[
https://issues.apache.org/jira/browse/FLINK-12602?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16846661#comment-16846661
]
Chesnay Schepler commented on FLINK-12602:
------------------------------------------
Ha, you got me there. flink-tests, flink-fs-tests and flink-yarn-tests also
don't need a suffix.
The purpose of the scala-suffix is to ensure that _users_ depending on a module
do not inadvertently pull in scala dependencies.
So, if you depend on module "MyModule", then there should be no transitive
dependency pulled in that depends on a specific scala version.
If it did you'd have no control over which version is actually pulled in.
Dependencies that have a {{test}} scope however are _not_ transitive and never
pulled in by the user.
That's why we can ignore those. This is why flink-avro does not have a suffix,
and why flink-tests in fact does not need one.
Now, let's consider what is wrong with the script.
As you laid out correctly, we do not check correctly for the scala suffix in
scala-free modules. With this change alone 11 modules are marked as having
wrong suffixes:
# flink-connector-hive
# flink-fs-tests
# flink-queryable-state-client-java
# flink-sql-connector-elasticsearch6
# flink-sql-connector-kafka
# flink-sql-connector-kafka-0.10
# flink-sql-connector-kafka-0.11
# flink-sql-connector-kafka-0.9
# flink-table-api-scala
# flink-tests
# flink-yarn-tests
fs-tests, tests, yarn-tests do not need a suffix because they only have test
dependencies.
table-api-scala does not need a suffix _right now_ but will obviously need it
in the future since it is supposed to contain scala code. As you pointed out
correctly, let's add the scala dependencies now to future-proof this puppy.
connector-hive and queryable-state-client-java are genuinely free of scala
dependencies, and we can remove the suffix as is.
The sql-connectors _do_ all need a suffix since they all rely on connectors
which also have one, which need it themselves since they rely on
runtime/streaming-java.
My suspicion is that these are tagged because the script works against
dependency-reduced poms, where provided dependencies are removed.
runtime/streaming-java are usually marked as provided, so they and their
transitive dependencies, including scala, are omitted from the pom.
As a result, as you pointed out, we not only have to check for scala
dependencies but for any (compile) dependencies with a scala-suffix.
> Correct the flink pom `artifactId` config and scala-free check logic
> --------------------------------------------------------------------
>
> Key: FLINK-12602
> URL: https://issues.apache.org/jira/browse/FLINK-12602
> Project: Flink
> Issue Type: Bug
> Components: Build System
> Affects Versions: 1.9.0
> Reporter: sunjincheng
> Priority: Major
>
> I find a shell issue in `verify_scala_suffixes.sh`(line 145) as follows:
> {code}
> grep "${module}_\d\+\.\d\+</artifactId>" "{}"
> {code}
> This code want to find out all modules that the module's `artifactId` with a
> `scala_binary_version` suffix.
> but the problem is our all `artifactId` value is in the pattern of
> `XXX_${scala.binary.version}`, such as:
> {code}
> <artifactId>flink-tests_${scala.binary.version}</artifactId>
> {code}
> then the result always empty, so this check did not take effect.
> I have already initiated a discussion of the issue. Please check the Mail
> thread here for details.
> http://mail-archives.apache.org/mod_mbox/flink-dev/201905.mbox/%3ccajsjtkw+8mcsc0fvneyaovl_ttrr_uuosx-tfgxj5gfqp1a...@mail.gmail.com%3E
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
