Migration to npe2#

This document details how to convert a plugin using the first generation napari-plugin-engine, to the new npe2 format.

The primary difference between the first generation and second generation plugin system relates to how napari discovers plugin functionality. In the first generation plugin engine, napari had to import plugin modules to search for hook implementations decorated with @napari_hook_implementation. In npe2, plugins declare their contributions statically with a manifest file.

Migrating using the npe2 command line tool#

npe2 provides a command line interface to help convert a napari-plugin-engine-style plugin.

1. Install npe2#

python -m pip install npe2

The npe2 command line tool provides a few commands to help you develop your plugin. In this case we’re going to use convert to modify a repository to fit the new pattern:

$ npe2 convert --help

Usage: npe2 convert [OPTIONS] PATH

  Convert first generation napari plugin to new (manifest) format.

Arguments:
  PATH  Path of a local repository to convert (package must also be installed
        in current environment). Or, the name of an installed package/plugin.
        If a package is provided instead of a directory, the new manifest will
        simply be printed to stdout.  [required]


Options:
  -n, --dry-runs  Just print manifest to stdout. Do not modify anything
                  [default: False]

  --help          Show this message and exit.

2. Ensure your plugin is intalled in your environment#

This step is critical: your first-generation plugin must be installed in your currently active environment for npe2 convert to find it.

Typically this will look something like:

conda activate your-env
cd path/to/your/plugin/repository
python -m pip install -e .

If npe2 convert cannot import your plugin, you will likely get an error like:

PackageNotFoundError: We tried hard! but could not detect a plugin named '...'

3. Convert your plugin with npe2 convert#

Warning

Executing npe2 convert . will modify the current directory!

The npe2 convert command will:

  1. Inspect your plugin for hook implementations, and generate an npe2-compatible manifest file, called napari.yaml.

  2. Modify your setup.cfg to use the new napari.manifest entry point, and include the manifest file in your package data.

Use the npe2 convert command, passing a path to a plugin repository (here, the current directory .)

# convert the current directory npe2 convert .
✔  Conversion complete!
New manifest at /Users/talley/Desktop/napari-animation/napari_animation/napari.yaml.
If you have any napari_plugin_engine imports or hook_implementation decorators, you may remove them now.

You are encouraged to inspect the newly-generated napari.yaml file. Refer to the manifest and contributions references pages for details on each field in the manifest.

Note

In some cases the conversion tool may not be able to completely convert your plugin. Notable cases include:

  • multi-layer writers using the napari_get_writer hook specification

  • locally scoped functions returned from napari_experimental_provide_function. All command contributions must have global python_paths.

Feel free to contact us on zulip or github if you need help converting!.

Now, update the local package metadata by repeating:

> python -m pip install -e .

The next time napari is run, your plugin should be discovered as an npe2 plugin.


Migration Reference#

This section goes into detail on the differences between first-generation and second-generation implementations. In many cases, this will be more detail than you need. If you are still struggling with a specific conversion after using npe2 convert and reading the contributions reference and guides, this section may be of help.

Existing napari-plugin-engine plugins expose functionality via hook implementations. These are functions decorated to indicate they fullfil a hook specification described by napari. Though there are some exceptions, most hook implementations can be straightforwardly mapped to npe2 contributions

npe2 provides a command-line tool that will generate plugin manifests by inspecting exposed hook implementations. Below, we will walk through the kinds of migrations npe2 convert helps with.

For each type of hook specification there is a corresponding section below with migration tips. Each lists the hook specifications that are relevant to that section and an example manifest. For details, refer to the Contributions references.

Readers#

Functions that acted as napari_get_reader hooks can be bound directly as the command for an npe2 reader.

napari_hook_spec#

def napari_get_reader(path: Union[str, List[str]]) -> Optional[ReaderFunction]

npe2 contributions#

name: napari
contributions:
  commands:
    - id: napari.get_reader
      python_name: napari.plugins._builtins:napari_get_reader
      title: Read data using napari's builtin reader
  readers:
    - command: napari.get_reader
      accepts_directories: true
      filename_patterns: ["*.csv", "*.npy"]

Writers: Single-layer writers#

Functions that act as single-layer writers like napari_write_image hooks can be bound directly as the command for an npe2 writer. The layer constraint(layer_types) and filename_extensions fields need to be populated.

Since these writers handle only one layer at a time, the layer_type is straightforward: ['image'] for an image writer, ['points'] for a point-set writer, etc.

The list of filename_extensions is used to determine how the writer is presented in napari’s “Save As” dialog.

napari_hook_spec#

def napari_write_image(path: str, data: Any, meta: dict) -> Optional[str]
def napari_write_labels(path: str, data: Any, meta: dict) -> Optional[str]
def napari_write_points(path: str, data: Any, meta: dict) -> Optional[str]
def napari_write_shapes(path: str, data: Any, meta: dict) -> Optional[str]
def napari_write_surfaces(path: str, data: Any, meta: dict) -> Optional[str]
def napari_write_vectors(path: str, data: Any, meta: dict) -> Optional[str]

Example npe2 contribution#

name: napari_svg
display_name: napari svg
contributions:
  commands:
    - id: napari_svg.write_image
      python_name: napari_svg.hook_implementations:napari_write_image
      title: Write Image as SVG
  writers:
    - command: napari_svg.write_image
      layer_types: ["image"]
      filename_extensions: [".svg"]

Writers: Multi-layer writers#

In npe2, the writer specification declares what file extensions and layer types are compatible with a writer. This is a departure from the behavior of the napari_get_writer which was responsible for rejecting data that was incompatible.

Usually, the npe2 writer command should be bound to one of the functions returned by napari_get_writer. From the example below, this is the writer function.

When migrating, you’ll need to fill out the layer_types and filename_extensions used by your writer. layer_types is a set of constraints describing the combinations of layer types acceptable by this writer. More about layer types can be found in the Writer contribution guide.

In the example below, the svg writer accepts a set of layers with 0 or more images, and 0 or more label layers, and so on. It will not accept surface layers, so if any surface layer is present this writer won’t be invoked.

Because layer type constraints are specified in the manifest, no plugin code has to be imported or run until a compatible writer is found.

napari_hook_spec#

def napari_get_writer(
    path: str, layer_types: List[str]
) -> Optional[WriterFunction]

Where the WriterFunction is something like:

def writer(
    path: str, layer_data: List[Tuple[Any, Dict, str]]
    ) -> List[str]

Example npe2 contribution#

name: napari_svg
display_name: napari SVG
entry_point: napari_svg
contributions:
  commands:
    - id: napari_svg.svg_writer
      title: Write SVG
      python_name: napari_svg.hook_implementations:writer
  writers:
    - command: napari_svg.svg_writer
      layer_types: ["image*", "labels*", "points*", "shapes*", "vectors*"]
      filename_extensions: [".svg"]

Widgets#

napari_experimental_provide_dock_widget hooks return another function that can be used to instantiate a widget and, optionally, arguments to be passed to that function.

In contrast the callable for an npe2 widget contribution is bound to the function actually instantiating the widget. It accepts only one argument: a napari Viewer proxy instance. The proxy restricts access to some Viewer functionality like private methods.

Similarly napari_experimental_provide_function hooks return ane or more functions to be wrapped with magicgui. In npe2, each of these functions should be added as a Command contribution with an associated Widget contribution. For each of these Widget contributions, the manifest autogenerate: true flag should be set so that npe2 knows to use magicgui.

napari_hook_spec#

def napari_experimental_provide_dock_widget() -> Union[
    AugmentedWidget, List[AugmentedWidget]
]

or

def napari_experimental_provide_function() -> Union[
    FunctionType, List[FunctionType]
]

Example npe2 contribution#

Dock Widget

name: napari-animation
display_name: animation
contributions:
  commands:
    - id: napari-animation.widget
      python_name: napari_animation._qt:AnimationWidget
      title: Make animation widget
  widgets:
    - command: napari_animation.widget
      display_name: Wizard

Function widget

name: my-function-plugin
display_name: My function plugin!
contributions:
  commands:
    - id: my-function-plugin.func
      python_name: my_function_plugin:my_typed_function
      title: Open widget for my function
  widgets:
    - command: my-function-plugin.func
      display_name: My function
      autogenerate: true # <-- will wrap my_typed_function with magicgui

Sample data providers#

Each sample returned from napari_provide_sample_data() should be bound as an individual sample data contribution.

napari_hook_spec#

def napari_provide_sample_data() -> Dict[str, Union[SampleData, SampleDict]]

Example npe2 contribution#

This example sample data provider:

def _generate_random_data(shape=(512, 512)):
    data = np.random.rand(*shape)
    return [(data, {'name': 'random data'})]

@napari_hook_implementation
def napari_provide_sample_data():
    return {
        'random data': _generate_random_data,
        'random image': 'https://picsum.photos/1024',
        'sample_key': {
            'display_name': 'Some Random Data (512 x 512)'
            'data': _generate_random_data,
        }
    }

Should be migrated to:

name: my-plugin
contributions:
  commands:
    - id: my-plugin.random
      title: Generate random data
      python_name: my_plugin:_generate_random_data
  sample_data:
    - display_name: Some Random Data (512 x 512)
      key: random data
      command: my-plugin.random
    - uri: https://picsum.photos/1024
      key: random image

Themes#

napari_experimental_provide_theme() hooks return a dictionary of theme properties. These properties can be directly embedded in npe2 theme contributions. This allows napari to read the theming data without running any code in the plugin package!

Example#

The theme provided by this hook:

def get_new_theme() -> Dict[str, Dict[str, Union[str, Tuple, List]]:
    # specify theme(s) that should be added to napari
    themes = {
        "super_dark": {
            "name": "super_dark",
            "background": "rgb(12, 12, 12)",
            "foreground": "rgb(65, 72, 81)",
            "primary": "rgb(90, 98, 108)",
            "secondary": "rgb(134, 142, 147)",
            "highlight": "rgb(106, 115, 128)",
            "text": "rgb(240, 241, 242)",
            "icon": "rgb(209, 210, 212)",
            "warning": "rgb(153, 18, 31)",
            "current": "rgb(0, 122, 204)",
            "syntax_style": "native",
            "console": "rgb(0, 0, 0)",
            "canvas": "black",
        }
    }
    return themes

becomes this theme contribution in the plugin manifest:

name: my-plugin
contributions:
  themes:
    - label: Super dark
      id: super_dark
      type: dark
      colors:
        background: "rgb(12, 12, 12)"
        foreground: "rgb(65, 72, 81)"
        primary: "rgb(90, 98, 108)"
        secondary: "rgb(134, 142, 147)"
        highlight: "rgb(106, 115, 128)"
        text: "rgb(240, 241, 242)"
        icon: "rgb(209, 210, 212)"
        warning: "rgb(153, 18, 31)"
        current: "rgb(0, 122, 204)"
        syntax_style: "native"
        console: "rgb(0, 0, 0)"
        canvas: "black"

The npe2 adapter#

Starting with napari v0.4.17+ napari will begin allowing users to use their npe1 plugins as if they were npe2 plugins using a “npe1 -> npe2 adaptor” (this option is at Preferences > Plugins > Use npe2 adaptor)

When this option is enabled and a plugin using the legacy plugin manager API is loaded for the first time, the plugin will be imported as usual and contributions will be discovered. A “shim” npe2 manifest representing the plugin’s contributions will be created and cached locally. On all future launches of napari, that cached manifest will be used and the plugin will not be imported immediately when napari boots.

Benefits#

The benefits for an end-user opting in to the npe2 adaptor are:

  • dramatically reduced time to load napari. By avoiding importing all plugins at launch, napari can boot significantly faster.

  • Plugins are imported lazily, only after one of their commands or menu items has been requested.

  • It will become clearer to the user which plugin has errored (if any), since it will only occur when the plugin’s functionality has been requested.

  • For napari, the internal codebase becomes much simpler since only the npe2 API needs to be used.

Caveats#

There are a couple npe1 features that are no longer supported in npe2, and the following things will be “ignored” for a user using the npe2 adaptor:

  • The “plugin sort order” preference is deprecated, and will not be used when loading npe2 plugins or npe1 plugins loaded with the npe2 adaptor.

  • arguments for add_dock_widget returned from napari_experimental_provide_dock_widget (such as area= or add_vertical_stretch=) will no longer do anything: area will always be 'right' and add_vertical_stretch will always be True.

There is nothing a plugin can do to prevent a user from using the npe2 adaptor, it is a user decision. Furthermore, as we deprecate the legacy napari-plugin-engine API, the npe2 adapter will likely become the only way that npe1 plugins are supported in the future, and the option to not use the npe2 adaptor will be removed.