In configure.com, the new IEEE default was a bit ahead of the configuration
question for Alpha. In numeric.c, catching potential overflow has been
simplified and better commented. I don't know of any reason the compiler
version would be relevant, so I've removed that from the ifdef. The maximum
exponent for single precision would be correct for the D_FLOAT double
format, but not for G_FLOAT; I've replaced it with NV_MAX_10_EXP, which will
reference DBL_MAX_10_EXP, which will be defined appropriately regardless of
format in float.h. The numeric tests have no complaints after this with a
non-IEEE build on Alpha.
--- configure.com;-0 Tue Aug 14 23:27:12 2001
+++ configure.com Thu Aug 16 18:21:44 2001
@@ -2352,9 +2352,10 @@
$ be_case_sensitive = "''ans'"
$! IEEE math?
$ echo ""
-$ echo "Perl normally uses G_FLOAT format floating point numbers"
-$ echo "internally, as do most things on VMS. You can, however, build"
-$ echo "with IEEE floating point numbers instead if you need to."
+$ echo "Perl normally uses IEEE format (T_FLOAT) floating point numbers"
+$ echo "internally on Alpha, but if you need G_FLOAT for binary compatibility"
+$ echo "with an external library or existing data, you may wish to disable"
+$ echo "the IEEE math option."
$ dflt = use_ieee_math
$ rp = "Use IEEE math? [''dflt'] "
$ GOSUB myread
--- numeric.c;-0 Wed Aug 15 09:14:09 2001
+++ numeric.c Thu Aug 16 18:21:45 2001
@@ -573,24 +573,21 @@
exponent = -exponent;
}
- /* Avoid %SYSTEM-F-FLTOVF_F sans VAXC$ESTABLISH.
- * In VAX VMS we by default use the D_FLOAT double format,
+ /* On OpenVMS VAX we by default use the D_FLOAT double format,
* and that format does not have *easy* capabilities [1] for
- * overflowing doubles 'silently' as IEEE fp does. Therefore we
- * need to detect early whether we would overflow (this is
- * the behaviour of the native string-to-float conversion routines,
- * and therefore the behaviour of native applications, too.)
+ * overflowing doubles 'silently' as IEEE fp does. We also need
+ * to support G_FLOAT on both VAX and Alpha, and though the exponent
+ * range is much larger than D_FLOAT it still doesn't do silent
+ * overflow. Therefore we need to detect early whether we would
+ * overflow (this is the behaviour of the native string-to-float
+ * conversion routines, and therefore of native applications, too).
*
- * [1] VAXC$EXTABLISH is the capability but it is basically a signal
- * handler setup routine, and one cannot return from a fp exception
- * handler and except much anything useful. */
-#if defined(VMS) && !defined(__IEEE_FP)
-# if defined(__DECC_VER) && __DECC_VER <= 50390006
- /* __F_FLT_MAX_10_EXP - 5 == 33 */
+ * [1] Trying to establish a condition handler to trap floating point
+ * exceptions is not a good idea. */
+#if defined(VMS) && !defined(__IEEE_FP) && defined(NV_MAX_10_EXP)
if (!negative &&
- (log10(value) + exponent) >= (__F_FLT_MAX_10_EXP - 5))
+ (log10(value) + exponent) >= (NV_MAX_10_EXP))
return NV_MAX;
-# endif
#endif
/* In UNICOS and in certain Cray models (such as T90) there is no
[end of patch]
