bug#1191: How to contribute back changes for Emacs on Solaris

[Dan Nicolaescu]

Ali Bahrami <ali_gnu@emvision.com> writes:

  > Dan Nicolaescu wrote:
  > <snip>
  > >
  > > I don't feel comfortable doing these changes without any testing at
  > > all.  In that case, the best would be to add a new file
  > > emacs/src/s/sol2-10.h that includes sol2-6.h and also contains your
  > > changes.
  > > Then change configure.in to use that new file for Solaris 10+.
  > 
  > I've done that, and built the result on Solaris Nevada (amd64),
  > Solaris 10 (sparc), and Solaris 9 (x86). It seems to work properly,
  > and to use dldump() for S10 and Nevada (aka OpenSolaris), while
  > using the generic dump for S9.
  > 
  > I rolled the amd64 configure change into this as well --- I needed it
  > to test, and it makes sense to do the complete deal in one go.
  > 
  > Here is the contents of src/s/sol2-10.h, in between the '====' lines:
  > 
  > ==========================================================================
  > /* Handle Solaris 2.10 */
  > 
  > #include "sol2-6.h"
  > 
  > /*
  >  * Use the Solaris dldump() function to dump emacs, instead of

Please mention here that the dldump function is used in unexsol.c,
otherwise people would have to guess what is going on.

  >  * the generic unexelf code. The resulting binary has a complete
  >  * symbol table, and is better for debugging and other observability
  >  * tools (debuggers, pstack, etc).


  >  * Note: Emacs used to use  dldump() before, but this was changed
  >  * in revision 1.3 of sol2-6.h, on September 13, 2002, with the comment:
  >  *
  >  *      dldump does not handle all the extensions used by GNU ld
  >  *
  >  * It is not clear what extensions this refers to. Testing with
  >  * OpenSolaris shows that it works on newer Solaris versions.
  >  * If a problem surfaces, please post the details to the OpenSolaris
  >  * mailing list given above.

This is not useful, you can omit it.

  >  * It is likely that dldump() works with older Solaris too,
  >  * but this has not been tested, and so, this change is for S10
                                                                ^^^
                                                                Solaris 10
  >  * and newer only at this time.
  >  */
  > 
  > #undef UNEXEC
  > #define UNEXEC unexsol.o

You also had a change to use system malloc.  Don't you want that in here too?


  > ==========================================================================
  > 
  > And here are the context diffs for the changes to configure.in
  > 
  > ==========================================================================
  > *** configure.in.orig   Sat Oct 18 17:40:27 2008
  > --- configure.in        Sat Oct 18 18:27:43 2008
  > ***************
  > *** 974,980 ****
  >     ;;
  > 
  >     *-auspex-sunos* | *-sun-sunos* | *-sun-bsd* | *-sun-solaris* \
  > !     | i[3456]86-*-solaris2* | i[3456]86-*-sunos5* | powerpc*-*-solaris2* \
  >       | rs6000-*-solaris2*)
  >       case "${canonical}" in
  >         m68*-sunos1* )  machine=sun1 ;;
  > --- 974,980 ----
  >     ;;
  > 
  >     *-auspex-sunos* | *-sun-sunos* | *-sun-bsd* | *-sun-solaris* \
  > !     | i[3456]86-*-solaris2* | x86_64-*-solaris2* | i[3456]86-*-sunos5* | 
power
  > pc*-*-solaris2* \
  >       | rs6000-*-solaris2*)
  >       case "${canonical}" in
  >         m68*-sunos1* )  machine=sun1 ;;
  > ***************
  > *** 988,993 ****
  > --- 988,1001 ----
  >         sparc* )                machine=sparc ;;
  >         * )             unported=yes ;;
  >       esac
  > + ## FIXME: make this into a proper fix that checks the compiler type,
  > + ## rather than relying on path. Or is /usr/ccs/lib/cpp a bad default now?
  > +     if [ "x$CC" = x/opt/SUNWspro/bin/cc ]; then
  > +         ## -Xs prevents spurious whitespace.
  > +       SOLARIS_NON_GNU_CPP="/opt/SUNWspro/bin/cc -E -Xs"
  > +     else
  > +       SOLARIS_NON_GNU_CPP=/usr/ccs/lib/cpp
  > +     fi
  >       case "${canonical}" in
  >         ## The Sun386 didn't get past 4.0.
  >         i[3456]86-*-sunos4        ) opsys=sunos4-0 ;;
  > ***************
  > *** 1028,1044 ****
  >                 NON_GNU_CPP=/usr/ccs/lib/cpp
  >                 RANLIB="ar -ts"
  >                 ;;
  > !       *-sunos5* | *-solaris* )
  >                 opsys=sol2-6
  > ! ## FIXME: make this into a proper fix that checks the compiler type,
  > ! ## rather than relying on path. Or is /usr/ccs/lib/cpp a bad default now?
  > !               if [ "x$CC" = x/opt/SUNWspro/bin/cc ]; then
  > !                 ## -Xs prevents spurious whitespace.
  > !                 NON_GNU_CPP="/opt/SUNWspro/bin/cc -E -Xs"
  > !               else
  > !                 NON_GNU_CPP=/usr/ccs/lib/cpp
  > !               fi
  >                 ;;
  >         *                         ) opsys=bsd4-2   ;;
  >       esac
  >       ## Watch out for a compiler that we know will not work.
  > --- 1036,1049 ----
  >                 NON_GNU_CPP=/usr/ccs/lib/cpp
  >                 RANLIB="ar -ts"
  >                 ;;
  > !       *-sunos5.[7-9] | *-solaris2.[7-9] )
  >                 opsys=sol2-6
  > !               NON_GNU_CPP=${SOLARIS_NON_GNU_CPP}
  >                 ;;
  > +       *-sunos5* | *-solaris* )
  > +               opsys=sol2-10
  > +               NON_GNU_CPP=${SOLARIS_NON_GNU_CPP}
  > +               ;;
  >         *                         ) opsys=bsd4-2   ;;
  >       esac
  >       ## Watch out for a compiler that we know will not work.
  > ==========================================================================


Can you generate this vs configure.in from CVS HEAD, that file has
changed a bit.