YTEP-1000: GitHub Migration

Abstract

Created: March 25, 2017 Author: Lots of folks

The primary source code and project management for yt should be moved from bitbucket to github, and as a result, from mercurial to git. This document outlines a timeline, conversion and import mechanism, and how to manage this transition.

Status

Completed

The yt steering committee has evaluated the broad strokes, and it was presented to the yt community at the beginning of March. There were no objections, and this document is to be iterated on to decide on the migration strategy and timeline.

Detailed Description

As discussed in the email to yt-dev, we should move from BitBucket to GitHub as a result of the pervasive network effect of GitHub and its impact on the community of yt developers.

The process of migration needs to be planned to minimize disruption to developer and user workflow. It will proceed in roughly three stages.

Stage 1

A clone is created on GitHub, and this is evaluated by the developers. The specifics of the conversion procedures (specifically, things like any branch filtering, tag conversions, embedding of hashes and so on) will be examined. A few particular items of note are that the changeset hashes in commit messages and issues are of particular interest. Any commit messsage filtering or file size reduction will be considered. This clone will not be final, and will not be open for pull requests, changes, etc. It will be a one-way and one-time sync.

Whatever methods are used to do this conversion will be written in this YTEP.

An example using fast-export can be found at http://github.com/yt-project/yt_fast_export . This includes git notes for each changeset hash, but they must be pulled specifically using git fetch origin “refs/notes/*:refs/notes/*” .

Possible timeline: Starting April 10, 2017

To Do items during this stage:

  • Create initial clone (this can be done manually)
  • Test migration of issues to GitHub (1:1 mapping of numbers)
  • Set up Jenkins jobs using git and GitHub plugins
  • Rewrite yt update
  • Update documentation
  • Update slack bot
  • Migrate supplemental repositories to GitHub

Stage 2

This is a brief stage, during which we wind down PRs from the Bitbucket repository. All PRs will be either accepted or declined. All PRs that are declined that are still “in progress” will need to be converted to github pull requests, the process for which will be documented. The simplest mechanism will be through hg export and git import, which will squash patches.

Possible timeline: Starting May 1, 2017. This stage is contingent on the Stage 1 To-Do’s being completed, but should be roughly three weeks after Stage 1 is entered.

To Do items during this stage:

  • Accept or decline all PRs on BitBucket. Those PRs that are not accepted by the conclusion will need to be moved to GitHub
  • Switch infrastructure over to GitHub
  • Open up PRs on GitHub and begin code review there
  • Migrate users (ask for their GH handle) and their access levels

Stage 3

This is the final stage. At this point, the switch will be flipped, and no more PRs or code review will be accepted on BitBucket.

Possible timeline: Starting May 15, 2017. This stage is contingent on the Stage 2 To-Do’s being completed, but should be roughly two weeks after Stage 2 is entered.

To Do items during this stage:

  • None that I can think of.

Once Stage 3 has been completed, the bitbucket repository will be marked read-only.

Progress and Notes

During the course of the stages, we will update this YTEP with notes on conversion processes, etc.

Backwards Compatibility

This will almost certainly not break internal APIs other than those identified in the “todo” section above, which are all related to project maintenance like updating and so on.

The developer workflow will break, but we are attempting to mitigate that through this YTEP.

Finally, the documentation (identified as needing to be updated) will be updated to reflect the new normal.