GTK provides powerful widgets to display and edit lists of data. This document gives an overview over the concepts and how they work together to allow developers to implement lists.
Lists are intended to be used whenever developers want to display many objects in roughly the same way.
Lists are perfectly fine to be used for very short list of only 2 or 3 elements, but generally scale fine to millions of items. Of course, the larger the list grows, the more care needs to be taken to choose the right data structures to keep things running well.
Lists are meant to be used with changing data, both with the items itself changing as well as the list adding and removing items. Of course, they work just as well with static data.
These terms are used throughout the documentation when talking about lists and you should be aware of what they refer to. These are often generic terms that have a specific meaning in this context.
Views display data from a model. A model is a GListModel and models can be provided in 3 ways or combinations thereof:
Many list models implementations already exist. There are models that provide specific data, like GtkDirectoryList. And there are models like GListStore that allow building lists manually.
Last but not least, developers are encouraged to create their own GListModel implementations. The interface is kept deliberately small to make this easy.
The same model can be used in multiple different views and wrapped with multiple different models at once.
The elements in a model are called items. All items are GObjects.
Every item in a model has a position which is the unsigned integer that describes where in the model the item is located. The first item in a model is at position 0. The position of an item can of course change as other items are added or removed from the model.
It is important to be aware of the difference between items and positions because the mapping from position to item is not permanent, so developers should think about whether they want to track items or positions when working with models. Oftentimes some things are really hard to do one way but very easy the other way.
The other important part of a view is a factory. Each factory is a GtkListItemFactory implementation that takes care of mapping the items of the model to widgets that can be shown in the view.
The way factories do this is by creating a listitem for each item that is currently in use. Listitems are always GtkListItem objects. They are only ever created by GTK and provide information about what item they are meant to display.
Different factory implementations use various different methods to allow developers to add the right widgets to listitems and to link those widgets with the item managed by the listitem. Finding a suitable factory implementation for the data displayed, the programming language and development environment is an important task that can simplify setting up the view tremendously.
Views support selections via a selection model. A selection model is an implementation of the GtkSelectionModel interface on top of the GListModel interface that allows marking each item in a model as either selected or not selected. Just like regular models, this can be implemented either by implementing GtkSelectionModel directly or by wrapping a model with one of the GTK models provided for this purposes, such as GtkNoSelection or GtkSingleSelection.
The behavior of selection models - ie which items they allow selecting and what effect this has on other items - is completely up to the selection model. As such, single-selections, multi-selections or sharing selection state between different selection models and/or views is possible. The selection state of an item is exposed in the listitem via the “selected” property.
Views and listitems also support activation. Activation means that double clicking or pressing enter while inside a focused row will cause the view to emit and activation signal such as “activate”. This provides an easy way to set up lists, but can also be turned off on listitems if undesired.
Both selections and activation are supported among other things via widget actions. This allows developers to add widgets to their lists that cause selections to change or to trigger activation via the GtkActionable interface. For a list of all supported actions see the relevant documentation.
While for short lists it is not a problem to instantiate widgets for every item in the model, once lists grow to thousands or millions of elements, this gets less feasible. Because of this, the views only create a limited amount of listitems and recycle them by binding them to new items. In general, views try to keep listitems available only for the items that can actually be seen on screen.
While this behavior allows views to scale effortlessly to huge lists, it has a few implication on what can be done with views. For example, it is not possible to query a view for a listitem used for a certain position - there might not be one and even if there is, that listitem might soon be recycled for a new position.
It is also important that developers save state they care about in the item and do not rely on the widgets they created as those widgets can be recycled for a new position at any time causing any state to be lost.
Another important requirement for views is that they need to know which items are not visible so they can be recycled. Views achieve that by implementing the GtkScrollable interface and expecting to be placed directly into a GtkScrolledWindow.
Of course, if you are only using models with few items, this is
not important and you can treat views like any other widget. But
if you use large lists and your performance suffers, you should be
aware of this. Views also allow tuning the number of listitems
they create such as with
developers running into performance problems should definitely
study the tradeoffs of those and experiment with them.
GTK offers a wide variety of wrapping models which change or supplement an existing model (or models) in some way. But when it comes to storing your actual data, there are only a few ready-made choices available: GListStore and GtkStringList.
GListStore is backed by a balanced tree and has performance characteristics that are expected for that data structure. It works reasonably well for dataset sizes in the 1,000,000 range, and can handle insertions and deletions. It uses a cached iter to make linear access to the items fast.
GtkStringList is not a general store - it can only handle strings.
It is backed by an dynamically allocated array and has performance
characteristics that are expected for that data structure.
GtkStringList is a good fit for any place where you would
char* and works best if the
dataset is not very dynamic.
If these models don’t fit your use case or scalability requirements, you should make a custom GListModel. It is a small interface and not very hard to implement.
For asymptotic performance comparisons between tree- and array-based implementations, see this article.
While GtkTreeView provided built-in support for trees, the list widgets, and in particular GListModel do not. This was a design choice because the common use case is displaying lists and not trees and it greatly simplifies the API interface provided.
However, GTK provides functionality to make trees look and behave like lists for the people who still want to display lists. This is achieved by using the GtkTreeListModel model to flatten a tree into a list. The GtkTreeExpander widget can then be used inside a listitem to allow users to expand and collapse rows and provide a similar experience to GtkTreeView.
Developers should refer to those objects’ API reference for more discussion on the topic.
One of the advantages of the new list widgets over GtkTreeViews and cell renderers is that they are fully themable using GTK CSS. This provides a lot of flexibility. The themes that ship with GTK provide a few predefined list styles that can be used in many situations:
This style of list is low density, spacious and uses an outline focus ring. It is suitable for lists of controls, e.g. in preference dialogs or settings panels. Use the .rich-list style class.
This style of list is medium density, using a full background to indicate focus and selection. Use the .navigation-sidebar style class.
This style of list is a high density table, similar in style to a traditional treeview. Individual cells can be selectable and editable. Use the .data-table style class.
Developers familiar with GtkTreeView may wonder how this way of doing lists compares to the way they know. This section will try to outline the similarities and differences between the two.
This new approach tries to provide roughly the same functionality as the old approach but often uses a very different approach to achieve these goals.
The main difference and one of the primary reasons for this new development is that items can be displayed using regular widgets and GtkCellRenderer is no longer necessary. This allows all benefits that widgets provide, such as complex layout and animating widgets and not only makes cell renderers obsolete, but also GtkCellArea.
The other big difference is the massive change to the data model. GtkTreeModel was a rather complex interface for a tree data structure and GListModel was deliberately designed to be a simple data structure for lists only. (See above) for how to still do trees with this new model.) Another big change is that the new model allows for bulk changes via the “items-changed” signal while GtkTreeModel only allows a single item to change at once. The goal here is of course to encourage implementation of custom list models.
Another consequence of the new model is that it is now easily possible to refer to the contents of a row in the model directly by keeping the item, while GtkTreeRowReference was a very slow mechanism to achieve the same. And because the items are real objects, developers can make them emit change signals causing listitems and their children to update, which wasn’t possible with GtkTreeModel.
The selection handling is also different. While selections used to be managed via custom code in each widget, selection state is now meant to be managed by the selection models. In particular this allows for complex use cases with specialized requirements.
Finally here’s a quick list of equivalent functionality to look for when transitioning code for easy lookup:
|GtkTreePath||guint position, GtkTreeListRow|