[
https://issues.apache.org/jira/browse/SPARK-13411?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Barry Becker updated SPARK-13411:
---------------------------------
Description:
I don't know if the behavior in 1.5.3 or 1.6.0 is correct, but its definitely
different.
Suppose I have a dataframe with a double column, "foo", that is all null valued.
If I do
val ext: DataFrame = df.agg(min("foo"), max("foo"),
count(col("foo")).alias("nonNullCount"))
In 1.5.2 I could do ext.getDouble(0) and get Double.NaN.
In 1.6.0, when I try this I get "value in null at index 0". Maybe the new
behavior is correct, but I think there is a typo in the message. It should say
"value is null at index 0".
Which behavior is correct? If 1.6.0 is correct, then it looks like I will need
to add isNull checks everywhere when retrieving values.
was:
I don't know if the behavior in 1.5.3 or 1.6.0 is correct, but its definitely
different.
Suppose I have a dataframe with a double column, "foo", that is all null valued.
If I do
val ext: DataFrame = df.agg(min("foo"), max("foo"),
count(col("foo")).alias("nonNullCount"))
In 1.5.2 I could do ext.getDouble(0) and get Double.NaN.
In 1.6.0, when I try this I get
Which is correct.
I think the 1.5.2 behavior is better otherwise I need to add special case
handling for when a column is all null.
> change in null aggregation behavior from spark 1.5.2 and 1.6.0
> ---------------------------------------------------------------
>
> Key: SPARK-13411
> URL: https://issues.apache.org/jira/browse/SPARK-13411
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 1.6.0
> Reporter: Barry Becker
>
> I don't know if the behavior in 1.5.3 or 1.6.0 is correct, but its definitely
> different.
> Suppose I have a dataframe with a double column, "foo", that is all null
> valued.
> If I do
> val ext: DataFrame = df.agg(min("foo"), max("foo"),
> count(col("foo")).alias("nonNullCount"))
> In 1.5.2 I could do ext.getDouble(0) and get Double.NaN.
> In 1.6.0, when I try this I get "value in null at index 0". Maybe the new
> behavior is correct, but I think there is a typo in the message. It should
> say "value is null at index 0".
> Which behavior is correct? If 1.6.0 is correct, then it looks like I will
> need to add isNull checks everywhere when retrieving values.
--
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
