Compress cgroup names based on some heuristics

[BenBE]

This is some WIP for #216. The current implementation handles most common cgroup namespace patterns encountered for system/user/machine slices, shortens scope identifiers, trims service names and handles both LXC and systemd-nspawn (SNC) containers.

The code probably doesn't win a beauty contest yet, but should suffice as a preview.

Points to clean/optimize:

• Shorten pattern output boilerplate (there's much repetition for string writing)
• Clean up code flow for some of the more complex rules
• Extract some helpers to ease parsing / string handling
• Improve multi-label awareness to simplify rules relying on multiple tokens (like user slices, SNC)
• Allow length determination mode to check for required buffer size
• Include some more legacy rules
• Document important shortening rules

[fasterit]

Very quick review:

• s.buf is not always free'd
• N/A will never be output as the initial / (slash) remains in cgroup_short
• This works reasonably well and makes the column much more useful
• Consider making it a separate column (CCGROUP) or configurable or switching together with m (merge command line). Rationale: it is a heuristic and it will have error cases. So the user should have choice / a fallback option. Making it a separate column would allow the other CGROUP column to be also insanely wide so it becomes at least useful.
• Please add a section to the htop man file for it including the cases it handles (systemd cgroup hierarchy, systemd nspawn, LXC as of now)
• (for a v2) colors would be nice!

[BenBE]

Very quick review:

• s.buf is not always free'd

Can you give a rough hint which case you noticed?

Edit: Okay, I found one place (successful return branch in LinuxProcessList.c), any others?

• N/A will never be output as the initial / (slash) remains in cgroup_short

hmmm, N/A as in process->cgroup == NULL should be handled. if String_eq(process->cgroup, "/") then having / displayed is expected behavior IMHO.

• This works reasonably well and makes the column much more useful

This aligns with the initial feedback I got from @lucaswerkmeister who provided me with the test cases for systemd-nspawn …

In the final patch I'll probably go to have this column 40 chars wide, as even with this compression there's still quite a few processes where this ain't enough. Though with #854 on its way there could be another route …

• Consider making it a separate column (CCGROUP) or configurable or switching together with m (merge command line). Rationale: it is a heuristic and it will have error cases. So the user should have choice / a fallback option. Making it a separate column would allow the other CGROUP column to be also insanely wide so it becomes at least useful.

Sounds like a good idea. Switching this together with command line merging is probably not a good choice though, as those features are too unrelated. But with two columns the user can choose the way to go. That aside, there's one fix we have to do for the column display anyway (namely the .30 width spec I added).

• Please add a section to the htop man file for it including the cases it handles (systemd cgroup hierarchy, systemd nspawn, LXC as of now)

Will do. Proposals for wording always welcome. ;-)

• (for a v2) colors would be nice!

Actually this shouldn't be too hard to implement given I already have the output code provided as a callback. Thus basically adding in another parameter for the color for the callback function and replacing char* buf by some RichString … Will provide this as a separate PR though.