CodexBar Issue #139 Main Fix Validation (Post-Fix vs Pre-Fix)

Date: 2026-02-19 Workspace: /Users/michalkrsik/windsurf_project_folder/CodexBar Branch: codex/perf-issue-139

Reference pre-fix report:

• /Users/michalkrsik/windsurf_project_folder/CodexBar/docs/perf-energy-issue-139-simulation-report-2026-02-19.md

Implemented Main Fix

File changed:

• /Users/michalkrsik/windsurf_project_folder/CodexBar/Sources/CodexBarCore/Providers/Codex/CodexStatusProbe.swift

Behavior change:

• Primary Codex PTY probe timeout reduced from 18s to 8s.
• Retry policy changed from retry on parseFailed OR timedOut to retry only on parseFailed.
• Parse retry timeout set to 4s.
• Timed-out runs now fail fast and wait for next scheduled refresh.

Post-Fix Validation Method

Target: main culprit path (Codex CLI failure path).

Practical simulation used:

• CODEX_CLI_PATH pointed to a fake codex script.
• Script behavior:
  • exits immediately for app-server args (forces RPC failure/fallback path),
  • otherwise busy-loops with no /status output (simulates heavy stuck CLI PTY behavior).
• Command run (3 times):
  • ./.build/debug/CodexBarCLI usage --provider codex --source cli --format json --pretty
• Collected:
  • wall time (/usr/bin/time -p),
  • sampled child CPU every 0.5s,
  • leftover child-process count after run.

Artifacts:

• /tmp/codexbar_main_fix_validation_after

Post-Fix Results (3 runs)

Run	Real (s)	Avg child CPU (%)	Max child CPU (%)	Remaining child procs
1	12.76	88.32	100.00	0
2	12.67	89.79	100.00	0
3	12.59	89.90	100.00	0
Mean	12.67	89.34	100.00	0

Side-by-Side Comparison Against Stored Pre-Fix Report

Pre-fix values are from the stored report's Culprit A simulation summary. Post-fix values are from the validation above.

Metric	Pre-fix (stored report)	Post-fix (this validation)	Delta
Failed-run duration (worst-case path)	42.00s (code-path budget before fix)	12.67s (measured mean)	-69.8%
Child CPU during failed run	113.32% avg	89.34% avg	-21.2%
Peak child CPU during failed run	115.90% max	100.00% max	-13.7%
Remaining child processes after failure	not captured in pre-fix report	0	improved

Derived CPU-time exposure index (avg CPU * duration):

• Pre-fix: 113.32 * 42.00 = 4759.44
• Post-fix: 89.34 * 12.67 = 1132.94
• Reduction: -76.2%

Conclusion

The implemented main fix materially reduces the failure-path runtime and overall CPU exposure. The heavy CLI process can still spike CPU while active, but it now lives for a much shorter window and is cleaned up after failure.