[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Builds disturbed by the usage of a 64bit kernel.



Dear all,

Starting from the observation that a pre-release debian pacakge of
root-system and the current xaralx (non-free) pacakge fail to build
from source on G5 powerpc machines but not on G4, I started to get
worried that that kind of failures would be more widespread.

The core of the problem is that the configure scripts sometimes mistake
the build system type to be powerpc64-unknown-linux-gnu instead of
powerpc-unknown-linux-gnu. I am wondering similar things happen on sparc
running a 32-bit userland with a 64-bits kernel, hence the crosspost.

To investigate a bit more, I am rebuilding the maching with a cowbuildd,
and in this email, I would like to give preliminary results before the
week-end.

Out of 500 packages already build, ~140 tried to detect the build system
type, and ~30 of them guessed wrongly (powerpc64 instead of powerpc).
Three of them failed to build and in one case the wrong guess is the
obvious reason:

 In file included from /usr/include/features.h:346,
                  from /usr/include/stdlib.h:25,
                  from argus.c:40:
 /usr/include/gnu/stubs.h:9:27: error: gnu/stubs-64.h: No such file or directory
 make[2]: *** [argus.o] Error 1

I will continue the experiment during the week-end, but I do not think that I
can manage to rebuild the whole archive.

I woudld be especially intersted by the opinion of the sparc porters: do you
use sparc64-powered buildd with 32-bit userlands? Do you often run in that kind
of problem?

Another think I am wondering is wether the pacakges built successfully with the
wrong build system type are useful on normal 32-bit processors.

Have a nice day,

-- 
Charles Plessy
http://charles.plessy.org
Wako, Saitama, Japan



Reply to: