| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 9378.1 | with -ieee, HUGE_VAL is +infinity
| HYDRA::NEWMAN | Chuck Newman, 508/467-5499 (DTN 297), MRO1-3/F26 | Thu Apr 03 1997 10:58 | 9 |
| My analysis (from looking at the .lis file) shows that when compiled
with -ieee, HUGE_VAL is +infinity (0x7ff0000000000000).
The compiler does the assignment correctly, but issues a warning that the
value isn't a valid number.
If they want the largest valid number, use DBL_MAX instead of HUGE_VAL.
-- Chuck Newman
|
| 9378.2 | HUGE_VAL is +Infinity in -ieee mode | SUBPAC::FARICELLI | | Thu Apr 03 1997 11:08 | 14 |
|
Apparently the author of math.h intended for HUGE_VAL to be +Infinity
when compiled in full ieee floating point mode; and the maximum
finite double value if not.
One way to get a +Infinity is to use a number larger than the
maximum (1.8e308 will do this), but this makes the compiler issue
a warning about exceeding the range of a double, even in
-ieee_with_no_inexact (-ieee) mode. Should it?
If you want the values of the maximum finite for double/float,
see /usr/include/float.h (FLT_MAX, DBL_MAX).
-- John Faricelli
|
| 9378.3 | Fixed in Steel | SMURF::GAF | Jerry Feldman, Unix Dev. Environment, DTN:381-2970 | Thu Apr 03 1997 17:11 | 20 |
| HUGE_VAL is required to be +Infinity in IEEE mode and DBL_MAX when not
in IEEE mode. The compiler warning has been with us since day 1 (eg.
Silver SSB).
In Steel, I change HUGE_VAL to point to the libc infinity value.
#if defined(_IEEE_FP)
extern unsigned int __DINFINITY[2];
# define HUGE_VAL (*((double *) (__DINFINITY)))
#else
# define HUGE_VAL 1.797693134862315708e308
#endif
This will suppress the warning, but could cause compilation errors in
some programs that initialize variables to HUGE_VAL.
In IEEE mode, all math functions that return HUGE_VAL return +/-
infinity.
Both the old C compiler, CFE, and the new C compiler, DECC, issue the
warning.
|
| 9378.4 | | VESPER::VESPER | OpenGL Alpha Geek | Thu Apr 03 1997 17:36 | 17 |
| > In Steel, I change HUGE_VAL to point to the libc infinity value.
> #if defined(_IEEE_FP)
> extern unsigned int __DINFINITY[2];
> # define HUGE_VAL (*((double *) (__DINFINITY)))
> #else
> # define HUGE_VAL 1.797693134862315708e308
> #endif
>
> This will suppress the warning, but could cause compilation errors in
> some programs that initialize variables to HUGE_VAL.
Ouch!
Correct according to the standard (``positive double expression''), but
I'd still rather see a constant expression.
Andy V
|
| 9378.5 | | SMURF::GAF | Jerry Feldman, Unix Dev. Environment, DTN:381-2970 | Fri Apr 04 1997 15:01 | 3 |
| re: .4
Unfortunately, there is no way to put a constant in there and not have
a compiler warning. Both CFE and DECC will generate a warning.
|
| 9378.6 | | RDGENG::CHAMBERLIN | Danger! Do not Reverse Polarity | Mon Apr 07 1997 06:01 | 10 |
| Thanks for the discussion - I didn't expect a simple query to generate so much!
re .3 and .4,
Personally I think I'd rather see a warning than have unexpailned errors later
on, but I guess we have to stick with the standards.
What do other suppliers do in this case?
/Ian
|
| 9378.7 | AIX, SUNOS, HP | SMURF::GAF | Jerry Feldman, Unix Dev. Environment, DTN:381-2970 | Mon Apr 07 1997 12:01 | 16 |
| IBM AIX references a double value:
#ifdef _ANSI_C_SOURCE
extern unsigned _DBLINF[2];
#define HUGE_VAL (*((double *)(_DBLINF)))
--------------------
SunOS:
#define HUGE_VAL __huge_val._d
--------------------
HP_UX uses a constant equivalent to the MAX_DOUBLE.
|