So, had a look into B::Deparse and I can confirm that it doesn't even try to handle __END__ .

As I expected it only tries to read the data from the <DATA> of the last package and always assumes that a __DATA__ line was leading.¹

That's the corresponding code chunk

        # Print __DATA__ section, if necessary
        no strict 'refs';
        my $laststash = defined $self->{'curcop'}
            ? $self->{'curcop'}->stash->NAME : $self->{'curstash'};
        if (defined *{$laststash."::DATA"}{IO}) {
            print "package $laststash;\n"
                unless $laststash eq $self->{'curstash'};
            print "__DATA__\n";
            print readline(*{$laststash."::DATA"});
        }
[download]

Theoretically it's not that difficult to decide if there was a preceding __END__ or __DATA__ , one only needs to seek back.

my $laststash = "main";
my $data=\*{$laststash."::DATA"};

$last = tell $data;
seek $data,-20,1;            # -20 is just a temporary hack for proof 
+of concept

my $endline;

$endline = readline($data) while tell $data < $last;

print $endline;
print <$data>;

#package Test;
# __DATA__
# data
# data
__END__
end
end
[download]

this will print

__END__
end
end
[download]

and after uncommenting __DATA__

__DATA__
# data
# data
__END__
end
end
[download]

The real issue is finding the real file handle, cause __END__ always belongs to *main::DATA no matter which package was used.

This issue needs some testing and even more documentation, cause I'm not 100% sure how multiple __END__ or __DATA__ sections in different files are handled.

I suppose (IIRC) that the last (current) __END__ dominates, but which filehandles are then even defined?

¹) could be called a bug!