Debian Codegraph @

upstream/debian-unstable/man3/overload.3perl

raw
1 .\" -*- mode: troff; coding: utf-8 -*-
2 .\" Automatically generated by Pod::Man 5.0102 (Pod::Simple 3.45)
3 .\"
4 .\" Standard preamble:
5 .\" ========================================================================
6 .de Sp \" Vertical space (when we can't use .PP)
7 .if t .sp .5v
8 .if n .sp
9 ..
10 .de Vb \" Begin verbatim text
11 .ft CW
12 .nf
13 .ne \\$1
14 ..
15 .de Ve \" End verbatim text
16 .ft R
17 .fi
18 ..
19 .\" \*(C` and \*(C' are quotes in nroff, nothing in troff, for use with C<>.
20 .ie n \{\
21 . ds C` ""
22 . ds C' ""
23 'br\}
24 .el\{\
25 . ds C`
26 . ds C'
27 'br\}
28 .\"
29 .\" Escape single quotes in literal strings from groff's Unicode transform.
30 .ie \n(.g .ds Aq \(aq
31 .el .ds Aq '
32 .\"
33 .\" If the F register is >0, we'll generate index entries on stderr for
34 .\" titles (.TH), headers (.SH), subsections (.SS), items (.Ip), and index
35 .\" entries marked with X<> in POD. Of course, you'll have to process the
36 .\" output yourself in some meaningful fashion.
37 .\"
38 .\" Avoid warning from groff about undefined register 'F'.
39 .de IX
40 ..
41 .nr rF 0
42 .if \n(.g .if rF .nr rF 1
43 .if (\n(rF:(\n(.g==0)) \{\
44 . if \nF \{\
45 . de IX
46 . tm Index:\\$1\t\\n%\t"\\$2"
47 ..
48 . if !\nF==2 \{\
49 . nr % 0
50 . nr F 2
51 . \}
52 . \}
53 .\}
54 .rr rF
55 .\" ========================================================================
56 .\"
57 .IX Title "overload 3perl"
58 .TH overload 3perl 2025-11-16 "perl v5.40.1" "Perl Programmers Reference Guide"
59 .\" For nroff, turn off justification. Always turn off hyphenation; it makes
60 .\" way too many mistakes in technical documents.
61 .if n .ad l
62 .nh
63 .SH NAME
64 overload \- Package for overloading Perl operations
65 .SH SYNOPSIS
66 .IX Header "SYNOPSIS"
67 .Vb 1
68 \& package SomeThing;
69 \&
70 \& use overload
71 \& \*(Aq+\*(Aq => \e&myadd,
72 \& \*(Aq\-\*(Aq => \e&mysub;
73 \& # etc
74 \& ...
75 \&
76 \& package main;
77 \& $a = SomeThing\->new( 57 );
78 \& $b = 5 + $a;
79 \& ...
80 \& if (overload::Overloaded $b) {...}
81 \& ...
82 \& $strval = overload::StrVal $b;
83 .Ve
84 .SH DESCRIPTION
85 .IX Header "DESCRIPTION"
86 This pragma allows overloading of Perl's operators for a class.
87 To overload built-in functions, see "Overriding Built-in Functions" in perlsub instead.
88 .SS Fundamentals
89 .IX Subsection "Fundamentals"
90 \fIDeclaration\fR
91 .IX Subsection "Declaration"
92 .PP
93 Arguments of the \f(CW\*(C`use overload\*(C'\fR directive are (key, value) pairs.
94 For the full set of legal keys, see "Overloadable Operations" below.
95 .PP
96 Operator implementations (the values) can be subroutines,
97 references to subroutines, or anonymous subroutines
98 \&\- in other words, anything legal inside a \f(CW\*(C`&{ ... }\*(C'\fR call.
99 Values specified as strings are interpreted as method names.
100 Thus
101 .PP
102 .Vb 5
103 \& package Number;
104 \& use overload
105 \& "\-" => "minus",
106 \& "*=" => \e&muas,
107 \& \*(Aq""\*(Aq => sub { ...; };
108 .Ve
109 .PP
110 declares that subtraction is to be implemented by method \f(CWminus()\fR
111 in the class \f(CW\*(C`Number\*(C'\fR (or one of its base classes),
112 and that the function \f(CWNumber::muas()\fR is to be used for the
113 assignment form of multiplication, \f(CW\*(C`*=\*(C'\fR.
114 It also defines an anonymous subroutine to implement stringification:
115 this is called whenever an object blessed into the package \f(CW\*(C`Number\*(C'\fR
116 is used in a string context (this subroutine might, for example,
117 return the number as a Roman numeral).
118 .PP
119 \fICalling Conventions and Magic Autogeneration\fR
120 .IX Subsection "Calling Conventions and Magic Autogeneration"
121 .PP
122 The following sample implementation of \f(CWminus()\fR (which assumes
123 that \f(CW\*(C`Number\*(C'\fR objects are simply blessed references to scalars)
124 illustrates the calling conventions:
125 .PP
126 .Vb 8
127 \& package Number;
128 \& sub minus {
129 \& my ($self, $other, $swap) = @_;
130 \& my $result = $$self \- $other; # *
131 \& $result = \-$result if $swap;
132 \& ref $result ? $result : bless \e$result;
133 \& }
134 \& # * may recurse once \- see table below
135 .Ve
136 .PP
137 Three arguments are passed to all subroutines specified in the
138 \&\f(CW\*(C`use overload\*(C'\fR directive (with exceptions \- see below, particularly
139 "nomethod").
140 .PP
141 The first of these is the operand providing the overloaded
142 operator implementation \-
143 in this case, the object whose \f(CWminus()\fR method is being called.
144 .PP
145 The second argument is the other operand, or \f(CW\*(C`undef\*(C'\fR in the
146 case of a unary operator.
147 .PP
148 The third argument is set to TRUE if (and only if) the two
149 operands have been swapped. Perl may do this to ensure that the
150 first argument (\f(CW$self\fR) is an object implementing the overloaded
151 operation, in line with general object calling conventions.
152 For example, if \f(CW$x\fR and \f(CW$y\fR are \f(CW\*(C`Number\*(C'\fRs:
153 .PP
154 .Vb 5
155 \& operation | generates a call to
156 \& ============|======================
157 \& $x \- $y | minus($x, $y, \*(Aq\*(Aq)
158 \& $x \- 7 | minus($x, 7, \*(Aq\*(Aq)
159 \& 7 \- $x | minus($x, 7, 1)
160 .Ve
161 .PP
162 Perl may also use \f(CWminus()\fR to implement other operators which
163 have not been specified in the \f(CW\*(C`use overload\*(C'\fR directive,
164 according to the rules for "Magic Autogeneration" described later.
165 For example, the \f(CW\*(C`use overload\*(C'\fR above declared no subroutine
166 for any of the operators \f(CW\*(C`\-\-\*(C'\fR, \f(CW\*(C`neg\*(C'\fR (the overload key for
167 unary minus), or \f(CW\*(C`\-=\*(C'\fR. Thus
168 .PP
169 .Vb 5
170 \& operation | generates a call to
171 \& ============|======================
172 \& \-$x | minus($x, 0, 1)
173 \& $x\-\- | minus($x, 1, undef)
174 \& $x \-= 3 | minus($x, 3, undef)
175 .Ve
176 .PP
177 Note the \f(CW\*(C`undef\*(C'\fRs:
178 where autogeneration results in the method for a standard
179 operator which does not change either of its operands, such
180 as \f(CW\*(C`\-\*(C'\fR, being used to implement an operator which changes
181 the operand ("mutators": here, \f(CW\*(C`\-\-\*(C'\fR and \f(CW\*(C`\-=\*(C'\fR),
182 Perl passes undef as the third argument.
183 This still evaluates as FALSE, consistent with the fact that
184 the operands have not been swapped, but gives the subroutine
185 a chance to alter its behaviour in these cases.
186 .PP
187 In all the above examples, \f(CWminus()\fR is required
188 only to return the result of the subtraction:
189 Perl takes care of the assignment to \f(CW$x\fR.
190 In fact, such methods should \fInot\fR modify their operands,
191 even if \f(CW\*(C`undef\*(C'\fR is passed as the third argument
192 (see "Overloadable Operations").
193 .PP
194 The same is not true of implementations of \f(CW\*(C`++\*(C'\fR and \f(CW\*(C`\-\-\*(C'\fR:
195 these are expected to modify their operand.
196 An appropriate implementation of \f(CW\*(C`\-\-\*(C'\fR might look like
197 .PP
198 .Vb 3
199 \& use overload \*(Aq\-\-\*(Aq => "decr",
200 \& # ...
201 \& sub decr { \-\-${$_[0]}; }
202 .Ve
203 .PP
204 If the "bitwise" feature is enabled (see feature), a fifth
205 TRUE argument is passed to subroutines handling \f(CW\*(C`&\*(C'\fR, \f(CW\*(C`|\*(C'\fR, \f(CW\*(C`^\*(C'\fR and \f(CW\*(C`~\*(C'\fR.
206 This indicates that the caller is expecting numeric behaviour. The fourth
207 argument will be \f(CW\*(C`undef\*(C'\fR, as that position (\f(CW$_[3]\fR) is reserved for use
208 by "nomethod".
209 .PP
210 \fIMathemagic, Mutators, and Copy Constructors\fR
211 .IX Subsection "Mathemagic, Mutators, and Copy Constructors"
212 .PP
213 The term 'mathemagic' describes the overloaded implementation
214 of mathematical operators.
215 Mathemagical operations raise an issue.
216 Consider the code:
217 .PP
218 .Vb 2
219 \& $a = $b;
220 \& \-\-$a;
221 .Ve
222 .PP
223 If \f(CW$a\fR and \f(CW$b\fR are scalars then after these statements
224 .PP
225 .Vb 1
226 \& $a == $b \- 1
227 .Ve
228 .PP
229 An object, however, is a reference to blessed data, so if
230 \&\f(CW$a\fR and \f(CW$b\fR are objects then the assignment \f(CW\*(C`$a = $b\*(C'\fR
231 copies only the reference, leaving \f(CW$a\fR and \f(CW$b\fR referring
232 to the same object data.
233 One might therefore expect the operation \f(CW\*(C`\-\-$a\*(C'\fR to decrement
234 \&\f(CW$b\fR as well as \f(CW$a\fR.
235 However, this would not be consistent with how we expect the
236 mathematical operators to work.
237 .PP
238 Perl resolves this dilemma by transparently calling a copy
239 constructor before calling a method defined to implement
240 a mutator (\f(CW\*(C`\-\-\*(C'\fR, \f(CW\*(C`+=\*(C'\fR, and so on.).
241 In the above example, when Perl reaches the decrement
242 statement, it makes a copy of the object data in \f(CW$a\fR and
243 assigns to \f(CW$a\fR a reference to the copied data.
244 Only then does it call \f(CWdecr()\fR, which alters the copied
245 data, leaving \f(CW$b\fR unchanged.
246 Thus the object metaphor is preserved as far as possible,
247 while mathemagical operations still work according to the
248 arithmetic metaphor.
249 .PP
250 Note: the preceding paragraph describes what happens when
251 Perl autogenerates the copy constructor for an object based
252 on a scalar.
253 For other cases, see "Copy Constructor".
254 .SS "Overloadable Operations"
255 .IX Subsection "Overloadable Operations"
256 The complete list of keys that can be specified in the \f(CW\*(C`use overload\*(C'\fR
257 directive are given, separated by spaces, in the values of the
258 hash \f(CW%overload::ops\fR:
259 .PP
260 .Vb 10
261 \& with_assign => \*(Aq+ \- * / % ** << >> x .\*(Aq,
262 \& assign => \*(Aq+= \-= *= /= %= **= <<= >>= x= .=\*(Aq,
263 \& num_comparison => \*(Aq< <= > >= == !=\*(Aq,
264 \& \*(Aq3way_comparison\*(Aq => \*(Aq<=> cmp\*(Aq,
265 \& str_comparison => \*(Aqlt le gt ge eq ne\*(Aq,
266 \& binary => \*(Aq& &= | |= ^ ^= &. &.= |. |.= ^. ^.=\*(Aq,
267 \& unary => \*(Aqneg ! ~ ~.\*(Aq,
268 \& mutators => \*(Aq++ \-\-\*(Aq,
269 \& func => \*(Aqatan2 cos sin exp abs log sqrt int\*(Aq,
270 \& conversion => \*(Aqbool "" 0+ qr\*(Aq,
271 \& iterators => \*(Aq<>\*(Aq,
272 \& filetest => \*(Aq\-X\*(Aq,
273 \& dereferencing => \*(Aq${} @{} %{} &{} *{}\*(Aq,
274 \& matching => \*(Aq~~\*(Aq,
275 \& special => \*(Aqnomethod fallback =\*(Aq,
276 .Ve
277 .PP
278 Most of the overloadable operators map one-to-one to these keys.
279 Exceptions, including additional overloadable operations not
280 apparent from this hash, are included in the notes which follow.
281 This list is subject to growth over time.
282 .PP
283 A warning is issued if an attempt is made to register an operator not found
284 above.
285 .IP \(bu 5
286 \&\f(CW\*(C`not\*(C'\fR
287 .Sp
288 The operator \f(CW\*(C`not\*(C'\fR is not a valid key for \f(CW\*(C`use overload\*(C'\fR.
289 However, if the operator \f(CW\*(C`!\*(C'\fR is overloaded then the same
290 implementation will be used for \f(CW\*(C`not\*(C'\fR
291 (since the two operators differ only in precedence).
292 .IP \(bu 5
293 \&\f(CW\*(C`neg\*(C'\fR
294 .Sp
295 The key \f(CW\*(C`neg\*(C'\fR is used for unary minus to disambiguate it from
296 binary \f(CW\*(C`\-\*(C'\fR.
297 .IP \(bu 5
298 \&\f(CW\*(C`++\*(C'\fR, \f(CW\*(C`\-\-\*(C'\fR
299 .Sp
300 Assuming they are to behave analogously to Perl's \f(CW\*(C`++\*(C'\fR and \f(CW\*(C`\-\-\*(C'\fR,
301 overloaded implementations of these operators are required to
302 mutate their operands.
303 .Sp
304 No distinction is made between prefix and postfix forms of the
305 increment and decrement operators: these differ only in the
306 point at which Perl calls the associated subroutine when
307 evaluating an expression.
308 .IP \(bu 5
309 \&\fIAssignments\fR
310 .Sp
311 .Vb 2
312 \& += \-= *= /= %= **= <<= >>= x= .=
313 \& &= |= ^= &.= |.= ^.=
314 .Ve
315 .Sp
316 Simple assignment is not overloadable (the \f(CW\*(Aq=\*(Aq\fR key is used
317 for the "Copy Constructor").
318 Perl does have a way to make assignments to an object do whatever
319 you want, but this involves using \fBtie()\fR, not overload \-
320 see "tie" in perlfunc and the "COOKBOOK" examples below.
321 .Sp
322 The subroutine for the assignment variant of an operator is
323 required only to return the result of the operation.
324 It is permitted to change the value of its operand
325 (this is safe because Perl calls the copy constructor first),
326 but this is optional since Perl assigns the returned value to
327 the left-hand operand anyway.
328 .Sp
329 An object that overloads an assignment operator does so only in
330 respect of assignments to that object.
331 In other words, Perl never calls the corresponding methods with
332 the third argument (the "swap" argument) set to TRUE.
333 For example, the operation
334 .Sp
335 .Vb 1
336 \& $a *= $b
337 .Ve
338 .Sp
339 cannot lead to \f(CW$b\fR's implementation of \f(CW\*(C`*=\*(C'\fR being called,
340 even if \f(CW$a\fR is a scalar.
341 (It can, however, generate a call to \f(CW$b\fR's method for \f(CW\*(C`*\*(C'\fR).
342 .IP \(bu 5
343 \&\fINon-mutators with a mutator variant\fR
344 .Sp
345 .Vb 2
346 \& + \- * / % ** << >> x .
347 \& & | ^ &. |. ^.
348 .Ve
349 .Sp
350 As described above,
351 Perl may call methods for operators like \f(CW\*(C`+\*(C'\fR and \f(CW\*(C`&\*(C'\fR in the course
352 of implementing missing operations like \f(CW\*(C`++\*(C'\fR, \f(CW\*(C`+=\*(C'\fR, and \f(CW\*(C`&=\*(C'\fR.
353 While these methods may detect this usage by testing the definedness
354 of the third argument, they should in all cases avoid changing their
355 operands.
356 This is because Perl does not call the copy constructor before
357 invoking these methods.
358 .IP \(bu 5
359 \&\f(CW\*(C`int\*(C'\fR
360 .Sp
361 Traditionally, the Perl function \f(CW\*(C`int\*(C'\fR rounds to 0
362 (see "int" in perlfunc), and so for floating-point-like types one
363 should follow the same semantic.
364 .IP \(bu 5
365 \&\fIString, numeric, boolean, and regexp conversions\fR
366 .Sp
367 .Vb 1
368 \& "" 0+ bool
369 .Ve
370 .Sp
371 These conversions are invoked according to context as necessary.
372 For example, the subroutine for \f(CW\*(Aq""\*(Aq\fR (stringify) may be used
373 where the overloaded object is passed as an argument to \f(CW\*(C`print\*(C'\fR,
374 and that for \f(CW\*(Aqbool\*(Aq\fR where it is tested in the condition of a flow
375 control statement (like \f(CW\*(C`while\*(C'\fR) or the ternary \f(CW\*(C`?:\*(C'\fR operation.
376 .Sp
377 Of course, in contexts like, for example, \f(CW\*(C`$obj + 1\*(C'\fR, Perl will
378 invoke \f(CW$obj\fR's implementation of \f(CW\*(C`+\*(C'\fR rather than (in this
379 example) converting \f(CW$obj\fR to a number using the numify method
380 \&\f(CW\*(Aq0+\*(Aq\fR (an exception to this is when no method has been provided
381 for \f(CW\*(Aq+\*(Aq\fR and "fallback" is set to TRUE).
382 .Sp
383 The subroutines for \f(CW\*(Aq""\*(Aq\fR, \f(CW\*(Aq0+\*(Aq\fR, and \f(CW\*(Aqbool\*(Aq\fR can return
384 any arbitrary Perl value.
385 If the corresponding operation for this value is overloaded too,
386 the operation will be called again with this value.
387 .Sp
388 As a special case if the overload returns the object itself then it will
389 be used directly. An overloaded conversion returning the object is
390 probably a bug, because you're likely to get something that looks like
391 \&\f(CW\*(C`YourPackage=HASH(0x8172b34)\*(C'\fR.
392 .Sp
393 .Vb 1
394 \& qr
395 .Ve
396 .Sp
397 The subroutine for \f(CW\*(Aqqr\*(Aq\fR is used wherever the object is
398 interpolated into or used as a regexp, including when it
399 appears on the RHS of a \f(CW\*(C`=~\*(C'\fR or \f(CW\*(C`!~\*(C'\fR operator.
400 .Sp
401 \&\f(CW\*(C`qr\*(C'\fR must return a compiled regexp, or a ref to a compiled regexp
402 (such as \f(CW\*(C`qr//\*(C'\fR returns), and any further overloading on the return
403 value will be ignored.
404 .IP \(bu 5
405 \&\fIIteration\fR
406 .Sp
407 If \f(CW\*(C`<>\*(C'\fR is overloaded then the same implementation is used
408 for both the \fIread-filehandle\fR syntax \f(CW\*(C`<$var>\*(C'\fR and
409 \&\fIglobbing\fR syntax \f(CW\*(C`<${var}>\*(C'\fR.
410 .IP \(bu 5
411 \&\fIFile tests\fR
412 .Sp
413 The key \f(CW\*(Aq\-X\*(Aq\fR is used to specify a subroutine to handle all the
414 filetest operators (\f(CW\*(C`\-f\*(C'\fR, \f(CW\*(C`\-x\*(C'\fR, and so on: see "\-X" in perlfunc for
415 the full list);
416 it is not possible to overload any filetest operator individually.
417 To distinguish them, the letter following the '\-' is passed as the
418 second argument (that is, in the slot that for binary operators
419 is used to pass the second operand).
420 .Sp
421 Calling an overloaded filetest operator does not affect the stat value
422 associated with the special filehandle \f(CW\*(C`_\*(C'\fR. It still refers to the
423 result of the last \f(CW\*(C`stat\*(C'\fR, \f(CW\*(C`lstat\*(C'\fR or unoverloaded filetest.
424 .Sp
425 This overload was introduced in Perl 5.12.
426 .IP \(bu 5
427 \&\fIMatching\fR
428 .Sp
429 The key \f(CW"~~"\fR allows you to override the smart matching logic used by
430 the \f(CW\*(C`~~\*(C'\fR operator and the switch construct (\f(CW\*(C`given\*(C'\fR/\f(CW\*(C`when\*(C'\fR). See
431 "Switch Statements" in perlsyn and feature.
432 .Sp
433 Unusually, the overloaded implementation of the smart match operator
434 does not get full control of the smart match behaviour.
435 In particular, in the following code:
436 .Sp
437 .Vb 2
438 \& package Foo;
439 \& use overload \*(Aq~~\*(Aq => \*(Aqmatch\*(Aq;
440 \&
441 \& my $obj = Foo\->new();
442 \& $obj ~~ [ 1,2,3 ];
443 .Ve
444 .Sp
445 the smart match does \fInot\fR invoke the method call like this:
446 .Sp
447 .Vb 1
448 \& $obj\->match([1,2,3],0);
449 .Ve
450 .Sp
451 rather, the smart match distributive rule takes precedence, so \f(CW$obj\fR is
452 smart matched against each array element in turn until a match is found,
453 so you may see between one and three of these calls instead:
454 .Sp
455 .Vb 3
456 \& $obj\->match(1,0);
457 \& $obj\->match(2,0);
458 \& $obj\->match(3,0);
459 .Ve
460 .Sp
461 Consult the match table in "Smartmatch Operator" in perlop for
462 details of when overloading is invoked.
463 .IP \(bu 5
464 \&\fIDereferencing\fR
465 .Sp
466 .Vb 1
467 \& ${} @{} %{} &{} *{}
468 .Ve
469 .Sp
470 If these operators are not explicitly overloaded then they
471 work in the normal way, yielding the underlying scalar,
472 array, or whatever stores the object data (or the appropriate
473 error message if the dereference operator doesn't match it).
474 Defining a catch-all \f(CW\*(Aqnomethod\*(Aq\fR (see below)
475 makes no difference to this as the catch-all function will
476 not be called to implement a missing dereference operator.
477 .Sp
478 If a dereference operator is overloaded then it must return a
479 \&\fIreference\fR of the appropriate type (for example, the
480 subroutine for key \f(CW\*(Aq${}\*(Aq\fR should return a reference to a
481 scalar, not a scalar), or another object which overloads the
482 operator: that is, the subroutine only determines what is
483 dereferenced and the actual dereferencing is left to Perl.
484 As a special case, if the subroutine returns the object itself
485 then it will not be called again \- avoiding infinite recursion.
486 .IP \(bu 5
487 \&\fISpecial\fR
488 .Sp
489 .Vb 1
490 \& nomethod fallback =
491 .Ve
492 .Sp
493 See "Special Keys for \f(CW\*(C`use overload\*(C'\fR".
494 .SS "Magic Autogeneration"
495 .IX Subsection "Magic Autogeneration"
496 If a method for an operation is not found then Perl tries to
497 autogenerate a substitute implementation from the operations
498 that have been defined.
499 .PP
500 Note: the behaviour described in this section can be disabled
501 by setting \f(CW\*(C`fallback\*(C'\fR to FALSE (see "fallback").
502 .PP
503 In the following tables, numbers indicate priority.
504 For example, the table below states that,
505 if no implementation for \f(CW\*(Aq!\*(Aq\fR has been defined then Perl will
506 implement it using \f(CW\*(Aqbool\*(Aq\fR (that is, by inverting the value
507 returned by the method for \f(CW\*(Aqbool\*(Aq\fR);
508 if boolean conversion is also unimplemented then Perl will
509 use \f(CW\*(Aq0+\*(Aq\fR or, failing that, \f(CW\*(Aq""\*(Aq\fR.
510 .PP
511 .Vb 10
512 \& operator | can be autogenerated from
513 \& |
514 \& | 0+ "" bool . x
515 \& =========|==========================
516 \& 0+ | 1 2
517 \& "" | 1 2
518 \& bool | 1 2
519 \& int | 1 2 3
520 \& ! | 2 3 1
521 \& qr | 2 1 3
522 \& . | 2 1 3
523 \& x | 2 1 3
524 \& .= | 3 2 4 1
525 \& x= | 3 2 4 1
526 \& <> | 2 1 3
527 \& \-X | 2 1 3
528 .Ve
529 .PP
530 Note: The iterator (\f(CW\*(Aq<>\*(Aq\fR) and file test (\f(CW\*(Aq\-X\*(Aq\fR)
531 operators work as normal: if the operand is not a blessed glob or
532 IO reference then it is converted to a string (using the method
533 for \f(CW\*(Aq""\*(Aq\fR, \f(CW\*(Aq0+\*(Aq\fR, or \f(CW\*(Aqbool\*(Aq\fR) to be interpreted as a glob
534 or filename.
535 .PP
536 .Vb 10
537 \& operator | can be autogenerated from
538 \& |
539 \& | < <=> neg \-= \-
540 \& =========|==========================
541 \& neg | 1
542 \& \-= | 1
543 \& \-\- | 1 2
544 \& abs | a1 a2 b1 b2 [*]
545 \& < | 1
546 \& <= | 1
547 \& > | 1
548 \& >= | 1
549 \& == | 1
550 \& != | 1
551 \&
552 \& * one from [a1, a2] and one from [b1, b2]
553 .Ve
554 .PP
555 Just as numeric comparisons can be autogenerated from the method
556 for \f(CW\*(Aq<=>\*(Aq\fR, string comparisons can be autogenerated from
557 that for \f(CW\*(Aqcmp\*(Aq\fR:
558 .PP
559 .Vb 3
560 \& operators | can be autogenerated from
561 \& ====================|===========================
562 \& lt gt le ge eq ne | cmp
563 .Ve
564 .PP
565 Similarly, autogeneration for keys \f(CW\*(Aq+=\*(Aq\fR and \f(CW\*(Aq++\*(Aq\fR is analogous
566 to \f(CW\*(Aq\-=\*(Aq\fR and \f(CW\*(Aq\-\-\*(Aq\fR above:
567 .PP
568 .Vb 6
569 \& operator | can be autogenerated from
570 \& |
571 \& | += +
572 \& =========|==========================
573 \& += | 1
574 \& ++ | 1 2
575 .Ve
576 .PP
577 And other assignment variations are analogous to
578 \&\f(CW\*(Aq+=\*(Aq\fR and \f(CW\*(Aq\-=\*(Aq\fR (and similar to \f(CW\*(Aq.=\*(Aq\fR and \f(CW\*(Aqx=\*(Aq\fR above):
579 .PP
580 .Vb 3
581 \& operator || *= /= %= **= <<= >>= &= ^= |= &.= ^.= |.=
582 \& \-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-||\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-
583 \& autogenerated from || * / % ** << >> & ^ | &. ^. |.
584 .Ve
585 .PP
586 Note also that the copy constructor (key \f(CW\*(Aq=\*(Aq\fR) may be
587 autogenerated, but only for objects based on scalars.
588 See "Copy Constructor".
589 .PP
590 \fIMinimal Set of Overloaded Operations\fR
591 .IX Subsection "Minimal Set of Overloaded Operations"
592 .PP
593 Since some operations can be automatically generated from others, there is
594 a minimal set of operations that need to be overloaded in order to have
595 the complete set of overloaded operations at one's disposal.
596 Of course, the autogenerated operations may not do exactly what the user
597 expects. The minimal set is:
598 .PP
599 .Vb 6
600 \& + \- * / % ** << >> x
601 \& <=> cmp
602 \& & | ^ ~ &. |. ^. ~.
603 \& atan2 cos sin exp log sqrt int
604 \& "" 0+ bool
605 \& ~~
606 .Ve
607 .PP
608 Of the conversions, only one of string, boolean or numeric is
609 needed because each can be generated from either of the other two.
610 .ie n .SS "Special Keys for ""use overload"""
611 .el .SS "Special Keys for \f(CWuse overload\fP"
612 .IX Subsection "Special Keys for use overload"
613 \fR\f(CI\*(C`nomethod\*(C'\fR\fI\fR
614 .IX Subsection "nomethod"
615 .PP
616 The \f(CW\*(Aqnomethod\*(Aq\fR key is used to specify a catch-all function to
617 be called for any operator that is not individually overloaded.
618 The specified function will be passed four parameters.
619 The first three arguments coincide with those that would have been
620 passed to the corresponding method if it had been defined.
621 The fourth argument is the \f(CW\*(C`use overload\*(C'\fR key for that missing
622 method. If the "bitwise" feature is enabled (see feature),
623 a fifth TRUE argument is passed to subroutines handling \f(CW\*(C`&\*(C'\fR, \f(CW\*(C`|\*(C'\fR, \f(CW\*(C`^\*(C'\fR and \f(CW\*(C`~\*(C'\fR to indicate that the caller is expecting numeric behaviour.
624 .PP
625 For example, if \f(CW$a\fR is an object blessed into a package declaring
626 .PP
627 .Vb 1
628 \& use overload \*(Aqnomethod\*(Aq => \*(Aqcatch_all\*(Aq, # ...
629 .Ve
630 .PP
631 then the operation
632 .PP
633 .Vb 1
634 \& 3 + $a
635 .Ve
636 .PP
637 could (unless a method is specifically declared for the key
638 \&\f(CW\*(Aq+\*(Aq\fR) result in a call
639 .PP
640 .Vb 1
641 \& catch_all($a, 3, 1, \*(Aq+\*(Aq)
642 .Ve
643 .PP
644 See "How Perl Chooses an Operator Implementation".
645 .PP
646 \fR\f(CI\*(C`fallback\*(C'\fR\fI\fR
647 .IX Subsection "fallback"
648 .PP
649 The value assigned to the key \f(CW\*(Aqfallback\*(Aq\fR tells Perl how hard
650 it should try to find an alternative way to implement a missing
651 operator.
652 .IP \(bu 4
653 defined, but FALSE
654 .Sp
655 .Vb 1
656 \& use overload "fallback" => 0, # ... ;
657 .Ve
658 .Sp
659 This disables "Magic Autogeneration".
660 .IP \(bu 4
661 \&\f(CW\*(C`undef\*(C'\fR
662 .Sp
663 In the default case where no value is explicitly assigned to
664 \&\f(CW\*(C`fallback\*(C'\fR, magic autogeneration is enabled.
665 .IP \(bu 4
666 TRUE
667 .Sp
668 The same as for \f(CW\*(C`undef\*(C'\fR, but if a missing operator cannot be
669 autogenerated then, instead of issuing an error message, Perl
670 is allowed to revert to what it would have done for that
671 operator if there had been no \f(CW\*(C`use overload\*(C'\fR directive.
672 .Sp
673 Note: in most cases, particularly the "Copy Constructor",
674 this is unlikely to be appropriate behaviour.
675 .PP
676 See "How Perl Chooses an Operator Implementation".
677 .PP
678 \fICopy Constructor\fR
679 .IX Subsection "Copy Constructor"
680 .PP
681 As mentioned above,
682 this operation is called when a mutator is applied to a reference
683 that shares its object with some other reference.
684 For example, if \f(CW$b\fR is mathemagical, and \f(CW\*(Aq++\*(Aq\fR is overloaded
685 with \f(CW\*(Aqincr\*(Aq\fR, and \f(CW\*(Aq=\*(Aq\fR is overloaded with \f(CW\*(Aqclone\*(Aq\fR, then the
686 code
687 .PP
688 .Vb 3
689 \& $a = $b;
690 \& # ... (other code which does not modify $a or $b) ...
691 \& ++$b;
692 .Ve
693 .PP
694 would be executed in a manner equivalent to
695 .PP
696 .Vb 4
697 \& $a = $b;
698 \& # ...
699 \& $b = $b\->clone(undef, "");
700 \& $b\->incr(undef, "");
701 .Ve
702 .PP
703 Note:
704 .IP \(bu 4
705 The subroutine for \f(CW\*(Aq=\*(Aq\fR does not overload the Perl assignment
706 operator: it is used only to allow mutators to work as described
707 here. (See "Assignments" above.)
708 .IP \(bu 4
709 As for other operations, the subroutine implementing '=' is passed
710 three arguments, though the last two are always \f(CW\*(C`undef\*(C'\fR and \f(CW\*(Aq\*(Aq\fR.
711 .IP \(bu 4
712 The copy constructor is called only before a call to a function
713 declared to implement a mutator, for example, if \f(CW\*(C`++$b;\*(C'\fR in the
714 code above is effected via a method declared for key \f(CW\*(Aq++\*(Aq\fR
715 (or 'nomethod', passed \f(CW\*(Aq++\*(Aq\fR as the fourth argument) or, by
716 autogeneration, \f(CW\*(Aq+=\*(Aq\fR.
717 It is not called if the increment operation is effected by a call
718 to the method for \f(CW\*(Aq+\*(Aq\fR since, in the equivalent code,
719 .Sp
720 .Vb 2
721 \& $a = $b;
722 \& $b = $b + 1;
723 .Ve
724 .Sp
725 the data referred to by \f(CW$a\fR is unchanged by the assignment to
726 \&\f(CW$b\fR of a reference to new object data.
727 .IP \(bu 4
728 The copy constructor is not called if Perl determines that it is
729 unnecessary because there is no other reference to the data being
730 modified.
731 .IP \(bu 4
732 If \f(CW\*(Aqfallback\*(Aq\fR is undefined or TRUE then a copy constructor
733 can be autogenerated, but only for objects based on scalars.
734 In other cases it needs to be defined explicitly.
735 Where an object's data is stored as, for example, an array of
736 scalars, the following might be appropriate:
737 .Sp
738 .Vb 1
739 \& use overload \*(Aq=\*(Aq => sub { bless [ @{$_[0]} ] }, # ...
740 .Ve
741 .IP \(bu 4
742 If \f(CW\*(Aqfallback\*(Aq\fR is TRUE and no copy constructor is defined then,
743 for objects not based on scalars, Perl may silently fall back on
744 simple assignment \- that is, assignment of the object reference.
745 In effect, this disables the copy constructor mechanism since
746 no new copy of the object data is created.
747 This is almost certainly not what you want.
748 (It is, however, consistent: for example, Perl's fallback for the
749 \&\f(CW\*(C`++\*(C'\fR operator is to increment the reference itself.)
750 .SS "How Perl Chooses an Operator Implementation"
751 .IX Subsection "How Perl Chooses an Operator Implementation"
752 Which is checked first, \f(CW\*(C`nomethod\*(C'\fR or \f(CW\*(C`fallback\*(C'\fR?
753 If the two operands of an operator are of different types and
754 both overload the operator, which implementation is used?
755 The following are the precedence rules:
756 .IP 1. 4
757 If the first operand has declared a subroutine to overload the
758 operator then use that implementation.
759 .IP 2. 4
760 Otherwise, if fallback is TRUE or undefined for the
761 first operand then see if the
762 rules for autogeneration
763 allows another of its operators to be used instead.
764 .IP 3. 4
765 Unless the operator is an assignment (\f(CW\*(C`+=\*(C'\fR, \f(CW\*(C`\-=\*(C'\fR, etc.),
766 repeat step (1) in respect of the second operand.
767 .IP 4. 4
768 Repeat Step (2) in respect of the second operand.
769 .IP 5. 4
770 If the first operand has a "nomethod" method then use that.
771 .IP 6. 4
772 If the second operand has a "nomethod" method then use that.
773 .IP 7. 4
774 If \f(CW\*(C`fallback\*(C'\fR is TRUE for both operands
775 then perform the usual operation for the operator,
776 treating the operands as numbers, strings, or booleans
777 as appropriate for the operator (see note).
778 .IP 8. 4
779 Nothing worked \- die.
780 .PP
781 Where there is only one operand (or only one operand with
782 overloading) the checks in respect of the other operand above are
783 skipped.
784 .PP
785 There are exceptions to the above rules for dereference operations
786 (which, if Step 1 fails, always fall back to the normal, built-in
787 implementations \- see Dereferencing), and for \f(CW\*(C`~~\*(C'\fR (which has its
788 own set of rules \- see \f(CW\*(C`Matching\*(C'\fR under "Overloadable Operations"
789 above).
790 .PP
791 Note on Step 7: some operators have a different semantic depending
792 on the type of their operands.
793 As there is no way to instruct Perl to treat the operands as, e.g.,
794 numbers instead of strings, the result here may not be what you
795 expect.
796 See "BUGS AND PITFALLS".
797 .SS "Losing Overloading"
798 .IX Subsection "Losing Overloading"
799 The restriction for the comparison operation is that even if, for example,
800 \&\f(CW\*(C`cmp\*(C'\fR should return a blessed reference, the autogenerated \f(CW\*(C`lt\*(C'\fR
801 function will produce only a standard logical value based on the
802 numerical value of the result of \f(CW\*(C`cmp\*(C'\fR. In particular, a working
803 numeric conversion is needed in this case (possibly expressed in terms of
804 other conversions).
805 .PP
806 Similarly, \f(CW\*(C`.=\*(C'\fR and \f(CW\*(C`x=\*(C'\fR operators lose their mathemagical properties
807 if the string conversion substitution is applied.
808 .PP
809 When you \fBchop()\fR a mathemagical object it is promoted to a string and its
810 mathemagical properties are lost. The same can happen with other
811 operations as well.
812 .SS "Inheritance and Overloading"
813 .IX Subsection "Inheritance and Overloading"
814 Overloading respects inheritance via the \f(CW@ISA\fR hierarchy.
815 Inheritance interacts with overloading in two ways.
816 .ie n .IP "Method names in the ""use overload"" directive" 4
817 .el .IP "Method names in the \f(CWuse overload\fR directive" 4
818 .IX Item "Method names in the use overload directive"
819 If \f(CW\*(C`value\*(C'\fR in
820 .Sp
821 .Vb 1
822 \& use overload key => value;
823 .Ve
824 .Sp
825 is a string, it is interpreted as a method name \- which may
826 (in the usual way) be inherited from another class.
827 .IP "Overloading of an operation is inherited by derived classes" 4
828 .IX Item "Overloading of an operation is inherited by derived classes"
829 Any class derived from an overloaded class is also overloaded
830 and inherits its operator implementations.
831 If the same operator is overloaded in more than one ancestor
832 then the implementation is determined by the usual inheritance
833 rules.
834 .Sp
835 For example, if \f(CW\*(C`A\*(C'\fR inherits from \f(CW\*(C`B\*(C'\fR and \f(CW\*(C`C\*(C'\fR (in that order),
836 \&\f(CW\*(C`B\*(C'\fR overloads \f(CW\*(C`+\*(C'\fR with \f(CW\*(C`\e&D::plus_sub\*(C'\fR, and \f(CW\*(C`C\*(C'\fR overloads
837 \&\f(CW\*(C`+\*(C'\fR by \f(CW"plus_meth"\fR, then the subroutine \f(CW\*(C`D::plus_sub\*(C'\fR will
838 be called to implement operation \f(CW\*(C`+\*(C'\fR for an object in package \f(CW\*(C`A\*(C'\fR.
839 .PP
840 Note that in Perl version prior to 5.18 inheritance of the \f(CW\*(C`fallback\*(C'\fR key
841 was not governed by the above rules. The value of \f(CW\*(C`fallback\*(C'\fR in the first
842 overloaded ancestor was used. This was fixed in 5.18 to follow the usual
843 rules of inheritance.
844 .SS "Run-time Overloading"
845 .IX Subsection "Run-time Overloading"
846 Since all \f(CW\*(C`use\*(C'\fR directives are executed at compile-time, the only way to
847 change overloading during run-time is to
848 .PP
849 .Vb 1
850 \& eval \*(Aquse overload "+" => \e&addmethod\*(Aq;
851 .Ve
852 .PP
853 You can also use
854 .PP
855 .Vb 1
856 \& eval \*(Aqno overload "+", "\-\-", "<="\*(Aq;
857 .Ve
858 .PP
859 though the use of these constructs during run-time is questionable.
860 .SS "Public Functions"
861 .IX Subsection "Public Functions"
862 Package \f(CW\*(C`overload.pm\*(C'\fR provides the following public functions:
863 .IP overload::StrVal(arg) 5
864 .IX Item "overload::StrVal(arg)"
865 Gives the string value of \f(CW\*(C`arg\*(C'\fR as in the
866 absence of stringify overloading. If you
867 are using this to get the address of a reference (useful for checking if two
868 references point to the same thing) then you may be better off using
869 \&\f(CWbuiltin::refaddr()\fR or \f(CWScalar::Util::refaddr()\fR, which are faster.
870 .IP overload::Overloaded(arg) 5
871 .IX Item "overload::Overloaded(arg)"
872 Returns true if \f(CW\*(C`arg\*(C'\fR is subject to overloading of some operations.
873 .IP overload::Method(obj,op) 5
874 .IX Item "overload::Method(obj,op)"
875 Returns \f(CW\*(C`undef\*(C'\fR or a reference to the method that implements \f(CW\*(C`op\*(C'\fR.
876 .Sp
877 Such a method always takes three arguments, which will be enforced if
878 it is an XS method.
879 .SS "Overloading Constants"
880 .IX Subsection "Overloading Constants"
881 For some applications, the Perl parser mangles constants too much.
882 It is possible to hook into this process via \f(CWoverload::constant()\fR
883 and \f(CWoverload::remove_constant()\fR functions.
884 .PP
885 These functions take a hash as an argument. The recognized keys of this hash
886 are:
887 .IP integer 8
888 .IX Item "integer"
889 to overload integer constants,
890 .IP float 8
891 .IX Item "float"
892 to overload floating point constants,
893 .IP binary 8
894 .IX Item "binary"
895 to overload octal and hexadecimal constants,
896 .IP q 8
897 .IX Item "q"
898 to overload \f(CW\*(C`q\*(C'\fR\-quoted strings, constant pieces of \f(CW\*(C`qq\*(C'\fR\- and \f(CW\*(C`qx\*(C'\fR\-quoted
899 strings and here-documents,
900 .IP qr 8
901 .IX Item "qr"
902 to overload constant pieces of regular expressions.
903 .PP
904 The corresponding values are references to functions which take three arguments:
905 the first one is the \fIinitial\fR string form of the constant, the second one
906 is how Perl interprets this constant, the third one is how the constant is used.
907 Note that the initial string form does not
908 contain string delimiters, and has backslashes in backslash-delimiter
909 combinations stripped (thus the value of delimiter is not relevant for
910 processing of this string). The return value of this function is how this
911 constant is going to be interpreted by Perl. The third argument is undefined
912 unless for overloaded \f(CW\*(C`q\*(C'\fR\- and \f(CW\*(C`qr\*(C'\fR\- constants, it is \f(CW\*(C`q\*(C'\fR in single-quote
913 context (comes from strings, regular expressions, and single-quote HERE
914 documents), it is \f(CW\*(C`tr\*(C'\fR for arguments of \f(CW\*(C`tr\*(C'\fR/\f(CW\*(C`y\*(C'\fR operators,
915 it is \f(CW\*(C`s\*(C'\fR for right-hand side of \f(CW\*(C`s\*(C'\fR\-operator, and it is \f(CW\*(C`qq\*(C'\fR otherwise.
916 .PP
917 Since an expression \f(CW"ab$cd,,"\fR is just a shortcut for \f(CW\*(Aqab\*(Aq . $cd . \*(Aq,,\*(Aq\fR,
918 it is expected that overloaded constant strings are equipped with reasonable
919 overloaded catenation operator, otherwise absurd results will result.
920 Similarly, negative numbers are considered as negations of positive constants.
921 .PP
922 Note that it is probably meaningless to call the functions \fBoverload::constant()\fR
923 and \fBoverload::remove_constant()\fR from anywhere but \fBimport()\fR and \fBunimport()\fR methods.
924 From these methods they may be called as
925 .PP
926 .Vb 6
927 \& sub import {
928 \& shift;
929 \& return unless @_;
930 \& die "unknown import: @_" unless @_ == 1 and $_[0] eq \*(Aq:constant\*(Aq;
931 \& overload::constant integer => sub {Math::BigInt\->new(shift)};
932 \& }
933 .Ve
934 .SH IMPLEMENTATION
935 .IX Header "IMPLEMENTATION"
936 What follows is subject to change RSN.
937 .PP
938 The table of methods for all operations is cached in magic for the
939 symbol table hash for the package. The cache is invalidated during
940 processing of \f(CW\*(C`use overload\*(C'\fR, \f(CW\*(C`no overload\*(C'\fR, new function
941 definitions, and changes in \f(CW@ISA\fR.
942 .PP
943 (Every SVish thing has a magic queue, and magic is an entry in that
944 queue. This is how a single variable may participate in multiple
945 forms of magic simultaneously. For instance, environment variables
946 regularly have two forms at once: their \f(CW%ENV\fR magic and their taint
947 magic. However, the magic which implements overloading is applied to
948 the stashes, which are rarely used directly, thus should not slow down
949 Perl.)
950 .PP
951 If a package uses overload, it carries a special flag. This flag is also
952 set when new functions are defined or \f(CW@ISA\fR is modified. There will be a
953 slight speed penalty on the very first operation thereafter that supports
954 overloading, while the overload tables are updated. If there is no
955 overloading present, the flag is turned off. Thus the only speed penalty
956 thereafter is the checking of this flag.
957 .PP
958 It is expected that arguments to methods that are not explicitly supposed
959 to be changed are constant (but this is not enforced).
960 .SH COOKBOOK
961 .IX Header "COOKBOOK"
962 Please add examples to what follows!
963 .SS "Two-face Scalars"
964 .IX Subsection "Two-face Scalars"
965 Put this in \fItwo_face.pm\fR in your Perl library directory:
966 .PP
967 .Vb 6
968 \& package two_face; # Scalars with separate string and
969 \& # numeric values.
970 \& sub new { my $p = shift; bless [@_], $p }
971 \& use overload \*(Aq""\*(Aq => \e&str, \*(Aq0+\*(Aq => \e&num, fallback => 1;
972 \& sub num {shift\->[1]}
973 \& sub str {shift\->[0]}
974 .Ve
975 .PP
976 Use it as follows:
977 .PP
978 .Vb 4
979 \& require two_face;
980 \& my $seven = two_face\->new("vii", 7);
981 \& printf "seven=$seven, seven=%d, eight=%d\en", $seven, $seven+1;
982 \& print "seven contains \*(Aqi\*(Aq\en" if $seven =~ /i/;
983 .Ve
984 .PP
985 (The second line creates a scalar which has both a string value, and a
986 numeric value.) This prints:
987 .PP
988 .Vb 2
989 \& seven=vii, seven=7, eight=8
990 \& seven contains \*(Aqi\*(Aq
991 .Ve
992 .SS "Two-face References"
993 .IX Subsection "Two-face References"
994 Suppose you want to create an object which is accessible as both an
995 array reference and a hash reference.
996 .PP
997 .Vb 12
998 \& package two_refs;
999 \& use overload \*(Aq%{}\*(Aq => \e&gethash, \*(Aq@{}\*(Aq => sub { $ {shift()} };
1000 \& sub new {
1001 \& my $p = shift;
1002 \& bless \e [@_], $p;
1003 \& }
1004 \& sub gethash {
1005 \& my %h;
1006 \& my $self = shift;
1007 \& tie %h, ref $self, $self;
1008 \& \e%h;
1009 \& }
1010 \&
1011 \& sub TIEHASH { my $p = shift; bless \e shift, $p }
1012 \& my %fields;
1013 \& my $i = 0;
1014 \& $fields{$_} = $i++ foreach qw{zero one two three};
1015 \& sub STORE {
1016 \& my $self = ${shift()};
1017 \& my $key = $fields{shift()};
1018 \& defined $key or die "Out of band access";
1019 \& $$self\->[$key] = shift;
1020 \& }
1021 \& sub FETCH {
1022 \& my $self = ${shift()};
1023 \& my $key = $fields{shift()};
1024 \& defined $key or die "Out of band access";
1025 \& $$self\->[$key];
1026 \& }
1027 .Ve
1028 .PP
1029 Now one can access an object using both the array and hash syntax:
1030 .PP
1031 .Vb 3
1032 \& my $bar = two_refs\->new(3,4,5,6);
1033 \& $bar\->[2] = 11;
1034 \& $bar\->{two} == 11 or die \*(Aqbad hash fetch\*(Aq;
1035 .Ve
1036 .PP
1037 Note several important features of this example. First of all, the
1038 \&\fIactual\fR type of \f(CW$bar\fR is a scalar reference, and we do not overload
1039 the scalar dereference. Thus we can get the \fIactual\fR non-overloaded
1040 contents of \f(CW$bar\fR by just using \f(CW$$bar\fR (what we do in functions which
1041 overload dereference). Similarly, the object returned by the
1042 \&\fBTIEHASH()\fR method is a scalar reference.
1043 .PP
1044 Second, we create a new tied hash each time the hash syntax is used.
1045 This allows us not to worry about a possibility of a reference loop,
1046 which would lead to a memory leak.
1047 .PP
1048 Both these problems can be cured. Say, if we want to overload hash
1049 dereference on a reference to an object which is \fIimplemented\fR as a
1050 hash itself, the only problem one has to circumvent is how to access
1051 this \fIactual\fR hash (as opposed to the \fIvirtual\fR hash exhibited by the
1052 overloaded dereference operator). Here is one possible fetching routine:
1053 .PP
1054 .Vb 8
1055 \& sub access_hash {
1056 \& my ($self, $key) = (shift, shift);
1057 \& my $class = ref $self;
1058 \& bless $self, \*(Aqoverload::dummy\*(Aq; # Disable overloading of %{}
1059 \& my $out = $self\->{$key};
1060 \& bless $self, $class; # Restore overloading
1061 \& $out;
1062 \& }
1063 .Ve
1064 .PP
1065 To remove creation of the tied hash on each access, one may an extra
1066 level of indirection which allows a non-circular structure of references:
1067 .PP
1068 .Vb 4
1069 \& package two_refs1;
1070 \& use overload
1071 \& \*(Aq%{}\*(Aq => sub { ${shift()}\->[1] },
1072 \& \*(Aq@{}\*(Aq => sub { ${shift()}\->[0] };
1073 \&
1074 \& sub new {
1075 \& my $p = shift;
1076 \& my $a = [@_];
1077 \& my %h;
1078 \& tie %h, $p, $a;
1079 \& bless \e [$a, \e%h], $p;
1080 \& }
1081 \& sub gethash {
1082 \& my %h;
1083 \& my $self = shift;
1084 \& tie %h, ref $self, $self;
1085 \& \e%h;
1086 \& }
1087 \&
1088 \& sub TIEHASH { my $p = shift; bless \e shift, $p }
1089 \& my %fields;
1090 \& my $i = 0;
1091 \& $fields{$_} = $i++ foreach qw{zero one two three};
1092 \& sub STORE {
1093 \& my $a = ${shift()};
1094 \& my $key = $fields{shift()};
1095 \& defined $key or die "Out of band access";
1096 \& $a\->[$key] = shift;
1097 \& }
1098 \& sub FETCH {
1099 \& my $a = ${shift()};
1100 \& my $key = $fields{shift()};
1101 \& defined $key or die "Out of band access";
1102 \& $a\->[$key];
1103 \& }
1104 .Ve
1105 .PP
1106 Now if \f(CW$baz\fR is overloaded like this, then \f(CW$baz\fR is a reference to a
1107 reference to the intermediate array, which keeps a reference to an
1108 actual array, and the access hash. The \fBtie()\fRing object for the access
1109 hash is a reference to a reference to the actual array, so
1110 .IP \(bu 4
1111 There are no loops of references.
1112 .IP \(bu 4
1113 Both "objects" which are blessed into the class \f(CW\*(C`two_refs1\*(C'\fR are
1114 references to a reference to an array, thus references to a \fIscalar\fR.
1115 Thus the accessor expression \f(CW\*(C`$$foo\->[$ind]\*(C'\fR involves no
1116 overloaded operations.
1117 .SS "Symbolic Calculator"
1118 .IX Subsection "Symbolic Calculator"
1119 Put this in \fIsymbolic.pm\fR in your Perl library directory:
1120 .PP
1121 .Vb 2
1122 \& package symbolic; # Primitive symbolic calculator
1123 \& use overload nomethod => \e&wrap;
1124 \&
1125 \& sub new { shift; bless [\*(Aqn\*(Aq, @_] }
1126 \& sub wrap {
1127 \& my ($obj, $other, $inv, $meth) = @_;
1128 \& ($obj, $other) = ($other, $obj) if $inv;
1129 \& bless [$meth, $obj, $other];
1130 \& }
1131 .Ve
1132 .PP
1133 This module is very unusual as overloaded modules go: it does not
1134 provide any usual overloaded operators, instead it provides an
1135 implementation for \f(CW"nomethod"\fR. In this example the \f(CW\*(C`nomethod\*(C'\fR
1136 subroutine returns an object which encapsulates operations done over
1137 the objects: \f(CW\*(C`symbolic\->new(3)\*(C'\fR contains \f(CW\*(C`[\*(Aqn\*(Aq, 3]\*(C'\fR, \f(CW\*(C`2 +
1138 symbolic\->new(3)\*(C'\fR contains \f(CW\*(C`[\*(Aq+\*(Aq, 2, [\*(Aqn\*(Aq, 3]]\*(C'\fR.
1139 .PP
1140 Here is an example of the script which "calculates" the side of
1141 circumscribed octagon using the above package:
1142 .PP
1143 .Vb 4
1144 \& require symbolic;
1145 \& my $iter = 1; # 2**($iter+2) = 8
1146 \& my $side = symbolic\->new(1);
1147 \& my $cnt = $iter;
1148 \&
1149 \& while ($cnt\-\-) {
1150 \& $side = (sqrt(1 + $side**2) \- 1)/$side;
1151 \& }
1152 \& print "OK\en";
1153 .Ve
1154 .PP
1155 The value of \f(CW$side\fR is
1156 .PP
1157 .Vb 2
1158 \& [\*(Aq/\*(Aq, [\*(Aq\-\*(Aq, [\*(Aqsqrt\*(Aq, [\*(Aq+\*(Aq, 1, [\*(Aq**\*(Aq, [\*(Aqn\*(Aq, 1], 2]],
1159 \& undef], 1], [\*(Aqn\*(Aq, 1]]
1160 .Ve
1161 .PP
1162 Note that while we obtained this value using a nice little script,
1163 there is no simple way to \fIuse\fR this value. In fact this value may
1164 be inspected in debugger (see perldebug), but only if
1165 \&\f(CW\*(C`bareStringify\*(C'\fR \fBO\fRption is set, and not via \f(CW\*(C`p\*(C'\fR command.
1166 .PP
1167 If one attempts to print this value, then the overloaded operator
1168 \&\f(CW""\fR will be called, which will call \f(CW\*(C`nomethod\*(C'\fR operator. The
1169 result of this operator will be stringified again, but this result is
1170 again of type \f(CW\*(C`symbolic\*(C'\fR, which will lead to an infinite loop.
1171 .PP
1172 Add a pretty-printer method to the module \fIsymbolic.pm\fR:
1173 .PP
1174 .Vb 8
1175 \& sub pretty {
1176 \& my ($meth, $a, $b) = @{+shift};
1177 \& $a = \*(Aqu\*(Aq unless defined $a;
1178 \& $b = \*(Aqu\*(Aq unless defined $b;
1179 \& $a = $a\->pretty if ref $a;
1180 \& $b = $b\->pretty if ref $b;
1181 \& "[$meth $a $b]";
1182 \& }
1183 .Ve
1184 .PP
1185 Now one can finish the script by
1186 .PP
1187 .Vb 1
1188 \& print "side = ", $side\->pretty, "\en";
1189 .Ve
1190 .PP
1191 The method \f(CW\*(C`pretty\*(C'\fR is doing object-to-string conversion, so it
1192 is natural to overload the operator \f(CW""\fR using this method. However,
1193 inside such a method it is not necessary to pretty-print the
1194 \&\fIcomponents\fR \f(CW$a\fR and \f(CW$b\fR of an object. In the above subroutine
1195 \&\f(CW"[$meth $a $b]"\fR is a catenation of some strings and components \f(CW$a\fR
1196 and \f(CW$b\fR. If these components use overloading, the catenation operator
1197 will look for an overloaded operator \f(CW\*(C`.\*(C'\fR; if not present, it will
1198 look for an overloaded operator \f(CW""\fR. Thus it is enough to use
1199 .PP
1200 .Vb 7
1201 \& use overload nomethod => \e&wrap, \*(Aq""\*(Aq => \e&str;
1202 \& sub str {
1203 \& my ($meth, $a, $b) = @{+shift};
1204 \& $a = \*(Aqu\*(Aq unless defined $a;
1205 \& $b = \*(Aqu\*(Aq unless defined $b;
1206 \& "[$meth $a $b]";
1207 \& }
1208 .Ve
1209 .PP
1210 Now one can change the last line of the script to
1211 .PP
1212 .Vb 1
1213 \& print "side = $side\en";
1214 .Ve
1215 .PP
1216 which outputs
1217 .PP
1218 .Vb 1
1219 \& side = [/ [\- [sqrt [+ 1 [** [n 1 u] 2]] u] 1] [n 1 u]]
1220 .Ve
1221 .PP
1222 and one can inspect the value in debugger using all the possible
1223 methods.
1224 .PP
1225 Something is still amiss: consider the loop variable \f(CW$cnt\fR of the
1226 script. It was a number, not an object. We cannot make this value of
1227 type \f(CW\*(C`symbolic\*(C'\fR, since then the loop will not terminate.
1228 .PP
1229 Indeed, to terminate the cycle, the \f(CW$cnt\fR should become false.
1230 However, the operator \f(CW\*(C`bool\*(C'\fR for checking falsity is overloaded (this
1231 time via overloaded \f(CW""\fR), and returns a long string, thus any object
1232 of type \f(CW\*(C`symbolic\*(C'\fR is true. To overcome this, we need a way to
1233 compare an object to 0. In fact, it is easier to write a numeric
1234 conversion routine.
1235 .PP
1236 Here is the text of \fIsymbolic.pm\fR with such a routine added (and
1237 slightly modified \fBstr()\fR):
1238 .PP
1239 .Vb 3
1240 \& package symbolic; # Primitive symbolic calculator
1241 \& use overload
1242 \& nomethod => \e&wrap, \*(Aq""\*(Aq => \e&str, \*(Aq0+\*(Aq => \e&num;
1243 \&
1244 \& sub new { shift; bless [\*(Aqn\*(Aq, @_] }
1245 \& sub wrap {
1246 \& my ($obj, $other, $inv, $meth) = @_;
1247 \& ($obj, $other) = ($other, $obj) if $inv;
1248 \& bless [$meth, $obj, $other];
1249 \& }
1250 \& sub str {
1251 \& my ($meth, $a, $b) = @{+shift};
1252 \& $a = \*(Aqu\*(Aq unless defined $a;
1253 \& if (defined $b) {
1254 \& "[$meth $a $b]";
1255 \& } else {
1256 \& "[$meth $a]";
1257 \& }
1258 \& }
1259 \& my %subr = (
1260 \& n => sub {$_[0]},
1261 \& sqrt => sub {sqrt $_[0]},
1262 \& \*(Aq\-\*(Aq => sub {shift() \- shift()},
1263 \& \*(Aq+\*(Aq => sub {shift() + shift()},
1264 \& \*(Aq/\*(Aq => sub {shift() / shift()},
1265 \& \*(Aq*\*(Aq => sub {shift() * shift()},
1266 \& \*(Aq**\*(Aq => sub {shift() ** shift()},
1267 \& );
1268 \& sub num {
1269 \& my ($meth, $a, $b) = @{+shift};
1270 \& my $subr = $subr{$meth}
1271 \& or die "Do not know how to ($meth) in symbolic";
1272 \& $a = $a\->num if ref $a eq _\|_PACKAGE_\|_;
1273 \& $b = $b\->num if ref $b eq _\|_PACKAGE_\|_;
1274 \& $subr\->($a,$b);
1275 \& }
1276 .Ve
1277 .PP
1278 All the work of numeric conversion is done in \f(CW%subr\fR and \fBnum()\fR. Of
1279 course, \f(CW%subr\fR is not complete, it contains only operators used in the
1280 example below. Here is the extra-credit question: why do we need an
1281 explicit recursion in \fBnum()\fR? (Answer is at the end of this section.)
1282 .PP
1283 Use this module like this:
1284 .PP
1285 .Vb 4
1286 \& require symbolic;
1287 \& my $iter = symbolic\->new(2); # 16\-gon
1288 \& my $side = symbolic\->new(1);
1289 \& my $cnt = $iter;
1290 \&
1291 \& while ($cnt) {
1292 \& $cnt = $cnt \- 1; # Mutator \*(Aq\-\-\*(Aq not implemented
1293 \& $side = (sqrt(1 + $side**2) \- 1)/$side;
1294 \& }
1295 \& printf "%s=%f\en", $side, $side;
1296 \& printf "pi=%f\en", $side*(2**($iter+2));
1297 .Ve
1298 .PP
1299 It prints (without so many line breaks)
1300 .PP
1301 .Vb 4
1302 \& [/ [\- [sqrt [+ 1 [** [/ [\- [sqrt [+ 1 [** [n 1] 2]]] 1]
1303 \& [n 1]] 2]]] 1]
1304 \& [/ [\- [sqrt [+ 1 [** [n 1] 2]]] 1] [n 1]]]=0.198912
1305 \& pi=3.182598
1306 .Ve
1307 .PP
1308 The above module is very primitive. It does not implement
1309 mutator methods (\f(CW\*(C`++\*(C'\fR, \f(CW\*(C`\-=\*(C'\fR and so on), does not do deep copying
1310 (not required without mutators!), and implements only those arithmetic
1311 operations which are used in the example.
1312 .PP
1313 To implement most arithmetic operations is easy; one should just use
1314 the tables of operations, and change the code which fills \f(CW%subr\fR to
1315 .PP
1316 .Vb 12
1317 \& my %subr = ( \*(Aqn\*(Aq => sub {$_[0]} );
1318 \& foreach my $op (split " ", $overload::ops{with_assign}) {
1319 \& $subr{$op} = $subr{"$op="} = eval "sub {shift() $op shift()}";
1320 \& }
1321 \& my @bins = qw(binary 3way_comparison num_comparison str_comparison);
1322 \& foreach my $op (split " ", "@overload::ops{ @bins }") {
1323 \& $subr{$op} = eval "sub {shift() $op shift()}";
1324 \& }
1325 \& foreach my $op (split " ", "@overload::ops{qw(unary func)}") {
1326 \& print "defining \*(Aq$op\*(Aq\en";
1327 \& $subr{$op} = eval "sub {$op shift()}";
1328 \& }
1329 .Ve
1330 .PP
1331 Since subroutines implementing assignment operators are not required
1332 to modify their operands (see "Overloadable Operations" above),
1333 we do not need anything special to make \f(CW\*(C`+=\*(C'\fR and friends work,
1334 besides adding these operators to \f(CW%subr\fR and defining a copy
1335 constructor (needed since Perl has no way to know that the
1336 implementation of \f(CW\*(Aq+=\*(Aq\fR does not mutate the argument \-
1337 see "Copy Constructor").
1338 .PP
1339 To implement a copy constructor, add \f(CW\*(C`\*(Aq=\*(Aq => \e&cpy\*(C'\fR to \f(CW\*(C`use overload\*(C'\fR
1340 line, and code (this code assumes that mutators change things one level
1341 deep only, so recursive copying is not needed):
1342 .PP
1343 .Vb 4
1344 \& sub cpy {
1345 \& my $self = shift;
1346 \& bless [@$self], ref $self;
1347 \& }
1348 .Ve
1349 .PP
1350 To make \f(CW\*(C`++\*(C'\fR and \f(CW\*(C`\-\-\*(C'\fR work, we need to implement actual mutators,
1351 either directly, or in \f(CW\*(C`nomethod\*(C'\fR. We continue to do things inside
1352 \&\f(CW\*(C`nomethod\*(C'\fR, thus add
1353 .PP
1354 .Vb 4
1355 \& if ($meth eq \*(Aq++\*(Aq or $meth eq \*(Aq\-\-\*(Aq) {
1356 \& @$obj = ($meth, (bless [@$obj]), 1); # Avoid circular reference
1357 \& return $obj;
1358 \& }
1359 .Ve
1360 .PP
1361 after the first line of \fBwrap()\fR. This is not a most effective
1362 implementation, one may consider
1363 .PP
1364 .Vb 1
1365 \& sub inc { $_[0] = bless [\*(Aq++\*(Aq, shift, 1]; }
1366 .Ve
1367 .PP
1368 instead.
1369 .PP
1370 As a final remark, note that one can fill \f(CW%subr\fR by
1371 .PP
1372 .Vb 10
1373 \& my %subr = ( \*(Aqn\*(Aq => sub {$_[0]} );
1374 \& foreach my $op (split " ", $overload::ops{with_assign}) {
1375 \& $subr{$op} = $subr{"$op="} = eval "sub {shift() $op shift()}";
1376 \& }
1377 \& my @bins = qw(binary 3way_comparison num_comparison str_comparison);
1378 \& foreach my $op (split " ", "@overload::ops{ @bins }") {
1379 \& $subr{$op} = eval "sub {shift() $op shift()}";
1380 \& }
1381 \& foreach my $op (split " ", "@overload::ops{qw(unary func)}") {
1382 \& $subr{$op} = eval "sub {$op shift()}";
1383 \& }
1384 \& $subr{\*(Aq++\*(Aq} = $subr{\*(Aq+\*(Aq};
1385 \& $subr{\*(Aq\-\-\*(Aq} = $subr{\*(Aq\-\*(Aq};
1386 .Ve
1387 .PP
1388 This finishes implementation of a primitive symbolic calculator in
1389 50 lines of Perl code. Since the numeric values of subexpressions
1390 are not cached, the calculator is very slow.
1391 .PP
1392 Here is the answer for the exercise: In the case of \fBstr()\fR, we need no
1393 explicit recursion since the overloaded \f(CW\*(C`.\*(C'\fR\-operator will fall back
1394 to an existing overloaded operator \f(CW""\fR. Overloaded arithmetic
1395 operators \fIdo not\fR fall back to numeric conversion if \f(CW\*(C`fallback\*(C'\fR is
1396 not explicitly requested. Thus without an explicit recursion \fBnum()\fR
1397 would convert \f(CW\*(C`[\*(Aq+\*(Aq, $a, $b]\*(C'\fR to \f(CW\*(C`$a + $b\*(C'\fR, which would just rebuild
1398 the argument of \fBnum()\fR.
1399 .PP
1400 If you wonder why defaults for conversion are different for \fBstr()\fR and
1401 \&\fBnum()\fR, note how easy it was to write the symbolic calculator. This
1402 simplicity is due to an appropriate choice of defaults. One extra
1403 note: due to the explicit recursion \fBnum()\fR is more fragile than \fBsym()\fR:
1404 we need to explicitly check for the type of \f(CW$a\fR and \f(CW$b\fR. If components
1405 \&\f(CW$a\fR and \f(CW$b\fR happen to be of some related type, this may lead to problems.
1406 .SS "\fIReally\fP Symbolic Calculator"
1407 .IX Subsection "Really Symbolic Calculator"
1408 One may wonder why we call the above calculator symbolic. The reason
1409 is that the actual calculation of the value of expression is postponed
1410 until the value is \fIused\fR.
1411 .PP
1412 To see it in action, add a method
1413 .PP
1414 .Vb 5
1415 \& sub STORE {
1416 \& my $obj = shift;
1417 \& $#$obj = 1;
1418 \& @$obj\->[0,1] = (\*(Aq=\*(Aq, shift);
1419 \& }
1420 .Ve
1421 .PP
1422 to the package \f(CW\*(C`symbolic\*(C'\fR. After this change one can do
1423 .PP
1424 .Vb 3
1425 \& my $a = symbolic\->new(3);
1426 \& my $b = symbolic\->new(4);
1427 \& my $c = sqrt($a**2 + $b**2);
1428 .Ve
1429 .PP
1430 and the numeric value of \f(CW$c\fR becomes 5. However, after calling
1431 .PP
1432 .Vb 1
1433 \& $a\->STORE(12); $b\->STORE(5);
1434 .Ve
1435 .PP
1436 the numeric value of \f(CW$c\fR becomes 13. There is no doubt now that the module
1437 symbolic provides a \fIsymbolic\fR calculator indeed.
1438 .PP
1439 To hide the rough edges under the hood, provide a \fBtie()\fRd interface to the
1440 package \f(CW\*(C`symbolic\*(C'\fR. Add methods
1441 .PP
1442 .Vb 3
1443 \& sub TIESCALAR { my $pack = shift; $pack\->new(@_) }
1444 \& sub FETCH { shift }
1445 \& sub nop { } # Around a bug
1446 .Ve
1447 .PP
1448 (the bug, fixed in Perl 5.14, is described in "BUGS"). One can use this
1449 new interface as
1450 .PP
1451 .Vb 3
1452 \& tie $a, \*(Aqsymbolic\*(Aq, 3;
1453 \& tie $b, \*(Aqsymbolic\*(Aq, 4;
1454 \& $a\->nop; $b\->nop; # Around a bug
1455 \&
1456 \& my $c = sqrt($a**2 + $b**2);
1457 .Ve
1458 .PP
1459 Now numeric value of \f(CW$c\fR is 5. After \f(CW\*(C`$a = 12; $b = 5\*(C'\fR the numeric value
1460 of \f(CW$c\fR becomes 13. To insulate the user of the module add a method
1461 .PP
1462 .Vb 1
1463 \& sub vars { my $p = shift; tie($_, $p), $_\->nop foreach @_; }
1464 .Ve
1465 .PP
1466 Now
1467 .PP
1468 .Vb 3
1469 \& my ($a, $b);
1470 \& symbolic\->vars($a, $b);
1471 \& my $c = sqrt($a**2 + $b**2);
1472 \&
1473 \& $a = 3; $b = 4;
1474 \& printf "c5 %s=%f\en", $c, $c;
1475 \&
1476 \& $a = 12; $b = 5;
1477 \& printf "c13 %s=%f\en", $c, $c;
1478 .Ve
1479 .PP
1480 shows that the numeric value of \f(CW$c\fR follows changes to the values of \f(CW$a\fR
1481 and \f(CW$b\fR.
1482 .SH AUTHOR
1483 .IX Header "AUTHOR"
1484 Ilya Zakharevich <\[email protected]\-state.edu\fR>.
1485 .SH "SEE ALSO"
1486 .IX Header "SEE ALSO"
1487 The \f(CW\*(C`overloading\*(C'\fR pragma can be used to enable or disable overloaded
1488 operations within a lexical scope \- see overloading.
1489 .SH DIAGNOSTICS
1490 .IX Header "DIAGNOSTICS"
1491 When Perl is run with the \fB\-Do\fR switch or its equivalent, overloading
1492 induces diagnostic messages.
1493 .PP
1494 Using the \f(CW\*(C`m\*(C'\fR command of Perl debugger (see perldebug) one can
1495 deduce which operations are overloaded (and which ancestor triggers
1496 this overloading). Say, if \f(CW\*(C`eq\*(C'\fR is overloaded, then the method \f(CW\*(C`(eq\*(C'\fR
1497 is shown by debugger. The method \f(CW\*(C`()\*(C'\fR corresponds to the \f(CW\*(C`fallback\*(C'\fR
1498 key (in fact a presence of this method shows that this package has
1499 overloading enabled, and it is what is used by the \f(CW\*(C`Overloaded\*(C'\fR
1500 function of module \f(CW\*(C`overload\*(C'\fR).
1501 .PP
1502 The module might issue the following warnings:
1503 .IP "Odd number of arguments for overload::constant" 4
1504 .IX Item "Odd number of arguments for overload::constant"
1505 (W) The call to overload::constant contained an odd number of arguments.
1506 The arguments should come in pairs.
1507 .IP "'%s' is not an overloadable type" 4
1508 .IX Item "'%s' is not an overloadable type"
1509 (W) You tried to overload a constant type the overload package is unaware of.
1510 .IP "'%s' is not a code reference" 4
1511 .IX Item "'%s' is not a code reference"
1512 (W) The second (fourth, sixth, ...) argument of overload::constant needs
1513 to be a code reference. Either an anonymous subroutine, or a reference
1514 to a subroutine.
1515 .IP "overload arg '%s' is invalid" 4
1516 .IX Item "overload arg '%s' is invalid"
1517 (W) \f(CW\*(C`use overload\*(C'\fR was passed an argument it did not
1518 recognize. Did you mistype an operator?
1519 .SH "BUGS AND PITFALLS"
1520 .IX Header "BUGS AND PITFALLS"
1521 .IP \(bu 4
1522 A pitfall when fallback is TRUE and Perl resorts to a built-in
1523 implementation of an operator is that some operators have more
1524 than one semantic, for example \f(CW\*(C`|\*(C'\fR:
1525 .Sp
1526 .Vb 5
1527 \& use overload \*(Aq0+\*(Aq => sub { $_[0]\->{n}; },
1528 \& fallback => 1;
1529 \& my $x = bless { n => 4 }, "main";
1530 \& my $y = bless { n => 8 }, "main";
1531 \& print $x | $y, "\en";
1532 .Ve
1533 .Sp
1534 You might expect this to output "12".
1535 In fact, it prints "<": the ASCII result of treating "|"
1536 as a bitwise string operator \- that is, the result of treating
1537 the operands as the strings "4" and "8" rather than numbers.
1538 The fact that numify (\f(CW\*(C`0+\*(C'\fR) is implemented but stringify
1539 (\f(CW""\fR) isn't makes no difference since the latter is simply
1540 autogenerated from the former.
1541 .Sp
1542 The only way to change this is to provide your own subroutine
1543 for \f(CW\*(Aq|\*(Aq\fR.
1544 .IP \(bu 4
1545 Magic autogeneration increases the potential for inadvertently
1546 creating self-referential structures.
1547 Currently Perl will not free self-referential
1548 structures until cycles are explicitly broken.
1549 For example,
1550 .Sp
1551 .Vb 2
1552 \& use overload \*(Aq+\*(Aq => \*(Aqadd\*(Aq;
1553 \& sub add { bless [ \e$_[0], \e$_[1] ] };
1554 .Ve
1555 .Sp
1556 is asking for trouble, since
1557 .Sp
1558 .Vb 1
1559 \& $obj += $y;
1560 .Ve
1561 .Sp
1562 will effectively become
1563 .Sp
1564 .Vb 1
1565 \& $obj = add($obj, $y, undef);
1566 .Ve
1567 .Sp
1568 with the same result as
1569 .Sp
1570 .Vb 1
1571 \& $obj = [\e$obj, \e$foo];
1572 .Ve
1573 .Sp
1574 Even if no \fIexplicit\fR assignment-variants of operators are present in
1575 the script, they may be generated by the optimizer.
1576 For example,
1577 .Sp
1578 .Vb 1
1579 \& "obj = $obj\en"
1580 .Ve
1581 .Sp
1582 may be optimized to
1583 .Sp
1584 .Vb 1
1585 \& my $tmp = \*(Aqobj = \*(Aq . $obj; $tmp .= "\en";
1586 .Ve
1587 .IP \(bu 4
1588 The symbol table is filled with names looking like line-noise.
1589 .IP \(bu 4
1590 This bug was fixed in Perl 5.18, but may still trip you up if you are using
1591 older versions:
1592 .Sp
1593 For the purpose of inheritance every overloaded package behaves as if
1594 \&\f(CW\*(C`fallback\*(C'\fR is present (possibly undefined). This may create
1595 interesting effects if some package is not overloaded, but inherits
1596 from two overloaded packages.
1597 .IP \(bu 4
1598 Before Perl 5.14, the relation between overloading and \fBtie()\fRing was broken.
1599 Overloading was triggered or not based on the \fIprevious\fR class of the
1600 \&\fBtie()\fRd variable.
1601 .Sp
1602 This happened because the presence of overloading was checked
1603 too early, before any \fBtie()\fRd access was attempted. If the
1604 class of the value \fBFETCH()\fRed from the tied variable does not
1605 change, a simple workaround for code that is to run on older Perl
1606 versions is to access the value (via \f(CW\*(C`() = $foo\*(C'\fR or some such)
1607 immediately after \fBtie()\fRing, so that after this call the \fIprevious\fR class
1608 coincides with the current one.
1609 .IP \(bu 4
1610 Barewords are not covered by overloaded string constants.
1611 .IP \(bu 4
1612 The range operator \f(CW\*(C`..\*(C'\fR cannot be overloaded.