This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: glibc 2.24 --- Starting soft/slush freeze


Hi all,

On 2016-07-05 14:26, Adhemerval Zanella wrote:
> Hi all,
> 
> At stated in a previous message we are now in soft/slushy freeze mode. 
> It happened some days off from define schedule mainly because I wanted
> people give some feedback for release blockers and desirables.

Yesterday I have uploaded a snapshot of the current HEAD to Debian
experimental, and it has been tried to build on many architectures [1].
We run the testsuite when building the package so I think it might be
interesting to give some feedback, even if we use additional patches
or different configurations than upstream. They are all build with
gcc version 5.4.0 20160609. It should also be noted that we mark as
XFAIL some tests [2] depending on the architecture, so the testsuite 
results have to be interpreted "regression from glibc 2.23 or new
test".

First of all I would say the results are quite positive, we usually have
more issues when uploading a released version. I guess less refactoring
has been done in this version compared to the previous one, so less 
"non-mainstream" architectures are broken.

The GNU libc builds and do not have regressions on the following
architectures:
- aarch64
- arm soft-float
- arm hard-float
- i686
- mips n64 le
- powerpc
- ppc64 le
- s390x
- x86_64

It builds on the following architectures but have regression in the
testsuite. I haven't investigated in details yet:
- alpha: Failure of math/test-idouble and math/test-ifloat. The ceil,
  floor and rint functions output a sNaN instead of a qNaN when being
  presented an sNaN as input.

- hppa: Failure of nptl/tst-default-attr, rt/tst-timer4 and
  rt/tst-timer5. It needs more investigation.

- mips o32 be and le: Failure of conform/XPG3/sys/stat.h/conform
  and conform/XPG4/sys/stat.h/conform. This is because the o32 ABI
  defines the stat.st_dev as unsigned long int instead of dev_t. We
  have the other conform tests involving stat.h marked as XFAIL
  already, so the same should probably be done there. However I don't
  understand why there are not seen in Joseph's tests [3].

- ppc64: Failure of math/test-double, math/test-float, mah/test-idouble
  and math/test-ifloat. The modf functions output a sNaN instead of a 
  qNaN when being presented an sNaN as input, and without setting the
  invalid exception. It seems to be bug #20240 which is supposed to be
  fixed, I'll try to investigate.

- sparc64: Failure of math/test-double, math/test-double-finite,
  math/test-float and math/test-float-finite. The ceil, floor and trunc
  functions set the inexact exception while they should not.

Aurelien

[1] https://buildd.debian.org/status/package.php?p=glibc&suite=experimental
[2] https://anonscm.debian.org/cgit/pkg-glibc/glibc.git/tree/debian/testsuite-xfail-debian.mk?h=glibc-2.24
[3] https://sourceware.org/glibc/wiki/Release/2.24

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien@aurel32.net                 http://www.aurel32.net

Attachment: signature.asc
Description: PGP signature


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]