Possibly offtopic : Binary only driver

[Mike Hearn]

On Sun, 21 Nov 2004 17:55:34 +0100, Arjan van de Ven wrote:
> it's not "the session is dead" but "the machine crashed, possibly
> corrupting the filesystem with it". That's not quite the same...

It's close enough. Boils down to the same thing. Corrupted
filesystems are bad because users might lose important work,
unsaved files due to an X server crash are bad because the users
might lose important work.

Drivers have to be reliable no matter where they are.
 
> bzzzzz. wrong ;)
> There's no way around this; esp since you can't see from a crash what
> caused it... this is why the kernel now prints a "tainted" thing so that
> the kernel developers can just ignore the bug/point the user to the bin
> only module vendor

Somehow the X team manage to debug their server with binary drivers.
Somehow the Wine team manage to debug Wine when running binary programs on
it. Claiming it's not possible says more about kernel development
mentality than any unarguable fact of programming.

> how is that different from "Ok our current driver only works with FC2.
> You'll have to wait a bit for a FC3 compatible driver" that you have now?

It's a matter of time: FC revs very quickly, and is only supported for a
year (well, maybe more with Fedora Legacy these days). Whereas 'stable'
kernel release cycles last a lot longer, or did.

> Esp since udev is NOT a kernel thing (although the 2.6 kernel more or
> less requires udev)

Well, the exact relationship isn't really important. There are other
examples of problems with out-of-tree drivers (not necessarily binary
only) eg ndiswrapper.

> but it's still choice.... where you claimed there wasn't any

A choice that makes nVidia avoidably uncompetitive in the Linux graphics
market isn't really a choice for them. Blame capitalism, don't shoot the
messenger ...