Re: FFI Code Generation Bug

[Sam S. <sd...@gn...>]

> * In message <291...@ww...>
> * On the subject of "FFI Code Generation Bug"
> * Sent on Tue, 17 Jun 2003 02:13:08 +0200 (MEST)
> * Honorable Scott Williams <or...@gm...> writes:
>
> If more than one foreign function has a user-defined struct for a
> return type, the C compiler gets confused. When the following file is
> compiled as a module:
> 
> ----[ testffi.lisp ]------
> (defpackage "TESTFFI"
>   (:use "FFI" "COMMON-LISP"))
> (in-package "TESTFFI")
> 
> (default-foreign-language :stdc)
> 
> (def-c-struct type1
>   (arg1 int))
> 
> (def-call-out func1
>   (:return-type type1))
> 
> (def-call-out func2
>   (:return-type type1))
> ---[ end testffi.lisp ]----
> 
> I get the following function definitions in the testffi.c file:
> 
> extern struct TYPE1 { int ARG1; } (func1)();
> extern struct TYPE1 { int ARG1; } (func2)();
> 
> which results in the following compiler error:
> 
> testffi.c:14: redefinition of `struct TYPE1'
> make[1]: *** [testffi.o] Error 1
> 
> This problem does not occur if TYPE1 is only used in argument declarations.
> A possible solution is to output a
> struct TYPE1 { int ARG1; };
> when the def-c-struct command is encountered, and then refer to that type as
> "struct TYPE1" instead of redeclaring the structure each time it is
> used.

indeed.  I am not sure why we are not doing that, but I _am_ sure there
was a compelling reason for that.
e.g., the appended patch will solve your problem, but might introduce
others.

I hope Joerg will chime in with his insight.


> Another possible solution is to avoid naming the struct, which would result in
> the following code:
> extern struct { int ARG1; } (func1)();
> extern struct { int ARG1; } (func2)();
> which does not cause a compiler error.

we did that before, it caused other problems.

> I don't know anything about the internal mechanisms clisp uses to
> generate code for the FFI, otherwise i'd fix it myself :o)

see TO-C-TYPEDECL in clisp/src/foreign1.lisp.


-- 
Sam Steingold (http://www.podval.org/~sds) running RedHat9 GNU/Linux
<http://www.camera.org> <http://www.iris.org.il> <http://www.memri.org/>
<http://www.mideasttruth.com/> <http://www.palestine-central.com/links.html>
A poet who reads his verse in public may have other nasty habits.



Index: foreign1.lisp
===================================================================
RCS file: /cvsroot/clisp/clisp/src/foreign1.lisp,v
retrieving revision 1.40
diff -u -w -b -u -b -w -i -B -r1.40 foreign1.lisp
--- foreign1.lisp	12 May 2003 10:43:18 -0000	1.40
+++ foreign1.lisp	17 Jun 2003 13:34:59 -0000
@@ -560,11 +560,8 @@
          (macrolet ((with-to-c ((class typename &key (tname '(gensym "t")))
                                 &body body)
                       `(let ((,typename (format nil "~A ~A" ,class ,tname)))
-                         (unwind-protect
-                              (progn (setf (gethash c-type *type-table*)
-                                           ,typename)
-                                     ,@body)
-                           (setf (gethash c-type *type-table*) nil)))))
+                         (setf (gethash c-type *type-table*) ,typename)
+                         ,@body)))
            (case (ctype-type c-type)
              (c-struct
               (cond ((sys::memq :typedef (svref c-type 2))