I use Log::Log4perl for logging, and Test::More and Test::Exception for testing. I write a number of negative tests (e.g. to see if a module reacts correctly when given incorrect input), and the output from these tests can be very verbose. This means the unexpected error messages (if any) will be far outnumbered by the expected ones, thus harder to detect when reading the output.

I'm looking for a strategy on how to suppress output from negative tests (warnings, output from die etc.), while keeping the output enabled for the positive tests (where output is unexpected and might indicate a problem).

The dilemma is probably best shown with an example:

foo.pl

# foo.pl
use strict;
use warnings;
use Log::Log4perl;

my $conf = qq(
  log4perl.category.Script                 = INFO, ScreenAppender
  log4perl.appender.ScreenAppender         = Log::Log4perl::Appender::
+Screen
  log4perl.appender.ScreenAppender.layout  = PatternLayout
  log4perl.appender.ScreenAppender.layout.ConversionPattern=[%p] %d %M
+ %F:%L - %m%n
);

Log::Log4perl::init( \$conf );
my $log = Log::Log4perl::->get_logger(q(Script));

# Add two numbers, no larger than 32 each
sub add {
  my ($i, $j) = @_;
  $log->logdie("type error") unless ($i and $j and $i =~ m/^\d+/ and $
+j =~ m/^\d+/);
  $log->warn("possible overflow") unless ($i < 2**5 and $j < 2**5);
  return $i + $j;
}

# When testing: Disable all logging,
# When not testing: call add()
caller() ? Log::Log4perl->appender_thresholds_adjust(7) : print add(@A
+RGV);

1;
__END__
[download]

foo.t

# foo.t
use strict;
use warnings;

use Test::More qw/no_plan/;
use Test::Exception;

require_ok(q{foo.pl});

# ------ negative tests for add() ------
throws_ok { add(); } qr/error/i, q{Expect error when no args};
throws_ok { add('a'); } qr/error/i, q{Expect error with non-numeric in
+put};
throws_ok { add('a1'); } qr/error/i, q{Expect error with non-numeric i
+nput};
throws_ok { add(1,'b'); } qr/error/i, q{Expect error with non-numeric 
+input};
throws_ok { add(1,'b1'); } qr/error/i, q{Expect error with non-numeric
+ input};

# ------ positive tests for add() ------
cmp_ok(add(2,2), q{==}, 4, q{Expect 2+2=4});

# Note: I want a warning here:
cmp_ok(add(1000,1000), q{==}, 2000, q{Expect 1000+1000=2000});
__END__
[download]

Run

$ prove --version
TAP::Harness v3.10 and Perl v5.10.0
$ prove foo.t
foo......ok
All tests successful.
Files=1, Tests=8,  0 wallclock secs ( 0.02 usr  0.00 sys +  0.08 cusr 
+ 0.00 csys =  0.10 CPU)
Result: PASS
[download]

The above code takes an all-or-nothing approach to the logging, which results in the wanted possible overflow warning to be suppressed.

Do any of the Enlightened Monks have thoughts on how to selectively suppress output from tests?