Reported problems building the source code

The following problems have been reported by readers building the source code. I am just collecting these and making them available: I do not have access to all the systems to try and duplicate the problems and/or fix them. Please let me know of any additional problems.

FreeBSD 2.2.5. The configure script does not detect the timespec structure that is defined in <sys/time.h>. The fix is to add #define HAVE_TIMESPEC_STRUCT 1 to config.h.

An alternate fix is to change the configure script, as noted here:

echo $ac_n "checking for timespec{}""... $ac_c" 1>&6
echo "configure:3077: checking for timespec{}" >&5
if eval "test \"`echo '$''{'ac_cv_have_timespec_struct'+set}'`\" = set";
then
  echo $ac_n "(cached) $ac_c" 1>&6
else
  cat > conftest.$ac_ext <<EOF
#line 3082 "configure"
#include "confdefs.h"
#include <time.h>
#include <sys/time.h>   /* <== added for FreeBSD 2.2.5 */
EOF
if (eval "$ac_cpp conftest.$ac_ext") 2>&5 |
  egrep "timespec" >/dev/null 2>&1; then
  rm -rf conftest*
  ac_cv_have_timespec_struct=yes
else
  rm -rf conftest*
  ac_cv_have_timespec_struct=no
fi
rm -f conftest*
 
fi
 
echo "$ac_t""$ac_cv_have_timespec_struct" 1>&6
if test $ac_cv_have_timespec_struct = yes ; then 
        cat >> confdefs.h <<\EOF
#define HAVE_TIMESPEC_STRUCT 1
EOF
  
fi

HP-UX 10.10 and 10.20 (PA-RISC): One reader reports the following changes:
I modified the Make.defines file created by configure:

# System = hppa1.1-hp-hpux10.2
                              
CC = cc                       
# I added -Aa flag (Ansi Mode Compilation) in CFLAGS variable
CFLAGS = -g -D_HPUX_SOURCE -Aa
LIBS = ../libunp.a                         
LIBS_XTI = ../libunpxti.a ../libunp.a -lxti

To build libgai I had problems with the h_errno variable undefined, so I found the definition of this variable in /usr/include/netdb.h, the code fragment in netdb.h is:

#ifdef _XOPEN_SOURCE_EXTENDED                                 
extern int h_errno;                                           
#endif

So I just added the -D_XOPEN_SOURCE_EXTENDED directive in CFLAGS in Make.defines but this didn't work so I did a dirty modification to the libgai/addrinfo.h file, I included the next line at the begining of addrinfo.h file:

extern int h_errno;

HP-UX B10.20 (9000/700 series): Another reader reports that defining _HPUX_SOURCE fixes numerous problems.
Linux, RedHat 5.0: Readers have reported problems compiling the code in the libgai/ directory: the NI_ constants referenced in the file gn_ipv46.c are undefined. The easiest solution is to take the definitions of the NI_ constants from the addrinfo.h header and put them into the gai_hdr.h header.
Linux systems with glibc2.0.6: One reader has reported that Figure 16.7 (line 22) encounters a problem on these systems. Apparently ioctl returns a value of 0 for ifc_len if it encounters an error, instead of ioctl returning -1. The easiest solution is to change line 12 to initialize lastlen to -1.
Another problem has been reported with glibc2: the configure script assumes that the presence of the addrinfo structure in <netdb.h> means that the NI_xxx names are defined in that header, which is not the case for these systems. You will need to define these constants yourself.
Linux systems with kernel 2.0.33: The header poll.h is apparently in the /usr/include/sys directory and not the expected /usr/include directory for #include <poll.h> that appears in the configure script and other headers. The easiest fix is to just make a link.
Linux, Slackware 3.3. Building the library in /lib generates a compiler error for line 102 of the file wrapsock.c complaining that the functions prototype for the Poll wrapper function is wrong. Installing the pthreads-1.0.tar.gz package solves the problem.
NeXTStep 3.3 on an Intel/NeXT platform: The following two fixes are needed. (1) Add -posix to the CC flag in the Make.defines file. (2) The function strdup is not provided by the standard C library so you must provide your own, add it to the lib directory and add strdup.o to LIB_OBJS in the Make.defines file.
Reliant Unix (used to be called SINIX). Add the following three lines to the config.guess file:
```
RM*:ReliantUNIX-*:*:*) 
echo mips-sni-sysv4 
exit 0 ;; 
```
They should be located near the SINIX section and the SINIX section should stay in place, as many older systems are still around.
SGI IRIX 6.3. I have received the following comments regarding the system identified by the configure script as mpis-sgi-irix6.3 on an SGI O2:
The configure script seems to work OK, but it does enable the HAVE_MSGHDR_MSG_CONTROL constant in config.h. This causes the code in lib/read_fd.c to fail on line 34 of page 387 in UNP. The constant SCM_RIGHTS on that line is not known on the SGI apparently. I removed the define in config.h and the code compiles OK.
The config.h file correctly attempts to substitute socklen_t with unsigned int, but for some reason the compilation in lib doesn't actually do the substitution properly, this is probably an SGI idiosyncrasy, I have seen many others. I think SGI is probably behind in their implementation of Posix.1g. The code compiles with warning messages.
The strangest error is that SGI haven't prototyped the pthread_create function properly. In their code and man page, the second argument is prototyped as pthread_attr_t *, i.e. without the const modifier. I have been chasing around all morning to see if I have missed something on this. I have the latest patch installed, but no dice. I have opened a case with tech support, and will let you know what happens.
Compiling in intro/ generates many warning errors saying that the weak definitions of xdr_* functions in /usr/lib/libnsl.so are preempted by weak definitions in /usr/lib/libc.so. So now we know!
Solaris 2.6 (and probably other versions, when using Sun's C compiler). Although I had early access to Solaris 2.6 for testing, this did not include Sun's C compiler, so I had to use gcc. If gcc is not being used, the configure script defines __STDC__ which is probably not correct.