* inexact->exact on nan and inf
@ 2003-09-23 23:50 Kevin Ryde
0 siblings, 0 replies; only message in thread
From: Kevin Ryde @ 2003-09-23 23:50 UTC (permalink / raw)
[-- Attachment #1: Type: text/plain, Size: 1433 bytes --]
* numbers.c (scm_inexact_to_exact): Don't depend on what double->long
cast gives for values bigger than a long, or for nan or inf.
* tests/numbers.test (inexact->exact): New tests.
This is merely a defensive proposal, it actually comes out ok on my
i386 debian already, since casting double->long gives 0x80000000 or
0x7FFFFFFF for values out of range, which of course don't pass
SCM_FIXABLE. But it doesn't seem wise to assume such values.
New code, for ease of contemplation,
/* The values SCM_MOST_POSITIVE_FIXNUM+1 and SCM_MOST_NEGATIVE_FIXNUM
are both powers of 2, so there's no rounding when making "double"
values from them. If plain SCM_MOST_POSITIVE_FIXNUM was used it
could get rounded on a 64-bit machine, hence the "+1".
The use of floor to force to an integer value ensures we don't
depend on how a double->long cast will round or how mpz_set_d will
round. For reference, double->long probably follows the hardware
rounding mode, whereas mpz_set_d truncates towards zero. */
double u = SCM_REAL_VALUE (z);
if (xisinf (u) || xisnan (u))
scm_num_overflow (s_scm_inexact_to_exact);
u = floor (u + 0.5);
if (u < (double) (SCM_MOST_POSITIVE_FIXNUM+1)
&& u >= (double) SCM_MOST_NEGATIVE_FIXNUM)
return SCM_MAKINUM ((long) u);
else
return scm_i_dbl2big (u);
[-- Attachment #2: numbers.c.inexact-nan.diff --]
[-- Type: text/plain, Size: 1366 bytes --]
--- numbers.c.~1.200.~ 1970-01-01 10:00:01.000000000 +1000
+++ numbers.c 2003-09-23 16:33:30.000000000 +1000
@@ -4337,14 +4337,25 @@
return z;
else if (SCM_REALP (z))
{
- double u = floor (SCM_REAL_VALUE (z) + 0.5);
- long lu = (long) u;
- if (SCM_FIXABLE (lu))
- return SCM_MAKINUM (lu);
- else if (!xisinf (u) && !xisnan (u))
- return scm_i_dbl2big (u);
- else
+ /* The values SCM_MOST_POSITIVE_FIXNUM+1 and SCM_MOST_NEGATIVE_FIXNUM
+ are both powers of 2, so there's no rounding when making "double"
+ values from them. If plain SCM_MOST_POSITIVE_FIXNUM was used it
+ could get rounded on a 64-bit machine, hence the "+1".
+
+ The use of floor to force to an integer value ensures we don't
+ depend on how a double->long cast will round or how mpz_set_d will
+ round. For reference, double->long probably follows the hardware
+ rounding mode, but mpz_set_d truncates towards zero. */
+
+ double u = SCM_REAL_VALUE (z);
+ if (xisinf (u) || xisnan (u))
scm_num_overflow (s_scm_inexact_to_exact);
+ u = floor (u + 0.5);
+ if (u < (double) (SCM_MOST_POSITIVE_FIXNUM+1)
+ && u >= (double) SCM_MOST_NEGATIVE_FIXNUM)
+ return SCM_MAKINUM ((long) u);
+ else
+ return scm_i_dbl2big (u);
}
else
SCM_WRONG_TYPE_ARG (1, z);
[-- Attachment #3: numbers.test.inexact.diff --]
[-- Type: text/plain, Size: 692 bytes --]
--- numbers.test.~1.32.~ 1970-01-01 10:00:01.000000000 +1000
+++ numbers.test 2003-09-22 16:34:30.000000000 +1000
@@ -2015,6 +2015,24 @@
;;; inexact->exact
;;;
+(with-test-prefix "inexact->exact"
+
+ (pass-if-exception exception:numerical-overflow "+inf"
+ (inexact->exact +.inf))
+
+ (pass-if-exception exception:numerical-overflow "-inf"
+ (inexact->exact -.inf))
+
+ (pass-if-exception exception:numerical-overflow "nan"
+ (inexact->exact +.nan))
+
+ (with-test-prefix "2.0**i to exact and back"
+ (do ((i 0 (1+ i))
+ (n 1.0 (* 2.0 n)))
+ ((> i 100))
+ (pass-if (list i n)
+ (= n (inexact->exact (exact->inexact n)))))))
+
;;;
;;; integer-length
;;;
[-- Attachment #4: Type: text/plain, Size: 142 bytes --]
_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-devel
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2003-09-23 23:50 UTC | newest]
Thread overview: (only message) (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2003-09-23 23:50 inexact->exact on nan and inf Kevin Ryde
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).