When will MS replace the NT-kernel in windows?

[magic]

But wouldn't you agree that it is a logical change and only a matter of time?

As for binary drivers, systemd-kernel will be refactored as a microkernel, with stable subsystem interfaces based on message passing. You will write your driver to process well formed UTF-8 INI messages and respond in kind. No ABIs to worry about, and Rust.NET (coming soon from Microsoft Research) drivers will seamlessly work on all architectures, present and future. For customers with legacy hardware, NT kernel will run as a service, allowing existing Windows drivers to be used.

I do hope he stops all software development work, and never does anything involving Linux again, because he does more harm than good.

I disagree. Considering recent trends in Windows he could only make it better.

[newbrain]

passive-agressive asshole

This is not acceptable. Kindly go fuck yourself.

That said, my post was admittedly sarcastic/tongue in cheek and I should have made that clear.

Benno Rice's opinion on systemd and Poettering, OTOH, is the important thing, and I find it quite worthy of being listened to: don't let the good part of post go to waste.

[nctnico]

The last thing I need today is some religious debate, so I'll just leave this here.

Agreed. Let's not re-iterate a debate that has been long gone & done. If everything that Lennart made is so bad, then why did his software end up in so many Linux distributions? I refuse to believe all those people are a bunch of idiotic sock puppets without any critical thinking skills for themselves. And who cares about how the code looks like? Some people can get totally worked up about C++ while the majority of the programmers can create excellent software using C++.

[Nominal Animal]

passive-agressive asshole

This is not acceptable. Kindly go fuck yourself.

You started it yourself, by calling my argument "religious".  I find that extremely, extremely offensive and aggravating.

It's just the same thing when I pointed out when Stephen Elop became Nokia's CEO, that his past career shows he is good at maintaining status quo but horrible at developing new products, I was called a "Linux conspiracy theorist".  Fuck that; that is just social games and painting an unwelcome but rational argument as irrational.

My opinion is irrelevant, and you can call it whatever you like.  But when you call my argument –– the post where I describe the underlying reasons for my opinions ––, "religious" or "fanboyism" or "conspiracy theory", you get me very, very angry.

Benno Rice's opinion on systemd and Poettering, OTOH, is the important thing

Then say that, instead of painting the entire argument as "religious".

[coromonadalix]

here we go again

i noticed recently  a lot  of arguments on many threads  and they end up closed ...................


CALM DOWN

[Nominal Animal]

If everything that Lennart made is so bad, then why did his software end up in so many Linux distributions?

Because he is socially very adept, as I already described.

It is at the core of the "negativity" against Poettering.  To those with mediocre to poor social skills –– and that covers a large majority of the FOSS software developers (because it is one arena they can shine, and the work is their hobby, with no social life to impact on their productivity) ––, it is like having an extremely attractive co-worker who gets all the raises and bonuses just because of how they look and interact with others, even when the quality of their actual work they get paid for is horrible, always needing others to come after them and clean things up.

This is all very well documented in Debian mailing lists, when Debian decided to adopt systemd.  It was a social decision, not a technical one.

Daniel J. Bernstein is just about the exact opposite of Poettering.  Not only is his code output of very high quality (the bug density is orders of magnitude closer to zero than in general), but his work in math and cryptography is excellent, too.  Yet, many consider him a nobody.  Before the turn of the millenium, qmail was orders of magnitude better (faster, fewer bugs) than sendmail, but just about all sendmail "fanbois" (that I knew of back then, all were mail server admins) hated DJB and qmail with a passion.  Similarly with djbdns and dnscache vs. bind.
He fits well within the math and crypto academia (they're all weirdos there anyway), but all software folks I know ignore him, because he hasn't marketed himself as well as the likes of Poettering.

It is the loudest squeaky wheel that gets the grease; the one that works quietly without fault for decades is always forgotten.

[nctnico]

Well, there are many things on this world which are technically inferior to alternatives (like VHS video tapes, electric cars, MS Windows, etc ,etc) but got hyped by people who got very rich of it. But it takes people buying the products to actually generate that money. Maybe technology in itself is not so important as you think.

And I don't think implementing systemd on any Linux distro is a result of social engineering but based on the simple fact that systemd is more advanced and a better fit (from a maintenance perspective) for modern day systems compared to the old sysv init. Regardless who wrote it.

[newbrain]

...

Since you blocked me, here is the text of the PM I tried to send you. The argument is closed for me.

I don't want to further pollute the thread, and that was never my intention, so this PM.
Please note that I was not directly addressing you and your arguments, it just happened that yours was the last post.

What I felt is that the thread was drifting to the usual sterile diatribe that you are, I am sure, very aware of.
And I admitted the post was clumsy, though done in good faith, and to bring the opinion of someone much more qualified than me.
Why resort to personal insult?

I do not always agree with you 100% - and that's fine - but in general you are one of the few "wall of text" posters that I take the time to read, as you have reasoned arguments.
Any other person with that answer would have been in my ignore list in a nanosecond, after being reported.
You were neither, take it for what it's worth.

[magic]

And I don't think implementing systemd on any Linux distro is a result of social engineering but based on the simple fact that systemd is more advanced and a better fit (from a maintenance perspective) for modern day systems compared to the old sysv init. Regardless who wrote it.

Yes, it gained entry into distribution by being marginally less PITA than sysvinit, and then expanded into an all-encompassing monstrosity.

Funny than Nominal mentioned DJB and DNS, because at one point systemd decided that they need to build their own DNS cache as well (obviously, everyone knows an init system needs one) and ended up repeating well known mistakes, bugs and security vulnerabilities that have long been studied and resolved by everybody else. Like, they could have avoided it by literally asking on StackOverflow "what rookie errors to avoid in a DNS implementation", maybe even by typing it into google.

And that's how Lennart's projects rise: bait distribution with promises of solving known problems, then dump loads of overengineered and unproven solutions on them and use others' production systems as testing sandbox.

Of course you get what you pay for, and that's why even free isn't cheap enough for me to buy anything from this guy. This and his arrogant attitude that everything is other people's problem, even if it could be better solved by systemd than anyone else. I recall arguing with Lennart personally about one of those things, utter waste of time.

[tooki]

It is at the core of the "negativity" against Poettering.  To those with mediocre to poor social skills –– and that covers a large majority of the FOSS software developers (because it is one arena they can shine, and the work is their hobby, with no social life to impact on their productivity)

Just to be clear, though: most open source software development today happens within the scope of people’s jobs. The days where most of it was coding people did on their own time are long gone.

See e.g. https://www.redhat.com/en/blog/evolution-open-source-contributors-hobbyists-professionals

[Karel]

My work is also my hobby. Am I a hobbyist or a professional? Or both?

[nctnico]

My work is also my hobby. Am I a hobbyist or a professional? Or both?

Dunno. You likely never retire though...

[Nominal Animal]

compared to the old sysv init

SysV init is the wrong comparison point, because although it was still widely in use, it was already being superceded by much better init systems.
At that point, we already had Upstart, Runit, and OpenRC in wide use for a number of years.  (Ubuntu 6.10 to 14.10 and Fedora 9 to Fedora 14 used Upstart by default, for example.)

The way systemd requires one to discover services via dbus instead of the service supervisor or init knowing whether a service is active or not –– I do not mean whether the service daemon is running, but whether it is providing specific services or not ––, is extremely fragile.  (This is crucial for dependency-based inits that can start services in parallel, thus speeding up the startup process significantly.)  It would be much better to have a public API (not ABI) for services to announce their state changes –– although because of some oddball services, this interface should be async-signal safe (available even from signal context, which excludes most of standard library facilities) –– which would be both modular (because the actual implementation can vary from dbus to whatever else) but also allow future enhancements.

Because systemd developers are against init agnosticism (they actively object to patches that would allow other init systems in e.g. Debian, and prefer to make packages they subsume systemd-only), we are stuck in the 2010 view of what an init system should be like on Debian- and Fedora-derivatives.  Fortunately, there are still distros that try to support others (which means they have to patch systemd themselves, which is not optimal –– staying to the vanilla upstream is typically a better choice), like Devuan and Gentoo and others.  It is notable that Devuan is not the only derivative Linux distribution that supports other init systems unlike their source; Artix is similar derivative from Arch.  I have not used Alma Linux myself (because I no longer maintain HPC clusters or server-class machines), but I've been told it too supports other init systems besides systemd.

The idea that systemd is anywhere near the top quality/technology in init systems is ridiculous to anyone who has actually compared it to openrc, runit, or s6.  (Anyone familiar with DJB will be delighted at s6 and its implementation. )
It is simply the most popular.  If you think popularity and quality are correlated, I think StackOverflow is a better place for you to interact, than here where the actual facts tend to matter more.  If you accept the popularity argument, then you'll see from my previous posts how that and the observed technical inferiority has directly lead to the 'Poettering hate'.  It is not personal, it is exactly about his output and interactions with others.

That said, as long as I can choose my init system and several are actively developed, I'm happy.  I'm even neutral-to-happy that systemd exists, because it is better than sysv, and it suffices for many people (and it therefore is a valid tool).


To circle back at the topic at hand, "replacing the kernel in Windows [with Linux]", this init sidetrack is actually quite illustrative.

It takes a lot of extra work to make services work with a different init system than they were developed for, unless the service is trivial, or designed to be init-agnostic.  We currently do not have an init-agnostic POSIX/Unix/BSD/Linux service state announcement API at all, so practical init-agnosticism is only possible for trivial services.

Similarly, run-time-monolithic kernels like Linux and the one Windows uses have their own syscall APIs, on top of which lay (userspace) libraries and services, all with a set of unstated assumptions and expectations tightly interwoven throughout the libraries and the kernel.  And there are external companies, driver vendors et cetera, that sometimes depend on very little known quirks of these libraries and services.

As to say Microsoft's ability to refactor or rewrite their libraries and services to work with a completely different kernel, one with a different set of assumptions and expectations and approaches, I'd like to point to the content of the OOXML standard and its attribute definitions (for example, autoSpaceLikeWord95, useWord97LineBreakRules, useWord2002TableStyleRules).  No, there is no documentation as to what these actually mean, other than their name: they mean exactly what the name says, nothing more.  Sure, this is a completely different business unit, but I do believe it is indicative of the culture of how quiet knowledge (like the internal design approaches of services and libraries) is segregated and likely often documented only as the implementation itself.  I could be wrong, though, never having worked for Microsoft myself.

[Nominal Animal]

Since you blocked me

Note that I only block members when I feel I cannot interact in a mutually beneficial way.  I was way too angry to interact with you, that's all; I could not have responded to the private message in any meaningful way at the time.  I've calmed down now.

Please note that I was not directly addressing you and your arguments, it just happened that yours was the last post.

Being labeled "religious" or "zealot/fanboy" or "conspiracy theorist" is one of the things that gets me unreasonably angry.  It stems from both my intent in interacting with others (more that in next paragraph), and past experiences, having been exactly right (because of rational analysis, not religion or belief in some silly conspiracies), but dismissed with a similar label; with afterwards the exact same people telling me to my face that "nobody reasonable could have predicted that".

Reason is what I always try to apply.  My opinions are worth absolutely nothing; but the reasons for those opinions are valuable, because they can be objectively evaluated and argued about.  I always try to explain those reasons, because arguing about opinions is less than useful –– it only generates strife, or logical fallacies like argument from authority –– but arguing about the reasons for opinions is informative and an opportunity to reflect, compare, and learn.

Why resort to personal insult?

Because I got so angry.

Avoiding an argument by labeling it as ridiculous or "religious debate" is one of those social discourse tricks that makes it easy to manipulate how others perceive something, while avoiding the rational-technical discussion of the points in said argument.  Here, me pointing out that hey, this isn't hate towards a person, but about their work output, and how it is perceived as being promulgated primarily via social discourse and social interaction tricks rather than technical merit, was simply brushed aside with a label "religious debate".

Perhaps that does not anger others, but myself, who see rationality and analytical thought as being the only thing keeping modern societies functional and occasionally out of war, and the main antagonist to that to be dogmatic and religious instinct-, feeling-, and belief-based "thinking", it was an insult I could not ignore.  It struck me at my very core.

[magic]

My latest systemd nonsense: a fresh Arch installation hasn't had systemd removed from it yet.
I had to hard poweroff the machine when systemd got stuck on shutdown trying to kill an unkillable task and wouldn't give up after a few minutes

And my first systemd experience: failure to boot, because a systemd component crashed when files it wrote on previous boot attempt were truncated due to hard reboot. Maybe they fixed this particular issue by now, I didn't check.

[Karel]

I had to hard poweroff the machine when systemd got stuck on shutdown trying to kill an unkillable task and wouldn't give up after a few minutes

Try OpenSuse Leap 15.4, that happens about 50% of the time when shutting down
Although it shutsdown after 2.5 minutes, it's annoying...
Leap 15.5 behaves better now.

[Nominal Animal]

The reason behind portability across Unix/BSD/Linux is Single Unix Specification and POSIX, which assume a very specific set of services/features the kernel must provide, including signals.  It also kinda-sorta assumes C, because POSIX and SuS system interfaces are defined in terms of their C bindings as part of the standard C library.

Microsoft's POSIX subsystem only supported the 1990 version of POSIX, IEEE Std 1003.1-1990 / ISO/IEC 9945-1:1990, which is only a small fraction of what the current/practical 2001 version (and later versions; current is from 2018) is.  There was no threads, minimal support for signals.  It really only existed to minimally fulfill FIPS-152, so that Windows NT could be considered for some US government purchases.  The later Windows Services for Unix targeted BSD (and thus Unix), and not POSIX.
Well over a decade after those were deprecated and no longer supported, Microsoft came up with Windows Subsystem for Linux.  WSL1 didn't work too well, so the much better (but still not comparable to running Linux in a virtual machine) WSL2 uses an actual Linux kernel port, virtualized under an internal hypervisor.

In other words, even to be able to run Linux binaries satisfactorily in Windows, Microsoft had to include a virtualized Linux kernel.  WSL1 was an attempt to do that via a lightweight process model and services forming a "library + services" layer on top of the Windows kernel, but it could not support all Linux syscalls.

I did not realize it before, but this alone should be sufficient indication of how different the Windows and Linux kernels are, making it unfeasible to replace one with another.

(Note: The POSIX subsystem for Windows NT was crap, but other than that, I am in no way saying that one thing is better than the other.  I'm only talking about the facts related to interoperability between kernels here.)

[nctnico]

Still, I think the fact that Apple has changed OS and platforms several times already (as pointed out by others), shows that it is possible to radically change what is under the hood of a computer without the users objecting to it to an extend that they buy something else.

[shapirus]

My latest systemd nonsense: a fresh Arch installation hasn't had systemd removed from it yet.
I had to hard poweroff the machine when systemd got stuck on shutdown trying to kill an unkillable task and wouldn't give up after a few minutes

It's a long-standing issue with systemd. It does have some settings to configure a maximum time to wait for a service to shut down, but by default they are set to infinity (or that's the practical outcome of whatever they are set to). That they are still not set to sane default, say, 5 seconds, with higher values set specifically only for those very few services which genuinely may need more time for a clean shutdown, is a good indicator of the systemd developers' attitude.

[shapirus]

without the users objecting to it to an extend that they buy something else.

They will never object regardless of how breaking the change is, because, you know, rounded corners, zero old fart's USB ports, non-existent repairability, and, of course, the logo on the lid... you won't find all those goodies neatly fitted in one package in other products.

[Karel]

I did not realize it before, but this alone should be sufficient indication of how different the Windows and Linux kernels are, making it unfeasible to replace one with another.

If one can run Linux programs on windows, one can also run windows programs on Linux. Shouldn't be too hard for MS.
Imagine a new windows OS based on a Linux kernel with some translator or virtualizer for classic windows programs.
If the cost of implementing such a solution is below the cost of keeping and maintaining the NT kernel, I wouldn't be surprised
if they would do that. Everything to please the shareholders...

[Karel]

[conspiracy mode on]
MS enforces the TPM  >=2.00 so that, in the future, nobody can tamper with the Linux kernel
(despite being open source, remember the term "Tivoization"?) used by future versions of windows.

https://en.wikipedia.org/wiki/Tivoization?useskin=vector

[conspiracy mode off]

[magic]

I'm starting to grow concerned that this isn't a joke thread and you might actually believe it

This paragraph is all you need to know about why no attempt at interoperability with Windows has ever succeeded fully, and none ever will.
Indeed, even Windows NT can't run all software that Windows 9x would run.

As to say Microsoft's ability to refactor or rewrite their libraries and services to work with a completely different kernel, one with a different set of assumptions and expectations and approaches, I'd like to point to the content of the OOXML standard and its attribute definitions (for example, autoSpaceLikeWord95, useWord97LineBreakRules, useWord2002TableStyleRules).  No, there is no documentation as to what these actually mean, other than their name: they mean exactly what the name says, nothing more.  Sure, this is a completely different business unit, but I do believe it is indicative of the culture of how quiet knowledge (like the internal design approaches of services and libraries) is segregated and likely often documented only as the implementation itself.  I could be wrong, though, never having worked for Microsoft myself.

[Nominal Animal]

Still, I think the fact that Apple has changed OS and platforms several times already (as pointed out by others), shows that it is possible to radically change what is under the hood of a computer without the users objecting to it to an extend that they buy something else.

That's what I pointed out too, noting that changing the entire OS is more feasible than just changing the kernel to a completely different one.

If one can run Linux programs on windows, one can also run windows programs on Linux.

Sure you can –– in a VM, running Windows in a virtual machine, like WSL2 does for Linux.  The main difference is that the Linux kernel runs happily headless, without any kind of display or input devices.

Note that MSYS2 and Cygwin are not binary-compatible, only support libraries and services that one can use to compile POSIXy/Linuxy C sources to work in Windows.  There are many quirks from devices to filesystem name case insensitivity that differs from typical POSIXy systems at runtime.

Imagine a new windows OS based on a Linux kernel with some translator or virtualizer for classic windows programs.
If the cost of implementing such a solution is below the cost of keeping and maintaining the NT kernel, I wouldn't be surprised
if they would do that. Everything to please the shareholders...

I've tried to show the reasons why I believe the cost of implementing such a solution would be immense, unrealistic.

Creating a new Windows OS based on a microkernel or a BSD-derived kernel, capable of running older Windows binaries (including all current OS APIs/ABIs and services) via virtualization (a derivative of the existing Windows kernels designed for backwards compatibility), now that would be interesting, and technically at least somewhat feasible.

However, it would require a deep change within Microsoft culture, though, making it unfeasible from a practical point of view.  If you look at Apple and early OS X, their attempts at making many things open source, they basically failed; I believe the underlying reason is that the open source approach was simply not compatible with the Apple company culture, costing more than it benefited the shareholders.  (We can discuss CUPS and whether it is an exception to this, but we'd need to look at its history to correctly assess it.)

What I believe would kill such efforts, is Microsoft's dependency on keeping major proprietary software houses and hardware vendors (developing their own drivers) on board.  Switching to a more POSIX-like kernel and interfaces would mean they'd have to port their code to that, which would lead to porting to additional systems like Linux (and obviously Mac OS, too) much easier, risking Windows' market share on the desktop in the future.

It would be too risky for the shareholders, I believe, no matter how interesting such a solution would be technically.

[Karel]

What I believe would kill such efforts, is Microsoft's dependency on keeping major proprietary software houses and hardware vendors (developing their own drivers) on board.  Switching to a more POSIX-like kernel and interfaces would mean they'd have to port their code to that, which would lead to porting to additional systems like Linux (and obviously Mac OS, too) much easier, risking Windows' market share on the desktop in the future.

And this is the first argument in this thread that makes sense from a business perspective.
All the rest is only about how difficult it is (technically). MS has done many big and costly changes,
either under the hood or above and some of them weren't welcomed by the users.
I believe this is really something that they could do. I'm not sure if it would make sense to me but I
wouldn't be surprised if they do it. All the signs are written on the wall.