You are absolutely right, it would! (and I didn't think of it... Just wanted to avoid checking that log file)

I guess this is the reason why this kind of tricks should be use while coding the script, and not in production...

But I wouldn't want to show such intimate stuff as warnings to the people who surf my site...

Update 2: When Apache waits too long for a script and there is still no output, it seems to kill the subprocess... Then we enter the END block, and the warnings are printed (I guess).

Update: If you are really concerned about this, you can always fork a child, and use the pipe to pass it the warnings... The child will store them in an array or a string, and pass them back when your end of the pipe closes... Much heavier, but feasible.

BEGIN {
    pipe(READ, WRITE);
    *STDERR = *WRITE;
    if(!open(DEBUG, "-|") { # child
        close WRITE; my @w = <READ>; print @w; exit }
    
}

END { close WRITE;
      print '<hr><b>Script warnings</b>',
            '<small><pre>', <DEBUG>, '</pre></small>'
}
[download]

(This is untested)

Question: is the END block executed by the child process? If so, where goes the probable warning about the closed DEBUG handle?