Exercise 3: debug=False Genuinely Skips the Reload Check — Possible Solution
====================================================================

THE GUARDED CHECK
------------------------------
    def make_maybe_reloading_app(debug):
        def app(environ, start_response):
            if debug and watcher.changed():
                cache.clear()
            if 'greeting.html' not in cache:
                with open(tpl_path) as f:
                    cache['greeting.html'] = f.read()
            body = cache['greeting.html'].encode('utf-8')
            start_response('200 OK', [('Content-type', 'text/plain'),
                                       ('Content-Length', str(len(body)))])
            return [body]
        return app

VERIFIED, REAL RESULT (debug=False)
------------------------------
    request 1:                                 b'v1 content'
    # the real template file is edited on disk while the server runs:
    #   'v2 content, edited while running'
    request 2, after edit, debug=False:        b'v1 content'   -- still stale

The debug=False app never once calls watcher.changed() - the `debug
and` short-circuit means the real filesystem stat() call inside
_safe_mtime() is never made at all when debug is off, exactly
reproducing Chapter 4's own original stale-cache finding, deliberately.

WHY SKIPPING THE stat() CALL IN PRODUCTION IS GENUINELY WORTH IT
------------------------------
watcher.changed() calls os.stat() on every watched path, on every
single incoming request. A single stat() call is fast, but a real,
busy production server can handle thousands of requests per second -
multiplying a small, real filesystem-syscall cost by every request,
forever, for a feature (auto-detecting a live code edit) that a real
production deployment has no legitimate reason to need, since nobody
is expected to be hand-editing template files on a live production
server in the first place. Gating the check behind the same debug
flag that already controls the traceback page keeps the real
per-request overhead at genuinely zero in the one mode where it would
otherwise run forever for no real benefit.

WHY THIS WORKS AS AN ANSWER
------------------------------
It verifies both real Apps side by side against the identical file
edit, confirming the debug=True case (already shown working in the
main chapter) and the debug=False case (deliberately, verifiably
stale) are genuinely different real behaviors controlled by one flag,
and grounds the "why skip it" reasoning in a concrete, real cost (a
per-request syscall) rather than an abstract performance concern.