[evanchooly]

Plan item 3.6 (review B11/A7).

The problem

For every property of every encoded entity, MorphiaPropertySerialization.shouldSerialize:

• read storeNulls, ignoreFinals and storeEmpties from the config (Boolean getters with defaults);
• then scanned the property's annotations, calling annotationType() on each, which is a JDK proxy call for reflection-mapped annotations, just to find @LoadOnly.

All of that is fixed once the property is mapped. The encoder also visited @LoadOnly properties that it would never write, and wrapped every document in a capturing lambda.

Change

• Precomputed flags. MorphiaPropertySerialization works out storeNulls, storeEmpties and neverSerialized (@LoadOnly, or a final field under ignoreFinals) in its constructor. MorphiaConfig is immutable and the annotations are fixed by then, so the results are identical. shouldSerialize is now a few boolean checks plus the empty-collection test.
• Encode order. CodecPlan.encodeOrder (from 3.0) also leaves out @LoadOnly properties. It already used == for the id, which the plan item also asked for.
• No lambda. EntityEncoder.encode writes the document's start and end directly instead of passing a capturing lambda to CodecHelper.document.

Tests

• TestCodecPlanEncodeOrder runs under both mappers on HarnessRenames. The encode order is the model's properties minus the id and the @LoadOnly legacy property, in model order.
• The harness's four fixture variants already pin shouldSerialize behavior under every storeNulls/storeEmpties combination, plus @LoadOnly, and are unchanged.
• The core and Kotlin suites pass with -Dmorphia.mapper=critter and with reflection: 1494 + 5 tests, 0 failures.

JMH

Expect a small gain. The review estimated "a few ns per property". This change is encode-only, and decode code is untouched.

Full CodecBenchmark, master (7bcb775f4b0) vs this branch

Same machine and settings as the earlier items, before and after back to back.

reflection

benchmark	model	before ns/op	after ns/op	Δ time	before B/op	after B/op	Δ alloc
decode	ARRAYS	71,154 ± 1,650	70,570 ± 1,201	-0.8%	74,858	74,858	-0.0%
decode	COLLECTIONS	8,720 ± 142	9,062 ± 356	+3.9%	16,704	16,696	-0.0%
decode	CONCRETE_COLLECTIONS	2,956 ± 60	3,064 ± 54	+3.7%	6,136	6,128	-0.1%
decode	LAZY_REFERENCE	3,293 ± 118	3,444 ± 94	+4.6%	7,315	7,293	-0.3%
decode	LIFECYCLE	1,111 ± 21	1,083 ± 14	-2.5%	2,840	2,840	-0.0%
decode	NESTED	3,018 ± 100	2,987 ± 50	-1.0%	5,656	5,672	+0.3%
decode	POLYMORPHIC	5,297 ± 99	5,406 ± 101	+2.1%	9,536	9,544	+0.1%
decode	SIMPLE	1,194 ± 63	1,167 ± 86	-2.2%	2,128	2,136	+0.4%
decodeInCursor	ARRAYS	73,808 ± 1,382	70,029 ± 1,333	-5.1%	74,546	74,546	-0.0%
decodeInCursor	COLLECTIONS	9,266 ± 510	9,049 ± 102	-2.3%	16,384	16,376	-0.0%
decodeInCursor	CONCRETE_COLLECTIONS	3,063 ± 131	3,020 ± 72	-1.4%	5,840	5,840	+0.0%
decodeInCursor	LAZY_REFERENCE	3,479 ± 156	3,386 ± 71	-2.7%	7,027	7,027	+0.0%
decodeInCursor	LIFECYCLE	1,242 ± 38	1,189 ± 23	-4.3%	2,528	2,528	-0.0%
decodeInCursor	NESTED	2,959 ± 105	3,081 ± 86	+4.1%	5,344	5,344	+0.0%
decodeInCursor	POLYMORPHIC	5,849 ± 167	5,408 ± 99	-7.5%	9,216	9,216	-0.0%
decodeInCursor	SIMPLE	1,171 ± 81	1,221 ± 31	+4.3%	1,824	1,832	+0.4%
encode	ARRAYS	109,617 ± 13,020	105,254 ± 3,797	-4.0%	135,598	135,587	-0.0%
encode	COLLECTIONS	7,838 ± 107	7,775 ± 87	-0.8%	7,336	7,336	+0.0%
encode	CONCRETE_COLLECTIONS	2,200 ± 118	2,158 ± 35	-1.9%	3,115	3,104	-0.3%
encode	LAZY_REFERENCE	1,909 ± 46	1,863 ± 31	-2.4%	3,843	3,832	-0.3%
encode	LIFECYCLE	850 ± 15	862 ± 19	+1.4%	3,064	3,064	+0.0%
encode	NESTED	2,349 ± 34	2,249 ± 29	-4.3%	1,856	1,856	-0.0%
encode	POLYMORPHIC	4,193 ± 65	4,225 ± 124	+0.8%	4,584	4,200	-8.4%
encode	SIMPLE	818 ± 16	787 ± 8	-3.9%	1,216	1,216	-0.0%

critter

benchmark	model	before ns/op	after ns/op	Δ time	before B/op	after B/op	Δ alloc
decode	ARRAYS	70,461 ± 1,642	71,336 ± 1,375	+1.2%	74,858	74,858	+0.0%
decode	COLLECTIONS	8,953 ± 196	9,215 ± 343	+2.9%	16,696	16,688	-0.0%
decode	CONCRETE_COLLECTIONS	2,997 ± 87	3,065 ± 20	+2.3%	6,107	6,128	+0.3%
decode	LAZY_REFERENCE	3,222 ± 97	3,289 ± 35	+2.1%	7,293	7,256	-0.5%
decode	LIFECYCLE	1,134 ± 43	1,148 ± 50	+1.2%	2,883	2,840	-1.5%
decode	NESTED	3,179 ± 30	3,182 ± 32	+0.1%	5,656	5,656	-0.0%
decode	POLYMORPHIC	5,697 ± 97	6,129 ± 329	+7.6%	9,528	9,528	+0.0%
decode	SIMPLE	1,138 ± 22	1,165 ± 45	+2.3%	2,128	2,136	+0.4%
decodeInCursor	ARRAYS	69,197 ± 1,206	72,537 ± 1,606	+4.8%	74,546	74,546	-0.0%
decodeInCursor	COLLECTIONS	9,152 ± 167	9,258 ± 195	+1.2%	16,384	16,384	+0.0%
decodeInCursor	CONCRETE_COLLECTIONS	3,080 ± 69	3,232 ± 54	+4.9%	5,840	5,840	+0.0%
decodeInCursor	LAZY_REFERENCE	3,248 ± 43	3,505 ± 129	+7.9%	6,971	6,963	-0.1%
decodeInCursor	LIFECYCLE	1,208 ± 21	1,201 ± 72	-0.6%	2,528	2,528	+0.0%
decodeInCursor	NESTED	3,221 ± 79	3,169 ± 54	-1.6%	5,344	5,344	-0.0%
decodeInCursor	POLYMORPHIC	5,796 ± 109	5,823 ± 108	+0.5%	9,216	9,216	+0.0%
decodeInCursor	SIMPLE	1,177 ± 29	1,226 ± 31	+4.2%	1,824	1,824	+0.0%
encode	ARRAYS	98,286 ± 2,463	102,225 ± 2,193	+4.0%	139,886	135,587	-3.1%
encode	COLLECTIONS	7,894 ± 80	8,008 ± 179	+1.4%	7,336	7,336	+0.0%
encode	CONCRETE_COLLECTIONS	2,057 ± 44	2,128 ± 19	+3.4%	3,104	3,125	+0.7%
encode	LAZY_REFERENCE	1,740 ± 36	1,770 ± 39	+1.7%	3,832	3,832	+0.0%
encode	LIFECYCLE	813 ± 29	867 ± 27	+6.7%	3,075	3,064	-0.3%
encode	NESTED	2,437 ± 24	2,415 ± 25	-0.9%	1,856	1,856	+0.0%
encode	POLYMORPHIC	4,235 ± 64	4,449 ± 194	+5.1%	4,200	4,200	-0.0%
encode	SIMPLE	782 ± 13	775 ± 9	-1.0%	1,216	1,216	-0.0%

This run was noisy: critter decode cells moved by up to ±8% even though decode code is unchanged, which sets the noise floor for this session. The encode deltas the review predicted are smaller than that, so I reran encode on its own.

encode only, master and this branch alternating, three rounds each

variant	model	master ns/op (3 runs)	3.6 ns/op (3 runs)	Δ median
reflection	SIMPLE	797, 839, 869	802, 824, 948	-1.7%
reflection	NESTED	2,354, 2,470, 3,668	2,337, 2,378, 2,536	-3.7%
reflection	POLYMORPHIC	4,191, 4,337, 4,350	4,195, 4,203, 4,809	-3.1%
reflection	LIFECYCLE	864, 901, 1,073	866, 876, 888	-2.8%
critter	SIMPLE	796, 798, 799	784, 870, 1,136	+9.1%
critter	NESTED	2,419, 2,492, 3,007	2,385, 2,434, 2,728	-2.3%
critter	POLYMORPHIC	4,386, 4,579, 4,642	4,217, 4,302, 4,447	-6.0%
critter	LIFECYCLE	859, 886, 1,065	845, 856, 880	-3.3%

Reading it:

• This branch is faster in 7 of 8 cells by median, by 2–6%. That's consistent with "a few ns per property".
• The exception is critter SIMPLE. Its median is pulled by a single 1,136 ns outlier, and its fastest run (784 ns) still beats master's (796 ns).
• Single-run outliers of up to +50% (e.g. 3,668 ns) show how noisy the machine was. Treat the gain as small and consistent in direction, not precisely sized.
• Allocation is unchanged, apart from run-to-run noise. The flags were never allocated, so this change saves work, not memory.

Decisions to review

1. Only the id and @LoadOnly properties leave the encode order, not every always-skipped property as the plan said. MergingEncoder, used by merge(), relies on visiting skipped properties. A final field skipped by ignoreFinals produces an $unset there. So those properties have to stay in the order. @LoadOnly properties are the only ones that neither encoder writes or unsets.
2. No benchmark-only proof of the gain. It's small and the machine was noisy. I've reported the medians and the raw runs rather than a single number.

[Copilot]

🟢 Approval recommended

The focused changes preserve serialization and merge semantics, with no unresolved blocking issues identified.

No open issues 🎉

What changed in this PR

This PR reduces per-property encoding work in Morphia by precomputing serialization decisions and avoiding unnecessary encoder operations, while preserving merge behavior.

Changes:

• Cache null, empty-value, and never-serialized flags when properties are mapped.
• Exclude @LoadOnly properties from encode order and write document boundaries directly without a capturing lambda.
• Add encode-order tests for both mappers and mark plan item 3.6 complete.

File	Description
design/​mapping-game-plan.md	Marks serialization optimization item 3.6 complete.
core/​src/​test/​java/​dev/​morphia/​mapping/​codec/​pojo/​TestCodecPlanEncodeOrder.java	Tests property order and id/@LoadOnly exclusion under both mappers.
core/​src/​main/​java/​dev/​morphia/​mapping/​codec/​pojo/​EntityEncoder.java	Replaces the document-wrapper lambda with direct writer calls.
core/​src/​main/​java/​dev/​morphia/​mapping/​codec/​pojo/​CodecPlan.java	Removes @LoadOnly properties from encode order.
core/​src/​main/​java/​dev/​morphia/​mapping/​codec/​MorphiaPropertySerialization.java	Precomputes flags used by serialization checks.

_{Review effort: Balanced}