Does Digital Mars support 80-bit long doubles (like Borland)? If it does, the "How Data is Stored" section of the "Compiler & Tools Guide Help is wrong: it says long doubles are 64 bits.

Norm

"Norm" <Norm_member@pathlink.com> wrote in message news:cehsr7$1nh1$1@digitaldaemon.com...
> Does Digital Mars support 80-bit long doubles

Yes.

> (like Borland)?

I don't know how Borland does it.

> If it does, the
> "How Data is Stored" section of the "Compiler & Tools Guide Help is wrong:
it
> says long doubles are 64 bits.

That obviously needs fixing <g>.

"Norm" <Norm_member@pathlink.com> wrote:

> Does Digital Mars support 80-bit long doubles (like Borland)? If it does, the "How Data is Stored" section of the "Compiler & Tools Guide Help is wrong: it says long doubles are 64 bits.

DMC does support it, but only under Win32. See <float.h> for definitions. Borland vs DMC use the same layout for long doubles. How could they *not* do? Most of the calculations are done in silicon, so Intel and it's IEEE standard has set the scene here.

From DMC's <float.h>:
#define LDBL_MIN        0x1p-16382L
#define LDBL_MIN_10_EXP -4932

and Borland:
#define LDBL_MIN            _tiny_ldble
#define LDBL_MIN_10_EXP     -4931

Someone is off by one here, but probably doesn't matter; the exponent is already 14 bits in long doubles AFAICR.

But I don't understand why Borland uses an extern var instead of a constant LDBL_MIN etc. Avoiding inline  math comparisons, linker problems?

--gv