The symfony/serializer component has several normalizer implementations which are rather redundant and hard to maintain. The idea is to refactor the whole process of translating between array and objects and back and rely on PropertyAccess for the logic of how to interact with objects. The discussion is summarized at symfony/symfony#30818.
The EU FOSSA Hackathon helped us a lot to validate this discussion and begin the refactoring of the serializer. Here is a detailed list on what we have achieved in regards to the plan during these two days:
We refactored the tests so that we can ensure a consistent behaviour and test the new normalizer when it is implemented. In that process, we found a couple of bugs that we fixed in Symfony 3.4 where applicable, and 4.2 for newer features.
A new extension point was provided to be able to extract properties from an object (and not only for a class). This is needed in order to implement future behaviour that needs the data of the object to filter the property list.
A default implementation using the PropertyListExtractorInterface was provided.
This new extension point has been added under the @experimental flag in order to be safe about BC policy if we need to change something.
A new interface and default implementation was added in order to instantiate objects when denormalizing. This can provide a new extension point to allow creating object without constructors, or using Doctrine to instantiate an object like it is done in API Platform.
There are still some points to cover in order to be BC on the behaviour of the current normalizers.
A pull request was created some days / week ago in order to extract read and write information on a property for a class. Extra work was added to this pull request in order to use this new system into the PropertyAccess component.
A draft pull request was created in order to see the big picture about this new normalizer system.
Some implementations were done on the new extraction system, a draft PR was created where it currently supports:
- Allowing attributes (
attributescontext key in existing normalizer) - Ignoring attributes (
ignored_attributescontext key in existing normalizer)
Other features and fixes were done on the serializer, like adding the possibility to populate object on childs, and fixes about incorrect cache when using ignored attributes.
Thanks to all people that have worked on these topics, it is not finished yet and will be continued but we made some huge steps forward having a rock solid normalization flow.