Give the date code one clock a test can freeze

[mircealungu]

Salvaged from a stale Claude worktree (.claude/worktrees/eloquent-lovelace-9b32b4, Aug 29). The rest of that worktree was superseded by master; this part never landed.

The problem

Every "is this today?" question in util/time.py read the wall clock directly, in two separate places:

def user_local_today(user):
    return datetime.now(user_zone(user)).date()

def server_now():
    return datetime.now(SERVER_TZ).replace(tzinfo=None)

The streak code's answers depend on the relationship between the server's date and the user's — that's the whole subject of the comment block at the top of the file, and of issue #587. But a test that builds its fixtures from the machine's own clock can only exercise whichever relationship happens to hold at the moment it runs. So the suite is green all afternoon and fails after local midnight, for reasons unrelated to the change under test. f162c541 ("Make the streak tests and the streak code agree on which day it is") went in for exactly that, and fixed the fixtures without removing the dependency on when you run them.

The change

Both functions now derive from a single _now_utc(), so patching that one function moves them together and a test can name the instant it means:

with clock_at("2026-03-01 12:00"):   # UTC
    ...

zeeguu/core/test/frozen_clock.py holds clock_at plus server_time_at, which gives the naive SERVER_TZ value the DB would hold for a given UTC instant.

The added friend-streak test runs one scenario at two pinned instants. Auckland is UTC+13 on that date, so at 06:00 UTC it is still the 1st there and at 12:00 UTC it is already the 2nd, while the server is on the 1st either way — the streak has to come out the same. Verified the seam actually bites:

2026-03-01 06:00 -> server_now: 2026-03-01 06:00:00 | auckland today: 2026-03-01
2026-03-01 12:00 -> server_now: 2026-03-01 12:00:00 | auckland today: 2026-03-02

Not included

The same worktree also reworked compute_current_streak to expire a streak only when it is stale in both friends' local calendars. Deliberately dropped: friend_streak_last_updated is written with server_now() and update_streak's continuation check (last_updated_date == server_yesterday) is server-frame, so reading it in two user calendars would let the UI show a streak that the increment path then resets to 1. Master's single-frame treatment is the correct one.

Behaviour

None changed — datetime.now(TZ) and datetime.now(utc).astimezone(TZ) are the same reading. get_server_time_utc() is left alone; it documents itself as unused.

zeeguu/core/test: 368 passed, 12 subtests passed.

🤖 Generated with Claude Code

[github-actions]

ArchLens - No architecturally relevant changes to the existing views