tytso@ATHENA.MIT.EDU Linux_Activists 11/07/91 14:27 (20 lines)
Subject: Re: nic.funet.fi unreachable
From: tytso@Athena.MIT.EDU (Theodore Ts'o)
A quick scan using Archie shows that none of the FTP sites for Linux
that it knows about are on the U.S. side of the Atlantic. In the hopes
of reducing inter-atlantic traffic and reducing the load on
nic.funet.fi, I've made Linux-0.10 available for anonymous FTP on
TSX-11.MIT.EDU (18.104.22.168). I will make an attempt to keeps things
I've just recently heard about Linux from the Hurd mailing list, and
from looking at the source code it looks very, very exciting. I haven't
managed to install it on my hard disk yet (it looks like I'll need to
blow away OS/2 in order to reclaim one of the four primary partitions
--- shucks), but just from looking at the source code there are a bunch
of things which look like interesting projects --- like supporting DOS
extended partitions and multiple threads per task. Now, all I need to
do is find some time to do some playing.... :-)
Subject: Weekend hacking and other randomness...
Date: Mon, 11 Nov 91 16:08:15 -0500
From: tytso@ATHENA.MIT.EDU (Theodore Ts'o)
Over the weekend, I've managed to put together some stuff which other
people might find useful. First of all, I've managed to build GNU diff,
GNU sed, Larry Wall's patch and tr. They pretty much built without
needing to make any changes --- the only real problem which I
encountered while trying to build them is that the configuration progam
which you need to run to build sed itself wants to uses sed to build the
Makefile. Oops. :-)
In any cases, if anyone wants to just snarf the binaries, they can be
found via anonymous FTP on TSX-11 in /pub/linux/binaries. If you want
to submit some something Linux-related for the FTP site, you can put the
file(s) in /pub/incoming and send me a quick note describing what you
I've made a quickie change to the tools/build.c program so that it can
optionally take a fourth argument which specifies what the root device
should be. Add to that a definition in the makefile for $(ROOTDEV), and
customizing kernels for a particular system becomes a snap! Since the
patch isn't that big, I'll include the diffs at the end of this message.
Other comments; maybe I'm doing something stupid, but I can't seem to
get mtools to want to write a file from Linux land to MS-DOS land. The
other way works fine, but when I try something like "mcopy foo.tar
c:foo.tar", the program just returns to the prompt without actually
performing the copy. Another bug: if you do "mcopy c:foo.tar .", the
file which will be created in your directory will be "???.???".
So you can't just specify a unix directory as the target; you actually
need to give the name that you want in the directory or the name will be
unrecognizable. This also happens if you try "mcopy a:*.* .". Final
problem: the program will hang if the destination file already exists;
you need to control-C out of it when this happens.
Also, trying to compile RCS for Linux apparently tickles a compiler bug
in gcc. When you try to compile ci.c, gcc gets an "unrecognizable insn"
error. When I mentioned to a friend of mine who is a gcc expert, he was
a bit surprised because he thought all of these bugs had been fixed in
version 1.40. His suggestion was that I should try compiling GCC 2.0,
which led me to discovering the filesystem corruption problems. (Which
are very repeatable, by the way; I can start with a newly mkfs'ed
partition and corrupt it just by untaring GCC 2.0.)
Another problem which was uncovered by trying to compile RCS: In
sys/wait.h, the definition for WIFEXITED(s) is missing a close
parenthesis at the end of the line.
That's all for now.... the patch tools/build.c are below.
Subject: Re: Weekend hacking and other randomness...
Date: Tue, 12 Nov 1991 08:39:35 +0200
From: Linus Benedict Torvalds < firstname.lastname@example.org>
In-Reply-To: Theodore Ts'o's message as of Nov 11, 16:08
To: email@example.com, Linuxfirstname.lastname@example.org
Theodore Ts'o: "Weekend hacking and other randomness..." (Nov 11, 16:08):
> Other comments; maybe I'm doing something stupid, but I can't seem to
> get mtools to want to write a file from Linux land to MS-DOS land.
Urgh. I should never have included the mtools package :-). The problem
is that the root-diskette filled up, and mcopy is just the front-end for
other programs. mwrite, which actually writes to msdos, didn't fit.
The only reason I had even those small mtools programs was so that linux
wouldn't be a closed system in the way the minix demo-disk is. I don't
like the suite very much, but as I didn't think of tar ...
> Also, trying to compile RCS for Linux apparently tickles a compiler bug
> in gcc.
All bugs are probably of my doing :-( I've hacked the machine
description to add the string instructions, and that's probably the
thing that wen't wrong.
Another problem with gcc-1.40 for linux is that it sometimes can give up
while optimizing. I don't know if it me again, but this time I think
it's the soft-float: gcc runs out of registers when optimizing. It
hasn't happened wery often, but it's very confusing when it does happen.
I hope 2.0 has a corrected soft-float, so that there is no need for
hacking on that too.
Re: file-system corruption: You were right. I think I've found the bug,
and will send out corrected "buffer.c" sometime today. Hopefully that
cures the problem. First kernel patch for this version: not too bad. I
haven't found the floppy driver problem yet, though.
And lastly: Everybody out there making changes to the kernel, please
stand by for the next kernel update... It will probably be in early
december, and I'd suggest you mail me any cdiffs for the kernel before
the end of november, so that I can try to incorporate them into the new
version. 0.11 will not be a major upgrade, just some minor fixes + the
ability to handle other screens than VGA. If things work out with
rename, that might be incorporated :-)