fetch

git splice fetch [--upstream <name>] [path…]

Fetches every branch of each splice's upstream, in parallel, into refs/splices/<key>/, and calls out which branch moved. status, diff, log and merge only use what was fetched last.

Someone pushes to pkg-a's upstream:

$ git clone -q "$WALKTHROUGH/upstream/pkg-a.git" ../pkg-a && echo "pkg-a: add a feature" >>../pkg-a/file.txt && git -C ../pkg-a commit -qam "pkg-a: add a feature" && git -C ../pkg-a push -q

$ git splice fetch
ok   vendor/pkg-a fetched (main moved 703b936..5099edf)

Looking at an upstream

fetch keeps each upstream's complete history in the monorepo, under refs/splices/<key>/<branch>, where the key is the upstream's short name in this repository, from its URL: https://git.example.com/pkg-a.git gets pkg-a (key renames it). Every Git command reads it as splices/<key>/<branch>, so there's nothing to clone:

$ git log --oneline splices/pkg-a/main
5099edf pkg-a: add a feature
703b936 pkg-a: a second commit, after the clone
9bb866a pkg-a: seed

$ git show splices/pkg-a/main:file.txt
pkg-a: seed
pkg-a: a second commit, after the clone
pkg-a: add a feature

For a checkout of the upstream, without a network round trip, add a worktree, and remove it when you're done. Commits made there don't reach upstream; changes belong in the monorepo, and push publishes them.

$ git worktree add -q --detach ../pkg-a-upstream splices/pkg-a/main && ls ../pkg-a-upstream
file.txt

$ git worktree remove ../pkg-a-upstream

git log --all shows the upstreams' histories too; git log --exclude='refs/splices/*' --all leaves them out.

Upstreams are fetched and pushed by URL, with no Git remote, so a plain git push can't send the whole monorepo to one by mistake (refs instead of remotes).