Perl-Style Accessors Considered Harmful

(a quick break-out page from the 'Accessors' section of CodingStyle).

We don't use traditional perl-style variable accessor methods very frequently (ie.

  sub foo {
    my ($self, $val) = @_;
    if (defined $val) {
      $self->{foo} = $val; 
    } else {
      return $val;
    }
  }

Instead, the more wordy Java/C++ style is preferred:

  sub get_foo {
    my ($self) = @_;
    return $val;
  }
  sub set_foo {
    my ($self, $val) = @_;
    $self->{foo} = $val;
  }

The perl style is considered a bad idea, because it can become a no-op, if the value being passed in is 'undef'. Here's how:

  • Let's say you have a perl-style accessor $self->foo(), which is used to access the value $self->{foo}.
  • $self->{foo} is currently eq 'bar'.
  • A caller comes along with a variable $, and wants to set the foo value to whatever's in $. They therefore call $self->foo($_).
  • The bug: if $_ is undef, that means that $self->foo(undef) is called. In a perl-style accessor, that is considered a 'get' operation instead of a 'set', so after that call, $self->{foo} is still set to 'bar'.

In other words, it's impossible to use a perl-style accessor to set a value to 'undef', and it's easy to accidentally perform a no-op instead of a set. This has bitten us in the past.

In the Java-style accessor, the source code itself mandates whether the operation is a set or a get; the data cannot affect which operation happens. Hence, it's safer.

  • No labels