Xenomai rootless and fixes

[hdiethelm]

This PR has 3 commits:

• Xenomai EVL: Abort on evl_attach_self() failed: Otherwhise, there where 1000's of error messages/s if something went wrong
• Xenomai EVL: Remove uneeded debug print
• Rootless for Xenomai and Xenomai EVL: As proposed in Rootless uspace preempt-rt (rebase of #918, fixes for #3928) #3964

This needs libevl >= 59-1 from https://github.com/hdiethelm/xenomai4-linuxcnc where the udev rules are installed. Of course, this can also be done manually. The rules are here: https://github.com/hdiethelm/xenomai4-linuxcnc/blob/lfs/libevl-debian/udev/90-libevl.rules

Tested and works both with Xenomai3 / Xenomai4 EVL.

[BsAtHome]

Why is the return type not bool?

[hdiethelm]

is_current_user_in_gid() has -1 on failure, 0 on not member, 1 on member.

It's a bit a mess in rtapi_app_main.cc with the return values. Sometimes 1 is pass, sometimes 0 and sometimes it is bool.

The detect_* functions return 1 on success. For example handle_command() and all commands have -1 on failure, 0 on success.

It would probably make sense to change all detect_* and related functions to bool?

[BsAtHome]

But you are using it like:

bool isInGid = is_current_user_in_gid(gid) > 0;

If this is supposed to distinguish between "user in gid", "user not in gid" or "error condition", then you should actually use the return value to distinguish the cases. However, you add this function and convert its return value to bool. You are not doing anything to distinguish error from not and treat errors like "user not in gid". That is fine by me... But then, my question stands: why is the return value not bool?

[hdiethelm]

f5316b6 changes all return values to bool in the detect_* code. I think this makes more sense.

is_current_user_in_gid() should never fail so the perror() handling should be enough. According to the man page, a wrong address or wrong size is the only reason and this is verified before. So out of memory is the only remaining reason.