single float results: a problem with C also, not just Fortran!

[Aaron Sloman]

I had somehow assumed that the interface between linux poplog
and C worked, because all sorts of things have been using it,
including all the X window facilities, such as XVed, rclib,
propsheet and other things.

So I assumed that the recently reported problem about a fortran function
failing to return a single float result to pop11 on linux even though
double floats all work fine was specific to fortran.

But I've just done a simple test which indicates that there is a problem
with the C interface also. (I am not a C programmer -- so I may have
made a mistake).

This is what I did, copying and modifying an earlier test program
provided by David Young:

I created a file called floatbaz.c containing only:

     double dbaz(double x)
         {return x-100.0;}

     float baz(float x)
         {return x-100.0;}

     int ibaz(int x)
         {return x-100;}

I compiled this with
     gcc -o floatbaz.so -fpic -shared floatbaz.c

which produced a file: floatbaz.so

I then did this in pop11:

 exload floatbaz [floatbaz]
     (language C)
     dbaz(1):dfloat,
     baz(1):float,
     ibaz(1):int
 endexload;


;;; testing dbaz produces expected result
 exacc dbaz(301.0) =>
 ** 201.0

;;; testing baz produces wrong result
 exacc baz(301.0) =>
 ** 0.0

 exacc baz(1.0) =>
 ** 0.0

;;; testing ibaz produces expected result
 exacc ibaz(301) =>
** 201

So integer results and double float results work fine, but not
single float results. Changing the exload command to use sfloat
    baz(1):sfloat,

makes no difference.

So the problem is there in C too (if what I have done is OK).

I apologise for not checking this earlier.

This is actually reassuring as it is likely that some relatively simple
fix to do with converting a machine float to a pop11 decimal will
fix the problem, whereas something that worked for C and not for Fortran
would be more difficult.

It looks as if this bit of the linux Poplog port was never tested
properly.

Aaron