Hacker News Reader

Comment by jandrese | Hacker News Reader

More likely they would require companies to publish specs suitable for driver authors. I’ve always thought that companies treating their register configuration and I/O offsets as a trade secret is ridiculous.

View on HN

This. Linked post is a misconception of how EU actually work. They will never "hardcode" something like "linux drivers" into the law. But, in theory, they could write a law like "If you want to sell your device in the EU, you must publish a freely available technical spec with enough level of detail to write a functioning driver".

This would not even be something completely new.

It would just revert the market for computer peripherals and interfaces to how it was before the launch of MS Windows 95.

In the MS-DOS days everything was provided with detailed technical information, so you could write your own device drivers, for any operating system.

This ended in 1995, after which most hardware vendors stopped providing documentation and replaced it with MS Windows device drivers.

The days spent trying to get drivers to work and deal with their bugs and conflicts were seriously grim.

Do you realize how much the PC ecosystem sucked back then?

Do you have any point of substance to make here? What do you mean by it sucked, and how does that relate to publishing driver specs?

Yes, it’s a dumb idealistic idea that never worked in practice. Geeks spent forever trying to get crappy drivers working IRQ conflicts fixed, etc.

It wasn’t until Windows 95 and WinHec and “Plug and Play” that x86 based PCs became usable for the mass market

That was largely hardware issues. In particular without a way to negotiate between drivers conflicts were doomed to happen regardless of who was writing the driver, especially with the relatively small number of resources to pull from.

And yet, Apple’s integrated solution didn’t have those issues with the Apple // in 1977 yet alone the Mac II with Nubus in 1987.

Yeah, PC clones were a right mess. The ISA bus was extremely primitive, even by the standards of the day. There is a reason the PCI bus was such a breath of fresh air.

Maybe, but that's a separate matter from publishing specs that would let someone write a driver.

How well did that work in the past when for profit companies had a motivation to write good drivers? You think it’s going to be a seamless mass market experience this time?

I don't see the connection. Two things appear to be true: 1. In the past, it was common to publish actual specs for talking to hardware. 2. In the past, drivers often sucked. You appear to be claiming that the first thing caused the second, but I can't see any reason for that to follow. It's not like hardware manufacturers couldn't, or indeed didn't, write their own official drivers, and on the other hand there's no reason to believe that a 3rd-party driver written using proper hardware docs wouldn't work well. In fact, it's quite reasonable to suggest that the improvement in quality is entirely tied to overall architecture improvements, mostly (as cousin comments point out) device enumeration standards and possibly OS improvements (preemptive multitasking and memory protection).

And how slow was the evolution of both Windows and the entire ecosystem compared to Apple that could release an OS, the hardware and the drivers simultaneously? When Apple releases new hardware it can upgrade everything in tandem. It’s especially apparent now that everything Apple sells runs some variant of the same processor from the AirPods, Watches, tablets, TV set top boxes, phones and even monitors?

And what would stop Apple from doing that while also having to publish documentation?

And the ability to use such a driver. Hell, even the spec wouldn't be such a strong requirement as people can reverse them. The issue is usually the firmware and signature checks.

This sounds much more reasonable to me. Banning unnecessary hardware DRM and requiring specs to get published would be a much more effective way to stimulate the market. Forcing manufacturers to publish Linux drivers would probably just end up in low-effort support on either side. The last person I want writing my Linux drivers is someone who doesn't care.

Yeah, you know it would end up with the manufacturers just releasing a binary driver for exactly one point release of the kernel and calling that box checked. Might even just be some NDIS thing where they wrap the Windows driver to get it mostly working.

This would be my expectation too. Though I’d think you’d need to go one further - hardware manuals AND a license to redistribute any necessary firmware binaries - the latter has been an issue with Nouveau for example.

In fact I think manuals are probably better in a lot of cases, OEM drivers are routinely terrible.

Intel publish datasheets which describe their network interface cards to this level. It's fantastic, literally anybody can write a driver for an Intel NIC for any OS. You can also debug existing drivers when you get strange hardware issues, which is part of my job. No other networking vendor does this.

Reply on news.ycombinator.com