testtmp.Sweep judged a template directory's owner dead with kill(pid,0)
on the pid in its name. A pid only means something inside one PID
namespace, and /tmp is shared across namespaces. A test binary run under
bwrap --unshare-pid got ESRCH for every host pid and deleted live
storetest templates mid-run ("copy template db: no such file"). In the
other direction, a pid recorded inside a sandbox can name an unrelated
host process and keep a dead owner's directory forever.
MkdirTemp now builds the directory under a staging name no sweep
matches, takes a flock on owner.lock, and renames it into place. The
lock follows the inode across the rename, and the fd is O_CLOEXEC and
held for the life of the process. Sweep removes a directory only when
it can take that lock. A directory with no owner.lock (made before this
change) keeps the old pid rule. Windows is unchanged.
Claude-Session: https://claude.ai/code/session_01Hk4vxsaKDm3E1LvsoHtHeV
* fix(test): template and fixture temp dirs no longer leak into /tmp (BUG-3412)
The store test templates are named for the owning pid, and each build
first removes the template dirs of processes that are gone (and pid-less
legacy dirs over a day old), so a test binary killed before TestMain's
Cleanup no longer leaves its template on the tmpfs for good. The fence
source tests remove their fixture dirs when the file finishes.
Claude-Session: https://claude.ai/code/session_01Q9WssCjFuHjSfLABoW4tmS
* fix(test): never sweep pid-less template dirs; age proves no owner gone (codex r1 on BUG-3412)
Claude-Session: https://claude.ai/code/session_01Q9WssCjFuHjSfLABoW4tmS
xarmian