jena-webfunction-plugin

Apache Jena binding for the WebAssembly Component Model. Registers ARQ filter functions under http://tegmentum.ai/ns/webfunction/ that load a WASM component from a URL and invoke its evaluate export.

Part of a three-binding family that all share one component ABI:

Binding	Repo
Stardog	tegmentum/stardog-webfunction-plugin
Apache Jena	you are here
Eclipse RDF4J	tegmentum/rdf4j-webfunction-plugin

The WIT world at src/main/wit/webfunction.wit (package stardog:webfunction@0.2.0) is byte-for-byte identical across the three repos, so a single Rust component runs unmodified under any of the three SPARQL engines. WASM runtime is webassembly4j (wasmtime provider).

Component runtime: webassembly4j (wasmtime provider). Component ABI shared with the Stardog binding — the WIT world at src/main/wit/webfunction.wit is package stardog:webfunction@0.2.0 and is loaded verbatim from the Stardog plugin for cross-framework component reuse.

SPARQL surfaces

The wf:call function is exposed through four different SPARQL surfaces; all back onto the same component's evaluate / aggregate-step / aggregate-finish exports.

Shape	Syntax	When to reach for it
Filter	BIND(wf:call(<url>, args...) AS ?x)	one value out of one wasm call
Aggregate	SELECT (<wf:call-agg>(<url>, ?v) AS ?sum)	reduce query rows to one value
Property	?x wf:call (<url> args...)	multi-row output, single subject variable
SERVICE	SERVICE <url> { BIND(...) }	multi-row, multi-var output

Usage

Auto-registered via Jena's JenaSubsystemLifecycle SPI on classpath. Then in SPARQL:

PREFIX wf: <http://tegmentum.ai/ns/webfunction/>
SELECT ?result WHERE {
  BIND(wf:call(<file:/path/to/component.wasm>, "stardog") AS ?result)
}

Performance

• Shared static Engine built once from WebFunctionConfig on first wf:call.
• ConcurrentHashMap<URL, Component> caches compiled components per URL. Repeat calls to the same wasm skip download + compile; only the per-call ComponentInstance is fresh.
• Cost: webfunctions.* system properties are read once at first use — changing them mid-run has no effect. Test-only JenaWasmInstance.resetCache() drops shared state for isolation.
• Bench (Darwin aarch64, to_upper component, warm cache):
  • evaluate: ~24 µs/op (42k ops/s)
  • instantiate: ~513 µs/op

Testing

Two paths depending on your environment:

mvn test — direct-in-JVM tests: TestWfCall, TestWfCallAgg, TestWfCallService, TestWfCallPropertyFunction, TestComponentBench. Runs WebFunctionInit.register() up-front and calls into Jena's ARQ registries directly, no HTTP server required.

mvn verify — additionally runs FusekiWasmIT via Testcontainers: boots a stain/jena-fuseki container, mounts the shaded plugin JAR from target/ into /fuseki/extra/ so Fuseki's JenaSystem.init() picks up the JenaSubsystemLifecycle SPI, drops the smoke-test wasm into /opt/wasm/, and POSTs a wf:call SPARQL query to the running server via QueryExecutionHTTP. Requirements: Docker running. On Apple Silicon, set DOCKER_DEFAULT_PLATFORM=linux/amd64 if the image is amd64-only.

The IT skips cleanly when: Docker is unavailable, the shaded JAR hasn't been built (mvn package), the wasm hasn't been built, or the shaded JAR doesn't include a natives/linux-x86_64/libwasmtime4j.so (needed for the container to load wasmtime — Fuseki runs Linux/amd64; dev laptops typically shade only their host classifier). Override the Fuseki image with -Dfuseki.image=..., the plugin JAR with -Dwf.plugin.jar=..., and the wasm with -Dwf.toUpper.wasm=....

Config (system properties)

• webfunctions.engine.provider (default wasmtime)
• webfunctions.engine.id
• webfunctions.fuel.limit
• webfunctions.memory.max.bytes
• webfunctions.timeout.millis
• webfunctions.exec.max.millis
• webfunctions.max.instances
• webfunctions.table.max.elements