Support macOS guests

[AkihiroSuda]

Usage:

limactl create --video template:macos
limactl start macos
limactl shell macos

The password prompt is shown during creating an instance, so as to run chown root:wheel ~/.lima/_mnt/0/Library/LaunchDaemons/..., which is required for the lima-macos-init launch daemon to run.

The password for GUI login is randomly generated and stored in /Users/${USER}.guest/password in the VM.

Fix #3618

TODOs (in follow-up PRs):

• Docs
• Switch back the default shell to zsh
• Graceful limactl stop (currently it has to be stopped from the guest)
• Propagate additional cloud-init data
  • Hostname
  • Timezone
  • DNS
  • CA Certificates
• Support non-plain mode
  • mounts
  • port forwards
• Let limactl sudoers generate the entry for chowning files
  (chown root:wheel ~/.lima/_mnt/0/Library/LaunchDaemons/...)
• Support headless mode (--video=false)
• Support macOS in hack/update-template.sh

---

Depends on:

• Introduce stable disk/iso filenames with symlink-based migration #4587
• Add pkg/plist for parsing XML-formatted plist documents #4601

[jandubois]

I don't have the energy to do a proper review today, but here are some things I noticed. Only the "boot" thing is a problem.

Another suggestion: maybe add a helper for inst.Config.OS != nil && *inst.Config.OS == limatype.DARWIN because that is used a lot. On second thought, is the nil check necessary? Because FillDefault guarantees that it isn't nil. We don't do nil checks for other fields in the driver code (or shouldn't have to).

[jandubois]

On second thought, is the nil check necessary?

I've wondered before why we use pointers for so many of the fields in LimaYAML.

For some, especially booleans, we want to be able to distinguish if the value was specified as the null value (i.e. false, 0, ""), or not provided at all. But for many values the null value is not a valid choice, so e.g. the OS could just be OS instead of *OS.

Or am I missing something?

[AkihiroSuda]

On second thought, is the nil check necessary? Because FillDefault guarantees that it isn't nil. We don't do nil checks for other fields in the driver code (or shouldn't have to).

Removed the nil check

I've wondered before why we use pointers for so many of the fields in LimaYAML.

Probably we wanted to treat all the types in the consistent manner.
That may help simplifying FillDefault with reflection in future?

[AkihiroSuda]

Split plist parser to:

• Add pkg/plist for parsing XML-formatted plist documents #4601