Types-Serialiser/Serialiser.pm

=head1 NAME

Types::Serialiser - simple data types for common serialisation formats

=encoding utf-8

=head1 SYNOPSIS

=head1 DESCRIPTION

This module provides some extra datatypes that are used by common
serialisation formats such as JSON or CBOR. The idea is to have a
repository of simple/small constants and containers that can be shared by
different implementations so they become interoperable between each other.

=cut

package Types::Serialiser;

use common::sense; # required to suppress annoying warnings

our $VERSION = '1.01';

=head1 SIMPLE SCALAR CONSTANTS

Simple scalar constants are values that are overloaded to act like simple
Perl values, but have (class) type to differentiate them from normal Perl
scalars. This is necessary because these have different representations in
the serialisation formats.

In the following, functions with zero or one arguments have a prototype of
C<()> and C<($)>, respectively, so act as constants and unary operators.

=head2 BOOLEANS (Types::Serialiser::Boolean class)

This type has only two instances, true and false. A natural representation
for these in Perl is C<1> and C<0>, but serialisation formats need to be
able to differentiate between them and mere numbers.

=over 4

=item $Types::Serialiser::true, Types::Serialiser::true

This value represents the "true" value. In most contexts is acts like
the number C<1>. It is up to you whether you use the variable form
(C<$Types::Serialiser::true>) or the constant form (C<Types::Serialiser::true>).

The constant is represented as a reference to a scalar containing C<1> -
implementations are allowed to directly test for this.

=item $Types::Serialiser::false, Types::Serialiser::false

This value represents the "false" value. In most contexts is acts like
the number C<0>. It is up to you whether you use the variable form
(C<$Types::Serialiser::false>) or the constant form (C<Types::Serialiser::false>).

The constant is represented as a reference to a scalar containing C<0> -
implementations are allowed to directly test for this.

=item Types::Serialiser::as_bool $value

Converts a Perl scalar into a boolean, which is useful syntactic
sugar. Strictly equivalent to:

   $value ? $Types::Serialiser::true : $Types::Serialiser::false

=item $is_bool = Types::Serialiser::is_bool $value

Returns true iff the C<$value> is either C<$Types::Serialiser::true> or
C<$Types::Serialiser::false>.

For example, you could differentiate between a perl true value and a
C<Types::Serialiser::true> by using this:

   $value && Types::Serialiser::is_bool $value

=item $is_true = Types::Serialiser::is_true $value

Returns true iff C<$value> is C<$Types::Serialiser::true>.

=item $is_false = Types::Serialiser::is_false $value

Returns false iff C<$value> is C<$Types::Serialiser::false>.

=back

=head2 ERROR (Types::Serialiser::Error class)

This class has only a single instance, C<error>. It is used to signal
an encoding or decoding error. In CBOR for example, and object that
couldn't be encoded will be represented by a CBOR undefined value, which
is represented by the error value in Perl.

=over 4

=item $Types::Serialiser::error, Types::Serialiser::error

This value represents the "error" value. Accessing values of this type
will throw an exception.

The constant is represented as a reference to a scalar containing C<undef>
- implementations are allowed to directly test for this.

=item $is_error = Types::Serialiser::is_error $value

Returns false iff C<$value> is C<$Types::Serialiser::error>.

=back

=cut

BEGIN {
   # for historical reasons, and to avoid extra dependencies in JSON::PP,
   # we alias *Types::Serialiser::Boolean with JSON::PP::Boolean.
   package JSON::PP::Boolean;

   *Types::Serialiser::Boolean:: = *JSON::PP::Boolean::;
}

{
   # this must done before blessing to work around bugs
   # in perl < 5.18 (it seems to be fixed in 5.18).
   package Types::Serialiser::BooleanBase;

   use overload
      "0+"     => sub { ${$_[0]} },
      "++"     => sub { $_[0] = ${$_[0]} + 1 },
      "--"     => sub { $_[0] = ${$_[0]} - 1 },
      fallback => 1;

   @Types::Serialiser::Boolean::ISA = Types::Serialiser::BooleanBase::;
}

our $true  = do { bless \(my $dummy = 1), Types::Serialiser::Boolean:: };
our $false = do { bless \(my $dummy = 0), Types::Serialiser::Boolean:: };
our $error = do { bless \(my $dummy    ), Types::Serialiser::Error::   };

sub true  () { $true  }
sub false () { $false }
sub error () { $error }

sub as_bool($) { $_[0] ? $true : $false }

sub is_bool  ($) {           UNIVERSAL::isa $_[0], Types::Serialiser::Boolean:: }
sub is_true  ($) {  $_[0] && UNIVERSAL::isa $_[0], Types::Serialiser::Boolean:: }
sub is_false ($) { !$_[0] && UNIVERSAL::isa $_[0], Types::Serialiser::Boolean:: }
sub is_error ($) {           UNIVERSAL::isa $_[0], Types::Serialiser::Error::   }

package Types::Serialiser::Error;

sub error {
   require Carp;
   Carp::croak ("caught attempt to use the Types::Serialiser::error value");
};

use overload
   "0+"     => \&error,
   "++"     => \&error,
   "--"     => \&error,
   fallback => 1;

=head1 NOTES FOR XS USERS

The recommended way to detect whether a scalar is one of these objects
is to check whether the stash is the C<Types::Serialiser::Boolean> or
C<Types::Serialiser::Error> stash, and then follow the scalar reference to
see if it's C<1> (true), C<0> (false) or C<undef> (error).

While it is possible to use an isa test, directly comparing stash pointers
is faster and guaranteed to work.

For historical reasons, the C<Types::Serialiser::Boolean> stash is
just an alias for C<JSON::PP::Boolean>. When printed, the classname
with usually be C<JSON::PP::Boolean>, but isa tests and stash pointer
comparison will normally work correctly (i.e. Types::Serialiser::true ISA
JSON::PP::Boolean, but also ISA Types::Serialiser::Boolean).

=head1 A GENERIC OBJECT SERIALIATION PROTOCOL

This section explains the object serialisation protocol used by
L<CBOR::XS>. It is meant to be generic enough to support any kind of
generic object serialiser.

This protocol is called "the Types::Serialiser object serialisation
protocol".

=head2 ENCODING

When the encoder encounters an object that it cannot otherwise encode (for
example, L<CBOR::XS> can encode a few special types itself, and will first
attempt to use the special C<TO_CBOR> serialisation protocol), it will
look up the C<FREEZE> method on the object.

Note that the C<FREEZE> method will normally be called I<during> encoding,
and I<MUST NOT> change the data structure that is being encoded in any
way, or it might cause memory corruption or worse.

If it exists, it will call it with two arguments: the object to serialise,
and a constant string that indicates the name of the data model. For
example L<CBOR::XS> uses C<CBOR>, and the L<JSON> and L<JSON::XS> modules
(or any other JSON serialiser), would use C<JSON> as second argument.

The C<FREEZE> method can then return zero or more values to identify the
object instance. The serialiser is then supposed to encode the class name
and all of these return values (which must be encodable in the format)
using the relevant form for Perl objects. In CBOR for example, there is a
registered tag number for encoded perl objects.

The values that C<FREEZE> returns must be serialisable with the serialiser
that calls it. Therefore, it is recommended to use simple types such as
strings and numbers, and maybe array references and hashes (basically, the
JSON data model). You can always use a more complex format for a specific
data model by checking the second argument, the data model.

The "data model" is not the same as the "data format" - the data model
indicates what types and kinds of return values can be returned from
C<FREEZE>. For example, in C<CBOR> it is permissible to return tagged CBOR
values, while JSON does not support these at all, so C<JSON> would be a
valid (but too limited) data model name for C<CBOR::XS>. similarly, a
serialising format that supports more or less the same data model as JSON
could use C<JSON> as data model without losing anything.

=head2 DECODING

When the decoder then encounters such an encoded perl object, it should
look up the C<THAW> method on the stored classname, and invoke it with the
classname, the constant string to identify the data model/data format, and
all the return values returned by C<FREEZE>.

=head2 EXAMPLES

See the C<OBJECT SERIALISATION> section in the L<CBOR::XS> manpage for
more details, an example implementation, and code examples.

Here is an example C<FREEZE>/C<THAW> method pair:

   sub My::Object::FREEZE {
      my ($self, $model) = @_;

      ($self->{type}, $self->{id}, $self->{variant})
   }

   sub My::Object::THAW {
      my ($class, $model, $type, $id, $variant) = @_;

      $class->new (type => $type, id => $id, variant => $variant)
   }

=head1 BUGS

The use of L<overload> makes this module much heavier than it should be
(on my system, this module: 4kB RSS, overload: 260kB RSS).

=head1 SEE ALSO

Currently, L<JSON::XS> and L<CBOR::XS> use these types.

=head1 AUTHOR

 Marc Lehmann <schmorp@schmorp.de>
 http://home.schmorp.de/

=cut

1

Revision:	1.10
Committed:	Tue Dec 1 01:47:20 2020 UTC (3 years, 7 months ago) by root
Branch:	MAIN
CVS Tags:	rel-1_01, HEAD
Changes since 1.9:	+13 -1 lines
Log Message:	1.01
#	User	Rev	Content
1	root	1.1	=head1 NAME
2
3			Types::Serialiser - simple data types for common serialisation formats
4
5			=encoding utf-8
6
7			=head1 SYNOPSIS
8
9			=head1 DESCRIPTION
10
11			This module provides some extra datatypes that are used by common
12			serialisation formats such as JSON or CBOR. The idea is to have a
13			repository of simple/small constants and containers that can be shared by
14			different implementations so they become interoperable between each other.
15
16			=cut
17
18			package Types::Serialiser;
19
20	root	1.4	use common::sense; # required to suppress annoying warnings
21
22	root	1.10	our $VERSION = '1.01';
23	root	1.1
24			=head1 SIMPLE SCALAR CONSTANTS
25
26			Simple scalar constants are values that are overloaded to act like simple
27			Perl values, but have (class) type to differentiate them from normal Perl
28			scalars. This is necessary because these have different representations in
29			the serialisation formats.
30
31	root	1.10	In the following, functions with zero or one arguments have a prototype of
32			C<()> and C<($)>, respectively, so act as constants and unary operators.
33
34	root	1.1	=head2 BOOLEANS (Types::Serialiser::Boolean class)
35
36			This type has only two instances, true and false. A natural representation
37			for these in Perl is C<1> and C<0>, but serialisation formats need to be
38			able to differentiate between them and mere numbers.
39
40			=over 4
41
42			=item $Types::Serialiser::true, Types::Serialiser::true
43
44			This value represents the "true" value. In most contexts is acts like
45			the number C<1>. It is up to you whether you use the variable form
46			(C<$Types::Serialiser::true>) or the constant form (C<Types::Serialiser::true>).
47
48			The constant is represented as a reference to a scalar containing C<1> -
49			implementations are allowed to directly test for this.
50
51			=item $Types::Serialiser::false, Types::Serialiser::false
52
53			This value represents the "false" value. In most contexts is acts like
54			the number C<0>. It is up to you whether you use the variable form
55			(C<$Types::Serialiser::false>) or the constant form (C<Types::Serialiser::false>).
56
57			The constant is represented as a reference to a scalar containing C<0> -
58			implementations are allowed to directly test for this.
59
60	root	1.10	=item Types::Serialiser::as_bool $value
61
62			Converts a Perl scalar into a boolean, which is useful syntactic
63			sugar. Strictly equivalent to:
64
65			$value ? $Types::Serialiser::true : $Types::Serialiser::false
66
67	root	1.1	=item $is_bool = Types::Serialiser::is_bool $value
68
69			Returns true iff the C<$value> is either C<$Types::Serialiser::true> or
70			C<$Types::Serialiser::false>.
71
72			For example, you could differentiate between a perl true value and a
73			C<Types::Serialiser::true> by using this:
74
75			$value && Types::Serialiser::is_bool $value
76
77			=item $is_true = Types::Serialiser::is_true $value
78
79			Returns true iff C<$value> is C<$Types::Serialiser::true>.
80
81			=item $is_false = Types::Serialiser::is_false $value
82
83			Returns false iff C<$value> is C<$Types::Serialiser::false>.
84
85			=back
86
87			=head2 ERROR (Types::Serialiser::Error class)
88
89			This class has only a single instance, C<error>. It is used to signal
90			an encoding or decoding error. In CBOR for example, and object that
91			couldn't be encoded will be represented by a CBOR undefined value, which
92			is represented by the error value in Perl.
93
94			=over 4
95
96			=item $Types::Serialiser::error, Types::Serialiser::error
97
98			This value represents the "error" value. Accessing values of this type
99			will throw an exception.
100
101			The constant is represented as a reference to a scalar containing C<undef>
102			- implementations are allowed to directly test for this.
103
104			=item $is_error = Types::Serialiser::is_error $value
105
106			Returns false iff C<$value> is C<$Types::Serialiser::error>.
107
108			=back
109
110			=cut
111
112	root	1.4	BEGIN {
113			# for historical reasons, and to avoid extra dependencies in JSON::PP,
114			# we alias *Types::Serialiser::Boolean with JSON::PP::Boolean.
115			package JSON::PP::Boolean;
116	root	1.6
117	root	1.4	Types::Serialiser::Boolean:: = JSON::PP::Boolean::;
118			}
119
120	root	1.6	{
121			# this must done before blessing to work around bugs
122			# in perl < 5.18 (it seems to be fixed in 5.18).
123			package Types::Serialiser::BooleanBase;
124
125			use overload
126			"0+" => sub { ${$_[0]} },
127			"++" => sub { $_[0] = ${$_[0]} + 1 },
128			"--" => sub { $_[0] = ${$_[0]} - 1 },
129			fallback => 1;
130
131			@Types::Serialiser::Boolean::ISA = Types::Serialiser::BooleanBase::;
132			}
133
134	root	1.1	our $true = do { bless \(my $dummy = 1), Types::Serialiser::Boolean:: };
135			our $false = do { bless \(my $dummy = 0), Types::Serialiser::Boolean:: };
136			our $error = do { bless \(my $dummy ), Types::Serialiser::Error:: };
137
138			sub true () { $true }
139			sub false () { $false }
140			sub error () { $error }
141
142	root	1.10	sub as_bool($) { $_[0] ? $true : $false }
143
144	root	1.1	sub is_bool ($) { UNIVERSAL::isa $_[0], Types::Serialiser::Boolean:: }
145			sub is_true ($) { $_[0] && UNIVERSAL::isa $_[0], Types::Serialiser::Boolean:: }
146			sub is_false ($) { !$_[0] && UNIVERSAL::isa $_[0], Types::Serialiser::Boolean:: }
147			sub is_error ($) { UNIVERSAL::isa $_[0], Types::Serialiser::Error:: }
148
149			package Types::Serialiser::Error;
150
151			sub error {
152			require Carp;
153	root	1.2	Carp::croak ("caught attempt to use the Types::Serialiser::error value");
154	root	1.1	};
155
156			use overload
157			"0+" => \&error,
158			"++" => \&error,
159			"--" => \&error,
160			fallback => 1;
161
162	root	1.2	=head1 NOTES FOR XS USERS
163
164			The recommended way to detect whether a scalar is one of these objects
165			is to check whether the stash is the C<Types::Serialiser::Boolean> or
166			C<Types::Serialiser::Error> stash, and then follow the scalar reference to
167			see if it's C<1> (true), C<0> (false) or C<undef> (error).
168
169			While it is possible to use an isa test, directly comparing stash pointers
170			is faster and guaranteed to work.
171
172	root	1.4	For historical reasons, the C<Types::Serialiser::Boolean> stash is
173			just an alias for C<JSON::PP::Boolean>. When printed, the classname
174	root	1.8	with usually be C<JSON::PP::Boolean>, but isa tests and stash pointer
175	root	1.4	comparison will normally work correctly (i.e. Types::Serialiser::true ISA
176			JSON::PP::Boolean, but also ISA Types::Serialiser::Boolean).
177
178	root	1.3	=head1 A GENERIC OBJECT SERIALIATION PROTOCOL
179
180			This section explains the object serialisation protocol used by
181			L<CBOR::XS>. It is meant to be generic enough to support any kind of
182			generic object serialiser.
183
184			This protocol is called "the Types::Serialiser object serialisation
185			protocol".
186
187			=head2 ENCODING
188
189			When the encoder encounters an object that it cannot otherwise encode (for
190			example, L<CBOR::XS> can encode a few special types itself, and will first
191			attempt to use the special C<TO_CBOR> serialisation protocol), it will
192			look up the C<FREEZE> method on the object.
193
194	root	1.9	Note that the C<FREEZE> method will normally be called I<during> encoding,
195			and I<MUST NOT> change the data structure that is being encoded in any
196			way, or it might cause memory corruption or worse.
197
198	root	1.8	If it exists, it will call it with two arguments: the object to serialise,
199	root	1.9	and a constant string that indicates the name of the data model. For
200			example L<CBOR::XS> uses C<CBOR>, and the L<JSON> and L<JSON::XS> modules
201	root	1.8	(or any other JSON serialiser), would use C<JSON> as second argument.
202	root	1.3
203			The C<FREEZE> method can then return zero or more values to identify the
204			object instance. The serialiser is then supposed to encode the class name
205			and all of these return values (which must be encodable in the format)
206	root	1.9	using the relevant form for Perl objects. In CBOR for example, there is a
207	root	1.3	registered tag number for encoded perl objects.
208
209	root	1.5	The values that C<FREEZE> returns must be serialisable with the serialiser
210			that calls it. Therefore, it is recommended to use simple types such as
211			strings and numbers, and maybe array references and hashes (basically, the
212			JSON data model). You can always use a more complex format for a specific
213	root	1.9	data model by checking the second argument, the data model.
214
215			The "data model" is not the same as the "data format" - the data model
216			indicates what types and kinds of return values can be returned from
217			C<FREEZE>. For example, in C<CBOR> it is permissible to return tagged CBOR
218			values, while JSON does not support these at all, so C<JSON> would be a
219			valid (but too limited) data model name for C<CBOR::XS>. similarly, a
220			serialising format that supports more or less the same data model as JSON
221			could use C<JSON> as data model without losing anything.
222	root	1.5
223	root	1.3	=head2 DECODING
224
225			When the decoder then encounters such an encoded perl object, it should
226			look up the C<THAW> method on the stored classname, and invoke it with the
227	root	1.8	classname, the constant string to identify the data model/data format, and
228			all the return values returned by C<FREEZE>.
229	root	1.3
230			=head2 EXAMPLES
231
232			See the C<OBJECT SERIALISATION> section in the L<CBOR::XS> manpage for
233			more details, an example implementation, and code examples.
234
235			Here is an example C<FREEZE>/C<THAW> method pair:
236
237			sub My::Object::FREEZE {
238	root	1.8	my ($self, $model) = @_;
239	root	1.3
240			($self->{type}, $self->{id}, $self->{variant})
241			}
242
243			sub My::Object::THAW {
244	root	1.8	my ($class, $model, $type, $id, $variant) = @_;
245	root	1.3
246	root	1.7	$class->new (type => $type, id => $id, variant => $variant)
247	root	1.3	}
248
249	root	1.1	=head1 BUGS
250
251			The use of L<overload> makes this module much heavier than it should be
252			(on my system, this module: 4kB RSS, overload: 260kB RSS).
253
254			=head1 SEE ALSO
255
256			Currently, L<JSON::XS> and L<CBOR::XS> use these types.
257
258			=head1 AUTHOR
259
260			Marc Lehmann <schmorp@schmorp.de>
261			http://home.schmorp.de/
262
263			=cut
264
265			1
266