In SE2.2, the mapping is:

INTEGER_8     -> int8_t (defined at stdint.h, ISO C 99)
INTEGER_16    -> int16_t 
INTEGER_32    -> int32_t
INTEGER_64    -> int64_t
CHARACTER     -> unsigned char
REAL_32       -> float
REAL_64       -> double
REAL_EXTENDED -> long double
BOOLEAN       -> char
POINTER       -> void*

The intXXX_t are defined in GNU systems (linux glibc, cygwin, and
similar) as:

int8_t  -> signed char
int16_t -> short int
int32_t -> int
int64_t -> long long int (in 32 bit machines)
int64_t -> long int (in 64 bit machines)

The following are aliases in SE:
  INTEGER -> INTEGER_32  -- Bound to change in future versions of SE.
  REAL    -> REAL_64

So, correcting your table:
> C Type  Eiffel type
> char  CHARACTER
Actually, signed char

> int   INTEGER (or INTEGER_32?)
right

> float REAL_32 
right

> double        REAL (or REAL_64???)
right. They are both the same

> short int     INTEGER_16
right

> long int      INTEGER_64
long int is 32-bit in 32-bit machines, so not in that case. see above

> long double   ???
REAL_EXTENDED

> Then how do you deal with types that have different length under
> different architectures (x86, x86-64, PPC, PPC64, x390, ARM et cetera)?

AFAIK, the only issue should be with "long int". Most of current
architectures/compiler have the same length for other types.


/usr/include/glib-2.0/glib/gtypes.h:

typedef char   gchar;
typedef short  gshort;
typedef long   glong;
typedef int    gint;
typedef gint   gboolean;

typedef unsigned char   guchar;
typedef unsigned short  gushort;
typedef unsigned long   gulong;
typedef unsigned int    guint;

typedef float   gfloat;
typedef double  gdouble;

typedef void* gpointer;
typedef const void *gconstpointer;