Forge repositories in database vs. Git repositories
Forge differs from Emacs packages like Magit and project, in that its first-class objects are remote repositories, not local file system paths. Forge "knows" about a set of remote repositories, which can be listed with M-x forge-list-repositories. Forge adds a remote repository to its known/tracked set upon explicit request, typically N f f from Magit.
A repository may be optionally associated with a local git worktree, but this is not exclusive. It is possible to use the same forge repository with multiple worktrees. The last used worktree will be remembered as the associated one.
Mapping from default-directory to a Forge repository is supported, according to fixed rules (see How Forge Detection Works). In short, if there is only one remote Forge uses that; otherwise it tries the value of forge.remote, then a remote named upstream, if that exists, and finally a remote named origin, if that exists.
Forge stores its state in a central database (forge-database-file). The database is largely cache-like, but not entirely. At least the following information cannot be re-fetched if the database is deleted:
- Choice of repositories, individual topics (in selective workflow)
- Choice of selective mode, or not having fetched topics before some date
- Associated worktree path
- Private per-topic status (unread, pending, done), topic marks (and mark definitions), notes