LWN: Comments on "How much GNU is there in GNU/Linux?"

How much GNU is there in GNU/Linux?

vonbrand — Sat, 28 Feb 2015 14:49:21 +0000

No new libc, just a fork. Remember the gcc vs egcs fiasco, which ended up by renaming egcs to GCC and silently shutting down gcc. No license change that time, but a GPLv4 might force it...

How much GNU is there in GNU/Linux?

farzin — Fri, 27 Feb 2015 23:58:29 +0000

three points

1) how has the researcher compared the contributions of GNU vs Linux vs etc. his graph? number of man hours? lines/characters of coding? sometimes you can optimize software by reducing the number of lines/characters. I remember a Steve Ballmer interview where he was frustrated over a contract with IBM. IBM wanted to pay Microsoft x amount of dollars for every x amount of lines of code. Of course, he argued that optimizing software sometimes means fewer lines of code.

2) Some in the comments probably haven't read the text under the graph either where it explicitly mentions that gnome was a GNU project too which would bring the percentage up to 14%.

3) GNU/Linux has evolved/changed through the course of time. It is important to select an appropriate method to judge the % of change. in molecular biology, DNA–DNA hybridization tells us the % of similarity between human and chimp is 98%, other methods of comparison
to take an example from evolutionary biology, you may have heard the popular notion that 98% of the human and chimp genome are identical. But this % is only the result of one method of genome-comparison (DNA–DNA hybridization), other (more accurate) methods give you a much lower %.

similarly in programming code, depending on the method of comparison, you can arrive at varying percentages.

How much GNU is there in GNU/Linux?

JanC_ — Mon, 06 Jun 2011 18:52:33 +0000

Well, there is a page that explains Ubuntu is based on Debian GNU/Linux...

GNU developed the catalysts which others use

coriordan — Sun, 05 Jun 2011 10:59:20 +0000

Thanks PO8, that's very interesting. I'll look into this more sometime.

Other people's accounts seem less optimistic about the other compilers that existed at the time. (RMS, Michael Tiemann, and others whose names I forget or don't know) I'll review those statements more critically next time.

Maybe those tools happened to work ok for your usage, but weren't good enough for larger/different projects? GNU was written in C, so it doesn't seem like the C++ frontend could be what made GCC important.

> Somebody would have, too, if

In all fairness, compiler development is littered with corpses of statements like this :-)

I think the fact that the GNU project actually did it, is what makes GNU special.

GNU developed the catalysts which others use

coriordan — Sun, 05 Jun 2011 10:51:47 +0000

One difference is that RMS's initial use of whatever system while writing GNU software was only as a temporary bootstrap, and it could have been any system (different developers surely used different systems - some Unix, some not).

GNU/Linux and the GNU development tools are developed by groups that communicate and collaborate to help each other. The GNU development tools are part of what makes GNU/Linux development continue.

Another metric besides LOC

neilbrown — Sun, 05 Jun 2011 03:39:47 +0000

> P.S. To answer the question asked by the title of this article, I suppose that there really isn't a better way to determine how much GNU in GNU/Linux other than LOC. Sigh.

Which leaves just a small step to the logical conclusion... it doesn't matter.

This isn't a contest. This is a collaboration. Let's just say a heart-felt "thank you" to every contributor, and not try to measure that which cannot be measured.

(Isn't that the whole point of ditching the 4-clause BSD license)

Another metric besides LOC

pr1268 — Sun, 05 Jun 2011 03:12:23 +0000

I'm not totally against using LOC as a metric, but it does seem to lend itself to abuse (see Tchernobog's comment above). The size of compressed source tarballs could be equally abused (consider if I threw in a 1 MB random bits file in my source tarball—that would certainly enlarge my contribution).

I'm unsure what else to use. I admit that this is a lame answer, but it's just that the stories of abusing LOC metrics permeate folklore far beyond the link in Tchernobog's comment.

I suppose a better way of measuring software code would be based on features added, tested, and verified to the code base (for new development), or defects fixed and verified (for maintenance). If an individual were to have an unusually low number of features added/defects fixed, then this could be investigated—perhaps it's not attributable to the individual being lazy or incompetent but rather the particular code is hairy. Conversely, if some "hot shot" adds a hundred features/fixes, then this too might be attributed to a bunch of simple additions/modifications, like cosmetic fixes or single-line updates (The Dilbert example I linked notwithstanding).

Please understand that I do not wish to preach management style/technique. Besides, I'm not currently in a managerial position, nor do I consider myself prepared for such a role. I'm just suggesting intuitively what might be better than LOC for software metrics.

P.S. To answer the question asked by the title of this article, I suppose that there really isn't a better way to determine how much GNU in GNU/Linux other than LOC. Sigh.

Another metric besides LOC

dlang — Sat, 04 Jun 2011 05:12:05 +0000

I get that you don't like LOC as a metric, but the only alternative that's been presented is the size of the compressed source .rpm files.

what metric would you propose to use instead?

GNU developed the catalysts which others use

fsateler — Sat, 04 Jun 2011 02:31:29 +0000

Well then it should be called <some old UNIX brand>/GNU/Linux, since without <some old UNIX brand> gcc, glibc and gdb would have never been possible to begin with!
Or perhaps WINE should be called Microsoft/WINE, since without MS Windows, WINE would not exist.

I think the argument does not make any sense at all.

Remove the GNU C library and then report how much of that software runs.

dmag — Fri, 03 Jun 2011 16:08:46 +0000

TCC did that years ago.

http://bellard.org/tcc/tccboot.html

How much GNU is there in GNU/Linux?

qubit — Fri, 03 Jun 2011 14:56:14 +0000

This succulently shows that right now, today, GLibC is more important then kernel.

This must be what they mean when they talk about the juicy details...

How much GNU is there in GNU/Linux?

vonbrand — Fri, 03 Jun 2011 14:53:37 +0000

Then why didn't RMS insist that it be called GNU/Solaris? All Sun machines I saw did have gcc as their choice C compiler, and ran assorted packages from a collection of GNU (and other) software packaged for its awfull package system... What about DG/UX (Data General's Unix), where the official C compiler was gcc, as were the development tools, and AFAIR much of the userland also, came directly from GNU? That was the state of the art Unix system in the 90's, much as today it's a Linux kernel beneath essentially the same sort of mix of applications from a variety of sources...

How much GNU is there in GNU/Linux?

vonbrand — Fri, 03 Jun 2011 14:46:18 +0000

Better get a grip on the BSD saga. The "freeing" (final removal of the last restrictions, really) was due to people at UCB, as part of a much earlier stream of "software should be shared" movement than GNU. Inspiration probably flowed from BSD (and similar) movement to RMS, not the other way around.

Remove the GNU C library and then report how much of that software runs.

vonbrand — Fri, 03 Jun 2011 14:35:52 +0000

Exactly. The BSD folks won't be thrilled because they don't agree with the FSF stance.

I for one would like it better if people just agreed on thanking for (software) gifts, wherever they come from (be it public domain, TeX, X, BSD, GNU, ...), and take their political views elsewhere. In particular, please don't insist on slapping your campaign flyer on everything that comes your way.

Another metric besides LOC

pr1268 — Fri, 03 Jun 2011 14:22:23 +0000

The LOC anecdote you linked reminded me of an orthogonal software engineering metric SNAFU that's equally silly: Fixing bugs for money! (A college professor told me that this actually happened at IBM sometime in the 1970s.)

Link to rejected proposal as "proof"

nix — Fri, 03 Jun 2011 09:51:10 +0000

That's definitely true, as planet.gnome.org never contained announcements/promotions of proprietary software anyway. His request was that it never mention proprietary software at all. As this amounted to an attempt to control what hundreds (?) of people write on their blogs it met with the sort of success you might expect.

How much GNU is there in GNU/Linux?

mpr22 — Fri, 03 Jun 2011 09:16:01 +0000

Conveniently, the hard-g pronunciation of "gnu" would be written /gnu/ in the IPA :)

GNU developed the catalysts which others use

PO8 — Fri, 03 Jun 2011 08:20:27 +0000

Just for the record, in 1991 Johnson PCC and dbx/adb were perfectly viable alternatives to GCC and GDB. Not nearly *as* good, mind you, but plenty good enough for the work that needed to be done. Technically, Johnson PCC was probably never "open source", but it was also tiny and simple, and would have been easy to replace with an open source equivalent. Somebody would have, too, if GCC hadn't been handed to them for free.

I was building commercial products using these tools as late as the late 1980s.

The big break for GCC actually came with the advent of C++. Up until GNU C++, the only available C++ implementation was AT&T's fairly horrific CFRONT preprocessor. Not sure I'm thanking the GNU folks for making this language viable, and at any rate it was never really used for fundamental infrastructure.

What did you think BSD used for tools before RMS helped them out, anyhow?

How much GNU is there in GNU/Linux?

tchernobog — Thu, 02 Jun 2011 22:51:19 +0000

I am sorry, but:

a) I said *I* will continue to name it GNU/Linux. You can do whatever you want, and I wont fire a shotgun at you. However, if asked, I will continue to motivate it in the same way.

b) nobody forces a name on Linux, except maybe Android people. The kernel still is named Linux. It's the whole system that's named differently. If one is to follow your "don't force names on things logic" you can't name your system Debian, Fedora, SuSE, Gentoo or Ubuntu, for that matters. By the way, I am free to call it Bonkers OS, if I like, since there is no trademarked name for the whole combination as an holistic system.

c) I am also interacting with Windows computers, indirectly. I do not go so far as to say I am using a Microsoft Windows system just because of that. I find *your* comment rather naive.

d) With my software engineer hat on: LoC *is* a poor metric, they not only discourage you to use that at school (at the university they fail you outright if you try to use that directly to estimate effort) and at work (Bill Gates is credited to criticize the IBM effort-estimation mode based on SLoC by saying: "Measuring programming progress by lines of code is like measuring aircraft building progress by weight."), but commons sense should help you too to understand it. Numbers will not help you when you use the wrong method to measure the wrong thing.

I do not have to show that the numbers themselves are wrong, as you suggest. Running the experiment again would be pointless, because its based on a wrong set of hypothesis.

I am claiming that this report shows nothing of interest to support your denigrating notion about the GNU project, to which you seem to have a personal grudge against.

LoC was introduced as a metric around 1960. Not only agreeing about what SLoC means is difficult, it also is hard to adjust that value for different languages and programming paradigms.

Look: two SLoC:

--------------------------

#include <iostream>
int main () { std::cout << "Hello world!" << std::endl; }

--------------------------

Look: more SLoC:

--------------------------

#include <iostream>

int
main ()
{
using namespace std;
cout << "Hello world!"
<< endl;
return 0;
}

--------------------------

A nice story is also here: http://www.folklore.org/StoryView.py?project=Macintosh&...

I advise reading the following book: N. E. Fenton, S. L. Pfleeger, "Software Metrics: a rigorous & practical approach", PWS Publishing 1998

Link to rejected proposal as "proof"

coriordan — Thu, 02 Jun 2011 22:28:40 +0000

> about 95% disagreeing with RMS

Maybe 95% of the volume, but that was mostly just two *very* loud people :-)

> RMS's request was similarly ignored

His request was that http://planet.gnome.org not be used to promote proprietary software. I didn't follow the whole issue but no announcements about proprietary software jumped out at me when I checked just now.

How much GNU is there in GNU/Linux?

pr1268 — Thu, 02 Jun 2011 20:54:08 +0000

In a fraction, 1/3. (Three out of nine letters when you include the '/'.)

Spoken, also 1/3 (syllables). Assuming you can pronounce "GNU" in one syllable. (I usually pronounce the 'G' hard, so it sounds like a mini-second syllable, as in "guh-NOO" [sorry, I don't know IPA].)

;-)

How much GNU is there in GNU/Linux?

nix — Thu, 02 Jun 2011 17:43:13 +0000

Quite so. This is the logic behind the FSF's adoption of things like the X Window System and a bunch of BSD utilities -- and, indeed, the Linux kernel itself -- under the GNU aegis: that they are freely available and necessary to produce a free operating system, thus we indicate that they are part of 'the GNU system', i.e., what we would now call a distribution. (You'll notice that this distribution is purely notional: there is nowhere you could download a running 'GNU system' from, you have to assemble it out of pieces)

Where the FSF goes haywire is that it considers that its distribution is a fundamental property of the programs that make it up, and therefore also of other distributions made out of the same pieces. You could as easily say that coreutils should be called RedHat/GNU coreutils because of the longtime employer of its maintainer: by extension, because RH maintains so much of the software that goes into Linux, Ubuntu should be called RedHat/Ubuntu. This is plainly mad, and RH would fight it because it's trademark dilution as well. (You could argue that coreutils's copyright is assigned to the FSF so it should be allowed to keep other entities out of its name, but X's copyright *isn't* assigned to the FSF yet the FSF feels happy to call it part of the GNU system. So copyright assignment cannot be relevant here.)

Here, as in many ways, RMS and the FSF are stuck in the past: in the era they come from, and in the proprietary world now, products really *did* get dubbed with the name of the vendor who was distributing them. The free software world stopped working like that fifteen years or more ago, but the FSF never changed its ways here. Product names are, more than ever, arbitrary monikers, not credit lists: but the FSF doesn't seem to have noticed.

Link to rejected proposal as "proof"

nix — Thu, 02 Jun 2011 17:30:46 +0000

However the thread itself ran about 95% disagreeing with RMS (you could feel the incredulity coming through the screen), 5% agreeing, and RMS's request was similarly ignored. That's... not the way he acts when making requests to projects in which the GNU Project holds the copyright.

How much GNU is there in GNU/Linux?

nix — Thu, 02 Jun 2011 17:26:02 +0000

Yeah, but it's hard to do much without glib these days. It's pretty much ubiquitous. libxml is even more so (though very much in maintenance mode, it makes minimal use of GNOME libraries, and is only really an official part of GNOME because its main developer was a GNOMEr when he started with it, I'd say: it's really a Veillard project.)

How much GNU is there in GNU/Linux?

nix — Thu, 02 Jun 2011 17:23:32 +0000

I didn't realise how dead it was. Most recent release 2008; ChangeLog lists a release on ?? February 2009, so presumably hasn't been touched much since then; cvsweb shows that even the 4.8.0 release notes were never completed. X server changelog last updated on Feb 1 2009. Not a single update that I can see after that date.

Zombies with shotgun wounds in their feet move more than this. This is a project with zero active developers. (I wonder what David Dawes is doing with his time, now he threw his co-developers out of the house because they wouldn't give him enough credit and discovered that when they've left the house is so very empty?)

GNU developed the catalysts which others use

nix — Thu, 02 Jun 2011 17:16:53 +0000

OK. So we should call it Carlini/Kosnik/Wakely/Linux (including extra contributor names in strict alphabetical order as their contributions rise above a certain threshold). If we get enough contributors within a sufficiently short timespan we can watch people choke to death as they try to speak the system's name!

How much GNU is there in GNU/Linux?

southey — Thu, 02 Jun 2011 16:12:50 +0000

This is a rather naive to say that you don't use all of the kernel. You (or perhaps the distro) have used the full kernel source to obtain a binary because the compiler has to go through the complete kernel source based on your input. Also while your particular system may not have all drivers, you and your system may directly interacting with systems that do (GPS, internet servers etc.).

If LoC is a very very poor metric, how can you say that [Linux] has a lot of GNU code in it?
As dlang said above, show that this is wrong. After all the code used is available at the linked site.

The comments in this article show that this naming scheme is just pure fud. Sure GNU project host a number of projects that are very useful to use. But since Linux is important to you, does Linus and other developers have the same right to call you [something]/tchernobog? I agree with mingo that names should not be forced on others (especially given you other comments about openness).

How much GNU is there in GNU/Linux?

tchernobog — Thu, 02 Jun 2011 13:39:29 +0000

Isn't anyone bothering to question how those numbers were achieved? LoC is a very very poor metric. It does not capture usefulness of code, its complexity, or how many commits were needed to get there.

If I were to choose between two pieces of code doing the same thing, I would choose the one doing it in less lines of code. It could be an index of better maintainability, perhaps. I am just saying that you can't tell anything from these numbers. They just start flamewars.

For instance, I don't use all of the kernel; most drivers do not apply to my hardware. I use most of the glibc functionality, though.

And I will continue to call it GNU/Linux, not because it has a lot of GNU code in it, but because I think that it is important that people remember they are using a free-as-in-speech system. Or in a few years we will end up with a lot of Linux (android, google OS, whatever) systems that will have a small free portion, and will be uncannily similar to the old windows world again. No changes allowed. DRM everywhere. Patents to prevent you reverse engineer it. Etc. Calling it GNU/Linux normally has some people asking me "what is GNU?"; that is the most important part.

How much GNU is there in GNU/Linux?

tchernobog — Thu, 02 Jun 2011 13:31:15 +0000

Well, you could say that no code comes directly from the FSF. The GNU project and the FSF are two distinct initiatives; one technical, the other legal.

Sorry, but no...

khim — Thu, 02 Jun 2011 08:28:23 +0000

EGCS was an experiment in new development methodology and never tried to split from GNU. All changes were always assigned to FSF and when EGCS was renamed to GCC it still accepted FSF's (actually RMS) decisions WRT to licenses, etc.

Basically it never tried to stray away from guidance of FSF so there never was any need to fight with it. Emacs/XEmacs was such a split while xemacs is quite successful you rarely can find it installed by default.

True, association with GNU is very often a burden and the only thing it gives the project is exposure - but this is quite important for free projects...

Remove the GNU C library and then report how much of that software runs.

freddyh — Thu, 02 Jun 2011 07:06:14 +0000

They indeed did, but the mentioned quote was from years before the Hurd was finished.
I don't need to remind you of how much the Hurd was delayed in the early 90's, do I?

How much GNU is there in GNU/Linux?

mingo — Thu, 02 Jun 2011 06:36:20 +0000

For a long time i shared your exact sentiment (i'd even place the 'pure luck' and 'good naming' proportion much higher than 40%), but then i had to admit that the phenomenal success of the "Git" project is somewhat of an outlier! :-)

So a proper, scalable development model (and, of course, good timing) matters a lot and can offset a bad name.

But yeah, i definitely agree that the "Freux" or "Freax" name would probably have been somewhat of an embarrassment with time, especially for desktop and business users.

(Just like dropping the cool minerals naming scheme (Beryl, Emerald, etc.) and going back to the dry and uninspiring "Compiz" name [for pure ego reasons!] was a disaster for that project, in terms of developer interest and growth.)

How much GNU is there in GNU/Linux?

nevets — Thu, 02 Jun 2011 03:01:06 +0000

The correct phrase is:

"How much GNU is there in Ubuntu"?

As someone else already mentioned, Ubuntu does not even mention that it is a Linux distribution (or Linux for that matter) in their web site.

How much GNU is there in GNU/Linux?

yarikoptic — Thu, 02 Jun 2011 02:23:34 +0000

GNU might vanish, Linux might get replaced, but even when dressed-up for a party it gets called Ubuntu, Debian will remain Debian! hooray!

GNU and Linux grew together and supported each other!

lxoliva — Thu, 02 Jun 2011 01:17:51 +0000

There is one pretty obvious flaw with the way the study that found a similarity was conducted: it disregards GNU packages that are in Debian and that haven't been repackaged by Ubuntu.

As for logos, the GNU tools I counted don't contain any AFAIK. Unlike Linux.

Now, if you want to compare a standard install, it would be just fair to apply the same standard to both sides, namely, discarding unusual drivers and architectures from Linux too. Or, if you want to claim they're important, then so are the less-than-common languages, architectures and operating systems supported by GCC, GDB and glibc, the architectures and object file formats supported by binutils, and all the options (or lack thereof) of Gnome.

Anyway, even if Linux *had* doubled in size in that time-frame and GNU hadn't grown at all, my very limited study with just a few GNU components showed they were 3 times larger that Linux, so these GNU tools alone would still be at least 50% larger than Linux.

How much GNU is there in GNU/Linux?

chad.netzer — Thu, 02 Jun 2011 01:05:50 +0000

Linux is just a cool name. I'd say roughly 40% of Linux's early success at attracting interested people was based on name alone. :)

How much GNU is there in GNU/Linux?

jmalcolm — Wed, 01 Jun 2011 22:43:20 +0000

You nailed it.

GNU and Linux grew together and supported each other!

dlang — Wed, 01 Jun 2011 20:01:13 +0000

note that the earlier comparison you point at is comparing the size of compressed source archives, while this one is comparing lines of source code. remember things like logos will not compress well (and will not count as many lines of code either)

they kernel hasn't quite doubled in that timeframe

2.6.21 is 10332702 lines
3.0-rc1 is 16700965 lines

but it's gown a lot. have the GNU tools been growing that fast?

in any case, unless someone points out a flaw in the tools (as opposed to just other studies that did other types of checks) I don't have reason to believe that the numbers presented in this article are wrong. the code to run this is available, so if you think there's a mistake in it, point it out.

given how large gcc and gdb are, I would be interested in seeing the results when run against the standard install rather than the entire repository

GNU and Linux grew together and supported each other!

lxoliva — Wed, 01 Jun 2011 19:11:36 +0000

http://fsfla.org/blogs/lxo/2007-05-21-gnu+linux has analysis that gets to very different results from those the cited report.

http://web.archive.org/web/20070928223909/http://www.slac... is another analysis that gets to a result that is even more disfavorable to Linux WRT GNU/Linux.

Now, considering that Linux didn't even double in size over the past 4 years, which would still put it ways behind the size of GNU back then, and that GNU also grew in this time-frame, I wonder how the report could possibly have arrived at a similar proportion of Linux and GNU.

But then, code size is not the whole story. Linus himself wrote, in the Linux 0.01 release announcement, that a kernel by itself gets you nowhere, and most of the tools used with linux are GNU software. So even he thought of linux as the kernel alone, and that GNU was essential to get a complete functional system.
http://www.kernel.org/pub/linux/kernel/Historic/old-versi...

This relationship could have changed like his opinion did, but it didn't: the combination of these two powerful developments fed each other and grew together and mostly inseparably, so neither party can reasonably claim credit alone.

That's why calling it GNU+Linux is just reasonable and fair. Calling it all Linux just feeds the misinformation for people who misperceive Linux as more than what it is. Of course those who seek this undeserved credit and want to spread misinformation will fight for it, finding other excuses to explain why it is fair to promote a small component of the whole over another equal-sized if not much bigger, older and just as historically-relevant component.

How much GNU is there in GNU/Linux?

Trelane — Wed, 01 Jun 2011 19:01:04 +0000

http://lwn.net/Articles/399370/

Discuss!