So everybody on i686 can expect *some* failures in the glibc math tests
with gcc-4. I've got the patch from Drepper's (whoops, from _Mr_
Drepper's) commit (thanks, Greg) which solves half of the failures.
Looking at fedora4, even with their ability to selectively pick fixes
from CVS they use 'make -k check'. I don't have a problem with that,
(gcc has failures, so do some versions of binutils on some platforms),
but the book's current wording expects no failures in the normal
case:
|Important
|
|In this section, the test suite for Glibc is considered critical. Do
|not skip it under any circumstance.
|
|Test the results:
|
|make check
- we then go on to talk about the reasons why some people will see
failures and say
|In general, the Glibc test suite is always expected to pass.
I propose to change this to something like
(i) add the patch - 2 failures in the math tests are better than 4
failures.
(ii) change the command to
make -k check >glibc-check-log 2>&1 ; grep Error glibc-check-log
(iii) add an explanation:
On i686 you can expect to see failures in the test-double and
test-idouble math tests with gcc4, as well as an expected (ignored)
failure in posix/annexc. These failures in the math tests appear to be
harmless.
(iv) reword the text after this.
The Glibc test suite is highly dependent on certain functions of the
host system, in particular the kernel. In certain circumstances, some
failures are unavoidable. This is a list of the most common issues:
(remove "In general, the Glibc test suite is always expected to pass.
However,")
* The math tests sometimes fail in other tests when running on systems
where the CPU is not a relatively new genuine Intel or authentic AMD.
Certain optimization settings are also known to be a factor here.
( add "in other tests")
( continue as now with reference to gettext)
The drawback to telling our users to log this is that the log takes
2.2MB. In context, I don't think that is significant (although I would
appreciate a reminder of _which_ version of the locales we should be
measuring : full or minimal ?).
Dislikes ? Objections ? Responses of "but it all passes on my
pentium-plus" ? Better wording ?
Ken
--
das eine Mal als Trag?die, das andere Mal als Farce
--
http://linuxfromscratch.org/mailman/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
