cgl@infidel.LANL.GOV (Chris Langton) (02/04/91)
Cellsim Bug Fix
========
This message is to announce the availability of a patch for Cellsim,
which fixes a problem encountered when running under SunOS 4.1 or newer.
The patch itself is appended at the end of this message.
This bug causes cellsim to hang --- and freezes the whole window system ---
if the cursor is moved out of the control panel while a run is in progress.
Cellsim Distribution List
-------------------------
This message is being sent to the CA mail-list. We are maintaining a Cellsim
distribution list, in order to keep Cellsim users up to date on new
releases and bug fixes. If you would like to be on this list, please
send a message to cgl@t13.lanl.gov. You will get two copies of this message
if you are on both lists - sorry!
For those unfamiliar with Cellsim
---------------------------------
Cellsim is a SunView-based cellular automata simulator, which can run
on a lattice of size up to 512x512. It can perform calculations
locally on a Sun workstation, or on a Connection Machine supercomputer
(either directly or through the network). On the CM, it is possible
to do some non-CA extensions, such as floating-point calculations and
so on. And of course, you'll get some fantastic speed on the CM.
Cellsim is still available for anonymous FTP, from
archive.cs.rpi.edu (128.213.10.18), in the directory "cellsim". The
tarfile "cellsim_2.5.tar.Z" contains the latest version of Cellsim,
V2.5 There are also some patches available in subdirectories there,
but the tarfile currently available has ALREADY HAD THE PATCHES
INSTALLED. (Sorry to shout, but I want to be sure no one misses that).
The future of Cellsim
---------------------
Cellsim was written by a few different authors, and the source-code is
really pretty ugly and getting hard to maintain at this point. I am
planning on eventually releasing a new, totally rewritten version,
that will probably run under X-windows (but which will be much easier
to port to other platforms than Cellsim is).
One of my main motivations for doing so is to take advantage of
several ideas I have for optimizing the calculations on the Connection
Machine, and taking advantage of some of the CM's features that are
not tapped into by the current Cellsim. I realize that not everyone
has access to a CM, so future Cellsims will still be able to run the
calculations locally as well.
The problem and the patch
-------------------------
Many people have noticed a pretty terrible bug in Cellsim, when
running on SunOS 4.1 or newer. The problem is that, if you start
running a rule, and move the mouse cursor outside of the Cellsim
window while the rule is running, SunView will essentially hang. You
won't be able to select any other windows, or stop the run by clicking
on the left mouse-button as you should be able to.
**NOTE: You should install this patch, even if you don't yet use SunOS
4.1, since in the future you may upgrade, and it's best to be
prepared. Also, any future patches to Cellsim may rely on this one
being already installed.
With the help of someone at Sun (Matthew Stier, to whom I am *very*
grateful for his help), I've fixed the problem in Cellsim.
Apparently, something was changed in the event-handler or notifier in
the SunView libraries when 4.1 came out. Actually, I'm not sure if
Cellsim *should* have worked even under 3.5 -- it may have been using
an undocumented property of SunView. I'm not familiar enough with the
low-level details of SunView, and I didn't write that part of Cellsim,
so I'm not sure about it.
Three routines in the "cellscr.c" file need to be changed, to fix
the problem. These routines are check_button(), set_run_mode(), and
unset_run_mode().
I've appended a patch at the end of this message, which you can
apply to your copy of cellscr.c (you will then need to recompile
Cellsim). To use the patch, save everything between the "==cut here=="
labels into a file. Assuming you call the file "4.1_patchfile", in the
same directory with cellscr.c, you can apply the patch by doing:
patch cellscr.c 4.1_patchfile
You should see a message indicating that 2 hunks succeeded.
Patch will make a backup copy of your original cellscr.c, called
"cellscr.c.orig", and will create a new, modified cellscr.c with the
patches installed. You should then be able to recompile Cellsim.
If you don't have the "patch" command, you can get a copy of the
three routines that were modified, via one of 2 methods:
1. Through anonymous FTP to archive.cs.rpi.edu (128.213.10.18). Use
the login name "anonymous", and your e-mail address as a password.
Look in the directory "cellsim/patch2", for a file named
"4.1_cellscr.c". You can replace your "cellscr.c" file with this
version, assuming you haven't made any local changes to your
cellscr.c.
If you *have* modified your cellscr.c file already, then you should get
the file "4.1_functions". This file contains only the three functions
that were changed. You should manually replace the old versions of
these routines in your cellscr.c file with these new versions.
The patchfile is also in the FTP archive, although you shouldn't
need it, since it's appended below.
2. If you aren't able to FTP, then send me a note requesting either
of the above files, and I will send it to you. I'm awfully busy
lately, so please use FTP if you can.
A linker bug
------------
Some people are encountering linker errors when trying to load in
computed-function rules (getting "ld: -T" errors). I haven't figured
this one out yet -- hopefully soon. Actually, someone else had that
problem more than a year ago, and we solved it then. The problem is,
I don't remember how we solved it, and I've moved so many times since
then, I can't find the bug-report where I recorded the solution! If
the person who had this problem is reading this, and remembers how we
fixed it, could you please let me know? Thanks. :-)
I realize I probably should have waited and fixed both bugs at once
with one patch, but the SunView problem is so severe, I wanted to make
this patch available as soon as possible. Also, I don't know if many
people are encountering the linker problem; it seems pretty rare.
Disclaimer
----------
I feel I better add this -- despite the fact that it is a tool for
doing computations on the Connection Machine (and Sun workstations),
Cellsim is in no way supported by Thinking Machines Corporation. I
worked on Cellsim for quite some time before joining TMC. If you have
any questions/comments/complaints about Cellsim, please direct them to
me (hiebeler@think.com) or Chris Langton (cgl@t13.lanl.gov), and not to
Thinking Machines Corporation or Los Alamos National Laboratory.
Here is the patch, below. Use everything between the "===cut here==="
delimiters (there is one at both ends -- don't forget to delete my
signature at the end).
===cut here===
*** cellscr_orig.c Thu Jan 31 08:36:19 1991
--- cellscr.c Thu Jan 31 08:39:09 1991
***************
*** 1707,1723 ****
set_run_mode()
{ /* disable panel & canvas input */
- int fd; /* window file descriptor */
frame_mask = *((Inputmask *) window_get(frame, WIN_PICK_INPUT_MASK));
! window_set(frame, WIN_CONSUME_PICK_EVENT, MS_LEFT, 0);
panel_mask = *((Inputmask *) window_get(panel, WIN_PICK_INPUT_MASK));
! window_set(panel, WIN_CONSUME_PICK_EVENT, WIN_NO_EVENTS, 0);
canvas_mask = *((Inputmask *) window_get(canvas, WIN_PICK_INPUT_MASK));
! window_set(canvas, WIN_CONSUME_PICK_EVENT, WIN_NO_EVENTS, 0);
! fd = (int) window_get(frame, WIN_FD);
! fcntl(fd, F_SETFL, FNDELAY);
! /* set non-blocking mode */
}
--- 1707,1719 ----
set_run_mode()
{ /* disable panel & canvas input */
frame_mask = *((Inputmask *) window_get(frame, WIN_PICK_INPUT_MASK));
! window_set(frame, WIN_CONSUME_PICK_EVENT, WIN_NO_EVENTS, 0);
panel_mask = *((Inputmask *) window_get(panel, WIN_PICK_INPUT_MASK));
! window_set(panel, WIN_CONSUME_PICK_EVENT, MS_LEFT, 0);
canvas_mask = *((Inputmask *) window_get(canvas, WIN_PICK_INPUT_MASK));
! window_set(canvas, WIN_CONSUME_PICK_EVENT, MS_LEFT, 0);
}
***************
*** 1732,1746 ****
check_button()
{ /* return TRUE iff the left button has
* been pressed */
! int err, button;
Event event;
! do { /* check all events in pending queue */
! err = window_read_event(frame, &event);
! button = ((err == 0) && (event_is_up(&event)) &&
(event_id(&event) == MS_LEFT));
! } while ((err == 0) && !button);
/* if no events remain, err will equal -1 */
! return (button);
}
--- 1728,1762 ----
check_button()
{ /* return TRUE iff the left button has
* been pressed */
! int err, button1, button2;
! int fd1, ff1, fd2, ff2;
Event event;
!
! /* set non-blocking mode in panel */
! fd1 = (int) window_get(panel, WIN_FD);
! ff1 = fcntl(fd1, F_GETFL, 0);
! fcntl(fd1, F_SETFL, FNDELAY);
! /* non-blocking mode in canvas */
! fd2 = (int) window_get(canvas, WIN_FD);
! ff2 = fcntl(fd2, F_GETFL, 0);
! fcntl(fd2, F_SETFL, FNDELAY);
! do {
! err = window_read_event(panel, &event);
! button1 = ((err == 0) && (event_is_up(&event)) &&
(event_id(&event) == MS_LEFT));
! } while ((err == 0) && !button1);
! do {
! err = window_read_event(canvas, &event);
! button2 = ((err == 0) && (event_is_up(&event)) &&
! (event_id(&event) == MS_LEFT));
! } while ((err == 0) && !button2);
/* if no events remain, err will equal -1 */
!
! /* Restore blocking mode in panel and canvas */
! fcntl(fd1, F_SETFL, ff1);
! fcntl(fd2, F_SETFL, ff2);
! /* if button was pressed in either panel or canvas, return 1 */
! return (button1 | button2);
}
===cut here===
--
Dave Hiebeler | Internet: hiebeler@think.com
Thinking Machines Corporation | Phone: (617) 234-4070 (work)
245 First Street |
Cambridge, MA 02142 USA |