Path: utzoo!attcan!uunet!munnari.oz.au!bruce!goanna!minyos!chudich!rcodi From: rcodi@chudich.co.rmit.oz (Ian Donaldson) Newsgroups: comp.sys.encore Subject: Re: Umax4.3 & NFS (& shared memory size) Message-ID: <1913@minyos.xx.rmit.oz> Date: 22 Nov 89 07:40:13 GMT References: <1536@bnlux0.bnl.gov> <1761@minyos.xx.rmit.oz> <1019@maxim.erbe.se> Sender: news@minyos.xx.rmit.oz Distribution: comp Lines: 111 prc@erbe.se (Robert Claeson) writes: >Even more funny, we went to UMAX V about one year ago and haven't regret it >since then. And the latest version of UMAX V, 2.2, even includes a complete ^^^^^^^^??? >4.3BSD environment with all 4.3BSD libraries, system calls, commands, job ^^^^ ???? >control, nroff/troff/ditroff, merged SYSV/BSD tty drivers and tape device >drivers and more. No, I'm quite happy with UMAX V (but then, I often have the >/bsd directory hierarchy first in my PATH). This is FAR from correct. Sure, under 2.2f, there are lots of BSD things, but the set is far from complete. For instance, the following indicates incompatability with BSD: cc no -M option chgrp no -R option chmod no -R option chown no -R option, can't set group & user at the same time cp no -p option df it reports 512 byte blocks, not 1k, output format is totally different du it reports 512 byte blocks, not 1k, doesn't work right over NFS last doesn't exist lastcomm doesn't exist lpc doesn't exist lpq doesn't exist lpr doesn't exist lprm doesn't exist lptest doesn't exist ls (it reports 512 byte blocks in the -s option, doesn't do multi-column output if output is to a tty, -g option default reversed) make pmake is closest (we symlink it to make) man SV man. Yuk! It doesn't even invoke a pager! mkdep doesn't exist pac doesn't exist ping doesn't exist ---<< serious deficiency pwd broken on NFS filesystems (it just hangs) ranlib doesn't exist (we've got a 1-line script to replace it) su doesn't check for group wheel permission before su to root doesn't set the environment to the target user tip doesn't exist gcore doesn't exist pi/px/pdx doen't exist dbx doesn't exist (cdb has a few dbx commands, but not many) script doesn't exist window doesn't exist users doesn't exist w doesn't exist pstat doesn't exist quotackeck -- the one supplied generates bad counts for random users binmail has serious security problems. ucbMail invokes the wrong vi (the AT&T one) which doesn't know about job control ps is the SV version... how do you tell the RSS of a process? vmstat doesn't exist (sar is not quite as nice & compact) iostat doesn't exist (sar is not quite as nice & compact) Now many of the above we ported from 4.3-tahoe to 2.2f and 2.2h, just to make the machine bearable. The lack of BSD network line printer system is really sad. We ported Berkeley lpd (after hacking it due to lack of Unix domain sockets) The above list isn't complete either. There are lots of other BSD things missing. We only ported the things we needed most. In addition, one MAJOR problem with UMAXV is that ordinary users can't mv directories around, other than rename them. Even over NFS!! Also, there are 2 rename(2) system calls (one in the AT&T library, and one in the BSD) and BOTH of them don't work, forcing you to write your own using link(2)/unlink(2) stat(2) returns bogus numbers for the blocks in a file, even over NFS, making du and 'ls -s' over NFS useless. We even ported 4.3-tahoe csh as the one supplied hung under 2.2f and crapped out under 2.2h (the problem is that the Encore C compiler is being smart with register allocation, which csh can't tolerate without modification.... use -q nocompilerregisters to work around it) The way BSD is merged with AT&T on the Encore leaves lots to be desired. For instance, many routines (such as statfs(2), ftok(3)) are only in the AT&T libraries, and you can't link code that is compiled in one universe with libraries compiled in the other due to kernel struct differences (mainly due to two different stat(2) structures). The way Sun have put it together in SunOS is FAR superior, although they started from BSD and added SV in where necessary. UMAXV is the reverse. However I must praise Encore for fixing a few major bugs for 2.2h that were present in 2.2f (eg: rlogind importing TERM). The list of bugs is still huge however. A few things broke in 2.2h too, notably NFS client mode and symbolic links, making NFS client mode useless and impossible to use in our environment (and probably everybody elses' too). NFS server mode is almost OK however except for lack of mv'ing directories and the blocksize reported by stat(2). Also, truncate(2) doesn't work over NFS and certain files (eg: a.out files) get mode 777 when you create them over NFS for some odd reason, indicating some low level NFS bug. Enough! Ian D