> Why did the first example generate the warnings?

Wrong question. Since sort is a built-in, it was able to silence the warnings!

> What is the recommended way to silence them?

Use them twice?

> Why is this not mentioned in List::Util?

Maybe because the author is frustrated about not beeing able to simulate a built-in?

Seriously ...

... it's not that easy to design functional constructs with special variables in code-blocks like sort does.

I tried it in hgrep and then I was confronted with many scoping problems, what if the block is defined within another package or if $a and $b are declared as lexicals?

Look into the code of Hash::MostUtils to have an idea about the complications to solve all of this.

I bet that List::Util couldn't solve all of this.

UPDATE

To prove my last guess:

This prints 55, but after uncommenting the second line its only 0 ...oops!

use List::Util qw/reduce/;
# my ($a,$b)=(0,0);
print reduce { $b+=$a } 1..10;
[download]

UPDATE

and here the same problem for List::MoreUtils

use warnings;
use strict;

# my ($a,$b)=(0,0);

use List::MoreUtils qw/pairwise/;
my @a = (1 .. 5);
my @b = (11 .. 15);
my @x = pairwise { $a + $b } @a, @b;   # returns 12, 14, 16, 18, 20

use Data::Dump;
dd \@x;
[download]

UPDATE

At least sort simply dies

my ($a,$b)=(0,0);
print sort { $b cmp $a } 1..10;
[download]

Can't use "my $b" in sort comparison at ... line 2.
[download]

Update

use warnings;
use strict;

our ($a,$b);

package Tst;

print sort { $b cmp $a } 1..10;
[download]

Use of uninitialized value $b in string comparison (cmp) at /home/lanx
+/B/PL/PM/ScopeListUtils.pl line 9.
Use of uninitialized value $a in string comparison (cmp) at /home/lanx
+/B/PL/PM/ScopeListUtils.pl line 9.
Use of uninitialized value $b in string comparison (cmp) at /home/lanx
+/B/PL/PM/ScopeListUtils.pl line 9.
Use of uninitialized value $a in string comparison (cmp) at /home/lanx
+/B/PL/PM/ScopeListUtils.pl line 9.
...
[download]