[
https://issues.apache.org/jira/browse/MSHADE-467?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17809323#comment-17809323
]
ASF GitHub Bot commented on MSHADE-467:
---------------------------------------
kriegaex commented on PR #210:
URL:
https://github.com/apache/maven-shade-plugin/pull/210#issuecomment-1903588421
I do not think it is pure chance that it happens in the new IT. Maybe, in
some other part of Maven, at least in some versions, there are other
concurrency issues we are hitting here by chance, just because we are testing
this setup. If you run the build a few times and it does not re-occur, maybe
the test is "just" flaky because of Maven or some situation on the GitHub
server. If it happens more often, maybe we can try my initial solution with the
reentrant lock and see if that one produces more stable results.
Looking at the stack trace below the [error message in the Maven 3.6.3 JDK 8
Windows
build](https://github.com/apache/maven-shade-plugin/actions/runs/7596727600/job/20718211260#step:8:8504),
it seems as if the Maven `MultiThreadedBuilder` is involved. Maybe its
resource management is suboptimal on a platform like Windows where open files
are locked.
> Dependency-reduced POM with missing exclusions in concurrent build
> ------------------------------------------------------------------
>
> Key: MSHADE-467
> URL: https://issues.apache.org/jira/browse/MSHADE-467
> Project: Maven Shade Plugin
> Issue Type: Bug
> Affects Versions: 3.4.1, 3.5.0, 3.5.1
> Reporter: Alexander Kriegisch
> Priority: Major
>
> As discussed in [this maven-users
> thread|https://lists.apache.org/thread/ljdbh9h0lqs1qsf9o8bnm35mtr85y4vr],
> when running Maven builds for [this
> reproducer|https://github.com/stl543/shadeMT] concurrently, some dependency
> exclusions are missing from the generated dependency-reduced POMs. Run the
> reproducer like this:
> {code:none}
> $ mvn clean package -T 4 | grep 'BUILD SUCCESS' && find . -name
> "dependency-reduced-pom.xml" -exec bash -c 'cat $0 | grep exclu | wc -l ' {}
> \;
> [INFO] BUILD SUCCESS
> 12
> 12
> 12
> 12
> {code}
> When running with {{-T 1}}, however, the correct result is 20.
> This concurrency issue might have been introduced by the MSHADE-413 fix, but
> I am not sure.
> I have a local fix, synchronising on the injected mojo parameter
> {{MavenSession session}} in the lower section of
> {{ShadeMojo::rewriteDependencyReducedPomIfWeHaveReduction}}. There might be a
> better way to fix the problem, but I it leave that up to the PR reviewers to
> come up with one, if any.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
