[Pkg-pciutils-discuss] Updating of pci.ids in Etch

[Matthew Wilcox]

Hi Andi, Martin

We have persistent requests to update pci.ids either automatically or
frequently, and as stable release managers, I wanted to ask your opinions
on how best to handle the situation.

Some misc facts:
 - pci.ids is currently 404k of the 660k pciutils binary package
 - The pciutils udeb contains a stripped-down pci.ids file which is a mere 236k
 - pci.ids is updated regularly upstream and can be retrieved from
   pciids.sf.net with a script (update-pciids).

Now, since pci.ids is only useful for mapping numbers to names, updating
pci.ids doesn't help us support more hardware.  It does help us get better
bug reports, and helps some users figure out where best to direct their
bug reports, or help themselves.

Some people want to update pci.ids on a regular basis.  However, I think
these people are misguided for a number of reasons.  If we have a cronjob
that goes and fetches a new pci.ids once a week from sourceforge, we're
effectively conducting a DDoS attack (see also: mythtv users and embedded
ntp implementations).  Also, updating pci.ids on a system which has all
of its hardware already recognised is pointless.

The people who actually benefit from pci.ids being updated are those doing
new installs, and they need it updated in d-i, which implies updating
the udeb.  I presume d-i gets rebuilt for each minor release, and if so,
that seems like the best time to push pci.ids updates into stable.

I'm told that volatile.debian.net isn't set up to support udebs, and
anyway, it's unclear to me that d-i should be pulling from volatile.
We can separate out pci.ids from pciutils easily enough, even making it
a separate source package, and binary-all so as to not require rebuilds.

But what do you want to do?  If you want to declare this an insufficiently
interesting problem to deal with, that's fine.  But in that case, I'd
like special dispensation to get pci.ids updated as close to release
as possible.  Maybe separating out pci.ids into binary-all would help
most with that.