Fix ContextWindowExceededError after 3 retries in react loop

[Copilot]

📝 Changes Description

_call_with_potential_trajectory_truncation was implicitly returning None after exhausting 3 retry attempts, causing AttributeError: 'NoneType' object has no attribute 'next_thought' when accessing the prediction.

Fix:

• Raise explicit ValueError when retries are exhausted
• The ValueError from truncate_trajectory (when trajectory cannot be truncated further) naturally propagates up
• Both errors are caught by forward()/aforward() which log and break gracefully, allowing the extract phase to proceed with the collected trajectory

# Before: would crash with AttributeError on None.next_thought
react = dspy.ReAct("question -> answer", tools=[...])
result = react(question="...")  # with very large context

# After: gracefully ends trajectory and proceeds to extraction

Changes:

• dspy/predict/react.py: Added explicit raise ValueError at end of retry loops in both sync and async versions
• tests/predict/test_react.py: Added combined test for sync and async context window exhaustion scenarios

✅ Contributor Checklist

• Pre-Commit checks are passing (locally and remotely)
• Title of your PR / MR corresponds to the required format
• Commit message follows required format {label}(dspy): {message}

⚠️ Warnings

None.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>[Bug] ContextWindowExceededError after 3 retries in react loop</issue_title>
<issue_description>### What happened?

Context

We get this error:NoneType' object has no attribute 'next_thought' when we work with observations/trajectories that continue to exceed the context window limit

• in the forward pass, we run _call_with_potential_trajectory_truncation() (code)

• this has a hardcoded limit of 3 retries.

• once that is hit, we return None (code) which causes attribute error over NoneType (code)

Question

1. What is the expected behavior? Doesn't a while loop until success make more sense?
2. On our end, we can monkey patch the truncation logic or even add our own logic for long trajectories but I think this is a bug

Steps to reproduce

explained above

DSPy version

3.0.3</issue_description>

<agent_instructions>Let's handle failure case (None) gracefully</agent_instructions>

Comments on the Issue (you are @copilot in this section)

• Fixes [Bug] ContextWindowExceededError after 3 retries in react loop #9103

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[TomeHirata]

@copilot We don't need this try-catch

[Copilot]

Removed the try-catch blocks in both sync and async methods. The ValueError from truncate_trajectory now naturally propagates up and is caught by forward(). Fixed in 2dd54fc.

[TomeHirata]

@copilot we don't need docstring in tests

[Copilot]

Removed the docstrings from both test functions. Fixed in fa8a09b.

[TomeHirata]

@copilot can we combine two tests by parameterizing?

[Copilot]

Combined the sync and async tests into a single test function. Fixed in e95c2dd.

[ianyu93]

Oh wow this is kinda cool