Hmm, here how I see things:
- Adapters are insanely great for end users (I include myself there) cause
it is easy to implement but they are limited by design cause you get a
model and return another model, so no in between processing (+ it can be
super verbose in some cases)
So
- We need a lower level API where you can, without creating any model,
serialize using jsonp + a thin layer on top integrating with the mapper
So we would get the business adapters (current ones) and the technical ones
("(De)Serializer"?). This would allow all users to do any fancy things when
needed - and even if 20% of the cases it is a real life need - keeping by
default a smooth API.
I don't think it is a lot of work since we already discussed it a bit in
this thread and cause jackson, the RI implementations (and probably other)
already have such an API.
Si it would need:
- JsonbGenerator (extends JsonGenerator { void write(String field, Object
object); void write(Object object) }), JsonbReader (extends JsonReader) { T
read(Type type)}
- Serializer { void serialize(T object, JsonbGenerator) }/Deserializer{ T
deserialize(JsonStructure json, JsonbReader reader) } API
- JsonbConfig with(De)Serializers() methods
Side note: this also handle more advanced cases I encountered working with
ElasticSearch where you need to modify the way the mapper serialize the
model at runtime. More generally anytime you try to provide a framework you
need such trick cause you don't fully control the model and you don't want
to rescan the whole model (= reimplement a part of the mapper).
Wdyt?
Romain Manni-Bucau
@rmannibucau
http://www.tomitribe.com
http://rmannibucau.wordpress.com
https://github.com/rmannibucau
2016-03-24 12:14 GMT+01:00 Roman Grigoriadi <roman.grigoriadi_at_oracle.com>:
> It depends if we include TypeWrapperAdapter into spec and API. If not,
> than yes it would be a proprietary of jsonb-ri.
>
> Roman.
>
>
> On 03/24/2016 12:09 PM, Romain Manni-Bucau wrote:
>
> Am I right saying it uses a proprietary API to work?
>
> If so I think my JsonbGenerator/JsonbReader are a standardization of such
> API and what Mark proposed as well. Then maybe we can discuss their
> integration in the spec?
> Le 24 mars 2016 12:03, "Roman Grigoriadi" <roman.grigoriadi_at_oracle.com> a
> écrit :
>
>> Hi,
>>
>> @Romain - I added a test -
>> org.eclipse.persistence.json.bind.adapters.PolymorphismAdapterTest.
>> It uses these pojos:
>>
>> public static class Animal {
>> public String name;}
>> public static class Dog extends Animal {
>> public String dogProperty;}
>> public static class Cat extends Animal {
>> public String catProperty;}
>> public static class Pojo {
>> public Animal animal; public List<Animal> listOfAnimals = new ArrayList<>();}
>>
>> and generates this JSON for accordingly instantiated pojo:
>>
>> {
>> "animal":{
>> "className":"org.eclipse.persistence.json.bind.adapters.PolymorphismAdapterTest$Dog",
>> "instance":{
>> "name":"Ralph",
>> "dogProperty":"Property of a Dog."
>> }
>> },
>> "listOfAnimals":[
>> {
>> "className":"org.eclipse.persistence.json.bind.adapters.PolymorphismAdapterTest$Cat",
>> "instance":{
>> "name":"Snowball",
>> "catProperty":"Property of a Cat."
>> }
>> },
>> {
>> "className":"org.eclipse.persistence.json.bind.adapters.PolymorphismAdapterTest$Dog",
>> "instance":{
>> "name":"Ralph",
>> "dogProperty":"Property of a Dog."
>> }
>> }
>> ]
>> }
>>
>> An
>> org.eclipse.persistence.json.bind.internal.unmarshaller.TypeWrapperItem
>> class is responsible for
>> *"how "Jsonb handles such event with creating instance of appropriate
>> type"". *Please see it at eclipselink repo
>> <http://git.eclipse.org/c/eclipselink/eclipselink.runtime.git>, master
>> branch. The Jsonb project is in eclipselink-runtime/jsonb folder.
>>
>> I don't know what are the internals of Johnzon, but I suppose it will
>> also include some custom handling for specified attribute in case of
>> TypeWrapper class. Maybe I just missed to specify explicitly in previous
>> communication, that I have to add this custom handling in JSONB. Of course
>> it will never work without it..
>>
>> Regards,
>> Roman
>>
>> On 03/23/2016 02:26 PM, Romain Manni-Bucau wrote:
>>
>> Will wait for the test cause I really don't see how "Jsonb handles such
>> event with creating instance of appropriate type" but wanted to share the
>> read before write solution works even if not elegant using the
>> typewrapper/adatper we spoke about:
>>
>>
>> public void ploymorphism() { // we run it since it checked list/item
>> conversion
>>
>> final Bar bar = new Bar(); // ~your Animal
>>
>> bar.value = 11;
>>
>>
>> final Bar2 bar2 = new Bar2();
>>
>> bar2.value = 21;
>>
>> bar2.value2 = 22;
>>
>>
>> final Polymorphism foo = new Polymorphism(); // the pojo with a
>> collection of bars
>>
>> foo.bars = new ArrayList<>(asList(bar, bar2));
>>
>>
>> final Jsonb jsonb = JsonbBuilder.create(
>>
>> new JsonbConfig()
>>
>> .setProperty("johnzon.readAttributeBeforeWrite", true) // NOT STANDARD,
>> enforces to use the getter to determine the type to deserialize
>>
>> .withPropertyOrderStrategy(PropertyOrderStrategy.LEXICOGRAPHICAL) /*
>> assertEquals() order */);
>>
>>
>> final String toString = jsonb.toJson(foo);
>>
>> assertEquals("{\"bars\":[" +
>>
>> "{\"type\":\"org.apache.johnzon.jsonb.AdapterTest$Bar\",\"value\":{\"
>> value\":11}}," +
>>
>> "{\"type\":\"org.apache.johnzon.jsonb.AdapterTest$Bar2\",\"value\":{\"
>> value\":21,\"value2\":22}}]}", toString);
>>
>>
>> final Polymorphism read = jsonb.fromJson(toString, Polymorphism.class);
>>
>> assertEquals(2, read.bars.size());
>>
>> assertEquals(11, read.bars.get(0).value);
>>
>> assertTrue(Bar.class == read.bars.get(0).getClass());
>>
>> assertEquals(21, read.bars.get(1).value);
>>
>> assertTrue(Bar2.class == read.bars.get(1).getClass());
>>
>> assertEquals(22, Bar2.class.cast(read.bars.get(1)).value2);
>>
>> }
>>
>>
>>
>>
>> 2016-03-23 14:18 GMT+01:00 Roman Grigoriadi <
>> <roman.grigoriadi_at_oracle.com>roman.grigoriadi_at_oracle.com>:
>>
>>>
>>>
>>> On 03/23/2016 02:02 PM, Romain Manni-Bucau wrote:
>>>
>>>
>>>
>>> 2016-03-23 13:39 GMT+01:00 Roman Grigoriadi <
>>> <roman.grigoriadi_at_oracle.com>roman.grigoriadi_at_oracle.com>:
>>>
>>>> I think [A] is misunderstood..
>>>>
>>>>
>>>>
>>>> *[A] Think I get it: the adapter will return Foo and then on this
>>>> instance Jsonb will populate fields. * That is not what I ment. An
>>>> adapter is at the end of the process, T instance is already populated both:
>>>> when adapter adapts it returns adapted result.
>>>> Above statement is maybe enough, but for sake of clarity let me revisit
>>>> whole process.
>>>>
>>>> 1. Lets have JSON document for a TypeWrapper<T>
>>>> {
>>>> "animal":{
>>>> "className":"com.foo.Dog",
>>>> "instance":{
>>>> "name":"Doggie",
>>>> "dogProperty":"Property of a Dog."
>>>> }
>>>> }
>>>> }
>>>>
>>>> During deserialization what happens first is a completed instance of a
>>>> TypeWrapper with a a fully populated T inside. This happens in following
>>>> steps:
>>>> 1. Jsonb encounters deserializing a TypeWrapper<T>, and processes it
>>>> custom TypeWrapperDeserializer.
>>>> 2. TypeWrapperDeserializer (implemented by us), creates an empty
>>>> instance of TypeWrapper.
>>>> 3. Jsonb reads and sets property "className" into TypeWrapper
>>>> 4. Jsonb loads a class by this classname and creates instance with
>>>> default construcotor in our case Dog or Cat and sets it to TypeWrapper
>>>> instance.
>>>>
>>>
>>> That's where I'm lost. Only way it could work - without putting this
>>> behavior in the spec - is if TypeWrapper overrides the setter of classname
>>> to also set the instance:
>>>
>>> public void setClassName(final String className) {
>>> this.className= className;
>>> try {
>>> this.value = Animal.class.cast(Thread.currentThread().getContextClassLoader().loadClass(className).newInstance());
>>> } catch (final InstantiationException | IllegalAccessException | ClassNotFoundException e) {
>>> throw new IllegalArgumentException(e);
>>> }
>>> }
>>>
>>> Nope, it is only the className set as a String in this moment.
>>> Next JsonParser enconters a JsonObject and pushes ObjectStarted event
>>> into JSONB. Jsonb handles such event with creating instance of appropriate
>>> type. That type is normally read from ClassModel (which is read from actual
>>> Field or ParameterizedType argument of a Collection). But in our case this
>>> type is swapped with a Class loaded dynamically by name.
>>>
>> So now we have a correct instance ready to be populated by common
>>> reflection deserialization.
>>>
>>> And then Jsonb before creating the Animal instance reads the field to
>>> extract its type which is likely not what is done since the type is deduced
>>> statically from the model class (field type) - this part is important to
>>> avoid surprises and open security issues.
>>>
>>> No need to create instance now, it is already fully populated. Just
>>> extract it from TypeWrapper as is.
>>>
>>>
>>> If I'm still wrong do you care writing a small test case showing that in
>>> action please?
>>>
>>> Sure I will implement a test and push it to our eclipselink git which
>>> hosts the jsonb-ri.
>>>
>>>
>>> 5. Jsonb parses a "ClassModel" for Dog or Cat class (reading all @Jsonb
>>>> annots etc if needed).
>>>> 6. Jsonb goes deeper in json document parsing json object for
>>>> "instance" property and populating all fields in Dog or Cat (doesn't really
>>>> care what is it as it is common reflection based process for all properties
>>>> inside).
>>>> 7. After above is completed an instance of TypeWrapper with fully
>>>> deserialized T inside is passed to an adapter method "T
>>>> adaptFromJson(TypeWrapper<T> obj)".
>>>> 8. Result from adapter is than set to a field, setter, collection, map,
>>>> array or whatever else is declared in domain model to unmarshall into.
>>>>
>>>> What returns an adapter is already ready to be set into model as is
>>>> with polymorphism working this way.
>>>>
>>>> Now I suppose this is exactly what:
>>>> [A]....*"Thought adapter should return a populated instance otherwise
>>>> no way to convert to business model properly IMO"* means ?
>>>>
>>>> Hope this helps and will bring some light.
>>>>
>>>> Thanks,
>>>> Roman
>>>>
>>>>
>>>>
>>>> On 03/23/2016 12:49 PM, Romain Manni-Bucau wrote:
>>>>
>>>>
>>>> 2016-03-23 12:29 GMT+01:00 Roman Grigoriadi <
>>>> <roman.grigoriadi_at_oracle.com>roman.grigoriadi_at_oracle.com>:
>>>>
>>>>> You load a Dog or Cat class by name, which appears in json and
>>>>> instantiate it with default constructor. This is what I suppose a custom
>>>>> deserializer for TypeWrapper would do.
>>>>>
>>>>>
>>>> [A] Think I get it: the adapter will return Foo and then on this
>>>> instance Jsonb will populate fields. Thought adapter should return a
>>>> populated instance otherwise no way to convert to business model properly
>>>> IMO (if you don't have an explicit model you maybe don't want a setter or
>>>> field to be used). That's why I was kind of blocked there.
>>>>
>>>>
>>>>> If you want to polymorphically adapt each animal in a collection you
>>>>> would have instead annotate Animal.class with @JsonbTypeAdapter, or pass an
>>>>> instance of AnimalAdapter to JsonbConfig.
>>>>>
>>>>> You are right, you can't add @JsonbTypeAdapter(AnimalAdapter.class) to
>>>>> List<Animal> as it would than have to be declared as: "extends
>>>>> TypeAdapter<List<Animal>>" and parent implementation would not work than.
>>>>>
>>>>>
>>>> Well adapter for items instead of the field in case of a collection is
>>>> something we can implement - we do in Johnzon - and which is useful IMO.
>>>> That said it doesn't solve the polymorphism issue. If [A] is solved it can
>>>> work but today we don't have any way to know if the adapter builds the
>>>> instance or just instantiate it if I got you right.
>>>>
>>>>
>>>>> Did I get your point?
>>>>>
>>>>>
>>>> Think so.
>>>>
>>>>
>>>>> Roman.
>>>>>
>>>>>
>>>>> On 03/23/2016 12:04 PM, Romain Manni-Bucau wrote:
>>>>>
>>>>> I don't get this part "Jsonb looks at animal field model and gets
>>>>> "toType" from annotated adapter".
>>>>>
>>>>> How do you create a Dog in your example instead of an Animal.
>>>>>
>>>>> Keep in mind you don't know all types from jsonb point of view. The
>>>>> real use case is more obvious on a collection - at least for me:
>>>>>
>>>>> public static class Pojo {
>>>>> @JsonbTypeAdapter(AnimalAdapter.class)
>>>>> public Collection<Animal> animals; // Dog, Cow, Cat, Horse...
>>>>> }
>>>>>
>>>>>
>>>>>
>>>>> Romain Manni-Bucau
>>>>> @rmannibucau
>>>>> <http://www.tomitribe.com>http://www.tomitribe.com
>>>>> <http://rmannibucau.wordpress.com>http://rmannibucau.wordpress.com
>>>>> <https://github.com/rmannibucau>https://github.com/rmannibucau
>>>>>
>>>>> 2016-03-23 11:51 GMT+01:00 Roman Grigoriadi <
>>>>> <roman.grigoriadi_at_oracle.com>roman.grigoriadi_at_oracle.com>:
>>>>>
>>>>>> You don't instantiate X in adapter you already have it. Here is
>>>>>> implementation of such adapter:
>>>>>>
>>>>>> class TypeAdapter<T> implements JsonbAdapter<T, TypeWrapper<T>> {
>>>>>>
>>>>>> @Override
>>>>>> public TypeWrapper<T> adaptToJson(T obj) throws Exception {
>>>>>> TypeWrapper<T> wrapper = new TypeWrapper<>();
>>>>>> wrapper.setClassName(obj.getClass().getName());
>>>>>> wrapper.setInstance(obj);
>>>>>> return wrapper;
>>>>>> }
>>>>>>
>>>>>> @Override
>>>>>> public T adaptFromJson(TypeWrapper<T> obj) throws Exception {
>>>>>> return obj.getInstance();
>>>>>> }
>>>>>> }
>>>>>>
>>>>>> class AnimalAdapter extends TypeAdapter<Animal> {}
>>>>>>
>>>>>> Than you have a pojo like this:
>>>>>>
>>>>>> public static class Pojo {
>>>>>> @JsonbTypeAdapter(AnimalAdapter.class)
>>>>>> public Animal animal;
>>>>>> }
>>>>>>
>>>>>> And json like this:
>>>>>>
>>>>>> {
>>>>>> "animal":{"className":"com.foo.Dog",name":"Doggie","dogProperty":"Property of a Dog."}}
>>>>>> }
>>>>>>
>>>>>>
>>>>>> What happens during deserialization in our implementation is:
>>>>>> JsonParser generates an event for started object at "animal" property.
>>>>>> Jsonb looks at animal field model and gets "toType" from annotated adapter.
>>>>>> Deserialization than happens for TypedWrapper as for common arbitrary user
>>>>>> Type, but with custom deserializer, reading class property first and
>>>>>> creating instance from it. After TypedWrapper<Animal> is fully
>>>>>> deserialized, it is passed to adapter instance, to retrieve an Animal,
>>>>>> which is finally set to Animal field.
>>>>>>
>>>>>> Regards,
>>>>>> Roman
>>>>>>
>>>>>>
>>>>>> On 03/23/2016 11:20 AM, Romain Manni-Bucau wrote:
>>>>>>
>>>>>> Hi Roman,
>>>>>>
>>>>>> can you fill some "..." please? typically at the moment "I know the
>>>>>> type is X". How with an adapter do you instantiate X and then let jsonb
>>>>>> populate it (without having another Jsonb instance)? That was the main
>>>>>> missing part of the puzzle.
>>>>>>
>>>>>>
>>>>>> Romain Manni-Bucau
>>>>>> @rmannibucau
>>>>>> <http://www.tomitribe.com>http://www.tomitribe.com
>>>>>> <http://rmannibucau.wordpress.com>http://rmannibucau.wordpress.com
>>>>>> <https://github.com/rmannibucau>https://github.com/rmannibucau
>>>>>>
>>>>>> 2016-03-23 11:14 GMT+01:00 Roman Grigoriadi <
>>>>>> <roman.grigoriadi_at_oracle.com>roman.grigoriadi_at_oracle.com>:
>>>>>>
>>>>>>> Hi,
>>>>>>>
>>>>>>> @Romain - what you and Dmitry suggested with adapters, will actully
>>>>>>> work in our case. We will have to to add a custom deserializer handler for
>>>>>>> a "TypedWrapper<T>" (similar to collection / array or reflection
>>>>>>> deserializers), which will read a class property first and on
>>>>>>> encountering a json object for T will create a correct type instance. The
>>>>>>> only condition would be, that class property must appear before inner
>>>>>>> instance in json object for "TypedWrapper<T>" as Eugen suggested:
>>>>>>>
>>>>>>> {"classProperty":"com.foo.Dog","instance":{...}}
>>>>>>>
>>>>>>> Adapters will than work in a common way to extract T from
>>>>>>> TypedWrapper and set it to a class model be it a field or a
>>>>>>> collection/array.
>>>>>>>
>>>>>>> I would also add required enumeration of allowed subtypes into
>>>>>>> TypedAdapter:
>>>>>>>
>>>>>>> class TypedAdapter<T> implements JsonbAdapter<T, TypedWrapper<T>> {
>>>>>>> TypedAdapter(Class[] allowedSubtypes) {...}
>>>>>>> ...
>>>>>>> }
>>>>>>>
>>>>>>> So we don't allow loading of any arbitrary class suggested by
>>>>>>> incoming json in case of not trusted third party.
>>>>>>>
>>>>>>> @Romain - adding support for custom serializers / deserializers in
>>>>>>> fashion of fasterxml with access to JsonGenerator/JsonParser would also
>>>>>>> solve the problem in similar fashion, but it would be more complicated for
>>>>>>> a client to implement logic driven by JsonParser events, than by providing
>>>>>>> predefined adapter.
>>>>>>>
>>>>>>> Thanks,
>>>>>>> Roman
>>>>>>>
>>>>>>> On 03/21/2016 09:36 PM, Romain Manni-Bucau wrote:
>>>>>>>
>>>>>>> Hi Dmitry,
>>>>>>>
>>>>>>> was my thought as well but it doesn't really work cause in adapter
>>>>>>> you can't relaunch the deserialization chain with the same jsonb instance.
>>>>>>> The serialization side is the easy one but the deserialization makes
>>>>>>> adapters API not enough to have a consistent behavior accross the
>>>>>>> deserialization. Making the adapter "JsonbAware" would work. We can
>>>>>>> probably use @Inject (CDI if there or just the JSR 330 as an optional
>>>>>>> dependency if not)
>>>>>>>
>>>>>>> Side note: I'd like we enable this impl but we don't provide them in
>>>>>>> the spec since it is likely dependent on your use case and saw cases where
>>>>>>> type was not enough and we will likely not handle all possible models.
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> Romain Manni-Bucau
>>>>>>> @rmannibucau
>>>>>>> <http://www.tomitribe.com>http://www.tomitribe.com
>>>>>>> <http://rmannibucau.wordpress.com>http://rmannibucau.wordpress.com
>>>>>>> <https://github.com/rmannibucau>https://github.com/rmannibucau
>>>>>>>
>>>>>>> 2016-03-21 21:27 GMT+01:00 Dmitry Kornilov <
>>>>>>> <dmitry.kornilov_at_oracle.com>dmitry.kornilov_at_oracle.com>:
>>>>>>>
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> This topic has been discussed already. Current version of the spec
>>>>>>>> is clear about polymorphism: *“**Deserialization into polymorphic
>>>>>>>> types is not supported by default mapping. [JSB-3.8-1]”. *The spec
>>>>>>>> is almost finished and I am strongly against going back and changing some
>>>>>>>> basic things now.
>>>>>>>>
>>>>>>>> On the other hand I understand that this use case exists. There is
>>>>>>>> a possible workaround to it using adapters. Sorry if there are any
>>>>>>>> mistakes, I am writing code without checking.
>>>>>>>>
>>>>>>>> We can create a generic wrapper to any type holding its type info
>>>>>>>> like this:
>>>>>>>>
>>>>>>>> class TypedWrapper<T> {
>>>>>>>> String type;
>>>>>>>> T obj;
>>>>>>>> }
>>>>>>>>
>>>>>>>> We can create a generic adapter which converts type to wrapper and
>>>>>>>> vice versa. I am not putting a full implementation here, I hope that the
>>>>>>>> idea is clear:
>>>>>>>>
>>>>>>>> class TypedAdapter implements JsonbAdapter<T, TypedWrapper<T>> {
>>>>>>>> TypedWrapper adaptToJson(T obj) {
>>>>>>>> TypedWrapper<T> t = new TypedWrapper<T>();
>>>>>>>> t.type = obj.getClass().getName();
>>>>>>>> t.obj = obj;
>>>>>>>> return t;
>>>>>>>> }
>>>>>>>>
>>>>>>>> T adaptFromJson(TypedWrapper<T> t) {
>>>>>>>> // Use reflection API to create an instance of object based
>>>>>>>> on t.type field value
>>>>>>>> }
>>>>>>>> }
>>>>>>>>
>>>>>>>> Now lets imagine that we have the following classes:
>>>>>>>>
>>>>>>>> class Animal {
>>>>>>>> }
>>>>>>>>
>>>>>>>> class Dog extends Animal {
>>>>>>>> }
>>>>>>>>
>>>>>>>> class Cat extends Animal {
>>>>>>>> }
>>>>>>>>
>>>>>>>> class Foo {
>>>>>>>> Animal animal;
>>>>>>>> }
>>>>>>>>
>>>>>>>> To properly handle serialization/deserialization of Foo we just
>>>>>>>> need to create an adapter which extends TypedAdapter from ab
>>>>>>>>
>>>>>>> ...
>
>
>