What about having the sort function qualify the names, eg. $main::a, to avoid that issue?

There is only an issue with sort if you are not using the fully qualified package name and you are within the scope of a my declaration. (Or an our declaration from a different package.)

Of course your co-workers may still want a clearer variable name...

Of course, one can skirt the issue in version 5.6+ using prototypes because then the arguments arrive in @_ as with ordinary subroutines (though it will be slower than the package variable method):

#!/usr/bin/perl -w
use strict;

package Foo;
sub numeric1 ($$)   {
    my($a, $b) = @_;  # could use any variable names
    $a <=> $b
}

# or just
sub numeric2 ($$) {$_[0] <=> $_[1]}

package main;
my($a, $b) = (13, 42);
my @array   = (3, 2, 5, 3, 4, 1);
my @sorted1 = sort Foo::numeric1 @array;
print "@sorted1\n";
my @sorted2 = sort Foo::numeric2 @array;
print "@sorted2\n";
[download]

Gotcha. So "best practice" would be to qualify the names in a function meant to be (re)used as a sort, and wish Perl generated a warning if a sort was called from a different package. Could do that myself, if a "debug mode" was enabled (a custom "use warning" setting, perhaps). Jumping through hoops to make the function work with the caller's package defeats the point of having $a,$b in the first place. So, create a fully-reusable (member of a class, whatnot) compare function that takes 2 args in the normal way; if someone wants to use it in a sort they call it from their local sort block: sort {Foo::compare($a,$b)} @list;. If you don't write functions meant to be directly used as a sort, the problem doesn't crop up.