Prevent Python SDK harness disconnect tracebacks from contaminating stderr assertions

[shiny-code-bot]

Finish Line

Expected client disconnects in the Python SDK app-server test harness do not print BrokenPipeError or connection-reset tracebacks to process stderr, and the full Python SDK test job passes reliably without weakening unrelated output assertions.

Current Status

State: completed and merged to main on August 23, 2026.

The repair landed through stacked PR #791 and default-branch PR #789. Final merge commit: 9e5acadfbd984332b09494495b5e647a26257018.

Implemented behavior:

• _ResponsesHttpServer.handle_error suppresses only expected BrokenPipeError and ConnectionResetError client disconnects.
• Every other exception delegates to the standard server error path and remains visible.
• Focused coverage proves both silent disconnect cases and unexpected-error delegation.
• No production SDK behavior, formatter assertion, workflow retry policy, or CI threshold was weakened.

Evidence:

• Focused harness tests: 3 passed.
• Full local Python SDK suite: 133 passed, 38 skipped.
• Automatic Validation reached terminal skipped with skip_reason=no_applicable_provider for the two Python test files.
• Background Review completed with no findings, overall_correctness=patch is correct, confidence 0.96.
• PR Silence expected Python SDK harness disconnects #791's stacked required CI passed, including Python SDK and repo checks.
• Combined PR Fix deterministic convergence report test clock #789 required CI passed after the stack was integrated.
• Fresh changed-files IntelliJ closeouts were GREEN for both the SDK repair and the exact combined Fix deterministic convergence report test clock #789 head.

#790's finish line is satisfied. #634 is no longer blocked by this issue; PR #789 is merged, and PR #788 can now update from repaired main before resuming #787.

Blocked by: none.

Last verified: August 23, 2026.

[shiny-code-bot]

Implementation handoff — PR #791

State: bounded repair implemented, committed, pushed, and opened as #791.

• Head: 837f70eb0a133d749d40e89c7d9ba6c9958524ae (test(sdk): silence expected harness disconnects)
• Branch: code/issue-790-sdk-broken-pipe
• Diff: 43 insertions across sdk/python/tests/app_server_harness.py and new focused sdk/python/tests/test_app_server_harness.py; no production SDK, formatter assertion, PR Fix deterministic convergence report test clock #789, workflow, or retry-policy changes
• Behavior: _ResponsesHttpServer.handle_error suppresses only BrokenPipeError and ConnectionResetError from sys.exc_info()[1]; every other exception delegates to super().handle_error

Local validation

From sdk/python:

• uv sync --group dev --frozen — passed
• uv run --frozen --no-sync pytest tests/test_app_server_harness.py — 3 passed
• uv run --frozen --no-sync ruff check --output-format=github tests/app_server_harness.py tests/test_app_server_harness.py — passed
• uv run --frozen --no-sync ruff format --check tests/app_server_harness.py tests/test_app_server_harness.py — 2 files already formatted
• uv run --frozen --no-sync pytest — 133 passed, 38 skipped
• Repository-required just fmt from codex-rs — passed with no unrelated tracked changes
• git diff --cached --check before commit — passed

Inspection and review evidence

• Initial changed-files IntelliJ IDEA 2026.2.1 closeout: RED/actionable_findings, 10 findings, lifecycle cleanup closed. The one finding mapped to the new hook (request: object) was fixed by matching the server socket request type. Eight warnings were on unchanged harness lines; the remaining new-file error was the prepared root IDE model not resolving pytest even though the checked-in SDK uv environment ran all tests successfully.
• Post-fix changed-files closeout: UNKNOWN/stale_results after the helper's single maintained internal retry; retry=false, cached findings withheld, lifecycle cleanup closed. Per the inspection contract, no further retry was attempted.
• Final bounded review found no breaking-surface, context, test-design, or change-size findings. The review skill's independent subagent mechanism was unavailable in this runtime.
• Automatic Validation: no separate automatic lifecycle evidence was observable in this session; all required manual gates are recorded above.
• Automatic Background Review: not yet observable for head 837f70eb0a133d749d40e89c7d9ba6c9958524ae at handoff time.

CI snapshot

At 2026-08-23T22:35:36Z, PR #791 had 10 check runs: 9 pending/in progress, 1 skipped (cla), and 0 failing. The target sdk / python-sdk job was in progress in workflow run 32671037844.

Next action: let PR #791 CI and review complete, then merge this repair before retrying PR #789. All original stop conditions remain preserved.

[shiny-code-bot]

Closure metadata reconciliation after PR #789 merged the completed #790 repair to main at 9e5acadfbd984332b09494495b5e647a26257018. The final Current Status contains complete validation, review, CI, and inspection evidence.