Sloppy header file testing in configure [with patch]

[Rhialto]

When compiling the version as of today on NetBSD 7, configure complains that IPv6 headers are not found, even though they are there. (NetBSD was one of the first OSes to support IPv6). It turns out the header file testing is sloppy; headers are attempted to be included even if they may not exists and are unneeded. Here's a diff. Attaching the file fails.

The first change, for net/if_var.h looks like a typo (the variable that is set does not match the file name). But it seems unused anyway.

diff --git a/configure.ac b/configure.ac
index d07f8e79..dc935598 100644
--- a/configure.ac
+++ b/configure.ac
@@ -737,7 +737,7 @@ AC_CHECK_HEADERS( linux/ipv6.h,   [hc_cv_have_linux_ipv6_h=yes],   [hc_cv_have_l
       #include <sys/socket.h>
     #endif
 ] )
-AC_CHECK_HEADERS( net/if_var.h,   [hc_cv_have_net_in_var_h=yes],   [hc_cv_have_net_in_var_h=no],
+AC_CHECK_HEADERS( net/if_var.h,   [hc_cv_have_net_if_var_h=yes],   [hc_cv_have_net_if_var_h=no],
 [
     #include <sys/types.h>
     #ifdef HAVE_SYS_SOCKET_H
@@ -753,8 +753,10 @@ AC_CHECK_HEADERS( netinet/in_var.h,  [hc_cv_have_netinet_in_var_h=yes],  [hc_cv_
     #ifdef HAVE_SYS_SOCKET_H
       #include <sys/socket.h>
     #endif
-    #ifdef HAVE_NET_IF_VAR_H
+    #ifdef HAVE_NET_IF_H
       #include <net/if.h>
+    #endif
+    #ifdef HAVE_NET_IF_VAR_H
       #include <net/if_var.h>
     #endif
     #ifdef HAVE_NETINET_IN_H
@@ -767,8 +769,10 @@ AC_CHECK_HEADERS( netinet6/in6_var.h,  [hc_cv_have_netinet6_in6_var_h=yes],  [hc
     #ifdef HAVE_SYS_SOCKET_H
       #include <sys/socket.h>
     #endif
-    #ifdef HAVE_NET_IF_VAR_H
+    #ifdef HAVE_NET_IF_H
       #include <net/if.h>
+    #endif
+    #ifdef HAVE_NET_IF_VAR_H
       #include <net/if_var.h>
     #endif
     #ifdef HAVE_NETINET_IN_VAR_H
@@ -1222,10 +1226,16 @@ AC_CHECK_MEMBERS( [struct in6_ifreq.ifr_ifru.ifru_flags],
     #ifdef HAVE_LINUX_IPV6_H
       #include <linux/ipv6.h>
     #endif
-    #ifdef HAVE_NETINET6_IN6_VAR_H
+    #ifdef HAVE_NET_IF_H
       #include <net/if.h>
+    #endif
+    #ifdef HAVE_NET_IF_VAR_H
       #include <net/if_var.h>
+    #endif
+    #ifdef HAVE_NETINET_IN_H
       #include <netinet/in.h>
+    #endif
+    #ifdef HAVE_NETINET6_IN6_VAR_H
       #include <netinet/in_var.h>  // Not the ipv6 file, please
     #endif
 ] )

[Rhialto]

The above diff is incorporated in my pull request #245.

[srorso]

Hi,

What happens when you use the CMake build procedure on NetBSD? That is not one of the systems I have tested; FreeBSD seems to work. So if you are willing to give it a try, I would be grateful.

Can you provide details on the version and the installation media used for NetBSD? I am open to building a VirtualBox NetBSD machine.

Also helpful for the optimization issue you raised is what version of gcc (or clang) is being used. I'm not sure about getting Autotools right, but I want the CMake build to handle this.

(Sidebar: Thanks Ivan for providing a workaround; I did not get an e-mail about the issue creation.)

First names are good here. Call me Steve.

Best Regards,
Steve Orso

[Rhialto]

Hi Steve,

I'm not so familiar with cmake, so I'm experimenting a little (my previous build attempts were simply using 1Stop).

My first try was to simply call cmake ../../hyperion; from the amd64/hyperion directory. That went through without complaints. It seems that the IPv6 detection that it did was fine, since the produced config.h file contains #define ENABLE_IPV6 1 which looks good. The -march=native is still there however; I'm trying to get around that next.

.../hercules/amd64/hyperion$ gmake
[  0%] Building C object CMakeFiles/hercs.dir/hsys.c.o
/mnt/vol1/rhialto/cvs/other/hercules/hyperion/hsys.c:1:0: error: bad value (native) for -march= switch
 /* HSYS.C       (c) Copyright Roger Bowler, 1999-2012                */
 ^
gmake[2]: *** [CMakeFiles/hercs.dir/build.make:63: CMakeFiles/hercs.dir/hsys.c.o] Error 1
gmake[1]: *** [CMakeFiles/Makefile2:2448: CMakeFiles/hercs.dir/all] Error 2
gmake: *** [Makefile:163: all] Error 2
.../hercules/amd64/hyperion$ 

I edited Herc31_COptsGNU.cmake to comment out line 196 that adds the -march=native. I like the remark

# Everything (almost) in Hercules must be compiled with position-independent
# code, so we shall just make it global.  On systems that support it of course.

since I identified it as a potential problem that SoftFloat may not be PIC even though it is linked into shared objects.

With the -march=native out of the way, everything builds and even runs (with judicious use of LD_LIBRARY_PATH). I tried the VM sixpack version 3 beta. I haven't tried it for more than a few minutes but it seems to work.

So the cmake build system is working well!

---

Regarding NetBSD: I am at version 7.0.1 but the latest is 7.1.1 (and 8 is supposed to be released "soon"). I expect the differences to be small; mainly security fixes and some fixed drivers. There are installation iso images at http://ftp.netbsd.org/pub/NetBSD/NetBSD-7.1.1/iso/NetBSD-7.1.1-amd64.iso with extensive instructions at http://ftp.netbsd.org/pub/NetBSD/NetBSD-7.1.1/amd64/INSTALL.html (but the simple version is: boot the iso image with a disk of a couple of GB, and follow the prompts).

These include X (as installation option) and the full C compiler.

Cmake and GNU make are not included in the base system, but pre-built packages are available from http://ftp.netbsd.org/pub/pkgsrc/packages/NetBSD/amd64/7.0/All. Using pkg_add http://ftp.netbsd.org/pub/pkgsrc/packages/NetBSD/amd64/7.0/All/gmake-4.1nb3.tgz or .../cmake-3.9.2nb1.tgz it should be possible to add gmake and cmake. Or look up pkgin in the INSTALL document.

[srorso]

Hi,

Thanks for the information. I will be on the road for a few days, so I will not have much time to work with NetBSD, but I did download the net install CD image and the html documentation. Light reading. -).

I am glad to hear that CMake works once you tackled the optimization flag. Once I get a test environment built, I will see if I can develop a robust correction for CMake. This will also give me a chance to test your configure.ac code in a variety of environments....Solaris was also sensitive to the ipv6 header tests.

SoftFloat-3a for Hercules is compiled with PIC; see line 472 in https://github.com/hercules-390/SoftFloat-3a/blob/master/CMakeLists.txt.

Best Regards,
Steve Orso

I will see if I can develop a robust correction for CMake.

cmake provides already the solution CheckCCompilerFlag no need to look at the compiler name , just check if the flag is supported the same approach can be used in configure.ac http://www.gnu.org/software/autoconf-archive/ax_check_compile_flag.html <http://www.gnu.org/software/autoconf-archive/ax_check_compile_flag.html>

[srorso]

Hi Rhialto,

(I saw your signature on another forum as ... "Rhialto" ... and take that to mean you are OK with "Rhialto" as a form of address...If that assumption is incorrect, please let me know and accept my apology in advance.)

Getting NetBSD running was a bit of a challenge for the following reasons,

1. There is no NetBSD software repository for 7.1.1. I overrode the missing repository with that for 7.1 and got messages about packages built for 7.0 being installed on 7.1.1. I submitted a problem report to NetBSD and fell back to your version 7.0.1.

2. Getting a root certificate store established to enable git clone using https was a non-trivial, and non-documented process. At least, I didn't find the documentation; Joachim Nilsson posted a procedure on his blog for NetBSD 6.1.5 and that procedure works on 7.0.1. (http://troglobit.com/howto/netbsd-pkgsrc/.

3. Learning curve specific to NetBSD. I had just rebuilt my FreeBSD, and NetBSD was a fair bit different. "Learning Curve" includes documentation generation; I create about 10 pages of install instructions specific to my VirtualBox based workstation farm for testing. I view those VirtualBox workstations as expendable, so documentation lets me rebuild one without creating variability in the setup.

4. My first assumption when building a new system is that if it goes badly, I did something wrong. It takes time to figure out, "no, it's really broken," and decide on a plan B. Or plan C. I was literally moments away from throwing in the towel when I found the root certificate store documentation.

I will be running a complete cycle of tests on all of my UNIX-like workstations and will include your pull request in the testing of the legacy GNU-Autotools build. I will then accept your PR separately so you receive the credit you have earned by creating the correction.

Best Regards,
Steve Orso

IMO the header detection logic is not sloppy , it just outdated, it suffers form a 15 year old logic that nobody cared to update in my current CMAKE build infrastructure sadly completely different from the one used in the current hyperion distribution there is only ONE header with special needs for APPLE CLANG to simplify the header detection logic I wrote a simple script which analyses the source tree to find out the includes and provide the proper info to write the proper login for the header detection ( I will run on FreeBSD and Linux Centos shortly and let Yo know the results ) here is the script just run it without arguments in the directory You want to analyse #! /usr/bin/env bash # - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - # _hdrcheck # Copyright Enrico Sorichetti 2018 - 2018 # Distributed under the Boost Software License, Version 1.0. # (See accompanying file LICENSE_1_0.txt or copy at # http://www.boost.org/LICENSE_1_0.txt) # _me="`basename ${0}`" _me="${_me%%.*}" # the name "${<the var>%%.*}" # the extension "${<the var>##*.}" IFS=$'\n' shopt -s extglob if [[ $# > 1 ]] ; then echo "too many arguments" exit elif [[ $# == 1 ]] ; then TOP="$1" else TOP="`pwd`" fi out="${_me}.$(basename ${TOP})" out="${out//+([.\/\ ])/_}.txt" rm -f ${out} tmp="`mktemp -t ${_me}`" src="`mktemp -t ${_me}`" src="`basename ${src}`" src="${src##*.}" ALL=$(find "${TOP}" -type f -name "*.[ch]") cat ${ALL} \ |sed -e 's+ ++g' \ |grep "^#include" \ |sed -e 's/#include<//' -e 's/#include"//' \ |sed -e 's/>.*$//' -e 's/".*$//'\ |grep ".h$" \ |sort |uniq \ "${tmp}" while read hdr ; do dir="`dirname ${hdr}`" if [ "$dir". == ".". ]; then dir="" else dir="/$dir" fi nam="`basename ${hdr}`" # special processing for config.h and platform.h if [ "${nam}". == "config.h". ]; then continue fi if [ "${nam}". == "platform.h". ]; then continue fi # check if the header is somewhere in the sources path fnd=$(find "${TOP}" -name "${nam}") if [ "${fnd}". != . ] ; then continue fi # try to compile a snippet using the header rm -f ${src}.c rm -f ${src}.o echo "#include <${hdr}>" > ${src}.c echo "int main(){}" >> ${src}.c cc -c ${src}.c 1>/dev/null 2>/dev/null RC=$? rm -f ${src}.o rm -f ${src}.c not="" if test -d /usr/include${dir} ; then fnd=$(find /usr/include"${dir}" -maxdepth 1 -name "${nam}" ) if [[ ${RC} == 0 ]] ; then if [ "${fnd}". == . ] ; then not="HIDDEN" fi else if [ "${fnd}". == . ] ; then not="NOT_FOUND" else not="HAS_SPECIAL_NEEDS" fi fi else if [[ ${RC} == 0 ]] ; then not="LOGIC_ERROR" else not="NOT_FOUND" fi fi printf "%-20s %d %s\n" $hdr ${RC} $not>>${out} done < ${tmp} exit and here the results for APPLE run from the directory containing decNumber,SoftFloat, Hyperion, spinhawk Mmsystem.h 1 NOT_FOUND SDKDDKVer.h 1 NOT_FOUND TT32if.h 1 NOT_FOUND Wincon.h 1 NOT_FOUND _dynamic_version.h 1 NOT_FOUND argz.h 1 NOT_FOUND arpa/inet.h 0 assert.h 0 byteswap.h 1 NOT_FOUND bzlib.h 0 commitinfo.h 1 NOT_FOUND conio.h 1 NOT_FOUND crtdbg.h 1 NOT_FOUND ctype.h 0 dbghelp.h 1 NOT_FOUND direct.h 1 NOT_FOUND dirent.h 0 dl.h 1 NOT_FOUND dld.h 1 NOT_FOUND dlfcn.h 0 dmalloc.h 1 NOT_FOUND errno.h 0 fcntl.h 0 hqa.h 1 NOT_FOUND iconv.h 0 ifaddrs.h 0 intrin.h 1 NOT_FOUND inttypes.h 0 io.h 1 NOT_FOUND kernel/image.h 1 NOT_FOUND libgen.h 0 limits.h 0 linux/if_tun.h 1 NOT_FOUND linux/ipv6.h 1 NOT_FOUND lmcons.h 1 NOT_FOUND mach-o/dyld.h 0 mach-o/getsect.h 0 mach/mach.h 0 malloc.h 1 NOT_FOUND math.h 0 memory.h 0 mstcpip.h 1 NOT_FOUND ndir.h 1 NOT_FOUND net/if.h 0 net/route.h 0 net/tcp.h 1 NOT_FOUND netdb.h 0 netinet/in.h 0 netinet/tcp.h 0 netinet6/in6_var.h 1 HAS_SPECIAL_NEEDS netioapi.h 1 NOT_FOUND oorexxapi.h 1 NOT_FOUND process.h 1 NOT_FOUND pthread.h 0 pwd.h 0 regex.h 0 regina/rexxsaa.h 1 NOT_FOUND rexx.h 1 NOT_FOUND rexxsaa.h 1 NOT_FOUND sched.h 0 setjmp.h 0 share.h 1 NOT_FOUND signal.h 0 stdarg.h 0 HIDDEN stdatomic.h 0 HIDDEN stdbool.h 0 HIDDEN stddef.h 0 stdint.h 0 stdio.h 0 stdlib.h 0 string.h 0 strings.h 0 sys/capability.h 1 NOT_FOUND sys/cygwin.h 1 NOT_FOUND sys/dir.h 0 sys/dl.h 1 NOT_FOUND sys/ioctl.h 0 sys/locking.h 1 NOT_FOUND sys/mman.h 0 sys/mount.h 0 sys/mtio.h 1 NOT_FOUND sys/ndir.h 1 NOT_FOUND sys/param.h 0 sys/prctl.h 1 NOT_FOUND sys/resource.h 0 sys/socket.h 0 sys/sockio.h 0 sys/stat.h 0 sys/sysctl.h 0 sys/time.h 0 sys/types.h 0 sys/uio.h 0 sys/un.h 0 sys/utsname.h 0 sys/wait.h 0 tchar.h 1 NOT_FOUND termios.h 0 time.h 0 tlhelp32.h 1 NOT_FOUND unistd.h 0 wincon.h 1 NOT_FOUND windows.h 1 NOT_FOUND winsock2.h 1 NOT_FOUND winuser.h 1 NOT_FOUND ws2tcpip.h 1 NOT_FOUND xmmintrin.h 0 HIDDEN zlib.h 0

[Rhialto]

Hi Steve,

yes, Rhialto is fine. It has been my handle on the Internet since the beginning.

I didn't anticipate that doing a basic setup would be that much work. Thank you for going through with it!

They should really find some way around the package version numbers. There is supposed to be very good binary compatibility when running older userland on newer kernels, but the package system checks this way too strictly. In fact I have disabled the check for myself, since I build packages in a 7.0 environment which I deliberately doe not upgrade (for stability) and then run them on the real system.
Thanks again!