On Sunday 17 May 2009 03:33:22 pk wrote: > Alan McKinnon wrote: > > As I see it, at the bottom of the stack you have a kernel and at the top > > a user space app (the X server will do for an example). Plug in a USB > > device that the app can use, and the kernel needs to make a node in /dev > > for it if it's not already there. The kernel should not be interrogating > > the device for all possible info - that is expensive - and doesn't need > > to. It only needs enough info to know what driver, major and minor > > numbers to use. X OTOH, can > > I couldn't agree more. And this is what Udev, as a user space app, does. > The only thing it doesn't handle is communicating with other user space > apps; this is currently Hals job. > > > the current model uses udev as the interface to the kernel's nodes and > > HAL as the interface to exactly what hardware you have. Seems pretty sane > > for the most usual use case. At some point in the stack you will need the > > OS-dependant part, my guess is the best place is between hal and udev. > > Only Linux uses > > Well, as I understand it this is what it looks like today: > > kernel <-> udev (or equivalent for non-linux kernel/OS) <-> hal <-> dbus > <-> user apps > > To me that seems a bit redundant...
No, there's nothing redundant in that. udev talks kernel-speak, hal talks userspace-speak. Hal could be distro-agnostic, udev can't be (not in a sane environment) and dbus is simply a transport layer for messages. That's five different functions going on, and none of them logically belong with any other in the same layer. > What I would like to see: > > kernel <-> udev <-> user apps Then X must talk to udev directly. Two problems: - only Linux has udev. Other OSes may not need, want or be willing to touch udev with a bargepole. - X is multi-platform. Good luck getting Keith to agree to make it essentially Linux only :-) > Or at the most: > > kernel <-> udev <-> daemon <-> user apps. But you have that in the current setup. Hal (for better or worse) is the daemon. dbus is simply a message transport and can be omitted from the conceptual diagram > > udev, but all OSes use something in that spot. And if not, they have > > static nodes. > > Yes, but if the developers could agree on a common API for the udev > daemon and it's equivalents on other platforms (what does BSD use?)... > Or if they could agree on using "Hal v2" (rewritten from scratch with no > or a minimum of dependencies). I don't think that's feasible. Easiest would be the bottom layer of hal has OS-specific code to talk to udev (or it's equivalent in other OSes). Or perhaps a plugin/module type system. > > Meanwhile we have an acknowledged problem with hal - it's too complex, > > too many things have been shoved into it that were never catered for in > > the design, configuration is horrific - and the devs are having their > > usual spirited debate about how best to approach a solution. This is > > perfectly normal and perfectly healthy > > Yes, I guess so. Since I'm (currently) not in the position to help out > I'll have to live with whatever they come up with. But sometimes it's a > bit frustrating... Sorry for the ranting. Hey, it could be worse. You could be forced to use Windows... -- alan dot mckinnon at gmail dot com
