openCypher conformance

GraphForge tests its Rust query engine against the vendored openCypher 2024.3 Technology Compatibility Kit (TCK). A scenario checks a query’s expected result or error. This describes compatibility with that corpus, rather than every extension offered by other graph databases.

Corpus and regression baseline

The corpus contains 3,897 upstream scenarios and one GraphForge regression scenario. The committed passing baseline lists all 3,898 identities. The corpus metadata records those counts; it is not a per-feature support matrix.

A full run must preserve every baseline case. A failure, skipped case, or missing identity causes the baseline check to fail. An unrelated improvement cannot cancel a regression. Baseline membership records the required passing set; consult the test output for results from a particular source commit.

Newly passing cases produce TCK XPASS warnings for review before inclusion in the baseline. Newly added cases outside the baseline are exploratory coverage: their failure alone does not fail the baseline check. This is the limited meaning of the runner’s “advisory” label. Existing baseline scenarios are required, and the runner does not use @skip-rust to exclude them.

The Rust BDD runner implements this set comparison. The PR workflow runs it in Rust Harness, Doc, and Feature Tests when the code-change filter selects that lane; its result feeds CI Gate. Prose-only changes do not run the whole corpus.

Run the checks

From a source checkout with the development prerequisites:

make test-tck

For a focused investigation, select feature filenames by substring:

TCK_ONLY=Temporal cargo test -p graphforge-api --test bdd

A focused run bypasses the whole-corpus baseline check because a subset cannot satisfy it. It is useful for debugging, but cannot establish full conformance. The normal CI run does not set TCK_ONLY.

When a reviewed change adds passing scenarios, a maintainer can regenerate the baseline and inspect the diff:

BLESS_TCK_BASELINE=1 cargo test -p graphforge-api --test bdd

Do not remove failing cases to make a run pass. Baseline edits need ordinary review, and the final full run must preserve the intended scenario set.

Correctness and performance are separate

Passing the TCK establishes results for those scenarios. It does not certify maximum graph size, workload latency, or production maturity.

The BDD runner’s per-scenario timing output is diagnostic. Performance comparisons use the separate Divan/BenchExec procedure with matched hardware, build, workload, and configuration. See benchmarking for the reproducible method and scale guidance for the claim-to-evidence table. Run results belong to their issue or CI run rather than becoming a current performance promise on this page.

For supported query syntax and examples, see the Cypher guide. Analyst verbs such as rank and cluster are separate APIs; their availability is not evidence that a Cypher scenario passes.