Return 422 when the evaluation function rejects a submission

[m-messer]

Summary

When the evaluation function reports that it can't process a submission (e.g. an unparseable expression), shimmy now responds with 422 Unprocessable Entity instead of 500.

• New supervisor.InvalidSubmissionError, returned by:
  • the RPC adapter when the worker sends JSON-RPC error code 422 (InvalidSubmissionRpcCode)
  • the file adapter when the worker sends error.code == "INVALID_SUBMISSION" (InvalidSubmissionFileCode)
• runtime: getErrorStatusCode maps InvalidSubmissionError to 422, and newErrorResponse returns only the evaluation function's message, without the wrapping context. If a worker response has no result, the worker's error message is now used instead of a generic unmarshal error.
• muEd ServeEvaluate: non-200 runtime responses are now converted to muEd ErrorResponse bodies (422/400 → VALIDATION_ERROR, anything else → INTERNAL_ERROR), instead of passing the legacy body through.
• mued_v0.1.0.yml: adds the 422-UnprocessableSubmission response to /evaluate.

CI change

Also disables the sandbox integration test job: nsjail fails to remount /proc/sys/fs/binfmt_misc on the current ubuntu-24.04 runner image, so every sandboxed launch exits 255. Pinning nsjail to an older commit didn't fix it and was reverted. The job should be turned back on once this is fixed.

Related

• toolkit-python: Report unprocessable submissions as invalid-submission errors (422) toolkit-python#14 sends these error codes from the worker side.
• compareSets: Raise on unparseable submissions so they surface as 422 compareSets#13 is the first evaluation function to use it.

🤖 Generated with Claude Code