github-actions[bot] commented on code in PR #64029:
URL: https://github.com/apache/doris/pull/64029#discussion_r3348619647
##########
be/src/exprs/function/function_convert_tz.cpp:
##########
@@ -212,6 +212,19 @@ class FunctionConvertTZ : public IFunction {
}
}
+ static std::pair<int64_t, int64_t> unix_timestamp_for_convert_tz(
+ const DateValueType& ts_value, const cctz::time_zone& from_tz) {
+ cctz::civil_second civil_time(ts_value.year(), ts_value.month(),
ts_value.day(),
+ ts_value.hour(), ts_value.minute(),
ts_value.second());
+ const auto lookup = from_tz.lookup(civil_time);
+ const bool skipped = lookup.kind ==
cctz::time_zone::civil_lookup::SKIPPED;
+ const auto tp = skipped ? lookup.trans : lookup.pre;
+
+ // Skipped civil times map to the transition instant. Do not keep the
+ // input fractional part inside a local time interval that never
existed.
+ return {tp.time_since_epoch().count(), skipped ? 0 :
ts_value.microsecond()};
+ }
Review Comment:
This changes runtime `convert_tz` to drop the DATETIMEV2 fractional part for
every skipped civil time. For example, with `debug_skip_fold_constant=true`,
`convert_tz('2021-03-28 02:30:00.003230', 'Europe/Paris', 'UTC')` now returns
the transition second with `.000000`, and the updated regression output accepts
that loss. The previous `DateV2Value::unix_timestamp(std::pair<...>)` path
converted the civil second but carried `ts_value.microsecond()` separately, and
other local-to-UTC paths such as `TimestampTzValue::convert_local_to_utc` still
preserve `dt.microsecond()`. Please keep the original microsecond in the
returned pair for `SKIPPED` as well, and update the regression/unit tests to
assert the fractional part is preserved.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
