The Dave wrote:
Actually, you gain expressivity but lose a lot of power when you do metaprogramming this way, as compared with TMP. With TMP you are working with first-class expressions so you can insert references to variables and other C++ entities right into your metaprogram. Also, different DSLs can be combined (e.g. as in Spirit and Phoenix) to do a more complicated job. That's really not so easy when you are working with strings.
Yes, I had already considered this. You would need to register C++ variables with the interpreter module to make them available to the interpreted code. It would be a pain in the neck, but could be encapsulated in a wrapper function. If the interpreter returns a function that is passable it becomes just as composable as any other function. That said, I don't think it's even remotely practical. The only reason to do it is because it is cool.
The desire of "modern language theorists" is to enable language extension without losing interoperability with the rest of the language, and I think what you can do with constexpr, while very cool, still falls far short of that.
Yes, well, I guess I got a little carried away. This constexr based idea for a limited and painfully difficult to achieve mixed language application (which we have plenty of practical examples of) is indeed a far cry from implementing compiler optimizations as a library. Regards, Luke