[
https://issues.apache.org/jira/browse/HIVE-11812?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15528773#comment-15528773
]
Chetna Chaudhari commented on HIVE-11812:
-----------------------------------------
Debugged this issue, issue exists because of {code}result.set((int)
(diffInMilliSeconds / (86400 * 1000)));{code} line of code. Integer division is
the issue, will need to round up division to get correct result. So replacing
above line of code with following :
{code}result.set((int) Math.round((double) diffInMilliSeconds / (86400 *
1000)));{code} .
> datediff sometimes returns incorrect results when called with dates
> -------------------------------------------------------------------
>
> Key: HIVE-11812
> URL: https://issues.apache.org/jira/browse/HIVE-11812
> Project: Hive
> Issue Type: Bug
> Components: UDF
> Affects Versions: 2.0.0
> Reporter: Nicholas Brenwald
> Assignee: Chetna Chaudhari
> Priority: Minor
>
> DATEDIFF returns an incorrect result when one of the arguments is a date
> type.
> The Hive Language Manual provides the following signature for datediff:
> {code}
> int datediff(string enddate, string startdate)
> {code}
> I think datediff should either throw an error (if date types are not
> supported), or return the correct result.
> To reproduce, create a table:
> {code}
> create table t (c1 string, c2 date);
> {code}
> Assuming you have a table x containing some data, populate table t with 1 row:
> {code}
> insert into t select '2015-09-15', '2015-09-15' from x limit 1;
> {code}
> Then run the following 12 test queries:
> {code}
> select datediff(c1, '2015-09-14') from t;
> select datediff(c1, '2015-09-15') from t;
> select datediff(c1, '2015-09-16') from t;
> select datediff('2015-09-14', c1) from t;
> select datediff('2015-09-15', c1) from t;
> select datediff('2015-09-16', c1) from t;
> select datediff(c2, '2015-09-14') from t;
> select datediff(c2, '2015-09-15') from t;
> select datediff(c2, '2015-09-16') from t;
> select datediff('2015-09-14', c2) from t;
> select datediff('2015-09-15', c2) from t;
> select datediff('2015-09-16', c2) from t;
> {code}
> The below table summarises the result. All results for column c1 (which is a
> string) are correct, but when using c2 (which is a date), two of the results
> are incorrect.
> || Test || Expected Result || Actual Result || Passed / Failed ||
> |datediff(c1, '2015-09-14')| 1 | 1| Passed |
> |datediff(c1, '2015-09-15')| 0 | 0| Passed |
> |datediff(c1, '2015-09-16') | -1 | -1| Passed |
> |datediff('2015-09-14', c1) | -1 | -1| Passed |
> |datediff('2015-09-15', c1)| 0 | 0| Passed |
> |datediff('2015-09-16', c1)| 1 | 1| Passed |
> |datediff(c2, '2015-09-14')| 1 | 0| {color:red}Failed{color} |
> |datediff(c2, '2015-09-15')| 0 | 0| Passed |
> |datediff(c2, '2015-09-16') | -1 | -1| Passed |
> |datediff('2015-09-14', c2) | -1 | 0 | {color:red}Failed{color} |
> |datediff('2015-09-15', c2)| 0 | 0| Passed |
> |datediff('2015-09-16', c2)| 1 | 1| Passed |
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
