sprintf %f panics with NaN #9027
Comments
From zefram@fysh.orgCreated by zefram@fysh.orgOn NetBSD, frexp() produces silly results if given a NaN argument. START_PATCH Inline Patch--- sv.c.orig 2006-08-15 13:37:41.000000000 +0100
+++ sv.c 2007-09-11 23:09:09.555166783 +0100
@@ -9138,7 +9138,7 @@
: SvNVx(argsv);
need = 0;
- if (c != 'e' && c != 'E') {
+ if (c != 'e' && c != 'E' && nv == nv) {
i = PERL_INT_MIN;
/* FIXME: if HAS_LONG_DOUBLE but not USE_LONG_DOUBLE this
will cast our (long double) to (double) */
This problem came up in CPAN testers reports on one of my modules. Perl Info |
From @TuxOn Tue, 11 Sep 2007 15:22:23 -0700, Zefram (via RT)
From my C-knowledge point of view, I completely fail to see the effect It is like writing if (1 == 1) {
-- |
|
The RT System itself - Status changed from 'new' to 'open' |
From @nwc10On Wed, Sep 12, 2007 at 08:50:30AM +0000, H.Merijn Brand wrote:
We'd better call you icc then. :-( NaN != NaN Nicholas Clark |
From @TuxOn Wed, 12 Sep 2007 09:53:08 +0100, Nicholas Clark <nick@ccl4.org> wrote:
tux will do
/me realizes again that he is never too old to learn new tricks -- |
From david@cantrell.org.ukOn Tue, Sep 11, 2007 at 03:22:23PM -0700, Zefram wrote:
'Twas I who generated the original bug report. I'm downloading blead as -- The voices said it's a good day to clean my weapons |
From @rgsOn 12/09/2007, via RT Zefram <perlbug-followup@perl.org> wrote:
now successfully tested. Thanks, applied as #31854. |
From @demerphqOn 9/12/07, Rafael Garcia-Suarez <rgarciasuarez@gmail.com> wrote:
Is it worth tweaking this to explain the point of the nv==nv check? yves |
From @rgsOn 12/09/2007, Rafael Garcia-Suarez <rgarciasuarez@gmail.com> wrote:
Further amended like this: Change 31855 by rgs@stcosmo on 2007/09/12 11:48:32 Amend change 31854 for when NAN_COMPARE_BROKEN Affected files ... ... //depot/perl/sv.c#1423 edit Differences ... ==== //depot/perl/sv.c#1423 (text) ==== @@ -9194,7 +9194,12 @@ need = 0; |
From perl5-porters@perl.orgOn Tue, Sep 11, 2007 at 03:22:23PM -0700, Zefram wrote:
This is only a partial fix. The affected module gets a little bit t/class_all..........ok 1/35panic: frexp at t/class_all.t line 9, <DATA> -- Just because it is possible to do this sort of thing |
From zefram@fysh.orgDavid Cantrell wrote:
Is that, by any chance, failing on an infinity? Try this test: perl -we 'printf "%f\n", 2**1e100' and for good measure: perl -we 'printf "%f\n", -2**1e100' If that's the problem, infinities will have to be detected in the sprintf -zefram |
From david@cantrell.org.ukOn Wed, Sep 12, 2007 at 04:57:17PM +0100, Zefram wrote:
Yup $ .../perl-5.10.0/bin/perl -we 'printf "%f\n", 2**1e100'
Same again -- I remember when computers were frustrating because they did |
From @nwc10On Wed, Sep 12, 2007 at 04:57:17PM +0100, Zefram wrote:
Given the spec in: http://www.opengroup.org/pubs/online/7908799/xsh/frexp.html perl is being buggy for not checking for NaN, +Inf or -Inf, because the Given this: $ perl -wle 'print 0 * "nan"' (and a non-broken nan comparison) it seems to be fairly easy to check for Nicholas Clark |
From @nwc10On Wed, Sep 12, 2007 at 06:20:48PM +0100, Nicholas Clark wrote:
And the converse is that 0 * anything else is 0, which is easier to check for, Nicholas Clark Change 31856 by nicholas@nicholas-saigo on 2007/09/12 18:38:54 Don't call frexp() on +Inf or -Inf either. (It's not just NaN it might Affected files ... ... //depot/perl/sv.c#1424 edit Differences ... ==== //depot/perl/sv.c#1424 (text) ==== @@ -9194,12 +9194,9 @@ need = 0; |
From @nwc10On Wed, Sep 12, 2007 at 07:40:08PM +0100, Nicholas Clark wrote:
I'm not sure how (or where) to write good regression tests for this, that don't Nicholas Clark |
From zefram@fysh.orgNicholas Clark wrote:
You don't need to know whether NaNs and infinities actually exist. use Test::More; -zefram |
From david@cantrell.org.ukOn Wed, Sep 12, 2007 at 07:40:08PM +0100, Nicholas Clark wrote:
Which works, for both NaNs and Infs, on NetBSD 2.1.0_STABLE/alpha at -- "Cynical" is a word used by the naive to describe the experienced. |
From p5p@spam.wizbit.beSmall history of the ticket: $ perl -we 'printf "%f\n", Inf/-Inf/Nan' (This was fixed by changes: 31855, 31856) On Wed Sep 12 13:59:27 2007, zefram@fysh.org wrote:
I checked t/op/sprintf.t <http://public.activestate.com/cgi-bin/ Were they added elsewhere? Or were there any problems with the test Kind regards, Bram |
From zefram@fysh.orgBram via RT wrote:
As far as I know, no one got around to it. My mail records don't include -zefram |
|
p5p@spam.wizbit.be - Status changed from 'open' to 'resolved' |
From p5p@perl.wizbit.beQuoting Zefram <zefram@fysh.org>:
Patch attached. Kind regards, Bram |
From p5p@perl.wizbit.beInline Patch--- old/t/op/sprintf2.t 2008-04-28 20:34:08.000000000 +0200
+++ new/t/op/sprintf2.t 2008-04-29 22:22:54.000000000 +0200
@@ -6,7 +6,7 @@
require './test.pl';
}
-plan tests => 1292;
+plan tests => 1295;
is(
sprintf("%.40g ",0.01),
@@ -134,3 +134,9 @@
}
}
+
+# test that %f doesn't panic with +Inf, -Inf, NaN [perl #45383]
+foreach my $n (2**1e100, -2**1e100, 2**1e100/2**1e100) { # +Inf, -Inf, NaN
+ eval { my $f = sprintf("%f", $n); };
+ is $@, "", "sprintf(\"%f\", $n)";
+} |
From @rgsThanks, applied. 2008/4/29 Bram <p5p@perl.wizbit.be>:
|
Migrated from rt.perl.org#45383 (status was 'resolved')
Searchable as RT45383$
The text was updated successfully, but these errors were encountered: