3 ms·

What do you mean by "kernel with no stable APIs"? Linux developers, as a policy, don't break user space. And since when Google became the representative of sta

by tagrun 5y ago

What do you mean by "kernel with no stable APIs"? Linux developers, as a policy, don't break user space. And since when Google became the representative of stable APIs? During 90s, the de facto desktop OS used a monolithic kernel, so clearly, having a monolithic kernel doesn't prevent common adoption on desktops. Fuchsia is a microkernel, which would play nice with phone vendors with binary blob drivers, but this is a licensing issue not a technical one.

tenebrisalietum 5y ago
Probably meant ABIs which causes proprietary drivers to be locked to the version of Linux they are shipped with.
- tagrun 5y ago
  That would be an entirely separate issue related to licensing (which historically led to the creation of GNU and free software foundation BTW).
  - trasz 5y ago
    How is the lack of stable kernel API/ABI related to licensing?
    - temac 5y ago
      There is an indirect relationship in that only out-of-tree kernel-space Linux drivers are inconvenienced by that, that lots of out-of-tree kernel-space Linux drivers are proprietary ones, and that most kernel-space Linux proprietary drivers are dubious on licensing grounds (to be clear, NVidia seems to be one of the most clean in that regard, but I'm not aware of tons of proprietary drivers as clean as they are). The thing is: it makes little sens for a kernel under GPLv2 to maintain an internal kernel space ABI for drivers (and I'm not even sure Linus would qualify with the licensing, but maybe I'm more moderate than he is). Ok I'm aware there is a mess at least on the Android side, but maybe the majority of it is questionable on licensing grounds to begin with (but the Linux Foundation is particularly lenient on licence compliance, btw, not surprising when you see the list of corporate members, and for some of the main ones their past practices - hm VMware hm...) And that being said, I would be curious to see tons of old Windows binary only kernel space drivers continuing to run on modern systems, or ironically even some user space drivers (IIRC, printer drivers are userspace but 32-bits ones do not work on 64-bits systems)... Ok there is more binary compat for a few major versions if you are lucky. But then MS can always randomly requires WDDM 2 for its next OS anyway. In the end, some old and/or exotic hardware work better (or at all) on recent Windows while some old and/or exotic hardware work better (or at all) under recent GNU/Linux distros, but I'm quite unsure this has anything to do with the precise kernel space API & ABI policies. What does not usually work is attempting to project a model on something that has vastly different premises. So yes, licensing can influence technical decisions. If your ecosystem is vastly built around binary only vendors, a more stable ABI is vastly more valuable than if you ecosystem is vastly built around source only free software vendors, AND pushing for in-tree integration on top of that...
snvzz 5y ago
>What do you mean by "kernel with no stable APIs"? Linux developers, as a policy, don't break user space. Sure, Linux tries hard for stable user<>kernel interface. Within the kernel, however, there's no such thing and it's an utter mess. >having a monolithic kernel doesn't prevent Even within monolithic kernels, and even after restricting to unix-like, there's those that are better structured internally (such as openbsd or dragonfly), and then there's Linux, which is an unworkable mess.

by tagrun 5y ago

tenebrisalietum 5y ago

Probably meant ABIs which causes proprietary drivers to be locked to the version of Linux they are shipped with.

tagrun 5y ago
That would be an entirely separate issue related to licensing (which historically led to the creation of GNU and free software foundation BTW).
- trasz 5y ago
  How is the lack of stable kernel API/ABI related to licensing?
  - temac 5y ago
    There is an indirect relationship in that only out-of-tree kernel-space Linux drivers are inconvenienced by that, that lots of out-of-tree kernel-space Linux drivers are proprietary ones, and that most kernel-space Linux proprietary drivers are dubious on licensing grounds (to be clear, NVidia seems to be one of the most clean in that regard, but I'm not aware of tons of proprietary drivers as clean as they are). The thing is: it makes little sens for a kernel under GPLv2 to maintain an internal kernel space ABI for drivers (and I'm not even sure Linus would qualify with the licensing, but maybe I'm more moderate than he is). Ok I'm aware there is a mess at least on the Android side, but maybe the majority of it is questionable on licensing grounds to begin with (but the Linux Foundation is particularly lenient on licence compliance, btw, not surprising when you see the list of corporate members, and for some of the main ones their past practices - hm VMware hm...) And that being said, I would be curious to see tons of old Windows binary only kernel space drivers continuing to run on modern systems, or ironically even some user space drivers (IIRC, printer drivers are userspace but 32-bits ones do not work on 64-bits systems)... Ok there is more binary compat for a few major versions if you are lucky. But then MS can always randomly requires WDDM 2 for its next OS anyway. In the end, some old and/or exotic hardware work better (or at all) on recent Windows while some old and/or exotic hardware work better (or at all) under recent GNU/Linux distros, but I'm quite unsure this has anything to do with the precise kernel space API & ABI policies. What does not usually work is attempting to project a model on something that has vastly different premises. So yes, licensing can influence technical decisions. If your ecosystem is vastly built around binary only vendors, a more stable ABI is vastly more valuable than if you ecosystem is vastly built around source only free software vendors, AND pushing for in-tree integration on top of that...

snvzz 5y ago

>What do you mean by "kernel with no stable APIs"? Linux developers, as a policy, don't break user space. Sure, Linux tries hard for stable user<>kernel interface. Within the kernel, however, there's no such thing and it's an utter mess. >having a monolithic kernel doesn't prevent Even within monolithic kernels, and even after restricting to unix-like, there's those that are better structured internally (such as openbsd or dragonfly), and then there's Linux, which is an unworkable mess.