[poketopa]

Summary

Fixes #1347.

Setting a container value whose elements are instances of different subtypes of the declared element type
(e.g. set("shapes", List.of(circle, square)) into List<Shape>) generated every element as the first
element's subtype. In strict mode the same call failed with
No matching results for given NodeResolvers. Invalid paths: [$.shapes[1].side].
This worked in 1.1.21 and regressed in 1.2.0.

(Optional): Description

When a container value is set, ValueAnalyzer builds two kinds of path resolvers from the value:

• a GenericTypeResolver for the container path ($.shapes), inferred from the value, and
• an InterfaceResolver per element path ($.shapes[0] → Circle, $.shapes[1] → Square).

GenericTypeResolverConverter.extractCollectionTypeArguments inferred the element type argument from the
first element only, so List<Shape> was rewritten to ArrayList<Circle> and every element node was
created as Circle before the per-element resolvers ran. The resolver for $.shapes[1] then saw a Circle
node, Circle.isAssignableFrom(Square) is false, and the Square value was projected onto a Circle node.
extractMapTypeArguments had the same problem with the first entry's key and value.

This change infers the element (and map key/value) type from all non-null elements:

• all elements share one class → that class, with its own type arguments (unchanged behaviour);
• elements have different classes → their most specific common superclass (Shape), so each element node
  keeps the declared type and the per-element InterfaceResolver picks its runtime type;
• the only common superclass is Object (e.g. implementations of an interface, or unrelated classes) →
  no type argument is inferred and the declared element type is kept.

How Has This Been Tested?

• ContainerElementSubtypeTest (java-tests): the two cases from Setting a list of mixed subtypes into an abstract-typed list generates every element as the first subtype (1.2.x regression) #1347 (default and strict mode), element
  order and element values preserved, mixed implementations of an interface, and mixed values in a
  Map<String, Shape>. All fail on main and pass with this change.
• GenericTypeResolverConverterTest (object-farm-api): same class, different subclasses, unrelated classes,
  null elements, map values.
• Regression command from CLAUDE.md:
  ./gradlew --continue clean :object-farm-api:check :fixture-monkey:test :fixture-monkey-tests:java-tests:test :fixture-monkey-tests:java-17-tests:test :fixture-monkey-tests:kotlin-tests:test (JDK 17)

Is the Document updated?

No. The behaviour matches what the documentation already describes for set with container values.

Note: #1337 also adds a GenericTypeResolverConverterTest (for a different method, createResolver).
The production changes do not overlap; if #1337 lands first I will rebase and merge the two test classes.

[CLAassistant]

All committers have signed the CLA.