downgrade PR builds maven from 3.3 to 3.2

[Kanstantsin Shautsou]

I see regressions with maven-3.3 itself for docker-plugin builds - maven-hpi (or shade or whatever else is happening) is picking unshaded jar into final package that ends with class errors/hell.

CB PR  wrong build:

 [INFO] --- maven-enforcer-plugin:1.0.1:display-info (display-info) @ docker-plugin ---

[INFO] Maven Version: 3.3.3

[INFO] JDK Version: 1.7.0_79 normalized as: 1.7.0-79

[INFO] OS Info: Arch: amd64 Family: unix Name: linux Version: 3.9.4-1-mansion

My local right build:

[INFO] --- maven-enforcer-plugin:1.0.1:display-info (display-info) @ docker-plugin ---

[INFO] Maven Version: 3.0.5

[INFO] JDK Version: 1.8.0_40 normalized as: 1.8.0-40

[INFO] OS Info: Arch: x86_64 Family: mac Name: mac os x Version: 10.10.4

[INFO]

[INFO] --- maven-enforcer-plugin:1.0.1:display-info (display-info) @ docker-plugin-parent ---

[INFO] Maven Version: 3.2.5

[INFO] JDK Version: 1.8.0_40 normalized as: 1.8.0-40

[INFO] OS Info: Arch: x86_64 Family: mac Name: mac os x Version: 10.10.4

Could somebody downgrade it to 3.2.*?

/cc @stephenc

[Kanstantsin Shautsou]

Note, java version unrelated, bad build locally reproduced:

[INFO] --- maven-enforcer-plugin:1.0.1:display-info (display-info) @ docker-plugin-parent ---

[INFO] Maven Version: 3.3.3

[James Nord]

Is this due to a maven 3.3.3 bug (ie can you reproduce it locally) or is this due to a bug in the jenkins maven plugin injecting something wierd?

[James Nord]

The version has been changed to 3.2.1 which was AFAICT the previously used version.

Generally we should be using the latest version - Kanstantsin can you see if this is a maven/maven plugin issue or a hpi issue so we can track this somewhere so we are not stuck on 3.2.1 forever more.

[Kanstantsin Shautsou]

As i wrote, it reproduced locally. 

git clone docker plugin and run mvn package using different maven versions and then compare docker-plugin/docker-plugin/target/docker-plugin.hpi/#WEB-INF/lib

https://www.dropbox.com/s/s9gwdbi4b15n9r9/Screenshot%202015-07-30%2013.29.09.png?dl=0 left side is wrong 3.3.3, right is expected.

Nevertheless 3.3 produces not the same results as 3.0-3.2

This can be related to the reason why @oleg-nenashed decided to implement this feature https://github.com/jenkinsci/maven-hpi-plugin/pull/18

[Kanstantsin Shautsou]

Let's wait for stephen comments, i found some filled bugs from him with 3.3 regressions. And frankly speaking i think that 3.3 messed somehow dependencies tree...

[James Nord]

possibly https://issues.apache.org/jira/browse/MNG-5807

[Stephen Connolly]

looks likely. If the shaded artifact was in a separate reactor you'd be fine. I think the bug here involves resolving these from the reactor. The bug is not fixed in 3.3.6-SNAPSHOT so it would be good if somebody had a fix for it before JvZ gets an alternative fix to MNG-5840 implemented and cuts 3.3.6

[Kanstantsin Shautsou]

Stephen, I believe in you!