On Aug 10, 2009, at 9:56 AM, Sidney San Martín wrote:

I can't find any documentation on the launchd checkin process. Using SampleD, launch.h, and other people's documentation of their discoveries it's possible to put together a working daemon, but with no guarantee that you're doing everything right. (Even though launchd is open source, its APIs need documentation.)

The SampleD project is fine. Its purpose is to illustrate how to check in.

launchd.plist(5) is the first place I even saw the check-in process mentioned: "The job must check-in to get a copy of the file descriptors using APIs outlined in launch(3)". launch(3) doesn't exist (<rdar://problem/7130390>). It seems like the following, at least, should be documented:

It won't be. We're going to deprecate that API. Just use the SampleD project. The comments are reliable as documentation of what's going on.

1) What check-in is and when it is necessary

If you don't know, you probably don't need it. :) Only use this API if your job advertises a socket.

2) The consequences of a job not checking in when launchd expects it

The only consequences are that the job won't receive the file descriptors for the sockets whose connection caused it to launch on- demand. But if your job isn't launch-on-demand, you don't need to worry.

3) What keys in the response dictionary are filled in at runtime

See SampleD.

4) How to memory manage launch data returned by check-in functions (i.e. what needs to be released)

See SampleD.

5) Error conditions and how to handle them

See SampleD.

What do you think? Has this come up before? Is there some secret store of documentation I'm missing? Is this API so intuitive that documentation is redundant?

No. We don't like that API very much and plan on getting rid of it and replacing it with something that is intuitive. So you're not really missing anything. -- Damien Sorresso BSD Engineering Apple Inc.