TCK Jakarta JSON-P 2.1 + JSON-B 3.0

This page consolidates the official TCK results for Jakarta JSON Processing 2.1 and Jakarta JSON Binding 3.0 run against Champollion. Daily source of truth: lien:https://codeberg.org/Vidocq/champollion/src/branch/main/TCK.md[TCK.md] in the repo.

The champollion-tck module is deliberately out of reactor (standalone Model 4.0.0 POM, no <parent>) to bypass ShrinkWrap Maven Resolver 3.3 which cannot parse Maven Model 4.1.0 POMs. Always invoke through the run-official-tck-*.sh scripts — never via mvn -pl champollion-tck.

Global summary

TCK	Total	PASS	Fails/Errors	Status
Jakarta JSON Processing 2.1	179 (api)	178 (99.4%)	1 ERROR (env)	✅ 100% of applicable tests PASS
Jakarta JSON Binding 3.0	~250	ongoing	variable	🔄 baseline being consolidated

TCK

Total

PASS

Fails/Errors

Status

Jakarta JSON Processing 2.1

179 (api)

178 (99.4%)

1 ERROR (env)

✅ 100% of applicable tests PASS

Jakarta JSON Binding 3.0

~250

ongoing

variable

🔄 baseline being consolidated

Last baseline date: 2026-05-03 (cf. lien:https://codeberg.org/Vidocq/champollion/src/branch/main/TCK.md[TCK.md] §First TCK run).

JSON-P 2.1 — detailed status

Metric	Baseline 2026-05-03	After M2.x + M3.4	After M6.x
Tests run	197	179 (api only)	179
PASS	65 (33%)	168 (94%)	178 (99.4%) ✅
FAIL	112	7	0
ERROR	20	4	1 (env)

The single remaining ERROR is JSONPSigTest.signatureTest — see Tests skipped due to environment below.

Major stability jump

The TCK’s JsonProviderTest.systemProperty() polluted System.getProperty("jakarta.json.provider") without cleanup, which caused JsonProvider.provider() to be shadowed by a mock in all subsequent tests. Fix: forkCount=1, reuseForks=false in surefire + isolating pluggability into its own profile. Detailed in TCK.md.

Recent fixes (M6.x)

ChampollionJsonObject.getString/getInt/getBoolean/isNull now throw NPE if the key is missing (spec §2.1.4).
ChampollionJsonObjectBuilder.remove(null) & addAll(null) throw NPE.
JsonProvider.createValue(Number) accepts Integer/Long/Double/Float/Short/Byte/BigDecimal.
createObjectBuilder(JsonObject|Map) & createArrayBuilder(JsonArray|Collection) overrides.
autoDetectingReader throws JsonException when the encoding cannot be determined (single byte 0x00 → jsonObjectUnknownEncoding.json).
ChampollionJsonPointer: the exception on a malformed ~n is deferred to resolution (TCK testResolvePathWithUnencodedTilde catches the exception in getValue(), not in createPointer()).

JSON-B 3.0 — known challenges

Test Status Justification

Test	Status	Justification
`JsonbBuilderTest.testCreateConfig`, `testWithConfig`	under investigation	`JsonbBuilder.newBuilder()` + `withConfig()` behaviour to dig into. Spec interpretation differs between Yasson and the TCK expectation.
others	under investigation	First global run to surface divergences. Per-test fine analysis on the next-release schedule.

JsonbBuilderTest.testCreateConfig, testWithConfig

under investigation

JsonbBuilder.newBuilder() + withConfig() behaviour to dig into. Spec interpretation differs between Yasson and the TCK expectation.

others

under investigation

First global run to surface divergences. Per-test fine analysis on the next-release schedule.

The contractual target remains 100% PASS before any structural merge on champollion-jsonb. The --static (codegen) mode must achieve the same score as runtime — validated by differential testing (cf. Internals).

Tests skipped due to environment

Test Reason

Test	Reason
`JSONPSigTest.signatureTest`	The official `jakarta-jsonp-tck-2.1.0.zip` does not ship the Eclipse signature file required (non-public artifact). Clean skip — `EX_CONFIG` (78). Not an applicative gap.

JSONPSigTest.signatureTest

The official jakarta-jsonp-tck-2.1.0.zip does not ship the Eclipse signature file required (non-public artifact). Clean skip — EX_CONFIG (78). Not an applicative gap.

Tests deliberately skipped (applicative)

As of today, none. Should a spec-interpretation divergence appear later, it will be documented here with spec citation, test hash, and reactivation plan — that’s the Champollion release discipline.

Reproduce locally

Install the official TCKs

The repo-root script downloads the official ZIPs from download.eclipse.org and installs them via mvn install:install-file with their genuine POMs (idempotent — won’t re-download if already present):

cd champollion
./install-tck.sh          # JSON-P + JSON-B
./install-tck.sh jsonp    # JSON-P 2.1 only
./install-tck.sh jsonb    # JSON-B 3.0 only

Maven coordinates installed:

jakarta.json:jakarta-json-tck:2.1.0
jakarta.json.bind:jakarta.json.bind-tck:3.0.0

Run the JSON-P 2.1 TCK

cd champollion
./run-official-tck-jsonp-2.1.sh             # smoke test (default)
./run-official-tck-jsonp-2.1.sh all         # full suite
./run-official-tck-jsonp-2.1.sh -Dtest=Foo  # targeted test

Report: champollion-tck/target/tck-report-jsonp.txt.

Run the JSON-B 3.0 TCK

cd champollion
./run-official-tck-jsonb-3.0.sh                    # smoke test (default)
./run-official-tck-jsonb-3.0.sh all                # full suite (runtime mode)
./run-official-tck-jsonb-3.0.sh all --static       # static codegen mode
./run-official-tck-jsonb-3.0.sh -Dtest=Foo         # targeted test

The --static flag recompiles the TCK fixtures with champollion-codegen-apt to validate runtime ↔ codegen consistency.

Report: champollion-tck/target/tck-report-jsonb.txt.

Behaviour when the TCK is not installed

The scripts check the presence of the TCK jar in ~/.m2/. If absent: exit 78 (EX_CONFIG, clean skip). The Forgejo CI job (.forgejo/workflows/ci.yml) treats this exit code as a skip and does not fail the build — useful while TCKs are not pre-installed on the runner.

Release discipline

No structural merge on champollion-jsonp or champollion-jsonb without a 100% TCK PASS.
Any change in champollion-codegen-apt must preserve the differential testing score before merge.
Disabled tests are documented in TCK.md with spec citation, test hash, and reactivation plan. No silent skip.