Description
In Spark 1.x releases we provided the CHANGES.txt file which proved useful to know exactly the differences and commit IDs between releases
In Spark 2.x we made packaging changes (moving make-distribution.sh to be under dev and different structure for the important Spark classes e.g. no big assembly jar), so I expect work wasn't done to ensure the changes file was still being provided
Note that users can view the release notes online (https://issues.apache.org/jira/secure/ReleaseNote.jspa?projectId=12315420&version=12336857) so I don't know if there is merit in including this file in the distribution, but I've had a look in the codebase and noticed the script used to generate said file is still present in master and definitely out of date:
https://github.com/apache/spark/blob/master/dev/create-release/generate-changelist.py, here we see the last commit is from the 27th of August 2014
SPARK_HOME = os.environ["SPARK_HOME"] NEW_RELEASE_VERSION = "1.0.0" PREV_RELEASE_GIT_TAG = "v0.9.1"
If we change this to
SPARK_HOME = os.environ["SPARK_HOME"] NEW_RELEASE_VERSION = "2.0.1" PREV_RELEASE_GIT_TAG = "v2.0.0"
we can create a CHANGES.txt.new file
So I think this file should be either be removed in 2.x. or updated across our currently maintained branches and then we'd need to run the make-distribution script to ensure the file is picked up and included in the distribution
I see no JIRAs mentioning "changes" under the spark project and no discussion of this on either mailing list. I downloaded the Spark 2.0.0 tgz (no 2.0.1 tgz at http://spark.apache.org/downloads.html) and verified there is no changes file here.