bug#17339: emacs-24.3.90: big build report

bug#17339: emacs-24.3.90: big build report

[Nelson H. F. Beebe]

I have a test lab with about 40 flavors of Unix running on current and
historical CPU architectures.  I have now done more than 70 build
attempts for the new emacs-24.3.90 release of 19-Apr-2014, and
happily, I can report success on most of them.  There remain a few
failures, and on some of the successful systems, I had to make
repeated tries to find option combinations that would produce a
successful build of emacs, including X11 support on all systems.

In NO CASE did I have to resort to editing any file from
emacs-24.3.90.

First, here is a brief list of systems for which I have successful
builds and installations; compiler choices, configure options and
environment variable settings are omitted, but I can supply them
offlist on request.

	--------------------------------------------------
	O/S version				CPU family
	--------------------------------------------------
	DragonFly 3.6				x86-64
	FreeBSD 9.1				x86
	FreeBSD 10.0				x86-64
	GNU/Linux ArchLinux			x86-64
	GNU/Linux Debian 6.0.9			x86-64
	GNU/Linux Debian 7.4			x86-64
	GNU/Linux Fedora 19			x86
	GNU/Linux Gentoo 2.2			Alpha
	GNU/Linux Gentoo 2.2			PowerPC-32
	GNU/Linux Gentoo 2.2			PowerPC-64
	GNU/Linux Gentoo 2.2			SPARC
	GNU/Linux Red Hat 5.10			IA-64
	GNU/Linux Red Hat 5.10			x86
	GNU/Linux Red Hat 5.10			x86-64
	GNU/Linux Red Hat 6.5			x86-64
	GNU/Linux Red Hat 7.0 beta		x86-64
	GNU/Linux Scientific Linux 6.5		x86-64
	GNU/Linux Slackware 14.0		x86-64
	GNU/Linux Ubuntu 12.04			x86
	GNU/Linux openSUSE 11.2			x86_64
	GNU/Linux openSUSE 12.3			x86_64
	GNU/Linux openSUSE 13.1			x86_64
	Mac OS X 10.5.8				PowerPC-32
	Mac OS X 10.7.5				x86-64
	MirBSD 10				x86
	OpenBSD 4.7				x86
	OpenBSD 5.1				x86
	OpenBSD 5.4				x86-64
	Solaris 10				SPARC
	Solaris 10				x86
	Solaris 10				x86-64
	Solaris 11				x86-64
	--------------------------------------------------

Here are the failing systems:

    GNU/Hurd x86:
	Compilation fails with

	process.c: In function 'handle_child_signal':
	process.c:6228:58: error: 'WCONTINUED' undeclared (first use in this function)
	    && child_status_changed (p->pid, &status, WUNTRACED | WCONTINUED))
								  ^
	process.c:6228:58: note: each undeclared identifier is reported only once for each function it appears in

	The symbol WCONTINUED is not defined in any file under /usr/include.

    NetBSD 5.0.2 x86:
	Compilation fails with

	    process.c: In function 'handle_child_signal':
	    process.c:6228:58: error: 'WCONTINUED' undeclared (first use in this function)
	That symbol is not found in any file under /usr/include.

    SGI IRIX 6.5 MIPS:
	/usr/local/bin/gcc -std=gnu99 -c -Demacs  -I. -I. -I../lib -I./../lib -g3 -O2  process.c
	process.c: In function `conv_sockaddr_to_lisp':
	process.c:2002: error: dereferencing pointer to incomplete type
	process.c:2003: error: dereferencing pointer to incomplete type
	process.c:2006: error: dereferencing pointer to incomplete type
	process.c: In function `get_lisp_to_sockaddr_size':
	process.c:2070: error: invalid application of `sizeof' to incomplete type `sockaddr_in6'
	process.c: In function `conv_lisp_to_sockaddr':
	process.c:2125: error: dereferencing pointer to incomplete type
	process.c:2126: error: dereferencing pointer to incomplete type
	process.c:2128: error: dereferencing pointer to incomplete type
	process.c: In function `server_accept_connection':
	process.c:4024: error: field `in6' has incomplete type

On all three systems, it looks like changes in emacs source code, or
perhaps the configure-time tests, will be needed.  I could see no
--enable-xxx, --disable-xxx, --with-xxx, or --without-xxx option that
would appear to affect the compilation issues in process.c.

Finally, here are problems for which I found workarounds:

(1) On several systems, linking failed with a missing symbol from the
    ImageMagick libraries:

	    undefined: PixelSetMagickColor

    The solution was to either

	    (a) add the --without-imagemagick option to configure

	    (b) build and install the latest ImageMagick release
		(6.8.9-0), and then retry the emacs build

    I've tried both approaches, and work in (b) is continuing.

(2) On NetBSD 6, configure tests for -lpng fail because that library does
    not exist; instead, the vendor provides only -lpng16.  I fooled the
    emacs configure script by creating this symlink:

		    /usr/local/lib/libpng.a: symbolic link to `/usr/pkg/lib/libpng16.a'

    so that -lpng is now found.

(3) On MirBSD, configure quits with the error

	    Emacs does not support `i386-unknown-mirbsd10' systems.

    The solution was to use the option --build=i386-unknown-openbsd4.9 in
    the configure step.  MirBSD is a derivative of OpenBSD 4.x, and that
    --build option trick has been necessary for a number of other software
    packages.

(4) On some systems (GNU/Linux on Alpha, SPARC, x86, PowerPC-32, and
    PowerPC-64), in the final "make install", I get this error very near
    the end:

	    /bin/bash: line 0: [: !=: unary operator expected
	    make[1]: Leaving directory '/local/build/bare/emacs-24.3.90/lib-src'

P.S. For a package of this size (410K lines of C, 1620K lines of
Lisp), the portability of emacs is wonderful!  My thanks go to the
great emacs development team who keep it building and working (almost)
everywhere.

-------------------------------------------------------------------------------
- Nelson H. F. Beebe                    Tel: +1 801 581 5254                  -
- University of Utah                    FAX: +1 801 581 4148                  -
- Department of Mathematics, 110 LCB    Internet e-mail: beebe@math.utah.edu  -
- 155 S 1400 E RM 233                       beebe@acm.org  beebe@computer.org -
- Salt Lake City, UT 84112-0090, USA    URL: http://www.math.utah.edu/~beebe/ -
-------------------------------------------------------------------------------

bug#17339: emacs-24.3.90: big build report

[Glenn Morris]

"Nelson H. F. Beebe" wrote:

> (1) On several systems, linking failed with a missing symbol from the
>     ImageMagick libraries:
>
> 	    undefined: PixelSetMagickColor

Please tell us what version of ImageMagick is installed on those systems.

> (2) On NetBSD 6, configure tests for -lpng fail because that library does
>     not exist; instead, the vendor provides only -lpng16.  I fooled the
>     emacs configure script by creating this symlink:
>
> 		    /usr/local/lib/libpng.a: symbolic link to `/usr/pkg/lib/libpng16.a'
>
>     so that -lpng is now found.

This has come up before; http://debbugs.gnu.org/11797#8 .
IMO it seems like a sub-optimal thing for NetBSD to do, but I guess they
have reasons.

Does: pkg-config --libs libpng

work on that system?

> (3) On MirBSD, configure quits with the error
>
> 	    Emacs does not support `i386-unknown-mirbsd10' systems.

I installed the following:

--- a/configure.ac	2014-04-13 22:51:08 +0000
+++ b/configure.ac	2014-04-29 14:49:11 +0000
@@ -505,7 +505,7 @@
   ;;
 
   ## OpenBSD ports
-  *-*-openbsd* )
+  *-*-openbsd* | *-*-mirbsd* )
     opsys=openbsd
   ;;

> (4) On some systems (GNU/Linux on Alpha, SPARC, x86, PowerPC-32, and
>     PowerPC-64), in the final "make install", I get this error very near
>     the end:
>
> 	    /bin/bash: line 0: [: !=: unary operator expected
> 	    make[1]: Leaving directory '/local/build/bare/emacs-24.3.90/lib-src'

I installed the following. If it does not fix it, please provide more
context from the build log.

--- a/lib-src/Makefile.in	2014-01-05 02:56:08 +0000
+++ b/lib-src/Makefile.in	2014-04-29 14:54:58 +0000
@@ -247,8 +247,8 @@
 	  chown ${gameuser} "$(DESTDIR)${gamedir}"; \
 	  chmod u=rwx,g=rwx,o=rx "$(DESTDIR)${gamedir}"; \
 	fi
-	if [ "`cd \"$(DESTDIR)${archlibdir}\" && /bin/pwd`" \
-             != "`cd ${srcdir} && /bin/pwd`" ]; then \
+	exp_archlibdir=`cd "$(DESTDIR)${archlibdir}" && /bin/pwd`; \
+	if [ "$$exp_archlibdir" != "`cd ${srcdir} && /bin/pwd`" ]; then \
 	  for file in ${SCRIPTS}; do \
 	    $(INSTALL_SCRIPT) ${srcdir}/$$file "$(DESTDIR)${archlibdir}/$$file"; \
 	  done ; \

bug#17339: emacs-24.3.90: big build report

[Glenn Morris]

"Nelson H. F. Beebe" wrote:

> 	process.c:6228:58: error: 'WCONTINUED' undeclared (first use in this function)
> 	    && child_status_changed (p->pid, &status, WUNTRACED | WCONTINUED))

I applied the following:

*** src/process.c	2014-04-16 13:27:28 +0000
--- src/process.c	2014-04-29 15:12:36 +0000
***************
*** 6225,6231 ****
        int status;
  
        if (p->alive
! 	  && child_status_changed (p->pid, &status, WUNTRACED | WCONTINUED))
  	{
  	  /* Change the status of the process that was found.  */
  	  p->tick = ++process_tick;
--- 6225,6235 ----
        int status;
  
        if (p->alive
! #ifndef WCONTINUED
!           && child_status_changed (p->pid, &status, WUNTRACED))
! #else
!           && child_status_changed (p->pid, &status, WUNTRACED | WCONTINUED))
! #endif
  	{
  	  /* Change the status of the process that was found.  */
  	  p->tick = ++process_tick;

bug#17339: emacs-24.3.90: big build report

[Glenn Morris]

Glenn Morris wrote:

> "Nelson H. F. Beebe" wrote:
[...]
>> (2) On NetBSD 6, configure tests for -lpng fail because that library does
>>     not exist; instead, the vendor provides only -lpng16.  I fooled the
>>     emacs configure script by creating this symlink:
>>
>> 		    /usr/local/lib/libpng.a: symbolic link to `/usr/pkg/lib/libpng16.a'
>>
>>     so that -lpng is now found.
>
> This has come up before; http://debbugs.gnu.org/11797#8 .

I applied the following:

--- a/configure.ac	2014-04-29 14:49:11 +0000
+++ b/configure.ac	2014-05-01 21:57:27 +0000
@@ -3101,8 +3101,15 @@
   fi
 
   if test "${HAVE_PNG}" = "yes"; then
-    AC_DEFINE(HAVE_PNG, 1, [Define to 1 if you have the png library (-lpng).])
-    LIBPNG="-lpng -lz -lm"
+    AC_DEFINE(HAVE_PNG, 1, [Define to 1 if you have the png library.])
+
+    dnl Some systems, eg NetBSD 6, only provide eg "libpng16", not "libpng".
+    lpng=`libpng-config --libs 2> /dev/null`
+    case $lpng in
+      -l*) : ;;
+      *) lpng="-lpng" ;;
+    esac
+    LIBPNG="$lpng -lz -lm"
 
     AC_CHECK_DECL(png_longjmp,
       [],
@@ -4976,7 +4983,7 @@
 echo "  Does Emacs use -ljpeg?                                  ${HAVE_JPEG}"
 echo "  Does Emacs use -ltiff?                                  ${HAVE_TIFF}"
 echo "  Does Emacs use a gif library?                           ${HAVE_GIF} $LIBGIF"
-echo "  Does Emacs use -lpng?                                   ${HAVE_PNG}"
+echo "  Does Emacs use a png library?                           ${HAVE_PNG} $LIBPNG"
 echo "  Does Emacs use -lrsvg-2?                                ${HAVE_RSVG}"
 echo "  Does Emacs use imagemagick?                             ${HAVE_IMAGEMAGICK}"
 

bug#17339: emacs-24.3.90: big build report

[Glenn Morris]

"Nelson H. F. Beebe" wrote:

>     SGI IRIX 6.5 MIPS:
> 	/usr/local/bin/gcc -std=gnu99 -c -Demacs  -I. -I. -I../lib -I./../lib -g3 -O2  process.c
> 	process.c: In function `conv_sockaddr_to_lisp':
> 	process.c:2002: error: dereferencing pointer to incomplete type
> 	process.c:2003: error: dereferencing pointer to incomplete type
> 	process.c:2006: error: dereferencing pointer to incomplete type
> 	process.c: In function `get_lisp_to_sockaddr_size':
> 	process.c:2070: error: invalid application of `sizeof' to incomplete type `sockaddr_in6'
> 	process.c: In function `conv_lisp_to_sockaddr':
> 	process.c:2125: error: dereferencing pointer to incomplete type
> 	process.c:2126: error: dereferencing pointer to incomplete type
> 	process.c:2128: error: dereferencing pointer to incomplete type
> 	process.c: In function `server_accept_connection':
> 	process.c:4024: error: field `in6' has incomplete type

IIUC, this is a failure to define the "sockaddr_in6" structure.
This is odd, because it is allegedly

http://techpubs.sgi.com/library/tpl/cgi-bin/getdoc.cgi?coll=0650&db=bks&srch=&fname=/SGI_Developer/IRIX_NetPG/sgi_html/ch03.html

defined in netinet/in.h, which process.c includes.

Can you find which .h file defines it on that system?
Or send netinet/in.h?

This is not a big deal, because we have had persistent problems with
SGI IRIX (eg http://debbugs.gnu.org/9684), and have dropped support for
it in Emacs 24.5.


In fact I see this looks like the same issue as http://debbugs.gnu.org/7643,
which you reported but never followed up to
I suggested something to try in

http://debbugs.gnu.org/7643#24

bug#17339: "undefined: PixelSetMagickColor" when building emacs-24.3.90

[Paul Eggert]

> Please tell us what version of ImageMagick is installed on those systems.

I went through the old ImageMagick versions (not a pleasant exercise) 
and found that PixelSetMagickColor was introduced in ImageMagick 6.3.5, 
dated September 2007.  The simplest fix is to require 6.3.5 or later, so 
I did that in emacs-24 bzr 117056.  If someone wants to port the Emacs 
source code to older versions of ImageMagick, please feel free.

bug#17339: SGI IRIX 6.5 in emacs-24.3.90: big build report

[Paul Eggert]

As luck would have it, I ported the emacs-24 branch to SGI IRIX 6.5 on 
April 13, the day after the emacs-24.3.90 pretest tarball was made.  I 
ran into the problem with sockaddr_in6 and fixed it in emacs-24 bzr 
116961 by tweaking src/conf_post.h to #define INET6 on IRIX 6.5.  So 
this should be fixed in the next pretest.  (This also should fix 
Bug#7643, which was marked done a while ago, I don't know why.)

The port requires IRIX cc, unfortunately.  I did not get undump to work 
with the old GCC installed at Utah.  I assume it's a simple fix for 
someone versed in undumping on IRIX, which I am not.  As far as I know 
this is the only remaining issue in Bug#17339.