From tchrist@jhereg.perl.com

  use Safe;
  $| = 1;

  my $DO_STRICT = 1;

  my $box = new Safe("Eval");
  $box->permit_only(qw{​:base_core});

  @​codes = (
  q{ $var->{'z'} = 12 },
  q{ defined $var->{'z'} },
  q{ $var = {} },
  q{ $var->{'y'} =~ /letter/ },
  q{ $var->{'z'} + 3 },
  q{ unlink $fred },
  q{ $var->{a}{b}{c} },
  q{ while (1) { } },
  q{ last },
  q{ return },
  q{ do 'grinch' },
  q{ require Help },
  );

  for $code (@​codes) {
  $i++;
  print "Running code #$i​: $code\n\t";
  $@​ = '';
  my $rv = $box->reval($code, $DO_STRICT);
  if (!defined $rv && $@​) {
  print("SAFE TRAP​: $@​\n");
  } else {
  if (defined $rv) {
  print qq(Retval = "$rv"\n\n);
  } else {
  print "Retval is <UNDEF>\n\n";
  }
  }
  }

Produces this output. Some of it is suspect.

  Running code #1​: $var->{'z'} = 12
  Global symbol "$var" requires explicit package name at (eval 2) line 1.
  Retval is <UNDEF>

  Running code #2​: defined $var->{'z'}
  Retval = ""

  Running code #3​: $var = {}
  SAFE TRAP​: anonymous hash trapped by operation mask at (eval 6) line 1.

  Running code #4​: $var->{'y'} =~ /letter/
  Retval = ""

  Running code #5​: $var->{'z'} + 3
  Retval = "3"

  Running code #6​: unlink $fred
  Global symbol "$fred" requires explicit package name at (eval 12) line 2.
  SAFE TRAP​: unlink trapped by operation mask at (eval 12) line 2.

  Running code #7​: $var->{a}{b}{c}
  Retval is <UNDEF>

  Running code #8​: while (1) { }
  SAFE TRAP​: loop exit trapped by operation mask at (eval 16) line 2.

  Running code #9​: last
  SAFE TRAP​: last trapped by operation mask at (eval 18) line 2.

  Running code #10​: return
  Retval is <UNDEF>

  Running code #11​: do 'grinch'
  SAFE TRAP​: do 'file' trapped by operation mask at (eval 22) line 2.

  Running code #12​: require Help
  SAFE TRAP​: require trapped by operation mask at (eval 24) line 3.

Issues​:

  Test code 1​: Why did the strict stuff leak out stderr, dang it!?
  Why did it not kill the compile?

  Test code 2​: Why did this autovivify and not trigger the
  same fault as test 3 did?
  Why is it not triggering the same complaint as
  test code 1 did?

  Test code 6​: More evil strict leakage out stderr.

  Text code 8​: I wrote it this way to show that it was not permissible,
  even though 1 while 1 is!

What do you think? This strict leakage is a real super pain.

--tom

--
  I'm sure that that could be indented more readably, but I'm scared of
  the awk parser. --Larry Wall in <6849@​jpl-devvax.JPL.NASA.GOV>