When timezone param of unix_timestamp is CST, ambiguous zone will lead different result.

[littlelittlewhite09]

Backend

VL (Velox)

Bug description

example:

set spark.sql.optimizer.excludedRules=org.apache.spark.sql.catalyst.optimizer.ConstantFolding;
select from_unixtime(
  unix_timestamp('Sun Jun 14 20:00:05 CST 2026', 'EEE MMM dd HH:mm:Ss z yyyy'),
  'yyyy-MM-dd HH:mm:ss');

vanilla: 2026-06-14 20:00:05
gluten: 2026-06-15 09:00:05

Gluten version

1.6.0

Spark version

3.5.8

Spark configurations

spark.sql.legacy.timeParserPolicy=LEGACY

System information

No response

Relevant logs

[philo-he]

Thanks for raising this issue. I think CST is not a standard or reliable timezone name. Velox has a number of these timezone cases that diverge from Spark. And I noticed even Spark discourages short name due to the ambiguity (See link). I'm not sure this is worth fixing.