Pharo/Squeak integration with git/mercurial

Mentor: George Herolyants
Second mentor: Göran Krampe
Level: Intermediate
Invited students: Viktor Kerkez, Joel Kitching
Students interested: Joel Kitching(very), Ian Corne, Dmitry Matveev, Mikaël Smaha

Source code management is highly important aspect of every software project. Currently in Squeak/Pharo world developer has very limited choice of such tools. The only available full-featured SCM tool is Monticello. Other alternatives such as changesets lack in some areas.
Monticello in its own turn has some disadvantages: it can only handle Smalltalk code while typical software project has many other artefacts (represented primarily as files) such as documentation, scripts, diagrams and so on, which would be nice to have in the same repository. This project aims to get round this and some other disadvantages by creating a possibility for developers to use modern distributed source code management tools, which are popular outside of the Smalltalk community. The most recognized are git and mercurial.

There is already SqueakSVN project but it aims to SVN integration. SVN is classic client-server SCM. But nowadays this kind of SCM's become less popular, while popularity of distributed SCM's (such as git/mercurial) is still growing. It is also crucial for open source projects like Pharo/Squeak to have distributed SCM to ease development process. Monticello is highly distributed and the new SCM must be distributed as well to successfully compete with Monticello. So git/mercurial fits this goal very well.

Technical Details

This project may go different ways towards it's goal, depending on decisions being made during it.

Initial steps will be:

  • Examining git/mercurial and choose one of them considering simplicity of API and it's ease of use, supported operating systems on and other important factors. There's also pure Smalltalk implementation of Git storage format on SqueakSource, so this step will require examining it as well.
  • Defining and implementing a file mapping format for the Smalltalk code structure (classes, methods and so on), in order to support versioning at the level of separate methods (there is already one defined by SqueakSVN project so this step may require examining it). Or examining the ability to avoid such a mapping and store code directly without needing a working copy on disk.
  • Studying how Squeak/Pharo handles system changes (changes or classes, methods). Then implement handlers for such changes to translate them to working copy using the mapping or through SCM API.
  • (optional but desirable) Implementing Smalltalk interface to the chosen SCM. It can be based on the SCM's command line interface or on it's API. These will require using OSProcess or FFI (or writing a plugin) respectively.

Additional steps may include:

  • Implementing import/export tool for Monticello repositories.
  • Implementing simple user interface for the chosen SCM so that programmer can work with it from within the Smalltalk environment.
  • Integration with widely used OmniBrowser would be even better.

In addition to this a local git/mercurial repository can be used instead of sources/changes files. This will require:

  • Exploring the current mechanism used in Pharo/Squeak to maintain method version history.
  • Replacing it with the developed system which will write changes to the local git/mercurial repository.

Benefits to the Student

The student will:

  • learn about modern source code management systems;
  • study their interfaces, both command line and API;
  • study Pharo/Squeak internal mechanisms related to handling of classes and methods changes and storing of method versions.
  • get experience in usage of FFI or writing plugins for Smalltalk VM;
  • get experience with making GUI in Pharo/Squeak;
  • enjoy writing programs in Smalltalk :)

Benefits to the Community

Developers will be able to choose SCM system grounding this choice on their own preferences. The community will gain access to some famous source code hosting services such as GitHub or Bitbucket. Among other benefits it will increase visibility of the community and the Smalltalk language in whole. It also will lower the entry barriers to newcomers and as a consequence will help community grow.

Getting rid of sources/changes files offers potentially unlimited method version history and avoids so called "condensing changes". And using local git/mercurial repository for this purpose will make managing changes more natural and uniform in small (every method accepting) and in large (changes at the project level).

Updated: 22.3.2010