[AbirAbbas]

What was wrong

A furrow serve daemon that outlived its client removed its socket path when it idled out, even if a newer daemon had since bound that same path. Afterwards the newer daemon kept listening and kept holding the lock, but no client could reach it: every dial failed with no such file or directory, and a freshly started daemon found the lock held and exited at once. codeaf hit this in a two-computer test (Agent-Field/CodeAF#1982).

What changed

• The server records the Unix socket's device and inode when it binds. At shutdown it removes the path only while it still names that socket, reading the entry without following symlinks. Idle, signalled and requested shutdowns all use this check.
• Windows cleanup and lock handling are unchanged.
• Release 0.3.5: a version bump only, the same edits as 0.3.4.

Tests

furrow-daemon/tests/socket_cleanup.rs (Unix only):

• an older daemon exiting after a newer one bound the same path leaves the newer socket and its lock in place;
• a replacement plain file or symlink survives;
• a missing path is fine;
• an untouched socket is still removed on exit, as before.

Three of these failed before the fix.

🤖 Generated with Claude Code

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}