Description of problem:
The Math::BigInt manual states that leading and trailing whitespace is
ignored, and in fact this has been the behaviour for many years.

However, in recent Math::BigInt (that has shipped in Linux
distributions... e.g. Fedora 25), this behaviour is broken, which breaks
our tools that rely on this.

With Math::BigInt 1.999727:

'use Math::BigInt; print Dumper(Math::BigInt->new("0x10\n\t\t\t"));';
$VAR1 = bless( {
		 'sign' => 'NaN',
		 'value' => [
			      0
			    ]
	       }, 'Math::BigInt' );

With Math::BigInt 1.9993:
'use Math::BigInt; print Dumper(Math::BigInt->new("0x10\n\t\t\t"));';
$VAR1 = bless( {
		 'sign' => '+',
		 'value' => [
			      16
			    ]
	       }, 'Math::BigInt' );

The first version to exhibit this buggy behaviour was 1.999712, with
1.999710 and 1.9997_11 both not having the bug. All versions after
1.999712, including the latest, 1.999807 exhibit the bug.

Version-Release number of selected component (if applicable):
perl-Math-BigInt-1.9997.27-1.fc25.noarch

How reproducible:
100%

Steps to Reproduce:
1. Math::BigInt->new("0x10\n\t\t\t")

Actual results:
print Dumper(Math::BigInt->new("0x10\n\t\t\t"));';
$VAR1 = bless( {
		 'sign' => 'NaN',
		 'value' => [
			      0
			    ]
	       }, 'Math::BigInt' );

Expected results:
'use Math::BigInt; print Dumper(Math::BigInt->new("0x10\n\t\t\t"));';
$VAR1 = bless( {
		 'sign' => '+',
		 'value' => [
			      16
			    ]
	       }, 'Math::BigInt' );

Additional info:

I've reported this upstream, so I'm largely filing this as a Fedora bug so that it's tracked for Fedora - as this means you cannot compile OpenPOWER firmware on Fedora 25.