1 # Abstract serialization services
3 The serialization services are based on the `serialize` and the `noserialize` annotations,
4 the `Serializable` interface and the implementations of `Serializer` and `Deserializer`.
6 ## The `serialize` annotation
8 A class annotated with `serialize` identifies it as a subclass of Serializable and
9 triggers the generation of customized serialization and deserialization services.
14 # Simple serializable class identifying a human
21 # Year of birth (`null` if unknown)
22 var birth: nullable Int
24 redef fun ==(o) do return o isa SELF and name == o.name and birth == o.birth
25 redef fun hash do return name.hash
29 The `Person` class also defines `==` and `hash`, this is optional but we will use it to make an important point.
30 By definition of a serializable class, an instance can be serialized to a stream, then deserialized.
31 The deserialized instance will not be the same instance, but they should be equal.
32 So, in this case, we can compare both instances with `==` to test their equality.
34 Some conditions applies to the classes that can be annotated as `serialize`.
35 All attributes of the class must be serializable, runtime errors will be
36 raised when trying to serialize non-serializable attributes.
38 In the class `Person`, all attributes are typed with classes the standards library.
39 These common types are defined defined as serializable by this project.
40 The attributes could also be typed with user-defined `serialize`
41 classes or any other subclass of `Serializable`.
44 # This `serialize` class is composed of two `serialize` attributes
51 redef fun ==(o) do return o isa SELF and partner_a == o.partner_a and partner_b == o.partner_b
52 redef fun hash do return partner_a.hash + 1024*partner_b.hash
56 ### Scope of the `serialize` annotation
58 `serialize` can annotate class definitions, modules and attributes:
60 * The annotation on a class applies only to the class definition,
61 only attributes declared locally will be serialized.
62 However, each definition of a class (a refinement or specialization) can be annotated with `serialize`.
64 * A module declaration annotated with `serialize` states that all its class definitions
65 and locally declared attributes are serializable.
68 module shared_between_clients is serialize
71 * Attribute annotated with `serialize` states that it is to be serialized, when the rest of the class does not.
72 The class will become subclass to `Serializable` but its attributes are not to be serialized by default.
73 Only the attributes with the `serialize` annotation will be serialized.
76 # Only serialize the `name`
78 var name: String is serialize
79 var avatar_path: String = "/somepath/"+name is lazy
83 ## The `noserialize` annotation
85 The `noserialize` annotation mark an exception in a `serialize` module or class definition.
87 * By default a module is `noserialize`. There is no need to declare it as such.
89 * A class definition annotated with `noserialize` within a `serialize` module will not be made serializable.
91 * A `noserialize` attribute within a class or module annotated with `serialize` will not serialize this attribute.
92 The class will still be made subclass of `Serializable` and it won't affect the other attributes.
93 The `noserialize` attribute will not be set at deserialization.
94 Usually, it will also be annotated with `lazy` to get its value by another mean after the object has been deserialized.
97 # Once again, only serialize the `name`
102 var avatar_path: String = "/somepath/"+name is noserialize, lazy
106 ## Custom serializable classes
108 The annotation `serialize` should be enough for most cases,
109 but in some cases you need more control over the serialization process.
111 For more control, create a subclass to `Serializable` and redefine `core_serialize_to`.
112 This method should use `Serializer::serialize_attribute` to serialize its components.
113 `serialize_attribute` works as a dictionary and organize attributes with a key.
115 You will also need to redefine `Deserializer::deserialize_class` to support this specific class.
116 The method should only act on known class names, and call super otherwise.
118 ### Example: the User class
120 The following example cannot use the `serialize` annotations
121 because some of the arguments to the `User` class need special treatment:
123 * The `name` attribute is perfectly normal, it can be serialized and deserialized
126 * The `password` attribute must be encrypted before being serialized,
127 and unencrypted on deserialization.
129 * The `avatar` attributes is kept as ASCII art in memory.
130 It could be serialized as such but it is cleaner to only
131 serialize the path to its source on the file system.
132 The data is reloaded on deserialization.
134 For this customization, the following code snippet implements
135 two serialization services: `User::core_serialize_to` and
136 `Deserializer::deserialize_class`.
139 module user_credentials
141 # User credentials for a website
148 # Clear text password
151 # User's avatar image as data blob
154 redef fun core_serialize_to(serializer: Serializer)
156 # This is the normal serialization process
157 serializer.serialize_attribute("name", name)
159 # Serialized an encrypted version of the password
161 # Obviously, `rot(13)` is not a good encrption
162 serializer.serialize_attribute("pass", password.rot(13))
164 # Do not serialize the image, only its path
165 serializer.serialize_attribute("avatar_path", avatar.path)
169 redef class Deserializer
170 redef fun deserialize_class(name)
172 if name == "User" then
173 # Deserialize normally
174 var user = deserialize_attribute("name")
177 var pass = deserialize_attribute("pass").rot(-13)
179 # Deserialize the path and load the avatar from the file system
180 var avatar_path = deserialize_attribute("avatar_path")
181 var avatar = new Image(avatar_path)
183 return new User(user, pass, avatar)
190 # An image loaded in memory as ASCII art
192 # Not really useful for this example, provided for consistency only.
194 # Path on the filesystem for `self`
197 # ASCII art composing this image
198 var ascii_art: String = path.read_all is lazy
203 See the documentation of the module `serialization::serialization` for more
204 information on the services to redefine.
206 ## Serialization services
208 The `serialize` annotation and the `Serializable` class are used on
209 classes specific to the business domain.
210 To write (and read) instances of these classes to a persistent format
211 you must use implementations of `Serializer` and `Deserializer`.
213 The main implementations of these services are `JsonSerializer` and `JsonDeserializer`,
214 from the `json_serialization` module.
217 import json_serialization
218 import user_credentials
220 # Data to be serialized and deserialized
221 var couple = new Partnership(
222 new Person("Alice", 1985, new Image("alice.png")),
223 new Person("Bob", null, new Image("bob.png")))
225 var path = "serialized_data.json"
226 var writer = new FileWriter(path)
227 var serializer = new JsonSerializer(writer)
228 serializer.serialize couple
231 var reader = new FileReader(path)
232 var deserializer = new JsonDeserializer(reader.to_s)
233 var deserialized_couple = deserializer.deserialize
236 assert couple == deserialize_couple
239 ## Limitations and TODO
241 The serialization has some limitations:
243 * Not enough classes from the standard library are supported.
244 This only requires someone to actually code the support.
245 It should not be especially hard for most classes, some can
246 simply declare the `serialize` annotation.
248 * A limitation of the Json parser prevents deserializing from files
249 with more than one object.
250 This could be improved in the future, but for now you should
251 serialize a single object to each filesand use different instances of
252 serializer and deserializer each time.
254 * The `serialize` annotation does not handle very well
255 complex constructors. This could be improved in the compiler.
256 For now, you may prefer to use `serialize` on simple classes,
257 of by using custom `Serializable`.
259 * The serialization uses only the short name of a class, not its qualified name.
260 This will cause problem when different classes using the same name.
261 This could be solved partially in the compiler and the library.
262 A special attention must be given to the consistency of the name across
263 the different programs sharing the serialized data.
265 * The serialization support in the compiler need some help to
266 deal with generic types. The solution is to use `nitserial`,
267 the next section explores this subject.
269 ## Dealing with generic types
271 One limitation of the serialization support in the compiler is with generic types.
272 For example, the `Array` class is generic and serializable.
273 However, the runtime types of Array instances are parameterized and are unknown to the compiler.
274 So the compiler won't support serializing instances of `Array[MySerializable]`.
276 The tool `nitserial` solves this problem at the level of user modules.
277 It does so by parsing a Nit module, group or project to find all known
278 parameterized types of generic classes.
279 It will then generating a Nit module to handle deserialization of these types.
281 Usage steps to serialize parameterized types:
283 * Write your program, let's call it `my_prog.nit`,
284 it must use some parameterized serializable types.
285 Let's say that you use `Array[MySerializable]`.
287 * Run nitserial using `nitserial my_prog.nit` to
288 generate the file `my_prog_serial.nit`.
290 * Compile your program by mixing in the generated module with:
291 `nitc my_prog.nit -m my_prog_serial.nit`
293 This was a simple example, in practical cases you may need
294 to use more than one generated file.
295 For example, on a client/server system, an instance can be created
296 server-side, serialized and the used client-side.
297 In this case, two files will be generated by nitserial,
298 one for the server and one for the client.
299 Both the files should be compiled with both the client and the server.