Re: [Gm2] complex types

[Gaius Mulley]

From: Gaius Mulley <address@hidden>
Subject: Re: [Gm2] complex types
To: Andreas Fischlin <address@hidden>
Cc: GM2 <address@hidden>

Andreas Fischlin <address@hidden> writes:

> Arn't C library conversion routines already longreal? MathLib0

Hi Andreas,

ahh I believe there could be a problem here..

> definition looks similar to this (within RAMSES/RASS):

> DEFINITION FOR C MODULE CMath; 
>
>  PROCEDURE sqrt(x : LONGREAL) : LONGREAL;
>  PROCEDURE sin (x : LONGREAL) : LONGREAL;
>  PROCEDURE cos (x : LONGREAL) : LONGREAL;
>  PROCEDURE exp (x : LONGREAL) : LONGREAL;
>  PROCEDURE log (x : LONGREAL) : LONGREAL;
>  PROCEDURE atan(x : LONGREAL) : LONGREAL;
>
> END CMath.

>    PROCEDURE Sin(x: REAL): REAL;
>    BEGIN
>      RETURN CMath.sin(x)
                    ^^^
>    END Sin;

I think there might be a portability problem with the above (in gm2) -
as the C function sin is defined to take and return a 'double' whereas
in gm2 LONGREAL maps onto a 'long double'.

cut from 'man 3 sin':

  #include <math.h>

  double sin(double x);
  float sinf(float x);
  long double sinl(long double x);

I guess it works on systems where sizeof(double)==sizeof(long double)
or with Modula-2 compilers where TSIZE(LONGREAL)==sizeof(double).  But
it will fail with gm2 as

LONGREAL == long double
REAL == double
SHORTREAL == float


For example:

address@hidden:~/GM2$  gm2 -g -fcpp -c sizes.mod
address@hidden:~/GM2$  gm2 -g -fcpp sizes.mod
address@hidden:~/GM2$ ./a.out

size of CHAR is 1 bytes
size of INTEGER is 4 bytes
size of CARDINAL is 4 bytes
size of LONGINT is 8 bytes
size of LONGCARD is 8 bytes
size of SHORTREAL is 4 bytes
size of REAL is 8 bytes
size of LONGREAL is 16 bytes
size of BITSET is 4 bytes

address@hidden:~/GM2$ gcc sizes.c
address@hidden:~/GM2$ ./a.out

size of char is: 1 bytes
size of int is: 4 bytes
size of unsigned int is: 4 bytes
size of long int is: 8 bytes
size of long unsigned int is: 8 bytes
size of long long int is: 8 bytes
size of long long unsigned int is: 8 bytes
size of float is: 4 bytes
size of double is: 8 bytes
size of long double is: 16 bytes

So specifically at position "^^^" we see that gm2 will attempt to pass
a "long double" parameter to a C function expecting a "double"

regards,
Gaius