Problem

The default_logger() function reads the global config and constructs a new Logger value on every call to birch.info, birch.error, etc.:

fn default_logger() -> Logger {
  let cfg = get_config()
  logger.new("app")
  |> logger.with_level(cfg.level)
  |> logger.with_handlers(cfg.handlers)
  |> logger.with_context(cfg.context)
}

This means every log statement allocates a new Logger, reads from persistent storage, and chains multiple builder calls — even if the config hasn't changed.

Impact

For applications with sparse logging (like CLIs), this is negligible. But for hot paths or high-throughput services, this per-call overhead adds up unnecessarily.

Proposal

Cache the default logger and invalidate it only when birch.configure() is called:

• Store the built logger alongside the config in global state
• On configure(), rebuild and cache the logger
• On default_logger(), return the cached logger directly

This is a pure internal optimization — no API changes needed.

Alternatives

• Use a generation counter: bump on configure(), rebuild logger only when the counter changes
• Accept the current behavior and document it as a tradeoff of the convenience API (users needing performance should use named loggers directly)