Weekly edition	Kernel	Security	Distributions	Contact Us	Search
Archives	Calendar	Subscribe	Write for LWN	LWN.net FAQ	Sponsors

I'll explain:

Posted Oct 24, 2006 17:59 UTC (Tue) by tetromino (subscriber, #33846)
In reply to: I'll explain: by hummassa
Parent article: GPLv3: What the Hackers Said (Linux Journal)

You know, if you load all the sources for a typical Debian/Apache web server, you'll get much more code from GNU than from Linux, hell you'll even get more code from Apache than from Linux, so why should you call it a "Linux server"??

Check your math. The 2.6.18 kernel has 229MB of source code (uncompressed). The GNU userland (i.e. glibc-2.5 and coreutils-6.3) has 125MB of uncomporessed source total. And Apache 2.0.59 is only 27MB...

Now, to be fair, gcc is ~256MB. However, the standard security practice in most organizations is to never install any compilers or development tools on production servers.

(Log in to post comments)

I'll explain:

Posted Oct 24, 2006 18:50 UTC (Tue) by oak (guest, #2786) [Link]

> Now, to be fair, gcc is ~256MB. However, the standard security practice
> in most organizations is to never install any compilers or development
> tools on production servers.

I've used Linux on my workstation >10 years. Workstation without a
toolchain is a toy, somewhat similar to default Windows installation.
For development work Gdb is pretty important and quite huge too...

I'll explain:

Posted Oct 24, 2006 19:00 UTC (Tue) by tetromino (subscriber, #33846) [Link]

Of course you need gcc on workstations and development machines. However, most sysadmins I've met are a bit paranoid about installing a compiler on a production server.

I'll explain:

Posted Oct 25, 2006 0:00 UTC (Wed) by emkey (guest, #144) [Link]

I've been a sysadmin for the better part of twenty years and I've never understood that particular foible. It is trivial to ftp, scp, etc over a binary once you have local access. And far less obvious potentially.

I'll explain:

Posted Oct 24, 2006 19:00 UTC (Tue) by stijn (subscriber, #570) [Link]

To be fully fair, the Linux kernel is designed to be compiled by gcc. This seems slightly more
relevant to me than the fact that you can pre-compile the kernel or remove gcc later on. The kernel
is intrinsically tied to gcc. So better include it in the balance after all.

I'll explain:

Posted Oct 24, 2006 21:02 UTC (Tue) by i3839 (subscriber, #31386) [Link]

If the kernel is intrinsically tied to gcc, then how come it can be compiled by some other compiler, like say the Intel one? Linux uses some gcc C extensions, but those can and often are also implemented by other compilers.

I'll explain:

Posted Oct 24, 2006 21:47 UTC (Tue) by stijn (subscriber, #570) [Link]

On how many architectures does the intel compiler work?

I'll explain:

Posted Oct 24, 2006 21:54 UTC (Tue) by k8to (subscriber, #15413) [Link]

How many nits can dance on the head of a pin?

This comment is not directed at anyone in particular, it is omnidirectional in this discussion.

I'll explain:

Posted Oct 24, 2006 22:36 UTC (Tue) by stijn (subscriber, #570) [Link]

The point I was making is that in the portability of Linux to many architectures gcc is a crucial
factor. It is all well to say that Linux can in principle be compiled by any suitable compiler and that
icc can compile Linux. The statement I made is that linux is intrinsically tied to gcc in the light of
Linux's famed portability. I thought that from the theme of this subthread the wording was clear
enough, if perhaps a little too implicit.

I'll explain:

Posted Oct 24, 2006 20:37 UTC (Tue) by nix (subscriber, #2304) [Link]

I'd be simply fascinated to hear how your servers work without libgcc,
libstdc++, et al (approx. 2.5 million lines as of GCC 4.1, although
runtime libraries for Java, Objective C, Ada et al account for about 1.5
million of those lines).

Any way you slice it that's a significant hunk of FSF-copyrighted code.

I'll explain:

Posted Oct 24, 2006 21:19 UTC (Tue) by tetromino (subscriber, #33846) [Link]

Actually, libstdc++-4.1.1 is only 24MB of uncompressed source code. The vast majority of the 256MB gcc source is the compiler itself, the java runtime, fortran, and ada junk, i.e. stuff that you probably wouldn't have installed on a server (unless you are using gcj in your webapps, of course).

I'll explain:

Posted Oct 25, 2006 7:16 UTC (Wed) by khim (subscriber, #9252) [Link]

And linux's 229MB contain 90% of junk not needed for your server as well (drivers for the things you'll never see installed on that server, filesystems not seen in the wild for years, etc).

I'll explain:

Posted Oct 25, 2006 8:13 UTC (Wed) by bronson (subscriber, #4806) [Link]

That's pretty much true for software in general. What do you suppose the percentage junk here is? Or here?

I'll explain:

Posted Oct 25, 2006 16:09 UTC (Wed) by nix (subscriber, #2304) [Link]

That's what my line counts corresponded to. I didn't just du the entire GCC source tree :) that would obviously give ridiculous results.

I'll explain:

Posted Oct 27, 2006 20:44 UTC (Fri) by cdmiller (subscriber, #2813) [Link]

"Now, to be fair, gcc is ~256MB. However, the standard security practice in most organizations is to never install any compilers or development tools on production servers."

If your server runs perl, php, whatever else, you have development tools installed, and usually you have gcc as well to install extensions to those tools. It's all moot anyhow, as the linux kernel with no userland is userless....