README.org

cffi-ops

Write CFFI stuff quickly without runtime overhead.

Introduction

CFFI is powerful, but using its API to write C-style code can sometimes be cumbersome because it requires you to repeatedly pass in types, unlike the dot operator in C that has some type inference capabilities.

This library provides CFFI with dot operator-like functionality at compile time, allowing you to write CFFI-related code as simple as C with just a small amount of FFI type declarations.

This library has been tested to work on SBCL, CCL, ECL, ABCL, and CLISP, and theoretically is portable across implementations that provide macroexpand-all.

Rules

Here is a comparison table between C syntax:

C	cffi-ops
x->y.z or x->y->z	(-> x y z) (Note that x, y, and z must be the same symbols used in defcstruct)
&x->y	(& (-> x y))
*x	([] x)
x[n]	([] x n)
&x[n] or x + n	(& ([] x n))
x.y = z	(setf (-> x y) z) if z is a variable
	(csetf (-> x y) z) if z is a CFFI pointer
A _a, *a = &_a	(clet ((a (foreign-alloca '(:struct A)))) ...)
A *a = malloc(sizeof(A))	(clet ((a (cffi:foreign-alloc '(:struct A)))) ...)
A _a = *b, *a = &_a	(clet ((a ([] b))) ...)
A *a = b	(clet ((a b)) ...)

Please note that since it is not possible to directly manipulate C compound types in Lisp, binding and assignment of compound types require the use of clet (or clet*) and csetf, which bind and operate on variables that are CFFI pointers.

And the symbol -> is directly exported from the arrow-macros package, so this library is fully compatible with arrow-macros, which means you can freely use all the macros (including ->) provided by arrow-macros inside or outside of clocally, clet, clet*, or csetf.

Example

For the following C code:

#include <stdlib.h>
#include <assert.h>

typedef struct {
  float x;
  float y;
  float z;
} Vector3;

typedef struct {
  Vector3 v1;
  Vector3 v2;
  Vector3 v3;  
} Matrix3;

void Vector3Add(Vector3 *output, const Vector3 *v1, const Vector3 *v2) {
  output->x = v1->x + v2->x;
  output->y = v1->y + v2->y;
  output->z = v1->z + v2->z;
}

int main(int argc, char *argv[]) {
  Matrix3 m1[3];
  m1[0].v1.x = 1.0;
  m1[0].v1.y = 2.0;
  m1[0].v1.z = 3.0;
  Matrix3 m2 = *m1;
  Vector3 *v1 = &m2.v1;
  Vector3 *v2 = malloc(sizeof(Vector3));
  *v2 = *v1;
  v2->x = 3.0;
  v2->z = 1.0;
  Vector3Add(v1, v1, v2);
  assert(v1->x == 4.0);
  assert(v1->y == 4.0);
  assert(v1->z == 4.0);
  free(v2);
  return 0;
}

The equivalent Lisp code (written using cffi-ops) is:

(defpackage cffi-ops-example
  (:use #:cl #:cffi #:cffi-ops))

(in-package #:cffi-ops-example)

(defcstruct vector3
  (x :float)
  (y :float)
  (z :float))

(defcstruct matrix3
  (v1 (:struct vector3))
  (v2 (:struct vector3))
  (v3 (:struct vector3)))

(defun vector3-add (output v1 v2)
  (clocally
    (declare (ctype (:pointer (:struct vector3)) output v1 v2))
    (setf (-> output x) (+ (-> v1 x) (-> v2 x))
          (-> output y) (+ (-> v1 y) (-> v2 y))
          (-> output z) (+ (-> v1 z) (-> v2 z)))))

(defun main ()
  (clet ((m1 (foreign-alloca '(:array (:struct matrix3) 3))))
    (setf (-> ([] m1 0) v1 x) 1.0
          (-> ([] m1 0) v1 y) 2.0
          (-> ([] m1 0) v1 z) 3.0)
    (clet* ((m2 ([] m1))
            (v1 (& (-> m2 v1)))
            (v2 (foreign-alloc '(:struct vector3))))
      (csetf ([] v2) ([] v1))
      (setf (-> v2 x) 3.0
            (-> v2 z) 1.0)
      (vector3-add v1 v1 v2)
      (assert (= (-> v1 x) 4.0))
      (assert (= (-> v1 y) 4.0))
      (assert (= (-> v1 z) 4.0))
      (foreign-free v2))))

And the equivalent Lisp code (written without using cffi-ops) is:

(defpackage cffi-example
  (:use #:cl #:cffi))

(in-package #:cffi-example)

(defcstruct vector3
  (x :float)
  (y :float)
  (z :float))

(defcstruct matrix3
  (v1 (:struct vector3))
  (v2 (:struct vector3))
  (v3 (:struct vector3)))

(declaim (inline memcpy))
(defcfun "memcpy" :void
  (dest :pointer)
  (src :pointer)
  (n :size))

(defun vector3-add (output v1 v2)
  (with-foreign-slots (((xout x) (yout y) (zout z)) output (:struct vector3))
    (with-foreign-slots (((x1 x) (y1 y) (z1 z)) v1 (:struct vector3))
      (with-foreign-slots (((x2 x) (y2 y) (z2 z)) v2 (:struct vector3))
        (setf xout (+ x1 x2) yout (+ y1 y2) zout (+ z1 z2))))))

(defun main ()
  (with-foreign-object (m1 '(:struct matrix3) 3)
    (with-foreign-slots ((x y z)
                         (foreign-slot-pointer
                          (mem-aptr m1 '(:struct matrix3) 0)
                          '(:struct matrix3) 'v1)
                         (:struct vector3))
      (setf x 1.0 y 2.0 z 3.0))
    (with-foreign-object (m2 '(:struct matrix3))
      (memcpy m2 m1 (foreign-type-size '(:struct matrix3)))
      (let ((v1 (foreign-slot-pointer m2 '(:struct matrix3) 'v1))
            (v2 (foreign-alloc '(:struct vector3))))
        (memcpy v2 v1 (foreign-type-size '(:struct vector3)))
        (with-foreign-slots ((x z) v2 (:struct vector3))
          (setf x 3.0 z 1.0))
        (vector3-add v1 v1 v2)
        (with-foreign-slots ((x y z) v1 (:struct vector3))
          (assert (= x 4.0))
          (assert (= y 4.0))
          (assert (= z 4.0)))
        (foreign-free v2)))))

Both of them should generate almost equivalent machine code in SBCL and have very similar performance.