I downloaded the latest glibc (2.2.2) and corresponding linuxthreads from
ftp.gnu.org. When I finally built it and ran "make check" the following
testing double (inline functions)
Failure: Test: yn (3, 0.7) == -15.819479052819633505
is: -1.58194790528196396195e+01 -0x1.fa392c2273c1d0000000p+3
should be: -1.58194790528196342905e+01 -0x1.fa392c2273c1a0000000p+3
difference: 5.32907051820075139403e-15 0x1.80000000000000000000p-48
ulp : 3.0000
max.ulp : 2.0000
Test suite completed:
2419 test cases plus 2211 tests for exception flags executed.
1 errors occurred.
This failure occured on i686 with RedHat 7.0, kernel-2.4.3, gcc-2.96-79,
make-3.79.1, perl-5.6.0-9, binutils-2.10.1
How severe is it? Can I use the built library? And how to fix it?
It does not happen in the glibc we ship (2.2.2-10) AFAIK.
Anyway, it is not something you should be thrilled about too much,
yn was not precise too much (2 units within last place) and the glibc
you were testing had the difference 3 units.
In glibc-2.2.3 (currently there is 2.3.3pre3) much of the double math
library has been rewritten, so it guarantees accurracy even in
the last bit.
Could you tell me where to find 2.2.2-10.
I tried wwwrpmfind.net and didn't find it.
Thanks for your attention!