[
https://issues.apache.org/jira/browse/SPARK-42335?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wei Guo updated SPARK-42335:
----------------------------
Description:
In PR [https://github.com/apache/spark/pull/29516], in order to fix some bugs,
univocity-parsers used by CSV dataSource was upgraded from 2.8.3 to 2.9.0, it
also involved a new feature of univocity-parsers that quoting values of the
first column that start with the comment character. It made a breaking for
users downstream that handing a whole row as input.
For codes:
{code:java}
Seq(("#abc", 1)).toDF.write.csv("/Users/guowei/comment_test") {code}
Before Spark 3.0，the content of output CSV files is shown as:
!image-2023-02-03-18-56-01-596.png!
After this change, the content is shown as:
!image-2023-02-03-18-56-10-083.png!
For users, they can't set comment option to '\u0000' to keep the behavior as
before because the new added `isCommentSet` check logic as follows:
{code:java}
val isCommentSet = this.comment != '\u0000'
def asWriterSettings: CsvWriterSettings = {
// other code
if (isCommentSet) {
format.setComment(comment)
}
// other code
}
{code}
It's better to pass the comment option through to univocity if users set it
explicitly in CSV dataSource.
After this change, the behavior as flows:
|id|code|2.4 and before|3.0 and after|this update|remark|
|1|Seq("#abc", "\u0000def", "xyz").toDF()
.write.option("comment", "\u0000").csv(path)|#abc
*def*
xyz|{color:#4c9aff}"#abc"{color}
{color:#4c9aff}*def*{color}
{color:#4c9aff}xyz{color}|{color:#4c9aff}#abc{color}
{color:#4c9aff}*"def"*{color}
{color:#4c9aff}xyz{color}|{color:#4c9aff}this update has a little bit
difference with 3.0{color}|
|2|Seq("#abc", "\u0000def", "xyz").toDF()
.write.option("comment", "#").csv(path)|#abc
*def*
xyz|"#abc"
*def*
xyz|"#abc"
*def*
xyz|the same|
|3|Seq("#abc", "\u0000def", "xyz").toDF()
.write.csv(path)|#abc
*def*
xyz|"#abc"
*def*
xyz|"#abc"
*def*
xyz|the same|
|4|{_}Seq{_}("#abc", "\u0000def", "xyz").toDF().write.text(path)
spark.read.option("comment", "\u0000").csv(path)|#abc
xyz|{color:#4c9aff}#abc{color}
{color:#4c9aff}\u0000def{color}
{color:#4c9aff}xyz{color}|{color:#4c9aff}#abc{color}
{color:#4c9aff}xyz{color}|{color:#4c9aff}this update has a little bit
difference with 3.0{color}|
|5|{_}Seq{_}("#abc", "\u0000def", "xyz").toDF().write.text(path)
spark.read.option("comment", "#").csv(path)|\u0000def
xyz|\u0000def
xyz|\u0000def
xyz|the same|
|6|{_}Seq{_}("#abc", "\u0000def", "xyz").toDF().write.text(path)
spark.read.csv(path)|#abc
xyz|#abc
\u0000def
xyz|#abc
\u0000def
xyz|the same|
was:
In PR [https://github.com/apache/spark/pull/29516], in order to fix some bugs,
univocity-parsers used by CSV dataSource was upgraded from 2.8.3 to 2.9.0, it
also involved a new feature of univocity-parsers that quoting values of the
first column that start with the comment character. It made a breaking for
users downstream that handing a whole row as input.
For codes:
{code:java}
Seq(("#abc", 1)).toDF.write.csv("/Users/guowei/comment_test") {code}
Before Spark 3.0，the content of output CSV files is shown as:
!image-2023-02-03-18-56-01-596.png!
After this change, the content is shown as:
!image-2023-02-03-18-56-10-083.png!
For users, they can't set comment option to '\u0000' to keep the behavior as
before because the new added `isCommentSet` check logic as follows:
{code:java}
val isCommentSet = this.comment != '\u0000'
def asWriterSettings: CsvWriterSettings = {
// other code
if (isCommentSet) {
format.setComment(comment)
}
// other code
}
{code}
It's better to pass the comment option through to univocity if users set it
explicitly in CSV dataSource.
|id|code|2.4 and before|3.0 and after|this update|remark|
|1|Seq("#abc", "\u0000def", "xyz").toDF()
.write.option("comment", "\u0000").csv(path)|#abc
*def*
xyz|{color:#4c9aff}"#abc"{color}
{color:#4c9aff}*def*{color}
{color:#4c9aff}xyz{color}|{color:#4c9aff}#abc{color}
{color:#4c9aff}*"def"*{color}
{color:#4c9aff}xyz{color}|{color:#4c9aff}this update has a little bit
difference with 3.0{color}|
|2|Seq("#abc", "\u0000def", "xyz").toDF()
.write.option("comment", "#").csv(path)|#abc
*def*
xyz|"#abc"
*def*
xyz|"#abc"
*def*
xyz|the same|
|3|Seq("#abc", "\u0000def", "xyz").toDF()
.write.csv(path)|#abc
*def*
xyz|"#abc"
*def*
xyz|"#abc"
*def*
xyz|the same|
|4|{_}Seq{_}("#abc", "\u0000def", "xyz").toDF().write.text(path)
spark.read.option("comment", "\u0000").csv(path)|#abc
xyz|{color:#4c9aff}#abc{color}
{color:#4c9aff}\u0000def{color}
{color:#4c9aff}xyz{color}|{color:#4c9aff}#abc{color}
{color:#4c9aff}xyz{color}|{color:#4c9aff}this update has a little bit
difference with 3.0{color}|
|5|{_}Seq{_}("#abc", "\u0000def", "xyz").toDF().write.text(path)
spark.read.option("comment", "#").csv(path)|\u0000def
xyz|\u0000def
xyz|\u0000def
xyz|the same|
|6|{_}Seq{_}("#abc", "\u0000def", "xyz").toDF().write.text(path)
spark.read.csv(path)|#abc
xyz|#abc
\u0000def
xyz|#abc
\u0000def
xyz|the same|
> Pass the comment option through to univocity if users set it explicitly in
> CSV dataSource
> -----------------------------------------------------------------------------------------
>
> Key: SPARK-42335
> URL: https://issues.apache.org/jira/browse/SPARK-42335
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 3.0.0, 3.1.0, 3.2.0, 3.3.0
> Reporter: Wei Guo
> Priority: Minor
> Fix For: 3.4.0
>
> Attachments: image-2023-02-03-18-56-01-596.png,
> image-2023-02-03-18-56-10-083.png
>
>
> In PR [https://github.com/apache/spark/pull/29516], in order to fix some
> bugs, univocity-parsers used by CSV dataSource was upgraded from 2.8.3 to
> 2.9.0, it also involved a new feature of univocity-parsers that quoting
> values of the first column that start with the comment character. It made a
> breaking for users downstream that handing a whole row as input.
>
> For codes:
> {code:java}
> Seq(("#abc", 1)).toDF.write.csv("/Users/guowei/comment_test") {code}
> Before Spark 3.0，the content of output CSV files is shown as:
> !image-2023-02-03-18-56-01-596.png!
> After this change, the content is shown as:
> !image-2023-02-03-18-56-10-083.png!
> For users, they can't set comment option to '\u0000' to keep the behavior as
> before because the new added `isCommentSet` check logic as follows:
> {code:java}
> val isCommentSet = this.comment != '\u0000'
> def asWriterSettings: CsvWriterSettings = {
> // other code
> if (isCommentSet) {
> format.setComment(comment)
> }
> // other code
> }
> {code}
> It's better to pass the comment option through to univocity if users set it
> explicitly in CSV dataSource.
>
> After this change, the behavior as flows:
> |id|code|2.4 and before|3.0 and after|this update|remark|
> |1|Seq("#abc", "\u0000def", "xyz").toDF()
> .write.option("comment", "\u0000").csv(path)|#abc
> *def*
> xyz|{color:#4c9aff}"#abc"{color}
> {color:#4c9aff}*def*{color}
> {color:#4c9aff}xyz{color}|{color:#4c9aff}#abc{color}
> {color:#4c9aff}*"def"*{color}
> {color:#4c9aff}xyz{color}|{color:#4c9aff}this update has a little bit
> difference with 3.0{color}|
> |2|Seq("#abc", "\u0000def", "xyz").toDF()
> .write.option("comment", "#").csv(path)|#abc
> *def*
> xyz|"#abc"
> *def*
> xyz|"#abc"
> *def*
> xyz|the same|
> |3|Seq("#abc", "\u0000def", "xyz").toDF()
> .write.csv(path)|#abc
> *def*
> xyz|"#abc"
> *def*
> xyz|"#abc"
> *def*
> xyz|the same|
> |4|{_}Seq{_}("#abc", "\u0000def", "xyz").toDF().write.text(path)
> spark.read.option("comment", "\u0000").csv(path)|#abc
> xyz|{color:#4c9aff}#abc{color}
> {color:#4c9aff}\u0000def{color}
> {color:#4c9aff}xyz{color}|{color:#4c9aff}#abc{color}
> {color:#4c9aff}xyz{color}|{color:#4c9aff}this update has a little bit
> difference with 3.0{color}|
> |5|{_}Seq{_}("#abc", "\u0000def", "xyz").toDF().write.text(path)
> spark.read.option("comment", "#").csv(path)|\u0000def
> xyz|\u0000def
> xyz|\u0000def
> xyz|the same|
> |6|{_}Seq{_}("#abc", "\u0000def", "xyz").toDF().write.text(path)
> spark.read.csv(path)|#abc
> xyz|#abc
> \u0000def
> xyz|#abc
> \u0000def
> xyz|the same|
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
