[pohl]

The original article makes an excellent point, and expresses it with a great deal of coherency: the reluctance of computer makers to engage in operating system development is based on an outmoded fear that was established in the days where interoperability was hard because we hadn't yet figured out things like open document formats, open network protocols, portable APIs, portable language runtimes, and virtual machines. A lot has changed since then, so why are they still so reluctant?

This is an excellent question.

Your response seems to have a very tenuous grasp on coherency. I'm trying to pull out what you're trying to say. It appears to hinge on the transitive verb "to matter", and I can find only one introductory paragraph where the author uses this verb, and he doesn't say anything objectionable the two times that he uses it:

He says that hardware and software both matter. I find that hard to refute. Then he goes on to say that if you asked him to say which matters more, he'd say software. I'm not surprised that he would say this, since he tends to be a "user experience" guy and I'm willing to grand him this premise for the rest of what he wrote.

Honestly I can't tell what counter-argument you're trying to make. Yes, Google has been successful. It certainly wasn't because of their hardware. Regardless, you're only responding to the setup of the thesis, not the thesis itself.

[balakk]

A lot has changed since then, so why are they still so reluctant?

It's this premise that I find inaccurate. Sure there's open everything, but writing an operating system from scratch to support all this is still hard. Hell, give me one commercial operating system written from scratch in the last decade.

The point is, why do it? Maintaining an operating system is big money. And you can't stop there - you've got to have a full-stack offering - business apps, fun apps, drivers, the whole set. Add to that support, interoperability with the world, backwards compatibility etc. It's a long-term commitment; you can't back out of it that easily.

The risk-to-reward ratio is pretty small; unless you have some earth-shaking innovation up your sleeve, and/or it reinforces/supports your business model significantly.

[cubicle67]

Why from scratch? If you remove that seemingly pointless requirement then there's quite a few examples, starting with the Litl OS which is based on Ubuntu

Heck, there's plenty of good solid starting points. What about BSD? Linux? Android?

[balakk]

Still doesn't matter. Unless you create an ecosystem around your software, I think it still isn't relevant in the larger scheme of things. For eg., can Maemo, Android, WebOS etc all share applications, APIs, drivers, and other infrastructure seamlessly?

The point is you need to end up creating an ecosystem around your offering, which is non-trivial. Even if you do, you may still end up as a niche player. Litl is nice; but how well do you think they'll do against traditional netbooks?

[steveklabnik]

Depending on how you architect your OS, it doesn't have to be impossible.

I have a little Exokernel project I devote my Saturdays to, and we'll actually be able to offer a POSIX-compliant libOS, and run anything that Linux does.

Or, there's always the hypervisor route, on desktops anyway.