System.Text.Kdl

C# support for KDL in the style of the System.Text.Json library.

This is a goal / passion project for me personally. I will be using it as a self-directed exercise to expand my skills with several key SDKs in modern C# and to "scratch the itch" of supporting/implementing a configuration language that I find interesting / compelling. I want to:

• Compare and contrast NewtonSoft.Json (with which I am very familiar) to System.Text.Json for posterity and performance
• Use the resulting knowledge of low allocation span-based parsers to create System.Text.Kdl primitives
• Build out additional features as relevant/appropriate

Must Have

• KdlReader - Fast forward-only parser/reader (analogous to Utf8JsonReader)
• Full spec compatibility (KDL 2.0) verified via unit test converage and the the well-defined test cases
• NuGet Package for easy / safe consumption
• KdlWriter - Low-level primitive writer (analogous to Utf8JsonWriter)

Should Have

• KdlDocument - In memory DOM for random access (analogous to JsonDocument)
• KdlSerializer - Tool for serializing/deserializing KDL to/from objects in memory (analogous to JsonConvert in NewtonSoft, or JsonSerializer in System.Text.Json)

May Have (especially if prioritized by sponsorship)

• KDL css-like style selectors implementation using KdlReader for efficient results
• Linq to objects style support (c.f. JObject, etc)
• Reflection based serialization (c.f. NewtonSoft.Json, System.Text.Json)
• Helper types and routines to bind KDL to asp.net core as a supported/bindable content-type?
• Generation based serialization (c.f. System.Text.Json)

Will Not Have (without explicit sponsorship)

• Strict Support for KDL 1.0. KDL 2.0 supercedes in every descernable way and supports much of what KDL 1.0 documents would have express.
• Assumptions about the platform consuming this assembly. Should work wherever dotnet core does.
• Backwards compatibility for older .NET or older .NET core. I will be simplifying by starting with dotnet 9.0 and beyond. Even a .net standard implementation seems to be no longer strictly necessary moving forward (in my reading).