team neusta: Converter and Populator Pattern within Pimcore
It is essential to understand the intention of a pattern to avoid misuse and confusion. The Converter and Populator pattern should be used whenever you have to transform objects of a specific type (system) into objects of a different type (system). Let me go on speaking about source and target types. The Converter and Populator pattern is an architectural framework that guides you through the transformation (aka mapping) of objects from one universe to another. Typically we use it to create View objects or Elasticsearch objects based upon our base system (e.g., Pimcore, Hybris)
As you can see in this sequence diagram, the conversion of objects will be done by the main component of our pattern called Converter.
But the Converter is more or less only the encapsulation of the steps necessary to "create a target object based upon the source object."
- Look up if already converted in the Converter cache.
- Potentially create a new target object (e.g., done here by ProductViewFactory)
- Fit the target object with new attribute values based upon the source object (e.g., done by several different ProductViewPopulators)
- Write back in the Cache.
Business entity type systems are mostly created to model the real world into your digital system. Relations between entities very often will be implemented by foreign keys or other technical instructions. A common example of the difference between data that should be stored in the system and data shown on the screen is, for example, a person's age. It makes no sense to save a person's age; otherwise, you have to think about a daily cron job that has to recalculate the age-based upon the current date. But in your view, you may only want to show the age of a person and not the birthdate. So transformation is necessary. This should give you just an idea of why two different type systems based upon the same entities could arise. Very often, I've seen implementations (sometimes even inside the Controllers) done by a simple (private) method inside the view action. I'll promise you this will grow and grow, and after a while, you can not handle changes without producing bugs or implement complex acceptance tests checking that your "internal" transformation works. So it's much better to accept the separation of the two type systems from the very beginning and start implementing this transformation by the Converter and Populator pattern.
The main reasons are the following:
- All transformation steps will be implemented in testable small units called populators.
- The already existing transformation will not be changed or touched by further transformations (Open-Closed-principle).
We are using the Converter and Populator pattern in all situations where we come in touch with different type systems:
View Type system
As I already mentioned above in the small birthday-age-example, I would recommend separating the view to our entities always from the business type system. Frontend developers will love you if you give them the chance to implement their frontends based upon their "own" type system (e.g., created by 90% string attributes, 5% integer values, and 5% rest).
Elastic Search Engine
The Elastic based description of our objects differs in very many points from our base implementation of business entities.
Imports from an external system (e.g., ERP systems like SAP)
With the fine-grain Populator, the mechanism is much easier to implement different interpretations of values from the external system.
Target Type Converter
The Target Type Converter provides the implementation of the convert method. If your programming language allows it, it is possible to implement a converter's single generic implementation, which can be staffed with different deployments of factories and populators based upon target and source type.
The Target Type is the type of objects which will be the results of the conversion. For example, it could be the ProductView class or a ProductElasticSearchDTO.
The Source Type is the type of objects which will be converted. For example, it could be the domain entity type Product.
Target Type Factory
The Target Type Factory creates an "empty" instance of the target object, which the given source object should fit. Very often, the implementation is just a simple new operator call.
Target Type Populator
The Target Type Populator is one component that provides a generic implementation of a populate method that fits at least one detailed attribute of our target type based upon the source typed object. This could be a simple combination of a target - type - setter - source type - getter call. But of course, in a more complex implementation, Populators use other Converters for Converter aggregated objects or use different services to get additional information necessary for the population. A little time ago, I always compared Populators to assembly-line workers doing only one step of the complete construction, but a colleague of mine pointed out that the difference between workers and populators is that the last group should work independently to each other and potentially in parallel.
But the basic idea is the same: each Populator is doing one step of the whole conversion.
The Target Type Converter will be fitted with the associated Target Type Factory (mostly one single instance) and the Target Type Populators (mostly several instances).
The TargetType Populators can work with a well-defined target object and do what they have to do:
- Call setters of the target object parameterized with getter calls from the source object
- Call service methods to determine particular calculated values and set the associated attribute(s) from the target object.
- Call itself a target type converter for an aggregated partial object inside the target object.
A positive consequence of this pattern is that you can reuse certain populations that come up several times. A typical example is aggregated partial objects, which could become a Converter-Populator construction itself. Some developers were complaining about this pattern's complexity, but most of the artifacts you need are generic, and you can reuse the code again and again. On the other side, 90% of the code needed in this pattern could be generated, so the Interface and method signatures are clear and well-defined. Last but not least, only the population itself differs depending on your transformation logic.
We use this pattern to transform Pimcore Data Objects into View objects for the Frontend (Twig) Developer. Although there are many useful Twig functions one can use to show values of the Pimcore object. It is much easier and more comfortable to define together with both sides of the development - Backend and Frontend - which values have to be shown and implement particular Data Transfer objects or View classes and covert and populate them based upon the Pimcore DataObjects.
Assume we have a facade which our Controller will call to get a view object based upon a Pimcore DataObject of type Firma, which is the german word for enterprises used by our customer (ubiquitous language):
Want to submit a guest post to Pimcore's Blog?
Submit a guest post and benefit from our network! With our newsletter, we reach more than 4 000 subscribers and attract more than 50 000 monthly visitors to our website, and we're always looking for more brilliant contributors to join our ranks. Contact us.