glibc VS illumos-gate

I am writing a short paper for school and the topic is Fast Inverse Square Root and alternatives. One of the questions is how the sqrt-function is implemented in glibc. Here is the code of that function.

The code isn't the easiest to read but in glibc it seems that the syscall is only performed if waiter are detected in userspace during an unlock operation

https://github.com/lattera/glibc/blob/master/nptl/pthread_mu...

You might still not realize, but floats are a large topic. Have a look at eg. this implementation here: https://github.com/lattera/glibc/blob/master/stdlib/strtod_l.c

This is a nice write up, thank you. However, I'm stilling interpreting this as a "fun trick" rather than the common sense method for solving the problem. For example, looking at the source code for htonl() from glibc: https://github.com/lattera/glibc/blob/master/inet/htonl.c

Second of all: The horrifying truth is that there is no such thing as a canonical text representation for IPv4 (source) (and yes... I am indeed citing the failed attempt to standardize this as my source for it not being standardized). Authoritatively speaking, all possible (non-binary) representations are equally invalid. In fact, text address resolution is typically delegated to the OS kernel, so what constitutes a "usable" address is liable to vary depending on if you're using Linux, OSX, Windows, or Other.

Optimized code gets really weird. The creators of strlen, for example, decided that iterating over each character to find the end was far too slow. So instead, they convert the character pointer into an int pointer and do bitwise manipulation with the int on two different magic numbers so they can check four/eight characters at once.

If you're using a cross-platform C compiler like gcc or clang, you're usually expected to switch to assembly. Here's the syscall instruction in glibc

Maybe you should include the math part of a libc statically with your code. glibc is one option, or dietlibc if you want it to be as small as possible.

It may become a footnote on Linux, but Linux isn't the only system out there -- and DTrace remains alive and well in many systems (not least in its reference implementation in illumos[0]).

[0] https://github.com/illumos/illumos-gate

Nobody's paid to have it pass Open Group Unix Branding certification tests

https://www.opengroup.org/openbrand/register/

so it can't use the UNIX™ trade mark.

But it's got the AT&T Unix kernel & userland sources contained in it.

PDP-11 Unix System III: https://www.tuhs.org/cgi-bin/utree.pl?file=SysIII/usr/src/ut...

IllumOS: https://github.com/illumos/illumos-gate/blob/b8169dedfa435c0...

> it seems like this bug might actually date back to the very beginning of ZFS with Sun

Looks like you might be right about that. The oldest commit referenced in the fix [0] was from 2006[1], which was just months after Sun released ZFS.

[0] https://github.com/openzfs/zfs/pull/15571

[1] https://github.com/illumos/illumos-gate/commit/c543ec060d

It could be good for in memory stores / log-structured merged trees / other data store applications although it isn't used much now-days. I find them simpler to implement and understand than red-black trees -- although that's a matter of taste I suppose. They beat red-black trees in read-heavy loads (i.e. writes / updates are more costly for AVL trees than for Red-Black trees although they beat R-B trees for read-heavy loads). You can find another implementation in Illumos (an open source Unix operating system) available here: https://github.com/illumos/illumos-gate/blob/master/usr/src/...

It's high time that the Illumos developers patched https://github.com/illumos/illumos-gate/blob/master/usr/src/... and https://github.com/illumos/illumos-gate/blob/master/usr/src/... to just set the error flag and return, and made the #ifndef TIOCSTI path in https://github.com/illumos/illumos-gate/blob/master/usr/src/... the only path.

Because by the looks of https://github.com/illumos/illumos-gate/blob/7c478bd95313f5f... the C shell was fixed years ago.

People are still actively working on Illumos. The last change was yesterday morning.

* https://illumos.org

People are still actively working on MirBSD. There's a CVS commit account that can be followed on the FediVerse.

* http://www.mirbsd.org

It's DragonFly BSD, not Dragon BSD, and the irony of that is that you missed FreeBSD, which is of course still going.

* https://dragonflybsd.org

* https://freebsd.org

As is GhostBSD, which tracks FreeBSD.

* https://ghostbsd.org

HardenedBSD is still going. Shawn Webb regularly talks about it on the FediVerse.

* https://hardenedbsd.org

Its successor is still out there: Illumos. Though it seems to be mainly focused on backwards compatibility for existing custom applications as it still enforces things like an 8 character username limit.

> Since then, illumos has rewritten all those components

But apparently kept the same license?

>> Most of the existing code is licensed under the CDDL and we expect new code will generally be under this license as well.[0]

[0] https://github.com/illumos/illumos-gate