fix: prevent empty tool names from overwriting valid names in SSE deltas

[timvw]

Summary

Adds defensive handling and additional test coverage for empty tool call function names in SSE streaming, complementing the root cause fix in PR #7594.

Context

PR #7594 already fixed the root cause by using get_or_insert_with() to prevent overwriting tool names. This PR builds on that fix by adding:

1. Defensive layer at emission point with debug logging
2. Additional test coverage for edge cases

Issues Addressed

• invalid_request_error: Invalid 'messages[XX].tool_calls[0].function.name': empty string. Expected a string with minimum length 1, but got an empty string instead. #7628 - Empty function name error with Azure OpenAI
• Empty function name cause Azure to reject the request #7094 - Azure rejecting requests due to empty function names
• Tool call streaming broken in 0.64.0 with local LLM providers #7517 - Tool call streaming broken with local LLM providers (LM Studio)
• Regression: Tool Name Lost in Streaming Chat Completions After PR #6996 #7579 - Regression: Tool names lost in streaming

Changes in This PR

1. Enhanced Documentation (lines 164-166)

Adds comment explaining the root cause and linking to issue #7579:

// Only update name if non-empty to prevent subsequent deltas
// from overwriting the initial name with empty strings.
// See: https://github.com/openai/codex/issues/7579

2. Defensive Layer at Emission Point (lines 228-251)

Even though the root cause is fixed, this adds defensive handling to skip any tool calls that somehow end up with empty/missing names, with detailed debug logging for troubleshooting:

match state.name {
    Some(name) if !name.is_empty() => { /* emit tool call */ }
    Some(name) if name.is_empty() => {
        debug!("Skipping tool call with empty name: call_id={}, arguments={}", ...);
    }
    None => {
        debug!("Skipping tool call with missing name: call_id={}, arguments={}", ...);
    }
}

This helps diagnose issues by logging:

• call_id of problematic tool calls
• arguments that were received
• Whether name was empty string vs completely missing

3. Additional Test Coverage

Adds skips_tool_calls_with_empty_or_missing_names test to verify the defensive layer works correctly when tool calls have empty or missing names.

Test Results

All tests passing (7/7):

• ✅ emits_multiple_tool_calls
• ✅ concatenates_tool_call_arguments_across_deltas
• ✅ emits_tool_calls_even_when_content_and_reasoning_present
• ✅ drops_partial_tool_calls_on_stop_finish_reason
• ✅ preserves_tool_call_name_when_empty_deltas_arrive (from PR fix: sse for chat #7594)
• ✅ preserves_name_when_subsequent_deltas_have_empty_names (this PR)
• ✅ skips_tool_calls_with_empty_or_missing_names (this PR)

Value Add

While PR #7594 fixed the root cause, this PR provides:

1. Defense in depth - Additional safety layer if the root cause fix is bypassed
2. Better observability - Debug logging helps diagnose streaming issues in production
3. Comprehensive testing - Edge case coverage for empty/missing names
4. Documentation - Links to GitHub issues explaining the problem

Testing

cargo test --package codex-api --lib sse::chat::tests

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[timvw]

I have read the CLA Document and I hereby sign the CLA

[etraut-openai]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. You're on a roll.

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".

[etraut-openai]

Thanks for the PR, but I think we'll pass on this one.