summaryrefslogtreecommitdiff
path: root/pod/perlpragma.pod
blob: 856014438e513eaa86e527563f9ebc4cc0fbe5d9 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
=head1 NAME

perlpragma - how to write a user pragma

=head1 DESCRIPTION

A pragma is a module which influences some aspect of the compile time or run
time behaviour of Perl, such as C<strict> or C<warnings>. With Perl 5.10 you
are no longer limited to the built in pragmata; you can now create user
pragmata that modify the behaviour of user functions within a lexical scope.

=head1 A basic example

For example, say you need to create a class implementing overloaded
mathematical operators, and would like to provide your own pragma that
functions much like C<use integer;> You'd like this code

    use MyMaths;
    
    my $l = MyMaths->new(1.2);
    my $r = MyMaths->new(3.4);
    
    print "A: ", $l + $r, "\n";
    
    use myint;
    print "B: ", $l + $r, "\n";
    
    {
        no myint;
        print "C: ", $l + $r, "\n";
    }
    
    print "D: ", $l + $r, "\n";
    
    no myint;
    print "E: ", $l + $r, "\n";

to give the output

    A: 4.6
    B: 4
    C: 4.6
    D: 4
    E: 4.6

I<i.e.>, where C<use myint;> is in effect, addition operations are forced
to integer, whereas by default they are not, with the default behaviour being
restored via C<no myint;>

The minimal implementation of the package C<MyMaths> would be something like
this:

    package MyMaths;
    use warnings;
    use strict;
    use myint();
    use overload '+' => sub {
        my ($l, $r) = @_;
	# Pass 1 to check up one call level from here
        if (myint::in_effect(1)) {
            int($$l) + int($$r);
        } else {
            $$l + $$r;
        }
    };
    
    sub new {
        my ($class, $value) = @_;
        bless \$value, $class;
    }
    
    1;

Note how we load the user pragma C<myint> with an empty list C<()> to
prevent its C<import> being called.

The interaction with the Perl compilation happens inside package C<myint>:

    package myint;
    
    use strict;
    use warnings;
    
    sub import {
        $^H{myint} = 1;
    }
    
    sub unimport {
        $^H{myint} = 0;
    }
    
    sub in_effect {
        my $level = shift // 0;
        my $hinthash = (caller($level))[10];
        return $hinthash->{myint};
    }
    
    1;

As pragmata are implemented as modules, like any other module, C<use myint;>
becomes

    BEGIN {
        require myint;
        myint->import();
    }

and C<no myint;> is

    BEGIN {
        require myint;
        myint->unimport();
    }

Hence the C<import> and C<unimport> routines are called at B<compile time>
for the user's code.

User pragmata store their state by writing to the magical hash C<%^H>,
hence these two routines manipulate it. The state information in C<%^H> is
stored in the optree, and can be retrieved read-only at runtime with C<caller()>,
at index 10 of the list of returned results. In the example pragma, retrieval
is encapsulated into the routine C<in_effect()>, which takes as parameter
the number of call frames to go up to find the value of the pragma in the
user's script. This uses C<caller()> to determine the value of
C<$^H{myint}> when each line of the user's script was called, and
therefore provide the correct semantics in the subroutine implementing the
overloaded addition.

=head1 Implementation details

The optree is shared between threads.  This means there is a possibility that
the optree will outlive the particular thread (and therefore the interpreter
instance) that created it, so true Perl scalars cannot be stored in the
optree.  Instead a compact form is used, which can only store values that are
integers (signed and unsigned), strings or C<undef> - references and
floating point values are stringified.  If you need to store multiple values
or complex structures, you should serialise them, for example with C<pack>.
The deletion of a hash key from C<%^H> is recorded, and as ever can be
distinguished from the existence of a key with value C<undef> with
C<exists>.

B<Don't> attempt to store references to data structures as integers which
are retrieved via C<caller> and converted back, as this will not be threadsafe.
Accesses would be to the structure without locking (which is not safe for
Perl's scalars), and either the structure has to leak, or it has to be
freed when its creating thread terminates, which may be before the optree
referencing it is deleted, if other threads outlive it.