I did a bit of studying (DBM-Deep-1.0016).
- write_value uses class DBM::Deep::Engine::Sector::Scalar for everything but references and undef.
- ::Scalar::_init receives the value and passes it to print_at.
- print_at expects a string of bytes. It's getting a string that contains non-bytes.
No encoding is done anywhere, as far as I've seen. Definitely a major bug. Two possible fixes:
- Have DBM::Deep::Engine::Sector::Scalar's _init encode values.
- Add another Sector type for strings with UTF8=1.
The latter should be simpler, more efficient, and allows the preservation of the UTF8 flag. Basically, adjust write_value and add
package DBM::Deep::Engine::Sector::Unicode;
use 5.006_000;
use strict;
use warnings FATAL => 'all';
no warnings 'recursion';
use base qw( DBM::Deep::Engine::Sector::Scalar );
sub type { $_[0]{engine}->SIG_UNICODE }
sub _init {
my $self = shift;
utf8::encode( $self->{data} )
if $] >= 5.008 && defined($self->{data});
$self->SUPER::_init();
}
sub data {
my $self = shift;
my $data = $self->SUPER::data();
utf8::decode( $data )
if $] >= 5.008;
return $data;
}
1;
__END__
And that's just for values. A separate fix is needed for the keys, I believe.
|