glibc-2.8 [was: Re: GCC-4.3.1, Linux-22.214.171.124]
Manuel Gonzalez Montoya
manuel.gonzalez.montoya at gmail.com
Fri Sep 5 13:41:17 PDT 2008
On Fri, Sep 5, 2008 at 6:11 AM, Robert Connolly
<robert at linuxfromscratch.org> wrote:
> I'm trying to find the most recent toolchain possible that will have all tests
> pass. I'm using Glibc-2.8, from:
> I get poor results with recent HJL Binutils and gcc43. I'm also using
> linux-126.96.36.199 headers, but Linux headers have never affected toolchain test
> suites in my experience.
> With Binutils-2.18, and gcc-4.2.5-20080827, I get perfect Binutils tests, but
> get three test failures from Glibc... iconvdata/bug-iconv6.out,
> iconvdata/tst-iconv7.out, and math/test-double.out.
> By dropping -mtune=native, from configparms, I lose a math/test-double.out
> failure from Glibc. My -mtune=native equals -mtune=prescott, according to
> gcc -v dummy.c, and -mtune=prescott has always caused toolchain failures for
> I retried with 'echo "CFLAGS += -march=i486 -mtune=i686" > configparms', and
> got all 3 failures, listed above. So math/test-double.out probably fails
> when -mtune is i686 or above (possibly i586 too).
> On my host system, with gcc41, and binutils-2.17, the glibc-2.8 test suite
> passes perfectly.
> Have any of you had all of glibc-2.8's tests pass with gcc42?
I think the exact math failure that you got is math/test-ildoubl,
thats because i am trying the same combination (glibc2.8, gcc4.3.2)
and got the same iconvdata test failures plus the math/test-ildoubl.
The math failure is caused by a gcc43 optimization and the result of a
function from the mpfr library. Take a look at
http://sourceware.org/ml/libc-alpha/2008-07/msg00024.html. The last
patch works for me, but somehow the patch has not find their way into
the cvs yet.
Regarding the iconvdata tests there are a couple of patches in the
last cvs snapshot but unfortunately they didn't fix the failures.
Im trying to find the patch mentioned here:
More information about the lfs-dev