Linux-Development-Sys Digest #574, Volume #6 Sat, 3 Apr 99 05:14:09 EST
Contents:
Re: How about /dev/web? (Joseph H Allen)
Re: Put another way: Linux from the Ground Up (Peter Pointner)
Re: help: cc/gcc/egcs broken (Andrew Robertson)
Re: How about /dev/web? (Alexander Viro)
help: cc/gcc/egcs broken ("Richard")
Re: Proposal: "Linux 2000 Platform" (Preston F. Crow)
Re: Idea: Make a seperate "i686" tree for Redhat Linux 6.0 (Enkidu)
Re: Took one guy 3 days, another 1 day, me 1 hour... (Mario Miyojim)
----------------------------------------------------------------------------
Crossposted-To: gnu.misc.discuss
From: [EMAIL PROTECTED] (Joseph H Allen)
Subject: Re: How about /dev/web?
Date: Sat, 3 Apr 1999 07:32:17 GMT
In article <7e3j9g$[EMAIL PROTECTED]>,
Alexander Viro <[EMAIL PROTECTED]> wrote:
>In article <[EMAIL PROTECTED]>,
>Joseph H Allen <[EMAIL PROTECTED]> wrote:
>[I wrote:]
>>>With SOCK_DGRAM you need packet boundaries *and* connectionless
>>>stuff. E.g. sendto(), even if your socket isn't connected.
>>This is easy. As with the tape device driver, Write() and read() give the
>>packet boundary. Open() gives the destination. Open() in this case does
>>not actually specify a connection, just the address for the write or read.
>>Alternatively, write() takes a packet containing both the address and the
>>data (you just open("/dev/dgram")). The second method is faster, but the
>>first is probably more useful for shell scripts.
>Erm... So you are doing open();write();close(); instead of sendto()? Cool.
>Which address will it be bound to? BTW, since when did you get select() in
>shell?
Read the whole paragraph- I gave two methods and indicated that one would be
fast while the other would be slow. The slow method is only to make it work
for shell scripts. I guess the fast method would work with shell scripts as
well, if you are willing to give the IP address in text format: 1.2.3.4:data
In any case it's possible- but there will be no standard way until somebody
actually tries it.
>>>You also need OOB stuff.
>>The OOB stuff is a real annoyance (incidentally, what uses it?). Here I
>>would stoop to using an ioctl().
>What will you do with passing descriptors via AF_UNIX (SCM_RIGHTS)?
I would do this with an ioctl as well. A more interesting question than the
details of the communication API is how you would pass the file descriptor
around if you assume the whole net driver is separate from the kernel (which
at least should be possible to do in principle).
>>> Try strace -elseek vi foo 2>/tmp/bar. And watch the show. *More*
>>>than sure that the same holds for EMACS (sorry, I don't have it on any
>>>of my boxen; on SunOS boxen in Uni it *definitely* uses lseek()).
>>Does it use it on 'foo' or the temporary swap file? (On SGI no lseeks are
> Both.
>>used at all in the above test). I've looked at the code for emacs, and know
>>that it is not true (it just reads the whole file into memory in one fell
> <boggle><Boggle><BOGGLE> WTF will happen if you have 40Mb file?
>RSS>40Mb??? Well, I knew that RMS had odd tastes, but *that*... Are you
>serious? No arbitrary limits, my ass... Size of swap isn't arbitrary, sure.
Emacs used to be limited to 8MB files because it used to use 9 bits of lisp
tag information out of every pointer. But be that as it may, most editors
which can handle files larger than swap do not edit the file in place, they
instead copy the file into a temporary paging file and edit it in place
there (the only unix editors with public source which handle files larger
than swap are nvi, vim and my editor joe; all work this way). You have to
copy the file one way or the other during an edit session, so you might as
well do it at the beginning and get a lines count in the process (especially
useful for vi since it organizes the file as lines). Also, it's safer to do
it this way in the absence of file locking. This fact may change as editors
start to use the mmap() feature along with some copy-on-write scheme, which
I plan on eventually doing with joe (even with mmap you have to copy the
whole file if you save it, but you can probably get it to the screen faster
and avoid double buffering the data (except in Linux which has seperate page
and disk buffers for no good reason, IMHO)).
Which reminds me of another fix I'd like to see done to Linux: you should be
able to set a limit on the amount of buffer space a process can cause to
become allocated. The idea is that you should be able to 'tar' a large
drive without causing all of your database server pages or buffers to be
paged out (which would cause the users to complain that the database is
slow). I would assume that this kind of working set limit would be useful
more generally as a way to reduce paging.
--
/* [EMAIL PROTECTED] (192.74.137.5) */ /* Joseph H. Allen */
int a[1817];main(z,p,q,r){for(p=80;q+p-80;p-=2*a[p])for(z=9;z--;)q=3&(r=time(0)
+r*57)/7,q=q?q-1?q-2?1-p%79?-1:0:p%79-77?1:0:p<1659?79:0:p>158?-79:0,q?!a[p+q*2
]?a[p+=a[p+=q]=q]=q:0:0;for(;q++-1817;)printf(q%79?"%c":"%c\n"," #"[!a[q-1]]);}
------------------------------
From: Peter Pointner <[EMAIL PROTECTED]>
Subject: Re: Put another way: Linux from the Ground Up
Date: Fri, 2 Apr 1999 17:19:22 GMT
[EMAIL PROTECTED] wrote:
> Hi everyone,
> Perhaps I didn't describe my question the right way. I have Linux running and
> have been using it just fine (RH 5.2)
> I am simply curious to build a Linux system WITHOUT using a distribution. I
> am perfectly fine with whatever it takes (time/compiling/etc). Can anyone
> show me the way? I have a spare system to play around with.
Hello,
you could simply start with an empty disk. Take a rescue floppy and create
a file system on it. Use your RH system and compile a kernel for your spare
system. Copy the kernel to your spare system. Find out what is needed to
boot that with lilo. When the kernel boots, it will complain, so find out
what's needed next (you could have a look in usr/src/linux/init/main.c).
And so on ...
Peter
------------------------------
From: Andrew Robertson <[EMAIL PROTECTED]>
Crossposted-To: alt.os.linux,comp.os.linux.x,comp.os.linux.setup,fsu.linux
Subject: Re: help: cc/gcc/egcs broken
Date: Sat, 03 Apr 1999 09:02:10 +0100
Richard wrote:
>
> I've been trying to compile a new kernel. But, every time I try compile
> anything, I get errors stating that the standard include files (*.h) can't
> be found. I've tried installing libstdc++, libstdc++-devel, egcs, gcc,
> egcs-c++, gcc-c++, etc... but I can't seem to get things to click. What
> packages do I need? (I want the latest ones that will work correctly.)
>
You need at least the glibc-devel.xxxx.rpm package. Others could be
useful
like ncurses-devel, svgalib-devel (maybe), etc
Andy
------------------------------
From: [EMAIL PROTECTED] (Alexander Viro)
Crossposted-To: gnu.misc.discuss
Subject: Re: How about /dev/web?
Date: 3 Apr 1999 03:06:57 -0500
In article <[EMAIL PROTECTED]>,
Joseph H Allen <[EMAIL PROTECTED]> wrote:
>>Erm... So you are doing open();write();close(); instead of sendto()? Cool.
>>Which address will it be bound to? BTW, since when did you get select() in
>>shell?
>
>Read the whole paragraph- I gave two methods and indicated that one would be
>fast while the other would be slow. The slow method is only to make it work
>for shell scripts. I guess the fast method would work with shell scripts as
>well, if you are willing to give the IP address in text format: 1.2.3.4:data
>In any case it's possible- but there will be no standard way until somebody
>actually tries it.
Sorry. What I mean is that the 'slow' variant will not work at all
(in SOCK_DGRAM you need *both* information about the source and target
addresses in/with the package). 'Fast' one will have unpleasant security
implications due to spoofing of source address.
>>What will you do with passing descriptors via AF_UNIX (SCM_RIGHTS)?
>
>I would do this with an ioctl as well. A more interesting question than the
Ahem... SCM_RIGHTS is not too pretty thing, but putting it into
ioctls...
>details of the communication API is how you would pass the file descriptor
>around if you assume the whole net driver is separate from the kernel (which
>at least should be possible to do in principle).
Sure. E.g. if you'll do <barf> streams. Lots'a'bloat and pretty slow.
You will need some primitive for descriptors-passing and pass them to
driver letting it pass the thing to recepient. Tons of extra context switches
for no apparent reason.
>Which reminds me of another fix I'd like to see done to Linux: you should be
>able to set a limit on the amount of buffer space a process can cause to
>become allocated. The idea is that you should be able to 'tar' a large
One problem: buffer space is heavily shared. And there is read-ahead.
And there is metadata stuff (*especially* heavily shared). And there is fork.
>drive without causing all of your database server pages or buffers to be
>paged out (which would cause the users to complain that the database is
>slow). I would assume that this kind of working set limit would be useful
>more generally as a way to reduce paging.
--
"You're one of those condescending Unix computer users!"
"Here's a nickel, kid. Get yourself a better computer" - Dilbert.
------------------------------
From: "Richard" <[EMAIL PROTECTED]>
Crossposted-To: alt.os.linux,comp.os.linux.x,comp.os.linux.setup,fsu.linux
Subject: help: cc/gcc/egcs broken
Date: Fri, 2 Apr 1999 23:32:15 -0500
I've been trying to compile a new kernel. But, every time I try compile
anything, I get errors stating that the standard include files (*.h) can't
be found. I've tried installing libstdc++, libstdc++-devel, egcs, gcc,
egcs-c++, gcc-c++, etc... but I can't seem to get things to click. What
packages do I need? (I want the latest ones that will work correctly.)
I'm running RH5.2 (without the devel packages). I know that some of the
devel packages are required, but I need to know which ones.
Thanks,
Richard
[EMAIL PROTECTED]
------------------------------
From: [EMAIL PROTECTED] (Preston F. Crow)
Crossposted-To: alt.os.linux,comp.os.linux.advocacy,comp.os.linux.misc
Subject: Re: Proposal: "Linux 2000 Platform"
Date: 1 Apr 1999 20:57:15 GMT
This is silly. (Well, it is April 1st.)
This is also corssposted to way too many newsgroups. Please keep it
in .advocacy where this political stuff belongs.
The only issue that should matter between distributions for
application support is which libraries are required. Since most
distributions are moving towards glibc2.1 with runtime support for
glibc2.0.7pre6 and libc5.4, applications are free to use any
widely-used libc.
Configuration files? Applications shouldn't care where configuration
files for other system components belong, because they should be
separate. I'm not about to install some application that needs to
hack my startup scripts.
RPM for package management? Yuck. Just use .tar.gz. No need to add
extra complexity for self-contained packages.
--PC
--
"And he [Christopher Robin] respects Owl, because you can't help respecting
anybody who can spell TUESDAY, even if he doesn't spell it right; but spelling
isn't everything. There are days when spelling Tuesday just doesn't matter."
-- _The House at Pooh Corner_ by A.A. Milne
------------------------------
From: Enkidu <[EMAIL PROTECTED]>
Crossposted-To:
comp.os.linux.misc,linux.redhat.misc,alt.linux,alt.os.linux,comp.os.linux.hardware
Subject: Re: Idea: Make a seperate "i686" tree for Redhat Linux 6.0
Date: Sat, 03 Apr 1999 20:54:55 +1200
Reply-To: [EMAIL PROTECTED]
Johan Kullstam wrote:
>
> Enkidu <[EMAIL PROTECTED]> writes:
>
> > It's a mixed blessing. Count the number of times there are questions
> > on this group from someone who has bought or downloaded Redhat, and
> > doesn't know how to partition a disk.
>
> is this a redhat problem or a generic linux problem? if debian,
> slackware &c were as popular i am sure we'd hear the same questions
> about them.
>
Good point, but having subscribed to these groups for a long time, I
still maintain that the number of *real dumb* questions correlates
pretty well with the rise of Redhat.
Does anyone want to do a count?
Cliff
------------------------------
From: Mario Miyojim <[EMAIL PROTECTED]>
Crossposted-To: comp.os.linux.advocacy
Subject: Re: Took one guy 3 days, another 1 day, me 1 hour...
Date: Fri, 02 Apr 1999 19:58:03 -0700
I have used Linux for 2 years now. Started with RedHat 2.01, the 4.1, then 5.1,
then 5.2. And I used my experience to install for other people, and orient friends
and family remotely. Each time my installation difficulties changed with the
environment and version (available hardware devices, my preferences, the
user's specifications, whether dual-boot or not, etc.).
My last installation of Linux at work was done in less than one hour, and it
was from the Mandrake 5.3 CDROM that I bought from www.cheapbytes.com.
This time is much more impressive, because there was no need to stop to think,
I knew exactly all the particulars of my hardware, and I did not have to run
rpm to install KDE 1.1. Everything was done automatically. It must be ideal
for Linux beginners migrating from the Windows domain. I think Mandrake
and Corel, and other initiatives are responses to long-time criticisms from
people who know nothing but Microsoft.
In the seventies there were discussions about what is better, command lines
or menu-oriented interfaces (curses). After extensive analysis of psychological
aspects of the problem, the conclusion was that there is no such thing as a
'better' interface. An interface that shows all choices (menus and GUIs) is
better for occasional users or beginners. Command lines are easier and
preferred by power users. Anyone can be a beginner one day and a power
user another day. Therefore, the best OS is the one that gives you choices.
That is how I use Linux. For some applications I prefer a GUI, either because
I am not familiar with it , or because it has a graphical nature, or because I am
in the mood today. But for other everyday applications, I simply go ahead and
quickly type the name of the application with proper switches, because it is
more specific and easy. Or I create command files or scripts for some
personalized procedures, which can be easily called by an icon; the advantage
is that the command or icon will doe exactly what I want it to do; if it fails, I
will know exactly why. That is why Linux is poised to dominate; it accommodates
the best of both worlds: the geeky and the occasional user, the game player and
the scientist, the writer and web developer, the learning child and the software
developer. It does not crash, so you can trust it. I am glad I am living this
historical moment, when something so complete is available at low cost to
us who know what is good in life.
These facts guarantee a looong life to Linux and shrt life to proprietary and
incomplete OSs. Remember the sequence? DOS, Windows 3, Windows 9x,
Windows 200x. How many more will the world have to endure?
Mario Miyojim
Charlie Ebert wrote:
> RHS Linux User wrote:
>
> > Rupert Said the following:
> >
> > >This marks the end of my first week with Linux.
> > >
> > >First, let me state: This is not a troll. I am going to criticize
> > >Linux, but not bash it. I have a great deal of respect for it and the
> > >enormous amount of effort so may have put in to developing and advocating
> > >it.
> > >
> > >I am a long time NT user and fan, and before that a big Mac guy -- ever
> > >since they first came out. I am a programmer, database developer, and
> > >web application developer. I am not a "newbie", but I'm not a hard-core
> > >geek either. My experience with Unix thus far has been using it lightly
> > >to serve Perl scripts through an ISP years ago.
> > >
> > >I also believe MS's days are numbered. Their core philosophies are old-
> > >hat and the world is changing. I do believe a publicly owned OS as Linux
> > >is will ultimately be the end of MS's reign -- it's pretty obvious Gates
> > >lacks the ability to really shift his way of thinking to the degree
> > >necessary. I do not believe that Linux will be the answer, because for
> > >it to be would require it appeal to the mainstream of users out there,
> > >not a small segment of geeks and sysadmins.
> > >
> > >It has taken me a full week to even get Linux installed and operating
> > >correctly with my machine's hardware. The network card itself took 3
> > >days and calling a Unix geek friend of mine over who ultimately found a
> > >solution so esoteric that I would have never been able to find it. The
> > >phrase, "God this sucks" was muttered many many times. Even trying to
> > >get the desired color depth, resolution, and refresh rate from my monitor
> > >was scary.
> > >
> > >
> > >-RS
> >
> > I can understand you point of view. It would be really disappointing if
> > I were want to install this really great operating system and have it working
> > in one day so I could play. It took me several days to install my version of
> > Linux too. Hell, it took me a week to put OS/2 on one of my machines also.
> > However the point I want to make is I was pretty sure that it would take time
> > to load this operating system. People who also loaded it up before me
> > warned me that it wouldn't be that easy. Especially since I never put such
> > a complicated piece of code on my computer before. One friend told me it would
> > be a great learning experience that it would be well worth the effort. And
> > he was right. I also bugged him alot too about setting up my machine before
> > do the installation, he is rather a well experience with Unix.
> >
> > If you must, be pissed, get mad, hate Linux. But remember this you probably
> > know alittle more about operating systems and how they work with computers.
> > The MAC os and Windows may make setting up real easy for you and you can play,
> > you really don't learn about what you are doing with your computer. Any idiot
> > can set up a Windows machine, they even do it for you if you buy a pre-installed
> > machine.
> >
> > The experience can be bad or good. It's up to you how you wish to look at it.
> >
> > John
>
> I don't understand folks. You must be installing slackware.
> I remember when I tried slackware two years ago it took me 2 weeks to get everthing
> set up.
> But in the process I learned a whole lot about LINUX. A whole lot.
>
> Today I have a copy of RedHat 5.2 on my machine.
> It took me about 1 hour to install and it got everything right on the first try.
>
> I don't think this is an issue anymore. RedHat has such a superior install it's
> silly.
> And further, everything they put on their release has been tested thoroughly so you
> have
> less chance of running into problems.
>
> It's like $12 over at best buy. And it's certainly a NT killer.
>
> I also read in here something about having to re-boot linux one to install
> software.
> What? I've never had to re-boot linux to install software. Not since I installed
> RedHat have
> I personally had to do that.
>
> Applix required NO re-boot.
> Cad Cam program had no re-boot.
> None of my editors did.
>
> There is no re-boot necessary with Linux on an install?
> What did you guys install which required a re-boot anyway?
>
> Charlie
------------------------------
** 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
******************************
