On Mon, 13 May 2002, MAEKAWA Masahide wrote:

> Julio Merino <jmmv@hispabsd.org> wrote:
>
> The mechanism to construct /dev device nodes dynamically is attractive.
>
> IMHO, FreeBSD's one is not good because it depends on many hard-coded
> majors/permissions. If I implement it, it runs as a daemon on userland
> and it is optional and we can run it in run-time. To run it in run-time,
> I must implement a mechanism to get majors <-> device in-kernel
> relationship from userland. (by using devprop?) It is not trivial.
>
> Anyway, out current NetBSD lacks some core mechanisms, I think.
>
> But to construct MAKEDEV from majors list, as Darren points out, is
> easier than it. It is not trivial, too.
>
> O.K. I will consider about this.

Is that a "I will consider this" about MAKEDEV, or about devfs?

If it's about MAKEDEV, please do if you can. That'd be cool as it would
further reduce the number of places we have to keep major numbers (and the
number of places we can get them out of sync).

If it's about devfs, please don't let devfs slow down the other work
you're doing. The devsw cleanup you're doing will make lkms much better,
and cure problems we've had in other places (like devices being left out
of the char<->block conversion). devfs will take a while, as the big thing
we need is a way to permanently identify a device (thus to make sure the
permissions you assign to the device one day come back to it even if it
has a different major or minor number).

Take care,

Bill