In the mad rush around the fact that Apple is, in fact, switching to Intel CPUs and despite my having to eat a helping of crow, the world does keep moving. And WebCore/WebKit is part of that world. What is WebCore? It’s an HTML rendering library based on KHTML. It’s open source and forms the basis of browsers like OmniWeb and Safari. In short, it’s the best-looking web engine out there and part of the family that’s the biggest [open-source] challenger to Mozilla’s Gecko engine. WebKit is a more developer-friendly framework that makes it easy to embed a web-browser view into your application. JavaScriptCore is, well, a JavaScript engine.
A few weeks back, there was a bit of commotion caused in the geek community (that is to say, on Slashdot) when one of the KHTML developers complained about people’s misinterpretations about how closely Apple works with them. Since KHTML is GPL software Apple has no obligation to return their changes to the KHTML team, yet many user of KHTML seemed to assume that was happening and blamed the KHTML developers for the fact that KHTML didn’t match WebCore feature for feature.
The reason KHTML lagged behind is because WebCore was developed Apple-style. Revisions to the code were tracked in Apple’s internal Source Control System (which the public doesn’t have access to), bugs were tracked in Radar, Apple’s internal bug-tracking system (which the public doesn’t have access to). All the public had access to was the opensource.apple.com server on which the source for the latest version of WebCore was stored. Obviously, this made it difficult for the KHTML team to know what changes had been made for what reasons and when those changes had happened.
That was then. Today, David Hyatt, leader of the WebCore/WebKit/JavaScriptCore/Safari team at Apple, announced that WebCore, WebKit and JavaScriptCore are moving from opensource.apple.com to a new, OpenDarwin-hosted server at webkit.opendarwin.org. The source is now hosted in a public CVS repository and the bugs are now tracked in a public database. WebKit has changed from an Apple project that gets released as open source, to an open source project that is managed by Apple employees. The difference is slight, but significant.
What does this mean to users? Not much, Safari will still be released as it was before, and new versions of WebKit, et al., will still be wrapped in those releases. What does this mean for geeky users? A lot; anytime they want, they can build a new version of WebKit, et al., for there system. Right now, the currently shipping version fails the Acid2 rendering test, the version in CVS passes. Getting that version before the rest of the kids on the block would be kind of nifty.
Posted on 7 June 2005 in Uncategorized
Comments Off
Some time ago, my Newton MessagePad 130 stopped. It stopped responding to input; it stopped providing output; it stopped turning on. It, simply, stopped. This left me rather saddened, as I rather like the Newton OS and the only feasible replacement for the 130 would be a MessagePad 2000 or 2100. And, well, those go for 200 dollars or more, these days.
There’s really not much you can do to bring a Newton back from the dead. The hardware was never very well documented by Apple — in fact, the Service Manual is basically comprised of “Step 1: make sure it’s really not working. Step 2: send it to Apple.” So, every once in a while I would come across the 130, pull it out, plug it in and hope. And, I would be saddened by the complete lack of response and put it away again.
That is, until today. For, when I plugged it in this time a faint buzzinng was heard and a second later the screen came alive with black pixels. Black picxels arranged in such a pattern as to tell me the machine was booting up and would be ready to use in mere moments. This made (and, indeed, makes) me very happy.
Now, the 130 has it’s limitations. The 2×00 series can use a 10 Base-T Ethernet PCMCIA card, or a 802.11b Wireless card, and even a BlueTooth card. The 130 can handle none of these things as it is forever limited to Newton OS 2.0. (It’s version 2.1 that supports TCP/IP and supports the fun drivers…)
Thus, my connection options are limited to serial and I shall have to accept this and buy a USB Serial adapter for my PowerBook. This is useful because, even though Apple killed the Newton over five years ago, there is a thriving community which provides the neccessary software to sync the Newton with Mac OS X, and provides all the nifty drivers for technology that has come into vogue since the Newton was officially killed.
As a side note, here are the (brief) specs of my beloved 130 compared to those of my beloved iPod:
| MessagePad | iPod |
|---|
| RAM | 1.17 MB | 32 MB |
|---|
| Storage | 1.29 MB NV-RAM | 10 GB hard drive |
|---|
| Screen | 5″ diagonal, 240×320 Black and White LCD | 2″ diagonal, 160×128 4 shade Greyscale LCD |
|---|
| CPU | ARM 610 RISC processor at 20 MHz | ARM7 RISC processor at 90 MHz |
|---|
| I/O | Pen w/ Natural Handwriting Recognition, External Keyboard, Serial Port, IR Port | Scroll Wheel, Five Buttons, FireWire, USB 2.0, iPod Remote
|
|---|
I’m planning on buying a PCMCIA CompactFlash adapter, and adding at least 32MB of Storage to the 130 that way. (Yes, that will bring the total amount of menory in the MessagePad up to the amount of RAM in the iPod.) It’s amazing what 10 years of progress can do.
Posted on 5 June 2005 in Uncategorized
1 Comment »
(This is quite definitely the longest blog entry I have every written, and certainly the longest treatise I have written in some time. It has also undergone very little (read: no) revision, as I started writing it Friday night and want to get it up at least one day ahead of WWDC 2005. I apologize in advance for the rambling, stilted, awkward style. I shall endeavour to create a summarized version for the front page, but for now the entire corpus shall have to dominate.)
[Two addenda added since the initial posting. One covers the Inquirer picking up the story, and the other covers a very good point brought up by John Gruber which could well explain everything while minimizing the amount of crow that has to consumed.]
C|Net’s made it ‘official’ (despite any of the parties involved commenting), so I guess I’ll join the maddening crowd and nail my trousers to the mast on this issue. Will Apple announce on Monday that they’re switching to IA-32, IA-64 or AMD64 processors? I think not. In order to explain why I don’t think that’s going to happen, I have to try and explain why it would.
The Case
When Steve Jobs unveiled the 2 GHz PowerMac G5s, he spent a great deal of time describing the amazing features of the PowerPC 970 processors. Not the least among those features was that they would be running at 3 GHz by the end of the year. That was 23 June, 2003; on that same day Intel brought out the 3.2 GHz Pentium 4. Two years later, Apple has just released new G5s running at… 2.7Ghz. The Pentium line, meanwhile has grown to include Hyperthreading, Dual Cores and boast speeds of up to 3.8Ghz. The PowerMac’s main advantage: dual-processor systems is fading away with the introduction of the Pentium D and its dual cores. IBM has failed to deliver the performance gains Apple expected, and doesn’t appear to be about to.
NeXTstep (and, later, OPENSTEP) was an operating system ported to a number of platforms. It also formed the basis of Mac OS X (the Cocoa classes are all prefixed with the characters ‘NS’ for ‘NextStep’). Initially developed on the Motorola 68k processors that powered the early Macintoshes, NeXTstep was eventually ported to the HP PA-RISC, the Sun SPARC and, most importantly, the x86 family. During the transition from NeXTstep/OPENSTEP to Mac OS X, the OS was re-christened with the code name ‘Rhapsody‘ and ported to the PowerPC. However, the x86 port was also maintained through both of the Developer Preview releases. The x86 port disappeared with the commercial release of Mac OS X Server and Mac OS X. Well, partially disappeared, as the open source Darwin OS (which forms the base of Mac OS X) has an actively maintained x86 port. It is widely believed that Apple has secretly maintained this port and that there is an x86 version of OS X hidden away in the bowels of 1 Infinite Loop. This belief is largely sustained through long-standing rumour.
The main objection to a change of CPU is the fact that none of the current software will run on it. Luckily, Apple has been through all this before. In the early ’90s, Apple switched from the Motorola 68k series to the PowerPC. That is widely regarded as the most graceful and seamless architecture transition in the history of computing. The seamlessness grew out of the fact that Apple embedded a 68k emulator deep within the PPC version of the Mac OS. This emulator allowed almost the entire body of existent Mac software to run on the PowerPC based machines despite being written for the 68k. Further, because the PowerPC was so much more powerful than the 68k, the software was able to run at speeds almost equal to what it achieved on an actual 68k!
It is worth noting that, before settling on the PowerPC as the 68k’s replacement, Apple considered a number of options including the x86. In fact, there was a secret project at Apple to port Macintosh System Software 7 (System 7 to its friends). The project was called ‘Star Trek’, as it was boldly taking the Macintosh where no Macintosh had gone before. However, the Star Trek project has little bearing on the current discussion, as the classic Mac OS was supplanted by the NeXTstep-based Mac OS X some time ago.
Transitioning from the PowerPC to the x86 would, thus, be simpler as there is an established pattern to be followed. The main requirement is that the existing PowerPC software can be run on the x86. Luckily, there is already an open source solution to this problem: PearPC. PearPC is a PowerPC emulator written for the x86 platform. It is quite capable, and allows for Mac OS X to be run on a Windows or Linux x86 PC. Integrating it into the Darwin kernel would be a bit of an engineering feat but quite doable. Especially if Steve Jobs decrees it essential to the existence of the company.
The second most important objection is that porting Mac OS X to the x86 platform would deprive Apple of its hardware revenue. After all, given the choice between the cost of an Apple brand system, and an off-the-shelf PC, consumers would flock to the less-expensive option. This fact was proven by the Mac OS licensing experiment in the mid-ninties when lost hardware revenue seriously damaged Apple’s financial standing. Thus, a simple port to the commodity x86 PC platform is out of the question.
Luckily, it isn’t necessary. Apple can design an x86-based machine that doesn’t use the generic PC BIOS. In fact, it would be quite logical. The current crop of Macs utilize the Open Firmware system. It is quite likely and x86-based mac would do likewise. Thus, while alternative operating system such as [Linux][linux] could be ported to the platform (as they have to the 68k- and PowerPC-based Macs), the version of Mac OS X that would run on that platform would be incompatible with the generic PC platform.
The last major sticking point is AltiVec (or, as Apple calls it, Velocity Engine), the PowerPC vector processing unit. There is equivalent architecture for the x86 in the MMX, SSE, SSE2 and SSE3 instruction sets. While AltiVec is generally given the advantage when compared to the x86 facilities, it cannot be denied that almost all the capabilities of AltiVec are present on the x86 side. Most vector code is written by hand, and thus would need to be re-written for the x86 facilities. This is in contrast to the non-vector code as almost all of that is written in a high level language such as Objective-C, C++, C, Java or Pascal and could be made to run native on a new platform merely by recompiling. This rewrite is not as onerous as it might be, by virtue of the fact that GCC4 has support for auto-vectorization. This means that the vector code could be rewritten in a high-level language and thus made fairly independent of the actual processor it will run on. In fact, the way programs are packaged on in Mac OS X means that they can contain compiled code for multiple architectures, that is, a single program could run natively on both PowerPC and x86, providing optimal performance for all Mac owners during the transition period when both architectures are in common use.
On a more technical note, the choice between IA-32 and AMD64 is a bit of a toss-up. Current chips from Intel and AMD support both architectures. However, the PowerPC 970 is a 64 bit processor, as is the AMD64 architecture. Thus, it it most likely Apple would go for AMD64 over IA-32. As for IA-64, it is fairly easy to disregard. Itanium has not been a massive success, the installed base is small — and getting smaller over time — and Intel has had problems scaling the clock speed. Basically, choosing Itanium would put Apple in the same position they were in with Motorola and the G4: depending on an unpopular processor which is improving at a very slow rate. Of the IA-32, IA-64 and AMD64, the IA-64 is the worst desirable choice.
Moving to the x86 would be a perfectly achievable move for Apple. The actual hardware requirements are met by the platform, and the transition could be made almost seamless from a user’s perspective as all their applications would continue to work. And the inevitable performance hit would be short-lived, ending as soon as the developer shipped a version compiled for x86. In addition, it would open the door for powerful Pentium or Opteron workstations as well as fast, cool (temperature-wise, that is) laptops with long battery life utilizing Intel’s Centrino platform. Which nicely demonstrates the final advantage: Apple would be able to take advantage of the incredibly amount of mind share Intel has gained through the ‘Intel Inside’ campaign.
The Rebuttal
The transition from 68k to PowerPC was not actually as seamless as it is remembered. One of the main problems with Mac OS 9 was the fact that almost a decade after the transition started, there were still swaths of 68k code in the operating system itself. The emulation layer was thorough enough to make almost all 68k software run on the PowerPC, unfortunately that meant there was little pressure to transition the more convoluted portions of the operating system to native code. Unfortunately this meant that the full potential of the PowerPC was unrealized under the Mac OS until the release of Mac OS X. While this condition would be less of a problem with Mac OS X due to its more porting-friendly design, it certainly cannot be dismissed entirely.
Requiring every developer to recompile their applications for a new platform is not a major request, but is still an imposition. That sort of requirement could rub the developers the wrong way and lead to some of them abandoning the Mac platform. That seems petty, but programmers are a fickle lot, and the Mac software world has a much higher population of ‘one man shops’ than in the Windows world.
Apple’s PowerPC hardware sales will plummet the moment a transition of this nature is announced. It is a very common occurrence for sales of a given model to slow when news of an upcoming revision hits the rumour grapevine. If Apple says that the entire line is going to be revamped and the current architecture dumped, why would anyone buy a PowerPC Mac? For exiting Mac users, the answer is that they wouldn’t, and for those thinking about switching from Windows, why would they buy a computer running on a chip that is ‘obviously’ inferior to the one they have. The only way Apple could make this transition without a serious drop in sales and the resulting hit to their coffers would be to launch at least one model of x86-based Mac at the same time they announce the switch, and have the rest of the line-up converted fairly quickly.
The C|Net article suggests that Apple would phase in the x86 chips slowly, starting with the low end (the Mac mini to be precise) and working their way up the performance ladder. This would, in my opinion, be quite the worst possible way of doing it. No matter how much testing has gone into an x86 version of Mac OS X and the corresponding hardware, there would be flaws and bugs. And, no matter how good the emulation layer, existing software would have to, by definition, run slower on the x86 machine. Thus, starting with the low end would mean Apple was thrusting a slow, buggy computer experience at the users it stands to lose the most with. For, the mini is targeted at ’switchers’, people coming to the Mac from Windows. If they are faced with an inferior experience, they are quite likely to write off their Mac experiment and Apple will have forever lost a customer. It is much more likely that we would see the introduction of an x86-based PowerMac, probably sold alongside the PowerPC-base PowerMac G5. This would mean they were starting at the high end and working their way down. This would make more sense, since it would insulate the most volatile portion of the user base from the change for as long as possible and put the tools for transitioning their applications to x86 in the hands of the developers who need to do it. (The majority of Mac developers use either a PowerBook, a PowerMac or both.) That said, the insulating effect would be minimal as the sales slump for PowerPC-based Macs would require the entire Mac line-up to be brought over as quickly as possible.
The emulation layer is a very dangerous land mine. The 68k emulation layer worked because the PowerPC could emulate a 68k processor at nearly 100% performance. The x86, unfortunately can only emulate a PowerPC orders of magnitude more slowly than a real PowerPC chip. That means that any program not recompiled for the x86 would run very sluggishly. They would, as Mac OS X on PearPC is, be usable, but the speed difference would be quite pronounced and damage the user experience. Damaging the user experience is the sort of behaviour that loses customers, and that is something that a computer maker with definite minority share of the marketplace really can’t afford to do.
The article also states quite clearly that Apple is switching to Intel processors. This seems odd, partially because the old rumours clearly name AMD’s Athlon processors as the platform Apple is keeping the ’secret x86 port’ running on, and because AMD’s Opterons use the same bus as the PowerPC G5. That bus is HyperTransport, and Apple is a member of the HyperTransport consortium. To use Intel processors would be to discard the time, effort and money Apple has pitched in to the development and marketing of HyperTransport. Thus, I think it would be much more likely for Apple to use AMD processors in an x86 Mac, yet the article names Intel as the company Apple is talking to.
“But,” you may well ask, “if Apple isn’t really considering transitioning why all the fuss all of a sudden?” And, that is a very good question. The answer is two-fold. Firstly, we are in the final few days before a Steve Jobs keynote address (the WWDC 2005 keynote on Monday morning) and, secondly, there’s a bit of a shake-up in the PowerPC world at present. Steve Jobs’s keynote speeches are the traditional time for Apple to announce new products. In fact every new product Apple has brought out since Jobs became CEO has been announced, by him, in a keynote speech. And the weeks before a keynote are always filled with a whirlwind of speculation and rumour about what he’s going to bring out. Yet, there are almost no rumours about what he’s going to announce this time. No expected speed-bumps for any models, no new models, no new, amazing software. Thus, the ground is ripe for any reasonable coherent rumour to take root and flourish. It seems the long-standing ‘Apple moves to Intel’ rumour has done just that. Apple may also be trying to make a point to IBM. The (tech) news has been filled by IBM and PowerPC recently. However, none of this attention has fallen on Apple. Instead, the press has been concentrating on Microsoft, Sony, Nintendo and IBM as all three next-generation games consoles are going to be based on one variation or another of an IBM PowerPC chip. Don’t forget that Apple is still waiting for the 3 GHz G5s that were supposed to be here a year ago. The fact that IBM is devoting a large amount of time and energy to the console makers has be a little irksome. So, why sow a few seeds about a switch away from the PowerPC? It’s sure to get IBM’s attention, after all, it’s worked before. When Apple was in talks with IBM to use the G5, they were also in talks with, you guessed it, Intel to use the, say it with me, x86 in the new line of PowerMacs. The competition from Intel spurred IBM into giving Apple what they wanted. In fact, the only reason we know Apple was in talks with Intel is that a few leaked internal IBM documents from the negotiations mention the fact. Dropping the PowerPC is a very big bargaining chip for Apple, and with IBM seemingly distracted by the console makers, one they may very well be playing.
That very same instability in the PowerPC market could actually turn out to be in Apple’s favour. The Cell processor could find a home in future Macs as a very powerful media co-processor. Being able to render complex effects on-the-fly and in real time would be a tremendous advantage in applications like Final Cut Pro and Soundtrack Pro, not to mention making the Mac far more attractive as a gaming platform. In addition, the experience gained in building the XBox 360’s triple core 3.2GHz processor could greatly accelerate progress on the G5. A 3+ GHz, multicore G5 would now be a matter of applying techniques already developed. The terrain is familiar, if you will.
Most importantly, however, is the fact that Apple could very well be talking to Intel and, in fact planning to use their chips, but not in the Mac family. Apple currently ships a computer powered by an AMD-manufactured 80486 (that’s an x86 chip, by the way): it’s the Airport Base Station. And there are a number of long-standing rumours about Apple products that could very well have Intel inside.
A Mac OS X tablet computer could be powered by Intel’s Centrino line; since it’s a tablet, it would be used with a different set of applications than a desktop or laptop Mac, so the need to compile for the x86 would be negligible as developers would be writing entirely new programs anyway. And, since using pen-based interface is slower than a mouse and keyboard driven one, the sluggishness of programs running on the PowerPC emulation layer would be less noticeable. And just imagine the bargaining power having an actual, shipping x86 version of Mac OS X would give Apple.
An Apple Cellphone (as unlikely as it is) could well be based on Intel’s Manitoba Cell Phone platform.
An Apple set-top-box could be an even better candidate than a Mac tablet. Since it would likely be a receiver for showing on TV video hosted on a Mac, it could have a very minimal interface, perhaps along the lines of the Airport Express, or a custom remote control-based on-screen one. There would, in fact be no need for such a device to run the Mac OS at all, it could easily be built on a custom operating system, Linux, any number of embedded OSes or even (though it would be slightly perverse) Windows.
So, since the technical hurdles are not to be taken lightly and are, in fact quite treacherous, I think a switch to x86 could only be the result of Apple completely losing faith in IBM’s ability to deliver competitive processors down the line. I don’t think that the current environment engenders that sort of thinking. And, when you combine that with the idea of this rumour being used as a bargaining device or being incorrect about the machines Intel processors are destined for, it seems very unlikely that Apple would decide, somewhat arbitrarily, to switch it x86 at this time.
In Conclusion
It ain’t gonna happen. The Mac is going to stay a PowerPC platform. There are simply no real, compelling arguments for the move that would counter the engineering expenses, consumer confusion, developer mind share problems and the inevitable sales slump. The transition from 68k to PowerPC really isn’t comparable to a PowerPC-to-x86 switch. The PowerPC was so much faster than the 68k that it gave Apple a lot of ‘wiggle room’ in which to hide all the potential pitfalls of the move. The x86 doesn’t provide anywhere near as much of a performance margin, which means there would be no room for error. Either everything would have to go right, or the transition would be remembered as a period of user frustration and software shortcomings. Apple’s viewed in a very positive light by most people right now, and risking that without a compelling reason would be foolish. If there is any weight to these rumours it’s because Apple has a non-Mac product forthcoming and it is using Intel processors. And, above all else, remember that Apple has just fired a very large and very noticeable warning shot across IBM’s bow. With only a few unsubstantiated rumours, Apple has taken the focus of PowerPC news off the next generation of consoles and onto themselves. If the rumours proved true, it would be a serious blow to IBM’s marketing of the PowerPC.
On a lighter note, switching to x86 will prevent Mac users being able to hold a snobbish holier-than-thou attitude to their Windows and Linux/x86 bretheren. In the words of an Anonymous Slashdot poster: “It’s going to be a terribly difficult rewrite for Apple as all their current operating systems depend on correct floating-point results.”
Now, it is simply a matter of ‘wait and see’: all will be revealed on Monday. In the meantime, if you’re not completely sick of this subject, John Gruber has taken on this idea twice, and both times handled it far better than I ever could. I would highly recommend reading his take on the current rumours, and his take on similar rumours two years ago. I do have to admit a Centrino-based tablet Mac would be very cool (even if the Tablet PC market is an almost guaranteed money loser), and both Gruber and I were wrong about the iPod shuffle…
Adendum
The Inquirer is saying that Apple will be starting the transition in the Portable line, rather than the desktops. They also say that Apple’s been cosying up to AMD as well, so a pure Intel x86 line-up is unlikely. This makes sens on both fronts. Apple’s weakest products, speed-wise, are the portables. Freescale simply hasn’t been delivering the clock speed growth Apple needs, and shoehorning the G5 into a portable that weighs less than a car is proving to be somewhat difficult. Centrino-based PowerBooks and iBooks are a tantalizing thought. In addition, the fact that the Inq has Apple shipping AMD-based machines fits as well, since the Opteron is a much more natural fit for the PowerMac line than the Pentium. At least in the way the chips are viewed, if not a technical sense. That said, this is the Inquirer we’re talking about here, and one must always take anything they say with a (very, very) large grain of salt.
Addendum 2
Gruber has weighed in again on the rumours. I’m happy to say that he hits a few of the same points I did, but he also has two very big points I missed. Most importantly, that if Apple isn’t switching to Intel, both C|Net and the Wall Street Journal are going to be caught with egg all over their faces and wildly inaccurate stories to explain. And if you think the Wall Street Journal would put itself in that position, you are very, very wrong. So, he suggests, Apple may very well be about to ship intel-made processors, but those processors would be PowerPC.
A fact I forgot to actually write down is that Apple shares IP right to the PowerPC with IBM and Motorola (well, probably Freescale, now). So it’s entirely possible that Apple could have the ability to hire some other chip maker to produce PowerPC chips. And, Intel is a damn good chip manufacturer. Thus, Gruber suggests Intel will start pumping out PowerPCs, and Apple will in fact be shipping machines based on Intel PowerPC chips. I have to admit, I think this scenario has legs.
Update: Gruber has posted a few more bit and pieces and said he’d put money on the idea of Apple buying PowerPCs from Intel — but only $1. Personally, I learned long ago not to put anything of material value on the line when it comes to Steve Jobs’s announcements.
Posted on 4 June 2005 in Uncategorized
2 Comments »
