Investigate and optimize persona class loading storage strategy

[dlqqq]

trafficstars

Summary

Currently, persona classes loaded from entry points are stored as an instance attribute (_persona_classes) in PersonaManager. We need to investigate whether entry points can change at runtime and optimize the storage strategy accordingly.

Background

In PR #1380, _persona_classes was converted from a class attribute to an instance attribute to support loading local personas from the .jupyter directory. The original assumption was that entry points are immutable at runtime, but @ellisonbg noted that entry points may change at runtime.

Investigation Needed

1. Verify entry point behavior: Determine if entry points can actually change during server runtime

   • Test dynamic installation/uninstallation of packages with entry points
   • Check if importlib_metadata.entry_points() reflects runtime changes
   • Document the expected behavior

2. Optimize storage strategy based on findings:

   • If entry points are truly immutable at runtime: Store them in a class attribute for better performance
   • If entry points can change: Keep current approach or implement a hybrid solution with cache invalidation

Proposed Investigation

1. Create test scenarios for dynamic entry point changes
2. Research importlib_metadata behavior with runtime package changes
3. Based on findings, implement the optimal storage strategy