Linhong Liu created SPARK-61579:
-----------------------------------

             Summary: Read/write-side CHAR/VARCHAR processing drops a nested 
NOT NULL field when rebuilding a struct
                 Key: SPARK-61579
                 URL: https://issues.apache.org/jira/browse/SPARK-61579
             Project: Spark
          Issue Type: Bug
          Components: SQL
    Affects Versions: 5.0.0
            Reporter: Linhong Liu


CharVarcharUtils.processStringForCharVarchar rebuilds a struct containing a 
CHAR/VARCHAR field as a named_struct of GetStructField(expr, i) values. Because 
the struct `expr` is nullable, every GetStructField is nullable, so a declared 
nested NOT NULL (e.g. STRUCT<c CHAR(2), z STRING NOT NULL>) is lost in the 
rebuilt type. The re-typed attribute is then wider than the original, which 
e.g. makes a Delta MERGE type check reject the assignment.

Fix: preserve each field's declared nullability when rebuilding the struct 
(wrap declared NOT NULL fields in KnownNotNull); the whole-struct-null case is 
still handled by the surrounding null guard. Found via correctness analysis.



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to