Maven WAR Plugin
  1. Maven WAR Plugin
  2. MWAR-215

class file JAR inconsistency (archiveClasses and attachClasses options)

    Details

    • Type: Improvement Improvement
    • Status: Open Open
    • Priority: Major Major
    • Resolution: Unresolved
    • Affects Version/s: 2.1-beta-1
    • Fix Version/s: None
    • Component/s: None
    • Labels:
      None
    • Testcase included:
      yes
    • Patch Submitted:
      Yes
    • Number of attachments :
      1

      Description

      Copy-paste from http://article.gmane.org/gmane.comp.jakarta.turbine.maven.devel/94789

      I have a couple of issues with the current WAR plugin and the way it
      packages class files in JAR files (archiveClasses and attachClasses
      configuration options), as these two options work independently of each
      other:

      • archiveClasses moves all class files (and related resources) from
        "classes" directory to JAR file inside WEB-INF/lib directory
      • attachClasses creates a new JAR file in "target" directory from same
        set of files and attaches it to the Maven project with some qualifier
        (by default it uses "classes" qualifier)
        When we deploy artifacts to Maven repository, this results in two
        different JAR files being deployed: one inside WAR and the other
        separate from WAR.

      Problems with this approach:

      • two different JAR files with different MD5/SHA1 checksums.
      • the naming convention for these two JAR files is different

      These issues really start to snag you when you need to perform updates
      after the initial deploy of the WAR. Consider the following scenario:

      • development team hands WAR artifact over to deployment team for deployment
      • development team needs to give an update to the webapp code (the
        dependencies have not changed). One approach is to give the whole WAR
        again, but a more reasonable approach would be to give only JAR file, as
        it contains all the changes and dependent JARs have not changed.

      Whole-WAR-approach becomes especially problematic, when the update needs
      to be uploaded over slow network links and you are in a hurry. However,
      giving only JAR file presents us with some problems:

      1. there is no easy way to check the currently deployed webapp JAR
        version, as the checksum of the embedded JAR file does not match any
        other JAR file in the Maven repository.
      2. as the naming convention differs, it is going to confuse the
        deployment team

      Solution to this mess? Unify the way "archiveClasses" and
      "attachClasses" functionalities work, so they would operate on the same
      JAR file. The way this would work:

      • if "archiveClasses" option is specified, it moves all class files to
        JAR file inside WEB-INF/lib directory (same as before). It would use the
        same naming convention as regular Maven JAR artifacts, with some qualifier.
      • if "attachClasses" option is specified, it attaches that same JAR
        file (as created in previous point) to the Maven project.
      • if "attachClasses" is specified, but no "archiveClasses" - nothing
        happens. Or maybe we should then implicitly turn on also the
        "archiveClasses" option?

      This has some implications for backwards compatibility:

      • naming convention of the embedded JAR file is changed
      • the attached JAR file is no longer created in the "target"
        directory, next to the WAR file (it is now attached directly from
        WEB-INF/lib directory)

        Issue Links

          Activity

          Hide
          Jason van Zyl added a comment -

          What are you plans to preserve backward compatibility? We are taking the same approach with plugins that we are with the Maven 3.x core. You must provide backward compatibility with a path to use the new features. It's unlikely this will be accepted if backward compatibility is not considered.

          Show
          Jason van Zyl added a comment - What are you plans to preserve backward compatibility? We are taking the same approach with plugins that we are with the Maven 3.x core. You must provide backward compatibility with a path to use the new features. It's unlikely this will be accepted if backward compatibility is not considered.
          Hide
          Stéphane Nicoll added a comment -

          Exactly what I said on the dev list.

          Show
          Stéphane Nicoll added a comment - Exactly what I said on the dev list.
          Hide
          Neeme Praks added a comment -

          I can rework the patch, I just need to know, what the behavior should be:

          1. by default turn on the consistent behavior (as the old behavior seems quite broken to me) and the user can enable the old behavior with some flag (e.g. legacyStyleClassesJar=true)
          2. by default leave the old behavior and the user can enable the new behavior with some flag (e.g. consistentClassesJar=true)

          I personally prefer the first approach, but you know better. Also, what should be the configuration option name, to switch between the new and old behavior?

          Show
          Neeme Praks added a comment - I can rework the patch, I just need to know, what the behavior should be: by default turn on the consistent behavior (as the old behavior seems quite broken to me) and the user can enable the old behavior with some flag (e.g. legacyStyleClassesJar=true) by default leave the old behavior and the user can enable the new behavior with some flag (e.g. consistentClassesJar=true) I personally prefer the first approach, but you know better. Also, what should be the configuration option name, to switch between the new and old behavior?
          Hide
          Dennis Lundberg added a comment -

          Usually when we change the behavior of a feature, the old behavior is the default. If someone wants the new behavior the need to change their configuration. That way we maintain backwards compatibility. So option 2 is what we want.

          Show
          Dennis Lundberg added a comment - Usually when we change the behavior of a feature, the old behavior is the default. If someone wants the new behavior the need to change their configuration. That way we maintain backwards compatibility. So option 2 is what we want.
          Hide
          Stéphane Nicoll added a comment -

          Think at it again, the backward compatibility issue is limited. What we could do is allow users to provide (optionally) the name of the embedded jar. As for the location of the file either we create it at the same location and copy it to WEB-INF/lib or we don't but it would only affect custom scripts relying on that specific location. If they do, these are probably old scripts before attachClasses came out so it's a good reason to upgrade.

          Show
          Stéphane Nicoll added a comment - Think at it again, the backward compatibility issue is limited. What we could do is allow users to provide (optionally) the name of the embedded jar. As for the location of the file either we create it at the same location and copy it to WEB-INF/lib or we don't but it would only affect custom scripts relying on that specific location. If they do, these are probably old scripts before attachClasses came out so it's a good reason to upgrade.
          Hide
          Dennis Lundberg added a comment -

          Sure, we can do it like that and make sure to document the new behavior properly.

          Show
          Dennis Lundberg added a comment - Sure, we can do it like that and make sure to document the new behavior properly.

            People

            • Assignee:
              Unassigned
              Reporter:
              Neeme Praks
            • Votes:
              3 Vote for this issue
              Watchers:
              4 Start watching this issue

              Dates

              • Created:
                Updated: