perf(shell): pixi shell is slow with fish

[TheSkyentist]

Checks

• [x] I have checked that this issue has not already been reported.

• [x] I have confirmed this bug exists on the latest version of pixi, using pixi --version.

Reproducible example

pixi info output:

System
------------
       Pixi version: 0.58.0
           Platform: osx-arm64
   Virtual packages: __unix=0=0
                   : __osx=15.6.1=0
                   : __archspec=1=m2
          Cache dir: /Users/user/Library/Caches/rattler/cache
       Auth storage: /Users/user/.rattler/credentials.json
   Config locations: /Users/user/.pixi/config.toml

Global
------------
            Bin dir: /Users/user/.pixi/bin
    Environment dir: /Users/user/.pixi/envs
       Manifest dir: /Users/user/.pixi/manifests/pixi-global.toml

Workspace
------------
               Name: new
            Version: 0.1.0
      Manifest file: /Users/user/Downloads/new/pixi.toml
       Last updated: 28-10-2025 17:24:03

Environments
------------
        Environment: default
           Features: default
           Channels: conda-forge
   Dependency count: 0
   Target platforms: osx-arm64
    Prefix location: /Users/user/Downloads/new/.pixi/envs/default

Issue description

When using fish (with an empty config), pixi shell takes a few seconds to spawn into a shell. By contrast, starting in zsh or bash the shell launches almost instantaneously.

Running pixi shell -vvv, allow debug commands print nearly immediately, with the last command being:

DEBUG pixi_cli::shell: Starting shell 'fish ' with source command: ' source "/var/folders/gq/kk0fb78d50vdr2kh_vjdml2syy_81s/T/pixi_env_2lP.fish"' which hangs until the shell spawns.

Expected behavior

It should start quickly, comparably with zsh or bash. Is there a setting that might ameliorate this? I've been using pixi with fish for a while now, and I believe it used to be faster in the past.

[Hofer-Julian]

Please provide the Pixi manifest where you experience the slow startup

[TheSkyentist]

This occurs with every pixi manifest, including a blank one:

[workspace] authors = ["TheSkyentist"] channels = ["conda-forge"] name = "new" platforms = ["osx-arm64"] version = "0.1.0"

[tasks]

[Hofer-Julian]

Huh, that is surprising. @tdejager tried it on his mac, and it was fast for him. You say pixi shell takes multiple seconds for you @TheSkyentist?

[TheSkyentist]

Yes, pixi shell takes about 2.5 seconds. @mfouesneau can confirm the same behavior on a similar M2 mac.

I've tried clearing/resetting all zsh and fish configs to no avail. @mfouesneau does not use fish and installed it for this test but seems the same behaviour.

In addition, simply running source /var/...etc....fish in a fish shell runs very fast, so I wonder if it is something after that, but I'm not an expert on how this works.

[tdejager]

Can you show me the output of: time fish --login -i -c exit. From outside of fish like from bash. And the same with --no-config?

[TheSkyentist]

From zsh:

fish --login -i -c exit 0.05s user 0.05s system 80% cpu 0.128 total

fish --login --no-config -i -c exit 0.01s user 0.01s system 73% cpu 0.033 total

From bash: (normal)

real 0m0.128s user 0m0.056s sys 0m0.050s

(no config)

real 0m0.035s user 0m0.011s sys 0m0.018s

From fish: (normal)

---

Executed in 104.94 millis fish external usr time 60.99 millis 0.31 millis 60.68 millis sys time 43.27 millis 2.10 millis 41.16 millis

(no config)

---

Executed in 27.80 millis fish external usr time 10.44 millis 0.30 millis 10.14 millis sys time 13.43 millis 1.87 millis 11.56 millis

Loading my config does take more time. However I have also tried replacing my fish config with a blank one and this does not speed up the time it takes to load pixi shell (at least noticeably).

[tdejager]

So strange.. Can you give the pixi shell timings? Do you have any setting in the pixi global configuration?

[tdejager]

Only other thing I can think of is using a sampling profiler for debugging.

[TheSkyentist]

I'm happy to run one! Do you have recommendations for which would be most helpful or might be best for interpreting what might be going on in this case?

[TheSkyentist]

Also, one more clue perhaps, pixi run ... runs without any slowdown.

[Hofer-Julian]

I just tried pixi shell with fish on my linux machine and I can confirm that it takes about two seconds for me, while bash is instantaneous.

[tdejager]

https://prefix.dev/channels/conda-forge/packages/samply is a good profiler to try out.
A pixi global install samply should work.

Some usage instructions for samly are here: https://github.com/mstange/samply
Might be that you need to use a debug version with pixi. To get some nice output. From the pixi project you can use pixi r install-as-debug pixi-dev or some other name, that should give you a globally installed version of a debug build. Or just use cargo if you have that.

[TheSkyentist]

Thanks @tdejager, I'm on vacation next week but will give this a try when I get back.

[tdejager]

Thanks for giving it a go, looking forward to what will turn up!

[Hofer-Julian]

@tdejager I don't think you want to use the debug install command still it has optimization at 1.

According to the docs, you want release profile with debug info enabled

[tdejager]

Also good!