I usually put all my cleanup code into an END block within the appropriate test file. That way, even if the test is interrupted/aborted for some reason, the cleanup code usually runs and I leave a clean environment.

This becomes a hassle if you set up stuff in one .t file that you depend on in another, but I avoid that (and since you're saying you randomize the execution order of your tests, it seems so do you).

If you really want to clean up before your tests run and not after (in which case I'd be interested to know why) you could put it into a BEGIN block of the relevant test.

(Note: this is meant as pseudocode - please ignore typos)

use strict;
use warninges;
use App::Prove;

#put set-up code here

my $oProve = App::Prove->new;
$oProve->process_args(@ARGV);
my $iExit = $oProve->run ? 0 : 1;

#put tear-down code here

#and we're done
exit($iExit);
[download]

Best, beth

Personally, I use :memory: databases with DBD::SQLite or I copy canned SQLite database files for every test. That way, I don't have to reset the database even when zeroing in on a bug by running a (set of) test repeatedly.