Hello, I'm reporting these make check failures on OSX Lion 10.7.4
for Homebrew, a package installer for Macs. I am trying to author
the updated install script for openmpi-1.6.0 using Homebrew, but
I get a few test failures using the default compiler, clang-3.1.
Note: I get no test failures using llvm-gcc-4.2.1 (5658, 2336.9).
My setup:
iMac from 2009
OSX Lion 10.7.4
Core i5 quad-core 64-bit nehalem
8MB RAM
XCode 4.3.2
clang 3.1 (build 318)
gfortran 4.2.4-5666.3
I took a look at your FAQ and searched both users and devel lists
for clang and OSX posts concerning test failures, but I only found
one thread where a clang bug was found concerning weak symbols.
http://www.open-mpi.org/community/lists/users/2010/10/14438.php
http://llvm.org/bugs/show_bug.cgi?id=8383
As the bug report on llvm.org is still open with no updates,
I just tested it, and clang-3.1 returns an error now instead
of compiling that:
$ clang -shared -fvisibility=hidden weak.c -o libweak.so && nm -D libweak.so | grep foo
weak.c:6:36: error: only weak aliases are supported on darwin
int foo_b(void) __attribute((weak, alias(("foo_c"))));
^
So here is what happened with me and openmpi-1.6.0 and make check:
configure terminal output - https://gist.github.com/2720855
config.log - https://raw.github.com/gist/2720840
make V=1 log - https://raw.github.com/gist/2720893
make V=1 check - https://gist.github.com/2720908
Let me know how I can help. I suppose the warnings between lines
745-800 of the log for make check in datatype are significant where there
are a couple of array out of bounds errors, and later the abort trap 6
messages. Also line 1811 of the configure terminal output mentioned
Darwin being an unknown architecture.
Okay that's about it. I hope it was ok to gist those logs as versus
load your mail server up, but I can fix that if you need it.
2bits
|