[
https://issues.apache.org/jira/browse/FLINK-19757?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17218976#comment-17218976
]
Jark Wu edited comment on FLINK-19757 at 10/22/20, 12:26 PM:
-------------------------------------------------------------
As I said above, your problem is because of the wrong return type of PROCTIME()
function. We should fix PROCTIME() return type to return a {{TIMESTAMP WITH
LOCAL TIME ZONE}} . This is not a trivial work, because that means we need to
support {{TIMESTAMP WITH LOCAL TIME ZONE}} type as time attribute for
time-based operations (e.g. windows).
Once the return type is fixed, the generated code will be correct.
was (Author: jark):
As I said above, your problem is because of the wrong return type of PROCTIME()
function. We should fix PROCTIME() return type to return a {{TIMESTAMP WITH
LOCAL TIME ZONE}} . This is not a trivial work, because that means we need to
support {{TIMESTAMP WITH LOCAL TIME ZONE}} type as time attribute for
time-based operations (e.g. windows).
The code generation is correct.
> TimeStampData can cause time inconsistent problem
> -------------------------------------------------
>
> Key: FLINK-19757
> URL: https://issues.apache.org/jira/browse/FLINK-19757
> Project: Flink
> Issue Type: Improvement
> Components: Table SQL / Runtime
> Affects Versions: 1.11.1
> Reporter: xiaogang zhou
> Priority: Major
> Labels: pull-request-available
>
> when we check jdk LocalDateTime code，we find that
>
> {code:java}
> // code placeholder
> public static LocalDateTime ofEpochSecond(long epochSecond, int nanoOfSecond,
> ZoneOffset offset) {
> Objects.requireNonNull(offset, "offset");
> NANO_OF_SECOND.checkValidValue(nanoOfSecond);
> long localSecond = epochSecond + offset.getTotalSeconds(); // overflow
> caught later
> long localEpochDay = Math.floorDiv(localSecond, SECONDS_PER_DAY);
> int secsOfDay = (int)Math.floorMod(localSecond, SECONDS_PER_DAY);
> LocalDate date = LocalDate.ofEpochDay(localEpochDay);
> LocalTime time = LocalTime.ofNanoOfDay(secsOfDay * NANOS_PER_SECOND +
> nanoOfSecond);
> return new LocalDateTime(date, time);
> }
> {code}
>
> offset.getTotalSeconds() they add the offset, but in the TimeStampData
> toLocalDateTime, we don't add a offset.
>
> I'd like to add a TimeZone.getDefault().getRawOffset() in the
> toLocalDateTime()
> and minus a TimeZone.getDefault().getRawOffset() in the
> fromLocalDateTime
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
