[comp.emacs] 18.51 on Sun386i

pml@serendip.UUCP (PM Lashley) (08/31/88)

I have just finished installing 18.51 on a Sun386i (SunOS 4.0.0 = BL8H.5)
and have the following observations.  (I have recently returned to the net.
If some of this is old news to you, please forgive the redundancy...)

1)  The etc/MACHINES file says:

>	Sun Roadrunner (m-sun386.h; s-bsd4-2.h?)

    I had better results with s-sunos4.h.


2)  The following patch was required to unexec.c

*** unexec.c~    Fri Jul  1 17:34:14 1988
--- unexec.c  Tue Aug 30 18:40:41 1988
***************
*** 264,277 ****
--- 264,279 ----
  #ifdef COFF
  #ifndef USG
  #ifndef STRIDE
  #ifndef UMAX
+ #ifndef sun386
  /* I have a suspicion that these are turned off on all systems
     and can be deleted.  Try it in version 19.  */
  #include <filehdr.h>
  #include <aouthdr.h>
  #include <scnhdr.h>
  #include <syms.h>
+ #endif /* not sun386 */
  #endif /* not UMAX */
  #endif /* Not STRIDE */
  #endif /* not USG */
  static long block_copy_start;               /* Old executable start point */


3) It was necessary to add the following lines to the end of m-sun386.h

/* Sun386i object files do not use the traditional initial underbar, so we */
/* must override the extra one from s-sunos4.h        */
#undef  LD_SWITCH_SYSTEM
#define LD_SWITCH_SYSTEM -e _start -Bstatic


4)  I am in the rare position of having two Sun workstations in my office.
I am normally logged in simultaniously on both of them under the same userid,
and through the wonders of NFS, using the same home directory.  This does not
interact well with the emacs server's socket (~/.emacs_server).  Does anyone
out there have any good reasons not to use gethostname() to make the socket
id unique to the host?  If that is done, what should be done about the single
reference within server.el?  Is there a reasonable way of obtaining the
hostname from lisp?  Would the preference for the new socket name be
a) .emacs_hostname, b) .emacs_server_hostname, c) .hostname_emacs_server,
d) .hostname_emacs, e) other, specify __________________ ?   Or would it be
better to add an (optional) environment variable to specify the socket name?


5)  In an attempt to regularize access to home directories, third party
software, etc. in an hetrogoneus network, the RoadRunner uses a Yellow Pages map and automount.  Under this scheme, emacs would normally be placed on a
server in directory /files/vol/emacs.  However this would be automounted
READ ONLY as /vol/emacs by the server and all network clients accessing the
application.  According to this technique, emacs should use the envrionment
variable emacs_ROOT instead of depending on any fixed installation point.
(For details on how emacs_ROOT would be automatically set up see the the text
at the end of /usr/etc/start_applic, and the Sun386i Advanced Administration
manual.)  /vol/emacs would contain additional subdirectories ./bin for shell
scripts, and ./bin.`arch` for binaries for each architecture supported.

Acceptance of this scheme raises several questions.  What mechanism would
be used to specify an environment variable to be prepended to the `built in'
paths?  (#define PATH_ENV_VAR emacs_ROOT ?)  How would this interact with
unexec?  Is there any reasonable way to extend this scheme to cover .o files?
What about version.el being updated for each architecture?


PMLashley	sun!cohesive!kla!pat
KLA Instruments	POBox 5016, 3901 Burton Drive
		Santa Clara, CA 95052
		408/988-6100