[Ecls-list] FFI: missing declaration garbles return value

[Michael Wildpaner]

Hi,

when declaring C foreign functions with the FFI, no function signature of 
the called function is fed to the C compiler.

A similar issue was already discussed under the subject "floats and uffi" 
in September.

If the calling convention of a particular platform does not coincide with 
the default declaration of a function, especially the default return type 
of 'int', garbage is produced.

To reproduce on Linux i386 with gcc:

foo.c:

 	double returns_double(void) {
 	  return 1.0;
 	}

use-foo.lisp:

 	(ffi:load-foreign-library "libfoo.so")
 	(ffi:def-function "returns_double" () :returning :double)

 	(format t "value is ~A~&" (returns-double))
 	(quit)

When testing with:

 	gcc  -Wall -g -O2 -fPIC -fstrict-aliasing -c foo.c -o foo.o
 	gcc -o libfoo.so foo.o -shared -lm

 	LD_LIBRARY_PATH=$PWD
 	export LD_LIBRARY

 	ecl -c use-foo.c -h use-foo.h -compile use-foo
 	ecl -load use-foo

the following output is produced:

 	value is 1.35607592d8

instead of the expected 1.0, obviously garbage and not a rounding error.

The generated use-foo.c contains the following line:

         value0=make_longfloat(returns_double()); NVALUES=1;

The implicit declaration of 'int returns_double()' makes the compiler 
generate a implicit cast of the return value from int to double. As the 
return value is already a double, it is trashed when calling 
make_longfloat (correctly declared as taking a double).

As a stop-gap measure, I will resort to (c-lines "#include \"foo.h\") to 
manually pre-declare the functions.

Do you think it would be feasible to make def-function auto-declare the 
functions it calls? The information on parameter types seems to be there.

Best wishes, Mike

-- 
If you don't breathe, there is no air.          DI Michael Wildpaner
If you don't walk, there is no earth.                  Ph.D. Student
If you don't speak, there is no world.
     -- Navajo (Dineh) wisdom