OSNews

"...The directory layout would not be based around Unix/Linux conventions. For example, why not store the fonts in '/fonts'? Again, I know that some Linux people find ``either /usr/opt/etc/X11/fonts or /etc/X11/usr/fonts depending on the phase of the moon'' a bit easier to remember than '/fonts'..."

I recommend the author to read a few Plan9 papers, especially this one:
http://plan9.bell-labs.com/sys/doc/names.html

Also the author should look more into OS X. It already has a lot of these features, especially the whole fonts bit. Fonts are stored in /Library/Fonts for all users, or ~/Library/Fonts for yourself. And it has an X server. And it has a decent GUI, but not nearly as efficient as xfce.

Right, Gentoo with it's 100's of USE flags, standard Linux lay-out, 'etc-update', 100's of different applications for the same thing etc... That is not what Mike wanted (not to bash Gentoo as I'm running it on non-x86 hardware...)

What Mike want sounds more like BeOS or Syllable with a different filesystem lay-out and the Mac OS X way of installing applications...

Nice article, I wonder how much work would really be needed to accomplish what was outlined. My OS dreams are somewhat more far-fetched that this.

I like the idea of the 'alias' symlink 'libs:'. However what I would like more is an aggregate symlink.
Sort of like a folder which concatenates lots of folders. So a /fonts folder could be defined such that `ls /fonts` would spew out the contents of /usr/fonts /usr/share/fonts /usr/X11R6/defoma/pango/fonts/...
New files created would automatically be placed in a predefined folder. Also some sort of file-name pecking order would need to be defined to stop collisions, but it would greatly simplify some things .

Maybe you were inspired by GoboLinux, or maybe you don't know about it. What you say about the filesystem layout has many similarities with what GoboLinux is aiming to achieve.

The idea of Gobo's filesystem layout is simplicity (also inspired by MacOSX). They have sensibly named directories, and applications are logically installed in a nice and tidy fashion. But in order to retain compatiblity with the 'legacy' layout, they have to have tonnes of symlinks pointing from the standard directories (/bin,/usr/bin etc.) which makes a mess.

Then to make it worse, they've hacked the kernel to hide those standard directories so you don't see them. In my opinion this is bad and it still doesn't remove the need for a good package manager.

Some interesting ideas, but I think I'd stick with Gentoo for now :-)

Edited 2006-10-12 19:14

Library Computer Access and Retrieval System

Once they have been installed/compiled, Linux applications sometimes leave behind no clue as to where they went!

Look into the --prefix option. If you want to know which files belong to a given program, install it to some temporary location such as ~/tmp_install.

My personal convention is to only ever install system wide programs with the package manager. (apt-get in my case.) If I need to compile and install something manually, I install it within my home directory, and each program within its own directory. For me, this is ~/local/programs

So whenever I run ./configure, I always use ./configure --prefix /home/david/local/programs/NameOfProgram

Once it's installed, I'll setup any necessary symlinks to my ~/local/usr/bin directory which is in my path.

That keeps compiled programs all contained in their own area. And since they are all installed to a single directory, if I want to get rid of the program, I just rm -rf ~/local/programs/NameOfProgram and it's gone. (If I created a symlink in ~/local/usr/bin, I'll delete that too.)

One that works and has functionality, Linux based but with the kernel modules more secluded and protected to keep the thing from hanging up.

Plus an interface that is slick (ie - kde) but very stable and apps that to not crash.

Plus, a speedy boot up time.

Personally, I'd take the concept of Squeak http://www.squeak.org and put it on top of the Linux kernel.

It wouldn't necessarily be based on Smalltalk. Actually, I'd rather have it based off of Slate http://slate.tunes.org. Slate is basically a Smalltalk/Self/Lisp hybrid that is a very interesting language.

I'd also probably skin another syntax like Applescript (or something as natural language as possible) on top of it too.

But the real interesting thing from a user perspective is that it would be "Turtles all the way down". What that means is that as much of the system (at least everything above the kernel) would be written in a highly, introspective language like Smalltalk or Slate or maybe even Lisp.

What that gives you is basically everything on your desktop is a "live object". It can be inspected and manipulated. Little applets are completely trivial to make for a non-programmer. You could have a little visual language too. See squeak again.

It might be hard to imagine, but for those of you that have worked in Smalltalk environments, you know what I mean. Of course it would be modern, GPU-accelerated with lots of eye-candy (unlike Squeak).

1) "It's not the hardware, you understand; the hardware is great; it's really moved on a lot since the 8 bit BBC Micro that I started with at age 7."

Very true. This is never stressed enough. Hardware is improving at a very fast rate, software is lagging behind badly. It is more than anything a kind of "denial". Microsoft has done nothing to promote XP 64bit, and I wonder how much they are going to in order to promote Vista 64.
The FOSS community isn't much better either. How often I have read: "64bit? What is the big deal?". SUSE has been the only exception. Mac OS X is also very progressive from this point of view.

2) "The second way of making the money go as far as possible would be to make the OS itself open source. This means that, if other people like the look of Mike OS, they can contribute to it. I don't care about personally profiting from the OS in monetary terms - I've still got four million Euros, remember. I'm not going to get dragged into an argument about the best licence to use; let's just say that it's a licence that means that anyone can contribute to the OS and also that, we can help ourselves to most of the FOSS freely available on other platforms.

Therefore, the Foundation is going start the MikeOS project and get it to a state where it is usable. This will use up the first million Euro and from that point onwards, it will have to survive on its own."

This point is so obvious that everybody should agree.
I can understand that if your name is Microsoft or Apple you might disagree, but if you are trying to create a tiny hobby OS, trying to keep it closed source shows a total lack of basic common sense.

3)Software installation: well, that is very similar to the OS X way or perhaps the PC-BSD one. It also vaguely reminds me of autopackage. But in any case he has got a very good point.

Mac OS X, but on a range of different hardware, and completely open-source.

The operative word here is "My".
The user wants the focus, simplicity and design of Mac OS, but the openess of Linux. You can't have both.

By being open source, everybody has the ability to input, and it's very difficult to get a comunity to agree on one way of doing things, one GUI, one of each subsystem. Only closed source, or controlled source can do this.

Is this a compendium (a summary with some considerations) or an addendum (a part joined subsequently) to Alternative OS contest?

I asked this, because this Dream OS article deals mainly with the characteristics featured by alternative OSes all grouped toghether to create a new mythical OS...

And by the way, just for the fact I am here...

Can I ask to moderators why the GEOS article won Alternative OS Contest?

Hasn't GEOS article being published online AFTER the deadline of 14 of July?????

A good compromise to all desires of the Author it is MorhpOS OS, except for the fact it is not Open Source...

...But, maybe in the future it could being released OPen Source as it has just happened with its GUI called Ambient.

Not only but MorphOS memory protection capabilities should be improved by finishing developing sandbox of QBOX environment directly related to its microkernel Quark, and not being only stucking with sandbox with Amiga API called ABOX.

So the author could get finally its dream OS ecoming real stuff.

From another point of view he had also considered AROS into its discussions. And sure AROS seems to me close to his desires, bus as being AmigaOS 3.1 Like, then AROS has no memory protection.

Well. But AROS is Open Source OS.

Any person could co-operate the project, join with other programmers in a distributed worldwide effort, and helping adding new features such as Memory Protection to AROS little, joung and promising OS.

So, when AROS will begin to deal with Protected Memory and other features, that actually are still missing, then Mr. Reed could at least make true his dream OS.

Please programmers. Help improving AROS:

http://www.aros.org

Does anybody still use programs on their OS?

I would very much love to see a "perfect" OS.
But I still prefer great programs running on it.

In practice, it's better to just 'fall in love' with your OS, in spite of all its shortcomings. That's the charm of it.

My dream OS would be a lot like Mike's, although I've often wondered if there isn't a way to get the best of both worlds situation with the RiscOS model and the Linux model, using NTFS's file system.

First, a hard link is a file with two names. It's the same exact data on the disk, with two different names/directories.

NTFS, if I recall correctly, has a feature that amounts to semi-hard links. It stores identical files as a single hard-linked file, but unlike a real hard link, it will create a new file if you change one of them. The ones that are unchanged continue to be the same data, the one you changed is now its own data in a different location on the disk.

Could you use such a thing to simultaneously have the ease of RiscOS drag'n'drop installation, but avoid the trouble of duplicated libraries taking up space?

The system I envision would store identical libraries (say, GTK+2.8.9) once and make fake hard links in each of the folders the file's supposedly in... Now, suppose you upgrade The Gimp to a new version that requires GTK+2.8.10. Now the files in The Gimp's folder are NOT hard-linked; they're GTK+2.8.10. All the others point remain pointing to 2.8.9 and remain working just as they used to.


The next difficulty in a RiscOS system is that a bug-fix to a library would have to be applied to EACH program that uses that library. (correct me if I'm wrong)
Suppose you make a special case where each library (or other shared file) on the system is ACTUALLY hard-linked to... uh, /usr. To upgrade all applications from 2.8.9 to 2.8.9a, you would apply the bugfix to that file. Yeah, I suppose you'd end up with the same problems if 2.8.9a broke application X.

Oh, and since one feature of hard (and "hard") links is that the data isn't removed from the disk until all the directory entries to it are removed, so you can still remove applications by deleting their folder.


All this work would probably put a serious strain on the filesystem.
Actually, all this could be done with symlinks (but that would involve adding more physical files, and keeping track of how many symlinks point to each file)



My only other 'big wish' is that the OS could have a bunch of, say, audio-playing interfaces, so that every player that connects to that audio interface could use all the plugins installed. Let's say you have an ancient music playing program that calls the audio interface. Add the Ogg Vorbis playing plugin to your system and... Voila, your music player can now play Ogg Vorbis because your system knows how to deal with it now. Yes, this would make audio players more uniform in terms of their playback... but there are other ways to differentiate yourself.

I know several OSes (such as BeOS) have done things like this... Windows does it with video codecs. All I have to do is install an XviD codec and WMP, Winamp and the rest now understand XviD.

Edited 2006-10-12 22:10

This description looks like PC-BSD, don't you think so?

all i can ever think when people whine on about how appdirs in linux would cure cancer or something i just cant help but think about ROX, sitting there with its full desktop of appdirs and its drag and drop saving and loading classes.
and of course gnustep.

First we get publicity in article and now we get publicity in front of the web page.

I am obliged to close a publicity before I can read something on this site each time I open the website or click on a link to see comment/read an article.

This site has too much publicity (and don't tell me that you gain no money with it), it is full of crap now. Thanks, I will save my time and not come back again. I hope this site will disappear from the net.

Edited 2006-10-13 05:14

I didn't find ur ideas worth; you have stressed less on the ideas but talked more about here and there.
these ideas are no-where near to throw away windows; not even any good linux distro these days;
what it will take to do that is completly outstanding OS, which no one has done so far.
but wait for some more time; i'll be putting it on the stage.

it's 2006 nex, so why is your computer really big hot and takes an age to actually do anything, you know, things like switch on and er switch off, what do you actually use your computer for? if it's so fikkin great why does it take so much time and wasted energy to do it? so ask yourself, if it's 2006 why is my computer a piece of spit... this rant goes on a bit more but you get the jist..

My dream OS would actually be a virtualized Mac OS 9 environment with each application running in its own sandbox, or Mac OS 9 environment and servers so that Apps can communicate with each other and the underlying system. Hopefully with parts re-written in i86 architecture with non-portable code running in seperate powerpc sandbox.

Mike, I hate to say it, but MikeOS isn't really very ambitious.

You're basically describing a custom Linux distro, and nothing more. As folks here have pointed out just about everything you desire exists in one form or another. It seems to me that your entire OS project would probably be concerned with picking the components and then making the required edits to sort out the directory structure. I'd expect this could be accomplished for considerably less than 1m Euros, and probably within just 6 months rather than a year.

What you're describing also seems to me to be fairly close to Haiku - excepting of course being based on the Linux kernel and running X-Windows.

Basically the guy wants an Amiga like OS with a few enhancements from the Unix world! what a shame! some people have no imagination at all.

The following text descibes my own ideal operating system. But before that, I would like to stretch the fact that hardware is far from ideal as the author says. The reason is that each device should be independent of O/S, all devices should have an object-oriented interface and the driver code should be in bytecode format loaded into their flashable roms, thus allowing very easy integration with the host O/S. Instead of that, what we get is a list of I/O ports or memory locations and a specification for some types of devices that is usually not implemented 100% and is fixed in the rom.

Let's go to the operating system...I want an operating system where:

1) there are no applications/drivers/filesystem whatsoever.

2) all data and code is stored in a relational database.

3) code is stored as functions.

4) data is automatically persisted and indepentent of code.

5) code is stored as bytecode; the kernel translates the bytecode to native code and caches the result, handling security as well.

6) programs are composed by the user with the help of a GUI, by chaining functions together. The GUI is 'function centric', i.e. it allows the user to process data using one or more functions. Let's name these programs 'tasks'.

7) tasks are full screen, with a small bar in the end of the screen which contains a view of open tasks. Clicking the other tasks brings them on the screen.

8) the system knows the data types of data, so it is possible to do queries across the system.

9) the system is distributed. Computations can run in any available node.

10) the system is reactive; state change can be intercepted and handled accordingly.

All the above is coordinated by the kernel (the only native code in the system).

Gold is a new operating system from scratch designed for desktop use and modern computing, you can see it here http://icculus.org/gold/

This operating system has really nice things, for example, see this about his GUI, http://www.jbit.net/~jbit/Golem-Stuff.txt

This is probably what you would want, it simply rocks!

I hope that the Linux developers are reading this. This guy's points are quite obvious if you've played around a little with Linux. I don't understand why is it so hard for the developers to figure them out.

You can't specify a meaningful assignment for 'libs:' if your library files are scattered over 1