Re: [gentoo-user] Beta test Gentoo with mdev instead of udev; version 5 - failure :-(

[Neil Bothwick <neil@digimed.co.uk>]

[-- Attachment #1: Type: text/plain, Size: 4878 bytes --]

On Tue, 13 Mar 2012 23:43:54 +0000, Alan Mackenzie wrote:

> > udev is not a device node system, it is a device manager. Requiring
> > drivers to handle it gets us into the same mess as Windows, where each
> > driver has to implement the same functionality itself. If a new modem
> > is released with a different USB ID, but using the same driver, your
> > way would require a new kernel, the current approach requires one
> > line to be added to a config file.  
> 
> Right!  Now this is beginning to look like a beginning of an answer to
> my lack of understanding.  ;-)
> 
> This config file - is this the udev "rules"?  Or am I getting mixed up
> with something else?

Yes. They are simple rules under /etc/udev/rules.d that determine what to
do when a device appears on the system, from giving it a specific name or
setting permissions to running a program. It is the latter that causes
problems when /usr is not mounted.

> > How so? It's central to the whole "when do we need /usr?" debate.  
> 
> I meant we'd already had a wide ranging discussion about early /usr.

Yes we didi, but it's not going to go away any time soon :(

> > > > You could use this to argue that /usr should be mounted before
> > > > udev is started, but you could just as well use it to argue that
> > > > udev should not be trying to run such rules at the boot
> > > > runlevel.  
> 
> > > Or that udev shouldn't have "rules".  I still don't understand the
> > > basic concept driving this thing.  My HDDs don't need rules - they
> > > just need a mapping from /dev/sd[ab] into device 8/0 and 8/16, and
> > > the appropriate drivers built into my kernel.  
> 
> > "I don't need it so no one needs it". It sounds like what you need is
> > mdev, but many people want or need more from a device manager. There
> > are many more and varied devices than simple hard disks.  
> 
> That's not fair.  I'm convinced _I_ don't need more than mdev; I'm still
> trying to get a handle on why other devices need more.

OK, here's another example. I have a box here with two USB>RS232
adaptors, connected to completely different devices. The programs
connecting to those devices need to be told which interface to
communicate with, but you can't know in advance which will grab ttyUSB0
and which ttyUSB1. udev rules mean I can give them persistent names
independent of the kernel assignments. One of the devices is a UPS,
imagine the problems if both were UPSes and the software didn't know
which was which - power fails to one and the software sends shutdown
commands to the computers that still have power and let the others fail
when the battery goes flat.

> > What you don't see is why *you* need it, and that's fair enough. Just
> > consider that it does things that others need, even if you don't.  
> 
> I'm not trying to be combative.  In fact, I'm trying not to be
> combative. I'm just trying to get some sort of grasp on what it is I
> don't yet see. I want to understand what udev offers that mdev can't,
> and I'm getting very frustrated about not being able to find the right
> questions to ask.

I wasn't suggesting you were being combative, hence the "fair enough".
You are seeing things from your perspective, as we all do (including the
udev devs who made this decision) and it sounds like you have no need for
any of this. Or maybe you do and don't know you are using it. Many
programs install udev rulesets - for example sane-backends installs a
1500+ line (including comments) rule file, libgphoto's is not much
smaller. all so you can connect a camera or scanner and not have to worry
about how it is recognised or configured.

> > But I still think the requirement for /usr to be mounted is a lazy, if
> > understandable, solution to the way udev's operations are implemented.
> > After all, the vast majority of PC Linux installations out there
> > already use an initramfs.  
> 
> They do, and I understand that one - it is the necessity to have a
> one-size-fits-all kernel in a binary distribution.  As gentooers, we
> don't suffer that constraint, therefore we don't (and shouldn't) need an
> initramfs, unless we want one.

On the other hand, Gentoo's policy has been to follow upstream as closely
as is practical, so if udev upstream wants /usr mounted early, that's
what we do - or do without udev. Both are reasonable choices and Walt
should be commended for his work on this, both on making it work and
raising awareness of the alternative. I haven't yet decided which way
I'll go, I'll probably give mdev a try on a non-desktop box, but it's
good to have a choice and may the best option win.

> > How do I set my laser printer to stun?  
> 
> How about a picture of Marilyn Monroe?

ROFL.


-- 
Neil Bothwick

Will we ever get out of this airport? asked Tom interminably.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 198 bytes --]