Due to some issues in Test::Builder, Test::Most's "die on failure" option has a curious behavior which was a right pain to work around, but basically means that we don't really die on failure. We can't because we'd lose our diagnostic information. Instead, we die when the next test is run or when the test run is finished. It's generally transparent to the end-user, but that causes a problem:
use Test::Most qw/die no_plan/;
# this triggers the die
# but these arguments are evaluated first
is factorial(2000), $some_big_assed_number;
Basically, we have to continue running until the next test. If there's expensive computation or something which changes the state of our system, the we may have to wait or have the system's state different from the state in which the failure occurred. This is really annoying and problematic. It does, however, add one tiny little benefit:
or explain $some_data;
That "or show $some_data" still triggers due to this bug, thus allowing us extra diagnostic information. It's a tiny saving grace, but a nice one nonetheless (and also lets us easily implement clean up code).
Note that Schwern is fixing the issue in Test::Builder 2.0.