Simon, really, get a grip:

 ┌─(9:56)──<~/dev/libc/glibc-package 2.7>──
 └[artemis] du debian/patches
 2,9M	debian/patches

Let's have a look:

  • debian/patches/i386/submitted-i686-timing.diff submitted in april 2006.
  • debian/patches/any/submitted-longdouble.diff submitted in july 2007.
  • debian/patches/locale/* most of them local to Debian, upstream locales are broken in many ways, upstream doesn't want to hear about our patches.

And so on. OTOH I don't think we are really forking the glibc, it's "just" that upstream doesn't care about many of the problem we care about (e.g. "embedded crap[1]" like mips or arm).

Go ask xorg packaging team, go ask moizilla packaging team, go ask OOo.org people go ask any team with a complex enough upstream (especially the ones we have to patch around a lot to make them use system libraries instead of internal copies) before asserting anything like "integrating quilt is a step in the wrong direction because it carves into stone that we are forking packages".

Notes

[1] upstreams wording