test(conformance): bump referee to 0.2.0-alpha.7; drop server-sse-polling from baseline (conformance#366)#2366
Conversation
…ling from baseline conformance#366 landed in alpha.7 — the referee now negotiates 2025-11-25 when probing the SSE priming/retry SHOULD checks, so server-sse-polling passes against the SDK's version-gated priming and is removed from the expected-failures baseline. tasks-status-notifications is still SKIPped unconditionally by the referee (0/0), so it stays intentionally absent. No other tasks-* changes at alpha.7. All six conformance legs (server, server:draft, server:extensions, server:2026, client:all, client:2026) exit 0 against the new pin.
|
@modelcontextprotocol/client
@modelcontextprotocol/codemod
@modelcontextprotocol/server
@modelcontextprotocol/server-legacy
@modelcontextprotocol/express
@modelcontextprotocol/fastify
@modelcontextprotocol/hono
@modelcontextprotocol/node
commit: |
There was a problem hiding this comment.
LGTM — straightforward conformance referee pin bump (alpha.6 → alpha.7) with the baseline reconciled per the documented procedure; test-only, no production code touched.
Extended reasoning...
Overview
This PR touches only the conformance test harness: test/conformance/package.json bumps the @modelcontextprotocol/conformance devDependency pin from 0.2.0-alpha.6 to 0.2.0-alpha.7, pnpm-lock.yaml is the matching lockfile update, and test/conformance/expected-failures.yaml removes the now-passing server-sse-polling entry and updates the alpha-version references in comments. No SDK package source code is modified.
Security risks
None. The change is a version pin of a test-only devDependency in a private workspace package plus a YAML baseline edit; nothing ships to consumers and no auth/crypto/transport code is touched.
Level of scrutiny
Low. This follows the exact reconciliation procedure documented at the top of expected-failures.yaml ("Newer conformance releases are adopted by deliberately bumping the package.json pin and reconciling this file in the same change"), mirroring the prior alpha.6 bump pattern. The baseline runner itself enforces correctness: stale entries fail CI, so an incorrect removal of server-sse-polling would surface immediately in the conformance legs.
Other factors
The PR description reports all six conformance legs exiting 0 against alpha.7, and the comment edits (alpha.6 → alpha.7, removal of the conformance#256/#366 hold note) are consistent with the diff. The missing changeset is fine since @modelcontextprotocol/test-conformance is private and unpublished. The bug-hunting system found no issues.
Bump the conformance referee pin to
0.2.0-alpha.7and reconcile baselines.Motivation and Context
alpha.7 carries conformance#366 (server-sse-polling now negotiates
2025-11-25for the priming/retry SHOULD checks). With it,server-sse-pollingpasses 3/3 against the SDK.How Has This Been Tested?
All 6 conformance legs exit 0:
server42/0,server:draft81/0,server:extensions136 pass / 27 expected (tasks-*),server:2026110/0,client:all438/0,client:2026374/0.Breaking Changes
None.
Types of changes
Checklist
Additional context
server-sse-pollingremoved fromexpected-failures.yaml(now passes).tasks-status-notificationsstill referee-side SKIP at alpha.7; comment updated.tasks-*baseline entries unchanged.