Talk:PMC proposals/Move Apertium to Github

From Apertium
Jump to navigation Jump to search

Reasons to Switch

  • GitHub’s excellent issue tracker
  • More people outside Apertium are far more familiar with Git vs. SVN (especially younger folks, see GCI/GSoC)
  • More people outside Apertium have GitHub accounts, easier to start-up for a new user
  • GitHub’s interface is far superior to SourceForge’s interface
  • Avoids SourceForge’s downtime (not so bad lately)
  • SourceForge gives an awful impression
  • More visibility as an FOSS project (people browse GitHub)

Prevailing Approaches

Common pros/cons are excluded for the sake of brevity.

Approach 1

Variant A

A monorepo with all the lingustic data, pairs and language modules. Other folders in SVN like the core engine and peripheral tools (e.g. APy) would live in their own repos.

Pros

  • Large-scale editing of e.g. 15 pairs is easy.
  • There are no meta-repos or submodules to deal with.
  • GitHub’s interface can be used directly.
  • Less need for extremely complicated git commands
  • Possible to do partial checkouts using SVN

Cons

  • The monorepo would be massive (> 3 GB).
    • Most devs (aside from the couple core devs) would have to use GitHub’s SVN bridge to work on a pair.
      • This is highly contradictory to working on GitHub
      • People new to Apertium will have to learn SVN, negating some reasons to switch
      • ALTERNATIVE VIEWPOINT: there is no "learning SVN", it's three commands.
    • Diluted usefulness of branches, PRs and hooks
    • GitHub doesn’t necessarily allow repos larger than 1 GB (unclear whether this limit refers to bare repo). If GitHub decides to stop us at some point after we switch, that’s really bad.
  • Everyone will disable email notifications (“watching” a repo) since there will be too much spam
  • Massive number of issue labels to curate and apply (non-members cannot tag an issue when submitting), reducing the effectiveness of the issue tracker
  • Commit access will continue to give write access to everything
  • Contradictory to the Git/GitHub philosophy (bad impression)

Variant B

Several monorepos, one for each of:

  • incubator
  • pairs
  • languages
  • tools

Pros

  • Large-scale editing of e.g. 15 pairs is easy.
  • There are no meta-repos or submodules to deal with.
  • GitHub’s interface can be used directly.
  • Less need for extremely complicated git commands
  • Possible to do partial checkouts using SVN

Cons

  • All the cons in Variant A, minus:
    • Repos will be smaller than the massive monorepo
  • Moving a package between release states wrecks its history
    • Or is more complicated to preserve the history ?

Approach 2

Individual repos for each pair, language module and tools. A couple of “meta-repos” that contain submodules pointing to collections of repos, e.g. apertium-staging would contain ~8 submodules pointing to each of the pairs in SVN’s /staging and apertium-all would have submodules to apertium-staging, apertium-incbuator, apertium-languages, etc. This hierarchy would be maintained via GitHub’s repo tags (a.k.a. “topics”), i.e. apertium-xxx-yyy could be marked with the incubator tag to end up in apertium-incubator.

Pros

  • Usable issue tracker for each repo
  • Fits into the Git/GitHub philosophy
  • People who wish to use Git can contribute using that (while it's still possible to use the SVN bridge for those who want that)
  • Familiar branching, PR and hooks that work as expected
  • Email notifications and watching repos is useful
  • An analogous change to 15 pairs will result in 15 different commits, each repo has its own history (both pro and con).
  • Granular permissions (not everyone has access to literally everything, especially useful for GCI/GSoC)
    • RESPONSE: Could be considered more bureaucratic

Cons

  • Harder for people who make changes to lots of pairs at the same time (i.e. couple of core devs)
    • Commands are more gnarly (git submodule can be pretty unintuitive)
      • RESPONSE: Possible to mitigate with aliases and cheat sheets
    • An analogous change to 15 pairs will result in 15 different commits, each repo has its own history (both pro and con).
  • Somewhat harder for people who use the meta-repos
    • RESPONSE: It’s really not that difficult to checkout (git submodule update --recursive --init) and pull updates to a meta-repo (git pull --recurse-submodules) and with aliases it can be even shorter.
  • Requires tooling to keep meta-repos up-to-date
    • RESPONSE: These are super simple scripts based on GitHub’s reliable API. Sushain is willing to write them and Tino is willing to host (and perhaps code review).
  • GitHub doesn’t provide a nice interface to view repos in a tree format
    • RESPONSE: Sushain will though! See this page that can be trivially finished to cover all our repos and is a very simple single HTML file (and pretty IMO). This page is automatically generated from the repo tags.

Related Concerns

  • Mailing lists - should probably be preserved on SourceForge for now until/unless we choose to switch to another solution or self-host them.
  • Existing issues - Sushain volunteers to manually transpose (or find an automatic solution) to moving our existing issues (pretty small #)