[SPARK-8436][SQL] Inconsistent behavior when converting a Timestamp column to Integer/Long and then convert back to Timestamp

[x1-]

abstract

Through casting Timestamp to Long and Long to Timestamp, we should use same time unit.
I aligned milliseconds.

issue

• Inconsistent behavior when converting a Timestamp column to Integer/Long and then convert back to Timestamp
• My JIRA name is https://issues.apache.org/jira/secure/ViewProfile.jspa?name=x1

[JoshRosen]

Jenkins, this is ok to test.

[SparkQA]

Test build #41321 has finished for PR 8339 at commit bd225ee.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[sarutak]

retest this please.

[sarutak]

@x1- The methods you changed are used from HiveQuerySuite so you should also modify tests in that test suite.
You can see which test cases you should modify in https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/41321/

[x1-]

@sarutak Thank you very much !
Now I modify tests, so later send PR again.

[sarutak]

I noticed the precision was changed by #7283 .

@rxin @marmbrus @davies do you think this change is reasonable?

[SparkQA]

Test build #41343 has finished for PR 8339 at commit bd225ee.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[x1-]

@sarutak
I tried casting Timestamp to Long and Long to Timestamp using Hive below.

hive> SELECT CAST(CAST(1.2 AS TIMESTAMP) AS DOUBLE);
OK
1.2
Time taken: 3.637 seconds, Fetched: 1 row(s)

hive> SELECT CAST(CAST(1200 AS TIMESTAMP) AS INT);
OK
1

This mean that Hive has contradiction in converting Timestamp to Int, and then Int to Timestamp.
Because of spark has different dialect, it is better way closing this PR, I think.

[x1-]

So, now I close this.