[bitc-dev] Why Type Classes?

[Eric Rannaud]

On Sun, 2008-11-16 at 00:44 +0100, Florian Weimer wrote:
> Given the Ada/GNAT experience, I don't think representation
> requirements at the language level are a good thing.  There are just
> too many ambiguities, and it's rather difficult to fit these
> requirements on existing backends (but GNAT managed to do it even for
> the JVM).
> 
> If you write your own backend, this may not be an issue.  But it will
> make porting to anything else very difficult.
> 
> IMHO, a library-based solution which combines address arithmetic with
> bare-metal loads and stores is more desirable.  But you'd need some
> form of compile-time meta-programming to ensure safety while
> preserving succinctness.

I believe this discussion around VMs has overlooked what I believe is
one primary design objective for BitC: a language capable of direct
hardware access, and capable of running on bare-metal. A language for
kernels and embedded platforms. I believe there is nothing in the
current language preventing a BitC program to be compiled to run in a
freestanding environment, for instance (and that would have been on
purpose).

If being able to target a VM is a very good thing, it cannot come at the
cost of losing the kind of low-level control you sometimes need in the
environments BitC is concerned about.

If all the hard low-level stuff is outsourced to a library, presumably
written in C, isn't it somewhat besides the point?