Linux-Development-Sys Digest #709, Volume #6 Fri, 14 May 99 00:14:56 EDT
Contents:
Re: Journaled Filesystem (Gianni Mariani)
Re: best distribution ("G. Sumner Hayes")
TCP/IP guru help needed (Modemch)
NCR53875, SMP, 2.2.x: comiles, does not boot (Bill Anderson)
Re: Translation of linux to minor languages ("Stefan Monnier "
<[EMAIL PROTECTED]>)
FireWire IEEE 1394b drivers? (Soeren Juelsgaard)
Re: Hostile Takeover of Linux (Alastair)
Re: USB Support (Christopher Browne)
Re: Glibc rant (Modemch)
Re: Translation of linux to minor languages ("Stefan Monnier "
<[EMAIL PROTECTED]>)
Hostile Takeover of Linux (dan)
Distributed processing without batch? (lon)
Re: tulip driver woes (was Re: Reliable (!) nic for 2.2 kernel?) (Ronald Cole)
Re: Hostile Takeover of Linux (Joe Pfeiffer)
Re: IOCTL Call help needed (Scott Lanning)
Re: Glibc rant (Juergen Heinzl)
IOCTL Call help needed (Masher)
SOS ([EMAIL PROTECTED])
----------------------------------------------------------------------------
From: Gianni Mariani <[EMAIL PROTECTED]>
Reply-To: [EMAIL PROTECTED]
Subject: Re: Journaled Filesystem
Date: Tue, 11 May 1999 15:35:00 GMT
This has nothing to do with a jfs.
If you have a glitch and the data wriiten to this disk is corrupted, then a
jfs isn't going to help. You need an fsck !
John Hughes wrote:
> [EMAIL PROTECTED] (bill davidsen) writes:
>
> > You might want to do some reading by someone who isn't pushing jfs.
> > Having used jfs heavily for about five years (AIX) and done some
> > reading, I can assure you that while it is useful for fast recovery from
> > a hard crash, it is *not* as good as a real fsck.
>
> Um, then you've got a crappy jfs.
>
> I don't see these problems with Veritas vxfs, even after running for
> years on crap hardware that crashes if somebody slams the door too hard.
>
> > The most common thing you miss using jfs after a crash is bitmap
> > problems, marking a blockj in use when it's free is just a little lost
> > space, while marking it free when it's not is corruption to come.
>
> Eurgh. Kill your jfs implementor.
>
> --
> John Hughes <[EMAIL PROTECTED]>,
> Atlantic Technologies Inc. Tel: +33-1-4313-3131
> 66 rue du Moulin de la Pointe, Fax: +33-1-4313-3139
> 75013 PARIS.
------------------------------
From: "G. Sumner Hayes" <[EMAIL PROTECTED]>
Subject: Re: best distribution
Date: Thu, 13 May 1999 18:53:33 -0400
octet wrote:
>
> I'm pretty new to Linux and would like to get some opinions from you
> folks. This is because I'm thinking about rolling out many Linux
> workstations to replace Windows workstations.
>
> 1. Which distribution is the oldest?
The oldest distributions (SLS, etc) no longer exist. Of the commonly
used distributions, I believe that Slackware, Red Hat, and Debian
emerged in that order; SUSE, Pacific Hi-Tech, and Caldera are younger,
but I'm not sure on their ordering.
> 2. Which one is the "technically" best distribution right now?
> 3. Which one is the best "over-all" distribution right now?
These are highly subjective questions. Personally, I would recommend
one of Red Hat or Debian for most situations, though if you need
good Japanese support or Netware interoperation you'd want to go
with Pacific Hi-Tech or Caldera. If you have staff who are familiar
with a particular distribution, you may as well go with that one. If
not, then Red Hat is the most widespread, and Debian is every bit as
good.
--Sumner
------------------------------
From: Modemch <[EMAIL PROTECTED]>
Crossposted-To: comp.protocols.tcp-ip
Subject: TCP/IP guru help needed
Date: 13 May 1999 18:32:48 -0400
Hi All.
I'm porting the BSD 4.3 TCP/IP stack to an embedded system, and have ran
into a problem. Under load, when the TCP window becomes 0, it never
recovers from this - subsequent acks from the sender get dropped because
it's sequence number is one segment size less from what is expected. I'm
using Linux, kernel 2.0.30 for development, and wondering whether it could
be a bug in that specific kernel, or in the BSD TCP/IP stack, but can't figure
it out. Here's a fragment of tcpdump output:
y.80 > x.1025: . 25633:25777(144) ack 0 win 31744 (ttl 64, id 48134)
y.80 > x.1025: . 25777:25921(144) ack 0 win 31744 (ttl 64, id 48135)
x.1025 > y.80: . ack 25921 win 288 (ttl 15, id 377)
y.80 > x.1025: . 25921:26065(144) ack 0 win 31744 (ttl 64, id 48182)
y.80 > x.1025: . 26065:26209(144) ack 0 win 31744 (ttl 64, id 48183)
x.1025 > y.80: . ack 26209 win 0 (ttl 15, id 378)
y.80 > x.1025: . ack 0 win 31744 (ttl 64, id 48282)
y.80 > x.1025: . ack 0 win 31744 (ttl 64, id 48396)
y.80 > x.1025: . ack 0 win 31744 (ttl 64, id 48602)
y.80 > x.1025: . ack 0 win 31744 (ttl 64, id 49009)
Those last acks are being dropped, and the connection does not
recover. This only happens when the window actually goes to 0, when it
just shrinks all extra packets are dropped, and everything goes well. I'm
totally at a loss here, I've been staring at this for the past 3 days.
Thanks in advance.
--
Regards,
Modemch
------------------------------
From: Bill Anderson <[EMAIL PROTECTED]>
Subject: NCR53875, SMP, 2.2.x: comiles, does not boot
Date: Thu, 13 May 1999 12:03:11 -0600
Ok, IIRC I posted a similiar post a while ago, but now I have further
info.
I have two of the above mentioned cards, running just fine under any
2.2.x when run in UP mode. When I switch to SMP, it compiles, but fails
to boot. It now gets all the way up to where it should either laod the
noduels (when compiled as modules), or run the card. At this point I get
a root panic; unabel to mount yadda-yadda. This is due to it not
'loading the drivers' in either case. Any ideas? It is sooo annoying to
not be able to use both of the cpus ...
------------------------------
From: "Stefan Monnier <[EMAIL PROTECTED]>"
<[EMAIL PROTECTED]>
Subject: Re: Translation of linux to minor languages
Date: 13 May 1999 19:20:50 -0400
>> Regexp matching can be done on UTF-8 just fine (no conversion
>> needed).
> you *do* need conversion. what does "." mean for a multi-byte char?
The regexp compiler needs a slight adjustment, but the string doesn't
need to be converted.
Stefan
------------------------------
From: Soeren Juelsgaard <[EMAIL PROTECTED]>
Subject: FireWire IEEE 1394b drivers?
Date: Thu, 13 May 1999 20:56:21 +0000
Reply-To: [EMAIL PROTECTED]
I have been looking for a firewire driver, but couldn't find any,
and before I run off and write one myself, I would like to know if
anyone is already working on a firewire driver, and if I can be of
any help?
Regards,
--
Soren Juelsgaard
DSP Engineer
------------------------------
From: [EMAIL PROTECTED] (Alastair)
Subject: Re: Hostile Takeover of Linux
Date: Thu, 13 May 1999 23:50:04 GMT
Reply-To: [EMAIL PROTECTED]
dan <[EMAIL PROTECTED]> wrote:
>
> snip ...
Maybe if you started making sense ...
--
Alastair
work : [EMAIL PROTECTED]
home : [EMAIL PROTECTED]
------------------------------
From: [EMAIL PROTECTED] (Christopher Browne)
Crossposted-To: redhat.general,comp.os.linux.misc
Subject: Re: USB Support
Reply-To: [EMAIL PROTECTED]
Date: Fri, 14 May 1999 00:44:51 GMT
On Thu, 13 May 1999 19:22:06 +0100, Chris <[EMAIL PROTECTED]> wrote:
>Really,
>
>What does minimal mean?
>
>Is it realistically unworkable but a public beta?
Minimal is more like:
"It can access the USB hub, and tell you what's on the hub. Actually
making use of the devices that are found may require drivers that
aren't There Yet..."
The value to this is that it allows you to at least see if the hub on
your motherboard is being detected, which is a decent first step.
The next step is likely to start deploying "device drivers" (which
may, or may not, be of the same sort as traditional Linux "device
drivers") to support doing "useful things" with the stuff that you
plug into the hub.
I would expect mice to be roughly the first devices supported;
keyboards coming a close second; rumor has it that Alan Cox is working
on "USB speaker" support, which might beat either or both.
There will doubtless be interest in a driver for the USB Zip drives;
that's different enough that it may take a bit longer...
A real interesting option would be that of USB-to-Serial convertors,
which apparently allow you to hook up a bunch of RS-232 ports to a USB
port. This is quite attractive as a way of adding RS-232 ports
without having to insert ISA cards, or consuming interrupts.
Unfortunately, RS-232 is weird enough that I'd hazard the guess that
supporting the ports bug-for-bug will make for a horrid code base when
building a driver to support this.
Note that it's not necessarily enough to just have a driver for a
device; it may also be necessary to have some higher level "device
management" code added in. After all, it would be cool to unplug a
failing USB keyboard and plug in a new one. But that means having to
break and re-establish connection between console and keyboard, which
might not work "out of the box."
--
"Who is General Failure and why is he reading my hard disk ?"
-- [EMAIL PROTECTED], Felix von Leitner
[EMAIL PROTECTED] <http://www.hex.net/~cbbrowne/hardware.html>
------------------------------
From: Modemch <[EMAIL PROTECTED]>
Subject: Re: Glibc rant
Date: 13 May 1999 18:50:07 -0400
[EMAIL PROTECTED] (Philipp Thomas) writes:
> On Mon, 10 May 1999 10:25:22 -0400, "Gerry S. Hayes" <[EMAIL PROTECTED]>
> wrote:
>
> >People like StarOffice may not have even known that when they built it
> >on Red Hat 5.0 (or whatever platform) that they were using a beta glibc.
>
> But they have had to be aware that they where calling an internal function
> (clearly visible by the name with leading underscores), which is a definitive
> nono regardless of whether its a development or release version.
Hmm... Now, yesterday, I was trying to compile Mesa 3.1, after installing
glibc-2.1 from an rpm, since gnome apps I downloaded from rawhide wouldn't
work. It would compile fine, but fail at link time with __setfpucw not
being defined. The funny thing is that this was defined in glibc.2.0.so,
and in glibc.2.1.so, but it still wouldn't link. But that's besides the
point. This function actually has a *manpage*, and it starts with
underscores. Now, I've got around the problem by re-writing the piece
without using __setfpucw, and using macros from fp_control.h, but - should
a function be considered internal, if it actually has a manpage describing
how it works? Btw, I still can't figure out why it wouldn't link.
--
Regards,
Modemch
------------------------------
From: "Stefan Monnier <[EMAIL PROTECTED]>"
<[EMAIL PROTECTED]>
Subject: Re: Translation of linux to minor languages
Date: 13 May 1999 19:27:57 -0400
> this is a problem. 95% solutions stacked upon other 95% solutions.
> after a while 0.95 * 0.95 * 0.95 ... becomes an unacceptibly small
> number.
That's actually not how it works in practice. In practice, the remaining 5%
of each one of those 95% solutions can be attained on a case by case basis in
ad-hoc ways, so you do get a solution no matter how many 95%-solutions you
stack up, but that solution requires more and more crufty hacks whose
interactions become harder and harder to understand.
And yet, Unix is loved, Tcl is loved, ...
> something like a sentence or paragraph of commentary. you do not
> always want huge file names. it'd be especially useful for annotating
> binary data files. one could even go so far as to put man pages in
> the meta data. it has a rather tenous linkage to the executable now.
> granted, some man pages cover multiple files and some man pages cover
> non file items. but for a lot of executables, i'd be nice if there
> were some compact synopsis of their usage bundled right there with the
> binary itself.
But this info might be bound to one file, several files or zero files
as you said. So it's not a simple matter of meta-data. It's really
a matter of being able to attach information in complex ways to multiple
objects with bidirectional pointers. That's sometimes called RDBM.
Stefan
------------------------------
From: dan <[EMAIL PROTECTED]>
Subject: Hostile Takeover of Linux
Date: Thu, 13 May 1999 18:36:07 +0000
Too bad guy's it was all for nothing.
Sylvan has tagged us all as idiots if we
don't get their Linux certification.
I feared it would be the demise of the Os
and now it is so. I guess we leared to
hold the rights to certification the next
time we build an Os. Nce work to all who
participated. But now with all the cheeters
entering the Linux field it will be sure
to be just another Microsoft and novell
"reboot it!" server with a bad track record.
Next time the next Linus will have to protect the
Os's trademark against these
Os killers certification processes.
I am sorry to say I have decided to start
work on another OS (open source of course)
with protection from the Sylvan Empire
(dark side).
Since Sylvan now has the say as to who can
be employed using linux and only if you use
Linux they way they say you should. Or in
short they now own it!
I have decided to start planning another os
but hold the rights as to whether or not anyone
can
"CERTIFY" people in the OS.
Making it fair for all to compete illiminating
the
Political B.S. of Sylvan
Which Linux has now fallen victim too.
Best of luck to Linux anyway. It was hard
work and truly is a great os but I can't
say I did'nt see it comming either. It almost
made
a great Os.
If any one has any ideas on how this protection
can
be implemented on an
already open source product. Maybe we should try
to
save it. Have a talk with Linus or something.
------------------------------
From: lon <[EMAIL PROTECTED]>
Subject: Distributed processing without batch?
Date: 14 May 1999 00:30:55 GMT
Perhaps a newbie issue, but I haven't picked up relevant info on FAQs,
etc...
I have setup a 10-node RH5.2 cluster to use as essentially
a single multiprocessor machine. This is a simple farm but not really
Beowulf-type, since PVM/MPI is not an option, given legacy binaries,
dynamically changing academic code and no real need for parallelization.
Our apps are generally fairly short (minimal io, virtually no swapping, < 5
mins dedicated run time each on generic PII 350, SCSI Ultra 2, 5 Gb RAM),
but we run each app > 10,000 times. What is needed is essentially a
non-SMP version of a multiprocessor box. Is it possible to have a user
login to a single server, submit an application, and have that application
transparently distributed to an available processor? We are using condor,
and have checked out DQS (minimally). Condor serves our needs reasonably
well, but the batch file requirements are superfluous, as all clients are
dedicated to the cluster. Moreover, condor gets pretty confused (i.e.,
hangs for hours, RH5.2 on dual 350 PII, 512 Mb RAM, Gateway ALR 7200
server, Celeron 366 clients) when asked to queue > 5000 apps. What is
really needed is a tranparent cluster alternative to a traditional SMP
kernel: submit a job and have it find its way to the least busy processor.
Is there any way to do this with a cluster? By its description, Mosix
(mosix.cs.huji.ac.il) fits the bill perfectly, but they aren't distributing
their code due to requests for funding...
Yes, I've read a bit about Beowulf, IPmasquerading, and am aware of
extreme linux, so please don't send me to the FAQs unless I've missed the
obvious.
-Lon
================== Posted via SearchLinux ==================
http://www.searchlinux.com
------------------------------
From: Ronald Cole <[EMAIL PROTECTED]>
Crossposted-To: comp.os.linux.networking
Subject: Re: tulip driver woes (was Re: Reliable (!) nic for 2.2 kernel?)
Date: 13 May 1999 17:12:52 -0700
"Rinaldi J. Montessi" <[EMAIL PROTECTED]> writes:
> tulip.c version 0.91 is available
> http://cesdis.gsfc.nasa.gov/linux/drivers/tulip.html
Gee, no kidding? Too bad it isn't in the distributed kernel...
--
Forte International, P.O. Box 1412, Ridgecrest, CA 93556-1412
Ronald Cole <[EMAIL PROTECTED]> Phone: (760) 499-9142
President, CEO Fax: (760) 499-9152
My PGP fingerprint: 15 6E C7 91 5F AF 17 C4 24 93 CB 6B EB 38 B5 E5
------------------------------
From: Joe Pfeiffer <[EMAIL PROTECTED]>
Subject: Re: Hostile Takeover of Linux
Date: 13 May 1999 20:21:48 -0600
dan <[EMAIL PROTECTED]> writes:
> Too bad guy's it was all for nothing.
> Sylvan has tagged us all as idiots if we
> don't get their Linux certification.
I know I'm going to be sorry I asked this, but... what in the *hell*
are you talking about?
--
Joseph J. Pfeiffer, Jr., Ph.D. Phone -- (505) 646-1605
Department of Computer Science FAX -- (505) 646-1002
New Mexico State University http://www.cs.nmsu.edu/~pfeiffer
------------------------------
From: [EMAIL PROTECTED] (Scott Lanning)
Subject: Re: IOCTL Call help needed
Date: 14 May 1999 02:58:23 GMT
Masher () wrote:
: I am trying to set the port speed on the serial port on a
: slackware box
:
: code at the moment is
:
: FILE *fp;
:
: fp = fopen("/dev/ttys0","r+");
:
: but what are the parameters for the ioctl call, cannot find
: helpful docs anywhere
Why not use termios? Then you can have that warm, fuzzy,
POSIXly-correct feeling. If you simply *must* use streams,
you can use fileno(fp) to get an fd. Use the terminal control
function tcgetattr to fill-out a termios struct, then you
can use the *convenience* function, cfsetispeed.
--
Scott Lanning: [EMAIL PROTECTED], http://physics.bu.edu/~slanning
------------------------------
From: [EMAIL PROTECTED] (Juergen Heinzl)
Subject: Re: Glibc rant
Date: Fri, 14 May 1999 03:21:57 GMT
In article <[EMAIL PROTECTED]>, Modemch wrote:
>[EMAIL PROTECTED] (Philipp Thomas) writes:
>
>> On Mon, 10 May 1999 10:25:22 -0400, "Gerry S. Hayes" <[EMAIL PROTECTED]>
>> wrote:
>>
>> >People like StarOffice may not have even known that when they built it
>> >on Red Hat 5.0 (or whatever platform) that they were using a beta glibc.
>>
>> But they have had to be aware that they where calling an internal function
>> (clearly visible by the name with leading underscores), which is a definitive
>> nono regardless of whether its a development or release version.
>
>Hmm... Now, yesterday, I was trying to compile Mesa 3.1, after installing
>glibc-2.1 from an rpm, since gnome apps I downloaded from rawhide wouldn't
>work. It would compile fine, but fail at link time with __setfpucw not
>being defined. The funny thing is that this was defined in glibc.2.0.so,
>and in glibc.2.1.so, but it still wouldn't link. But that's besides the
>point. This function actually has a *manpage*, and it starts with
>underscores. Now, I've got around the problem by re-writing the piece
>without using __setfpucw, and using macros from fp_control.h, but - should
>a function be considered internal, if it actually has a manpage describing
>how it works? Btw, I still can't figure out why it wouldn't link.
Mind the manual pages still describe libc5 functions and you had better
use the info pages here or some other reference. In addition see that the
manual pages says "i386 Linux Man Page", say it is a somewhat system
specific function and to rely on it can spell trouble. The new glibc comes
with a much richer set of functions than anything before and to document
all these will take some time and volunteers.
Moving from a libc5 to a libc6 system does not mean to hit some magic
button and voila.
Some other stuff needs work to to make it compile like the latest smail
version which requires just some new define's (like many other systems)
or GNU cpio (which is caused by a pretty old bug and can result in a
complete data loss) and so on.
If you decide to go the "compile it on my own" way you ought to be able
to work out things like that. "Use the source Luke" eh ? Yes, sure 8)
Not meant personally though.
Minor remark regarding compatibility ... the latest Applix patch became
necessary due to a bug that is *now* caught and *only* by the glibc-2.1.x
and even other *nix platforms do not catch it. Say other programmes might
fail too, simply because of the same or other bugs that now show up.
Add to that all the introduction of the new Unix98 style PTY's which need
additional adjustments to some programmes (rxvt, ddd, telnetd) and you
will see that complete backwards compatibility is not really possible
if one does not wants to end up with a big mess to support buggy code,
obsolete features or simply hacks that had better go away.
If you want to avoid all `problems' wait about 6 months or so but in the
meantime be prepared to have to fire up the editor or to install some
vendor specific patches. You can help others by working things out and
I think this is the better way to go compared to wasting time with working
out who is to blame for what if at all.
Cheers,
Juergen
--
\ Real name : J�rgen Heinzl \ no flames /
\ EMail Private : [EMAIL PROTECTED] \ send money instead /
------------------------------
From: (Masher)
Subject: IOCTL Call help needed
Date: Thu, 13 May 1999 22:28:36 GMT
I am trying to set the port speed on the serial port on a slackware
box
code at the moment is
FILE *fp;
fp = fopen("/dev/ttys0","r+");
but what are the parameters for the ioctl call, cannot find helpful
docs anywhere
not in man pages nor in .h file
------------------------------
From: [EMAIL PROTECTED]
Crossposted-To:
comp.os.development.apps,comp.os.linux.misc,comp.os.linux.setup,linux.dev.config
Subject: SOS
Reply-To: [EMAIL PROTECTED]
Date: Fri, 14 May 1999 01:06:18 GMT
Any Linux guru out there pls. help.
When i try to make the newly downloaded gtk+, it give me this error
msg.
libtool:link:'../..glib-1.2.2/gmodule.la' is not a valid libtool
archive.
When i try to locate the gmoudle.la, it is not on my system. What
should i do?
I would appreciate if anyone ever encountered this msg before help me
solve the problem.
Thanks in advanced
Andrew
------------------------------
** FOR YOUR REFERENCE **
The service address, to which questions about the list itself and requests
to be added to or deleted from it should be directed, is:
Internet: [EMAIL PROTECTED]
You can send mail to the entire list (and comp.os.linux.development.system) via:
Internet: [EMAIL PROTECTED]
Linux may be obtained via one of these FTP sites:
ftp.funet.fi pub/Linux
tsx-11.mit.edu pub/linux
sunsite.unc.edu pub/Linux
End of Linux-Development-System Digest
******************************
