RE: Building CC++ on Mac OS X

[Ian Gough]

OK, found the problem.

If I call the small sched_getscheduler test program test.c, then it
compiles with no errors or warnings. If I call the same program but
named test.cpp, then gcc complains about an undefined identifier.

The autoconf test creates conftest.c when testing for sched_getscheduler
(and probably other functions too) and so the compiler never complains
about missing functions. Using the -std=c99 options to gcc causes
warnings to be reported on the c files.

[commoncpp2 613] cat test.c
#include <sched.h>

int main ()
{
  int policy;
  policy = sched_getscheduler(0);
  return 0;
}
[commoncpp2 614] cat test.cpp
#include <sched.h>

int main ()
{
  int policy;
  policy = sched_getscheduler(0);
  return 0;
}
[commoncpp2 615] gcc -c -g -O2 test.cpp
test.cpp: In function `int main()':
test.cpp:6: `sched_getscheduler' undeclared (first use this function)
test.cpp:6: (Each undeclared identifier is reported only once for each
function
   it appears in.)
[commoncpp2 616] gcc -c -g -O2 test.c
[commoncpp2 617]
[commoncpp2 618] gcc -c -g -O2 -std=c99 test.c
test.c: In function `main':
test.c:6: warning: implicit declaration of function `sched_getscheduler'


> -----Original Message-----
> From: Federico Montesino Pouzols 
> [mailto:address@hidden On Behalf Of Federico 
> Montesino Pouzols
> Sent: Wednesday, January 29, 2003 4:18 PM
> To: Ian Gough
> Cc: 'bug-commoncpp'
> Subject: Re: Building CC++ on Mac OS X
> 
> 
> 
>       Yes, the semaphore stuff is still pending, I expect to 
> fix it soon. As for the sched_getscheduler issue, it is weird 
> that it is detected, since the test is like this:
> 
>         AC_TRY_COMPILE(
>                 [#include <sched.h>],
>                 [
>                 int policy;
>                 policy = sched_getscheduler(0);
>                 ],
>                 [AC_DEFINE(HAVE_SCHED_GETSCHEDULER)])
> 
>       Could you try to compile on your system a simple C 
> program like the one above? It should not compile, but it 
> seems that autoconf is able to succesfully compile it.
> 
> On Mon, Jan 27, 2003 at 11:10:46AM -0500, Ian Gough wrote:
> > Grabbed a new CVS snapshot this morning and tried to build 
> after these 
> > changes and I still get compilation failures:
> > 
> > g++ -DHAVE_CONFIG_H -I. -I. -I../include -I../src 
> > g++ -DCCXX_EXPORT_LIBRARY
> > -I/usr/local/include -D_GNU_SOURCE -Wall -ansi -pedantic 
> -I../include 
> > -c thread.cpp -MT thread.lo -MD -MP -MF .deps/thread.TPlo  
> -fno-common 
> > -DPIC -o .libs/thread.lo
> > cc1plus: warning: changing search order for system directory 
> > "/usr/local/include"
> > cc1plus: warning:   as it has already been specified as a non-system
> > directory
> > thread.cpp: In constructor `ost::Thread::Thread(int, long unsigned
> > int)':
> > thread.cpp:536: `sched_getscheduler' undeclared (first use this
> > function)
> > thread.cpp:536: (Each undeclared identifier is reported 
> only once for 
> > each
> >    function it appears in.)
> > thread.cpp:554: `sched_getparam' undeclared (first use this 
> function)
> > make[1]: *** [thread.lo] Error 1
> > make: *** [all-recursive] Error 1
> > 
> > Checking if HAVE_SCHED_GETSCHEDULER is defined in one of the header 
> > files shows that it is defined in config.h for some reason.
> > 
> > [/var/root/src/revgen/xGrid/ccxx 552] grep -r HAVE_SCHED * 
> > acconfig.h:#undef HAVE_SCHED_GETSCHEDULER acconfig.h:#ifdef 
> > HAVE_SCHED_H config.log:#define HAVE_SCHED_H 1
> > config.log:#define HAVE_SCHED_GETSCHEDULER 1
> > 
> config.status:${ac_dA}HAVE_SCHED_H${ac_dB}HAVE_SCHED_H${ac_dC}
> 1${ac_dD}
> > 
> config.status:${ac_dA}HAVE_SCHED_GETSCHEDULER${ac_dB}HAVE_SCHE
D_GETSCHED
> > ULER${ac_dC}1${ac_dD}
> > 
> config.status:${ac_uA}HAVE_SCHED_H${ac_uB}HAVE_SCHED_H${ac_uC}
> 1${ac_uD}
> > 
> config.status:${ac_uA}HAVE_SCHED_GETSCHEDULER${ac_uB}HAVE_SCHE
D_GETSCHED
> > ULER${ac_uC}1${ac_uD}
> > configure:#define HAVE_SCHED_GETSCHEDULER 1
> > include/cc++/config.h:#define HAVE_SCHED_GETSCHEDULER 1
> > include/cc++/config.h:#define HAVE_SCHED_H 1
> > include/cc++/config.h:#ifdef HAVE_SCHED_H
> > include/cc++/config.tmp:#define HAVE_SCHED_GETSCHEDULER 1
> > include/cc++/config.tmp:#define HAVE_SCHED_H 1
> > include/cc++/config.tmp:#ifdef HAVE_SCHED_H
> > include/config.h:#define HAVE_SCHED_GETSCHEDULER 1
> > include/config.h:#define HAVE_SCHED_H 1
> > include/config.h:#ifdef HAVE_SCHED_H
> > include/config.h.in:#undef HAVE_SCHED_GETSCHEDULER
> > include/config.h.in:#undef HAVE_SCHED_H
> > include/config.h.in:#ifdef HAVE_SCHED_H
> > m4/ost_pthread.m4:              
> [AC_DEFINE(HAVE_SCHED_GETSCHEDULER)])
> > m4/ost_pthread.m4:dnl #undef HAVE_SCHED_GETSCHEDULER
> > m4/ost_pthread.m4:dnl #ifdef HAVE_SCHED_H
> > src/thread.cpp:#ifdef HAVE_SCHED_GETSCHEDULER
> > src/thread.cpp:#endif // ifdef HAVE_SCHED_GETSCHEDULER
> > 
> > But /usr/include/sched.h shows that there is no 
> sched_getscheduler() 
> > function.
> > 
> > #ifndef _SCHED_H_
> > #define _SCHED_H_
> > 
> > #include <pthread_impl.h>
> > 
> > /*
> >  * Scheduling paramters
> >  */
> > #ifndef __POSIX_LIB__
> > struct sched_param { int sched_priority;  char 
> > opaque[__SCHED_PARAM_SIZE__]; }; #endif
> > 
> > extern int sched_yield(void);
> > extern int sched_get_priority_min(int);
> > extern int sched_get_priority_max(int);
> > #endif /* _SCHED_H_ */
> > 
> > I grepped for sched_getscheduler and turned up no hits, so I am not 
> > sure why the autoconf system thinks that it exists.
> > 
> > I then defined __FreeBSD__ to get around this problem, but ran into 
> > another compile problem.
> > 
> > g++ -DHAVE_CONFIG_H -I. -I. -I../include -D__FreeBSD__ -I../src
> > -DCCXX_EXPORT_LIBRARY -I/usr/local/include -D_GNU_SOURCE 
> -Wall -ansi 
> > -pedantic -I../include -c semaphore.cpp -MT semaphore.lo 
> -MD -MP -MF 
> > .deps/semaphore.TPlo  -fno-common -DPIC -o .libs/semaphore.lo
> > cc1plus: warning: changing search order for system directory 
> > "/usr/local/include"
> > cc1plus: warning:   as it has already been specified as a non-system
> > directory
> > /usr/include/sys/sem.h: In destructor `virtual
> > ost::Semaphore::~Semaphore()':
> > /usr/include/sys/sem.h:242: too few arguments to function `int 
> > ost::semctl(int,
> >    int, int, ost::semun)'
> > semaphore.cpp:100: at this point in file
> > /usr/include/sys/sem.h: In member function `int
> > ost::Semaphore::getValue()':
> > /usr/include/sys/sem.h:242: too few arguments to function `int 
> > ost::semctl(int,
> >    int, int, ost::semun)'
> > semaphore.cpp:126: at this point in file
> > make: *** [semaphore.lo] Error 1
> > 
> > Mac OS X does not seem to have SYSV semaphores and an 
> incomplete set 
> > of POSIX semaphore functions. It does NOT have sem_init and 
> > sem_destroy, but it does have sem_open and sem_close.
> > 
> > > -----Original Message-----
> > > From: Federico Montesino Pouzols
> > > [mailto:address@hidden On Behalf Of Federico 
> > > Montesino Pouzols
> > > Sent: Wednesday, January 22, 2003 3:20 AM
> > > To: Ian Gough
> > > Cc: 'bug-commoncpp'
> > > Subject: Re: Building CC++ on Mac OS X
> > > 
> > > 
> > > 
> > >   Good, I have commited some modifications that should
> > > fix this issue.
> > > 
> > > On Tue, Jan 21, 2003 at 09:47:57AM -0500, Ian Gough wrote:
> > > > >       Um, I have just looked at this and actually there is
> > > already some
> > > > > autoconf stuff. Could you confirm if, in the config.h of
> > > cc++, the
> > > > > symbol HAVE_SOCKLEN_T is undefined?
> > > > > 
> > > > 
> > > > It is commented out. From include/cc++/config.h
> > > > 
> > > > #define HAVE_INET_ATON 1
> > > > #define HAVE_INET_SOCKETS 1
> > > > #define HAVE_UNIX_SOCKETS 1
> > > > /* #undef HAVE_SOCKLEN_T */
> > > > /* #undef HAVE_SYS_LIBCSYS_H */
> > > > /* #undef HAVE_SYS_STREAM_H */
> > > > /* #undef HAVE_SYS_POLL_H */
> > > > /* #undef HAVE_POLL_H */
> > > > 
> > > 
> > 
> > 
> > 
> > _______________________________________________
> > Bug-commoncpp mailing list
> > address@hidden 
> > http://mail.gnu.org/mailman/listinfo/bug-commoncpp
>