Addendum

I said above "it should be possible to do e.g. $painting->artist, but I think that could be solved with the existing DB schema using some more advanced DBIx features".

You can add missing pseudo-relationships very easily by turning your ResultSet class into a Moo class and adding object methods. For this demonstration, add the following code to Painting.pm, Play.pm and Poem.pm (after the "do not add anything above this line" line!):

use Moo;
sub artist { $_[0]->artwork->artist }            # returns a DBIx obj
sub artist_name { $_[0]->artwork->artist->name } # returns a string
sub name { $_[0]->artwork->name }                # returns a string
[download]

... and add the following to Artist.pm and Comment.pm:

use Moo;
sub language { $_[0]->country->language->name } # string
[download]

Now in order to get "Poems not by English artists that tobyink doesn't hate", the query code remains unchanged (because the new accessors are DBIx-side, not SQL-side):

$rs = $db->resultset('Poem')->search({
    'user.name'     => 'tobyink',
    'comments.text' => { '!=' => 'bad' },   # DBIx handles this NOT
    'country.name'  => \'!= country_2.name' # literal SQL for this NOT
}, {
    prefetch => {
        artwork => [
            { artist   => 'country' },
            { comments => { user => 'country' } },
        ],                          # table aliased here to 'country_2
+'
    },
});
[download]

But the application code changes from:

  
while ( my $poem = $rs->next ) {
    say sprintf(q{tobyink doesn't hate "%s" (%s)},
        $poem->artwork->name,
        $poem->artwork->artist->name,
    );
}
[download]

... to:

  
while ( my $poem = $rs->next ) {
    say sprintf(q{tobyink doesn't hate "%s" (%s)},
        $poem->name,
        $poem->artist_name,
    );
}
[download]

You can use the new accessors throughout the application, e.g. for making a comparison similar to the one in "Comments by Users who speak the same Language as the Artist" in application code rather than SQL:

if ( $comment->language eq $play->artist->language ) { ... }
[download]

And you can add arbitrary methods to the classes, not just accessors. For example, if you are building a REST application where your DB tables and ORM classes represent resources exposed via API endpoints, you can add a custom chained deletion flow, or a standard method that builds the HTTP response object:

use Moo;

sub endpoint_data {
    return { map { $_ => $_[0]->$_ } qw/only these fields/ };
}
[download]

... so you can always call $obj->endpoint_data when you are finished processing the logic for the request.

Hope this helps!