Slashdot Mirror

What do you think some of the major concerns/design issues are? I'm talking about nuts and bolts...

Many of the important theoretic issues have been addressed at the nuts-and-bolts level by the Mozart Programming System. Specifically, if you read Distributed Programming in Mozart - A Tutorial Introduction you'll have an idea of the kind of distributed programming power provided by a network of Mozart systems.

The key to Mozart's power is its use of ultra-light-weight threads that can share single-assignment distributed variables within heirarchical computation spaces. What this means is you can have unlimited "processes" that are waiting on all sorts of things all over the network -- and failures are easily confined to the minimum logical spaces.

By "ultra-light-weight threads" I mean a virtual unification of process structure with data structure.

The future of the Internet is in what I call "rational programming" derived from a revival of Bertrand Russell's Relation Arithmetic. Rational programming is a classically applicable branch of relation arithmetic's sub theory of quantum software (as opposed to the hardware-oriented technology of quantum computing). By classically applicable I mean it is applies to conventional computing systems -- not just quantum information systems. Rational programming will subsume what Tim Berners Lee calls the semantic web. The basic problem Tim (and just about everyone back through Bertrand Russell) fails to perceive is that logic is irrational. John McCarthy's signature line says it all about this kind of approach: "He who refuses to do arithmetic is doomed to talk nonsense." More on this a bit later, but first some history, because he who fails to learn from history is doomed to repeat its nonsense:

When I invented the precursor to Postscript (an audacious claim that I can back up -- it started as a replacement for NAPLPS which I proposed while Manager of Interactive Architectures for Viewdata Corp of America back in November of 1981 -- the Xerox PARC guys found my approach of what they called a "tokenized Forth" communication protocol to be an intriguing way to encode text and graphics), I was interested in having a Forth virtual machine migrate into silicon (ala Novix) so it could evolve from mere graphics rendering into a distributed Smalltalk VM environment (ala Squeak) as videotex terminal/personal computer capacities increased. But I was _not_ interested in object-oriented programming as the long-term semantics of distributed programming environments. (I still have some of the hardcopy of the communiques with Xerox PARC and others from this period.)

Rather, relational semantics were what I saw as the ultimate direction for distributed programming. I had a bit of a go at Tony Hoare's "communicating sequential processes" paradigm and its Transputer realization because he was, at least, starting with the hard problem of parallelism rather than making like the drunk looking for his keys under the light post the way everyone else seemed to be doing (and still are, save for Mozart, since threads, etc. are always an afterthought). But, because there were other hard problems like abstraction, transactions and persistence that he ignored, I christened his approach "Occam's Chainsaw Massacre" in my communiques (in honor of his distributed programming language "Occam") and dropped it in favor of relational programming, which has inherent parallelism resulting from both dependency and indeterminacy. (BTW: Dr. Hoare seems to have finally come to his senses about this issue.)

Unfortunately, the only researcher doing hardcore work on relational programming (meaning, getting to the root of relational semantics in a way that Codd had failed to do) at the time was Bruce MacLennan, then, of The Naval Postgraduate School, and he just didn't have the glamour of Alan Kay at places like Xerox PARC to attract the attention of guys like Steve Jobs. Bruce had a bit of a blind-spot, too, when it came to transactions and persistence, which I attempted to remedy by bringing David P. Reed's work on distributed transactions for the ARPAnet to him, but although he wrote a white paper on a predicate calculus (close to a relational) implementation of Reed's thesis (MIT/LCS/TR-205), he didn't really "get it", IMHO. Reed and MacLennan abandoned their work for other pursuits (ironically, Reed was chief scientist at Lotus while Notes was being developed but did not contribute his ideas on distributed synchronization to that development despite the fact that we had a mutual acquaintance from my Plato days by the name of Ray Ozzie -- so, I share some of the blame for this failure) even as Steve Jobs botched the embryonic object oriented world by abandoning Smalltalk and giving us, instead, a lineage consisting of Object Pascal on the Lisa/Mac which begat Objective C on Jobs's NeXT which begat Java at Sun via Naughton and Gosling's experience with NeXT.

This brings us to the present -- a world in which Javascript-based technologies like Tibet promise to not only salvage the object oriented aspect of the Internet from the birth defects of Jobs's spawn, but actually provide an advance over Smalltalk in the same lineage as CLOS and Self. But it is also a world in which there is growing confusion over the proper role of "metadata" in the form of XML -- particularly when it comes to speech acts and distributed inference. I would call Tibet "the next major Internet advance" except for the fact that the basic idea for a Tibet-like system has been around and well understood since the early 1980's. When it is finally released, Tibet (or a system like it) will put the Internet back on track. I call that a "recovery", not an "advance".

We are now poised to move forward with type inference based on full blown inference engines, thereby dispensing with the nonterminating arguments over statically vs dynamically typed languages that allowed Steve Jobs's spawn to get its nose in the tent. If you want to declare a "type" in a declarative language, just make another declaration and let the inference engine figure out what it can do with that information prior to run time. See how easy that was? Well, there is more to it than that, but not that much: Assertions have implications and assertions made prior to run time have implications prior to run time. Live with it and don't repeat the mistakes of the past.

The confusion over semantic webs, and the reason Berners Lee et al will fail, is essentially the same as the confusion that has beleaguered all inferential systems such as logic programming and "artificial intelligence" over the years: logic is irrational and the real world demands rationality -- otherwise nothing makes sense. By "rationality" I mean that reasoning must literally incorporate "ratios" -- or, as John McCarthy would put it, doing arithmetic so things make sense. By making sense, I mean there is a sense in which one interprets the sea of assertions that clearly dominates for a particular purpose. With logic not only are you limited to 0 and 1 as effective quantities; you have no adequate theoretic basis from which to derive more accurate quantities with which to make sense by taking ratios and determining which inferences are dominant.

Fuzzy logic and expert systems incorporating probabilities have typically failed because they are not based in the first principles of probability and statistics. As Gauss, the premiere probability theorist put it, "Mathematics is the study of relations." He didn't say, "Mathematics is the study of multisets." There are good reasons that relational databases, and not set manipulation languages, have come to dominate business applications -- and Gauss was aware of these differences when he began to derive his laws of probability. Subsequent axiomatizations of mathematics based on set theory were similarly misguided and have led to the idea that "fuzzy sets" are the way to introduce rationality into programming. Rather than sets, relations are the foundation, not just of mathematics but of rationality in the same sense that Gauss realized when he derived his theory of probability from the study of relations.

Rationality allows for judgment which is recognized as inherently fallible -- but which allows one to procede without exponentiating all possible paths of inference. Judgment also allows various identities to limit sharing of information to that needed -- thereby creating speech acts and a basis for rational measures of credibility associated with those identities. Since credit-rating is a degeneration of credibility, it should come as no shock that the invention of negative numbers, originating as they did with the Arabic invention of double entry account keeping, has its analog in something that might be called "logical debt" with which negative probabilities are associated.

And now we have come to the "quantum" aspect of rational programming. It is precisely the "credibility debt" aspect of rational programming that corresponds, in mathematical detail, to the various equations of quantum mechanics and their negative probability amplitudes. (Von Neumann's quantum logic failed to properly incorporate logical debt which has led to much confusion.) Logical debt is important to distributed programming for the same reason debt is important to financial networks. Logical debt is a way of handling poor synchronization of information flow in the same way that financial debt is a way of handling poor synchronization of cash flow. As in any rational system, there are both limits to credit and limits to credibilty that influence one's judgments and actions, including speech acts.

The object oriented folks may, in a sense, have the last laugh here because when we divide up inference into identities that engage in speech acts, we are reintroducing the notion of objects that hide information via exchange of speech act messages that can be thought of as "setters" (assertions) and "getters" (queries). However, I believe it is only fair to recognize that the excellent intuitions of Johan Dahl and Kristen Nygaard did need the added insights and rigor of philosophers like J. L. Austin and T. Etter.

Python isn't compellingly better than the Perl/Javascript combo. Perl/Javascript already has such a dominant position that by the time Python's growth rate matters (Javascript's growth rate is much higher), Perl/Javascript will:

Already have displaced Java (not to mention C++) on servers and clients.

Start being displaced by a next generation technology -- thus crowding out any potential play for dominance Python might have had.

For a look at a next generation technology, any of the distributed declarative semantics systems would do (the most commonly referenced ones being RDF-like). However, at present, my money is on something at least as capable as Mozart.

In late 1981, I was given the responsibility to develop an authoring system for the Viewtron videotex network planned for nation-wide deployment by AT&T and Knight-Ridder due to my prior work at the Plato project. At about that time, the cover story for Byte Magazine by Larry Tesler of Xerox PARC was about Smalltalk. Since I had been looking for a decent language upon which to base a network programming environment, Dennis Hall, then technical director of the Viewtron pilot, arranged a trip to Xerox PARC to see a demonstration of their system. We met with the Xerox PARC Smalltalk team in November of 1981.

We were having difficulty with the standards committee controlling the North American Presentation Level Protocol Syntax -- the graphics protocol upon which the Viewtron videotex terminals built by Western Electric were based. Specifically, there wasn't enough programmability. The Western Electric terminal was so limited in capacity that we had to fit the graphics interpreter into a very few number of bytes, and could afford only a few thousand bytes of dynamically downloadable store. I had been enamored with Forth ever since the Byte magazine article about it about a year or so earlier (my first digital purchase was an HP35 so reverse polish didn't bother me perhaps as much as it should have). Even so, I was hunting around for options. Jim Thompson, another senior staff member with the Viewtron project, was also interested in Forth -- enough so that he had subscribed to the Forth newsletter, which he shared with me. Jim was supposed to develop a menu system to run on the central system. I had specifically asked that his menu system never achieve Turing Machine equivalence, because I knew what sort of horrors lay in wait for us if it did. Nevertheless, Jim eventually implemented GOLFBAL "Game Oriented Language for Business And Leisure" -- and it was a Forth derivative. I had rejected Forth as anything but the low level protocol and engine for the telesoftware graphics system and was fairly horrified to discover what he had done. In any case, it was this immersion in Forth we brought with us to our meeting with the Xerox PARC folks.

Now, I swear on a stack of bibles that after I met with the PARC folks and discussed the problems of graphics communications, I had no idea the industry could end up being stuck with Postscript as a type-setting standard. I can say this for a certainty because:

I wanted to see a Novix-style reduction-to-hardware of the Forth virtual machine so that Forth would become the macro assembly language. Then we could use the Forth silicon machine to start running dynamically downloaded Smalltalk -- or some similar high level language -- compiled for the Forth stack machine which would provide much more powerful graphics specifications than Forth itself.

I never imagined the Smalltalk guys would actually depart from Smalltalk itself as a graphical specification language.

By the time the PARC guys spun off Adobe with Postscript and its Forth-like engine, I had become more interested in constraint/relational programming semantics than object oriented semantics because it more naturally fits graphics description, distribution, nondeterminism and parallelism not to mention databases.

It was summer of 1982 when I met with Tesler for the last time -- and he had just left PARC to go work on Lisa. We were sitting in the empty Astrodome, I think it was, next to the convention center where the Commodore 64 was being introduced to the world market as part of the precursor to Comdex. 64K of memory! At any rate, Tesler and I discussed the reason he had abandoned Smalltalk for the Lisa. I had thought that type inference coupled with artful use of assembly language libraries would be sufficient on the Motorola 68000 family, but Tesler was insistent that Object Pascal was necessary for adequate speed. Frankly, I was apalled that Tesler had so easily abandoned Smalltalk with type inference since he had made specific mention of it as an optimization technique in his Byte article. But in a recent email exchange about this history, he told me type inference was never of much interest to him -- that others at Apple were hooked on Object Pascal.

The horrifying thing about all this is that when Steve Jobs took off from Apple to found NeXT, instead of correcting the nonsense with Postscript and going straight for Smalltalk with type inference, he repeated the mistake, only this time with Objective-C. Then, as I understand it, Objective-C was the precursor to Java with its reliance on declaration rather than inference for type checking. This despite the fact that Sun already had the Self programming language in house with type inference and dynamic optimization technologies that realized the potential of Smalltalk at along last. Unfortunately the only technology to make it bigtime from Sun's Self project was the Hotspot JVM.

Although these aren't exactly the same mistakes over and over, we're still struggling to get a decent, widely-used dynamically typed language "for everyone" that includes a pure OO library for graphics. Python isn't easily deployable and although I'm a Perl bigot, even I realize we're unlikely to get Perlscript installed in every browser anytime soon. Anyway I'm partial to prototype languages like Self when it comes to Smalltalk offspring. I do have hopes for TIBET as a way of turning Javascript into a powerful programming system across many platforms -- as outrageous as that sounds. I know Bill Edney and Scott Shattuck were some of the first NeXT hackers, but we can all pray for a swift recovery. This isn't an official announcement or anything -- but Bill and Scott did do a presentation at Hackers so I figure I can mention it in the mode of a "hot rumor".

As I said, I'm more into constraint/relational stuff these days myself, but it sure would be nice if someone brought the power originally in Smalltalk the ubiquity it deserved almost 20 years ago.

Systems in which light-weight threads are first order constructs, such as Mozart illustrate why relational programming will eventually subsume functional or procedural programming:

Functions are special cases of relations.

It's important to build relational semantics (light-weight threads with logic variables or their equivalent) in at the kernel. Otherwise, you end up kludging around, either recreating it at the higher levels or malanalyzing your relational task to fit your functional tool.

Open source studies like this one are increasing awareness of the need for light weight threads.

That's good.

The next step will be for people to recognize that what they are doing with all those threads is essentially relational in nature so they can really address the impedance mismatch between relational database and object oriented programming.