`w`: Shows incorrect value in the `WHAT` column

[alxndrv]

There is a discrepancy in the information output by this implementation of w and the reference implementation:

$ cargo run -q w  
 21:22:38 up  1:06,  1 user,  load average: 0.09, 0.27, 0.58
USER     TTY       LOGIN@   IDLE   JCPU   PCPU  WHAT
alx      tty1      20:16    1:06m  0.00   0     /bin/sh/usr/bin/niri-session-l

$ w
 21:22:38 up  1:06,  1 user,  load average: 0.09, 0.26, 0.58
USER     TTY       LOGIN@   IDLE   JCPU   PCPU  WHAT
alx      tty1      20:16    1:06m  0.00s   ?    systemctl --user --wait start niri.service

1. The value of WHAT seems to be stripping away all whitespace from the command it's trying to display.
   /bin/sh/usr/bin/niri-session-l is clearly supposed to be this shell script invocation: /bin/sh /usr/bin/niri-session -l

2. The actual content of the column refers to two slightly different steps in the invocation chain.
   In this specific case /usr/bin/niri-session is a bash script which is invoked by the display manager at login-time.
   systemctl --user --wait start niri.service is something that is called from inside this bash script.

Looking at these processes with ps, I can confirm that both are still running, and the /bin/sh invocation is indeed higher up in the invocation chain (it has a lower PID).
Seeing how there's two different processes being shown in the WHAT, I'd also assume that the PCPU values would also have a discrepancy if they anything else than zero 🙂

[alxndrv]

The whitespace issue seems to stem directly from how the kernel exposes the value in /proc/<pid>/cmdline:

$ cat /proc/1161/cmdline 
/bin/sh/usr/bin/niri-session-l% 

$ xxd /proc/1161/cmdline
00000000: 2f62 696e 2f73 6800 2f75 7372 2f62 696e  /bin/sh./usr/bin
00000010: 2f6e 6972 692d 7365 7373 696f 6e00 2d6c  /niri-session.-l
00000020: 00                                       .

Simply reading it as a string makes the whitespace disappear but the binary shows that there's actually a NULL byte in place of every whitespace which can be parsed.

[alxndrv]

The session process being wrong seems to share the root cause with #183. Both are seemingly caused by the fact that the uutils implementation has no systemd awareness, where as procps-ng includes explicit checks for various init systems.

As for the whitespace issue, opened a PR for it: #666.
Feel free to close this one if you feel that it's not sufficiently different from #183.