> For more info see: > <http://www.freedesktop.org/wiki/Software/systemd/separate-usr-is-broken>.
No. Anything on freedesktop.org, and *especially* from the systemd guys, is worth linking to for more information. It is incredibly biased towards huge desktop systems and behemoth software; it is full of bad engineering decisions and (voluntary or involuntary) ignorance as far as smaller systems are concerned. The reality about /usr is that there is *no reason at all* why a separate /usr should be broken; historically, a lot of systems DID have a separate /usr, and that is the very reason why /usr is indeed separate: when mass storage was scarce and expensive, big networked installations had a NFS-mounted /usr, and / contained the necessary software to boot up to the point where /usr could be mounted. Unix was actually *designed* to have a separate /usr. So: * either the article authors are misinformed * or the article authors are dishonest * or their distributions (Fedora, Ubuntu...) chose to install critical software into /usr, which is a mistake * or their distributions chose to treat some /usr software as critical (which is very probable considering they are willing to use a spaghetti monster as process 1, despite all good engineering practices). Please do not refer to freedesktop.org anymore, except as an example of how not to design software. > devtmpfs: the reason to use devtmpfs that I have heard is that it > allows the information about devices that the kernel already has, to > be used directly rather than figuring it out again in userspace (so > you do half the work). Yes, it does that - without the correct Unix permissions, so some userspace tweaking is still necessary. It is a half-hearted attempt at kernel device management, and arbitrarily decides that /dev will be a tmpfs mounting point (which is not a bad idea in itself, but the kernel should not decide how users organize their filesystems). I'd rather have a full userspace implementation like udev that lets me mount /dev whenever I want, as the filesystem type I want. I have done without devtmpfs for years and feel no need to switch. > It also avoids some race-conditions with node > creation if nodes are created from userspace. Race conditions only happen if userspace helpers are forked in parallel as the hotplug system does. If userspace helpers are forked sequentially (which *should* totally be a kernel option), or the information is sent sequentially to a userspace daemon (like with the netlink system), there is no race condition. -- Laurent _______________________________________________ busybox mailing list [email protected] http://lists.busybox.net/mailman/listinfo/busybox
