Design questions

What follows are some questions that came up when I was thinking about the design for an ANSI CL linear algebra library.

• How slow is CLOS?
  • How much does it depend on the implementation?
  • How much dispatching can be done at compile-time? Answer: Not so much, as was pointed out to me by P.B. on comp.lang.lisp. The thing is, you can change the class hierarchy pretty much arbitrarily in the interpreter, so there's not much you can do at compile time. A more compile-time based approach may require building our own class system, but that carries its own share of problems with it, and some may question whether it is the Lisp Way.
  • Do we need a macro system like C++ templates to get specializations based on e.g. matrix size (for operations tuned for fixed-size vectors)?
• How slow are foreign array accesses?
  • Are Lisp compilers smart enough to optimize away the function call?
  • How much does it depend on the implementation?
• Are there popular Lisp distributions that don't support finalization? (Finalization is essential if we use dynamically allocated C arrays as the underlying storage medium.)
• Series functions (/ generators?)
  • Can we use custom scanners to support things like in-place (destructive) copy from e.g. a strided matrix view into compact storage?
  • Do series functions give performance comparable to that of a hand-rolled routine?