Maven doesn't try to download a dependency when it already exists a version with classifier locally

Something else to add : If I launch after the site build, another build which doesn't produce the main artefact (for example : mvn sonar), it will also fail trying to download dependency. I think that it can be related to maven-metadata and more probably to the maven maven-metadata-local.xml which is generated.
After the mvn site, I have in my local repository in org\apache\maven\it\test-metadata-local-clover\module1\1.0.0-SNAPSHOT:

maven-metadata-central.xml
maven-metadata-local.xml
maven-metadata-remote-repo.xml
maven-metadata-remote-repo.xml.sha1
module1-1.0.0-SNAPSHOT-clover.jar
module1-1.0.0-SNAPSHOT.pom

and in each metadata file :

• maven-metadata-central.xml

  <?xml version="1.0" encoding="UTF-8"?>
  <metadata xsi:schemaLocation="http://maven.apache.org/METADATA/1.0.0 http://maven.apache.org/xsd/metadata-1.0.0.xsd" xmlns="http://maven.apache.org/METADATA/1.0.0"
      xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
    <groupId>org.apache.maven.it.test-metadata-local-clover</groupId>
    <artifactId>module1</artifactId>
    <version>1.0.0-SNAPSHOT</version>
  </metadata>
  

• maven-metadata-local.xml

  <?xml version="1.0" encoding="UTF-8"?>
  <metadata xsi:schemaLocation="http://maven.apache.org/METADATA/1.0.0 http://maven.apache.org/xsd/metadata-1.0.0.xsd" xmlns="http://maven.apache.org/METADATA/1.0.0"
      xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
    <groupId>org.apache.maven.it.test-metadata-local-clover</groupId>
    <artifactId>module1</artifactId>
    <version>1.0.0-SNAPSHOT</version>
    <versioning>
      <snapshot>
        <localCopy>true</localCopy>
      </snapshot>
      <lastUpdated>20090423144418</lastUpdated>
    </versioning>
  </metadata>
  

• maven-metadata-remote-repo.xml

  <?xml version="1.0" encoding="UTF-8"?>
  <metadata xsi:schemaLocation="http://maven.apache.org/METADATA/1.0.0 http://maven.apache.org/xsd/metadata-1.0.0.xsd" xmlns="http://maven.apache.org/METADATA/1.0.0"
      xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
    <groupId>org.apache.maven.it.test-metadata-local-clover</groupId>
    <artifactId>module1</artifactId>
    <version>1.0.0-SNAPSHOT</version>
    <versioning>
      <snapshot>
        <timestamp>20090423.140624</timestamp>
        <buildNumber>1</buildNumber>
      </snapshot>
      <lastUpdated>20090423140625</lastUpdated>
    </versioning>
  </metadata>
  

  Do you think that maven-metadata-local.xml is good ? It says that there's a local copy whereas there's no module1-1.0.0-SNAPSHOT.jar in the directory.

The problem is that when the Maven resolve the artifact version with maven-metadata-local.xml when it exists and it resolves it to 1.0.0-SNAPSHOT whereas only module1-1.0.0-SNAPSHOT-clover.jar (in this exemple). Then when it try to access the file artifact (module1-1.0.0-SNAPSHOT.jar) and fails, it tries to get the same version on remote repositories whereas it should have try to resolve the artifact version with data found in maven-metadata-$

.xml before trying to get it from each repository.

I have attached a small patch (base on revision 805295 in branch maven-2.2.x) which might not be the best way to correct this bug : while resolving artifact version, it loads local metadata only if artifact's file exists locally.

A possible correction for bug MNG-4142

I've encountered this problem as well when doing distributed builds (using Hudson).
When one build server (#1) builds and deploys an artifact the maven-metadata-local.xml will contain the localCopy attribute:
<versioning>
<snapshot>
<localCopy>true</localCopy>
</snapshot>
<lastUpdated>20091127124227</lastUpdated>
</versioning>

Which means that if another build server builds the same module AFTER server #1 and produce a new artifact in the repository, machine #1 will never download this new SNAPSHOT version when trying to build artifacts depending on the artifact.
This is a huge problem (and a scary one!) for us and I would like to see a fix for this soon

I just ran into this issue myself (with Maven 2.2.1). A build on one of my build machines was failing due to maven not downloading a newer version of a dependency that was available on Nexus because it had previously installed an earlier version of that same dependency into the local repository.

When I removed the <localCopy>true</localCopy> element from the metadata of the non-updating dependency the problem went away - and dutifully came back when I re-added it.

Peter, I agree - this is a scary problem for anybody trying to implement a distributed build system. What I don't understand is why more people are not encountering this issue.

For now I'll attempt a workaround this by adding a cron job, per build agent, to frequently change <localCopy> back to false. If anybody has a better suggestion, I'd love to hear it.

Hi,

I would like to ask why the attached patch is not being accepted?
There are currently over three dozen votes for this and it is apparent that people are hitting the problem.

In addition, perhaps the title of the bug is not quite correct. This doesn't just happen for artifacts with classifiers. In our case it happens for POM files where there are several inherited POMs.
We're using Hudson and Nexus had to limit the parents to build on one Hudson node, while all modules which depend on them - on the rest of the nodes.

I also think this is a huge and very obscure problem (esp. with distributed builds) and, provided that there is already a patch, I believe it's about time for it to be reviewed.

Regards,

Martin

Were you able to test if the problem exists with maven 3 ?
I agree that this issue is really annoying but for now there is no 2.X version release scheduled.
That's why there was no feedback from dev team.

Bonjour Arnaud!

As per your request, I have tested this with Maven 3.0 today. I am afraid the problem still persists. Today this actually happened with a completely different project/setup/machine under both Maven 2.2.1 and 3.0.

This is not just annoying, it's actually a show-stopper when it happens to a developer who's not expecting it, as they spend quite some time trying to figure it out.

Please review for the release of Maven 3.1, (if a 2.x version will not be further released).

Regards,

Martin

I've the same problem but with standard SNAPSHOT dependencies, not using classifier, and with 3.0.3.

This bug/feature doesn't really have much to do with classifiers. I believe the original bug report filer just thought it had something to do with the problem.
Perhaps somebody should rename the bug to "Maven doesn't try to download SNAPSHOT artifacts, if you've installed them locally; even if you pass -U".

I believe that by using -U the local metadata should get "unlocked" (which currently doesn't do this) for the locally installed SNAPSHOT artifacts.

alright, good then, that's my problem!

Can't wait for the fix to come, hopefully as early as 3.0.4...

We're a year after the latest comment, could someone update the Fix version, that'd give some hope.

Our team is severely being affected by this problem, and it seems to be larger than the description suggests.
We have SNAPSHOT update problems on a daily basis, even for deps than haven't been installed locally.
A CI server deploy the artifact, and in some case no one seem to get the updated artifact.

Thanks for pointing me to other bug report if my problem is out of scope here.

I have replied to numerous posts on Stack Overflow concerning this particular issue. People keep hitting this and are unhappy with the current behavior. In the company I work for, we had to make Jenkins use private repositories for each project which would be removed before every build, just for the sake of sanity and for us not losing our minds when a build fails due to this reason.

The problem here is also observed under Maven 3. Furthermore, it's not a classifier problem. It happens to whatever dependencies you have to modules which you've installed locally.

My suggestion would be to make the -U option "unlock" the Maven metadata files so that they are fetched from the remote repository.

I would also appreciate some response from the you guys. What's the problem here that keeps this issue unscheduled for future releases? We haven't received an official comment for a long while. :-|

Removing a private repository before each build would indeed solve my problem
I'm considering the option, since I'm losing my mind with all the build failing due to SNAPSHOT update problem.

Developers new to Maven in our team don't even understand "SNAPSHOT update", they just think they have to install manually each SNAPSHOT dependency... which messes up even more the whole thing.

And I insist, our CI server build agents don't install SNAPSHOT locally, and still they fail to retrieve the latest snapshots in many cases.

It's not a locally installed artifact update problem ONLY.