(dired-get-filename): Complain only about "." and "..",

[emacs.git] / etc / MACHINES

This is a list of the status of GNU Emacs on various machines and systems.

For each system and machine, we give the configuration name you should
pass to the `configure' script to prepare to build Emacs for that
system/machine.

The `configure' script uses the configuration name to decide which
machine and operating system description files `src/config.h' should
include.  The machine description files are all in `src/m', and have
names similar to, but not identical to, the machine names used in
configuration names.  The operating system files are all in `src/s',
and are named similarly.  See the `configure' script if you need to
know which configuration names use which machine and operating system
description files.

If you add support for a new configuration, add a section to this
file, and then edit the `configure' script to tell it which
configuration name(s) should select your new machine description and
system description files.

\f
Here are the configurations Emacs is intended to work with, with the
corresponding configuration names.  You can postpend version numbers
to operating system names (i.e. sunos4.1) or architecture names (i.e.
hppa1.1).  If you leave out the version number, the `configure' script
will configure Emacs for the latest version it knows about.

Acorn RISCiX (arm-acorn-riscix1.2)

  Emacs 19.29 has changes that ought to support RISCiX 1.2.

  Due to a bug in the RISCiX C compiler (3.4.5), emacs must
  be built with gcc (versions 2.5.8 onwards).

  In addition, you will need GNU sed and GNU make, as the RISCiX release
  versions of these utilities cannot cope with building emacs-19!

    GNU sed should be configured with:

    env 'DEFS=-Dgetopt=gnu_getopt -Dopterr=gnu_opterr -Doptind=gnu_optind \
            -Doptarg=gnu_optarg' ./configure

    GNU make (3.72+) should be configured with:

    env 'CFLAGS=-Dgetopt=gnu_getopt -Dopterr=gnu_opterr -Doptind=gnu_optind \
              -Doptarg=gnu_optarg' ./configure

  Emacs may be configured to use the X toolkit, by adding --with-x-toolkit
  to the configure command.  If you do this, you will need to edit the line
  in src/Makefile which defines LIBW (about line 59) to read:

  LIBW= -lXaw_n

  This ensures that the non-shared widget library is used.

  It is unlikely that this version of emacs will work with RISCiX 1.1.

Alliant (fx80-alliant-bsd):

  18.52 worked on system version 4.  Previous Emacs versions were
  known to work on previous system versions.

  If you are using older versions of their operating system, you may
  need to edit `src/config.h' to use `m/alliant1.h' (on version 1) or
  `m/alliant.h' (on versions 2 and 3).

Alliant FX/2800 (i860-alliant-bsd)

  Known to work with 19.26 and OS version 2.2, compiler version 1.3.

Alpha (DEC) running OSF/1 or GNU/Linux (alpha-dec-osf1, alpha-dec-linux-gnu)

  For OSF/1 (aka Digital Unix) version 4.0, update 386,
  it is reported that you need to run configure this way:

      configure --x-includes=/usr/include --x-libraries=/usr/shlib

  DEC C compiler version 5.9 (DEC C V5.9-005 on Digital UNIX V4.0f)
  is reported to produce bogus binaries of Emacs 21.2 when the
  command-line switches "-O4 -arch ev6 -tune ev6" are used.  Using
  just -O4 produces a good executable.

  For 4.0 revision 564, and 4.0A and 4.0B, Emacs 20 seems to work
  with no special configuration options.  However, if you use GCC as
  your compiler, you will need version 2.8.1 or later, as older
  versions fail to build with a message "Invalid dimension for the
  charset-ID 160".

  Note that the X11 libraries on GNU/Linux systems for the Alpha are
  said to have bugs that prevent Emacs from working with X (as of
  November 1995).  Recent releases work (July 2000).

Altos 3068 (m68k-altos-sysv)

  18.52 was said to work, provided you don't compile unexec.c with -O.

Amdahl UTS (580-amdahl-sysv)

  Small changes for 18.38 were merged in 18.39.  It is mostly
  working, but at last report a bug sometimes causes Emacs to
  grab very large amounts of memory.  No fix or explanation
  has yet been reported.  It may be possible to find this bug
  if you find which Emacs command it happens within and then
  run that command with a breakpoint set at malloc.

  The 5.2u370 compiler is so brain damaged that it is not
  even worth trying to use it.  Success was obtained with the
  uts native C compiler on uts version 5.2.5.

Apple Macintosh running Mac OS X

  For installation on all versions of the Mac OS platform, see the file
  mac/INSTALL.

Apple PowerPC Macintosh running GNU/Linux

  There are special considerations for a variety of this system which
  is known as the ``Yellow Dog [GNU/]Linux'': Emacs may crash during
  dumping.  To solve this, edit the header file src/m/macppc.h in the
  Emacs distribution, and remove the "#if 0" and "#endif" directives
  which surround the following block near the end of the file:

    #if 0  /* This breaks things on PPC GNU/Linux ecept for Yellowdog,
              even with identical GCC, as, ld.  Let's take it out until we
              know what's really going on here.  */
    /* GCC 2.95 and newer on GNU/Linux PPC changed the load address to
       0x10000000.  */
    #if defined __linux__
    #if __GNUC__ > 2 || (__GNUC__ == 2 && __GNUC_MINOR__ >= 95)
    #define DATA_SEG_BITS  0x10000000
    #endif
    #endif
    #endif /* 0 */

  After that, reconfigure and rebuild Emacs.  It should now build
  successfully.

Apollo running X Windows (m68k-apollo-bsd)

  Apollo version now supports dumping.  It has been tested on SR10.3 and
  SR10.4.  It certainly requires at least SR10.0, and maybe SR10.2.  Be sure
  to build in the BSD environment.

  By default, everything is compiled with the switch "-W0,-opt,2".  Don't try
  to change this to full optimization (-O).  The full optimizer (in Domain CC
  6.7, 6.8 and 6.9) generates some bad code in several modules which causes
  the emacs window, under X, to be refreshed with each keystroke.

  The configuration stuff should work for the most part.  However, some Domain
  installations may have to edit src/Makefile manually after it is created.
  There are too many versions of both cc and X to automate this easily.

  In `lib-src/Makefile', emacsclient and emacsserver compile and work fine
  under CC 6.9.  They now probably work under other versions of the compiler,
  as well.

  The Apollo Domain CC compiler will issue quite a few warning messages,
  mostly complaining about incompatible pointers.  In general, these are
  harmless and can be ignored.  If you discover otherwise, please submit a bug
  report identifying the problem in detail.

  When you try to dump emacs, you may get the message ".rwdi section needs
  relocation."  This means you are linking with some code that has compressed
  data sections.  In some cases this comes from linking with X libraries.  Try
  using shared X libraries instead.  With some versions of Domain/OS this is
  as simple as removing the "-lX11" from the LIBX line in src/Makefile.

  When running the configure script, use the configuration name
  "m68k-apollo-bsd".  You will also need to use the "-with-gcc=no" and
  "-with-x" options.  Depending upon your site configuration, you may have to
  use other configure options, as well.  Examine the INSTALL file for other
  configure options.

  Check out the file 'lisp/x-apollo.el'.  To use it, add

        (load "x-apollo")

  to your .emacs file.  It provides useful default Apollo function key
  bindings.

AT&T 3b2, 3b5, 3b15, 3b20 (we32k-att-sysv)

  Emacs will probably not work with certain kernel constants too small.

  In param.h CDLIMIT should be at least (1L << 12) in order to allow
  processes to write up to 2 Mbyte files.  This parameter is configurable
  by normal means in /etc/master.d/kernel; examine that file for the
  symbol CDLIMIT or ULIMIT, and raise it by several powers of 2.  Then
  do normal kernel rebuild things via "cd /boot; mkboot -k KERNEL" and so
  forth.

  In seg.h NSEGP and STACKSEG should be at least 16 and 4 respectively
  to allow processes with total size of up to 2Mbytes.
  However, I'm told it is unlikely this would fail to be true.

  The MAXMEM may also prevent Emacs from running.  The file
  3B-MAXMEM in this directory explains how to increase MAXMEM.

  On some of these machines, you may need to define IN_SCCS_ID
  in config.h to make Emacs work.  Supposedly you can tell whether
  this is necessary by checking something in /usr/include/sys/time.h;
  we do not know precisely what.

AT&T 7300 or 3b1 (m68k-att-sysv)

  18.52 worked.  If you have strange troubles with dumping
  Emacs, delete the last few lines from `src/m/7300.h' and recompile.
  These lines are supposed to produce a sharable executable.

  `src/m/7300.h' defines SHORTNAMES because operating system versions
  older than 3.5 did not support long symbol names.  Version 3.5 does
  support them, so you can remove the #define SHORTNAMES in that
  version.

Bull DPX/2 models 2nn or 3nn (m68k-bull-sysv3)

  Minor fixes merged into 19.19, which should work with CC or GCC.

  You should compile with all the POSIX stuff: undef _SYSV and define
  _POSIX_SOURCE, _XOPEN_SOURCE and _BULL_SOURCE.

  On bos2.00.45 there is a bug that makes the F_SETOWN fcntl
  call enters in an infinite loop. F_SETOWN_BUG has been defined to avoid
  calling it.

Bull DPX/20 (rs6000-bull-bosx)

  Version 19 works.

Bull sps7 (m68k-bull-sysv2)

  Changes partially merged in version 19, but some fixes are probably required.

CCI 5/32, 6/32

  See "Tahoe".

Celerity (celerity-celerity-bsd4.2)

  Version 18.49 worked.  This configuration name is a hack, because we
  don't know the processor used by Celerities.  If someone
  who uses a Celerity could get in touch with us, we can teach
  config.sub a better name for the configuration.

Clipper (clipper-???)

  Version 19 has support for some brand of clipper system.  If you
  have successfully built Emacs 19 on some sort of clipper system, let
  us know so we can flesh out this entry.

  Note that the Orion 105 is also a clipper, but some system-related
  parameters are different.

Convex (c1-convex-bsd, c2-convex-bsd, c32-convex-bsd, c34-convex-bsd,
        c38-convex-bsd)

  Support updated and residual bugs fixed in 19.26.

Cubix QBx/386 (i386-cubix-sysv)

  Changes merged in 19.1.  Systems before 2/A/0 may fail to compile etags.c
  due to a compiler bug.

Cydra 5 (cydra-cydrome-sysv)

  18.51 worked in one version of their operating system but stopped
  working in a newer version.  This has not been fixed.

Data General Aviion (m88k-dg-dgux)

  19.23 works; however, the GCC provided with DGUX 5.4R3.00 fails to
  compile src/emacs.c.  GCC 2.5.8 does work.
  The 19.26 pretest was reported to work; no word on which compiler.
  System versions other than DGUX 5.4R3.00 have not been tested.

  DGUX 5.4R3.10 works with 19.29 and 19.30.

  DGUX R4.11 contains changes to the stdio internals and it doesn't work
  with versions before 20.2 without patches.  20.2 works in interactive
  mode but usually fails in batch mode.  The problem is that using
  stderr in the dumped emacs usually leads to a segmentation fault.
  Only m88k has been tested.

DECstation (mips-dec-ultrix or mips-dec-osf)

  This machine is the older Mips-based DECstation.
  Emacs should now work on the Alpha CPU.

  19.25 works on Ultrix 4.2.  The 19.26 pretest was reported to work
  on Ultrix 4.2a and on 4.4.

  One user reported 19.25 did not work at all with --with-x-toolkit
  using X11R5 patch level 10, but worked ok with X11R5 pl26.

  See under Ultrix for problems using X windows on Ultrix.
  Note that this is a MIPS machine.

  For Ultrix versions 4.1 or earlier, you may need to define
  SYSTEM_MALLOC in `src/m/pmax.h', because XvmsAlloc.o in libX11.a seems
  to insist on defining malloc itself.

  For Ultrix versions prior to 4.0, you may need to delete
  the definition of START_FILES from `src/m/pmax.h'.

Motorola Delta 147 (m68k-motorola-sysv)

  The EMacs 19.26 pretest was reported to work.

  Motorola Delta boxes running System V/68 release 3.
  Tested on 147 board with SVR3V7, no X and gcc.
  Tested on 167 board with SVR3V7, no X, cc, gnucc and gcc.
  Reports say it works with X too.

  The installation script chooses the compiler itself.  gnucc is
  preferred.

Motorola Delta 187 (m88k-motorola-sysv,
                    m88k-motorola-sysvr4, or
                    m88k-motorola-m88kbcs)

  The 19.26 pretest was reported to run on SVR3.  However, if you
  use --with-x-toolkit on svr3, you will have problems compiling some
  files because time.h and sys/time.h get included twice.
  One fix is to edit those files to protect against multiple inclusion.

  As of version 19.13, Emacs was reported to run under SYSVr3 and SYSVr4.

Dual running System V (m68k-dual-sysv)

  As of 17.46, this worked except for a few changes
  needed in unexec.c.

Dual running Uniplus (m68k-dual-uniplus)

  Worked, as of 17.51.

Elxsi 6400 (elxsi-elxsi-sysv)

  Changes for 12.0 release are in 19.1.
  Dumping should work now.

Encore machine (ns16k-encore-bsd)

  This machine bizarrely uses 4.2BSD modified to use the COFF format
  for object files.  Works (as of 18.40).  For the APC processor you
  must enable two lines at the end of `src/s/umax.h', which are commented
  out in the file as distributed.

  WARNING: If you compile Emacs with the "-O" compiler switch, you
  must also use the "-q enter_exits" switch so that all functions have
  stack frames.  Otherwise routines that call `alloca' all lose.

  A kernel bug in some system versions causes input characters to be lost
  occasionally.

Fujitsu DS/90 (sparc-fujitsu-sysv4)

  Changes merged in 20.3.

GEC 63 (local-gec63-usg5.2)

  Changes are partially merged in version 18, but certainly require
  more work.  Let us know if you get this working, and we'll give it a
  real configuration name.

Gould Power Node (pn-gould-bsd4.2 or pn-gould-bsd4.3)

  18.36 worked on versions 1.2 and 2.0 of the operating system.

  On UTX/32 2.0, use pn-gould-bsd4.3.

  On UTX/32 1.2 and UTX/32S 1.0, use pn-gould-bsd4.2 and note that
  compiling `lib-src/sorted-doc' tickles a compiler bug: remove the -g
  flag to cc in the makefile.

  UTX/32 1.3 has a bug in the bcopy library routine.  Fix it by
  #undef BSTRING in `src/m/gould.h'.

  Version 19 incorporates support for releases 2.1 and later of UTX/32.
  A site running a pre-release of 2.1 should #define RELEASE2_1 in config.h.

Gould NP1 (np1-gould-bsd)

  Version 19 supposedly works.

Harris Night Hawk (m68k-harris-cxux or m88k-harris-cxux)

  This port was added in 19.23.  The configuration actually tested was
  a Night Hawk 4800 running CX/UX 7.0.

  If you have GCC ported and want to build with it, you probably need to
  change things (like compiler switches) defined in the s/cxux.h file.

  If you have X11R6 installed in /usr/lib, configure will fail to find
  it and may find X11R5 instead.  To work around this problem, use
  --x-libraries=/usr/lib when you run configure.

  With CX/UX 7.0 and later releases, you need to build after setting the
  SDE_TARGET environment variable to COFF (a port using ELF and shared
  libraries has not yet been done).

Harris Power PC (powerpc-harris-powerunix)

  Patches have been merged in 19.31.

Honeywell XPS100 (xps100-honeywell-sysv)

  Config file added in version 19.

Hewlett-Packard 9000 series 200 or 300 (m68k-hp-bsd or m68k-hp-hpux
                                        or m68k-hp-netbsd)

  These machines are 68000-series CPUs running HP/UX
  (a derivative of sysV with some BSD features) or BSD 4.3 ported by Utah.
  The operating system suffix determines which system Emacs is built for.

  Series 200 HPUX runs Emacs only if it has the "HP/UX upgrade".

  Version 19 works under BSD.  The 19.26 pretest was reported
  to work on HPUX 9.  19.31 works on HPUX 10.01, but there are
  some problems on 10.10 which have not been resolved.  Emacs 19.34
  works on HPUX 10.20 provided you compile with GCC; with the HP C
  compiler, subprocess commands do not work.

  On HPUX 9, Emacs sometimes crashes with SIGBUS or SIGSEGV after you
  delete a frame.  We think this is due to a bug in the X libraries
  provided by HP.  With the alternative X libraries in
  /usr/contrib/mitX11R5/lib, the problem does not happen.

  If you are running HP/UX release 8.0 or later, you need the optional
  "C/ANSI C" software in order to build Emacs (older releases of HP/UX
  do not require any special software).  If the file "/etc/filesets/C"
  exists on your machine, you have this software, otherwise you do not.

  Note that HP has used two incompatible assembler syntaxes,
  and has recently changed the format of C function frames.
  `src/crt0.c' and `src/alloca.s' have been conditionalised for the new
  assembler and new function-entry sequence.  You may need to define
  OLD_HP_ASSEMBLER if you are using an older hpux version.  If you
  have an official (bought from HP) series 300 machine you have
  the new assembler.  Kernels that are 5.+ or later have new
  assembler.  A Series 200 that has been upgraded to a 68010
  processor and a 5.+ kernel has the new compiler.

  Define C_SWITCH_MACHINE to be +X to make a version of Emacs that
  runs on both 68010 and 68020 based HP/UX's.

  Define HPUX_68010 if you are using the new assembler, for
  a system that has a 68010 without a 68881.  This is to say,
  a s200 (upgraded) or s310.

  Define the symbol HPUX_NET if you have the optional network features
  that include the `netunam' system call.  This is referred to as
  Network Services (NS/9000) in HP literature.

HP 9000 series 500: not supported.

  The series 500 has a seriously incompatible memory architecture
  which relocates data in memory during execution of a program,
  and support for it would be difficult to implement.

HP 9000 series 700 or 800 (Spectrum) (hppa1.0-hp-hpux or hppa1.1-hp-hpux
                                      or ...hpux9shr, or ...-nextstep)

  Use hppa1.1 for the 700 series and hppa1.0 for the 800
  series machines.  (Emacs may not actually care which one you use.)

  Support for NextSTEP was added in 19.31.

  Emacs 20 may work on HPUX 10.  You need patch PHSS_6202 to install
  the Xaw and Xmu libraries.  On HPUX 10.20 you may need to compile with GCC;
  when Emacs was compiled with HP's C compiler, HP92453-01 A.10.32.03,
  the subprocess features failed to work.

  19.26 is believed to work on HPUX 9 provided you compile with GCC.
  As of version 19.16, Emacs was reported to build (using GCC) and run
  on HP 9000/700 series machines running HP/UX versions 8.07 and 9.01.
  The HP compiler is known to fail on some versions if you use +O3,
  but it may work with lower optimization levels.

  Use hppa1.1-hp-hpux9shr to use shared libraries on HPUX version 9.
  You may need to create the X libraries libXaw.a and libXmu.a from
  the MIT X distribute, and you may need to edit src/Makefile's
  definition of LIBXT to look like this:

     LIBXT= $(LIBW) -lXmu -lXt $(LIBXTR6) -lXext

  Some people report trouble using the GNU memory allocator under
  HP/UX version 9.  The problems often manifest as lots of ^@'s in the
  buffer.

  We are told that these problems go away if you obtain the latest
  patches for the HP/UX C compiler.  James J Dempsey
  <jjd@spserv.bbn.com> says that this set of versions works for him:
    /bin/cc:
            HP92453-01 A.09.28 HP C Compiler
    /lib/ccom:
            HP92453-01 A.09.28 HP C Compiler
             HP-UX SLLIC/OPTIMIZER  HP-UX.09.00.23  02/18/93
             Ucode Code Generator - HP-UX.09.00.23.5 (patch)  2/18/93

  For 700 series machines, the HP-UX patch needed is known as
  PHSS_2653.  (Perhaps for 800 series machines as well; we don't
  know.)  If you are on the Internet, you should be able to obtain
  this patch by using telnet to access the machine
  support.mayfield.hp.com and logging in as "hpslreg" and following
  the instructions there.  Or you may be able to use this
  web site:

    HP Patch Server: http://support.mayfield.hp.com/patches/html/patches.html
    HP Support Line: http://support.mayfield.hp.com

  Please do not ask FSF for further support on this.  If you have any
  trouble obtaining the patch, contact HP Software Support.

  If your buffer fills up with nulls (^@) at some point, it could well
  be that problem.  That problem does not happen when people use GCC
  to compile Emacs.  On the other hand, the HP compiler version 9.34
  was reported to work for the 19.26 pretest.  9.65 was also reported to work.

  If you turn on the DSUSP character (delayed suspend),
  Emacs 19.26 does not know how to turn it off on HPUX.
  You need to turn it off manually.

  If you are running HP/UX release 8.0 or later, you need the optional
  "C/ANSI C" software in order to build Emacs (older releases of HP/UX
  do not require any special software).  If the file "/etc/filesets/C"
  exists on your machine, you have this software, otherwise you do not.

High Level Hardware Orion (orion-highlevel-bsd)

  This is the original microprogrammed hardware.
  Machine description file ought to work.

High Level Hardware Orion 1/05 (clipper-highlevel-bsd)

  Changes merged in 18.52.  This is the one with the Clipper cpu.
  Note that systems which lack NFS need LOAD_AVE_TYPE changed to `double'.

  C compiler has a bug; it loops compiling eval.c.
  Compile it by hand without optimization.

HITACHI SR2001/SR2201 series (hppa1.1-hitachi-hiuxwe2)

  These machines are based on PA architecture running HI-UX/MPP
  (based on OSF1. `MPP' stands for `Massively Parallel Processor').

  Emacs 19.34 is believed to work; its pretest was tested
  both on SR2001 (output of `uname -rv' is `00-01-BB 0') and
  SR2201 (`02-00 0').

  Emacs 20.7 was reported to build on a system whose `uname -rs'
  output is `HI-UX/MPP 03-04'.

  The machine description file is `src/m/sr2k.h' is based on
  `src/m/hp800.h'. The system description file is `src/s/hiuxwe2.h'
  based on `src/s/osf1.h'. Note that this system doesn't use COFF.

IBM PS/2 (i386-ibm-aix1.1 or i386-ibm-aix1.2)

  Changes merged in version 19.  You may need to copy
  /usr/lib/samples/hft/hftctl.c to the Emacs src directory.

  i386-ibm-aix1.1 may not work with certain new X window managers, and
  may be suboptimal.

IBM RS/6000 (rs6000-ibm-aix*)

  Emacs 19.26 is believed to work; its pretest was tested.

  Compiling with the system's `cc' and CFLAGS containing `-O5' might
  fail because libXbsd isn't found.  This is a compiler bug;
  re-configure Emacs so that it isn't compiled with `-O5'.

  On AIX 4.3.x and 4.4, compiling with /bin/c89 fails because it
  treats certain warnings as errors.  Use `cc' instead.

  At last report, Emacs didn't run well on terminals.  Informed
  persons say that the tty VMIN and VTIME settings have been
  corrupted; if you have a fix, please send it to us.

  Compiling with -O using the IBM compiler has been known
  to make Emacs work incorrectly.  It's reported that on
  AIX 3.2.5 with an IBM compiler earlier than 1.03.00.14,
  cc -O fails for some files.  You need to install any
  PTF containing APAR #IX42810 to bring the compiler to
  the 1.03.00.14 level to allow optimized compiles.

  There are reports that IBM compiler versions earlier than 1.03.00.02
  fail even without -O.  However, another report said that compiler
  version 1.02.01.00 did work, on AIX 3.2.4, with Emacs 19.31.

  As of 19.11, if you strip the Emacs executable, it ceases to work.

  If you are using AIX 3.2.3, you may get a core dump when loading
  ange-ftp.  You may be able to fix the problem by defining LIBS_TERMCAP
  as -ltermcap -lcurses.  Please tell us if this fails to work.

  If anyone can fix the above problems, or confirm that they don't happen
  with certain versions of various programs, we would appreciate it.

IBM RT/PC (romp-ibm-bsd or romp-ibm-aix)

  Use romp-ibm-bsd for the 4.2-like system and romp-ibm-aix for AIX.
  19.22 is reported to work under bsd.  We don't know about AIX.

  On BSD, if you have trouble, try compiling with a different compiler.

  On AIX, the file /usr/lib/samples/hft/hftctl.c must be compiled into
  hftctl.o, with this result left in the src directory (hftctl.c is
  part of the standard AIX distribution).

  window.c must not be compiled with -O on AIX.

IBM System/390 running GNU/Linux (s390-*-linux-gnu)

  As of Emacs 21.2, a 31-bit only version is supported on this
  system.

Integrated Solutions `Optimum V' (m68k-isi-bsd4.2 or -bsd4.3)

  18.52 said to work on some sort of ISI machine.
  Version 18.45 worked (running on a Optimum V (VME bus, 68020)
  BSD 4.2 (3.05e) system).  18.42 is reported to work on
  a Qbus 68010 system.  Has not been tried on `WorkStation' `Cluster
  Compute Node' `Cluster WorkStation' or `Server Node'  (Love the
  StudLYCaps)

  Compilation with -O is rumored to break something.

  On recent system versions, you may need to undefine the macro UMAX
  in `lib-src/loadst.c' and `src/getpagesize.h'.  They stupidly defined this
  in a system header file, which confuses Emacs (which thinks that UMAX
  indicates the Umax operating system).

Intel 386 (i386-*-isc,          i386-*-esix,     i386-*-bsdi2,
           i386-*-xenix,        i386-*-freebsd,  i386-*-linux-gnu,
           i386-*-sol2.4,       i386-*-sysv3,    i386-intsys-sysv,
           i386-*-sysv4,        i386-*-sysv4.2,
           i386-*-sysv5.3,      i386-*-bsd4.2,   i386-*-cygwin,
           i386-*-sco3.2v4,     i386-*-bsd386,   i386-*-386bsd,
           i386-*-msdos,        i386-*-windowsnt.
           i386... can be replaced with i486... or i586...)

  In the above configurations, * means that the manufacturer's name
  you specify does not matter, and you can use any name you like
  (but it should not contain any dashes or stars).

  When using the ISC configurations, be sure to specify the isc
        version number - for example, if you're running ISC 3.0, use
        i386-unknown-isc3.0 as your configuration name.
  Use i386-*-esix for Esix; Emacs runs as of version 19.6.
  Use i386-*-linux-gnu for GNU/Linux systems; Emacs runs as of version 19.26.
  Use i386-*-cygwin for Cygwin; Emacs builds as of version 21.4, in both X11
  and non-X11 modes.  (The Cygwin site has source and binaries for 21.2.)
  Use i386-intsys-sysv for Integrated Solutions 386 machines.
  It may also be correct for Microport systems.
  Use i386-*-sco3.2v4 for SCO 3.2v4; Emacs runs as of version 19.26.

  On GNU/Linux systems, Emacs 19.23 was said to work properly with libc
  version 4.5.21, but not with 4.5.19.  If your system uses QMAGIC
  for the executable format, you must edit config.h to define LINUX_QMAGIC.

  On GNU/Linux, configure may fail to put these definitions in config.h:

    #define HAVE_GETTIMEOFDAY
    #define HAVE_MKDIR
    #define HAVE_RMDIR
    #define HAVE_XSCREENNUMBEROFSCREEN

  To work around the problem, add those definitions by hand.
  It is possible that this problem happens only with X11R6.
  Newer system versions have fixed it.

  The 19.26 pretest was reported to work on SVR4.3 and on Freebsd.

  19.29 is reported to crash when using Motif on Solaris 2.5.
  The reasons are not yet known.

  Use i386-*-bsdiN for BSDI BSD/OS version N; Emacs runs as of version 19.23.
  In some system versions, `make' is broken; use GNU make instead.
  Shell bugs in version 1.0 of BSD/OS cause configure
  to do the wrong thing with --with-x-toolkit; the workaround is to edit
  configure to run another shell such as bash.

  For System V release 3, use i386-*-sysv3.
  For System V release 4, use i386-*-sysv4.
  For System V release 4.2, use i386-*-sysv4.2.

  If you are using Xenix, see notes at end under Xenix.
  If you are using Esix, see notes at end under Esix.
  If you are using SCO Unix, see notes at end under SCO.

  On 386bsd, NetBSD and FreeBSD, at one time, it was necessary to use
  GNU make, not the system's make.  Assuming it's installed as gmake,
  do `gmake install MAKE=gmake'.  However, more recently it is
  reported that using the system Make on NetBSD 1.3.1 works ok.

  If you are using System V release 4.2, you may find that `cc -E'
  puts spurious spaces in `src/xmakefile'.  If that happens,
  specify CPP=/lib/cpp as an option when you run make.
  There is no problem if you compile with GCC.

  Note that use of Linux with GCC 2.4 and the DLL 4.4 libraries
  requires the experimental "net 2" network patches (no relation to
  Berkeley Net 2).  There is a report that (some version of) Linux
  requires including `/usr/src/linux/include/linux' in buffer.c
  but no coherent explanation of why that might be so.  If it is so,
  in current versions of Linux, something else should probably be changed.

  Some sysV.3 systems seem to have bugs in `opendir';
  for them, alter `config.h' to define NONSYSTEM_DIR_LIBRARY
  and undefine SYSV_SYSTEM_DIR.

  If you use optimization on V.3, you may need the option -W2,'-y 0'
  to prevent certain faulty optimization.

  On 386/ix, to link with shared libraries, add #define USG_SHARED_LIBRARIES
  to config.h.

  On SCO, there are problems in regexp matching when Emacs is compiled
  with the system compiler.  The compiler version is "Microsoft C
  version 6", SCO 4.2.0h Dev Sys Maintenance Supplement 01/06/93;
  Quick C Compiler Version 1.00.46 (Beta).  The solution is to compile
  with GCC.

  On ISC systems (2.02 and more recent), don't try to use the versions
  of X that come with the system; use XFree86 instead.

  There is no consistency in the handling of certain system header files
  on V.3.

  Some versions have sys/sioctl.h, and require it in sysdep.c.
  But some versions do not have sys/sioctl.h.
  For a given version of the system, this may depend on whether you have
  X Windows or TCP/IP.  Define or undefine NO_SIOCTL_H in config.h
  according to whether you have the file.

  Likewise, some versions have been known to need sys/ttold.h, sys/stream.h,
  and sys/ptem.h included in sysdep.c.  If your system has these files,
  try defining NEED_PTEM_H in config.h if you have trouble without it.

  You may find that adding -I/usr/X/include or -I/usr/netinclude or both
  to CFLAGS avoids compilation errors on certain systems.

  Some versions convince sysdep.c to try to use `struct tchars'
  but define `struct tc' instead; add `#define tchars tc'
  to config.h to solve this problem.

Iris 2500 and Iris 2500 Turbo (m68k-sgi-iris3.5 or m68k-sgi-iris3.6)

  Version 18 was said to work; use m68k-sgi-iris3.5 for system version 2.5
  and m68k-sgi-iris3.6 for system version 3.6.
  Note that the 3030 is the same as the Iris 2500 Turbo.

Iris 4D (mips-sgi-irix[456].*)

  You can build a 64-bit executable (with larger maximum buffer size)
  on Irix 6.5 by specifying the 64-bit ABI using the `-64' compiler
  flag or otherwise (see cc(1)).  This may work on earlier Irix 6
  systems if you edit src/s/irix6-0.h following irix6-5.h.

  Building Emacs 21.1 and 21.2 on versions of Irix before 6.5.10,
  especially when Emacs is built with GCC, was reported to have subtle
  problems such as being unable to print to stdout under the -batch
  command-line option.  Building with the native compiler or upgrading
  the OS to a newer version solves these problems.  There's evidence
  that these problems are actually related to the runtime libraries
  (before IRIX 6.5.10, the IRIX runtimes were based on the MIPSpro 7.2
  compilers), so installing patches for the runtime from
  http://www.sgi.com/support/patch_intro.html could solve the problem
  even without upgrading the OS.  The dump process is the crucial
  step that needs the upgraded runtime, so a workaround is to dump
  Emacs on a machine with a newer OS, then copy the binary to the
  older OS.

  If compiling with GCC on Irix 6 yields an error "conflicting types
  for `initstate'", install GCC 2.95 or a newer version, and this
  problem should go away.  It is possible that this problem results
  from upgrading the operating system without reinstalling GCC; so you
  could also try reinstalling the same version of GCC, and telling us
  whether that fixes the problem.

  The 19.26 pretest was reported to work on IRIX 4.0.5 and 5.2.
  19.23 was reported to work on IRIX 5.2, but you may need to install
  the "compiler_dev.hdr.internal" subsystem in order to compile unexelfsgi.c.
  19.22 was known to work on all Silicon Graphics machines running
  IRIX 4.0.5 or IRIX 5.1.

  Compiling with -O using IRIX compilers prior to 3.10.1 may not work.
  Don't use -O or use GCC instead.

  Most IRIX 3.3 systems do not have an ANSI C compiler, but a few do.
  Compile Emacs 18 with the -cckr switch on these machines.

  There is a bug in IRIX 3.3 that can sometimes leave ptys owned by root
  with a permission of 622.  This causes malfunctions in use of
  subprocesses of Emacs.  Irix versions 4.0 and later with GNU Emacs
  versions 18.59 and later fix this bug.

Masscomp (m68k-masscomp-rtu)

  18.36 worked on a 5500DP running RTU v3.1a and compiler version 3.2
  with minor fixes that are included in 18.37.  However, bizarre behavior
  was reported for 18.36 on a Masscomp (model and version unknown but probably
  a 68020 system).  The report sounds like a compiler bug.

  A compiler bug affecting statements like
    unsigned char k; unsigned char *p;... x = p[k];
  has been reported for "C version 1.2 under RTU 3.1".  We do not wish
  to take the time to install the numerous workarounds required to
  compensate for this bug.

  For RTU version 3.1, define FIRST_PTY_LETTER to be 'p' in `src/s/rtu.h'
  (or #undef and redefine it in config.h) so that ptys will be used.

  GNU Emacs is said to have no chance of compiling on RTU versions
  prior to v3.0.

Megatest (m68k-megatest-bsd)

  Emacs 15 worked; do not have any reports about Emacs 16 or 17
  but any new bugs are probably not difficult.

Mips (mips-mips-riscos, mips-mips-riscos4.0, or mips-mips-bsd)

  The C compiler on Riscos 4.51 dumps core trying to optimize
  parts of Emacs.  Try without optimization or try GCC.

  Meanwhile, the linker on that system returns success even if
  there are undefined symbols; as a result, configure gets the
  wrong answers to various questions.  No work-around is known
  except to edit src/config.h by hand to indicate which functions
  don't exist.

  Use mips-mips-riscos4.0 for RISCOS version 4.
  Use mips-mips-bsd with the BSD world.

  Note that the proper configuration names for DECstations are
  mips-dec-ultrix and mips-dec-osf.

  If you are compiling with GCC, then you must run fixincludes;
  the alternative of using -traditional won't work because
  the definition of SIGN_EXTEND_CHAR uses the keyword `signed'.

  If the SYSV world is the default, then you probably need the following
  line in etc/Makefile:

    CFLAGS= -g -systype bsd43

  Some operating systems on MIPS machines give SIGTRAP for division by
  zero instead of the usual signals.  The only real solution is to fix
  the system to give a proper signal.

  In the meantime, you can change init_data in data.c if you wish.
  Change it to handle SIGTRAP as well as SIGFPE.  But this will have a
  great disadvantage: you will not be able to run Emacs under a
  debugger.  I think crashing on division by zero is a lesser problem.

  dsg@mitre.org reported needing to use --x-libraries=/bsd43/usr/lib
  on a riscos4bsd site.  But it is not clear whether this is needed in
  general or only because of quirks on a particular site.

National Semiconductor 32000 (ns32k-ns-genix)

  This is for a complete machine from National Semiconductor,
  running Genix.  Changes merged in version 19.

NCR Tower 32 (m68k-ncr-sysv2 or m68k-ncr-sysv3)

  If you are running System V release 2, use m68k-ncr-sysv2.
  If you are running System V release 3, use m68k-ncr-sysv3.

  These both worked as of 18.56.  If you change `src/ymakefile' so that
  CFLAGS includes C_OPTIMIZE_SWITCH rather than C_DEBUG_SWITCH, check
  out the comments in `src/m/tower32.h' (for System V release 2) or
  `src/m/tower32v3.h' (for System V release 3) about this.

  There is a report that compilation with -O did not work with 18.54
  under System V release 2.

NCR Intel system (i386-ncr-sysv4.2)

  This system works in 19.31, but if you don't link it with GNU ld,
  you may need to set LD_RUN_PATH at link time to specify where
  to find the X libraries.

NEC EWS4800 (mips-nec-sysv4)

  This system works in 20.4, but you should use the compiler
  /usr/abiccs/bin/cc (MIPS ABI MODE).

NeXT (m68k-next-nextstep)

  Emacs 19 has not been tested extensively yet, but it seems to work
  in a NeXTStep 3.0 terminal window, and under the X server called
  co-Xist.  You may need to specify -traditional when src/Makefile
  builds xmakefile.

  NeXT users might want to implement direct operation with NeXTStep,
  but from the point of view of the GNU project, that is a
  distraction.

  Thanks to Thorsten Ohl for working on the NeXT port of Emacs 19.

Nixdorf Targon 31 (m68k-nixdorf-sysv)

  Machine description file for version 17 is included in 18
  but whether it works is not known.
  `src/unexec.c' bombs if compiled with -O.
  Note that the "Targon 35" is really a Pyramid.

Nu (TI or LMI) (m68k-nu-sysv)

  Version 18 is believed to work.

Paragon OSF/1 (i860-intel-osf1)

  Changes merged in 19.29.

  There is a bug in OSF/1 make which claims there is a syntax error
  in the src/xmakefile.  You can successfully build emacs with:

        pmake MAKE=pmake

Plexus (m68k-plexus-sysv)

  Worked as of 17.56.

Pmax (DEC Mips)  (mips-dec-ultrix or mips-dec-osf1)

  See under DECstation, above.

Prime EXL (i386-prime-sysv)

  Minor changes merged in 19.1.

Pyramid (pyramid-pyramid-bsd)

  The 19.26 pretest was observed to work on OSx 5.0, but it is necessary
  to edit gmalloc.c.  You must add #include <sys/types.h> at the top,
  and delete the #define for size_t.

  You need to build Emacs in the Berkeley universe with
  the `ucb' command, as in `ucb make' or `ucb build-install'.

  In OSx 4.0, it seems necessary to add the following two lines
  to `src/m/pyramid.h':
     #define _longjmp longjmp
     #define _setjmp setjmp

  In Pyramid system 2.5 there has been a compiler bug making
  Emacs crash just after screen-splitting with Qnil containing 0.
  A compiler that fixes this is Pyramid customer number 8494,
  internal number 1923.

  Some versions of the pyramid compiler get fatal
  errors when the -gx compiler switch is used; if this
  happens to you, change `src/m/pyramid.h' to define
  C_DEBUG_SWITCH with an empty definition.

  Some old system versions may require you to define PYRAMID_OLD
  in when alloca.s is preprocessed, in order to define _longjmp and _setjmp.

Sequent Balance (ns32k-sequent-bsd4.2 or ns32k-sequent-bsd4.3)

  Emacs 18.51 worked on system version 3.0.  18.52 is said to work.
  Delete some lines at the end of `src/m/sequent.h' for earlier system
  versions.

Sequent Symmetry (i386-sequent-bsd, i386-sequent-ptx, i386-sequent-ptx4)

  19.33 has changes to support ptx 4 (a modified SVR4).

  Emacs 19 should work on Dynix (BSD).  However, if you compile with
  the Sequent compiler, you may find Emacs does not restore the
  terminal settings on exit.  If this happens, compile with GCC.

  Emacs 19.27 contains patches that should support
  DYNIX/ptx 1.4 and 2.1 with the native cc compiler.

  GCC can't compile src/process.c due to a non-standard Sequent asm
  keyword extension supported by cc and used for the network byte/word
  swapping functions in the PTX /usr/include/netinet/in.h file.  GCC
  2.5.8 includes the file <sys/byteorder.h> which can be included into
  netinet/in.h to perform these byte/word swapping functions in the
  same manner.  Patches have been submitted to the FSF against GCC
  2.6.0 to fix this problem and allow Emacs to be built with GCC.

  If your machine does not have TCP/IP installed, you will have to edit the
  src/s/ptx.h file and comment out #define TCPIP_INSTALLED.

Siemens Nixdorf RM600 and RM400 (mips-siemens-sysv4)

  Changes merged in 19.29.  This configuration should also work for
  Pyramid MIS Server running DC-OSX 1.x.  The version configured with
  `--with-x' works without any modifications, but `--with-x-toolkit'
  works only if the Athena library and the Toolkit library are linked
  statically.  For this, edit `src/Makefile' after the `configure' run
  and modify the lines with `-lXaw' and `-lXt' as follows:

    LIBW= /usr/lib/libXaw.a
    LIBXT= $(LIBW) -lXmu /usr/lib/libXt.a $(LIBXTR6) -lXext

  In addition, `--with-x-toolkit=motif' works only
  if the Motif library and the Toolkit library are linked statically.
  To do this, edit `src/Makefile' after the `configure' run
  and modify the lines with `-lXm' and `-lXt' as follows:

    LIBW= /usr/lib/libXm.a /usr/ccs/lib/libgen.a
    LIBXT= $(LIBW) -lXmu /usr/lib/libXt.a $(LIBXTR6) -lXext

SONY News (m68k-sony-bsd4.2 or m68k-sony-bsd4.3)

  18.52 worked.  Use m68k-sony-bsd4.3 for system release 3.

SONY News 3000 series (RISC NEWS) (mips-sony-bsd)

  The 19.26 pretest is reported to work.

  Some versions of the operating system give SIGTRAP for division by zero
  instead of the usual signals.  This causes division by zero
  to make Emacs crash.  The system should be fixed to give the proper signal.
  Changing Emacs is not a proper solution, because it would prevent
  Emacs from working under any debugger.  But you can change init_data
  in data.c if you wish.

Stardent i860 (i860-stardent-sysv4.0)

  19.26 pretest reported to work.

Stardent 1500 or 3000

  See Titan.

Stride (m68k-stride-sysv)

  Works (most recent news for 18.30) on their release 2.0.
  For release 2.2, see the end of `src/m/stride.h'.
  It may be possible to run on their V.1 system but changes
  in the s- file would be needed.

Sun 3, Sun 4 (sparc), Sun 386 (m68k-sun-sunos, sparc-sun-sunos, i386-sun-sunos,
                               sparc-sun-sunos4.1.3noshr, sparc-sun-solaris2.*,
                               i386-sun-solaris2.*, sparc*-*-linux-gnu)

  To build a 64-bit Emacs (with larger maximum buffer size and
  including large file support) on a Solaris system which supports
  64-bit executables, use the Sun compiler, configuring something like
  this (see the cc documentation for information on 64-bit
  compilation):  env CC="cc -xarch=v9" ./configure

  As of version 2.95, GCC doesn't support the 64-bit ABI properly, but
  later releases may.

  Some versions of Solaris 8 have a bug in their XIM (X Input Method)
  implementation which causes Emacs to dump core when one of several
  frames is closed.  To avoid this, either install patch 108773-12
  (for Sparc) or 108874-12 (for x86), or configure Emacs with the
  `--with-xim=no' switch (you can use Leim input methods instead).

  On Solaris 2.7, building Emacs with WorkShop Compilers 5.0 98/12/15
  C 5.0 failed, apparently with non-default CFLAGS, most probably due to
  compiler bugs.  Using Sun Solaris 2.7 Sun WorkShop 6 update 1 C
  release was reported to work without problems.  It worked OK on
  another system with Solaris 8 using apparently the same 5.0 compiler
  and the default CFLAGS.

  Emacs 21.1 and 21.2 built with Sun's ProWorks PC3.0.1 compiler on
  Intel/Solaris 8 was reported to abort and dump core during startup.
  Using GCC or a newer SUN compiler (Sun WokShop 6 update 2 C 5.3
  2001/05/15) solves the problem.

  Emacs 20.5 and later work on SPARC GNU/Linux with the 32-bit ABI.
  As of release 2.95, GCC doesn't work properly with the 64-bit ABI
  (applicable on UltraSPARC), but that isn't the default mode.

  Emacs 20.3 fails to build on Solaris 2.5 if you use GCC 2.7.2.3.
  Installing GCC 2.8 fixes the problem.

  19.32 works on Solaris 2.4 and 2.5.  On Solaris 2.5
  you may need one of these patches to prevent Emacs from crashing
  when it starts up:
        103093-03: [README] SunOS 5.5: kernel patch (2140557 bytes)
        102832-01: [README] OpenWindows 3.5: Xview Jumbo Patch (4181613 bytes)
        103242-04: [README] SunOS 5.5: linker patch (595363 bytes)

  There are reports that using SunSoft cc with -xO4 -xdepend produces
  bad code for some part of Emacs.

  Emacs works ok Sunos 4.1.x
  provided you completely replace your C shared library
  using one of the SunOS 4.1.x jumbo replacement patches from Sun.
  Here are the patch numbers for Sunos 4.1.3:
   100890-10   SunOS 4.1.3: domestic libc jumbo patch
   100891-10   SunOS 4.1.3: international libc jumbo patch

  Some people report that Emacs crashes immediately on startup when
  used with a non-X terminal, but we think this is due to compiling
  with GCC and failing to use GCC's "fixed" system header files.

  Some Sun versions of X windows use the clipboard, not the selections,
  for transferring text between clients.  The Cut, Paste and Copy items
  in the menu bar Edit menu work with the clipboard.

  It's important to include the SunOS version number in the
  configuration name.  For example, for SunOS release 4.0 on a Sun 3,
  use `m68k-sun-sunos4.0'; for SunOS release 4.1 on a Sparc, use
  `sparc-sun-sunos4.1'.  For SunOS release 4.1.3 on a Sparc, use
  `sparc-sun-sunos4.1.3'.  Note that shared libraries are now
  used by default on SunOS 4.1.

  A user reported irreproducible segmentation faults when using 19.29
  on Solaris 2.3 and 2.4 after compiling it with the Sun compiler.
  The problem went away when GCC 2.7.0 was used instead.  We do not know
  whether anything in Emacs is partly to blame for this.

  X11R6 is set up to make shared libraries only, on Sunos 4.
  Therefore, in order to link Emacs, you need to create static X libraries.
  To do this, rebuild X11 after setting
    #define ForceNormalLib YES
    #define SeparateSharedCompile YES
  in site.def (after #ifdef AfterVendorCF).

  Use `m68k' for the 68000-based Sun boxes, `sparc' for Sparcstations,
  and `i386' for Sun Roadrunners.  i386 calls for Sunos4.0.

  If you compile with Sun's ANSI compiler acc, you need additional options
  when linking temacs, such as
     /usr/lang/SC2.0.1/values-Xt.o -L/usr/lang/SC2.0.1/cg87 -L/usr/lang/SC2.0.1
  (those should be added just before the libraries) and you need to
  add -lansi just before -lc.  The precise file names depend on the
  compiler version, so we cannot easily arrange to supply them.

  On SunOS 4.1.1, do not use /usr/5bin/cc.  You can use gcc or/usr/bin/cc.
  Make sure the environment variable LD_LIBRARY_PATH is not defined.

  Some people report crashes on SunOS 4.1.3 if SYSTEM_MALLOC is defined.
  Others have reported that Emacs works if SYSTEM_MALLOC is defined, and not
  if it is undefined.  So far we do not know why results vary in this way.
  The sources are set up so that SYSTEM_MALLOC is defined; if that crashes,
  or if you want the benefit of the relocating memory allocator, you can
  try enabling the #undef SYSTEM_MALLOC in src/s/sunos4-1-3.h.

  On Solaris 2, you need to install patch 100947-02 to fix a system bug.
  Presumably this patch comes from Sun.  You must alter the definition of
  LD_SWITCH_SYSTEM if your X11 libraries are not in /usr/openwin/lib.
  You must make sure that /usr/ucblib is not in your LD_LIBRARY_PATH.

  On Solaris 2.2, with a multiprocessor SparcCenter 1000, Emacs 19.17 is
  reported to hang sometimes if it exits while it has one or more
  subprocesses (e.g. the `wakeup' subprocess used by `display-time').
  Emacs and its subprocesses become zombies, and in their zombie state
  slow down their host and disable rlogin and telnet.  This is most
  likely due to a bug in Solaris 2.2's multiprocessor support,
  rather than an Emacs bug.

  On Solaris, do not use /usr/ucb/cc.  Use /opt/SUNWspro/bin/cc.  Make
  sure that /usr/ccs/bin and /opt/SUNWspro/bin are in your PATH before
  /usr/ucb.  (Most free software packages have the same requirement on
  Solaris.)

  If you have trouble using open-network-stream, get the distribution
  of `bind' (the BSD name-server), build libresolv.a, and link Emacs
  with -lresolv, by copying the #definition of LIBS_SYSTEM in
  src/s/sunos4-1.h to src/config.h.  This problem is due to obsolete
  software in the nonshared standard library.

  If you want to use SunWindows, define HAVE_SUN_WINDOWS
  in config.h to enable a special interface called `emacstool'.
  The definition must *precede* the #include "machine.h".
  System version 3.2 is required for this facility to work.

  We recommend that you instead use the X window system, which
  has technical advantages, is an industry standard, and is also
  free software.  The FSF does not support the SunWindows code;
  we installed it only on the understanding we would not let it
  divert our efforts from what we think is important.

  If you are compiling for X windows, and the X window library was
  compiled to use the 68881, then you must edit config.h according
  the comments at the end of `src/m/sun3.h'.

  Note that Emacs on a Sun is not really as big as it looks.
  As dumped, it includes around 200k of zeros between the
  original text section and the original data section
  (now remapped as part of the text).  These are never
  swapped in.

  To build a single Emacs that will run on Sun 2 and Sun 3
  HARDWARE, just build it on the Sun 2.

  On Sunos 4.1.3, the word is that Emacs can loop infinitely
  on startup with X due perhaps to a bug in Sunos.  Installing all of
  these Sun patches fixes the problem.  We don't know which of them
  are really relevant.

    100075-11 100224-06 100347-03 100482-05 100557-02 100623-03 100804-03
    101080-01 100103-12 100249-09 100496-02 100564-07 100630-02 100891-10
    101134-01 100170-09 100296-04 100377-09 100507-04 100567-04 100650-02
    101070-01 101145-01 100173-10 100305-15 100383-06 100513-04 100570-05
    100689-01 101071-03 101200-02 100178-09 100338-05 100421-03 100536-02
    100584-05 100784-01 101072-01 101207-01

Tadpole 68K (m68k-tadpole-sysv)

  Changes merged in 19.1.

  You may need to edit Makefile to change the variables LIBDIR and
  BINDIR from /usr/local to /usr/contrib.

  To give movemail access to /usr/mail, you may need to execute

    chmod 2755 etc/movemail; chgrp mail etc/movemail

Tahoe (tahoe-tahoe-bsd4.2 or tahoe-tahoe-bsd4.3)

  18.52 was known to work on some Tahoes, but a compiler bug intervenes
  on others.  Some Emacs versions have worked in Unisys 1r4
  (not in 1r3) and CCI I.21.

  If you have trouble compiling `lib-src/loadst.c', turn off the definition
  of DKSTAT_HEADER_FILE in `src/m/tahoe.h'.

Tandem Integrity S2 (mips-tandem-sysv)

  Changes merged in 18.56 but subprocess support is turned off.
  You will probably want to see if you can make subprocesses work.

  You must edit `lib-src/Makefile' to define LOADLIBES = -mld.

Tektronix XD88 (m88k-tektronix-sysv3*)

  The 19.26 pretest was reported to work.
  Minor changes merged in 19.19.

Tektronix 16000 box (6130?) (ns16k-tektronix-bsd)

  Emacs 17.61 worked.

Tektronix 4300 (m68k-tektronix-bsd)

  Emacs 19.26 pretest reported to work.

Titan P2 or P3 (titan-titan-sysv)

  Changes probably merged in version 19.

Ustation E30 (SS5E) (m68k-unisys-unipl)

  Changes merged in 18.52; don't know whether they work.

Vaxen running Berkeley Unix (vax-dec-bsd4.1, vax-dec-bsd4.2, vax-dec-bsd4.3),
              Ultrix (vax-dec-ultrix),
              System V (vax-dec-sysv0, vax-dec-sysv2), or
              VMS (vax-dec-vms)

  Works.

  See under Ultrix for problems using X windows on Ultrix (vax-dec-ultrix).

  18.27 worked on System V rel 2 (vax-dec-sysv2).

  18.36 worked on System V rel 0 (vax-dec-sysv0).

  Richard Levitte <levitte@e.kth.se> distributes a set of patches to
  Emacs 18.59 to make it work nicely under VMS.  Emacs 19 probably
  won't work very well, or even compile.  Levitte is working on a
  port, so these problems should be fixed in the near future.

Whitechapel MG1 (ns16k-whitechapel-?)

  May work.  Supposedly no changes were needed except in `src/m/mg1.h'
  file.  I do not know what Unix version runs on them.

Wicat (m68k-wicat-sysv)

  Changes merged as of 18.6; whether they work is unknown.
  See comments in `src/m/wicat.h' for things you should change
  depending on the system and compiler version you have.
\f
Here are notes about some of the systems supported:

Berkeley 4.1 (bsd4.1)

  Works on vaxes.

Berkeley 4.2 (bsd4.2)

  Works on several machines.

Berkeley 4.3 (bsd4.3)

  Works, on Vaxes at least.

Esix

  The following was written for Emacs 18.59 and has been
  slightly adapted for Emacs 19.  It may need more change to be correct.

  Use s/usg5-4.h for Esix System V 4.0.[34] systems if you also have
  XFree86.  If you insist on using the Esix X Window libraries, good
  luck.  s/esix5r4.h provides a starting point, but doesn't seem to
  work consistently.  The basic problems involve the need to load
  -lX11 *last* in the link command, and even then some things break.
  You get best results by installing XFree86 and forgetting about the
  Esix stuff unless you want to run IXI xdt3, which really only needs
  the Esix X11 shared libraries.

  To compile with XFree86, make sure that your LD_LIBRARY_PATH
  contains /usr/X386/lib.  Be careful if you also have the Esix X
  Window libraries that /usr/X386/lib appears *first* in the
  LD_LIBRARY_PATH.  Then define C_SWITCH_X_SYSTEM -I/usr/X386/include.

Linux (actually GNU/Linux)

  Most of the complete systems which use the Linux kernel are close
  enough to the GNU system to be considered variant GNU systems.  We
  call them "Linux-based GNU systems," or GNU/Linux for short.

  It is not coincidence that many of the other components used with
  Linux--including GNU Emacs--were developed specifically for the GNU
  project.  The GNU project was launched in 1984 to develop a free
  complete Unix-like operating system.  To reach this goal, we had to
  develop whatever system components were not available as freely
  redistributable software from some other source.

  The GNU project wants users of GNU/Linux systems to be aware of how
  these systems relate to the GNU project, because that will help
  spread the GNU idea that software should be free--and thus encourage
  people to write more free software.  See the file LINUX-GNU in this
  directory for more explanation.

Microport

  See under "Intel 386".

MSDOS

  For installation on MSDOS, see the file INSTALL (search for `MSDOG',
  near the end of the file).  See the "MS-DOS" chapter of the manual
  for information about using Emacs on MSDOS.

SCO Unix
  If you have TCP but not X, you need to edit src/s/sco4.h
  to define HAVE_SOCKETS.

  If you are using MMDF instead of sendmail, you need to remove
  /usr/lib/sendmail or modify lisp/paths.el before compiling.
  lisp/paths.el (which is loaded during the build) will attempt to use
  sendmail if it exists.

  If you are using SMAIL, you need to define the macro
  SMAIL in config.h.

System V rel 0 (usg5.0)

  Works, on Vaxes and 3bxxx's.
  There are some problems in 18.37 due to shortnames/cccp problems:
  use the emacs 17 cpp if you have it.

System V rel 2 (usg5.2)

  Works on various machines.
  On some (maybe all) machines the library -lPW exists and contains
  a version of `alloca'.  On these machines, to use it, put
     #define HAVE_ALLOCA
     #define LIB_STANDARD -lPW -lc
  in the `src/m/MACHINENAME.h' file for the machine.

  If you find that the character Meta-DEL makes Emacs crash,
  find where function init_sys_modes in sysdep.c sets sg.c_cc[VQUIT]
  and make it store 7 there.  I have as yet no evidence of whether
  this problem, known in HP/UX, exists in other system V versions.

System V rel 2.2 (usg5.2.2)

  In 5.2.2 AT&T undid, incompatibly, their previous incompatible
  change to the way the nlist library is called.  A different s- file
  is used to enable the other interface.

  They call themselves the right choice--can't they choose?

  Emacs version 18 unexec is currently not working properly
  on 5.2.2.  Nobody knows why yet.  A workaround is to define
  NO_REMAP.  It is not yet known whether this applies to all
  machines running 5.2.2.

System V rel 3 (usg5.3)

  Some versions of this system support ptys and BSD-style sockets.
  On such systems, you should define HAVE_PTYS and HAVE_SOCKETS in config.h.

  If you want to link Emacs with shared libraries, define
  USG_SHARED_LIBRARIES.

  You may have to add ANSI idempotence #-lines to your sys/types.h
  file to get Emacs to compile correctly.  This may be necessary on
  other pre-ANSI systems as well.

  On an AT&T 6386WGS using System V Release 3.2 and X11R3, the X support
  cannot be made to work.  Whether or not the GNU relocating malloc is
  used, the symptom is that the first call Emacs makes to sbrk(0) returns
  (char *)-1.  Sorry, you're stuck with character-only mode.  Try
  installing Xfree86 to fix this.

System V rel 4.0.3 and 4.0.4 (usg5.4)

  Supported, including shared libraries for ELF, but ptys do not work
  because TIOCGPGRP fails to work on ptys (but Dell 2.2 seems to have
  fixed this).  This failure is probably due to a misunderstanding of
  the consequences of the POSIX spec: many system designers mistakenly
  think that POSIX requires this feature to fail.  This is untrue;
  ptys are an extension, and POSIX says that extensions *when used*
  may change the action of standard facilities in any fashion.

  If you get compilation errors about wrong number of
  arguments to getpgrp, define GETPGRP_NO_ARG.

  The standard C preprocessor may generate xmakefile incorrectly.  However,
  /lib/cpp will work, so use `make CPP=/lib/cpp'.  Standard cpp
  seems to work OK under Dell 2.2.

  Some versions 3 and earlier of V.4, on the Intel 386 and 860, had
  problems in the X11 libraries.  These prevent Emacs from working
  with X.  You can use Emacs with X provided your copy of X is based
  on X11 release 4 or newer, or is Dell's 2.2 (which is a 4.0.3).
  Unfortunately, the only way you can tell whether your X11 library is
  new enough is to try compiling Emacs to use X.  If emacs runs, your
  X11 library is new enough.

  In this context, GSV4 and GSV4i are alternate names for X11R4.
  OL2.* is X11R3 based.  OL3 is in between X11R3 and X11R4, and may or
  may not work, depending on who made the Unix system.  If the library
  libXol is part of the X distribution, then you have X11R3 and Emacs
  won't work with X.

  Most versions of V.4 support sockets.  If `/usr/lib/libsocket.so'
  exists, your system supports them.  If yours does not, you must add
  #undef HAVE_SOCKETS in config.h, after the inclusion of s-usg5-4.h.
  (Any system that supports Internet should implement sockets.)

Ultrix (bsd4.3)

  Recent versions of Ultrix appear to support the features of Berkeley 4.3.
  Ultrix was at the BSD 4.2 level for a long time after BSD 4.3 came out.

  Ultrix 3.0 has incompatibilities in its X library if you have the
  Ultrix version of X (UWS version 2.0).  To solve them, you need to
  prevent XvmsAlloc.o in Xlib from being used.  Israel Pinkas says:

    I added the following lines to config.h after the X defines:

    #if defined(ultrix) && defined(X11)
    #define OBJECTS_SYSTEM calloc.o
    #endif

    Then I ran the following:

    ar x /usr/lib/libc.a calloc.o

  The problem is said to be gone in UWS version 2.1.

Uniplus 5.2 (unipl5.2)

  Works, on Dual machines at least.

VMS (vmsM.N)

  Richard Levitte <levitte@e.kth.se> distributes a set of patches to
  Emacs 18.59 to make it work nicely under VMS.  Emacs 19 probably
  won't work very well, or even compile.  Levitte is working on a
  port, so these problems should be fixed in the near future.

  Note that Emacs for VMS is usually distributed in a special VMS
  distribution.  See the file ../vms/VMSINSTALL for info on moving
  Unix distributions to VMS, and other VMS-related topics.

Windows NT/95/98/ME/2000

  For installation on all versions of the MS-Windows platform, see the
  file nt/INSTALL.

Xenix (xenix)

  Should work in 18.50, but you will need to edit the files
  `lib-src/Makefile' and `src/ymakefile'
  (see the comments that mention "Xenix" for what to change.)
  Compiling Emacs with -O is said not to work.

  If you want Emacs to work with Smail (installed as /usr/bin/smail)
  then add the line   #define SMAIL  to config.h.

  The file etc/XENIX suggests some useful things to do to Xenix
  to make the Emacs meta key work.
\f
Local variables:
mode: indented-text
fill-prefix: "  "
End: