real_double logarithm test fails due to ulp() bug

[jdemeyer]

This is a doctest error on Solaris SPARC and FreeBSD introduced by #12557:

sage -t  --long -force_lib devel/sage/sage/rings/real_double.pyx
**********************************************************************
File "/home/buildbot/build/sage/mark-1/mark_full/build/sage-5.1.beta5/devel/sage-main/sage/rings/real_double.pyx", line 1741:
    sage: all( check_error(x) for x in sxrange(0.01, 2.00, 0.01) )
Expected:
    True
Got:
    False
**********************************************************************

It turns out that the bug is actually in the ulp() method which returns a value which is a factor 2 too small. Since in all other systems (except Solaris SPARC and FreeBSD) the RDF log() method is accurate up to 0.5 ulp for these inputs, the tests for #12557 passed.

Component: numerical

Author: Jeroen Demeyer

Reviewer: Volker Braun

Merged: sage-5.1.beta6

Issue created by migration from https://trac.sagemath.org/ticket/13134

[jdemeyer]

Author: Jeroen Demeyer

[jdemeyer]

comment:2

Attachment: 13134_ulp.patch.gz

[kcrisman]

comment:4

A similar error was reported on FreeBSD here; I've asked whether this patch fixes it.

[vbraun]

comment:7

Looks good to me!

[vbraun]

Reviewer: Volker Braun

[jdemeyer]

Merged: sage-5.1.beta6