Add a performance comparison of two plugin versions (1.5.2 vs. current)

[codeling]

Summary

Adds tests/performance/, a local (not CI) harness comparing what the plugin costs a Joomla site in two versions - by default tag 1.5.2 vs. HEAD - against a site without bfstop.

• setup-sites.sh installs three identical Joomla 5.4.8 sites (none / baseline / current), each plugin from the release zip its own deploy.sh zip builds, with default settings.
• bench.php serves them with PHP's built-in server (OPcache on) and measures, via a small auto_prepend_file (prepend.php):
  • installed footprint (zip, files, PHP lines, empty tables)
  • per-request server time, CPU, peak memory, PHP files, SQL queries for page views, failed logins, an address getting blocked, and blocked requests - repeated with differently sized tables
  • throughput / latency percentiles under concurrent load
  • database growth per failed login
  • the queries a request sends to bfstop's tables (MariaDB general log)
• Writes report.md + results.json + raw per-request metrics.
• run.sh does both steps; README.md documents requirements, options and caveats.

Notes

• Needs MariaDB (1.5.2 only has MySQL SQL scripts; data is seeded with the sequence engine).
• Requests carry a browser User-Agent and use a new user name per failed login: without that the current version's risk score / account throttle add deliberate multi-second sleeps that would swamp the measurement (documented in the README).
• The old plugin is run on Joomla 5 (it doesn't support 6).

Test plan

• --quick runs end to end (sequential, throughput, growth, query trace) and the report renders
• setup-sites.sh verified from scratch and re-run on existing sites
• php -l on the new files
• Full-size run (default settings) - results to be posted separately

🤖 Generated with Claude Code

https://claude.ai/code/session_01PtohRckRkeL2TVCfEhc7oJ

---

Generated by Claude Code