This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
linknamespace failures with massively parallel builds
- From: Florian Weimer <fweimer at redhat dot com>
- To: GNU C Library <libc-alpha at sourceware dot org>
- Date: Thu, 21 May 2015 18:07:00 +0200
- Subject: linknamespace failures with massively parallel builds
- Authentication-results: sourceware.org; auth=none
I see a lot of failures when running make check with large -j numbers
(on hardware which has a matching number of threads). Here are some
example:
FAIL: conform/XPG4/setjmp.h/linknamespace
FAIL: conform/XPG4/signal.h/linknamespace
FAIL: conform/XPG4/string.h/linknamespace
FAIL: conform/XPG4/strings.h/linknamespace
FAIL: conform/XPG4/stropts.h/linknamespace
FAIL: conform/XPG4/sys/ipc.h/linknamespace
FAIL: conform/XPG4/sys/mman.h/linknamespace
FAIL: conform/XPG4/sys/msg.h/linknamespace
Does anyone else see this?
linknamespace.pl exits with 1, with the following output:
==> conform/XPG4/termios.h/linknamespace.out <==
[initial] cfgetispeed
[initial] cfgetospeed
[initial] cfsetispeed
[initial] cfsetospeed
[initial] cfgetispeed -> [libc.a(speed.o)] __libc_errno ->
[libc.a(errno.o)] errno
[initial] tcdrain
[initial] tcflow
[initial] tcflush
[initial] tcgetattr
[initial] tcsendbreak
[initial] tcsetattr
==> conform/XPG4/termios.h/linknamespace.test-result <==
FAIL: conform/XPG4/termios.h/linknamespace
original exit status 1
I looked for file system races as an explanation, but I don't see them.
--
Florian Weimer / Red Hat Product Security