Optimize protobuf encoding and add CI performance reports

[Fyzu]

Protobuf WKT encoding performs avoidable reflective reads, and numeric map keys allocate temporary strings. This PR uses concrete WKT getters and direct quoted numeric-key writes, fixes codec edge cases, and adds repeatable CI comparisons so future PRs and default-branch pushes report their performance impact.

Encoding changes

• Use concrete Timestamp/Duration/Struct/Value/ListValue getters with DynamicMessage fallback, one exact-size uint64 buffer, cached descriptors/WKT classification, and presence checks before reflective reads.
• Fix deprecated fields omitted by typed encoding, unsigned map keys printed as signed values, and custom JSON names producing invalid JSON or uncompilable generated codecs.
• Preserve numeric/boolean map-key spelling under writer settings; add regression fixtures and a 30-case codegen/typed/reflection × UTF-16/UTF-8 JMH matrix.

The full same-compiler CI comparison on Java 21/25 measured Struct throughput +95–116%, codegen/typed maps +12–29%, and reflection Timestamp +25–29%. Struct allocations fell about 44%; map allocations fell 26–39% across all paths. Of 60 comparisons, 31 show improvement signals, 29 are inconclusive, and none show regression signals or allocation growth alerts. These are per-workload advisory signals, not an overall application speedup. Full CI results and raw artifacts supersede the earlier mixed-compiler follow-up; the historical local reports and compact CSV evidence remain in docs/performance-port-2026-09-05.md and docs/performance-results/.

Performance CI

• Compare exact PR base/head SHAs and pre-/post-push SHAs on main, using stock JMH on Java 21 and 25.
• Clean-build both revisions with the same compiler, shared candidate benchmark inputs, and isolated Maven repositories; measure sequentially on the same runner, alternating variant order.
• Retain raw samples, environment, source identities, build logs and Markdown reports for 90 days. Job summaries are available on every run.
• A separate trusted workflow_run publisher validates JSON artifacts, adds commit checks, and updates one PR comment while rejecting stale results. It never executes PR/artifact code with write credentials.
• Throughput alerts require ≥10% change and separated JMH intervals and remain advisory. The 14 allocation budgets stay enforced; missing/nonfinite GC measurements now fail and allocation reports are uploaded.

The publisher starts automatically once its workflow exists on main. This introducing PR already gets measurement job summaries and artifacts. No external account, JMH fork, branch-protection change, or paid runner is required. docs/performance-ci.md compares GitHub-native reporting, dedicated runners, CodSpeed, Bencher, and github-action-benchmark and explains protocol/limits.

Validation

• Encoding: clean Maven verification on Corretto 21.0.11 and 25.0.4; 496 tests passed on each with -Werror.
• Current commit passed build, all 14 allocation checks, and the official protobuf conformance suite: CI run.
• CI tooling: 15 tests cover measurement validation, thresholds, invalid allocation data, artifact bounds/identity, token redirects, and stale/missing PR reports.
• End-to-end local smoke run clean-built main and candidate and successfully measured three cases with identical JDK/input sources; smoke output is explicitly excluded from performance evidence.
• Actionlint, shell syntax checks, Spotless, and whitespace checks passed.
• Full Java 21/25 CI comparisons passed on this revision: 30 cases per JVM, all 60 summaries independently reconstructed from raw JMH result files and validated. Performance run.

Context reuse, packed field names, primitive-list rewrites, and CodSpeed infrastructure remain deferred.

[Fyzu]

First full CI performance report

Performance run: passed on Java 21 and 25. All 60 base/head comparisons were validated against the raw JMH results. Each JVM comparison clean-built both revisions with the same compiler and benchmark inputs, then measured both on one runner.

JVM	Improvement signals	Regression signals	Inconclusive	Allocation growth alerts
Java 21	15	0	15	0
Java 25	16	0	14	0

Observed changes across the two within-JVM comparisons:

• Struct: throughput +95–116%, with allocation down 44%.
• Codegen/typed maps: throughput +12–29%, with allocation down 29–39%. Reflection maps allocate 26–34% less, while timing stays inconclusive under the alert rule.
• Reflection Timestamp: throughput +25–29%, with allocation down 18–19%. Typed Timestamp improves +40–45% on Java 25; its Java 21 timing remains inconclusive despite higher measured means.
• Simple-message timings remain inconclusive. There are no throughput regression signals or relative allocation growth alerts in this run.

These are advisory signals, not a universal speedup or a proof that smaller regressions are absent. A timing signal requires at least 10% change and non-overlapping JMH 99.9% intervals. The existing 14 absolute allocation budgets, build, and official conformance job also passed on this commit.

Both jobs used two forks, 3 × 1s warmup, 4 × 1s measurement, one thread, a 256 MiB heap, and the GC profiler. Java 21 ran on AMD EPYC 9V74; Java 25 ran on AMD EPYC 7763. Each base/head pair shared its runner; absolute throughput must not be compared between JVM jobs. Raw results, environment, build logs, and reports are available in the workflow artifacts for 90 days.

This first report is posted manually because GitHub starts the automatic publisher only after its workflow exists on main. After merge, it will update one report comment per PR and attach a report check to PR/default-branch commits.

Java 21: all 30 comparisons

Java 21 performance

Base: cb106d5 → candidate: 2facf0a
Shared benchmark source: 0a6e1acc0ab8a278090e2943d1fc820ac55b122c

Throughput alerts are advisory. Existing allocation budgets are enforced separately.
A timing signal needs at least 10% change and separated JMH 99.9% intervals; otherwise it is inconclusive.
Allocation alerts need both >5% and >16 B/op growth (or >16 B/op from zero).

Benchmark	Base ops/s	Candidate ops/s	Change	Timing	B/op base → candidate	Allocation
simpleCodegenUtf16	10,370,842 ±341,682	10,414,523 ±170,627	+0.4%	inconclusive	295.5 → 295.5	within alert threshold
simpleCodegenUtf8	10,241,226 ±72,012	10,207,074 ±266,454	-0.3%	inconclusive	271.5 → 271.5	within alert threshold
simpleTypedUtf16	7,943,958 ±39,171	7,766,646 ±414,158	-2.2%	inconclusive	295.5 → 295.5	within alert threshold
simpleTypedUtf8	8,363,583 ±181,199	8,390,696 ±178,066	+0.3%	inconclusive	271.5 → 271.5	within alert threshold
simpleReflectionUtf16	4,079,940 ±87,313	4,104,177 ±89,500	+0.6%	inconclusive	340.0 → 340.0	within alert threshold
simpleReflectionUtf8	4,014,666 ±178,880	4,127,033 ±24,504	+2.8%	inconclusive	316.0 → 316.0	within alert threshold
complexCodegenUtf16	698,043 ±12,166	710,305 ±30,278	+1.8%	inconclusive	1529.9 → 1481.9	within alert threshold
complexCodegenUtf8	692,663 ±15,013	709,149 ±19,545	+2.4%	inconclusive	1505.9 → 1457.9	within alert threshold
complexTypedUtf16	586,471 ±9,004	638,543 ±13,273	+8.9%	inconclusive	1393.9 → 1289.9	within alert threshold
complexTypedUtf8	611,716 ±31,875	687,931 ±5,818	+12.5%	improvement signal	1369.9 → 1265.9	within alert threshold
complexReflectionUtf16	321,765 ±19,919	365,878 ±8,730	+13.7%	improvement signal	1809.9 → 1353.9	within alert threshold
complexReflectionUtf8	334,000 ±5,222	389,007 ±25,672	+16.5%	improvement signal	1785.9 → 1329.9	within alert threshold
mapCodegenUtf16	76,413 ±3,349	87,047 ±5,906	+13.9%	improvement signal	8241.2 → 5025.3	within alert threshold
mapCodegenUtf8	67,909 ±3,733	82,759 ±5,461	+21.9%	improvement signal	8217.3 → 5001.2	within alert threshold
mapTypedUtf16	56,780 ±520	64,529 ±598	+13.6%	improvement signal	7586.5 → 4805.6	within alert threshold
mapTypedUtf8	58,561 ±3,755	65,371 ±2,249	+11.6%	improvement signal	7478.5 → 4781.6	within alert threshold
mapReflectionUtf16	41,492 ±2,788	45,309 ±1,391	+9.2%	inconclusive	11474.2 → 7634.2	within alert threshold
mapReflectionUtf8	42,588 ±2,987	46,605 ±2,715	+9.4%	inconclusive	11450.0 → 7610.2	within alert threshold
structCodegenUtf16	331,791 ±2,023	694,107 ±16,765	+109.2%	improvement signal	1304.9 → 735.7	within alert threshold
structCodegenUtf8	328,503 ±2,508	642,553 ±14,084	+95.6%	improvement signal	1280.9 → 711.7	within alert threshold
structTypedUtf16	328,649 ±1,653	693,528 ±6,308	+111.0%	improvement signal	1304.9 → 735.7	within alert threshold
structTypedUtf8	313,182 ±22,329	649,290 ±8,114	+107.3%	improvement signal	1280.9 → 711.7	within alert threshold
structReflectionUtf16	312,735 ±3,067	664,742 ±17,004	+112.6%	improvement signal	1304.9 → 735.7	within alert threshold
structReflectionUtf8	309,704 ±6,502	631,736 ±5,040	+104.0%	improvement signal	1280.9 → 711.7	within alert threshold
timestampCodegenUtf16	3,851,822 ±217,874	3,950,015 ±44,748	+2.5%	inconclusive	464.0 → 464.0	within alert threshold
timestampCodegenUtf8	3,381,086 ±640,889	4,027,935 ±410,233	+19.1%	inconclusive	440.0 → 440.0	within alert threshold
timestampTypedUtf16	2,038,832 ±471,660	2,784,428 ±313,894	+36.6%	inconclusive	568.0 → 464.0	within alert threshold
timestampTypedUtf8	2,504,749 ±12,815	3,011,387 ±966,249	+20.2%	inconclusive	544.0 → 440.0	within alert threshold
timestampReflectionUtf16	1,970,770 ±16,752	2,475,058 ±68,447	+25.6%	improvement signal	568.0 → 464.0	within alert threshold
timestampReflectionUtf8	2,010,862 ±12,434	2,597,959 ±15,150	+29.2%	improvement signal	544.0 → 440.0	within alert threshold

Java 25: all 30 comparisons

Java 25 performance

Base: cb106d5 → candidate: 2facf0a
Shared benchmark source: 0a6e1acc0ab8a278090e2943d1fc820ac55b122c

Throughput alerts are advisory. Existing allocation budgets are enforced separately.
A timing signal needs at least 10% change and separated JMH 99.9% intervals; otherwise it is inconclusive.
Allocation alerts need both >5% and >16 B/op growth (or >16 B/op from zero).

Benchmark	Base ops/s	Candidate ops/s	Change	Timing	B/op base → candidate	Allocation
simpleCodegenUtf16	10,578,497 ±280,965	10,458,581 ±184,067	-1.1%	inconclusive	295.5 → 295.5	within alert threshold
simpleCodegenUtf8	11,291,915 ±374,536	10,947,881 ±210,398	-3.0%	inconclusive	271.5 → 271.5	within alert threshold
simpleTypedUtf16	7,363,052 ±463,955	7,468,600 ±220,017	+1.4%	inconclusive	295.5 → 295.5	within alert threshold
simpleTypedUtf8	7,509,394 ±162,979	7,217,980 ±969,836	-3.9%	inconclusive	271.5 → 271.5	within alert threshold
simpleReflectionUtf16	3,812,740 ±138,725	3,834,762 ±88,143	+0.6%	inconclusive	340.0 → 340.0	within alert threshold
simpleReflectionUtf8	3,958,051 ±16,564	3,878,738 ±75,244	-2.0%	inconclusive	316.0 → 316.0	within alert threshold
complexCodegenUtf16	705,035 ±2,513	711,964 ±9,761	+1.0%	inconclusive	1529.9 → 1481.9	within alert threshold
complexCodegenUtf8	695,071 ±2,171	702,646 ±8,148	+1.1%	inconclusive	1505.9 → 1457.9	within alert threshold
complexTypedUtf16	600,122 ±5,221	645,933 ±34,480	+7.6%	inconclusive	1393.9 → 1289.9	within alert threshold
complexTypedUtf8	606,470 ±3,308	652,121 ±45,106	+7.5%	inconclusive	1369.9 → 1265.9	within alert threshold
complexReflectionUtf16	340,317 ±3,584	391,090 ±7,457	+14.9%	improvement signal	1785.9 → 1353.9	within alert threshold
complexReflectionUtf8	338,562 ±7,945	397,554 ±17,700	+17.4%	improvement signal	1761.9 → 1329.9	within alert threshold
mapCodegenUtf16	97,083 ±3,129	120,789 ±3,934	+24.4%	improvement signal	6719.4 → 4589.5	within alert threshold
mapCodegenUtf8	95,506 ±7,868	123,368 ±9,674	+29.2%	improvement signal	6575.4 → 4565.5	within alert threshold
mapTypedUtf16	71,887 ±993	82,843 ±3,198	+15.2%	improvement signal	6815.4 → 4805.5	within alert threshold
mapTypedUtf8	73,203 ±1,630	82,443 ±2,778	+12.6%	improvement signal	6875.5 → 4781.5	within alert threshold
mapReflectionUtf16	45,740 ±1,595	48,694 ±2,462	+6.5%	inconclusive	10268.1 → 7634.2	within alert threshold
mapReflectionUtf8	48,927 ±7,088	51,296 ±1,837	+4.8%	inconclusive	10243.9 → 7610.2	within alert threshold
structCodegenUtf16	346,039 ±20,824	748,367 ±35,609	+116.3%	improvement signal	1304.9 → 735.7	within alert threshold
structCodegenUtf8	351,087 ±4,581	686,471 ±37,699	+95.5%	improvement signal	1280.9 → 711.7	within alert threshold
structTypedUtf16	344,243 ±11,440	738,578 ±4,930	+114.6%	improvement signal	1304.9 → 735.7	within alert threshold
structTypedUtf8	345,069 ±7,138	701,693 ±1,833	+103.3%	improvement signal	1280.9 → 711.7	within alert threshold
structReflectionUtf16	339,522 ±939	708,234 ±33,833	+108.6%	improvement signal	1304.9 → 735.7	within alert threshold
structReflectionUtf8	327,948 ±12,777	667,536 ±12,051	+103.5%	improvement signal	1280.9 → 711.7	within alert threshold
timestampCodegenUtf16	4,133,560 ±15,560	4,135,420 ±9,009	+0.0%	inconclusive	464.0 → 464.0	within alert threshold
timestampCodegenUtf8	4,799,459 ±106,949	4,803,623 ±11,117	+0.1%	inconclusive	440.0 → 440.0	within alert threshold
timestampTypedUtf16	2,492,100 ±42,363	3,483,652 ±17,731	+39.8%	improvement signal	568.0 → 464.0	within alert threshold
timestampTypedUtf8	2,728,228 ±133,774	3,943,016 ±24,407	+44.5%	improvement signal	544.0 → 440.0	within alert threshold
timestampReflectionUtf16	1,958,363 ±234,616	2,454,709 ±82,159	+25.3%	improvement signal	568.0 → 464.0	within alert threshold
timestampReflectionUtf8	2,188,362 ±20,057	2,814,535 ±60,239	+28.6%	improvement signal	544.0 → 440.0	within alert threshold