[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: RFC: DKMS - Dynamic Kernel Module Support



Le jeudi 11 septembre 2008 à 20:02 +0200, David Paleino a écrit :
> On Thu, 11 Sep 2008 17:52:39 +0000, Tzafrir Cohen wrote:
> > On Thu, Sep 11, 2008 at 07:43:39PM +0200, Josselin Mouette wrote:
> > > Yes, and if dkps depends on linux-headers-2.6-$subarch, that will do the
> > > trick at least for the default kernel. (Depending on just
> > > linux-headers-2.6 is not enough, since linux-headers-2.6.xx-y-$subarch
> > > provides it).
> > 
> > I think you meant:
> > 
> >   depend on linux-headers-2.6.26-1-all
> > 
> > There is no linux-headers-2.6-all-latest

No, I meant what I wrote. If you depend on a specific version of the
kernel, the whole point of this package is moot.

> > For i386 the situation is particualrily bad, as the -all will pull a
> > hosts of other packages.

You need a OR dependency, not to bring all of them. For example, for
i386 this becomes:
Depends: linux-headers-2.6-686 | linux-headers-2.6-486 | linux-headers-2.6-amd64 | ...

All of this is very bad. We need a way to express “I need the headers
installed for all installed kernel packages”. The idea of package groups
only partially solves this; it guarantees the headers are upgraded with
the image, but it does not provide a good way to install the *good*
headers.

For example, let’s say d-i installed the linux-image-2.6-amd64 image,
because you have a 64bit CPU. If you install a package using dkms, and
it pulls the headers. But which headers? APT is not clever enough to
install the ones corresponding to an existing kernel image; it will just
install the first one in the list.

> Sure, but who said to get -all?

At least getting -all guarantees that you have the correct one, so this
is by far not the worst idea that has been thrown. Having a
linux-headers-2.6-all package depending on the default version of -all
would do the trick for most situations, I think.

> apt-get is able to determine the architecture he's running on, right? Anyways,
> dkms is a shells script, it could use dpkg-architecture to get the right string
> to append to the package name. And, with the idea I exposed before, of those
> "triggers", that should be feasible (i.e. "mark the package
> linux-headers-$version-`dpkg-architecture | grep blabla` as to be installed").
> 
> Am I just saying non-sense things? :(

Yes. 

You cannot install packages in a triggered script, or in whatever way
that will be determined from within a package itself. You need to get
*all* the requirements through package dependencies so that it can go in
a single APT run.

Cheers,
-- 
 .''`.
: :' :      We are debian.org. Lower your prices, surrender your code.
`. `'       We will add your hardware and software distinctiveness to
  `-        our own. Resistance is futile.

Attachment: signature.asc
Description: Ceci est une partie de message =?ISO-8859-1?Q?num=E9riquement?= =?ISO-8859-1?Q?_sign=E9e?=


Reply to: