3 ms·
I mean it as a driver developer. Its trivial to do things that are difficult or impossible anywhere else. The environment has good support for driver developer
by JoeAltmaier 11y ago
I mean it as a driver developer. Its trivial to do things that are difficult or impossible anywhere else. The environment has good support for driver developers.
I would dispute the 'total count' argument. Windows doesn't require rebuilding the kernel to get device support for some vendor's hardware. The total count of Windows machines actually succeeding in supporting any given device could be many times higher.
- sudioStudio64 11y agoYou are assuming that your years of experience would lend weight to your assertion...not so with some of these guys. They know what they know and one thing they know is that you, sir, are wrong. I would be interested to hear more about what you like about doing driver development on windows if you have a minute. I've only just started playing around with the platform sdk...just building samples at this point.
- JoeAltmaier 11y agoConfused; what guys? Driver APIs on Windows have professional support for managing all the resources drivers have to deal with - handles, page mapping, protection and ownership, timing and threading. All that is do-it-yourself on other platforms. And the DDK ships with debugging tools. Again, do-it-yourself on other platforms.
- sudioStudio64 11y agoOh, sorry. I was just saying that some of the people on HN won't listen sometimes. I didn't realize that all of that had to be done on your own with other platforms. Wow.
- pgeorgi 11y agoWindows doesn't require rebuilding the kernel (although I guess there are few Linux drivers that are mutually exclusive, and just building all of them is standard practice for many distributions), but every now and then it requires rebuilding the drivers (which may or may not happen). Linux's integrated design makes that the kernel developers' problem, not the users'.
- lqdc13 11y agoIt becomes the users' problem when driver support is not available for the new hardware they just bought. So they have to hunt down the PR on github and build it themselves.
- pgeorgi 11y agoMy scanner and audio interface still work on Linux (and yes, I had to look if they're supported), no matter what updates come my way. Upgrading from XP to Windows 8 disabled them. For the audio interface there's even a Win8.1 driver, but its reliability is a joke. The DDK takes care of everything? Not for this vendor.
- pjc50 11y agoYou really ought to give an example for this. Having said that, the main advantage for developers has to be the relative stability of the device driver API. Linux explicitly doesn't have one, as part of the war on closed source drivers, which makes maintaining an out-of-tree driver a hassle.
- dredmorbius 11y agoI mean it as a driver developer. Fair enough. I would dispute the 'total count' argument. Here's Greg's claim: "Linux today supports more hardware devices than any other operating system in the history of the world." http://www.linuxfoundation.org/collaborate/workgroups/technical-advisory-board-tab/linuxdevicedrivermodel http://www.linuxfoundation.org/collaborate/workgroups/techni... I don't know what data it's based on, but I strongly suspect he's correct on that. Again: my contention isn't that Linux does better on desktop-grade hardware, if anything, that's its weak point. But that's entirely due to vendor leverage by Microsoft, not technical shortcomings. Windows doesn't require rebuilding the kernel... Two elements to that: 1. Neither does linux in many cases. You'll have to compile the kernel module itself, of course (or someone will). That's the case for Windows as well. Generally, for stock kernels, device support is via modules, and you simply have to provide the module. In practice, it's already present, and device autodetection loads modules at boot time. This means that, e.g., you can take a hard drive out of one system, put it in another (of the same general CPU architecture), and it simply boots and works. 2. Drivers are sketchy, mkay? Not allowing any arbitrary code to be loaded into kernel space offers security advantages. It's on of those security / convenience trade-offs. The total count of Windows machines Counting Android as Linux, you're close to losing, if not already lost, with that metric as well. Quick search suggests that Android bypassed Windows in early 2014: http://www.phonearena.com/news/Android-installed-base-might-outgrow-total-Windows-devices-in-early-2014_id33310 http://www.phonearena.com/news/Android-installed-base-might-... Present count is 1.675 billion Android devices: http://www.statisticbrain.com/android-phone-statistics/ http://www.statisticbrain.com/android-phone-statistics/ Windows as a share of all computing devices fell behind the combined iOS and Android count in 2012: http://www.zdnet.com/article/windows-has-fallen-behind-apple-ios-and-google-android/ http://www.zdnet.com/article/windows-has-fallen-behind-apple...
- balls187 11y ago> I don't know what data it's based on, but I strongly suspect he's correct on that. That is called "Confirmation Bias."
- dredmorbius 11y ago
- rlpb 11y ago> I mean it as a driver developer. Its trivial to do things that are difficult or impossible anywhere else. What is a "driver developer"? What's your end goal here - to enable the hardware on a particular OS, or do you have the added constraint of wanting to maintain control of your "driver" by keeping it proprietary, and so you have the additional requirement of a pluggable architecture instead of just contributing to the upstream Linux kernel? In other words, do the obstacles you claim exist only exist because you are trying to write non-free drivers?
- weland 11y agoI haven't written Windows drivers, so I cannot talk about your other points -- but you're not quite right on this one: > Windows doesn't require rebuilding the kernel to get device support for some vendor's hardware. Neither does Linux. You have to compile the module (i.e. the driver itself), of course, but you can modprobe it without rebuilding the kernel. This has always been the case. A long time ago (during the days of kernel 2.4 or so) there were people (ahem, mostly using Gentoo) who tried to shave off a few milliseconds when booting by building everything into the kernel so as not to waste time loading modules. In that case, sure, you have to rebuild the kernel, but that is not a common approach. I... have to wonder, where did you get the idea? Back then (I was... um... I was also frequenting the Gentoo establishments...) Windows was our laughing stock because you had to reboot your computer when installing drivers.
- JoeAltmaier 11y agoWriting Infiniband drivers for Windows, Linux and Sun. So that dates me. It was trivial on Windows, difficult on Linux and impossible on Sun machines. I used to joke "Linux will never get widespread until they have a Have Disk button" meaning I can get a driver from a vendor with a push of a button during installation. Having to build anything is a distraction just to use a new device.
- weland 11y agoI don't think I ever rebuilt the entire kernel when writing Linux drivers. I dimly recall one instance when I was dabbling with/hooking through the network stack and had to do it for some reason, but I don't recall the reason. Then again, I didn't develop drivers for pre-2.6 kernels, so I may be missing something specific to your codebase. > I used to joke "Linux will never get widespread until they have a Have Disk button" meaning I can get a driver from a vendor with a push of a button during installation. The chasm between communities is remarkable :). At the time, I liked Linux for precisely the opposite thing: it came with all the drivers I needed out of the box. No fiddling with CDs and floppies after installation. Except for the odd device that required me to download a driver from the vendor's website (and even then I didn't need to reboot my computer, at least!) I was good to go. Whereas, after installing Windows on a PC, the first thing to do was spend an hour or so installing drivers for all that stuff that made it possible to have I/O more advanced than 640x480, 16 colors and PC speaker. Granted, though, my experience then was limited to desktop computers. Linux hardware support was far less diverse than it is today. There was no fiddling to do if you had stuff that was supported; but the chances of it being supported weren't always good.