Ok, this sounds reasonable.Hmm, this causes some other design issues. We will have a SnakSerializer which has to take a $dataValueSerializer. Of course we should not specify a concrete type because of flexibility, but a problem of this solution is that wrong Serializers that cannot handle DataValues will only throw an exception at runtime but do not show a warning when "compiling". Do you have any suggestions how we can avoid those issues?
Good observation. This, a decrease in type safety, is a typical trade-off one makes when creating boundaries. This means the code constructing a SnakSerializer is responsible for making sure it is feeding in a Serializer implementation that can serialize data values. A small price to pay for the decoupling and flexibility gained.
Thanks for the example. I think this will be the next step after we have finished the serializers and deserializers.Also note that not a lot of code should be constructing these service objects, and know about the dependencies. In fact, this likely should only be done in one place, which is a factory provided by the library. An example of this can be seen here: https://github.com/wmde/AskSerialization/blob/master/src/Ask/DeserializerFactory.php
I see that if we write good tests all problems with types will be found at least when running the tests. This sounds very reasonable. Thank you for your explanations. :-)Another important thing to realize is that if one writes the right component and integration tests, any incorrect wiring up of the serializes will be found.