NAP-6 — Contributable Menus

Author

Draga Doncila Pop <ddoncila@gmail.com>

Created

2022-12-02

Resolution

TBD

Resolved

TBD

Status

Draft

Type

Standards Track

Version effective

TBD

Abstract

Since the initial release of npe2 infrastructure has been in place for plugin developers to declare menu contributions. These contributions add new items to menus in napari that have been deemed contributable.

Until now, only the layer context menu (available through right clicking on a layer in the layer list) has been contributable, but much discussion has occurred on the list of menus we wish to open up for contribution and the guiding principles behind the organization of this list.

This NAP defines an overall structure for contributable menus, an initial list of contributable menus that are to be opened up for plugin developers, and a process for users and plugin developers to propose new contributable menus to be added to the existing list.

Motivation and Scope

Currently plugin developers can provide processing and analysis extensions strictly through dock widget contributions. These are exposed to the user under the Plugins menu, either directly at the top level if a plugin only provides a single widget, or in a submenu labelled with the plugin’s display name when a plugin provides multiple widgets.

This Plugins menu quickly becomes difficult to parse with increasing number of plugins installed in an environment, and does not provide sufficient structure for a user to be able to quickly and coherently navigate through the extensions available to them.

A napari viewer with many plugins installed quickly becomes unwieldy.

The vast majority of plugins (217 out of 263) available today provide at least one widget contribution. Of these, 140 provide just a single widget 19 provide more than five widgets and 9 provide more than 10. This means that while it is important to provide structure within an individual plugin’s widgets, we must also provide cross-plugin structure so that users with many plugins installed can find widgets by the action they want to perform, rather than by hunting across endless plugin submenus, or attempting to discern what a plugin’s widget might do from its title.

Without meaningful places to put their contributions, plugin developers are coming up with their own way to organize contributions, whether through numbering widgets in their menu, mangling names to achieve a certain order, or coming up with their own unsupported solutions for adding new menu items.

The goal of this NAP, therefore, is to provide a structured set of contributable menus that is easy to navigate, semantically organized and intuitive for both users and plugin developers.

Note

It is highly likely that with growing numbers of plugins, widgets and menus, menu navigation itself becomes more burdensome when hunting for a specific action. Searchability of menu items is not within scope for this NAP, but will be made available to users via a command palette.

What is a Menu Contribution?

A MenuItem contribution in the npe2 manifest adds a new item to one of the napari menus (defined by an ID). When this item is clicked, the associated command is executed. Additionally, enablement clauses can be defined that control when this menu item is available for clicking and when it is disabled. Similarly, a when clause can be used to control whether the menu item is visible in the menu at all.

In addition to the menu items themselves, Submenu contributions can also be defined, which add a new submenu to a contributable menu which can be populated with new MenuItem contributions.

Currently, the only napari menu to which items can be contributed in this way is the layer context menu, accessible by right clicking on a layer in the LayerList as shown in the screenshot below. The new menu items and submenu are produced by the following code snippet:

name: napari-demo
display_name: Demo plugin

contributions:
  commands:
    - id: napari-demo.menu_item
      title: A new menu item
      python_name: napari_demo:menu_item
    - id: napari-demo.submenu_item
      title: A new item in a submenu
      python_name: napari_demo:submenu_item

  menus:
    napari/layers/context:
      - submenu: context_submenu
      - command: napari-demo.submenu_item
    hello_world:
      - command: napari-demo.menu_item

  submenus:
    - id: context_submenu
      label: A new submenu

This NAP proposes new menu IDs and new top level menus to open for contribution.

Right click layer context menu with new menu item and submenu contributed by a plugin

What do Menu Contributions do?

MenuItem contributions can be thought of as auxiliary contributions that provide a dispatch mechanism for other existing contributions. Currently these would strictly be widget contributions, but this mechanism can easily be extended to other commands, which can take as input napari objects like specific layers, or the Viewer, and produce output the Viewer uses - currently this would be new layers.

Moving forward, new contribution types could be defined that allow plugin developers to run context aware commands that interact with different Viewer components without the need for a widget.

For example, LayerEditor contributions could take the currently active/selected Layers and edit the underlying data, while LayerGenerator contributions could take the same input and create new layers in the viewer. By providing dedicated contributions for such actions, napari can enforce rules about layer editing and layer generation more strictly than for widget contributions, which, if taking the Viewer, can perform arbitrary actions upon all objects within the Viewer.

We therefore propose a menu structure that would be easily extensible with these new contribution types and provide intuitive locations for both plugin developers to add their functionality, and users to find it.

Detailed Description

We propose an initial set of contributable menus organized by the napari object being acted upon by the actions within the menu, and the likely output of those actions.

The Layers Menu

Currently the foremost example of such an object is the napari Layers, and this menu therefore contains five submenus organized by the types of processing the user may wish to perform on the selected Layer or Layers.

The Layers submenus are organized to give the user an immediate feeling of what might happen to their Layers as a result of clicking one of these menu items.

  1. Visualization - Items in this submenu allow you to generate visualizations from selected layer or layers. They do not change the layer data.

  2. Measure - Items in this submenu provide utilities for summarising information about your layer’s data.

  3. Edit - The items in this submenu change the data of your layer through Filters or Transformations. Additionally Annotation Tools provide a location for convenience layer editing tools e.g. Labels split/merge actions. Items in this submenu should not generate new layers, but rather act upon the existing layer data.

  4. Generate - Items in this submenu are the main analysis actions you can take on your layer. These items should add new layers to the viewer based on analyses and processing of data in your selected layer(s). The five proposed submenus are Projection, Segmentation, Classification, Registration and Tracks.

Many of the actions in this menu exist in the right click layer context menu. These items should be replicated in the Layers menu as needed, both to aid discoverability and to ensure users are not met with empty menus on initially opening napari. It’s also possible that exceedingly common operations e.g. thresholds, are provided in the future by napari itself.

The Acquisition Menu

In addition to the Layers menu, we add Acquisition as a top level menu.

Acquisition will contain widgets and utilities for interfacing with microscopes and other types of cameras.

The I/O Utilities Menu

A cursory analysis of widget names revealed a minimum of 17 plugins provide widgets dedicated to importing and exporting of data, features, models and/or other material supporting analysis.

These widgets usually require more choices from the user than are currently possible via the reader and writer interfaces. Although many discussions have been raised about expanding the opening and saving options in napari to support more complex choices (#1637, #2801, #4611, #4882…), we are not presently close to providing a unified interface for complex file opening/saving. Additionally, default napari opening and saving is entirely focused on reading data into layer, but there are many other reasons a user may wish to read a file or save some output from the Viewer.

It is likely, therefore, that some plugins will always have bespoke interfaces for importing and exporting various file formats. These interfaces will be exposed via the new File->I/O Utilities menu.

Plugin Submenus

The goal of the newly proposed menus is to provide a natural place where generally applicable actions can be semantically organized and easy to locate. However, many napari plugins contain an assortment of highly specialized widgets (that often interact with each other) that support highly specific, and sometimes ordered, workflows and analyses.

It may never make sense for such plugins to distribute their widgets across the different napari menus, particularly when they are designed to work in concert on specific data formats or layer types.

We therefore give plugin developers full control over their own submenu under Plugins->My Plugin. Plugin developers can organize all contributions under this submenu as they see fit, including adding their own submenus of arbitrary depth.

Complete Set of Proposed Contributable napari Menus

File
├─ ...
├─ IO Utilities
Layers
├─ Visualization
├─ Edit
│  ├─ Annotation Tools
│  ├─ Filter
│  ├─ Transform
├─ Measure
├─ Generate
│  ├─ Registration
│  ├─ Projection
│  ├─ Segmentation
│  ├─ Tracks
│  ├─ Classification
Acquisition

As a case study, we take four plugins offering between 9 and 14 widget contributions and arrange their widgets in these menus: empanada-napari, napari-stracking, napari-mm3 and napari-clemreg. Where a plugin’s widgets don’t naturally fit into one of the proposed menus, they are left in the plugin’s own submenu. Note that we have arranged these widgets purely based on title and cursory inspection of the documentation, so this should not be considered a concrete proposal for the structure of these plugins.

File
├─ ...
├─ IO Utilities
│  ├─ Export 2D segmentations (empanada-napari)
│  ├─ Store training dataset (empanada-napari)
│  ├─ nd2ToTIFF (napari-mm3)
Layers
├─ Visualization
│  ├─ SFilterTrack (napari-stracking)
├─ Edit
│  ├─ Annotation Tools
│  │  ├─ Merge Labels (empanada-napari)
│  │  ├─ Delete Labels (empanada-napari)
│  │  ├─ Split Labels (empanada-napari)
│  │  ├─ Jump to label (empanada-napari)
│  │  ├─ Find next available label (empanada-napari)
│  │  ├─ Pick training patches (empanada-napari)
│  ├─ Filter
│  ├─ Transform
│  │  ├─ make_image_warping (napari-clemreg)
├─ Measure
│  ├─ SParticlesProperties (napari-stracking)
│  ├─ STracksFeatures (napari-stracking)
├─ Generate
│  ├─ Registration
│  ├─ Projection
│  ├─ Segmentation
│  │  ├─ 2D Inference (empanada-napari)
│  │  ├─ 3D Inference (empanada-napari)
│  │  ├─ make_log_segmentation (napari-clemreg)
│  │  ├─ make_clean_binary_segmentation (napari-clemreg)
│  │  ├─ SegmentOtsu (napari-mm3)
│  │  ├─ SegmentUnet (napari-mm3)
│  │  ├─ Foci (napari-mm3)
│  │  ├─ SDetectorDog (napari-stracking)
│  │  ├─ SDetectorDoh (napari-stracking)
│  │  ├─ SDetectorLog (napari-stracking)
│  │  ├─ SDetectorSeg (napari-stracking)
│  ├─ Tracks
│  │  ├─ SLinkerShortestPath (napari-stracking)
│  │  ├─ Tracks (napari-mm3)
│  ├─ Classification
│  ├─ make_point_cloud_sampling (napari-clemreg)
│  ├─ make_point_cloud_registration (napari-clemreg)
Acquisition
Plugins
├─ empanada-napari
│  ├─ Finetune a model
│  ├─ Train a model
│  ├─ Register a model
│  ├─ Get model info
├─ napari-stracking
│  ├─ SScale
│  ├─ SPipeline
├─ napari-mm3
│  ├─ Compile
│  ├─ PickChannels
│  ├─ Subtract
│  ├─ Annotate
│  ├─ Colors
├─ napari-clemreg
│  ├─ mask_roi
│  ├─ make_data_preprocessing
│  ├─ train_model
│  ├─ predict_from_model

Items that Don’t Fit?

Where a plugin developer feels that none of the submenus of a given menu are suitable for their purpose, they should add their item to the deepest applicable menu. For example, a widget that takes a layer and produces a new layer through random perturbations would not fit under Layers -> Generate -> Segmentation but could fit under Layers -> Generate.

Where a plugin developer feels no top level menu or submenu is suitable for their purpose, they should add their item to their own plugin’s submenu under Plugins -> Your Plugin, and consider requesting a new contributable menu via the process described below.

The top level menu bar in napari is not open for contribution, and new top level menus can only be added via the process described below.

Process for Expanding Set of Contributable Menus

New contributable menus or submenus will be added either following periodic analysis of all plugin contributions, or following user request upon core developer consensus.

Periodic Contribution Analysis

As the number of plugins, types of contributions and Viewer interactions grows, it is important that the set of contributable menus is periodically assessed to add new submenus as required.

Every 6 months to 1 year, the core developers will perform an analysis on the total set of menu item and widget contributions of all plugins, and derive new groupings to ensure that the length of each submenu remains managable.

For example, consider the Layers -> Generate -> Segmentation menu. If analysis of the plugin ecosystem reveals 40 different contributions for Watershed Segmentation, a new Watershed submenu would be added under Layers -> Generate -> Segmentation -> Watershed.

User Request

napari users or plugin developers can at any moment raise an issue on the napari repository to open up an existing menu for contribution or to add a new submenu to any of the currently contributable menus.

Core developers will assess the proposed menu/submenu based on its generality, the number of existing plugins that may contribute to this menu, whether the proposed menu is sufficiently meaningful to be immediately understandable by users and other plugin developers, and whether there is significant overlap with existing contributable menus.

Once core developer consensus on adding the menu item is achieved, a Pull Request will be raised opening up this menu for contribution. The user proposing the menu is not responsible for opening this Pull Request, though they may, if they wish.

Implementation

  • Add list of menus to napari/constants/_menus.py, warning for invalid IDs and new top level menus as per #5153

  • Implement menu contribution interface that allows plugins to refer to their own plugin submenu

  • Add functionality for opening widgets from a menu item

  • Expose menu contribution reference in npe2

  • Write menu contribution guide in npe2

Backward Compatibility

This work does not have any backward compatibility considerations for existing features.

In future, backward compatibility concerns could arise when a menu name/ID is changed in napari, or when napari removes a menu that was previously contributable.

In the first case, no change is required in plugin manifests, as napari can simply maintain logic for migrating old IDs to new ones.

If a contributable napari submenu is removed, this should be highlighted first by a deprecation warning. Once the submenu is deprecated, contributions that refer to this submenu should instead be placed in the higher level menu. For example, if Layers -> Generate -> Segmentation is removed, existing contributions referring to this ID will be placed under Layers -> Generate.

If a contributable napari top level menu is removed, this should be highlighted first by a deprecation warning. Before the submenu is deprecated, core developers must work to identify plugins that will need migration and aid the migration process by opening issues and PRs as required. Top level menus should not be removed without a clear migration guide of where these contributions should be placed in the future. After the menu is deprecated, contributions referring to this menu should raise a warning, and be placed in the plugin’s own submenu at the highest level.

Plugin contributions to non contributable menus will raise warnings and be placed in the plugin’s submenu at the top level.

Future Work

As mentioned above, a key feature to support rapid browsing for actions is the search functionality via the command palette. This is actively being worked on and is essential for navigation.

Once more contribution types are exposed for users, it’s important that users are aware why certain actions are disabled when the user doesn’t meet the requisite context declared in the contribution’s enablement clause. Since the syntax for declaring these contexts is strictly defined, we should be able to surface information to suers about what is required for the action to be enabled and functional. For example, an action could declare itself enabled only when a points layer and an image layer are selected. If the user has only selected an image layer, we could indicate the missing context to the user e.g. “Action takes a points and image layer, but no points layer is selected”.

A desired attribute of these menu items is that users always know what will happen when they click a menu item. Does a widget open? Is the layer edited? Is a new layer added to the Viewer? Once more contribution types are exposed, we should be able to either add this information as metadata in the manifest file, or infer it from return type annotations of contribution commands, and also expose this to the user.

Finally, the number of actions in each menu is heavily dependent on the plugins installed in the user’s environment. Given a complete set of contributable menus, we could dynamically inspect how many menu items each submenu contains, and group them appropriately for the user while limiting unnecessary depth. For example, if the user’s environment has six plugins installed that each provide a Watershed segmentation, we could display a Layers -> Generate -> Segmentation -> Watershed submenu. If the user has just one Watershed segmentation plugin installed, this submenu would not appear. This would require very careful design to ensure the user still knows what to expect when they load up the Viewer.

Alternatives

The main alternative is the proposed Tools menu from npe2 #161.

This is a single top level menu containing the same submenus as our proposed list, but organized roughly in order of when actions may be performed in a standard image processing workflow.

General feedback from the community and the core developers is that this menu structure, while mostly containing individual submenus that make sense:

  • is too long and therefore difficult to parse at a glance

  • does not give the user a good indication of what inputs an action takes and what its output will be

  • is not semantically structured and is rather just a one stop shop for “plugin stuff”

  • will be difficult to extend further in meaningful ways as we develop more complex viewer interactions and plugin contributions e.g. multi canvas

Discussion

  • May 8 2022: npe2 #160 is opened and merged during core dev hackathon. Allows arbitrary menu locations in npe2 to support plugins contributing to other plugins, etc. Validation would happen elsewhere.

  • May 8 2022: npe2 #161 is opened with almost instant approvals. Initial feedback is that it’s difficult for people to know the input/output of menu items, suggests creating a NAP. Complexity arises with desire to declare contributable menus but still allow plugins to contribute to other plugin’s menus.

  • Jun 2 2022: npe2 #161 After further discussion (on zulip and in PR), this schema is identified as potentialy too limiting and there is mention that #160 may need to be reverted. A NAP is once again suggested as this is an influential decision with lots of opinions.

  • Jun 13 2022: npe2 #161 is closed and #160 is reverted, with comment for follow up over in the napari repo.

  • Sep 30 2022: napari #5153 opened with same list as in npe2, minimal discussion and input.

  • Oct 28 2022: napari #5153 discussion on core devs zulip stream begins. Developers mostly agree on the inidividual menu items but don’t like how deep the Tools menu already is, and the lack of semantic meaning in its structure.

References and Footnotes

All NAPs should be declared as dedicated to the public domain with the CC0 license 1, as in Copyright, below, with attribution encouraged with CC0+BY 2.