Debugging performance issues in zsh
03 Apr 2023If I add too much code in my .zshrc
, my zsh takes longer to load. As I use a lot of different terminal windows (splitting one main tmux window), I want those split to happen quickly. This is why I'm trying very hard to keep the loading time of zsh under 150ms.
Hyperfine
One way to evaluate the current loading speed and track potential regressions and improvement is to use hyperfine
hyperfine --warmup 3 "zsh -i -c exit"
This will benchmark how long it will take on average to load zsh. The -i
makes it load in interactive mode (so, by sourcing ~/.zshrc
), and -c exit
makes it execute the exit command.
Note that if you have any commands running asynchronously in the background (like prompt optimization), they will purposefuly not be included in the time.
zprof
The other debugging tool is to use zprof
, which is the zsh profiler.
Include zmodload zsh/zprof
at the top of your ~/.zshrc
file, and zprof
at the bottom. Next time you'll open zsh, you'll see a table like this:
num calls time self name
-----------------------------------------------------------------------------------
1) 1 240,22 240,22 65,76% 240,22 240,22 65,76% oroshi_tools_pyenv
2) 1 72,98 72,98 19,98% 72,98 72,98 19,98% oroshi_theming_index
3) 1 14,77 14,77 4,04% 14,77 14,77 4,04% compinit
4) 1 14,12 14,12 3,86% 14,12 14,12 3,86% oroshi-completion-styling
5) 1 5,85 5,85 1,60% 5,85 5,85 1,60% oroshi_tools_fzf
6) 1 4,95 4,95 1,35% 4,95 4,95 1,35% _zsh_highlight_bind_widgets
7) 1 4,56 4,56 1,25% 3,01 3,01 0,82% oroshi_tools_z
8) 1 2,27 2,27 0,62% 2,27 2,27 0,62% _zsh_highlight_load_highlighters
9) 2 2,17 1,09 0,60% 2,17 1,09 0,60% promptinit
10) 1 2,10 2,10 0,58% 2,10 2,10 0,58% oroshi_tools_ls
11) 7 1,51 0,22 0,41% 1,51 0,22 0,41% add-zsh-hook
12) 7 0,64 0,09 0,17% 0,64 0,09 0,17% compdef
13) 2 0,63 0,31 0,17% 0,63 0,31 0,17% is-at-least
14) 1 14,85 14,85 4,07% 0,09 0,09 0,02% oroshi_completion_compinit
This shows where most of the time is spend, the number of times a specific function is called, and if you scroll down you'll see details of the stacktrace of each sub command.
In this example, it is clear that my oroshi_tools_pyenv
method is too slow, and I need to optimize it.
Note that this only tracks the time to run functions, not the time to source files, so if you need to benchmark a specific sourced file, wrap it in a function that you invoke immediatly.
Want to add something ? Feel free to get in touch on Twitter : @pixelastic