For SG’s earlier review of the Beta 2, scroll down in http://www.freedom-to-tinker.com/?cat=18.

For a treat, see http://lukego.livejournal.com/, Luke Gorrie’s report of his trip around the world with an XO and a unicycle. (Thanks to Vadim Nasardinov for pointing me at this.) Learn about hacking the XO way down at the Forth level!

The deadline for the “Get one, give one” deal has been extended to the
end of the year.

My old friend Brian Silverman, whom I have not seen in years, has co-authored a paper that helps explain the kind of educational philosophy behind the OLPC project, althhough it is not specifically about OLPC. See http://llk.media.mit.edu/papers/IDC-2005.pdf.

[...][...]

I’ve played with it a bit, and I was impressed by its development environment. I don’t think it plays well with SVN, although it would presumably be possible to code something based on its facility for saving your code to a text file (filing out). This didn’t seem to be working at all for me though! I may have an unstable version or something. Squeak does have its own repository system called Monticello that I believe provides similar facilities to other source control systems.

It’s got a graphical debugger that you can edit your program through, which is a nice touch. Another thing that struck me is that writing code in Squeak is tightly bound to the GUI. While I believe that it’s possible to write all your code in Sqeak’s Notepad-esque “Workspace”, a programmer would normally use the system browser to create new objects and add methods to them. It also has a built-in testing framework, a method browser that lets you search by name or by desired effect (!), and some other nifty stuff too, iirc.

The text editor itself isn’t powerful, which is kind of a shame. I would have liked to see simple macros (in the vi sense, not the Lisp sense) at least, as well as search-and-replace. Presumably lacking these isn’t a huge impediment to the Smalltalk programmer, but that makes it largely useless for anything else.

I do think they’ve crafted a development system far superior to many. There’s a slew of things that other IDEs do that it doesn’t, but it’s mostly things that aren’t hugely useful to a Squeak programmer. They’ve really managed to put the “Integrated” in IDE.

I can’t tell anything about the performance of the implementation; I know that L Peter Deutsch invented JIT compilers for Smalltalk originally (in 1984) but I don’t know if this has one. Apparently compared to Smalltalk-80 it’s public domain and written in itself, has blocks as true closures, block-local variables, it’s based on a VM that targets dynamic languages. Ah, it can generate C and claims to be as fast as commercial Smalltalks. It has better facilities for graphics and sound and networking. It’s ported to all major platforms.

The original OOPSLA-1997 paper talks about the implementation, as least as it was at that time. The implementation has a clever scheme to minimize memory overhead per object. Pointers are direct (no object table to indirect through). The GC is incremental and compacting though not generational. Aha, there’s no JIT compiler, and the benchmarks seem to show that it’s more than a factor of 5 slower than ParcPlace Smalltalk. They were focusing on keeping the implementation simple and easy to port. Of course, someone could do a JIT compiler someday, as could the kind of optimizations used in the Self compiler. This paper is from some time ago and so there may have been improvements since.

There are apparently two interesting Web frameworks (at least) called Seaside and Aida. In Common Lisp, there is UnCommon Web, and coming soon, we hope, Lisp on Lines.

It certainly looks like it could be the basis of a LispM successor. The LispM was Lisp all the way down through the operating system, but there are few realistic venues for that kind of thing these days. My best prognostication at this point is that we won’t see anything like this until the pendulum swings back to having interest in innovative operating systems, and someone does a good one that uses an object-granularity capability architecture for security.

The LispM had all of its applications written in Lisp, and they had some extensibility; I don’t know whether you can write new commands for the Squeak tools. I doubt that their editor is anything like as powerful as Zmacs (the LispM Emacs, one of my own major contributions). I’m not sure what its story is on threads and synchronization. Is there a profiler? Does it play well with source control systems like Subversion, given that it’s not oriented toward code source files? (That’s not particularly a LispM thing, but it’s important these days.) There isn’t a built-in object-oriented database system a la Statice (also one of my own contributions, which may be why I thought of it). I don’t see a foreign function call interface (also not a LispM thing of course), sometimes very useful (high-performance crypto libraries, fast access to relational database systems, etc). Some people might be put off by its use of its own widgets rather than those of the underlying operating system (a contentious issue in the Java world). And, of course, no Lisp macros. Only a 32-bit address space, which limits the degree to which you can build your whole world in it. (I found a 2004 blog posting saying that they’re working on a 64-bit version.) No multiple inheritance or multimethods or method combination, none of which is all that essential but when you need them, they’re useful.

Python, as a language, also has a lot in common with Lisp and Smalltalk. A major advantage over both, from a practical point of view, is its “conventional” lexical syntax. That means no macros in the Lisp sense. Smalltalk has an unconventional syntax as well as no macros. Python’s popularity results in more libraries, although for some reason it seems to be weak in the IDE area; I don’t know why.

As far as being LispM-like, I think one of the most important things about Squeak is that the sources are available, so that you can learn about, debug, and even modify what you are given. By having a wide range of applications in Lisp with sources available, the advantage of available sources went a lot farther. We’re seeing this with the Python code in the XO, which is supposed to have a key on the keyboard that immediately takes you to the source code of whatever you’re using/doing. I think this is one of the things that was so attractive about LispM’s. There’s a nice paper about one person’s experience with LispM’s at http://www.gia.ist.utl.pt/~aml/Links/LispOS.txt that stresses this. This is the main reason I drew the parallel between Lisp machines and the XO.

Dan, what do you think of Squeak as a LispM successor?

glad to hear you enjoyed the Google talk! As you imply you’re within walking distance to the Media Lab, it’d be a pleasure to have you visit us at the OLPC offices sometime. One of the _explicit_ reasons I joined OLPC was to try and recreate some of the magic of Lisp machines — it’d be fantastic to talk about this with you. I’m out of town for a couple of weeks, but drop me an e-mail if you’d like to grab lunch when I’m back.

Cheers,
Ivan.

Ex-MIT’s Negroponte is naturally very happy about this project, since, as he said, everybody was very interested about these machines … but nobody was signing checks. He was reticent initially, since he hoped to start with bigger countries.

I don’t think that the kids receiving these computers will care about the differences between Python and Lisp, at least not initially