Benchmarks

Performance results for the Branchline interpreter and VM.

Latest release

Runtime and allocation (mean, p95)

Source data: CSV.

Runtime	Dataset	Mean (us/op)	p95 (us/op)	Alloc (B/op)	Params
Interpreter pathExpressions	Small	0.483	0.461	784.215	collectionType=hash
Interpreter pathExpressions	Small	0.571	0.512	784.247	collectionType=persistent
Interpreter pathExpressions	Medium	0.523	0.491	784.216	collectionType=hash
Interpreter pathExpressions	Medium	0.546	0.521	784.245	collectionType=persistent
Interpreter pathExpressions	Large	0.503	0.491	784.199	collectionType=hash
Interpreter pathExpressions	Large	0.531	0.530	784.219	collectionType=persistent
Interpreter arrayComprehensions	Small	1.208	1.132	1376.501	collectionType=hash
Interpreter arrayComprehensions	Small	1.164	1.112	1376.494	collectionType=persistent
Interpreter arrayComprehensions	Medium	6.854	7.182	5114.281	collectionType=hash
Interpreter arrayComprehensions	Medium	7.167	7.368	5170.192	collectionType=persistent
Interpreter arrayComprehensions	Large	39.359	49.856	32776.972	collectionType=hash
Interpreter arrayComprehensions	Large	43.578	54.464	32833.489	collectionType=persistent
Interpreter typicalTransform	Small	1.334	1.282	848.453	collectionType=hash
Interpreter typicalTransform	Small	1.457	1.392	824.501	collectionType=persistent
Interpreter typicalTransform	Medium	10.816	10.720	2322.185	collectionType=hash
Interpreter typicalTransform	Medium	11.197	11.088	2370.793	collectionType=persistent
Interpreter typicalTransform	Large	53.675	63.040	8746.021	collectionType=hash
Interpreter typicalTransform	Large	55.447	64.960	8802.413	collectionType=persistent
Interpreter deepNestedRead	Small	9.269	9.152	2665.923	collectionType=hash
Interpreter deepNestedRead	Small	9.069	9.042	2682.221	collectionType=persistent
Interpreter deepNestedSet	Small	12.168	13.104	15635.672	collectionType=hash
Interpreter deepNestedSet	Small	11.887	12.080	14667.436	collectionType=persistent
Interpreter nestedPathSetInLoop	Small	2.420	2.352	4240.803	collectionType=hash
Interpreter nestedPathSetInLoop	Small	2.601	2.484	4480.860	collectionType=persistent
Interpreter nestedPlusAssign	Small	2.247	2.132	3200.921	collectionType=hash
Interpreter nestedPlusAssign	Small	2.329	2.232	3200.910	collectionType=persistent
Interpreter stdlibCascadeAppend	Small	1.736	1.692	4368.649	collectionType=hash
Interpreter stdlibCascadeAppend	Small	2.059	1.944	4368.796	collectionType=persistent
Interpreter deepNestedRead	Medium	182.155	193.280	38087.847	collectionType=hash
Interpreter deepNestedRead	Medium	178.057	191.232	40577.373	collectionType=persistent
Interpreter deepNestedSet	Medium	228.287	244.480	272693.282	collectionType=hash
Interpreter deepNestedSet	Medium	231.112	243.136	275179.276	collectionType=persistent
Interpreter nestedPathSetInLoop	Medium	22.532	24.032	41478.981	collectionType=hash
Interpreter nestedPathSetInLoop	Medium	23.735	25.376	43935.609	collectionType=persistent
Interpreter nestedPlusAssign	Medium	19.404	20.608	38892.923	collectionType=hash
Interpreter nestedPlusAssign	Medium	19.764	20.864	38989.209	collectionType=persistent
Interpreter stdlibCascadeAppend	Medium	14.795	15.744	46076.122	collectionType=hash
Interpreter stdlibCascadeAppend	Medium	14.889	15.808	46052.503	collectionType=persistent
Interpreter deepNestedRead	Large	2251.729	2289.664	440685.226	collectionType=hash
Interpreter deepNestedRead	Large	2326.345	2400.256	452727.240	collectionType=persistent
Interpreter deepNestedSet	Large	2592.467	2648.678	3640846.138	collectionType=hash
Interpreter deepNestedSet	Large	2526.443	2578.022	3352928.793	collectionType=persistent
Interpreter nestedPathSetInLoop	Large	115.437	124.288	204716.771	collectionType=hash
Interpreter nestedPathSetInLoop	Large	124.110	133.888	216776.171	collectionType=persistent
Interpreter nestedPlusAssign	Large	95.046	104.576	200656.079	collectionType=hash
Interpreter nestedPlusAssign	Large	107.789	117.248	200737.816	collectionType=persistent
Interpreter stdlibCascadeAppend	Large	80.099	88.960	242584.604	collectionType=hash
Interpreter stdlibCascadeAppend	Large	79.098	88.064	242585.160	collectionType=persistent
VM pathExpressions	Small	1.042	0.982	704.448	n/a
VM pathExpressions	Medium	1.044	0.982	704.437	n/a
VM pathExpressions	Large	1.053	1.002	704.433	n/a
VM arrayComprehensions	Small	2.561	2.472	800.730	n/a
VM arrayComprehensions	Medium	14.435	14.416	803.594	n/a
VM arrayComprehensions	Large	94.766	104.576	5669.857	n/a
VM typicalTransform	Small	7.083	6.872	1065.880	n/a
VM typicalTransform	Medium	59.049	67.072	4676.051	n/a
VM typicalTransform	Large	305.319	321.536	20721.367	n/a
VM canonicalMutationStyleLegacySelfAppendVm	Small	2.986	3.124	2241.013	n/a
VM canonicalMutationStylePlusAssignVm	Small	2.993	2.792	1688.908	n/a
VM xmlToJsonOrderVm	Small	33.675	49.472	51818.792	n/a
VM canonicalMutationStyleLegacySelfAppendVm	Medium	25.484	27.744	25927.565	n/a
VM canonicalMutationStylePlusAssignVm	Medium	26.937	29.728	46695.393	n/a
VM xmlToJsonOrderVm	Medium	34.191	49.152	51818.896	n/a
VM canonicalMutationStyleLegacySelfAppendVm	Large	168.689	180.480	136499.536	n/a
VM canonicalMutationStylePlusAssignVm	Large	178.479	191.488	1030747.775	n/a
VM xmlToJsonOrderVm	Large	35.921	53.376	51819.904	n/a
Other canonicalMutationStyleLegacySelfAppend	Small	1.680	2.204	1768.637	n/a
Other canonicalMutationStylePlusAssign	Small	1.164	1.082	1264.499	n/a
Other contractInferAndDiff	Small	103.823	119.552	287370.776	n/a
Other conversionLossAudit	Small	27.397	39.040	43520.262	n/a
Other inspectNormalizeLoop	Small	67.967	84.736	129357.913	n/a
Other xmlToJsonOrder	Small	33.778	50.432	52850.521	n/a
Other canonicalMutationStyleLegacySelfAppend	Medium	12.798	13.568	21164.296	n/a
Other canonicalMutationStylePlusAssign	Medium	9.382	9.664	16338.842	n/a
Other contractInferAndDiff	Medium	106.501	124.160	287903.258	n/a
Other conversionLossAudit	Medium	28.580	44.736	44481.982	n/a
Other inspectNormalizeLoop	Medium	65.785	79.744	127314.812	n/a
Other xmlToJsonOrder	Medium	33.421	47.808	52883.819	n/a
Other canonicalMutationStyleLegacySelfAppend	Large	62.418	71.680	112523.158	n/a
Other canonicalMutationStylePlusAssign	Large	46.266	55.424	88514.924	n/a
Other contractInferAndDiff	Large	102.831	117.888	287363.754	n/a
Other conversionLossAudit	Large	28.202	38.464	44481.521	n/a
Other inspectNormalizeLoop	Large	69.578	89.472	127149.840	n/a
Other xmlToJsonOrder	Large	33.713	48.064	52851.104	n/a

Interpreter vs VM ratio (mean)

Benchmark	Dataset	Interpreter (us/op)	VM (us/op)	Ratio (VM/Interp)
pathExpressions	Small	0.571	1.042	1.824x
pathExpressions	Medium	0.546	1.044	1.911x
pathExpressions	Large	0.531	1.053	1.982x
arrayComprehensions	Small	1.164	2.561	2.201x
arrayComprehensions	Medium	7.167	14.435	2.014x
arrayComprehensions	Large	43.578	94.766	2.175x
typicalTransform	Small	1.457	7.083	4.860x
typicalTransform	Medium	11.197	59.049	5.274x
typicalTransform	Large	55.447	305.319	5.506x

Release history

• perf-ce58599 (prerelease)
• perf-7285ce0 (prerelease)
• snapshot-e206245 (prerelease)
• snapshot-c502022 (prerelease)
• v0.10.0 (prerelease)

Full archive: Release benchmark history.

Methodology

Benchmark results are split by measurement type so the public pages do not collapse different costs into one speed claim.

Measurement	What it means	Comparison rule
Cross-engine JSON execution	JSON-like transformations that can run through Branchline, a Kotlin baseline, and configured JSONata engines.	Compare only rows with equivalent transformation intent and a recorded validation state.
Interpreter vs VM execution	Branchline interpreter and VM throughput over the same JMH datasets.	Use the same machine, JDK, inputs, and release artifacts; watch for VM fallback or overhead notes.
Branchline-specific measurements	XML conversion, contract inference or diff, inspect/normalize, conversion-loss audit, and canonical mutation style.	Treat as Branchline product-surface measurements unless a fair external-engine analog exists.
Publication health	Release summary Markdown, generated HTML pages, and CSV assets.	Verify links point to generated HTML directories and CSV assets sit next to each release page.

Public benchmark rows should identify the case id, workload category, whether the case is product-representative, engine, score, error margin, allocation when available, expected-failure or unavailable reason, validation state, and notes.

What we measure

Area	Interpreter benchmark	VM benchmark
Path expressions	InterpreterTransformBenchmark.pathExpressions	VMTransformBenchmark.pathExpressions
Array comprehensions	InterpreterTransformBenchmark.arrayComprehensions	VMTransformBenchmark.arrayComprehensions
Typical transform	InterpreterTransformBenchmark.typicalTransform	VMTransformBenchmark.typicalTransform

Datasets: Small (10x5), Medium (100x10), Large (500x25).

Representative case categories

Category	Status	Public framing
ETL and API normalization	Planned Tier 1 cross-engine cases such as etl-order-normalization and api-envelope-shaping.	Representative JSON execution comparison after semantic validation is available.
Large filter/map/aggregate	Planned Tier 1 cross-engine case large-filter-map-aggregate.	Collection throughput comparison, separate from parse or compile cost.
Contract drift and governance	Planned Tier 1 JSON case contract-drift-projection; Tier 2 contract inference/diff work remains Branchline-specific.	Execution can be compared only for equivalent JSON projection; contract analysis is a Branchline-specific measurement.
Numeric precision invoice handling	Planned Tier 1 cross-engine case numeric-precision-invoice.	Requires notes for numeric equivalence and large-id handling before speed claims.
XML, inspect, and conversion audit	Planned Tier 2 Branchline-specific cases.	Product-surface measurements, not JSONata speed comparisons by default.

Current release summaries may still contain micro or suite-shaped cases. Treat them as coverage and plumbing evidence until representative cases and semantic validation are present.

Run locally

./gradlew :interpreter-benchmarks:jmh :vm-benchmarks:jmh

Notes

• Compare runs on the same machine and JDK.
• Report deltas against the previous release.
• Missing suites, unavailable engines, expected failures, semantic mismatches, and VM fallback must be visible in the release notes or generated tables.