[Pkg-xen-devel] About hotplug/udev

[Ralph Passgang]

Am Montag, 20. Februar 2006 19:31 schrieb Julien Danjou:
> On Mon, Feb 20, 2006 at 06:47:49PM +0100, Ralph Passgang wrote:
> > on sarge? sarge's udev is too old (that's why I added this dependency
> > at all).
>
> Yes, it was on sarge.
> But this dependency is satisfied on sarge with hotplug version
> included. So it seems not ok.

yeah, what to do about that?

on sarge either hotplug or hotplug and udev ist installed. udev cannot be 
installed alone (as long as you stay with the default sarge version).

> > If you want to use udev on sarge in combination with xen then you have to
> > use a backported udev in any case.
>
> This is what I done, but I had to discover it by myself, because the
> dependency was satisfied by hotplug.
> Is hotplug too old on Sarge ?

no, hotplug is fine.

hotplug works on sarge, etch and sid... udev only on etch and sid.

that's why I tried to help a bit with setting the hotplug | udev dependency to 
"hotplug | udev (>= 0.0.59)". Even if it doesn't prevent a user from 
installing sarge's udev version, it might help to realize faster what is 
going on. I haven't realized that a conflict might be better.

> > were you forced to remove udev to install my xen3 packages? if so, then
> > we should change the dependcy to "hotplug | udev" again, because it's not
> > a problem to have sarge's udev version installed, but in this case
> > hotplug is used by xen and udev ignored.
>
> No, if I recall correctly, I had to upgrade it. But I think I did not
> try to use only hotplug instead of udev. I know that I wanted udev
> installed.

so, we seems to need a Conflicht for "udev (< 0.059)". That would prevent 
sarge users from even triying sarge udev version in conbination with xen...

> Maybe there is no problem currently, but since I had some troubles with
> your packages and udev/hotplug last week end, I prefer to tell it right
> now. ;-)

absoloutly correct. I created these packages for me and the company I am 
working for, thinking adam will release his version soon. As this has not 
happend for some weeks I decided to publish it so debian users has an easy 
alternative to compile xen from source.

I never had tested these packages on all possible combination and haven't 
really had a policy in mind or something like this. the must stuff was still 
adams works. But with xen 2.0.6 there was (afaik) no hotplug | udev at all, 
as well as there were no xenstore and no vmxloader or even pae and amd64 and 
so on.

I was quiet happy to get all this working, but I always knew that there is 
more work needed to get it perfect, so I always asked for feedback.

if you have something that can made be better, just do it or at least tell/ask 
on the ml :)

> Anyway this is only a Sarge issue.

correct, but if possible we should help backporters and sarge users... to be 
honest, of course we do this for sid and etch, but the most users that will 
use xen in an production enviroment in combination with debian will most 
likely use sarge and will be happy about good backports.

so if we can make it better then I did it with my dependency, then go ahead...

> > We could add a udev version check into xendomains or the xend
> > init-script, so that sarge users will be informed that their udev version
> > is to old and hotplug will be used.
>
> Could be fine, but in fact we have to follow unstable, not really sarge ;)

correct, I already also dislike my idea, but not because of that, just because 
I don't really want to hack xendomains for something stupid then printing 
warning.

if this is important it has to be documented well, that should be enough (in 
combination with the best possible dependencies & conflicts).

> > but that rembers me of something that we should discuss too:
>
> Fine ! ;)
>
> > we have "hotplug" as build-dependcy. by doing this we can be sure to have
> > the hotplug etc scripts installed by the normal make all/dist call. But
> > because we want to have also udev support I added the "make udev-install"
> > line in the rules file. But is it really clever to go this way? Wouldn't
> > it be better to not build-depend on hotplug (I guess for compiling it's
> > not needed in any way) and then call "make hotplug-install" & "make
> > udev-install"?
>
> Really really really clever.
> Having a build-dep on hotplug or udev is ugly I think. (I did not see
> this). Furthermore, hotplus is now provided by udev, IIRC.

because udev provides hotplug it compiled anyways, indeed... :)

but you hadn't any hotplug scripts installed then, so my old approach was 
broken. it was just really working (as expected) on buildhost without udev. 
but that is fixed in svn :)

but interessting, this is why "Provide" is complexer than a "X | Y" 
dependency. We have discussed that topic for "xen-hypervisor | 
xen-hypervisor-pae" vs "provides: xen-hypervisor" (for the pae version) 
already.

Now I am against the "Provides: xen-hypervisor" thing. Possible future 
dom0 .deb kernels could have "depends: xen-hypervisor" or "depends: 
xen-hypervisor-pae" to be sure just to be installed when the correct 
hypervisor is installed. if we use the provide instead, then we would need to 
use conflicts in the kernels to prevent that situation... is much more 
complicated, isn't it? (sorry for bringing that up again)...

--Ralph