Don't ask – don't tell: Palm clarifies OS strategy
Clear as mud
Analysis Talk about rotten timing. As Palm's PalmSource developer conference was due to kick off yesterday, much of central San Jose, which was hosting the jamboree, suffered a power brown-out.
Traffic lights were dark as motorists tried to negotiate three-lane wide four-way intersections without guidance, and hundreds wer e trapped in the Convention Center's underground parking lot.
This threw the official business into disarray, and keynotes were pushed into the afternoon. But all was not lost. Selected soft news media had been given an NDA-briefing the previous morning, and dutifully filed the reports to fill the void. The Register, along with swathes of the international press was excluded from the briefings. Why, we don't know.
Perhaps it's because we're highly selective about NDAs, although that doesn't account for why the non-US press was shunned en bloc. (This is like a European firm telling the New York Times or the Wall Street Journal to go to hell, preferring to brief the local Lichtenstein newswire).
Perhaps it's because Palm genuinely doesn't see a market outside the United States - a view supported by the cancellation of the m705 in the UK for an example. Or perhaps because it's lazy, and knows it can easily buy pages of US coverage simply by making a call.
Or perhaps it's because The Register tends to ask the wrong sort of questions: questions that have brought you a succession of scoops on Palm's increasingly frayed business relationships, and analysis which were considered rank heresy two years ago, but which are now conventional wisdom among the brokers who make or break Palm stock.
Finally yesterday, Palm announced anew operating system, after many years of false starts. But the clarification it seems, is a measure to buy some time, and amounts to blowing a cloud of smoke over the proceedings.
How so? Well Palm's got a new hardware platform. It's ARM, and it's cannily rustled up a choice of suppliers. It's got a new homegrown operating system too, which is much more sophisticated than the patchwork manquÈ devised by Jeff Hawkins as he devised the first, brilliant low-cost Palm. Only Palm's decided that it's not going to tell developers about the new OS, or let them access it, until it quite sees fit. And that isn't now.
Palm OS 5.0 is API compatible with Palm OS 4.1, it transpires. Or it should be. However, it seems that your applications might not work on ARM - particularly if, as so many brilliantly improvisational Palm programmers have done, they peek and poke around the memory heap. This has led to a platform which is essentially un-migratable.
Now a company in this position can go one of two ways. It can do an Apple, and simply port the new OS to a new processor, and not make any promises about enhanced functionality. It can sell it merely as a performance upgrade.
In a minor modern miracle, this is exactly what Apple did in 1994, as it moved from 68000 to PowerPC. It promised developers the same crazy freedoms as before, offered no new CPU-specific features, and as a result, compromised raw performance and potential for backward compatibility. This took place while Palm's software chief David Nagel was managing the doomed Copland OS into oblivion: Copland took up thousands of man year's work at Apple before it was abandoned.
The other option is to virtualize the legacy sessions, much as Intel did so spectacularly with the 80386. This allowed DOS sessions to be multiplexed, while new registers controlled the sessions, and carefully added on a new native instruction set.
Again, there's a downside, in that the legacy APIs really do look like legacy APIs. That makes the transition look awful painful to loyal developers, who ask for some kind of carrot in return for the pain of modifying their applications to run native.
Palm's opted for the middle path, but it's done so in a most peculiar manner.
Instead of evanglising the riches of its great new modern OS, with selective forays into the technical press, Palm has decided to postpone the painful migration. Palm is presenting OS5.0 as a more or less compatible version of OS4.1 that runs a whole lot quicker.
In the meantime, the same restrictions apply: Palm developers are obliged to invent ingenious hacks, such as the one that overcomes the limit of four sockets open at any one time. We put this specific point to Steve Sakoman yesterday, knowing that Steve had nurtured the brilliant BeOS for a decade, only to find himself trying to teach a moribund organization to walk on its hind legs for the first time. To his credit, he promised that fixes would be forthcoming.
So that means the migration is postponed until PalmOS version 6.0, scheduled for any time in the next two years. (Six doesn't have a name yet, officially at least). Migration from PalmOS to a protected memory, fully multi-threaded OS will require a major rewrite of many applications, as we've consistently pointed out. But Palm's roadmap is still blank: it won't release the new APIs until such a time as it thinks good and proper.
Developers we spoke to after the announcement - ranging from the military to some key verticals - were disappointed to be given only half roadmap. APIs are the lifeblood of a developer forum, and Palm's failure to disclose - let alone evangelize - the new, funky 32bit APIs dismayed more than a few. Because they know they're going to be busy.
First they have to clean up their code so the Dragonball apps will run on the new ARM platform. Then they'll have to make severe modifications to the code they've just fixed to take advantage of the new OS features, discarding those changes. And those modifications are as yet unknown, until Palm publishes ARM APIs. Which is a lot of work, for what exactly? (We hope to be able to tell you real soon. Right now Palm has a brand new OS, but it's not going to let developers use it. So don't ask, OK, and we won't tell? ®
Sponsored: Benefits from the lessons learned in HPC