Problem & question with AIX 3.1.3

[Chin Fang]

In article <1991May24.131032.16139 at news.larc.nasa.gov> jcburt at ipsun.larc.nasa.gov (John Burton) writes:
>I've got a minor problem with my RS/6000 running AIX 3.1.3.
>When I login to the system console (which is our primary workstation)
>I usually start my Xwindow environment with an "open xinit"
>(i.e. open a virtual screen). Every so often (not consistently at all)
>when I'm in Xwindows using Motif, the screen will blank and then display
>the login prompt (just like I logged completely off in one fell swoop).
>If I press ALT + CTRL/ACT I get my windows screen back and everything
>seems fine. The obvious question is, what did I do to cause this...well,

The entire virtual terminal and multiple servers on one machine is a BIG 
JOKE!  Why IBM wants to run more than one server on one machine?  320's are
powerful but not that powerful :-(.  If I see someone do this on any 320,
I will freeze her/his account!   The virtual terminal switching is a big
kludge.  How can you expect people having console security with this?

A better kludge would be presenting a new login in prompt when Alt-Act is
pressed.

Next, consider this,  I need to run some FEM codes which may require me
to allocate big memory to my FORTRAN COMMONs, now I got 16Megs on each
machine, so even with one X I am fine.  Now suddenly a jerk fires up two
X servers and my response immediately becomes terrible.  Whose fault?

Upgrading memory capacity is not the point since I don't control budget.

I have to use tons of kludge to prevent people using this.  I am serious.
we have to spend hours a day sometimes to fend off off-campus system breakers.
So anything that may create security hole is of my concern.

>I don't know...I wasn't even touching the keyboard or mouse (just reading
>a News article with my hands away from the machine. There were no other
>users on the system at the time...) Any suggestions? or is AIX3.1.3 just that
>flakey?
>

>From my own experience with 12 320s daily, all running 3.1.3, my own answer
to you is yes. It's flaky, both the OS and X server.  I don't use any IBM 
supplied clients, except xde and info.

Not even Motif mwm.  I use twm, it's flaky too running under IBM X, but
in general is tolerable.

>The question I have is...WHEN IS IBM GOING TO GET WITH IT AND UPGRADE TO
>X11R4 and MOTIF 1.1?????
>

You can get Dan Greening's two patches for X11R4 clients/libs from 
export.lcs.mit.edu.  cd contrib, then get ibm.rs6000,fix[1-2]. and follow 
the instructions in the patch files.

>Its kinda frustrating to have many X11 compliant packages available, but
>not able to use them because they NO LONGER SUPPORT X11R3!!!
>

You can partially overcome this by building R4 clients/libs and try to port
other things youself.  I have done so and am still doing.  Other than all 
R4 standard clients that come in Tape #1, You can port quite a lot other 
things both for X and non-X environment without too much trouble.  The 
most time consuming part for me is to deal with IBM's headers for C srcs.

For instance, yesterday I was porting elm 2.3 with 11 patches applied,  I
spent almost two hours editing out some standard C function declarations in
elm src in order to build this guy using cc.  Kind of tiring.  So afterwards
I built acm on my 320, the multiple user airieal combat game, just to get some
relief.  After dinner, I put in Xtetris too and wasted a lot time playing with
it (I didn't tell my boss  :-)   

A short conclusion is that even bsdcc discussed in /usr/lpp/bos/bsdport 
can fail you quite spectecularly for BSD codes.  I built perl 4.003 using 
just cc with my own twicking of config.sh and only had one of the group
tests failed.  My own feeling is that between cc and bsdcc, there is really
no clear winner as to which one is better for porting BSD codes.

Note, Dan's patches may be handy for many things, but not for all!  A case is 
xlock.  If you use xmkmf after you have installed Dan's libs/includes/clients
(seriously, this would take you one morning to do from building to done),
you will NOT be able to build xlock.  I hacked the Makefile that came with
xlock in the contrib/clients/xlock dir and just built it using cc -O without any
trouble at all.  Another thing is that if you want to use the very handy
xwebster, you have to port Xw HP wedget lib first, get it from contrib/
of export.lcs.mit.edu.  Not too much hassle to set it up.

Dont' worry about IBM.  There is a will, there is a way.  And there are also
Suns available for a far more smoother computing environment.

But as far as the hardware is concerned, thumb up to IBM's Austin Div.
Good job.  Sun and Dec are no match.  (but how about HP 9000s?... later..)

Regards,

Chin Fang
Mechanical Engineering Department
Stanford University
fangchin at leland.stanford.edu