On Wed, Mar 16, 2005 at 11:12:29AM -0800, Thomas Bushnell BSG wrote: > > To be in SCC, under the proposal we're all discussing, an arch must > have build 50% of the archive, not counting arch-specific packages. > > The Debian Hurd project has another category that should be excluded > because they are kernel-specific. (The current list on the web page > is update, makedev, ld.so, modconf, modutils, netbase, pcmcia-cs, > procps, ppp, pppconfig, setserial. There are surely more.) > > Ideally we would have a new field in the control file to specify > Kernel (or System), which would normally be "any", but for packages > like these would be "linux". But that's a hornets nest of potential > problems. > > We could ask the maintainer of update (say) to change it from > "Architecture: any" to "Architecture: i386 mk68 ...". But that's > tedious, brittle, and we all hate it. > > What would really win, of course, is "Architecture: !hurd-i386". But > negative declarations are currently not yet supported. They should > be. > > None of this is insoluble, and I assume that it won't be a serious > problem taking account of it, but it does mean that applying the 50% > and 90% threshholds will require more than simply looking at > statistics and applying them, because we don't currently have a > fully-robust way to indicate the relevant kind of "arch specific".
The 'type-handling' package by Robert Millan appears to already be in
use for addressing exactly this sort of thing (since, as far as I am
aware, he wrote it for working on various BSD ports), at least in terms of
simplifying the Architecture: entry. See the cdbs source package for some
useful things this can deal with.
Teaching the build statistics stuff to understand what the real number of
packages for a given architecture for is a separate question, of course.
--
Joel Aelwyn <[EMAIL PROTECTED]> ,''`.
: :' :
`. `'
`-
signature.asc
Description: Digital signature
