Open Launchd | Hacker News Reader

Open Launchd

github.com

65 pointsbpierre27 comments

Is this related to the recent "OSX 10.10 OpenSource" post? Someone did claim that Apple stopped posting the source code for launchd...

I don't think so, no. A launchd port to FreeBSD is a project that has been hanging around and attempted multiple times by various parties with differing success (I linked to a working, though early launchd_xml port below).

This one in particular is the openlaunchd port first started by R. Tyler back in 2005 when he was a GSoC student, but was abandoned and restarted 8 years later in 2013, announced here: http://unethicalblogger.com/2013/12/03/scratchiest-neckbeard...

    [...] FreeBSD unlocks some hacking that I've wanted to
    do, such as:

    Restarting my work with launchd on top of FreeBSD as a
    viable alternative to rc.d (aside: systemd which many
    Linux hackers are familiar with can be considered a
    poor impersonation of launchd, which is, in my opinion,
    a pretty stellar job management daemon)

That said, I don't think openlaunchd is entirely dead yet. I think a launchd port to FreeBSD is inevitable, whether it's this, launchd_xml or some other one (I'm fairly certain there's some non-public ports), but as a whole it's been a protracted ordeal.

Being that I am a Mac guy, I always sort of wondered about launchd as a reasonable middle ground between the godawful current init mess and the looming all-encompassing Stygian Lovecraftian systemd mess.

Apple originally wanted other Unix-likes to consider launchd. Ubuntu was considering it and I think even had it briefly as an option, but ultimately they decided to go with Upstart. This was mostly because the original license terms for launchd were deemed too restrictive (Apple Open Source License had some extra terms that were somewhat unpopular). Apple changed the license terms a year or two later, but it was too late.

I would love to see this on BSD though.

link to the project homepage: https://wiki.freebsd.org/launchd

Not really an apple user, can anyone tell me what the benefit to porting this is? Is it much better than rc.d or sysvinit?

Faster boots, on-demand running of rc.d scripts, &c.

Yes. Launchd is arguably the first modern unix init system, with systemd and upstart being it's conceptual children.

It eschews shell scripts in favor of non-turing complete launch configs, manages process lifetimes in addition to just starting them and does bootup in parallel using pre-created sockets for synchronization.

It is quite barebones compared to systemd, but fundamentally it uses similar approaches to fulfill the same goals.

Solaris SMF came out at around the same time as launchd, which looks a bit more like systemd. I don't think there is one single origin to these systems.

Was there any work done to replace XML for configuration? You may call it superficial, but I'd make that a priority over pretty much everything else. Many people won't get past that bullet point.

Last commit made 5 months ago, and no updates on the project anywhere. Seems abandoned.

The FreeBSD wiki page[0] has an updates page, with this, from a month and a half ago:

> 20140910: Outback Dingo has made available an older port of pfSense's launchd_xml.

[0] https://wiki.freebsd.org/launchd

And a link to the repository: https://github.com/outbackdingo/launchd_xml

Note that the pfsense / outbackdingo launchd did something rather naughty -- it put the xml parser back in pid 1 which was intentionally not there by design at Apple.

tl;dr even if you could convince the FreeBSD devs that launchd is a great idea they wouldn't accept this as-is.

doesn't launchd on osx also parse plist xml files? How are those handled?

OS X has a binary plist format, too. Maybe it uses that to bootstrap and does xml parsing in sub-processes.

There's another binary that does the processing and the data is sent to the launchd daemon over a Mach-specific IPC channel that FreeBSD doesn't have.

The launchd_xml code is mildly toxic for openlaunchd, it's derived off of the Apple Public Source License codebase instead of the newer Apache 2.0 licensed codebase that openlaunchd derives from.

IMO the ASL is going to be required to get FreeBSD to adopt openlaunchd at some point

Not quite abandoned, more a question of time. The upstream code is sufficiently undocumented and without tests that in some cases my approach has been to #ifdef stuff out until I can at least get some units compiling.

A test suite for this code would have made this whole effort far easier :(

I wonder if this could be used as an alternative to systemd?

It does most of the same things - the objections to systemd around covering multiple overlapping areas in a single project would also apply to launchd (and the Solaris SMF).

The best way to make cron / at / sysvinit / syslog / daemonize / xinetd / supervisord fans happy is to let them keep using cron / at / sysvinit / syslog / daemonize / xinetd / supervisord.

The biggest difference is that Apple hasn't demonstrated hostility/irrationality/utter dickery at people who report bugs on their tracker. The same cannot be said of systemd.

Is there even an open bug tracker for launchd?

Not really "open", it's write only.

http://bugreport.apple.com

Thanks, I tried searching for a launchd bug tracker but didn't find this. I'm still not sure if this is a fair comparison as we can't actually see how apple has responded to similar concerns (if any).

Fact is most people don't care about the init system as long as it works.

Even still, launchd covers far less ground than systemd, all things considered. It's just inetd, crond, atd and watchdogd, mostly.

And initd itself. That's five things, and Apple care less about what you have to say re: 'Unix philosophy' that LP does.

Reply on news.ycombinator.com