Java - Self expansion of QueryParameters without reflection

[andreaTP]

I was reviewing this issue, and this proposal is aimed to fully remove all the usages of reflection in Java.

The QueryParameter type classes are currently using annotations, in combination with reflection up into abstractions (not easily swappable in user-land).
I do believe that a generated approach is preferable over the current implementation, it will also more closely resemble the design of the serialization/deserialization of the objects.

Code currently generated

class GetQueryParameters {
    /** Select properties to be returned */
    @QueryParameter(name = "%24select")
    @jakarta.annotation.Nullable public String[] select;

    /** Search items by search phrases */
    @QueryParameter(name = "%24search")
    @jakarta.annotation.Nullable public String search;
}

Proposal

class GetQueryParameters {
    /** Select properties to be returned */
    @jakarta.annotation.Nullable public String[] select;

    /** Search items by search phrases */
    @jakarta.annotation.Nullable public String search;

    public Map<String, Object> toQueryParameters() {
      return Map.of(
         "%24select", select,
         "%24search", search);
    }
}

the corresponding implementation of addQueryParameter simplifies to:

    public void addQueryParameters(@Nullable final Map<String, Object> parameters) {
        if (parameters == null) return;
        queryParameters.putAll(parameters);
    }

and the rest should be changed accordingly to make use of the new instance method.
After this change, we can safely delete the QueryParameter annotation, and the generated LOCs will be comparable with the current implementation.

I'm happy to hear feedback on this proposal and I can volunteer to implement it as I believe that removing reflection is one key aspect that will determine the success of Kiota in the Java ecosystem (e.g. being able to execute Kiota with zero configuration on native-image is the end goal).

[baywet]

Hi @andreaTP,
Thanks for the reminder on this.
I think this is promising, two things this proposal doesn't account for:

• We need to check for null values before adding them in the addQueryParameters implementation.
• The toQueryParameters method probably needs to have entries for query paramters where the URI Template parameter name matches the symbol name (I don't believe we emit annotations today when that's the case)

Other than that, if we don't anticipate a significant increase in size, we should proceed. And yes, please take it on if you can, this is time sensitive as this would be on the red-path for the Microsoft Graph Java SDK GA.

[andreaTP]

I don't think the increased size is going to be humongous, also, the advantage of completely removing reflection is really appealing!

I'll take it 👍