Hyukjin Kwon created SPARK-60994:
------------------------------------
Summary: Type a decimal literal of the Python client as PySpark
does
Key: SPARK-60994
URL: https://issues.apache.org/jira/browse/SPARK-60994
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
In PySpark Connect, {{lit(Decimal(...))}} sends a decimal literal of
{{DecimalType()}}, of precision 10 and scale 0, which the server widens to the
precision and scale of the value: {{lit(Decimal("1.5"))}} is a
{{decimal(10,1)}} and {{lit(Decimal(0))}} a {{decimal(10,0)}}. The Rust-backed
Python client sends the precision and scale of the value, so these literals are
a {{decimal(2,1)}} and a {{decimal(1,0)}}, and the expressions over them have
other types: {{lit(Decimal("1.5")) * lit(Decimal("2.25"))}} is a
{{decimal(6,3)}} rather than a {{decimal(21,3)}}, and so may overflow where
PySpark's does not. The seconds of {{make_interval}}, {{make_dt_interval}} and
{{try_make_interval}}, which default to {{lit(Decimal(0))}}, differ the same
way. Type a decimal literal as PySpark does.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
