Even on the desktop, what's changed?

I would expect that Microsoft will be finding ways to move out from under IIS. I believe Indigo is a path for that, but I am not thoroughly versed in this stuff yet. I am less certain they even see a need out from under SQL Server. I am less certain of that myself. It's getting better, it's competitive, and I expect it makes some money. I don't know much about that either. I'd expect some people somewhere inside that large company are finding alternate paths, though.

But even on the desktop, what's changed?

I have a friend who developed some significant client/server software ten years ago, that had a long fruitful life. He's been considering a rewrite, and lately wondered if much of Longhorn obviated the basis for that rewrite. Was there any value in this software above Longhorn, or would it be trivialized?

We discussed Avalon primarily, but considered most of what was presented at the PDC. And we looked at a picture of what his software does. Clearly Longhorn is not a step backward for his kind of application. But just as clearly it was not a giant leap forward.

A few of the grungiest bits would be easier with Longhorn. The bulk of the application would still require good engineering, and with that would be not significantly more difficult to build, nor less functional to run, on any other major OS platform.

So even on the desktop, what has changed? Not as much as might first appear if you're already building significantly sized applications in a "managed runtime", i.e Java, Smalltalk, or Python.

This one is a bit of a ramble...

We need to find a way to lead, not just follow.

Yeah. I am not convinced that what Linux needs is a WinFS-like file system.

But one of the open source world's strengths is the number of programmers who are free to pursue many ideas. A large corporation with a significant cash cow like Windows or SQL Server will always make decisions that leverage those legacies.

I am convinced the hardware, wireless, and Internet platforms are slowly entering their own new level. And there are already object models that run on the most significant platforms. One is Java, which is as good as it needs to be. The innovation needed for the future is above the basic object model.

The next most significant object models are Python and Smalltalk. I don't think we need one object model. And above a narrow band of componentry, the future is almost certainly not to be based in an object model. A simple message passing model will do fine.

I don't think the competition or the worldwide customer base will capitulate completely to the news out of Redmond. But to be sure the dominant platform is finally taking a few significant steps forward.

A Heap O'Wha?

Reading about Reaps reminds me just a little about this cool thing unfortunately called Stalin...

Stalin also does global static life-time analysis for all allocated data. This allows much temporary allocated storage to be reclaimed without garbage collection.

Stalin does "whole program optimization". One of the results is that complete lifetimes of allocated objects very often are known ahead of time. Rather than force manual allocation, and rather than use a general garbage collector for all objects, Stalin can explicitly free objects at the point where it can prove the object is no longer referenced.

Because this is "whole program optimization" the software become inflexible to updates without re-analyzing the whole program. That's not too bad in many cases, especially considering the "whole program" may just be one or a set of components but not everything.

Stalin can also do interesting representation analyses, "unbox" values like numbers, and generally eliminate run-time type checking and dispatching. Perhaps the GNU Java Compiler can move toward these optimizations, and perhaps researchers will continue to look at ideas beyond the currently popular virtual machines; ideas that balance the need for dynamic programming and updates with transparent optimizations that get the most out of the hardware.