IntelliJ warning : "abstract value used in trait may cause errors during initialization" - what's the problem with LAZY abstract val-s?

Question

IntelliJ gives me the warning that "abstract value in trait may cause errors during initialization" - as shown in the picture below.

• What does this mean ?
• Should I take this warning seriously ?
• Should I use def-s instead of lazy val-s ?
• Are they not the same in this case ? Since we are talking here about dependency injection, which creates a static - possibly cyclical - object graph.
• But ! because the lazy modifier is there, it does not matter if the graph is cyclical or not. Right ?
• The values themselves will simply behave as they were a memoized function right ?
• Since we are talking about references here (in a static object graph) where references are directed edges and objects are nodes, what's the problem here ?
• Could you please give an example where using lazy val-s instead of def-s can cause problems during initialization ?

Thank you for reading :)

The whole codebase is on github.

This is the particular line that is problematic (on github).

I also copy paste the problematic code here:

trait EntityMarshallers[V<:EntityValue[V]]{
  implicit lazy val decoderEntityV:    Decoder[Entity[V]]
  implicit lazy val encoderEntityV:    Encoder[Entity[V]]
  implicit lazy val _encoderV:         Encoder[V]
  implicit lazy val classTag:          ClassTag[V]

}

This is the IntelliJ warning :