Glibc-2.8

[Greg Schafer]

Hi List,

There is a whacky testsuite issue with this release. Essentially, two iconv
tests will fail - bug-iconv6 and tst-iconv7. The fixes are already upstream,
but when the backported patches are applied, the 2 tests will pass, however,
a whole bunch of other tests will fail.

After some lengthy investigation, I managed to discover the cause and
engineer a workaround. Essentially, it's an ordering issue ie: the order in
which the various libc sub-directories are built. More details here:

http://sourceware.org/ml/libc-help/2008-09/msg00035.html

The workaround is to do this immediately prior to running the testsuite:

cp -v ../glibc-$GLIBC_VER/iconvdata/gconv-modules iconvdata

There is also a failure in the math area. It probably only affects 32-bit
x86 and is already reported but so far no action by upstream:

http://sourceware.org/ml/libc-alpha/2008-07/msg00024.html

Probably not worth worrying about as a) glibc devs are pretty much x86_64
only these days and b) it appears to be a problem in the testsuite itself if
I'm reading it correctly.

Incidentally, the current glibc-2.6-x86-math-tests-1.patch is not needed for
GCC-4.3 and therefore appears to be a GCC-4.2-only thing. Therefore I'll
change the condition to [[ $GCC_VER == 4.2* && $DIY_TARGET == i?86* ]].

Last problem is a build failure in Inetutils due to Glibc header cleanups.
Already fixed upstream but we can easily use a sed:

sed -i.bak '/<signal/a #include <stddef.h>' libicmp/icmp_timestamp.c

Regards
Greg