add cached eval

[casperdcl]

Attempts to cache runs of underlying commands so that e.g. #{gpu_bg_color}#{gpu_percentage} only calls nvidia-smi once per update.

• fixes update overhead/efficiency #27
• also fixes (maybe) Cygwin blink #5
• related status-interval not respected #15

Notes:

• runs cached commands a maximum of once every 2 seconds
• caches based on date +%s and basename command, ignoring command arguments. This could be a problem if running the same command with different arguments (not currently a problem)

details

While testing this, I think I've found an upstream problem (with tmux or tpm). Basically

set -g status-right "%H:%M:%S #(sleep 1)"
set -g status-interval 5

will ignore status-interval and result in one update per second (!) which appears to be #15.

1. status-interval not respected #15 (status-interval not respected) looks like an upstream bug
2. each update, commands may be run in parallel? e.g. set -g status-right #(write_hello_to_debug_file) #(write_bye_to_debug_file) won't guarantee hello is written before bye. This means that caching within an update may not work. Needs investigating... it's possible that some higher level code (outside this repo) handles caching of #(external commands)
3. oddly, before this PR, I get updates more than 1/s, but after it's about once every 5 sec, respecting status-interval. This seems to fix (maybe) status-interval not respected #15 but no there's not clear reason why... I suspect some higher level logic may be doing something like:

   commands = find_pattern("#(*)", status_right)
   while True:
       execution_time = exec(commands)
       if execution_time < 1:
           sleep(status_interval)

debugging

in ~/.tmux.conf:

set -g status-right '#{cpu_bg_color}#{cpu_percentage}#[bg=green][#{ram_bg_color}#{ram_percentage}#[bg=green]] #{gpu_bg_color}#{gpu_percentage}#[bg=green][#{gram_bg_color}#{gram_percentage}#[bg=green]]'

print to a debug file in ~/.tmux/plugins/tmux-cpu/debug.log:

diff --git a/scripts/helpers.sh b/scripts/helpers.sh
index df31cfb..d1e3e84 100644
--- a/scripts/helpers.sh
+++ b/scripts/helpers.sh
@@ -99,7 +99,9 @@ cached_eval(){
   local val="$(get_cache_val "$key")"
   if [ -z "$val" ]; then
     put_cache_val "$key" "$($command "${@:2}")"
+    echo miss "$key" >> ~/.tmux/plugins/tmux-cpu/debug.log
   else
     echo -n "$val"
+    echo hit "$key" >> ~/.tmux/plugins/tmux-cpu/debug.log
   fi
 }

• observation: will print miss nvidia-smi 4 times in quick succession each update
• expected: miss nvidia-smi once, then hit nvidia-smi 3 times

tail -f ~/.tmux/plugins/tmux-cpu/debug.log

[ctjhoa]

Sorry, I have very few times to work on my open source projects. I'll try to review your work tomorrow.

[ctjhoa]

1 comment.
It looks good to me otherwise

[ctjhoa]

chmod 0700 is not wanted on global system tmp directory

[casperdcl]

it's never global. get_tmp_dir always returns an EUID-specific subdir.

[casperdcl]

one of the main reasons I looked at tmux-copycat's implementation for reference is because they would've fixed any problems with it by now if it was wrong.

[ctjhoa]

From man tmux

tmux stores the server socket in a directory under TMUX_TMPDIR or /tmp if it is unset.

So either TMUX_TMPDIR or /tmp should exist

[ctjhoa]

Maybe that's why tmux-copycat does that if inside tmux env TMUX_TMPDIR turn into TMPDIR

[ctjhoa]

Sorry for the back and forth

[casperdcl]

maybe; though I don't really see much point in reverting - surely checking TMP & TEMP is fine?

[ctjhoa]

I'm ok with both solution