PDL::Objects - Man Page

There are basically two reasons for subclassing ndarrays. The first is simply that you want to be able to use your own routines like

  $ndarray->something

but don't want to mess up the PDL namespace (a worthy goal, indeed!). The other is that you wish to provide special handling of some functions or more information about the data the ndarray contains. The first case you can do with

  package BAR;
  our @ISA=qw/PDL/;
  sub foo {my($this) = @_; fiddle;}

  package main;
  $x = BAR->pdl(5);
  $x->foo;

However, because a PDL object is an opaque reference to a C struct, it is not instantly possible to extend the PDL class by e.g. extra data via subclassing.

typemaps turns Perl values into the C entities needed by the C code. PDL's one uses SvPDLV to extract a pdl* from a Perl value. That knows how to deal with:

• "simple" blessed scalars that have a C pointer
• blessed hash-refs that have a PDL member that's a C pointer as above

• blessed hash-refs that have a PDL member that's a Perl code-ref. It will be called, and is expected to return an ndarray that conforms to one of cases 1-2 or 4-5 above. As of 2.094, the original hash-ref will be passed as the first argument. This means you can give a method as shown below, rather than needing to make a closure each time. Example from t/subclass.t:

    package PDL::Derived2;
    # This is a array of ones of dim 'Coeff'
    # All that is stored initially is "Coeff", the
    # PDL array is only realised when a boring PDL
    # function is called on it. One can imagine methods
    # in PDL::Derived2 doing manipulation on the Coeffs
    # rather than actualizing the data.
    our @ISA = qw/PDL/;
    sub new {
      my $class = shift;
      bless {
        Coeff=>shift,
        PDL=>\&cache,
        SomethingElse=>42,
      }, $class;
    }
    # Actualize the value (demonstrating cacheing)
    # One can imagine expiring the cache if say, Coeffs change
    sub cache {
      my $self = shift;
      return $self->{Cache} if exists $self->{Cache};
      $self->{Cache} = PDL->ones(@$self{qw(Coeff Coeff)})+2;
    }

• "simple" Perl data, either a scalar or an array-reference
• blessed hash-refs that are a Math::Complex (or subclass) object - special case of the "scalar" above

The rest of this document deals with the second case above.

To enable subclassing, make a package and bless a hash-ref into it with a PDL member. Make that package inherit from PDL, and redefine the method initialize.

  package FOO;
  our ISA = qw(PDL::Hash);
  sub initialize {
    my $class = shift;
    my $self = $class->SUPER::initialize(@_);
    $self->{creation_time} = time(); # necessary extension :-)
    $self;
  }

All PDL constructors will call initialize() to make sure that your extensions are added by all PDL constructors automatically.

Do remember that if you subclass a class that is subclassed from an ndarray, you need to call SUPER::initialize. Make sure it is callable as an instance method, e.g. by copying data from $class if ref $class is true.

You can find some simple examples of PDL subclassing in the PDL distribution test-case files. Look in t/subclass.t.

For PDL Functions where the output is created and returned, PDL will call the subclassed object's initialize method on either instance or class to create the output object. (See PDL::Indexing for a discussion on Output Auto-Creation.) This behavior is summarized as follows:

PDL will call initialize on the first input argument if it is an object, else on PDL to create the output object. In the spirit of the Perl philosophy of making Easy Things Easy, this behavior enables PDL-subclassed objects to be written without having to overload the many simple PDL functions in this category.