最简单的学习方法是让Spark解释这个计划。

scala> spark.sql("SELECT 'true' = true").explain
== Physical Plan ==
*(1) Project [true AS (CAST(true AS BOOLEAN) = true)#53]
+- *(1) Scan OneRowRelation[]

scala> spark.sql("SELECT '2022-01-02' > date('2022-01-03')").explain
== Physical Plan ==
*(1) Project [false AS (CAST(2022-01-02 AS DATE) > CAST(2022-01-03 AS DATE))#55]
+- *(1) Scan OneRowRelation[]

scala> spark.sql("SELECT '2022-08-12 01:00:00.123456' < timestamp('2019-08-12 01:00:00.123456')").explain
== Physical Plan ==
*(1) Project [false AS (CAST(2022-08-12 01:00:00.123456 AS TIMESTAMP) < CAST(2019-08-12 01:00:00.123456 AS TIMESTAMP))#57]
+- *(1) Scan OneRowRelation[]

我们可以清楚地看到，在所有情况下，Spark都试图将string转换为其他类型（boolean、date、timestamp、int ......）。