1 |
On Sun, Sep 21, 2014 at 10:02 PM, Frank Peters <frank.peters@×××××××.net> wrote: |
2 |
> On Sun, 21 Sep 2014 20:45:17 -0400 |
3 |
> Rich Freeman <rich0@g.o> wrote: |
4 |
>> You can create device nodes using mknod, and I'd be |
5 |
>> shocked if that ever went away. |
6 |
>> |
7 |
> |
8 |
> But now certain static USB nodes, in particular those for |
9 |
> scanners, have been removed in favor of dynamic allocation |
10 |
> using udev or its equivalents. When this happened I was |
11 |
> certainly shocked, and it could be the beginning of a trend. |
12 |
|
13 |
Fair point, although to some extent this reflects the nature of how |
14 |
modern devices work. Back in the day you had a few serial/parallel |
15 |
ports and you could tell which one was which because they all used |
16 |
different IO ports or IRQs that were hard-coded into the designs. Now |
17 |
you just have one USB host controller which is really the only actual |
18 |
true hardware device on the system and everything that is hooked up to |
19 |
it is virtualized. Plug-and-play really did away with the way device |
20 |
nodes tended to work, and systems like udev are probably the cleanest |
21 |
solution. I for one am happy that I haven't had to configure an IRQ |
22 |
since the 90s. |
23 |
|
24 |
> |
25 |
>> |
26 |
>> Just what is it that you actually |
27 |
>> need the kernel to do for you that you don't think will still be |
28 |
>> around in 20 years? Linus is VERY conservative about removing system |
29 |
>> calls. |
30 |
>> |
31 |
> |
32 |
> There are things which are not system calls that could easily be |
33 |
> changed. It is not too far fetched to consider a time if and when |
34 |
> systemd became so popular and entrenched that the kernel would be |
35 |
> hard-coded to pass control only to systemd and nothing else. |
36 |
|
37 |
That seems extremely unlikely. How many people ran anything other |
38 |
than sysvinit as their init for the 15 years or so before upstart came |
39 |
along? Making the kernel dependent on systemd would defeat the whole |
40 |
purpose of having a separation between userspace and kernelspace. |
41 |
|
42 |
> |
43 |
>> |
44 |
>> If the whole world moves to systemd the biggest problem you'll have is |
45 |
>> that you'll have to write your own service startup scripts, but from |
46 |
>> the sound of things you're doing that anyway. Most of the services |
47 |
>> you probably run aren't linux-exclusive either, so while it seems |
48 |
>> likely that many will start reporting their status to systemd it seems |
49 |
>> unlikely that they will refuse to work without it. |
50 |
>> |
51 |
> |
52 |
> There are a growing number of applications that will no longer compile |
53 |
> without either dbus or udev. In fact, even though I don't use them, |
54 |
> I had to install both eudev and dbus in order to be able to use certain |
55 |
> applications (I just substituted a symlink to /bin/true in place of |
56 |
> dbus-launch to keep that unnecessary daemon from starting). |
57 |
|
58 |
Well, it seems likely that dbus will be a kernel module before long, |
59 |
so it will be readily available. I'm sure there are plenty of |
60 |
programs that don't work if you don't have any number of kernel |
61 |
options disabled. Kdbus is viewed as the future standard mechanism |
62 |
for linux inter-process communication, so programs relying on it |
63 |
should be as surprising as programs that rely on ptys. |
64 |
|
65 |
Much of the issue boils down to the linux world becoming more |
66 |
complex/functional. Back when you could assume that your printer was |
67 |
attached to a parallel port and spoke postscript things were simpler. |
68 |
Today people want to plug in their USB headset and have the computer |
69 |
know to use the USB headset for their teleconference and put the |
70 |
output in the speakers when the phone rings. That just isn't going to |
71 |
work with a world where you output a sound by directing a .au file to |
72 |
a device node. |
73 |
|
74 |
-- |
75 |
Rich |