fix: emit stream finalization errors

[HAYDEN-OAI]

Summary

• route errors thrown while emitting final stream events through the normal stream error handler
• cover malformed OpenAI-compatible streams that finish without an assistant message

Root cause

EventStream._run() handled executor rejections, but _emitFinal() ran inside a fulfilled .then() callback whose returned promise was ignored. If finalization threw, such as when no assistant message was produced, the rejection escaped instead of emitting the stream's error event.

Impact

Consumers using .on('error', ...) now receive finalization failures consistently, and awaiting done() rejects with the same error rather than leaving an unhandled rejection that can crash the process.

Validation

• pnpm run format
• pnpm exec tsc --noEmit
• pnpm exec jest tests/lib/ChatCompletionStream.test.ts --runInBand
• git diff --check

Resolves #553

[openai-sdks]

OkTest Summary

✅ 237/237 SDK tests passed in 10.695s for Node SDK PR #1972.

Test results — 42 files

Test	Result	Time
tests/chat-completions-complex-body.test.ts	✅ Passed	198ms
tests/chat-completions-create.test.ts	✅ Passed	254ms
tests/chat-completions-stream.test.ts	✅ Passed	230ms
tests/files-content-binary.test.ts	✅ Passed	168ms
tests/files-create-multipart.test.ts	✅ Passed	163ms
tests/files-list-pagination.test.ts	✅ Passed	202ms
tests/initialize-config.test.ts	✅ Passed	190ms
tests/instance-isolation.test.ts	✅ Passed	103ms
tests/models-list.test.ts	✅ Passed	133ms
tests/responses-background-lifecycle.test.ts	✅ Passed	232ms
tests/responses-body-method-errors.test.ts	✅ Passed	324ms
tests/responses-cancel-timeout.test.ts	✅ Passed	216ms
tests/responses-cancel.test.ts	✅ Passed	185ms
tests/responses-compact-retries.test.ts	✅ Passed	263ms
tests/responses-compact.test.ts	✅ Passed	289ms
tests/responses-create-advanced-stream.test.ts	✅ Passed	116ms
tests/responses-create-advanced.test.ts	✅ Passed	169ms
tests/responses-create-disconnect.test.ts	✅ Passed	1.13s
tests/responses-create-errors.test.ts	✅ Passed	230ms
tests/responses-create-malformed-api-responses.test.ts	✅ Passed	191ms
tests/responses-create-retries.test.ts	✅ Passed	311ms
tests/responses-create-stream-failures.test.ts	✅ Passed	234ms
tests/responses-create-stream-timeout.test.ts	✅ Passed	2.177s
tests/responses-create-stream-wire.test.ts	✅ Passed	2.621s
tests/responses-create-stream.test.ts	✅ Passed	80ms
tests/responses-create-terminal-states.test.ts	✅ Passed	294ms
tests/responses-create-timeout.test.ts	✅ Passed	202ms
tests/responses-create.test.ts	✅ Passed	147ms
tests/responses-delete.test.ts	✅ Passed	212ms
tests/responses-input-items-errors.test.ts	✅ Passed	207ms
tests/responses-input-items-list.test.ts	✅ Passed	168ms
tests/responses-input-items-options.test.ts	✅ Passed	215ms
tests/responses-input-tokens-count-timeout.test.ts	✅ Passed	234ms
tests/responses-input-tokens-count.test.ts	✅ Passed	241ms
tests/responses-malformed-inputs.test.ts	✅ Passed	2.293s
tests/responses-not-found-errors.test.ts	✅ Passed	248ms
tests/responses-parse.test.ts	✅ Passed	161ms
tests/responses-retrieve-retries.test.ts	✅ Passed	256ms
tests/responses-retrieve.test.ts	✅ Passed	241ms
tests/responses-stored-method-errors.test.ts	✅ Passed	600ms
tests/retry-behavior.test.ts	✅ Passed	3.011s
tests/sdk-error-shape.test.ts	✅ Passed	292ms

View OkTest run #28817115989

_{SDK merge (1a9d8e41924f) · head (1e9094e8c3d6) · base (95b54e589491) · OkTest (7fc645091b1a)}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 13a871bdba

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[jbeckwith-oai]

Comprehensive review completed; no blocking or nonblocking findings.

Reproduction: I applied this PR’s regression test to base commit 95b54e58 and ran jest tests/lib/ChatCompletionStream.test.ts --runInBand -t "emits finalization failures as errors". The original behavior reproduced exactly: _emitFinal() threw OpenAIError: stream ended without producing a ChatCompletionMessage with role=assistant, the error did not reach the stream error path, done() remained pending, and the test failed after the 30s timeout.

Validation on head 1e9094e8:

• jest tests/lib/ChatCompletionStream.test.ts --runInBand: 6/6 passed, including the malformed user-role-only stream regression.
• jest tests/lib --runInBand: 215/215 passed across 15 suites.
• Direct EventStream probes confirmed synchronous executor throws, asynchronous executor rejections, and _emitFinal() throws each emit exactly one error, then end, and reject done() with the matching message; the success path still emits end and resolves done().
• A throwing end listener still surfaces independently while done() remains resolved and errored remains false, so follow-up commit 1e9094e8 correctly preserves pre-existing end-listener behavior and fixes the concern raised on the first revision.
• CI is green across Node 20/22/24/26, build, lint, examples, ecosystem, CodeQL, Agents SDK regression, and detect-breaking-changes; OkTest reports 237/237 SDK tests passed.

Compatibility: the patch changes no public API, exports, types, or data shapes. The shared EventStream change is behaviorally scoped to routing executor/finalization failures through the already documented error lifecycle. The extra microtask remains inside the existing deferred setTimeout startup and does not reorder observable stream events. Existing error normalization/cause preservation is retained. The only newly handled cases are the reported finalization rejection and synchronous executor throws; end listener exceptions are intentionally outside the handler. This is an appropriate bug fix with low breaking-change risk.