Part twelve of a series of posts about automated testing.

Take this contrived example of a class with configuration read from a file on disk:

package ToyWrapper;

use Config::Tiny;

sub new {
    my $class = shift;
    my %args = @_;
    my $config = Config::Tiny->new->read($args{config_file});

    bless {
        config => $config,
    }, $class;

sub wrap_toy {
    my $self = shift;
    my $use_bow = $self->config->{use_bow};


When it comes to testing the toy wrapper, each test will need to set up a slightly different configuration, to see how the behaviour is affected. But we’re making life very hard for ourselves - we need to somehow intercept the filesystem calls made by a third-party module, before we can even begin testing.

Instead, let your class ask for just the config object to be injected at construction time, and let some separate code take responsibility for reading files off disk. Because there’s no unusual behaviour needed in the constructor, it disappears when using Moo or Moose:

package ToyWrapper;

use Moo;

has config => (is => 'ro', required => 1);

sub wrap_toy { ... }


Constructing objects for testing now becomes trivial. The same principle applies to any sort of resource access at construction time; don’t do it, and avoid similar ideas such as init() methods. These generally point to the need for a separate class to do object instantiation.

Further reading