Python 3.12 deprecated utcnow(). Your timestamps were already wrong. The warning is new; the bug is old. Since 3.12, these emit a DeprecationWarning:
datetime.utcnow()
datetime.utcfromtimestamp(ts)
What was wrong with them
Both return a naive datetime: the clock reading in UTC, with tzinfo=None. The value doesn't say it's UTC. Two things follow:
>>> from datetime import datetime, timezone
>>> datetime.utcnow() < datetime.now(timezone.utc)
TypeError: can't compare offset-naive and offset-aware datetimes
And anything that later treats a naive datetime as local time shifts it by your server's UTC offset. That's the classic "the nightly job ran five hours late" bug, and it only shows up on servers that aren't set to UTC.
The replacements
from datetime import datetime, timezone
- now = datetime.utcnow()
+ now = datetime.now(timezone.utc)
- created = datetime.utcfromtimestamp(ts)
+ created = datetime.fromtimestamp(ts, timezone.utc)
Both return aware datetimes, with UTC attached. Comparisons, arithmetic and .isoformat() (which now ends in +00:00) all behave.
The part people miss: stored values
If your database or files already hold naive UTC values, mixing them with new aware ones raises the same TypeError. Pick one:
- Make old values aware when you read them:
value.replace(tzinfo=timezone.utc). - Or migrate the column to a timezone-aware type once, then switch the code.
And check anything that serialises timestamps: the +00:00 suffix can break a strict parser on the other end.
Make the next one loud
pytest -W error::DeprecationWarning
Every deprecated call your tests touch now fails with a traceback pointing at your line, so the next change like this shows up long before the release that removes it.
Versyka finds every call a deprecation like this touches and opens the change as one tested pull request, but grep -rn "utcnow\|utcfromtimestamp" . gets you the list in a second.
How many utcnow() calls does your codebase have? Guess first, then grep.
Top comments (0)