[evanchooly]

Plan item 4.4 (#19, A6): one generic-type resolver for both mappers.

Changes

• New TypeResolver (internal). For each class it binds every type variable its supertypes declare to the type the hierarchy gives it. The binding is by TypeVariable identity, not by name or position. It follows superclasses and interfaces, and the bindings are cached per class with a ClassValue. resolve(context, type) substitutes those bindings through ParameterizedType, GenericArrayType and wildcards. A generic array whose component resolves to a class becomes that array class, so T[] with T = String is String[], the same as a declared String[].
• TypeData.get uses it for the type arguments it reports, replacing the positional TypeParameters. TypeData.Builder.build() no longer wraps a list that the constructor copies anyway.
• The reflective mapper resolves every field and getter type against the entity: EntityModel.getTypeData(Type). This replaces the by-name parameterization map, propagateTypes and the 3-arg getTypeData, which only handled a bare T.
• Critter's PropertyModelGenerator calls the same resolver and drops its own, which kept only plain-class bindings. GenerationUtils.emitTypeData now emits WildCardTypeData rather than flattening wildcards.
• The game plan marks 4.4 done.

Fixes

Case	Before	After
class ReorderedMap<V,K> extends HashMap<K,V>, field ReorderedMap<String,Integer>	key String, value Integer: swapped	key Integer, value String
Base<T>{T value; List<T> values;}, Sub extends Base<List<String>>	reflection List<String> / List<Object>; critter Object for value	List<String> / List<List<String>> in both
Middle<T,M> extends Root<List<T>>, Leaf extends Middle<String,Integer>, Root's T value	reflection List<Object>, critter Object	List<String> in both
ArrayBase<T>{T[] array;}, StringArray extends ArrayBase<String>	Object (array)	String[]
List<? extends Number>, List<? super Integer> in critter	List<Number>, List<Object>	List<? extends Number>, List<? extends Object>, as in reflection

Tests

• New TestGenericResolution covers each row above, plus getter-based properties, round trips for the reordered map, the List<String> binding and a List<? super Integer> holding a string and an integer, and an unbound T[]. Each assertion runs against whichever mapper the build selects; under critter it also asserts that critter built the model (a CritterEntityModel) instead of falling back to reflection, except for the array entities critter doesn't support yet. The new resolution cases failed before this change with both mappers. Without the emitTypeData change, the two wildcard cases still fail under critter.
• TypeDataTest gains a binding through an interface (Holder<T>) and a subclass of its implementor.
• Generated models: I ran the critter AOT leg locally (generate-test-models, then the tests against the generated models) for these tests.
• Full suites: core, validation and Kotlin: 0 failures with -Dmorphia.mapper=critter and with reflection.

Decisions to review

1. Critter shares the resolver instead of getting a second one over classfile signatures, as the plan expected. PropertyModelGenerator already reads each member's generic type through reflection; only typeData(String), which nothing outside tests calls, parses signatures. One resolver keeps the mappers from drifting. Generated models still need no reflection at runtime, since the resolved TypeData is emitted as constants.
2. ? super X still decodes as its upper bound, Object, contrary to the plan, which listed it as a bug. A List<? super Integer> can hold a List<Object> with strings in it, and decoding it as Integer threw ClassCastException (caught in review; the mixed-list round trip now covers it). The plan entry records this.
3. Interface bindings are included. That covers a getter declared in a generic interface. They can't conflict with superclass bindings, because bindings are keyed by identity.
4. EntityModel.getTypeData(Class, TypeData, Type) is replaced by getTypeData(Type). The class is @MorphiaInternal and revapi excludes it.
5. Still open, not regressions:
   • Critter skips array-typed properties in both runtime and AOT mode, so the array cases only exercise reflection.
   • A type variable of an enclosing class (Outer<T>.Inner) still resolves to Object.
   • Mutually recursive generics (Foo extends Base<List<Foo>>) still recurse in TypeData.get, as before.

Benchmarks

CodecBenchmark, before (master at 1ef42e8) and after, both mappers, -prof gc. This step only changes how TypeData is built at mapping time, so the encode/decode path shouldn't move, and allocation doesn't: every B/op is within 1%. The first full run showed reflection 3–12% slower across the board, encode included, and critter 2–7% faster, the kind of uniform shift that means machine noise. Rerunning reflection over three alternating rounds puts every median within ±1.3% (the branch's outliers in round 1 coincide with test runs I had going).

reflection: full run

benchmark	model	before ns/op	after ns/op	Δ time	before B/op	after B/op
decode	-	25,241 ± 503	25,391 ± 580	+0.6%	47,433	47,769
decode	ARRAYS	26,438 ± 184	27,285 ± 438	+3.2%	15,569	15,569
decode	COLLECTIONS	8,038 ± 57	8,976 ± 232	+11.7%	16,704	16,712
decode	CONCRETE_COLLECTIONS	2,635 ± 45	2,672 ± 39	+1.4%	5,664	5,672
decode	LAZY_REFERENCE	2,566 ± 22	2,694 ± 70	+5.0%	6,072	6,072
decode	LIFECYCLE	682 ± 32	721 ± 51	+5.7%	1,704	1,704
decode	NESTED	2,752 ± 48	2,904 ± 95	+5.5%	5,656	5,648
decode	POLYMORPHIC	4,900 ± 93	5,184 ± 167	+5.8%	9,544	9,544
decode	SIMPLE	1,183 ± 236	1,086 ± 88	-8.2%	2,112	2,112
decodeInCursor	ARRAYS	26,835 ± 232	28,781 ± 347	+7.3%	15,257	15,257
decodeInCursor	COLLECTIONS	8,371 ± 70	8,816 ± 218	+5.3%	16,392	16,400
decodeInCursor	CONCRETE_COLLECTIONS	2,781 ± 40	2,855 ± 95	+2.7%	5,360	5,360
decodeInCursor	LAZY_REFERENCE	2,759 ± 27	2,870 ± 57	+4.0%	5,784	5,784
decodeInCursor	LIFECYCLE	696 ± 42	735 ± 37	+5.7%	1,392	1,392
decodeInCursor	NESTED	2,692 ± 42	3,220 ± 472	+19.6%	5,344	5,344
decodeInCursor	POLYMORPHIC	4,994 ± 130	5,549 ± 591	+11.1%	9,224	9,216
decodeInCursor	SIMPLE	1,025 ± 45	1,138 ± 85	+11.1%	1,800	1,808
encode	-	21,723 ± 250	22,253 ± 271	+2.4%	22,769	22,769
encode	ARRAYS	31,955 ± 1,106	33,271 ± 1,108	+4.1%	98,721	98,721
encode	COLLECTIONS	7,520 ± 89	8,250 ± 466	+9.7%	7,336	7,335
encode	CONCRETE_COLLECTIONS	2,078 ± 38	2,127 ± 23	+2.4%	3,125	3,104
encode	LAZY_REFERENCE	1,823 ± 33	1,812 ± 29	-0.6%	3,843	3,832
encode	LIFECYCLE	371 ± 4	384 ± 9	+3.6%	1,144	1,144
encode	NESTED	2,243 ± 28	2,346 ± 67	+4.6%	1,856	1,856
encode	POLYMORPHIC	4,051 ± 42	4,205 ± 90	+3.8%	4,200	4,200
encode	SIMPLE	751 ± 13	799 ± 12	+6.4%	1,216	1,216

critter: full run

benchmark	model	before ns/op	after ns/op	Δ time	before B/op	after B/op
decode	-	27,235 ± 354	26,722 ± 351	-1.9%	47,517	47,601
decode	ARRAYS	27,411 ± 244	26,500 ± 225	-3.3%	15,569	15,569
decode	COLLECTIONS	9,073 ± 183	8,893 ± 111	-2.0%	16,712	16,712
decode	CONCRETE_COLLECTIONS	2,789 ± 59	2,795 ± 95	+0.2%	5,664	5,672
decode	LAZY_REFERENCE	2,723 ± 72	2,627 ± 56	-3.5%	6,072	6,072
decode	LIFECYCLE	694 ± 13	665 ± 8	-4.2%	1,704	1,704
decode	NESTED	3,165 ± 64	3,064 ± 149	-3.2%	5,656	5,648
decode	POLYMORPHIC	5,680 ± 89	5,547 ± 229	-2.4%	9,552	9,552
decode	SIMPLE	1,090 ± 32	1,082 ± 53	-0.7%	2,112	2,112
decodeInCursor	ARRAYS	27,372 ± 323	27,052 ± 247	-1.2%	15,257	15,257
decodeInCursor	COLLECTIONS	9,045 ± 135	9,078 ± 191	+0.4%	16,400	16,400
decodeInCursor	CONCRETE_COLLECTIONS	2,804 ± 52	2,839 ± 99	+1.2%	5,360	5,360
decodeInCursor	LAZY_REFERENCE	2,788 ± 72	2,732 ± 41	-2.0%	5,784	5,784
decodeInCursor	LIFECYCLE	705 ± 31	690 ± 30	-2.0%	1,392	1,392
decodeInCursor	NESTED	3,256 ± 76	3,127 ± 30	-4.0%	5,344	5,352
decodeInCursor	POLYMORPHIC	5,639 ± 91	5,572 ± 125	-1.2%	9,240	9,240
decodeInCursor	SIMPLE	1,067 ± 33	1,081 ± 60	+1.4%	1,800	1,800
encode	-	22,591 ± 613	21,875 ± 290	-3.2%	22,769	22,769
encode	ARRAYS	32,933 ± 256	31,660 ± 298	-3.9%	98,721	98,721
encode	COLLECTIONS	7,798 ± 82	7,631 ± 85	-2.1%	7,336	7,336
encode	CONCRETE_COLLECTIONS	2,097 ± 27	2,047 ± 16	-2.4%	3,104	3,104
encode	LAZY_REFERENCE	1,717 ± 27	1,663 ± 20	-3.2%	3,832	3,832
encode	LIFECYCLE	378 ± 17	352 ± 7	-6.9%	1,144	1,144
encode	NESTED	2,381 ± 19	2,283 ± 13	-4.1%	1,856	1,856
encode	POLYMORPHIC	4,194 ± 56	4,125 ± 41	-1.7%	4,200	4,200
encode	SIMPLE	760 ± 56	727 ± 4	-4.4%	1,216	1,216

Reflection rerun, 3 alternating rounds (ns/op per round, Δ of medians):

benchmark	model	before	after	Δ
decode	-	24,257, 24,485, 24,355	24,581, 24,756, 24,596	+1.0%
decode	COLLECTIONS	8,282, 8,433, 8,138	11,812, 8,392, 8,198	+1.3%
decode	NESTED	2,677, 2,738, 2,693	4,384, 2,717, 2,618	+0.9%
decode	POLYMORPHIC	4,904, 4,945, 4,853	6,331, 4,967, 4,954	+1.3%
decode	SIMPLE	1,081, 1,009, 1,028	1,495, 1,037, 1,011	+1.0%
encode	-	21,472, 21,421, 21,443	21,687, 21,400, 21,431	-0.1%
encode	COLLECTIONS	7,473, 7,646, 7,506	7,726, 7,359, 7,450	-0.7%
encode	NESTED	2,212, 2,196, 2,220	2,283, 2,234, 2,237	+1.1%
encode	POLYMORPHIC	4,044, 3,989, 3,996	4,025, 4,047, 4,000	+0.7%
encode	SIMPLE	750, 756, 745	1,020, 759, 752	+1.3%

[Copilot]

🟡 Changes recommended

Lower-bound codec selection breaks valid mixed-value contravariant collections, while the Critter tests can mask generation failures by accepting reflective fallback.

2 open issues

• Preserve dynamic codecs for contravariant collection elements · New
• Assert generated models in non-array mapper tests · New

What changed in this PR

This PR centralizes generic-type resolution for Morphia’s reflective and Critter mappers, addressing plan item 4.4 and reducing differences in their type metadata.

Changes:

• Adds a cached resolver for inherited type variables, nested generics, arrays, and wildcards.
• Integrates shared resolution into both mappers and preserves wildcard metadata in generated models.
• Adds regression tests for generic bindings and round trips, and marks the plan item complete.

File	Description
design/​mapping-game-plan.md	Marks item 4.4 complete and explains the shared-resolver approach.
core/​src/​test/​java/​dev/​morphia/​test/​mapping/​TestGenericResolution.java	Adds generic-resolution and round-trip coverage.
core/​src/​test/​java/​dev/​morphia/​test/​mapping/​codec/​pojo/​TypeDataTest.java	Tests inherited interface bindings.
core/​src/​main/​java/​dev/​morphia/​mapping/​conventions/​MethodDiscovery.java	Resolves getter types against the entity.
core/​src/​main/​java/​dev/​morphia/​mapping/​conventions/​FieldDiscovery.java	Resolves field types against the entity.
core/​src/​main/​java/​dev/​morphia/​mapping/​codec/​pojo/​WildCardTypeData.java	Exposes wildcard construction and bound direction.
core/​src/​main/​java/​dev/​morphia/​mapping/​codec/​pojo/​TypeResolver.java	Implements cached hierarchy bindings and type substitution.
core/​src/​main/​java/​dev/​morphia/​mapping/​codec/​pojo/​TypeData.java	Uses shared argument resolution and preserves lower bounds.
core/​src/​main/​java/​dev/​morphia/​mapping/​codec/​pojo/​EntityModel.java	Replaces name-based resolution with the shared resolver.
core/​src/​main/​java/​dev/​morphia/​critter/​parser/​generator/​PropertyModelGenerator.java	Replaces Critter’s separate resolver.
core/​src/​main/​java/​dev/​morphia/​critter/​parser/​generator/​GenerationUtils.java	Emits wildcard metadata in generated models.

_{Review effort: Balanced}

[Copilot]

🟢 Approval recommended

No active review issues remain to block approval of the shared generic-type resolution change.

No open issues 🎉

2 resolved since last review

• Assert generated models in non-array mapper tests
• Preserve dynamic codecs for contravariant collection elements

_{Review effort: Balanced}