[
https://issues.apache.org/jira/browse/SPARK-16464?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15384059#comment-15384059
]
Neil Dewar commented on SPARK-16464:
------------------------------------
Hi Dongjoon,
When a second column has been created with an identical name as an existing
column, I don't think one version of the column effectively hides the other.
Going back to the original code that I included in the post:
`sdfCar <- createDataFrame(sqlContext, mtcars)
sdfCar1 <- withColumn(sdfCar, "isEfficient", sdfCar$mpg<=20)
sdfCar1 <- withColumn(sdfCar1, "isEfficient", ifelse(sdfCar1$mpg ==
sdfCar1$mpg,1,0)) `
If I run `str(sdfCar1)` it displays both versions of the column with the same
name, so the original is not hidden.
If I run functions that run on DataFrames such as `summary(sdfCar1)` or
`columns(sdfCar1)`, they fail reporting
> Reference 'isEfficient' is ambiguous, could be: isEfficient#62,
> isEfficient#63.
Same thing if I run functions that operate on a column, such as
`var(sdfCar1$isEfficient)`
`factorial(sdfCar1$isEfficient)`
`log10(sdfCar1$isEfficient)`
`sdfCar2 <- subset(sdfCar1, sdfCar1$isEfficient < 0.5)`
`sdfCar1 <- withColumn(sdfCar1, "newCol", ifelse(sdfCar1$isEfficient > 30))`
They all fail reporting the error:
> Reference 'isEfficient' is ambiguous, could be: isEfficient#62,
> isEfficient#63.
My point here is that the behavior of SparkR is unstable when withColumn() has
been used to add multiple columns of the same name to a DataFrame.
As most (if not all) SparkR functions expect a DataFrame should not have
multiple columns with the same name, I think `withColumn()` should prevent this
from occurring.
> withColumn() allows illegal creation of duplicate column names on DataFrame
> ---------------------------------------------------------------------------
>
> Key: SPARK-16464
> URL: https://issues.apache.org/jira/browse/SPARK-16464
> Project: Spark
> Issue Type: Bug
> Components: SparkR, SQL
> Affects Versions: 1.6.1
> Environment: Databricks.com
> Reporter: Neil Dewar
> Priority: Minor
>
> If I take an existing DataFrame, I am permitted to use withColumn() to create
> a duplicate column name. I assume this should be illegal, and withColumn
> should be prevented from permitting this. Some functions subsequently fail
> due to the duplicate column names. Example:
> sdfCar <- createDataFrame(sqlContext, mtcars)
> sdfCar1 <- withColumn(sdfCar, "isEfficient", sdfCar$mpg<=20)
> sdfCar1 <- withColumn(sdfCar1, "isEfficient", ifelse(sdfCar1$mpg ==
> sdfCar1$mpg,1,0))
> sdfCar2 <- subset(sdfCar1, select=sdfCar1$isEfficient)
> # subset() command fails with message: "Reference 'isEfficient' is ambiguous"
> Note: I only know if this is SparkR - it might affect other languages APIs.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
