signal-bot/signal_bot/pack_command.py
Krzysztof kuhy Rudnicki 405541be9b
Some checks are pending
CI / check (push) Waiting to run
CI / web (push) Waiting to run
release: 0.2.6 «Stephanus I» — dodano !analityka
One row per command attempt, not two aggregates written only on success.

command_uses holds last-use only and daily_uses counts successes; both are
rate-limiter state and both are written by store.record_use, which the
dispatcher calls only when a handler returns cleanly. A gate refusal and a
crash returned before it, so a command that failed every time looked exactly
like one nobody ran, and nothing knew which command a counted message was.

command_events now takes every attempt with its outcome, its duration and
the item it served; upstream_calls takes every outbound request made while
serving one. The two writes stay separate on purpose -- recording a refusal
as a use is what slid the cooldown window forward every time somebody was
told to wait -- and test_dispatcher_outcomes asserts last_used does not move.

Backfilled from journald joined against message_events, with daily_uses
filling the residual so the two sources cannot double-count. Two things were
wrong on the first attempt and both were found by measuring rather than by
reasoning: the window has to be symmetric, because sent_at_ms comes from the
sender's phone and the journal stamp from this host (194 of 299 lines landed
*before* their own message, by up to 3.2 s); and matching has to be a global
closest-pair assignment, because per-row greedy cannot separate two commands
seconds apart. Together those took the dropped rows from 137 to zero, and
recovered 418 invocations back to 29 August.

Also: !analityka in the group, twelve Polish lines, naming at most three
people and only ones whose display name has been seen; an Analytics tab with
Nivo charts; and continuous latency measurement, which answers the open
question in DOCS-latency.md -- a text send is 235 ms and an image send about
2.2 s, so the upload really is where the time goes.

EPHEMERAL_TABLES was doing two unrelated jobs and is now two tuples: what a
schema bump rebuilds, and what leaving a group clears. daily_uses left the
first (it is a backfill source now) and stayed in the second, and
_drop_reshaped took over the repair a bump used to do for it.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T4QTp3h5Q3MvfrwGV8xqPf
2026-08-31 23:19:51 +02:00

137 lines
5.3 KiB
Python

# Copyright (c) 2026 Krzysztof Rudnicki. MIT licence.
"""The photo commands a private pack defines, and how one message is answered.
One command per entry, so the cooldown, the log line and the rate-limit row
are per person -- and every one of them is named by the entry's pseudonym
(``pack:p1``), never by what was typed. Reading the journal or the database
tells you somebody asked for a photo, and nothing about whose.
Several names in one message mean "a photo with all of them": the trigger
takes the rest of the line as further names, each resolved through the same
table the trigger came from, and the sets are intersected. Where an entry
offers a choice, the choice is made among the options that actually have a
photo -- picking blindly and then reporting failure would be a "no" the
album could have answered.
"""
from __future__ import annotations
import base64
import secrets
from typing import TYPE_CHECKING, Final
from signal_bot.analytics import probe
from signal_bot.pack import candidate_keysets, media_type
if TYPE_CHECKING:
from collections.abc import Sequence
from signal_bot.context import CommandContext
from signal_bot.decorators import Decorator
from signal_bot.pack import Entry
from signal_bot.pack_holder import PackHolder
from signal_bot.registry import CommandRegistry, Handler
# The prefix every pack command's name carries, so one glance at a log line
# or a command_uses row says where it came from without saying who.
NAME_PREFIX: Final = "pack:"
NO_PHOTO: Final = "BEEP BOOP NIE MAM TAKIEGO ZDJĘCIA."
# What an argument resolver yields for a word it cannot turn into a key. It
# can never collide with a real one: pack_load requires person keys to be
# ascii alphanumeric, and a custom key is "t" followed by digits.
UNKNOWN: Final = "?"
# Deliberately says nothing about who the entry is: !pomoc prints this next
# to the trigger, and the trigger is the only place a name may appear.
ONE_PERSON: Final = "losowe zdjęcie tej osoby"
EVERYBODY: Final = "losowe zdjęcie, na którym są oni wszyscy"
EITHER: Final = "losowe zdjęcie jednego z nich"
def description_for(entry: Entry) -> str:
"""What !pomoc says an entry does, without naming anybody."""
if len(entry.options) > 1:
return EITHER
return ONE_PERSON if len(entry.options[0]) == 1 else EVERYBODY
def pick(items: Sequence[str]) -> str:
"""One item, uniformly at random.
secrets rather than random for the reason animals.paged_url gives: the
choice is not security-sensitive, but ruff's ALL rule set flags `random`
and this repo forbids the noqa that would silence it.
"""
return items[secrets.randbelow(len(items))]
def pack_handler(holder: PackHolder, entry: Entry) -> Handler:
"""Build the handler that answers one entry's trigger.
Reads ``holder.pack`` at call time rather than closing over the album it
was built with: a photo !add_photo stored a second ago has to be visible
to this person's own command, and a restart is not an acceptable price.
"""
async def handler(ctx: CommandContext) -> None:
pack = holder.pack
wanted = candidate_keysets(pack, entry, ctx.args)
viable = [keys for keys in wanted or () if pack.matching(keys)]
if not viable:
# One line and no photo, for a name nobody has, a combination
# nobody photographed, and a tail of six names alike: from the
# group's side those are the same question with the same answer.
await ctx.reply(NO_PHOTO)
return
filename = pick(pack.matching(pick_keys(viable)))
# The stored name is content-addressed -- pack_store.content_name is
# the first 128 bits of the file's SHA-256 -- so it identifies the
# photograph without describing it or anyone in it. That is what
# makes "which photo keeps coming back" answerable at all.
probe.note_item(filename)
payload = base64.b64encode(pack.media_path(filename).read_bytes()).decode("ascii")
await ctx.reply_media(payload, media_type(filename), filename)
return handler
def pick_keys(viable: Sequence[frozenset[str]]) -> frozenset[str]:
"""One of the readings that has photos, chosen at random."""
return viable[secrets.randbelow(len(viable))]
def register_pack_commands(
registry: CommandRegistry,
holder: PackHolder,
*,
decorate: Decorator | None = None,
) -> None:
"""Register one command per pack entry, scoped to the pack's groups.
The entries and the groups are read once, here, and never again: a swap
only ever replaces Pack *data*. If it could add a command it could also
add one to the wrong group, which is the failure this album exists to
make impossible.
Args:
registry: Registry to populate.
holder: The album, in the mutable box every handler reads through.
decorate: Optional callable applied to each handler, used to attach
the cooldown gate without hard-wiring policy into the commands.
"""
pack = holder.pack
for entry in pack.entries:
handler = pack_handler(holder, entry)
if decorate is not None:
handler = decorate(handler)
registry.register(
f"{NAME_PREFIX}{entry.key}",
entry.triggers,
handler,
description_for(entry),
groups=pack.groups,
takes_args=True,
)