pandas/epoch-from-int-cast
| Rule | epoch-from-int-cast |
|---|---|
| Pack | pandas |
| Severity | error |
| Fix | unsafe, with xpd check --fix --unsafe-fixes |
| Library versions | every version |
pandas 3 stores datetimes in microseconds by default, so casting a datetime to an int and dividing by 10**9 is off by a factor of 1,000. Subtract the epoch and divide by a pd.Timedelta.
seconds = df["time"].astype("int64") // 10**9seconds = (df["time"] - pd.Timestamp("1970-01-01")) // pd.Timedelta("1s")What the rule reports
Section titled “What the rule reports”The rule reports df["d"].astype("int64") // 10**9, index.asi8 // 10**9 and df["d"].values.astype("float64") / 1e9, which are off by a factor of 1,000 now that datetimes default to microseconds. It also reports .view(np.int64) divided the same way.
How to fix it
Section titled “How to fix it”xpd check --fix --unsafe-fixes applies the fixes of this rule. Check the result, because an unsafe fix can change what the code does.
The unsafe fix writes (col - pd.Timestamp("1970-01-01")) // pd.Timedelta("1s"), with "1ms" for 10**6 and "1us" for 10**3. The result changes on purpose, a column with a time zone raises TypeError, because the epoch has none, a missing value gives NaN, and the fix of asi8 and .values returns an Index or a Series where the old code returned a NumPy array.
The replacement assumes dates without a time zone, and for dates with a time zone you subtract pd.Timestamp("1970-01-01", tz="UTC") instead. A missing timestamp stays missing.
Message
Section titled “Message”xpd prints this message with the issue. A name that starts with $ stands for code from your file.
Datetimes default to microseconds in pandas 3, so this cast no longer gives nanoseconds. Write ($X - pd.Timestamp("1970-01-01")) // $UNIT.Browse the other rules of the pandas pack.