[SPARK-12115] [SparkR] Change numPartitions() to getNumPartitions() to be consistent with Scala/Python

[yanboliang]

Change numPartitions() to getNumPartitions() to be consistent with Scala/Python.
Note: If we can not catch up with 1.6 release, it will be breaking change for 1.7 that we also need to explain in release note.

cc @sun-rui @felixcheung @shivaram

[felixcheung]

This is actually not exported from SparkR - since it is first integrated in Spark 1.4, SparkR is exporting a smaller/different set of API.
You can see in https://github.com/apache/spark/blob/master/R/pkg/NAMESPACE

While it is possible to access this with Spark:::numPartitions(), it has been available since Spark 1.4 so this rename is actually going to be a breaking change (of an internal API).
So I'd vote for no change in SparkR.

[SparkQA]

Test build #47121 has finished for PR 10123 at commit 6870073.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[srowen]

Yeah, it's unfortunately because until a change 3 days ago, we had 3 different methods in 4 languages for this simple function. Now everything but R uses getNumPartitions. It's not worth breaking something but may be worth adding/deprecating a method.

[yanboliang]

@felixcheung Thanks for your comments. I got it's not an exposed API, but I think to provide consistent function name is necessary especially when we want to expose RDD API someday. I think another solution is to add getNumPartitions as an alias of numPartitions which will not cause breaking change, and we can expose getNumPartitions when we want to expose RDD API. Looking forward to other members comments.

[yanboliang]

@srowen I think we have the consistent idea that to provide both getNumPartitions and numPartitions at SparkR side, and mark numPartitions as deprecated.

[sun-rui]

+1 @yanboliang

[SparkQA]

Test build #47135 has finished for PR 10123 at commit ed691eb.

• This patch fails some tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #47137 has finished for PR 10123 at commit 94c596d.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[shivaram]

Yeah this looks fine. Since it wasn't a publicly exposed API I don't think the backwards compatibility matters that much -- but for right now I'm +1 on just deprecating numPartitions.

[shivaram]

Could you add a note that this is deprecated here ?

Also this is the title of the document, so its better to have this be something like Gets the number of partitions of an RDD cc @felixcheung

[felixcheung]

It is, although we are not generating doc for internal API via @noRd below...
Still, better to be descriptive.

[felixcheung]

In fact, if we are deprecating, consider adding

.Deprecated(newFuncSig, old = oldFuncSig)

call to the deprecated version

[SparkQA]

Test build #47179 has finished for PR 10123 at commit c17ce80.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[shivaram]

LGTM. Merging this. Thanks @yanboliang