Thanks, Travis!
On 10/13/12 15:36, Travis Vitek wrote:
Liviu,
I built the library and tests before and after your change with the 64-bit
flag, and I saw no differences in the number of failed tests between the
builds. I've attached the output of 'gmake -k runall' before and after your
change to STDCXX-1072 in case you want to look over them.
I wrote up the following test case to prove that I could reproduce the issue
using the compile/link options used to build the stdcxx tests. The code is
below..
[vitek@andromeda] 138 % cat t.cpp
#include <thread.h>
#include <synch.h>
#include <malloc.h>
int main ()
{
int *ip;
mutex_t *mp;
ip = (int*)malloc(sizeof (int) + sizeof (mutex_t));
mp = (mutex_t*)(ip + 1);
mutex_init(mp, USYNC_THREAD | LOCK_ROBUST, 0);
mutex_lock(mp);
mutex_unlock(mp);
mutex_destroy(mp);
free(ip);
}
[vitek@andromeda] 139 % gmake t
CC -c -D_RWSTDDEBUG -mt -I/amd/homes/vitek/tmp/stdcxx-4.2.x/include \
-I/build/vitek/stdcxx-4.2.x_sunpro-5.8_sunos-5.10-patched/include \
-I/amd/homes/vitek/tmp/stdcxx-4.2.x/tests/include \
-library=%none -g -m64 +w -errtags -erroff=hidef t.cpp
CC t.o -o t -L/build/vitek/stdcxx-4.2.x_sunpro-5.8_sunos-5.10-patched/rwtest
\
-lrwtest15S -library=%none -mt -m64 \
-L/build/vitek/stdcxx-4.2.x_sunpro-5.8_sunos-5.10-patched/lib \
-lstd15S -lm
Bus Error (core dumped)
[vitek@andromeda] 140 %
I also tested with POSIX mutexes and saw the same behavior.
Travis
From: Liviu Nicoara
Sent: Thursday, October 11, 2012 5:28 AM
To: dev@stdcxx.apache.org
Subject: Re: [PATCH] Re: STDCXX-1072 SPARC V8 mutex alignment requirements
I applied the patch on 4.2.x. If someone with access to a SPARC machine
could give it a runall and post the results here it would be awesome. I
will postpone closing the incident until then.
Thanks!
Liviu