ICA shows unrepeatable cp instruction

Ken Moffat ken at linuxfromscratch.org
Fri Dec 30 17:31:16 PST 2005

   I'm still playing with my farce scripts for ICA on ppc, to see just 
exactly how repeatable (or otherwise) our build turns out to be 
(preliminary discussion was on lfs-dev, apropos the alphabetic branch), 
and yes, it is indeed something of a farce :)

   My first attempts highlighted some errors in my scripts, and in my ICA 
method (specifically, you *really* don't want to rebuild a base system 
in place *after* adding things like lynx and berkelydb that man and perl 
can use), so the current method is: cross-build from x86_64 to ppc (in 
my case, I then chroot on ppc from an existing system - I like having X 
available), build the final-system, then rebuild it without referencing 
programs in /tools.  Now, I'm starting to analyse my second attempts.

  What is interesting is that my second (in-place) build of the 
final-system dumps the asm headers from Linux-libc-headers into a 
subdirectory.  The first build put the asm-ppc headers in 
/usr/include/asm (i.e. it created a new asm directory) and the second 
build put them into /usr/include/asm/asm-ppc.

  Now, my first thought about this is "well, that's just the way it is, 
we don't tell people to build in-place" but then I realised that if, for 
some reason, a builder repeats this command, they will end up with a 
similar subdirectory of unnecessary headers.

   Funnily enough, LFS-svn look as if it will have the same issue. 
Should we change to e.g.

   cp -R include/asm-${ARCH}/* /usr/include/asm

or is there another more elegant way of doing this ?  To be clear, I'm 
only advocating a change because I hate avoidable "only do this once" 

   das eine Mal als Tragödie, das andere Mal als Farce

More information about the cross-lfs mailing list