OSNews

Programming languages have already become OS more than 20 years ago... Remember the LISP Machines? Or Alan Kay's DynaBook?

I feel the author should have investigated these languages, too (even tho I understand he focused on lightweight languages).

There is also a project, Croquet, that aims to create a 3D Operating Environment using Squeak Smalltalk on top of Windows/Linux/MacOS. You may find more info here: http://www.opencroquet.org/summary.html

... to see how languages are evolving and going to evolve in the near future. Also, like Valkadesh pointed out, the merge between the "OS" and the programming language has been done in the past.

I still doubt, and almost hope, that PL do not entirely become "OS" and neither does "OS"es become only a single available thru a single PL.

Up to now it is possible to contribute applications for almost any OSes using the language of your choice, be it assembly, C, C++, Cobol, Fortran, Pascal or the myriad of others. Maybe there is a way to make the concepts and generally known methods of programming evolve as to provide more isolation while still allowing the programmer to select the language used to develop.

Squeak/Smalltalk anyone? Damn creepy eyes..... I wish I could turn them off :-p

Guido is removing rexec from Python due to security problems. See this message:

http://mail.python.org/pipermail/python-announce-list/2003-January/...

Don't forget FORTH.

Don't bash what you don't know. I am the first to admit that Squeak's user interface is... primitive at best, but Squeak (and Smalltalk) remains one of the most productive environments I've ever used. Do yourself a favor, download Cincom VisualWorks (free for non-commercial), give it a try, and you'll find that the whole VS.Net vs Eclipse/NetBeans thing is just a childish quarrel.

Well, ain't FORTH a language for creating OSes? ;-P

This article provides a good summary of why PLs should act
more like OSes --- thanks!

I think it covers only a small corner of the features a PL
should provide, though. In particular, does Perl, Python, or
Ruby offer a way to terminate a "process'? That crucial feature
is usually missing. (The Java specifiers famously revoked the
ability to kill threads in Java.)

The Lisp machine and Sqeak are about providing a higher-
level interface to application programmers, and I'm a fan of
both. However, I believe they provide only Python-like control
over "processes" through lexical scope --- again, no termination.
The implementors of those systems had the right (IMHO)
bias toward cooperation, but didn't provide convenient isolation
for those cases where it's needed.

The article commendably mentions a priori resource limits as
a problem. I'll point out that the cited LL2 talk doesn't address
that problem, either, because it's an area of ongoing research
in our world. (It's easy to limit resource consumption in an system
biased toward isolation, not so easy in one that is biased toward
cooperation, at least not without reversing the bias. See KaffeOS
for a Java-based example of the former.)

BASIC was The WORD!
OS and Language, hehe.

a language that uses a VM could become an OS if you just make the OS the VM, but how is that diffrent than the old micro computers from the 70's that ran on a basic ROM? same thing as if you ran a machine on a Java VM or a python Interpretor etc.

Some archives of broadcasts from last year's workshop are available here:

http://technetcast.ddj.com/tnc_catalog.html?item_id=1295

<blockquote>Programming languages have already become OS more than 20 years ago... Remember the LISP Machines? Or Alan Kay's DynaBook?

I feel the author should have investigated these languages, too (even tho I understand he focused on lightweight languages).</blockquote>

Thanks, Valkadesh. Perhaps the title should have been "PL Will Become Multiuser OS" or something like that. My focus is on security/isolation.

Hi Matthew, nice to see you here.

<blockquote>The Java specifiers famously revoked the ability to kill threads in Java.</blockquote>

Yes, I've heard. Anyone knows the story behind this?

I'd like to add that it seems Perl6 will replace Safe.pm with something ala VMS (with some emphasis on the ability to do resource limiting).

Note quite an OS, note quite a programming language (FRED), note quite an office suite ... very powerful and easy to
use.

Runs on HP200LX or a P4 but you can't order it on the Web!

Go check it out :
http://www.framework.com

I wouldn't say Squeak's interface is primitive per se, but in a lot of ways it is different that what people are used to. I'd agree that is visually primitive looking- it doesn't have the polished widget and window decoration look like Mac OS X has, for example. However, the GUI system is quite advanced and lets you do a lot of things that you can't do in any other OS that is as available to my knowledge.

But yeah, Squeak is an OS. Most people run it as an OS hosted on top of another OS, but it has the facilities of any OS. Hardware drivers were added as a part of the Squeak OS- drivers written entirely in Smalltalk- imagine the easy of debugging!

Yes, Perl or Python could be created into OSes, but they aren't such as a natural choice as are Lisp, Smalltalk or Forth- designed to be in such a position, where Perl and Python (and many other "little" languages) were designed to be a part of a greater environment, Unix, Windows, etc. and as such may require a lot more work to be made into an OS. And when that work is done, it still may not feel like it was the best thing to do...

i've often thought that (similar to Smalltalk) objects should be fully protected processes. This doesnt make sense for small objects, so some would be inlined as threads and some even smaller ones (Integer, Array, etc...) would be inlined straight to object code.

instead of having objects and protection orthogonal (the current system) combine the two and you've got instant protection from most multi-threading bugs because you've defined one object to be one thread of execution. any cross-object calling would be automatically protected because the compiler could insert locks automatically.

what i meant was, in smalltalk when you want a process, you subclass Process. I dont think it is possible to create a process any other way. which is sort-of similar to what i was proposing...

I think the article is good in the sense it addresses the fact that most applications need OS kernel like functionality and most languages simply don't offer an easy way to implement this; and unlike real OS kernels the enermous manpower to overcome these issues isn't in place.

Some of the examples were simply false. Databases like Oracle most certainly do offer the the features he spoke off in terms of preventing loops and bad code from creating problemss. While safe.pm may not offer sandboxing YACC itself can run in any reasonable VM and thus so can perl; and message passing between Perl's is not hard (they can pass arbitrarily complex datastructes using datadumper for example).

Has everyone forgotten about Oberon? Anyway, I don't agree with much of anything that is said in the "read more..." section.

A programming language is just a language. An operating system manages a computer's resources for programs and users. A programming library provides commonly used functions for programmers. A program performs a specific function for end-users.

And generally, it shouldn't get more complicated than that.

I think, author mistake in requiring protective features.
For a long time Forth used not to protect himself but rather
to use safe code. For opensource OS, provided by Forth, such
"paranoya" is not a good thing. I think there are too many
changes since DOS' time and there is no need to protect self
from virii. All that protection now seems an illness.
Why OS is to protect from safe code running? That only adds
useless code and complexity. Of course, where it is needed
(where unknown buggy code can appear), safe compilation and
execution can be provided by supporting VOCABULARY with safe
versions of words.
"Is it an OS, that is frightened by itself!" (about muKernels).

I think next PLs are good choice for single-PL (SPL) OS:
1) Forth; 2) Smalltalk; 3) LISP; 4) ML class p.l.; 5) other,
see requirements below.
The p.l. used in SPL OS should be highly extensible and easy
to control, it should be interpreted in any case (for being
useful to operator/user as a shell). It should be able to
control hardware. For the first time, that is all.
BASIC is not extensible, thus I don't take it in account anymore.
I have not heard Oberon to be interpreted.
I'd say that any "algoloid" like C, Pascal, PL/1-PL/M is not
a choice for SPL OS.

You don't subclass Process to create one. You instantiate Process.

In Smalltalk, the easiest way to make a process, that is, instaniate Process with some body of code is to send the #newProcess message to a block.

That is, if you want a thread to sit in the background and print "hello" pausing a second in between, you could say:

| aBlock p |
aBlock := [
[true] whileTrue: [
Transcript show: 'hello'; cr.
(Delay forSeconds: 0.5) wait.
]
].

"To run the process as the foreground thread/process"
p := aBlock newProcess.
" ... or ... "
"or run it in the background!"
p := aBlock fork.

"To start, pause or kill the process:"
p resume.
p suspend.
p terminate.

"lower the process's priority."
p priority: 1.

"<blockquote>The Java specifiers famously revoked the ability to kill threads in Java.</blockquote>

Yes, I've heard. Anyone knows the story behind this?"

I think it was because resources like semaphores, open files, sockets, etc. aren't tracked by thread, and so when a thread is killed, those things aren't freed at the same time. This can lead to resource leaks, and deadlocks (as locked semaphores may never be unlocked).

Operating systems spend a lot of time tracking program resources to ensure they are released when the program exits. Programming languages could potentially do that too, but nothing (OS or language) can keep track of things it doesn't know about. Duplicating this in both the OS and language runtime can get a little heavy, and error prone if one doesn't match the other.

Better might be to increast integration between languages and OS (e.g. sometimes Java threads are mapped to OS threads).

<blockquote><blockquote>The Java specifiers famously revoked the ability to kill threads in Java.</blockquote>

Yes, I've heard. Anyone knows the story behind this?
</blockquote>

Killing a thread could leave an object in a locked state with
no way to unlock it. For much code, this is nothing new, because
a program can always grab a lock and loop forever. But killing a
thread could also leave system objects (normally only manipulated
by "trusted" code) locked, and the system had no way to protect
against this problem. I believe that's the reason that thread
termination was removed from Java.

The alternative is to add a kernel--user distintion to the run-time
system, where thread kills only happen while executing
user code. (Kernel code always returns "quickly" to avoid
delaying a kill). Adding a kernel--user distinction is one way that
Godmar Back's KaffeOS improved Java, so that Godmar was able
to support thread kills.

Look at emacs, If thats not a OS I dont know what is.

Like many of you pointed out, several languages are already OS's. And most OS's are written in PL's. Take both Windows and Linux being written in C++ for instance. I believe this entire situation is more of a marriage than an either/or situation. We can't have one without the other.

I think the author should switch to a real programming language like Delphi, C++ or Java instead of focusing on lightweight languages Pearl, Python and Ruby and then he should rewrite the article in about a year. All 3 of these languages have excellent threading models (although Delphi's encapsulates C++) and lend themselves to isolation and compartmentalization very well.

I live for PERL, but I've heard that .NET has a bunch of security stuff in it as far as what code is and is not allowed to do. If you check out http://www.gotdotnet.com/terrarium, you can see an example of this in the form oa game.

Terrarium is this screensaver/game where you write AI code to create plants or animals, and then compile it. Your compiled code is then sent via P2P networking to other machines. Part of this is to demonstrate that if you try to put anything 'dangerous' in your code, it won't be accepted and can't be run. That, and there's code signing, something to do with public key encryption. I wish I could talk about it more intelligently - I guess I have some reading to do.

You end with a remark that Tcl has a sandbox. A strong one, btw. It also has a Virtual File System. The ability to run multiple threads in total isolation, with an event loop for things like file/socket I/O and user interfaces. One could argue that these qualify as coming closer to a full-blown OS than what you describe.

Your bias shows in stopping at Perl, Python, and Ruby.

I haven't seen this mentioned yet, but java is getting exactly the features you describe:
http://jcp.org/en/jsr/detail?id=121

"...All conformant implementations must guarantee at least isolation of Java state (i.e. logically disjoint heap spaces, per application mutable static fields, etc). Additional forms of isolation possible include separation of JNI state and separation of process state...Research by Sun and IBM has demonstrated these additional forms of isolation and sharing..."