How I went from 1.2s to 0.15s shell startup by lazy-loading conda
Recently I got a bit annoyed, sometimes it took ages to start up a new terminal on my macbook pro. Pretty annoying, sitting there waiting for something as basic as a terminal to startup.
My Setup
- macOS
- Ghostty terminal
- oh-my-zsh with powerlevel10k
- Only plugin:
git
Pretty basic, right? Should be fast enough.
Profiling
So the first thing I went ahead with, was some profiling. Pretty standard when you are working with performance issues, but to be honest, I've never had to profile my shell before.
First, I got a rough number:
$ time zsh -i -c exit
zsh -i -c exit 0.58s user 0.26s system 69% cpu 1.219 total1.2 seconds. Well that seems faster than what I was experiencing, but still; that's an insane amount for a shell.
I then generated a breakdown by adding zsh's built-in profiler to ~/.zshrc:
# add to the very top of .zshrc
zmodload zsh/zprof
# add to the very bottom of .zshrc
zprofRestart the shell and you get a function-level breakdown:
num calls time self name
-------------------------------------------------------------------
1) 1 539.93 87.63% 539.90 87.63% __conda_activate
2) 22 36.39 5.91% 31.56 5.12% _omz_source
3) 2 17.18 2.79% 17.18 2.79% compaudit
4) 1 10.93 1.77% 5.77 0.94% powerlevel10k
5) 1 22.90 3.72% 5.71 0.93% compinitConda was the culprit; 540ms, 87.6% of total startup. Turns out it runs conda shell.zsh hook on every shell start, even if you haven't touched conda in weeks... and I just use it for some AI work.
Applying some fixes
1. Lazy-load conda
Instead of running conda init on every shell startup, I wrap it so it only initializes when you actually run conda or activate:
# Lazy conda - only init when needed
lazy_conda() {
unfunction conda activate 2>/dev/null
__conda_setup="$('/opt/homebrew/Caskroom/miniconda/base/bin/conda' 'shell.zsh' 'hook' 2> /dev/null)"
if [ $? -eq 0 ]; then
eval "$__conda_setup"
else
if [ -f "/opt/homebrew/Caskroom/miniconda/base/etc/profile.d/conda.sh" ]; then
. "/opt/homebrew/Caskroom/miniconda/base/etc/profile.d/conda.sh"
else
export PATH="/opt/homebrew/Caskroom/miniconda/base/bin:$PATH"
fi
fi
unset __conda_setup
}
conda() { lazy_conda && conda "$@" }
activate() { lazy_conda && conda activate "$@" }This saved around ~540ms, and really solved all my issues. A tool I used a few times, was dragging down my normal workflow. And with lazy loading it still works, so nothing lost, but gained A LOT!
2. Lazy-load bun completions
Bun completions were sourced eagerly on every shell start. Wrapped them with compdef so they only load when you actually tab-complete a bun command.
_lazy_bun_completion() {
unfunction _lazy_bun_completion 2>/dev/null
[ -s "$HOME/.bun/_bun" ] && source "$HOME/.bun/_bun"
}
compdef _lazy_bun_completion bun3. Housekeeping
Consolidated 6 separate export PATH=... lines (including a duplicate $HOME/.local/bin) into a single declaration at the top. This happens when tools add their own PATH exports, you might not want to do this.
Also removed ~50 lines of commented-out oh-my-zsh template comments. Doesn't affect performance, but makes the config actually readable.
Results
So my final profiling showed ~0.15s might seem bad, but seems instant for me, and...
1.2s → 0.15s, that's 8x faster.
The entire win came from one fix; lazy-loading conda. The rest (bun completions, PATH consolidation, boilerplate cleanup) were just some clean up, while I was touching this file. Only the bun one might have some effect.
| Component | Before | After |
|---|---|---|
| Conda | 540ms | 0ms (deferred) |
| oh-my-zsh | 36ms | 32ms |
| compinit | 23ms | 15ms |
| powerlevel10k | 11ms | 17ms |
| Other | 590ms | 86ms |
| Total | 1.2s | 0.15s |
I expected to dig deep to fix some issue but luckily it was just one very slow thing, that I use rarely.
In the future I might dig deeper in optimizing my shell, but for now it works again. The "Other" component is also mostly a mystery, but zprof only tracks the top level functions; I have no need to dig deeper now, maybe you can enlighten me about optimizing my shell in the comments!
Thanks for reading.