langchain - 💡(How to fix) Fix bug: LLM response .choices[0] accessed without empty-list guard

Description

Several files in the langchain codebase access response.choices[0] or response.choices[0].message without first checking whether the choices list is empty. When an LLM provider returns an empty choices list (content filtering, quota, streaming edge cases), this produces an IndexError instead of a meaningful error.

Affected patterns

# Bad — crashes with IndexError if choices is empty
text = response.choices[0].text
content = response.choices[0].message.content

# Good
if not response.choices:
    raise ValueError("LLM returned empty response")
text = response.choices[0].text

Reproducible

Any LLM call that triggers content filtering (e.g., OpenAI content_filter finish_reason) will return choices=[]. Azure OpenAI and Gemini 2.5 Flash also return empty choices on some safety events.

Fix

Add an empty-choices guard before each unguarded access. A PR implementing this fix is available at #37516 — requesting maintainer review.