Hacker News new | past | comments | ask | show | jobs | submit login

"We really are using a 1970s era operating system well past its sell-by date. We get a lot done, and we have fun, but let's face it, the fundamental design of Unix is older than many of the readers of Slashdot, while lots of different, great ideas about computing and networks have been developed in the last 30 years. Using Unix is the computing equivalent of listening only to music by David Cassidy."

Rob Pike 2004, https://interviews.slashdot.org/story/04/10/18/1153211/rob-p...




>while lots of different, great ideas about computing and networks have been developed in the last 30 years. Using Unix is the computing equivalent of listening only to music by David Cassidy.

The problem here is that there aren't a lot of alternatives. You could use Windows, which is like listening only to music by MC Hammer, or you could use a Mac, which is like listening only to music by Duran Duran.

Because of software/backwards compatibility concerns, and how dependent everything is on the underlying OS, it's really hard to change anything in the OS, especially the fundamental design. It'd be nice to make a clean-sheet new OS, but good luck getting anyone to adopt it: look at how well Plan9 and BeOS fared.


> which is like listening only to music by Duran Duran

You say that like it were a bad thing.


i'm hoping you are being sarcastic, but i do not want to live in a world where i only listen to my favorite band


It's OK at first but once you get beyond the shiny, it's quite limiting.


The thing is UNIX, by definition, is never going to move beyond its original design, meaning POSIX + C.

Whereas Mac OS, Windows, iOS, Android, ChromeOS have moved into more productive language runtimes, with rich frameworks, improving safety across OS layers, even if they have a few bumps along the way.


That's not true. Windows in particular is greatly limited by past design choices, such as filename limitations, the security model, the fact that many programs have to be run as administrator because that's how they were written years ago, etc.

There's nothing preventing you from running different language runtimes and such on Linux/Unix systems; people do it all the time. Have you not noticed Mono? It's been around for many years. Plus frameworks like Qt; that certainly wasn't around before the late 90s.


Windows 10 already sorted out the filename limitations.

A few more releases and in around 10 years, Win32 will be drinking beers with Carbon.

Mono and Qt don't change the architecture of UNIX and their adoption across UNIX variants isn't a game changer.


>Windows 10 already sorted out the filename limitations.

No, it hasn't. Filenames are still case-insensitive (and in a terrible way, where it seems to remember how they were first typed but that can never be changed), backslashes are still used for path separators instead of escaping characters, and the worst of all is that drive letters are still in use, which is an utterly archaic concept from the days of systems with dual floppy drives. Also, try making a file with a double-quote character in it, or a question mark. I've run into trouble before copying files from a Linux system to a Windows system because of the reserved characters on Windows.

>Mono and Qt don't change the architecture of UNIX and their adoption across UNIX variants isn't a game changer.

Nothing you've mentioned has changed the architecture of Windows. The fundamental architecture of Windows hasn't changed at all since WinNT 4.0 (or maybe 3.5); it just has an ugly new UI slapped on top and some slightly different administration tools.


Windows NT (the core OS) suffers none of those problems.

The Windows (Win32) environment suffers those limitations. It also suffers 20+ years of strong binary compatibility, broad hardware support, and consistent reliability that systems of similar class (e.g. Linux desktops) can't match.

If it makes you feel any better, drive letters are a convenient illusion made possible by the Win32 subsystem; NT has no such concept and mounts file systems into the object hierarchy (NT is fundamentally object oriented - a more modern and flexible design than is provided by UNIX).

The fundamental architecture of Windows, the kernel, hasn't changed in ages because it doesn't need to; it is far more sophisticated than UNIX will ever be and far more sophisticated than you will ever need. The fundamental architecture of Win32 hasn't changed since 32-bits was an exciting concept and it won't change because the market has said loud and clear that they want Windows-level compatibility. See Windows RT and The Year of the Linux Desktop for evidence that users aren't clamoring to ditch Win32 in favor of something more pure.


> The fundamental architecture of Windows, the kernel, hasn't changed in ages because it doesn't need to; it is far more sophisticated than UNIX will ever be and far more sophisticated than you will ever need.

Another case in point: it allowed MS to write a layer on top of the NT kernel to run unmodified Linux binaries. (Windows subsystem for Linux).


Linux has the same thing, called WINE. It already works quite well, and if they had access to the Windows and Win32 source code and a staff of full-time developers, it'd work flawlessly.

An API translation layer doesn't prove that a kernel is "more sophisticated" than another; that's just fanboyism.


> consistent reliability

Is that a claim you can substantiate? I admit I stopped paying attention around the time that scale-up of expensive servers with high minimum specs + expensive software licenses was overtaken by scale-out approaches + open source, but NT never struck me as being particularly stable in the face of badly written software and drivers. Has it improved a lot in this dimension over the past 15 years?


Is there a possibility to use the NT system without the flaws (?) of Win32? I hardly know anything about Windows programming, and I'm a bit curious now, having read your comment.


Windows exposes personalities above the kernel, in the NT days that meant OS/2 1.x, POSIX and Win32.

Nowadays on Windows 10 it means UWP, Win32 and Linux syscalls.

In theory someone could call the ntdll.dll and create a new personality but those APIs are undocumented and only possibly made available to Microsoft partners.


Complaints of a UNIX refugee on Windows.

For us, Windows pathanmes are just fine.

As for Windows architecture, maybe you should spend some hours reading Windows Internals book series, BUILD and Channel 9 sessions about MinWin, Drawbridge, Picoprocesses, UWP, User Space Drivers, Secure Kernel,....


>you should spend some hours reading Windows Internals book series

I'd love to read the source code myself to see how it works instead


Google "davec apcobj.c"; that'll point you in the direction of some of Dave's brilliant work that dates back to 1989.

(Three versions of NT have been leaked; NT 4.0, Windows 2000, an the Windows Research Kit (which is Win2k3) -- they are all trivial to find online (first page of Google results).)


https://www.microsoft.com/en-us/sharedsource/

Not exactly "you can get it if you want it", but not "you can't even get it over our dead body", either.


You can check out ReactOS which works like Windows but is open source.


> Windows 10 already sorted out the filename limitations.

Only for new applications. These written for the old ABIs still trip over after the 240th character, even though the FileSystem supports much more.

To put in another words: I belive the inherent unix limitations (process model for terminals, process signalling, lack of structure) are still less limiting than DOS assumptions about the consumer hardware and applications of the 80's.


You are moving the goal posts from OS limitations to limitations of old applications.

Unix applications written using old assumptions (e.g. ?14? characters max for symbol names in libraries, ?16? bit address space, assuming the C library knows of gets) can have problems on modern systems, too.


No one still uses Unix applications from the 1970s.

Lots of people still use Windows applications with the limitations mentioned here, either because they refuse to give up their 1990s Windows applications, or the weird little ISVs refuse to update their code.


Old, badly written, programs requiring admin was a problem the first year after Vista was released. Today, not anymore.

All programs of interest have migrated and there is also a compatibility layer redirecting older programs to write to a fake system directory.


What do you mean by unix in this context? Everything save for windows bears a strong kinship with Unix and every OS is more than capable of running additional runtimes and frameworks beyond C.


UNIX and C are symbiotic, regardless of whatever runs on top, only POSIX and C are common to any UNIX.

Windows roots are on VMS, not UNIX. There is hardly anything UNIX related on its architecture, regarding kernel design.


It was an OpenVMS derivative wiyh code copied or clean-slated against a modified form of its behavior. However, I heard the networking stack was from BSD.



Thanks for the tip.


The IP stack is. There's even still an etc/ directory buried in the Windows tree to support it.


This is no longer true, the TCP/IP stack was rewritten on Windows Vista.

https://technet.microsoft.com/en-us/network/bb545475.aspx


>only POSIX and C are common to any UNIX.

What about something like this: https://www.redox-os.org/

Its written in Rust not C. In fact, according to the github stats, there is no C.

>Rust 72.4%

>Shell 13.2%

>Makefile 12.5%

>TeX 1.9%


Yes, but Redox is not UNIX.


It's why everyone focuses on making something on some cross platform abstraction layer or something similar. In many ways, things like amazon lambda are a 'new os'.


So which OS is like listening only to music by Simon and Garfunkel?


On the other hand, the record player he first heard David Cassidy on will still work in the next house he buys. That's because, once we settle on foundational stuff like electricity delivery, we don't break it every time we have good ideas about it. It's the same with everything, like math or computer hardware: after a period of diverse experiments, the foundations solidify and we build immense structures on top of them. It's a net win, overall, even if we get stuck in local maximums for too long sometimes.

In view of that, it's only natural that big segments of software beyond the OS like word processors are also stagnant. We don't actually need a diverse marketplace of competing word processing ideas anymore... the problem is fundamentally solved as far as the public is concerned. It's not as exciting for software developers, but it's totally natural for it to happen.


I think WordPress, wikis and Google (Docs) would disagree with you about word processing being a solved problem. And they seem to do well enough. They offer a fundamentally different notion of what you're trying to do when you're putting prose down. (Personally I hate these systems like Microsoft Word and Google Docs that force you to think in terms of a page. Especially because they don't give you any tools to actually make things lay out properly. But it's okay, because people have been fundamentally rethinking the issue for years now.)

Now if you do want page layout, I think the people who like to do that use something different — not Word. So I don't think they've really cornered any market, unless the market is people who want to use Word.


But just because an idea is old doesn't mean it is necessarily bad. Some ideas like Unix Philosophy have stood the test of time. The fact that this statement was made in 2004 and that Unix is still going stronger 13 years later is proof.


Inertia is only proof of inertia.


Suggesting that Unix has been continuing just by inertia doesn't explain why Apple adopted Unix for macOS or why most newer computers nowadays run a Unix OS.


Apple did not adopt UNIX per se.

Apple failed miserably to create a new OS, decided to buy a company instead and they happened to get the one where Steve Jobs was.

NextSTEP was partly based on UNIX, because it was competing against Solaris and needed some compatibility for easing the port of applications.

It used a micro-kernel like architecture, drivers where written in Objective-C, the whole userspace used the Foundation libraries and the bundles concept, the GUI was based on Postscript engine.

All very little to do with what a UNIX is.

Also if Gassé didn't ask the crazy amount of money he did, Mac OS X would probably be based on BeOS, which didn't have anything to do with UNIX.


Apple took NeXTSTEP and with minimal changes had macOS certified as UNIX, so it is actually UNIX, unlike many UNIX-like operating systems.


I suggested nothing of the sort. I am criticizing your use of "still going strong" as proof of good design. And there are many possible explanations of why Apple adopted Unix. Inertia is certainly one of them! (If you want to write a new operating system, and all your programmers are familiar with unix, and you don't have enough money to start from scratch... you start from unix)


Intiertia among developers. Also since when is Apple know for their software architectural decision-making lol.


Apple didn't make that decision; NeXT did.


The fundamental design of von Neumann architecture is even older and Unix can be treated as just another layer of abstraction on top of it. It cannot really get past its sell-by date.


> We really are using a 1970s era operating system ...

Try and read it this way:

"We really are using a 1970s era operating system ..."


Is it more like listening to music made with a Stratocaster or Les Paul? Lots of great ideas have been applied to guitars too, but lost of people still use the old ones.


Further evidence that Rob Pike is a second-rate engineer and a lousy writer.


Hm? I mean I don't know how one could think that and spearheaded go, but that paragraph seems reasonable to me.


He said that because back then he was working on a new OS. This was PR.


And why do you think he was working on a new OS, instead of improving Unix?


Better question: why did he abandon it?


Wrong. He was working for Google and becoming reacquainted with Linux at the time he said that.




Registration is open for Startup School 2019. Classes start July 22nd.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: