Gnu/Linux Considered Harmful

[rhb]

Seriously, who is responsible for that sh*t and what is wrong with them?

Programmer's remorse: a piece of software can always be improved! Some programmers just can't stop working on a piece of software.

I'm extremely proud to have written two 15,000 line libraries which never had a bug reported against them during the 8 or so years the package was supported.  They continued in use for another 6-8 years without any support at all until they were completely obsolete.

[rhb]

Harry Spencer said it best : "Those who do not understand Unix are condemned to reinvent it, poorly."

Actually, it's Henry, not Harry.  I tried a Linux distro around '95.  I had a Sun 3/60 and a 3/110 in a rather *interesting* configuration.  It was rather complex to set up, but was pretty neat.  Lots of fancy tricks with NFS and NIS.

I installed Linux, connected it to the network and played around for a day or two.  My conclusion was that while no one could beat Windows NT as a commercial venture, Linux just might do it for the sheer craziness of it.  As it has turned out I was right.  MS is incorporating binary compatibility with Linux.

[malagas_on_fire]

From a desktop PC view Gnu Linux may become to be what other OSes has to offer because of tendency to serve the GUI interface and nowadays the gaming... at least some distros have more consistency / stability or less change than others, as following distrowatch... and you can see fragmentation if that's the subject of this topic. You trade stable versions for latest drivers .... Very good for compreend the nature of the OS.

The embedded linux has grown pretty well and did make linux very suitable for low power SBC , not so scattered so far .... but not very documented in some cases...

The unix ... pick you're flavour or the one that suits your needs. Buy a proper book for shell programming

Windows ... it is designed to work  on a PC, user interface, simplicity and driver multi compatibility. 

MacOS never really used so no opinion at al in this subject.. sorry.

[rhb]

It seems like most of the heat in this rant is properly directed at the distribution vendors, not at Linux (the kernel) or (most of) gnu utilities.

No.  It is gratuitous changes in syntax and semantics breaking things.

[rhb]

My biggest complaint is Octave which I used for many years.  It cannot be compiled on Solaris because the autoconf crap is so screwed up.

Would you say Autoconf is such a mess that it would be worthwhile transitioning to something newer like CMake?

Oh, please, no.  Gnu make(1)  works just fine if you know how to use it. KISS

[emece67]

.

[rhb]

It seems like most of the heat in this rant is properly directed at the distribution vendors, not at Linux (the kernel) or (most of) gnu utilities.

I'm building and using my ATE/calibration system using GNU tools: Octave, gawk, nc, ssh. My system is CLI-based, so I know what's going on, and so it runs smoothly on lightweight hosts like rpis and old laptops. Just recently I've automated much of my testing/calibration process using makefiles (make clean; make config; make run; make postrun; make report). The GNU tools, and Linux, have been a godsend.

My focus is thermometry (resistance and optical). Not ready to post any work here yet, but getting close.

You have your make(1) model *exactly* right.  I'm looking forward to seeing your work.

[rhb]

The worst thing is: Linux is being used as an example in many OS design courses. Amazingly it was written as an operational alternative to Minix. Minix was conceived as a teaching tool setting performance and functionality aside, while Linux was the opposite. And now the dirty hack has become the teaching tool of choice. To me it's like teaching Excel instead of programming and SQL instead of data structures.

I am sure most Linux users will not fully understand the relevance of this example and they will dismiss it as the complaints of a fastidious BSD idiot fanboy.

End of rant, congratulations if you didn't fall asleep!

Very nice narrative.   And a good explanation of why Linus would have flunked an OS course under Andy Tanebaum

[rhb]

Where can I start to describe what's wrong with Linux?

Nowhere. Everything goes smooth as long as you use a stable & well tested distribution in a PC with hardware which is supported by Linux (the latter is much less of an issue nowadays) and do a training so you know what you are doing. I've been running Linux servers at customers since around 1996 using Debian until I quit doing company networks. However before letting Linux loose on my customers I took a one day hands-on training on how to install and configure a Linux server. Even today Debian is one of the better distributions if you want to do serious work with Linux.

Just one minor problem.  Most of the corporate users use Red Hat or Suse.  So large, expensive commercial packages are only supported on those.

[sokoloff]

It seems like most of the heat in this rant is properly directed at the distribution vendors, not at Linux (the kernel) or (most of) gnu utilities.

No.  It is gratuitous changes in syntax and semantics breaking things.

In the kernel, in gnu utils, or in the distro? I'd wager the proportions are less than 1%, less than 4%, greater than 95%, respectively.

The kernel seems to have a very strong "don't break userspace" ethos.

[rhb]

If you do not recognize the allusion in the subject line or know what an allusion is, *please* just skip this.

I was expecting something SJW related!

SJW?  Yet another acronym I don't recognize.  please explain.

The allusion is "GOTO Considered Harmful" by Wirth.

Edit:  I'm going to have to stop.  I haven't even made it through the first page.  But the comments are *great*.  However it happened Dave has managed to recreate Usenet circa 1990 and earlier.  Absolutely brilliant people.

[sokoloff]

Social Justice Warrior

[rhb]

Social Justice Warrior

ROFL  Never expected that!

[Marco]

Oh, please, no.  Gnu make(1)  works just fine if you know how to use it. KISS

So tell, me ... without an incredible mount of ad-hoc scripting (ie. recreating a meta-build system on the fly, as with all such endeavours, poorly) how would you allow building something of the complexity of Octave with the same amount of instructions as you can with a meta-build system? (When it works, obviously.)

This is not some POSIX only, or POSIX+X dependent application. You can't appeal to the authority of ancients on this one, these kinds of graphical cross-platform applications have no parallel in early Unix. Even a multi-architecture OS is trivial to build compared to large cross platform graphical applications. New problems, new solutions ... to say they are not necessary is putting you in the same shoes as those who are changing the ancient ways of dealing with long standing problems, ie. pretending you are the smartest person in the room. It's always tempting to do so isn't it?

[0culus]

[snip]

The worst thing is: Linux is being used as an example in many OS design courses. Amazingly it was written as an operational alternative to Minix. Minix was conceived as a teaching tool setting performance and functionality aside, while Linux was the opposite. And now the dirty hack has become the teaching tool of choice. To me it's like teaching Excel instead of programming and SQL instead of data structures.

I am sure most Linux users will not fully understand the relevance of this example and they will dismiss it as the complaints of a fastidious BSD idiot fanboy.

End of rant, congratulations if you didn't fall asleep!

The two OS courses I've taken both used the MIT curriculum, which uses xv6 (basically a reimplementation of AT&T Unix) and JOS (another flavor of the same). I use Linux a lot, but I'm damned if I want to get into the internals. I've found the kernel code to be horrifyingly difficult to read and understand.

I mostly use macOS these days and virtualize everything else I need anyway.

[rhodges]

No.  It is gratuitous changes in syntax and semantics breaking things.

I decided to have a look at Xen, and for whatever reason, it seemed like a good idea to do it with Centos 7...

What the hell?!

[root@xen1 rh]# ifconfig
bash: ifconfig: command not found
[root@xen1 rh]# netstat -rn
bash: netstat: command not found
[root@xen1 rh]# arp -na
bash: arp: command not found

They removed important commands and replaced them with "ip". Bastards.

[Marco]

They removed important commands and replaced them with "ip". Bastards.

Ah well, at least this one isn't on Redhat. It was those damn Russians.

[rhb]

Oh, please, no.  Gnu make(1)  works just fine if you know how to use it. KISS

So tell, me ... without an incredible mount of ad-hoc scripting (ie. recreating a meta-build system on the fly, as with all such endeavours, poorly) how would you allow building something of the complexity of Octave with the same amount of instructions as you can with a meta-build system? (When it works, obviously.)

This is not some POSIX only, or POSIX+X dependent application. You can't appeal to the authority of ancients on this one, these kinds of graphical cross-platform applications have no parallel in early Unix. Even a multi-architecture OS is trivial to build compared to large cross platform graphical applications. New problems, new solutions ... to say they are not necessary is putting you in the same shoes as those who are changing the ancient ways of dealing with long standing problems, ie. pretending you are the smartest person in the room. It's always tempting to do so isn't it?

The range of supported platforms for Octave is rather small.  The last time I tried to build it, they didn't even allow Solaris as a choice when submitting a bug report.

First and foremost.  I neither need nor want the GUI.  A terminal window and gnuplot worked just fine for many years.  So there is *no* reason not to allow building a fringe system without the GUI stuff.  I'm going to try to keep up with this thread, but it's getting really hard.  There are a lot of great comments and a lot of very nuanced aspects to the issues. It is growing at an insane rate and it's not  bunch of stupid comments.  In fact, I've not read any comments I would regard as stupid.  Ignorant maybe, as some of us have a long history with *nix and some much more limited.  But I am *very* impressed by what Dave has created.  I was put off by the clown aspect of his video persona, but he has built an incredible forum.

As a bit of perspective, this is a partial list of *nix systems I've worked with.  I'm sure there were some others, but these are what I can recall at the moment.

Unix platforms or clones:

Minix
Coherent
SGI Irix
IBM AIX
Intergraph CLIX ( pure Sys V on the Clipper chip)
HP-UX "snakes" series
DEC Ultrix
Sun 386i ( rebranded Interactive 386 Sys V)
SunOS 4.x
Solaris
Illumos (nee OpenSolaris)
Intel i386 hypercube
Intel i860 hypecube
Evans and Sutherland ?
Alliant ?
FreeBDS
OpenBSD
DEC/Compaq Tru64 (aka OSF)
Mac OS X

Linux distros:

Slackware
Mandrake
Red Hat
Suse
Fedora
CentOS
Debian
Ubuntu

The list of non-*nix systems is almost as long.

[KE5FX]

Can't really pin that one on millennials. I'm gen x and even most of the guys my dad's age have been using IDEs for decades. You don't *have* to change to the latest one each year. I've been using the same versions of ISE and Quartus I learned back when I first started messing with FPGAs.

That's a huge mistake in the FPGA world.  Those IDEs have gotten so much better (and faster) over the past few years it's not even funny.

[james_s]

Can't really pin that one on millennials. I'm gen x and even most of the guys my dad's age have been using IDEs for decades. You don't *have* to change to the latest one each year. I've been using the same versions of ISE and Quartus I learned back when I first started messing with FPGAs.

That's a huge mistake in the FPGA world.  Those IDEs have gotten so much better (and faster) over the past few years it's not even funny.

Those "better and faster" IDEs don't support the parts I'm using so that's a bit of a moot point isn't it?

That and from what little I've seen, Vivado is most certainly not faster than ISE. What I've got works just fine for what I'm doing, I know how to use it and it gets the job done. How is it a mistake to stick with what I know? I'm recreating 30+ year old hardware, I don't need the latest or greatest.

[KE5FX]

That and from what little I've seen, Vivado is most certainly not faster than ISE. What I've got works just fine for what I'm doing, I know how to use it and it gets the job done. How is it a mistake to stick with what I know? I'm recreating 30+ year old hardware, I don't need the latest or greatest.

Depends.  In the brand-X world, if you're in production with old Spartan3 hardware you're on borrowed time anyway.  If you're in production with Spartan6, you are indeed stuck with ISE.  If you're working with 7-series, you're presumably using Vivado, and the last thing you want to use is an ancient build of Vivado.  Newer versions are much less buggy. 

In any event, recent builds of Vivado are absolutely faster than ISE ever was, that question isn't even close.

For Intel/Altera, I never used anything older than Quartus 16.0, but I found it helpful to move to 16.1 for reasons I don't actually remember at this point. 

I have no problem using a five-year-old build of Windows and a three-year-old build of the MSVC IDE, since those products have only gotten worse over time.  But I've always found that trying to stick with older versions of FPGA tools is an exercise in false economy.

[james_s]

I'm not "in production" though, I'm building one-offs for myself and open source stuff, mostly using Cyclone II's, some Spartan3 and Spartan6 but those are drastic overkill in most cases. I'm stuck at 13.0sp1 for Quartus as that's the last one that supports the parts I'm using. Anyway my point was that you don't *have* to update if you don't want to. The stuff I'm using works every bit as well as it did when I started, and I'm making the same sorts of projects I was then, bigger/faster/newer parts would just be a waste when $13 dev boards are readily available and I rarely fill more than 50% of the logic.

[malagas_on_fire]

Here is an example of a daily usage of linux for an applicantion, e.g. Kicad 5 on Debian Stretch x64 :


You need to dig further documentation and online guidance when errors appear and have some background experience on compiling prgrams.. the infamous "ldconfig" 

https://www.eevblog.com/forum/kicad/kicad-5-looks-official-now/msg1711409/#msg1711409

[bd139]

This is why I usually have a policy of "from distribution repo only".

[janoc]

Oh no, not another circle jerk about Linux being unusable 

My biggest complaint is Octave which I used for many years.  It cannot be compiled on Solaris because the autoconf crap is so screwed up.

Would you say Autoconf is such a mess that it would be worthwhile transitioning to something newer like CMake?

Oh, please, no.  Gnu make(1)  works just fine if you know how to use it. KISS

Please come back once you had to maintain a large software system that has to be both multi-platform and support multiple toolchains (e.g. Visual C++ and gcc/clang).

Octave is that sort of system and it could certainly benefit from a more modern build system. Make and autoconf are absolutely horrid for applications like that - the original complaint about Solaris (a basically dead OS with little support these days) is only a case in the point, autoconf was notorious for these compatibility issues and hacked up config scripts that broke as soon as the system was a bit different than what the authors expected. If you have never needed that, fine. But don't assume that nobody else does.


The rest I am not going to quote one by one but what I see as common complaints here:

* Lack of compatibility between Linux distros - the are actually compatible fairly well but if you are worried about this, pick one that is well supported and stick with it (and software distribution issues are addressed below).

* Systemd bashing - can we, please, finally drop this nonsense fight? Most people who complain about it here seem to do so basically only because:

a) it is different than what they were used to
b) it is "not unix philosophy"/Lennart Poettering is an ass -  , that's more a political than technical discussion
c) it has to be some evil conspiracy pushed on everyone by Redhat -  - it actually improves compatibility between the distros (complaint above - before every distro used to have different init setup, different init scripts, different way of handling system services and supporting that when writing and distributing software was a nightmare).

Also most people have no clue what systemd really does and why (no, it isn't just an init system and there are good reasons for it). Yes, it is different than just a bunch of scripts symlinked from /etc/rc.d but it is not that different or complicated to use and there are plenty of advantages (such as automatic supervision of your services).

And cdev - you are surely aware that the binary logging is not the default on most installations, right? And that you can forward the logs into regular text files, such as /var/log/messages, exactly as before? (And most distributions do exactly that, fyi).

If all you need is a bare bones Linux server that runs somewhere in a container you obviously don't need systemd. But that is not all uses, especially not for systems that are meant to be able to support a desktop or are expected to deal with device hotplugging, dynamic network reconfiguration (wifi ...) and so on. General purpose distributions need to be able to handle that and systemd has made that much simpler.

Lennart has addressed these complaints here:
http://0pointer.de/blog/projects/the-biggest-myths.html

You don't need to agree with him but at least read that first before bashing something with bogus arguments.


* "FPGA/MCU tools work on one distro but not another and I have to keep a separate VM for each!" - sorry, complain to the vendor of those tools. It is not Linux's fault that the vendor doesn't put in the effort to package their software properly, even though it is perfectly possible to do so, even supporting multiple distributions - see e.g. Flatpak or AppImage or how Steam distributes their games on Linux.

When a Windows installer makes a mess of your system by installing all sorts of random crap all over the machine (drivers, background services, updater, DRM ...) and/or fails at it, nobody seems to blame Windows for it neither.