9 $ {$package . "::OVERLOAD"}{dummy}++; # Register with magic by touching.
10 *{$package . "::()"} = \&nil; # Make it findable via fetchmethod.
12 if ($_ eq 'fallback') {
16 if (not ref $sub and $sub !~ /::/) {
17 $ {$package . "::(" . $_} = $sub;
20 #print STDERR "Setting `$ {'package'}::\cO$_' to \\&`$sub'.\n";
21 *{$package . "::(" . $_} = \&{ $sub };
24 ${$package . "::()"} = $fb; # Make it findable too (fallback only).
28 $package = (caller())[0];
29 # *{$package . "::OVERLOAD"} = \&OVERLOAD;
31 $package->overload::OVERLOAD(@_);
35 $package = (caller())[0];
36 ${$package . "::OVERLOAD"}{dummy}++; # Upgrade the table
39 if ($_ eq 'fallback') {
40 undef $ {$package . "::()"};
42 delete $ {$package . "::"}{"(" . $_};
49 $package = ref $package if ref $package;
55 return undef unless $globref;
56 my $sub = \&{*$globref};
57 return $sub if $sub ne \&nil;
58 return shift->can($ {*$globref});
61 sub OverloadedStringify {
63 $package = ref $package if ref $package;
65 ov_method mycan($package, '(""'), $package
66 or ov_method mycan($package, '(0+'), $package
67 or ov_method mycan($package, '(bool'), $package
68 or ov_method mycan($package, '(nomethod'), $package;
73 $package = ref $package if ref $package;
74 #my $meth = $package->can('(' . shift);
75 ov_method mycan($package, '(' . shift), $package;
76 #return $meth if $meth ne \&nil;
81 my $package = ref $_[0];
82 return "$_[0]" unless $package;
83 bless $_[0], overload::Fake; # Non-overloaded package
85 bless $_[0], $package; # Back
86 $package . substr $str, index $str, '=';
90 (OverloadedStringify($_[0])) ?
95 sub mycan { # Real can would leave stubs.
96 my ($package, $meth) = @_;
97 return \*{$package . "::$meth"} if defined &{$package . "::$meth"};
99 foreach $p (@{$package . "::ISA"}) {
100 my $out = mycan($p, $meth);
114 %ops = ( with_assign => "+ - * / % ** << >> x .",
115 assign => "+= -= *= /= %= **= <<= >>= x= .=",
116 str_comparison => "< <= > >= == !=",
117 '3way_comparison'=> "<=> cmp",
118 num_comparison => "lt le gt ge eq ne",
122 func => "atan2 cos sin exp abs log sqrt",
123 conversion => 'bool "" 0+',
125 dereferencing => '${} @{} %{} &{} *{}',
126 special => 'nomethod fallback =');
129 # Arguments: what, sub
132 $^H |= $constants{$_[0]} | 0x20000;
137 sub remove_constant {
138 # Arguments: what, sub
141 $^H &= ~ $constants{$_[0]};
152 overload - Package for overloading perl operations
165 $a = new SomeThing 57;
168 if (overload::Overloaded $b) {...}
170 $strval = overload::StrVal $b;
172 =head1 CAVEAT SCRIPTOR
174 Overloading of operators is a subject not to be taken lightly.
175 Neither its precise implementation, syntax, nor semantics are
176 100% endorsed by Larry Wall. So any of these may be changed
177 at some point in the future.
181 =head2 Declaration of overloaded functions
183 The compilation directive
190 declares function Number::add() for addition, and method muas() in
191 the "class" C<Number> (or one of its base classes)
192 for the assignment form C<*=> of multiplication.
194 Arguments of this directive come in (key, value) pairs. Legal values
195 are values legal inside a C<&{ ... }> call, so the name of a
196 subroutine, a reference to a subroutine, or an anonymous subroutine
197 will all work. Note that values specified as strings are
198 interpreted as methods, not subroutines. Legal keys are listed below.
200 The subroutine C<add> will be called to execute C<$a+$b> if $a
201 is a reference to an object blessed into the package C<Number>, or if $a is
202 not an object from a package with defined mathemagic addition, but $b is a
203 reference to a C<Number>. It can also be called in other situations, like
204 C<$a+=7>, or C<$a++>. See L<MAGIC AUTOGENERATION>. (Mathemagical
205 methods refer to methods triggered by an overloaded mathematical
208 Since overloading respects inheritance via the @ISA hierarchy, the
209 above declaration would also trigger overloading of C<+> and C<*=> in
210 all the packages which inherit from C<Number>.
212 =head2 Calling Conventions for Binary Operations
214 The functions specified in the C<use overload ...> directive are called
215 with three (in one particular case with four, see L<Last Resort>)
216 arguments. If the corresponding operation is binary, then the first
217 two arguments are the two arguments of the operation. However, due to
218 general object calling conventions, the first argument should always be
219 an object in the package, so in the situation of C<7+$a>, the
220 order of the arguments is interchanged. It probably does not matter
221 when implementing the addition method, but whether the arguments
222 are reversed is vital to the subtraction method. The method can
223 query this information by examining the third argument, which can take
224 three different values:
230 the order of arguments is as in the current operation.
234 the arguments are reversed.
238 the current operation is an assignment variant (as in
239 C<$a+=7>), but the usual function is called instead. This additional
240 information can be used to generate some optimizations. Compare
241 L<Calling Conventions for Mutators>.
245 =head2 Calling Conventions for Unary Operations
247 Unary operation are considered binary operations with the second
248 argument being C<undef>. Thus the functions that overloads C<{"++"}>
249 is called with arguments C<($a,undef,'')> when $a++ is executed.
251 =head2 Calling Conventions for Mutators
253 Two types of mutators have different calling conventions:
257 =item C<++> and C<-->
259 The routines which implement these operators are expected to actually
260 I<mutate> their arguments. So, assuming that $obj is a reference to a
263 sub incr { my $n = $ {$_[0]}; ++$n; $_[0] = bless \$n}
265 is an appropriate implementation of overloaded C<++>. Note that
267 sub incr { ++$ {$_[0]} ; shift }
269 is OK if used with preincrement and with postincrement. (In the case
270 of postincrement a copying will be performed, see L<Copy Constructor>.)
272 =item C<x=> and other assignment versions
274 There is nothing special about these methods. They may change the
275 value of their arguments, and may leave it as is. The result is going
276 to be assigned to the value in the left-hand-side if different from
279 This allows for the same method to be used as overloaded C<+=> and
280 C<+>. Note that this is I<allowed>, but not recommended, since by the
281 semantic of L<"Fallback"> Perl will call the method for C<+> anyway,
282 if C<+=> is not overloaded.
286 B<Warning.> Due to the presense of assignment versions of operations,
287 routines which may be called in assignment context may create
288 self-referential structures. Currently Perl will not free self-referential
289 structures until cycles are C<explicitly> broken. You may get problems
290 when traversing your structures too.
294 use overload '+' => sub { bless [ \$_[0], \$_[1] ] };
296 is asking for trouble, since for code C<$obj += $foo> the subroutine
297 is called as C<$obj = add($obj, $foo, undef)>, or C<$obj = [\$obj,
298 \$foo]>. If using such a subroutine is an important optimization, one
299 can overload C<+=> explicitly by a non-"optimized" version, or switch
300 to non-optimized version if C<not defined $_[2]> (see
301 L<Calling Conventions for Binary Operations>).
303 Even if no I<explicit> assignment-variants of operators are present in
304 the script, they may be generated by the optimizer. Say, C<",$obj,"> or
305 C<',' . $obj . ','> may be both optimized to
307 my $tmp = ',' . $obj; $tmp .= ',';
309 =head2 Overloadable Operations
311 The following symbols can be specified in C<use overload> directive:
315 =item * I<Arithmetic operations>
317 "+", "+=", "-", "-=", "*", "*=", "/", "/=", "%", "%=",
318 "**", "**=", "<<", "<<=", ">>", ">>=", "x", "x=", ".", ".=",
320 For these operations a substituted non-assignment variant can be called if
321 the assignment variant is not available. Methods for operations "C<+>",
322 "C<->", "C<+=>", and "C<-=>" can be called to automatically generate
323 increment and decrement methods. The operation "C<->" can be used to
324 autogenerate missing methods for unary minus or C<abs>.
326 See L<"MAGIC AUTOGENERATION">, L<"Calling Conventions for Mutators"> and
327 L<"Calling Conventions for Binary Operations">) for details of these
330 =item * I<Comparison operations>
332 "<", "<=", ">", ">=", "==", "!=", "<=>",
333 "lt", "le", "gt", "ge", "eq", "ne", "cmp",
335 If the corresponding "spaceship" variant is available, it can be
336 used to substitute for the missing operation. During C<sort>ing
337 arrays, C<cmp> is used to compare values subject to C<use overload>.
339 =item * I<Bit operations>
341 "&", "^", "|", "neg", "!", "~",
343 "C<neg>" stands for unary minus. If the method for C<neg> is not
344 specified, it can be autogenerated using the method for
345 subtraction. If the method for "C<!>" is not specified, it can be
346 autogenerated using the methods for "C<bool>", or "C<\"\">", or "C<0+>".
348 =item * I<Increment and decrement>
352 If undefined, addition and subtraction methods can be
353 used instead. These operations are called both in prefix and
356 =item * I<Transcendental functions>
358 "atan2", "cos", "sin", "exp", "abs", "log", "sqrt",
360 If C<abs> is unavailable, it can be autogenerated using methods
361 for "E<lt>" or "E<lt>=E<gt>" combined with either unary minus or subtraction.
363 =item * I<Boolean, string and numeric conversion>
365 "bool", "\"\"", "0+",
367 If one or two of these operations are not overloaded, the remaining ones can
368 be used instead. C<bool> is used in the flow control operators
369 (like C<while>) and for the ternary "C<?:>" operation. These functions can
370 return any arbitrary Perl value. If the corresponding operation for this value
371 is overloaded too, that operation will be called again with this value.
377 If not overloaded, the argument will be converted to a filehandle or
378 glob (which may require a stringification). The same overloading
379 happens both for the I<read-filehandle> syntax C<E<lt>$varE<gt>> and
380 I<globbing> syntax C<E<lt>${var}E<gt>>.
382 =item * I<Dereferencing>
384 '${}', '@{}', '%{}', '&{}', '*{}'.
386 If not overloaded, the argument will be dereferenced I<as is>, thus
387 should be of correct type. These functions should return a reference
388 of correct type, or another object with overloaded dereferencing.
392 "nomethod", "fallback", "=",
394 see L<SPECIAL SYMBOLS FOR C<use overload>>.
398 See L<"Fallback"> for an explanation of when a missing method can be
401 A computer-readable form of the above table is available in the hash
402 %overload::ops, with values being space-separated lists of names:
404 with_assign => '+ - * / % ** << >> x .',
405 assign => '+= -= *= /= %= **= <<= >>= x= .=',
406 str_comparison => '< <= > >= == !=',
407 '3way_comparison'=> '<=> cmp',
408 num_comparison => 'lt le gt ge eq ne',
412 func => 'atan2 cos sin exp abs log sqrt',
413 conversion => 'bool "" 0+',
415 dereferencing => '${} @{} %{} &{} *{}',
416 special => 'nomethod fallback ='
418 =head2 Inheritance and overloading
420 Inheritance interacts with overloading in two ways.
424 =item Strings as values of C<use overload> directive
428 use overload key => value;
430 is a string, it is interpreted as a method name.
432 =item Overloading of an operation is inherited by derived classes
434 Any class derived from an overloaded class is also overloaded. The
435 set of overloaded methods is the union of overloaded methods of all
436 the ancestors. If some method is overloaded in several ancestor, then
437 which description will be used is decided by the usual inheritance
440 If C<A> inherits from C<B> and C<C> (in this order), C<B> overloads
441 C<+> with C<\&D::plus_sub>, and C<C> overloads C<+> by C<"plus_meth">,
442 then the subroutine C<D::plus_sub> will be called to implement
443 operation C<+> for an object in package C<A>.
447 Note that since the value of the C<fallback> key is not a subroutine,
448 its inheritance is not governed by the above rules. In the current
449 implementation, the value of C<fallback> in the first overloaded
450 ancestor is used, but this is accidental and subject to change.
452 =head1 SPECIAL SYMBOLS FOR C<use overload>
454 Three keys are recognized by Perl that are not covered by the above
459 C<"nomethod"> should be followed by a reference to a function of four
460 parameters. If defined, it is called when the overloading mechanism
461 cannot find a method for some operation. The first three arguments of
462 this function coincide with the arguments for the corresponding method if
463 it were found, the fourth argument is the symbol
464 corresponding to the missing method. If several methods are tried,
465 the last one is used. Say, C<1-$a> can be equivalent to
467 &nomethodMethod($a,1,1,"-")
469 if the pair C<"nomethod" =E<gt> "nomethodMethod"> was specified in the
470 C<use overload> directive.
472 If some operation cannot be resolved, and there is no function
473 assigned to C<"nomethod">, then an exception will be raised via die()--
474 unless C<"fallback"> was specified as a key in C<use overload> directive.
478 The key C<"fallback"> governs what to do if a method for a particular
479 operation is not found. Three different cases are possible depending on
480 the value of C<"fallback">:
487 substituted method (see L<MAGIC AUTOGENERATION>). If this fails, it
488 then tries to calls C<"nomethod"> value; if missing, an exception
493 The same as for the C<undef> value, but no exception is raised. Instead,
494 it silently reverts to what it would have done were there no C<use overload>
497 =item * defined, but FALSE
499 No autogeneration is tried. Perl tries to call
500 C<"nomethod"> value, and if this is missing, raises an exception.
504 B<Note.> C<"fallback"> inheritance via @ISA is not carved in stone
505 yet, see L<"Inheritance and overloading">.
507 =head2 Copy Constructor
509 The value for C<"="> is a reference to a function with three
510 arguments, i.e., it looks like the other values in C<use
511 overload>. However, it does not overload the Perl assignment
512 operator. This would go against Camel hair.
514 This operation is called in the situations when a mutator is applied
515 to a reference that shares its object with some other reference, such
521 To make this change $a and not change $b, a copy of C<$$a> is made,
522 and $a is assigned a reference to this new object. This operation is
523 done during execution of the C<++$a>, and not during the assignment,
524 (so before the increment C<$$a> coincides with C<$$b>). This is only
525 done if C<++> is expressed via a method for C<'++'> or C<'+='> (or
526 C<nomethod>). Note that if this operation is expressed via C<'+'>
527 a nonmutator, i.e., as in
532 then C<$a> does not reference a new copy of C<$$a>, since $$a does not
533 appear as lvalue when the above code is executed.
535 If the copy constructor is required during the execution of some mutator,
536 but a method for C<'='> was not specified, it can be autogenerated as a
537 string copy if the object is a plain scalar.
543 The actually executed code for
546 Something else which does not modify $a or $b....
552 Something else which does not modify $a or $b....
553 $a = $a->clone(undef,"");
556 if $b was mathemagical, and C<'++'> was overloaded with C<\&incr>,
557 C<'='> was overloaded with C<\&clone>.
561 Same behaviour is triggered by C<$b = $a++>, which is consider a synonym for
564 =head1 MAGIC AUTOGENERATION
566 If a method for an operation is not found, and the value for C<"fallback"> is
567 TRUE or undefined, Perl tries to autogenerate a substitute method for
568 the missing operation based on the defined operations. Autogenerated method
569 substitutions are possible for the following operations:
573 =item I<Assignment forms of arithmetic operations>
575 C<$a+=$b> can use the method for C<"+"> if the method for C<"+=">
578 =item I<Conversion operations>
580 String, numeric, and boolean conversion are calculated in terms of one
581 another if not all of them are defined.
583 =item I<Increment and decrement>
585 The C<++$a> operation can be expressed in terms of C<$a+=1> or C<$a+1>,
586 and C<$a--> in terms of C<$a-=1> and C<$a-1>.
590 can be expressed in terms of C<$aE<lt>0> and C<-$a> (or C<0-$a>).
594 can be expressed in terms of subtraction.
598 C<!> and C<not> can be expressed in terms of boolean conversion, or
599 string or numerical conversion.
601 =item I<Concatenation>
603 can be expressed in terms of string conversion.
605 =item I<Comparison operations>
607 can be expressed in terms of its "spaceship" counterpart: either
608 C<E<lt>=E<gt>> or C<cmp>:
610 <, >, <=, >=, ==, != in terms of <=>
611 lt, gt, le, ge, eq, ne in terms of cmp
615 <> in terms of builtin operations
617 =item I<Dereferencing>
619 ${} @{} %{} &{} *{} in terms of builtin operations
621 =item I<Copy operator>
623 can be expressed in terms of an assignment to the dereferenced value, if this
624 value is a scalar and not a reference.
628 =head1 Losing overloading
630 The restriction for the comparison operation is that even if, for example,
631 `C<cmp>' should return a blessed reference, the autogenerated `C<lt>'
632 function will produce only a standard logical value based on the
633 numerical value of the result of `C<cmp>'. In particular, a working
634 numeric conversion is needed in this case (possibly expressed in terms of
637 Similarly, C<.=> and C<x=> operators lose their mathemagical properties
638 if the string conversion substitution is applied.
640 When you chop() a mathemagical object it is promoted to a string and its
641 mathemagical properties are lost. The same can happen with other
644 =head1 Run-time Overloading
646 Since all C<use> directives are executed at compile-time, the only way to
647 change overloading during run-time is to
649 eval 'use overload "+" => \&addmethod';
653 eval 'no overload "+", "--", "<="';
655 though the use of these constructs during run-time is questionable.
657 =head1 Public functions
659 Package C<overload.pm> provides the following public functions:
663 =item overload::StrVal(arg)
665 Gives string value of C<arg> as in absence of stringify overloading.
667 =item overload::Overloaded(arg)
669 Returns true if C<arg> is subject to overloading of some operations.
671 =item overload::Method(obj,op)
673 Returns C<undef> or a reference to the method that implements C<op>.
677 =head1 Overloading constants
679 For some application Perl parser mangles constants too much. It is possible
680 to hook into this process via overload::constant() and overload::remove_constant()
683 These functions take a hash as an argument. The recognized keys of this hash
690 to overload integer constants,
694 to overload floating point constants,
698 to overload octal and hexadecimal constants,
702 to overload C<q>-quoted strings, constant pieces of C<qq>- and C<qx>-quoted
703 strings and here-documents,
707 to overload constant pieces of regular expressions.
711 The corresponding values are references to functions which take three arguments:
712 the first one is the I<initial> string form of the constant, the second one
713 is how Perl interprets this constant, the third one is how the constant is used.
714 Note that the initial string form does not
715 contain string delimiters, and has backslashes in backslash-delimiter
716 combinations stripped (thus the value of delimiter is not relevant for
717 processing of this string). The return value of this function is how this
718 constant is going to be interpreted by Perl. The third argument is undefined
719 unless for overloaded C<q>- and C<qr>- constants, it is C<q> in single-quote
720 context (comes from strings, regular expressions, and single-quote HERE
721 documents), it is C<tr> for arguments of C<tr>/C<y> operators,
722 it is C<s> for right-hand side of C<s>-operator, and it is C<qq> otherwise.
724 Since an expression C<"ab$cd,,"> is just a shortcut for C<'ab' . $cd . ',,'>,
725 it is expected that overloaded constant strings are equipped with reasonable
726 overloaded catenation operator, otherwise absurd results will result.
727 Similarly, negative numbers are considered as negations of positive constants.
729 Note that it is probably meaningless to call the functions overload::constant()
730 and overload::remove_constant() from anywhere but import() and unimport() methods.
731 From these methods they may be called as
736 die "unknown import: @_" unless @_ == 1 and $_[0] eq ':constant';
737 overload::constant integer => sub {Math::BigInt->new(shift)};
740 B<BUGS> Currently overloaded-ness of constants does not propagate
743 =head1 IMPLEMENTATION
745 What follows is subject to change RSN.
747 The table of methods for all operations is cached in magic for the
748 symbol table hash for the package. The cache is invalidated during
749 processing of C<use overload>, C<no overload>, new function
750 definitions, and changes in @ISA. However, this invalidation remains
751 unprocessed until the next C<bless>ing into the package. Hence if you
752 want to change overloading structure dynamically, you'll need an
753 additional (fake) C<bless>ing to update the table.
755 (Every SVish thing has a magic queue, and magic is an entry in that
756 queue. This is how a single variable may participate in multiple
757 forms of magic simultaneously. For instance, environment variables
758 regularly have two forms at once: their %ENV magic and their taint
759 magic. However, the magic which implements overloading is applied to
760 the stashes, which are rarely used directly, thus should not slow down
763 If an object belongs to a package using overload, it carries a special
764 flag. Thus the only speed penalty during arithmetic operations without
765 overloading is the checking of this flag.
767 In fact, if C<use overload> is not present, there is almost no overhead
768 for overloadable operations, so most programs should not suffer
769 measurable performance penalties. A considerable effort was made to
770 minimize the overhead when overload is used in some package, but the
771 arguments in question do not belong to packages using overload. When
772 in doubt, test your speed with C<use overload> and without it. So far
773 there have been no reports of substantial speed degradation if Perl is
774 compiled with optimization turned on.
776 There is no size penalty for data if overload is not used. The only
777 size penalty if overload is used in some package is that I<all> the
778 packages acquire a magic during the next C<bless>ing into the
779 package. This magic is three-words-long for packages without
780 overloading, and carries the cache table if the package is overloaded.
782 Copying (C<$a=$b>) is shallow; however, a one-level-deep copying is
783 carried out before any operation that can imply an assignment to the
784 object $a (or $b) refers to, like C<$a++>. You can override this
785 behavior by defining your own copy constructor (see L<"Copy Constructor">).
787 It is expected that arguments to methods that are not explicitly supposed
788 to be changed are constant (but this is not enforced).
790 =head1 Metaphor clash
792 One may wonder why the semantic of overloaded C<=> is so counter intuitive.
793 If it I<looks> counter intuitive to you, you are subject to a metaphor
796 Here is a Perl object metaphor:
798 I< object is a reference to blessed data>
800 and an arithmetic metaphor:
802 I< object is a thing by itself>.
804 The I<main> problem of overloading C<=> is the fact that these metaphors
805 imply different actions on the assignment C<$a = $b> if $a and $b are
806 objects. Perl-think implies that $a becomes a reference to whatever
807 $b was referencing. Arithmetic-think implies that the value of "object"
808 $a is changed to become the value of the object $b, preserving the fact
809 that $a and $b are separate entities.
811 The difference is not relevant in the absence of mutators. After
812 a Perl-way assignment an operation which mutates the data referenced by $a
813 would change the data referenced by $b too. Effectively, after
814 C<$a = $b> values of $a and $b become I<indistinguishable>.
816 On the other hand, anyone who has used algebraic notation knows the
817 expressive power of the arithmetic metaphor. Overloading works hard
818 to enable this metaphor while preserving the Perlian way as far as
819 possible. Since it is not not possible to freely mix two contradicting
820 metaphors, overloading allows the arithmetic way to write things I<as
821 far as all the mutators are called via overloaded access only>. The
822 way it is done is described in L<Copy Constructor>.
824 If some mutator methods are directly applied to the overloaded values,
825 one may need to I<explicitly unlink> other values which references the
830 $b = $a; # $b is "linked" to $a
832 $a = $a->clone; # Unlink $b from $a
835 Note that overloaded access makes this transparent:
838 $b = $a; # $b is "linked" to $a
839 $a += 4; # would unlink $b automagically
841 However, it would not make
844 $a = 4; # Now $a is a plain 4, not 'Data'
846 preserve "objectness" of $a. But Perl I<has> a way to make assignments
847 to an object do whatever you want. It is just not the overload, but
848 tie()ing interface (see L<perlfunc/tie>). Adding a FETCH() method
849 which returns the object itself, and STORE() method which changes the
850 value of the object, one can reproduce the arithmetic metaphor in its
851 completeness, at least for variables which were tie()d from the start.
853 (Note that a workaround for a bug may be needed, see L<"BUGS">.)
857 Please add examples to what follows!
859 =head2 Two-face scalars
861 Put this in F<two_face.pm> in your Perl library directory:
863 package two_face; # Scalars with separate string and
865 sub new { my $p = shift; bless [@_], $p }
866 use overload '""' => \&str, '0+' => \&num, fallback => 1;
873 my $seven = new two_face ("vii", 7);
874 printf "seven=$seven, seven=%d, eight=%d\n", $seven, $seven+1;
875 print "seven contains `i'\n" if $seven =~ /i/;
877 (The second line creates a scalar which has both a string value, and a
878 numeric value.) This prints:
880 seven=vii, seven=7, eight=8
883 =head2 Two-face references
885 Suppose you want to create an object which is accessible as both an
886 array reference, and a hash reference, similar to the builtin
887 L<array-accessible-as-a-hash|perlref/"Pseudo-hashes: Using an array as
888 a hash"> builtin Perl type. Let us make it better than the builtin
889 type, there will be no restriction that you cannot use the index 0 of
893 use overload '%{}' => \&gethash, '@{}' => sub { $ {shift()} };
901 tie %h, ref $self, $self;
905 sub TIEHASH { my $p = shift; bless \ shift, $p }
908 $fields{$_} = $i++ foreach qw{zero one two three};
910 my $self = ${shift()};
911 my $key = $fields{shift()};
912 defined $key or die "Out of band access";
913 $$self->[$key] = shift;
916 my $self = ${shift()};
917 my $key = $fields{shift()};
918 defined $key or die "Out of band access";
922 Now one can access an object using both the array and hash syntax:
924 my $bar = new two_refs 3,4,5,6;
926 $bar->{two} == 11 or die 'bad hash fetch';
928 Note several important features of this example. First of all, the
929 I<actual> type of $bar is a scalar reference, and we do not overload
930 the scalar dereference. Thus we can get the I<actual> non-overloaded
931 contents of $bar by just using C<$$bar> (what we do in functions which
932 overload dereference). Similarly, the object returned by the
933 TIEHASH() method is a scalar reference.
935 Second, we create a new tied hash each time the hash syntax is used.
936 This allows us not to worry about a possibility of a reference loop,
937 would would lead to a memory leak.
939 Both these problems can be cured. Say, if we want to overload hash
940 dereference on a reference to an object which is I<implemented> as a
941 hash itself, the only problem one has to circumvent is how to access
942 this I<actual> hash (as opposed to the I<virtual> exhibited by
943 overloaded dereference operator). Here is one possible fetching routine:
946 my ($self, $key) = (shift, shift);
947 my $class = ref $self;
948 bless $self, 'overload::dummy'; # Disable overloading of %{}
949 my $out = $self->{$key};
950 bless $self, $class; # Restore overloading
954 To move creation of the tied hash on each access, one may an extra
955 level of indirection which allows a non-circular structure of references:
958 use overload '%{}' => sub { ${shift()}->[1] },
959 '@{}' => sub { ${shift()}->[0] };
965 bless \ [$a, \%h], $p;
970 tie %h, ref $self, $self;
974 sub TIEHASH { my $p = shift; bless \ shift, $p }
977 $fields{$_} = $i++ foreach qw{zero one two three};
980 my $key = $fields{shift()};
981 defined $key or die "Out of band access";
986 my $key = $fields{shift()};
987 defined $key or die "Out of band access";
991 Now if $baz is overloaded like this, then C<$bar> is a reference to a
992 reference to the intermediate array, which keeps a reference to an
993 actual array, and the access hash. The tie()ing object for the access
994 hash is also a reference to a reference to the actual array, so
1000 There are no loops of references.
1004 Both "objects" which are blessed into the class C<two_refs1> are
1005 references to a reference to an array, thus references to a I<scalar>.
1006 Thus the accessor expression C<$$foo-E<gt>[$ind]> involves no
1007 overloaded operations.
1011 =head2 Symbolic calculator
1013 Put this in F<symbolic.pm> in your Perl library directory:
1015 package symbolic; # Primitive symbolic calculator
1016 use overload nomethod => \&wrap;
1018 sub new { shift; bless ['n', @_] }
1020 my ($obj, $other, $inv, $meth) = @_;
1021 ($obj, $other) = ($other, $obj) if $inv;
1022 bless [$meth, $obj, $other];
1025 This module is very unusual as overloaded modules go: it does not
1026 provide any usual overloaded operators, instead it provides the L<Last
1027 Resort> operator C<nomethod>. In this example the corresponding
1028 subroutine returns an object which encapsulates operations done over
1029 the objects: C<new symbolic 3> contains C<['n', 3]>, C<2 + new
1030 symbolic 3> contains C<['+', 2, ['n', 3]]>.
1032 Here is an example of the script which "calculates" the side of
1033 circumscribed octagon using the above package:
1036 my $iter = 1; # 2**($iter+2) = 8
1037 my $side = new symbolic 1;
1041 $side = (sqrt(1 + $side**2) - 1)/$side;
1045 The value of $side is
1047 ['/', ['-', ['sqrt', ['+', 1, ['**', ['n', 1], 2]],
1048 undef], 1], ['n', 1]]
1050 Note that while we obtained this value using a nice little script,
1051 there is no simple way to I<use> this value. In fact this value may
1052 be inspected in debugger (see L<perldebug>), but ony if
1053 C<bareStringify> B<O>ption is set, and not via C<p> command.
1055 If one attempts to print this value, then the overloaded operator
1056 C<""> will be called, which will call C<nomethod> operator. The
1057 result of this operator will be stringified again, but this result is
1058 again of type C<symbolic>, which will lead to an infinite loop.
1060 Add a pretty-printer method to the module F<symbolic.pm>:
1063 my ($meth, $a, $b) = @{+shift};
1064 $a = 'u' unless defined $a;
1065 $b = 'u' unless defined $b;
1066 $a = $a->pretty if ref $a;
1067 $b = $b->pretty if ref $b;
1071 Now one can finish the script by
1073 print "side = ", $side->pretty, "\n";
1075 The method C<pretty> is doing object-to-string conversion, so it
1076 is natural to overload the operator C<""> using this method. However,
1077 inside such a method it is not necessary to pretty-print the
1078 I<components> $a and $b of an object. In the above subroutine
1079 C<"[$meth $a $b]"> is a catenation of some strings and components $a
1080 and $b. If these components use overloading, the catenation operator
1081 will look for an overloaded operator C<.>, if not present, it will
1082 look for an overloaded operator C<"">. Thus it is enough to use
1084 use overload nomethod => \&wrap, '""' => \&str;
1086 my ($meth, $a, $b) = @{+shift};
1087 $a = 'u' unless defined $a;
1088 $b = 'u' unless defined $b;
1092 Now one can change the last line of the script to
1094 print "side = $side\n";
1098 side = [/ [- [sqrt [+ 1 [** [n 1 u] 2]] u] 1] [n 1 u]]
1100 and one can inspect the value in debugger using all the possible
1103 Something is is still amiss: consider the loop variable $cnt of the
1104 script. It was a number, not an object. We cannot make this value of
1105 type C<symbolic>, since then the loop will not terminate.
1107 Indeed, to terminate the cycle, the $cnt should become false.
1108 However, the operator C<bool> for checking falsity is overloaded (this
1109 time via overloaded C<"">), and returns a long string, thus any object
1110 of type C<symbolic> is true. To overcome this, we need a way to
1111 compare an object to 0. In fact, it is easier to write a numeric
1114 Here is the text of F<symbolic.pm> with such a routine added (and
1115 slightly modified str()):
1117 package symbolic; # Primitive symbolic calculator
1119 nomethod => \&wrap, '""' => \&str, '0+' => \#
1121 sub new { shift; bless ['n', @_] }
1123 my ($obj, $other, $inv, $meth) = @_;
1124 ($obj, $other) = ($other, $obj) if $inv;
1125 bless [$meth, $obj, $other];
1128 my ($meth, $a, $b) = @{+shift};
1129 $a = 'u' unless defined $a;
1136 my %subr = ( n => sub {$_[0]},
1137 sqrt => sub {sqrt $_[0]},
1138 '-' => sub {shift() - shift()},
1139 '+' => sub {shift() + shift()},
1140 '/' => sub {shift() / shift()},
1141 '*' => sub {shift() * shift()},
1142 '**' => sub {shift() ** shift()},
1145 my ($meth, $a, $b) = @{+shift};
1146 my $subr = $subr{$meth}
1147 or die "Do not know how to ($meth) in symbolic";
1148 $a = $a->num if ref $a eq __PACKAGE__;
1149 $b = $b->num if ref $b eq __PACKAGE__;
1153 All the work of numeric conversion is done in %subr and num(). Of
1154 course, %subr is not complete, it contains only operators used in the
1155 example below. Here is the extra-credit question: why do we need an
1156 explicit recursion in num()? (Answer is at the end of this section.)
1158 Use this module like this:
1161 my $iter = new symbolic 2; # 16-gon
1162 my $side = new symbolic 1;
1166 $cnt = $cnt - 1; # Mutator `--' not implemented
1167 $side = (sqrt(1 + $side**2) - 1)/$side;
1169 printf "%s=%f\n", $side, $side;
1170 printf "pi=%f\n", $side*(2**($iter+2));
1172 It prints (without so many line breaks)
1174 [/ [- [sqrt [+ 1 [** [/ [- [sqrt [+ 1 [** [n 1] 2]]] 1]
1176 [/ [- [sqrt [+ 1 [** [n 1] 2]]] 1] [n 1]]]=0.198912
1179 The above module is very primitive. It does not implement
1180 mutator methods (C<++>, C<-=> and so on), does not do deep copying
1181 (not required without mutators!), and implements only those arithmetic
1182 operations which are used in the example.
1184 To implement most arithmetic operations is easy, one should just use
1185 the tables of operations, and change the code which fills %subr to
1187 my %subr = ( 'n' => sub {$_[0]} );
1188 foreach my $op (split " ", $overload::ops{with_assign}) {
1189 $subr{$op} = $subr{"$op="} = eval "sub {shift() $op shift()}";
1191 my @bins = qw(binary 3way_comparison num_comparison str_comparison);
1192 foreach my $op (split " ", "@overload::ops{ @bins }") {
1193 $subr{$op} = eval "sub {shift() $op shift()}";
1195 foreach my $op (split " ", "@overload::ops{qw(unary func)}") {
1196 print "defining `$op'\n";
1197 $subr{$op} = eval "sub {$op shift()}";
1200 Due to L<Calling Conventions for Mutators>, we do not need anything
1201 special to make C<+=> and friends work, except filling C<+=> entry of
1202 %subr, and defining a copy constructor (needed since Perl has no
1203 way to know that the implementation of C<'+='> does not mutate
1204 the argument, compare L<Copy Constructor>).
1206 To implement a copy constructor, add C<'=' => \&cpy> to C<use overload>
1207 line, and code (this code assumes that mutators change things one level
1208 deep only, so recursive copying is not needed):
1212 bless [@$self], ref $self;
1215 To make C<++> and C<--> work, we need to implement actual mutators,
1216 either directly, or in C<nomethod>. We continue to do things inside
1217 C<nomethod>, thus add
1219 if ($meth eq '++' or $meth eq '--') {
1220 @$obj = ($meth, (bless [@$obj]), 1); # Avoid circular reference
1224 after the first line of wrap(). This is not a most effective
1225 implementation, one may consider
1227 sub inc { $_[0] = bless ['++', shift, 1]; }
1231 As a final remark, note that one can fill %subr by
1233 my %subr = ( 'n' => sub {$_[0]} );
1234 foreach my $op (split " ", $overload::ops{with_assign}) {
1235 $subr{$op} = $subr{"$op="} = eval "sub {shift() $op shift()}";
1237 my @bins = qw(binary 3way_comparison num_comparison str_comparison);
1238 foreach my $op (split " ", "@overload::ops{ @bins }") {
1239 $subr{$op} = eval "sub {shift() $op shift()}";
1241 foreach my $op (split " ", "@overload::ops{qw(unary func)}") {
1242 $subr{$op} = eval "sub {$op shift()}";
1244 $subr{'++'} = $subr{'+'};
1245 $subr{'--'} = $subr{'-'};
1247 This finishes implementation of a primitive symbolic calculator in
1248 50 lines of Perl code. Since the numeric values of subexpressions
1249 are not cached, the calculator is very slow.
1251 Here is the answer for the exercise: In the case of str(), we need no
1252 explicit recursion since the overloaded C<.>-operator will fall back
1253 to an existing overloaded operator C<"">. Overloaded arithmetic
1254 operators I<do not> fall back to numeric conversion if C<fallback> is
1255 not explicitly requested. Thus without an explicit recursion num()
1256 would convert C<['+', $a, $b]> to C<$a + $b>, which would just rebuild
1257 the argument of num().
1259 If you wonder why defaults for conversion are different for str() and
1260 num(), note how easy it was to write the symbolic calculator. This
1261 simplicity is due to an appropriate choice of defaults. One extra
1262 note: due to the explicit recursion num() is more fragile than sym():
1263 we need to explicitly check for the type of $a and $b. If components
1264 $a and $b happen to be of some related type, this may lead to problems.
1266 =head2 I<Really> symbolic calculator
1268 One may wonder why we call the above calculator symbolic. The reason
1269 is that the actual calculation of the value of expression is postponed
1270 until the value is I<used>.
1272 To see it in action, add a method
1277 @$obj->[0,1] = ('=', shift);
1280 to the package C<symbolic>. After this change one can do
1282 my $a = new symbolic 3;
1283 my $b = new symbolic 4;
1284 my $c = sqrt($a**2 + $b**2);
1286 and the numeric value of $c becomes 5. However, after calling
1288 $a->STORE(12); $b->STORE(5);
1290 the numeric value of $c becomes 13. There is no doubt now that the module
1291 symbolic provides a I<symbolic> calculator indeed.
1293 To hide the rough edges under the hood, provide a tie()d interface to the
1294 package C<symbolic> (compare with L<Metaphor clash>). Add methods
1296 sub TIESCALAR { my $pack = shift; $pack->new(@_) }
1298 sub nop { } # Around a bug
1300 (the bug is described in L<"BUGS">). One can use this new interface as
1302 tie $a, 'symbolic', 3;
1303 tie $b, 'symbolic', 4;
1304 $a->nop; $b->nop; # Around a bug
1306 my $c = sqrt($a**2 + $b**2);
1308 Now numeric value of $c is 5. After C<$a = 12; $b = 5> the numeric value
1309 of $c becomes 13. To insulate the user of the module add a method
1311 sub vars { my $p = shift; tie($_, $p), $_->nop foreach @_; }
1316 symbolic->vars($a, $b);
1317 my $c = sqrt($a**2 + $b**2);
1320 printf "c5 %s=%f\n", $c, $c;
1323 printf "c13 %s=%f\n", $c, $c;
1325 shows that the numeric value of $c follows changes to the values of $a
1330 Ilya Zakharevich E<lt>F<ilya@math.mps.ohio-state.edu>E<gt>.
1334 When Perl is run with the B<-Do> switch or its equivalent, overloading
1335 induces diagnostic messages.
1337 Using the C<m> command of Perl debugger (see L<perldebug>) one can
1338 deduce which operations are overloaded (and which ancestor triggers
1339 this overloading). Say, if C<eq> is overloaded, then the method C<(eq>
1340 is shown by debugger. The method C<()> corresponds to the C<fallback>
1341 key (in fact a presence of this method shows that this package has
1342 overloading enabled, and it is what is used by the C<Overloaded>
1343 function of module C<overload>).
1347 Because it is used for overloading, the per-package hash %OVERLOAD now
1348 has a special meaning in Perl. The symbol table is filled with names
1349 looking like line-noise.
1351 For the purpose of inheritance every overloaded package behaves as if
1352 C<fallback> is present (possibly undefined). This may create
1353 interesting effects if some package is not overloaded, but inherits
1354 from two overloaded packages.
1356 Relation between overloading and tie()ing is broken. Overloading is
1357 triggered or not basing on the I<previous> class of tie()d value.
1359 This happens because the presence of overloading is checked too early,
1360 before any tie()d access is attempted. If the FETCH()ed class of the
1361 tie()d value does not change, a simple workaround is to access the value
1362 immediately after tie()ing, so that after this call the I<previous> class
1363 coincides with the current one.
1365 B<Needed:> a way to fix this without a speed penalty.
1367 Barewords are not covered by overloaded string constants.
1369 This document is confusing. There are grammos and misleading language
1370 used in places. It would seem a total rewrite is needed.