netcoreapp3.1 and will update to net6.0 if it is released in November. And since RabbitMQ.Client library exposes ReadOnlyMemory<byte> as a payload would be useful to read it right from the memory without allocating string or array first. Therefore speaks nothing against using Span and Memory in our custom implementation, if not in Utf8Json itself yet.

• I've heard in some comments somebody has already a fork with a customized writer/reader, which we could reuse?
• JsonReader.ReadPropertyNameSegmentRaw() allocates a string, which will be not used by instantiating of object. Ideally, I would compare the array range with the expected property name string and just advance the offset. Does anybody already uses this way JsonReader.ReadStringSegmentRaw() and JsonReader.ReadPropertyNameSegmentRaw() in your serializers? The latter messed up the offset in my test and I can't figure out why.
• how do you check whether the property name is right else?