TypeManager vs. ObjectMapper

[ndr-brt]

I see around the codebase a lot of direct usages of the ObjectMapper class/instances (e.g. CrawlerJob)
For my understanding, for ser/des object from/to json TypeManager should be used.
So, I wanted to know if are there reasons to avoid it and using ObjectMapper otherwise if we just need to use TypeManager everywhere.

[jimmarino]

If the ObjectManager derives from the TypeManager, I think it is OK (it makes test fixtures easier as well). If a new ObjectManager() is used somewhere (other than a test), that should be changed to pull the OM from the TypeManager.

[ndr-brt]

In those cases the ObjectMapper is instantiate directly, so I think it'll better to refer to TypeManager instead.

[jimmarino]

If a class only needs ObjectMapper, I think it is fine to use that directly since it makes texting more succinct. Also, ObjectMapper contains more methods than TypeManager. We added some basic de/serialization to TypeManager as a convenience for not having to catch checked exceptions.

I would tend to leave this decision up to specific implementations.

[ndr-brt]

Why not use TypeManager as the only component for serdes, adding functionality on its interface when needed?
I think the coexistence of two ways to do serdes could confuse, and I think that's why there are some unexpected instantiations of ObjectMapper here and there.

[jimmarino]

I don't think TypeManager should reproduce the methods of ObjectMapper. If it causes confusion, I'd rather remove the convenience methods that wrap common ObjectMapper methods to throw unchecked exceptions.

My preference would be to not pass TypeManager to classes that do not specifically require one of its methods as doing so does not provide any value.

The use of new ObjectMapper() in classes other than tests should not be done, so it would be good to correct that in a small, targeted PR.