1PERLOP(1)              Perl Programmers Reference Guide              PERLOP(1)
2
3
4

NAME

6       perlop - Perl operators and precedence
7

DESCRIPTION

9       In Perl, the operator determines what operation is performed,
10       independent of the type of the operands.  For example "$x + $y" is
11       always a numeric addition, and if $x or $y do not contain numbers, an
12       attempt is made to convert them to numbers first.
13
14       This is in contrast to many other dynamic languages, where the
15       operation is determined by the type of the first argument.  It also
16       means that Perl has two versions of some operators, one for numeric and
17       one for string comparison.  For example "$x == $y" compares two numbers
18       for equality, and "$x eq $y" compares two strings.
19
20       There are a few exceptions though: "x" can be either string repetition
21       or list repetition, depending on the type of the left operand, and "&",
22       "|", "^" and "~" can be either string or numeric bit operations.
23
24   Operator Precedence and Associativity
25       Operator precedence and associativity work in Perl more or less like
26       they do in mathematics.
27
28       Operator precedence means some operators group more tightly than
29       others.  For example, in "2 + 4 * 5", the multiplication has higher
30       precedence, so "4 * 5" is grouped together as the right-hand operand of
31       the addition, rather than "2 + 4" being grouped together as the left-
32       hand operand of the multiplication. It is as if the expression were
33       written "2 + (4 * 5)", not "(2 + 4) * 5". So the expression yields "2 +
34       20 == 22", rather than "6 * 5 == 30".
35
36       Operator associativity defines what happens if a sequence of the same
37       operators is used one after another: usually that they will be grouped
38       at the left or the right. For example, in "9 - 3 - 2", subtraction is
39       left associative, so "9 - 3" is grouped together as the left-hand
40       operand of the second subtraction, rather than "3 - 2" being grouped
41       together as the right-hand operand of the first subtraction. It is as
42       if the expression were written "(9 - 3) - 2", not "9 - (3 - 2)". So the
43       expression yields "6 - 2 == 4", rather than "9 - 1 == 8".
44
45       For simple operators that evaluate all their operands and then combine
46       the values in some way, precedence and associativity (and parentheses)
47       imply some ordering requirements on those combining operations. For
48       example, in "2 + 4 * 5", the grouping implied by precedence means that
49       the multiplication of 4 and 5 must be performed before the addition of
50       2 and 20, simply because the result of that multiplication is required
51       as one of the operands of the addition. But the order of operations is
52       not fully determined by this: in "2 * 2 + 4 * 5" both multiplications
53       must be performed before the addition, but the grouping does not say
54       anything about the order in which the two multiplications are
55       performed. In fact Perl has a general rule that the operands of an
56       operator are evaluated in left-to-right order. A few operators such as
57       "&&=" have special evaluation rules that can result in an operand not
58       being evaluated at all; in general, the top-level operator in an
59       expression has control of operand evaluation.
60
61       Some comparison operators, as their associativity, chain with some
62       operators of the same precedence (but never with operators of different
63       precedence).  This chaining means that each comparison is performed on
64       the two arguments surrounding it, with each interior argument taking
65       part in two comparisons, and the comparison results are implicitly
66       ANDed.  Thus "$x < $y <= $z" behaves exactly like
67       "$x < $y && $y <= $z", assuming that "$y" is as simple a scalar as it
68       looks.  The ANDing short-circuits just like "&&" does, stopping the
69       sequence of comparisons as soon as one yields false.
70
71       In a chained comparison, each argument expression is evaluated at most
72       once, even if it takes part in two comparisons, but the result of the
73       evaluation is fetched for each comparison.  (It is not evaluated at all
74       if the short-circuiting means that it's not required for any
75       comparisons.)  This matters if the computation of an interior argument
76       is expensive or non-deterministic.  For example,
77
78           if($x < expensive_sub() <= $z) { ...
79
80       is not entirely like
81
82           if($x < expensive_sub() && expensive_sub() <= $z) { ...
83
84       but instead closer to
85
86           my $tmp = expensive_sub();
87           if($x < $tmp && $tmp <= $z) { ...
88
89       in that the subroutine is only called once.  However, it's not exactly
90       like this latter code either, because the chained comparison doesn't
91       actually involve any temporary variable (named or otherwise): there is
92       no assignment.  This doesn't make much difference where the expression
93       is a call to an ordinary subroutine, but matters more with an lvalue
94       subroutine, or if the argument expression yields some unusual kind of
95       scalar by other means.  For example, if the argument expression yields
96       a tied scalar, then the expression is evaluated to produce that scalar
97       at most once, but the value of that scalar may be fetched up to twice,
98       once for each comparison in which it is actually used.
99
100       In this example, the expression is evaluated only once, and the tied
101       scalar (the result of the expression) is fetched for each comparison
102       that uses it.
103
104           if ($x < $tied_scalar < $z) { ...
105
106       In the next example, the expression is evaluated only once, and the
107       tied scalar is fetched once as part of the operation within the
108       expression.  The result of that operation is fetched for each
109       comparison, which normally doesn't matter unless that expression result
110       is also magical due to operator overloading.
111
112           if ($x < $tied_scalar + 42 < $z) { ...
113
114       Some operators are instead non-associative, meaning that it is a syntax
115       error to use a sequence of those operators of the same precedence.  For
116       example, "$x .. $y .. $z" is an error.
117
118       Perl operators have the following associativity and precedence, listed
119       from highest precedence to lowest.  Operators borrowed from C keep the
120       same precedence relationship with each other, even where C's precedence
121       is slightly screwy.  (This makes learning Perl easier for C folks.)
122       With very few exceptions, these all operate on scalar values only, not
123       array values.
124
125           left        terms and list operators (leftward)
126           left        ->
127           nonassoc    ++ --
128           right       **
129           right       ! ~ \ and unary + and -
130           left        =~ !~
131           left        * / % x
132           left        + - .
133           left        << >>
134           nonassoc    named unary operators
135           chained     < > <= >= lt gt le ge
136           chain/na    == != eq ne <=> cmp ~~
137           nonassoc    isa
138           left        &
139           left        | ^
140           left        &&
141           left        || //
142           nonassoc    ..  ...
143           right       ?:
144           right       = += -= *= etc. goto last next redo dump
145           left        , =>
146           nonassoc    list operators (rightward)
147           right       not
148           left        and
149           left        or xor
150
151       In the following sections, these operators are covered in detail, in
152       the same order in which they appear in the table above.
153
154       Many operators can be overloaded for objects.  See overload.
155
156   Terms and List Operators (Leftward)
157       A TERM has the highest precedence in Perl.  They include variables,
158       quote and quote-like operators, any expression in parentheses, and any
159       function whose arguments are parenthesized.  Actually, there aren't
160       really functions in this sense, just list operators and unary operators
161       behaving as functions because you put parentheses around the arguments.
162       These are all documented in perlfunc.
163
164       If any list operator ("print()", etc.) or any unary operator
165       ("chdir()", etc.)  is followed by a left parenthesis as the next token,
166       the operator and arguments within parentheses are taken to be of
167       highest precedence, just like a normal function call.
168
169       In the absence of parentheses, the precedence of list operators such as
170       "print", "sort", or "chmod" is either very high or very low depending
171       on whether you are looking at the left side or the right side of the
172       operator.  For example, in
173
174           @ary = (1, 3, sort 4, 2);
175           print @ary;         # prints 1324
176
177       the commas on the right of the "sort" are evaluated before the "sort",
178       but the commas on the left are evaluated after.  In other words, list
179       operators tend to gobble up all arguments that follow, and then act
180       like a simple TERM with regard to the preceding expression.  Be careful
181       with parentheses:
182
183           # These evaluate exit before doing the print:
184           print($foo, exit);  # Obviously not what you want.
185           print $foo, exit;   # Nor is this.
186
187           # These do the print before evaluating exit:
188           (print $foo), exit; # This is what you want.
189           print($foo), exit;  # Or this.
190           print ($foo), exit; # Or even this.
191
192       Also note that
193
194           print ($foo & 255) + 1, "\n";
195
196       probably doesn't do what you expect at first glance.  The parentheses
197       enclose the argument list for "print" which is evaluated (printing the
198       result of "$foo & 255").  Then one is added to the return value of
199       "print" (usually 1).  The result is something like this:
200
201           1 + 1, "\n";    # Obviously not what you meant.
202
203       To do what you meant properly, you must write:
204
205           print(($foo & 255) + 1, "\n");
206
207       See "Named Unary Operators" for more discussion of this.
208
209       Also parsed as terms are the "do {}" and "eval {}" constructs, as well
210       as subroutine and method calls, and the anonymous constructors "[]" and
211       "{}".
212
213       See also "Quote and Quote-like Operators" toward the end of this
214       section, as well as "I/O Operators".
215
216   The Arrow Operator
217       ""->"" is an infix dereference operator, just as it is in C and C++.
218       If the right side is either a "[...]", "{...}", or a "(...)" subscript,
219       then the left side must be either a hard or symbolic reference to an
220       array, a hash, or a subroutine respectively.  (Or technically speaking,
221       a location capable of holding a hard reference, if it's an array or
222       hash reference being used for assignment.)  See perlreftut and perlref.
223
224       Otherwise, the right side is a method name or a simple scalar variable
225       containing either the method name or a subroutine reference, and the
226       left side must be either an object (a blessed reference) or a class
227       name (that is, a package name).  See perlobj.
228
229       The dereferencing cases (as opposed to method-calling cases) are
230       somewhat extended by the "postderef" feature.  For the details of that
231       feature, consult "Postfix Dereference Syntax" in perlref.
232
233   Auto-increment and Auto-decrement
234       "++" and "--" work as in C.  That is, if placed before a variable, they
235       increment or decrement the variable by one before returning the value,
236       and if placed after, increment or decrement after returning the value.
237
238           $i = 0;  $j = 0;
239           print $i++;  # prints 0
240           print ++$j;  # prints 1
241
242       Note that just as in C, Perl doesn't define when the variable is
243       incremented or decremented.  You just know it will be done sometime
244       before or after the value is returned.  This also means that modifying
245       a variable twice in the same statement will lead to undefined behavior.
246       Avoid statements like:
247
248           $i = $i ++;
249           print ++ $i + $i ++;
250
251       Perl will not guarantee what the result of the above statements is.
252
253       The auto-increment operator has a little extra builtin magic to it.  If
254       you increment a variable that is numeric, or that has ever been used in
255       a numeric context, you get a normal increment.  If, however, the
256       variable has been used in only string contexts since it was set, and
257       has a value that is not the empty string and matches the pattern
258       "/^[a-zA-Z]*[0-9]*\z/", the increment is done as a string, preserving
259       each character within its range, with carry:
260
261           print ++($foo = "99");      # prints "100"
262           print ++($foo = "a0");      # prints "a1"
263           print ++($foo = "Az");      # prints "Ba"
264           print ++($foo = "zz");      # prints "aaa"
265
266       "undef" is always treated as numeric, and in particular is changed to 0
267       before incrementing (so that a post-increment of an undef value will
268       return 0 rather than "undef").
269
270       The auto-decrement operator is not magical.
271
272   Exponentiation
273       Binary "**" is the exponentiation operator.  It binds even more tightly
274       than unary minus, so "-2**4" is "-(2**4)", not "(-2)**4".  (This is
275       implemented using C's pow(3) function, which actually works on doubles
276       internally.)
277
278       Note that certain exponentiation expressions are ill-defined: these
279       include "0**0", "1**Inf", and "Inf**0".  Do not expect any particular
280       results from these special cases, the results are platform-dependent.
281
282   Symbolic Unary Operators
283       Unary "!" performs logical negation, that is, "not".  See also "not"
284       for a lower precedence version of this.
285
286       Unary "-" performs arithmetic negation if the operand is numeric,
287       including any string that looks like a number.  If the operand is an
288       identifier, a string consisting of a minus sign concatenated with the
289       identifier is returned.  Otherwise, if the string starts with a plus or
290       minus, a string starting with the opposite sign is returned.  One
291       effect of these rules is that "-bareword" is equivalent to the string
292       "-bareword".  If, however, the string begins with a non-alphabetic
293       character (excluding "+" or "-"), Perl will attempt to convert the
294       string to a numeric, and the arithmetic negation is performed.  If the
295       string cannot be cleanly converted to a numeric, Perl will give the
296       warning Argument "the string" isn't numeric in negation (-) at ....
297
298       Unary "~" performs bitwise negation, that is, 1's complement.  For
299       example, "0666 & ~027" is 0640.  (See also "Integer Arithmetic" and
300       "Bitwise String Operators".)  Note that the width of the result is
301       platform-dependent: "~0" is 32 bits wide on a 32-bit platform, but 64
302       bits wide on a 64-bit platform, so if you are expecting a certain bit
303       width, remember to use the "&" operator to mask off the excess bits.
304
305       Starting in Perl 5.28, it is a fatal error to try to complement a
306       string containing a character with an ordinal value above 255.
307
308       If the "bitwise" feature is enabled via "use feature 'bitwise'" or "use
309       v5.28", then unary "~" always treats its argument as a number, and an
310       alternate form of the operator, "~.", always treats its argument as a
311       string.  So "~0" and "~"0"" will both give 2**32-1 on 32-bit platforms,
312       whereas "~.0" and "~."0"" will both yield "\xff".  Until Perl 5.28,
313       this feature produced a warning in the "experimental::bitwise"
314       category.
315
316       Unary "+" has no effect whatsoever, even on strings.  It is useful
317       syntactically for separating a function name from a parenthesized
318       expression that would otherwise be interpreted as the complete list of
319       function arguments.  (See examples above under "Terms and List
320       Operators (Leftward)".)
321
322       Unary "\" creates references.  If its operand is a single sigilled
323       thing, it creates a reference to that object.  If its operand is a
324       parenthesised list, then it creates references to the things mentioned
325       in the list.  Otherwise it puts its operand in list context, and
326       creates a list of references to the scalars in the list provided by the
327       operand.  See perlreftut and perlref.  Do not confuse this behavior
328       with the behavior of backslash within a string, although both forms do
329       convey the notion of protecting the next thing from interpolation.
330
331   Binding Operators
332       Binary "=~" binds a scalar expression to a pattern match.  Certain
333       operations search or modify the string $_ by default.  This operator
334       makes that kind of operation work on some other string.  The right
335       argument is a search pattern, substitution, or transliteration.  The
336       left argument is what is supposed to be searched, substituted, or
337       transliterated instead of the default $_.  When used in scalar context,
338       the return value generally indicates the success of the operation.  The
339       exceptions are substitution ("s///") and transliteration ("y///") with
340       the "/r" (non-destructive) option, which cause the return value to be
341       the result of the substitution.  Behavior in list context depends on
342       the particular operator.  See "Regexp Quote-Like Operators" for details
343       and perlretut for examples using these operators.
344
345       If the right argument is an expression rather than a search pattern,
346       substitution, or transliteration, it is interpreted as a search pattern
347       at run time.  Note that this means that its contents will be
348       interpolated twice, so
349
350           '\\' =~ q'\\';
351
352       is not ok, as the regex engine will end up trying to compile the
353       pattern "\", which it will consider a syntax error.
354
355       Binary "!~" is just like "=~" except the return value is negated in the
356       logical sense.
357
358       Binary "!~" with a non-destructive substitution ("s///r") or
359       transliteration ("y///r") is a syntax error.
360
361   Multiplicative Operators
362       Binary "*" multiplies two numbers.
363
364       Binary "/" divides two numbers.
365
366       Binary "%" is the modulo operator, which computes the division
367       remainder of its first argument with respect to its second argument.
368       Given integer operands $m and $n: If $n is positive, then "$m % $n" is
369       $m minus the largest multiple of $n less than or equal to $m.  If $n is
370       negative, then "$m % $n" is $m minus the smallest multiple of $n that
371       is not less than $m (that is, the result will be less than or equal to
372       zero).  If the operands $m and $n are floating point values and the
373       absolute value of $n (that is "abs($n)") is less than "(UV_MAX + 1)",
374       only the integer portion of $m and $n will be used in the operation
375       (Note: here "UV_MAX" means the maximum of the unsigned integer type).
376       If the absolute value of the right operand ("abs($n)") is greater than
377       or equal to "(UV_MAX + 1)", "%" computes the floating-point remainder
378       $r in the equation "($r = $m - $i*$n)" where $i is a certain integer
379       that makes $r have the same sign as the right operand $n (not as the
380       left operand $m like C function "fmod()") and the absolute value less
381       than that of $n.  Note that when "use integer" is in scope, "%" gives
382       you direct access to the modulo operator as implemented by your C
383       compiler.  This operator is not as well defined for negative operands,
384       but it will execute faster.
385
386       Binary "x" is the repetition operator.  In scalar context, or if the
387       left operand is neither enclosed in parentheses nor a "qw//" list, it
388       performs a string repetition.  In that case it supplies scalar context
389       to the left operand, and returns a string consisting of the left
390       operand string repeated the number of times specified by the right
391       operand.  If the "x" is in list context, and the left operand is either
392       enclosed in parentheses or a "qw//" list, it performs a list
393       repetition.  In that case it supplies list context to the left operand,
394       and returns a list consisting of the left operand list repeated the
395       number of times specified by the right operand.  If the right operand
396       is zero or negative (raising a warning on negative), it returns an
397       empty string or an empty list, depending on the context.
398
399           print '-' x 80;             # print row of dashes
400
401           print "\t" x ($tab/8), ' ' x ($tab%8);      # tab over
402
403           @ones = (1) x 80;           # a list of 80 1's
404           @ones = (5) x @ones;        # set all elements to 5
405
406   Additive Operators
407       Binary "+" returns the sum of two numbers.
408
409       Binary "-" returns the difference of two numbers.
410
411       Binary "." concatenates two strings.
412
413   Shift Operators
414       Binary "<<" returns the value of its left argument shifted left by the
415       number of bits specified by the right argument.  Arguments should be
416       integers.  (See also "Integer Arithmetic".)
417
418       Binary ">>" returns the value of its left argument shifted right by the
419       number of bits specified by the right argument.  Arguments should be
420       integers.  (See also "Integer Arithmetic".)
421
422       If "use integer" (see "Integer Arithmetic") is in force then signed C
423       integers are used (arithmetic shift), otherwise unsigned C integers are
424       used (logical shift), even for negative shiftees.  In arithmetic right
425       shift the sign bit is replicated on the left, in logical shift zero
426       bits come in from the left.
427
428       Either way, the implementation isn't going to generate results larger
429       than the size of the integer type Perl was built with (32 bits or 64
430       bits).
431
432       Shifting by negative number of bits means the reverse shift: left shift
433       becomes right shift, right shift becomes left shift.  This is unlike in
434       C, where negative shift is undefined.
435
436       Shifting by more bits than the size of the integers means most of the
437       time zero (all bits fall off), except that under "use integer" right
438       overshifting a negative shiftee results in -1.  This is unlike in C,
439       where shifting by too many bits is undefined.  A common C behavior is
440       "shift by modulo wordbits", so that for example
441
442           1 >> 64 == 1 >> (64 % 64) == 1 >> 0 == 1  # Common C behavior.
443
444       but that is completely accidental.
445
446       If you get tired of being subject to your platform's native integers,
447       the "use bigint" pragma neatly sidesteps the issue altogether:
448
449           print 20 << 20;  # 20971520
450           print 20 << 40;  # 5120 on 32-bit machines,
451                            # 21990232555520 on 64-bit machines
452           use bigint;
453           print 20 << 100; # 25353012004564588029934064107520
454
455   Named Unary Operators
456       The various named unary operators are treated as functions with one
457       argument, with optional parentheses.
458
459       If any list operator ("print()", etc.) or any unary operator
460       ("chdir()", etc.)  is followed by a left parenthesis as the next token,
461       the operator and arguments within parentheses are taken to be of
462       highest precedence, just like a normal function call.  For example,
463       because named unary operators are higher precedence than "||":
464
465           chdir $foo    || die;       # (chdir $foo) || die
466           chdir($foo)   || die;       # (chdir $foo) || die
467           chdir ($foo)  || die;       # (chdir $foo) || die
468           chdir +($foo) || die;       # (chdir $foo) || die
469
470       but, because "*" is higher precedence than named operators:
471
472           chdir $foo * 20;    # chdir ($foo * 20)
473           chdir($foo) * 20;   # (chdir $foo) * 20
474           chdir ($foo) * 20;  # (chdir $foo) * 20
475           chdir +($foo) * 20; # chdir ($foo * 20)
476
477           rand 10 * 20;       # rand (10 * 20)
478           rand(10) * 20;      # (rand 10) * 20
479           rand (10) * 20;     # (rand 10) * 20
480           rand +(10) * 20;    # rand (10 * 20)
481
482       Regarding precedence, the filetest operators, like "-f", "-M", etc. are
483       treated like named unary operators, but they don't follow this
484       functional parenthesis rule.  That means, for example, that
485       "-f($file).".bak"" is equivalent to "-f "$file.bak"".
486
487       See also "Terms and List Operators (Leftward)".
488
489   Relational Operators
490       Perl operators that return true or false generally return values that
491       can be safely used as numbers.  For example, the relational operators
492       in this section and the equality operators in the next one return 1 for
493       true and a special version of the defined empty string, "", which
494       counts as a zero but is exempt from warnings about improper numeric
495       conversions, just as "0 but true" is.
496
497       Binary "<" returns true if the left argument is numerically less than
498       the right argument.
499
500       Binary ">" returns true if the left argument is numerically greater
501       than the right argument.
502
503       Binary "<=" returns true if the left argument is numerically less than
504       or equal to the right argument.
505
506       Binary ">=" returns true if the left argument is numerically greater
507       than or equal to the right argument.
508
509       Binary "lt" returns true if the left argument is stringwise less than
510       the right argument.
511
512       Binary "gt" returns true if the left argument is stringwise greater
513       than the right argument.
514
515       Binary "le" returns true if the left argument is stringwise less than
516       or equal to the right argument.
517
518       Binary "ge" returns true if the left argument is stringwise greater
519       than or equal to the right argument.
520
521       A sequence of relational operators, such as "$x < $y <= $z", performs
522       chained comparisons, in the manner described above in the section
523       "Operator Precedence and Associativity".  Beware that they do not chain
524       with equality operators, which have lower precedence.
525
526   Equality Operators
527       Binary "==" returns true if the left argument is numerically equal to
528       the right argument.
529
530       Binary "!=" returns true if the left argument is numerically not equal
531       to the right argument.
532
533       Binary "eq" returns true if the left argument is stringwise equal to
534       the right argument.
535
536       Binary "ne" returns true if the left argument is stringwise not equal
537       to the right argument.
538
539       A sequence of the above equality operators, such as "$x == $y == $z",
540       performs chained comparisons, in the manner described above in the
541       section "Operator Precedence and Associativity".  Beware that they do
542       not chain with relational operators, which have higher precedence.
543
544       Binary "<=>" returns -1, 0, or 1 depending on whether the left argument
545       is numerically less than, equal to, or greater than the right argument.
546       If your platform supports "NaN"'s (not-a-numbers) as numeric values,
547       using them with "<=>" returns undef.  "NaN" is not "<", "==", ">", "<="
548       or ">=" anything (even "NaN"), so those 5 return false.  "NaN != NaN"
549       returns true, as does "NaN !=" anything else.  If your platform doesn't
550       support "NaN"'s then "NaN" is just a string with numeric value 0.
551
552           $ perl -le '$x = "NaN"; print "No NaN support here" if $x == $x'
553           $ perl -le '$x = "NaN"; print "NaN support here" if $x != $x'
554
555       (Note that the bigint, bigrat, and bignum pragmas all support "NaN".)
556
557       Binary "cmp" returns -1, 0, or 1 depending on whether the left argument
558       is stringwise less than, equal to, or greater than the right argument.
559
560       Binary "~~" does a smartmatch between its arguments.  Smart matching is
561       described in the next section.
562
563       The two-sided ordering operators "<=>" and "cmp", and the smartmatch
564       operator "~~", are non-associative with respect to each other and with
565       respect to the equality operators of the same precedence.
566
567       "lt", "le", "ge", "gt" and "cmp" use the collation (sort) order
568       specified by the current "LC_COLLATE" locale if a "use locale" form
569       that includes collation is in effect.  See perllocale.  Do not mix
570       these with Unicode, only use them with legacy 8-bit locale encodings.
571       The standard "Unicode::Collate" and "Unicode::Collate::Locale" modules
572       offer much more powerful solutions to collation issues.
573
574       For case-insensitive comparisons, look at the "fc" in perlfunc case-
575       folding function, available in Perl v5.16 or later:
576
577           if ( fc($x) eq fc($y) ) { ... }
578
579   Class Instance Operator
580       Binary "isa" evaluates to true when the left argument is an object
581       instance of the class (or a subclass derived from that class) given by
582       the right argument.  If the left argument is not defined, not a blessed
583       object instance, nor does not derive from the class given by the right
584       argument, the operator evaluates as false. The right argument may give
585       the class either as a bareword or a scalar expression that yields a
586       string class name:
587
588           if( $obj isa Some::Class ) { ... }
589
590           if( $obj isa "Different::Class" ) { ... }
591           if( $obj isa $name_of_class ) { ... }
592
593       This is an experimental feature and is available from Perl 5.31.6 when
594       enabled by "use feature 'isa'". It emits a warning in the
595       "experimental::isa" category.
596
597   Smartmatch Operator
598       First available in Perl 5.10.1 (the 5.10.0 version behaved
599       differently), binary "~~" does a "smartmatch" between its arguments.
600       This is mostly used implicitly in the "when" construct described in
601       perlsyn, although not all "when" clauses call the smartmatch operator.
602       Unique among all of Perl's operators, the smartmatch operator can
603       recurse.  The smartmatch operator is experimental and its behavior is
604       subject to change.
605
606       It is also unique in that all other Perl operators impose a context
607       (usually string or numeric context) on their operands, autoconverting
608       those operands to those imposed contexts.  In contrast, smartmatch
609       infers contexts from the actual types of its operands and uses that
610       type information to select a suitable comparison mechanism.
611
612       The "~~" operator compares its operands "polymorphically", determining
613       how to compare them according to their actual types (numeric, string,
614       array, hash, etc.).  Like the equality operators with which it shares
615       the same precedence, "~~" returns 1 for true and "" for false.  It is
616       often best read aloud as "in", "inside of", or "is contained in",
617       because the left operand is often looked for inside the right operand.
618       That makes the order of the operands to the smartmatch operand often
619       opposite that of the regular match operator.  In other words, the
620       "smaller" thing is usually placed in the left operand and the larger
621       one in the right.
622
623       The behavior of a smartmatch depends on what type of things its
624       arguments are, as determined by the following table.  The first row of
625       the table whose types apply determines the smartmatch behavior.
626       Because what actually happens is mostly determined by the type of the
627       second operand, the table is sorted on the right operand instead of on
628       the left.
629
630        Left      Right      Description and pseudocode
631        ===============================================================
632        Any       undef      check whether Any is undefined
633                       like: !defined Any
634
635        Any       Object     invoke ~~ overloading on Object, or die
636
637        Right operand is an ARRAY:
638
639        Left      Right      Description and pseudocode
640        ===============================================================
641        ARRAY1    ARRAY2     recurse on paired elements of ARRAY1 and ARRAY2[2]
642                       like: (ARRAY1[0] ~~ ARRAY2[0])
643                               && (ARRAY1[1] ~~ ARRAY2[1]) && ...
644        HASH      ARRAY      any ARRAY elements exist as HASH keys
645                       like: grep { exists HASH->{$_} } ARRAY
646        Regexp    ARRAY      any ARRAY elements pattern match Regexp
647                       like: grep { /Regexp/ } ARRAY
648        undef     ARRAY      undef in ARRAY
649                       like: grep { !defined } ARRAY
650        Any       ARRAY      smartmatch each ARRAY element[3]
651                       like: grep { Any ~~ $_ } ARRAY
652
653        Right operand is a HASH:
654
655        Left      Right      Description and pseudocode
656        ===============================================================
657        HASH1     HASH2      all same keys in both HASHes
658                       like: keys HASH1 ==
659                                grep { exists HASH2->{$_} } keys HASH1
660        ARRAY     HASH       any ARRAY elements exist as HASH keys
661                       like: grep { exists HASH->{$_} } ARRAY
662        Regexp    HASH       any HASH keys pattern match Regexp
663                       like: grep { /Regexp/ } keys HASH
664        undef     HASH       always false (undef can't be a key)
665                       like: 0 == 1
666        Any       HASH       HASH key existence
667                       like: exists HASH->{Any}
668
669        Right operand is CODE:
670
671        Left      Right      Description and pseudocode
672        ===============================================================
673        ARRAY     CODE       sub returns true on all ARRAY elements[1]
674                       like: !grep { !CODE->($_) } ARRAY
675        HASH      CODE       sub returns true on all HASH keys[1]
676                       like: !grep { !CODE->($_) } keys HASH
677        Any       CODE       sub passed Any returns true
678                       like: CODE->(Any)
679
680       Right operand is a Regexp:
681
682        Left      Right      Description and pseudocode
683        ===============================================================
684        ARRAY     Regexp     any ARRAY elements match Regexp
685                       like: grep { /Regexp/ } ARRAY
686        HASH      Regexp     any HASH keys match Regexp
687                       like: grep { /Regexp/ } keys HASH
688        Any       Regexp     pattern match
689                       like: Any =~ /Regexp/
690
691        Other:
692
693        Left      Right      Description and pseudocode
694        ===============================================================
695        Object    Any        invoke ~~ overloading on Object,
696                             or fall back to...
697
698        Any       Num        numeric equality
699                        like: Any == Num
700        Num       nummy[4]    numeric equality
701                        like: Num == nummy
702        undef     Any        check whether undefined
703                        like: !defined(Any)
704        Any       Any        string equality
705                        like: Any eq Any
706
707       Notes:
708
709       1. Empty hashes or arrays match.
710       2. That is, each element smartmatches the element of the same index in
711       the other array.[3]
712       3. If a circular reference is found, fall back to referential equality.
713       4. Either an actual number, or a string that looks like one.
714
715       The smartmatch implicitly dereferences any non-blessed hash or array
716       reference, so the "HASH" and "ARRAY" entries apply in those cases.  For
717       blessed references, the "Object" entries apply.  Smartmatches involving
718       hashes only consider hash keys, never hash values.
719
720       The "like" code entry is not always an exact rendition.  For example,
721       the smartmatch operator short-circuits whenever possible, but "grep"
722       does not.  Also, "grep" in scalar context returns the number of
723       matches, but "~~" returns only true or false.
724
725       Unlike most operators, the smartmatch operator knows to treat "undef"
726       specially:
727
728           use v5.10.1;
729           @array = (1, 2, 3, undef, 4, 5);
730           say "some elements undefined" if undef ~~ @array;
731
732       Each operand is considered in a modified scalar context, the
733       modification being that array and hash variables are passed by
734       reference to the operator, which implicitly dereferences them.  Both
735       elements of each pair are the same:
736
737           use v5.10.1;
738
739           my %hash = (red    => 1, blue   => 2, green  => 3,
740                       orange => 4, yellow => 5, purple => 6,
741                       black  => 7, grey   => 8, white  => 9);
742
743           my @array = qw(red blue green);
744
745           say "some array elements in hash keys" if  @array ~~  %hash;
746           say "some array elements in hash keys" if \@array ~~ \%hash;
747
748           say "red in array" if "red" ~~  @array;
749           say "red in array" if "red" ~~ \@array;
750
751           say "some keys end in e" if /e$/ ~~  %hash;
752           say "some keys end in e" if /e$/ ~~ \%hash;
753
754       Two arrays smartmatch if each element in the first array smartmatches
755       (that is, is "in") the corresponding element in the second array,
756       recursively.
757
758           use v5.10.1;
759           my @little = qw(red blue green);
760           my @bigger = ("red", "blue", [ "orange", "green" ] );
761           if (@little ~~ @bigger) {  # true!
762               say "little is contained in bigger";
763           }
764
765       Because the smartmatch operator recurses on nested arrays, this will
766       still report that "red" is in the array.
767
768           use v5.10.1;
769           my @array = qw(red blue green);
770           my $nested_array = [[[[[[[ @array ]]]]]]];
771           say "red in array" if "red" ~~ $nested_array;
772
773       If two arrays smartmatch each other, then they are deep copies of each
774       others' values, as this example reports:
775
776           use v5.12.0;
777           my @a = (0, 1, 2, [3, [4, 5], 6], 7);
778           my @b = (0, 1, 2, [3, [4, 5], 6], 7);
779
780           if (@a ~~ @b && @b ~~ @a) {
781               say "a and b are deep copies of each other";
782           }
783           elsif (@a ~~ @b) {
784               say "a smartmatches in b";
785           }
786           elsif (@b ~~ @a) {
787               say "b smartmatches in a";
788           }
789           else {
790               say "a and b don't smartmatch each other at all";
791           }
792
793       If you were to set "$b[3] = 4", then instead of reporting that "a and b
794       are deep copies of each other", it now reports that "b smartmatches in
795       a".  That's because the corresponding position in @a contains an array
796       that (eventually) has a 4 in it.
797
798       Smartmatching one hash against another reports whether both contain the
799       same keys, no more and no less.  This could be used to see whether two
800       records have the same field names, without caring what values those
801       fields might have.  For example:
802
803           use v5.10.1;
804           sub make_dogtag {
805               state $REQUIRED_FIELDS = { name=>1, rank=>1, serial_num=>1 };
806
807               my ($class, $init_fields) = @_;
808
809               die "Must supply (only) name, rank, and serial number"
810                   unless $init_fields ~~ $REQUIRED_FIELDS;
811
812               ...
813           }
814
815       However, this only does what you mean if $init_fields is indeed a hash
816       reference. The condition "$init_fields ~~ $REQUIRED_FIELDS" also allows
817       the strings "name", "rank", "serial_num" as well as any array reference
818       that contains "name" or "rank" or "serial_num" anywhere to pass
819       through.
820
821       The smartmatch operator is most often used as the implicit operator of
822       a "when" clause.  See the section on "Switch Statements" in perlsyn.
823
824       Smartmatching of Objects
825
826       To avoid relying on an object's underlying representation, if the
827       smartmatch's right operand is an object that doesn't overload "~~", it
828       raises the exception ""Smartmatching a non-overloaded object breaks
829       encapsulation"".  That's because one has no business digging around to
830       see whether something is "in" an object.  These are all illegal on
831       objects without a "~~" overload:
832
833           %hash ~~ $object
834              42 ~~ $object
835          "fred" ~~ $object
836
837       However, you can change the way an object is smartmatched by
838       overloading the "~~" operator.  This is allowed to extend the usual
839       smartmatch semantics.  For objects that do have an "~~" overload, see
840       overload.
841
842       Using an object as the left operand is allowed, although not very
843       useful.  Smartmatching rules take precedence over overloading, so even
844       if the object in the left operand has smartmatch overloading, this will
845       be ignored.  A left operand that is a non-overloaded object falls back
846       on a string or numeric comparison of whatever the "ref" operator
847       returns.  That means that
848
849           $object ~~ X
850
851       does not invoke the overload method with "X" as an argument.  Instead
852       the above table is consulted as normal, and based on the type of "X",
853       overloading may or may not be invoked.  For simple strings or numbers,
854       "in" becomes equivalent to this:
855
856           $object ~~ $number          ref($object) == $number
857           $object ~~ $string          ref($object) eq $string
858
859       For example, this reports that the handle smells IOish (but please
860       don't really do this!):
861
862           use IO::Handle;
863           my $fh = IO::Handle->new();
864           if ($fh ~~ /\bIO\b/) {
865               say "handle smells IOish";
866           }
867
868       That's because it treats $fh as a string like
869       "IO::Handle=GLOB(0x8039e0)", then pattern matches against that.
870
871   Bitwise And
872       Binary "&" returns its operands ANDed together bit by bit.  Although no
873       warning is currently raised, the result is not well defined when this
874       operation is performed on operands that aren't either numbers (see
875       "Integer Arithmetic") nor bitstrings (see "Bitwise String Operators").
876
877       Note that "&" has lower priority than relational operators, so for
878       example the parentheses are essential in a test like
879
880           print "Even\n" if ($x & 1) == 0;
881
882       If the "bitwise" feature is enabled via "use feature 'bitwise'" or "use
883       v5.28", then this operator always treats its operands as numbers.
884       Before Perl 5.28 this feature produced a warning in the
885       "experimental::bitwise" category.
886
887   Bitwise Or and Exclusive Or
888       Binary "|" returns its operands ORed together bit by bit.
889
890       Binary "^" returns its operands XORed together bit by bit.
891
892       Although no warning is currently raised, the results are not well
893       defined when these operations are performed on operands that aren't
894       either numbers (see "Integer Arithmetic") nor bitstrings (see "Bitwise
895       String Operators").
896
897       Note that "|" and "^" have lower priority than relational operators, so
898       for example the parentheses are essential in a test like
899
900           print "false\n" if (8 | 2) != 10;
901
902       If the "bitwise" feature is enabled via "use feature 'bitwise'" or "use
903       v5.28", then this operator always treats its operands as numbers.
904       Before Perl 5.28. this feature produced a warning in the
905       "experimental::bitwise" category.
906
907   C-style Logical And
908       Binary "&&" performs a short-circuit logical AND operation.  That is,
909       if the left operand is false, the right operand is not even evaluated.
910       Scalar or list context propagates down to the right operand if it is
911       evaluated.
912
913   C-style Logical Or
914       Binary "||" performs a short-circuit logical OR operation.  That is, if
915       the left operand is true, the right operand is not even evaluated.
916       Scalar or list context propagates down to the right operand if it is
917       evaluated.
918
919   Logical Defined-Or
920       Although it has no direct equivalent in C, Perl's "//" operator is
921       related to its C-style "or".  In fact, it's exactly the same as "||",
922       except that it tests the left hand side's definedness instead of its
923       truth.  Thus, "EXPR1 // EXPR2" returns the value of "EXPR1" if it's
924       defined, otherwise, the value of "EXPR2" is returned.  ("EXPR1" is
925       evaluated in scalar context, "EXPR2" in the context of "//" itself).
926       Usually, this is the same result as "defined(EXPR1) ? EXPR1 : EXPR2"
927       (except that the ternary-operator form can be used as a lvalue, while
928       "EXPR1 // EXPR2" cannot).  This is very useful for providing default
929       values for variables.  If you actually want to test if at least one of
930       $x and $y is defined, use "defined($x // $y)".
931
932       The "||", "//" and "&&" operators return the last value evaluated
933       (unlike C's "||" and "&&", which return 0 or 1).  Thus, a reasonably
934       portable way to find out the home directory might be:
935
936           $home =  $ENV{HOME}
937                 // $ENV{LOGDIR}
938                 // (getpwuid($<))[7]
939                 // die "You're homeless!\n";
940
941       In particular, this means that you shouldn't use this for selecting
942       between two aggregates for assignment:
943
944           @a = @b || @c;            # This doesn't do the right thing
945           @a = scalar(@b) || @c;    # because it really means this.
946           @a = @b ? @b : @c;        # This works fine, though.
947
948       As alternatives to "&&" and "||" when used for control flow, Perl
949       provides the "and" and "or" operators (see below).  The short-circuit
950       behavior is identical.  The precedence of "and" and "or" is much lower,
951       however, so that you can safely use them after a list operator without
952       the need for parentheses:
953
954           unlink "alpha", "beta", "gamma"
955                   or gripe(), next LINE;
956
957       With the C-style operators that would have been written like this:
958
959           unlink("alpha", "beta", "gamma")
960                   || (gripe(), next LINE);
961
962       It would be even more readable to write that this way:
963
964           unless(unlink("alpha", "beta", "gamma")) {
965               gripe();
966               next LINE;
967           }
968
969       Using "or" for assignment is unlikely to do what you want; see below.
970
971   Range Operators
972       Binary ".." is the range operator, which is really two different
973       operators depending on the context.  In list context, it returns a list
974       of values counting (up by ones) from the left value to the right value.
975       If the left value is greater than the right value then it returns the
976       empty list.  The range operator is useful for writing "foreach (1..10)"
977       loops and for doing slice operations on arrays.  In the current
978       implementation, no temporary array is created when the range operator
979       is used as the expression in "foreach" loops, but older versions of
980       Perl might burn a lot of memory when you write something like this:
981
982           for (1 .. 1_000_000) {
983               # code
984           }
985
986       The range operator also works on strings, using the magical auto-
987       increment, see below.
988
989       In scalar context, ".." returns a boolean value.  The operator is
990       bistable, like a flip-flop, and emulates the line-range (comma)
991       operator of sed, awk, and various editors.  Each ".." operator
992       maintains its own boolean state, even across calls to a subroutine that
993       contains it.  It is false as long as its left operand is false.  Once
994       the left operand is true, the range operator stays true until the right
995       operand is true, AFTER which the range operator becomes false again.
996       It doesn't become false till the next time the range operator is
997       evaluated.  It can test the right operand and become false on the same
998       evaluation it became true (as in awk), but it still returns true once.
999       If you don't want it to test the right operand until the next
1000       evaluation, as in sed, just use three dots ("...") instead of two.  In
1001       all other regards, "..." behaves just like ".." does.
1002
1003       The right operand is not evaluated while the operator is in the "false"
1004       state, and the left operand is not evaluated while the operator is in
1005       the "true" state.  The precedence is a little lower than || and &&.
1006       The value returned is either the empty string for false, or a sequence
1007       number (beginning with 1) for true.  The sequence number is reset for
1008       each range encountered.  The final sequence number in a range has the
1009       string "E0" appended to it, which doesn't affect its numeric value, but
1010       gives you something to search for if you want to exclude the endpoint.
1011       You can exclude the beginning point by waiting for the sequence number
1012       to be greater than 1.
1013
1014       If either operand of scalar ".." is a constant expression, that operand
1015       is considered true if it is equal ("==") to the current input line
1016       number (the $. variable).
1017
1018       To be pedantic, the comparison is actually "int(EXPR) == int(EXPR)",
1019       but that is only an issue if you use a floating point expression; when
1020       implicitly using $. as described in the previous paragraph, the
1021       comparison is "int(EXPR) == int($.)" which is only an issue when $.  is
1022       set to a floating point value and you are not reading from a file.
1023       Furthermore, "span" .. "spat" or "2.18 .. 3.14" will not do what you
1024       want in scalar context because each of the operands are evaluated using
1025       their integer representation.
1026
1027       Examples:
1028
1029       As a scalar operator:
1030
1031           if (101 .. 200) { print; } # print 2nd hundred lines, short for
1032                                      #  if ($. == 101 .. $. == 200) { print; }
1033
1034           next LINE if (1 .. /^$/);  # skip header lines, short for
1035                                      #   next LINE if ($. == 1 .. /^$/);
1036                                      # (typically in a loop labeled LINE)
1037
1038           s/^/> / if (/^$/ .. eof());  # quote body
1039
1040           # parse mail messages
1041           while (<>) {
1042               $in_header =   1  .. /^$/;
1043               $in_body   = /^$/ .. eof;
1044               if ($in_header) {
1045                   # do something
1046               } else { # in body
1047                   # do something else
1048               }
1049           } continue {
1050               close ARGV if eof;             # reset $. each file
1051           }
1052
1053       Here's a simple example to illustrate the difference between the two
1054       range operators:
1055
1056           @lines = ("   - Foo",
1057                     "01 - Bar",
1058                     "1  - Baz",
1059                     "   - Quux");
1060
1061           foreach (@lines) {
1062               if (/0/ .. /1/) {
1063                   print "$_\n";
1064               }
1065           }
1066
1067       This program will print only the line containing "Bar".  If the range
1068       operator is changed to "...", it will also print the "Baz" line.
1069
1070       And now some examples as a list operator:
1071
1072           for (101 .. 200) { print }      # print $_ 100 times
1073           @foo = @foo[0 .. $#foo];        # an expensive no-op
1074           @foo = @foo[$#foo-4 .. $#foo];  # slice last 5 items
1075
1076       Because each operand is evaluated in integer form, "2.18 .. 3.14" will
1077       return two elements in list context.
1078
1079           @list = (2.18 .. 3.14); # same as @list = (2 .. 3);
1080
1081       The range operator in list context can make use of the magical auto-
1082       increment algorithm if both operands are strings, subject to the
1083       following rules:
1084
1085       •   With one exception (below), if both strings look like numbers to
1086           Perl, the magic increment will not be applied, and the strings will
1087           be treated as numbers (more specifically, integers) instead.
1088
1089           For example, "-2".."2" is the same as "-2..2", and "2.18".."3.14"
1090           produces "2, 3".
1091
1092       •   The exception to the above rule is when the left-hand string begins
1093           with 0 and is longer than one character, in this case the magic
1094           increment will be applied, even though strings like "01" would
1095           normally look like a number to Perl.
1096
1097           For example, "01".."04" produces "01", "02", "03", "04", and
1098           "00".."-1" produces "00" through "99" - this may seem surprising,
1099           but see the following rules for why it works this way.  To get
1100           dates with leading zeros, you can say:
1101
1102               @z2 = ("01" .. "31");
1103               print $z2[$mday];
1104
1105           If you want to force strings to be interpreted as numbers, you
1106           could say
1107
1108               @numbers = ( 0+$first .. 0+$last );
1109
1110           Note: In Perl versions 5.30 and below, any string on the left-hand
1111           side beginning with "0", including the string "0" itself, would
1112           cause the magic string increment behavior. This means that on these
1113           Perl versions, "0".."-1" would produce "0" through "99", which was
1114           inconsistent with "0..-1", which produces the empty list. This also
1115           means that "0".."9" now produces a list of integers instead of a
1116           list of strings.
1117
1118       •   If the initial value specified isn't part of a magical increment
1119           sequence (that is, a non-empty string matching
1120           "/^[a-zA-Z]*[0-9]*\z/"), only the initial value will be returned.
1121
1122           For example, "ax".."az" produces "ax", "ay", "az", but "*x".."az"
1123           produces only "*x".
1124
1125       •   For other initial values that are strings that do follow the rules
1126           of the magical increment, the corresponding sequence will be
1127           returned.
1128
1129           For example, you can say
1130
1131               @alphabet = ("A" .. "Z");
1132
1133           to get all normal letters of the English alphabet, or
1134
1135               $hexdigit = (0 .. 9, "a" .. "f")[$num & 15];
1136
1137           to get a hexadecimal digit.
1138
1139       •   If the final value specified is not in the sequence that the
1140           magical increment would produce, the sequence goes until the next
1141           value would be longer than the final value specified. If the length
1142           of the final string is shorter than the first, the empty list is
1143           returned.
1144
1145           For example, "a".."--" is the same as "a".."zz", "0".."xx" produces
1146           "0" through "99", and "aaa".."--" returns the empty list.
1147
1148       As of Perl 5.26, the list-context range operator on strings works as
1149       expected in the scope of "use feature 'unicode_strings". In previous
1150       versions, and outside the scope of that feature, it exhibits "The
1151       "Unicode Bug"" in perlunicode: its behavior depends on the internal
1152       encoding of the range endpoint.
1153
1154       Because the magical increment only works on non-empty strings matching
1155       "/^[a-zA-Z]*[0-9]*\z/", the following will only return an alpha:
1156
1157           use charnames "greek";
1158           my @greek_small =  ("\N{alpha}" .. "\N{omega}");
1159
1160       To get the 25 traditional lowercase Greek letters, including both
1161       sigmas, you could use this instead:
1162
1163           use charnames "greek";
1164           my @greek_small =  map { chr } ( ord("\N{alpha}")
1165                                               ..
1166                                            ord("\N{omega}")
1167                                          );
1168
1169       However, because there are many other lowercase Greek characters than
1170       just those, to match lowercase Greek characters in a regular
1171       expression, you could use the pattern "/(?:(?=\p{Greek})\p{Lower})+/"
1172       (or the experimental feature "/(?[ \p{Greek} & \p{Lower} ])+/").
1173
1174   Conditional Operator
1175       Ternary "?:" is the conditional operator, just as in C.  It works much
1176       like an if-then-else.  If the argument before the "?" is true, the
1177       argument before the ":" is returned, otherwise the argument after the
1178       ":" is returned.  For example:
1179
1180           printf "I have %d dog%s.\n", $n,
1181                   ($n == 1) ? "" : "s";
1182
1183       Scalar or list context propagates downward into the 2nd or 3rd
1184       argument, whichever is selected.
1185
1186           $x = $ok ? $y : $z;  # get a scalar
1187           @x = $ok ? @y : @z;  # get an array
1188           $x = $ok ? @y : @z;  # oops, that's just a count!
1189
1190       The operator may be assigned to if both the 2nd and 3rd arguments are
1191       legal lvalues (meaning that you can assign to them):
1192
1193           ($x_or_y ? $x : $y) = $z;
1194
1195       Because this operator produces an assignable result, using assignments
1196       without parentheses will get you in trouble.  For example, this:
1197
1198           $x % 2 ? $x += 10 : $x += 2
1199
1200       Really means this:
1201
1202           (($x % 2) ? ($x += 10) : $x) += 2
1203
1204       Rather than this:
1205
1206           ($x % 2) ? ($x += 10) : ($x += 2)
1207
1208       That should probably be written more simply as:
1209
1210           $x += ($x % 2) ? 10 : 2;
1211
1212   Assignment Operators
1213       "=" is the ordinary assignment operator.
1214
1215       Assignment operators work as in C.  That is,
1216
1217           $x += 2;
1218
1219       is equivalent to
1220
1221           $x = $x + 2;
1222
1223       although without duplicating any side effects that dereferencing the
1224       lvalue might trigger, such as from "tie()".  Other assignment operators
1225       work similarly.  The following are recognized:
1226
1227           **=    +=    *=    &=    &.=    <<=    &&=
1228                  -=    /=    |=    |.=    >>=    ||=
1229                  .=    %=    ^=    ^.=           //=
1230                        x=
1231
1232       Although these are grouped by family, they all have the precedence of
1233       assignment.  These combined assignment operators can only operate on
1234       scalars, whereas the ordinary assignment operator can assign to arrays,
1235       hashes, lists and even references.  (See "Context" and "List value
1236       constructors" in perldata, and "Assigning to References" in perlref.)
1237
1238       Unlike in C, the scalar assignment operator produces a valid lvalue.
1239       Modifying an assignment is equivalent to doing the assignment and then
1240       modifying the variable that was assigned to.  This is useful for
1241       modifying a copy of something, like this:
1242
1243           ($tmp = $global) =~ tr/13579/24680/;
1244
1245       Although as of 5.14, that can be also be accomplished this way:
1246
1247           use v5.14;
1248           $tmp = ($global =~  tr/13579/24680/r);
1249
1250       Likewise,
1251
1252           ($x += 2) *= 3;
1253
1254       is equivalent to
1255
1256           $x += 2;
1257           $x *= 3;
1258
1259       Similarly, a list assignment in list context produces the list of
1260       lvalues assigned to, and a list assignment in scalar context returns
1261       the number of elements produced by the expression on the right hand
1262       side of the assignment.
1263
1264       The three dotted bitwise assignment operators ("&.=" "|.=" "^.=") are
1265       new in Perl 5.22.  See "Bitwise String Operators".
1266
1267   Comma Operator
1268       Binary "," is the comma operator.  In scalar context it evaluates its
1269       left argument, throws that value away, then evaluates its right
1270       argument and returns that value.  This is just like C's comma operator.
1271
1272       In list context, it's just the list argument separator, and inserts
1273       both its arguments into the list.  These arguments are also evaluated
1274       from left to right.
1275
1276       The "=>" operator (sometimes pronounced "fat comma") is a synonym for
1277       the comma except that it causes a word on its left to be interpreted as
1278       a string if it begins with a letter or underscore and is composed only
1279       of letters, digits and underscores.  This includes operands that might
1280       otherwise be interpreted as operators, constants, single number
1281       v-strings or function calls.  If in doubt about this behavior, the left
1282       operand can be quoted explicitly.
1283
1284       Otherwise, the "=>" operator behaves exactly as the comma operator or
1285       list argument separator, according to context.
1286
1287       For example:
1288
1289           use constant FOO => "something";
1290
1291           my %h = ( FOO => 23 );
1292
1293       is equivalent to:
1294
1295           my %h = ("FOO", 23);
1296
1297       It is NOT:
1298
1299           my %h = ("something", 23);
1300
1301       The "=>" operator is helpful in documenting the correspondence between
1302       keys and values in hashes, and other paired elements in lists.
1303
1304           %hash = ( $key => $value );
1305           login( $username => $password );
1306
1307       The special quoting behavior ignores precedence, and hence may apply to
1308       part of the left operand:
1309
1310           print time.shift => "bbb";
1311
1312       That example prints something like "1314363215shiftbbb", because the
1313       "=>" implicitly quotes the "shift" immediately on its left, ignoring
1314       the fact that "time.shift" is the entire left operand.
1315
1316   List Operators (Rightward)
1317       On the right side of a list operator, the comma has very low
1318       precedence, such that it controls all comma-separated expressions found
1319       there.  The only operators with lower precedence are the logical
1320       operators "and", "or", and "not", which may be used to evaluate calls
1321       to list operators without the need for parentheses:
1322
1323           open HANDLE, "< :encoding(UTF-8)", "filename"
1324               or die "Can't open: $!\n";
1325
1326       However, some people find that code harder to read than writing it with
1327       parentheses:
1328
1329           open(HANDLE, "< :encoding(UTF-8)", "filename")
1330               or die "Can't open: $!\n";
1331
1332       in which case you might as well just use the more customary "||"
1333       operator:
1334
1335           open(HANDLE, "< :encoding(UTF-8)", "filename")
1336               || die "Can't open: $!\n";
1337
1338       See also discussion of list operators in "Terms and List Operators
1339       (Leftward)".
1340
1341   Logical Not
1342       Unary "not" returns the logical negation of the expression to its
1343       right.  It's the equivalent of "!" except for the very low precedence.
1344
1345   Logical And
1346       Binary "and" returns the logical conjunction of the two surrounding
1347       expressions.  It's equivalent to "&&" except for the very low
1348       precedence.  This means that it short-circuits: the right expression is
1349       evaluated only if the left expression is true.
1350
1351   Logical or and Exclusive Or
1352       Binary "or" returns the logical disjunction of the two surrounding
1353       expressions.  It's equivalent to "||" except for the very low
1354       precedence.  This makes it useful for control flow:
1355
1356           print FH $data              or die "Can't write to FH: $!";
1357
1358       This means that it short-circuits: the right expression is evaluated
1359       only if the left expression is false.  Due to its precedence, you must
1360       be careful to avoid using it as replacement for the "||" operator.  It
1361       usually works out better for flow control than in assignments:
1362
1363           $x = $y or $z;              # bug: this is wrong
1364           ($x = $y) or $z;            # really means this
1365           $x = $y || $z;              # better written this way
1366
1367       However, when it's a list-context assignment and you're trying to use
1368       "||" for control flow, you probably need "or" so that the assignment
1369       takes higher precedence.
1370
1371           @info = stat($file) || die;     # oops, scalar sense of stat!
1372           @info = stat($file) or die;     # better, now @info gets its due
1373
1374       Then again, you could always use parentheses.
1375
1376       Binary "xor" returns the exclusive-OR of the two surrounding
1377       expressions.  It cannot short-circuit (of course).
1378
1379       There is no low precedence operator for defined-OR.
1380
1381   C Operators Missing From Perl
1382       Here is what C has that Perl doesn't:
1383
1384       unary & Address-of operator.  (But see the "\" operator for taking a
1385               reference.)
1386
1387       unary * Dereference-address operator.  (Perl's prefix dereferencing
1388               operators are typed: "$", "@", "%", and "&".)
1389
1390       (TYPE)  Type-casting operator.
1391
1392   Quote and Quote-like Operators
1393       While we usually think of quotes as literal values, in Perl they
1394       function as operators, providing various kinds of interpolating and
1395       pattern matching capabilities.  Perl provides customary quote
1396       characters for these behaviors, but also provides a way for you to
1397       choose your quote character for any of them.  In the following table, a
1398       "{}" represents any pair of delimiters you choose.
1399
1400           Customary  Generic        Meaning        Interpolates
1401               ''       q{}          Literal             no
1402               ""      qq{}          Literal             yes
1403               ``      qx{}          Command             yes*
1404                       qw{}         Word list            no
1405               //       m{}       Pattern match          yes*
1406                       qr{}          Pattern             yes*
1407                        s{}{}      Substitution          yes*
1408                       tr{}{}    Transliteration         no (but see below)
1409                        y{}{}    Transliteration         no (but see below)
1410               <<EOF                 here-doc            yes*
1411
1412               * unless the delimiter is ''.
1413
1414       Non-bracketing delimiters use the same character fore and aft, but the
1415       four sorts of ASCII brackets (round, angle, square, curly) all nest,
1416       which means that
1417
1418           q{foo{bar}baz}
1419
1420       is the same as
1421
1422           'foo{bar}baz'
1423
1424       Note, however, that this does not always work for quoting Perl code:
1425
1426           $s = q{ if($x eq "}") ... }; # WRONG
1427
1428       is a syntax error.  The "Text::Balanced" module (standard as of v5.8,
1429       and from CPAN before then) is able to do this properly.
1430
1431       There can (and in some cases, must) be whitespace between the operator
1432       and the quoting characters, except when "#" is being used as the
1433       quoting character.  "q#foo#" is parsed as the string "foo", while
1434       "q #foo#" is the operator "q" followed by a comment.  Its argument will
1435       be taken from the next line.  This allows you to write:
1436
1437           s {foo}  # Replace foo
1438             {bar}  # with bar.
1439
1440       The cases where whitespace must be used are when the quoting character
1441       is a word character (meaning it matches "/\w/"):
1442
1443           q XfooX # Works: means the string 'foo'
1444           qXfooX  # WRONG!
1445
1446       The following escape sequences are available in constructs that
1447       interpolate, and in transliterations whose delimiters aren't single
1448       quotes ("'").
1449
1450           Sequence     Note  Description
1451           \t                  tab               (HT, TAB)
1452           \n                  newline           (NL)
1453           \r                  return            (CR)
1454           \f                  form feed         (FF)
1455           \b                  backspace         (BS)
1456           \a                  alarm (bell)      (BEL)
1457           \e                  escape            (ESC)
1458           \x{263A}     [1,8]  hex char          (example shown: SMILEY)
1459           \x1b         [2,8]  restricted range hex char (example: ESC)
1460           \N{name}     [3]    named Unicode character or character sequence
1461           \N{U+263D}   [4,8]  Unicode character (example: FIRST QUARTER MOON)
1462           \c[          [5]    control char      (example: chr(27))
1463           \o{23072}    [6,8]  octal char        (example: SMILEY)
1464           \033         [7,8]  restricted range octal char  (example: ESC)
1465
1466       [1] The result is the character specified by the hexadecimal number
1467           between the braces.  See "[8]" below for details on which
1468           character.
1469
1470           Only hexadecimal digits are valid between the braces.  If an
1471           invalid character is encountered, a warning will be issued and the
1472           invalid character and all subsequent characters (valid or invalid)
1473           within the braces will be discarded.
1474
1475           If there are no valid digits between the braces, the generated
1476           character is the NULL character ("\x{00}").  However, an explicit
1477           empty brace ("\x{}") will not cause a warning (currently).
1478
1479       [2] The result is the character specified by the hexadecimal number in
1480           the range 0x00 to 0xFF.  See "[8]" below for details on which
1481           character.
1482
1483           Only hexadecimal digits are valid following "\x".  When "\x" is
1484           followed by fewer than two valid digits, any valid digits will be
1485           zero-padded.  This means that "\x7" will be interpreted as "\x07",
1486           and a lone "\x" will be interpreted as "\x00".  Except at the end
1487           of a string, having fewer than two valid digits will result in a
1488           warning.  Note that although the warning says the illegal character
1489           is ignored, it is only ignored as part of the escape and will still
1490           be used as the subsequent character in the string.  For example:
1491
1492             Original    Result    Warns?
1493             "\x7"       "\x07"    no
1494             "\x"        "\x00"    no
1495             "\x7q"      "\x07q"   yes
1496             "\xq"       "\x00q"   yes
1497
1498       [3] The result is the Unicode character or character sequence given by
1499           name.  See charnames.
1500
1501       [4] "\N{U+hexadecimal number}" means the Unicode character whose
1502           Unicode code point is hexadecimal number.
1503
1504       [5] The character following "\c" is mapped to some other character as
1505           shown in the table:
1506
1507            Sequence   Value
1508              \c@      chr(0)
1509              \cA      chr(1)
1510              \ca      chr(1)
1511              \cB      chr(2)
1512              \cb      chr(2)
1513              ...
1514              \cZ      chr(26)
1515              \cz      chr(26)
1516              \c[      chr(27)
1517                                # See below for chr(28)
1518              \c]      chr(29)
1519              \c^      chr(30)
1520              \c_      chr(31)
1521              \c?      chr(127) # (on ASCII platforms; see below for link to
1522                                #  EBCDIC discussion)
1523
1524           In other words, it's the character whose code point has had 64
1525           xor'd with its uppercase.  "\c?" is DELETE on ASCII platforms
1526           because "ord("?") ^ 64" is 127, and "\c@" is NULL because the ord
1527           of "@" is 64, so xor'ing 64 itself produces 0.
1528
1529           Also, "\c\X" yields " chr(28) . "X"" for any X, but cannot come at
1530           the end of a string, because the backslash would be parsed as
1531           escaping the end quote.
1532
1533           On ASCII platforms, the resulting characters from the list above
1534           are the complete set of ASCII controls.  This isn't the case on
1535           EBCDIC platforms; see "OPERATOR DIFFERENCES" in perlebcdic for a
1536           full discussion of the differences between these for ASCII versus
1537           EBCDIC platforms.
1538
1539           Use of any other character following the "c" besides those listed
1540           above is discouraged, and as of Perl v5.20, the only characters
1541           actually allowed are the printable ASCII ones, minus the left brace
1542           "{".  What happens for any of the allowed other characters is that
1543           the value is derived by xor'ing with the seventh bit, which is 64,
1544           and a warning raised if enabled.  Using the non-allowed characters
1545           generates a fatal error.
1546
1547           To get platform independent controls, you can use "\N{...}".
1548
1549       [6] The result is the character specified by the octal number between
1550           the braces.  See "[8]" below for details on which character.
1551
1552           If a character that isn't an octal digit is encountered, a warning
1553           is raised, and the value is based on the octal digits before it,
1554           discarding it and all following characters up to the closing brace.
1555           It is a fatal error if there are no octal digits at all.
1556
1557       [7] The result is the character specified by the three-digit octal
1558           number in the range 000 to 777 (but best to not use above 077, see
1559           next paragraph).  See "[8]" below for details on which character.
1560
1561           Some contexts allow 2 or even 1 digit, but any usage without
1562           exactly three digits, the first being a zero, may give unintended
1563           results.  (For example, in a regular expression it may be confused
1564           with a backreference; see "Octal escapes" in perlrebackslash.)
1565           Starting in Perl 5.14, you may use "\o{}" instead, which avoids all
1566           these problems.  Otherwise, it is best to use this construct only
1567           for ordinals "\077" and below, remembering to pad to the left with
1568           zeros to make three digits.  For larger ordinals, either use
1569           "\o{}", or convert to something else, such as to hex and use
1570           "\N{U+}" (which is portable between platforms with different
1571           character sets) or "\x{}" instead.
1572
1573       [8] Several constructs above specify a character by a number.  That
1574           number gives the character's position in the character set encoding
1575           (indexed from 0).  This is called synonymously its ordinal, code
1576           position, or code point.  Perl works on platforms that have a
1577           native encoding currently of either ASCII/Latin1 or EBCDIC, each of
1578           which allow specification of 256 characters.  In general, if the
1579           number is 255 (0xFF, 0377) or below, Perl interprets this in the
1580           platform's native encoding.  If the number is 256 (0x100, 0400) or
1581           above, Perl interprets it as a Unicode code point and the result is
1582           the corresponding Unicode character.  For example "\x{50}" and
1583           "\o{120}" both are the number 80 in decimal, which is less than
1584           256, so the number is interpreted in the native character set
1585           encoding.  In ASCII the character in the 80th position (indexed
1586           from 0) is the letter "P", and in EBCDIC it is the ampersand symbol
1587           "&".  "\x{100}" and "\o{400}" are both 256 in decimal, so the
1588           number is interpreted as a Unicode code point no matter what the
1589           native encoding is.  The name of the character in the 256th
1590           position (indexed by 0) in Unicode is "LATIN CAPITAL LETTER A WITH
1591           MACRON".
1592
1593           An exception to the above rule is that "\N{U+hex number}" is always
1594           interpreted as a Unicode code point, so that "\N{U+0050}" is "P"
1595           even on EBCDIC platforms.
1596
1597       NOTE: Unlike C and other languages, Perl has no "\v" escape sequence
1598       for the vertical tab (VT, which is 11 in both ASCII and EBCDIC), but
1599       you may use "\N{VT}", "\ck", "\N{U+0b}", or "\x0b".  ("\v" does have
1600       meaning in regular expression patterns in Perl, see perlre.)
1601
1602       The following escape sequences are available in constructs that
1603       interpolate, but not in transliterations.
1604
1605           \l          lowercase next character only
1606           \u          titlecase (not uppercase!) next character only
1607           \L          lowercase all characters till \E or end of string
1608           \U          uppercase all characters till \E or end of string
1609           \F          foldcase all characters till \E or end of string
1610           \Q          quote (disable) pattern metacharacters till \E or
1611                       end of string
1612           \E          end either case modification or quoted section
1613                       (whichever was last seen)
1614
1615       See "quotemeta" in perlfunc for the exact definition of characters that
1616       are quoted by "\Q".
1617
1618       "\L", "\U", "\F", and "\Q" can stack, in which case you need one "\E"
1619       for each.  For example:
1620
1621        say"This \Qquoting \ubusiness \Uhere isn't quite\E done yet,\E is it?";
1622        This quoting\ Business\ HERE\ ISN\'T\ QUITE\ done\ yet\, is it?
1623
1624       If a "use locale" form that includes "LC_CTYPE" is in effect (see
1625       perllocale), the case map used by "\l", "\L", "\u", and "\U" is taken
1626       from the current locale.  If Unicode (for example, "\N{}" or code
1627       points of 0x100 or beyond) is being used, the case map used by "\l",
1628       "\L", "\u", and "\U" is as defined by Unicode.  That means that case-
1629       mapping a single character can sometimes produce a sequence of several
1630       characters.  Under "use locale", "\F" produces the same results as "\L"
1631       for all locales but a UTF-8 one, where it instead uses the Unicode
1632       definition.
1633
1634       All systems use the virtual "\n" to represent a line terminator, called
1635       a "newline".  There is no such thing as an unvarying, physical newline
1636       character.  It is only an illusion that the operating system, device
1637       drivers, C libraries, and Perl all conspire to preserve.  Not all
1638       systems read "\r" as ASCII CR and "\n" as ASCII LF.  For example, on
1639       the ancient Macs (pre-MacOS X) of yesteryear, these used to be
1640       reversed, and on systems without a line terminator, printing "\n" might
1641       emit no actual data.  In general, use "\n" when you mean a "newline"
1642       for your system, but use the literal ASCII when you need an exact
1643       character.  For example, most networking protocols expect and prefer a
1644       CR+LF ("\015\012" or "\cM\cJ") for line terminators, and although they
1645       often accept just "\012", they seldom tolerate just "\015".  If you get
1646       in the habit of using "\n" for networking, you may be burned some day.
1647
1648       For constructs that do interpolate, variables beginning with ""$"" or
1649       ""@"" are interpolated.  Subscripted variables such as $a[3] or
1650       "$href->{key}[0]" are also interpolated, as are array and hash slices.
1651       But method calls such as "$obj->meth" are not.
1652
1653       Interpolating an array or slice interpolates the elements in order,
1654       separated by the value of $", so is equivalent to interpolating
1655       "join $", @array".  "Punctuation" arrays such as "@*" are usually
1656       interpolated only if the name is enclosed in braces "@{*}", but the
1657       arrays @_, "@+", and "@-" are interpolated even without braces.
1658
1659       For double-quoted strings, the quoting from "\Q" is applied after
1660       interpolation and escapes are processed.
1661
1662           "abc\Qfoo\tbar$s\Exyz"
1663
1664       is equivalent to
1665
1666           "abc" . quotemeta("foo\tbar$s") . "xyz"
1667
1668       For the pattern of regex operators ("qr//", "m//" and "s///"), the
1669       quoting from "\Q" is applied after interpolation is processed, but
1670       before escapes are processed.  This allows the pattern to match
1671       literally (except for "$" and "@").  For example, the following
1672       matches:
1673
1674           '\s\t' =~ /\Q\s\t/
1675
1676       Because "$" or "@" trigger interpolation, you'll need to use something
1677       like "/\Quser\E\@\Qhost/" to match them literally.
1678
1679       Patterns are subject to an additional level of interpretation as a
1680       regular expression.  This is done as a second pass, after variables are
1681       interpolated, so that regular expressions may be incorporated into the
1682       pattern from the variables.  If this is not what you want, use "\Q" to
1683       interpolate a variable literally.
1684
1685       Apart from the behavior described above, Perl does not expand multiple
1686       levels of interpolation.  In particular, contrary to the expectations
1687       of shell programmers, back-quotes do NOT interpolate within double
1688       quotes, nor do single quotes impede evaluation of variables when used
1689       within double quotes.
1690
1691   Regexp Quote-Like Operators
1692       Here are the quote-like operators that apply to pattern matching and
1693       related activities.
1694
1695       "qr/STRING/msixpodualn"
1696               This operator quotes (and possibly compiles) its STRING as a
1697               regular expression.  STRING is interpolated the same way as
1698               PATTERN in "m/PATTERN/".  If "'" is used as the delimiter, no
1699               variable interpolation is done.  Returns a Perl value which may
1700               be used instead of the corresponding "/STRING/msixpodualn"
1701               expression.  The returned value is a normalized version of the
1702               original pattern.  It magically differs from a string
1703               containing the same characters: "ref(qr/x/)" returns "Regexp";
1704               however, dereferencing it is not well defined (you currently
1705               get the normalized version of the original pattern, but this
1706               may change).
1707
1708               For example,
1709
1710                   $rex = qr/my.STRING/is;
1711                   print $rex;                 # prints (?si-xm:my.STRING)
1712                   s/$rex/foo/;
1713
1714               is equivalent to
1715
1716                   s/my.STRING/foo/is;
1717
1718               The result may be used as a subpattern in a match:
1719
1720                   $re = qr/$pattern/;
1721                   $string =~ /foo${re}bar/;   # can be interpolated in other
1722                                               # patterns
1723                   $string =~ $re;             # or used standalone
1724                   $string =~ /$re/;           # or this way
1725
1726               Since Perl may compile the pattern at the moment of execution
1727               of the "qr()" operator, using "qr()" may have speed advantages
1728               in some situations, notably if the result of "qr()" is used
1729               standalone:
1730
1731                   sub match {
1732                       my $patterns = shift;
1733                       my @compiled = map qr/$_/i, @$patterns;
1734                       grep {
1735                           my $success = 0;
1736                           foreach my $pat (@compiled) {
1737                               $success = 1, last if /$pat/;
1738                           }
1739                           $success;
1740                       } @_;
1741                   }
1742
1743               Precompilation of the pattern into an internal representation
1744               at the moment of "qr()" avoids the need to recompile the
1745               pattern every time a match "/$pat/" is attempted.  (Perl has
1746               many other internal optimizations, but none would be triggered
1747               in the above example if we did not use "qr()" operator.)
1748
1749               Options (specified by the following modifiers) are:
1750
1751                   m   Treat string as multiple lines.
1752                   s   Treat string as single line. (Make . match a newline)
1753                   i   Do case-insensitive pattern matching.
1754                   x   Use extended regular expressions; specifying two
1755                       x's means \t and the SPACE character are ignored within
1756                       square-bracketed character classes
1757                   p   When matching preserve a copy of the matched string so
1758                       that ${^PREMATCH}, ${^MATCH}, ${^POSTMATCH} will be
1759                       defined (ignored starting in v5.20) as these are always
1760                       defined starting in that release
1761                   o   Compile pattern only once.
1762                   a   ASCII-restrict: Use ASCII for \d, \s, \w and [[:posix:]]
1763                       character classes; specifying two a's adds the further
1764                       restriction that no ASCII character will match a
1765                       non-ASCII one under /i.
1766                   l   Use the current run-time locale's rules.
1767                   u   Use Unicode rules.
1768                   d   Use Unicode or native charset, as in 5.12 and earlier.
1769                   n   Non-capture mode. Don't let () fill in $1, $2, etc...
1770
1771               If a precompiled pattern is embedded in a larger pattern then
1772               the effect of "msixpluadn" will be propagated appropriately.
1773               The effect that the "/o" modifier has is not propagated, being
1774               restricted to those patterns explicitly using it.
1775
1776               The "/a", "/d", "/l", and "/u" modifiers (added in Perl 5.14)
1777               control the character set rules, but "/a" is the only one you
1778               are likely to want to specify explicitly; the other three are
1779               selected automatically by various pragmas.
1780
1781               See perlre for additional information on valid syntax for
1782               STRING, and for a detailed look at the semantics of regular
1783               expressions.  In particular, all modifiers except the largely
1784               obsolete "/o" are further explained in "Modifiers" in perlre.
1785               "/o" is described in the next section.
1786
1787       "m/PATTERN/msixpodualngc"
1788       "/PATTERN/msixpodualngc"
1789               Searches a string for a pattern match, and in scalar context
1790               returns true if it succeeds, false if it fails.  If no string
1791               is specified via the "=~" or "!~" operator, the $_ string is
1792               searched.  (The string specified with "=~" need not be an
1793               lvalue--it may be the result of an expression evaluation, but
1794               remember the "=~" binds rather tightly.)  See also perlre.
1795
1796               Options are as described in "qr//" above; in addition, the
1797               following match process modifiers are available:
1798
1799                g  Match globally, i.e., find all occurrences.
1800                c  Do not reset search position on a failed match when /g is
1801                   in effect.
1802
1803               If "/" is the delimiter then the initial "m" is optional.  With
1804               the "m" you can use any pair of non-whitespace (ASCII)
1805               characters as delimiters.  This is particularly useful for
1806               matching path names that contain "/", to avoid LTS (leaning
1807               toothpick syndrome).  If "?" is the delimiter, then a match-
1808               only-once rule applies, described in "m?PATTERN?" below.  If
1809               "'" (single quote) is the delimiter, no variable interpolation
1810               is performed on the PATTERN.  When using a delimiter character
1811               valid in an identifier, whitespace is required after the "m".
1812
1813               PATTERN may contain variables, which will be interpolated every
1814               time the pattern search is evaluated, except for when the
1815               delimiter is a single quote.  (Note that $(, $), and $| are not
1816               interpolated because they look like end-of-string tests.)  Perl
1817               will not recompile the pattern unless an interpolated variable
1818               that it contains changes.  You can force Perl to skip the test
1819               and never recompile by adding a "/o" (which stands for "once")
1820               after the trailing delimiter.  Once upon a time, Perl would
1821               recompile regular expressions unnecessarily, and this modifier
1822               was useful to tell it not to do so, in the interests of speed.
1823               But now, the only reasons to use "/o" are one of:
1824
1825               1.  The variables are thousands of characters long and you know
1826                   that they don't change, and you need to wring out the last
1827                   little bit of speed by having Perl skip testing for that.
1828                   (There is a maintenance penalty for doing this, as
1829                   mentioning "/o" constitutes a promise that you won't change
1830                   the variables in the pattern.  If you do change them, Perl
1831                   won't even notice.)
1832
1833               2.  you want the pattern to use the initial values of the
1834                   variables regardless of whether they change or not.  (But
1835                   there are saner ways of accomplishing this than using
1836                   "/o".)
1837
1838               3.  If the pattern contains embedded code, such as
1839
1840                       use re 'eval';
1841                       $code = 'foo(?{ $x })';
1842                       /$code/
1843
1844                   then perl will recompile each time, even though the pattern
1845                   string hasn't changed, to ensure that the current value of
1846                   $x is seen each time.  Use "/o" if you want to avoid this.
1847
1848               The bottom line is that using "/o" is almost never a good idea.
1849
1850       The empty pattern "//"
1851               If the PATTERN evaluates to the empty string, the last
1852               successfully matched regular expression is used instead.  In
1853               this case, only the "g" and "c" flags on the empty pattern are
1854               honored; the other flags are taken from the original pattern.
1855               If no match has previously succeeded, this will (silently) act
1856               instead as a genuine empty pattern (which will always match).
1857
1858               Note that it's possible to confuse Perl into thinking "//" (the
1859               empty regex) is really "//" (the defined-or operator).  Perl is
1860               usually pretty good about this, but some pathological cases
1861               might trigger this, such as "$x///" (is that "($x) / (//)" or
1862               "$x // /"?) and "print $fh //" ("print $fh(//" or
1863               "print($fh //"?).  In all of these examples, Perl will assume
1864               you meant defined-or.  If you meant the empty regex, just use
1865               parentheses or spaces to disambiguate, or even prefix the empty
1866               regex with an "m" (so "//" becomes "m//").
1867
1868       Matching in list context
1869               If the "/g" option is not used, "m//" in list context returns a
1870               list consisting of the subexpressions matched by the
1871               parentheses in the pattern, that is, ($1, $2, $3...)  (Note
1872               that here $1 etc. are also set).  When there are no parentheses
1873               in the pattern, the return value is the list "(1)" for success.
1874               With or without parentheses, an empty list is returned upon
1875               failure.
1876
1877               Examples:
1878
1879                open(TTY, "+</dev/tty")
1880                   || die "can't access /dev/tty: $!";
1881
1882                <TTY> =~ /^y/i && foo();       # do foo if desired
1883
1884                if (/Version: *([0-9.]*)/) { $version = $1; }
1885
1886                next if m#^/usr/spool/uucp#;
1887
1888                # poor man's grep
1889                $arg = shift;
1890                while (<>) {
1891                   print if /$arg/o; # compile only once (no longer needed!)
1892                }
1893
1894                if (($F1, $F2, $Etc) = ($foo =~ /^(\S+)\s+(\S+)\s*(.*)/))
1895
1896               This last example splits $foo into the first two words and the
1897               remainder of the line, and assigns those three fields to $F1,
1898               $F2, and $Etc.  The conditional is true if any variables were
1899               assigned; that is, if the pattern matched.
1900
1901               The "/g" modifier specifies global pattern matching--that is,
1902               matching as many times as possible within the string.  How it
1903               behaves depends on the context.  In list context, it returns a
1904               list of the substrings matched by any capturing parentheses in
1905               the regular expression.  If there are no parentheses, it
1906               returns a list of all the matched strings, as if there were
1907               parentheses around the whole pattern.
1908
1909               In scalar context, each execution of "m//g" finds the next
1910               match, returning true if it matches, and false if there is no
1911               further match.  The position after the last match can be read
1912               or set using the "pos()" function; see "pos" in perlfunc.  A
1913               failed match normally resets the search position to the
1914               beginning of the string, but you can avoid that by adding the
1915               "/c" modifier (for example, "m//gc").  Modifying the target
1916               string also resets the search position.
1917
1918       "\G assertion"
1919               You can intermix "m//g" matches with "m/\G.../g", where "\G" is
1920               a zero-width assertion that matches the exact position where
1921               the previous "m//g", if any, left off.  Without the "/g"
1922               modifier, the "\G" assertion still anchors at "pos()" as it was
1923               at the start of the operation (see "pos" in perlfunc), but the
1924               match is of course only attempted once.  Using "\G" without
1925               "/g" on a target string that has not previously had a "/g"
1926               match applied to it is the same as using the "\A" assertion to
1927               match the beginning of the string.  Note also that, currently,
1928               "\G" is only properly supported when anchored at the very
1929               beginning of the pattern.
1930
1931               Examples:
1932
1933                   # list context
1934                   ($one,$five,$fifteen) = (`uptime` =~ /(\d+\.\d+)/g);
1935
1936                   # scalar context
1937                   local $/ = "";
1938                   while ($paragraph = <>) {
1939                       while ($paragraph =~ /\p{Ll}['")]*[.!?]+['")]*\s/g) {
1940                           $sentences++;
1941                       }
1942                   }
1943                   say $sentences;
1944
1945               Here's another way to check for sentences in a paragraph:
1946
1947                my $sentence_rx = qr{
1948                   (?: (?<= ^ ) | (?<= \s ) )  # after start-of-string or
1949                                               # whitespace
1950                   \p{Lu}                      # capital letter
1951                   .*?                         # a bunch of anything
1952                   (?<= \S )                   # that ends in non-
1953                                               # whitespace
1954                   (?<! \b [DMS]r  )           # but isn't a common abbr.
1955                   (?<! \b Mrs )
1956                   (?<! \b Sra )
1957                   (?<! \b St  )
1958                   [.?!]                       # followed by a sentence
1959                                               # ender
1960                   (?= $ | \s )                # in front of end-of-string
1961                                               # or whitespace
1962                }sx;
1963                local $/ = "";
1964                while (my $paragraph = <>) {
1965                   say "NEW PARAGRAPH";
1966                   my $count = 0;
1967                   while ($paragraph =~ /($sentence_rx)/g) {
1968                       printf "\tgot sentence %d: <%s>\n", ++$count, $1;
1969                   }
1970                }
1971
1972               Here's how to use "m//gc" with "\G":
1973
1974                   $_ = "ppooqppqq";
1975                   while ($i++ < 2) {
1976                       print "1: '";
1977                       print $1 while /(o)/gc; print "', pos=", pos, "\n";
1978                       print "2: '";
1979                       print $1 if /\G(q)/gc;  print "', pos=", pos, "\n";
1980                       print "3: '";
1981                       print $1 while /(p)/gc; print "', pos=", pos, "\n";
1982                   }
1983                   print "Final: '$1', pos=",pos,"\n" if /\G(.)/;
1984
1985               The last example should print:
1986
1987                   1: 'oo', pos=4
1988                   2: 'q', pos=5
1989                   3: 'pp', pos=7
1990                   1: '', pos=7
1991                   2: 'q', pos=8
1992                   3: '', pos=8
1993                   Final: 'q', pos=8
1994
1995               Notice that the final match matched "q" instead of "p", which a
1996               match without the "\G" anchor would have done.  Also note that
1997               the final match did not update "pos".  "pos" is only updated on
1998               a "/g" match.  If the final match did indeed match "p", it's a
1999               good bet that you're running an ancient (pre-5.6.0) version of
2000               Perl.
2001
2002               A useful idiom for "lex"-like scanners is "/\G.../gc".  You can
2003               combine several regexps like this to process a string part-by-
2004               part, doing different actions depending on which regexp
2005               matched.  Each regexp tries to match where the previous one
2006               leaves off.
2007
2008                $_ = <<'EOL';
2009                   $url = URI::URL->new( "http://example.com/" );
2010                   die if $url eq "xXx";
2011                EOL
2012
2013                LOOP: {
2014                    print(" digits"),       redo LOOP if /\G\d+\b[,.;]?\s*/gc;
2015                    print(" lowercase"),    redo LOOP
2016                                                   if /\G\p{Ll}+\b[,.;]?\s*/gc;
2017                    print(" UPPERCASE"),    redo LOOP
2018                                                   if /\G\p{Lu}+\b[,.;]?\s*/gc;
2019                    print(" Capitalized"),  redo LOOP
2020                                             if /\G\p{Lu}\p{Ll}+\b[,.;]?\s*/gc;
2021                    print(" MiXeD"),        redo LOOP if /\G\pL+\b[,.;]?\s*/gc;
2022                    print(" alphanumeric"), redo LOOP
2023                                           if /\G[\p{Alpha}\pN]+\b[,.;]?\s*/gc;
2024                    print(" line-noise"),   redo LOOP if /\G\W+/gc;
2025                    print ". That's all!\n";
2026                }
2027
2028               Here is the output (split into several lines):
2029
2030                line-noise lowercase line-noise UPPERCASE line-noise UPPERCASE
2031                line-noise lowercase line-noise lowercase line-noise lowercase
2032                lowercase line-noise lowercase lowercase line-noise lowercase
2033                lowercase line-noise MiXeD line-noise. That's all!
2034
2035       "m?PATTERN?msixpodualngc"
2036               This is just like the "m/PATTERN/" search, except that it
2037               matches only once between calls to the "reset()" operator.
2038               This is a useful optimization when you want to see only the
2039               first occurrence of something in each file of a set of files,
2040               for instance.  Only "m??"  patterns local to the current
2041               package are reset.
2042
2043                   while (<>) {
2044                       if (m?^$?) {
2045                                           # blank line between header and body
2046                       }
2047                   } continue {
2048                       reset if eof;       # clear m?? status for next file
2049                   }
2050
2051               Another example switched the first "latin1" encoding it finds
2052               to "utf8" in a pod file:
2053
2054                   s//utf8/ if m? ^ =encoding \h+ \K latin1 ?x;
2055
2056               The match-once behavior is controlled by the match delimiter
2057               being "?"; with any other delimiter this is the normal "m//"
2058               operator.
2059
2060               In the past, the leading "m" in "m?PATTERN?" was optional, but
2061               omitting it would produce a deprecation warning.  As of
2062               v5.22.0, omitting it produces a syntax error.  If you encounter
2063               this construct in older code, you can just add "m".
2064
2065       "s/PATTERN/REPLACEMENT/msixpodualngcer"
2066               Searches a string for a pattern, and if found, replaces that
2067               pattern with the replacement text and returns the number of
2068               substitutions made.  Otherwise it returns false (a value that
2069               is both an empty string ("") and numeric zero (0) as described
2070               in "Relational Operators").
2071
2072               If the "/r" (non-destructive) option is used then it runs the
2073               substitution on a copy of the string and instead of returning
2074               the number of substitutions, it returns the copy whether or not
2075               a substitution occurred.  The original string is never changed
2076               when "/r" is used.  The copy will always be a plain string,
2077               even if the input is an object or a tied variable.
2078
2079               If no string is specified via the "=~" or "!~" operator, the $_
2080               variable is searched and modified.  Unless the "/r" option is
2081               used, the string specified must be a scalar variable, an array
2082               element, a hash element, or an assignment to one of those; that
2083               is, some sort of scalar lvalue.
2084
2085               If the delimiter chosen is a single quote, no variable
2086               interpolation is done on either the PATTERN or the REPLACEMENT.
2087               Otherwise, if the PATTERN contains a "$" that looks like a
2088               variable rather than an end-of-string test, the variable will
2089               be interpolated into the pattern at run-time.  If you want the
2090               pattern compiled only once the first time the variable is
2091               interpolated, use the "/o" option.  If the pattern evaluates to
2092               the empty string, the last successfully executed regular
2093               expression is used instead.  See perlre for further explanation
2094               on these.
2095
2096               Options are as with "m//" with the addition of the following
2097               replacement specific options:
2098
2099                   e   Evaluate the right side as an expression.
2100                   ee  Evaluate the right side as a string then eval the
2101                       result.
2102                   r   Return substitution and leave the original string
2103                       untouched.
2104
2105               Any non-whitespace delimiter may replace the slashes.  Add
2106               space after the "s" when using a character allowed in
2107               identifiers.  If single quotes are used, no interpretation is
2108               done on the replacement string (the "/e" modifier overrides
2109               this, however).  Note that Perl treats backticks as normal
2110               delimiters; the replacement text is not evaluated as a command.
2111               If the PATTERN is delimited by bracketing quotes, the
2112               REPLACEMENT has its own pair of quotes, which may or may not be
2113               bracketing quotes, for example, "s(foo)(bar)" or "s<foo>/bar/".
2114               A "/e" will cause the replacement portion to be treated as a
2115               full-fledged Perl expression and evaluated right then and
2116               there.  It is, however, syntax checked at compile-time.  A
2117               second "e" modifier will cause the replacement portion to be
2118               "eval"ed before being run as a Perl expression.
2119
2120               Examples:
2121
2122                   s/\bgreen\b/mauve/g;              # don't change wintergreen
2123
2124                   $path =~ s|/usr/bin|/usr/local/bin|;
2125
2126                   s/Login: $foo/Login: $bar/; # run-time pattern
2127
2128                   ($foo = $bar) =~ s/this/that/;      # copy first, then
2129                                                       # change
2130                   ($foo = "$bar") =~ s/this/that/;    # convert to string,
2131                                                       # copy, then change
2132                   $foo = $bar =~ s/this/that/r;       # Same as above using /r
2133                   $foo = $bar =~ s/this/that/r
2134                               =~ s/that/the other/r;  # Chained substitutes
2135                                                       # using /r
2136                   @foo = map { s/this/that/r } @bar   # /r is very useful in
2137                                                       # maps
2138
2139                   $count = ($paragraph =~ s/Mister\b/Mr./g);  # get change-cnt
2140
2141                   $_ = 'abc123xyz';
2142                   s/\d+/$&*2/e;               # yields 'abc246xyz'
2143                   s/\d+/sprintf("%5d",$&)/e;  # yields 'abc  246xyz'
2144                   s/\w/$& x 2/eg;             # yields 'aabbcc  224466xxyyzz'
2145
2146                   s/%(.)/$percent{$1}/g;      # change percent escapes; no /e
2147                   s/%(.)/$percent{$1} || $&/ge;       # expr now, so /e
2148                   s/^=(\w+)/pod($1)/ge;       # use function call
2149
2150                   $_ = 'abc123xyz';
2151                   $x = s/abc/def/r;           # $x is 'def123xyz' and
2152                                               # $_ remains 'abc123xyz'.
2153
2154                   # expand variables in $_, but dynamics only, using
2155                   # symbolic dereferencing
2156                   s/\$(\w+)/${$1}/g;
2157
2158                   # Add one to the value of any numbers in the string
2159                   s/(\d+)/1 + $1/eg;
2160
2161                   # Titlecase words in the last 30 characters only
2162                   substr($str, -30) =~ s/\b(\p{Alpha}+)\b/\u\L$1/g;
2163
2164                   # This will expand any embedded scalar variable
2165                   # (including lexicals) in $_ : First $1 is interpolated
2166                   # to the variable name, and then evaluated
2167                   s/(\$\w+)/$1/eeg;
2168
2169                   # Delete (most) C comments.
2170                   $program =~ s {
2171                       /\*     # Match the opening delimiter.
2172                       .*?     # Match a minimal number of characters.
2173                       \*/     # Match the closing delimiter.
2174                   } []gsx;
2175
2176                   s/^\s*(.*?)\s*$/$1/;        # trim whitespace in $_,
2177                                               # expensively
2178
2179                   for ($variable) {           # trim whitespace in $variable,
2180                                               # cheap
2181                       s/^\s+//;
2182                       s/\s+$//;
2183                   }
2184
2185                   s/([^ ]*) *([^ ]*)/$2 $1/;  # reverse 1st two fields
2186
2187                   $foo !~ s/A/a/g;    # Lowercase all A's in $foo; return
2188                                       # 0 if any were found and changed;
2189                                       # otherwise return 1
2190
2191               Note the use of "$" instead of "\" in the last example.  Unlike
2192               sed, we use the \<digit> form only in the left hand side.
2193               Anywhere else it's $<digit>.
2194
2195               Occasionally, you can't use just a "/g" to get all the changes
2196               to occur that you might want.  Here are two common cases:
2197
2198                   # put commas in the right places in an integer
2199                   1 while s/(\d)(\d\d\d)(?!\d)/$1,$2/g;
2200
2201                   # expand tabs to 8-column spacing
2202                   1 while s/\t+/' ' x (length($&)*8 - length($`)%8)/e;
2203
2204               While "s///" accepts the "/c" flag, it has no effect beyond
2205               producing a warning if warnings are enabled.
2206
2207   Quote-Like Operators
2208       "q/STRING/"
2209       'STRING'
2210           A single-quoted, literal string.  A backslash represents a
2211           backslash unless followed by the delimiter or another backslash, in
2212           which case the delimiter or backslash is interpolated.
2213
2214               $foo = q!I said, "You said, 'She said it.'"!;
2215               $bar = q('This is it.');
2216               $baz = '\n';                # a two-character string
2217
2218       "qq/STRING/"
2219       "STRING"
2220           A double-quoted, interpolated string.
2221
2222               $_ .= qq
2223                (*** The previous line contains the naughty word "$1".\n)
2224                           if /\b(tcl|java|python)\b/i;      # :-)
2225               $baz = "\n";                # a one-character string
2226
2227       "qx/STRING/"
2228       "`STRING`"
2229           A string which is (possibly) interpolated and then executed as a
2230           system command, via /bin/sh or its equivalent if required.  Shell
2231           wildcards, pipes, and redirections will be honored.  Similarly to
2232           "system", if the string contains no shell metacharacters then it
2233           will executed directly.  The collected standard output of the
2234           command is returned; standard error is unaffected.  In scalar
2235           context, it comes back as a single (potentially multi-line) string,
2236           or "undef" if the shell (or command) could not be started.  In list
2237           context, returns a list of lines (however you've defined lines with
2238           $/ or $INPUT_RECORD_SEPARATOR), or an empty list if the shell (or
2239           command) could not be started.
2240
2241           Because backticks do not affect standard error, use shell file
2242           descriptor syntax (assuming the shell supports this) if you care to
2243           address this.  To capture a command's STDERR and STDOUT together:
2244
2245               $output = `cmd 2>&1`;
2246
2247           To capture a command's STDOUT but discard its STDERR:
2248
2249               $output = `cmd 2>/dev/null`;
2250
2251           To capture a command's STDERR but discard its STDOUT (ordering is
2252           important here):
2253
2254               $output = `cmd 2>&1 1>/dev/null`;
2255
2256           To exchange a command's STDOUT and STDERR in order to capture the
2257           STDERR but leave its STDOUT to come out the old STDERR:
2258
2259               $output = `cmd 3>&1 1>&2 2>&3 3>&-`;
2260
2261           To read both a command's STDOUT and its STDERR separately, it's
2262           easiest to redirect them separately to files, and then read from
2263           those files when the program is done:
2264
2265               system("program args 1>program.stdout 2>program.stderr");
2266
2267           The STDIN filehandle used by the command is inherited from Perl's
2268           STDIN.  For example:
2269
2270               open(SPLAT, "stuff")   || die "can't open stuff: $!";
2271               open(STDIN, "<&SPLAT") || die "can't dupe SPLAT: $!";
2272               print STDOUT `sort`;
2273
2274           will print the sorted contents of the file named "stuff".
2275
2276           Using single-quote as a delimiter protects the command from Perl's
2277           double-quote interpolation, passing it on to the shell instead:
2278
2279               $perl_info  = qx(ps $$);            # that's Perl's $$
2280               $shell_info = qx'ps $$';            # that's the new shell's $$
2281
2282           How that string gets evaluated is entirely subject to the command
2283           interpreter on your system.  On most platforms, you will have to
2284           protect shell metacharacters if you want them treated literally.
2285           This is in practice difficult to do, as it's unclear how to escape
2286           which characters.  See perlsec for a clean and safe example of a
2287           manual "fork()" and "exec()" to emulate backticks safely.
2288
2289           On some platforms (notably DOS-like ones), the shell may not be
2290           capable of dealing with multiline commands, so putting newlines in
2291           the string may not get you what you want.  You may be able to
2292           evaluate multiple commands in a single line by separating them with
2293           the command separator character, if your shell supports that (for
2294           example, ";" on many Unix shells and "&" on the Windows NT "cmd"
2295           shell).
2296
2297           Perl will attempt to flush all files opened for output before
2298           starting the child process, but this may not be supported on some
2299           platforms (see perlport).  To be safe, you may need to set $|
2300           ($AUTOFLUSH in "English") or call the "autoflush()" method of
2301           "IO::Handle" on any open handles.
2302
2303           Beware that some command shells may place restrictions on the
2304           length of the command line.  You must ensure your strings don't
2305           exceed this limit after any necessary interpolations.  See the
2306           platform-specific release notes for more details about your
2307           particular environment.
2308
2309           Using this operator can lead to programs that are difficult to
2310           port, because the shell commands called vary between systems, and
2311           may in fact not be present at all.  As one example, the "type"
2312           command under the POSIX shell is very different from the "type"
2313           command under DOS.  That doesn't mean you should go out of your way
2314           to avoid backticks when they're the right way to get something
2315           done.  Perl was made to be a glue language, and one of the things
2316           it glues together is commands.  Just understand what you're getting
2317           yourself into.
2318
2319           Like "system", backticks put the child process exit code in $?.  If
2320           you'd like to manually inspect failure, you can check all possible
2321           failure modes by inspecting $? like this:
2322
2323               if ($? == -1) {
2324                   print "failed to execute: $!\n";
2325               }
2326               elsif ($? & 127) {
2327                   printf "child died with signal %d, %s coredump\n",
2328                       ($? & 127),  ($? & 128) ? 'with' : 'without';
2329               }
2330               else {
2331                   printf "child exited with value %d\n", $? >> 8;
2332               }
2333
2334           Use the open pragma to control the I/O layers used when reading the
2335           output of the command, for example:
2336
2337             use open IN => ":encoding(UTF-8)";
2338             my $x = `cmd-producing-utf-8`;
2339
2340           "qx//" can also be called like a function with "readpipe" in
2341           perlfunc.
2342
2343           See "I/O Operators" for more discussion.
2344
2345       "qw/STRING/"
2346           Evaluates to a list of the words extracted out of STRING, using
2347           embedded whitespace as the word delimiters.  It can be understood
2348           as being roughly equivalent to:
2349
2350               split(" ", q/STRING/);
2351
2352           the differences being that it only splits on ASCII whitespace,
2353           generates a real list at compile time, and in scalar context it
2354           returns the last element in the list.  So this expression:
2355
2356               qw(foo bar baz)
2357
2358           is semantically equivalent to the list:
2359
2360               "foo", "bar", "baz"
2361
2362           Some frequently seen examples:
2363
2364               use POSIX qw( setlocale localeconv )
2365               @EXPORT = qw( foo bar baz );
2366
2367           A common mistake is to try to separate the words with commas or to
2368           put comments into a multi-line "qw"-string.  For this reason, the
2369           "use warnings" pragma and the -w switch (that is, the $^W variable)
2370           produces warnings if the STRING contains the "," or the "#"
2371           character.
2372
2373       "tr/SEARCHLIST/REPLACEMENTLIST/cdsr"
2374       "y/SEARCHLIST/REPLACEMENTLIST/cdsr"
2375           Transliterates all occurrences of the characters found (or not
2376           found if the "/c" modifier is specified) in the search list with
2377           the positionally corresponding character in the replacement list,
2378           possibly deleting some, depending on the modifiers specified.  It
2379           returns the number of characters replaced or deleted.  If no string
2380           is specified via the "=~" or "!~" operator, the $_ string is
2381           transliterated.
2382
2383           For sed devotees, "y" is provided as a synonym for "tr".
2384
2385           If the "/r" (non-destructive) option is present, a new copy of the
2386           string is made and its characters transliterated, and this copy is
2387           returned no matter whether it was modified or not: the original
2388           string is always left unchanged.  The new copy is always a plain
2389           string, even if the input string is an object or a tied variable.
2390
2391           Unless the "/r" option is used, the string specified with "=~" must
2392           be a scalar variable, an array element, a hash element, or an
2393           assignment to one of those; in other words, an lvalue.
2394
2395           If the characters delimiting SEARCHLIST and REPLACEMENTLIST are
2396           single quotes ("tr'SEARCHLIST'REPLACEMENTLIST'"), the only
2397           interpolation is removal of "\" from pairs of "\\".
2398
2399           Otherwise, a character range may be specified with a hyphen, so
2400           "tr/A-J/0-9/" does the same replacement as
2401           "tr/ACEGIBDFHJ/0246813579/".
2402
2403           If the SEARCHLIST is delimited by bracketing quotes, the
2404           REPLACEMENTLIST must have its own pair of quotes, which may or may
2405           not be bracketing quotes; for example, "tr[aeiouy][yuoiea]" or
2406           "tr(+\-*/)/ABCD/".
2407
2408           Characters may be literals, or (if the delimiters aren't single
2409           quotes) any of the escape sequences accepted in double-quoted
2410           strings.  But there is never any variable interpolation, so "$" and
2411           "@" are always treated as literals.  A hyphen at the beginning or
2412           end, or preceded by a backslash is also always considered a
2413           literal.  Escape sequence details are in the table near the
2414           beginning of this section.
2415
2416           Note that "tr" does not do regular expression character classes
2417           such as "\d" or "\pL".  The "tr" operator is not equivalent to the
2418           tr(1) utility.  "tr[a-z][A-Z]" will uppercase the 26 letters "a"
2419           through "z", but for case changing not confined to ASCII, use "lc",
2420           "uc", "lcfirst", "ucfirst" (all documented in perlfunc), or the
2421           substitution operator "s/PATTERN/REPLACEMENT/" (with "\U", "\u",
2422           "\L", and "\l" string-interpolation escapes in the REPLACEMENT
2423           portion).
2424
2425           Most ranges are unportable between character sets, but certain ones
2426           signal Perl to do special handling to make them portable.  There
2427           are two classes of portable ranges.  The first are any subsets of
2428           the ranges "A-Z", "a-z", and "0-9", when expressed as literal
2429           characters.
2430
2431             tr/h-k/H-K/
2432
2433           capitalizes the letters "h", "i", "j", and "k" and nothing else, no
2434           matter what the platform's character set is.  In contrast, all of
2435
2436             tr/\x68-\x6B/\x48-\x4B/
2437             tr/h-\x6B/H-\x4B/
2438             tr/\x68-k/\x48-K/
2439
2440           do the same capitalizations as the previous example when run on
2441           ASCII platforms, but something completely different on EBCDIC ones.
2442
2443           The second class of portable ranges is invoked when one or both of
2444           the range's end points are expressed as "\N{...}"
2445
2446            $string =~ tr/\N{U+20}-\N{U+7E}//d;
2447
2448           removes from $string all the platform's characters which are
2449           equivalent to any of Unicode U+0020, U+0021, ... U+007D, U+007E.
2450           This is a portable range, and has the same effect on every platform
2451           it is run on.  In this example, these are the ASCII printable
2452           characters.  So after this is run, $string has only controls and
2453           characters which have no ASCII equivalents.
2454
2455           But, even for portable ranges, it is not generally obvious what is
2456           included without having to look things up in the manual.  A sound
2457           principle is to use only ranges that both begin from, and end at,
2458           either ASCII alphabetics of equal case ("b-e", "B-E"), or digits
2459           ("1-4").  Anything else is unclear (and unportable unless "\N{...}"
2460           is used).  If in doubt, spell out the character sets in full.
2461
2462           Options:
2463
2464               c   Complement the SEARCHLIST.
2465               d   Delete found but unreplaced characters.
2466               r   Return the modified string and leave the original string
2467                   untouched.
2468               s   Squash duplicate replaced characters.
2469
2470           If the "/d" modifier is specified, any characters specified by
2471           SEARCHLIST  not found in REPLACEMENTLIST are deleted.  (Note that
2472           this is slightly more flexible than the behavior of some tr
2473           programs, which delete anything they find in the SEARCHLIST,
2474           period.)
2475
2476           If the "/s" modifier is specified, sequences of characters, all in
2477           a row, that were transliterated to the same character are squashed
2478           down to a single instance of that character.
2479
2480            my $a = "aaaba"
2481            $a =~ tr/a/a/s     # $a now is "aba"
2482
2483           If the "/d" modifier is used, the REPLACEMENTLIST is always
2484           interpreted exactly as specified.  Otherwise, if the
2485           REPLACEMENTLIST is shorter than the SEARCHLIST, the final
2486           character, if any, is replicated until it is long enough.  There
2487           won't be a final character if and only if the REPLACEMENTLIST is
2488           empty, in which case REPLACEMENTLIST is copied from SEARCHLIST.
2489           An empty REPLACEMENTLIST is useful for counting characters in a
2490           class, or for squashing character sequences in a class.
2491
2492               tr/abcd//            tr/abcd/abcd/
2493               tr/abcd/AB/          tr/abcd/ABBB/
2494               tr/abcd//d           s/[abcd]//g
2495               tr/abcd/AB/d         (tr/ab/AB/ + s/[cd]//g)  - but run together
2496
2497           If the "/c" modifier is specified, the characters to be
2498           transliterated are the ones NOT in SEARCHLIST, that is, it is
2499           complemented.  If "/d" and/or "/s" are also specified, they apply
2500           to the complemented SEARCHLIST.  Recall, that if REPLACEMENTLIST is
2501           empty (except under "/d") a copy of SEARCHLIST is used instead.
2502           That copy is made after complementing under "/c".  SEARCHLIST is
2503           sorted by code point order after complementing, and any
2504           REPLACEMENTLIST  is applied to that sorted result.  This means that
2505           under "/c", the order of the characters specified in SEARCHLIST is
2506           irrelevant.  This can lead to different results on EBCDIC systems
2507           if REPLACEMENTLIST contains more than one character, hence it is
2508           generally non-portable to use "/c" with such a REPLACEMENTLIST.
2509
2510           Another way of describing the operation is this: If "/c" is
2511           specified, the SEARCHLIST is sorted by code point order, then
2512           complemented.  If REPLACEMENTLIST is empty and "/d" is not
2513           specified, REPLACEMENTLIST is replaced by a copy of SEARCHLIST (as
2514           modified under "/c"), and these potentially modified lists are used
2515           as the basis for what follows.  Any character in the target string
2516           that isn't in SEARCHLIST is passed through unchanged.  Every other
2517           character in the target string is replaced by the character in
2518           REPLACEMENTLIST that positionally corresponds to its mate in
2519           SEARCHLIST, except that under "/s", the 2nd and following
2520           characters are squeezed out in a sequence of characters in a row
2521           that all translate to the same character.  If SEARCHLIST is longer
2522           than REPLACEMENTLIST, characters in the target string that match a
2523           character in SEARCHLIST that doesn't have a correspondence in
2524           REPLACEMENTLIST are either deleted from the target string if "/d"
2525           is specified; or replaced by the final character in REPLACEMENTLIST
2526           if "/d" isn't specified.
2527
2528           Some examples:
2529
2530            $ARGV[1] =~ tr/A-Z/a-z/;   # canonicalize to lower case ASCII
2531
2532            $cnt = tr/*/*/;            # count the stars in $_
2533            $cnt = tr/*//;             # same thing
2534
2535            $cnt = $sky =~ tr/*/*/;    # count the stars in $sky
2536            $cnt = $sky =~ tr/*//;     # same thing
2537
2538            $cnt = $sky =~ tr/*//c;    # count all the non-stars in $sky
2539            $cnt = $sky =~ tr/*/*/c;   # same, but transliterate each non-star
2540                                       # into a star, leaving the already-stars
2541                                       # alone.  Afterwards, everything in $sky
2542                                       # is a star.
2543
2544            $cnt = tr/0-9//;           # count the ASCII digits in $_
2545
2546            tr/a-zA-Z//s;              # bookkeeper -> bokeper
2547            tr/o/o/s;                  # bookkeeper -> bokkeeper
2548            tr/oe/oe/s;                # bookkeeper -> bokkeper
2549            tr/oe//s;                  # bookkeeper -> bokkeper
2550            tr/oe/o/s;                 # bookkeeper -> bokkopor
2551
2552            ($HOST = $host) =~ tr/a-z/A-Z/;
2553             $HOST = $host  =~ tr/a-z/A-Z/r; # same thing
2554
2555            $HOST = $host =~ tr/a-z/A-Z/r   # chained with s///r
2556                          =~ s/:/ -p/r;
2557
2558            tr/a-zA-Z/ /cs;                 # change non-alphas to single space
2559
2560            @stripped = map tr/a-zA-Z/ /csr, @original;
2561                                            # /r with map
2562
2563            tr [\200-\377]
2564               [\000-\177];                 # wickedly delete 8th bit
2565
2566            $foo !~ tr/A/a/    # transliterate all the A's in $foo to 'a',
2567                               # return 0 if any were found and changed.
2568                               # Otherwise return 1
2569
2570           If multiple transliterations are given for a character, only the
2571           first one is used:
2572
2573            tr/AAA/XYZ/
2574
2575           will transliterate any A to X.
2576
2577           Because the transliteration table is built at compile time, neither
2578           the SEARCHLIST nor the REPLACEMENTLIST are subjected to double
2579           quote interpolation.  That means that if you want to use variables,
2580           you must use an "eval()":
2581
2582            eval "tr/$oldlist/$newlist/";
2583            die $@ if $@;
2584
2585            eval "tr/$oldlist/$newlist/, 1" or die $@;
2586
2587       "<<EOF"
2588           A line-oriented form of quoting is based on the shell "here-
2589           document" syntax.  Following a "<<" you specify a string to
2590           terminate the quoted material, and all lines following the current
2591           line down to the terminating string are the value of the item.
2592
2593           Prefixing the terminating string with a "~" specifies that you want
2594           to use "Indented Here-docs" (see below).
2595
2596           The terminating string may be either an identifier (a word), or
2597           some quoted text.  An unquoted identifier works like double quotes.
2598           There may not be a space between the "<<" and the identifier,
2599           unless the identifier is explicitly quoted.  The terminating string
2600           must appear by itself (unquoted and with no surrounding whitespace)
2601           on the terminating line.
2602
2603           If the terminating string is quoted, the type of quotes used
2604           determine the treatment of the text.
2605
2606           Double Quotes
2607               Double quotes indicate that the text will be interpolated using
2608               exactly the same rules as normal double quoted strings.
2609
2610                      print <<EOF;
2611                   The price is $Price.
2612                   EOF
2613
2614                      print << "EOF"; # same as above
2615                   The price is $Price.
2616                   EOF
2617
2618           Single Quotes
2619               Single quotes indicate the text is to be treated literally with
2620               no interpolation of its content.  This is similar to single
2621               quoted strings except that backslashes have no special meaning,
2622               with "\\" being treated as two backslashes and not one as they
2623               would in every other quoting construct.
2624
2625               Just as in the shell, a backslashed bareword following the "<<"
2626               means the same thing as a single-quoted string does:
2627
2628                       $cost = <<'VISTA';  # hasta la ...
2629                   That'll be $10 please, ma'am.
2630                   VISTA
2631
2632                       $cost = <<\VISTA;   # Same thing!
2633                   That'll be $10 please, ma'am.
2634                   VISTA
2635
2636               This is the only form of quoting in perl where there is no need
2637               to worry about escaping content, something that code generators
2638               can and do make good use of.
2639
2640           Backticks
2641               The content of the here doc is treated just as it would be if
2642               the string were embedded in backticks.  Thus the content is
2643               interpolated as though it were double quoted and then executed
2644               via the shell, with the results of the execution returned.
2645
2646                      print << `EOC`; # execute command and get results
2647                   echo hi there
2648                   EOC
2649
2650           Indented Here-docs
2651               The here-doc modifier "~" allows you to indent your here-docs
2652               to make the code more readable:
2653
2654                   if ($some_var) {
2655                     print <<~EOF;
2656                       This is a here-doc
2657                       EOF
2658                   }
2659
2660               This will print...
2661
2662                   This is a here-doc
2663
2664               ...with no leading whitespace.
2665
2666               The delimiter is used to determine the exact whitespace to
2667               remove from the beginning of each line.  All lines must have at
2668               least the same starting whitespace (except lines only
2669               containing a newline) or perl will croak.  Tabs and spaces can
2670               be mixed, but are matched exactly.  One tab will not be equal
2671               to 8 spaces!
2672
2673               Additional beginning whitespace (beyond what preceded the
2674               delimiter) will be preserved:
2675
2676                   print <<~EOF;
2677                     This text is not indented
2678                       This text is indented with two spaces
2679                               This text is indented with two tabs
2680                     EOF
2681
2682               Finally, the modifier may be used with all of the forms
2683               mentioned above:
2684
2685                   <<~\EOF;
2686                   <<~'EOF'
2687                   <<~"EOF"
2688                   <<~`EOF`
2689
2690               And whitespace may be used between the "~" and quoted
2691               delimiters:
2692
2693                   <<~ 'EOF'; # ... "EOF", `EOF`
2694
2695           It is possible to stack multiple here-docs in a row:
2696
2697                  print <<"foo", <<"bar"; # you can stack them
2698               I said foo.
2699               foo
2700               I said bar.
2701               bar
2702
2703                  myfunc(<< "THIS", 23, <<'THAT');
2704               Here's a line
2705               or two.
2706               THIS
2707               and here's another.
2708               THAT
2709
2710           Just don't forget that you have to put a semicolon on the end to
2711           finish the statement, as Perl doesn't know you're not going to try
2712           to do this:
2713
2714                  print <<ABC
2715               179231
2716               ABC
2717                  + 20;
2718
2719           If you want to remove the line terminator from your here-docs, use
2720           "chomp()".
2721
2722               chomp($string = <<'END');
2723               This is a string.
2724               END
2725
2726           If you want your here-docs to be indented with the rest of the
2727           code, use the "<<~FOO" construct described under "Indented Here-
2728           docs":
2729
2730               $quote = <<~'FINIS';
2731                  The Road goes ever on and on,
2732                  down from the door where it began.
2733                  FINIS
2734
2735           If you use a here-doc within a delimited construct, such as in
2736           "s///eg", the quoted material must still come on the line following
2737           the "<<FOO" marker, which means it may be inside the delimited
2738           construct:
2739
2740               s/this/<<E . 'that'
2741               the other
2742               E
2743                . 'more '/eg;
2744
2745           It works this way as of Perl 5.18.  Historically, it was
2746           inconsistent, and you would have to write
2747
2748               s/this/<<E . 'that'
2749                . 'more '/eg;
2750               the other
2751               E
2752
2753           outside of string evals.
2754
2755           Additionally, quoting rules for the end-of-string identifier are
2756           unrelated to Perl's quoting rules.  "q()", "qq()", and the like are
2757           not supported in place of '' and "", and the only interpolation is
2758           for backslashing the quoting character:
2759
2760               print << "abc\"def";
2761               testing...
2762               abc"def
2763
2764           Finally, quoted strings cannot span multiple lines.  The general
2765           rule is that the identifier must be a string literal.  Stick with
2766           that, and you should be safe.
2767
2768   Gory details of parsing quoted constructs
2769       When presented with something that might have several different
2770       interpretations, Perl uses the DWIM (that's "Do What I Mean") principle
2771       to pick the most probable interpretation.  This strategy is so
2772       successful that Perl programmers often do not suspect the ambivalence
2773       of what they write.  But from time to time, Perl's notions differ
2774       substantially from what the author honestly meant.
2775
2776       This section hopes to clarify how Perl handles quoted constructs.
2777       Although the most common reason to learn this is to unravel
2778       labyrinthine regular expressions, because the initial steps of parsing
2779       are the same for all quoting operators, they are all discussed
2780       together.
2781
2782       The most important Perl parsing rule is the first one discussed below:
2783       when processing a quoted construct, Perl first finds the end of that
2784       construct, then interprets its contents.  If you understand this rule,
2785       you may skip the rest of this section on the first reading.  The other
2786       rules are likely to contradict the user's expectations much less
2787       frequently than this first one.
2788
2789       Some passes discussed below are performed concurrently, but because
2790       their results are the same, we consider them individually.  For
2791       different quoting constructs, Perl performs different numbers of
2792       passes, from one to four, but these passes are always performed in the
2793       same order.
2794
2795       Finding the end
2796           The first pass is finding the end of the quoted construct.  This
2797           results in saving to a safe location a copy of the text (between
2798           the starting and ending delimiters), normalized as necessary to
2799           avoid needing to know what the original delimiters were.
2800
2801           If the construct is a here-doc, the ending delimiter is a line that
2802           has a terminating string as the content.  Therefore "<<EOF" is
2803           terminated by "EOF" immediately followed by "\n" and starting from
2804           the first column of the terminating line.  When searching for the
2805           terminating line of a here-doc, nothing is skipped.  In other
2806           words, lines after the here-doc syntax are compared with the
2807           terminating string line by line.
2808
2809           For the constructs except here-docs, single characters are used as
2810           starting and ending delimiters.  If the starting delimiter is an
2811           opening punctuation (that is "(", "[", "{", or "<"), the ending
2812           delimiter is the corresponding closing punctuation (that is ")",
2813           "]", "}", or ">").  If the starting delimiter is an unpaired
2814           character like "/" or a closing punctuation, the ending delimiter
2815           is the same as the starting delimiter.  Therefore a "/" terminates
2816           a "qq//" construct, while a "]" terminates both "qq[]" and "qq]]"
2817           constructs.
2818
2819           When searching for single-character delimiters, escaped delimiters
2820           and "\\" are skipped.  For example, while searching for terminating
2821           "/", combinations of "\\" and "\/" are skipped.  If the delimiters
2822           are bracketing, nested pairs are also skipped.  For example, while
2823           searching for a closing "]" paired with the opening "[",
2824           combinations of "\\", "\]", and "\[" are all skipped, and nested
2825           "[" and "]" are skipped as well.  However, when backslashes are
2826           used as the delimiters (like "qq\\" and "tr\\\"), nothing is
2827           skipped.  During the search for the end, backslashes that escape
2828           delimiters or other backslashes are removed (exactly speaking, they
2829           are not copied to the safe location).
2830
2831           For constructs with three-part delimiters ("s///", "y///", and
2832           "tr///"), the search is repeated once more.  If the first delimiter
2833           is not an opening punctuation, the three delimiters must be the
2834           same, such as "s!!!" and "tr)))", in which case the second
2835           delimiter terminates the left part and starts the right part at
2836           once.  If the left part is delimited by bracketing punctuation
2837           (that is "()", "[]", "{}", or "<>"), the right part needs another
2838           pair of delimiters such as "s(){}" and "tr[]//".  In these cases,
2839           whitespace and comments are allowed between the two parts, although
2840           the comment must follow at least one whitespace character;
2841           otherwise a character expected as the start of the comment may be
2842           regarded as the starting delimiter of the right part.
2843
2844           During this search no attention is paid to the semantics of the
2845           construct.  Thus:
2846
2847               "$hash{"$foo/$bar"}"
2848
2849           or:
2850
2851               m/
2852                 bar       # NOT a comment, this slash / terminated m//!
2853                /x
2854
2855           do not form legal quoted expressions.   The quoted part ends on the
2856           first """ and "/", and the rest happens to be a syntax error.
2857           Because the slash that terminated "m//" was followed by a "SPACE",
2858           the example above is not "m//x", but rather "m//" with no "/x"
2859           modifier.  So the embedded "#" is interpreted as a literal "#".
2860
2861           Also no attention is paid to "\c\" (multichar control char syntax)
2862           during this search.  Thus the second "\" in "qq/\c\/" is
2863           interpreted as a part of "\/", and the following "/" is not
2864           recognized as a delimiter.  Instead, use "\034" or "\x1c" at the
2865           end of quoted constructs.
2866
2867       Interpolation
2868           The next step is interpolation in the text obtained, which is now
2869           delimiter-independent.  There are multiple cases.
2870
2871           "<<'EOF'"
2872               No interpolation is performed.  Note that the combination "\\"
2873               is left intact, since escaped delimiters are not available for
2874               here-docs.
2875
2876           "m''", the pattern of "s'''"
2877               No interpolation is performed at this stage.  Any backslashed
2878               sequences including "\\" are treated at the stage to "parsing
2879               regular expressions".
2880
2881           '', "q//", "tr'''", "y'''", the replacement of "s'''"
2882               The only interpolation is removal of "\" from pairs of "\\".
2883               Therefore "-" in "tr'''" and "y'''" is treated literally as a
2884               hyphen and no character range is available.  "\1" in the
2885               replacement of "s'''" does not work as $1.
2886
2887           "tr///", "y///"
2888               No variable interpolation occurs.  String modifying
2889               combinations for case and quoting such as "\Q", "\U", and "\E"
2890               are not recognized.  The other escape sequences such as "\200"
2891               and "\t" and backslashed characters such as "\\" and "\-" are
2892               converted to appropriate literals.  The character "-" is
2893               treated specially and therefore "\-" is treated as a literal
2894               "-".
2895
2896           "", "``", "qq//", "qx//", "<file*glob>", "<<"EOF""
2897               "\Q", "\U", "\u", "\L", "\l", "\F" (possibly paired with "\E")
2898               are converted to corresponding Perl constructs.  Thus,
2899               "$foo\Qbaz$bar" is converted to
2900               "$foo . (quotemeta("baz" . $bar))" internally.  The other
2901               escape sequences such as "\200" and "\t" and backslashed
2902               characters such as "\\" and "\-" are replaced with appropriate
2903               expansions.
2904
2905               Let it be stressed that whatever falls between "\Q" and "\E" is
2906               interpolated in the usual way.  Something like "\Q\\E" has no
2907               "\E" inside.  Instead, it has "\Q", "\\", and "E", so the
2908               result is the same as for "\\\\E".  As a general rule,
2909               backslashes between "\Q" and "\E" may lead to counterintuitive
2910               results.  So, "\Q\t\E" is converted to "quotemeta("\t")", which
2911               is the same as "\\\t" (since TAB is not alphanumeric).  Note
2912               also that:
2913
2914                 $str = '\t';
2915                 return "\Q$str";
2916
2917               may be closer to the conjectural intention of the writer of
2918               "\Q\t\E".
2919
2920               Interpolated scalars and arrays are converted internally to the
2921               "join" and "." catenation operations.  Thus, "$foo XXX '@arr'"
2922               becomes:
2923
2924                 $foo . " XXX '" . (join $", @arr) . "'";
2925
2926               All operations above are performed simultaneously, left to
2927               right.
2928
2929               Because the result of "\Q STRING \E" has all metacharacters
2930               quoted, there is no way to insert a literal "$" or "@" inside a
2931               "\Q\E" pair.  If protected by "\", "$" will be quoted to become
2932               "\\\$"; if not, it is interpreted as the start of an
2933               interpolated scalar.
2934
2935               Note also that the interpolation code needs to make a decision
2936               on where the interpolated scalar ends.  For instance, whether
2937               "a $x -> {c}" really means:
2938
2939                 "a " . $x . " -> {c}";
2940
2941               or:
2942
2943                 "a " . $x -> {c};
2944
2945               Most of the time, the longest possible text that does not
2946               include spaces between components and which contains matching
2947               braces or brackets.  because the outcome may be determined by
2948               voting based on heuristic estimators, the result is not
2949               strictly predictable.  Fortunately, it's usually correct for
2950               ambiguous cases.
2951
2952           the replacement of "s///"
2953               Processing of "\Q", "\U", "\u", "\L", "\l", "\F" and
2954               interpolation happens as with "qq//" constructs.
2955
2956               It is at this step that "\1" is begrudgingly converted to $1 in
2957               the replacement text of "s///", in order to correct the
2958               incorrigible sed hackers who haven't picked up the saner idiom
2959               yet.  A warning is emitted if the "use warnings" pragma or the
2960               -w command-line flag (that is, the $^W variable) was set.
2961
2962           "RE" in "m?RE?", "/RE/", "m/RE/", "s/RE/foo/",
2963               Processing of "\Q", "\U", "\u", "\L", "\l", "\F", "\E", and
2964               interpolation happens (almost) as with "qq//" constructs.
2965
2966               Processing of "\N{...}" is also done here, and compiled into an
2967               intermediate form for the regex compiler.  (This is because, as
2968               mentioned below, the regex compilation may be done at execution
2969               time, and "\N{...}" is a compile-time construct.)
2970
2971               However any other combinations of "\" followed by a character
2972               are not substituted but only skipped, in order to parse them as
2973               regular expressions at the following step.  As "\c" is skipped
2974               at this step, "@" of "\c@" in RE is possibly treated as an
2975               array symbol (for example @foo), even though the same text in
2976               "qq//" gives interpolation of "\c@".
2977
2978               Code blocks such as "(?{BLOCK})" are handled by temporarily
2979               passing control back to the perl parser, in a similar way that
2980               an interpolated array subscript expression such as
2981               "foo$array[1+f("[xyz")]bar" would be.
2982
2983               Moreover, inside "(?{BLOCK})", "(?# comment )", and a
2984               "#"-comment in a "/x"-regular expression, no processing is
2985               performed whatsoever.  This is the first step at which the
2986               presence of the "/x" modifier is relevant.
2987
2988               Interpolation in patterns has several quirks: $|, $(, $), "@+"
2989               and "@-" are not interpolated, and constructs $var[SOMETHING]
2990               are voted (by several different estimators) to be either an
2991               array element or $var followed by an RE alternative.  This is
2992               where the notation "${arr[$bar]}" comes handy: "/${arr[0-9]}/"
2993               is interpreted as array element "-9", not as a regular
2994               expression from the variable $arr followed by a digit, which
2995               would be the interpretation of "/$arr[0-9]/".  Since voting
2996               among different estimators may occur, the result is not
2997               predictable.
2998
2999               The lack of processing of "\\" creates specific restrictions on
3000               the post-processed text.  If the delimiter is "/", one cannot
3001               get the combination "\/" into the result of this step.  "/"
3002               will finish the regular expression, "\/" will be stripped to
3003               "/" on the previous step, and "\\/" will be left as is.
3004               Because "/" is equivalent to "\/" inside a regular expression,
3005               this does not matter unless the delimiter happens to be
3006               character special to the RE engine, such as in "s*foo*bar*",
3007               "m[foo]", or "m?foo?"; or an alphanumeric char, as in:
3008
3009                 m m ^ a \s* b mmx;
3010
3011               In the RE above, which is intentionally obfuscated for
3012               illustration, the delimiter is "m", the modifier is "mx", and
3013               after delimiter-removal the RE is the same as for
3014               "m/ ^ a \s* b /mx".  There's more than one reason you're
3015               encouraged to restrict your delimiters to non-alphanumeric,
3016               non-whitespace choices.
3017
3018           This step is the last one for all constructs except regular
3019           expressions, which are processed further.
3020
3021       parsing regular expressions
3022           Previous steps were performed during the compilation of Perl code,
3023           but this one happens at run time, although it may be optimized to
3024           be calculated at compile time if appropriate.  After preprocessing
3025           described above, and possibly after evaluation if concatenation,
3026           joining, casing translation, or metaquoting are involved, the
3027           resulting string is passed to the RE engine for compilation.
3028
3029           Whatever happens in the RE engine might be better discussed in
3030           perlre, but for the sake of continuity, we shall do so here.
3031
3032           This is another step where the presence of the "/x" modifier is
3033           relevant.  The RE engine scans the string from left to right and
3034           converts it into a finite automaton.
3035
3036           Backslashed characters are either replaced with corresponding
3037           literal strings (as with "\{"), or else they generate special nodes
3038           in the finite automaton (as with "\b").  Characters special to the
3039           RE engine (such as "|") generate corresponding nodes or groups of
3040           nodes.  "(?#...)" comments are ignored.  All the rest is either
3041           converted to literal strings to match, or else is ignored (as is
3042           whitespace and "#"-style comments if "/x" is present).
3043
3044           Parsing of the bracketed character class construct, "[...]", is
3045           rather different than the rule used for the rest of the pattern.
3046           The terminator of this construct is found using the same rules as
3047           for finding the terminator of a "{}"-delimited construct, the only
3048           exception being that "]" immediately following "[" is treated as
3049           though preceded by a backslash.
3050
3051           The terminator of runtime "(?{...})" is found by temporarily
3052           switching control to the perl parser, which should stop at the
3053           point where the logically balancing terminating "}" is found.
3054
3055           It is possible to inspect both the string given to RE engine and
3056           the resulting finite automaton.  See the arguments
3057           "debug"/"debugcolor" in the "use re" pragma, as well as Perl's -Dr
3058           command-line switch documented in "Command Switches" in perlrun.
3059
3060       Optimization of regular expressions
3061           This step is listed for completeness only.  Since it does not
3062           change semantics, details of this step are not documented and are
3063           subject to change without notice.  This step is performed over the
3064           finite automaton that was generated during the previous pass.
3065
3066           It is at this stage that "split()" silently optimizes "/^/" to mean
3067           "/^/m".
3068
3069   I/O Operators
3070       There are several I/O operators you should know about.
3071
3072       A string enclosed by backticks (grave accents) first undergoes double-
3073       quote interpolation.  It is then interpreted as an external command,
3074       and the output of that command is the value of the backtick string,
3075       like in a shell.  In scalar context, a single string consisting of all
3076       output is returned.  In list context, a list of values is returned, one
3077       per line of output.  (You can set $/ to use a different line
3078       terminator.)  The command is executed each time the pseudo-literal is
3079       evaluated.  The status value of the command is returned in $? (see
3080       perlvar for the interpretation of $?).  Unlike in csh, no translation
3081       is done on the return data--newlines remain newlines.  Unlike in any of
3082       the shells, single quotes do not hide variable names in the command
3083       from interpretation.  To pass a literal dollar-sign through to the
3084       shell you need to hide it with a backslash.  The generalized form of
3085       backticks is "qx//", or you can call the "readpipe" in perlfunc
3086       function.  (Because backticks always undergo shell expansion as well,
3087       see perlsec for security concerns.)
3088
3089       In scalar context, evaluating a filehandle in angle brackets yields the
3090       next line from that file (the newline, if any, included), or "undef" at
3091       end-of-file or on error.  When $/ is set to "undef" (sometimes known as
3092       file-slurp mode) and the file is empty, it returns '' the first time,
3093       followed by "undef" subsequently.
3094
3095       Ordinarily you must assign the returned value to a variable, but there
3096       is one situation where an automatic assignment happens.  If and only if
3097       the input symbol is the only thing inside the conditional of a "while"
3098       statement (even if disguised as a "for(;;)" loop), the value is
3099       automatically assigned to the global variable $_, destroying whatever
3100       was there previously.  (This may seem like an odd thing to you, but
3101       you'll use the construct in almost every Perl script you write.)  The
3102       $_ variable is not implicitly localized.  You'll have to put a
3103       "local $_;" before the loop if you want that to happen.  Furthermore,
3104       if the input symbol or an explicit assignment of the input symbol to a
3105       scalar is used as a "while"/"for" condition, then the condition
3106       actually tests for definedness of the expression's value, not for its
3107       regular truth value.
3108
3109       Thus the following lines are equivalent:
3110
3111           while (defined($_ = <STDIN>)) { print; }
3112           while ($_ = <STDIN>) { print; }
3113           while (<STDIN>) { print; }
3114           for (;<STDIN>;) { print; }
3115           print while defined($_ = <STDIN>);
3116           print while ($_ = <STDIN>);
3117           print while <STDIN>;
3118
3119       This also behaves similarly, but assigns to a lexical variable instead
3120       of to $_:
3121
3122           while (my $line = <STDIN>) { print $line }
3123
3124       In these loop constructs, the assigned value (whether assignment is
3125       automatic or explicit) is then tested to see whether it is defined.
3126       The defined test avoids problems where the line has a string value that
3127       would be treated as false by Perl; for example a "" or a "0" with no
3128       trailing newline.  If you really mean for such values to terminate the
3129       loop, they should be tested for explicitly:
3130
3131           while (($_ = <STDIN>) ne '0') { ... }
3132           while (<STDIN>) { last unless $_; ... }
3133
3134       In other boolean contexts, "<FILEHANDLE>" without an explicit "defined"
3135       test or comparison elicits a warning if the "use warnings" pragma or
3136       the -w command-line switch (the $^W variable) is in effect.
3137
3138       The filehandles STDIN, STDOUT, and STDERR are predefined.  (The
3139       filehandles "stdin", "stdout", and "stderr" will also work except in
3140       packages, where they would be interpreted as local identifiers rather
3141       than global.)  Additional filehandles may be created with the "open()"
3142       function, amongst others.  See perlopentut and "open" in perlfunc for
3143       details on this.
3144
3145       If a "<FILEHANDLE>" is used in a context that is looking for a list, a
3146       list comprising all input lines is returned, one line per list element.
3147       It's easy to grow to a rather large data space this way, so use with
3148       care.
3149
3150       "<FILEHANDLE>"  may also be spelled "readline(*FILEHANDLE)".  See
3151       "readline" in perlfunc.
3152
3153       The null filehandle "<>" is special: it can be used to emulate the
3154       behavior of sed and awk, and any other Unix filter program that takes a
3155       list of filenames, doing the same to each line of input from all of
3156       them.  Input from "<>" comes either from standard input, or from each
3157       file listed on the command line.  Here's how it works: the first time
3158       "<>" is evaluated, the @ARGV array is checked, and if it is empty,
3159       $ARGV[0] is set to "-", which when opened gives you standard input.
3160       The @ARGV array is then processed as a list of filenames.  The loop
3161
3162           while (<>) {
3163               ...                     # code for each line
3164           }
3165
3166       is equivalent to the following Perl-like pseudo code:
3167
3168           unshift(@ARGV, '-') unless @ARGV;
3169           while ($ARGV = shift) {
3170               open(ARGV, $ARGV);
3171               while (<ARGV>) {
3172                   ...         # code for each line
3173               }
3174           }
3175
3176       except that it isn't so cumbersome to say, and will actually work.  It
3177       really does shift the @ARGV array and put the current filename into the
3178       $ARGV variable.  It also uses filehandle ARGV internally.  "<>" is just
3179       a synonym for "<ARGV>", which is magical.  (The pseudo code above
3180       doesn't work because it treats "<ARGV>" as non-magical.)
3181
3182       Since the null filehandle uses the two argument form of "open" in
3183       perlfunc it interprets special characters, so if you have a script like
3184       this:
3185
3186           while (<>) {
3187               print;
3188           }
3189
3190       and call it with "perl dangerous.pl 'rm -rfv *|'", it actually opens a
3191       pipe, executes the "rm" command and reads "rm"'s output from that pipe.
3192       If you want all items in @ARGV to be interpreted as file names, you can
3193       use the module "ARGV::readonly" from CPAN, or use the double bracket:
3194
3195           while (<<>>) {
3196               print;
3197           }
3198
3199       Using double angle brackets inside of a while causes the open to use
3200       the three argument form (with the second argument being "<"), so all
3201       arguments in "ARGV" are treated as literal filenames (including "-").
3202       (Note that for convenience, if you use "<<>>" and if @ARGV is empty, it
3203       will still read from the standard input.)
3204
3205       You can modify @ARGV before the first "<>" as long as the array ends up
3206       containing the list of filenames you really want.  Line numbers ($.)
3207       continue as though the input were one big happy file.  See the example
3208       in "eof" in perlfunc for how to reset line numbers on each file.
3209
3210       If you want to set @ARGV to your own list of files, go right ahead.
3211       This sets @ARGV to all plain text files if no @ARGV was given:
3212
3213           @ARGV = grep { -f && -T } glob('*') unless @ARGV;
3214
3215       You can even set them to pipe commands.  For example, this
3216       automatically filters compressed arguments through gzip:
3217
3218           @ARGV = map { /\.(gz|Z)$/ ? "gzip -dc < $_ |" : $_ } @ARGV;
3219
3220       If you want to pass switches into your script, you can use one of the
3221       "Getopts" modules or put a loop on the front like this:
3222
3223           while ($_ = $ARGV[0], /^-/) {
3224               shift;
3225               last if /^--$/;
3226               if (/^-D(.*)/) { $debug = $1 }
3227               if (/^-v/)     { $verbose++  }
3228               # ...           # other switches
3229           }
3230
3231           while (<>) {
3232               # ...           # code for each line
3233           }
3234
3235       The "<>" symbol will return "undef" for end-of-file only once.  If you
3236       call it again after this, it will assume you are processing another
3237       @ARGV list, and if you haven't set @ARGV, will read input from STDIN.
3238
3239       If what the angle brackets contain is a simple scalar variable (for
3240       example, $foo), then that variable contains the name of the filehandle
3241       to input from, or its typeglob, or a reference to the same.  For
3242       example:
3243
3244           $fh = \*STDIN;
3245           $line = <$fh>;
3246
3247       If what's within the angle brackets is neither a filehandle nor a
3248       simple scalar variable containing a filehandle name, typeglob, or
3249       typeglob reference, it is interpreted as a filename pattern to be
3250       globbed, and either a list of filenames or the next filename in the
3251       list is returned, depending on context.  This distinction is determined
3252       on syntactic grounds alone.  That means "<$x>" is always a "readline()"
3253       from an indirect handle, but "<$hash{key}>" is always a "glob()".
3254       That's because $x is a simple scalar variable, but $hash{key} is
3255       not--it's a hash element.  Even "<$x >" (note the extra space) is
3256       treated as "glob("$x ")", not "readline($x)".
3257
3258       One level of double-quote interpretation is done first, but you can't
3259       say "<$foo>" because that's an indirect filehandle as explained in the
3260       previous paragraph.  (In older versions of Perl, programmers would
3261       insert curly brackets to force interpretation as a filename glob:
3262       "<${foo}>".  These days, it's considered cleaner to call the internal
3263       function directly as "glob($foo)", which is probably the right way to
3264       have done it in the first place.)  For example:
3265
3266           while (<*.c>) {
3267               chmod 0644, $_;
3268           }
3269
3270       is roughly equivalent to:
3271
3272           open(FOO, "echo *.c | tr -s ' \t\r\f' '\\012\\012\\012\\012'|");
3273           while (<FOO>) {
3274               chomp;
3275               chmod 0644, $_;
3276           }
3277
3278       except that the globbing is actually done internally using the standard
3279       "File::Glob" extension.  Of course, the shortest way to do the above
3280       is:
3281
3282           chmod 0644, <*.c>;
3283
3284       A (file)glob evaluates its (embedded) argument only when it is starting
3285       a new list.  All values must be read before it will start over.  In
3286       list context, this isn't important because you automatically get them
3287       all anyway.  However, in scalar context the operator returns the next
3288       value each time it's called, or "undef" when the list has run out.  As
3289       with filehandle reads, an automatic "defined" is generated when the
3290       glob occurs in the test part of a "while", because legal glob returns
3291       (for example, a file called 0) would otherwise terminate the loop.
3292       Again, "undef" is returned only once.  So if you're expecting a single
3293       value from a glob, it is much better to say
3294
3295           ($file) = <blurch*>;
3296
3297       than
3298
3299           $file = <blurch*>;
3300
3301       because the latter will alternate between returning a filename and
3302       returning false.
3303
3304       If you're trying to do variable interpolation, it's definitely better
3305       to use the "glob()" function, because the older notation can cause
3306       people to become confused with the indirect filehandle notation.
3307
3308           @files = glob("$dir/*.[ch]");
3309           @files = glob($files[$i]);
3310
3311       If an angle-bracket-based globbing expression is used as the condition
3312       of a "while" or "for" loop, then it will be implicitly assigned to $_.
3313       If either a globbing expression or an explicit assignment of a globbing
3314       expression to a scalar is used as a "while"/"for" condition, then the
3315       condition actually tests for definedness of the expression's value, not
3316       for its regular truth value.
3317
3318   Constant Folding
3319       Like C, Perl does a certain amount of expression evaluation at compile
3320       time whenever it determines that all arguments to an operator are
3321       static and have no side effects.  In particular, string concatenation
3322       happens at compile time between literals that don't do variable
3323       substitution.  Backslash interpolation also happens at compile time.
3324       You can say
3325
3326             'Now is the time for all'
3327           . "\n"
3328           .  'good men to come to.'
3329
3330       and this all reduces to one string internally.  Likewise, if you say
3331
3332           foreach $file (@filenames) {
3333               if (-s $file > 5 + 100 * 2**16) {  }
3334           }
3335
3336       the compiler precomputes the number which that expression represents so
3337       that the interpreter won't have to.
3338
3339   No-ops
3340       Perl doesn't officially have a no-op operator, but the bare constants 0
3341       and 1 are special-cased not to produce a warning in void context, so
3342       you can for example safely do
3343
3344           1 while foo();
3345
3346   Bitwise String Operators
3347       Bitstrings of any size may be manipulated by the bitwise operators ("~
3348       | & ^").
3349
3350       If the operands to a binary bitwise op are strings of different sizes,
3351       | and ^ ops act as though the shorter operand had additional zero bits
3352       on the right, while the & op acts as though the longer operand were
3353       truncated to the length of the shorter.  The granularity for such
3354       extension or truncation is one or more bytes.
3355
3356           # ASCII-based examples
3357           print "j p \n" ^ " a h";            # prints "JAPH\n"
3358           print "JA" | "  ph\n";              # prints "japh\n"
3359           print "japh\nJunk" & '_____';       # prints "JAPH\n";
3360           print 'p N$' ^ " E<H\n";            # prints "Perl\n";
3361
3362       If you are intending to manipulate bitstrings, be certain that you're
3363       supplying bitstrings: If an operand is a number, that will imply a
3364       numeric bitwise operation.  You may explicitly show which type of
3365       operation you intend by using "" or "0+", as in the examples below.
3366
3367           $foo =  150  |  105;        # yields 255  (0x96 | 0x69 is 0xFF)
3368           $foo = '150' |  105;        # yields 255
3369           $foo =  150  | '105';       # yields 255
3370           $foo = '150' | '105';       # yields string '155' (under ASCII)
3371
3372           $baz = 0+$foo & 0+$bar;     # both ops explicitly numeric
3373           $biz = "$foo" ^ "$bar";     # both ops explicitly stringy
3374
3375       This somewhat unpredictable behavior can be avoided with the "bitwise"
3376       feature, new in Perl 5.22.  You can enable it via
3377       "use feature 'bitwise'" or "use v5.28".  Before Perl 5.28, it used to
3378       emit a warning in the "experimental::bitwise" category.  Under this
3379       feature, the four standard bitwise operators ("~ | & ^") are always
3380       numeric.  Adding a dot after each operator ("~. |. &. ^.") forces it to
3381       treat its operands as strings:
3382
3383           use feature "bitwise";
3384           $foo =  150  |  105;        # yields 255  (0x96 | 0x69 is 0xFF)
3385           $foo = '150' |  105;        # yields 255
3386           $foo =  150  | '105';       # yields 255
3387           $foo = '150' | '105';       # yields 255
3388           $foo =  150  |. 105;        # yields string '155'
3389           $foo = '150' |. 105;        # yields string '155'
3390           $foo =  150  |.'105';       # yields string '155'
3391           $foo = '150' |.'105';       # yields string '155'
3392
3393           $baz = $foo &  $bar;        # both operands numeric
3394           $biz = $foo ^. $bar;        # both operands stringy
3395
3396       The assignment variants of these operators ("&= |= ^= &.= |.= ^.=")
3397       behave likewise under the feature.
3398
3399       It is a fatal error if an operand contains a character whose ordinal
3400       value is above 0xFF, and hence not expressible except in UTF-8.  The
3401       operation is performed on a non-UTF-8 copy for other operands encoded
3402       in UTF-8.  See "Byte and Character Semantics" in perlunicode.
3403
3404       See "vec" in perlfunc for information on how to manipulate individual
3405       bits in a bit vector.
3406
3407   Integer Arithmetic
3408       By default, Perl assumes that it must do most of its arithmetic in
3409       floating point.  But by saying
3410
3411           use integer;
3412
3413       you may tell the compiler to use integer operations (see integer for a
3414       detailed explanation) from here to the end of the enclosing BLOCK.  An
3415       inner BLOCK may countermand this by saying
3416
3417           no integer;
3418
3419       which lasts until the end of that BLOCK.  Note that this doesn't mean
3420       everything is an integer, merely that Perl will use integer operations
3421       for arithmetic, comparison, and bitwise operators.  For example, even
3422       under "use integer", if you take the sqrt(2), you'll still get
3423       1.4142135623731 or so.
3424
3425       Used on numbers, the bitwise operators ("&" "|" "^" "~" "<<" ">>")
3426       always produce integral results.  (But see also "Bitwise String
3427       Operators".)  However, "use integer" still has meaning for them.  By
3428       default, their results are interpreted as unsigned integers, but if
3429       "use integer" is in effect, their results are interpreted as signed
3430       integers.  For example, "~0" usually evaluates to a large integral
3431       value.  However, "use integer; ~0" is "-1" on two's-complement
3432       machines.
3433
3434   Floating-point Arithmetic
3435       While "use integer" provides integer-only arithmetic, there is no
3436       analogous mechanism to provide automatic rounding or truncation to a
3437       certain number of decimal places.  For rounding to a certain number of
3438       digits, "sprintf()" or "printf()" is usually the easiest route.  See
3439       perlfaq4.
3440
3441       Floating-point numbers are only approximations to what a mathematician
3442       would call real numbers.  There are infinitely more reals than floats,
3443       so some corners must be cut.  For example:
3444
3445           printf "%.20g\n", 123456789123456789;
3446           #        produces 123456789123456784
3447
3448       Testing for exact floating-point equality or inequality is not a good
3449       idea.  Here's a (relatively expensive) work-around to compare whether
3450       two floating-point numbers are equal to a particular number of decimal
3451       places.  See Knuth, volume II, for a more robust treatment of this
3452       topic.
3453
3454           sub fp_equal {
3455               my ($X, $Y, $POINTS) = @_;
3456               my ($tX, $tY);
3457               $tX = sprintf("%.${POINTS}g", $X);
3458               $tY = sprintf("%.${POINTS}g", $Y);
3459               return $tX eq $tY;
3460           }
3461
3462       The POSIX module (part of the standard perl distribution) implements
3463       "ceil()", "floor()", and other mathematical and trigonometric
3464       functions.  The "Math::Complex" module (part of the standard perl
3465       distribution) defines mathematical functions that work on both the
3466       reals and the imaginary numbers.  "Math::Complex" is not as efficient
3467       as POSIX, but POSIX can't work with complex numbers.
3468
3469       Rounding in financial applications can have serious implications, and
3470       the rounding method used should be specified precisely.  In these
3471       cases, it probably pays not to trust whichever system rounding is being
3472       used by Perl, but to instead implement the rounding function you need
3473       yourself.
3474
3475   Bigger Numbers
3476       The standard "Math::BigInt", "Math::BigRat", and "Math::BigFloat"
3477       modules, along with the "bignum", "bigint", and "bigrat" pragmas,
3478       provide variable-precision arithmetic and overloaded operators,
3479       although they're currently pretty slow.  At the cost of some space and
3480       considerable speed, they avoid the normal pitfalls associated with
3481       limited-precision representations.
3482
3483               use 5.010;
3484               use bigint;  # easy interface to Math::BigInt
3485               $x = 123456789123456789;
3486               say $x * $x;
3487           +15241578780673678515622620750190521
3488
3489       Or with rationals:
3490
3491               use 5.010;
3492               use bigrat;
3493               $x = 3/22;
3494               $y = 4/6;
3495               say "x/y is ", $x/$y;
3496               say "x*y is ", $x*$y;
3497               x/y is 9/44
3498               x*y is 1/11
3499
3500       Several modules let you calculate with unlimited or fixed precision
3501       (bound only by memory and CPU time).  There are also some non-standard
3502       modules that provide faster implementations via external C libraries.
3503
3504       Here is a short, but incomplete summary:
3505
3506         Math::String           treat string sequences like numbers
3507         Math::FixedPrecision   calculate with a fixed precision
3508         Math::Currency         for currency calculations
3509         Bit::Vector            manipulate bit vectors fast (uses C)
3510         Math::BigIntFast       Bit::Vector wrapper for big numbers
3511         Math::Pari             provides access to the Pari C library
3512         Math::Cephes           uses the external Cephes C library (no
3513                                big numbers)
3514         Math::Cephes::Fraction fractions via the Cephes library
3515         Math::GMP              another one using an external C library
3516         Math::GMPz             an alternative interface to libgmp's big ints
3517         Math::GMPq             an interface to libgmp's fraction numbers
3518         Math::GMPf             an interface to libgmp's floating point numbers
3519
3520       Choose wisely.
3521
3522
3523
3524perl v5.32.1                      2021-05-31                         PERLOP(1)
Impressum