Giorgio Agrelli <giorgio_a@libero.it> wrote:
Matthias-Christian Ott wrote:
Clemens Koller <clemens.koller@anagramm.de> wrote:
Giorgio Lando schrieb:
A little distribution should not try to work on multiple architectures,
Sorry, that's total nonsense!
The CRUX philosophy (stay simple and functional) is the key to be portable to other architectures.
This the result of portability and shows that porting CRUX is a (hopefully) trivial task. Thus we could merge quite quickly.
I don't agree on that. I think you misunderstood Clemens words. It's not trivial (even if in the latest years it became much easier). Archs like ppc are not really one arch, but a dozen of archs who share almost-similar processors. It was not trivial to write a toolchain able to build 64bit kernels for G5 and Pseries. It was not trivial to make an iso that was able to boot on 6 different kinds of powerpc machines, and our last work, making crux-ppc run on sam440ep turned out into writing kernel patches over the denx-tree and rtc drivers writing (cjg is taking care of that).
With other architectures it may be complicated (like ARM, ...), but it is not impossible and other distributions have and had the same problems. But this is no reason for forking a separate project.
I think Clemens meant that if a port is kept as easy as possible, then it's easy to find out when a port needs to be modified for a specific arch... if you keep putting flag variables in the port, in the footprints etc, you will never know if that port builds just fine in *every* case. If you put conditions in footprints it's very hard to give a simple look at the footprint and understand that it contains arch-specific files...
I think Pkgsrc does this also and has no real problems with it. -- Matthias-Christian Ott