Bug: Cache fails on new version of OpenAI API

[devjeetr]

The code for persisting the response cache fails in two places:

1. TiCoder/src/main.py

   Line 403 in 3362f00

   json.dump(config.codex_query_response_log, f)

2. TiCoder/src/main.py

   Line 441 in 3362f00

   json.dump(config.codex_query_response_log, f)

This happens because the response object returned by OpenAI is a BaseModel from pydantic which isn't automatically JSON serializable.

It's unclear if a specific version of the OpenAI client library is expected beyond what's specified in requirements.txt. I've tested this with openai==1.78.1. But I believe this failure will occur for all recent versions of the openai client that return responses as pydantic models.

Possible Fix

1. Update response => response.model_dump() if it is a pydantic object in

   TiCoder/src/query_chat_model.py

   Line 370 in 3362f00

   v = (k, response, current_time)

2. Parse response into ChatCompletion in

   TiCoder/src/query_chat_model.py

   Line 318 in 3362f00

   resp = config.codex_query_response_log[str(k)][1]

   
   This is needed because call sites expect the response to be a structured object.

I have a PR implementing this that I will attach here.