ReflectionCache not Serializable

[elgbar]

Ref issue #245, it it possible to make ReflectionCache Serializable?

It should be relatively easy, but I would suggest to only serialize it as empty.
Serializing actual contents may be complicated and unlikely to be worth the hassle.

Originally posted by @cowtowncoder in #245 (comment)

I do need to pass a configured instance of an ObjectMapper between applications.

If this is infeasible, any other suggestions for how to deal with configuring ObjectMapper cross application?

[cowtowncoder]

Yes I would think so. Someone would need to make the change, test serializability (databind has some tests, just initialize mapper with module, serialize + deserialize back, use for reading/writing [simple cases ok]), and provide a patch.

With 3.x things will change a bit to make things much easier so that modules do not usually need to do anything to keep them serializable (mapper serializes modules and NOT things modules register; problem with 2.x is that mappers are serialized with full state, 3.0 only serializes configuration).

[k163377]

@elgbar
Sorry for the long wait.

Is this still a necessary issue?
Personally, I thought it could be accomplished by making KotlinModule.Builder to Serializable.

[cowtowncoder]

I think the problem is that even if Builder is serializable, there are use cases where ObjectMappers are passed in a way that requires JDK serialization.

If there is need for ReflectionCache to be java.io.Serializable, it should almost certainly be done in a way that nothing is serialized (actual cache fields are marked as transient) and that on deserialization an empty ReflectionCache is constructed instead (in a state that works for mapper & Kotlin module).

A reasonable example of how to do that can be found from jackson-databind, class LRUMap (in package com.fasterxml.jackson.databind.util) -- aside from transient fields, addition of:

    protected Object readResolve() {
        return new LRUMap<K,V>(_initialEntries, _maxEntries);
    }

handles reconstruction (in this case retaining configuration settings; that might not be needed with ReflectionCache)

[elgbar]

The application I was wringing at the time is now EOL, so I don't personally need it now.

The solution I came up with was including a file in an Uber jar with the absolute classpath to an object instance which I used to do serialisation stuff with.

But as a feature I still see it's usage

[cowtowncoder]

I agree: an ObjectMapper with Kotlin module registered should be JDK serializable.

[k163377]

As far as I understood it, it looked like simply inheriting Serializable and setting serialVersionUID, so I created a PR anyway.
#634

[k163377]

It will be released in 2.15.

[cowtowncoder]

@k163377 One follow-up thing: could you please merge this from 2.15 into master? Due to differences in Jackson 2.x/3.0 APIs there are some conflicts and I am not knowledgeable enough to resolve those.

Some things are due to renaming:

• LRUMap becomes SimpleLookupCache
• JsonSerializer/JsonDeserializer -> ValueSerializer/ValueDeserializer

and then there is Java package name change from com.fasterxml.jackson.* into tools.jackson.*.

This merge should be done after merging every new feature in 2.15. Similarly, if a fix goes in an earlier branch (backport), like 2.14, it needs to be merged to all later branches (to 2.15, then master)

[k163377]

@cowtowncoder Created #636

[cowtowncoder]

Thank you @k163377 -- I merged it; I know there are pre-existing failures in master but I don't think this significantly changed the situation. And we don't want codebases to further diverge.