Re: [issues] Eglibc issue on PPC?

["Joseph S. Myers" <joseph@xxxxxxxxxxxxxxxx>]

On Fri, 3 Jul 2009, Jim Gifford wrote:

> Been looking at this issue for a few weeks, and everyone has pointed this out
> to be a glibc/eglibc issue. Want your take on this. Using the latest GCC
> 4.4.0. Building from a x86_64 to powerpc G3.
> 
> The build method used is the cross-lfs dev book, temporary system gcc.
> http://cross-lfs.org/view/svn/ppc/temp-system/gcc.html
> our eglibc instructions are at
> http://cross-lfs.org/view/svn/ppc/cross-tools/eglibc.html

If you have trouble with a build procedure from a third party you should 
raise the issue with them, not with the EGLIBC developers.  The supported 
EGLIBC build procedure is that described in the EGLIBC.cross-building 
file.  Note that this does not discuss how to build a native toolchain to 
run on the target system, only a cross toolchain, and does not involve any 
modification to the native toolchain on the host/build system.

>      gcc -c -DIN_GCC    -DGENERATOR_FILE -I. -Ibuild
[...]

"gcc" should be a native compiler.

>      -I/tools/include -I/tools/include
[...]

>      In file included from /tools/include/endian.h:37,

It does not make sense to use a native compiler with -I options pointing 
to a directory containing files with the same names as system headers.  A 
native compiler needs to get files such as endian.h and ctype.h from the 
native header directory, /usr/include.  Any build procedure involving 
using a native compiler with non-native system headers is broken.  If you 
want to build a native toolchain to run on the target, you need to use an 
intermediate cross toolchain to do so, and the compilers in such a 
toolchain will be appropriately named for cross compilers and will be 
configured to find their system headers automatically without needing such 
-I options.

-- 
Joseph S. Myers
joseph@xxxxxxxxxxxxxxxx