BASICS

Getting a procs config going should be as easy as (on Debian):

apt install nim         # See also https://nim-lang.org/
nimble install procs    # Also installs $HOME/etc/procs conf
$HOME/.nimble/bin/procs # gives a help message

The Nim experience can sometimes have fairly rough-hewn edges, though. So far, though, something like the above has worked for me on Gentoo Linux, Debian, and Android Termux. procs only supports Linux /proc queries at the moment. If the above nimble does not work, you can maybe just:

git clone https://github.com/c-blake/cligen
git clone https://github.com/c-blake/procs
cd procs; nim c -d:release --path:../cligen procs
mv procs SOMEWHERE_IN_$PATH
cp -r configs/cb0 ~/.config/procs

Here are some screenshots with many related kernel threads merged (command used is shown in the output):

and with all kernel threads merged into one row:

Merging/coloring categories/kinds are all very user-defined. In the above, kernel threads are underlined and processes marked as runnable are bolded, and my terminal makes bold default foreground color render as orange.

GENERAL COMMENTS

This program is a mix of various procps/top/vmstat/pidof/pgrep/pkill features. It does environment-variable-driven themed display of process and system-wide metadata colorized by builtin process traits and also based upon user-defined process categories. It also supports "merging" or "rolling up" statistics for processes related to each other in user-defined ways, e.g. kernel threads or firefox processes. Conceptually, this is similar to what already happens with process statistics for a multi-threaded program, but the relationship between merged procs can be less reliant upon kernel categories.

Configuration is similar enough to https://github.com/c-blake/lc/ that they can share theme files via symlinks. Some ideas like username abbreviation and the kind/color systems carry over almost exactly. Unlike lc, procs is intended to also be a user-friendly API/library interface to process statistics/data. So, it can perhaps facilitate other new utility programs. Its only non-stdlib dependency is cligen.

Though written in Nim, not C, this API/multicommand is about as efficient or faster. procs tries hard not to make unnecessary system calls. E.g., with a format of just '%p %c' it will only open & read /proc/*/stat files.¹ Like lc, procs display is more of a "ps construction toolkit" than a pool of pre-packaged formats. Fancy configs can create more work/slow things down. Such is true with almost any featureful program. I have timed a basic process listing as taking about 57% the run-time of the C-based procps ps, though there are surely environments/configurations where ps can be faster.

ASAP mode

One feature more unique to procs display is its ASAP mode. For output styles with no sorting or merging, process rows are written to stdout as soon as the data is collected. This lowers user-perceived "latency to first output" by a very large multiple. That can help on a system that is struggling to make progress/schedule the procs display process any CPU time.

ASAP style of flow also applies to procs find --actions=kill (more briefly pk) , for example, to send signals as quickly as possible to misbehaving matching programs which may be grinding a system to a near halt. The pgrep/pkill in procps (at least as of version 3.3.15) read & select all processes before acting upon any. While hopefully rare, when ASAP action matters it can be very helpful.

Aliases

If you create hard-links or sym-links from the procs executable to any of { "pd", "pf", "pk", "pw", "scroll", "sc" }, then the multi-command can be bypassed and those commands activate, respectively, procs display, procs find, procs find -akill, and procs find -await. Being a cligen multi-command you can also type the shortest unique prefix for most things, e.g., procs f -ak.

Replacing "top" like functionality with pd and sc

procs display --delay 1 or more briefly pd -d1 aids a similar use case but different theory of operation to top -ib. procs is not an interactive program and has no compile/run-time curses/ncurses/terminal dependency. All coloring/merging ideas generally available in pd are used for a differential report. You can log to a file and look at a nicely embellished report later.

pd also allows user-defined sense of "idle". You can use traits besides CPU activity like RAM/IO activity, and even things independent of having been scheduled such as signal masks, nice value, etc. It does not print system-wide stats each iteration - that is what procs scrollsy (or sc) is for. top always felt "over bundled" to me.

This is kind of new/unusual/abstract. So, here is a screenshot (p=pd -sb with my configs/cb0 config) of GNU yes cruising along at 100 GB/s (no need for pv!). A relevant part of configs/cb0/style is -DJ><R -oDJ><R which diffs by cumulative jiffies, write, read & RAM and then sorts by the same. (You could reverse said sort order/etc. if you like..)

For system-wide statistics you can use procs scrollsy (aka sc). With default format in configs/cb0/config:[scrollsy], it makes output like this:

top is really just a combination of pd -d1 & sc with usually less history accessible via terminal scrollback. With two terminal windows you can get the same data (much more really) in a more open/re-analyzable format.

Actions on unrelated processes

wait/Wait actions of procs find (or pw) are more unusual functionality. The selected set of processes is checked for lack of existence (via a 0 signal) each delay separated interval. procs exits when either any or all (lower or uppercase) of the processes have failed to exist at least once. Up to fast PID recycling, this recreates features of the Bash wait/wait -n builtin for processes unrelated to the wait-er.

procs is definitely a work in progress, but a nice enough bundle of useful ideas to share. With so many features and just me as a user, there are surely many bugs.

Footnotes

1. Meanwhile, Linux procps ps (/bin/ps on most Linux) opens, reads, parses & closes both /proc/PID/stat and /proc/PID/status. This makes procs display aka pd roughly 2X faster. Using a /n -> /dev/null symlink, PROCS_CONFIG=/n tim 'pd -f%p\ %c>/n' '/bin/ps ax>/n' gives (2.8576 +- 0.0070)e-03 pd -f%p\ %c>/n & (5.0281 +- 0.0063)e-03 /bin/ps ax>/n. Adding process typology & highlights back (not using PROCS_CONFIG=/n) slows pd down to 3.9739 +- 0.0044 ms, still 1.27X faster than stock Linux ps. ↩