Yang Jie created SPARK-60451:
--------------------------------
Summary: Cast.contextIndependentFoldable is true for a special
datetime string cast to DATE or TIMESTAMP_NTZ
Key: SPARK-60451
URL: https://issues.apache.org/jira/browse/SPARK-60451
Project: Spark
Issue Type: Bug
Components: SQL
Affects Versions: 5.0.0
Reporter: Yang Jie
Cast.contextIndependentFoldable decides from the types alone
(Cast.scala:716-718). A string cast to DATE or TIMESTAMP_NTZ needs no time
zone, so it counts as foldable without external context. But a special datetime
value such as 'today', 'now' or 'epoch' is replaced only by the optimizer rule
SpecialDatetimeValues, and an ordinary Cast.eval does not know it.
V2ExpressionBuilder folds every contextIndependentFoldable sub-expression, so
building a V2 expression from an analyzed expression evaluates such a cast as
an ordinary cast. On a V2 catalog:
{noformat}
CREATE TABLE t (id INT, d DATE, CONSTRAINT c CHECK (d <= CAST('today' AS
DATE)));
-- ANSI on: fails with CAST_INVALID_INPUT
-- ANSI off: stores the predicate d <= NULL, so INSERT INTO t VALUES (1,
DATE'9999-12-31') passes
CREATE TABLE t (id INT, d DATE DEFAULT CAST('today' AS DATE));
-- ANSI on: fails with CAST_INVALID_INPUT
-- ANSI off: INSERT INTO t(id) VALUES (1) writes NULL instead of the current
date
{noformat}
The implicit cast in CHECK (d <= 'today') behaves the same. REPLACE TABLE
converts its constraints the same way as CREATE TABLE, and
GeneratedColumnExpression.toV2 runs the same builder on its analyzed expression.
A fix could make contextIndependentFoldable false for a foldable string child
whose value is a special datetime value. DateTimeUtils.extractSpecialValue
recognizes these values; it is private now.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
