This avoids, probably among other things, leaving the lock file in place
after a SIGINT interrupts a sync DB update.
Fixes regression introduced in 4f8ae2b.
Signed-off-by: Dave Reisner <email@example.com>
Dan this seems to work (surviving a variety of -Sy and -Syu <pkg> testcases)
and pactest doesn't complain, but you know this portion of the code way
better than I do. Does this seem sane?