sepuri sai krishna created SPARK-59860:
------------------------------------------
Summary: Fail on overflow in from_unixtime instead of wrapping
around
Key: SPARK-59860
URL: https://issues.apache.org/jira/browse/SPARK-59860
Project: Spark
Issue Type: Bug
Components: SQL
Affects Versions: 5.0.0
Reporter: sepuri sai krishna
{{from_unixtime}} converts seconds to microseconds with a plain {{Long}}
multiplication, so for inputs beyond about 9.2 * 10^12 seconds the result
silently wraps around and it returns a wrong timestamp string instead of
failing. This happens with ANSI mode on and off, and in both codegen and
interpreted mode:
{code:sql}
SET spark.sql.session.timeZone=UTC;
SELECT from_unixtime(9223372036854); -- +294247-01-10 04:00:54 (largest
value that fits, correct)
SELECT from_unixtime(9223372036855); -- -290308-12-21 19:59:05
SELECT from_unixtime(9223372036854775807); -- 1969-12-31 23:59:59
{code}
One second past the largest value that fits jumps back by about 584,000 years,
and the maximum long value comes out as a plausible looking 1969 timestamp.
{{timestamp_seconds}} performs the same conversion and fails with
{{DATETIME_OVERFLOW}} on these inputs, with ANSI mode on and off:
{code:sql}
SELECT timestamp_seconds(9223372036855);
-- [DATETIME_OVERFLOW] Datetime operation overflow: create a TIMESTAMP from
9223372036855L seconds since the epoch.
{code}
{{from_unixtime}} should fail with the same error.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
