feat: add --rerun flag to rerun optimization from stored trace

[mohammedahmed18]

Summary

• Adds --rerun <trace_id> CLI flag that passes rerun_trace_id through all AI service client methods (optimize_code, optimize_python_code_line_profiler, generate_regression_tests, optimize_code_refinement, code_repair)
• Companion to codeflash-ai/codeflash-internal#TBD which adds server-side handling

Test plan

• Run codeflash --rerun <existing_trace_id> --file <path> --function <name> --server local against local aiservice with the companion PR deployed
• Verify AI service returns stored results instantly (no LLM latency)
• Verify full optimization pipeline still completes (discovery, test execution, benchmarking)

🤖 Generated with Claude Code

[claude]

Claude finished @mohammedahmed18's task in 3m 12s —— View job

---

PR Review Summary

• Triage PR size → SMALL (31 additions, 3 deletions across 4 files)
• Run lint/typecheck (prek + mypy)
• Resolve stale review threads → none found
• Code review
• Duplicate detection
• Check codeflash-ai[bot] optimization PRs

---

Prek Checks

✅ prek passed — no lint/format issues.

mypy reports 148 pre-existing errors in function_optimizer.py and one other file; none are introduced by this PR.

---

Code Review

Classification: SMALL — clean parameter threading, no complex logic.

Findings: No bugs, no security issues, no breaking API changes.

Observations:

1. Consistent threading — rerun_trace_id is correctly passed through all five AI service call sites:

   • optimize_code → generate_optimizations (function_optimizer.py:1867)
   • optimize_python_code_line_profiler → determine_best_candidate (function_optimizer.py:1361)
   • generate_regression_tests → verifier.py:74
   • optimize_code_refinement → process_single_candidate (function_optimizer.py:1295)
   • code_repair → repair_optimization (function_optimizer.py:1491)

2. Property design (function_optimizer.py:880) — using getattr(self.args, "rerun", None) if self.args else None safely handles both the args=None and attribute-missing cases. Clean approach.

3. No tests — acceptable here since correctness depends on a companion server-side PR (codeflash-internal) and there's no mock surface to test against locally. The test plan correctly calls for integration testing against local aiservice.

Duplicate Detection

No duplicates detected. The rerun_trace_id parameter is a new cross-cutting concern applied consistently.

Test Coverage

Skipped (SMALL PR).

---

codeflash-ai[bot] Optimization PRs

PRs #1895, #1891, and #1890 all have CI failures (unit-tests, JS optimization tests, tracer-replay, java-tracer-e2e). The same checks fail identically across all three PRs targeting different base commits, indicating pre-existing failures on main. Left comments on each; keeping open until base branch CI is fixed.

---

Last updated: 2026-03-27T16:20Z