Test and Deploy

Tips for testing napari plugins

Testing is a big topic! If you are completely new to writing tests in python, consider reading this post on Getting Started With Testing in Python

We recommend using pytest for testing your plugin. Aim for 100% test coverage!

The make_napari_viewer fixture

Testing a napari Viewer requires some setup and teardown each time. We have created a pytest fixture called make_napari_viewer that you can use (this requires that you have napari installed in your environment).

To use a fixture in pytest, you simply include the name of the fixture in the test parameters (oddly enough, you don’t need to import it!). For example, to create a napari viewer for testing:

def test_something_with_a_viewer(make_napari_viewer):
    viewer = make_napari_viewer()
    ...  # carry on with your test

Prefer smaller unit tests when possible

The most common issue people run into when designing tests for napari plugins is that they try to test everything as a full “integration test”, starting from the napari event or action that would trigger their plugin to do something. For example, let’s say you have a dock widget that connects a mouse callback to the viewer:

class MyWidget:
    def __init__(self, viewer: 'napari.Viewer'):
        self._viewer = viewer

        def _on_mouse_move(viewer, event):
            if 'Shift' in event.modifiers:

def napari_experimental_provide_dock_widget():
    return MyWidget

You might think that you need to somehow simulate a mouse movement in napari in order to test this, but you don’t! Just trust that napari will call this function with a Viewer and an Event when a mouse move has been made, and otherwise leave napari out of it.

Instead, focus on “unit testing” your code: just call the function directly with objects that emulate, or “mock” the objects that your function expects to receive from napari. You may also need to slightly reorganize your code. Let’s modify the above widget to make it easier to test:

class MyWidget:
    def __init__(self, viewer: 'napari.Viewer'):
        self._viewer = viewer
        # connecting to a method rather than a local function
        # makes it easier to test

    def _on_mouse_move(self, viewer, event):
        if 'Shift' in event.modifiers:

To test this, we can often just instantiate the widget with our own viewer, and then call the methods directly. As for the event object, notice that all we care about in this plugin is that it has a modifiers attribute that may or may not contain the string "Shift". So let’s just fake it!

class FakeEvent:
    modifiers = {'Shift'}

def test_mouse_callback(make_napari_viewer):
    viewer = make_napari_viewer()
    wdg = MyWidget(viewer)
    wdg._on_mouse_move(viewer, FakeEvent())
    # assert that what you expect to happen actually happened!

Preparing for release

To help users find your plugin, make sure to use the Framework :: napari classifier in your package’s core metadata. (If you used the cookiecutter, this has already been done for you.)

Once your package is listed on PyPI (and includes the Framework :: napari classifier), it will also be visible on the napari hub. To ensure you are providing the relevant metadata and description for your plugin, see the following documentation in the napari hub GitHub’s docs folder:

The hub

For more about the napari hub, see the napari hub About page. To learn more about the hub’s development process, see the napari hub GitHub’s Wiki.

If you want your plugin to be available on PyPI, but not visible on the napari hub, you can add a .napari/config.yml file to the root of your repository with a visibility key. For details, see the customization guide.

Finally, once you have curated your package metadata and description, you can preview your metadata, and check any missing fields using the napari hub preview page service. Check out this guide for instructions on how to set it up.


When you are ready to share your plugin, upload the Python package to PyPI after which it will be installable using pip install <yourpackage>, or (assuming you added the Framework :: napari classifier) in the builtin plugin installer dialog.

If you used the Cookiecutter template, you can also setup automated deployments on github for every tagged commit.

What about conda?

While you are free to distribute your plugin on anaconda cloud in addition to or instead of PyPI, the built-in napari plugin installer doesn’t currently install from conda. In this case, you may guide your users to install your package on the command line using conda in your readme or documentation.

A future version of napari and the napari stand-alone application may support directly installing from conda.

When you are ready for users, announce your plugin on the Image.sc forum.