I used the words may and easy in "... you may end up with no (easy) way ..." quite deliberately as I have no knowledge of what platform the OP is using. Regardless, we both seem to be in agreement that an "Exit" button is a good idea.

"I prefer one that doesn't call exit"

Why do you have that preference? Do you perhaps think that exit in a Tk callback refers to CORE::exit?

Take a look at this from Tk::exit:

"If calling exit from code invoked via a Tk callback then this Tk version of exit cleans up more reliably than using the perl exit."

and a little further down

"... Tk::exit is imported by default ..."

or simply look in your copy of Tk.pm:

...
@EXPORT    = qw(Exists Ev exit MainLoop DoOneEvent tkinit);
...
[download]

As for destroy() (documented in Tk::Widget), you'd only need to use that (on your MainWindow object) if you had code after MainLoop which you needed to execute. Here's an example of the difference:

#!/usr/bin/env perl

use strict;
use warnings;

use Tk;

my $mw = MainWindow->new;

$mw->Button(-text => 'Exit', -command => sub { exit })->pack;
$mw->Button(-text => 'Destroy', -command => sub { $mw->destroy })->pac
+k;

MainLoop;

print "After MainLoop\n";
[download]

Sample runs:

$ pm_tk_exit_vs_destroy.pl  # Using 'Exit' button
$ pm_tk_exit_vs_destroy.pl  # Using 'Destroy' button
After MainLoop
$
[download]