Skip to content

brainglobe/napari-experimental

napari-experimental

License BSD-3 PyPI Python Version tests codecov napari hub

Overview

Features

This plugin allows napari layers to be organised into 'layer groups'. Groups can hold any number of layers, in addition to further sub-groups. This follows suggestions from the corresponding issue on the main napari repository.

Main features:

  • Creation of group layers
  • Drag and drop layers/groups to re-organise them
  • Sync changes in layer order from the plugin to the main napari LayerList
  • Toggle visibility of layers and entire groups through the 'eye' icon or right click menu

Ethos

The aim of this plugin is to provide an entirely separate way to interact with layers in napari. While using it, the main layer list should only be used to add/remove layers, with all re-ordering, re-naming etc done directly within the plugin itself.

To aid this, the plugin contains its own independent layer controls, as well as right click menus and other features.

Outlook

Ultimately, the goal of this plugin is to provide a way to experiment with group layers independent from the main napari codebase. Hopefully, parts of this plugin widget will be incorporated back into napari, replacing the existing LayerList.

Implementation Details

This section is a halfway point between full developer documentation and throwing docstrings in a contributor's face. We have been diligent in adding docstrings to our methods, however we recommend you first read about model/view programming and in particular tree models and views. We also recommend you read the docstrings for Groups and Nodes in the napari codebase, for context on how napari currently handles such structures. You may also want to review this article on composite design patterns.

The docstrings, plus the explanations in this section, should then afford you enough information into how the plugin is operating.

Additionally, the "enter debugger" button has been left within the plugin. This is intended for developer use and should be removed when the plugin is ready for release! Adding a breakpoint within the GroupLayerWidget._enter_debug method allows a developer to enter a debug context with the plugin's widget as self, whilst running napari.

Key Classes

The key classes implemented in this plugin are

  • GroupLayerNode, GroupLayer: These are the Python classes that provide the tree-structure that group layers require. We expand on these below.
  • QtGroupLayerControls, QtGroupLayerControlsContainer: These classes are used to build the "control box" that the plugin provides when selecting a layer within the plugin. For all intents and purposes it mimics the existing napari layer viewer context window, but also reacts when selecting a group layer.
  • QtGroupLayerModel, QtGroupLayerView: These subclass from the appropriate Qt abstract classes, and provide the model/tree infrastructure for working with GroupLayers. Beyond this, they do not contain any remarkable functionality beyond patching certain methods for consistency with the data being handled / displayed.

Groups and Nodes

GroupLayerNode (GLN) and GroupLayer (GL) inherit from the base napari classes Node and Group (respectively). The GLNs act as wrappers for Layers - each one tracks (a pointer to) one particular Layer through the .layer attribute and uses its information when rendering in the GUI. GLs allow us to organise GLNs into groups, but they themselves must also be instances of GLN. For this reason, the convention we have adopted is that GLs have the .layer attribute set to None (since they do not track an individual layer that is associated to them specifically). Furthermore, the is_group method is explicitly defined on both GLNs and GLs, which provides a way to distinguish between the two classes when traversing the tree structure.

It should also be noted that we have made the decision not to make GLNs inherit from Layer, nor make the existing Layer class subclass from Node.

  • The latter would require changes to core napari code, so is outside the scope of this plugin.
  • The former option then introduces event conflicts (within the constructors) when GL inherits from both Group and Layer. Again, fixing these issues requires either edits to the core napari codebase, or regurgitating a lot of the core napari codebase inside a new class for just a couple of line changes.
  • From a separation of concerns perspective, it is cleaner to have the group layer functionality impose a structure on top of the existing LayerList that napari tracks, rather than duplicate this information so it can be stored in the GL, and then have to deal with changes to one of the objects being reflected back on the "original" it came from.

Differences in Indexing Conventions

A key difference between a flat LayerList and our GL structure is how they are indexed, and how the structures are rendered in the napari viewer based on this indexing. LayerList is indexed from 0 ascending, and renders in the napari viewer with the layer at index 0 at the bottom of the LayerList. Layers with higher indices are rendered on top of those with a lower index, with the item assigned the highest index appearing at the top of the layer viewer window.

By contrast, GL uses NestedIndexes to track the position of GLNs; these are tuples of integers that can be of arbitrary length (so long as the GL has the structure to match), and should be interpreted as subsequent accesses of objects in the tree:

  • (0, 1) refers to the item at index 1 of the (sub)-GL at index 0 of the root GL.
  • (2,) refers to the item at index 2 of the root GL.

Note that an "item" can be a GLN or a GL (use the is_group method to distinguish if necessary)! Again, items are added to a GL using the lowest available index. GLs can also assign a flat index order to their elements by starting from the root tree and counting upwards from 0, descending into sub-GLs and exhausting their items when they are encountered (see GroupLayer.flat_index_order).

However GLs render with index 0 at the top of the view, and higher indices below them. In order to preserve the user's intuitive understanding of "layers higher up in the display appear on top of lower layers", it is thus necessary to pass the reversed order of layers back to the viewer after a rearrangement event in the GL viewer.

Todo

Desirable features

  • Expand right click menu options to include:
    • Add selected items to new group
    • Add selected items to existing group
    • All existing right-click options for layers from the main LayerList
  • Add independent buttons to add/remove layers to the plugin (this would make it fully independent of the main LayerList). Also, style the Add empty layer group button to match these.

Known bugs and issues (non-breaking, but poor UX)

  • The open/closed state of group layers doesn't persist on drag and drop #23

Known bugs and issues (breaking)

Development Tasks

  • Create a standalone docs site that expands on the implementation details section.
    • Remove said section and transfer its contents to the docs site
    • Link to the docs site in the README.

Installation

You can install napari-experimental via pip:

pip install napari-experimental

To install the latest development version:

pip install git+https://github.com/brainglobe/napari-experimental.git

You can also install a version of the package that uses the latest version of napari (fetched from https://github.com/napari/napari):

pip install napari-experimental[napari-latest]

Contributing

Contributions are very welcome. Tests can be run with tox, please ensure the coverage at least stays the same before you submit a pull request.

License

Distributed under the terms of the [BSD-3] license, "napari-experimental" is free and open source software

Issues

If you encounter any problems, please file an issue along with a detailed description.

About

No description, website, or topics provided.

Resources

License

Code of conduct

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages