Jason Newton's series targets setups with many worktrees of one repository, such as a fleet of agents that each need an isolated checkout. Objects are shared, but every 'git worktree add' rewrites all tracked files, and a branch can be checked out in only one worktree. Patch 1 adds '--reflink', which on a copy-on-write filesystem reflinks the current worktree's files and index and then runs 'git reset --hard' to rewrite only the paths that differ. A helper in copy.c uses FICLONE on Linux and clonefile() on macOS.
Newton has since sent follow-ups. One fixes a Windows build failure caused by -Werror=unused-function. Another makes reflink_file() refuse on every Windows version, citing git-lfs issue 6312: FSCTL_DUPLICATE_EXTENTS_TO_FILE can publish zeros when the source data is still cached, and no fixed Windows build is known. With that patch, '--reflink' and worktree.reflink=true warn and fall back to a regular checkout, while worktree.reflink=auto stays silent. A further patch has 'git worktree add <path>' record an auto-created branch in worktrees/<id>/owned-branch, so that 'git worktree remove' deletes it.
The review thread has been mostly about the branch behavior. Newton argued that a detached HEAD, the workflow suggested to him, leaves status, push, rename, upstream and symbolic-ref with no branch to work with. An ephemeral branch is a normal refs/heads/ ref while the worktree exists, so he wants it to be opt-in rather than a change of default.
The discussion also turned on conduct. Eric Sunshine said that feeding a reviewer's comments into an agent and pasting the LLM-generated reply is not a good way to convince stakeholders, and that Newton should explain the value in his own words. Phillip Wood said one of Newton's replies fell far short of the standard of constructive discussion expected on the list. He said it attacked the reviewer and did not explain what problem the patch solves. Newton answered that reviewers should not assume which parts are LLM-written, that he has little tolerance left, and that some tools have worked around the current behavior outside Git.
No reviewer has endorsed the approach in the excerpts. Whether the shared-branch and owned-branch behavior can be justified over '--detach' remains open.