diet-guard/diet_guard/_cli_gate.py
Krzysztof kuhy Rudnicki b53ac92c59 perf(gate): narrow the pre-lock pull and defer the heavy imports
The gate's pre-lock refresh ran a *full* sync tick -- peer logs, budget,
both food banks, a rebuild and a push -- in front of the fullscreen lock.
Measured against the live remote: ~18-27s, of which the curated-bank loop
alone was 17s spent fetching nothing (absent on Firebase for every peer, so
each of ~21 peers fell through to a GitHub round trip returning `{}`).

It was also failing: five `sync unavailable (network error trying to refresh
the session)` entries in the last five days, one of them a meal that logged
locally but never published.

The gate has one question -- "has a peer already logged this slot?" -- and
only the peer logs can answer it. `_sync_refresh` does that and nothing
else: one bulk revision-map read, a fetch of only the peers whose revision
moved, merge, re-sign, write. No budget, no banks, no push. The full tick
still runs, after the window closes, so nothing is permanently skipped --
the exhaustive pass just moves off the path the user waits on.

Three state invariants have their own regression tests, because each is
silent when broken: `pushed_rev` survives verbatim (this pass never pushes,
and overwriting it would make the device stop publishing forever),
`peer_revs` is merged rather than replaced (only changed peers are visited),
and the log is written before the state.

Separately, `gate --check` was importing an HTTP stack and a GUI it never
uses. `requests` (~78ms) arrived via `_cli_sync` -> `crdt_sync` and via
`_resolve` -> `_estimator_off`; both are now resolved through module-level
`__getattr__` (PEP 562) and `import_module`, which keeps every existing
`patch.object` target and needs no lint suppression.

Measured: `gate --check` 216ms -> ~110ms; narrow pull 92ms against the live
remote. Accepted trade, documented in `_sync_refresh`: a budget edited on
the phone is not visible in the lock dashboard until the post-window tick.
It cannot affect the lock decision, which is why it is safe to defer.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LJ5YnA3CERitQ57HjMu2Qz
2026-08-21 12:58:44 +02:00

133 lines
6.0 KiB
Python

"""CLI handler for the ``gate`` subcommand.
Split out from :mod:`diet_guard._cli` to keep that module under the repo's
500-line cap (see ``CLAUDE.md``'s "feat: split oversized modules" history).
The gate's actual window logic already lives in ``_gatelock*.py``; this is
just the thin CLI glue, same as ``_cli_sync.py`` is for ``sync``.
"""
from __future__ import annotations
from importlib import import_module
import sys
from typing import TYPE_CHECKING
from diet_guard._gate import gate_is_due
from diet_guard._gatelock_lockfile import acquire_gate_lock, release_gate_lock
from diet_guard._gatelock_support import wait_for_display
from diet_guard._sync_events import publish_after_log
if TYPE_CHECKING:
from collections.abc import Callable
# Names resolved lazily below. Importing ``_gatelock`` eagerly costs ~120ms:
# it reaches ``_resolve`` -> ``_estimator_off`` -> ``requests`` (~78ms) on top
# of tkinter. The common gate tick is *not due* and opens no window, so it must
# not pay for a GUI and an HTTP stack it never touches.
#
# A module-level ``__getattr__`` (PEP 562) rather than a function-local import
# because seven tests do ``patch.object(_cli_gate, "MealGate", ...)`` and
# ``scripts/check_patch_targets.py`` resolves targets with ``hasattr`` -- both
# go through ``__getattr__`` and keep working, while a function-local import
# would make the attribute invisible to both.
_LAZY_ATTRS = {
"MealGate": ("diet_guard._gatelock", "MealGate"),
# ``_sync_refresh`` reaches ``crdt_sync`` -> ``requests``; the not-due
# tick never pulls, so it must not pay that import either.
"pull_peer_logs": ("diet_guard._sync_refresh", "pull_peer_logs"),
}
def __getattr__(name: str) -> object:
"""Resolve the deferred GUI import on first attribute access."""
target = _LAZY_ATTRS.get(name)
if target is None:
msg = f"module {__name__!r} has no attribute {name!r}"
raise AttributeError(msg)
module_name, attr = target
return getattr(import_module(module_name), attr)
def _should_lock(emit: Callable[[str], None]) -> bool:
"""Decide whether to lock, pulling from sync first if a lock looks due.
This wraps the pure, local :func:`~diet_guard._gate.gate_is_due` predicate
with one network step: the cheap local check runs first, and only when it
says a lock is due do we pull (via :func:`pull_peer_logs`, which fails
closed) and re-read the freshly written log. So a meal just logged on the
phone can clear the lock, while a not-due tick never touches the network
at all.
The pull is deliberately the *narrow* one -- peer logs only, no budget, no
food banks, no push. The full tick measured ~21s against the live remote,
which is not something to sit in front of a fullscreen lock; it still runs
in :func:`cmd_gate` once the window has closed. Only the peer logs can
answer "has a peer already logged this slot?", which is the sole question
this predicate asks.
"""
if not gate_is_due():
return False
# Through the module object so the lazy hook above resolves it, and so a
# test's ``patch.object`` still wins.
reason = sys.modules[__name__].pull_peer_logs()
if reason is not None:
emit(f"{reason}; using local log.")
return gate_is_due()
def cmd_gate(emit: Callable[[str], None], *, check: bool, demo: bool) -> int:
"""Run the log-to-unlock gate.
Three modes: ``--check`` is a headless decision (no window) whose exit code
a timer reads; ``--demo`` always shows a safe demo window; bare ``gate``
shows the real lock only when one is due. Both real modes route through
:func:`_should_lock`, which -- only when a lock is otherwise due -- pulls
the shared log first (via :func:`pull_peer_logs`) so a meal just logged on
the phone unlocks without a manual re-entry. A flock guard stops a second
window from stacking on top of the first, and a window-opening mode first
waits for the X display so a session-start launch never crashes unshown.
Args:
emit: A one-line output sink (``_cli._emit``, passed in rather than
imported -- see ``_cli_sync.cmd_sync`` for why).
check: Headless mode -- print and return an exit code, open no window.
demo: Use safe demo mode (local grab + close button) for the window.
Returns:
For ``--check``: 0 if not due, 1 if a lock is due. Otherwise 0.
"""
if check:
due = _should_lock(emit)
emit("due (a lock is warranted)" if due else "ok (no lock needed)")
return 1 if due else 0
if not demo and not _should_lock(emit):
emit("ok - no lock needed right now.")
return 0
handle = acquire_gate_lock()
if handle is None:
emit("the gate is already running.")
return 0
try:
# At session start the timer can fire before the X display/auth cookie
# is ready; wait it out so the window opens instead of crashing on a
# "couldn't connect to display" TclError (see _gatelock.wait_for_display).
if not wait_for_display():
emit("display not ready yet; will retry on the next timer tick.")
return 0
# Via the module object, not a bare name: ``MealGate`` is resolved by
# the lazy ``__getattr__`` above, and a test's ``patch.object`` sets a
# real module attribute that must win over it.
sys.modules[__name__].MealGate(demo_mode=demo).run()
finally:
release_gate_lock(handle)
# Publish only once ``run()`` has returned -- i.e. the mainloop has ended and
# the window is gone. Deliberately NOT inside the unlock handler: that runs
# on the Tk thread with the lock still up, so a slow or hanging network call
# would keep the user locked in behind a successful log. Here the worst case
# is a late publish. One call site covers every unlock reason, including the
# "Synced from another device" path, which hooking the log itself would miss.
reason = publish_after_log()
if reason is not None:
emit(f"logged locally, not yet published ({reason}).")
return 0