test(bdd): Gherkin/BDD e2e pilot on the existing Vitest runner#1038
test(bdd): Gherkin/BDD e2e pilot on the existing Vitest runner#1038branarakic wants to merge 2 commits into
Conversation
Self-contained `bdd/` package that runs Gherkin .feature specs via @amiceli/vitest-cucumber (Vitest-native — one runner, one CI lane, no standalone Cucumber): - entity-predicate.feature: pure smoke spec over the OT-RFC-43/44 dual-read invariant in packages/core/src/entity-predicate.ts. Exposed as the package's `test` script so it runs under the standard `turbo test` lane (27 step-tests). - ka-lifecycle.feature: @DevNet KA lifecycle (create -> write -> finalize -> promote -> publish) mirroring devnet/v10-core-flows fullPublish(); tag-gated to skip cleanly when no .devnet is present (run via `test:devnet`). Route/payload surface of the @DevNet spec cross-checked against the live daemon handlers (assertion.ts, memory.ts, status.ts). See bdd/README.md for the rationale, tradeoffs, and how to extend. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
| "test:all": "vitest run --config vitest.config.ts" | ||
| }, | ||
| "devDependencies": { | ||
| "@amiceli/vitest-cucumber": "^6.5.0", |
There was a problem hiding this comment.
🔴 Bug: This PR adds a new external dependency but does not update pnpm-lock.yaml. All of the repo's CI jobs install with pnpm install --frozen-lockfile, so they will fail as soon as this package lands. Regenerate and commit the lockfile alongside this manifest change.
There was a problem hiding this comment.
This looks like a false positive: pnpm-lock.yaml was committed in the initial commit (3 @amiceli/vitest-cucumber entries) and pnpm install --frozen-lockfile passes (exit 0) on this branch. 9448c877e adds the new @origintrail-official/dkg-core workspace dep with the lockfile updated alongside; frozen install still passes.
| import { | ||
| isAssertionEntityPredicate, | ||
| isEntityPredicate, | ||
| } from '../../packages/core/src/entity-predicate.js'; |
There was a problem hiding this comment.
🟡 Issue: Importing ../../packages/core/src/... directly bypasses the workspace package graph, so Turbo/PNPM do not see bdd as depending on dkg-core. That can let turbo test reuse a cached smoke-spec result after packages/core/src/entity-predicate.ts changes, which defeats the purpose of this regression check. Prefer a declared workspace dependency on @origintrail-official/dkg-core and import through its public export.
There was a problem hiding this comment.
Good catch — fixed in 9448c877e. Now imports the helpers via the public @origintrail-official/dkg-core export, declared as a workspace:* dependency, plus a bdd/turbo.json setting test to dependsOn: ["^build"]. turbo run test --filter=@origintrail-official/dkg-bdd --dry now shows @origintrail-official/dkg-core#build as a dependency, so the smoke cache invalidates when that code changes.
| if (!apiPort) return { available: false, reason: 'no apiPort in node config' }; | ||
|
|
||
| return { | ||
| available: true, |
There was a problem hiding this comment.
🟡 Issue: detectDevnet() marks the devnet as available based only on .devnet/node1/config.json, so a stopped but still-provisioned cluster will be treated as runnable. In that state test:all will include the @devnet scenario and then fail in BeforeAllScenarios, even though the README says it should self-skip without a cluster. Gate availability on a live status probe as well, or downgrade an offline node to a skip instead of an error.
There was a problem hiding this comment.
Fixed in 9448c877e. Availability is now gated on a live /api/status probe (3s timeout), not just on-disk provisioning. Verified by simulation: with a provisioned-but-stopped node (config present, port dead), test:all skips cleanly (@devnet scenarios skipped — node1 provisioned but not responding ...) and exits 0 instead of failing in BeforeAllScenarios.
- Import isEntityPredicate/isAssertionEntityPredicate via the public @origintrail-official/dkg-core export and declare it as a workspace dependency; add bdd/turbo.json so `test` dependsOn `^build`. Turbo now sees bdd -> dkg-core and invalidates the smoke-spec cache when that code changes (the relative `../../packages/core/src` import bypassed the package graph and could serve a stale cached pass). - Gate @DevNet availability on a live /api/status probe (with a 3s timeout), not just on-disk provisioning, so a stopped-but-provisioned cluster skips cleanly instead of failing in BeforeAllScenarios. Verified by simulation. The review's "pnpm-lock.yaml not updated" finding is a false positive: the lockfile was committed in the initial commit (3 @amiceli entries) and `pnpm install --frozen-lockfile` passes (exit 0). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
1 participant
What & why
A pilot evaluating Gherkin/BDD end-to-end tests for the DKG monorepo — describing system behaviour in human-readable
Given/When/Then— before committing to it more broadly.It deliberately runs on the project's existing Vitest runner via
@amiceli/vitest-cucumber(peerDepvitest ^4.0.4, matches our^4.0.18): one runner, one CI lane, no standalone Cucumber, no Jest typings.Self-contained in a new
bdd/workspace package — nothing existing is changed except adding the workspace member + lockfile entry.What's included
bdd/features/entity-predicate.featureScenario Outlineover the OT-RFC-43/44dkg:entity↔ legacydkg:rootEntitydual-read invariant (exercises the realpackages/core/src/entity-predicate.ts)testscript → runs underturbo test(CI). ✅ 27 step-tests greenbdd/features/ka-lifecycle.featurecreate → write → finalize → promote → publishas readable behaviour; mirrorsdevnet/v10-core-flowsfullPublish()@devnet, tag-gated → self-skips cleanly with no cluster; run viatest:devnetVerification
turbo test(the CI path), not just direct vitest.@devnetspec's every route / request payload / response field was cross-checked against the live daemon handlers (assertion.ts,memory.ts,status.ts) and matchesfullPublish()—write→written,promote→promotedCount,finalize→merkleRoot/eip712Digest,publish→kaId/status,/api/status→identityId. It has not been executed against a live cluster in this pilot.auth.tokenparsing (matches the harness's#-comment handling).Scope / how to extend
This is intentionally narrow — a working template + proposal (see
bdd/README.mdfor the full rationale and tradeoffs). It does not rewrite the 9 existing devnet Vitest suites. If we expand, the highest readable-spec value is the KA-routes unification and conviction-staking flows. Feedback on the ergonomics welcome.🤖 Generated with Claude Code