Case Classes with Option[Long] parameters don't deserialize properly

[kscaldef]

It seems that for small numbers, an Option[Long] in a case class actually gets deserialized with a runtime type of int. This can then manifest in a couple ways. Maps using these values as keys may not behave as expected, and conversions to Java Longs will throw a ClassCastException:

[info]   java.lang.ClassCastException: java.lang.Integer cannot be cast to java.lang.Long
[info]   at scala.runtime.BoxesRunTime.unboxToLong(Unknown Source)

This only seems to affect case classes. Option[Long] by itself seems to behave correctly.

[christophercurrie]

This uncovers one of the dirtier parts of Scala. The JVM doesn't support using an intrinsic like long as a type parameter, and Long is just long in the Scala type system. At one point I think that they mapped it to the boxed types in the type parameter but that ended up being a source of bugs, so they changed to Object.

For 2.2, I've added patches to databind to support using @JsonDeserialize as a workaround:

class OptionLongHolder
{
  @JsonDeserialize(contentAs=classOf[java.lang.Long])
  var optLong: Option[Long]
}

This works, as it forces Jackson to use Long for deserialization rather than the narrowest natural integer.

[magro]

We're currently tied to jackson 2.1.x. Is it somehow possible to get this backported?

[christophercurrie]

I'll look into it. AFAIK, core Jackson doesn't plan on any further maintenance on 2.1.x, but I'll see what I can do about an interim release.

[magro]

Ok, thanx!

[PanzerKunst]

Hello,

Just wanted to thank you for the fix in 2.2. Got a bit stressed when I noticed that conversion bug, so I was relieved to see that it's solved using the @JsonDeserialize(contentAs=classOf[java.lang.Long]) annotation.

[eugeniyk]

What is the workaround if your case class is generated for you? I don't have an access to apply attribute.

[devshorts]

@christophercurrie is it possible to create an optional module that uses what finatra has to work around this? It looks like they use the scala signature parsing https://github.com/twitter/finatra/tree/develop/jackson/src/main/scala/com/twitter/finatra/json/internal/caseclass/jackson

Having to annotate works fine, but its problematic when you are trying to distribute a model class (for example) which now pulls in all of jackson databind. Had the annotation been in jackson-annotations it would be less problematic, but leaking a dependency like that really sucks.

I'd be really open to having a separate module (if we don't want to include that in the main one) that provided for option[long] parsing via signature parsing in the interim, until the deprecation comes around for 2.10>