From: Junio C Hamano <gitster@pobox.com>
Date: Fri, 13 Jul 2007 18:23:07 GMT
Subject: Re: [PATCH] lockfile.c: schedule remove_lock_file only once.
Message-ID: <7vabu0noxg.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <20070713141450.GA8392MdfPADPa@greensroom.kotnet.org>

Sven Verdoolaege <skimo@kotnet.org> writes:

> Removing a lockfile once should be enough.

Yeah.  I wonder what we were smoking.  415e96c8 which introduces
the atexit to index.c does:

    int hold_index_file_for_update(struct cache_file *cf, const char *path)
    {
           sprintf(cf->lockfile, "%s.lock", path);
           cf->next = cache_file_list;
           cache_file_list = cf;
           if (!cf->next) {
                   signal(SIGINT, remove_lock_file_on_signal);
                   atexit(remove_lock_file);
           }
           return open(cf->lockfile, O_RDWR | O_CREAT | O_EXCL, 0600);
    }

whose intent is exactly "do this once, only for the first one".

The reason we do not use lk->next but instead check lk->on_list,
and the reason why we do not remove the lock from the list, are
described in 1084b845.

But your "fire atexit() once" fix is needed.  Thanks.