The following Benchmarked code snipped shows a 10-15% performance increase by simply wrapping the log calls. Note that it is just a scalar being logged (suppressed, actually), not even an array, an especially not a super-long array. I don't find it to be particularly bloated either:
Read more... (2 kB)

 $ perl  l4p_bench.pl
 x86_64-linux
                          Rate        fib_log_buffer fib_log_buffer_is
+_chk
 fib_log_buffer        15605/s                    --                  
+-11%
 fib_log_buffer_is_chk 17454/s                   12%                  
+  --
[download]

--
No matter how great and destructive your problems may seem now, remember, you've probably only seen the tip of them. [1]

Then add this to your test:

sub fib_no_log {
  my @fibonacci = (0, 1);
  my ($n, $sum) = (1, 0);
  while ($n < 1_000_000) {
    $n = $fibonacci[$#fibonacci] + $fibonacci[ $#fibonacci - 1 ];
    push @fibonacci, $n;
    $sum += $n if (($n % 2) == 0);
  }
  return $sum;
}

...

    'fib_no_log' => sub { fib_no_log() },
[download]

and you will get this result:

                        Rate fib_log_buffer fib_log_buffer_is_chk    f
+ib_no_log
fib_log_buffer        11712/s             --                  -13%    
+      -68%
fib_log_buffer_is_chk 13525/s            15%                    --    
+      -63%
fib_no_log            36540/s           212%                  170%    
+        --
[download]

Why? Because even the call to is_debug takes twice as long as the rest, the "meat" of your loop. Obviously you should have a perl_critic rule to warn from use of log4perl at all. Or avoid artificial benchmarks ;-)

How would Perl::Critic find out that @super_long_array is in fact a super long array? Note that the negative impact is only when the error message is really long and takes ages to expand. In the normal case warnings or error messages would be relatively short and the impact nonexistant.

Perl::Critic is a static code analyzer and can of course, by definition, not know the runtime length of arrays.

The intention is to create Perl::Critic policy to help track down all logging calls, and then based on the output, help identify which ones are likely to have a negative impact on performance.

--
No matter how great and destructive your problems may seem now, remember, you've probably only seen the tip of them. [1]

I first wanted to reply that this would be nonsense because nobody puts big arrays into error logs (and I still stand by that). A perl_critic rule that produces 99% false positives is not a good perl_critic rule. But then it occured to me that debug-logs might have such calls more frequently.

So you got a point, but only in so far as logging level 'debug' is concerned. You might flag all calls to debug that don't have such a if-clause and you actually would have a somewhat sensible rule IMO.

No because the string will be interpolated on the way in, so you won't avoid the hit. The only way to avoid the hit is to ensure the string never gets interpolated in the first place (it's not the log call that's slow, it's the string interpolation of a huge array).

I wouldn't bother doing the check first, I just change the code to:

$logger->error(sub {"Erroneous array: @super_long_array"});
[download]

then I believe it does not get expanded until inside Log::Log4perl when it decides to log or not.

So now all you need is a Perl Critic rule for using sub {} in logging ;-)