AER-4590 Avoid redundant work in JSONObjectHandle

[JornC]

This contains some proven non-controversial optimizations in the JSON parser framework.

---

LLM-generated technical summary

Two changes in JSONObjectHandle, both provably equivalent.

has() enumerated the object twice per call. It was:

return getInner() != null && keySet() != null && keySet().contains(key);

keySet() delegates to GWT's JSONObject.keySet(), which on every call runs a full for-in with hasOwnProperty, allocating a fresh String[] and an anonymous AbstractSet. Nothing is cached — size() even carries a comment about having to recheck for foreign changes.

Crucially, the returned set overrides contains to delegate straight back to containsKey, so the key array it just built is never read:

public boolean contains(Object o) {
  return (o instanceof String) && containsKey((String) o);
}

So both enumerations were pure waste, and inner.containsKey(key) reaches the identical native key in jsObject. Equivalence holds by construction rather than by argument, including for keys whose value is JSON null and for prototype-chain names such as toString (which has() already reported as present — that pre-existing quirk is preserved).

keySet() can never return null, so that guard was dead. A key != null check was added to keep has(null) returning false: without it, null in jsObject coerces to the property name "null".

In one AERIUS Calculator boot this is called ~15,400 times, so ~30,800 object enumerations disappear.

getStringOrDefault allocated a wrapper around its own state. It called new JSONObjectHandle(inner).getString(key). inner is the class's only field and the constructor merely assigns it, so the new instance was identical to this; getString(key) is the same call without the allocation.

Verified against the GWT sources shipped in org.gwtproject:gwt-user:2.13.1; JSONObject.java is byte-identical across 2.10.0, 2.12.1, 2.13.0 and 2.13.1, so this does not depend on the GWT version in use.

Two things a reviewer should know. The module contains no test files, so mvn test passes having run nothing — compilation is the only automated evidence here, and real validation is a client boot. And the runtime saving has not been measured in a browser; the change is justified by strictly removing work, not by a benchmark.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[BertScholten]

LGTM