[SPARK-18489] Implicit type conversion during comparision between Integer type column and String type column

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Duplicate
• Component/s: SQL
• Assignee: Unassigned
• Reporter: Bipul Kumar
• Created: 17/Nov/16 11:29
• Updated: 07/Feb/20 17:23
• Resolved: 17/Nov/16 11:50

Description

Suppose I have a dataframe with schema:

root
 |-- _c0: integer (nullable = true)
 |-- _c1: double (nullable = true)
 |-- _c2: string (nullable = true)

and data:

+---+---+----+
|_c0|_c1| _c2|
+---+---+----+
|  1|1.0|   1|
|  2|1.0|   s|
|  3|3.1|null|
+---+---+----+

if the following operations are carried out:

df.where("_c1==_c2").show
+---+---+---+
|_c0|_c1|_c2|
+---+---+---+
|  1|1.0|  1|
+---+---+---+

df.where("_c1<>_c2").show   or   df.where("_c1!=_c2").show 
+---+---+---+
|_c0|_c1|_c2|
+---+---+---+
+---+---+---+

So the related operation results are ambiguous
Here the stringified numeric values are being Implicitly casted where the others are just ignored instead of throwing an exception
In my view these things can lead to incorrect results if dataset is not properly observed.

Also SQL-99 standard discourages implicit casting to avoid such things.
https://users.dcc.uchile.cl/~cgutierr/cursos/BD/standards.pdf

The same implicit casting is also there for UDFs and aggregation functions.

Issue Links

• duplicates: SPARK-17913 Filter/join expressions can return incorrect results when comparing strings to longs