hi!
when i was optimizing time functions in osicat i was pondering that cffi:defcfun could add an ftype declaim to the emitted defun based on the ffi types.
it may be only a heuristic, but most of the time the return value could be annotated which would help the compiler.
it would also be useful if there was a way to tell defcfun to declare the emitted defun to be inline.
</random ideas>
On Tue, May 13, 2008 at 4:51 PM, Attila Lendvai attila.lendvai@gmail.com wrote:
it may be only a heuristic, but most of the time the return value could be annotated which would help the compiler.
Sounds straightforward for integer types, and there's a pointer type somewhere as well. What about floats? From the manual:
"On most but not all systems, :float and :double represent a Lisp single-float and double-float, respectively. It is not so useful to consider the relationship between Lisp types and C types as isomorphic, as simply to recognize the relationship, and relative precision, among each respective category."
I don't remember what the implications are.
it would also be useful if there was a way to tell defcfun to declare the emitted defun to be inline.
Why not let the user decide that?
"Luís Oliveira" luismbo@gmail.com writes:
What about floats? From the manual:
"On most but not all systems, :float and :double represent a Lisp single-float and double-float, respectively. It is not so useful to consider the relationship between Lisp types and C types as isomorphic, as simply to recognize the relationship, and relative precision, among each respective category."
I don't remember what the implications are.
If I recall correctly, it means that because Lisp's float types are not specified in terms of C's, thinking of C:floats and C:doubles as L:single-floats and L:double-floats is bound to lead to confusion.
I suppose the information about float types, and perhaps all types, could come from CFFI-SYS.
-
Attila Lendvai -
Luís Oliveira -
Stephen Compall