On Saturday, 5. May 2001 01:16, you wrote:
> > > One way to address this has been prototyped for printers: let the
> > > hotplug subsystem tell the scanner subsystem about device types,
> > > since it's already got that data available:
> >
> > This does not help you on installation, which causes most problems.
>
> Depends on the particulars. I've worked with hotplug-like systems
> where there's a straightforward bit of logic that goes like
>
> if (first time I've seen this device) {
> run first-time setup program;
> }
Installing the device this works great.
It's useless during installation of the software.
Sorry about the confusion.
> > > Or, it could go the other way around. If you're scanning /dev/...
> > > nodes, add a corresponding ioctl to find which USB device is
> > > involved, and then use the existing USB support to find VID/PID.
> >
> > This is very helpful.
> > We should do this as generic as possible.
>
> I find it no less helpful than doing it the other way around, in fact.
> Both are generic approaches. One of them works well with
> today's hotplug, which seems healthy to me.
Well, actually no, they are not.
Working through usbdevfs is limited to usb obviously.
There are eg. cameras with both, usb and firewire.
Secondly going through the device node is easier
Open device
Check type via ioctl
vs.
Scan through usbdevfs
Try opening indicated device
recheck device was not replugged
>
> > > And there's also Alan's suggestion, to do scanner drivers entirely in
> > > usermode, focussing development work on usbdevfs and APIs to it.
> > > (I think the Java USB API is more powerful than libusb, but that can
> > > just be a hacker challenge ... :)
> >
> > 1. No persistent device permissions
> > 2. Some scanners have interrupt endpoints
> > 3. Very hard to use without hotplugging, which is beta at best
> > 4. There was this little problem with control messages and usbdevfs
>
> Johannes commented on #2/#4 ... Re #1, device permissions do
> need more work in any scenario (particularly with the current mount
> options bug!!) but hotplugging can easily chmod things. Re #3, that's
> just not true ... the usermode API wouldn't depend on hotplugging
> except maybe for device permissions.
Which are essential.
With the current model I can do 'chmod 660 /dev/usbscanner0' and the issue is
settled. You can't beat that for simplicity ;-)
All the usual tools work, etc.
The hotplugging tools are great, but in many cases utter overkill.
> > In this case I disagree with Alan.
>
> You just want to create and fix a different set of bugs ... not
> that I'm sure what it would be! :)
I find enough bugs, without any alterations ;-)
I keep seeing race conditions. I hope this doesn't get unhealthy.
Regards
Oliver
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
http://lists.sourceforge.net/lists/listinfo/linux-usb-devel
