policy on deserializeImpl JSONValue

[gulugulubing]

When revising the JSONValue's deserializeImpl function, it looks like

item.put(newElem);

in onArrayElement() can't handle JSONValue[]. Now the codes there are suggested by AI.

[schveiguy]

For sure we need to figure out why put doesn't work. I don't want to have special cases here. Can you post the errors that occur?

[gulugulubing]

For sure we need to figure out why put doesn't work. I don't want to have special cases here. Can you post the errors that occur?

My fault. In deserializing JSONValue array, I should call deserializeImpl:

        case ArrayStart:
            item.type = JSONType.Array;
            item.array = null;
            deserializeImpl(policy, tokenizer, item.array);
            break;

other than call deserializeArray directly. Thus item.put(newElem) works. I fixed it in the new commit.

[schveiguy]

In deserializing JSONValue array, I should call deserializeImpl

Ahh!!! ok, because deserializeImpl is the thing that decides to use Appender for deserializeArray.

OK, so let's standardize on calling deserializeImpl for all types. Even object.

And make sure to call it via policy.deserializeImpl(...) to give the policy a chance to intercept.

[schveiguy]

Nice work!

[gulugulubing]

I am thinking about adding maxDepth check into defaultDeserializePolicy. The problem is how to store the change of the depth? If we assign a field to store the depth in defaultDeserializePolicy then check and modify it in onXXXBegin() and onXXXEnd(), it looks like all the customized policies will also need to define the depth field. We couldn't provide a shim for them like relPol because relPol is constant while depth is changing.

So it looks like it's not worthy to add maxDepth as official option into defaultDeserializePolicy. Users who want it could implement their own onXXXBegin() and onXXXEnd() with maxDepth check.

[schveiguy]

If we assign a field to store the depth in defaultDeserializePolicy then check and modify it in onXXXBegin() and onXXXEnd(), it looks like all the customized policies will also need to define the depth field.

You would add a shim in DefaultDeserializationPolicy just like onField, which calls the module. Notice how relPol is only used as a shim to deal with legacy code. This is a new concept, so we don't need that legacy.

Something like

auto onObjectBegin(JT, T)(ref JT tokenizer, ref T item) {
   if(++depth > maxDepth)
      throw new ex(...);
   return .onObjectBegin(this, tokenizer, item);
}

void onObjectEnd(JT, T, C)(ref JT tokenizer, ref T item, ref C context) {
   --depth;
   .onObjectEnd(this, tokenizer, item, context);
}

take a look at how the onField hook works to implement the release after a member.

[gulugulubing]

You would add a shim in DefaultDeserializationPolicy just like onField, which calls the module. Notice how relPol is only used as a shim to deal with legacy code. This is a new concept, so we don't need that legacy.

Thank you, I got it. I have done it in #62