[
https://issues.apache.org/jira/browse/FLINK-33430?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17782164#comment-17782164
]
Matthias Pohl commented on FLINK-33430:
---------------------------------------
I don't fully understand the intention of this Jira issue. We're not using
{{edu.illinois:nondex-maven-plugin:2.1.1:nondex}} as part of our build system.
Therefore, the test instability your mentioning isn't present. Looking through
the code, it appears that the production code is also deterministically
generating the String (I haven't checked the entire code, though, tbh).
Based on what I found about the {{edu.illinois:nondex-maven-plugin}} (from
[this Github source|https://github.com/TestingResearchIllinois/NonDex]), it
verifies that the code behaves in the same way among different JVM
implementations. Is this correct?
If you think that it makes sense to add this plugin to Flink's build pipeline
we should actually approach this on a more systematic level:
1. Discussing why this plugin should be added in the first place (e.g. in the
dev ML). Where are the benefits to the Flink users?
2. If the community agrees we should add the plugin to the build pipeline and
fix all tests in one go.
WDYT?
Your approach of fixing a test that's not failing in our CI right now but just
running it with a plugin
> Fix Test Failures
> -----------------
>
> Key: FLINK-33430
> URL: https://issues.apache.org/jira/browse/FLINK-33430
> Project: Flink
> Issue Type: Bug
> Reporter: dserfe
> Priority: Minor
> Labels: pull-request-available
>
> This issue is to fix test failures caused by flakiness.
> Steps to reproduce test failures:
> 1)Run the following commands
> {{}}
> {code:java}
> {code}
> {{mvn -pl flink-tests edu.illinois:nondex-maven-plugin:2.1.1:nondex
> -Dtest=org.apache.flink.test.misc.GenericTypeInfoTest#testSerializerTree}}
>
> 2) Then we found the following test failures:
> {{}}
> {code:java}
> [ERROR] GenericTypeInfoTest.testSerializerTree:53 Expected: "GenericTypeInfo
> (PojoWithCollectionGeneric)\n pojos:java.util.List\n key:int\n
> sqlDate:java.sql.Date\n bigInt:java.math.BigInteger\n
> bigDecimalKeepItNull:java.math.BigDecimal\n intVal:java.math.BigInteger\n
> scale:int\n scalaBigInt:scala.math.BigInt\n bigInteger:java.math.BigInteger\n
> mixed:java.util.List\n
> makeMeGeneric:org.apache.flink.test.operators.util.CollectionDataSets$PojoWithDateAndEnum\n
> group:java.lang.String\n date:java.util.Date\n
> cat:org.apache.flink.test.operators.util.CollectionDataSets$Category (is
> enum)\n" but: was "GenericTypeInfo (PojoWithCollectionGeneric)\n key:int\n
> bigDecimalKeepItNull:java.math.BigDecimal\n intVal:java.math.BigInteger\n
> scale:int\n bigInt:java.math.BigInteger\n sqlDate:java.sql.Date\n
> pojos:java.util.List\n scalaBigInt:scala.math.BigInt\n
> bigInteger:java.math.BigInteger\n
> makeMeGeneric:org.apache.flink.test.operators.util.CollectionDataSets$PojoWithDateAndEnum\n
> date:java.util.Date\n
> cat:org.apache.flink.test.operators.util.CollectionDataSets$Category (is
> enum)\n group:java.lang.String\n mixed:java.util.List\n" {code}
> The root cause of the test failure is due to assuming a consistent order of
> {{{}serTree{}}}, but the elements of it are actually not consistent, to fix
> the test failure:
> _change the assertion to compare two lists, which can ignore the orders of
> elements_
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
