[VL][1.2][Result mismatch] Cast string to integral type does not ignore ISO control characters

[wForget]

Backend

VL (Velox)

Bug description

1. create test dataset

   create table t1 as select url_decode('111111%00') as c1;
   

2. cast to integral type
   check sqls:

   select cast(c1 as int) from t1;
   select cast(c1 as bigint) from t1;
   

   gluten 1.2 + spark 3.5:

   NULL
   NULL
   

   vanilla spark:

   111111
   111111
   

3. cast to fractional type
   check sqls:

   select cast(c1 as float) from t1;
   select cast(c1 as double) from t1;
   

   gluten 1.2 + spark 3.5:

   111111.0
   111111.0
   

   vanilla spark:

   111111.0
   111111.0
   

Related PRs: apache/spark#41535

Spark version

None

Spark configurations

No response

System information

No response

Relevant logs

No response

[wForget]

I will try to fix it, it seems to be related to this code:

https://github.com/facebookincubator/velox/blob/main/velox/functions/sparksql/specialforms/SparkCastHooks.cpp#L70-L75

[jackylee-ch]

String type conversions were also fixed in #1569 . They used trim to remove invalid characters, but that didn't handle invalid characters inside a String. It would be great to fix this in velox so we can remove #1569 in gluten. cc @philo-he

[wForget]

String type conversions were also fixed in #1569 . They used trim to remove invalid characters, but that didn't handle invalid characters inside a String. It would be great to fix this in velox so we can remove #1569 in gluten. cc @philo-he

Thank you for the information. Since the trim behavior for different types of cast in spark is different, it seems that extending #1569 is a simpler solution, can we fix it that way first?