PolyType-based TAML serializer for .NET?

[AArnott]

Rather than re-invent all the rules for serialization including attributes, how to interpret various types, etc., what about basing your new TAML serializer on PolyType?

A fully developed PolyType-based serializer is Nerdbank.MessagePack, and while there's a lot of code there to support the messagepack binary format with high performance, serializers can be relatively small and simple (like the CBOR, JSON and Xml example serializers here).

If you're interested in this, but don't have the time for it, would you accept a PR?

CC: @eiriktsarpalis

[AArnott]

I'm whipping up a prototype. It will only be partially functional until I get the go ahead to keep going from you.

[AArnott]

main...AArnott:polytype

[csharpfritz]

This would definitely help with specifying data types and helping with an eventual schema feature.

Additionally, there was a request for a TAML -> YAML converter utility to help with adoption. I can see PolyType helping with that utility tool as well

[AArnott]

Great. I'll keep developing it then.
Even with PolyType, most of the work in implementing a serializer is repetitive, leaving only the actual formatting of the bytes to be unique (and other traits like non-support for multi-line strings). So between that and your interest in a YAML converter utility as well, all based on PolyType, I think it's time for me to resurrect my PolyType based multi-format serializer which I started early this year and then abandoned.

In particular if you're comfortable with only target net10.0 as I see you're currently doing, this is a particularly viable option.

[csharpfritz]

There's notes about coming support for multi-line text in #5

[AArnott]

Regarding schemas, Nerdbank.MessagePack demonstrates how a schema can be automatically generated for a given type graph. In the case of that library, we express the schema using JSON Schema. But it could be anything, really.
The multi-format serializer I'm picking back up will have schema generation built-in, in the same manner.

[AArnott]

Here it is:

https://github.com/AArnott/ShapeShift/tree/main/src/ShapeShift.Taml

ShapeShift is brand new, though it's borrowing heavily from Nerdbank.MessagePack, so its API is very unstable. As such, I'd prefer to keep all encoder/decoder libraries in the same repo while we iterate on it. TAML I realize is also iterating quickly and it can be useful to iterate the spec and code at the same time, so I'm interested to hear your thoughts on this.

[AArnott]

There are many more features (and thousands of tests) that I still plan to port from Nerdbank.MessagePack over to ShapeShift, and apply them such that as many of them as possible automatically run on all the ShapeShift encoders.
Work over the next few days or weeks.