fix: handle null response.output in parse_response

[gavin913-lss]

What breaks

parse_response() in src/openai/lib/_parsing/_responses.py crashes with TypeError when response.output is None.

The chatgpt.com Codex backend (https://chatgpt.com/backend-api/codex) sometimes sends response.output: null in the consolidated response.completed event — even when valid response.output_item.done events were streamed earlier. The SDK then raises TypeError: 'NoneType' object is not iterable inside parse_response().

Fixes #3325

Root cause

Line 61 does for output in response.output: without checking for None. When the Codex backend sends response.output: null, iterating over None crashes.

Fix

Add a fallback to empty list when response.output is None:

for output in response.output or []:

This is safe because:

• If response.output is a valid list, it iterates normally
• If response.output is None, it iterates over nothing (no output items to parse)
• The earlier streamed events already populated the output items, so the final consolidated event having null output is a no-op

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: dadb254510

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Preserve streamed output when final payload omits it

When the streaming response.completed event has event.response.output == None (the scenario described by this fix), this fallback makes parse_response() build a final ParsedResponse with output=[]. In the streaming path, ResponseStreamState.accumulate_event() calls parse_response(response=event.response) for response.completed instead of parsing the accumulated snapshot, so get_final_response() and the emitted completed event lose all output items/text/tool calls that arrived via earlier response.output_item.added and delta events. This avoids the crash but turns valid streamed responses into empty final responses; the fallback should preserve the accumulated snapshot or otherwise not replace missing final output with an empty list.

Useful? React with 👍 / 👎.