Uploaded image for project: 'Jenkins'
  1. Jenkins
  2. JENKINS-33274

"Base build" link for first build of branch project

    XMLWordPrintable

    Details

    • Similar Issues:

      Description

      The first build of a branch project should have a UI link like Previous Build but pointing to the build on the project corresponding to the base branch (such as master for Git) which its SCM revision links to. Should use SCMSource.parentHeads; major SCM plugins needs to implement this API. When present, the base build should be used as a baseline for determining the SCM changelog, so that a user can see the changes made in the SCM commit initiating a branch. An API should also be introduced to core so that plugins such as junit can use this backlink rather than previousBuild for purposes of comparing test results; the same could be true of static analysis, code coverage, etc.

      Merely changing Run.getPreviousBuild to return a build from another Job is unsafe; many plugins may assume that the Job pairs match, or that getNextBuild is the inverse operation, etc. Perhaps we need an API addition to Run such as

      /**
       * Gets the previous build to be used for purposes of comparison of SCM commits, test results, and other reports.
       * If {@link #getPreviousBuild} is not null, that should be returned.
       * Otherwise, this may instead return a build of a {@link Job} of the same type representing a base branch from which this branch was forked.
       */
      public @CheckForNull RunT getPreviousLogicalBuild() {
          return getPreviousBuild();
      }
      

      which would most closely preserve existing semantics: build-to-build comparisons.

      Additionally/alternately:

      /**
       * Gets a build to be used for purposes of comparison of SCM commits, test results, and other reports.
       * By default, {@link #getPreviousBuild}.
       * For a job representing a branch forked from a base branch, this should return the build of the branch point.
       * Branch points may change as a base branch is manually remerged into a feature branch.
       * And jobs may be automatically building against the proposed merge of a branch,
       * in which case the base build is likely to be the latest build of the base branch.
       */
      public @CheckForNull RunT getBaseBuild() {
          return getPreviousBuild();
      }
      

      which would allow a plugin to display the cumulative changes from the base branch, which may be more relevant.
      (For example, in a build of a PR you probably want to know whether there are FindBugs regressions relative to the base branch, not whether the 13th commit to the PR introduced regressions relative to the 12th commit.)

        Attachments

          Issue Links

            Activity

            Hide
            abayer Andrew Bayer added a comment -

            I'd say implement both - I can think of more places I'd use the base build rather than previous logical build, but the semantics of the two would be different enough that they'd probably both be useful.

            Where would these implementations live? In things like MultibranchJob?

            Show
            abayer Andrew Bayer added a comment - I'd say implement both - I can think of more places I'd use the base build rather than previous logical build, but the semantics of the two would be different enough that they'd probably both be useful. Where would these implementations live? In things like MultibranchJob ?
            Hide
            abayer Andrew Bayer added a comment -

            This is not actually that worthwhile a feature to add, in practice, and is hairier to implement than I thought two years ago when I first looked into it, so I'm closing this as won't fix.

            Show
            abayer Andrew Bayer added a comment - This is not actually that worthwhile a feature to add, in practice, and is hairier to implement than I thought two years ago when I first looked into it, so I'm closing this as won't fix.
            Hide
            teilo James Nord added a comment -

            can you define worthwhile?

            I would say not introducing new spotbugs issues into a build is very worthwhile.  

            given a PR build has no comparision spotbugs will treat all issues as new, which means you have to fix every single issue, or just ignore them and hope you don;t introduce new ones (when you are starting on a non greenfield project that has say 1000 spotbugs issues and you want to ratchet them down over time, or at least not allow new issues to be introduced).

             

            There are other use cases as well but that above is pretty worthwhile.  (you can s/spotbugs/checkstyle/) and there are likely many other use cases too that require this support.

            Show
            teilo James Nord added a comment - can you define worthwhile? I would say not introducing new spotbugs issues into a build is very worthwhile.   given a PR build has no comparision spotbugs will treat all issues as new, which means you have to fix every single issue, or just ignore them and hope you don;t introduce new ones (when you are starting on a non greenfield project that has say 1000 spotbugs issues and you want to ratchet them down over time, or at least not allow new issues to be introduced).   There are other use cases as well but that above is pretty worthwhile.  (you can s/spotbugs/checkstyle/) and there are likely many other use cases too that require this support.
            Hide
            abayer Andrew Bayer added a comment -

            Reopening at James Nord's request, but I don't expect that we'll be working on this at any time in the foreseeable future.

            Show
            abayer Andrew Bayer added a comment - Reopening at James Nord 's request, but I don't expect that we'll be working on this at any time in the foreseeable future.
            Hide
            jglick Jesse Glick added a comment -

            Yeah the only real workaround for the example James Nord gave is to use Sonar to track regressions and give up on doing it from Jenkins.

            Show
            jglick Jesse Glick added a comment - Yeah the only real workaround for the example James Nord gave is to use Sonar to track regressions and give up on doing it from Jenkins.
            Hide
            jglick Jesse Glick added a comment -

            Sounds like Ulli Hafner has implemented something like this privately in warnings-ng.

            Show
            jglick Jesse Glick added a comment - Sounds like Ulli Hafner has implemented something like this privately in warnings-ng .

              People

              • Assignee:
                Unassigned
                Reporter:
                jglick Jesse Glick
              • Votes:
                4 Vote for this issue
                Watchers:
                12 Start watching this issue

                Dates

                • Created:
                  Updated: