Git Branching Models

Our current branching model is not optimal and makes keeping track of versions hard. This page discusses various branching models.

Models

Current SVN-like Model

• master is the development branch which contains incompatible code as well
• A version branch e.g. 1.5.2 is created when releasing a new stable version
• 'Stable commits' are merged from master frequently (==> rolling release like)

Pros

• Easy management of incompatible code (==> rolling-release works well)

Cons

• Prone to losing track of differences between master and version branches
• Doesn't integrate well with popular Git models implemented on e.g. GitHub

Feature Branches

• New, maybe incompatible features are implemented in separate branches and merged as soon as they are stable
• master is mostly stable
• When releasing a new version, a tag is created from master
• Release intervals are smaller

Example (adding a big feature)

• Current version is 1.5
• Someone wants to implement complicated things which are netcode-incompatible with 1.5
• Someone creates feature/complicated-things branched from master
• Work is now done on feature/complicated-things
• Once feature/complicated-things is ready it needs to wait for 1.6
• (time passes)
• (1.6 is ready for release)
• master is tagged as 1.5.x-final marking the end of 1.5 compatibility
• master version number is increased to 1.6
• feature/complicated-things is merged into master
• master build gets released to the public

Pros

• "Git"-way (==> integrates well with GitHub)
• No weird cherry-picking is necessary

Cons

• We have to modify our nightly system and build server settings to support feature branches somehow (required for properly distributing tests)

Comments

• My concerns over Feature Branches are largely based upon any added bureaucracy or complexity to making contributions. --Talidan (talk) 18:04, 9 August 2016 (UTC)