Thoughts on Herb Sutter's Metaclasses?

[Chris Katko]

One more note: It seems like this would be heaven for trying out 
new language features without having to "manually" add them into 
the compiler first. By hijacking the syntax to AST stage, we can 
add new constructs with real-functioning code, and have others 
evaluate themselves and unit test them without having to 
"imagine" whether the code would actually work. It'd be on a real 
binary, using a real (unmodified!) compiler.

And you can also have your own "in-house" specializations. Which 
is no different than expecting people to use your API. Some API's 
have method chaining, and you use it. Some have pass-by-name, and 
you use it.