Skip Menu |
 

This queue is for tickets about the Math-Float128 CPAN distribution.

Report information
The Basics
Id: 132376
Status: open
Priority: 0/
Queue: Math-Float128

People
Owner: Nobody in particular
Requestors: SREZIC [...] cpan.org
Cc:
AdminCc:

Bug Information
Severity: (no value)
Broken in: 0.12
Fixed in: (no value)



Subject: Test suite fails on Fedora 31
Download (untitled) / with headers
text/plain 410b
t/03arith.t and t/13constants.t fail on my Fedora 31 smoker: ... Expected 1 Got 8 t/03arith.t .................. Failed 1/20 subtests ... M_1_PIq: 3.18309886183790671537767526745028737e-01 M_2_PIq: 6.36619772367581343075535053490057474e-01 M_SQRT2q: 1.41421356237309504880168872420969798e+00 t/13constants.t .............. Failed 3/23 subtests ... The t/03arith.t failure also happens of Ubuntu 19.10.
Download (untitled) / with headers
text/plain 546b
On 2020-04-19 03:41:49, SREZIC wrote: Show quoted text
> t/03arith.t and t/13constants.t fail on my Fedora 31 smoker: > > ... > > Expected 1 > Got 8 > t/03arith.t .................. > Failed 1/20 subtests > ... > M_1_PIq: 3.18309886183790671537767526745028737e-01 > M_2_PIq: 6.36619772367581343075535053490057474e-01 > M_SQRT2q: 1.41421356237309504880168872420969798e+00 > t/13constants.t .............. > Failed 3/23 subtests > ... > > The t/03arith.t failure also happens of Ubuntu 19.10.
On debian/testing (bullseye, version 11) both tests also fail.
Download (untitled) / with headers
text/plain 1.5k
On Sun Apr 19 03:49:20 2020, SREZIC wrote: Show quoted text
> On 2020-04-19 03:41:49, SREZIC wrote:
> > t/03arith.t and t/13constants.t fail on my Fedora 31 smoker: > > > > ... > > > > Expected 1 > > Got 8 > > t/03arith.t .................. > > Failed 1/20 subtests > > ... > > M_1_PIq: 3.18309886183790671537767526745028737e-01 > > M_2_PIq: 6.36619772367581343075535053490057474e-01 > > M_SQRT2q: 1.41421356237309504880168872420969798e+00 > > t/13constants.t .............. > > Failed 3/23 subtests > > ... > > > > The t/03arith.t failure also happens of Ubuntu 19.10.
> > On debian/testing (bullseye, version 11) both tests also fail.
Hi Slaven,I got the same FAIL reports from Peter Acklam's smoke a few days ago. The 03arith.t failure is a bug in that test script. I've assumed that signbit() will return 1 if the signbit is set ... but I should instead be checking whether it returns a true value or not. Looks to me that the values quoted in your test report for M_1_PIq, M_2_PIq and M_SQRT2q are correct, whereas other versions of quadmath.h are assigning less precise (and therefor incorrect, IMO) approximations.I'll take steps to address those anomalies.My initial thoughts were to simply increase $eps (in t/13constants.t) from 0 to 1e-33, but there's probably a more open and honorable way to deal with it ;-) I reported the issue regarding M_SQRT2q a while back at https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83800. The gcc developers seemed  to take a somewhat hazy view as to whether it should be considered a bug, but it seems that it has now been fixed - in some places, at least. Thanks for the report Cheers,Rob
Download (untitled) / with headers
text/plain 423b
Fixed in https://github.com/sisyphus/math-float128 I've allowed a tolerance of 1e-33, but only where it's currently needed - ie for M_1_PIq, M_2_PIq and M_SQRT2q. For the other constants I've stayed with zero tolerance. On the downside that means that some of them might, in future, also need to allow a little tolerance. The upside is that, this way, future testing will inform us when an anomaly arises. Cheers, Rob


This service is sponsored and maintained by Best Practical Solutions and runs on Perl.org infrastructure.

Please report any issues with rt.cpan.org to rt-cpan-admin@bestpractical.com.