The London Perl and Raku Workshop takes place on 26th Oct 2024. If your company depends on Perl, please consider sponsoring and/or attending.


Class::Classless -- framework for classless OOP


  use strict;
  use Class::Classless;
  my $ob1 = $Class::Classless::ROOT->clone;
  $ob1->{'NAME'} = 'Ob1';
  $ob1->{'stuff'} = 123;
  $ob1->{'Thing'} = 789;
  my $ob2 = $ob1->clone;
  $ob2->{'NAME'} = 'Ob2';
  printf "ob1 stuff: <%s>\n", $ob1->{'stuff'};
  printf "ob2 stuff: <%s>\n", $ob2->{'stuff'};
  printf "ob1 Thing: <%s>\n", $ob1->{'Thing'};
  printf "ob2 Thing: <%s>\n", $ob2->{'Thing'};
  $ob1->{'METHODS'}{'zaz'} =  sub {
     print "Zaz! on ", $_[0]{'NAME'}, "\n";

This prints the following:

    ob1 stuff: <123>
    ob2 stuff: <123>
    ob1 Thing: <789>
    ob2 Thing: <>
    Zaz! on Ob1
    Zaz! on Ob2
       'stuff', 123, 
       'NAME', 'Ob1', 
       'Thing', 789, 
       'METHODS', { 'zaz', 'CODE(0x20068360)' }, 
       'PARENTS', [ 'ROOT' ], 
       'stuff', 123, 
       'NAME', 'Ob2', 
       'METHODS', {  }, 
       'PARENTS', [ 'Ob1' ], 


In class-based OOP frameworks, methods are applicable to objects by virtue of objects belonging to classes that either provide those methods, or inherit them from classes that do.

In classless OOP frameworks (AKA delegation-and-prototypes frameworks), what methods an object is capable of is basically an attribute of that object. That is, in Perl terms: instead of methods being entries in the symbol table of the package/class the object belongs to, they are entries in a hash table inside the object. Inheritance is implemented not by having classes inheriting from other classes (via ISA lists), but by having objects inherit from other objects (via PARENTS lists).

In class-based OOP frameworks, you get new objects by calling constructors. In a classless framework, you get new objects by copying ("cloning") an existing object -- and the new clone becomes a child (inheritor) of the original object. (Where do you get the one original object? The language provides one, which has no parents, and which contains some general purpose methods like "clone".)


Each classless object is a reference to a hash, containing:

* an entry 'PARENTS', which is a reference to a list of this node's parents. (For ROOT, this will be an empty list; for most nodes, there will be just one item in this list; with multiple parents, you get multiple inheritance.)

* An entry 'NAME', which is initialized to a unique value (like "x_11") when the object has just been created by cloning. The 'NAME' attribute is not required, and deleting it is harmless.

* An entry 'METHODS', which is a reference to a hash that maps method names (e.g., "funk") to coderefs or to constant values. When you call $foo->funk(@stuff), Class::Classless's dispatcher looks to see if there's a $foo->{'METHODS'}{'funk'}. If so, and if it's a coderef, then that coderef is called with ($foo, $callstate, @stuff) as its parameter list (See the section "What A Method Sees", below, for an explanation of this). If there's a $foo->{'METHODS'}{'funk'} and it's not a coderef, then the value is returned, possibly with automagical dereferencing. (See the section "Constant Methods", below.) But, finally, if there is no such method, Class::Classless's dispatcher looks in $foo's parent to see if there's a $foo_parent->{'METHODS'}{'funk'}, and so on up the inheritance tree. If no 'funk' method is found in $foo or any of $foo's ancestors, Class::Classless dies with an error to that effect. (But see the section on the NO_FAIL attribute, below.)

* Anything else you want to put in the hash. I provide no inherent mechanism for accessing attributes (unlike, say, Self, which can automagically treat method calls as accessors, roughly speaking), so you're down to setting with $a->{'foo'} = VAL, reading with $a->{'foo'}, and possibly testing for the attribute with an exists($a->{'foo'}). (However, do have a look at the get_i, set_i, and exists_i methods, below.)


ROOT provides various methods you might find helpful:

* $thing->clone -- makes a new object based on an existing one. The only way you get to produce new objects is to clone existing ones. Existing objects are either clones of ROOT, or clones of clones of ROOT, and so on. A newly cloned object has a copy of all its parent's attributes whose names don't match /^[A-Z]/s (i.e., that don't begin with a letter between ASCII capital A and ASCII capital Z, inclusive). The new object is then initialized with a per-session-unique name like "x_12"; its PARENT attribute is set to a list containing its one parent; and its 'METHODS' attribute is set to an empty hash. (Note that the copying of parent attributes is not a deep copy -- the parent has foo => [bar, baz], then the child will have a reference to that same list, not a copy of that list!)

(Also, if $thing->is_lineage_memoized is true, the clone will have a memoized lineage too. And note that $Class::Classless::ROOT has lineage memoization off. See the description of "$thing->memoize_lineage", below, for a description of what this all means.)

* $thing->polyclone($thing2, $thing3...) -- makes a new object based on $thing, $thing2, $thing3, etc. Attributes in $thing overrride those in $thing2, and so on. The PARENTS list will consist of $thing, $thing2, $thing3, etc., in that order. Also, if $thing->is_lineage_memoized is true, the clone will have a memoized lineage too.

* $thing->get_i('attrib') -- ("get, with inheritance"). $thing->get_i('foo') returns the value of the 'foo' attribute for $thing. If there is no $thing->{'foo'}, it looks for a 'foo' attribute in each of $thing's ancestors. Returns the first one found. If none are found, returns undef. (But note that undef could result if $thing->{'foo'} or $some_parent->{'foo'} is undef.)

* $thing->exists_i('attrib') -- ("exists, with inheritance"). $thing->exists('foo') returns true if either $thing or any of its ancestors contains a 'foo' attribute (as tested with simply exists($node->{'foo'})). Otherwise, returns false.

* $thing->put_i('attrib', VALUE) -- ("put, with inheritance"). put_i looks across $thing and its ancestors, and for the first one that contains an 'attrib' attribute, sets its value to VALUE, and then returns VALUE. If neither $thing nor any of its ancestors contain a 'attrib' attribute, this will set $thing->{'attrib'} = VALUE and return VALUE, but will warn (via carp) if $^W (warnings, usually from giving Perl a -w switch) is true.

* $thing->EXAMINE -- prints a somewhat simpleminded dump of the contents of the object. Like a cheapo version of Data::Dumper's Dump() function.

* $thing->FLATTEN -- deletes all attributes (and their values) in the object whose names do not match /^[A-Z]/s (i.e., whose names don't begin with a letter between ASCII capital A and ASCII capital Z, inclusive). You can use this if you don't need an object's data, but don't feel bold enough to destroy it, because it may have clone-children that would be orphaned (a bad thing) if this node lost its PARENT attribute, say.

* $thing->allcan('baz') -- returns the list (in order) of all 'baz' methods in $thing's ISA tree. This may be an empty list. (Note that the NO_FAIL attribute has no effect on the allcan method.) Note especially that the magic dereferencing magic for constant method values is not triggered. That is, what allcan('baz') returns is simply a list of the values of $x->{'METHODS'}{'baz'} wherever such a METHODS entry exists, for all objects in $thing's inheritance tree.

* $thing->howcan('baz') -- just like allcan, but the list returned consists of pairs of values, where each pair consists of 1) the object that provides the 'baz' method, followed by 2) the value it provides for that method. (Remember that that value may be a coderef, or it may be any kind of other reference (which will not be magically resolved as it would have been by the dispatcher -- see the section "Constant Methods", or it may be any nonreference scalar value -- including 0 or undef!) The pairs are in order. You can read this list into a hash that maps from the methods to the method-values, but of course then you lose the ordering.

* $thing->can('baz') -- if $thing is capable of the method 'baz', this returns true, otherwise it returns false. Do not try to override the can method. (Note that the NO_FAIL attribute has no effect on the can method.) Note also that this does NOT return the method's value, as it did in the first version of Class::Classless, which (like Perl's normal object system) would return the (coderef) value of the method 'baz' for the first object in $thing's tree that provided such a method.

That worked then, since all method values under the first version of Class::Classless had to be coderefs (which were, of course, true in a boolean context). However, now that a Class::Classless method have have a constant value that is false, having can() return that value would be indistinguishable from having it return any false value meant as a signal the object incapable of the method. In short, can() simply has to return either true or false now. If you need the value of the methods, use allcan() or howcan().

* $thing->VERSION -- same as $thing->get_i('VERSION'). Note that ROOT has an entry of 'VERSION' => '0.00'. Do not try to override the VERSION method.

* $thing->VERSION(version_number) -- dies if $thing->VERSION is less than version_number. Otherwise returns $thing->VERSION.

* $thing->isa($thing2) -- returns true if $thing2 is in $thing's ISA tree -- i.e., if it's an ancestor of $thing. (Also returns true if $thing2 is $thing.) Otherwise returns false. Do not try to override the isa method.

* $thing->ISA_TREE -- returns $thing's ISA tree, linearized -- i.e., the list of nodes, in order, starting with $thing (and presumably ending with $ROOT), that you would search thru for method calls on $thing, or get_i calls on $thing. Do not try to override the ISA_TREE method.

* $thing->memoize_lineage -- makes this object eligible for having its ISA_TREE cached. Normally, every method call on an object causes the routine ISA_TREE to be called, so that Class::Classless knows where to look for methods, and in what order. You can avoid this having to happen each time by causing the results of $thing->ISA_TREE to be memoized (cached); then, subsequent method calls on $thing will just use the cached linearization. This means, however, that you must not change any of $thing's ancestry (who its parents are, or any of its parents' parents, etc.), or the changes will not be noticed. (If you do want to change any such thing, unmemoize the lineage first, as below. Also remember that you will need to unmemoize the lineages of all existing clones, too.)

(The ISA_TREE cache happens to be stored in $thing->{'ISA_CACHE'}.)

$thing->memoize_lineage has no effect if memoization is already on. This always returns $thing, which makes it convenient for calling on newly cloned objects:

  $thing = $foo->clone->memoize_lineage;

Note that as described above, the normal behavior of $foo->clone is to turn on ISA_TREE memoization for any new clones of $foo if $foo has its ISA_TREE memoization turned on.

* $thing->unmemoize_lineage -- this turns off the above-mentioned ISA_TREE cache for $thing. Has no effect if lineage-memoization is already off. Like $thing->memoize_lineage, this returns $thing. Think carefully about how you use this. It's never going to be a problem if the only way you call it is as:

  $thing = $foo->clone->unmemoize_lineage;

I.e., when you want a new object whose lineage you want to be free to alter later without having to worry about caching. (And when in doubt, leave caching off.)

However, note that this is wrong:

  $thing = $foo->clone->memoize_lineage;
  push @{$thing->{'PARENTS'}}, $yorp;

...because the 'unmemoize_lineage' call on $thing will be using an already out-of-date cache of its old ISA_TREE. That is likely to be harmless, though, unless $yorp overrides the normal 'unmemoize_lineage' method. But this is better:

  $thing = $foo->clone->memoize_lineage;
  push @{$thing->{'PARENTS'}}, $yorp;

But consider this harder case:

  $thing = $foo->clone->memoize_lineage;
  $zaz = $thing->clone; # so it will have memoization 
  ...more stuff...
  push @{$thing->{'PARENTS'}}, $yorp;

Even though you correctly turned off $thing's cache at the right moment, you forgot about $zaz's cache, which was and still is out of date.

* $thing->is_lineage_memoized -- returns true iff $thing is using lineage memoization.

* $thing->DESTROY -- this is here to trap DESTROY calls that Perl makes when it's about to deallocate an object, either when the object's reference count goes to 0, or at global destruction time. Currently it's a no-op, for many annoyingly complicated reasons. Do not try to override the DESTROY method! If you don't know what DESTROY methods are for anyway, don't worry about it.


I expect that most methods (i.e., things in the $foo->{'METHODS'} hash) will be coderefs. However, if you want the value of a method to be a constant, I figure there's no point in making you say:

  $foo->{'METHODS'}{'funk'} = sub { 7 };

just so $foo->funk can return the constant value 7.

So instead, I've made it so that when you call $foo->funk, and Class::Classless finds that $foo->{'METHODS'}{'funk'} exists, or that $some_ancestor->{'METHODS'}{'funk'} exists, it takes that value and decides what to do with that value, like so:

* Unless that value (which, by the way, is free to be undef!) is a reference, then it's a constant, so return it. That means that if you set $foo->{'METHODS'}{'funk'} = 7, then $foo->funk will always return 7.

* If it's an unblessed coderef, call it with arguments as explained in the "What a Method Sees" section, below. Note that blessed coderefs (as rare at they are) are not called.

* Otherwise, it must be some sort other sort of constant to return, which happens to be a reference.

* If it's a reference of the class '_deref_array', then it's array-dereferenced before being returned. So if you wanted $foo->band_members to return a constant list ('Ad Rock', 'MCA', 'Mike D'), you can do it with: $foo->{'METHODS'}{'band_members'} = bless [ 'Ad Rock', 'MCA', 'Mike D'], '_deref_array'. When you call $foo->band_members then, Class::Classless's dispatcher will basically say: return(@{$foo->{'METHODS'}{'band_members'}});

* If it's a reference of the class '_deref_scalar', then it's scalar-dereferenced before being returned. This is not as immediately and obviously useful as the same trick with '_deref_array', but it does make possible a few tricks. First off, you can have something like:

    my $counter = 0;
    bless $counter, '_deref_scalar';
    $fee->{'METHODS'}{'counter_value'} = \$counter;
    $fye->{'METHODS'}{'counter_value'} = \$counter;
    $foe->{'METHODS'}{'counter_value'} = \$counter;

to have these all share the same value, which you'd get from going $fee->counter_value, $fye->counter_value, or $foe->counter_value.

Second off, suppose (as unlikely as it is) you actually wanted a constant value to be returned -- but the value you want returned is an unblessed coderef! If you just stuck that value in $foo->{'METHODS'}, it'd get called instead of returned as a constant. Well, you can just go:

    my $cr = sub { ...whatever... };
    $foo->{'METHODS'}{'zaz'} = bless \$cr, '_deref_scalar';

So when you call $foo->zaz, Class::Classless sees a scalar of class '_deref_scalar', and returns it, like return(${$foo->{'METHODS'}{'zaz'}}). That value is, of course, your coderef.

* And finally, if the value in $foo->{'METHODS'}{'funk'} was a reference, but was neither an unblessed coderef, nor a reference of class '_deref_array', nor of class '_deref_scalar', then it's just returned.


Under Perl's normal object system, when you call

  $foo->bar($x, @y ...)

bar's @_ will consist of

  ($foo, $x, @y ...)

So normally the first thing bar will do is something like:

  my($obj, $first, @rest) = @_;


  my $obj  = shift @_;
  my $first = shift @_;
  my @rest = @_;

However, subs called as methods by Class::Classless's dispatcher have one extra argument; $_[1] is the "callstate", an object created every time you call a Class::Classless object, and belonging to the class 'Class::Classless::CALLSTATE'. Normally all you'd ever want to do with it is say:

  $callstate->NEXT('foo', $bar, @baz)

which is equivalent to $callstate->SUPER::foo($bar, @baz) under Perl's normal object system. See the section "More on NEXT".

So, in other words, the first line of a Class::Classless method to be called as

  $foo->bar($x, @y ...)

would be

  my($obj, $callstate, $first, @rest) = @_;

or the like.


I considered making some sort of mechanism for having private attributes versus inherited attributes, but decided on just letting the user work it out with get_i, set_i, and exists_i; onto this I added the feature that attributes whose names start with a character in the ASCII range [A-Z] (as opposed to [a-z], or anything else) don't get copied by the clone method, and also aren't deleted by the FLATTEN method. That's the complete extent of the special treatment that Class::Classless accords to attributes whose names start with [A-Z].

The upshot of this is that you can have something like "class data" by just taking a generic object (i.e., one you expect to be cloned) and setting attributes in it like

    $generic->{'Interface'} = 'Tk';

then all clones of that attribute can effectively 'share' that value like so...

    # send in the clones...
    $w1 = $generic->clone;
    $w2 = $generic->clone;
    $w3 = $generic->clone;
    print $w1->get_i('Interface');  # to read it
    print $w2->get_i('Interface');  # to read it (same value)
    print $w3->get_i('Interface');  # to read it (same value)
    print $w2->put_i('Interface', 'VT320');  # to set it

and even this, if this makes any useful sense:

    print $whatever->exists_i('Interface');  # to make sure it exists

However, to repeat myself somewhat, the only reason this is shared is that clone didn't copy the 'Interface' method when it made clones of $generic, so calling get_i on any of the children so produced will find the attribute not in the children, but will fall back on finding it in $generic->{'Interface'}.

But if you go and set $w1->{'Interface'} (as opposed to using set_i), then $w1->get_i('Interface') will get you the value of $w1->{'Interface'}, not the value of $generic->{'Interface'}. In other words, you'd be overriding the value you'd still be getting at with $generic->{'Interface'}, $w2->get_i('Interface'), $w3->get_i('Interface'), or even (uninterestingly) $generic->get_i('Interface').

And in any case, you can really share data by virtue of the fact that the clone method (at least, not the default clone method) doesn't do copying of references (AKA "deep copying") -- so you can just have all the objects that you want to share data simply have a reference to a common piece of data:

    my $bar = 123;
    $w->{'foo'} = \$bar;
    # Then any clones of $w will have a reference to that value --
    #  not to copies of it!
    # Similarly:
    $w->{'zaz'} = [5,6,7];
    $w->{'quux'} = {a => 11, b => 12};


If all you want is single-inheritance, you can skip this section, since things will work as you expect: objects inherit from their parents, and so on, all the way back to a parentless object (i.e., ROOT).

As to how this works with multiple inheritance, consider first how Perl's built-in mechanism for class inheritance works: first, a depth-first search of the ISA tree, and then falling back to the class UNIVERSAL, which is the implicit root for all classes.

Class::Classless's system is different -- consider this case:

       /  \
     A      X
     |      /
     B    /
      \ /

Here, Perl's depth-first search would linearize the tree (i.e., convert it to a flat list consisting of search path) as:

    C   B   A   Y   X   Root/Universal

However, I think this is just not the right way to do things. The point of X being a child of Y is so that X can have a chance to override Y. Perl's normal depth-first search doesn't allow that in cases like this. So my rule is: search over ancestors depth-first, but never search a node until you've searched all its children (that is, children that are still ancestors of the node you've built this tree for -- any other children are irrelevant). So I linearize that list as:

    C   B   A   X   Y   Root/Universal

So X does override Y. (And Root/Universal is not a special case in the searching rule.)

Now, fatal errors may result with bizarre trees -- namely ones with cyclicity in them, such as: X's parents are A and B, A's parent is B, and B's parent is A. But in some cases Class::Classless might just try to ignore the cyclic part. So just don't make any cyclic trees, OK?


If you call $thing->zaz and there is no 'zaz' method that $thing is capable of, then normally Class::Classless with throw a fatal error. However, if $thing->get_i{'NO_FAIL'} is true, then a no-operation (like sub { return; } ) simply results.

(NO_FAIL also controls what happens if you call $thing->NEXT('zaz') and there is no NEXT 'zaz' method; if NO_FAIL is true, a no-operation results; otherwise, a fatal error results. See the section "More on NEXT", below.)

Implementationally, the way this is implemented is that when you call a method, a routine of Class::Classless's called the dispatcher looks figures out the linearization of the inheritance tree of the target object of the method call, and then, one-at-a-time, goes over the objects in the linearization, looking for an object whose METHODS hash contains an entry for the name of the method. ("Linearization" meaning simply a list of objects, in the order in which they should be searched.)

Each call also creates an object, called a "callstate" object, one of whose attributes is called "no_fail" (note lowercase), and whose value starts out being undef. If the dispatcher, while going thru the linearization and looking at the METHODS, sees an object with a defined 'NO_FAIL' attribute (note uppercase), it uses that value (the value of the first object in the list with a defined NO_FAIL attribute) to set the no_fail attribute of the callstate. If it finishes searching the list and hasn't seen an object with a METHODS entry for the method it's dispatching for, one of two things will happen: if no_fail is set to true, the dispatcher will act as if it found the method and its value was sub{return}. Otherwise, the dispatcher will die with a fatal error like:

  Can't find method foo in OBJECT_NAME or any ancestors

So, normally, the only way for the no_fail attribute of the callstate to be usefully set is for the dispatcher to have seen an object with a NO_FAIL attribute set. In other words, if you want method lookup in an object to be unfailing, set $x->{'NO_FAIL'} = 1 for it or any of its ancestors; and if you want to override that for a descendant, set its $y->{'NO_FAIL'} = 0. (Note that just for sake of sanity, the NO_FAIL of $ROOT is set to 0.)

But in the case of using callstate->NEXT call to continue a method dispatch (i.e., getting the dispatcher to pick up where it left off), you may want to control the callstate's no_fail attribute directly, regardless of the NO_FAIL attributes of any of the objects the dispatcher's seen so far. In that case, you can use the $callstate->set_no_fail_true to set no_fail to true (i.e., lookup failures from NEXTing off of this callstate don't generate fatal errors). See the section on callstates, below, for more options.


Every time you call a method on a Class::Classless object (whether normally, or via a $callstate->NEXT(...) call), a new Class::Classless::CALLSTATE object is created, and passed as $_[1] to that method. Besides this being the way I happen to implement $callstate->NEXT(methodname, arguments) (by recording the state of the dispatcher for later resumption), you can use this object to get metainformation about this method call. You can access that information like so:

* $callstate->target -- the object that was the target of the method call. Same as the $_[0] that the method sees.

* $callstate->found_name -- the name this method was called as.

* $callstate->lineage -- the list of objects representing the linearization of the target object's ISA tree. (Same as $obj->ISA_TREE.)

* $callstate->home -- the object the called method was found in.

* $callstate->sub_found -- the routine that is being called. Same as $callstate->home->{'METHODS'}{$callstate->target}.

* $callstate->found_depth -- the number representing the index in the $callstate->lineage list where this method was found. In other words, $callstate->home is ($callstate->lineage)[$callstate->found_depth].

* $callstate->set_no_fail_true -- set the no_fail attribute of this callstate to true -- meaning failure is impossible for any NEXT calls based on this call. (Obviously it's meaningless to consider failure of the current method -- it was already found, otherwise how could there be code that's accessing its callstate!) I expect this is useful for cases where you want to NEXT, but aren't sure that there is a next method in the tree. With the no_fail set, failure in the NEXT lookup will act as if it triggered a method consisting of just sub { return; }.

* $callstate->set_no_fail_false -- set the no_fail attribute of this callstate to true -- meaning failure is possible for any NEXT calls in the contituation of the current call state. I don't anticipate this being useful, but I provide it for completeness.

* $callstate->set_no_fail_undef -- set the no_fail attribute of this callstate to undef -- meaning that failure is possible, but that this value can be set by the next object in the linearization of the inheritance tree. I don't anticipate this being useful, but I provide it for completeness.

* $callstate->no_fail -- returns the value of no_fail attribute of this callstate so far. See the section "The NO_FAIL attribute", above. I don't anticipate this being useful, but I provide it for completeness.

* $callstate->via_next -- return true the current method was called via $callstate->NEXT. Otherwise returns false.

The whole callstate mechanism (used by the above methods as well as by the NEXT method) assumes you don't change the object's ISA tree (or any of the METHODS hashes in any part of the ISA tree) in the middle of the call. If you do, the information in $callstate will be out of synch with reality (since it contains the linearization as of the beginning of the call)), which is fine as long as you don't use it for anything (like NEXTing) after that point, in that call.


Calling $callstate->NEXT is the mechanism I allow for doing what Perl's built-in object system does with SUPER:: calls, and like what some object systems do with "before- and after-demons".

The basic syntax to NEXT is as follows:

  $callstate->NEXT( method_name , ...arguments... );

However, if you call it with a method_name of undef, it will use the current value of $callstate->found_name, i.e., the name the currently running method was found as. Note that this can come to de undefined in two ways -- either by the parameter list being null, as in either of:


or by being explicitly undef:

  $callstate->NEXT(undef, $foo, $bar);

In either case, the undef is interpreted as $callstate->found_name. I offer this as just a (hopefully) convenient shortcut.

Now, if you call NEXT and there is no method with the desired name in the remainder of the linearization of the inheritance tree, what happens depends on the no_fail attribute; if you want to insure that the NEXT will not fail (since failing would mean a fatal error), you can set the callstate's no_fail attribute to true:


(which means it can't fail.)

Note, by the way, that NEXTing never automatically copies the argument list of the current method for the next one. You have to do that yourself. There's many ways to do it, but consider something like:

  $x->{"METHODS"}{"foo"} = sub {
    my($o, $cs) = splice(@_,0,2);
    # then copy arguments from @_, but don't change @_ any further:
    my($zaz, @foo) = @_
    # then you can pass on the arguments still in @_
      # undef to mean 'the name I was called as'

If you forgot and just said $cs->NEXT() or (pointlessly) $cs->NEXT(undef), then the next 'foo' method would have nothing in its argument list after its usual two first items (the target object and the callstate).

A further note: currently, each method call (whether normal, or via a NEXT) creates a new callstate object. However, when NEXTing, the attributes of the current callstate object are copied into the new callstate object -- except for the via_next attribute, which is forced to true, of course.


This module does what it does by blessing all "Class::Classless" objects into a class (Class::Classless::X, in point of fact) that provides no methods except for an AUTOLOAD method that intercepts all method calls and does the dispatching. This is how I fiendishly usurp Perl's normal method dispatching scheme. (Actually I do provide other methods upfront: can, VERSION, isa, DESTROY, and ISA_TREE, as I basically have to, it turns out.)

Consult the source for details. It's not that long.


* The moral of this module is that if you don't like the object framework that comes with a language, quit your bitching and just make your own! And the meta-moral is that object systems aren't black boxes that have to be fused with the language itself.

* Note that the can you may export from UNIVERSAL has nothing at all to do with the can that you should be using for Class::Classless objects. The only way you should call can on classless objects is like $obj->can('foo').

* How to test if something is a classless object: ref($obj) eq 'Class::Classless::X'

* Don't make cyclic trees. I don't go to extreme lengths to stop you from doing so, but don't expect sane behavior if you do.

* The reason the $callstate->NEXT('foo') is called NEXT is because it starts looking in the next object in the linearization of the ISA_TREE. This next object is not necessarily an ancestor (i.e., a superior object) of the current object -- in the above section, X is A's next node, altho A is clearly not a superior node.

* Don't try to derive new classes from any of the classes that Class::Classless defines. First off, it may not work, for any reading of "work". Second off, what's the point?

* Note that there's currently no mechanism for parent objects to know what their children are. However, if you needed this, you could override the clone method with something that would track this. But note that this would create circular data structures, complicating garbage collection -- you'd have to explicitly destroy objects, the way you have to with Tree::DAG_Node nodes.

* Why don't I let objects define their own DESTROY methods? One short reason: this unpredictably and intermittently triggers a strange bug in Perl's garbage collection system during global destruction. Better, longer reason: I don't see any way to make sure that, during global destruction, Perl never destroys a parent before its children. If a parent is destroyed before its children, and that parent provides a DESTROY that the children inherit, then when it comes time for the children to be destroyed, the DESTROY method they planned on using would have become inaccessible. This seems an intractable problem.

* Callstate objects were added as an afterthought. They are meant to be small and inexpensive, not extensible. I can't imagine a use for them other than the uses outlined in the documentation -- i.e., getting at (or sometimes modifying) an attribute of the current state of the method dispatcher. If you're considering any other use of callstate objects, email me -- I'd be interested in hearing what you have in mind.

* While I was writing Class::Classless, I read up on Self. To quote FOLDOC (, Self is/was "a small, dynamically typed object-oriented language, based purely on prototypes and delegation. Self was developed by the Self Group at Sun Microsystems Laboratories, Inc. and Stanford University. It is an experimental exploratory programming language." For more information, see


To Marx, a classless society never meant the absolute equality of result, but merely the absence of artificial barriers between social groups. According to David McClellan, a Marx scholar, Marx "had a dynamic or subjective element in his definition of class; a class only existed when it was conscious of itself as such, and this always implied common hostility to another social group." In The Thought of Karl Marx, (New York: Harper & Row, 1971) p. 155.


The thanks for the quote as well as for thinking of the name "Class::Classless" go to Veblen, who can be seen making that secret potato soup of his at

Thanks to my many minions in EFNet #perl for help, suggestions, and encouragement. Especial thanks to Merlyn, Halfjack, and Skrewtape for assuring me that the idea of objects-without-class wasn't just some Felliniesque fever dream I had, but is a concept that has precedent in other programming languages.

And thanks to Damian Conway for stritching the brines of his poor students with this module.


For information on Perl's classy OOP system, see perlobj, perltoot, UNIVERSAL, and Damian Conway's excellent book Object Oriented Perl from Manning Press.


Copyright (c) 1999, 2000 Sean M. Burke. All rights reserved.

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself.


Sean M. Burke,