MacOS Silicon M1 Chip - Installing Napari

[chrislearns]

Let me begin by saying I apologize if this is the wrong place for this. If there is a better place please let me know and I will move it.

I spent way too long trying to get napari to run on my 2021 M1 MacBook Pro, so I wanted to save the next poor soul the struggle. I have reproducible, successful steps below. Note, I am by no means an expert in this realm and all my software dev experience is in Swift/the Apple ecosystem. If anyone has suggestions for a better way to do this, please (a) comment below, (b) make sure it is reproducible, and (c) please explain your steps and any sources you have as I have done. I'm sure any good dev in this field may understand what you may type, but this may be a good resource for some less tech-savvy individuals.

MY STEPS

• Step 1 - Install Anacaonda:

  1. https://www.anaconda.com/download
  2. Download the M1/M2 Distribution
  3. Install the .pkg (when I did this, I was personally running Anaconda3-2023.03-1-MacOSX-arm64.pkg)

• Step 2 - Setup x86 env: Source
  NOTE: My env is named someRandomEnv and you should change it to whatever you would like
  $ CONDA_SUBDIR=osx-64 conda create -n someRandomEnv python=3.9
$ conda activate someRandomEnv
$ conda config --env --set subdir osx-64

• Step 3 - Install pyqt separately: Source
  Install pyqt5
  $ pip install pyqt5

• Step 4 - Install Napari: Source
  $ python -m pip install "napari[all]"

• Step 5 - Test:
  $ napari

[psobolewskiPhD]

Hi @chrislearns sorry to hear you had troubles with napari on your M1!
We actually updated the installation instructions for arm64 macOS, but until there is a release they are on the dev version of the docs:
https://napari.org/dev/tutorials/fundamentals/installation.html

Basically, conda (or even better mamba) install from conda-forge channel should JustWork(tm). For pip, you need to install a Qt backend first, like pyqt from conda-forge and then pip install napari (without the all, see https://napari.org/dev/tutorials/fundamentals/installation.html#choosing-a-different-qt-backend)

Was there a specific package you needed that required using Rosetta2 emulation (x86 environment on arm64)?

[chrislearns]

@psobolewskiPhD no worries at all, I was just excited to have it working! So I believe I tried it and specifically I tried that section you mentioned with "Choose a different backend" which is how I arrived at Step 3, ultimately, but evenso I kept getting this error:

ImportError: dlopen(/Users/chrisguirguis/anaconda3/envs/napari-env/lib/python3.9/site-packages/vispy/visuals/text/_sdf_cpu.cpython-39-darwin.so, 0x0002): tried: '/Users/chrisguirguis/anaconda3/envs/napari-env/lib/python3.9/site-packages/vispy/visuals/text/_sdf_cpu.cpython-39-darwin.so' (mach-o file, but is an incompatible architecture (have 'x86_64', need 'arm64')), '/System/Volumes/Preboot/Cryptexes/OS/Users/chrisguirguis/anaconda3/envs/napari-env/lib/python3.9/site-packages/vispy/visuals/text/_sdf_cpu.cpython-39-darwin.so' (no such file), '/Users/chrisguirguis/anaconda3/envs/napari-env/lib/python3.9/site-packages/vispy/visuals/text/_sdf_cpu.cpython-39-darwin.so' (mach-o file, but is an incompatible architecture (have 'x86_64', need 'arm64'))

So that was when I tried looking for a solution and found what became Step 1 from above.

[chrislearns]

@psobolewskiPhD honestly, this may be a bit ignorant on my part, but I'm not entirely sure regarding anaconda for env vs for installing. The whole anaconda/conda-forge/mamba part of this has always been somewhat confusing for me. I just went through the exact 5 steps above to get it working.

At some point I did try the mamba thing you are mentioning because I saw your post here and my computer really didn't like it. I'd have to go back and dig to find the errors but I ended up having to uninstall everything including anaconda and try again from scratch, which is when I tried the steps above.

Thank you again! It's nice to see an active community - I kept seeing your name pop up too in different places!

[psobolewskiPhD]

Ah what I meant was just using conda to manage your environments and then pip to install everything:

conda create ...
conda activate ...
pip install ...

vs. using conda to install as well, e.g.:

conda create ...
conda activate ...
conda install napari

In the latter approach you are installing binaries from conda, which can come from default channel (anaconda) or from conda-forge which is a community run distribution (see: https://conda-forge.org/docs/user/introduction.html )
In the former approach, you get python from conda, but then install package as wheels (binaries) from pypi if available, else from source, also from pypi.
My experience on arm64 has been that wheels of dependencies can be lacking on pypi, while building from source isn't always reliable, so installing binaries from conda-forge using conda/mamba tends to be the simplest solution.

You can use:
conda info
to see what channels you're using (sources for the binaries). For example here's my miniforge3 (miniconda from conda-forge):

     active environment : base
    active env location : /Users/piotrsobolewski/Dev/miniforge3
            shell level : 1
       user config file : /Users/piotrsobolewski/.condarc
 populated config files : /Users/piotrsobolewski/Dev/miniforge3/.condarc
                          /Users/piotrsobolewski/.condarc
          conda version : 23.3.1
    conda-build version : not installed
         python version : 3.10.8.final.0
       virtual packages : __archspec=1=arm64
                          __osx=13.3.1=0
                          __unix=0=0
       base environment : /Users/piotrsobolewski/Dev/miniforge3  (writable)
      conda av data dir : /Users/piotrsobolewski/Dev/miniforge3/etc/conda
  conda av metadata url : None
           channel URLs : https://conda.anaconda.org/conda-forge/osx-arm64
                          https://conda.anaconda.org/conda-forge/noarch
          package cache : /Users/piotrsobolewski/Dev/miniforge3/pkgs
                          /Users/piotrsobolewski/.conda/pkgs
       envs directories : /Users/piotrsobolewski/Dev/miniforge3/envs
                          /Users/piotrsobolewski/.conda/envs
               platform : osx-arm64
             user-agent : conda/23.3.1 requests/2.27.1 CPython/3.10.8 Darwin/22.4.0 OSX/13.3.1
                UID:GID : 501:20
             netrc file : None
           offline mode : False

You can manually specify channels when installing, e.g.:
conda install -c conda-forge napari

There can be some binary incompatibilities (https://conda-forge.org/docs/user/tipsandtricks.html) when mixing channels, so one can set channel priorities to conda-forge:

conda config --add channels conda-forge 
conda config --set channel_priority strict 

[chrislearns]

@psobolewskiPhD ahh yes! Okay so I initially tried installing via conda install napari but it stalled for a long time on the solving part of the process, so I went through and tried changing the solver as mentioned here to libmamba solver and again I tried even using mamba but that ended up running into another slew of issues based around an error that was something like this: zstandard could not be imported. running without .conda support.

Now, to be clear, I definitely acknowledge that some of the struggle is due to my own ignorance in this realm. But your tips above are good to have. And I may try reinstalling using some of what you said above on another M1 to see how it does.

[psobolewskiPhD]

I suspect that issue may have been related to conda-forge vs. default channels.

And you are certainly not alone in this, I think the conda/anaconda/conda-forge/pip interactions are all non-obvious and a major issue.
We listed a few solutions here, if you scroll down to conda-forge and the Note
https://napari.org/dev/tutorials/fundamentals/installation.html#install-as-python-package-recommended

I think we should maybe add an admonition regarding anaconda/miniconda and conda-forge.

If you have a chance, I would be curious using your arm64 anaconda if you could try:

conda create -y -n napari-CF -c conda-forge python=3.9 napari

You may need to use the libmamba solver though to get the install to resolve...

[psobolewskiPhD]

It works with the classic solver, but it takes ~10 minutes. With mamba it's ~1 min.