home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
3D Images
/
3D Images.iso
/
programs
/
amiga
/
rayshade
/
inetray
/
history
< prev
next >
Wrap
Text File
|
1995-01-12
|
11KB
|
229 lines
======================================================================
H I S T O R Y
doc: Mon May 18 09:52:34 1992
dlm: Thu Aug 19 21:13:19 1993
(c) 1992 ant@ips.id.ethz.ch
uE-Info: 228 38 T 0 0 72 3 2 8 ofnI
======================================================================
05-18-92 [0.1.0]: - Added Version numbering
- Changed output from stdout to stderr
- Graceful continue when xdr_???() fail.
- Changed some output
05-19-92 : - Changed start: keyword to params: keyword
- nobcast: entries are started automatically
- RELEASED
05-20-92 [0.1.1]: - changed misspelled error message
- ignored hosts reacted to terminateAll() and exited
with Dispatcher disappeared.
05-22-92 : - added SNAKE_QUIRK (utils.c)
well, somewhere in here, I released patch #1.
06-05-92 [0.2.0]: - removed inetray.relay totally
- added directed broadcasts
- added arguments to inetray.ping (Use List)
- changed syntax of .inetrayrc (Use Lists)
06-13-92 [0.2.1]: - changed terminateAll() in comm.c to allow exit even
if some servers don't shut down correctly.
06-14-92 - inetray did not work with at(1). Fixed that.
06-15-92 - Added PRINTSCHEDULE flag to inetray.c to allow
checking of the printschedule (which was all right).
06-19-92 - Added a Rescheduling after half the image is completed.
All outstanding blocks on machines which have not yet
delivered a result are given priority.
- Changed something in inetray.c: If, while writing, some
msg arrives from a worker or server, writing is
interrupted.
- Removed an error in comm.c: When clnt_create() does
not work, the machine should just be ignored.
- added example mole.ray
- RELEASED patch.1
06-21-92 [0.2.2]: - changed comm.c to ignore ``program not registered''
errors on calls to START and TERMINATE. This mostly
makes inetray.ping nicer.
- added BLOCKSPERSERVER to config.h
06-24-92 - modified rpc.inetrayd.c: an obscure kernel bug on
some SUNs made wait give back a wrong pid's status.
Therefore, I modified rpc.inetrayd.c such that even
under those circumstances the worker is killed.
Note: It seems that this bug only surfaces on
heavily loaded machines.
06-26-92 - After rereading K&R last nite I found out that I
could get rid of wPrint() in utils.c - so I did.
- added MINBLOCKSIZE, MAXBLOCKSIZE & BLOCKSPERSERVER
to .inetrayrc; with those you can tweak the
scheduler to be more permissive about relative
machine speeds.
- changed scheduler, so that last block is smaller
than all the others.
- changed the handling of the SUN4 kernel bug
described above a bit. Should be less of a problem
now.
- RELEASED
06-28-92 [0.2.3] - the program netinfo has been added to allow users to
find the real addresses of their machines. This
program is much simpler to use than ifconfig.
- RELEASED
06-29-92 [0.2.4] - changed Makefile; does not use the install command
any more but is portable.
07-01-92 - Heh, heh; the ``kernel bug'' was not really a kernel
bug after all: popen also forks to call the cpp,
this process is never waited for. So I added that to
rpc.inetrayd.c.
07-03-92 [1.0.0] - Decided that it's time to release Inetray
officially. [0.2.4] is therefore never released.
- Changed something in comm.c to display the internet
numbers nicely (columns).
- Bug in comm.c when gethostbyaddr() returns NULL due
to a system error.
- RELEASED
07-05-92 [1.0.1] - There was a bug with blocksize == 1. This bug was
introduced with version 0.2.1 when the scheduler
changed. The case of blocksize == 1 is now handled
separately in irtrace.c.
02-19-93 [1.0.2] - illegal uid reported in syslog was wrong (rpc.inetrayd.c)
- rc-file error reported on commandline error (rcfile.c)
02-26-93 - Add /tmp_mnt patch sent by roland@jts.com
03-18-93 - changed email addr
04-26-93 - to disallow ``inetray < foo'' initSelf() (inetray.c)
was changed.
- changes to disallow ``inetray'' (i.e. reading the
input from the keyboard) mainly in inetray.c (also
common.h and irtrace.c changed due to change in
scope of constants IN,OUT,ERR). Doc in POO.
- Note: patch.2 has never be released.
PLAN: Version 2.0.0 should be usable as an explorer module. In a first
step Inetray should run in a pipe and with its std-input
re-directed. This means that the std input should be sent to
the workers (before or after CPP?). Since no worker should be
able to write files, the input should probably be buffered in a
``life buffer''. The main remaining problem is that the input is
to be buffered on the dispatcher machine before running the
parser there. This is a problem when the std-input is a TTY: how
can I find out (without parsing the options myself) if the input
is expected from the keyboard? It would be acceptable to
disallow this mode of operation completely. This was achieved
with the last change described above!
04-28-93 [1.0.3] - Intermediate version. With this version, the stdin
problem was solved (see POO).
- Note: patch.3 has never be released.
05-07-93 [1.1.0] - Inetray now is completely compatible to Rayshade.
Stdin is sent to all workers before rendering starts.
Referenced files must still be accessible as before
(#include and heighfield files).
05-10-93 - All the docu has been brought up to date.
- The new version has been made.
05-13-93 [1.1.1] - replace STARTDELAY by rTimeout (settable)
- kill lBuffer with SIGUSR1 (instead of SIGINT). Like
this ^C from the keyboard should work.
05-17-93 - add dummy pipe to fill() (lbuffer.c) to detect death
of parent on dispatcher. Nobody writes to statusP, so
when select() returns ready to read it means that the
parent has died.
05-21-93 [1.2.0] - allow multiple servers to be started when inetd is
used. This requires some changes in the *.x files,
inetray.start.c rpc.inetrayd.c and comm.c.
- add version number to the info passed back to the
dispatcher.
- seems the CSCS net drops many packets. This is
especially annoying for STARTFRAME,TRACEBLOCK and
TERMINATE.
05-24-93 - some changes to allow packets to be lost and to
compile some statistics concerning this.
05-25-93 - Continue working on the ``lost packets'' problem. Not
so shure any more it's really a network problem.
Could just be that the worker with id 0 somehow takes
the other worker(s) on the same host down with him
(signals? stderr redirection?).
05-27-93 - Fucking embarassing: CSCS net does NOT drop packets!
What was happening was that I forgot to svc_destroy()
the olde service for the secondary servers; like
this, all secodary servers were also listening to the
port of the primary server and thus it was pure luck
who got the request.
06-02-93 - servers started with inetray.start did not unregister
themselves before exiting (which they should). This
is due to a bug in rpc.inetrayd.c where exit()
instead of exitClean() was called a few times.
- inetray.start now checks first if there is already an
inetray.start running on the same machine before
starting up. It does this by issueing a
callrpc(`hostname`,STARTER,IRSV1,0) and checking the
return value.
06-03-93 - Due to a weird Ultrix (bug?) feature, the open syslog
does not seem to survive a close(2) (as done by
dup2() or close()). To re-activate the syslog it has
to be closed and opened again (rpc.inetrayd.c).
- Some minor correction to inetray.start.c (visual
improvement).
06-14-93 - changed registering such that the PID is always
visible (up to 5 digits).
- removed syslog bugs from inetray.start.c (forgot to
put LOG_ERR to some messages).
06-17-93 - changed order of initialization in inetray.c such
that having no write permission in the directory does
not lead to a total crash.
06-18-93 - added the irstart shellscript to the distribution.
Not documented so far.
06-22-93 - allowed sendrpc() to have its very own timeout. This
meant duplicating a lot of SUN RPC code.
07-08-93 - Don't read /dev/tty if parent dead (i.e. was started
from make, for example.
- Add waitpid() whenever rpc.inetrayd checks for the
dispatcher. Like this, it also exits if the child is
killed due to a rayshade internal error (hopefully).
07-09-93 - Replace all wait() by waitpid() in rpc.inetrayd.c.
This is cleaner.
- Changed inetray.c such that the first interrupt leads
to a shutDown(0) if the father has died (and thus the
pgrp has changed).
07-10-93 - Added stderr redirection test to rpc.inetrayd.c since
redirection seems to work only spuriously.
- Added a cleanup handler to inetray.start.c which
deregisters itself after a signal.
- Changed rpc.inetrayd.c such that empty error files
are unlinked on exit.
- Added NOTEMPNAM_QUIRK for the Convex.
07-11-93 [2.0.0] - Changed scheduling. Removed rescheduling after
completing half the image. On the first pass, double
blocks are rendered; once the first pass is over,
only single blocks are calculated. This should
improve overall performance if images contain parts
with greatly differing complexity.
07-20-93 - Started adding the NOSYSLOG_QUIRK
- Added the generr utility
- Added the depend target to the Makefile
07-21-93 - Continued on the stuff above
07-22-93 - Started moving it to SunOS 5 (Solaris). What a bummer
but it'll help me clean up the syscalls I use.
- Bug in inetray.start: the warning saying that there
already is a version running works only every other
time (i.e. it deregistrates itself anyway!)
07-23-93 - Removed machine specific quirks (SNAKE & AUX).
- Updated Makefile
- Updated doc
08-17-93 - Somewhere along the line the whole version got
SysVRel4-clean. For the NeXT (Mach) I added a few BSD
syscalls and generally moved things in the Makefile
around a lot.
08-19-93 - Bug in rpc.inetrayd.c: during wait the syscall can be
interrupted by the timer. This might well be a
general problem!!!!