Hacker News Reader

Comment by TheDong | Hacker News Reader

I would place "driver support" as on the same level.

Porting a web browser to work on your platform (chromium, webkit) doesn't require rewriting the whole thing, and chromium etc already are designed in a way to isolate browser code from OS-specific code (as we can see by chromium code for windows/linux/macOS/android being mostly shared).

On the other hand, making it so your OS works with the large number of common USB devices, networking chips, etc... it's a fantastical effort which linux only barely manages to accomplish.

View on HN

Drivers and mind share are the only serious problems. Emulating Posix is not a debilitating chore. The browser can run adequately under OS emulation. Getting Vulkan ported is doable. Anyway, ordinary CPU cores are good enough for all but game graphics.

But the overwhelming mass of devices to drive is the show-stopper. Probably the best that can be done is to implement drivers for Qemu virtual devices, and run under a hypervisor that has drivers for real things.

Drivers are mostly not too hard, you just have to do them. USB seems like a challenge though. If your driver model isn't too wild, you can repurpose drivers from the BSDs and Linux. And maybe even windows networking drivers if ndis works for you.

A 1.0 launch really doesn't need to support everything. It just needs to support a large enough handful of common devices. Hell, Apple OSes barely support any hardware.

> Drivers are mostly not too hard

Except when they are, which is anytime something is poorly documented, which is pretty much all the time. Let's not even get started on vendors who don't WANT you to write drivers for their hardware.

> Hell, Apple OSes barely support any hardware.

Would be interesting to see an OS target specifically Apple HW to make it a bit easier on the devs.

For any hardware you're likely to support on v1, there's probably a FreeBSD driver, a NetBSD driver, a Linux driver, and if you're lucky a leaked datasheet that's easy to search for. And a windows driver if you're desperate enough to disassemble that.

If that's not documentation, I dunno what is. Each of them will be wrong, of course, but they're all likely to be wrong differently, so you can learn from the differences.

This can be somewhat mitigated by running these OSes as VM guests, under Linux (or Windows) supervisor. That way, you only need to implement the virt devices. This makes some things not achievable (e.g. 0.7s startup time), but can make bootstrap phase of growth much easier.

“Linux is a free set of buggy device drivers” https://news.ycombinator.com/item?id=13660437

And a POSIX layer.

Reply on news.ycombinator.com