Volume Number: 13 (1997)
Issue Number: 3
Column Tag: Rhapsody
A Quick Take On Apple's Purchase Of NeXT
By Dave Klingler
What are the pros and cons of Apple's integration of NeXT technologies with the MacOS
A Late 1996 Christmas Present for the Rest of Us
My first reaction to the news of NeXT's sale to Apple was a groan of despair. For those who don't remember, Steve Jobs used to make having an Apple computer a roller coaster ride. Having anything at all to do with NeXT was the same way. Steve doesn't believe in color, floppy disks, games or self-restraint when it comes to offending the entire computer industry.
When I found out that Steve would be strictly an adviser to Amelio and Roizen, I breathed a cautious sigh of relief. I have met Steve Jobs and spoken to him twice. My personal reaction was that he was an individual of very high integrity and strong convictions, and I liked him a great deal. Steve can sometimes be right on when devising "insanely great" technologies. You can get an idea what life is like for me when you consider that I'm addicted to using my monochrome cube. Pain and pleasure all mixed into one.
I think the best thing to come out of the deal is that Apple will get Avi Tevanian, and, presumably, Jean-Marie Hullot. Tevanian did much of the original Mach port to the NeXT platform, and Hullot wrote a fair portion of the NeXT operating system and Interface Builder. I'd gladly work for Apple if it meant a chance to work with either one of these two guys. I've developed a strong respect for the rest of NeXT's software engineers as well.
If Apple slashes the cost of NeXT's products, such as the OS and WebObjects, I will be overjoyed. The products are superior, and their new lease on life would mean that I don't have to switch platforms for a while longer. NeXT's OS is arguably the most immediately relevant part of the deal for Mac developers, so I'm going to tell some of what I know about it.
As most any Mac developer knows by now, Apple spent some time considering a number of options for the development of the next MacOS. These included purchase or integration of BeOS, Solaris or NeXT's OpenStep for Mach.
OpenStep for Mach (formerly called NeXTStep) is in many ways a far more elegant OS than the BeOS. It is based on the Mach microkernel with a BSD 4.4 personality for unix compatibility. The Display Postscript (DPS) server is tightly integrated into the OS just on top of the kernel, and so is Objective C. What makes OpenStep for Mach sort of magical is the sum of its parts: Mach + DPS + Objective C = an extremely elegant development environment. I'll refer to it from this point on as OpenStep, but keep in mind that the development environment has been decoupled from the OS and is available as OpenStep for Windows NT and Solaris as well.
For those of you who might be worried about DPS's speed, it's not really an issue. NeXT and Adobe optimized the DPS server so well that one never sees any sort of redraw. Note that I said "server"; DPS is a client-server system, so you can host a display on a different system than the one that's actually doing the drawing calculations.
The Mach microkernel is a multiprocessor-capable threading OS kernel. It has been ported to almost every processor extant, one of those being the PowerPC. Anyone who does NT development will notice that the NT kernel API is almost identical to Mach's, other than the Graphics Device Interface (NT's GDI was recently tied into the kernel in a controversial move by Microsoft to make NT faster for everyday users, thus changing the NT kernel's status as a microkernel.) MkLinux is based on Mach with a Linux personality similar to the BSD personality that OpenStep uses. If you've gotten the idea that Mach is very flexible, you're right.
Objective C is superior in many ways to C++. It is a true object-oriented language that functions sort of like a cross between C and Smalltalk. You will find that Java is more similar to Objective C than it is to C++. The nice thing about Objective C in relation to OpenStep is that the runtime object server is part of the core OS, so you get a very high percentage of code reuse.
As an example, I probably have in the neighborhood of 100 instances of the Text class currently running on the screen of the NeXT cube I am using to type this article. All of them are using the same piece of code in memory, and only the data for each object differs. Applications that use common object classes such as Text, Window and Button can use hardly any memory, because it is possible for them to use very little new code. OpenStep can be very efficient, and the only development system even remotely comparable is Smalltalk. In short, OpenStep is a developer's dream - keeping NeXT in business despite arguably the worst overall smart-but-dumb management in the computer industry.
Finally, there are some wonderful surprises in store for Macintosh developers. The first is that NeXT has enhanced OpenStep with some pseudo-realtime capabilities for developers wanting to do animation. These capabilities take the form of the DPSTimedEntry, a request to the operating system to call a certain function at a certain interval. Since OpenStep is a preemptive multitasking system, DPSTimedEntries are not truly realtime, but the system usually has no trouble achieving granularity fine enough for seamless animations.
The second surprise is that OpenStep's 3DKit is based on Renderman, the scene description language from Steve Jobs' other company, Pixar. You may recall Renderman's scene description language was used to create the movie Toy Story, and there truly is no substitute. Some of the best modeling software available is built on OpenStep and 3DKit, and many Mac developers will be ecstatic to gain access to Renderman. It's built into OpenStep's development libraries.
One of OpenStep's greatest weaknesses is its unix file system, and the BeOS is far better than OpenStep in this area. From the point of view of a Mac programmer, the unix file system might seem great at first because of its reliability, support for essentially unlimited path- and filename-size, and great support for security. Unfortunately, however, the unix file system has grown pretty archaic in many ways and it is arguably unsuitable for a personal computer. Long ago NeXT promised OpenStep developers that there was a completely revamped object-oriented file system on the way, similar to the BeOS or Newton OS. It never arrived. Hopefully Apple will bring it out.
The BeOS also has it all over OpenStep when it comes to device management. BeOS devices are just as object-oriented as the rest of the system, which means that a BeOS device isn't actually instantiated until it's needed. It can be linked and unlinked on the fly, and I have to say I was completely blown away once by the sight of a Be computer rebooting before its (still-warm) monitor could warm up. The system booted with only the devices it needed, saving the rest until an app used them.
The last place BeOS is really stellar is in its multiprocessor support. There's no real excuse for the lack of it in OpenStep; NeXT just never got around to releasing a threadsafe version of the AppKit (the large library of object classes). So, we have an OS running on a multiprocessor-capable kernel with multithreading applications but no multiprocessing because nobody ever got around to revamping the object libraries. It's a sure bet someone will fix this problem quickly, in fact, NeXT had a version of NeXTStep running on a multiprocessor machine before they got out of the hardware business. Some OpenStep people get sort of steamed when you bring up this topic.
The new version of OpenStep, 4.0, is also no longer POSIX-compliant and that broke a lot of applications. Lots of OpenStep people get pretty steamed about this topic as well.
OpenStep is far less flexible than MacOS when it comes to game or multimedia development. Something like RAVE+Game Sprockets is needed badly.
Last, but not at all least, the OpenStep Workspace Manager GUI needs a revamp as much as the Finder's GUI. In some ways it is more friendly, but in some ways it really needs many of the same changes the Macintosh Finder is getting. Again, not a hard problem to solve.
That summarizes Apple's acquisition of NeXT Software from the point of view of a longtime Mac fanatic and OpenStep developer. I still really can't figure out what at NeXT was worth $400 million, but it's a great PR move and I think it was done right. I'm happy to have Apple supporting my favorite OS, and I can now postpone my port of Objective C to the Macintosh yet another time. Like many OpenStep developers, I got a better Christmas present than I could ever have imagined, and I think everyone who will use a computer in the next few years did also.