The difference with $Data::Dumper::Useperl = 1; is related to strings that can be represented entirely without utf8. On ingest eval makes the usual perl heuristic about utf8, and gets it wrong:
#!/usr/bin/perl -w
use utf8; # so source code is utf-8 encoded
use Data::Dumper;
$data1 = 'ä ☺'; # a-umlaut, space, smiley
$data2 = 'ä '; # a-umlaut, space, space
$Data::Dumper::Useperl = 1;
$dump1 = Dumper( $data1 );
print $dump1;
$dump2 = Dumper( $data2 );
print $dump2;
print "\n";
$Data::Dumper::Useperl = 0;
$dump1 = Dumper( $data1 );
print $dump1;
$dump2 = Dumper( $data2 );
print $dump2;
print "\n";
Output
$VAR1 = "\x{e4} \x{263a}";
$VAR1 = 'ä ';
$VAR1 = "\x{e4} \x{263a}";
$VAR1 = "\x{e4} ";
