Re: Kernel-package, fix version 2 ...

[Sven Luther <sven.luther@wanadoo.fr>]

On Mon, Oct 24, 2005 at 11:25:30AM +0200, Jonas Smedegaard wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> On Mon, 24 Oct 2005 10:58:06 +0200
> Sven Luther <sven.luther@wanadoo.fr> wrote:
> 
> > Oh, right, altough your code disregard the fact that i need to check
> > for two return values of the --supported* case.
> 
> It also seems, Sven, that you do not follow the the logic laid out here
> (which we agreed on, except for the use of debconf[1]):
> http://minbar.dodds.net/~vorlon/k-p-pseudocode
> 
> If I understand that correctly, then why don't you?

I don't get this, my yesterday upload indeed didn't follow this, but todays do
it ok, with the exception that we don't wait and ask in case of failure. The
procedure is :

  1) if --supported is supported and returns 0, we keep the tool.
  2) if --supported is supported and returns 2, we drop the tool.
  3) if --supported is not supported, then we do explicit checks builtin into
     k-p for the thre major tools, and drop it if it fails.

This is basically my understanding of Steve's pseudo-code, and Steve also
seemed to not contradict this this morning, i think.

> And it was pointed out earlier to separate arguments in system() calls.

Yep. But i did, did i not ? The code is :

  my @ramdisklist =
    grep {
      my $ret = system ("$_ --supported-host-version=$hostversion --supported-target-version=$version 1>/dev/null 2>&1");
      -x and (
        $ret == 0 or (
          $ret == 1 and (
            (($ramdisk eq "/usr/sbin/mkinitrd") and
             (system ("dpkg", "--compare-versions", "$version", "lt", "2.6.13") == 0)) or
            (($ramdisk eq "/usr/sbin/mkinitramfs") and
             (system ("dpkg", "--compare-versions", "$version", "lt", "2.6.12") == 1)) or
            (($ramdisk eq "/usr/sbin/mkinitrd.yaird") and
             (system ("dpkg", "--compare-versions", "$version", "lt", "2.6.8") == 1) and
             (system ("dpkg", "--compare-versions", "$hostversion", "lt", "2.6.8") == 1))
          )
        )
      )
    }
    (split (/ /, $ramdisk));

As i understand, you can do the separating argument trick if you don't
redirect as the first call does, i am no perl expert though, but that is what
i was told (by you maybe ? don't remember).

> I offer to take a look at all of this, and also look at the perl coding
> style already used by Manoj and mimic that as I best can. Would you
> like that, or would you rather do this yourself?

Cool with me, i would rather someone else had a second look.

> If you agree, I will post so SVN but not upload a new NMU without
> your acknowledgement, Sven.

Ok, i uploaded .3 with the above code, since we need to upload -rc5, but this
can still be changed.

Don't forget to mirror the change in the preinst script too.

> [1] I do not claim that we agreed _not_ to use debconf, just that we
> didn't agree on using it ;-)

Well, Steve made some comment about it being time for k-p to be debconfified,
i still don't get what you have against debconfification, but i agree that it
should be done separatedly from the other fixes.

Friendly,

Sven Luther