Langue: en

Version: 2009-05-17 (debian - 07/07/09)

Section: 3 (Bibliothèques de fonctions)


Class::XSAccessor - Generate fast XS accessors without runtime compilation


   package MyClass;
   use Class::XSAccessor
     constructor => 'new',
     getters => {
       get_foo => 'foo', # 'foo' is the hash key to access
       get_bar => 'bar',
     setters => {
       set_foo => 'foo',
       set_bar => 'bar',
     accessors => {
       foo => 'foo',
       bar => 'bar',
     predicates => {
       has_foo => 'foo',
       has_bar => 'bar',
     true => [ 'is_token', 'is_whitespace' ],
     false => [ 'significant' ];
   # The imported methods are implemented in fast XS.
   # normal class code here.


Class::XSAccessor implements fast read, write and read/write accessors in XS. Additionally, it can provide predicates such as "has_foo()" for testing whether the attribute "foo" is defined in the object. It only works with objects that are implemented as ordinary hashes. Class::XSAccessor::Array implements the same interface for objects that use arrays for their internal representation.

Since version 0.10, the module can also generate simple constructors (implemented in XS) for you. Simply supply the "constructor => 'constructor_name'" option or the "constructors => ['new', 'create', 'spawn']" option. These constructors do the equivalent of the following perl code:

   sub new {
     my $class = shift;
     return bless { @_ }, ref($class)||$class;

That means they can be called on objects and classes but will not clone objects entirely. Parameters to "new()" are added to the object.

The XS accessor methods were between 1.6 and 2.5 times faster than typical pure-perl accessors in some simple benchmarking. The lower factor applies to the potentially slightly obscure "sub set_foo_pp {$_[0]->{foo} = $_[1]}", so if you usually write clear code, a factor of two speed-up is a good estimate.

The method names may be fully qualified. In the example of the synopsis, you could have written "MyClass::get_foo" instead of "get_foo". This way, you can install methods in classes other than the current class. See also: The "class" option below.

By default, the setters return the new value that was set and the accessors (mutators) do the same. You can change this behaviour with the "chained" option, see below. The predicates obviously return a boolean.

Since version 1.01, you can generate extremely simply methods which simply return true or false (and always do so). If that seems like a really superfluous thing to you, then think of a large class hierarchy with interfaces such as PPI. This is implemented as the "true" and "false" options, see synopsis.


In addition to specifying the types and names of accessors, you can add options which modify behaviour. The options are specified as key/value pairs just as the accessor declaration. Example:
   use Class::XSAccessor
     getters => {
       get_foo => 'foo',
     replace => 1;

The list of available options is:


Set this to a true value to prevent "Class::XSAccessor" from complaining about replacing existing subroutines.


Set this to a true value to change the return value of setters and mutators (when called with an argument). If "chained" is enabled, the setters and accessors/mutators will return the object. Mutators called without an argument still return the value of the associated attribute.

As with the other options, "chained" affects all methods generated in the same "use Class::XSAccessor ..." statement.


By default, the accessors are generated in the calling class. Using the "class" option, you can explicitly specify where the methods are to be generated.


Probably wouldn't work if your objects are tied hashes. But that's a strange thing to do anyway.

Scary code exploiting strange XS features.

If you think writing an accessor in XS should be a laughably simple exercise, then please contemplate how you could instantiate a new XS accessor for a new hash key that's only known at run-time. Note that compiling C code at run-time a la Inline::C is a no go.

Threading. With version 1.00, a memory leak has been fixed that would leak a small amount of memory if you loaded "Class::XSAccessor"-based classes in a subthread that hadn't been loaded in the ``main'' thread before. If the subthread then terminated, a hash key and an int per associated method used ot be lost. Note that this mattered only if classes were only loaded in a sort of throw-away thread.

In the new implementation as of 1.00, the memory will not be released again either in the above situation. But it will be recycled when the same class or a similar class is loaded again in any thread.





Steffen Mueller, <>

Chocolateboy, <>

Copyright (C) 2008-2009 by Steffen Mueller

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself, either Perl version 5.8 or, at your option, any later version of Perl 5 you may have available.