Systemd's Networkd Now Supports Bridging

[zanny]

And systemd is LGPL, not GPL. LGPL is fine for non-critical code but init is critical. If systemd wins, what can stop RedHat from adding proprietary "extensions" and suddenly making all of us who don't pay them second class citizens?

It is called forking. You could even fork upstart. Systemd could never go actual proprietary because the source is copyrighted under hundreds of contributors, and if they made any effort to do anything but the most popular move, half their hobbyist devs would flee. People write proprietary drivers for the linux kernel all the time anyway. Yes, I would have preferred to see systemd under pure GPL, but my understanding is lgpl is so other projects like the BSDs could fork it in the future if they wanted without license conflicts if their apis become so de facto other operating systems require them.

See: ffmpeg / libav, openoffice / libreoffice, mariadb / mysql.

[Teho]

Begone, troll. Kernel development is not controlled by corporations and hijacking isn't about how much code you contribute.

Eh? Only 13,6% of contributions to Linux kernel are done by unpaid developers. All the major maintainers work for companies or foundations that are directly funded by corporations (Linaro, Linux Foundation...). I'm still curious how systemd is supposedly "hijacking" anything.

And systemd is LGPL, not GPL. LGPL is fine for non-critical code but init is critical. If systemd wins, what can stop RedHat from adding proprietary "extensions" and suddenly making all of us who don't pay them second class citizens?

Even in this hypothetical scenario the "extensions" would need to be separate files and services from the ones that are under LGPLv2.1+ and everyone has the same right to do it as Red Hat.

[erendorn]

Monolithic?
Developed by corporations?
Originally a pet project of a single guy?
Incompatible with other competing open source project?
Compatible with proprietary modules?

The Linux kernel is dangerous. And it kills kitten!

[benalib]

what's the right definition of bloatware? it's systemd
it's bloated and become more bloated with more dependencies and too many unnecessary features with monolithic design
soon it will be larger than linux itself

[Teho]

what's the right definition of bloatware? it's systemd

systemd contains roughly 150k lines of code. It's written entirely in C (the XML is for man pages, "automake" is for bash and ZSH completition). It's split in to roughly 80 binaries each of which essentially do "one thing and one thing well". Everything but the systemd (init (PID1)), systemd-udev and systemd-journald is optional and things like PAM integration can be disabled. Each of the remaining modules (systemd-{hostanmed,timedated,localed,logind,networkd,socke t-proxyd...} can be disabled on both compile and run time. Most code exist outside of PID1 and thanks to sharing code between various ways of starting services (times, sockets, dbus, service files...) it's a lot simpler than using cronie, xined, atd... in pararell.

In comparison Linux is a monolithic kernel with 17 million lines of code that run in ring 0.

According to Ohloh.net the rsyslog syslog implementation contains 137k lines of code in C.

it's bloated and become more bloated with more dependencies and too many unnecessary features with monolithic design

According to the systemd README it has huge list of five dependencies: dbus, libcap, fsck, agetty and glibc, all of which you would probably have installed anyway. Here's a breakdown of systemd's depdencies. It's good to note that systemd 209 will drop dependency on libdbus and it will drop the dbus dependency with kdbus sometime next year.

soon it will be larger than linux itself

...yeah, no.

[tomegun]

What would be considered basic and what would be considered advanced? If it works as a replacement to the interfaces file, it is enough for me.

At first, we want to focus on stuff that has a more-or-less static configuration. So if each of your interfaces always connect to only one network (i.e., your machine is not a phone/laptop), we probably want to support it.

At first we target static IP addresses and bridging, next step will be DHCPv4. After that we will see depending on what people decide to work on, but DHCPv6, IPv4LL, DNAv4, teaming, vlan, ... are the kinds of things I'm interested in.

[tomegun]

It is called forking. You could even fork upstart. Systemd could never go actual proprietary because the source is copyrighted under hundreds of contributors, and if they made any effort to do anything but the most popular move, half their hobbyist devs would flee. People write proprietary drivers for the linux kernel all the time anyway. Yes, I would have preferred to see systemd under pure GPL, but my understanding is lgpl is so other projects like the BSDs could fork it in the future if they wanted without license conflicts if their apis become so de facto other operating systems require them.

See: ffmpeg / libav, openoffice / libreoffice, mariadb / mysql.

systemd is (mostly) LGPL because we provide some libs, and we want to be able to move code around freely. The libraries we provide are required to be LGPL as we want to allow people to link against them even from proprietary apps.

[intellivision]

I can tell you all now that once networkd gets a stable release, development will mysteriously stop on Conman, at least outside the scope of systemd.
That's what I don't like about its design, with every component integrated, the original projects from where they are merged from are halted, since it's in systemd, and then conveniently Linux only.

[Guest]

In comparison Linux is a monolithic kernel with 17 million lines of code that run in ring 0.

I didn't know I run entire Linux kernel on my PC. Does it mean I'm using AMD driver instead of NVIDIA sometimes?

[Pajn]

It's split in to roughly 80 binaries each of which essentially do "one thing and one thing well". Everything but the systemd (init (PID1)), systemd-udev and systemd-journald is optional and things like PAM integration can be disabled. Each of the remaining modules (systemd-{hostanmed,timedated,localed,logind,networkd,socke t-proxyd...} can be disabled on both compile and run time. Most code exist outside of PID1 and thanks to sharing code between various ways of starting services (times, sockets, dbus, service files...) it's a lot simpler than using cronie, xined, atd... in pararell.
[/QUOTE]
And I can take journald and run it with say, Upstart?
or udev and run it with OpenRC?

No, I can't. And that's what makes it monolithic.
I don't care how much it clutters my file system if I still can't use the pieces separately.