#546 Mirror Forgejo and Gitea, and commits, branches and forks (#538)

[Anonymous]

Originally created by: athanclark

Part of [#538]. Adds Forgejo and Gitea, and commits, branches and forks, and changes the importer to one repository per timeline with --only/--except. This is stacked on [#544], and needs [#545] (128-character tags) to merge first. GitLab follows in its own PR.

Repositories named on their host

A repository is now a reference: owner/name on github.com, and <platform>:<host>/<path> elsewhere.

	Repository	Item	Person
github.com	github/acme/app	github/acme/app#412	github/mona
Forgejo / Gitea	forgejo:codeberg.org/acme/app	forgejo:codeberg.org/acme/app#5	forgejo:codeberg.org/mona
GitHub Enterprise	github:ghe.example.com/acme/app	…#5
GitLab (next PR)	gitlab/group/sub/app, or gitlab:<host>/… self-hosted	…#12 for issues, …!7 for merge requests

• IDs name the host too (fj:codeberg.org:acme:app:5).
• Routing rules match references, with * for everything under an owner or group.
• Mention titles are written as the platform would from the item: #12, then owner/name#12, then the whole reference across hosts.

Forgejo and Gitea

server/forgejo-source.mjs translates Forgejo's timelines and webhook deliveries into GitHub's shapes, so all the mapping is shared.

• Receiver: it tells platforms apart by their headers, each with its own secret. FORGEJO_WEBHOOK_SECRET is checked against the bare hex HMAC.
• Importer: it uses Forgejo's API, token auth (FORGEJO_TOKEN) and limit paging.
• Labels: Forgejo's deliveries say only that labels changed, so the duration's tags follow without a moment. The importer still records label moments from timelines.

Checked against git.araliasoftware.com (Forgejo 16, read-only): a dry run of athan/openchronology-publicity read items, merges, renames, mentions and links in 6 requests.

Commits, branches and forks

Commits are moments on their repository, at the time they were authored, titled with the message's first line.

• Tags: the platform, the repository's tag, commit, the branch as named (feat/new_feature), and the author's account.
• No hashes: the hash is in the ID and the address, never in a tag. Email addresses are never copied.
• Links: a commit that names an item is linked to it.

Which branch a commit is tagged with:

• Importer: it reads the default branch first, then each other branch only as far as its own commits, and skips branches whose head it has already read. A full import of this repository read 833 commits in 26 requests; before the skip it took 292.
• Receiver: it tags each push with its branch, and keeps commits that GitHub marks as already pushed elsewhere.

Branches are moments when created or deleted (live), or when they began, at their first commit of their own (imported). Forks are moments too.

Choosing what to import

Items, commits, branches and forks are now kinds like the events.

• Importer: --only commits or --except forks,branches, replacing --events and --no-events.
• Routing rules: "only" and "except", the same way.
• Without items: their durations and links are left out, and --only commits reads no items at all.
• One repository per timeline: --repo and --timeline are each given once, and a second --repo is refused.

Tests

• Unit: 26 GitHub and Forgejo tests pass, against fake APIs. They cover references, tags and IDs, Forgejo's timelines, deliveries, signatures, receiver and importer, commits and their branches, branch and fork moments, and the command's arguments. npm test passes apart from contents.test.mjs, which fails locally with git ENOBUFS.
• Real dry runs: against this repository, both commits only and everything, and against the Forgejo repository above.

🤖 Generated with Claude Code

https://claude.ai/code/session_011fFY8eruEi1xAkkMBni13B