Edge case with trailing text unreported

With HTML::Parser v3.76.

Consider the following chunk of data:

Hello world! <span class="highlight">Isn't this wonderful</span> really?

Creating an object, such as:

# using curry and assuming this runs within a module that acts as a simple wrapper, nothing fancy.
my $p = HTML::Parser->new(
        api_version => 3, 
        start_h => [ $self->curry::add_start, 'self, tagname, attr, attrseq, text, column, line, offset, offset_end'],
        end_h   => [ $self->curry::add_end,   'self, tagname, attr, attrseq, text, column, line, offset, offset_end' ],
        marked_sections => 1,
        comment_h => [ $self->curry::add_comment, 'self, text, column, line, offset, offset_end'],
        declaration_h => [ $self->curry::add_declaration, 'self, text, column, line, offset, offset_end'],
        default_h => [ $self->curry::add_default, 'self, tagname, attr, attrseq, text, column, line, offset, offset_end'],
        text_h => [ $self->curry::add_text, 'self, text, column, line, offset, offset_end'],
        empty_element_tags => 1,
        end_document_h => [ $self->curry::end_document, 'self, skipped_text'],
);
$p->parse( $html );
sub add_text
{
    my $self = shift( @_ );
    print( "got '", $_[1], "'\n" );
}

And this would yield:

got 'Hello world! '
got 'Isn't this wonderful'

However, ' really?' is not being reported.
One has to explicitly call $p->eof to have the trailing text reported.
If this is an intended feature, then it ought to be made clear in the documentation. However, I think one should not have to call eof to get that last trailing text.

Start by reproducing the HTML::Parser v3.76 example and tracing how the parse call handles the trailing text compared with an explicit eof call. Determine whether the expected result is automatic reporting of trailing text or clarified documentation, then add coverage for the example and verify the chosen behavior.