Fascinating. Isn't reflection in these VM languages pretty slow usually? Won't you pay that cost on access each time? Still, C# seems to be capable of a lot if you can do this ergonomically. Very cool.
That really depends which reflection APIs are being used.
For C#, you can use type reflection to get the data or shape of a piece of code or a type (GetType and friends). That isn't super fast, and isn't intended to be.
The other major C# reflection API is Reflection.Emit, which allows you emit IL at runtime. The Emit->IL->JIT process is slower than executing the code outright the first time, but once you've done it, you can get code that's as fast as as if it were compiled from source. And, in the case, you might end up with faster code, since it allows you to do transforms on how the data is laid out and compiled.
Notably, the .NET regex engine exposes options for compiling the regex down to IL if it's going to be used a lot, so a library like Trill using reflection to accelerate code by vectorizing it, while keeping better ergonomics makes a lot of sense. You still pay the cost of compiling at runtime, since it's not a compile-time transform. Those do exist for .NET, in the form of Rosalyn plugins, but that's a bit more involved still.
