On Wed, Nov 9, 2022 at 5:16 PM Larry McVoy <lm(a)mcvoy.com> wrote: Amen bro ... Corporate pride -- but you were right. In the end, it comes down to: Simple Economics matters more the Sophisticated Engineering but corporate pride wanted to believe my cool engineering is going to crush the competition. ᐧ

What I find incredibly interesting any time the concept of fragmentation comes up is how did several versions of UNIX with slightly differing interfaces create such a headache for UNIX vendors and users in the day, but now we've got a Linux/BSD landscape out there with still pretty significant differences between distributions and UNIX's progeny seem to be doing just fine. Were users looking for different things from their computers in the 90s vs today? Have folks just gotten more used to variability in computing environments and just accept it as part of the plan? What comes to mind for me is the different init systems, desktop environments, networking tools, user management tools, and basically that anything that isn't lore in POSIX seems to be up in the air these days. However, you go back to when SVR4 derivatives were king, they all had the same init, the same useradd, the same /etc/passwd, the same ifconfig. Maybe some of the snazzier new features were pretty variable, but the most basic stuff like starting your system, creating a user, seeing if you were connected to a network, essential administrative functions, were relatively consistent. Nowadays I have to wonder if my init system is runlevel based or some systemd monstrosity. I have to question whether I can rely on useradd or some other tool being present or if I should forgo it all and just edit the /etc files directly. Heck, I couldn't say which but I seem to recall a distro I played around with in the past year where this actually didn't work, I had to research whatever arcane user management tools they shipped with that one because whatever they chose broke with convention so much. I have to pray it has ifconfig or else go look up the docs for iproute2 and iw because nobody can make up their mind on what to replace ifconfig with, just that it has to go and replacing it haphazardly and non-universally is better than fixing/modernizing it. Not looking to start some great debate over which of these components is ideal of course, just remarking at the fact that in the early 90s, if you were on a contemporary UNIX system, you'd probably have no trouble modifying system init, adding users, networks, etc., but today I sit down at an unknown Linux machine and I have no confidence that the particular flavor of system administration that I'm used to will be even remotely represented in the subset of tools that particular distro ships. Luckily, it's free, so perhaps that is what has made the difference, folks are more willing to deal with variability when they aren't paying for what should be a consistent experience, but regardless, the fragmentation in Linux world today feels like it is much more severe than UNIX was in the past, but that's also looking through a lens upon a time I certainly wasn't cognizant of this stuff in. Anywho, that was definitely an informative read, thanks for the share. As someone who is constantly trying to dial in my own personal Linux distro, the questions of standardization and uniformity feature in my mind often. - Matt G. ------- Original Message ------- On Wednesday, November 9th, 2022 at 2:51 PM, Joseph Holsten <joseph(a)josephholsten.com> wrote:

Two things, I think: a) Today most identifiably Unix software is “sourceware”, and so the differences between Linuxes, *BSD, and macOS are fairly easily taken care of (eg. with autotools). b) A lot of Unix software is now distributed (more or less) by the OS vendor. Packaging has hidden the portability problem from the end user. iOS and Android are the most popular end-user Unix systems. None of these concerns matter there — they’ve got two completely incompatible layered APIs that hide the fact they’re Unix from applications and users. Even if you restrict the discussion to non-mobile systems, macOS and ChromeOS probably top the list for end-user systems. Again, none of these concerns matter. I mention this to make the point that “Unix” is not what it once was, both in technical terms, and in commercial success. d

I agree. The differences between desktop/server Linux distributions are largely invisible to application code. THe most obvious exceptions are locations for files. I’ve found it’s basically easier to deal with that dynamically (in the application code), rather than doing an autoconf-based distro switch. Init script vs. units, etc, are minor issues too. Sometimes I’ve found a particular platform doesn’t have a package for eg. the right SSL library, or the right version of something, but that’s fairly rare, and mostly able to be worked around by being conservative in dependency choices. Alpine’s use of musl rather than glibc was a bigger problem, but musl is increasingly glibc compatible. Other libc versions used by embedded platforms, create more problems, but then those applications tend to be fairly narrowly targeted anyway. But all that said … it’s a heap easier than making source code compile across every common Unix in the mid-to-late 90’s. d

On Thu, Nov 10, 2022 at 1:14 AM Theodore Ts'o <tytso(a)mit.edu> wrote: It's all a matter of degree and what things bite you. This is why we had to develop the cluster checker because Linux is all over the place Linux cluster configurations they needed to test. BTW: that had been a developer during the UNIX wars and pure Fortran on VMS/IBM etc days. I suspect they would claim it was not much better for them. Is it better in some areas? Sure, in another no. It depends if it's something that bites you and you care about. Things like the "systemd wars" did not help which I would contend was not much less glorious that the Unix wars -- just different players. In the end, it's all about who is making the call of what is different/what is changing. I'm too old to get too religious about this. The 'sameness' is because of UNIX not because of Linux. Ken and Dennis's core ideas created an industry and set of systems that are all 'close enough that we all can get work done. In the end, at any given time, one group or any other will have the incentive to have their way lead the market - and the hard part for many of us to accept is that the incentive is most often *economic not technical*. Larry's old paper nailed the basic issue. It was use the core ideas of Ken and Dennis - (call it anything you want) - and make it freely available to remove the economic barriers. This is what Linux exploited, and that is what made it successful. I'm pleased to say, that Larry's 'Sourceware' came about --> today its in the form of Linux. But the ideas (the core IP) is called UNIX. And yes there are specific differences. But in general; they are close enough. ᐧ