ATNF VLBI Wiki

A zoom link will be provided via email to the difx-users list. The meeting will be recorded and made available shortly after conclusion.

• Date: Monday November 16 2200 UTC
• Title: Migrating from SVN to git: challenges and opportunities
• Agenda:
  • Opening (5 minutes, Adam Deller)
  • Automated migration - current status (30 min + 20 minutes questions, Greg Lindahl)
    • Includes both git migration and continuous integration
  • Discussion session (30 minutes, moderated by Greg)
    • Includes the context of work on GPU-DiFX currently being undertaken (under git)
    • Potential workflows for using git and how they differ to how we have traditionally used svn
    • Goal: go / no-go decision on migration, along with a timeframe
  • Planning for future sessions (15 minutes, Adam Deller)
    • Dates and times
      • option 1 - get in before Thanksgiving, Tuesday November 24 ~0800 UTC or ~1600 UTC
      • option 2 - wait until after Thanksgiving, Tuesday December 1 ~0800 UTC or ~1600 UTC
    • Topics
      • Request from Leonid to discuss mixed-polarisation (post-)correlation
    • Organisers (Adam is happy to coordinate zoom links etc, but would appreciate sharing the coordination around).

As transcribed by Adam, please let me know of any errors that need to be corrected.

• The DiFX codebase could use a clean-up in conjunction with the transition to git.
  • This should involve minimally removing a few large files that are used for testing (placing them in a different repo)
  • Another option is to split out a couple of packages that are used more widely beyond DiFX itself (potentially mark5access, vdifio, codifio)
• It is highly desirable that we retain the email notifications of commits after transitioning to git (Greg noted that this is possible, but would need to look into how.)
• Documentation of the git workflow to be used will be essential. Greg suggests that a 1-page “cookbook” should be provided on the github site, for developers to follow.
• Where should we store the binary data for running continuous integration tests?
  • Small datasets can test most functionality and could be hosted on github.
  • Hosting in a place physically close to where the CI will run (california) would be advantageous
  • GPU tests would be done in a more ad hoc way
  • It would be possible to host larger datasets for a more lengthy comparison at an individual correlator site and have those (larger) tests be run locally in a non-continuous way (at release time, say). Adam's post-meeting note: It would be good to have each of Bonn, NRAO, Haystack, and Swinburne (plus of course any other interested sites) commit to hosting one larger dataset for infrequent testing.
• The one-way conversion and history preservation from SVN to git that Greg has provided means that it is not necessarily essential to have a hard date to work towards for a jump to git. As long as people don't habitually have lots of unchecked-in code, anyway. The following should be a minimum for a “go-live” making git the primary repo:
  • Testing datasets have been created and are available
  • Documentation on how to check in / push / merge on git is available. Ideally this would be on the github page itself.
  • Continuous integration is live
• The SVN site should stay available, read-only, after the transition to git is complete. Obviously with a big warning to direct people to the new site.
• There should probably be at least one more minor DiFX release prior to the transition to git, which would include bugfixes for VGOS complex LSB modes, and Jan's outputbands capabilities.
• Propose to have a working group for git transition that identifies final structure for git repo, which would report back a proposal at one of the future mini-workshops (and via email) for endorsement. Proposed members: Greg L. (chair), Adam D., Shaoguang, Chris P. or Cormac R., Mark W., Geoff C., a representative from Bonn. Others welcome.
• Github accessibility is not consistent in China. Is this a roadblock?

• All difx-developers to provide Greg L. with github usernames
• Working group membership should be finalised and tasks allocated.
• Adam to follow up with Shaoguang about potential barriers to using github in China.