[SPARK-8206][SQL][WIP]Add function round

[zhichao-li]

No description provided.

[zhichao-li]

cc @chenghao-intel

[SparkQA]

Test build #34972 has finished for PR 6836 at commit 4069bba.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
  • case class Round(left: Expression, right: Expression)

[chenghao-intel]

This probably not necessary, as both of the parameters are literals(constant value).

[markhamstra]

I don't believe this is quite right, since setScale in BigDecimal will handle negative values differently than SQL ROUND(column_name, decimals) will handle a negative value of decimals -- every SQL engine that I'm familiar with will treat negative decimals as if it were 0.

[chenghao-intel]

Still some other issues need to be addressed:

• From the GenericUDFRound of Hive, seems the scale is a constant value, we need to it in the same way.
• The output dataType can be either decimal or double.

[markhamstra]

Actually, it's worth considering whether in the common case of rounding to zero decimal places we should be producing integral values.

[zhichao-li]

@markhamstra@chenghao-intel For hive it would produce double for such case I guess we should keep the same ?

[chenghao-intel]

We'd better to keep the same behavior as Hive does, otherwise it will causes inconsistencies for existed code.

[SparkQA]

Test build #35254 has finished for PR 6836 at commit 2d5d402.

• This patch fails Scala style tests.
• This patch does not merge cleanly.
• This patch adds the following public classes (experimental):
  • case class Round(left: Expression, right: Expression)

[SparkQA]

Test build #35500 has finished for PR 6836 at commit cd30012.

• This patch passes all tests.
• This patch does not merge cleanly.
• This patch adds the following public classes (experimental):
  • case class Round(left: Expression, right: Expression)

[yjshen]

Ah.... I didn't notice there are two JIRA for round, #6938 is my working version, behave as what Hive's GenericUDFRound does.

[yjshen]

How about just act as Hive's GenericUDFRound (#6938 takes the same logic to behave like Hive's counterparts) ? If so, I think we can merge these two.

[chenghao-intel]

can you rebase this please?

[SparkQA]

Test build #35547 has finished for PR 6836 at commit 87164a1.

• This patch passes all tests.
• This patch does not merge cleanly.
• This patch adds the following public classes (experimental):
  • case class Round(left: Expression, right: Expression)

[yjshen]

GenericUDFRound would return null if Float.isNaN(f) or Float.isInfinite(f)

[zhichao-li]

Good catch, would add that same as hive.

[SparkQA]

Test build #35613 has finished for PR 6836 at commit 4e66e8d.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
  • case class Round(left: Expression, right: Expression)

[SparkQA]

Test build #35626 has finished for PR 6836 at commit 918376d.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
  • case class Round(left: Expression, right: Expression)