[BEAM-3279] Deprecate and remove Coder.consistentWithEquals

[AlexKbit]

Hi @kennknowles !
Could you please review my changes:

1. Deprecate Coder.consistentWithEquals method and add javadoc
2. Remove overriding from coders in the java-core

I'm not sure about removal of overriding from :
org.apache.beam.sdk.coders.MapCoder#consistentWithEquals
org.apache.beam.sdk.coders.ListCoder#consistentWithEquals

because it breaks the tests below:
org.apache.beam.sdk.coders.MapCoderTest > testConsistentWithEquals FAILED
org.apache.beam.sdk.coders.ListCoderTest > testConsistentWithEquals FAILED

I can rewrite these tests if it's necessary.

Post-Commit Tests Status (on master branch)

Lang	SDK	Apex	Dataflow	Flink	Gearpump	Samza	Spark
Go		---	---	---	---	---	---
Java
Python		---			---	---	---

See .test-infra/jenkins/README for trigger phrase, status and link of all Jenkins jobs.

[kennknowles]

Thanks for working on this!

[kennknowles]

Great. This looks like a good improvement.

[AlexKbit]

@kennknowles this change will ready to merge, If you don't have other comments

[AlexKbit]

@kennknowles PR squashed & ready to merge.
Thank you for your time!

[kennknowles]

@AlexKbit I opened a pull request to you for minor changes. Did you see it?

[AlexKbit]

@kennknowles Do you mean tests with hardcoded expected result for consistentWithEquals?
Yep. I removed tests from ListCoderTest & MapCoderTest.

[kennknowles]

https://github.com/AlexKbit/beam/pull/1

[AlexKbit]

@kennknowles should i do squash with this changes again?

[kennknowles]

I can squash them for you. Thanks!

[kennknowles]

Thanks! If you want to follow-up there is good test coverage missing here: each coder could have a test, shared in CoderProperties, that the equality on encoded bytes matches the equality on structuralValue.

[AlexKbit]

Thank you! Can you create another ticket for this task(assign to me)?

[kennknowles]

https://issues.apache.org/jira/browse/BEAM-6904

[AlexKbit]

Thank you! 😄

[kennknowles]

Congratulations! This change (which is good) exposed a critical bug in the core SDK. We may have to roll this back, hopefully only temporarily, while we figure out what to do.

[kennknowles]

To roll this forward, we need to do a few things:

• understand which core transforms are affected when the serialized coder changes
• preferred: replace all Java serialized coders in the core SDK with StructuredCoder
  • for runners like Dataflow (and I think Flink) that do in-place update, make them OK with the change